macvlan: Support creating macvtaps from macvlans
[linux-2.6-microblaze.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/flow.h>
30 #include <net/xfrm.h>
31 #include <net/ip.h>
32 #ifdef CONFIG_XFRM_STATISTICS
33 #include <net/snmp.h>
34 #endif
35
36 #include "xfrm_hash.h"
37
38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
40 #define XFRM_MAX_QUEUE_LEN      100
41
42 DEFINE_MUTEX(xfrm_cfg_mutex);
43 EXPORT_SYMBOL(xfrm_cfg_mutex);
44
45 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
46 static struct dst_entry *xfrm_policy_sk_bundles;
47 static DEFINE_RWLOCK(xfrm_policy_lock);
48
49 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
50 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
51                                                 __read_mostly;
52
53 static struct kmem_cache *xfrm_dst_cache __read_mostly;
54
55 static void xfrm_init_pmtu(struct dst_entry *dst);
56 static int stale_bundle(struct dst_entry *dst);
57 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
58 static void xfrm_policy_queue_process(unsigned long arg);
59
60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
61                                                 int dir);
62
63 static inline bool
64 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
65 {
66         const struct flowi4 *fl4 = &fl->u.ip4;
67
68         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
69                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
70                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
71                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
72                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
73                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
74 }
75
76 static inline bool
77 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
78 {
79         const struct flowi6 *fl6 = &fl->u.ip6;
80
81         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
82                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
83                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
84                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
85                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
86                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
87 }
88
89 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
90                          unsigned short family)
91 {
92         switch (family) {
93         case AF_INET:
94                 return __xfrm4_selector_match(sel, fl);
95         case AF_INET6:
96                 return __xfrm6_selector_match(sel, fl);
97         }
98         return false;
99 }
100
101 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
102 {
103         struct xfrm_policy_afinfo *afinfo;
104
105         if (unlikely(family >= NPROTO))
106                 return NULL;
107         rcu_read_lock();
108         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
109         if (unlikely(!afinfo))
110                 rcu_read_unlock();
111         return afinfo;
112 }
113
114 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
115 {
116         rcu_read_unlock();
117 }
118
119 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
120                                                   const xfrm_address_t *saddr,
121                                                   const xfrm_address_t *daddr,
122                                                   int family)
123 {
124         struct xfrm_policy_afinfo *afinfo;
125         struct dst_entry *dst;
126
127         afinfo = xfrm_policy_get_afinfo(family);
128         if (unlikely(afinfo == NULL))
129                 return ERR_PTR(-EAFNOSUPPORT);
130
131         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
132
133         xfrm_policy_put_afinfo(afinfo);
134
135         return dst;
136 }
137
138 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
139                                                 xfrm_address_t *prev_saddr,
140                                                 xfrm_address_t *prev_daddr,
141                                                 int family)
142 {
143         struct net *net = xs_net(x);
144         xfrm_address_t *saddr = &x->props.saddr;
145         xfrm_address_t *daddr = &x->id.daddr;
146         struct dst_entry *dst;
147
148         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
149                 saddr = x->coaddr;
150                 daddr = prev_daddr;
151         }
152         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
153                 saddr = prev_saddr;
154                 daddr = x->coaddr;
155         }
156
157         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
158
159         if (!IS_ERR(dst)) {
160                 if (prev_saddr != saddr)
161                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
162                 if (prev_daddr != daddr)
163                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
164         }
165
166         return dst;
167 }
168
169 static inline unsigned long make_jiffies(long secs)
170 {
171         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
172                 return MAX_SCHEDULE_TIMEOUT-1;
173         else
174                 return secs*HZ;
175 }
176
177 static void xfrm_policy_timer(unsigned long data)
178 {
179         struct xfrm_policy *xp = (struct xfrm_policy*)data;
180         unsigned long now = get_seconds();
181         long next = LONG_MAX;
182         int warn = 0;
183         int dir;
184
185         read_lock(&xp->lock);
186
187         if (unlikely(xp->walk.dead))
188                 goto out;
189
190         dir = xfrm_policy_id2dir(xp->index);
191
192         if (xp->lft.hard_add_expires_seconds) {
193                 long tmo = xp->lft.hard_add_expires_seconds +
194                         xp->curlft.add_time - now;
195                 if (tmo <= 0)
196                         goto expired;
197                 if (tmo < next)
198                         next = tmo;
199         }
200         if (xp->lft.hard_use_expires_seconds) {
201                 long tmo = xp->lft.hard_use_expires_seconds +
202                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
203                 if (tmo <= 0)
204                         goto expired;
205                 if (tmo < next)
206                         next = tmo;
207         }
208         if (xp->lft.soft_add_expires_seconds) {
209                 long tmo = xp->lft.soft_add_expires_seconds +
210                         xp->curlft.add_time - now;
211                 if (tmo <= 0) {
212                         warn = 1;
213                         tmo = XFRM_KM_TIMEOUT;
214                 }
215                 if (tmo < next)
216                         next = tmo;
217         }
218         if (xp->lft.soft_use_expires_seconds) {
219                 long tmo = xp->lft.soft_use_expires_seconds +
220                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
221                 if (tmo <= 0) {
222                         warn = 1;
223                         tmo = XFRM_KM_TIMEOUT;
224                 }
225                 if (tmo < next)
226                         next = tmo;
227         }
228
229         if (warn)
230                 km_policy_expired(xp, dir, 0, 0);
231         if (next != LONG_MAX &&
232             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
233                 xfrm_pol_hold(xp);
234
235 out:
236         read_unlock(&xp->lock);
237         xfrm_pol_put(xp);
238         return;
239
240 expired:
241         read_unlock(&xp->lock);
242         if (!xfrm_policy_delete(xp, dir))
243                 km_policy_expired(xp, dir, 1, 0);
244         xfrm_pol_put(xp);
245 }
246
247 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
248 {
249         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
250
251         if (unlikely(pol->walk.dead))
252                 flo = NULL;
253         else
254                 xfrm_pol_hold(pol);
255
256         return flo;
257 }
258
259 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
260 {
261         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
262
263         return !pol->walk.dead;
264 }
265
266 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
267 {
268         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
269 }
270
271 static const struct flow_cache_ops xfrm_policy_fc_ops = {
272         .get = xfrm_policy_flo_get,
273         .check = xfrm_policy_flo_check,
274         .delete = xfrm_policy_flo_delete,
275 };
276
277 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
278  * SPD calls.
279  */
280
281 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
282 {
283         struct xfrm_policy *policy;
284
285         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
286
287         if (policy) {
288                 write_pnet(&policy->xp_net, net);
289                 INIT_LIST_HEAD(&policy->walk.all);
290                 INIT_HLIST_NODE(&policy->bydst);
291                 INIT_HLIST_NODE(&policy->byidx);
292                 rwlock_init(&policy->lock);
293                 atomic_set(&policy->refcnt, 1);
294                 skb_queue_head_init(&policy->polq.hold_queue);
295                 setup_timer(&policy->timer, xfrm_policy_timer,
296                                 (unsigned long)policy);
297                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
298                             (unsigned long)policy);
299                 policy->flo.ops = &xfrm_policy_fc_ops;
300         }
301         return policy;
302 }
303 EXPORT_SYMBOL(xfrm_policy_alloc);
304
305 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
306
307 void xfrm_policy_destroy(struct xfrm_policy *policy)
308 {
309         BUG_ON(!policy->walk.dead);
310
311         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
312                 BUG();
313
314         security_xfrm_policy_free(policy->security);
315         kfree(policy);
316 }
317 EXPORT_SYMBOL(xfrm_policy_destroy);
318
319 static void xfrm_queue_purge(struct sk_buff_head *list)
320 {
321         struct sk_buff *skb;
322
323         while ((skb = skb_dequeue(list)) != NULL)
324                 kfree_skb(skb);
325 }
326
327 /* Rule must be locked. Release descentant resources, announce
328  * entry dead. The rule must be unlinked from lists to the moment.
329  */
330
331 static void xfrm_policy_kill(struct xfrm_policy *policy)
332 {
333         policy->walk.dead = 1;
334
335         atomic_inc(&policy->genid);
336
337         if (del_timer(&policy->polq.hold_timer))
338                 xfrm_pol_put(policy);
339         xfrm_queue_purge(&policy->polq.hold_queue);
340
341         if (del_timer(&policy->timer))
342                 xfrm_pol_put(policy);
343
344         xfrm_pol_put(policy);
345 }
346
347 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
348
349 static inline unsigned int idx_hash(struct net *net, u32 index)
350 {
351         return __idx_hash(index, net->xfrm.policy_idx_hmask);
352 }
353
354 static struct hlist_head *policy_hash_bysel(struct net *net,
355                                             const struct xfrm_selector *sel,
356                                             unsigned short family, int dir)
357 {
358         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
359         unsigned int hash = __sel_hash(sel, family, hmask);
360
361         return (hash == hmask + 1 ?
