Merge tag 'backlight-for-linus-3.15' of git://git.kernel.org/pub/scm/linux/kernel...
[sfrench/cifs-2.6.git] / net / xfrm / xfrm_policy.c
1 /*
2  * xfrm_policy.c
3  *
4  * Changes:
5  *      Mitsuru KANDA @USAGI
6  *      Kazunori MIYAZAWA @USAGI
7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8  *              IPv6 support
9  *      Kazunori MIYAZAWA @USAGI
10  *      YOSHIFUJI Hideaki
11  *              Split up af-specific portion
12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
13  *
14  */
15
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
28 #include <net/dst.h>
29 #include <net/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 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
43 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
44                                                 __read_mostly;
45
46 static struct kmem_cache *xfrm_dst_cache __read_mostly;
47
48 static void xfrm_init_pmtu(struct dst_entry *dst);
49 static int stale_bundle(struct dst_entry *dst);
50 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
51 static void xfrm_policy_queue_process(unsigned long arg);
52
53 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
54                                                 int dir);
55
56 static inline bool
57 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
58 {
59         const struct flowi4 *fl4 = &fl->u.ip4;
60
61         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
62                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
63                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
64                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
65                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
66                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
67 }
68
69 static inline bool
70 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
71 {
72         const struct flowi6 *fl6 = &fl->u.ip6;
73
74         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
75                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
76                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
77                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
78                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
79                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
80 }
81
82 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
83                          unsigned short family)
84 {
85         switch (family) {
86         case AF_INET:
87                 return __xfrm4_selector_match(sel, fl);
88         case AF_INET6:
89                 return __xfrm6_selector_match(sel, fl);
90         }
91         return false;
92 }
93
94 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
95 {
96         struct xfrm_policy_afinfo *afinfo;
97
98         if (unlikely(family >= NPROTO))
99                 return NULL;
100         rcu_read_lock();
101         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
102         if (unlikely(!afinfo))
103                 rcu_read_unlock();
104         return afinfo;
105 }
106
107 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
108 {
109         rcu_read_unlock();
110 }
111
112 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
113                                                   const xfrm_address_t *saddr,
114                                                   const xfrm_address_t *daddr,
115                                                   int family)
116 {
117         struct xfrm_policy_afinfo *afinfo;
118         struct dst_entry *dst;
119
120         afinfo = xfrm_policy_get_afinfo(family);
121         if (unlikely(afinfo == NULL))
122                 return ERR_PTR(-EAFNOSUPPORT);
123
124         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
125
126         xfrm_policy_put_afinfo(afinfo);
127
128         return dst;
129 }
130
131 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
132                                                 xfrm_address_t *prev_saddr,
133                                                 xfrm_address_t *prev_daddr,
134                                                 int family)
135 {
136         struct net *net = xs_net(x);
137         xfrm_address_t *saddr = &x->props.saddr;
138         xfrm_address_t *daddr = &x->id.daddr;
139         struct dst_entry *dst;
140
141         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
142                 saddr = x->coaddr;
143                 daddr = prev_daddr;
144         }
145         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
146                 saddr = prev_saddr;
147                 daddr = x->coaddr;
148         }
149
150         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
151
152         if (!IS_ERR(dst)) {
153                 if (prev_saddr != saddr)
154                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
155                 if (prev_daddr != daddr)
156                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
157         }
158
159         return dst;
160 }
161
162 static inline unsigned long make_jiffies(long secs)
163 {
164         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
165                 return MAX_SCHEDULE_TIMEOUT-1;
166         else
167                 return secs*HZ;
168 }
169
170 static void xfrm_policy_timer(unsigned long data)
171 {
172         struct xfrm_policy *xp = (struct xfrm_policy *)data;
173         unsigned long now = get_seconds();
174         long next = LONG_MAX;
175         int warn = 0;
176         int dir;
177
178         read_lock(&xp->lock);
179
180         if (unlikely(xp->walk.dead))
181                 goto out;
182
183         dir = xfrm_policy_id2dir(xp->index);
184
185         if (xp->lft.hard_add_expires_seconds) {
186                 long tmo = xp->lft.hard_add_expires_seconds +
187                         xp->curlft.add_time - now;
188                 if (tmo <= 0)
189                         goto expired;
190                 if (tmo < next)
191                         next = tmo;
192         }
193         if (xp->lft.hard_use_expires_seconds) {
194                 long tmo = xp->lft.hard_use_expires_seconds +
195                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
196                 if (tmo <= 0)
197                         goto expired;
198                 if (tmo < next)
199                         next = tmo;
200         }
201         if (xp->lft.soft_add_expires_seconds) {
202                 long tmo = xp->lft.soft_add_expires_seconds +
203                         xp->curlft.add_time - now;
204                 if (tmo <= 0) {
205                         warn = 1;
206                         tmo = XFRM_KM_TIMEOUT;
207                 }
208                 if (tmo < next)
209                         next = tmo;
210         }
211         if (xp->lft.soft_use_expires_seconds) {
212                 long tmo = xp->lft.soft_use_expires_seconds +
213                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
214                 if (tmo <= 0) {
215                         warn = 1;
216                         tmo = XFRM_KM_TIMEOUT;
217                 }
218                 if (tmo < next)
219                         next = tmo;
220         }
221
222         if (warn)
223                 km_policy_expired(xp, dir, 0, 0);
224         if (next != LONG_MAX &&
225             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
226                 xfrm_pol_hold(xp);
227
228 out:
229         read_unlock(&xp->lock);
230         xfrm_pol_put(xp);
231         return;
232
233 expired:
234         read_unlock(&xp->lock);
235         if (!xfrm_policy_delete(xp, dir))
236                 km_policy_expired(xp, dir, 1, 0);
237         xfrm_pol_put(xp);
238 }
239
240 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
241 {
242         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
243
244         if (unlikely(pol->walk.dead))
245                 flo = NULL;
246         else
247                 xfrm_pol_hold(pol);
248
249         return flo;
250 }
251
252 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
253 {
254         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
255
256         return !pol->walk.dead;
257 }
258
259 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
260 {
261         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
262 }
263
264 static const struct flow_cache_ops xfrm_policy_fc_ops = {
265         .get = xfrm_policy_flo_get,
266         .check = xfrm_policy_flo_check,
267         .delete = xfrm_policy_flo_delete,
268 };
269
270 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
271  * SPD calls.
