[XFRM]: remove xerr_idxp from __xfrm_policy_check()
[linux-2.6.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 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;
782
783         again1:
784                 hlist_for_each_entry(pol, entry,
785                                      &xfrm_policy_inexact[dir], bydst) {
786                         if (pol->type != type)
787                                 continue;
788                         hlist_del(&pol->bydst);
789                         hlist_del(&pol->byidx);
790                         write_unlock_bh(&xfrm_policy_lock);
791
792                         xfrm_policy_kill(pol);
793
794                         write_lock_bh(&xfrm_policy_lock);
795                         goto again1;
796                 }
797
798                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
799         again2:
800                         hlist_for_each_entry(pol, entry,
801                                              xfrm_policy_bydst[dir].table + i,
802                                              bydst) {
803                                 if (pol->type != type)
804                                         continue;
805                                 hlist_del(&pol->bydst);
806                                 hlist_del(&pol->byidx);
807                                 write_unlock_bh(&xfrm_policy_lock);
808
809                                 xfrm_policy_kill(pol);
810
811                                 write_lock_bh(&xfrm_policy_lock);
812                                 goto again2;
813                         }
814                 }
815
816                 xfrm_policy_count[dir] = 0;
817         }
818         atomic_inc(&flow_cache_genid);
819         write_unlock_bh(&xfrm_policy_lock);
820 }
821 EXPORT_SYMBOL(xfrm_policy_flush);
822
823 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
824                      void *data)
825 {
826         struct xfrm_policy *pol;
827         struct hlist_node *entry;
828         int dir, count, error;
829
830         read_lock_bh(&xfrm_policy_lock);
831         count = 0;
832         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
833                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
834                 int i;
835
836                 hlist_for_each_entry(pol, entry,
837                                      &xfrm_policy_inexact[dir], bydst) {
838                         if (pol->type == type)
839                                 count++;
840                 }
841                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
842                         hlist_for_each_entry(pol, entry, table + i, bydst) {
843                                 if (pol->type == type)
844                                         count++;
845                         }
846                 }
847         }
848
849         if (count == 0) {
850                 error = -ENOENT;
851                 goto out;
852         }
853
854         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
855                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
856                 int i;
857
858                 hlist_for_each_entry(pol, entry,
859                                      &xfrm_policy_inexact[dir], bydst) {
860                         if (pol->type != type)
861                                 continue;
862                         error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
863                         if (error)
864                                 goto out;
865                 }
866                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
867                         hlist_for_each_entry(pol, entry, table + i, bydst) {
868                                 if (pol->type != type)
869                                         continue;
870                                 error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
871                                 if (error)
872                                         goto out;
873                         }
874                 }
875         }
876         error = 0;
877 out:
878         read_unlock_bh(&xfrm_policy_lock);
879         return error;
880 }
881 EXPORT_SYMBOL(xfrm_policy_walk);
882
883 /* Find policy to apply to this flow. */
884
885 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
886                              u8 type, u16 family, int dir)
887 {
888         struct xfrm_selector *sel = &pol->selector;
889         int match;
890
891         if (pol->family != family ||
892             pol->type != type)
893                 return 0;
894
895         match = xfrm_selector_match(sel, fl, family);
896         if (match) {
897                 if (!security_xfrm_policy_lookup(pol, fl->secid, dir))
898                         return 1;
899         }
900
901         return 0;
902 }
903
904 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
905                                                      u16 family, u8 dir)
906 {
907         struct xfrm_policy *pol, *ret;
908         xfrm_address_t *daddr, *saddr;
909         struct hlist_node *entry;
910         struct hlist_head *chain;
911         u32 priority = ~0U;
912
913         daddr = xfrm_flowi_daddr(fl, family);
914         saddr = xfrm_flowi_saddr(fl, family);
915         if (unlikely(!daddr || !saddr))
916                 return NULL;
917
918         read_lock_bh(&xfrm_policy_lock);
919         chain = policy_hash_direct(daddr, saddr, family, dir);
920         ret = NULL;
921         hlist_for_each_entry(pol, entry, chain, bydst) {
922                 if (xfrm_policy_match(pol, fl, type, family, dir)) {
923                         ret = pol;
924                         priority = ret->priority;
925                         break;
926                 }
927         }
928         chain = &xfrm_policy_inexact[dir];
929         hlist_for_each_entry(pol, entry, chain, bydst) {
930                 if (xfrm_policy_match(pol, fl, type, family, dir) &&
931                     pol->priority < priority) {
932                         ret = pol;
933                         break;
934                 }
935         }
936         if (ret)
937                 xfrm_pol_hold(ret);
938         read_unlock_bh(&xfrm_policy_lock);
939
940         return ret;
941 }
942
943 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
944                                void **objp, atomic_t **obj_refp)
945 {
946         struct xfrm_policy *pol;
947
948 #ifdef CONFIG_XFRM_SUB_POLICY
949         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
950         if (pol)
951                 goto end;
952 #endif
953         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
954
955 #ifdef CONFIG_XFRM_SUB_POLICY
956 end:
957 #endif