362                 &net->xfrm.policy_inexact[dir] :
363                 net->xfrm.policy_bydst[dir].table + hash);
364 }
365
366 static struct hlist_head *policy_hash_direct(struct net *net,
367                                              const xfrm_address_t *daddr,
368                                              const xfrm_address_t *saddr,
369                                              unsigned short family, int dir)
370 {
371         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
372         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
373
374         return net->xfrm.policy_bydst[dir].table + hash;
375 }
376
377 static void xfrm_dst_hash_transfer(struct hlist_head *list,
378                                    struct hlist_head *ndsttable,
379                                    unsigned int nhashmask)
380 {
381         struct hlist_node *tmp, *entry0 = NULL;
382         struct xfrm_policy *pol;
383         unsigned int h0 = 0;
384
385 redo:
386         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
387                 unsigned int h;
388
389                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
390                                 pol->family, nhashmask);
391                 if (!entry0) {
392                         hlist_del(&pol->bydst);
393                         hlist_add_head(&pol->bydst, ndsttable+h);
394                         h0 = h;
395                 } else {
396                         if (h != h0)
397                                 continue;
398                         hlist_del(&pol->bydst);
399                         hlist_add_after(entry0, &pol->bydst);
400                 }
401                 entry0 = &pol->bydst;
402         }
403         if (!hlist_empty(list)) {
404                 entry0 = NULL;
405                 goto redo;
406         }
407 }
408
409 static void xfrm_idx_hash_transfer(struct hlist_head *list,
410                                    struct hlist_head *nidxtable,
411                                    unsigned int nhashmask)
412 {
413         struct hlist_node *tmp;
414         struct xfrm_policy *pol;
415
416         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
417                 unsigned int h;
418
419                 h = __idx_hash(pol->index, nhashmask);
420                 hlist_add_head(&pol->byidx, nidxtable+h);
421         }
422 }
423
424 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
425 {
426         return ((old_hmask + 1) << 1) - 1;
427 }
428
429 static void xfrm_bydst_resize(struct net *net, int dir)
430 {
431         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
432         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
433         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
434         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
435         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
436         int i;
437
438         if (!ndst)
439                 return;
440
441         write_lock_bh(&xfrm_policy_lock);
442
443         for (i = hmask; i >= 0; i--)
444                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
445
446         net->xfrm.policy_bydst[dir].table = ndst;
447         net->xfrm.policy_bydst[dir].hmask = nhashmask;
448
449         write_unlock_bh(&xfrm_policy_lock);
450
451         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
452 }
453
454 static void xfrm_byidx_resize(struct net *net, int total)
455 {
456         unsigned int hmask = net->xfrm.policy_idx_hmask;
457         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
458         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
459         struct hlist_head *oidx = net->xfrm.policy_byidx;
460         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
461         int i;
462
463         if (!nidx)
464                 return;
465
466         write_lock_bh(&xfrm_policy_lock);
467
468         for (i = hmask; i >= 0; i--)
469                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
470
471         net->xfrm.policy_byidx = nidx;
472         net->xfrm.policy_idx_hmask = nhashmask;
473
474         write_unlock_bh(&xfrm_policy_lock);
475
476         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
477 }
478
479 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
480 {
481         unsigned int cnt = net->xfrm.policy_count[dir];
482         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
483
484         if (total)
485                 *total += cnt;
486
487         if ((hmask + 1) < xfrm_policy_hashmax &&
488             cnt > hmask)
489                 return 1;
490
491         return 0;
492 }
493
494 static inline int xfrm_byidx_should_resize(struct net *net, int total)
495 {
496         unsigned int hmask = net->xfrm.policy_idx_hmask;
497
498         if ((hmask + 1) < xfrm_policy_hashmax &&
499             total > hmask)
500                 return 1;
501
502         return 0;
503 }
504
505 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
506 {
507         read_lock_bh(&xfrm_policy_lock);
508         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
509         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
510         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
511         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
512         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
513         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
514         si->spdhcnt = net->xfrm.policy_idx_hmask;
515         si->spdhmcnt = xfrm_policy_hashmax;
516         read_unlock_bh(&xfrm_policy_lock);
517 }
518 EXPORT_SYMBOL(xfrm_spd_getinfo);
519
520 static DEFINE_MUTEX(hash_resize_mutex);
521 static void xfrm_hash_resize(struct work_struct *work)
522 {
523         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
524         int dir, total;
525
526         mutex_lock(&hash_resize_mutex);
527
528         total = 0;
529         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
530                 if (xfrm_bydst_should_resize(net, dir, &total))
531                         xfrm_bydst_resize(net, dir);
532         }
533         if (xfrm_byidx_should_resize(net, total))
534                 xfrm_byidx_resize(net, total);
535
536         mutex_unlock(&hash_resize_mutex);
537 }
538
539 /* Generate new index... KAME seems to generate them ordered by cost
540  * of an absolute inpredictability of ordering of rules. This will not pass. */
541 static u32 xfrm_gen_index(struct net *net, int dir)
542 {
543         static u32 idx_generator;
544
545         for (;;) {
546                 struct hlist_head *list;
547                 struct xfrm_policy *p;
548                 u32 idx;
549                 int found;
550
551                 idx = (idx_generator | dir);
552                 idx_generator += 8;
553                 if (idx == 0)
554                         idx = 8;
555                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
556                 found = 0;
557                 hlist_for_each_entry(p, list, byidx) {
558                         if (p->index == idx) {
559                                 found = 1;
560                                 break;
561                         }
562                 }
563                 if (!found)
564                         return idx;
565         }
566 }
567
568 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
569 {
570         u32 *p1 = (u32 *) s1;
571         u32 *p2 = (u32 *) s2;
572         int len = sizeof(struct xfrm_selector) / sizeof(u32);
573         int i;
574
575         for (i = 0; i < len; i++) {
576                 if (p1[i] != p2[i])
577                         return 1;
578         }
579
580         return 0;
581 }
582
583 static void xfrm_policy_requeue(struct xfrm_policy *old,
584                                 struct xfrm_policy *new)
585 {
586         struct xfrm_policy_queue *pq = &old->polq;
587         struct sk_buff_head list;
588
589         __skb_queue_head_init(&list);
590
591         spin_lock_bh(&pq->hold_queue.lock);
592         skb_queue_splice_init(&pq->hold_queue, &list);
593         if (del_timer(&pq->hold_timer))
594                 xfrm_pol_put(old);
595         spin_unlock_bh(&pq->hold_queue.lock);
596
597         if (skb_queue_empty(&list))
598                 return;
599
600         pq = &new->polq;
601
602         spin_lock_bh(&pq->hold_queue.lock);
603         skb_queue_splice(&list, &pq->hold_queue);
604         pq->timeout = XFRM_QUEUE_TMO_MIN;
605         if (!mod_timer(&pq->hold_timer, jiffies))
606                 xfrm_pol_hold(new);
607         spin_unlock_bh(&pq->hold_queue.lock);
608 }
609
610 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
611                                    struct xfrm_policy *pol)
612 {
613         u32 mark = policy->mark.v & policy->mark.m;
614
615         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
616                 return true;
617
618         if ((mark & pol->mark.m) == pol->mark.v &&
619             policy->priority == pol->priority)
620                 return true;
621
622         return false;
623 }
624
625 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
626 {
627         struct net *net = xp_net(policy);
628         struct xfrm_policy *pol;
629         struct xfrm_policy *delpol;
630         struct hlist_head *chain;
631         struct hlist_node *newpos;
632
633         write_lock_bh(&xfrm_policy_lock);
634         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
635         delpol = NULL;
636         newpos = NULL;
637         hlist_for_each_entry(pol, chain, bydst) {
638                 if (pol->type == policy->type &&
639                     !selector_cmp(&pol->selector, &policy->selector) &&
640                     xfrm_policy_mark_match(policy, pol) &&
641                     xfrm_sec_ctx_match(pol->security, policy->security) &&
642                     !WARN_ON(delpol)) {
643                         if (excl) {
644                                 write_unlock_bh(&xfrm_policy_lock);
645                                 return -EEXIST;
646                         }
647                         delpol = pol;
648                         if (policy->priority > pol->priority)
649                                 continue;
650                 } else if (policy->priority >= pol->priority) {
651                         newpos = &pol->bydst;
652                         continue;
653                 }
654                 if (delpol)
655                         break;
656         }
657         if (newpos)
658                 hlist_add_after(newpos, &policy->bydst);
659         else
660                 hlist_add_head(&policy->bydst, chain);
661         xfrm_pol_hold(policy);
662         net->xfrm.policy_count[dir]++;
663         atomic_inc(&flow_cache_genid);
664
665         /* After previous checking, family can either be AF_INET or AF_INET6 */
666         if (policy->family == AF_INET)
667                 rt_genid_bump_ipv4(net);
668         else
669                 rt_genid_bump_ipv6(net);
670
671         if (delpol) {
672                 xfrm_policy_requeue(delpol, policy);
673                 __xfrm_policy_unlink(delpol, dir);
674         }
675         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
676         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
677         policy->curlft.add_time = get_seconds();
678         policy->curlft.use_time = 0;
679         if (!mod_timer(&policy->timer, jiffies + HZ))
680                 xfrm_pol_hold(policy);
681         list_add(&policy->walk.all, &net->xfrm.policy_all);
682         write_unlock_bh(&xfrm_policy_lock);
683
684         if (delpol)
685                 xfrm_policy_kill(delpol);
686         else if (xfrm_bydst_should_resize(net, dir, NULL))
687                 schedule_work(&net->xfrm.policy_hash_work);
688
689         return 0;
690 }
691 EXPORT_SYMBOL(xfrm_policy_insert);
692
693 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
694                                           int dir, struct xfrm_selector *sel,
695                                           struct xfrm_sec_ctx *ctx, int delete,
696                                           int *err)
697 {
698         struct xfrm_policy *pol, *ret;
699         struct hlist_head *chain;
700
701         *err = 0;
702         write_lock_bh(&xfrm_policy_lock);
703         chain = policy_hash_bysel(net, sel, sel->family, dir);
704         ret = NULL;
705         hlist_for_each_entry(pol, chain, bydst) {
706                 if (pol->type == type &&
707                     (mark & pol->mark.m) == pol->mark.v &&
708                     !selector_cmp(sel, &pol->selector) &&
709                     xfrm_sec_ctx_match(ctx, pol->security)) {
710                         xfrm_pol_hold(pol);
711                         if (delete) {
712                                 *err = security_xfrm_policy_delete(
713                                                                 pol->security);
714                                 if (*err) {
715                                         write_unlock_bh(&xfrm_policy_lock);
716                                         return pol;
717                                 }
718                                 __xfrm_policy_unlink(pol, dir);
719                         }
720                         ret = pol;
721                         break;
722                 }
723         }
724         write_unlock_bh(&xfrm_policy_lock);
725
726         if (ret && delete)
727                 xfrm_policy_kill(ret);
728         return ret;
729 }
730 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
731
732 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
733                                      int dir, u32 id, int delete, int *err)
734 {
735         struct xfrm_policy *pol, *ret;
736         struct hlist_head *chain;
737
738         *err = -ENOENT;
739         if (xfrm_policy_id2dir(id) != dir)
740                 return NULL;
741
742         *err = 0;
743         write_lock_bh(&xfrm_policy_lock);
744         chain = net->xfrm.policy_byidx + idx_hash(net, id);
745         ret = NULL;
746         hlist_for_each_entry(pol, chain, byidx) {
747                 if (pol->type == type && pol->index == id &&
748                     (mark & pol->mark.m) == pol->mark.v) {
749                         xfrm_pol_hold(pol);
750                         if (delete) {
751                                 *err = security_xfrm_policy_delete(
752                                                                 pol->security);
753                                 if (*err) {
754                                         write_unlock_bh(&xfrm_policy_lock);
755                                         return pol;
756                                 }
757                                 __xfrm_policy_unlink(pol, dir);
758                         }
759                         ret = pol;
760                         break;
761                 }
762         }
763         write_unlock_bh(&xfrm_policy_lock);
764
765         if (ret && delete)
766                 xfrm_policy_kill(ret);
767         return ret;
768 }
769 EXPORT_SYMBOL(xfrm_policy_byid);
770
771 #ifdef CONFIG_SECURITY_NETWORK_XFRM
772 static inline int
773 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
774 {
775         int dir, err = 0;
776
777         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
778                 struct xfrm_policy *pol;
779                 int i;
780
781                 hlist_for_each_entry(pol,
782                                      &net->xfrm.policy_inexact[dir], bydst) {
783                         if (pol->type != type)
784                                 continue;
785                         err = security_xfrm_policy_delete(pol->security);
786                         if (err) {
787                                 xfrm_audit_policy_delete(pol, 0,
788                                                          audit_info->loginuid,
789                                                          audit_info->sessionid,
790                                                          audit_info->secid);
791                                 return err;
792                         }
793                 }
794                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
795                         hlist_for_each_entry(pol,
796                                              net->xfrm.policy_bydst[dir].table + i,
797                                              bydst) {
798                                 if (pol->type != type)
799                                         continue;
800                                 err = security_xfrm_policy_delete(
801                                                                 pol->security);
802                                 if (err) {
803                                         xfrm_audit_policy_delete(pol, 0,
804                                                         audit_info->loginuid,
805                                                         audit_info->sessionid,
806                                                         audit_info->secid);
807                                         return err;
808                                 }
809                         }
810                 }
811         }
812         return err;
813 }
814 #else
815 static inline int
816 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
817 {
818         return 0;
819 }
820 #endif
821
822 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
823 {
824         int dir, err = 0, cnt = 0;
825
826         write_lock_bh(&xfrm_policy_lock);
827
828         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
829         if (err)
830                 goto out;
831
832         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
833                 struct xfrm_policy *pol;
834                 int i;
835
836         again1:
837                 hlist_for_each_entry(pol,
838                                      &net->xfrm.policy_inexact[dir], bydst) {
839                         if (pol->type != type)
840                                 continue;
841                         __xfrm_policy_unlink(pol, dir);
842                         write_unlock_bh(&xfrm_policy_lock);
843                         cnt++;
844
845                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
846                                                  audit_info->sessionid,
847                                                  audit_info->secid);
848
849                         xfrm_policy_kill(pol);
850
851                         write_lock_bh(&xfrm_policy_lock);
852                         goto again1;
853                 }
854
855                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
856         again2:
857                         hlist_for_each_entry(pol,
858                                              net->xfrm.policy_bydst[dir].table + i,
859                                              bydst) {
860                                 if (pol->type != type)
861                                         continue;
862                                 __xfrm_policy_unlink(pol, dir);
863                                 write_unlock_bh(&xfrm_policy_lock);
864                                 cnt++;
865
866                                 xfrm_audit_policy_delete(pol, 1,
867                                                          audit_info->loginuid,
868                                                          audit_info->sessionid,
869                                                          audit_info->secid);
870                                 xfrm_policy_kill(pol);
871
872                                 write_lock_bh(&xfrm_policy_lock);
873                                 goto again2;
874                         }
875                 }
876
877         }
878         if (!cnt)
879                 err = -ESRCH;
880 out:
881         write_unlock_bh(&xfrm_policy_lock);
882         return err;
883 }
884 EXPORT_SYMBOL(xfrm_policy_flush);
885
886 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
887                      int (*func)(struct xfrm_policy *, int, int, void*),
888                      void *data)
889 {
890         struct xfrm_policy *pol;
891         struct xfrm_policy_walk_entry *x;
892         int error = 0;
893
894         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
895             walk->type != XFRM_POLICY_TYPE_ANY)
896                 return -EINVAL;
897
898         if (list_empty(&walk->walk.all) && walk->seq != 0)
899                 return 0;
900
901         write_lock_bh(&xfrm_policy_lock);
902         if (list_empty(&walk->walk.all))
903                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
904         else
905                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
906         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
907                 if (x->dead)
908                         continue;
909                 pol = container_of(x, struct xfrm_policy, walk);
910                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
911                     walk->type != pol->type)
912                         continue;
913                 error = func(pol, xfrm_policy_id2dir(pol->index),
914                              walk->seq, data);
915                 if (error) {
916                         list_move_tail(&walk->walk.all, &x->all);
917                         goto out;
918                 }
919                 walk->seq++;
920         }
921         if (walk->seq == 0) {
922                 error = -ENOENT;
923                 goto out;
924         }
925         list_del_init(&walk->walk.all);
926 out:
927         write_unlock_bh(&xfrm_policy_lock);
928         return error;
929 }
930 EXPORT_SYMBOL(xfrm_policy_walk);
931
932 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
933 {
934         INIT_LIST_HEAD(&walk->walk.all);
935         walk->walk.dead = 1;
936         walk->type = type;
937         walk->seq = 0;
938 }
939 EXPORT_SYMBOL(xfrm_policy_walk_init);
940
941 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
942 {
943         if (list_empty(&walk->walk.all))
944                 return;
945
946         write_lock_bh(&xfrm_policy_lock);
947         list_del(&walk->walk.all);
948         write_unlock_bh(&xfrm_policy_lock);
949 }
950 EXPORT_SYMBOL(xfrm_policy_walk_done);
951
952 /*
953  * Find policy to apply to this flow.
