xfrm: SP lookups signature with mark
[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/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
28 #include <net/dst.h>
29 #include <net/xfrm.h>
30 #include <net/ip.h>
31 #ifdef CONFIG_XFRM_STATISTICS
32 #include <net/snmp.h>
33 #endif
34
35 #include "xfrm_hash.h"
36
37 DEFINE_MUTEX(xfrm_cfg_mutex);
38 EXPORT_SYMBOL(xfrm_cfg_mutex);
39
40 static DEFINE_RWLOCK(xfrm_policy_lock);
41
42 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
43 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
44
45 static struct kmem_cache *xfrm_dst_cache __read_mostly;
46
47 static HLIST_HEAD(xfrm_policy_gc_list);
48 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
49
50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
52 static void xfrm_init_pmtu(struct dst_entry *dst);
53
54 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
55                                                 int dir);
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 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
92                                                   xfrm_address_t *saddr,
93                                                   xfrm_address_t *daddr,
94                                                   int family)
95 {
96         struct xfrm_policy_afinfo *afinfo;
97         struct dst_entry *dst;
98
99         afinfo = xfrm_policy_get_afinfo(family);
100         if (unlikely(afinfo == NULL))
101                 return ERR_PTR(-EAFNOSUPPORT);
102
103         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
104
105         xfrm_policy_put_afinfo(afinfo);
106
107         return dst;
108 }
109
110 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
111                                                 xfrm_address_t *prev_saddr,
112                                                 xfrm_address_t *prev_daddr,
113                                                 int family)
114 {
115         struct net *net = xs_net(x);
116         xfrm_address_t *saddr = &x->props.saddr;
117         xfrm_address_t *daddr = &x->id.daddr;
118         struct dst_entry *dst;
119
120         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
121                 saddr = x->coaddr;
122                 daddr = prev_daddr;
123         }
124         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
125                 saddr = prev_saddr;
126                 daddr = x->coaddr;
127         }
128
129         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
130
131         if (!IS_ERR(dst)) {
132                 if (prev_saddr != saddr)
133                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
134                 if (prev_daddr != daddr)
135                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
136         }
137
138         return dst;
139 }
140
141 static inline unsigned long make_jiffies(long secs)
142 {
143         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
144                 return MAX_SCHEDULE_TIMEOUT-1;
145         else
146                 return secs*HZ;
147 }
148
149 static void xfrm_policy_timer(unsigned long data)
150 {
151         struct xfrm_policy *xp = (struct xfrm_policy*)data;
152         unsigned long now = get_seconds();
153         long next = LONG_MAX;
154         int warn = 0;
155         int dir;
156
157         read_lock(&xp->lock);
158
159         if (xp->walk.dead)
160                 goto out;
161
162         dir = xfrm_policy_id2dir(xp->index);
163
164         if (xp->lft.hard_add_expires_seconds) {
165                 long tmo = xp->lft.hard_add_expires_seconds +
166                         xp->curlft.add_time - now;
167                 if (tmo <= 0)
168                         goto expired;
169                 if (tmo < next)
170                         next = tmo;
171         }
172         if (xp->lft.hard_use_expires_seconds) {
173                 long tmo = xp->lft.hard_use_expires_seconds +
174                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
175                 if (tmo <= 0)
176                         goto expired;
177                 if (tmo < next)
178                         next = tmo;
179         }
180         if (xp->lft.soft_add_expires_seconds) {
181                 long tmo = xp->lft.soft_add_expires_seconds +
182                         xp->curlft.add_time - now;
183                 if (tmo <= 0) {
184                         warn = 1;
185                         tmo = XFRM_KM_TIMEOUT;
186                 }
187                 if (tmo < next)
188                         next = tmo;
189         }
190         if (xp->lft.soft_use_expires_seconds) {
191                 long tmo = xp->lft.soft_use_expires_seconds +
192                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
193                 if (tmo <= 0) {
194                         warn = 1;
195                         tmo = XFRM_KM_TIMEOUT;
196                 }
197                 if (tmo < next)
198                         next = tmo;
199         }
200
201         if (warn)
202                 km_policy_expired(xp, dir, 0, 0);
203         if (next != LONG_MAX &&
204             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
205                 xfrm_pol_hold(xp);
206
207 out:
208         read_unlock(&xp->lock);
209         xfrm_pol_put(xp);
210         return;
211
212 expired:
213         read_unlock(&xp->lock);
214         if (!xfrm_policy_delete(xp, dir))
215                 km_policy_expired(xp, dir, 1, 0);
216         xfrm_pol_put(xp);
217 }
218
219
220 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
221  * SPD calls.
222  */
223
224 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
225 {
226         struct xfrm_policy *policy;
227
228         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
229
230         if (policy) {
231                 write_pnet(&policy->xp_net, net);
232                 INIT_LIST_HEAD(&policy->walk.all);
233                 INIT_HLIST_NODE(&policy->bydst);
234                 INIT_HLIST_NODE(&policy->byidx);
235                 rwlock_init(&policy->lock);
236                 atomic_set(&policy->refcnt, 1);
237                 setup_timer(&policy->timer, xfrm_policy_timer,
238                                 (unsigned long)policy);
239         }
240         return policy;
241 }
242 EXPORT_SYMBOL(xfrm_policy_alloc);
243
244 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
245
246 void xfrm_policy_destroy(struct xfrm_policy *policy)
247 {
248         BUG_ON(!policy->walk.dead);
249
250         BUG_ON(policy->bundles);
251
252         if (del_timer(&policy->timer))
253                 BUG();
254
255         security_xfrm_policy_free(policy->security);
256         kfree(policy);
257 }
258 EXPORT_SYMBOL(xfrm_policy_destroy);
259
260 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
261 {
262         struct dst_entry *dst;
263
264         while ((dst = policy->bundles) != NULL) {
265                 policy->bundles = dst->next;
266                 dst_free(dst);
267         }
268
269         if (del_timer(&policy->timer))
270                 atomic_dec(&policy->refcnt);
271
272         if (atomic_read(&policy->refcnt) > 1)
273                 flow_cache_flush();
274
275         xfrm_pol_put(policy);
276 }
277
278 static void xfrm_policy_gc_task(struct work_struct *work)
279 {
280         struct xfrm_policy *policy;
281         struct hlist_node *entry, *tmp;
282         struct hlist_head gc_list;
283
284         spin_lock_bh(&xfrm_policy_gc_lock);
285         gc_list.first = xfrm_policy_gc_list.first;
286         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
287         spin_unlock_bh(&xfrm_policy_gc_lock);
288
289         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
290                 xfrm_policy_gc_kill(policy);
291 }
292 static DECLARE_WORK(xfrm_policy_gc_work, xfrm_policy_gc_task);
293
294 /* Rule must be locked. Release descentant resources, announce
295  * entry dead. The rule must be unlinked from lists to the moment.
296  */
297
298 static void xfrm_policy_kill(struct xfrm_policy *policy)
299 {
300         int dead;
301
302         write_lock_bh(&policy->lock);
303         dead = policy->walk.dead;
304         policy->walk.dead = 1;
305         write_unlock_bh(&policy->lock);
306
307         if (unlikely(dead)) {
308                 WARN_ON(1);
309                 return;
310         }
311
312         spin_lock_bh(&xfrm_policy_gc_lock);
313         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
314         spin_unlock_bh(&xfrm_policy_gc_lock);
315
316         schedule_work(&xfrm_policy_gc_work);
317 }
318
319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
320
321 static inline unsigned int idx_hash(struct net *net, u32 index)
322 {
323         return __idx_hash(index, net->xfrm.policy_idx_hmask);
324 }
325
326 static struct hlist_head *policy_hash_bysel(struct net *net, struct xfrm_selector *sel, unsigned short family, int dir)
327 {
328         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
329         unsigned int hash = __sel_hash(sel, family, hmask);
330
331         return (hash == hmask + 1 ?
