cfaf17c8851e7ef63699cfe81af23bbde00d7041
[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 #include <linux/audit.h>
29 #include <linux/cache.h>
30
31 #include "xfrm_hash.h"
32
33 int sysctl_xfrm_larval_drop __read_mostly;
34
35 DEFINE_MUTEX(xfrm_cfg_mutex);
36 EXPORT_SYMBOL(xfrm_cfg_mutex);
37
38 static DEFINE_RWLOCK(xfrm_policy_lock);
39
40 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
41 EXPORT_SYMBOL(xfrm_policy_count);
42
43 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
44 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
45
46 static struct kmem_cache *xfrm_dst_cache __read_mostly;
47
48 static struct work_struct xfrm_policy_gc_work;
49 static HLIST_HEAD(xfrm_policy_gc_list);
50 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
51
52 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
53 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
54 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
55 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
56
57 static inline int
58 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
59 {
60         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
61                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
62                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
63                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
64                 (fl->proto == sel->proto || !sel->proto) &&
65                 (fl->oif == sel->ifindex || !sel->ifindex);
66 }
67
68 static inline int
69 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
70 {
71         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
72                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
73                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
74                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
75                 (fl->proto == sel->proto || !sel->proto) &&
76                 (fl->oif == sel->ifindex || !sel->ifindex);
77 }
78
79 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
80                     unsigned short family)
81 {
82         switch (family) {
83         case AF_INET:
84                 return __xfrm4_selector_match(sel, fl);
85         case AF_INET6:
86                 return __xfrm6_selector_match(sel, fl);
87         }
88         return 0;
89 }
90
91 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
92 {
93         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
94         struct xfrm_type **typemap;
95         int err = 0;
96
97         if (unlikely(afinfo == NULL))
98                 return -EAFNOSUPPORT;
99         typemap = afinfo->type_map;
100
101         if (likely(typemap[type->proto] == NULL))
102                 typemap[type->proto] = type;
103         else
104                 err = -EEXIST;
105         xfrm_policy_unlock_afinfo(afinfo);
106         return err;
107 }
108 EXPORT_SYMBOL(xfrm_register_type);
109
110 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
111 {
112         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
113         struct xfrm_type **typemap;
114         int err = 0;
115
116         if (unlikely(afinfo == NULL))
117                 return -EAFNOSUPPORT;
118         typemap = afinfo->type_map;
119
120         if (unlikely(typemap[type->proto] != type))
121                 err = -ENOENT;
122         else
123                 typemap[type->proto] = NULL;
124         xfrm_policy_unlock_afinfo(afinfo);
125         return err;
126 }
127 EXPORT_SYMBOL(xfrm_unregister_type);
128
129 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
130 {
131         struct xfrm_policy_afinfo *afinfo;
132         struct xfrm_type **typemap;
133         struct xfrm_type *type;
134         int modload_attempted = 0;
135
136 retry:
137         afinfo = xfrm_policy_get_afinfo(family);
138         if (unlikely(afinfo == NULL))
139                 return NULL;
140         typemap = afinfo->type_map;
141
142         type = typemap[proto];
143         if (unlikely(type && !try_module_get(type->owner)))
144                 type = NULL;
145         if (!type && !modload_attempted) {
146                 xfrm_policy_put_afinfo(afinfo);
147                 request_module("xfrm-type-%d-%d",
148                                (int) family, (int) proto);
149                 modload_attempted = 1;
150                 goto retry;
151         }
152
153         xfrm_policy_put_afinfo(afinfo);
154         return type;
155 }
156
157 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
158                     unsigned short family)
159 {
160         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
161         int err = 0;
162
163         if (unlikely(afinfo == NULL))
164                 return -EAFNOSUPPORT;
165
166         if (likely(afinfo->dst_lookup != NULL))
167                 err = afinfo->dst_lookup(dst, fl);
168         else
169                 err = -EINVAL;
170         xfrm_policy_put_afinfo(afinfo);
171         return err;
172 }
173 EXPORT_SYMBOL(xfrm_dst_lookup);
174
175 void xfrm_put_type(struct xfrm_type *type)
176 {
177         module_put(type->owner);
178 }
179
180 int xfrm_register_mode(struct xfrm_mode *mode, int family)
181 {
182         struct xfrm_policy_afinfo *afinfo;
183         struct xfrm_mode **modemap;
184         int err;
185
186         if (unlikely(mode->encap >= XFRM_MODE_MAX))
187                 return -EINVAL;
188
189         afinfo = xfrm_policy_lock_afinfo(family);
190         if (unlikely(afinfo == NULL))
191                 return -EAFNOSUPPORT;
192
193         err = -EEXIST;
194         modemap = afinfo->mode_map;
195         if (likely(modemap[mode->encap] == NULL)) {
196                 modemap[mode->encap] = mode;
197                 err = 0;
198         }
199
200         xfrm_policy_unlock_afinfo(afinfo);
201         return err;
202 }
203 EXPORT_SYMBOL(xfrm_register_mode);
204
205 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
206 {
207         struct xfrm_policy_afinfo *afinfo;
208         struct xfrm_mode **modemap;
209         int err;
210
211         if (unlikely(mode->encap >= XFRM_MODE_MAX))
212                 return -EINVAL;
213
214         afinfo = xfrm_policy_lock_afinfo(family);
215         if (unlikely(afinfo == NULL))
216                 return -EAFNOSUPPORT;
217
218         err = -ENOENT;
219         modemap = afinfo->mode_map;
220         if (likely(modemap[mode->encap] == mode)) {
221                 modemap[mode->encap] = NULL;
222                 err = 0;
223         }
224
225         xfrm_policy_unlock_afinfo(afinfo);
226         return err;
227 }
228 EXPORT_SYMBOL(xfrm_unregister_mode);
229
230 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
231 {
232         struct xfrm_policy_afinfo *afinfo;
233         struct xfrm_mode *mode;
234         int modload_attempted = 0;
235
236         if (unlikely(encap >= XFRM_MODE_MAX))
237                 return NULL;
238
239 retry:
240         afinfo = xfrm_policy_get_afinfo(family);
241         if (unlikely(afinfo == NULL))
242                 return NULL;
243
244         mode = afinfo->mode_map[encap];
245         if (unlikely(mode && !try_module_get(mode->owner)))
246                 mode = NULL;
247         if (!mode && !modload_attempted) {
248                 xfrm_policy_put_afinfo(afinfo);
249                 request_module("xfrm-mode-%d-%d", family, encap);
250                 modload_attempted = 1;
251                 goto retry;
252         }
253
254         xfrm_policy_put_afinfo(afinfo);
255         return mode;
256 }
257
258 void xfrm_put_mode(struct xfrm_mode *mode)
259 {
260         module_put(mode->owner);
261 }
262
263 static inline unsigned long make_jiffies(long secs)
264 {
265         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
266                 return MAX_SCHEDULE_TIMEOUT-1;
267         else
268                 return secs*HZ;
269 }
270
271 static void xfrm_policy_timer(unsigned long data)
272 {
273         struct xfrm_policy *xp = (struct xfrm_policy*)data;
274         unsigned long now = get_seconds();
275         long next = LONG_MAX;
276         int warn = 0;
277         int dir;
278
279         read_lock(&xp->lock);
280
281         if (xp->dead)
282                 goto out;
283
284         dir = xfrm_policy_id2dir(xp->index);
285
286         if (xp->lft.hard_add_expires_seconds) {
287                 long tmo = xp->lft.hard_add_expires_seconds +
288                         xp->curlft.add_time - now;
289                 if (tmo <= 0)
290                         goto expired;
291                 if (tmo < next)
292                         next = tmo;
293         }
294         if (xp->lft.hard_use_expires_seconds) {
295                 long tmo = xp->lft.hard_use_expires_seconds +
296                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
297                 if (tmo <= 0)
298                         goto expired;
299                 if (tmo < next)
300                         next = tmo;
301         }
302         if (xp->lft.soft_add_expires_seconds) {
303                 long tmo = xp->lft.soft_add_expires_seconds +
304                         xp->curlft.add_time - now;
305                 if (tmo <= 0) {
306                         warn = 1;
307                         tmo = XFRM_KM_TIMEOUT;
308                 }
309                 if (tmo < next)
310                         next = tmo;
311         }
312         if (xp->lft.soft_use_expires_seconds) {
313                 long tmo = xp->lft.soft_use_expires_seconds +
314                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
315                 if (tmo <= 0) {
316                         warn = 1;
317                         tmo = XFRM_KM_TIMEOUT;
318                 }
319                 if (tmo < next)
320                         next = tmo;
321         }
322
323         if (warn)
324                 km_policy_expired(xp, dir, 0, 0);
325         if (next != LONG_MAX &&
326             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
327                 xfrm_pol_hold(xp);
328
329 out:
330         read_unlock(&xp->lock);
331         xfrm_pol_put(xp);
332         return;
333
334 expired:
335         read_unlock(&xp->lock);
336         if (!xfrm_policy_delete(xp, dir))
337                 km_policy_expired(xp, dir, 1, 0);
338         xfrm_pol_put(xp);
339 }
340
341
342 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
343  * SPD calls.