954  *
955  * Returns 0 if policy found, else an -errno.
956  */
957 static int xfrm_policy_match(const struct xfrm_policy *pol,
958                              const struct flowi *fl,
959                              u8 type, u16 family, int dir)
960 {
961         const struct xfrm_selector *sel = &pol->selector;
962         int ret = -ESRCH;
963         bool match;
964
965         if (pol->family != family ||
966             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
967             pol->type != type)
968                 return ret;
969
970         match = xfrm_selector_match(sel, fl, family);
971         if (match)
972                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
973                                                   dir);
974
975         return ret;
976 }
977
978 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
979                                                      const struct flowi *fl,
980                                                      u16 family, u8 dir)
981 {
982         int err;
983         struct xfrm_policy *pol, *ret;
984         const xfrm_address_t *daddr, *saddr;
985         struct hlist_head *chain;
986         u32 priority = ~0U;
987
988         daddr = xfrm_flowi_daddr(fl, family);
989         saddr = xfrm_flowi_saddr(fl, family);
990         if (unlikely(!daddr || !saddr))
991                 return NULL;
992
993         read_lock_bh(&xfrm_policy_lock);
994         chain = policy_hash_direct(net, daddr, saddr, family, dir);
995         ret = NULL;
996         hlist_for_each_entry(pol, chain, bydst) {
997                 err = xfrm_policy_match(pol, fl, type, family, dir);
998                 if (err) {
999                         if (err == -ESRCH)
1000                                 continue;
1001                         else {
1002                                 ret = ERR_PTR(err);
1003                                 goto fail;
1004                         }
1005                 } else {
1006                         ret = pol;
1007                         priority = ret->priority;
1008                         break;
1009                 }
1010         }
1011         chain = &net->xfrm.policy_inexact[dir];
1012         hlist_for_each_entry(pol, chain, bydst) {
1013                 err = xfrm_policy_match(pol, fl, type, family, dir);
1014                 if (err) {
1015                         if (err == -ESRCH)
1016                                 continue;
1017                         else {
1018                                 ret = ERR_PTR(err);
1019                                 goto fail;
1020                         }
1021                 } else if (pol->priority < priority) {
1022                         ret = pol;
1023                         break;
1024                 }
1025         }
1026         if (ret)
1027                 xfrm_pol_hold(ret);
1028 fail:
1029         read_unlock_bh(&xfrm_policy_lock);
1030
1031         return ret;
1032 }
1033
1034 static struct xfrm_policy *
1035 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1036 {
1037 #ifdef CONFIG_XFRM_SUB_POLICY
1038         struct xfrm_policy *pol;
1039
1040         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1041         if (pol != NULL)
1042                 return pol;
1043 #endif
1044         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1045 }
1046
1047 static int flow_to_policy_dir(int dir)
1048 {
1049         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1050             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1051             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1052                 return dir;
1053
1054         switch (dir) {
1055         default:
1056         case FLOW_DIR_IN:
1057                 return XFRM_POLICY_IN;
1058         case FLOW_DIR_OUT:
1059                 return XFRM_POLICY_OUT;
1060         case FLOW_DIR_FWD:
1061                 return XFRM_POLICY_FWD;
1062         }
1063 }
1064
1065 static struct flow_cache_object *
1066 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1067                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1068 {
1069         struct xfrm_policy *pol;
1070
1071         if (old_obj)
1072                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1073
1074         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1075         if (IS_ERR_OR_NULL(pol))
1076                 return ERR_CAST(pol);
1077
1078         /* Resolver returns two references:
1079          * one for cache and one for caller of flow_cache_lookup() */
1080         xfrm_pol_hold(pol);
1081
1082         return &pol->flo;
1083 }
1084
1085 static inline int policy_to_flow_dir(int dir)
1086 {
1087         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1088             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1089             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1090                 return dir;
1091         switch (dir) {
1092         default:
1093         case XFRM_POLICY_IN:
1094                 return FLOW_DIR_IN;
1095         case XFRM_POLICY_OUT:
1096                 return FLOW_DIR_OUT;
1097         case XFRM_POLICY_FWD:
1098                 return FLOW_DIR_FWD;
1099         }
1100 }
1101
1102 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1103                                                  const struct flowi *fl)
1104 {
1105         struct xfrm_policy *pol;
1106
1107         read_lock_bh(&xfrm_policy_lock);
1108         if ((pol = sk->sk_policy[dir]) != NULL) {
1109                 bool match = xfrm_selector_match(&pol->selector, fl,
1110                                                  sk->sk_family);
1111                 int err = 0;
1112
1113                 if (match) {
1114                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1115                                 pol = NULL;
1116                                 goto out;
1117                         }
1118                         err = security_xfrm_policy_lookup(pol->security,
1119                                                       fl->flowi_secid,
1120                                                       policy_to_flow_dir(dir));
1121                         if (!err)
1122                                 xfrm_pol_hold(pol);
1123                         else if (err == -ESRCH)
1124                                 pol = NULL;
1125                         else
1126                                 pol = ERR_PTR(err);
1127                 } else
1128                         pol = NULL;
1129         }
1130 out:
1131         read_unlock_bh(&xfrm_policy_lock);
1132         return pol;
1133 }
1134
1135 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1136 {
1137         struct net *net = xp_net(pol);
1138         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1139                                                      pol->family, dir);
1140
1141         list_add(&pol->walk.all, &net->xfrm.policy_all);
1142         hlist_add_head(&pol->bydst, chain);
1143         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1144         net->xfrm.policy_count[dir]++;
1145         xfrm_pol_hold(pol);
1146
1147         if (xfrm_bydst_should_resize(net, dir, NULL))
1148                 schedule_work(&net->xfrm.policy_hash_work);
1149 }
1150
1151 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1152                                                 int dir)
1153 {
1154         struct net *net = xp_net(pol);
1155
1156         if (hlist_unhashed(&pol->bydst))
1157                 return NULL;
1158
1159         hlist_del(&pol->bydst);
1160         hlist_del(&pol->byidx);
1161         list_del(&pol->walk.all);
1162         net->xfrm.policy_count[dir]--;
1163
1164         return pol;
1165 }
1166
1167 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1168 {
1169         write_lock_bh(&xfrm_policy_lock);
1170         pol = __xfrm_policy_unlink(pol, dir);
1171         write_unlock_bh(&xfrm_policy_lock);
1172         if (pol) {
1173                 xfrm_policy_kill(pol);
1174                 return 0;
1175         }
1176         return -ENOENT;
1177 }
1178 EXPORT_SYMBOL(xfrm_policy_delete);
1179
1180 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1181 {
1182         struct net *net = xp_net(pol);
1183         struct xfrm_policy *old_pol;
1184
1185 #ifdef CONFIG_XFRM_SUB_POLICY
1186         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1187                 return -EINVAL;
1188 #endif
1189
1190         write_lock_bh(&xfrm_policy_lock);
1191         old_pol = sk->sk_policy[dir];
1192         sk->sk_policy[dir] = pol;
1193         if (pol) {
1194                 pol->curlft.add_time = get_seconds();
1195                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1196                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1197         }
1198         if (old_pol) {
1199                 if (pol)
1200                         xfrm_policy_requeue(old_pol, pol);
1201
1202                 /* Unlinking succeeds always. This is the only function
1203                  * allowed to delete or replace socket policy.