332                 &net->xfrm.policy_inexact[dir] :
333                 net->xfrm.policy_bydst[dir].table + hash);
334 }
335
336 static struct hlist_head *policy_hash_direct(struct net *net, xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
337 {
338         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
339         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
340
341         return net->xfrm.policy_bydst[dir].table + hash;
342 }
343
344 static void xfrm_dst_hash_transfer(struct hlist_head *list,
345                                    struct hlist_head *ndsttable,
346                                    unsigned int nhashmask)
347 {
348         struct hlist_node *entry, *tmp, *entry0 = NULL;
349         struct xfrm_policy *pol;
350         unsigned int h0 = 0;
351
352 redo:
353         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
354                 unsigned int h;
355
356                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
357                                 pol->family, nhashmask);
358                 if (!entry0) {
359                         hlist_del(entry);
360                         hlist_add_head(&pol->bydst, ndsttable+h);
361                         h0 = h;
362                 } else {
363                         if (h != h0)
364                                 continue;
365                         hlist_del(entry);
366                         hlist_add_after(entry0, &pol->bydst);
367                 }
368                 entry0 = entry;
369         }
370         if (!hlist_empty(list)) {
371                 entry0 = NULL;
372                 goto redo;
373         }
374 }
375
376 static void xfrm_idx_hash_transfer(struct hlist_head *list,
377                                    struct hlist_head *nidxtable,
378                                    unsigned int nhashmask)
379 {
380         struct hlist_node *entry, *tmp;
381         struct xfrm_policy *pol;
382
383         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
384                 unsigned int h;
385
386                 h = __idx_hash(pol->index, nhashmask);
387                 hlist_add_head(&pol->byidx, nidxtable+h);
388         }
389 }
390
391 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
392 {
393         return ((old_hmask + 1) << 1) - 1;
394 }
395
396 static void xfrm_bydst_resize(struct net *net, int dir)
397 {
398         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
399         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
400         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
401         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
402         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
403         int i;
404
405         if (!ndst)
406                 return;
407
408         write_lock_bh(&xfrm_policy_lock);
409
410         for (i = hmask; i >= 0; i--)
411                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
412
413         net->xfrm.policy_bydst[dir].table = ndst;
414         net->xfrm.policy_bydst[dir].hmask = nhashmask;
415
416         write_unlock_bh(&xfrm_policy_lock);
417
418         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
419 }
420
421 static void xfrm_byidx_resize(struct net *net, int total)
422 {
423         unsigned int hmask = net->xfrm.policy_idx_hmask;
424         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
425         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
426         struct hlist_head *oidx = net->xfrm.policy_byidx;
427         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
428         int i;
429
430         if (!nidx)
431                 return;
432
433         write_lock_bh(&xfrm_policy_lock);
434
435         for (i = hmask; i >= 0; i--)
436                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
437
438         net->xfrm.policy_byidx = nidx;
439         net->xfrm.policy_idx_hmask = nhashmask;
440
441         write_unlock_bh(&xfrm_policy_lock);
442
443         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
444 }
445
446 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
447 {
448         unsigned int cnt = net->xfrm.policy_count[dir];
449         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
450
451         if (total)
452                 *total += cnt;
453
454         if ((hmask + 1) < xfrm_policy_hashmax &&
455             cnt > hmask)
456                 return 1;
457
458         return 0;
459 }
460
461 static inline int xfrm_byidx_should_resize(struct net *net, int total)
462 {
463         unsigned int hmask = net->xfrm.policy_idx_hmask;
464
465         if ((hmask + 1) < xfrm_policy_hashmax &&
466             total > hmask)
467                 return 1;
468
469         return 0;
470 }
471
472 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
473 {
474         read_lock_bh(&xfrm_policy_lock);
475         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
476         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
477         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
478         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
479         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
480         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
481         si->spdhcnt = net->xfrm.policy_idx_hmask;
482         si->spdhmcnt = xfrm_policy_hashmax;
483         read_unlock_bh(&xfrm_policy_lock);
484 }
485 EXPORT_SYMBOL(xfrm_spd_getinfo);
486
487 static DEFINE_MUTEX(hash_resize_mutex);
488 static void xfrm_hash_resize(struct work_struct *work)
489 {
490         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
491         int dir, total;
492
493         mutex_lock(&hash_resize_mutex);
494
495         total = 0;
496         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
497                 if (xfrm_bydst_should_resize(net, dir, &total))
498                         xfrm_bydst_resize(net, dir);
499         }
500         if (xfrm_byidx_should_resize(net, total))
501                 xfrm_byidx_resize(net, total);
502
503         mutex_unlock(&hash_resize_mutex);
504 }
505
506 /* Generate new index... KAME seems to generate them ordered by cost
507  * of an absolute inpredictability of ordering of rules. This will not pass. */
508 static u32 xfrm_gen_index(struct net *net, int dir)
509 {
510         static u32 idx_generator;
511
512         for (;;) {
513                 struct hlist_node *entry;
514                 struct hlist_head *list;
515                 struct xfrm_policy *p;
516                 u32 idx;
517                 int found;
518
519                 idx = (idx_generator | dir);
520                 idx_generator += 8;
521                 if (idx == 0)
522                         idx = 8;
523                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
524                 found = 0;
525                 hlist_for_each_entry(p, entry, list, byidx) {
526                         if (p->index == idx) {
527                                 found = 1;
528                                 break;
529                         }
530                 }
531                 if (!found)
532                         return idx;
533         }
534 }
535
536 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
537 {
538         u32 *p1 = (u32 *) s1;
539         u32 *p2 = (u32 *) s2;
540         int len = sizeof(struct xfrm_selector) / sizeof(u32);
541         int i;
542
543         for (i = 0; i < len; i++) {
544                 if (p1[i] != p2[i])
545                         return 1;
546         }
547
548         return 0;
549 }
550
551 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
552 {
553         struct net *net = xp_net(policy);
554         struct xfrm_policy *pol;
555         struct xfrm_policy *delpol;
556         struct hlist_head *chain;
557         struct hlist_node *entry, *newpos;
558         struct dst_entry *gc_list;
559
560         write_lock_bh(&xfrm_policy_lock);
561         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
562         delpol = NULL;
563         newpos = NULL;
564         hlist_for_each_entry(pol, entry, chain, bydst) {
565                 if (pol->type == policy->type &&
566                     !selector_cmp(&pol->selector, &policy->selector) &&
567                     xfrm_sec_ctx_match(pol->security, policy->security) &&
568                     !WARN_ON(delpol)) {
569                         if (excl) {
570                                 write_unlock_bh(&xfrm_policy_lock);
571                                 return -EEXIST;
572                         }
573                         delpol = pol;
574                         if (policy->priority > pol->priority)
575                                 continue;
576                 } else if (policy->priority >= pol->priority) {
577                         newpos = &pol->bydst;
578                         continue;
579                 }
580                 if (delpol)
581                         break;
582         }
583         if (newpos)
584                 hlist_add_after(newpos, &policy->bydst);
585         else
586                 hlist_add_head(&policy->bydst, chain);
587         xfrm_pol_hold(policy);
588         net->xfrm.policy_count[dir]++;
589         atomic_inc(&flow_cache_genid);
590         if (delpol)
591                 __xfrm_policy_unlink(delpol, dir);
592         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
593         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
594         policy->curlft.add_time = get_seconds();
595         policy->curlft.use_time = 0;
596         if (!mod_timer(&policy->timer, jiffies + HZ))
597                 xfrm_pol_hold(policy);
598         list_add(&policy->walk.all, &net->xfrm.policy_all);
599         write_unlock_bh(&xfrm_policy_lock);
600
601         if (delpol)
602                 xfrm_policy_kill(delpol);
603         else if (xfrm_bydst_should_resize(net, dir, NULL))
604                 schedule_work(&net->xfrm.policy_hash_work);
605
606         read_lock_bh(&xfrm_policy_lock);
607         gc_list = NULL;
608         entry = &policy->bydst;
609         hlist_for_each_entry_continue(policy, entry, bydst) {
610                 struct dst_entry *dst;
611
612                 write_lock(&policy->lock);
613                 dst = policy->bundles;
614                 if (dst) {
615                         struct dst_entry *tail = dst;
616                         while (tail->next)
617                                 tail = tail->next;
618                         tail->next = gc_list;
619                         gc_list = dst;
620
621                         policy->bundles = NULL;
622                 }
623                 write_unlock(&policy->lock);
624         }
625         read_unlock_bh(&xfrm_policy_lock);
626
627         while (gc_list) {
628                 struct dst_entry *dst = gc_list;
629
630                 gc_list = dst->next;
631                 dst_free(dst);
632         }
633
634         return 0;
635 }
636 EXPORT_SYMBOL(xfrm_policy_insert);
637
638 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
639                                           int dir, struct xfrm_selector *sel,
640                                           struct xfrm_sec_ctx *ctx, int delete,
641                                           int *err)
642 {
643         struct xfrm_policy *pol, *ret;
644         struct hlist_head *chain;
645         struct hlist_node *entry;
646
647         *err = 0;
648         write_lock_bh(&xfrm_policy_lock);
649         chain = policy_hash_bysel(net, sel, sel->family, dir);
650         ret = NULL;
651         hlist_for_each_entry(pol, entry, chain, bydst) {
652                 if (pol->type == type &&
653                     !selector_cmp(sel, &pol->selector) &&
654                     xfrm_sec_ctx_match(ctx, pol->security)) {
655                         xfrm_pol_hold(pol);
656                         if (delete) {
657                                 *err = security_xfrm_policy_delete(
658                                                                 pol->security);
659                                 if (*err) {
660                                         write_unlock_bh(&xfrm_policy_lock);
661                                         return pol;
662                                 }
663                                 __xfrm_policy_unlink(pol, dir);
664                         }
665                         ret = pol;
666                         break;
667                 }
668         }
669         write_unlock_bh(&xfrm_policy_lock);
670
671         if (ret && delete) {
672                 atomic_inc(&flow_cache_genid);
673                 xfrm_policy_kill(ret);
674         }
675         return ret;
676 }
677 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
678
679 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
680                                      int dir, u32 id, int delete, int *err)
681 {
682         struct xfrm_policy *pol, *ret;
683         struct hlist_head *chain;
684         struct hlist_node *entry;
685
686         *err = -ENOENT;
687         if (xfrm_policy_id2dir(id) != dir)
688                 return NULL;
689
690         *err = 0;
691         write_lock_bh(&xfrm_policy_lock);
692         chain = net->xfrm.policy_byidx + idx_hash(net, id);
693         ret = NULL;
694         hlist_for_each_entry(pol, entry, chain, byidx) {
695                 if (pol->type == type && pol->index == id) {
696                         xfrm_pol_hold(pol);
697                         if (delete) {
698                                 *err = security_xfrm_policy_delete(
699                                                                 pol->security);
700                                 if (*err) {
701                                         write_unlock_bh(&xfrm_policy_lock);
702                                         return pol;
703                                 }
704                                 __xfrm_policy_unlink(pol, dir);
705                         }
706                         ret = pol;
707                         break;
708                 }
709         }
710         write_unlock_bh(&xfrm_policy_lock);
711
712         if (ret && delete) {
713                 atomic_inc(&flow_cache_genid);
714                 xfrm_policy_kill(ret);
715         }
716         return ret;
717 }
718 EXPORT_SYMBOL(xfrm_policy_byid);
719
720 #ifdef CONFIG_SECURITY_NETWORK_XFRM
721 static inline int
722 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
723 {
724         int dir, err = 0;
725
726         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
727                 struct xfrm_policy *pol;