344  */
345
346 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
347 {
348         struct xfrm_policy *policy;
349
350         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
351
352         if (policy) {
353                 INIT_HLIST_NODE(&policy->bydst);
354                 INIT_HLIST_NODE(&policy->byidx);
355                 rwlock_init(&policy->lock);
356                 atomic_set(&policy->refcnt, 1);
357                 init_timer(&policy->timer);
358                 policy->timer.data = (unsigned long)policy;
359                 policy->timer.function = xfrm_policy_timer;
360         }
361         return policy;
362 }
363 EXPORT_SYMBOL(xfrm_policy_alloc);
364
365 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
366
367 void __xfrm_policy_destroy(struct xfrm_policy *policy)
368 {
369         BUG_ON(!policy->dead);
370
371         BUG_ON(policy->bundles);
372
373         if (del_timer(&policy->timer))
374                 BUG();
375
376         security_xfrm_policy_free(policy);
377         kfree(policy);
378 }
379 EXPORT_SYMBOL(__xfrm_policy_destroy);
380
381 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
382 {
383         struct dst_entry *dst;
384
385         while ((dst = policy->bundles) != NULL) {
386                 policy->bundles = dst->next;
387                 dst_free(dst);
388         }
389
390         if (del_timer(&policy->timer))
391                 atomic_dec(&policy->refcnt);
392
393         if (atomic_read(&policy->refcnt) > 1)
394                 flow_cache_flush();
395
396         xfrm_pol_put(policy);
397 }
398
399 static void xfrm_policy_gc_task(struct work_struct *work)
400 {
401         struct xfrm_policy *policy;
402         struct hlist_node *entry, *tmp;
403         struct hlist_head gc_list;
404
405         spin_lock_bh(&xfrm_policy_gc_lock);
406         gc_list.first = xfrm_policy_gc_list.first;
407         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
408         spin_unlock_bh(&xfrm_policy_gc_lock);
409
410         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
411                 xfrm_policy_gc_kill(policy);
412 }
413
414 /* Rule must be locked. Release descentant resources, announce
415  * entry dead. The rule must be unlinked from lists to the moment.
416  */
417
418 static void xfrm_policy_kill(struct xfrm_policy *policy)
419 {
420         int dead;
421
422         write_lock_bh(&policy->lock);
423         dead = policy->dead;
424         policy->dead = 1;
425         write_unlock_bh(&policy->lock);
426
427         if (unlikely(dead)) {
428                 WARN_ON(1);
429                 return;
430         }
431
432         spin_lock(&xfrm_policy_gc_lock);
433         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
434         spin_unlock(&xfrm_policy_gc_lock);
435
436         schedule_work(&xfrm_policy_gc_work);
437 }
438
439 struct xfrm_policy_hash {
440         struct hlist_head       *table;
441         unsigned int            hmask;
442 };
443
444 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
445 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
446 static struct hlist_head *xfrm_policy_byidx __read_mostly;
447 static unsigned int xfrm_idx_hmask __read_mostly;
448 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
449
450 static inline unsigned int idx_hash(u32 index)
451 {
452         return __idx_hash(index, xfrm_idx_hmask);
453 }
454
455 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
456 {
457         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
458         unsigned int hash = __sel_hash(sel, family, hmask);
459
460         return (hash == hmask + 1 ?
461                 &xfrm_policy_inexact[dir] :
462                 xfrm_policy_bydst[dir].table + hash);
463 }
464
465 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
466 {
467         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
468         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
469
470         return xfrm_policy_bydst[dir].table + hash;
471 }
472
473 static void xfrm_dst_hash_transfer(struct hlist_head *list,
474                                    struct hlist_head *ndsttable,
475                                    unsigned int nhashmask)
476 {
477         struct hlist_node *entry, *tmp;
478         struct xfrm_policy *pol;
479
480         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
481                 unsigned int h;
482
483                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
484                                 pol->family, nhashmask);
485                 hlist_add_head(&pol->bydst, ndsttable+h);
486         }
487 }
488
489 static void xfrm_idx_hash_transfer(struct hlist_head *list,
490                                    struct hlist_head *nidxtable,
491                                    unsigned int nhashmask)
492 {
493         struct hlist_node *entry, *tmp;
494         struct xfrm_policy *pol;
495
496         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
497                 unsigned int h;
498
499                 h = __idx_hash(pol->index, nhashmask);
500                 hlist_add_head(&pol->byidx, nidxtable+h);
501         }
502 }
503
504 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
505 {
506         return ((old_hmask + 1) << 1) - 1;
507 }
508
509 static void xfrm_bydst_resize(int dir)
510 {
511         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
512         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
513         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
514         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
515         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
516         int i;
517
518         if (!ndst)
519                 return;
520
521         write_lock_bh(&xfrm_policy_lock);
522
523         for (i = hmask; i >= 0; i--)
524                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
525
526         xfrm_policy_bydst[dir].table = ndst;
527         xfrm_policy_bydst[dir].hmask = nhashmask;
528
529         write_unlock_bh(&xfrm_policy_lock);
530
531         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
532 }
533
534 static void xfrm_byidx_resize(int total)
535 {
536         unsigned int hmask = xfrm_idx_hmask;
537         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
538         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
539         struct hlist_head *oidx = xfrm_policy_byidx;
540         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
541         int i;
542
543         if (!nidx)
544                 return;
545
546         write_lock_bh(&xfrm_policy_lock);
547
548         for (i = hmask; i >= 0; i--)
549                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
550
551         xfrm_policy_byidx = nidx;
552         xfrm_idx_hmask = nhashmask;
553
554         write_unlock_bh(&xfrm_policy_lock);
555
556         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
557 }
558
559 static inline int xfrm_bydst_should_resize(int dir, int *total)
560 {
561         unsigned int cnt = xfrm_policy_count[dir];
562         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
563
564         if (total)
565                 *total += cnt;
566
567         if ((hmask + 1) < xfrm_policy_hashmax &&
568             cnt > hmask)
569                 return 1;
570
571         return 0;
572 }
573
574 static inline int xfrm_byidx_should_resize(int total)
575 {
576         unsigned int hmask = xfrm_idx_hmask;
577
578         if ((hmask + 1) < xfrm_policy_hashmax &&
579             total > hmask)
580                 return 1;
581
582         return 0;
583 }
584
585 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
586 {
587         read_lock_bh(&xfrm_policy_lock);
588         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
589         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
590         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
591         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
592         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
593         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
594         si->spdhcnt = xfrm_idx_hmask;
595         si->spdhmcnt = xfrm_policy_hashmax;
596         read_unlock_bh(&xfrm_policy_lock);
597 }
598 EXPORT_SYMBOL(xfrm_spd_getinfo);
599
600 static DEFINE_MUTEX(hash_resize_mutex);
601 static void xfrm_hash_resize(struct work_struct *__unused)
602 {
603         int dir, total;
604
605         mutex_lock(&hash_resize_mutex);
606
607         total = 0;
608         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
609                 if (xfrm_bydst_should_resize(dir, &total))
610                         xfrm_bydst_resize(dir);
611         }
612         if (xfrm_byidx_should_resize(total))
613                 xfrm_byidx_resize(total);
614
615         mutex_unlock(&hash_resize_mutex);
616 }
617
618 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
619
620 /* Generate new index... KAME seems to generate them ordered by cost
621  * of an absolute inpredictability of ordering of rules. This will not pass. */
622 static u32 xfrm_gen_index(u8 type, int dir)
623 {
624         static u32 idx_generator;
625
626         for (;;) {
627                 struct hlist_node *entry;
628                 struct hlist_head *list;
629                 struct xfrm_policy *p;
630                 u32 idx;
631                 int found;
632
633                 idx = (idx_generator | dir);
634                 idx_generator += 8;
635                 if (idx == 0)
636                         idx = 8;
637                 list = xfrm_policy_byidx + idx_hash(idx);
638                 found = 0;
639                 hlist_for_each_entry(p, entry, list, byidx) {
640                         if (p->index == idx) {
641                                 found = 1;
642                                 break;
643                         }
644                 }
645                 if (!found)
646                         return idx;
647         }
648 }
649
650 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
651 {
652         u32 *p1 = (u32 *) s1;
653         u32 *p2 = (u32 *) s2;
654         int len = sizeof(struct xfrm_selector) / sizeof(u32);
655         int i;
656
657         for (i = 0; i < len; i++) {
658                 if (p1[i] != p2[i])
659                         return 1;
660         }
661
662         return 0;
663 }
664
665 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
666 {
667         struct xfrm_policy *pol;
668         struct xfrm_policy *delpol;
669         struct hlist_head *chain;
670         struct hlist_node *entry, *newpos;
671         struct dst_entry *gc_list;
672
673         write_lock_bh(&xfrm_policy_lock);
674         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
675         delpol = NULL;
676         newpos = NULL;
677         hlist_for_each_entry(pol, entry, chain, bydst) {
678                 if (pol->type == policy->type &&
679                     !selector_cmp(&pol->selector, &policy->selector) &&
680                     xfrm_sec_ctx_match(pol->security, policy->security) &&
681                     !WARN_ON(delpol)) {
682                         if (excl) {
683                                 write_unlock_bh(&xfrm_policy_lock);
684                                 return -EEXIST;
685                         }
686                         delpol = pol;
687                         if (policy->priority > pol->priority)
688                                 continue;
689                 } else if (policy->priority >= pol->priority) {
690                         newpos = &pol->bydst;
691                         continue;
692                 }
693                 if (delpol)
694                         break;
695         }
696         if (newpos)
697                 hlist_add_after(newpos, &policy->bydst);
698         else
699                 hlist_add_head(&policy->bydst, chain);
700         xfrm_pol_hold(policy);
701         xfrm_policy_count[dir]++;
702         atomic_inc(&flow_cache_genid);
703         if (delpol) {
704                 hlist_del(&delpol->bydst);
705                 hlist_del(&delpol->byidx);
706                 xfrm_policy_count[dir]--;
707         }
708         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