1204                  */
1205                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1206         }
1207         write_unlock_bh(&xfrm_policy_lock);
1208
1209         if (old_pol) {
1210                 xfrm_policy_kill(old_pol);
1211         }
1212         return 0;
1213 }
1214
1215 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1216 {
1217         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1218
1219         if (newp) {
1220                 newp->selector = old->selector;
1221                 if (security_xfrm_policy_clone(old->security,
1222                                                &newp->security)) {
1223                         kfree(newp);
1224                         return NULL;  /* ENOMEM */
1225                 }
1226                 newp->lft = old->lft;
1227                 newp->curlft = old->curlft;
1228                 newp->mark = old->mark;
1229                 newp->action = old->action;
1230                 newp->flags = old->flags;
1231                 newp->xfrm_nr = old->xfrm_nr;
1232                 newp->index = old->index;
1233                 newp->type = old->type;
1234                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1235                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1236                 write_lock_bh(&xfrm_policy_lock);
1237                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1238                 write_unlock_bh(&xfrm_policy_lock);
1239                 xfrm_pol_put(newp);
1240         }
1241         return newp;
1242 }
1243
1244 int __xfrm_sk_clone_policy(struct sock *sk)
1245 {
1246         struct xfrm_policy *p0 = sk->sk_policy[0],
1247                            *p1 = sk->sk_policy[1];
1248
1249         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1250         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1251                 return -ENOMEM;
1252         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1253                 return -ENOMEM;
1254         return 0;
1255 }
1256
1257 static int
1258 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1259                unsigned short family)
1260 {
1261         int err;
1262         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1263
1264         if (unlikely(afinfo == NULL))
1265                 return -EINVAL;
1266         err = afinfo->get_saddr(net, local, remote);
1267         xfrm_policy_put_afinfo(afinfo);
1268         return err;
1269 }
1270
1271 /* Resolve list of templates for the flow, given policy. */
1272
1273 static int
1274 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1275                       struct xfrm_state **xfrm, unsigned short family)
1276 {
1277         struct net *net = xp_net(policy);
1278         int nx;
1279         int i, error;
1280         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1281         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1282         xfrm_address_t tmp;
1283
1284         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1285                 struct xfrm_state *x;
1286                 xfrm_address_t *remote = daddr;
1287                 xfrm_address_t *local  = saddr;
1288                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1289
1290                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1291                     tmpl->mode == XFRM_MODE_BEET) {
1292                         remote = &tmpl->id.daddr;
1293                         local = &tmpl->saddr;
1294                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1295                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1296                                 if (error)
1297                                         goto fail;
1298                                 local = &tmp;
1299                         }
1300                 }
1301
1302                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1303
1304                 if (x && x->km.state == XFRM_STATE_VALID) {
1305                         xfrm[nx++] = x;
1306                         daddr = remote;
1307                         saddr = local;
1308                         continue;
1309                 }
1310                 if (x) {
1311                         error = (x->km.state == XFRM_STATE_ERROR ?
1312                                  -EINVAL : -EAGAIN);
1313                         xfrm_state_put(x);
1314                 }
1315                 else if (error == -ESRCH)
1316                         error = -EAGAIN;
1317
1318                 if (!tmpl->optional)
1319                         goto fail;
1320         }
1321         return nx;
1322
1323 fail:
1324         for (nx--; nx>=0; nx--)
1325                 xfrm_state_put(xfrm[nx]);
1326         return error;
1327 }
1328
1329 static int
1330 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1331                   struct xfrm_state **xfrm, unsigned short family)
1332 {
1333         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1334         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1335         int cnx = 0;
1336         int error;
1337         int ret;
1338         int i;
1339
1340         for (i = 0; i < npols; i++) {
1341                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1342                         error = -ENOBUFS;
1343                         goto fail;
1344                 }
1345
1346                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1347                 if (ret < 0) {
1348                         error = ret;
1349                         goto fail;
1350                 } else
1351                         cnx += ret;
1352         }
1353
1354         /* found states are sorted for outbound processing */
1355         if (npols > 1)
1356                 xfrm_state_sort(xfrm, tpp, cnx, family);
1357
1358         return cnx;
1359
1360  fail:
1361         for (cnx--; cnx>=0; cnx--)
1362                 xfrm_state_put(tpp[cnx]);
1363         return error;
1364
1365 }
1366
1367 /* Check that the bundle accepts the flow and its components are
1368  * still valid.
1369  */
1370
1371 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1372 {
1373         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1374         int tos;
1375
1376         if (!afinfo)
1377                 return -EINVAL;
1378
1379         tos = afinfo->get_tos(fl);
1380
1381         xfrm_policy_put_afinfo(afinfo);
1382
1383         return tos;
1384 }
1385
1386 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1387 {
1388         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1389         struct dst_entry *dst = &xdst->u.dst;
1390
1391         if (xdst->route == NULL) {
1392                 /* Dummy bundle - if it has xfrms we were not
1393                  * able to build bundle as template resolution failed.
1394                  * It means we need to try again resolving. */
1395                 if (xdst->num_xfrms > 0)
1396                         return NULL;
1397         } else if (dst->flags & DST_XFRM_QUEUE) {
1398                 return NULL;
1399         } else {
1400                 /* Real bundle */
1401                 if (stale_bundle(dst))
1402                         return NULL;
1403         }
1404
1405         dst_hold(dst);
1406         return flo;
1407 }
1408
1409 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1410 {
1411         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1412         struct dst_entry *dst = &xdst->u.dst;
1413
1414         if (!xdst->route)
1415                 return 0;
1416         if (stale_bundle(dst))
1417                 return 0;
1418
1419         return 1;
1420 }
1421
1422 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1423 {
1424         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1425         struct dst_entry *dst = &xdst->u.dst;
1426
1427         dst_free(dst);
1428 }
1429
1430 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1431         .get = xfrm_bundle_flo_get,
1432         .check = xfrm_bundle_flo_check,
1433         .delete = xfrm_bundle_flo_delete,
1434 };
1435
1436 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1437 {
1438         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1439         struct dst_ops *dst_ops;
1440         struct xfrm_dst *xdst;
1441
1442         if (!afinfo)
1443                 return ERR_PTR(-EINVAL);
1444
1445         switch (family) {
1446         case AF_INET:
1447                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1448                 break;
1449 #if IS_ENABLED(CONFIG_IPV6)
1450         case AF_INET6:
1451                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1452                 break;
1453 #endif
1454         default:
1455                 BUG();
1456         }
1457         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1458
1459         if (likely(xdst)) {
1460                 struct dst_entry *dst = &xdst->u.dst;
1461
1462                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1463                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1464                 if (afinfo->init_dst)
1465                         afinfo->init_dst(net, xdst);
1466         } else
1467                 xdst = ERR_PTR(-ENOBUFS);
1468
1469         xfrm_policy_put_afinfo(afinfo);
1470
1471         return xdst;
1472 }
1473
1474 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1475                                  int nfheader_len)
1476 {
1477         struct xfrm_policy_afinfo *afinfo =
1478                 xfrm_policy_get_afinfo(dst->ops->family);
1479         int err;
1480
1481         if (!afinfo)
1482                 return -EINVAL;
1483
1484         err = afinfo->init_path(path, dst, nfheader_len);
1485
1486         xfrm_policy_put_afinfo(afinfo);
1487
1488         return err;
1489 }
1490
1491 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1492                                 const struct flowi *fl)
1493 {
1494         struct xfrm_policy_afinfo *afinfo =
1495                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1496         int err;
1497
1498         if (!afinfo)
1499                 return -EINVAL;
1500
1501         err = afinfo->fill_dst(xdst, dev, fl);
1502
1503         xfrm_policy_put_afinfo(afinfo);
1504
1505         return err;
1506 }
1507
1508
1509 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1510  * all the metrics... Shortly, bundle a bundle.
1511  */
1512
1513 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1514                                             struct xfrm_state **xfrm, int nx,
1515                                             const struct flowi *fl,
1516                                             struct dst_entry *dst)
1517 {
1518         struct net *net = xp_net(policy);
1519         unsigned long now = jiffies;
1520         struct net_device *dev;
1521         struct xfrm_mode *inner_mode;
1522         struct dst_entry *dst_prev = NULL;
1523         struct dst_entry *dst0 = NULL;
1524         int i = 0;
1525         int err;
1526         int header_len = 0;
1527         int nfheader_len = 0;
1528         int trailer_len = 0;
1529         int tos;
1530         int family = policy->selector.family;
1531         xfrm_address_t saddr, daddr;
1532
1533         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1534
1535         tos = xfrm_get_tos(fl, family);
1536         err = tos;
1537         if (tos < 0)
1538                 goto put_states;
1539
1540         dst_hold(dst);
1541
1542         for (; i < nx; i++) {
1543                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1544                 struct dst_entry *dst1 = &xdst->u.dst;
1545
1546                 err = PTR_ERR(xdst);
1547                 if (IS_ERR(xdst)) {
1548                         dst_release(dst);
1549                         goto put_states;
1550                 }
1551
1552                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1553                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1554                                                         xfrm_af2proto(family));
1555                         if (!inner_mode) {
1556                                 err = -EAFNOSUPPORT;
1557                                 dst_release(dst);
1558                                 goto put_states;
1559                         }
1560                 } else
1561                         inner_mode = xfrm[i]->inner_mode;
1562
1563                 if (!dst_prev)
1564                         dst0 = dst1;
1565                 else {
1566                         dst_prev->child = dst_clone(dst1);
1567                         dst1->flags |= DST_NOHASH;
1568                 }
1569
1570                 xdst->route = dst;
1571                 dst_copy_metrics(dst1, dst);
1572
1573                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1574                         family = xfrm[i]->props.family;
1575                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1576                                               family);
1577                         err = PTR_ERR(dst);
1578                         if (IS_ERR(dst))
1579                                 goto put_states;
1580                 } else
1581                         dst_hold(dst);
1582
1583                 dst1->xfrm = xfrm[i];
1584                 xdst->xfrm_genid = xfrm[i]->genid;
1585
1586                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1587                 dst1->flags |= DST_HOST;
1588                 dst1->lastuse = now;
1589
1590                 dst1->input = dst_discard;
1591                 dst1->output = inner_mode->afinfo->output;
1592
1593                 dst1->next = dst_prev;
1594                 dst_prev = dst1;
1595
1596                 header_len += xfrm[i]->props.header_len;
1597                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1598                         nfheader_len += xfrm[i]->props.header_len;
1599                 trailer_len += xfrm[i]->props.trailer_len;
1600         }
1601
1602         dst_prev->child = dst;
1603         dst0->path = dst;
1604
1605         err = -ENODEV;
1606         dev = dst->dev;
1607         if (!dev)
1608                 goto free_dst;
1609
1610         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1611         xfrm_init_pmtu(dst_prev);
1612
1613         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1614                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1615
1616                 err = xfrm_fill_dst(xdst, dev, fl);
1617                 if (err)
1618                         goto free_dst;
1619
1620                 dst_prev->header_len = header_len;
1621                 dst_prev->trailer_len = trailer_len;
1622                 header_len -= xdst->u.dst.xfrm->props.header_len;
1623                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1624         }
1625
1626 out:
1627         return dst0;
1628
1629 put_states:
1630         for (; i < nx; i++)
1631                 xfrm_state_put(xfrm[i]);
1632 free_dst:
1633         if (dst0)
1634                 dst_free(dst0);
1635         dst0 = ERR_PTR(err);