958         if ((*objp = (void *) pol) != NULL)
959                 *obj_refp = &pol->refcnt;
960 }
961
962 static inline int policy_to_flow_dir(int dir)
963 {
964         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
965             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
966             XFRM_POLICY_FWD == FLOW_DIR_FWD)
967                 return dir;
968         switch (dir) {
969         default:
970         case XFRM_POLICY_IN:
971                 return FLOW_DIR_IN;
972         case XFRM_POLICY_OUT:
973                 return FLOW_DIR_OUT;
974         case XFRM_POLICY_FWD:
975                 return FLOW_DIR_FWD;
976         };
977 }
978
979 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
980 {
981         struct xfrm_policy *pol;
982
983         read_lock_bh(&xfrm_policy_lock);
984         if ((pol = sk->sk_policy[dir]) != NULL) {
985                 int match = xfrm_selector_match(&pol->selector, fl,
986                                                 sk->sk_family);
987                 int err = 0;
988
989                 if (match)
990                   err = security_xfrm_policy_lookup(pol, fl->secid, policy_to_flow_dir(dir));
991
992                 if (match && !err)
993                         xfrm_pol_hold(pol);
994                 else
995                         pol = NULL;
996         }
997         read_unlock_bh(&xfrm_policy_lock);
998         return pol;
999 }
1000
1001 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1002 {
1003         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1004                                                      pol->family, dir);
1005
1006         hlist_add_head(&pol->bydst, chain);
1007         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1008         xfrm_policy_count[dir]++;
1009         xfrm_pol_hold(pol);
1010
1011         if (xfrm_bydst_should_resize(dir, NULL))
1012                 schedule_work(&xfrm_hash_work);
1013 }
1014
1015 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1016                                                 int dir)
1017 {
1018         if (hlist_unhashed(&pol->bydst))
1019                 return NULL;
1020
1021         hlist_del(&pol->bydst);
1022         hlist_del(&pol->byidx);
1023         xfrm_policy_count[dir]--;
1024
1025         return pol;
1026 }
1027
1028 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1029 {
1030         write_lock_bh(&xfrm_policy_lock);
1031         pol = __xfrm_policy_unlink(pol, dir);
1032         write_unlock_bh(&xfrm_policy_lock);
1033         if (pol) {
1034                 if (dir < XFRM_POLICY_MAX)
1035                         atomic_inc(&flow_cache_genid);
1036                 xfrm_policy_kill(pol);
1037                 return 0;
1038         }
1039         return -ENOENT;
1040 }
1041 EXPORT_SYMBOL(xfrm_policy_delete);
1042
1043 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1044 {
1045         struct xfrm_policy *old_pol;
1046
1047 #ifdef CONFIG_XFRM_SUB_POLICY
1048         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1049                 return -EINVAL;
1050 #endif
1051
1052         write_lock_bh(&xfrm_policy_lock);
1053         old_pol = sk->sk_policy[dir];
1054         sk->sk_policy[dir] = pol;
1055         if (pol) {
1056                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
1057                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1058                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1059         }
1060         if (old_pol)
1061                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1062         write_unlock_bh(&xfrm_policy_lock);
1063
1064         if (old_pol) {
1065                 xfrm_policy_kill(old_pol);
1066         }
1067         return 0;
1068 }
1069
1070 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1071 {
1072         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1073
1074         if (newp) {
1075                 newp->selector = old->selector;
1076                 if (security_xfrm_policy_clone(old, newp)) {
1077                         kfree(newp);
1078                         return NULL;  /* ENOMEM */
1079                 }
1080                 newp->lft = old->lft;
1081                 newp->curlft = old->curlft;
1082                 newp->action = old->action;
1083                 newp->flags = old->flags;
1084                 newp->xfrm_nr = old->xfrm_nr;
1085                 newp->index = old->index;
1086                 newp->type = old->type;
1087                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1088                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1089                 write_lock_bh(&xfrm_policy_lock);
1090                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1091                 write_unlock_bh(&xfrm_policy_lock);
1092                 xfrm_pol_put(newp);
1093         }
1094         return newp;
1095 }
1096
1097 int __xfrm_sk_clone_policy(struct sock *sk)
1098 {
1099         struct xfrm_policy *p0 = sk->sk_policy[0],
1100                            *p1 = sk->sk_policy[1];
1101
1102         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1103         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1104                 return -ENOMEM;
1105         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1106                 return -ENOMEM;
1107         return 0;
1108 }
1109
1110 /* Resolve list of templates for the flow, given policy. */
1111
1112 static int
1113 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1114                       struct xfrm_state **xfrm,
1115                       unsigned short family)
1116 {
1117         int nx;
1118         int i, error;
1119         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1120         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1121
1122         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1123                 struct xfrm_state *x;
1124                 xfrm_address_t *remote = daddr;
1125                 xfrm_address_t *local  = saddr;
1126                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1127
1128                 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1129                         remote = &tmpl->id.daddr;
1130                         local = &tmpl->saddr;
1131                 }
1132
1133                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1134
1135                 if (x && x->km.state == XFRM_STATE_VALID) {
1136                         xfrm[nx++] = x;
1137                         daddr = remote;
1138                         saddr = local;
1139                         continue;
1140                 }
1141                 if (x) {
1142                         error = (x->km.state == XFRM_STATE_ERROR ?