272  */
273
274 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
275 {
276         struct xfrm_policy *policy;
277
278         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
279
280         if (policy) {
281                 write_pnet(&policy->xp_net, net);
282                 INIT_LIST_HEAD(&policy->walk.all);
283                 INIT_HLIST_NODE(&policy->bydst);
284                 INIT_HLIST_NODE(&policy->byidx);
285                 rwlock_init(&policy->lock);
286                 atomic_set(&policy->refcnt, 1);
287                 skb_queue_head_init(&policy->polq.hold_queue);
288                 setup_timer(&policy->timer, xfrm_policy_timer,
289                                 (unsigned long)policy);
290                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
291                             (unsigned long)policy);
292                 policy->flo.ops = &xfrm_policy_fc_ops;
293         }
294         return policy;
295 }
296 EXPORT_SYMBOL(xfrm_policy_alloc);
297
298 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
299
300 void xfrm_policy_destroy(struct xfrm_policy *policy)
301 {
302         BUG_ON(!policy->walk.dead);
303
304         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
305                 BUG();
306
307         security_xfrm_policy_free(policy->security);
308         kfree(policy);
309 }
310 EXPORT_SYMBOL(xfrm_policy_destroy);
311
312 static void xfrm_queue_purge(struct sk_buff_head *list)
313 {
314         struct sk_buff *skb;
315
316         while ((skb = skb_dequeue(list)) != NULL)
317                 kfree_skb(skb);
318 }
319
320 /* Rule must be locked. Release descentant resources, announce
321  * entry dead. The rule must be unlinked from lists to the moment.
322  */
323
324 static void xfrm_policy_kill(struct xfrm_policy *policy)
325 {
326         policy->walk.dead = 1;
327
328         atomic_inc(&policy->genid);
329
330         if (del_timer(&policy->polq.hold_timer))
331                 xfrm_pol_put(policy);
332         xfrm_queue_purge(&policy->polq.hold_queue);
333
334         if (del_timer(&policy->timer))
335                 xfrm_pol_put(policy);
336
337         xfrm_pol_put(policy);
338 }
339
340 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
341
342 static inline unsigned int idx_hash(struct net *net, u32 index)
343 {
344         return __idx_hash(index, net->xfrm.policy_idx_hmask);
345 }
346
347 static struct hlist_head *policy_hash_bysel(struct net *net,
348                                             const struct xfrm_selector *sel,
349                                             unsigned short family, int dir)
350 {
351         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
352         unsigned int hash = __sel_hash(sel, family, hmask);
353
354         return (hash == hmask + 1 ?
355                 &net->xfrm.policy_inexact[dir] :
356                 net->xfrm.policy_bydst[dir].table + hash);
357 }
358
359 static struct hlist_head *policy_hash_direct(struct net *net,
360                                              const xfrm_address_t *daddr,
361                                              const xfrm_address_t *saddr,
362                                              unsigned short family, int dir)
363 {
364         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
365         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
366
367         return net->xfrm.policy_bydst[dir].table + hash;
368 }
369
370 static void xfrm_dst_hash_transfer(struct hlist_head *list,
371                                    struct hlist_head *ndsttable,
372                                    unsigned int nhashmask)
373 {
374         struct hlist_node *tmp, *entry0 = NULL;
375         struct xfrm_policy *pol;
376         unsigned int h0 = 0;
377
378 redo:
379         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
380                 unsigned int h;
381
382                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
383                                 pol->family, nhashmask);
384                 if (!entry0) {
385                         hlist_del(&pol->bydst);
386                         hlist_add_head(&pol->bydst, ndsttable+h);
387                         h0 = h;
388                 } else {
389                         if (h != h0)
390                                 continue;
391                         hlist_del(&pol->bydst);
392                         hlist_add_after(entry0, &pol->bydst);
393                 }
394                 entry0 = &pol->bydst;
395         }
396         if (!hlist_empty(list)) {
397                 entry0 = NULL;
398                 goto redo;
399         }
400 }
401
402 static void xfrm_idx_hash_transfer(struct hlist_head *list,
403                                    struct hlist_head *nidxtable,
404                                    unsigned int nhashmask)
405 {
406         struct hlist_node *tmp;
407         struct xfrm_policy *pol;
408
409         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
410                 unsigned int h;
411
412                 h = __idx_hash(pol->index, nhashmask);
413                 hlist_add_head(&pol->byidx, nidxtable+h);
414         }
415 }
416
417 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
418 {
419         return ((old_hmask + 1) << 1) - 1;
420 }
421
422 static void xfrm_bydst_resize(struct net *net, int dir)
423 {
424         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
425         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
426         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
427         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
428         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
429         int i;
430
431         if (!ndst)
432                 return;
433
434         write_lock_bh(&net->xfrm.xfrm_policy_lock);
435
436         for (i = hmask; i >= 0; i--)
437                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
438
439         net->xfrm.policy_bydst[dir].table = ndst;
440         net->xfrm.policy_bydst[dir].hmask = nhashmask;
441
442         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
443
444         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
445 }
446
447 static void xfrm_byidx_resize(struct net *net, int total)
448 {
449         unsigned int hmask = net->xfrm.policy_idx_hmask;
450         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
451         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
452         struct hlist_head *oidx = net->xfrm.policy_byidx;
453         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
454         int i;
455
456         if (!nidx)
457                 return;
458
459         write_lock_bh(&net->xfrm.xfrm_policy_lock);
460
461         for (i = hmask; i >= 0; i--)
462                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
463
464         net->xfrm.policy_byidx = nidx;
465         net->xfrm.policy_idx_hmask = nhashmask;
466
467         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
468
469         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
470 }
471
472 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
473 {
474         unsigned int cnt = net->xfrm.policy_count[dir];
475         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
476
477         if (total)
478                 *total += cnt;
479
480         if ((hmask + 1) < xfrm_policy_hashmax &&
481             cnt > hmask)
482                 return 1;
483
484         return 0;
485 }
486
487 static inline int xfrm_byidx_should_resize(struct net *net, int total)
488 {
489         unsigned int hmask = net->xfrm.policy_idx_hmask;
490
491         if ((hmask + 1) < xfrm_policy_hashmax &&
492             total > hmask)
493                 return 1;
494
495         return 0;
496 }
497
498 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
499 {
500         read_lock_bh(&net->xfrm.xfrm_policy_lock);
501         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
502         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
503         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
504         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
505         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
506         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
507         si->spdhcnt = net->xfrm.policy_idx_hmask;
508         si->spdhmcnt = xfrm_policy_hashmax;
509         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
510 }
511 EXPORT_SYMBOL(xfrm_spd_getinfo);
512
513 static DEFINE_MUTEX(hash_resize_mutex);