1636         goto out;
1637 }
1638
1639 static int inline
1640 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1641 {
1642         if (!*target) {
1643                 *target = kmalloc(size, GFP_ATOMIC);
1644                 if (!*target)
1645                         return -ENOMEM;
1646         }
1647         memcpy(*target, src, size);
1648         return 0;
1649 }
1650
1651 static int inline
1652 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1653 {
1654 #ifdef CONFIG_XFRM_SUB_POLICY
1655         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1656         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1657                                    sel, sizeof(*sel));
1658 #else
1659         return 0;
1660 #endif
1661 }
1662
1663 static int inline
1664 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1665 {
1666 #ifdef CONFIG_XFRM_SUB_POLICY
1667         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1668         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1669 #else
1670         return 0;
1671 #endif
1672 }
1673
1674 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1675                                 struct xfrm_policy **pols,
1676                                 int *num_pols, int *num_xfrms)
1677 {
1678         int i;
1679
1680         if (*num_pols == 0 || !pols[0]) {
1681                 *num_pols = 0;
1682                 *num_xfrms = 0;
1683                 return 0;
1684         }
1685         if (IS_ERR(pols[0]))
1686                 return PTR_ERR(pols[0]);
1687
1688         *num_xfrms = pols[0]->xfrm_nr;
1689
1690 #ifdef CONFIG_XFRM_SUB_POLICY
1691         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1692             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1693                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1694                                                     XFRM_POLICY_TYPE_MAIN,
1695                                                     fl, family,
1696                                                     XFRM_POLICY_OUT);
1697                 if (pols[1]) {
1698                         if (IS_ERR(pols[1])) {
1699                                 xfrm_pols_put(pols, *num_pols);
1700                                 return PTR_ERR(pols[1]);
1701                         }
1702                         (*num_pols) ++;
1703                         (*num_xfrms) += pols[1]->xfrm_nr;
1704                 }
1705         }
1706 #endif
1707         for (i = 0; i < *num_pols; i++) {
1708                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1709                         *num_xfrms = -1;
1710                         break;
1711                 }
1712         }
1713
1714         return 0;
1715
1716 }
1717
1718 static struct xfrm_dst *
1719 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1720                                const struct flowi *fl, u16 family,
1721                                struct dst_entry *dst_orig)
1722 {
1723         struct net *net = xp_net(pols[0]);
1724         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1725         struct dst_entry *dst;
1726         struct xfrm_dst *xdst;
1727         int err;
1728
1729         /* Try to instantiate a bundle */
1730         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1731         if (err <= 0) {
1732                 if (err != 0 && err != -EAGAIN)
1733                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1734                 return ERR_PTR(err);
1735         }
1736
1737         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1738         if (IS_ERR(dst)) {
1739                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1740                 return ERR_CAST(dst);
1741         }
1742
1743         xdst = (struct xfrm_dst *)dst;
1744         xdst->num_xfrms = err;
1745         if (num_pols > 1)
1746                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1747         else
1748                 err = xfrm_dst_update_origin(dst, fl);
1749         if (unlikely(err)) {
1750                 dst_free(dst);
1751                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1752                 return ERR_PTR(err);
1753         }
1754
1755         xdst->num_pols = num_pols;
1756         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1757         xdst->policy_genid = atomic_read(&pols[0]->genid);
1758
1759         return xdst;
1760 }
1761
1762 static void xfrm_policy_queue_process(unsigned long arg)
1763 {
1764         int err = 0;
1765         struct sk_buff *skb;
1766         struct sock *sk;
1767         struct dst_entry *dst;
1768         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1769         struct xfrm_policy_queue *pq = &pol->polq;
1770         struct flowi fl;
1771         struct sk_buff_head list;
1772
1773         spin_lock(&pq->hold_queue.lock);
1774         skb = skb_peek(&pq->hold_queue);
1775         if (!skb) {
1776                 spin_unlock(&pq->hold_queue.lock);
1777                 goto out;
1778         }
1779         dst = skb_dst(skb);
1780         sk = skb->sk;
1781         xfrm_decode_session(skb, &fl, dst->ops->family);
1782         spin_unlock(&pq->hold_queue.lock);
1783
1784         dst_hold(dst->path);
1785         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1786                           sk, 0);
1787         if (IS_ERR(dst))
1788                 goto purge_queue;
1789
1790         if (dst->flags & DST_XFRM_QUEUE) {
1791                 dst_release(dst);
1792
1793                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1794                         goto purge_queue;
1795
1796                 pq->timeout = pq->timeout << 1;
1797                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1798                         xfrm_pol_hold(pol);
1799         goto out;
1800         }
1801
1802         dst_release(dst);
1803
1804         __skb_queue_head_init(&list);
1805
1806         spin_lock(&pq->hold_queue.lock);
1807         pq->timeout = 0;
1808         skb_queue_splice_init(&pq->hold_queue, &list);
1809         spin_unlock(&pq->hold_queue.lock);
1810
1811         while (!skb_queue_empty(&list)) {
1812                 skb = __skb_dequeue(&list);
1813
1814                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1815                 dst_hold(skb_dst(skb)->path);
1816                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1817                                   &fl, skb->sk, 0);
1818                 if (IS_ERR(dst)) {
1819                         kfree_skb(skb);
1820                         continue;
1821                 }
1822
1823                 nf_reset(skb);
1824                 skb_dst_drop(skb);
1825                 skb_dst_set(skb, dst);
1826
1827                 err = dst_output(skb);
1828         }
1829
1830 out:
1831         xfrm_pol_put(pol);
1832         return;
1833
1834 purge_queue:
1835         pq->timeout = 0;
1836         xfrm_queue_purge(&pq->hold_queue);
1837         xfrm_pol_put(pol);
1838 }
1839
1840 static int xdst_queue_output(struct sk_buff *skb)
1841 {
1842         unsigned long sched_next;
1843         struct dst_entry *dst = skb_dst(skb);
1844         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1845         struct xfrm_policy *pol = xdst->pols[0];
1846         struct xfrm_policy_queue *pq = &pol->polq;
1847         const struct sk_buff *fclone = skb + 1;
1848
1849         if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
1850                      fclone->fclone == SKB_FCLONE_CLONE)) {
1851                 kfree_skb(skb);
1852                 return 0;
1853         }
1854
1855         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1856                 kfree_skb(skb);
1857                 return -EAGAIN;
1858         }
1859
1860         skb_dst_force(skb);
1861
1862         spin_lock_bh(&pq->hold_queue.lock);
1863
1864         if (!pq->timeout)
1865                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1866
1867         sched_next = jiffies + pq->timeout;
1868
1869         if (del_timer(&pq->hold_timer)) {
1870                 if (time_before(pq->hold_timer.expires, sched_next))
1871                         sched_next = pq->hold_timer.expires;
1872                 xfrm_pol_put(pol);
1873         }
1874
1875         __skb_queue_tail(&pq->hold_queue, skb);
1876         if (!mod_timer(&pq->hold_timer, sched_next))
1877                 xfrm_pol_hold(pol);
1878
1879         spin_unlock_bh(&pq->hold_queue.lock);
1880
1881         return 0;
1882 }
1883
1884 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1885                                                  struct dst_entry *dst,
1886                                                  const struct flowi *fl,
1887                                                  int num_xfrms,
1888                                                  u16 family)
1889 {
1890         int err;
1891         struct net_device *dev;
1892         struct dst_entry *dst1;
1893         struct xfrm_dst *xdst;
1894
1895         xdst = xfrm_alloc_dst(net, family);
1896         if (IS_ERR(xdst))
1897                 return xdst;
1898
1899         if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0 ||
1900             (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP))
1901                 return xdst;
1902
1903         dst1 = &xdst->u.dst;
1904         dst_hold(dst);
1905         xdst->route = dst;
1906
1907         dst_copy_metrics(dst1, dst);
1908
1909         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1910         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1911         dst1->lastuse = jiffies;
1912
1913         dst1->input = dst_discard;
1914         dst1->output = xdst_queue_output;
1915
1916         dst_hold(dst);
1917         dst1->child = dst;
1918         dst1->path = dst;
1919
1920         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1921
1922         err = -ENODEV;
1923         dev = dst->dev;
1924         if (!dev)
1925                 goto free_dst;
1926
1927         err = xfrm_fill_dst(xdst, dev, fl);
1928         if (err)
1929                 goto free_dst;
1930
1931 out:
1932         return xdst;
1933
1934 free_dst:
1935         dst_release(dst1);
1936         xdst = ERR_PTR(err);
1937         goto out;
1938 }
1939
1940 static struct flow_cache_object *
1941 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1942                    struct flow_cache_object *oldflo, void *ctx)
1943 {
1944         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1945         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1946         struct xfrm_dst *xdst, *new_xdst;
1947         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1948
1949         /* Check if the policies from old bundle are usable */
1950         xdst = NULL;
1951         if (oldflo) {
1952                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1953                 num_pols = xdst->num_pols;
1954                 num_xfrms = xdst->num_xfrms;
1955                 pol_dead = 0;
1956                 for (i = 0; i < num_pols; i++) {
1957                         pols[i] = xdst->pols[i];
1958                         pol_dead |= pols[i]->walk.dead;
1959                 }
1960                 if (pol_dead) {
1961                         dst_free(&xdst->u.dst);
1962                         xdst = NULL;
1963                         num_pols = 0;
1964                         num_xfrms = 0;
1965                         oldflo = NULL;
1966                 }
1967         }
1968
1969         /* Resolve policies to use if we couldn't get them from
1970          * previous cache entry */
1971         if (xdst == NULL) {
1972                 num_pols = 1;
1973                 pols[0] = __xfrm_policy_lookup(net, fl, family,
1974                                                flow_to_policy_dir(dir));
1975                 err = xfrm_expand_policies(fl, family, pols,
1976                                            &num_pols, &num_xfrms);
1977                 if (err < 0)
1978                         goto inc_error;
1979                 if (num_pols == 0)
1980                         return NULL;
1981                 if (num_xfrms <= 0)
1982                         goto make_dummy_bundle;
1983         }
1984
1985         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1986         if (IS_ERR(new_xdst)) {
1987                 err = PTR_ERR(new_xdst);
1988                 if (err != -EAGAIN)
1989                         goto error;
1990                 if (oldflo == NULL)
1991                         goto make_dummy_bundle;
1992                 dst_hold(&xdst->u.dst);
1993                 return oldflo;
1994         } else if (new_xdst == NULL) {
1995                 num_xfrms = 0;
1996                 if (oldflo == NULL)
1997                         goto make_dummy_bundle;
1998                 xdst->num_xfrms = 0;
1999                 dst_hold(&xdst->u.dst);
2000                 return oldflo;
2001         }
2002
2003         /* Kill the previous bundle */
2004         if (xdst) {
2005                 /* The policies were stolen for newly generated bundle */
2006                 xdst->num_pols = 0;
2007                 dst_free(&xdst->u.dst);
2008         }
2009
2010         /* Flow cache does not have reference, it dst_free()'s,
2011          * but we do need to return one reference for original caller */
2012         dst_hold(&new_xdst->u.dst);
2013         return &new_xdst->flo;
2014
2015 make_dummy_bundle:
2016         /* We found policies, but there's no bundles to instantiate:
2017          * either because the policy blocks, has no transformations or
2018          * we could not build template (no xfrm_states).*/
2019         xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
2020         if (IS_ERR(xdst)) {
2021                 xfrm_pols_put(pols, num_pols);
2022                 return ERR_CAST(xdst);
2023         }
2024         xdst->num_pols = num_pols;
2025         xdst->num_xfrms = num_xfrms;
2026         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
2027
2028         dst_hold(&xdst->u.dst);
2029         return &xdst->flo;
2030
2031 inc_error:
2032         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2033 error:
2034         if (xdst != NULL)
2035                 dst_free(&xdst->u.dst);
2036         else
2037                 xfrm_pols_put(pols, num_pols);
2038         return ERR_PTR(err);
2039 }
2040
2041 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2042                                         struct dst_entry *dst_orig)
2043 {
2044         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2045         struct dst_entry *ret;
2046
2047         if (!afinfo) {
2048                 dst_release(dst_orig);
2049                 return ERR_PTR(-EINVAL);
2050         } else {
2051                 ret = afinfo->blackhole_route(net, dst_orig);
2052         }
2053         xfrm_policy_put_afinfo(afinfo);
2054
2055         return ret;
2056 }
2057
2058 /* Main function: finds/creates a bundle for given flow.
2059  *
2060  * At the moment we eat a raw IP route. Mostly to speed up lookups
2061  * on interfaces with disabled IPsec.