709         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
710         policy->curlft.add_time = get_seconds();
711         policy->curlft.use_time = 0;
712         if (!mod_timer(&policy->timer, jiffies + HZ))
713                 xfrm_pol_hold(policy);
714         write_unlock_bh(&xfrm_policy_lock);
715
716         if (delpol)
717                 xfrm_policy_kill(delpol);
718         else if (xfrm_bydst_should_resize(dir, NULL))
719                 schedule_work(&xfrm_hash_work);
720
721         read_lock_bh(&xfrm_policy_lock);
722         gc_list = NULL;
723         entry = &policy->bydst;
724         hlist_for_each_entry_continue(policy, entry, bydst) {
725                 struct dst_entry *dst;
726
727                 write_lock(&policy->lock);
728                 dst = policy->bundles;
729                 if (dst) {
730                         struct dst_entry *tail = dst;
731                         while (tail->next)
732                                 tail = tail->next;
733                         tail->next = gc_list;
734                         gc_list = dst;
735
736                         policy->bundles = NULL;
737                 }
738                 write_unlock(&policy->lock);
739         }
740         read_unlock_bh(&xfrm_policy_lock);
741
742         while (gc_list) {
743                 struct dst_entry *dst = gc_list;
744
745                 gc_list = dst->next;
746                 dst_free(dst);
747         }
748
749         return 0;
750 }
751 EXPORT_SYMBOL(xfrm_policy_insert);
752
753 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
754                                           struct xfrm_selector *sel,
755                                           struct xfrm_sec_ctx *ctx, int delete,
756                                           int *err)
757 {
758         struct xfrm_policy *pol, *ret;
759         struct hlist_head *chain;
760         struct hlist_node *entry;
761
762         *err = 0;
763         write_lock_bh(&xfrm_policy_lock);
764         chain = policy_hash_bysel(sel, sel->family, dir);
765         ret = NULL;
766         hlist_for_each_entry(pol, entry, chain, bydst) {
767                 if (pol->type == type &&
768                     !selector_cmp(sel, &pol->selector) &&
769                     xfrm_sec_ctx_match(ctx, pol->security)) {
770                         xfrm_pol_hold(pol);
771                         if (delete) {
772                                 *err = security_xfrm_policy_delete(pol);
773                                 if (*err) {
774                                         write_unlock_bh(&xfrm_policy_lock);
775                                         return pol;
776                                 }
777                                 hlist_del(&pol->bydst);
778                                 hlist_del(&pol->byidx);
779                                 xfrm_policy_count[dir]--;
780                         }
781                         ret = pol;
782                         break;
783                 }
784         }
785         write_unlock_bh(&xfrm_policy_lock);
786
787         if (ret && delete) {
788                 atomic_inc(&flow_cache_genid);
789                 xfrm_policy_kill(ret);
790         }
791         return ret;
792 }
793 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
794
795 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
796                                      int *err)
797 {
798         struct xfrm_policy *pol, *ret;
799         struct hlist_head *chain;
800         struct hlist_node *entry;
801
802         *err = -ENOENT;
803         if (xfrm_policy_id2dir(id) != dir)
804                 return NULL;
805
806         *err = 0;
807         write_lock_bh(&xfrm_policy_lock);
808         chain = xfrm_policy_byidx + idx_hash(id);
809         ret = NULL;
810         hlist_for_each_entry(pol, entry, chain, byidx) {
811                 if (pol->type == type && pol->index == id) {
812                         xfrm_pol_hold(pol);
813                         if (delete) {
814                                 *err = security_xfrm_policy_delete(pol);
815                                 if (*err) {
816                                         write_unlock_bh(&xfrm_policy_lock);
817                                         return pol;
818                                 }
819                                 hlist_del(&pol->bydst);
820                                 hlist_del(&pol->byidx);
821                                 xfrm_policy_count[dir]--;
822                         }
823                         ret = pol;
824                         break;
825                 }
826         }
827         write_unlock_bh(&xfrm_policy_lock);
828
829         if (ret && delete) {
830                 atomic_inc(&flow_cache_genid);
831                 xfrm_policy_kill(ret);
832         }
833         return ret;
834 }
835 EXPORT_SYMBOL(xfrm_policy_byid);
836
837 #ifdef CONFIG_SECURITY_NETWORK_XFRM
838 static inline int
839 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
840 {
841         int dir, err = 0;
842
843         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
844                 struct xfrm_policy *pol;
845                 struct hlist_node *entry;
846                 int i;
847
848                 hlist_for_each_entry(pol, entry,
849                                      &xfrm_policy_inexact[dir], bydst) {
850                         if (pol->type != type)
851                                 continue;
852                         err = security_xfrm_policy_delete(pol);
853                         if (err) {
854                                 xfrm_audit_log(audit_info->loginuid,
855                                                audit_info->secid,
856                                                AUDIT_MAC_IPSEC_DELSPD, 0,
857                                                pol, NULL);
858                                 return err;
859                         }
860                 }
861                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
862                         hlist_for_each_entry(pol, entry,
863                                              xfrm_policy_bydst[dir].table + i,
864                                              bydst) {
865                                 if (pol->type != type)
866                                         continue;
867                                 err = security_xfrm_policy_delete(pol);
868                                 if (err) {
869                                         xfrm_audit_log(audit_info->loginuid,
870                                                        audit_info->secid,
871                                                        AUDIT_MAC_IPSEC_DELSPD,
872                                                        0, pol, NULL);
873                                         return err;
874                                 }
875                         }
876                 }
877         }
878         return err;
879 }
880 #else
881 static inline int
882 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
883 {
884         return 0;
885 }
886 #endif
887
888 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
889 {
890         int dir, err = 0;
891
892         write_lock_bh(&xfrm_policy_lock);
893
894         err = xfrm_policy_flush_secctx_check(type, audit_info);
895         if (err)
896                 goto out;
897
898         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
899                 struct xfrm_policy *pol;
900                 struct hlist_node *entry;
901                 int i, killed;
902
903                 killed = 0;
904         again1:
905                 hlist_for_each_entry(pol, entry,
906                                      &xfrm_policy_inexact[dir], bydst) {
907                         if (pol->type != type)
908                                 continue;
909                         hlist_del(&pol->bydst);
910                         hlist_del(&pol->byidx);
911                         write_unlock_bh(&xfrm_policy_lock);
912
913                         xfrm_audit_log(audit_info->loginuid, audit_info->secid,
914                                        AUDIT_MAC_IPSEC_DELSPD, 1, pol, NULL);
915
916                         xfrm_policy_kill(pol);
917                         killed++;
918
919                         write_lock_bh(&xfrm_policy_lock);
920                         goto again1;
921                 }
922
923                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
924         again2:
925                         hlist_for_each_entry(pol, entry,
926                                              xfrm_policy_bydst[dir].table + i,
927                                              bydst) {
928                                 if (pol->type != type)
929                                         continue;
930                                 hlist_del(&pol->bydst);
931                                 hlist_del(&pol->byidx);
932                                 write_unlock_bh(&xfrm_policy_lock);
933
934                                 xfrm_audit_log(audit_info->loginuid,
935                                                audit_info->secid,
936                                                AUDIT_MAC_IPSEC_DELSPD, 1,
937                                                pol, NULL);
938
939                                 xfrm_policy_kill(pol);
940                                 killed++;
941
942                                 write_lock_bh(&xfrm_policy_lock);
943                                 goto again2;
944                         }
945                 }
946
947                 xfrm_policy_count[dir] -= killed;
948         }
949         atomic_inc(&flow_cache_genid);
950 out:
951         write_unlock_bh(&xfrm_policy_lock);
952         return err;
953 }
954 EXPORT_SYMBOL(xfrm_policy_flush);
955
956 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
957                      void *data)
958 {
959         struct xfrm_policy *pol, *last = NULL;
960         struct hlist_node *entry;
961         int dir, last_dir = 0, count, error;
962
963         read_lock_bh(&xfrm_policy_lock);
964         count = 0;
965
966         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
967                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
968                 int i;
969
970                 hlist_for_each_entry(pol, entry,
971                                      &xfrm_policy_inexact[dir], bydst) {
972                         if (pol->type != type)
973                                 continue;
974                         if (last) {
975                                 error = func(last, last_dir % XFRM_POLICY_MAX,
976                                              count, data);
977                                 if (error)
978                                         goto out;
979                         }
980                         last = pol;
981                         last_dir = dir;
982                         count++;
983                 }
984                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
985                         hlist_for_each_entry(pol, entry, table + i, bydst) {
986                                 if (pol->type != type)
987                                         continue;
988                                 if (last) {
989                                         error = func(last, last_dir % XFRM_POLICY_MAX,
990                                                      count, data);
991                                         if (error)
992                                                 goto out;
993                                 }
994                                 last = pol;
995                                 last_dir = dir;
996                                 count++;
997                         }
998                 }
999         }
1000         if (count == 0) {
1001                 error = -ENOENT;
1002                 goto out;
1003         }
1004         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
1005 out:
1006         read_unlock_bh(&xfrm_policy_lock);
1007         return error;
1008 }
1009 EXPORT_SYMBOL(xfrm_policy_walk);
1010
1011 /*
1012  * Find policy to apply to this flow.
1013  *
1014  * Returns 0 if policy found, else an -errno.