514 static void xfrm_hash_resize(struct work_struct *work)
515 {
516         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
517         int dir, total;
518
519         mutex_lock(&hash_resize_mutex);
520
521         total = 0;
522         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
523                 if (xfrm_bydst_should_resize(net, dir, &total))
524                         xfrm_bydst_resize(net, dir);
525         }
526         if (xfrm_byidx_should_resize(net, total))
527                 xfrm_byidx_resize(net, total);
528
529         mutex_unlock(&hash_resize_mutex);
530 }
531
532 /* Generate new index... KAME seems to generate them ordered by cost
533  * of an absolute inpredictability of ordering of rules. This will not pass. */
534 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
535 {
536         static u32 idx_generator;
537
538         for (;;) {
539                 struct hlist_head *list;
540                 struct xfrm_policy *p;
541                 u32 idx;
542                 int found;
543
544                 if (!index) {
545                         idx = (idx_generator | dir);
546                         idx_generator += 8;
547                 } else {
548                         idx = index;
549                         index = 0;
550                 }
551
552                 if (idx == 0)
553                         idx = 8;
554                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
555                 found = 0;
556                 hlist_for_each_entry(p, list, byidx) {
557                         if (p->index == idx) {
558                                 found = 1;
559                                 break;
560                         }
561                 }
562                 if (!found)
563                         return idx;
564         }
565 }
566
567 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
568 {
569         u32 *p1 = (u32 *) s1;
570         u32 *p2 = (u32 *) s2;
571         int len = sizeof(struct xfrm_selector) / sizeof(u32);
572         int i;
573
574         for (i = 0; i < len; i++) {
575                 if (p1[i] != p2[i])
576                         return 1;
577         }
578
579         return 0;
580 }
581
582 static void xfrm_policy_requeue(struct xfrm_policy *old,
583                                 struct xfrm_policy *new)
584 {
585         struct xfrm_policy_queue *pq = &old->polq;
586         struct sk_buff_head list;
587
588         __skb_queue_head_init(&list);
589
590         spin_lock_bh(&pq->hold_queue.lock);
591         skb_queue_splice_init(&pq->hold_queue, &list);
592         if (del_timer(&pq->hold_timer))
593                 xfrm_pol_put(old);
594         spin_unlock_bh(&pq->hold_queue.lock);
595
596         if (skb_queue_empty(&list))
597                 return;
598
599         pq = &new->polq;
600
601         spin_lock_bh(&pq->hold_queue.lock);
602         skb_queue_splice(&list, &pq->hold_queue);
603         pq->timeout = XFRM_QUEUE_TMO_MIN;
604         if (!mod_timer(&pq->hold_timer, jiffies))
605                 xfrm_pol_hold(new);
606         spin_unlock_bh(&pq->hold_queue.lock);
607 }
608
609 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
610                                    struct xfrm_policy *pol)
611 {
612         u32 mark = policy->mark.v & policy->mark.m;
613
614         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
615                 return true;
616
617         if ((mark & pol->mark.m) == pol->mark.v &&
618             policy->priority == pol->priority)
619                 return true;
620
621         return false;
622 }
623
624 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
625 {
626         struct net *net = xp_net(policy);
627         struct xfrm_policy *pol;
628         struct xfrm_policy *delpol;
629         struct hlist_head *chain;
630         struct hlist_node *newpos;
631
632         write_lock_bh(&net->xfrm.xfrm_policy_lock);
633         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
634         delpol = NULL;
635         newpos = NULL;
636         hlist_for_each_entry(pol, chain, bydst) {
637                 if (pol->type == policy->type &&
638                     !selector_cmp(&pol->selector, &policy->selector) &&
639                     xfrm_policy_mark_match(policy, pol) &&
640                     xfrm_sec_ctx_match(pol->security, policy->security) &&
641                     !WARN_ON(delpol)) {
642                         if (excl) {
643                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
644                                 return -EEXIST;
645                         }
646                         delpol = pol;
647                         if (policy->priority > pol->priority)
648                                 continue;
649                 } else if (policy->priority >= pol->priority) {
650                         newpos = &pol->bydst;
651                         continue;
652                 }
653                 if (delpol)
654                         break;
655         }
656         if (newpos)
657                 hlist_add_after(newpos, &policy->bydst);
658         else
659                 hlist_add_head(&policy->bydst, chain);
660         xfrm_pol_hold(policy);
661         net->xfrm.policy_count[dir]++;
662         atomic_inc(&net->xfrm.flow_cache_genid);
663
664         /* After previous checking, family can either be AF_INET or AF_INET6 */
665         if (policy->family == AF_INET)
666                 rt_genid_bump_ipv4(net);
667         else
668                 rt_genid_bump_ipv6(net);
669
670         if (delpol) {
671                 xfrm_policy_requeue(delpol, policy);
672                 __xfrm_policy_unlink(delpol, dir);
673         }
674         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
675         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
676         policy->curlft.add_time = get_seconds();
677         policy->curlft.use_time = 0;
678         if (!mod_timer(&policy->timer, jiffies + HZ))
679                 xfrm_pol_hold(policy);
680         list_add(&policy->walk.all, &net->xfrm.policy_all);
681         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
682
683         if (delpol)
684                 xfrm_policy_kill(delpol);
685         else if (xfrm_bydst_should_resize(net, dir, NULL))
686                 schedule_work(&net->xfrm.policy_hash_work);
687
688         return 0;
689 }
690 EXPORT_SYMBOL(xfrm_policy_insert);
691
692 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
693                                           int dir, struct xfrm_selector *sel,
694                                           struct xfrm_sec_ctx *ctx, int delete,
695                                           int *err)
696 {
697         struct xfrm_policy *pol, *ret;
698         struct hlist_head *chain;
699
700         *err = 0;
701         write_lock_bh(&net->xfrm.xfrm_policy_lock);
702         chain = policy_hash_bysel(net, sel, sel->family, dir);
703         ret = NULL;
704         hlist_for_each_entry(pol, chain, bydst) {
705                 if (pol->type == type &&
706                     (mark & pol->mark.m) == pol->mark.v &&
707                     !selector_cmp(sel, &pol->selector) &&
708                     xfrm_sec_ctx_match(ctx, pol->security)) {
709                         xfrm_pol_hold(pol);
710                         if (delete) {
711                                 *err = security_xfrm_policy_delete(
712                                                                 pol->security);
713                                 if (*err) {
714                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
715                                         return pol;
716                                 }
717                                 __xfrm_policy_unlink(pol, dir);
718                         }
719                         ret = pol;
720                         break;
721                 }
722         }
723         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
724
725         if (ret && delete)
726                 xfrm_policy_kill(ret);
727         return ret;
728 }
729 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
730
731 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
732                                      int dir, u32 id, int delete, int *err)
733 {
734         struct xfrm_policy *pol, *ret;
735         struct hlist_head *chain;
736
737         *err = -ENOENT;
738         if (xfrm_policy_id2dir(id) != dir)
739                 return NULL;
740
741         *err = 0;
742         write_lock_bh(&net->xfrm.xfrm_policy_lock);