1143                                  -EINVAL : -EAGAIN);
1144                         xfrm_state_put(x);
1145                 }
1146
1147                 if (!tmpl->optional)
1148                         goto fail;
1149         }
1150         return nx;
1151
1152 fail:
1153         for (nx--; nx>=0; nx--)
1154                 xfrm_state_put(xfrm[nx]);
1155         return error;
1156 }
1157
1158 static int
1159 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1160                   struct xfrm_state **xfrm,
1161                   unsigned short family)
1162 {
1163         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1164         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1165         int cnx = 0;
1166         int error;
1167         int ret;
1168         int i;
1169
1170         for (i = 0; i < npols; i++) {
1171                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1172                         error = -ENOBUFS;
1173                         goto fail;
1174                 }
1175
1176                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1177                 if (ret < 0) {
1178                         error = ret;
1179                         goto fail;
1180                 } else
1181                         cnx += ret;
1182         }
1183
1184         /* found states are sorted for outbound processing */
1185         if (npols > 1)
1186                 xfrm_state_sort(xfrm, tpp, cnx, family);
1187
1188         return cnx;
1189
1190  fail:
1191         for (cnx--; cnx>=0; cnx--)
1192                 xfrm_state_put(tpp[cnx]);
1193         return error;
1194
1195 }
1196
1197 /* Check that the bundle accepts the flow and its components are
1198  * still valid.
1199  */
1200
1201 static struct dst_entry *
1202 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1203 {
1204         struct dst_entry *x;
1205         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1206         if (unlikely(afinfo == NULL))
1207                 return ERR_PTR(-EINVAL);
1208         x = afinfo->find_bundle(fl, policy);
1209         xfrm_policy_put_afinfo(afinfo);
1210         return x;
1211 }
1212
1213 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1214  * all the metrics... Shortly, bundle a bundle.
1215  */
1216
1217 static int
1218 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1219                    struct flowi *fl, struct dst_entry **dst_p,
1220                    unsigned short family)
1221 {
1222         int err;
1223         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1224         if (unlikely(afinfo == NULL))
1225                 return -EINVAL;
1226         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1227         xfrm_policy_put_afinfo(afinfo);
1228         return err;
1229 }
1230
1231
1232 static int stale_bundle(struct dst_entry *dst);
1233
1234 /* Main function: finds/creates a bundle for given flow.
1235  *
1236  * At the moment we eat a raw IP route. Mostly to speed up lookups
1237  * on interfaces with disabled IPsec.
1238  */
1239 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1240                 struct sock *sk, int flags)
1241 {
1242         struct xfrm_policy *policy;
1243         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1244         int npols;
1245         int pol_dead;
1246         int xfrm_nr;
1247         int pi;
1248         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1249         struct dst_entry *dst, *dst_orig = *dst_p;
1250         int nx = 0;
1251         int err;
1252         u32 genid;
1253         u16 family;
1254         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1255
1256 restart:
1257         genid = atomic_read(&flow_cache_genid);
1258         policy = NULL;
1259         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1260                 pols[pi] = NULL;
1261         npols = 0;
1262         pol_dead = 0;
1263         xfrm_nr = 0;
1264
1265         if (sk && sk->sk_policy[1])
1266                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1267
1268         if (!policy) {
1269                 /* To accelerate a bit...  */
1270                 if ((dst_orig->flags & DST_NOXFRM) ||
1271                     !xfrm_policy_count[XFRM_POLICY_OUT])
1272                         return 0;
1273
1274                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1275                                            dir, xfrm_policy_lookup);
1276         }
1277
1278         if (!policy)
1279                 return 0;
1280
1281         family = dst_orig->ops->family;
1282         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
1283         pols[0] = policy;
1284         npols ++;
1285         xfrm_nr += pols[0]->xfrm_nr;
1286
1287         switch (policy->action) {
1288         case XFRM_POLICY_BLOCK:
1289                 /* Prohibit the flow */
1290                 err = -EPERM;
1291                 goto error;
1292
1293         case XFRM_POLICY_ALLOW:
1294 #ifndef CONFIG_XFRM_SUB_POLICY
1295                 if (policy->xfrm_nr == 0) {
1296                         /* Flow passes not transformed. */
1297                         xfrm_pol_put(policy);
1298                         return 0;
1299                 }
1300 #endif
1301
1302                 /* Try to find matching bundle.