728                 struct hlist_node *entry;
729                 int i;
730
731                 hlist_for_each_entry(pol, entry,
732                                      &net->xfrm.policy_inexact[dir], bydst) {
733                         if (pol->type != type)
734                                 continue;
735                         err = security_xfrm_policy_delete(pol->security);
736                         if (err) {
737                                 xfrm_audit_policy_delete(pol, 0,
738                                                          audit_info->loginuid,
739                                                          audit_info->sessionid,
740                                                          audit_info->secid);
741                                 return err;
742                         }
743                 }
744                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
745                         hlist_for_each_entry(pol, entry,
746                                              net->xfrm.policy_bydst[dir].table + i,
747                                              bydst) {
748                                 if (pol->type != type)
749                                         continue;
750                                 err = security_xfrm_policy_delete(
751                                                                 pol->security);
752                                 if (err) {
753                                         xfrm_audit_policy_delete(pol, 0,
754                                                         audit_info->loginuid,
755                                                         audit_info->sessionid,
756                                                         audit_info->secid);
757                                         return err;
758                                 }
759                         }
760                 }
761         }
762         return err;
763 }
764 #else
765 static inline int
766 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
767 {
768         return 0;
769 }
770 #endif
771
772 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
773 {
774         int dir, err = 0, cnt = 0;
775         struct xfrm_policy *dp;
776
777         write_lock_bh(&xfrm_policy_lock);
778
779         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
780         if (err)
781                 goto out;
782
783         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
784                 struct xfrm_policy *pol;
785                 struct hlist_node *entry;
786                 int i;
787
788         again1:
789                 hlist_for_each_entry(pol, entry,
790                                      &net->xfrm.policy_inexact[dir], bydst) {
791                         if (pol->type != type)
792                                 continue;
793                         dp = __xfrm_policy_unlink(pol, dir);
794                         write_unlock_bh(&xfrm_policy_lock);
795                         if (dp)
796                                 cnt++;
797
798                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
799                                                  audit_info->sessionid,
800                                                  audit_info->secid);
801
802                         xfrm_policy_kill(pol);
803
804                         write_lock_bh(&xfrm_policy_lock);
805                         goto again1;
806                 }
807
808                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
809         again2:
810                         hlist_for_each_entry(pol, entry,
811                                              net->xfrm.policy_bydst[dir].table + i,
812                                              bydst) {
813                                 if (pol->type != type)
814                                         continue;
815                                 dp = __xfrm_policy_unlink(pol, dir);
816                                 write_unlock_bh(&xfrm_policy_lock);
817                                 if (dp)
818                                         cnt++;
819
820                                 xfrm_audit_policy_delete(pol, 1,
821                                                          audit_info->loginuid,
822                                                          audit_info->sessionid,
823                                                          audit_info->secid);
824                                 xfrm_policy_kill(pol);
825
826                                 write_lock_bh(&xfrm_policy_lock);
827                                 goto again2;
828                         }
829                 }
830
831         }
832         if (!cnt)
833                 err = -ESRCH;
834         atomic_inc(&flow_cache_genid);
835 out:
836         write_unlock_bh(&xfrm_policy_lock);
837         return err;
838 }
839 EXPORT_SYMBOL(xfrm_policy_flush);
840
841 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
842                      int (*func)(struct xfrm_policy *, int, int, void*),
843                      void *data)
844 {
845         struct xfrm_policy *pol;
846         struct xfrm_policy_walk_entry *x;
847         int error = 0;
848
849         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
850             walk->type != XFRM_POLICY_TYPE_ANY)
851                 return -EINVAL;
852
853         if (list_empty(&walk->walk.all) && walk->seq != 0)
854                 return 0;
855
856         write_lock_bh(&xfrm_policy_lock);
857         if (list_empty(&walk->walk.all))
858                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
859         else
860                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
861         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
862                 if (x->dead)
863                         continue;
864                 pol = container_of(x, struct xfrm_policy, walk);
865                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
866                     walk->type != pol->type)
867                         continue;
868                 error = func(pol, xfrm_policy_id2dir(pol->index),
869                              walk->seq, data);
870                 if (error) {
871                         list_move_tail(&walk->walk.all, &x->all);
872                         goto out;
873                 }
874                 walk->seq++;
875         }
876         if (walk->seq == 0) {
877                 error = -ENOENT;
878                 goto out;
879         }
880         list_del_init(&walk->walk.all);
881 out:
882         write_unlock_bh(&xfrm_policy_lock);
883         return error;
884 }
885 EXPORT_SYMBOL(xfrm_policy_walk);
886
887 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
888 {
889         INIT_LIST_HEAD(&walk->walk.all);
890         walk->walk.dead = 1;
891         walk->type = type;
892         walk->seq = 0;
893 }
894 EXPORT_SYMBOL(xfrm_policy_walk_init);
895
896 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
897 {
898         if (list_empty(&walk->walk.all))
899                 return;
900
901         write_lock_bh(&xfrm_policy_lock);
902         list_del(&walk->walk.all);
903         write_unlock_bh(&xfrm_policy_lock);
904 }
905 EXPORT_SYMBOL(xfrm_policy_walk_done);
906
907 /*
908  * Find policy to apply to this flow.
909  *
910  * Returns 0 if policy found, else an -errno.
911  */
912 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
913                              u8 type, u16 family, int dir)
914 {
915         struct xfrm_selector *sel = &pol->selector;
916         int match, ret = -ESRCH;
917
918         if (pol->family != family ||
919             pol->type != type)
920                 return ret;
921
922         match = xfrm_selector_match(sel, fl, family);
923         if (match)
924                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
925                                                   dir);
926
927         return ret;
928 }
929
930 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
931                                                      struct flowi *fl,
932                                                      u16 family, u8 dir)
933 {
934         int err;
935         struct xfrm_policy *pol, *ret;
936         xfrm_address_t *daddr, *saddr;
937         struct hlist_node *entry;
938         struct hlist_head *chain;
939         u32 priority = ~0U;
940
941         daddr = xfrm_flowi_daddr(fl, family);
942         saddr = xfrm_flowi_saddr(fl, family);
943         if (unlikely(!daddr || !saddr))
944                 return NULL;
945
946         read_lock_bh(&xfrm_policy_lock);
947         chain = policy_hash_direct(net, daddr, saddr, family, dir);
948         ret = NULL;
949         hlist_for_each_entry(pol, entry, chain, bydst) {
950                 err = xfrm_policy_match(pol, fl, type, family, dir);
951                 if (err) {
952                         if (err == -ESRCH)
953                                 continue;
954                         else {
955                                 ret = ERR_PTR(err);
956                                 goto fail;
957                         }
958                 } else {
959                         ret = pol;
960                         priority = ret->priority;
961                         break;
962                 }
963         }
964         chain = &net->xfrm.policy_inexact[dir];
965         hlist_for_each_entry(pol, entry, chain, bydst) {
966                 err = xfrm_policy_match(pol, fl, type, family, dir);
967                 if (err) {
968                         if (err == -ESRCH)
969                                 continue;
970                         else {
971                                 ret = ERR_PTR(err);
972                                 goto fail;
973                         }
974                 } else if (pol->priority < priority) {
975                         ret = pol;
976                         break;
977                 }
978         }
979         if (ret)
980                 xfrm_pol_hold(ret);
981 fail:
982         read_unlock_bh(&xfrm_policy_lock);
983
984         return ret;
985 }
986
987 static int xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
988                               u8 dir, void **objp, atomic_t **obj_refp)
989 {
990         struct xfrm_policy *pol;
991         int err = 0;
992
993 #ifdef CONFIG_XFRM_SUB_POLICY
994         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
995         if (IS_ERR(pol)) {
996                 err = PTR_ERR(pol);
997                 pol = NULL;
998         }
999         if (pol || err)
1000                 goto end;
1001 #endif
1002         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1003         if (IS_ERR(pol)) {
1004                 err = PTR_ERR(pol);
1005                 pol = NULL;
1006         }
1007 #ifdef CONFIG_XFRM_SUB_POLICY
1008 end:
1009 #endif
1010         if ((*objp = (void *) pol) != NULL)
1011                 *obj_refp = &pol->refcnt;
1012         return err;
1013 }
1014
1015 static inline int policy_to_flow_dir(int dir)
1016 {
1017         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1018             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1019             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1020                 return dir;
1021         switch (dir) {
1022         default:
1023         case XFRM_POLICY_IN:
1024                 return FLOW_DIR_IN;
1025         case XFRM_POLICY_OUT:
1026                 return FLOW_DIR_OUT;
1027         case XFRM_POLICY_FWD:
1028                 return FLOW_DIR_FWD;
1029         }
1030 }
1031
1032 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1033 {
1034         struct xfrm_policy *pol;
1035
1036         read_lock_bh(&xfrm_policy_lock);
1037         if ((pol = sk->sk_policy[dir]) != NULL) {
1038                 int match = xfrm_selector_match(&pol->selector, fl,
1039                                                 sk->sk_family);
1040                 int err = 0;
1041
1042                 if (match) {
1043                         err = security_xfrm_policy_lookup(pol->security,
1044                                                       fl->secid,
1045                                                       policy_to_flow_dir(dir));
1046                         if (!err)
1047                                 xfrm_pol_hold(pol);
1048                         else if (err == -ESRCH)
1049                                 pol = NULL;
1050                         else
1051                                 pol = ERR_PTR(err);
1052                 } else
1053                         pol = NULL;
1054         }
1055         read_unlock_bh(&xfrm_policy_lock);
1056         return pol;
1057 }
1058
1059 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1060 {
1061         struct net *net = xp_net(pol);
1062         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1063                                                      pol->family, dir);
1064
1065         list_add(&pol->walk.all, &net->xfrm.policy_all);
1066         hlist_add_head(&pol->bydst, chain);
1067         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1068         net->xfrm.policy_count[dir]++;
1069         xfrm_pol_hold(pol);
1070
1071         if (xfrm_bydst_should_resize(net, dir, NULL))
1072                 schedule_work(&net->xfrm.policy_hash_work);
1073 }
1074
1075 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1076                                                 int dir)
1077 {
1078         struct net *net = xp_net(pol);
1079
1080         if (hlist_unhashed(&pol->bydst))
1081                 return NULL;
1082
1083         hlist_del(&pol->bydst);
1084         hlist_del(&pol->byidx);
1085         list_del(&pol->walk.all);
1086         net->xfrm.policy_count[dir]--;
1087
1088         return pol;
1089 }
1090
1091 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1092 {
1093         write_lock_bh(&xfrm_policy_lock);
1094         pol = __xfrm_policy_unlink(pol, dir);
1095         write_unlock_bh(&xfrm_policy_lock);
1096         if (pol) {
1097                 if (dir < XFRM_POLICY_MAX)
1098                         atomic_inc(&flow_cache_genid);
1099                 xfrm_policy_kill(pol);
1100                 return 0;
1101         }
1102         return -ENOENT;
1103 }
1104 EXPORT_SYMBOL(xfrm_policy_delete);
1105
1106 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1107 {
1108         struct net *net = xp_net(pol);
1109         struct xfrm_policy *old_pol;
1110
1111 #ifdef CONFIG_XFRM_SUB_POLICY
1112         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1113                 return -EINVAL;