743         chain = net->xfrm.policy_byidx + idx_hash(net, id);
744         ret = NULL;
745         hlist_for_each_entry(pol, chain, byidx) {
746                 if (pol->type == type && pol->index == id &&
747                     (mark & pol->mark.m) == pol->mark.v) {
748                         xfrm_pol_hold(pol);
749                         if (delete) {
750                                 *err = security_xfrm_policy_delete(
751                                                                 pol->security);
752                                 if (*err) {
753                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
754                                         return pol;
755                                 }
756                                 __xfrm_policy_unlink(pol, dir);
757                         }
758                         ret = pol;
759                         break;
760                 }
761         }
762         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
763
764         if (ret && delete)
765                 xfrm_policy_kill(ret);
766         return ret;
767 }
768 EXPORT_SYMBOL(xfrm_policy_byid);
769
770 #ifdef CONFIG_SECURITY_NETWORK_XFRM
771 static inline int
772 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
773 {
774         int dir, err = 0;
775
776         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
777                 struct xfrm_policy *pol;
778                 int i;
779
780                 hlist_for_each_entry(pol,
781                                      &net->xfrm.policy_inexact[dir], bydst) {
782                         if (pol->type != type)
783                                 continue;
784                         err = security_xfrm_policy_delete(pol->security);
785                         if (err) {
786                                 xfrm_audit_policy_delete(pol, 0,
787                                                          audit_info->loginuid,
788                                                          audit_info->sessionid,
789                                                          audit_info->secid);
790                                 return err;
791                         }
792                 }
793                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
794                         hlist_for_each_entry(pol,
795                                              net->xfrm.policy_bydst[dir].table + i,
796                                              bydst) {
797                                 if (pol->type != type)
798                                         continue;
799                                 err = security_xfrm_policy_delete(
800                                                                 pol->security);
801                                 if (err) {
802                                         xfrm_audit_policy_delete(pol, 0,
803                                                         audit_info->loginuid,
804                                                         audit_info->sessionid,
805                                                         audit_info->secid);
806                                         return err;
807                                 }
808                         }
809                 }
810         }
811         return err;
812 }
813 #else
814 static inline int
815 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
816 {
817         return 0;
818 }
819 #endif
820
821 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
822 {
823         int dir, err = 0, cnt = 0;
824
825         write_lock_bh(&net->xfrm.xfrm_policy_lock);
826
827         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
828         if (err)
829                 goto out;
830
831         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
832                 struct xfrm_policy *pol;
833                 int i;
834
835         again1:
836                 hlist_for_each_entry(pol,
837                                      &net->xfrm.policy_inexact[dir], bydst) {
838                         if (pol->type != type)
839                                 continue;
840                         __xfrm_policy_unlink(pol, dir);
841                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
842                         cnt++;
843
844                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
845                                                  audit_info->sessionid,
846                                                  audit_info->secid);
847
848                         xfrm_policy_kill(pol);
849
850                         write_lock_bh(&net->xfrm.xfrm_policy_lock);
851                         goto again1;
852                 }
853
854                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
855         again2:
856                         hlist_for_each_entry(pol,
857                                              net->xfrm.policy_bydst[dir].table + i,
858                                              bydst) {
859                                 if (pol->type != type)
860                                         continue;
861                                 __xfrm_policy_unlink(pol, dir);
862                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
863                                 cnt++;
864
865                                 xfrm_audit_policy_delete(pol, 1,
866                                                          audit_info->loginuid,
867                                                          audit_info->sessionid,
868                                                          audit_info->secid);
869                                 xfrm_policy_kill(pol);
870
871                                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
872                                 goto again2;
873                         }
874                 }
875
876         }
877         if (!cnt)
878                 err = -ESRCH;
879 out:
880         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
881         return err;
882 }
883 EXPORT_SYMBOL(xfrm_policy_flush);
884
885 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
886                      int (*func)(struct xfrm_policy *, int, int, void*),
887                      void *data)
888 {
889         struct xfrm_policy *pol;
890         struct xfrm_policy_walk_entry *x;
891         int error = 0;
892
893         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
894             walk->type != XFRM_POLICY_TYPE_ANY)
895                 return -EINVAL;
896
897         if (list_empty(&walk->walk.all) && walk->seq != 0)
898                 return 0;
899
900         write_lock_bh(&net->xfrm.xfrm_policy_lock);
901         if (list_empty(&walk->walk.all))
902                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
903         else
904                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
905         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
906                 if (x->dead)
907                         continue;
908                 pol = container_of(x, struct xfrm_policy, walk);
909                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
910                     walk->type != pol->type)
911                         continue;
912                 error = func(pol, xfrm_policy_id2dir(pol->index),
913                              walk->seq, data);
914                 if (error) {
915                         list_move_tail(&walk->walk.all, &x->all);
916                         goto out;
917                 }
918                 walk->seq++;
919         }
920         if (walk->seq == 0) {
921                 error = -ENOENT;
922                 goto out;
923         }
924         list_del_init(&walk->walk.all);
925 out:
926         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
927         return error;
928 }
929 EXPORT_SYMBOL(xfrm_policy_walk);
930
931 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
932 {
933         INIT_LIST_HEAD(&walk->walk.all);
934         walk->walk.dead = 1;
935         walk->type = type;
936         walk->seq = 0;
937 }
938 EXPORT_SYMBOL(xfrm_policy_walk_init);
939
940 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
941 {
942         if (list_empty(&walk->walk.all))
943                 return;
944
945         write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
946         list_del(&walk->walk.all);
947         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
948 }
949 EXPORT_SYMBOL(xfrm_policy_walk_done);
950
951 /*
952  * Find policy to apply to this flow.