1303                  *
1304                  * LATER: help from flow cache. It is optional, this
1305                  * is required only for output policy.
1306                  */
1307                 dst = xfrm_find_bundle(fl, policy, family);
1308                 if (IS_ERR(dst)) {
1309                         err = PTR_ERR(dst);
1310                         goto error;
1311                 }
1312
1313                 if (dst)
1314                         break;
1315
1316 #ifdef CONFIG_XFRM_SUB_POLICY
1317                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1318                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1319                                                             fl, family,
1320                                                             XFRM_POLICY_OUT);
1321                         if (pols[1]) {
1322                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1323                                         err = -EPERM;
1324                                         goto error;
1325                                 }
1326                                 npols ++;
1327                                 xfrm_nr += pols[1]->xfrm_nr;
1328                         }
1329                 }
1330
1331                 /*
1332                  * Because neither flowi nor bundle information knows about
1333                  * transformation template size. On more than one policy usage
1334                  * we can realize whether all of them is bypass or not after
1335                  * they are searched. See above not-transformed bypass
1336                  * is surrounded by non-sub policy configuration, too.
1337                  */
1338                 if (xfrm_nr == 0) {
1339                         /* Flow passes not transformed. */
1340                         xfrm_pols_put(pols, npols);
1341                         return 0;
1342                 }
1343
1344 #endif
1345                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1346
1347                 if (unlikely(nx<0)) {
1348                         err = nx;
1349                         if (err == -EAGAIN && flags) {
1350                                 DECLARE_WAITQUEUE(wait, current);
1351
1352                                 add_wait_queue(&km_waitq, &wait);
1353                                 set_current_state(TASK_INTERRUPTIBLE);
1354                                 schedule();
1355                                 set_current_state(TASK_RUNNING);
1356                                 remove_wait_queue(&km_waitq, &wait);
1357
1358                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1359
1360                                 if (nx == -EAGAIN && signal_pending(current)) {
1361                                         err = -ERESTART;
1362                                         goto error;
1363                                 }
1364                                 if (nx == -EAGAIN ||
1365                                     genid != atomic_read(&flow_cache_genid)) {
1366                                         xfrm_pols_put(pols, npols);
1367                                         goto restart;
1368                                 }
1369                                 err = nx;
1370                         }
1371                         if (err < 0)
1372                                 goto error;
1373                 }
1374                 if (nx == 0) {
1375                         /* Flow passes not transformed. */
1376                         xfrm_pols_put(pols, npols);
1377                         return 0;
1378                 }
1379
1380                 dst = dst_orig;
1381                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1382
1383                 if (unlikely(err)) {
1384                         int i;
1385                         for (i=0; i<nx; i++)
1386                                 xfrm_state_put(xfrm[i]);
1387                         goto error;
1388                 }
1389
1390                 for (pi = 0; pi < npols; pi++) {
1391                         read_lock_bh(&pols[pi]->lock);
1392                         pol_dead |= pols[pi]->dead;
1393                         read_unlock_bh(&pols[pi]->lock);
1394                 }
1395
1396                 write_lock_bh(&policy->lock);
1397                 if (unlikely(pol_dead || stale_bundle(dst))) {
1398                         /* Wow! While we worked on resolving, this
1399                          * policy has gone. Retry. It is not paranoia,
1400                          * we just cannot enlist new bundle to dead object.
1401                          * We can't enlist stable bundles either.