1015  */
1016 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
1017                              u8 type, u16 family, int dir)
1018 {
1019         struct xfrm_selector *sel = &pol->selector;
1020         int match, ret = -ESRCH;
1021
1022         if (pol->family != family ||
1023             pol->type != type)
1024                 return ret;
1025
1026         match = xfrm_selector_match(sel, fl, family);
1027         if (match)
1028                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
1029
1030         return ret;
1031 }
1032
1033 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
1034                                                      u16 family, u8 dir)
1035 {
1036         int err;
1037         struct xfrm_policy *pol, *ret;
1038         xfrm_address_t *daddr, *saddr;
1039         struct hlist_node *entry;
1040         struct hlist_head *chain;
1041         u32 priority = ~0U;
1042
1043         daddr = xfrm_flowi_daddr(fl, family);
1044         saddr = xfrm_flowi_saddr(fl, family);
1045         if (unlikely(!daddr || !saddr))
1046                 return NULL;
1047
1048         read_lock_bh(&xfrm_policy_lock);
1049         chain = policy_hash_direct(daddr, saddr, family, dir);
1050         ret = NULL;
1051         hlist_for_each_entry(pol, entry, chain, bydst) {
1052                 err = xfrm_policy_match(pol, fl, type, family, dir);
1053                 if (err) {
1054                         if (err == -ESRCH)
1055                                 continue;
1056                         else {
1057                                 ret = ERR_PTR(err);
1058                                 goto fail;
1059                         }
1060                 } else {
1061                         ret = pol;
1062                         priority = ret->priority;
1063                         break;
1064                 }
1065         }
1066         chain = &xfrm_policy_inexact[dir];
1067         hlist_for_each_entry(pol, entry, chain, bydst) {
1068                 err = xfrm_policy_match(pol, fl, type, family, dir);
1069                 if (err) {
1070                         if (err == -ESRCH)
1071                                 continue;
1072                         else {
1073                                 ret = ERR_PTR(err);
1074                                 goto fail;
1075                         }
1076                 } else if (pol->priority < priority) {
1077                         ret = pol;
1078                         break;
1079                 }
1080         }
1081         if (ret)
1082                 xfrm_pol_hold(ret);
1083 fail:
1084         read_unlock_bh(&xfrm_policy_lock);
1085
1086         return ret;
1087 }
1088
1089 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1090                                void **objp, atomic_t **obj_refp)
1091 {
1092         struct xfrm_policy *pol;
1093         int err = 0;
1094
1095 #ifdef CONFIG_XFRM_SUB_POLICY
1096         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1097         if (IS_ERR(pol)) {
1098                 err = PTR_ERR(pol);
1099                 pol = NULL;
1100         }
1101         if (pol || err)
1102                 goto end;
1103 #endif
1104         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1105         if (IS_ERR(pol)) {
1106                 err = PTR_ERR(pol);
1107                 pol = NULL;
1108         }
1109 #ifdef CONFIG_XFRM_SUB_POLICY
1110 end:
1111 #endif
1112         if ((*objp = (void *) pol) != NULL)
1113                 *obj_refp = &pol->refcnt;
1114         return err;
1115 }
1116
1117 static inline int policy_to_flow_dir(int dir)
1118 {
1119         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1120             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1121             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1122                 return dir;
1123         switch (dir) {
1124         default:
1125         case XFRM_POLICY_IN:
1126                 return FLOW_DIR_IN;
1127         case XFRM_POLICY_OUT:
1128                 return FLOW_DIR_OUT;
1129         case XFRM_POLICY_FWD:
1130                 return FLOW_DIR_FWD;
1131         }
1132 }
1133
1134 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1135 {
1136         struct xfrm_policy *pol;
1137
1138         read_lock_bh(&xfrm_policy_lock);
1139         if ((pol = sk->sk_policy[dir]) != NULL) {
1140                 int match = xfrm_selector_match(&pol->selector, fl,
1141                                                 sk->sk_family);
1142                 int err = 0;
1143
1144                 if (match) {
1145                         err = security_xfrm_policy_lookup(pol, fl->secid,
1146                                         policy_to_flow_dir(dir));
1147                         if (!err)
1148                                 xfrm_pol_hold(pol);
1149                         else if (err == -ESRCH)
1150                                 pol = NULL;
1151                         else
1152                                 pol = ERR_PTR(err);
1153                 } else
1154                         pol = NULL;
1155         }
1156         read_unlock_bh(&xfrm_policy_lock);
1157         return pol;
1158 }
1159
1160 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1161 {
1162         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1163                                                      pol->family, dir);
1164
1165         hlist_add_head(&pol->bydst, chain);
1166         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1167         xfrm_policy_count[dir]++;
1168         xfrm_pol_hold(pol);
1169
1170         if (xfrm_bydst_should_resize(dir, NULL))
1171                 schedule_work(&xfrm_hash_work);
1172 }
1173
1174 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1175                                                 int dir)
1176 {
1177         if (hlist_unhashed(&pol->bydst))
1178                 return NULL;
1179
1180         hlist_del(&pol->bydst);
1181         hlist_del(&pol->byidx);
1182         xfrm_policy_count[dir]--;
1183
1184         return pol;
1185 }
1186
1187 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1188 {
1189         write_lock_bh(&xfrm_policy_lock);
1190         pol = __xfrm_policy_unlink(pol, dir);
1191         write_unlock_bh(&xfrm_policy_lock);
1192         if (pol) {
1193                 if (dir < XFRM_POLICY_MAX)
1194                         atomic_inc(&flow_cache_genid);
1195                 xfrm_policy_kill(pol);
1196                 return 0;
1197         }
1198         return -ENOENT;
1199 }
1200 EXPORT_SYMBOL(xfrm_policy_delete);
1201
1202 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1203 {
1204         struct xfrm_policy *old_pol;
1205
1206 #ifdef CONFIG_XFRM_SUB_POLICY
1207         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1208                 return -EINVAL;
1209 #endif
1210
1211         write_lock_bh(&xfrm_policy_lock);
1212         old_pol = sk->sk_policy[dir];
1213         sk->sk_policy[dir] = pol;
1214         if (pol) {
1215                 pol->curlft.add_time = get_seconds();
1216                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1217                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1218         }
1219         if (old_pol)
1220                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1221         write_unlock_bh(&xfrm_policy_lock);
1222
1223         if (old_pol) {
1224                 xfrm_policy_kill(old_pol);
1225         }
1226         return 0;
1227 }
1228
1229 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1230 {
1231         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1232
1233         if (newp) {
1234                 newp->selector = old->selector;
1235                 if (security_xfrm_policy_clone(old, newp)) {
1236                         kfree(newp);
1237                         return NULL;  /* ENOMEM */
1238                 }
1239                 newp->lft = old->lft;
1240                 newp->curlft = old->curlft;
1241                 newp->action = old->action;
1242                 newp->flags = old->flags;
1243                 newp->xfrm_nr = old->xfrm_nr;
1244                 newp->index = old->index;
1245                 newp->type = old->type;
1246                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1247                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1248                 write_lock_bh(&xfrm_policy_lock);
1249                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1250                 write_unlock_bh(&xfrm_policy_lock);
1251                 xfrm_pol_put(newp);
1252         }
1253         return newp;
1254 }
1255
1256 int __xfrm_sk_clone_policy(struct sock *sk)
1257 {
1258         struct xfrm_policy *p0 = sk->sk_policy[0],
1259                            *p1 = sk->sk_policy[1];
1260
1261         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1262         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1263                 return -ENOMEM;
1264         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1265                 return -ENOMEM;
1266         return 0;
1267 }
1268
1269 static int
1270 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1271                unsigned short family)
1272 {
1273         int err;
1274         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1275
1276         if (unlikely(afinfo == NULL))
1277                 return -EINVAL;
1278         err = afinfo->get_saddr(local, remote);
1279         xfrm_policy_put_afinfo(afinfo);
1280         return err;
1281 }
1282
1283 /* Resolve list of templates for the flow, given policy. */
1284
1285 static int
1286 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1287                       struct xfrm_state **xfrm,
1288                       unsigned short family)
1289 {
1290         int nx;
1291         int i, error;
1292         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1293         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1294         xfrm_address_t tmp;
1295
1296         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1297                 struct xfrm_state *x;
1298                 xfrm_address_t *remote = daddr;
1299                 xfrm_address_t *local  = saddr;
1300                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1301
1302                 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1303                         remote = &tmpl->id.daddr;
1304                         local = &tmpl->saddr;
1305                         family = tmpl->encap_family;
1306                         if (xfrm_addr_any(local, family)) {
1307                                 error = xfrm_get_saddr(&tmp, remote, family);
1308                                 if (error)
1309                                         goto fail;
1310                                 local = &tmp;
1311                         }
1312                 }
1313
1314                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1315
1316                 if (x && x->km.state == XFRM_STATE_VALID) {
1317                         xfrm[nx++] = x;
1318                         daddr = remote;
1319                         saddr = local;
1320                         continue;
1321                 }
1322                 if (x) {
1323                         error = (x->km.state == XFRM_STATE_ERROR ?
1324                                  -EINVAL : -EAGAIN);
1325                         xfrm_state_put(x);
1326                 }
1327
1328                 if (!tmpl->optional)
1329                         goto fail;
1330         }
1331         return nx;
1332
1333 fail:
1334         for (nx--; nx>=0; nx--)
1335                 xfrm_state_put(xfrm[nx]);
1336         return error;
1337 }
1338
1339 static int
1340 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1341                   struct xfrm_state **xfrm,
1342                   unsigned short family)
1343 {
1344         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1345         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1346         int cnx = 0;
1347         int error;
1348         int ret;
1349         int i;
1350
1351         for (i = 0; i < npols; i++) {
1352                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1353                         error = -ENOBUFS;
1354                         goto fail;
1355                 }
1356
1357                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1358                 if (ret < 0) {
1359                         error = ret;
1360                         goto fail;
1361                 } else
1362                         cnx += ret;
1363         }
1364
1365         /* found states are sorted for outbound processing */
1366         if (npols > 1)
1367                 xfrm_state_sort(xfrm, tpp, cnx, family);
1368
1369         return cnx;
1370
1371  fail:
1372         for (cnx--; cnx>=0; cnx--)
1373                 xfrm_state_put(tpp[cnx]);
1374         return error;
1375
1376 }
1377
1378 /* Check that the bundle accepts the flow and its components are
1379  * still valid.