953  *
954  * Returns 0 if policy found, else an -errno.
955  */
956 static int xfrm_policy_match(const struct xfrm_policy *pol,
957                              const struct flowi *fl,
958                              u8 type, u16 family, int dir)
959 {
960         const struct xfrm_selector *sel = &pol->selector;
961         int ret = -ESRCH;
962         bool match;
963
964         if (pol->family != family ||
965             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
966             pol->type != type)
967                 return ret;
968
969         match = xfrm_selector_match(sel, fl, family);
970         if (match)
971                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
972                                                   dir);
973
974         return ret;
975 }
976
977 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
978                                                      const struct flowi *fl,
979                                                      u16 family, u8 dir)
980 {
981         int err;
982         struct xfrm_policy *pol, *ret;
983         const xfrm_address_t *daddr, *saddr;
984         struct hlist_head *chain;
985         u32 priority = ~0U;
986
987         daddr = xfrm_flowi_daddr(fl, family);
988         saddr = xfrm_flowi_saddr(fl, family);
989         if (unlikely(!daddr || !saddr))
990                 return NULL;
991
992         read_lock_bh(&net->xfrm.xfrm_policy_lock);
993         chain = policy_hash_direct(net, daddr, saddr, family, dir);
994         ret = NULL;
995         hlist_for_each_entry(pol, chain, bydst) {
996                 err = xfrm_policy_match(pol, fl, type, family, dir);
997                 if (err) {
998                         if (err == -ESRCH)
999                                 continue;
1000                         else {
1001                                 ret = ERR_PTR(err);
1002                                 goto fail;
1003                         }
1004                 } else {
1005                         ret = pol;
1006                         priority = ret->priority;
1007                         break;
1008                 }
1009         }
1010         chain = &net->xfrm.policy_inexact[dir];
1011         hlist_for_each_entry(pol, chain, bydst) {
1012                 err = xfrm_policy_match(pol, fl, type, family, dir);
1013                 if (err) {
1014                         if (err == -ESRCH)
1015                                 continue;
1016                         else {
1017                                 ret = ERR_PTR(err);
1018                                 goto fail;
1019                         }
1020                 } else if (pol->priority < priority) {
1021                         ret = pol;
1022                         break;
1023                 }
1024         }
1025         if (ret)
1026                 xfrm_pol_hold(ret);
1027 fail:
1028         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1029
1030         return ret;
1031 }
1032
1033 static struct xfrm_policy *
1034 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1035 {
1036 #ifdef CONFIG_XFRM_SUB_POLICY
1037         struct xfrm_policy *pol;
1038
1039         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1040         if (pol != NULL)
1041                 return pol;
1042 #endif
1043         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1044 }
1045
1046 static int flow_to_policy_dir(int dir)
1047 {
1048         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1049             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1050             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1051                 return dir;
1052
1053         switch (dir) {
1054         default:
1055         case FLOW_DIR_IN:
1056                 return XFRM_POLICY_IN;
1057         case FLOW_DIR_OUT:
1058                 return XFRM_POLICY_OUT;
1059         case FLOW_DIR_FWD:
1060                 return XFRM_POLICY_FWD;
1061         }
1062 }
1063
1064 static struct flow_cache_object *
1065 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1066                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1067 {
1068         struct xfrm_policy *pol;
1069
1070         if (old_obj)
1071                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1072
1073         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1074         if (IS_ERR_OR_NULL(pol))
1075                 return ERR_CAST(pol);
1076
1077         /* Resolver returns two references:
1078          * one for cache and one for caller of flow_cache_lookup() */
1079         xfrm_pol_hold(pol);
1080
1081         return &pol->flo;
1082 }
1083
1084 static inline int policy_to_flow_dir(int dir)
1085 {
1086         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1087             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1088             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1089                 return dir;
1090         switch (dir) {
1091         default:
1092         case XFRM_POLICY_IN:
1093                 return FLOW_DIR_IN;
1094         case XFRM_POLICY_OUT:
1095                 return FLOW_DIR_OUT;
1096         case XFRM_POLICY_FWD:
1097                 return FLOW_DIR_FWD;
1098         }
1099 }
1100
1101 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1102                                                  const struct flowi *fl)
1103 {
1104         struct xfrm_policy *pol;
1105         struct net *net = sock_net(sk);
1106
1107         read_lock_bh(&net->xfrm.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(&net->xfrm.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_init(&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         struct net *net = xp_net(pol);
1170
1171         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1172         pol = __xfrm_policy_unlink(pol, dir);
1173         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1174         if (pol) {
1175                 xfrm_policy_kill(pol);
1176                 return 0;
1177         }
1178         return -ENOENT;
1179 }
1180 EXPORT_SYMBOL(xfrm_policy_delete);
1181
1182 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1183 {
1184         struct net *net = xp_net(pol);
1185         struct xfrm_policy *old_pol;
1186
1187 #ifdef CONFIG_XFRM_SUB_POLICY
1188         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1189                 return -EINVAL;
1190 #endif
1191
1192         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1193         old_pol = sk->sk_policy[dir];
1194         sk->sk_policy[dir] = pol;
1195         if (pol) {
1196                 pol->curlft.add_time = get_seconds();
1197                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1198                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1199         }
1200         if (old_pol) {
1201                 if (pol)
1202                         xfrm_policy_requeue(old_pol, pol);
1203
1204                 /* Unlinking succeeds always. This is the only function
1205                  * allowed to delete or replace socket policy.