1402                          */
1403                         write_unlock_bh(&policy->lock);
1404                         if (dst)
1405                                 dst_free(dst);
1406
1407                         err = -EHOSTUNREACH;
1408                         goto error;
1409                 }
1410                 dst->next = policy->bundles;
1411                 policy->bundles = dst;
1412                 dst_hold(dst);
1413                 write_unlock_bh(&policy->lock);
1414         }
1415         *dst_p = dst;
1416         dst_release(dst_orig);
1417         xfrm_pols_put(pols, npols);
1418         return 0;
1419
1420 error:
1421         dst_release(dst_orig);
1422         xfrm_pols_put(pols, npols);
1423         *dst_p = NULL;
1424         return err;
1425 }
1426 EXPORT_SYMBOL(xfrm_lookup);
1427
1428 static inline int
1429 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1430 {
1431         struct xfrm_state *x;
1432         int err;
1433
1434         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1435                 return 0;
1436         x = skb->sp->xvec[idx];
1437         if (!x->type->reject)
1438                 return 0;
1439         xfrm_state_hold(x);
1440         err = x->type->reject(x, skb, fl);
1441         xfrm_state_put(x);
1442         return err;
1443 }
1444
1445 /* When skb is transformed back to its "native" form, we have to
1446  * check policy restrictions. At the moment we make this in maximally
1447  * stupid way. Shame on me. :-) Of course, connected sockets must
1448  * have policy cached at them.
1449  */
1450
1451 static inline int
1452 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x, 
1453               unsigned short family)
1454 {
1455         if (xfrm_state_kern(x))
1456                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1457         return  x->id.proto == tmpl->id.proto &&
1458                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1459                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1460                 x->props.mode == tmpl->mode &&
1461                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1462                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1463                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1464                   xfrm_state_addr_cmp(tmpl, x, family));
1465 }
1466
1467 /*
1468  * 0 or more than 0 is returned when validation is succeeded (either bypass
1469  * because of optional transport mode, or next index of the mathced secpath
1470  * state with the template.
1471  * -1 is returned when no matching template is found.
1472  * Otherwise "-2 - errored_index" is returned.
1473  */
1474 static inline int
1475 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1476                unsigned short family)
1477 {
1478         int idx = start;
1479
1480         if (tmpl->optional) {
1481                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1482                         return start;
1483         } else
1484                 start = -1;
1485         for (; idx < sp->len; idx++) {
1486                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1487                         return ++idx;
1488                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1489                         if (start == -1)
1490                                 start = -2-idx;
1491                         break;
1492                 }
1493         }
1494         return start;
1495 }
1496
1497 int
1498 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1499 {
1500         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1501         int err;
1502
1503         if (unlikely(afinfo == NULL))
1504                 return -EAFNOSUPPORT;
1505
1506         afinfo->decode_session(skb, fl);
1507         err = security_xfrm_decode_session(skb, &fl->secid);
1508         xfrm_policy_put_afinfo(afinfo);
1509         return err;
1510 }
1511 EXPORT_SYMBOL(xfrm_decode_session);
1512
1513 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1514 {
1515         for (; k < sp->len; k++) {
1516                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1517                         *idxp = k;
1518                         return 1;
1519                 }
1520         }
1521
1522         return 0;
1523 }
1524
1525 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 
1526                         unsigned short family)
1527 {
1528         struct xfrm_policy *pol;
1529         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1530         int npols = 0;
1531         int xfrm_nr;
1532         int pi;
1533         struct flowi fl;
1534         u8 fl_dir = policy_to_flow_dir(dir);
1535         int xerr_idx = -1;
1536
1537         if (xfrm_decode_session(skb, &fl, family) < 0)
1538                 return 0;
1539         nf_nat_decode_session(skb, &fl, family);
1540
1541         /* First, check used SA against their selectors. */
1542         if (skb->sp) {
1543                 int i;
1544
1545                 for (i=skb->sp->len-1; i>=0; i--) {
1546                         struct xfrm_state *x = skb->sp->xvec[i];
1547                         if (!xfrm_selector_match(&x->sel, &fl, family))
1548                                 return 0;
1549                 }
1550         }
1551
1552         pol = NULL;
1553         if (sk && sk->sk_policy[dir])
1554                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1555
1556         if (!pol)
1557                 pol = flow_cache_lookup(&fl, family, fl_dir,
1558                                         xfrm_policy_lookup);
1559
1560         if (!pol) {
1561                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1562                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1563                         return 0;
1564                 }
1565                 return 1;
1566         }
1567
1568         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1569
1570         pols[0] = pol;
1571         npols ++;
1572 #ifdef CONFIG_XFRM_SUB_POLICY
1573         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1574                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1575                                                     &fl, family,
1576                                                     XFRM_POLICY_IN);
1577                 if (pols[1]) {
1578                         pols[1]->curlft.use_time = (unsigned long)xtime.tv_sec;
1579                         npols ++;
1580                 }
1581         }
1582 #endif
1583
1584         if (pol->action == XFRM_POLICY_ALLOW) {
1585                 struct sec_path *sp;
1586                 static struct sec_path dummy;
1587                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1588                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1589                 struct xfrm_tmpl **tpp = tp;
1590                 int ti = 0;
1591                 int i, k;
1592
1593                 if ((sp = skb->sp) == NULL)
1594                         sp = &dummy;
1595
1596                 for (pi = 0; pi < npols; pi++) {
1597                         if (pols[pi] != pol &&
1598                             pols[pi]->action != XFRM_POLICY_ALLOW)
1599                                 goto reject;
1600                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1601                                 goto reject_error;
1602                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1603                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1604                 }
1605                 xfrm_nr = ti;
1606                 if (npols > 1) {
1607                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1608                         tpp = stp;
1609                 }
1610
1611                 /* For each tunnel xfrm, find the first matching tmpl.