1380  */
1381
1382 static struct dst_entry *
1383 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1384 {
1385         struct dst_entry *x;
1386         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1387         if (unlikely(afinfo == NULL))
1388                 return ERR_PTR(-EINVAL);
1389         x = afinfo->find_bundle(fl, policy);
1390         xfrm_policy_put_afinfo(afinfo);
1391         return x;
1392 }
1393
1394 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1395  * all the metrics... Shortly, bundle a bundle.
1396  */
1397
1398 static int
1399 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1400                    struct flowi *fl, struct dst_entry **dst_p,
1401                    unsigned short family)
1402 {
1403         int err;
1404         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1405         if (unlikely(afinfo == NULL))
1406                 return -EINVAL;
1407         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1408         xfrm_policy_put_afinfo(afinfo);
1409         return err;
1410 }
1411
1412 static int inline
1413 xfrm_dst_alloc_copy(void **target, void *src, int size)
1414 {
1415         if (!*target) {
1416                 *target = kmalloc(size, GFP_ATOMIC);
1417                 if (!*target)
1418                         return -ENOMEM;
1419         }
1420         memcpy(*target, src, size);
1421         return 0;
1422 }
1423
1424 static int inline
1425 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1426 {
1427 #ifdef CONFIG_XFRM_SUB_POLICY
1428         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1429         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1430                                    sel, sizeof(*sel));
1431 #else
1432         return 0;
1433 #endif
1434 }
1435
1436 static int inline
1437 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1438 {
1439 #ifdef CONFIG_XFRM_SUB_POLICY
1440         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1441         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1442 #else
1443         return 0;
1444 #endif
1445 }
1446
1447 static int stale_bundle(struct dst_entry *dst);
1448
1449 /* Main function: finds/creates a bundle for given flow.
1450  *
1451  * At the moment we eat a raw IP route. Mostly to speed up lookups
1452  * on interfaces with disabled IPsec.
1453  */
1454 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1455                   struct sock *sk, int flags)
1456 {
1457         struct xfrm_policy *policy;
1458         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1459         int npols;
1460         int pol_dead;
1461         int xfrm_nr;
1462         int pi;
1463         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1464         struct dst_entry *dst, *dst_orig = *dst_p;
1465         int nx = 0;
1466         int err;
1467         u32 genid;
1468         u16 family;
1469         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1470
1471 restart:
1472         genid = atomic_read(&flow_cache_genid);
1473         policy = NULL;
1474         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1475                 pols[pi] = NULL;
1476         npols = 0;
1477         pol_dead = 0;
1478         xfrm_nr = 0;
1479
1480         if (sk && sk->sk_policy[1]) {
1481                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1482                 if (IS_ERR(policy))
1483                         return PTR_ERR(policy);
1484         }
1485
1486         if (!policy) {
1487                 /* To accelerate a bit...  */
1488                 if ((dst_orig->flags & DST_NOXFRM) ||
1489                     !xfrm_policy_count[XFRM_POLICY_OUT])
1490                         return 0;
1491
1492                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1493                                            dir, xfrm_policy_lookup);
1494                 if (IS_ERR(policy))
1495                         return PTR_ERR(policy);
1496         }
1497
1498         if (!policy)
1499                 return 0;
1500
1501         family = dst_orig->ops->family;
1502         policy->curlft.use_time = get_seconds();
1503         pols[0] = policy;
1504         npols ++;
1505         xfrm_nr += pols[0]->xfrm_nr;
1506
1507         switch (policy->action) {
1508         case XFRM_POLICY_BLOCK:
1509                 /* Prohibit the flow */
1510                 err = -EPERM;
1511                 goto error;
1512
1513         case XFRM_POLICY_ALLOW:
1514 #ifndef CONFIG_XFRM_SUB_POLICY
1515                 if (policy->xfrm_nr == 0) {
1516                         /* Flow passes not transformed. */
1517                         xfrm_pol_put(policy);
1518                         return 0;
1519                 }
1520 #endif
1521
1522                 /* Try to find matching bundle.
1523                  *
1524                  * LATER: help from flow cache. It is optional, this
1525                  * is required only for output policy.
1526                  */
1527                 dst = xfrm_find_bundle(fl, policy, family);
1528                 if (IS_ERR(dst)) {
1529                         err = PTR_ERR(dst);
1530                         goto error;
1531                 }
1532
1533                 if (dst)
1534                         break;
1535
1536 #ifdef CONFIG_XFRM_SUB_POLICY
1537                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1538                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1539                                                             fl, family,
1540                                                             XFRM_POLICY_OUT);
1541                         if (pols[1]) {
1542                                 if (IS_ERR(pols[1])) {
1543                                         err = PTR_ERR(pols[1]);
1544                                         goto error;
1545                                 }
1546                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1547                                         err = -EPERM;
1548                                         goto error;
1549                                 }
1550                                 npols ++;
1551                                 xfrm_nr += pols[1]->xfrm_nr;
1552                         }
1553                 }
1554
1555                 /*
1556                  * Because neither flowi nor bundle information knows about
1557                  * transformation template size. On more than one policy usage
1558                  * we can realize whether all of them is bypass or not after
1559                  * they are searched. See above not-transformed bypass
1560                  * is surrounded by non-sub policy configuration, too.
1561                  */
1562                 if (xfrm_nr == 0) {
1563                         /* Flow passes not transformed. */
1564                         xfrm_pols_put(pols, npols);
1565                         return 0;
1566                 }
1567
1568 #endif
1569                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1570
1571                 if (unlikely(nx<0)) {
1572                         err = nx;
1573                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1574                                 /* EREMOTE tells the caller to generate
1575                                  * a one-shot blackhole route.
1576                                  */
1577                                 xfrm_pol_put(policy);
1578                                 return -EREMOTE;
1579                         }
1580                         if (err == -EAGAIN && flags) {
1581                                 DECLARE_WAITQUEUE(wait, current);
1582
1583                                 add_wait_queue(&km_waitq, &wait);
1584                                 set_current_state(TASK_INTERRUPTIBLE);
1585                                 schedule();
1586                                 set_current_state(TASK_RUNNING);
1587                                 remove_wait_queue(&km_waitq, &wait);
1588
1589                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1590
1591                                 if (nx == -EAGAIN && signal_pending(current)) {
1592                                         err = -ERESTART;
1593                                         goto error;
1594                                 }
1595                                 if (nx == -EAGAIN ||
1596                                     genid != atomic_read(&flow_cache_genid)) {
1597                                         xfrm_pols_put(pols, npols);
1598                                         goto restart;
1599                                 }
1600                                 err = nx;
1601                         }
1602                         if (err < 0)
1603                                 goto error;
1604                 }
1605                 if (nx == 0) {
1606                         /* Flow passes not transformed. */
1607                         xfrm_pols_put(pols, npols);
1608                         return 0;
1609                 }
1610
1611                 dst = dst_orig;
1612                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1613
1614                 if (unlikely(err)) {
1615                         int i;
1616                         for (i=0; i<nx; i++)
1617                                 xfrm_state_put(xfrm[i]);
1618                         goto error;
1619                 }
1620
1621                 for (pi = 0; pi < npols; pi++) {
1622                         read_lock_bh(&pols[pi]->lock);
1623                         pol_dead |= pols[pi]->dead;
1624                         read_unlock_bh(&pols[pi]->lock);
1625                 }
1626
1627                 write_lock_bh(&policy->lock);
1628                 if (unlikely(pol_dead || stale_bundle(dst))) {
1629                         /* Wow! While we worked on resolving, this
1630                          * policy has gone. Retry. It is not paranoia,
1631                          * we just cannot enlist new bundle to dead object.
1632                          * We can't enlist stable bundles either.
1633                          */
1634                         write_unlock_bh(&policy->lock);
1635                         if (dst)
1636                                 dst_free(dst);
1637
1638                         err = -EHOSTUNREACH;
1639                         goto error;
1640                 }
1641
1642                 if (npols > 1)
1643                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1644                 else
1645                         err = xfrm_dst_update_origin(dst, fl);
1646                 if (unlikely(err)) {
1647                         write_unlock_bh(&policy->lock);
1648                         if (dst)
1649                                 dst_free(dst);
1650                         goto error;
1651                 }
1652
1653                 dst->next = policy->bundles;
1654                 policy->bundles = dst;
1655                 dst_hold(dst);
1656                 write_unlock_bh(&policy->lock);
1657         }
1658         *dst_p = dst;
1659         dst_release(dst_orig);
1660         xfrm_pols_put(pols, npols);
1661         return 0;
1662
1663 error:
1664         dst_release(dst_orig);
1665         xfrm_pols_put(pols, npols);
1666         *dst_p = NULL;
1667         return err;
1668 }
1669 EXPORT_SYMBOL(__xfrm_lookup);
1670
1671 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1672                 struct sock *sk, int flags)
1673 {
1674         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1675
1676         if (err == -EREMOTE) {
1677                 dst_release(*dst_p);
1678                 *dst_p = NULL;
1679                 err = -EAGAIN;
1680         }
1681
1682         return err;
1683 }
1684 EXPORT_SYMBOL(xfrm_lookup);
1685
1686 static inline int
1687 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1688 {
1689         struct xfrm_state *x;
1690         int err;
1691
1692         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1693                 return 0;
1694         x = skb->sp->xvec[idx];
1695         if (!x->type->reject)
1696                 return 0;
1697         xfrm_state_hold(x);
1698         err = x->type->reject(x, skb, fl);
1699         xfrm_state_put(x);
1700         return err;
1701 }
1702
1703 /* When skb is transformed back to its "native" form, we have to
1704  * check policy restrictions. At the moment we make this in maximally
1705  * stupid way. Shame on me. :-) Of course, connected sockets must
1706  * have policy cached at them.