1206                  */
1207                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1208         }
1209         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1210
1211         if (old_pol) {
1212                 xfrm_policy_kill(old_pol);
1213         }
1214         return 0;
1215 }
1216
1217 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1218 {
1219         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1220         struct net *net = xp_net(old);
1221
1222         if (newp) {
1223                 newp->selector = old->selector;
1224                 if (security_xfrm_policy_clone(old->security,
1225                                                &newp->security)) {
1226                         kfree(newp);
1227                         return NULL;  /* ENOMEM */
1228                 }
1229                 newp->lft = old->lft;
1230                 newp->curlft = old->curlft;
1231                 newp->mark = old->mark;
1232                 newp->action = old->action;
1233                 newp->flags = old->flags;
1234                 newp->xfrm_nr = old->xfrm_nr;
1235                 newp->index = old->index;
1236                 newp->type = old->type;
1237                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1238                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1239                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1240                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1241                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1242                 xfrm_pol_put(newp);
1243         }
1244         return newp;
1245 }
1246
1247 int __xfrm_sk_clone_policy(struct sock *sk)
1248 {
1249         struct xfrm_policy *p0 = sk->sk_policy[0],
1250                            *p1 = sk->sk_policy[1];
1251
1252         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1253         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1254                 return -ENOMEM;
1255         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1256                 return -ENOMEM;
1257         return 0;
1258 }
1259
1260 static int
1261 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1262                unsigned short family)
1263 {
1264         int err;
1265         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1266
1267         if (unlikely(afinfo == NULL))
1268                 return -EINVAL;
1269         err = afinfo->get_saddr(net, local, remote);
1270         xfrm_policy_put_afinfo(afinfo);
1271         return err;
1272 }
1273
1274 /* Resolve list of templates for the flow, given policy. */
1275
1276 static int
1277 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1278                       struct xfrm_state **xfrm, unsigned short family)
1279 {
1280         struct net *net = xp_net(policy);
1281         int nx;
1282         int i, error;
1283         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1284         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1285         xfrm_address_t tmp;
1286
1287         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1288                 struct xfrm_state *x;
1289                 xfrm_address_t *remote = daddr;
1290                 xfrm_address_t *local  = saddr;
1291                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1292
1293                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1294                     tmpl->mode == XFRM_MODE_BEET) {
1295                         remote = &tmpl->id.daddr;
1296                         local = &tmpl->saddr;
1297                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1298                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1299                                 if (error)
1300                                         goto fail;
1301                                 local = &tmp;
1302                         }
1303                 }
1304
1305                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1306
1307                 if (x && x->km.state == XFRM_STATE_VALID) {
1308                         xfrm[nx++] = x;
1309                         daddr = remote;
1310                         saddr = local;
1311                         continue;
1312                 }
1313                 if (x) {
1314                         error = (x->km.state == XFRM_STATE_ERROR ?
1315                                  -EINVAL : -EAGAIN);
1316                         xfrm_state_put(x);
1317                 } else if (error == -ESRCH) {
1318                         error = -EAGAIN;
1319                 }
1320
1321                 if (!tmpl->optional)
1322                         goto fail;
1323         }
1324         return nx;
1325
1326 fail:
1327         for (nx--; nx >= 0; nx--)
1328                 xfrm_state_put(xfrm[nx]);
1329         return error;
1330 }
1331
1332 static int
1333 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1334                   struct xfrm_state **xfrm, unsigned short family)
1335 {
1336         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1337         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1338         int cnx = 0;
1339         int error;
1340         int ret;
1341         int i;
1342
1343         for (i = 0; i < npols; i++) {
1344                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1345                         error = -ENOBUFS;
1346                         goto fail;
1347                 }
1348
1349                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1350                 if (ret < 0) {
1351                         error = ret;
1352                         goto fail;
1353                 } else
1354                         cnx += ret;
1355         }
1356
1357         /* found states are sorted for outbound processing */
1358         if (npols > 1)
1359                 xfrm_state_sort(xfrm, tpp, cnx, family);
1360
1361         return cnx;
1362
1363  fail:
1364         for (cnx--; cnx >= 0; cnx--)
1365                 xfrm_state_put(tpp[cnx]);
1366         return error;
1367
1368 }
1369
1370 /* Check that the bundle accepts the flow and its components are
1371  * still valid.
1372  */
1373
1374 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1375 {
1376         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1377         int tos;
1378
1379         if (!afinfo)
1380                 return -EINVAL;
1381
1382         tos = afinfo->get_tos(fl);
1383
1384         xfrm_policy_put_afinfo(afinfo);
1385
1386         return tos;
1387 }
1388
1389 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1390 {
1391         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1392         struct dst_entry *dst = &xdst->u.dst;
1393
1394         if (xdst->route == NULL) {
1395                 /* Dummy bundle - if it has xfrms we were not
1396                  * able to build bundle as template resolution failed.