1612                  * For each tmpl before that, find corresponding xfrm.
1613                  * Order is _important_. Later we will implement
1614                  * some barriers, but at the moment barriers
1615                  * are implied between each two transformations.
1616                  */
1617                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1618                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1619                         if (k < 0) {
1620                                 if (k < -1)
1621                                         /* "-2 - errored_index" returned */
1622                                         xerr_idx = -(2+k);
1623                                 goto reject;
1624                         }
1625                 }
1626
1627                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1628                         goto reject;
1629
1630                 xfrm_pols_put(pols, npols);
1631                 return 1;
1632         }
1633
1634 reject:
1635         xfrm_secpath_reject(xerr_idx, skb, &fl);
1636 reject_error:
1637         xfrm_pols_put(pols, npols);
1638         return 0;
1639 }
1640 EXPORT_SYMBOL(__xfrm_policy_check);
1641
1642 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1643 {
1644         struct flowi fl;
1645
1646         if (xfrm_decode_session(skb, &fl, family) < 0)
1647                 return 0;
1648
1649         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1650 }
1651 EXPORT_SYMBOL(__xfrm_route_forward);
1652
1653 /* Optimize later using cookies and generation ids. */
1654
1655 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1656 {
1657         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1658          * to "-1" to force all XFRM destinations to get validated by
1659          * dst_ops->check on every use.  We do this because when a
1660          * normal route referenced by an XFRM dst is obsoleted we do
1661          * not go looking around for all parent referencing XFRM dsts
1662          * so that we can invalidate them.  It is just too much work.
1663          * Instead we make the checks here on every use.  For example:
1664          *
1665          *      XFRM dst A --> IPv4 dst X
1666          *
1667          * X is the "xdst->route" of A (X is also the "dst->path" of A
1668          * in this example).  If X is marked obsolete, "A" will not
1669          * notice.  That's what we are validating here via the
1670          * stale_bundle() check.
1671          *
1672          * When a policy's bundle is pruned, we dst_free() the XFRM
1673          * dst which causes it's ->obsolete field to be set to a
1674          * positive non-zero integer.  If an XFRM dst has been pruned
1675          * like this, we want to force a new route lookup.
1676          */
1677         if (dst->obsolete < 0 && !stale_bundle(dst))
1678                 return dst;
1679
1680         return NULL;
1681 }
1682
1683 static int stale_bundle(struct dst_entry *dst)
1684 {
1685         return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1686 }
1687
1688 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1689 {
1690         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1691                 dst->dev = &loopback_dev;
1692                 dev_hold(&loopback_dev);
1693                 dev_put(dev);
1694         }
1695 }
1696 EXPORT_SYMBOL(xfrm_dst_ifdown);
1697
1698 static void xfrm_link_failure(struct sk_buff *skb)
1699 {
1700         /* Impossible. Such dst must be popped before reaches point of failure. */
1701         return;
1702 }
1703
1704 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1705 {
1706         if (dst) {
1707                 if (dst->obsolete) {
1708                         dst_release(dst);
1709                         dst = NULL;
1710                 }
1711         }
1712         return dst;
1713 }
1714
1715 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1716 {
1717         struct dst_entry *dst, **dstp;
1718
1719         write_lock(&pol->lock);
1720         dstp = &pol->bundles;
1721         while ((dst=*dstp) != NULL) {
1722                 if (func(dst)) {
1723                         *dstp = dst->next;
1724                         dst->next = *gc_list_p;
1725                         *gc_list_p = dst;
1726                 } else {
1727                         dstp = &dst->next;
1728                 }
1729         }
1730         write_unlock(&pol->lock);
1731 }
1732
1733 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1734 {
1735         struct dst_entry *gc_list = NULL;
1736         int dir;
1737
1738         read_lock_bh(&xfrm_policy_lock);
1739         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1740                 struct xfrm_policy *pol;
1741                 struct hlist_node *entry;
1742                 struct hlist_head *table;
1743                 int i;
1744
1745                 hlist_for_each_entry(pol, entry,