2062  */
2063 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2064                               const struct flowi *fl,
2065                               struct sock *sk, int flags)
2066 {
2067         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2068         struct flow_cache_object *flo;
2069         struct xfrm_dst *xdst;
2070         struct dst_entry *dst, *route;
2071         u16 family = dst_orig->ops->family;
2072         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2073         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2074
2075 restart:
2076         dst = NULL;
2077         xdst = NULL;
2078         route = NULL;
2079
2080         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2081                 num_pols = 1;
2082                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2083                 err = xfrm_expand_policies(fl, family, pols,
2084                                            &num_pols, &num_xfrms);
2085                 if (err < 0)
2086                         goto dropdst;
2087
2088                 if (num_pols) {
2089                         if (num_xfrms <= 0) {
2090                                 drop_pols = num_pols;
2091                                 goto no_transform;
2092                         }
2093
2094                         xdst = xfrm_resolve_and_create_bundle(
2095                                         pols, num_pols, fl,
2096                                         family, dst_orig);
2097                         if (IS_ERR(xdst)) {
2098                                 xfrm_pols_put(pols, num_pols);
2099                                 err = PTR_ERR(xdst);
2100                                 goto dropdst;
2101                         } else if (xdst == NULL) {
2102                                 num_xfrms = 0;
2103                                 drop_pols = num_pols;
2104                                 goto no_transform;
2105                         }
2106
2107                         dst_hold(&xdst->u.dst);
2108
2109                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2110                         xdst->u.dst.next = xfrm_policy_sk_bundles;
2111                         xfrm_policy_sk_bundles = &xdst->u.dst;
2112                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2113
2114                         route = xdst->route;
2115                 }
2116         }
2117
2118         if (xdst == NULL) {
2119                 /* To accelerate a bit...  */
2120                 if ((dst_orig->flags & DST_NOXFRM) ||
2121                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2122                         goto nopol;
2123
2124                 flo = flow_cache_lookup(net, fl, family, dir,
2125                                         xfrm_bundle_lookup, dst_orig);
2126                 if (flo == NULL)
2127                         goto nopol;
2128                 if (IS_ERR(flo)) {
2129                         err = PTR_ERR(flo);
2130                         goto dropdst;
2131                 }
2132                 xdst = container_of(flo, struct xfrm_dst, flo);
2133
2134                 num_pols = xdst->num_pols;
2135                 num_xfrms = xdst->num_xfrms;
2136                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
2137                 route = xdst->route;
2138         }
2139
2140         dst = &xdst->u.dst;
2141         if (route == NULL && num_xfrms > 0) {
2142                 /* The only case when xfrm_bundle_lookup() returns a
2143                  * bundle with null route, is when the template could
2144                  * not be resolved. It means policies are there, but
2145                  * bundle could not be created, since we don't yet
2146                  * have the xfrm_state's. We need to wait for KM to
2147                  * negotiate new SA's or bail out with error.*/
2148                 if (net->xfrm.sysctl_larval_drop) {
2149                         dst_release(dst);
2150                         xfrm_pols_put(pols, drop_pols);
2151                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2152
2153                         return make_blackhole(net, family, dst_orig);
2154                 }
2155                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
2156                         DECLARE_WAITQUEUE(wait, current);
2157
2158                         add_wait_queue(&net->xfrm.km_waitq, &wait);
2159                         set_current_state(TASK_INTERRUPTIBLE);
2160                         schedule();
2161                         set_current_state(TASK_RUNNING);
2162                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
2163
2164                         if (!signal_pending(current)) {
2165                                 dst_release(dst);
2166                                 goto restart;
2167                         }
2168
2169                         err = -ERESTART;
2170                 } else
2171                         err = -EAGAIN;
2172
2173                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2174                 goto error;
2175         }
2176
2177 no_transform:
2178         if (num_pols == 0)
2179                 goto nopol;
2180
2181         if ((flags & XFRM_LOOKUP_ICMP) &&
2182             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2183                 err = -ENOENT;
2184                 goto error;
2185         }
2186
2187         for (i = 0; i < num_pols; i++)
2188                 pols[i]->curlft.use_time = get_seconds();
2189
2190         if (num_xfrms < 0) {
2191                 /* Prohibit the flow */
2192                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2193                 err = -EPERM;
2194                 goto error;
2195         } else if (num_xfrms > 0) {
2196                 /* Flow transformed */
2197                 dst_release(dst_orig);
2198         } else {
2199                 /* Flow passes untransformed */
2200                 dst_release(dst);
2201                 dst = dst_orig;
2202         }
2203 ok:
2204         xfrm_pols_put(pols, drop_pols);
2205         if (dst && dst->xfrm &&
2206             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2207                 dst->flags |= DST_XFRM_TUNNEL;
2208         return dst;
2209
2210 nopol:
2211         if (!(flags & XFRM_LOOKUP_ICMP)) {
2212                 dst = dst_orig;
2213                 goto ok;
2214         }
2215         err = -ENOENT;
2216 error:
2217         dst_release(dst);
2218 dropdst:
2219         dst_release(dst_orig);
2220         xfrm_pols_put(pols, drop_pols);
2221         return ERR_PTR(err);
2222 }
2223 EXPORT_SYMBOL(xfrm_lookup);
2224
2225 static inline int
2226 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2227 {
2228         struct xfrm_state *x;
2229
2230         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2231                 return 0;
2232         x = skb->sp->xvec[idx];
2233         if (!x->type->reject)
2234                 return 0;
2235         return x->type->reject(x, skb, fl);
2236 }
2237
2238 /* When skb is transformed back to its "native" form, we have to
2239  * check policy restrictions. At the moment we make this in maximally
2240  * stupid way. Shame on me. :-) Of course, connected sockets must
2241  * have policy cached at them.
2242  */
2243
2244 static inline int
2245 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2246               unsigned short family)
2247 {
2248         if (xfrm_state_kern(x))
2249                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2250         return  x->id.proto == tmpl->id.proto &&
2251                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2252                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2253                 x->props.mode == tmpl->mode &&
2254                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2255                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2256                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2257                   xfrm_state_addr_cmp(tmpl, x, family));
2258 }
2259
2260 /*
2261  * 0 or more than 0 is returned when validation is succeeded (either bypass
2262  * because of optional transport mode, or next index of the mathced secpath
2263  * state with the template.
2264  * -1 is returned when no matching template is found.
2265  * Otherwise "-2 - errored_index" is returned.
2266  */
2267 static inline int
2268 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2269                unsigned short family)
2270 {
2271         int idx = start;
2272
2273         if (tmpl->optional) {
2274                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2275                         return start;
2276         } else
2277                 start = -1;
2278         for (; idx < sp->len; idx++) {
2279                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2280                         return ++idx;
2281                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2282                         if (start == -1)
2283                                 start = -2-idx;
2284                         break;
2285                 }
2286         }
2287         return start;
2288 }
2289
2290 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2291                           unsigned int family, int reverse)
2292 {
2293         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2294         int err;
2295
2296         if (unlikely(afinfo == NULL))
2297                 return -EAFNOSUPPORT;
2298
2299         afinfo->decode_session(skb, fl, reverse);
2300         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2301         xfrm_policy_put_afinfo(afinfo);
2302         return err;
2303 }
2304 EXPORT_SYMBOL(__xfrm_decode_session);
2305
2306 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2307 {
2308         for (; k < sp->len; k++) {
2309                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2310                         *idxp = k;
2311                         return 1;
2312                 }
2313         }
2314
2315         return 0;
2316 }
2317
2318 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2319                         unsigned short family)
2320 {
2321         struct net *net = dev_net(skb->dev);
2322         struct xfrm_policy *pol;
2323         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2324         int npols = 0;
2325         int xfrm_nr;
2326         int pi;
2327         int reverse;
2328         struct flowi fl;
2329         u8 fl_dir;
2330         int xerr_idx = -1;
2331
2332         reverse = dir & ~XFRM_POLICY_MASK;
2333         dir &= XFRM_POLICY_MASK;
2334         fl_dir = policy_to_flow_dir(dir);
2335
2336         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2337                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2338                 return 0;
2339         }
2340
2341         nf_nat_decode_session(skb, &fl, family);
2342
2343         /* First, check used SA against their selectors. */
2344         if (skb->sp) {
2345                 int i;
2346
2347                 for (i=skb->sp->len-1; i>=0; i--) {
2348                         struct xfrm_state *x = skb->sp->xvec[i];
2349                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2350                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2351                                 return 0;
2352                         }
2353                 }
2354         }
2355
2356         pol = NULL;
2357         if (sk && sk->sk_policy[dir]) {
2358                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2359                 if (IS_ERR(pol)) {
2360                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2361                         return 0;
2362                 }
2363         }
2364
2365         if (!pol) {
2366                 struct flow_cache_object *flo;
2367
2368                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2369                                         xfrm_policy_lookup, NULL);
2370                 if (IS_ERR_OR_NULL(flo))
2371                         pol = ERR_CAST(flo);
2372                 else
2373                         pol = container_of(flo, struct xfrm_policy, flo);
2374         }
2375
2376         if (IS_ERR(pol)) {
2377                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2378                 return 0;
2379         }
2380
2381         if (!pol) {
2382                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2383                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2384                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2385                         return 0;
2386                 }
2387                 return 1;
2388         }
2389
2390         pol->curlft.use_time = get_seconds();
2391
2392         pols[0] = pol;
2393         npols ++;
2394 #ifdef CONFIG_XFRM_SUB_POLICY
2395         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2396                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2397                                                     &fl, family,
2398                                                     XFRM_POLICY_IN);
2399                 if (pols[1]) {
2400                         if (IS_ERR(pols[1])) {
2401                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2402                                 return 0;
2403                         }
2404                         pols[1]->curlft.use_time = get_seconds();
2405                         npols ++;
2406                 }
2407         }
2408 #endif
2409
2410         if (pol->action == XFRM_POLICY_ALLOW) {
2411                 struct sec_path *sp;
2412                 static struct sec_path dummy;
2413                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2414                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2415                 struct xfrm_tmpl **tpp = tp;
2416                 int ti = 0;
2417                 int i, k;
2418
2419                 if ((sp = skb->sp) == NULL)
2420                         sp = &dummy;
2421
2422                 for (pi = 0; pi < npols; pi++) {
2423                         if (pols[pi] != pol &&
2424                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2425                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2426                                 goto reject;
2427                         }
2428                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2429                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2430                                 goto reject_error;
2431                         }
2432                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2433                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2434                 }
2435                 xfrm_nr = ti;
2436                 if (npols > 1) {
2437                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2438                         tpp = stp;
2439                 }
2440
2441                 /* For each tunnel xfrm, find the first matching tmpl.
2442                  * For each tmpl before that, find corresponding xfrm.
2443                  * Order is _important_. Later we will implement
2444                  * some barriers, but at the moment barriers
2445                  * are implied between each two transformations.