1707  */
1708
1709 static inline int
1710 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1711               unsigned short family)
1712 {
1713         if (xfrm_state_kern(x))
1714                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1715         return  x->id.proto == tmpl->id.proto &&
1716                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1717                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1718                 x->props.mode == tmpl->mode &&
1719                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1720                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1721                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1722                   xfrm_state_addr_cmp(tmpl, x, family));
1723 }
1724
1725 /*
1726  * 0 or more than 0 is returned when validation is succeeded (either bypass
1727  * because of optional transport mode, or next index of the mathced secpath
1728  * state with the template.
1729  * -1 is returned when no matching template is found.
1730  * Otherwise "-2 - errored_index" is returned.
1731  */
1732 static inline int
1733 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1734                unsigned short family)
1735 {
1736         int idx = start;
1737
1738         if (tmpl->optional) {
1739                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1740                         return start;
1741         } else
1742                 start = -1;
1743         for (; idx < sp->len; idx++) {
1744                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1745                         return ++idx;
1746                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1747                         if (start == -1)
1748                                 start = -2-idx;
1749                         break;
1750                 }
1751         }
1752         return start;
1753 }
1754
1755 int
1756 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1757 {
1758         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1759         int err;
1760
1761         if (unlikely(afinfo == NULL))
1762                 return -EAFNOSUPPORT;
1763
1764         afinfo->decode_session(skb, fl);
1765         err = security_xfrm_decode_session(skb, &fl->secid);
1766         xfrm_policy_put_afinfo(afinfo);
1767         return err;
1768 }
1769 EXPORT_SYMBOL(xfrm_decode_session);
1770
1771 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1772 {
1773         for (; k < sp->len; k++) {
1774                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1775                         *idxp = k;
1776                         return 1;
1777                 }
1778         }
1779
1780         return 0;
1781 }
1782
1783 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1784                         unsigned short family)
1785 {
1786         struct xfrm_policy *pol;
1787         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1788         int npols = 0;
1789         int xfrm_nr;
1790         int pi;
1791         struct flowi fl;
1792         u8 fl_dir = policy_to_flow_dir(dir);
1793         int xerr_idx = -1;
1794
1795         if (xfrm_decode_session(skb, &fl, family) < 0)
1796                 return 0;
1797         nf_nat_decode_session(skb, &fl, family);
1798
1799         /* First, check used SA against their selectors. */
1800         if (skb->sp) {
1801                 int i;
1802
1803                 for (i=skb->sp->len-1; i>=0; i--) {
1804                         struct xfrm_state *x = skb->sp->xvec[i];
1805                         if (!xfrm_selector_match(&x->sel, &fl, family))
1806                                 return 0;
1807                 }
1808         }
1809
1810         pol = NULL;
1811         if (sk && sk->sk_policy[dir]) {
1812                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1813                 if (IS_ERR(pol))
1814                         return 0;
1815         }
1816
1817         if (!pol)
1818                 pol = flow_cache_lookup(&fl, family, fl_dir,
1819                                         xfrm_policy_lookup);
1820
1821         if (IS_ERR(pol))
1822                 return 0;
1823
1824         if (!pol) {
1825                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1826                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1827                         return 0;
1828                 }
1829                 return 1;
1830         }
1831
1832         pol->curlft.use_time = get_seconds();
1833
1834         pols[0] = pol;
1835         npols ++;
1836 #ifdef CONFIG_XFRM_SUB_POLICY
1837         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1838                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1839                                                     &fl, family,
1840                                                     XFRM_POLICY_IN);
1841                 if (pols[1]) {
1842                         if (IS_ERR(pols[1]))
1843                                 return 0;
1844                         pols[1]->curlft.use_time = get_seconds();
1845                         npols ++;
1846                 }
1847         }
1848 #endif
1849
1850         if (pol->action == XFRM_POLICY_ALLOW) {
1851                 struct sec_path *sp;
1852                 static struct sec_path dummy;
1853                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1854                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1855                 struct xfrm_tmpl **tpp = tp;
1856                 int ti = 0;
1857                 int i, k;
1858
1859                 if ((sp = skb->sp) == NULL)
1860                         sp = &dummy;
1861
1862                 for (pi = 0; pi < npols; pi++) {
1863                         if (pols[pi] != pol &&
1864                             pols[pi]->action != XFRM_POLICY_ALLOW)
1865                                 goto reject;
1866                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1867                                 goto reject_error;
1868                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1869                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1870                 }
1871                 xfrm_nr = ti;
1872                 if (npols > 1) {
1873                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1874                         tpp = stp;
1875                 }
1876
1877                 /* For each tunnel xfrm, find the first matching tmpl.
1878                  * For each tmpl before that, find corresponding xfrm.
1879                  * Order is _important_. Later we will implement
1880                  * some barriers, but at the moment barriers
1881                  * are implied between each two transformations.
1882                  */
1883                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1884                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1885                         if (k < 0) {
1886                                 if (k < -1)
1887                                         /* "-2 - errored_index" returned */
1888                                         xerr_idx = -(2+k);
1889                                 goto reject;
1890                         }
1891                 }
1892
1893                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1894                         goto reject;
1895
1896                 xfrm_pols_put(pols, npols);
1897                 return 1;
1898         }
1899
1900 reject:
1901         xfrm_secpath_reject(xerr_idx, skb, &fl);
1902 reject_error:
1903         xfrm_pols_put(pols, npols);
1904         return 0;
1905 }
1906 EXPORT_SYMBOL(__xfrm_policy_check);
1907
1908 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1909 {
1910         struct flowi fl;
1911
1912         if (xfrm_decode_session(skb, &fl, family) < 0)
1913                 return 0;
1914
1915         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1916 }
1917 EXPORT_SYMBOL(__xfrm_route_forward);
1918
1919 /* Optimize later using cookies and generation ids. */
1920
1921 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1922 {
1923         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1924          * to "-1" to force all XFRM destinations to get validated by
1925          * dst_ops->check on every use.  We do this because when a
1926          * normal route referenced by an XFRM dst is obsoleted we do
1927          * not go looking around for all parent referencing XFRM dsts
1928          * so that we can invalidate them.  It is just too much work.
1929          * Instead we make the checks here on every use.  For example:
1930          *
1931          *      XFRM dst A --> IPv4 dst X
1932          *
1933          * X is the "xdst->route" of A (X is also the "dst->path" of A
1934          * in this example).  If X is marked obsolete, "A" will not
1935          * notice.  That's what we are validating here via the
1936          * stale_bundle() check.
1937          *
1938          * When a policy's bundle is pruned, we dst_free() the XFRM
1939          * dst which causes it's ->obsolete field to be set to a
1940          * positive non-zero integer.  If an XFRM dst has been pruned
1941          * like this, we want to force a new route lookup.
1942          */
1943         if (dst->obsolete < 0 && !stale_bundle(dst))
1944                 return dst;
1945
1946         return NULL;
1947 }
1948
1949 static int stale_bundle(struct dst_entry *dst)
1950 {
1951         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1952 }
1953
1954 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1955 {
1956         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1957                 dst->dev = &loopback_dev;
1958                 dev_hold(&loopback_dev);
1959                 dev_put(dev);
1960         }
1961 }
1962 EXPORT_SYMBOL(xfrm_dst_ifdown);
1963
1964 static void xfrm_link_failure(struct sk_buff *skb)
1965 {
1966         /* Impossible. Such dst must be popped before reaches point of failure. */
1967         return;
1968 }
1969
1970 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1971 {
1972         if (dst) {
1973                 if (dst->obsolete) {
1974                         dst_release(dst);
1975                         dst = NULL;
1976                 }
1977         }
1978         return dst;
1979 }
1980
1981 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1982 {
1983         struct dst_entry *dst, **dstp;
1984
1985         write_lock(&pol->lock);
1986         dstp = &pol->bundles;
1987         while ((dst=*dstp) != NULL) {
1988                 if (func(dst)) {
1989                         *dstp = dst->next;
1990                         dst->next = *gc_list_p;
1991                         *gc_list_p = dst;
1992                 } else {
1993                         dstp = &dst->next;
1994                 }
1995         }
1996         write_unlock(&pol->lock);
1997 }
1998
1999 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2000 {
2001         struct dst_entry *gc_list = NULL;
2002         int dir;
2003
2004         read_lock_bh(&xfrm_policy_lock);
2005         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2006                 struct xfrm_policy *pol;
2007                 struct hlist_node *entry;
2008                 struct hlist_head *table;
2009                 int i;
2010
2011                 hlist_for_each_entry(pol, entry,
2012                                      &xfrm_policy_inexact[dir], bydst)
2013                         prune_one_bundle(pol, func, &gc_list);
2014
2015                 table = xfrm_policy_bydst[dir].table;
2016                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2017                         hlist_for_each_entry(pol, entry, table + i, bydst)
2018                                 prune_one_bundle(pol, func, &gc_list);
2019                 }
2020         }
2021         read_unlock_bh(&xfrm_policy_lock);
2022
2023         while (gc_list) {
2024                 struct dst_entry *dst = gc_list;
2025                 gc_list = dst->next;
2026                 dst_free(dst);
2027         }
2028 }
2029
2030 static int unused_bundle(struct dst_entry *dst)
2031 {
2032         return !atomic_read(&dst->__refcnt);
2033 }
2034
2035 static void __xfrm_garbage_collect(void)
2036 {
2037         xfrm_prune_bundles(unused_bundle);
2038 }
2039
2040 static int xfrm_flush_bundles(void)
2041 {
2042         xfrm_prune_bundles(stale_bundle);
2043         return 0;
2044 }
2045
2046 void xfrm_init_pmtu(struct dst_entry *dst)
2047 {
2048         do {
2049                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2050                 u32 pmtu, route_mtu_cached;
2051
2052                 pmtu = dst_mtu(dst->child);
2053                 xdst->child_mtu_cached = pmtu;
2054
2055                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2056
2057                 route_mtu_cached = dst_mtu(xdst->route);
2058                 xdst->route_mtu_cached = route_mtu_cached;
2059
2060                 if (pmtu > route_mtu_cached)
2061                         pmtu = route_mtu_cached;
2062
2063                 dst->metrics[RTAX_MTU-1] = pmtu;
2064         } while ((dst = dst->next));
2065 }
2066
2067 EXPORT_SYMBOL(xfrm_init_pmtu);
2068
2069 /* Check that the bundle accepts the flow and its components are
2070  * still valid.