1114 #endif
1115
1116         write_lock_bh(&xfrm_policy_lock);
1117         old_pol = sk->sk_policy[dir];
1118         sk->sk_policy[dir] = pol;
1119         if (pol) {
1120                 pol->curlft.add_time = get_seconds();
1121                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1122                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1123         }
1124         if (old_pol)
1125                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1126         write_unlock_bh(&xfrm_policy_lock);
1127
1128         if (old_pol) {
1129                 xfrm_policy_kill(old_pol);
1130         }
1131         return 0;
1132 }
1133
1134 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1135 {
1136         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1137
1138         if (newp) {
1139                 newp->selector = old->selector;
1140                 if (security_xfrm_policy_clone(old->security,
1141                                                &newp->security)) {
1142                         kfree(newp);
1143                         return NULL;  /* ENOMEM */
1144                 }
1145                 newp->lft = old->lft;
1146                 newp->curlft = old->curlft;
1147                 newp->action = old->action;
1148                 newp->flags = old->flags;
1149                 newp->xfrm_nr = old->xfrm_nr;
1150                 newp->index = old->index;
1151                 newp->type = old->type;
1152                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1153                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1154                 write_lock_bh(&xfrm_policy_lock);
1155                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1156                 write_unlock_bh(&xfrm_policy_lock);
1157                 xfrm_pol_put(newp);
1158         }
1159         return newp;
1160 }
1161
1162 int __xfrm_sk_clone_policy(struct sock *sk)
1163 {
1164         struct xfrm_policy *p0 = sk->sk_policy[0],
1165                            *p1 = sk->sk_policy[1];
1166
1167         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1168         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1169                 return -ENOMEM;
1170         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1171                 return -ENOMEM;
1172         return 0;
1173 }
1174
1175 static int
1176 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1177                unsigned short family)
1178 {
1179         int err;
1180         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1181
1182         if (unlikely(afinfo == NULL))
1183                 return -EINVAL;
1184         err = afinfo->get_saddr(net, local, remote);
1185         xfrm_policy_put_afinfo(afinfo);
1186         return err;
1187 }
1188
1189 /* Resolve list of templates for the flow, given policy. */
1190
1191 static int
1192 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1193                       struct xfrm_state **xfrm,
1194                       unsigned short family)
1195 {
1196         struct net *net = xp_net(policy);
1197         int nx;
1198         int i, error;
1199         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1200         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1201         xfrm_address_t tmp;
1202
1203         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1204                 struct xfrm_state *x;
1205                 xfrm_address_t *remote = daddr;
1206                 xfrm_address_t *local  = saddr;
1207                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1208
1209                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1210                     tmpl->mode == XFRM_MODE_BEET) {
1211                         remote = &tmpl->id.daddr;
1212                         local = &tmpl->saddr;
1213                         family = tmpl->encap_family;
1214                         if (xfrm_addr_any(local, family)) {
1215                                 error = xfrm_get_saddr(net, &tmp, remote, family);
1216                                 if (error)
1217                                         goto fail;
1218                                 local = &tmp;
1219                         }
1220                 }
1221
1222                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1223
1224                 if (x && x->km.state == XFRM_STATE_VALID) {
1225                         xfrm[nx++] = x;
1226                         daddr = remote;
1227                         saddr = local;
1228                         continue;
1229                 }
1230                 if (x) {
1231                         error = (x->km.state == XFRM_STATE_ERROR ?
1232                                  -EINVAL : -EAGAIN);
1233                         xfrm_state_put(x);
1234                 }
1235                 else if (error == -ESRCH)
1236                         error = -EAGAIN;
1237
1238                 if (!tmpl->optional)
1239                         goto fail;
1240         }
1241         return nx;
1242
1243 fail:
1244         for (nx--; nx>=0; nx--)
1245                 xfrm_state_put(xfrm[nx]);
1246         return error;
1247 }
1248
1249 static int
1250 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1251                   struct xfrm_state **xfrm,
1252                   unsigned short family)
1253 {
1254         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1255         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1256         int cnx = 0;
1257         int error;
1258         int ret;
1259         int i;
1260
1261         for (i = 0; i < npols; i++) {
1262                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1263                         error = -ENOBUFS;
1264                         goto fail;
1265                 }
1266
1267                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1268                 if (ret < 0) {
1269                         error = ret;
1270                         goto fail;
1271                 } else
1272                         cnx += ret;
1273         }
1274
1275         /* found states are sorted for outbound processing */
1276         if (npols > 1)
1277                 xfrm_state_sort(xfrm, tpp, cnx, family);
1278
1279         return cnx;
1280
1281  fail:
1282         for (cnx--; cnx>=0; cnx--)
1283                 xfrm_state_put(tpp[cnx]);
1284         return error;
1285
1286 }
1287
1288 /* Check that the bundle accepts the flow and its components are
1289  * still valid.
1290  */
1291
1292 static struct dst_entry *
1293 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1294 {
1295         struct dst_entry *x;
1296         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1297         if (unlikely(afinfo == NULL))
1298                 return ERR_PTR(-EINVAL);
1299         x = afinfo->find_bundle(fl, policy);
1300         xfrm_policy_put_afinfo(afinfo);
1301         return x;
1302 }
1303
1304 static inline int xfrm_get_tos(struct flowi *fl, int family)
1305 {
1306         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1307         int tos;
1308
1309         if (!afinfo)
1310                 return -EINVAL;
1311
1312         tos = afinfo->get_tos(fl);
1313
1314         xfrm_policy_put_afinfo(afinfo);
1315
1316         return tos;
1317 }
1318
1319 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1320 {
1321         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1322         struct dst_ops *dst_ops;
1323         struct xfrm_dst *xdst;
1324
1325         if (!afinfo)
1326                 return ERR_PTR(-EINVAL);
1327
1328         switch (family) {
1329         case AF_INET:
1330                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1331                 break;
1332 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1333         case AF_INET6:
1334                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1335                 break;
1336 #endif
1337         default:
1338                 BUG();
1339         }
1340         xdst = dst_alloc(dst_ops) ?: ERR_PTR(-ENOBUFS);
1341
1342         xfrm_policy_put_afinfo(afinfo);
1343
1344         return xdst;
1345 }
1346
1347 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1348                                  int nfheader_len)
1349 {
1350         struct xfrm_policy_afinfo *afinfo =
1351                 xfrm_policy_get_afinfo(dst->ops->family);
1352         int err;
1353
1354         if (!afinfo)
1355                 return -EINVAL;
1356
1357         err = afinfo->init_path(path, dst, nfheader_len);
1358
1359         xfrm_policy_put_afinfo(afinfo);
1360
1361         return err;
1362 }
1363
1364 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1365 {
1366         struct xfrm_policy_afinfo *afinfo =
1367                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1368         int err;
1369
1370         if (!afinfo)
1371                 return -EINVAL;
1372
1373         err = afinfo->fill_dst(xdst, dev);
1374
1375         xfrm_policy_put_afinfo(afinfo);
1376
1377         return err;
1378 }
1379
1380 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1381  * all the metrics... Shortly, bundle a bundle.
1382  */
1383
1384 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1385                                             struct xfrm_state **xfrm, int nx,
1386                                             struct flowi *fl,
1387                                             struct dst_entry *dst)
1388 {
1389         struct net *net = xp_net(policy);
1390         unsigned long now = jiffies;
1391         struct net_device *dev;
1392         struct dst_entry *dst_prev = NULL;
1393         struct dst_entry *dst0 = NULL;
1394         int i = 0;
1395         int err;
1396         int header_len = 0;
1397         int nfheader_len = 0;
1398         int trailer_len = 0;
1399         int tos;
1400         int family = policy->selector.family;
1401         xfrm_address_t saddr, daddr;
1402
1403         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1404
1405         tos = xfrm_get_tos(fl, family);
1406         err = tos;
1407         if (tos < 0)
1408                 goto put_states;
1409
1410         dst_hold(dst);
1411
1412         for (; i < nx; i++) {
1413                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1414                 struct dst_entry *dst1 = &xdst->u.dst;
1415
1416                 err = PTR_ERR(xdst);
1417                 if (IS_ERR(xdst)) {
1418                         dst_release(dst);
1419                         goto put_states;
1420                 }
1421
1422                 if (!dst_prev)
1423                         dst0 = dst1;
1424                 else {
1425                         dst_prev->child = dst_clone(dst1);
1426                         dst1->flags |= DST_NOHASH;
1427                 }
1428
1429                 xdst->route = dst;
1430                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1431
1432                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1433                         family = xfrm[i]->props.family;
1434                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1435                                               family);
1436                         err = PTR_ERR(dst);
1437                         if (IS_ERR(dst))
1438                                 goto put_states;
1439                 } else
1440                         dst_hold(dst);
1441
1442                 dst1->xfrm = xfrm[i];
1443                 xdst->genid = xfrm[i]->genid;
1444
1445                 dst1->obsolete = -1;
1446                 dst1->flags |= DST_HOST;
1447                 dst1->lastuse = now;
1448
1449                 dst1->input = dst_discard;
1450                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1451
1452                 dst1->next = dst_prev;
1453                 dst_prev = dst1;
1454
1455                 header_len += xfrm[i]->props.header_len;
1456                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1457                         nfheader_len += xfrm[i]->props.header_len;
1458                 trailer_len += xfrm[i]->props.trailer_len;
1459         }
1460
1461         dst_prev->child = dst;
1462         dst0->path = dst;
1463
1464         err = -ENODEV;
1465         dev = dst->dev;
1466         if (!dev)
1467                 goto free_dst;
1468
1469         /* Copy neighbour for reachability confirmation */
1470         dst0->neighbour = neigh_clone(dst->neighbour);
1471
1472         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1473         xfrm_init_pmtu(dst_prev);
1474
1475         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1476                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1477
1478                 err = xfrm_fill_dst(xdst, dev);
1479                 if (err)
1480                         goto free_dst;
1481
1482                 dst_prev->header_len = header_len;
1483                 dst_prev->trailer_len = trailer_len;
1484                 header_len -= xdst->u.dst.xfrm->props.header_len;
1485                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1486         }
1487
1488 out:
1489         return dst0;
1490
1491 put_states:
1492         for (; i < nx; i++)
1493                 xfrm_state_put(xfrm[i]);
1494 free_dst:
1495         if (dst0)
1496                 dst_free(dst0);
1497         dst0 = ERR_PTR(err);
1498         goto out;
1499 }
1500
1501 static int inline
1502 xfrm_dst_alloc_copy(void **target, void *src, int size)
1503 {
1504         if (!*target) {
1505                 *target = kmalloc(size, GFP_ATOMIC);
1506                 if (!*target)
1507                         return -ENOMEM;
1508         }
1509         memcpy(*target, src, size);
1510         return 0;
1511 }
1512
1513 static int inline
1514 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1515 {
1516 #ifdef CONFIG_XFRM_SUB_POLICY
1517         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1518         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1519                                    sel, sizeof(*sel));
1520 #else
1521         return 0;
1522 #endif
1523 }
1524
1525 static int inline
1526 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1527 {
1528 #ifdef CONFIG_XFRM_SUB_POLICY
1529         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1530         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1531 #else
1532         return 0;
1533 #endif
1534 }
1535
1536 static int stale_bundle(struct dst_entry *dst);
1537
1538 /* Main function: finds/creates a bundle for given flow.