1746                                      &xfrm_policy_inexact[dir], bydst)
1747                         prune_one_bundle(pol, func, &gc_list);
1748
1749                 table = xfrm_policy_bydst[dir].table;
1750                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1751                         hlist_for_each_entry(pol, entry, table + i, bydst)
1752                                 prune_one_bundle(pol, func, &gc_list);
1753                 }
1754         }
1755         read_unlock_bh(&xfrm_policy_lock);
1756
1757         while (gc_list) {
1758                 struct dst_entry *dst = gc_list;
1759                 gc_list = dst->next;
1760                 dst_free(dst);
1761         }
1762 }
1763
1764 static int unused_bundle(struct dst_entry *dst)
1765 {
1766         return !atomic_read(&dst->__refcnt);
1767 }
1768
1769 static void __xfrm_garbage_collect(void)
1770 {
1771         xfrm_prune_bundles(unused_bundle);
1772 }
1773
1774 static int xfrm_flush_bundles(void)
1775 {
1776         xfrm_prune_bundles(stale_bundle);
1777         return 0;
1778 }
1779
1780 void xfrm_init_pmtu(struct dst_entry *dst)
1781 {
1782         do {
1783                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1784                 u32 pmtu, route_mtu_cached;
1785
1786                 pmtu = dst_mtu(dst->child);
1787                 xdst->child_mtu_cached = pmtu;
1788
1789                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1790
1791                 route_mtu_cached = dst_mtu(xdst->route);
1792                 xdst->route_mtu_cached = route_mtu_cached;
1793
1794                 if (pmtu > route_mtu_cached)
1795                         pmtu = route_mtu_cached;
1796
1797                 dst->metrics[RTAX_MTU-1] = pmtu;
1798         } while ((dst = dst->next));
1799 }
1800
1801 EXPORT_SYMBOL(xfrm_init_pmtu);
1802
1803 /* Check that the bundle accepts the flow and its components are
1804  * still valid.
1805  */
1806
1807 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family, int strict)
1808 {
1809         struct dst_entry *dst = &first->u.dst;
1810         struct xfrm_dst *last;
1811         u32 mtu;
1812
1813         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1814             (dst->dev && !netif_running(dst->dev)))
1815                 return 0;
1816
1817         last = NULL;
1818
1819         do {
1820                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1821
1822                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1823                         return 0;
1824                 if (fl && !security_xfrm_flow_state_match(fl, dst->xfrm))
1825                         return 0;
1826                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1827                         return 0;
1828                 if (xdst->genid != dst->xfrm->genid)
1829                         return 0;
1830
1831                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
1832                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1833                         return 0;
1834
1835                 mtu = dst_mtu(dst->child);
1836                 if (xdst->child_mtu_cached != mtu) {
1837                         last = xdst;
1838                         xdst->child_mtu_cached = mtu;
1839                 }
1840
1841                 if (!dst_check(xdst->route, xdst->route_cookie))
1842                         return 0;
1843                 mtu = dst_mtu(xdst->route);
1844                 if (xdst->route_mtu_cached != mtu) {
1845                         last = xdst;
1846                         xdst->route_mtu_cached = mtu;
1847                 }
1848
1849                 dst = dst->child;
1850         } while (dst->xfrm);
1851
1852         if (likely(!last))
1853                 return 1;
1854
1855         mtu = last->child_mtu_cached;
1856         for (;;) {
1857                 dst = &last->u.dst;
1858
1859                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1860                 if (mtu > last->route_mtu_cached)
1861                         mtu = last->route_mtu_cached;
1862                 dst->metrics[RTAX_MTU-1] = mtu;
1863
1864                 if (last == first)
1865                         break;
1866
1867                 last = last->u.next;
1868                 last->child_mtu_cached = mtu;
1869         }
1870
1871         return 1;
1872 }
1873
1874 EXPORT_SYMBOL(xfrm_bundle_ok);
1875
1876 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1877 {
1878         int err = 0;
1879         if (unlikely(afinfo == NULL))
1880                 return -EINVAL;
1881         if (unlikely(afinfo->family >= NPROTO))
1882                 return -EAFNOSUPPORT;
1883         write_lock_bh(&xfrm_policy_afinfo_lock);
1884         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1885                 err = -ENOBUFS;
1886         else {
1887                 struct dst_ops *dst_ops = afinfo->dst_ops;
1888                 if (likely(dst_ops->kmem_cachep == NULL))