2071  */
2072
2073 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2074                 struct flowi *fl, int family, int strict)
2075 {
2076         struct dst_entry *dst = &first->u.dst;
2077         struct xfrm_dst *last;
2078         u32 mtu;
2079
2080         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2081             (dst->dev && !netif_running(dst->dev)))
2082                 return 0;
2083 #ifdef CONFIG_XFRM_SUB_POLICY
2084         if (fl) {
2085                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2086                         return 0;
2087                 if (first->partner &&
2088                     !xfrm_selector_match(first->partner, fl, family))
2089                         return 0;
2090         }
2091 #endif
2092
2093         last = NULL;
2094
2095         do {
2096                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2097
2098                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2099                         return 0;
2100                 if (fl && pol &&
2101                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2102                         return 0;
2103                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2104                         return 0;
2105                 if (xdst->genid != dst->xfrm->genid)
2106                         return 0;
2107
2108                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
2109                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2110                         return 0;
2111
2112                 mtu = dst_mtu(dst->child);
2113                 if (xdst->child_mtu_cached != mtu) {
2114                         last = xdst;
2115                         xdst->child_mtu_cached = mtu;
2116                 }
2117
2118                 if (!dst_check(xdst->route, xdst->route_cookie))
2119                         return 0;
2120                 mtu = dst_mtu(xdst->route);
2121                 if (xdst->route_mtu_cached != mtu) {
2122                         last = xdst;
2123                         xdst->route_mtu_cached = mtu;
2124                 }
2125
2126                 dst = dst->child;
2127         } while (dst->xfrm);
2128
2129         if (likely(!last))
2130                 return 1;
2131
2132         mtu = last->child_mtu_cached;
2133         for (;;) {
2134                 dst = &last->u.dst;
2135
2136                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2137                 if (mtu > last->route_mtu_cached)
2138                         mtu = last->route_mtu_cached;
2139                 dst->metrics[RTAX_MTU-1] = mtu;
2140
2141                 if (last == first)
2142                         break;
2143
2144                 last = (struct xfrm_dst *)last->u.dst.next;
2145                 last->child_mtu_cached = mtu;
2146         }
2147
2148         return 1;
2149 }
2150
2151 EXPORT_SYMBOL(xfrm_bundle_ok);
2152
2153 #ifdef CONFIG_AUDITSYSCALL
2154 /* Audit addition and deletion of SAs and ipsec policy */
2155
2156 void xfrm_audit_log(uid_t auid, u32 sid, int type, int result,
2157                     struct xfrm_policy *xp, struct xfrm_state *x)
2158 {
2159
2160         char *secctx;
2161         u32 secctx_len;
2162         struct xfrm_sec_ctx *sctx = NULL;
2163         struct audit_buffer *audit_buf;
2164         int family;
2165         extern int audit_enabled;
2166
2167         if (audit_enabled == 0)
2168                 return;
2169
2170         BUG_ON((type == AUDIT_MAC_IPSEC_ADDSA ||
2171                 type == AUDIT_MAC_IPSEC_DELSA) && !x);
2172         BUG_ON((type == AUDIT_MAC_IPSEC_ADDSPD ||
2173                 type == AUDIT_MAC_IPSEC_DELSPD) && !xp);
2174
2175         audit_buf = audit_log_start(current->audit_context, GFP_ATOMIC, type);
2176         if (audit_buf == NULL)
2177                 return;
2178
2179         switch(type) {
2180         case AUDIT_MAC_IPSEC_ADDSA:
2181                 audit_log_format(audit_buf, "SAD add: auid=%u", auid);
2182                 break;
2183         case AUDIT_MAC_IPSEC_DELSA:
2184                 audit_log_format(audit_buf, "SAD delete: auid=%u", auid);
2185                 break;
2186         case AUDIT_MAC_IPSEC_ADDSPD:
2187                 audit_log_format(audit_buf, "SPD add: auid=%u", auid);
2188                 break;
2189         case AUDIT_MAC_IPSEC_DELSPD:
2190                 audit_log_format(audit_buf, "SPD delete: auid=%u", auid);
2191                 break;
2192         default:
2193                 return;
2194         }
2195
2196         if (sid != 0 &&
2197                 security_secid_to_secctx(sid, &secctx, &secctx_len) == 0)
2198                 audit_log_format(audit_buf, " subj=%s", secctx);
2199         else
2200                 audit_log_task_context(audit_buf);
2201
2202         if (xp) {
2203                 family = xp->selector.family;
2204                 if (xp->security)
2205                         sctx = xp->security;
2206         } else {
2207                 family = x->props.family;
2208                 if (x->security)
2209                         sctx = x->security;
2210         }
2211
2212         if (sctx)
2213                 audit_log_format(audit_buf,
2214                                 " sec_alg=%u sec_doi=%u sec_obj=%s",
2215                                 sctx->ctx_alg, sctx->ctx_doi, sctx->ctx_str);
2216
2217         switch(family) {
2218         case AF_INET:
2219                 {
2220                         struct in_addr saddr, daddr;
2221                         if (xp) {
2222                                 saddr.s_addr = xp->selector.saddr.a4;
2223                                 daddr.s_addr = xp->selector.daddr.a4;
2224                         } else {
2225                                 saddr.s_addr = x->props.saddr.a4;
2226                                 daddr.s_addr = x->id.daddr.a4;
2227                         }
2228                         audit_log_format(audit_buf,
2229                                          " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2230                                          NIPQUAD(saddr), NIPQUAD(daddr));
2231                 }
2232                         break;
2233         case AF_INET6:
2234                 {
2235                         struct in6_addr saddr6, daddr6;
2236                         if (xp) {
2237                                 memcpy(&saddr6, xp->selector.saddr.a6,
2238                                         sizeof(struct in6_addr));
2239                                 memcpy(&daddr6, xp->selector.daddr.a6,
2240                                         sizeof(struct in6_addr));
2241                         } else {
2242                                 memcpy(&saddr6, x->props.saddr.a6,
2243                                         sizeof(struct in6_addr));
2244                                 memcpy(&daddr6, x->id.daddr.a6,
2245                                         sizeof(struct in6_addr));
2246                         }
2247                         audit_log_format(audit_buf,
2248                                          " src=" NIP6_FMT " dst=" NIP6_FMT,
2249                                          NIP6(saddr6), NIP6(daddr6));
2250                 }
2251                 break;
2252         }
2253
2254         if (x)
2255                 audit_log_format(audit_buf, " spi=%lu(0x%lx) protocol=%s",
2256                                 (unsigned long)ntohl(x->id.spi),
2257                                 (unsigned long)ntohl(x->id.spi),
2258                                 x->id.proto == IPPROTO_AH ? "AH" :
2259                                 (x->id.proto == IPPROTO_ESP ?