1397                  * It means we need to try again resolving. */
1398                 if (xdst->num_xfrms > 0)
1399                         return NULL;
1400         } else if (dst->flags & DST_XFRM_QUEUE) {
1401                 return NULL;
1402         } else {
1403                 /* Real bundle */
1404                 if (stale_bundle(dst))
1405                         return NULL;
1406         }
1407
1408         dst_hold(dst);
1409         return flo;
1410 }
1411
1412 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1413 {
1414         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1415         struct dst_entry *dst = &xdst->u.dst;
1416
1417         if (!xdst->route)
1418                 return 0;
1419         if (stale_bundle(dst))
1420                 return 0;
1421
1422         return 1;
1423 }
1424
1425 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1426 {
1427         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1428         struct dst_entry *dst = &xdst->u.dst;
1429
1430         dst_free(dst);
1431 }
1432
1433 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1434         .get = xfrm_bundle_flo_get,
1435         .check = xfrm_bundle_flo_check,
1436         .delete = xfrm_bundle_flo_delete,
1437 };
1438
1439 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1440 {
1441         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1442         struct dst_ops *dst_ops;
1443         struct xfrm_dst *xdst;
1444
1445         if (!afinfo)
1446                 return ERR_PTR(-EINVAL);
1447
1448         switch (family) {
1449         case AF_INET:
1450                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1451                 break;
1452 #if IS_ENABLED(CONFIG_IPV6)
1453         case AF_INET6:
1454                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1455                 break;
1456 #endif
1457         default:
1458                 BUG();
1459         }
1460         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1461
1462         if (likely(xdst)) {
1463                 struct dst_entry *dst = &xdst->u.dst;
1464
1465                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1466                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1467                 if (afinfo->init_dst)
1468                         afinfo->init_dst(net, xdst);
1469         } else
1470                 xdst = ERR_PTR(-ENOBUFS);
1471
1472         xfrm_policy_put_afinfo(afinfo);
1473
1474         return xdst;
1475 }
1476
1477 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1478                                  int nfheader_len)
1479 {
1480         struct xfrm_policy_afinfo *afinfo =
1481                 xfrm_policy_get_afinfo(dst->ops->family);
1482         int err;
1483
1484         if (!afinfo)
1485                 return -EINVAL;
1486
1487         err = afinfo->init_path(path, dst, nfheader_len);
1488
1489         xfrm_policy_put_afinfo(afinfo);
1490
1491         return err;
1492 }
1493
1494 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1495                                 const struct flowi *fl)
1496 {
1497         struct xfrm_policy_afinfo *afinfo =
1498                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1499         int err;
1500
1501         if (!afinfo)
1502                 return -EINVAL;
1503
1504         err = afinfo->fill_dst(xdst, dev, fl);
1505
1506         xfrm_policy_put_afinfo(afinfo);
1507
1508         return err;
1509 }
1510
1511
1512 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1513  * all the metrics... Shortly, bundle a bundle.
1514  */
1515
1516 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1517                                             struct xfrm_state **xfrm, int nx,
1518                                             const struct flowi *fl,
1519                                             struct dst_entry *dst)
1520 {
1521         struct net *net = xp_net(policy);
1522         unsigned long now = jiffies;
1523         struct net_device *dev;
1524         struct xfrm_mode *inner_mode;
1525         struct dst_entry *dst_prev = NULL;
1526         struct dst_entry *dst0 = NULL;
1527         int i = 0;
1528         int err;
1529         int header_len = 0;
1530         int nfheader_len = 0;
1531         int trailer_len = 0;
1532         int tos;
1533         int family = policy->selector.family;
1534         xfrm_address_t saddr, daddr;
1535
1536         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1537
1538         tos = xfrm_get_tos(fl, family);
1539         err = tos;
1540         if (tos < 0)
1541                 goto put_states;
1542
1543         dst_hold(dst);
1544
1545         for (; i < nx; i++) {
1546                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1547                 struct dst_entry *dst1 = &xdst->u.dst;
1548
1549                 err = PTR_ERR(xdst);
1550                 if (IS_ERR(xdst)) {
1551                         dst_release(dst);
1552                         goto put_states;
1553                 }
1554
1555                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1556                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1557                                                         xfrm_af2proto(family));
1558                         if (!inner_mode) {
1559                                 err = -EAFNOSUPPORT;
1560                                 dst_release(dst);
1561                                 goto put_states;
1562                         }
1563                 } else
1564                         inner_mode = xfrm[i]->inner_mode;
1565
1566                 if (!dst_prev)
1567                         dst0 = dst1;
1568                 else {
1569                         dst_prev->child = dst_clone(dst1);
1570                         dst1->flags |= DST_NOHASH;
1571                 }
1572
1573                 xdst->route = dst;
1574                 dst_copy_metrics(dst1, dst);
1575
1576                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1577                         family = xfrm[i]->props.family;
1578                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1579                                               family);
1580                         err = PTR_ERR(dst);
1581                         if (IS_ERR(dst))
1582                                 goto put_states;
1583                 } else
1584                         dst_hold(dst);
1585
1586                 dst1->xfrm = xfrm[i];
1587                 xdst->xfrm_genid = xfrm[i]->genid;
1588
1589                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1590                 dst1->flags |= DST_HOST;
1591                 dst1->lastuse = now;
1592
1593                 dst1->input = dst_discard;
1594                 dst1->output = inner_mode->afinfo->output;
1595
1596                 dst1->next = dst_prev;
1597                 dst_prev = dst1;
1598
1599                 header_len += xfrm[i]->props.header_len;
1600                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1601                         nfheader_len += xfrm[i]->props.header_len;
1602                 trailer_len += xfrm[i]->props.trailer_len;
1603         }
1604
1605         dst_prev->child = dst;
1606         dst0->path = dst;
1607
1608         err = -ENODEV;
1609         dev = dst->dev;
1610         if (!dev)
1611                 goto free_dst;
1612
1613         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1614         xfrm_init_pmtu(dst_prev);
1615
1616         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1617                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1618