1539  *
1540  * At the moment we eat a raw IP route. Mostly to speed up lookups
1541  * on interfaces with disabled IPsec.
1542  */
1543 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1544                   struct sock *sk, int flags)
1545 {
1546         struct xfrm_policy *policy;
1547         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1548         int npols;
1549         int pol_dead;
1550         int xfrm_nr;
1551         int pi;
1552         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1553         struct dst_entry *dst, *dst_orig = *dst_p;
1554         int nx = 0;
1555         int err;
1556         u32 genid;
1557         u16 family;
1558         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1559
1560 restart:
1561         genid = atomic_read(&flow_cache_genid);
1562         policy = NULL;
1563         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1564                 pols[pi] = NULL;
1565         npols = 0;
1566         pol_dead = 0;
1567         xfrm_nr = 0;
1568
1569         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1570                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1571                 err = PTR_ERR(policy);
1572                 if (IS_ERR(policy)) {
1573                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1574                         goto dropdst;
1575                 }
1576         }
1577
1578         if (!policy) {
1579                 /* To accelerate a bit...  */
1580                 if ((dst_orig->flags & DST_NOXFRM) ||
1581                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1582                         goto nopol;
1583
1584                 policy = flow_cache_lookup(net, fl, dst_orig->ops->family,
1585                                            dir, xfrm_policy_lookup);
1586                 err = PTR_ERR(policy);
1587                 if (IS_ERR(policy)) {
1588                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1589                         goto dropdst;
1590                 }
1591         }
1592
1593         if (!policy)
1594                 goto nopol;
1595
1596         family = dst_orig->ops->family;
1597         pols[0] = policy;
1598         npols ++;
1599         xfrm_nr += pols[0]->xfrm_nr;
1600
1601         err = -ENOENT;
1602         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1603                 goto error;
1604
1605         policy->curlft.use_time = get_seconds();
1606
1607         switch (policy->action) {
1608         default:
1609         case XFRM_POLICY_BLOCK:
1610                 /* Prohibit the flow */
1611                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1612                 err = -EPERM;
1613                 goto error;
1614
1615         case XFRM_POLICY_ALLOW:
1616 #ifndef CONFIG_XFRM_SUB_POLICY
1617                 if (policy->xfrm_nr == 0) {
1618                         /* Flow passes not transformed. */
1619                         xfrm_pol_put(policy);
1620                         return 0;
1621                 }
1622 #endif
1623
1624                 /* Try to find matching bundle.
1625                  *
1626                  * LATER: help from flow cache. It is optional, this
1627                  * is required only for output policy.
1628                  */
1629                 dst = xfrm_find_bundle(fl, policy, family);
1630                 if (IS_ERR(dst)) {
1631                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1632                         err = PTR_ERR(dst);
1633                         goto error;
1634                 }
1635
1636                 if (dst)
1637                         break;
1638
1639 #ifdef CONFIG_XFRM_SUB_POLICY
1640                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1641                         pols[1] = xfrm_policy_lookup_bytype(net,
1642                                                             XFRM_POLICY_TYPE_MAIN,
1643                                                             fl, family,
1644                                                             XFRM_POLICY_OUT);
1645                         if (pols[1]) {
1646                                 if (IS_ERR(pols[1])) {
1647                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1648                                         err = PTR_ERR(pols[1]);
1649                                         goto error;
1650                                 }
1651                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1652                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1653                                         err = -EPERM;
1654                                         goto error;
1655                                 }
1656                                 npols ++;
1657                                 xfrm_nr += pols[1]->xfrm_nr;
1658                         }
1659                 }
1660
1661                 /*
1662                  * Because neither flowi nor bundle information knows about
1663                  * transformation template size. On more than one policy usage
1664                  * we can realize whether all of them is bypass or not after
1665                  * they are searched. See above not-transformed bypass
1666                  * is surrounded by non-sub policy configuration, too.
1667                  */
1668                 if (xfrm_nr == 0) {
1669                         /* Flow passes not transformed. */
1670                         xfrm_pols_put(pols, npols);
1671                         return 0;
1672                 }
1673
1674 #endif
1675                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1676
1677                 if (unlikely(nx<0)) {
1678                         err = nx;
1679                         if (err == -EAGAIN && net->xfrm.sysctl_larval_drop) {
1680                                 /* EREMOTE tells the caller to generate
1681                                  * a one-shot blackhole route.
1682                                  */
1683                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1684                                 xfrm_pol_put(policy);
1685                                 return -EREMOTE;
1686                         }
1687                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1688                                 DECLARE_WAITQUEUE(wait, current);
1689
1690                                 add_wait_queue(&net->xfrm.km_waitq, &wait);
1691                                 set_current_state(TASK_INTERRUPTIBLE);
1692                                 schedule();
1693                                 set_current_state(TASK_RUNNING);
1694                                 remove_wait_queue(&net->xfrm.km_waitq, &wait);
1695
1696                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1697
1698                                 if (nx == -EAGAIN && signal_pending(current)) {
1699                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1700                                         err = -ERESTART;
1701                                         goto error;
1702                                 }
1703                                 if (nx == -EAGAIN ||
1704                                     genid != atomic_read(&flow_cache_genid)) {
1705                                         xfrm_pols_put(pols, npols);
1706                                         goto restart;
1707                                 }
1708                                 err = nx;
1709                         }
1710                         if (err < 0) {
1711                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1712                                 goto error;
1713                         }
1714                 }
1715                 if (nx == 0) {
1716                         /* Flow passes not transformed. */
1717                         xfrm_pols_put(pols, npols);
1718                         return 0;
1719                 }
1720
1721                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1722                 err = PTR_ERR(dst);
1723                 if (IS_ERR(dst)) {
1724                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1725                         goto error;
1726                 }
1727
1728                 for (pi = 0; pi < npols; pi++) {
1729                         read_lock_bh(&pols[pi]->lock);
1730                         pol_dead |= pols[pi]->walk.dead;
1731                         read_unlock_bh(&pols[pi]->lock);
1732                 }
1733
1734                 write_lock_bh(&policy->lock);
1735                 if (unlikely(pol_dead || stale_bundle(dst))) {
1736                         /* Wow! While we worked on resolving, this
1737                          * policy has gone. Retry. It is not paranoia,
1738                          * we just cannot enlist new bundle to dead object.
1739                          * We can't enlist stable bundles either.
1740                          */
1741                         write_unlock_bh(&policy->lock);
1742                         dst_free(dst);
1743
1744                         if (pol_dead)
1745                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLDEAD);
1746                         else
1747                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1748                         err = -EHOSTUNREACH;
1749                         goto error;
1750                 }
1751
1752                 if (npols > 1)
1753                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1754                 else
1755                         err = xfrm_dst_update_origin(dst, fl);
1756                 if (unlikely(err)) {
1757                         write_unlock_bh(&policy->lock);
1758                         dst_free(dst);
1759                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1760                         goto error;
1761                 }
1762
1763                 dst->next = policy->bundles;
1764                 policy->bundles = dst;
1765                 dst_hold(dst);
1766                 write_unlock_bh(&policy->lock);
1767         }
1768         *dst_p = dst;
1769         dst_release(dst_orig);
1770         xfrm_pols_put(pols, npols);
1771         return 0;
1772
1773 error:
1774         xfrm_pols_put(pols, npols);
1775 dropdst:
1776         dst_release(dst_orig);
1777         *dst_p = NULL;
1778         return err;
1779
1780 nopol:
1781         err = -ENOENT;
1782         if (flags & XFRM_LOOKUP_ICMP)
1783                 goto dropdst;
1784         return 0;
1785 }
1786 EXPORT_SYMBOL(__xfrm_lookup);
1787
1788 int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1789                 struct sock *sk, int flags)
1790 {
1791         int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1792
1793         if (err == -EREMOTE) {
1794                 dst_release(*dst_p);
1795                 *dst_p = NULL;
1796                 err = -EAGAIN;
1797         }
1798
1799         return err;
1800 }
1801 EXPORT_SYMBOL(xfrm_lookup);
1802
1803 static inline int
1804 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1805 {
1806         struct xfrm_state *x;
1807
1808         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1809                 return 0;
1810         x = skb->sp->xvec[idx];
1811         if (!x->type->reject)
1812                 return 0;
1813         return x->type->reject(x, skb, fl);
1814 }
1815
1816 /* When skb is transformed back to its "native" form, we have to
1817  * check policy restrictions. At the moment we make this in maximally
1818  * stupid way. Shame on me. :-) Of course, connected sockets must
1819  * have policy cached at them.
1820  */
1821
1822 static inline int
1823 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1824               unsigned short family)
1825 {
1826         if (xfrm_state_kern(x))
1827                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1828         return  x->id.proto == tmpl->id.proto &&
1829                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1830                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1831                 x->props.mode == tmpl->mode &&
1832                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1833                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1834                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1835                   xfrm_state_addr_cmp(tmpl, x, family));
1836 }
1837
1838 /*
1839  * 0 or more than 0 is returned when validation is succeeded (either bypass
1840  * because of optional transport mode, or next index of the mathced secpath
1841  * state with the template.