2260                                 "ESP" : "IPCOMP"));
2261
2262         audit_log_format(audit_buf, " res=%u", result);
2263         audit_log_end(audit_buf);
2264 }
2265
2266 EXPORT_SYMBOL(xfrm_audit_log);
2267 #endif /* CONFIG_AUDITSYSCALL */
2268
2269 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2270 {
2271         int err = 0;
2272         if (unlikely(afinfo == NULL))
2273                 return -EINVAL;
2274         if (unlikely(afinfo->family >= NPROTO))
2275                 return -EAFNOSUPPORT;
2276         write_lock_bh(&xfrm_policy_afinfo_lock);
2277         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2278                 err = -ENOBUFS;
2279         else {
2280                 struct dst_ops *dst_ops = afinfo->dst_ops;
2281                 if (likely(dst_ops->kmem_cachep == NULL))
2282                         dst_ops->kmem_cachep = xfrm_dst_cache;
2283                 if (likely(dst_ops->check == NULL))
2284                         dst_ops->check = xfrm_dst_check;
2285                 if (likely(dst_ops->negative_advice == NULL))
2286                         dst_ops->negative_advice = xfrm_negative_advice;
2287                 if (likely(dst_ops->link_failure == NULL))
2288                         dst_ops->link_failure = xfrm_link_failure;
2289                 if (likely(afinfo->garbage_collect == NULL))
2290                         afinfo->garbage_collect = __xfrm_garbage_collect;
2291                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2292         }
2293         write_unlock_bh(&xfrm_policy_afinfo_lock);
2294         return err;
2295 }
2296 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2297
2298 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2299 {
2300         int err = 0;
2301         if (unlikely(afinfo == NULL))
2302                 return -EINVAL;
2303         if (unlikely(afinfo->family >= NPROTO))
2304                 return -EAFNOSUPPORT;
2305         write_lock_bh(&xfrm_policy_afinfo_lock);
2306         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2307                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2308                         err = -EINVAL;
2309                 else {
2310                         struct dst_ops *dst_ops = afinfo->dst_ops;
2311                         xfrm_policy_afinfo[afinfo->family] = NULL;
2312                         dst_ops->kmem_cachep = NULL;
2313                         dst_ops->check = NULL;
2314                         dst_ops->negative_advice = NULL;
2315                         dst_ops->link_failure = NULL;
2316                         afinfo->garbage_collect = NULL;
2317                 }
2318         }
2319         write_unlock_bh(&xfrm_policy_afinfo_lock);
2320         return err;
2321 }
2322 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2323
2324 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2325 {
2326         struct xfrm_policy_afinfo *afinfo;
2327         if (unlikely(family >= NPROTO))
2328                 return NULL;
2329         read_lock(&xfrm_policy_afinfo_lock);
2330         afinfo = xfrm_policy_afinfo[family];
2331         if (unlikely(!afinfo))
2332                 read_unlock(&xfrm_policy_afinfo_lock);
2333         return afinfo;
2334 }
2335
2336 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2337 {
2338         read_unlock(&xfrm_policy_afinfo_lock);
2339 }
2340
2341 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2342 {
2343         struct xfrm_policy_afinfo *afinfo;
2344         if (unlikely(family >= NPROTO))
2345                 return NULL;
2346         write_lock_bh(&xfrm_policy_afinfo_lock);
2347         afinfo = xfrm_policy_afinfo[family];
2348         if (unlikely(!afinfo))
2349                 write_unlock_bh(&xfrm_policy_afinfo_lock);
2350         return afinfo;
2351 }
2352
2353 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2354 {
2355         write_unlock_bh(&xfrm_policy_afinfo_lock);
2356 }
2357
2358 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2359 {
2360         switch (event) {
2361         case NETDEV_DOWN:
2362                 xfrm_flush_bundles();
2363         }
2364         return NOTIFY_DONE;
2365 }
2366
2367 static struct notifier_block xfrm_dev_notifier = {
2368         xfrm_dev_event,
2369         NULL,
2370         0
2371 };
2372
2373 static void __init xfrm_policy_init(void)
2374 {
2375         unsigned int hmask, sz;
2376         int dir;
2377
2378         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2379                                            sizeof(struct xfrm_dst),
2380                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2381                                            NULL, NULL);
2382
2383         hmask = 8 - 1;
2384         sz = (hmask+1) * sizeof(struct hlist_head);
2385
2386         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2387         xfrm_idx_hmask = hmask;
2388         if (!xfrm_policy_byidx)
2389                 panic("XFRM: failed to allocate byidx hash\n");
2390
2391         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2392                 struct xfrm_policy_hash *htab;
2393
2394                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2395
2396                 htab = &xfrm_policy_bydst[dir];
2397                 htab->table = xfrm_hash_alloc(sz);
2398                 htab->hmask = hmask;
2399                 if (!htab->table)
2400                         panic("XFRM: failed to allocate bydst hash\n");
2401         }
2402
2403         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2404         register_netdevice_notifier(&xfrm_dev_notifier);
2405 }
2406
2407 void __init xfrm_init(void)
2408 {
2409         xfrm_state_init();
2410         xfrm_policy_init();
2411         xfrm_input_init();
2412 }
2413
2414 #ifdef CONFIG_XFRM_MIGRATE
2415 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2416                                        struct xfrm_selector *sel_tgt)
2417 {
2418         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2419                 if (sel_tgt->family == sel_cmp->family &&
2420                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2421                                   sel_cmp->family) == 0 &&
2422                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2423                                   sel_cmp->family) == 0 &&
2424                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2425                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2426                         return 1;
2427                 }
2428         } else {
2429                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2430                         return 1;
2431                 }
2432         }
2433         return 0;
2434 }
2435
2436 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2437                                                      u8 dir, u8 type)
2438 {
2439         struct xfrm_policy *pol, *ret = NULL;
2440         struct hlist_node *entry;
2441         struct hlist_head *chain;
2442         u32 priority = ~0U;
2443
2444         read_lock_bh(&xfrm_policy_lock);
2445         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2446         hlist_for_each_entry(pol, entry, chain, bydst) {
2447                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2448                     pol->type == type) {
2449                         ret = pol;
2450                         priority = ret->priority;
2451                         break;
2452                 }
2453         }
2454         chain = &xfrm_policy_inexact[dir];
2455         hlist_for_each_entry(pol, entry, chain, bydst) {
2456                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2457                     pol->type == type &&
2458                     pol->priority < priority) {
2459                         ret = pol;
2460                         break;
2461                 }
2462         }
2463
2464         if (ret)
2465                 xfrm_pol_hold(ret);
2466
2467         read_unlock_bh(&xfrm_policy_lock);
2468
2469         return ret;
2470 }
2471
2472 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2473 {
2474         int match = 0;
2475
2476         if (t->mode == m->mode && t->id.proto == m->proto &&
2477             (m->reqid == 0 || t->reqid == m->reqid)) {
2478                 switch (t->mode) {
2479                 case XFRM_MODE_TUNNEL:
2480                 case XFRM_MODE_BEET:
2481                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2482                                           m->old_family) == 0 &&
2483                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2484                                           m->old_family) == 0) {
2485                                 match = 1;
2486                         }
2487                         break;
2488                 case XFRM_MODE_TRANSPORT:
2489                         /* in case of transport mode, template does not store
2490                            any IP addresses, hence we just compare mode and
2491                            protocol */
2492                         match = 1;
2493                         break;
2494                 default:
2495                         break;
2496                 }
2497         }
2498         return match;
2499 }
2500
2501 /* update endpoint address(es) of template(s) */
2502 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2503                                struct xfrm_migrate *m, int num_migrate)
2504 {
2505         struct xfrm_migrate *mp;
2506         struct dst_entry *dst;
2507         int i, j, n = 0;
2508
2509         write_lock_bh(&pol->lock);
2510         if (unlikely(pol->dead)) {
2511                 /* target policy has been deleted */
2512                 write_unlock_bh(&pol->lock);
2513                 return -ENOENT;
2514         }
2515
2516         for (i = 0; i < pol->xfrm_nr; i++) {
2517                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2518                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2519                                 continue;
2520                         n++;
2521                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL)
2522                                 continue;
2523                         /* update endpoints */
2524                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2525                                sizeof(pol->xfrm_vec[i].id.daddr));
2526                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2527                                sizeof(pol->xfrm_vec[i].saddr));
2528                         pol->xfrm_vec[i].encap_family = mp->new_family;
2529                         /* flush bundles */
2530                         while ((dst = pol->bundles) != NULL) {
2531                                 pol->bundles = dst->next;
2532                                 dst_free(dst);
2533                         }
2534                 }
2535         }
2536
2537         write_unlock_bh(&pol->lock);
2538
2539         if (!n)
2540                 return -ENODATA;
2541
2542         return 0;
2543 }
2544
2545 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2546 {
2547         int i, j;
2548
2549         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2550                 return -EINVAL;
2551
2552         for (i = 0; i < num_migrate; i++) {
2553                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2554                                    m[i].old_family) == 0) &&
2555                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2556                                    m[i].old_family) == 0))
2557                         return -EINVAL;
2558                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2559                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2560                         return -EINVAL;
2561
2562                 /* check if there is any duplicated entry */
2563                 for (j = i + 1; j < num_migrate; j++) {
2564                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2565                                     sizeof(m[i].old_daddr)) &&
2566                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2567                                     sizeof(m[i].old_saddr)) &&
2568                             m[i].proto == m[j].proto &&
2569                             m[i].mode == m[j].mode &&
2570                             m[i].reqid == m[j].reqid &&
2571                             m[i].old_family == m[j].old_family)
2572                                 return -EINVAL;
2573                 }
2574         }
2575
2576         return 0;
2577 }
2578
2579 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2580                  struct xfrm_migrate *m, int num_migrate)
2581 {
2582         int i, err, nx_cur = 0, nx_new = 0;
2583         struct xfrm_policy *pol = NULL;
2584         struct xfrm_state *x, *xc;
2585         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2586         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2587         struct xfrm_migrate *mp;
2588
2589         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2590                 goto out;
2591
2592         /* Stage 1 - find policy */
2593         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2594                 err = -ENOENT;
2595                 goto out;
2596         }
2597
2598         /* Stage 2 - find and update state(s) */
2599         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2600                 if ((x = xfrm_migrate_state_find(mp))) {
2601                         x_cur[nx_cur] = x;
2602                         nx_cur++;
2603                         if ((xc = xfrm_state_migrate(x, mp))) {
2604                                 x_new[nx_new] = xc;
2605                                 nx_new++;
2606                         } else {
2607                                 err = -ENODATA;
2608                                 goto restore_state;
2609                         }
2610                 }
2611         }
2612
2613         /* Stage 3 - update policy */
2614         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2615                 goto restore_state;
2616
2617         /* Stage 4 - delete old state(s) */
2618         if (nx_cur) {
2619                 xfrm_states_put(x_cur, nx_cur);
2620                 xfrm_states_delete(x_cur, nx_cur);
2621         }
2622
2623         /* Stage 5 - announce */
2624         km_migrate(sel, dir, type, m, num_migrate);
2625
2626         xfrm_pol_put(pol);
2627
2628         return 0;
2629 out:
2630         return err;
2631
2632 restore_state:
2633         if (pol)
2634                 xfrm_pol_put(pol);
2635         if (nx_cur)
2636                 xfrm_states_put(x_cur, nx_cur);
2637         if (nx_new)
2638                 xfrm_states_delete(x_new, nx_new);
2639
2640         return err;
2641 }
2642 EXPORT_SYMBOL(xfrm_migrate);
2643 #endif