1889                         dst_ops->kmem_cachep = xfrm_dst_cache;
1890                 if (likely(dst_ops->check == NULL))
1891                         dst_ops->check = xfrm_dst_check;
1892                 if (likely(dst_ops->negative_advice == NULL))
1893                         dst_ops->negative_advice = xfrm_negative_advice;
1894                 if (likely(dst_ops->link_failure == NULL))
1895                         dst_ops->link_failure = xfrm_link_failure;
1896                 if (likely(afinfo->garbage_collect == NULL))
1897                         afinfo->garbage_collect = __xfrm_garbage_collect;
1898                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1899         }
1900         write_unlock_bh(&xfrm_policy_afinfo_lock);
1901         return err;
1902 }
1903 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1904
1905 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1906 {
1907         int err = 0;
1908         if (unlikely(afinfo == NULL))
1909                 return -EINVAL;
1910         if (unlikely(afinfo->family >= NPROTO))
1911                 return -EAFNOSUPPORT;
1912         write_lock_bh(&xfrm_policy_afinfo_lock);
1913         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1914                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1915                         err = -EINVAL;
1916                 else {
1917                         struct dst_ops *dst_ops = afinfo->dst_ops;
1918                         xfrm_policy_afinfo[afinfo->family] = NULL;
1919                         dst_ops->kmem_cachep = NULL;
1920                         dst_ops->check = NULL;
1921                         dst_ops->negative_advice = NULL;
1922                         dst_ops->link_failure = NULL;
1923                         afinfo->garbage_collect = NULL;
1924                 }
1925         }
1926         write_unlock_bh(&xfrm_policy_afinfo_lock);
1927         return err;
1928 }
1929 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1930
1931 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1932 {
1933         struct xfrm_policy_afinfo *afinfo;
1934         if (unlikely(family >= NPROTO))
1935                 return NULL;
1936         read_lock(&xfrm_policy_afinfo_lock);
1937         afinfo = xfrm_policy_afinfo[family];
1938         if (unlikely(!afinfo))
1939                 read_unlock(&xfrm_policy_afinfo_lock);
1940         return afinfo;
1941 }
1942
1943 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1944 {
1945         read_unlock(&xfrm_policy_afinfo_lock);
1946 }
1947
1948 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
1949 {
1950         struct xfrm_policy_afinfo *afinfo;
1951         if (unlikely(family >= NPROTO))
1952                 return NULL;
1953         write_lock_bh(&xfrm_policy_afinfo_lock);
1954         afinfo = xfrm_policy_afinfo[family];
1955         if (unlikely(!afinfo))
1956                 write_unlock_bh(&xfrm_policy_afinfo_lock);
1957         return afinfo;
1958 }
1959
1960 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
1961 {
1962         write_unlock_bh(&xfrm_policy_afinfo_lock);
1963 }
1964
1965 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1966 {
1967         switch (event) {
1968         case NETDEV_DOWN:
1969                 xfrm_flush_bundles();
1970         }
1971         return NOTIFY_DONE;
1972 }
1973
1974 static struct notifier_block xfrm_dev_notifier = {
1975         xfrm_dev_event,
1976         NULL,
1977         0
1978 };
1979
1980 static void __init xfrm_policy_init(void)
1981 {
1982         unsigned int hmask, sz;
1983         int dir;
1984
1985         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1986                                            sizeof(struct xfrm_dst),
1987                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1988                                            NULL, NULL);
1989
1990         hmask = 8 - 1;
1991         sz = (hmask+1) * sizeof(struct hlist_head);
1992
1993         xfrm_policy_byidx = xfrm_hash_alloc(sz);
1994         xfrm_idx_hmask = hmask;
1995         if (!xfrm_policy_byidx)
1996                 panic("XFRM: failed to allocate byidx hash\n");
1997
1998         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1999                 struct xfrm_policy_hash *htab;
2000
2001                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2002
2003                 htab = &xfrm_policy_bydst[dir];
2004                 htab->table = xfrm_hash_alloc(sz);
2005                 htab->hmask = hmask;
2006                 if (!htab->table)
2007                         panic("XFRM: failed to allocate bydst hash\n");
2008         }
2009
2010         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
2011         register_netdevice_notifier(&xfrm_dev_notifier);
2012 }
2013
2014 void __init xfrm_init(void)
2015 {
2016         xfrm_state_init();
2017         xfrm_policy_init();
2018         xfrm_input_init();
2019 }
2020