1842  * -1 is returned when no matching template is found.
1843  * Otherwise "-2 - errored_index" is returned.
1844  */
1845 static inline int
1846 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1847                unsigned short family)
1848 {
1849         int idx = start;
1850
1851         if (tmpl->optional) {
1852                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1853                         return start;
1854         } else
1855                 start = -1;
1856         for (; idx < sp->len; idx++) {
1857                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1858                         return ++idx;
1859                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1860                         if (start == -1)
1861                                 start = -2-idx;
1862                         break;
1863                 }
1864         }
1865         return start;
1866 }
1867
1868 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1869                           unsigned int family, int reverse)
1870 {
1871         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1872         int err;
1873
1874         if (unlikely(afinfo == NULL))
1875                 return -EAFNOSUPPORT;
1876
1877         afinfo->decode_session(skb, fl, reverse);
1878         err = security_xfrm_decode_session(skb, &fl->secid);
1879         xfrm_policy_put_afinfo(afinfo);
1880         return err;
1881 }
1882 EXPORT_SYMBOL(__xfrm_decode_session);
1883
1884 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1885 {
1886         for (; k < sp->len; k++) {
1887                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1888                         *idxp = k;
1889                         return 1;
1890                 }
1891         }
1892
1893         return 0;
1894 }
1895
1896 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1897                         unsigned short family)
1898 {
1899         struct net *net = dev_net(skb->dev);
1900         struct xfrm_policy *pol;
1901         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1902         int npols = 0;
1903         int xfrm_nr;
1904         int pi;
1905         int reverse;
1906         struct flowi fl;
1907         u8 fl_dir;
1908         int xerr_idx = -1;
1909
1910         reverse = dir & ~XFRM_POLICY_MASK;
1911         dir &= XFRM_POLICY_MASK;
1912         fl_dir = policy_to_flow_dir(dir);
1913
1914         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1915                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
1916                 return 0;
1917         }
1918
1919         nf_nat_decode_session(skb, &fl, family);
1920
1921         /* First, check used SA against their selectors. */
1922         if (skb->sp) {
1923                 int i;
1924
1925                 for (i=skb->sp->len-1; i>=0; i--) {
1926                         struct xfrm_state *x = skb->sp->xvec[i];
1927                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1928                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
1929                                 return 0;
1930                         }
1931                 }
1932         }
1933
1934         pol = NULL;
1935         if (sk && sk->sk_policy[dir]) {
1936                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1937                 if (IS_ERR(pol)) {
1938                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1939                         return 0;
1940                 }
1941         }
1942
1943         if (!pol)
1944                 pol = flow_cache_lookup(net, &fl, family, fl_dir,
1945                                         xfrm_policy_lookup);
1946
1947         if (IS_ERR(pol)) {
1948                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1949                 return 0;
1950         }
1951
1952         if (!pol) {
1953                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1954                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1955                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
1956                         return 0;
1957                 }
1958                 return 1;
1959         }
1960
1961         pol->curlft.use_time = get_seconds();
1962
1963         pols[0] = pol;
1964         npols ++;
1965 #ifdef CONFIG_XFRM_SUB_POLICY
1966         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1967                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
1968                                                     &fl, family,
1969                                                     XFRM_POLICY_IN);
1970                 if (pols[1]) {
1971                         if (IS_ERR(pols[1])) {
1972                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1973                                 return 0;
1974                         }
1975                         pols[1]->curlft.use_time = get_seconds();
1976                         npols ++;
1977                 }
1978         }
1979 #endif
1980
1981         if (pol->action == XFRM_POLICY_ALLOW) {
1982                 struct sec_path *sp;
1983                 static struct sec_path dummy;
1984                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1985                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1986                 struct xfrm_tmpl **tpp = tp;
1987                 int ti = 0;
1988                 int i, k;
1989
1990                 if ((sp = skb->sp) == NULL)
1991                         sp = &dummy;
1992
1993                 for (pi = 0; pi < npols; pi++) {
1994                         if (pols[pi] != pol &&
1995                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1996                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
1997                                 goto reject;
1998                         }
1999                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2000                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2001                                 goto reject_error;
2002                         }
2003                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2004                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2005                 }
2006                 xfrm_nr = ti;
2007                 if (npols > 1) {
2008                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2009                         tpp = stp;
2010                 }
2011
2012                 /* For each tunnel xfrm, find the first matching tmpl.
2013                  * For each tmpl before that, find corresponding xfrm.
2014                  * Order is _important_. Later we will implement
2015                  * some barriers, but at the moment barriers
2016                  * are implied between each two transformations.
2017                  */
2018                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2019                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2020                         if (k < 0) {
2021                                 if (k < -1)
2022                                         /* "-2 - errored_index" returned */
2023                                         xerr_idx = -(2+k);
2024                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2025                                 goto reject;
2026                         }
2027                 }
2028
2029                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2030                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2031                         goto reject;
2032                 }
2033
2034                 xfrm_pols_put(pols, npols);
2035                 return 1;
2036         }
2037         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2038
2039 reject:
2040         xfrm_secpath_reject(xerr_idx, skb, &fl);
2041 reject_error:
2042         xfrm_pols_put(pols, npols);
2043         return 0;
2044 }
2045 EXPORT_SYMBOL(__xfrm_policy_check);
2046
2047 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2048 {
2049         struct net *net = dev_net(skb->dev);
2050         struct flowi fl;
2051         struct dst_entry *dst;
2052         int res;
2053
2054         if (xfrm_decode_session(skb, &fl, family) < 0) {
2055                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2056                 return 0;
2057         }
2058
2059         dst = skb_dst(skb);
2060
2061         res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
2062         skb_dst_set(skb, dst);
2063         return res;
2064 }
2065 EXPORT_SYMBOL(__xfrm_route_forward);
2066
2067 /* Optimize later using cookies and generation ids. */
2068
2069 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2070 {
2071         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2072          * to "-1" to force all XFRM destinations to get validated by
2073          * dst_ops->check on every use.  We do this because when a
2074          * normal route referenced by an XFRM dst is obsoleted we do
2075          * not go looking around for all parent referencing XFRM dsts
2076          * so that we can invalidate them.  It is just too much work.
2077          * Instead we make the checks here on every use.  For example:
2078          *
2079          *      XFRM dst A --> IPv4 dst X
2080          *
2081          * X is the "xdst->route" of A (X is also the "dst->path" of A
2082          * in this example).  If X is marked obsolete, "A" will not
2083          * notice.  That's what we are validating here via the
2084          * stale_bundle() check.
2085          *
2086          * When a policy's bundle is pruned, we dst_free() the XFRM
2087          * dst which causes it's ->obsolete field to be set to a
2088          * positive non-zero integer.  If an XFRM dst has been pruned
2089          * like this, we want to force a new route lookup.
2090          */
2091         if (dst->obsolete < 0 && !stale_bundle(dst))
2092                 return dst;
2093
2094         return NULL;
2095 }
2096
2097 static int stale_bundle(struct dst_entry *dst)
2098 {
2099         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2100 }
2101
2102 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2103 {
2104         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2105                 dst->dev = dev_net(dev)->loopback_dev;
2106                 dev_hold(dst->dev);
2107                 dev_put(dev);
2108         }
2109 }
2110 EXPORT_SYMBOL(xfrm_dst_ifdown);
2111
2112 static void xfrm_link_failure(struct sk_buff *skb)
2113 {
2114         /* Impossible. Such dst must be popped before reaches point of failure. */
2115         return;
2116 }
2117
2118 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2119 {
2120         if (dst) {
2121                 if (dst->obsolete) {
2122                         dst_release(dst);
2123                         dst = NULL;
2124                 }
2125         }
2126         return dst;
2127 }
2128
2129 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2130 {
2131         struct dst_entry *dst, **dstp;
2132
2133         write_lock(&pol->lock);
2134         dstp = &pol->bundles;
2135         while ((dst=*dstp) != NULL) {
2136                 if (func(dst)) {
2137                         *dstp = dst->next;
2138                         dst->next = *gc_list_p;
2139                         *gc_list_p = dst;
2140                 } else {
2141                         dstp = &dst->next;
2142                 }
2143         }
2144         write_unlock(&pol->lock);
2145 }
2146
2147 static void xfrm_prune_bundles(struct net *net, int (*func)(struct dst_entry *))
2148 {
2149         struct dst_entry *gc_list = NULL;
2150         int dir;
2151
2152         read_lock_bh(&xfrm_policy_lock);
2153         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2154                 struct xfrm_policy *pol;
2155                 struct hlist_node *entry;
2156                 struct hlist_head *table;
2157                 int i;
2158
2159                 hlist_for_each_entry(pol, entry,
2160                                      &net->xfrm.policy_inexact[dir], bydst)
2161                         prune_one_bundle(pol, func, &gc_list);
2162
2163                 table = net->xfrm.policy_bydst[dir].table;
2164                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2165                         hlist_for_each_entry(pol, entry, table + i, bydst)
2166                                 prune_one_bundle(pol, func, &gc_list);
2167                 }
2168         }
2169         read_unlock_bh(&xfrm_policy_lock);
2170
2171         while (gc_list) {
2172                 struct dst_entry *dst = gc_list;
2173                 gc_list = dst->next;
2174                 dst_free(dst);
2175         }
2176 }
2177
2178 static int unused_bundle(struct dst_entry *dst)
2179 {
2180         return !atomic_read(&dst->__refcnt);
2181 }
2182
2183 static void __xfrm_garbage_collect(struct net *net)
2184 {
2185         xfrm_prune_bundles(net, unused_bundle);
2186 }
2187
2188 static int xfrm_flush_bundles(struct net *net)
2189 {
2190         xfrm_prune_bundles(net, stale_bundle);
2191         return 0;
2192 }
2193
2194 static void xfrm_init_pmtu(struct dst_entry *dst)
2195 {
2196         do {
2197                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2198                 u32 pmtu, route_mtu_cached;
2199
2200                 pmtu = dst_mtu(dst->child);
2201                 xdst->child_mtu_cached = pmtu;
2202
2203                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2204
2205                 route_mtu_cached = dst_mtu(xdst->route);
2206                 xdst->route_mtu_cached = route_mtu_cached;
2207
2208                 if (pmtu > route_mtu_cached)
2209                         pmtu = route_mtu_cached;
2210
2211                 dst->metrics[RTAX_MTU-1] = pmtu;
2212         } while ((dst = dst->next));
2213 }
2214
2215 /* Check that the bundle accepts the flow and its components are
2216  * still valid.