2446                  */
2447                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2448                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2449                         if (k < 0) {
2450                                 if (k < -1)
2451                                         /* "-2 - errored_index" returned */
2452                                         xerr_idx = -(2+k);
2453                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2454                                 goto reject;
2455                         }
2456                 }
2457
2458                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2459                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2460                         goto reject;
2461                 }
2462
2463                 xfrm_pols_put(pols, npols);
2464                 return 1;
2465         }
2466         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2467
2468 reject:
2469         xfrm_secpath_reject(xerr_idx, skb, &fl);
2470 reject_error:
2471         xfrm_pols_put(pols, npols);
2472         return 0;
2473 }
2474 EXPORT_SYMBOL(__xfrm_policy_check);
2475
2476 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2477 {
2478         struct net *net = dev_net(skb->dev);
2479         struct flowi fl;
2480         struct dst_entry *dst;
2481         int res = 1;
2482
2483         if (xfrm_decode_session(skb, &fl, family) < 0) {
2484                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2485                 return 0;
2486         }
2487
2488         skb_dst_force(skb);
2489
2490         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2491         if (IS_ERR(dst)) {
2492                 res = 0;
2493                 dst = NULL;
2494         }
2495         skb_dst_set(skb, dst);
2496         return res;
2497 }
2498 EXPORT_SYMBOL(__xfrm_route_forward);
2499
2500 /* Optimize later using cookies and generation ids. */
2501
2502 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2503 {
2504         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2505          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2506          * get validated by dst_ops->check on every use.  We do this
2507          * because when a normal route referenced by an XFRM dst is
2508          * obsoleted we do not go looking around for all parent
2509          * referencing XFRM dsts so that we can invalidate them.  It
2510          * is just too much work.  Instead we make the checks here on
2511          * every use.  For example:
2512          *
2513          *      XFRM dst A --> IPv4 dst X
2514          *
2515          * X is the "xdst->route" of A (X is also the "dst->path" of A
2516          * in this example).  If X is marked obsolete, "A" will not
2517          * notice.  That's what we are validating here via the
2518          * stale_bundle() check.
2519          *
2520          * When a policy's bundle is pruned, we dst_free() the XFRM
2521          * dst which causes it's ->obsolete field to be set to
2522          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2523          * this, we want to force a new route lookup.
2524          */
2525         if (dst->obsolete < 0 && !stale_bundle(dst))
2526                 return dst;
2527
2528         return NULL;
2529 }
2530
2531 static int stale_bundle(struct dst_entry *dst)
2532 {
2533         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2534 }
2535
2536 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2537 {
2538         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2539                 dst->dev = dev_net(dev)->loopback_dev;
2540                 dev_hold(dst->dev);
2541                 dev_put(dev);
2542         }
2543 }
2544 EXPORT_SYMBOL(xfrm_dst_ifdown);
2545
2546 static void xfrm_link_failure(struct sk_buff *skb)
2547 {
2548         /* Impossible. Such dst must be popped before reaches point of failure. */
2549 }
2550
2551 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2552 {
2553         if (dst) {
2554                 if (dst->obsolete) {
2555                         dst_release(dst);
2556                         dst = NULL;
2557                 }
2558         }
2559         return dst;
2560 }
2561
2562 static void __xfrm_garbage_collect(struct net *net)
2563 {
2564         struct dst_entry *head, *next;
2565
2566         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2567         head = xfrm_policy_sk_bundles;
2568         xfrm_policy_sk_bundles = NULL;
2569         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2570
2571         while (head) {
2572                 next = head->next;
2573                 dst_free(head);
2574                 head = next;
2575         }
2576 }
2577
2578 void xfrm_garbage_collect(struct net *net)
2579 {
2580         flow_cache_flush();
2581         __xfrm_garbage_collect(net);
2582 }
2583 EXPORT_SYMBOL(xfrm_garbage_collect);
2584
2585 static void xfrm_garbage_collect_deferred(struct net *net)
2586 {
2587         flow_cache_flush_deferred();
2588         __xfrm_garbage_collect(net);
2589 }
2590
2591 static void xfrm_init_pmtu(struct dst_entry *dst)
2592 {
2593         do {
2594                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2595                 u32 pmtu, route_mtu_cached;
2596
2597                 pmtu = dst_mtu(dst->child);
2598                 xdst->child_mtu_cached = pmtu;
2599
2600                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2601
2602                 route_mtu_cached = dst_mtu(xdst->route);
2603                 xdst->route_mtu_cached = route_mtu_cached;
2604
2605                 if (pmtu > route_mtu_cached)
2606                         pmtu = route_mtu_cached;
2607
2608                 dst_metric_set(dst, RTAX_MTU, pmtu);
2609         } while ((dst = dst->next));
2610 }
2611
2612 /* Check that the bundle accepts the flow and its components are
2613  * still valid.
2614  */
2615
2616 static int xfrm_bundle_ok(struct xfrm_dst *first)
2617 {
2618         struct dst_entry *dst = &first->u.dst;
2619         struct xfrm_dst *last;
2620         u32 mtu;
2621
2622         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2623             (dst->dev && !netif_running(dst->dev)))
2624                 return 0;
2625
2626         if (dst->flags & DST_XFRM_QUEUE)
2627                 return 1;
2628
2629         last = NULL;
2630
2631         do {
2632                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2633
2634                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2635                         return 0;
2636                 if (xdst->xfrm_genid != dst->xfrm->genid)
2637                         return 0;
2638                 if (xdst->num_pols > 0 &&
2639                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2640                         return 0;
2641
2642                 mtu = dst_mtu(dst->child);
2643                 if (xdst->child_mtu_cached != mtu) {
2644                         last = xdst;
2645                         xdst->child_mtu_cached = mtu;
2646                 }
2647
2648                 if (!dst_check(xdst->route, xdst->route_cookie))
2649                         return 0;
2650                 mtu = dst_mtu(xdst->route);
2651                 if (xdst->route_mtu_cached != mtu) {
2652                         last = xdst;
2653                         xdst->route_mtu_cached = mtu;
2654                 }
2655
2656                 dst = dst->child;
2657         } while (dst->xfrm);
2658
2659         if (likely(!last))
2660                 return 1;
2661
2662         mtu = last->child_mtu_cached;
2663         for (;;) {
2664                 dst = &last->u.dst;
2665
2666                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2667                 if (mtu > last->route_mtu_cached)
2668                         mtu = last->route_mtu_cached;
2669                 dst_metric_set(dst, RTAX_MTU, mtu);
2670
2671                 if (last == first)
2672                         break;
2673
2674                 last = (struct xfrm_dst *)last->u.dst.next;
2675                 last->child_mtu_cached = mtu;
2676         }
2677
2678         return 1;
2679 }
2680
2681 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2682 {
2683         return dst_metric_advmss(dst->path);
2684 }
2685
2686 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2687 {
2688         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2689
2690         return mtu ? : dst_mtu(dst->path);
2691 }
2692
2693 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2694                                            struct sk_buff *skb,
2695                                            const void *daddr)
2696 {
2697         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2698 }
2699
2700 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2701 {
2702         struct net *net;
2703         int err = 0;
2704         if (unlikely(afinfo == NULL))
2705                 return -EINVAL;
2706         if (unlikely(afinfo->family >= NPROTO))
2707                 return -EAFNOSUPPORT;
2708         spin_lock(&xfrm_policy_afinfo_lock);
2709         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2710                 err = -ENOBUFS;
2711         else {
2712                 struct dst_ops *dst_ops = afinfo->dst_ops;
2713                 if (likely(dst_ops->kmem_cachep == NULL))
2714                         dst_ops->kmem_cachep = xfrm_dst_cache;
2715                 if (likely(dst_ops->check == NULL))
2716                         dst_ops->check = xfrm_dst_check;
2717                 if (likely(dst_ops->default_advmss == NULL))
2718                         dst_ops->default_advmss = xfrm_default_advmss;
2719                 if (likely(dst_ops->mtu == NULL))
2720                         dst_ops->mtu = xfrm_mtu;
2721                 if (likely(dst_ops->negative_advice == NULL))
2722                         dst_ops->negative_advice = xfrm_negative_advice;
2723                 if (likely(dst_ops->link_failure == NULL))
2724                         dst_ops->link_failure = xfrm_link_failure;
2725                 if (likely(dst_ops->neigh_lookup == NULL))
2726                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2727                 if (likely(afinfo->garbage_collect == NULL))
2728                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2729                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2730         }
2731         spin_unlock(&xfrm_policy_afinfo_lock);
2732
2733         rtnl_lock();
2734         for_each_net(net) {
2735                 struct dst_ops *xfrm_dst_ops;
2736
2737                 switch (afinfo->family) {
2738                 case AF_INET:
2739                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2740                         break;
2741 #if IS_ENABLED(CONFIG_IPV6)
2742                 case AF_INET6:
2743                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2744                         break;
2745 #endif
2746                 default:
2747                         BUG();
2748                 }
2749                 *xfrm_dst_ops = *afinfo->dst_ops;
2750         }
2751         rtnl_unlock();
2752
2753         return err;
2754 }
2755 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2756
2757 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2758 {
2759         int err = 0;
2760         if (unlikely(afinfo == NULL))
2761                 return -EINVAL;
2762         if (unlikely(afinfo->family >= NPROTO))
2763                 return -EAFNOSUPPORT;
2764         spin_lock(&xfrm_policy_afinfo_lock);
2765         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2766                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2767                         err = -EINVAL;
2768                 else
2769                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2770                                          NULL);
2771         }
2772         spin_unlock(&xfrm_policy_afinfo_lock);
2773         if (!err) {
2774                 struct dst_ops *dst_ops = afinfo->dst_ops;
2775
2776                 synchronize_rcu();
2777
2778                 dst_ops->kmem_cachep = NULL;
2779                 dst_ops->check = NULL;
2780                 dst_ops->negative_advice = NULL;
2781                 dst_ops->link_failure = NULL;
2782                 afinfo->garbage_collect = NULL;
2783         }
2784         return err;
2785 }
2786 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2787
2788 static void __net_init xfrm_dst_ops_init(struct net *net)
2789 {
2790         struct xfrm_policy_afinfo *afinfo;
2791
2792         rcu_read_lock();
2793         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2794         if (afinfo)
2795                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2796 #if IS_ENABLED(CONFIG_IPV6)
2797         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2798         if (afinfo)
2799                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2800 #endif
2801         rcu_read_unlock();
2802 }
2803
2804 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2805 {
2806         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2807
2808         switch (event) {
2809         case NETDEV_DOWN:
2810                 xfrm_garbage_collect(dev_net(dev));
2811         }
2812         return NOTIFY_DONE;
2813 }
2814
2815 static struct notifier_block xfrm_dev_notifier = {
2816         .notifier_call  = xfrm_dev_event,
2817 };
2818
2819 #ifdef CONFIG_XFRM_STATISTICS
2820 static int __net_init xfrm_statistics_init(struct net *net)
2821 {
2822         int rv;
2823
2824         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2825                           sizeof(struct linux_xfrm_mib),
2826                           __alignof__(struct linux_xfrm_mib)) < 0)
2827                 return -ENOMEM;
2828         rv = xfrm_proc_init(net);
2829         if (rv < 0)
2830                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2831         return rv;
2832 }
2833
2834 static void xfrm_statistics_fini(struct net *net)
2835 {
2836         xfrm_proc_fini(net);
2837         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2838 }
2839 #else
2840 static int __net_init xfrm_statistics_init(struct net *net)
2841 {
2842         return 0;
2843 }
2844
2845 static void xfrm_statistics_fini(struct net *net)
2846 {
2847 }
2848 #endif
2849
2850 static int __net_init xfrm_policy_init(struct net *net)
2851 {
2852         unsigned int hmask, sz;
2853         int dir;
2854
2855         if (net_eq(net, &init_net))
2856                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2857                                            sizeof(struct xfrm_dst),
2858                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2859                                            NULL);
2860
2861         hmask = 8 - 1;
2862         sz = (hmask+1) * sizeof(struct hlist_head);
2863
2864         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2865         if (!net->xfrm.policy_byidx)
2866                 goto out_byidx;
2867         net->xfrm.policy_idx_hmask = hmask;
2868
2869         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2870                 struct xfrm_policy_hash *htab;
2871
2872                 net->xfrm.policy_count[dir] = 0;
2873                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2874
2875                 htab = &net->xfrm.policy_bydst[dir];