2217  */
2218
2219 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2220                 struct flowi *fl, int family, int strict)
2221 {
2222         struct dst_entry *dst = &first->u.dst;
2223         struct xfrm_dst *last;
2224         u32 mtu;
2225
2226         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2227             (dst->dev && !netif_running(dst->dev)))
2228                 return 0;
2229 #ifdef CONFIG_XFRM_SUB_POLICY
2230         if (fl) {
2231                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2232                         return 0;
2233                 if (first->partner &&
2234                     !xfrm_selector_match(first->partner, fl, family))
2235                         return 0;
2236         }
2237 #endif
2238
2239         last = NULL;
2240
2241         do {
2242                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2243
2244                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2245                         return 0;
2246                 if (fl && pol &&
2247                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2248                         return 0;
2249                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2250                         return 0;
2251                 if (xdst->genid != dst->xfrm->genid)
2252                         return 0;
2253
2254                 if (strict && fl &&
2255                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2256                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2257                         return 0;
2258
2259                 mtu = dst_mtu(dst->child);
2260                 if (xdst->child_mtu_cached != mtu) {
2261                         last = xdst;
2262                         xdst->child_mtu_cached = mtu;
2263                 }
2264
2265                 if (!dst_check(xdst->route, xdst->route_cookie))
2266                         return 0;
2267                 mtu = dst_mtu(xdst->route);
2268                 if (xdst->route_mtu_cached != mtu) {
2269                         last = xdst;
2270                         xdst->route_mtu_cached = mtu;
2271                 }
2272
2273                 dst = dst->child;
2274         } while (dst->xfrm);
2275
2276         if (likely(!last))
2277                 return 1;
2278
2279         mtu = last->child_mtu_cached;
2280         for (;;) {
2281                 dst = &last->u.dst;
2282
2283                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2284                 if (mtu > last->route_mtu_cached)
2285                         mtu = last->route_mtu_cached;
2286                 dst->metrics[RTAX_MTU-1] = mtu;
2287
2288                 if (last == first)
2289                         break;
2290
2291                 last = (struct xfrm_dst *)last->u.dst.next;
2292                 last->child_mtu_cached = mtu;
2293         }
2294
2295         return 1;
2296 }
2297
2298 EXPORT_SYMBOL(xfrm_bundle_ok);
2299
2300 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2301 {
2302         struct net *net;
2303         int err = 0;
2304         if (unlikely(afinfo == NULL))
2305                 return -EINVAL;
2306         if (unlikely(afinfo->family >= NPROTO))
2307                 return -EAFNOSUPPORT;
2308         write_lock_bh(&xfrm_policy_afinfo_lock);
2309         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2310                 err = -ENOBUFS;
2311         else {
2312                 struct dst_ops *dst_ops = afinfo->dst_ops;
2313                 if (likely(dst_ops->kmem_cachep == NULL))
2314                         dst_ops->kmem_cachep = xfrm_dst_cache;
2315                 if (likely(dst_ops->check == NULL))
2316                         dst_ops->check = xfrm_dst_check;
2317                 if (likely(dst_ops->negative_advice == NULL))
2318                         dst_ops->negative_advice = xfrm_negative_advice;
2319                 if (likely(dst_ops->link_failure == NULL))
2320                         dst_ops->link_failure = xfrm_link_failure;
2321                 if (likely(afinfo->garbage_collect == NULL))
2322                         afinfo->garbage_collect = __xfrm_garbage_collect;
2323                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2324         }
2325         write_unlock_bh(&xfrm_policy_afinfo_lock);
2326
2327         rtnl_lock();
2328         for_each_net(net) {
2329                 struct dst_ops *xfrm_dst_ops;
2330
2331                 switch (afinfo->family) {
2332                 case AF_INET:
2333                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2334                         break;
2335 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2336                 case AF_INET6:
2337                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2338                         break;
2339 #endif
2340                 default:
2341                         BUG();
2342                 }
2343                 *xfrm_dst_ops = *afinfo->dst_ops;
2344         }
2345         rtnl_unlock();
2346
2347         return err;
2348 }
2349 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2350
2351 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2352 {
2353         int err = 0;
2354         if (unlikely(afinfo == NULL))
2355                 return -EINVAL;
2356         if (unlikely(afinfo->family >= NPROTO))
2357                 return -EAFNOSUPPORT;
2358         write_lock_bh(&xfrm_policy_afinfo_lock);
2359         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2360                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2361                         err = -EINVAL;
2362                 else {
2363                         struct dst_ops *dst_ops = afinfo->dst_ops;
2364                         xfrm_policy_afinfo[afinfo->family] = NULL;
2365                         dst_ops->kmem_cachep = NULL;
2366                         dst_ops->check = NULL;
2367                         dst_ops->negative_advice = NULL;
2368                         dst_ops->link_failure = NULL;
2369                         afinfo->garbage_collect = NULL;
2370                 }
2371         }
2372         write_unlock_bh(&xfrm_policy_afinfo_lock);
2373         return err;
2374 }
2375 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2376
2377 static void __net_init xfrm_dst_ops_init(struct net *net)
2378 {
2379         struct xfrm_policy_afinfo *afinfo;
2380
2381         read_lock_bh(&xfrm_policy_afinfo_lock);
2382         afinfo = xfrm_policy_afinfo[AF_INET];
2383         if (afinfo)
2384                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2385 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2386         afinfo = xfrm_policy_afinfo[AF_INET6];
2387         if (afinfo)
2388                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2389 #endif
2390         read_unlock_bh(&xfrm_policy_afinfo_lock);
2391 }
2392
2393 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2394 {
2395         struct xfrm_policy_afinfo *afinfo;
2396         if (unlikely(family >= NPROTO))
2397                 return NULL;
2398         read_lock(&xfrm_policy_afinfo_lock);
2399         afinfo = xfrm_policy_afinfo[family];
2400         if (unlikely(!afinfo))
2401                 read_unlock(&xfrm_policy_afinfo_lock);
2402         return afinfo;
2403 }
2404
2405 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2406 {
2407         read_unlock(&xfrm_policy_afinfo_lock);
2408 }
2409
2410 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2411 {
2412         struct net_device *dev = ptr;
2413
2414         switch (event) {
2415         case NETDEV_DOWN:
2416                 xfrm_flush_bundles(dev_net(dev));
2417         }
2418         return NOTIFY_DONE;
2419 }
2420
2421 static struct notifier_block xfrm_dev_notifier = {
2422         .notifier_call  = xfrm_dev_event,
2423 };
2424
2425 #ifdef CONFIG_XFRM_STATISTICS
2426 static int __net_init xfrm_statistics_init(struct net *net)
2427 {
2428         int rv;
2429
2430         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2431                           sizeof(struct linux_xfrm_mib)) < 0)
2432                 return -ENOMEM;
2433         rv = xfrm_proc_init(net);
2434         if (rv < 0)
2435                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2436         return rv;
2437 }
2438
2439 static void xfrm_statistics_fini(struct net *net)
2440 {
2441         xfrm_proc_fini(net);
2442         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2443 }
2444 #else
2445 static int __net_init xfrm_statistics_init(struct net *net)
2446 {
2447         return 0;
2448 }
2449
2450 static void xfrm_statistics_fini(struct net *net)
2451 {
2452 }
2453 #endif
2454
2455 static int __net_init xfrm_policy_init(struct net *net)
2456 {
2457         unsigned int hmask, sz;
2458         int dir;
2459
2460         if (net_eq(net, &init_net))
2461                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2462                                            sizeof(struct xfrm_dst),
2463                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2464                                            NULL);
2465
2466         hmask = 8 - 1;
2467         sz = (hmask+1) * sizeof(struct hlist_head);
2468
2469         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2470         if (!net->xfrm.policy_byidx)
2471                 goto out_byidx;
2472         net->xfrm.policy_idx_hmask = hmask;
2473
2474         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2475                 struct xfrm_policy_hash *htab;
2476
2477                 net->xfrm.policy_count[dir] = 0;
2478                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2479
2480                 htab = &net->xfrm.policy_bydst[dir];
2481                 htab->table = xfrm_hash_alloc(sz);
2482                 if (!htab->table)
2483                         goto out_bydst;
2484                 htab->hmask = hmask;
2485         }
2486
2487         INIT_LIST_HEAD(&net->xfrm.policy_all);
2488         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2489         if (net_eq(net, &init_net))
2490                 register_netdevice_notifier(&xfrm_dev_notifier);
2491         return 0;
2492
2493 out_bydst:
2494         for (dir--; dir >= 0; dir--) {
2495                 struct xfrm_policy_hash *htab;
2496
2497                 htab = &net->xfrm.policy_bydst[dir];
2498                 xfrm_hash_free(htab->table, sz);
2499         }
2500         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2501 out_byidx:
2502         return -ENOMEM;
2503 }
2504
2505 static void xfrm_policy_fini(struct net *net)
2506 {
2507         struct xfrm_audit audit_info;
2508         unsigned int sz;
2509         int dir;
2510
2511         flush_work(&net->xfrm.policy_hash_work);
2512 #ifdef CONFIG_XFRM_SUB_POLICY
2513         audit_info.loginuid = -1;
2514         audit_info.sessionid = -1;
2515         audit_info.secid = 0;
2516         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2517 #endif
2518         audit_info.loginuid = -1;
2519         audit_info.sessionid = -1;
2520         audit_info.secid = 0;
2521         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2522         flush_work(&xfrm_policy_gc_work);
2523
2524         WARN_ON(!list_empty(&net->xfrm.policy_all));
2525
2526         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2527                 struct xfrm_policy_hash *htab;
2528
2529                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2530
2531                 htab = &net->xfrm.policy_bydst[dir];
2532                 sz = (htab->hmask + 1);
2533                 WARN_ON(!hlist_empty(htab->table));
2534                 xfrm_hash_free(htab->table, sz);
2535         }
2536
2537         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2538         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2539         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2540 }
2541
2542 static int __net_init xfrm_net_init(struct net *net)
2543 {
2544         int rv;
2545
2546         rv = xfrm_statistics_init(net);
2547         if (rv < 0)
2548                 goto out_statistics;
2549         rv = xfrm_state_init(net);
2550         if (rv < 0)
2551                 goto out_state;
2552         rv = xfrm_policy_init(net);
2553         if (rv < 0)
2554                 goto out_policy;
2555         xfrm_dst_ops_init(net);
2556         rv = xfrm_sysctl_init(net);
2557         if (rv < 0)
2558                 goto out_sysctl;
2559         return 0;
2560
2561 out_sysctl:
2562         xfrm_policy_fini(net);
2563 out_policy:
2564         xfrm_state_fini(net);
2565 out_state:
2566         xfrm_statistics_fini(net);
2567 out_statistics:
2568         return rv;