2876                 htab->table = xfrm_hash_alloc(sz);
2877                 if (!htab->table)
2878                         goto out_bydst;
2879                 htab->hmask = hmask;
2880         }
2881
2882         INIT_LIST_HEAD(&net->xfrm.policy_all);
2883         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2884         if (net_eq(net, &init_net))
2885                 register_netdevice_notifier(&xfrm_dev_notifier);
2886         return 0;
2887
2888 out_bydst:
2889         for (dir--; dir >= 0; dir--) {
2890                 struct xfrm_policy_hash *htab;
2891
2892                 htab = &net->xfrm.policy_bydst[dir];
2893                 xfrm_hash_free(htab->table, sz);
2894         }
2895         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2896 out_byidx:
2897         return -ENOMEM;
2898 }
2899
2900 static void xfrm_policy_fini(struct net *net)
2901 {
2902         struct xfrm_audit audit_info;
2903         unsigned int sz;
2904         int dir;
2905
2906         flush_work(&net->xfrm.policy_hash_work);
2907 #ifdef CONFIG_XFRM_SUB_POLICY
2908         audit_info.loginuid = INVALID_UID;
2909         audit_info.sessionid = -1;
2910         audit_info.secid = 0;
2911         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2912 #endif
2913         audit_info.loginuid = INVALID_UID;
2914         audit_info.sessionid = -1;
2915         audit_info.secid = 0;
2916         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2917
2918         WARN_ON(!list_empty(&net->xfrm.policy_all));
2919
2920         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2921                 struct xfrm_policy_hash *htab;
2922
2923                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2924
2925                 htab = &net->xfrm.policy_bydst[dir];
2926                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2927                 WARN_ON(!hlist_empty(htab->table));
2928                 xfrm_hash_free(htab->table, sz);
2929         }
2930
2931         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2932         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2933         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2934 }
2935
2936 static int __net_init xfrm_net_init(struct net *net)
2937 {
2938         int rv;
2939
2940         rv = xfrm_statistics_init(net);
2941         if (rv < 0)
2942                 goto out_statistics;
2943         rv = xfrm_state_init(net);
2944         if (rv < 0)
2945                 goto out_state;
2946         rv = xfrm_policy_init(net);
2947         if (rv < 0)
2948                 goto out_policy;
2949         xfrm_dst_ops_init(net);
2950         rv = xfrm_sysctl_init(net);
2951         if (rv < 0)
2952                 goto out_sysctl;
2953         return 0;
2954
2955 out_sysctl:
2956         xfrm_policy_fini(net);
2957 out_policy:
2958         xfrm_state_fini(net);
2959 out_state:
2960         xfrm_statistics_fini(net);
2961 out_statistics:
2962         return rv;
2963 }
2964
2965 static void __net_exit xfrm_net_exit(struct net *net)
2966 {
2967         xfrm_sysctl_fini(net);
2968         xfrm_policy_fini(net);
2969         xfrm_state_fini(net);
2970         xfrm_statistics_fini(net);
2971 }
2972
2973 static struct pernet_operations __net_initdata xfrm_net_ops = {
2974         .init = xfrm_net_init,
2975         .exit = xfrm_net_exit,
2976 };
2977
2978 void __init xfrm_init(void)
2979 {
2980         register_pernet_subsys(&xfrm_net_ops);
2981         xfrm_input_init();
2982 }
2983
2984 #ifdef CONFIG_AUDITSYSCALL
2985 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2986                                          struct audit_buffer *audit_buf)
2987 {
2988         struct xfrm_sec_ctx *ctx = xp->security;
2989         struct xfrm_selector *sel = &xp->selector;
2990
2991         if (ctx)
2992                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2993                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2994
2995         switch(sel->family) {
2996         case AF_INET:
2997                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2998                 if (sel->prefixlen_s != 32)
2999                         audit_log_format(audit_buf, " src_prefixlen=%d",
3000                                          sel->prefixlen_s);
3001                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3002                 if (sel->prefixlen_d != 32)
3003                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3004                                          sel->prefixlen_d);
3005                 break;
3006         case AF_INET6:
3007                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3008                 if (sel->prefixlen_s != 128)
3009                         audit_log_format(audit_buf, " src_prefixlen=%d",
3010                                          sel->prefixlen_s);
3011                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3012                 if (sel->prefixlen_d != 128)
3013                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3014                                          sel->prefixlen_d);
3015                 break;
3016         }
3017 }
3018
3019 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
3020                            kuid_t auid, u32 sessionid, u32 secid)
3021 {
3022         struct audit_buffer *audit_buf;
3023
3024         audit_buf = xfrm_audit_start("SPD-add");
3025         if (audit_buf == NULL)
3026                 return;
3027         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3028         audit_log_format(audit_buf, " res=%u", result);
3029         xfrm_audit_common_policyinfo(xp, audit_buf);
3030         audit_log_end(audit_buf);
3031 }
3032 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3033
3034 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3035                               kuid_t auid, u32 sessionid, u32 secid)
3036 {
3037         struct audit_buffer *audit_buf;
3038
3039         audit_buf = xfrm_audit_start("SPD-delete");
3040         if (audit_buf == NULL)
3041                 return;
3042         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3043         audit_log_format(audit_buf, " res=%u", result);
3044         xfrm_audit_common_policyinfo(xp, audit_buf);
3045         audit_log_end(audit_buf);
3046 }
3047 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3048 #endif
3049
3050 #ifdef CONFIG_XFRM_MIGRATE
3051 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3052                                         const struct xfrm_selector *sel_tgt)
3053 {
3054         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3055                 if (sel_tgt->family == sel_cmp->family &&
3056                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3057                                     sel_cmp->family) &&
3058                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3059                                     sel_cmp->family) &&
3060                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3061                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3062                         return true;
3063                 }
3064         } else {
3065                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3066                         return true;
3067                 }
3068         }
3069         return false;
3070 }
3071
3072 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3073                                                      u8 dir, u8 type)
3074 {
3075         struct xfrm_policy *pol, *ret = NULL;
3076         struct hlist_head *chain;
3077         u32 priority = ~0U;
3078
3079         read_lock_bh(&xfrm_policy_lock);
3080         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
3081         hlist_for_each_entry(pol, chain, bydst) {
3082                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3083                     pol->type == type) {
3084                         ret = pol;
3085                         priority = ret->priority;
3086                         break;
3087                 }
3088         }
3089         chain = &init_net.xfrm.policy_inexact[dir];
3090         hlist_for_each_entry(pol, chain, bydst) {
3091                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3092                     pol->type == type &&
3093                     pol->priority < priority) {
3094                         ret = pol;
3095                         break;
3096                 }
3097         }
3098
3099         if (ret)
3100                 xfrm_pol_hold(ret);
3101
3102         read_unlock_bh(&xfrm_policy_lock);
3103
3104         return ret;
3105 }
3106
3107 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3108 {
3109         int match = 0;
3110
3111         if (t->mode == m->mode && t->id.proto == m->proto &&
3112             (m->reqid == 0 || t->reqid == m->reqid)) {
3113                 switch (t->mode) {
3114                 case XFRM_MODE_TUNNEL:
3115                 case XFRM_MODE_BEET:
3116                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3117                                             m->old_family) &&
3118                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3119                                             m->old_family)) {
3120                                 match = 1;
3121                         }
3122                         break;
3123                 case XFRM_MODE_TRANSPORT:
3124                         /* in case of transport mode, template does not store
3125                            any IP addresses, hence we just compare mode and
3126                            protocol */
3127                         match = 1;
3128                         break;
3129                 default:
3130                         break;
3131                 }
3132         }
3133         return match;
3134 }
3135
3136 /* update endpoint address(es) of template(s) */
3137 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3138                                struct xfrm_migrate *m, int num_migrate)
3139 {
3140         struct xfrm_migrate *mp;
3141         int i, j, n = 0;
3142
3143         write_lock_bh(&pol->lock);
3144         if (unlikely(pol->walk.dead)) {
3145                 /* target policy has been deleted */
3146                 write_unlock_bh(&pol->lock);
3147                 return -ENOENT;
3148         }
3149
3150         for (i = 0; i < pol->xfrm_nr; i++) {
3151                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3152                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3153                                 continue;
3154                         n++;
3155                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3156                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3157                                 continue;
3158                         /* update endpoints */
3159                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3160                                sizeof(pol->xfrm_vec[i].id.daddr));
3161                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3162                                sizeof(pol->xfrm_vec[i].saddr));
3163                         pol->xfrm_vec[i].encap_family = mp->new_family;
3164                         /* flush bundles */
3165                         atomic_inc(&pol->genid);
3166                 }
3167         }
3168
3169         write_unlock_bh(&pol->lock);
3170
3171         if (!n)
3172                 return -ENODATA;
3173
3174         return 0;
3175 }
3176
3177 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3178 {
3179         int i, j;
3180
3181         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3182                 return -EINVAL;
3183
3184         for (i = 0; i < num_migrate; i++) {
3185                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3186                                     m[i].old_family) &&
3187                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3188                                     m[i].old_family))
3189                         return -EINVAL;
3190                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3191                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3192                         return -EINVAL;
3193
3194                 /* check if there is any duplicated entry */
3195                 for (j = i + 1; j < num_migrate; j++) {
3196                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3197                                     sizeof(m[i].old_daddr)) &&
3198                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3199                                     sizeof(m[i].old_saddr)) &&
3200                             m[i].proto == m[j].proto &&
3201                             m[i].mode == m[j].mode &&
3202                             m[i].reqid == m[j].reqid &&
3203                             m[i].old_family == m[j].old_family)
3204                                 return -EINVAL;
3205                 }
3206         }
3207
3208         return 0;
3209 }
3210
3211 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3212                  struct xfrm_migrate *m, int num_migrate,
3213                  struct xfrm_kmaddress *k)
3214 {
3215         int i, err, nx_cur = 0, nx_new = 0;
3216         struct xfrm_policy *pol = NULL;
3217         struct xfrm_state *x, *xc;
3218         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3219         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3220         struct xfrm_migrate *mp;
3221
3222         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3223                 goto out;
3224
3225         /* Stage 1 - find policy */
3226         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
3227                 err = -ENOENT;
3228                 goto out;
3229         }
3230
3231         /* Stage 2 - find and update state(s) */
3232         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3233                 if ((x = xfrm_migrate_state_find(mp))) {
3234                         x_cur[nx_cur] = x;
3235                         nx_cur++;
3236                         if ((xc = xfrm_state_migrate(x, mp))) {
3237                                 x_new[nx_new] = xc;
3238                                 nx_new++;
3239                         } else {
3240                                 err = -ENODATA;
3241                                 goto restore_state;
3242                         }
3243                 }
3244         }
3245
3246         /* Stage 3 - update policy */
3247         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3248                 goto restore_state;
3249
3250         /* Stage 4 - delete old state(s) */
3251         if (nx_cur) {
3252                 xfrm_states_put(x_cur, nx_cur);
3253                 xfrm_states_delete(x_cur, nx_cur);
3254         }
3255
3256         /* Stage 5 - announce */
3257         km_migrate(sel, dir, type, m, num_migrate, k);
3258
3259         xfrm_pol_put(pol);
3260
3261         return 0;
3262 out:
3263         return err;
3264
3265 restore_state:
3266         if (pol)
3267                 xfrm_pol_put(pol);
3268         if (nx_cur)
3269                 xfrm_states_put(x_cur, nx_cur);
3270         if (nx_new)
3271                 xfrm_states_delete(x_new, nx_new);
3272
3273         return err;
3274 }
3275 EXPORT_SYMBOL(xfrm_migrate);
3276 #endif