2569 }
2570
2571 static void __net_exit xfrm_net_exit(struct net *net)
2572 {
2573         xfrm_sysctl_fini(net);
2574         xfrm_policy_fini(net);
2575         xfrm_state_fini(net);
2576         xfrm_statistics_fini(net);
2577 }
2578
2579 static struct pernet_operations __net_initdata xfrm_net_ops = {
2580         .init = xfrm_net_init,
2581         .exit = xfrm_net_exit,
2582 };
2583
2584 void __init xfrm_init(void)
2585 {
2586         register_pernet_subsys(&xfrm_net_ops);
2587         xfrm_input_init();
2588 }
2589
2590 #ifdef CONFIG_AUDITSYSCALL
2591 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2592                                          struct audit_buffer *audit_buf)
2593 {
2594         struct xfrm_sec_ctx *ctx = xp->security;
2595         struct xfrm_selector *sel = &xp->selector;
2596
2597         if (ctx)
2598                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2599                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2600
2601         switch(sel->family) {
2602         case AF_INET:
2603                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2604                 if (sel->prefixlen_s != 32)
2605                         audit_log_format(audit_buf, " src_prefixlen=%d",
2606                                          sel->prefixlen_s);
2607                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2608                 if (sel->prefixlen_d != 32)
2609                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2610                                          sel->prefixlen_d);
2611                 break;
2612         case AF_INET6:
2613                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2614                 if (sel->prefixlen_s != 128)
2615                         audit_log_format(audit_buf, " src_prefixlen=%d",
2616                                          sel->prefixlen_s);
2617                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2618                 if (sel->prefixlen_d != 128)
2619                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2620                                          sel->prefixlen_d);
2621                 break;
2622         }
2623 }
2624
2625 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2626                            uid_t auid, u32 sessionid, u32 secid)
2627 {
2628         struct audit_buffer *audit_buf;
2629
2630         audit_buf = xfrm_audit_start("SPD-add");
2631         if (audit_buf == NULL)
2632                 return;
2633         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2634         audit_log_format(audit_buf, " res=%u", result);
2635         xfrm_audit_common_policyinfo(xp, audit_buf);
2636         audit_log_end(audit_buf);
2637 }
2638 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2639
2640 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2641                               uid_t auid, u32 sessionid, u32 secid)
2642 {
2643         struct audit_buffer *audit_buf;
2644
2645         audit_buf = xfrm_audit_start("SPD-delete");
2646         if (audit_buf == NULL)
2647                 return;
2648         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2649         audit_log_format(audit_buf, " res=%u", result);
2650         xfrm_audit_common_policyinfo(xp, audit_buf);
2651         audit_log_end(audit_buf);
2652 }
2653 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2654 #endif
2655
2656 #ifdef CONFIG_XFRM_MIGRATE
2657 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2658                                        struct xfrm_selector *sel_tgt)
2659 {
2660         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2661                 if (sel_tgt->family == sel_cmp->family &&
2662                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2663                                   sel_cmp->family) == 0 &&
2664                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2665                                   sel_cmp->family) == 0 &&
2666                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2667                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2668                         return 1;
2669                 }
2670         } else {
2671                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2672                         return 1;
2673                 }
2674         }
2675         return 0;
2676 }
2677
2678 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2679                                                      u8 dir, u8 type)
2680 {
2681         struct xfrm_policy *pol, *ret = NULL;
2682         struct hlist_node *entry;
2683         struct hlist_head *chain;
2684         u32 priority = ~0U;
2685
2686         read_lock_bh(&xfrm_policy_lock);
2687         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2688         hlist_for_each_entry(pol, entry, chain, bydst) {
2689                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2690                     pol->type == type) {
2691                         ret = pol;
2692                         priority = ret->priority;
2693                         break;
2694                 }
2695         }
2696         chain = &init_net.xfrm.policy_inexact[dir];
2697         hlist_for_each_entry(pol, entry, chain, bydst) {
2698                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2699                     pol->type == type &&
2700                     pol->priority < priority) {
2701                         ret = pol;
2702                         break;
2703                 }
2704         }
2705
2706         if (ret)
2707                 xfrm_pol_hold(ret);
2708
2709         read_unlock_bh(&xfrm_policy_lock);
2710
2711         return ret;
2712 }
2713
2714 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2715 {
2716         int match = 0;
2717
2718         if (t->mode == m->mode && t->id.proto == m->proto &&
2719             (m->reqid == 0 || t->reqid == m->reqid)) {
2720                 switch (t->mode) {
2721                 case XFRM_MODE_TUNNEL:
2722                 case XFRM_MODE_BEET:
2723                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2724                                           m->old_family) == 0 &&
2725                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2726                                           m->old_family) == 0) {
2727                                 match = 1;
2728                         }
2729                         break;
2730                 case XFRM_MODE_TRANSPORT:
2731                         /* in case of transport mode, template does not store
2732                            any IP addresses, hence we just compare mode and
2733                            protocol */
2734                         match = 1;
2735                         break;
2736                 default:
2737                         break;
2738                 }
2739         }
2740         return match;
2741 }
2742
2743 /* update endpoint address(es) of template(s) */
2744 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2745                                struct xfrm_migrate *m, int num_migrate)
2746 {
2747         struct xfrm_migrate *mp;
2748         struct dst_entry *dst;
2749         int i, j, n = 0;
2750
2751         write_lock_bh(&pol->lock);
2752         if (unlikely(pol->walk.dead)) {
2753                 /* target policy has been deleted */
2754                 write_unlock_bh(&pol->lock);
2755                 return -ENOENT;
2756         }
2757
2758         for (i = 0; i < pol->xfrm_nr; i++) {
2759                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2760                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2761                                 continue;
2762                         n++;
2763                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2764                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2765                                 continue;
2766                         /* update endpoints */
2767                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2768                                sizeof(pol->xfrm_vec[i].id.daddr));
2769                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2770                                sizeof(pol->xfrm_vec[i].saddr));
2771                         pol->xfrm_vec[i].encap_family = mp->new_family;
2772                         /* flush bundles */
2773                         while ((dst = pol->bundles) != NULL) {
2774                                 pol->bundles = dst->next;
2775                                 dst_free(dst);
2776                         }
2777                 }
2778         }
2779
2780         write_unlock_bh(&pol->lock);
2781
2782         if (!n)
2783                 return -ENODATA;
2784
2785         return 0;
2786 }
2787
2788 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2789 {
2790         int i, j;
2791
2792         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2793                 return -EINVAL;
2794
2795         for (i = 0; i < num_migrate; i++) {
2796                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2797                                    m[i].old_family) == 0) &&
2798                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2799                                    m[i].old_family) == 0))
2800                         return -EINVAL;
2801                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2802                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2803                         return -EINVAL;
2804
2805                 /* check if there is any duplicated entry */
2806                 for (j = i + 1; j < num_migrate; j++) {
2807                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2808                                     sizeof(m[i].old_daddr)) &&
2809                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2810                                     sizeof(m[i].old_saddr)) &&
2811                             m[i].proto == m[j].proto &&
2812                             m[i].mode == m[j].mode &&
2813                             m[i].reqid == m[j].reqid &&
2814                             m[i].old_family == m[j].old_family)
2815                                 return -EINVAL;
2816                 }
2817         }
2818
2819         return 0;
2820 }
2821
2822 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2823                  struct xfrm_migrate *m, int num_migrate,
2824                  struct xfrm_kmaddress *k)
2825 {
2826         int i, err, nx_cur = 0, nx_new = 0;
2827         struct xfrm_policy *pol = NULL;
2828         struct xfrm_state *x, *xc;
2829         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2830         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2831         struct xfrm_migrate *mp;
2832
2833         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2834                 goto out;
2835
2836         /* Stage 1 - find policy */
2837         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2838                 err = -ENOENT;
2839                 goto out;
2840         }
2841
2842         /* Stage 2 - find and update state(s) */
2843         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2844                 if ((x = xfrm_migrate_state_find(mp))) {
2845                         x_cur[nx_cur] = x;
2846                         nx_cur++;
2847                         if ((xc = xfrm_state_migrate(x, mp))) {
2848                                 x_new[nx_new] = xc;
2849                                 nx_new++;
2850                         } else {
2851                                 err = -ENODATA;
2852                                 goto restore_state;
2853                         }
2854                 }
2855         }
2856
2857         /* Stage 3 - update policy */
2858         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2859                 goto restore_state;
2860
2861         /* Stage 4 - delete old state(s) */
2862         if (nx_cur) {
2863                 xfrm_states_put(x_cur, nx_cur);
2864                 xfrm_states_delete(x_cur, nx_cur);
2865         }
2866
2867         /* Stage 5 - announce */
2868         km_migrate(sel, dir, type, m, num_migrate, k);
2869
2870         xfrm_pol_put(pol);
2871
2872         return 0;
2873 out:
2874         return err;
2875
2876 restore_state:
2877         if (pol)
2878                 xfrm_pol_put(pol);
2879         if (nx_cur)
2880                 xfrm_states_put(x_cur, nx_cur);
2881         if (nx_new)
2882                 xfrm_states_delete(x_new, nx_new);
2883
2884         return err;
2885 }
2886 EXPORT_SYMBOL(xfrm_migrate);
2887 #endif