1619                 err = xfrm_fill_dst(xdst, dev, fl);
1620                 if (err)
1621                         goto free_dst;
1622
1623                 dst_prev->header_len = header_len;
1624                 dst_prev->trailer_len = trailer_len;
1625                 header_len -= xdst->u.dst.xfrm->props.header_len;
1626                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1627         }
1628
1629 out:
1630         return dst0;
1631
1632 put_states:
1633         for (; i < nx; i++)
1634                 xfrm_state_put(xfrm[i]);
1635 free_dst:
1636         if (dst0)
1637                 dst_free(dst0);
1638         dst0 = ERR_PTR(err);
1639         goto out;
1640 }
1641
1642 #ifdef CONFIG_XFRM_SUB_POLICY
1643 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1644 {
1645         if (!*target) {
1646                 *target = kmalloc(size, GFP_ATOMIC);
1647                 if (!*target)
1648                         return -ENOMEM;
1649         }
1650
1651         memcpy(*target, src, size);
1652         return 0;
1653 }
1654 #endif
1655
1656 static int xfrm_dst_update_parent(struct dst_entry *dst,
1657                                   const struct xfrm_selector *sel)
1658 {
1659 #ifdef CONFIG_XFRM_SUB_POLICY
1660         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1661         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1662                                    sel, sizeof(*sel));
1663 #else
1664         return 0;
1665 #endif
1666 }
1667
1668 static int xfrm_dst_update_origin(struct dst_entry *dst,
1669                                   const struct flowi *fl)
1670 {
1671 #ifdef CONFIG_XFRM_SUB_POLICY
1672         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1673         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1674 #else
1675         return 0;
1676 #endif
1677 }
1678
1679 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1680                                 struct xfrm_policy **pols,
1681                                 int *num_pols, int *num_xfrms)
1682 {
1683         int i;
1684
1685         if (*num_pols == 0 || !pols[0]) {
1686                 *num_pols = 0;
1687                 *num_xfrms = 0;
1688                 return 0;
1689         }
1690         if (IS_ERR(pols[0]))
1691                 return PTR_ERR(pols[0]);
1692
1693         *num_xfrms = pols[0]->xfrm_nr;
1694
1695 #ifdef CONFIG_XFRM_SUB_POLICY
1696         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1697             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1698                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1699                                                     XFRM_POLICY_TYPE_MAIN,
1700                                                     fl, family,
1701                                                     XFRM_POLICY_OUT);
1702                 if (pols[1]) {
1703                         if (IS_ERR(pols[1])) {
1704                                 xfrm_pols_put(pols, *num_pols);
1705                                 return PTR_ERR(pols[1]);
1706                         }
1707                         (*num_pols)++;
1708                         (*num_xfrms) += pols[1]->xfrm_nr;
1709                 }
1710         }
1711 #endif
1712         for (i = 0; i < *num_pols; i++) {
1713                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1714                         *num_xfrms = -1;
1715                         break;
1716                 }
1717         }
1718
1719         return 0;
1720
1721 }
1722
1723 static struct xfrm_dst *
1724 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1725                                const struct flowi *fl, u16 family,
1726                                struct dst_entry *dst_orig)
1727 {
1728         struct net *net = xp_net(pols[0]);
1729         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1730         struct dst_entry *dst;
1731         struct xfrm_dst *xdst;
1732         int err;
1733
1734         /* Try to instantiate a bundle */
1735         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1736         if (err <= 0) {
1737                 if (err != 0 && err != -EAGAIN)
1738                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1739                 return ERR_PTR(err);
1740         }
1741
1742         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1743         if (IS_ERR(dst)) {
1744                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1745                 return ERR_CAST(dst);
1746         }
1747
1748         xdst = (struct xfrm_dst *)dst;
1749         xdst->num_xfrms = err;
1750         if (num_pols > 1)
1751                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1752         else
1753                 err = xfrm_dst_update_origin(dst, fl);
1754         if (unlikely(err)) {
1755                 dst_free(dst);
1756                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1757                 return ERR_PTR(err);
1758         }
1759
1760         xdst->num_pols = num_pols;
1761         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1762         xdst->policy_genid = atomic_read(&pols[0]->genid);
1763
1764         return xdst;
1765 }
1766
1767 static void xfrm_policy_queue_process(unsigned long arg)
1768 {
1769         int err = 0;
1770         struct sk_buff *skb;
1771         struct sock *sk;
1772         struct dst_entry *dst;
1773         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1774         struct xfrm_policy_queue *pq = &pol->polq;
1775         struct flowi fl;
1776         struct sk_buff_head list;
1777
1778         spin_lock(&pq->hold_queue.lock);
1779         skb = skb_peek(&pq->hold_queue);
1780         if (!skb) {
1781                 spin_unlock(&pq->hold_queue.lock);
1782                 goto out;
1783         }
1784         dst = skb_dst(skb);
1785         sk = skb->sk;
1786         xfrm_decode_session(skb, &fl, dst->ops->family);
1787         spin_unlock(&pq->hold_queue.lock);
1788
1789         dst_hold(dst->path);
1790         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1791                           sk, 0);
1792         if (IS_ERR(dst))
1793                 goto purge_queue;
1794
1795         if (dst->flags & DST_XFRM_QUEUE) {
1796                 dst_release(dst);
1797
1798                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1799                         goto purge_queue;
1800
1801                 pq->timeout = pq->timeout << 1;
1802                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1803                         xfrm_pol_hold(pol);
1804         goto out;
1805         }
1806
1807         dst_release(dst);
1808
1809         __skb_queue_head_init(&list);
1810
1811         spin_lock(&pq->hold_queue.lock);
1812         pq->timeout = 0;
1813         skb_queue_splice_init(&pq->hold_queue, &list);
1814         spin_unlock(&pq->hold_queue.lock);
1815
1816         while (!skb_queue_empty(&list)) {
1817                 skb = __skb_dequeue(&list);
1818
1819                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1820                 dst_hold(skb_dst(skb)->path);
1821                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1822                                   &fl, skb->sk, 0);
1823                 if (IS_ERR(dst)) {
1824                         kfree_skb(skb);
1825                         continue;
1826                 }
1827
1828                 nf_reset(skb);
1829                 skb_dst_drop(skb);
1830                 skb_dst_set(skb, dst);
1831
1832                 err = dst_output(skb);
1833         }
1834
1835 out:
1836         xfrm_pol_put(pol);
1837         return;
1838
1839 purge_queue:
1840         pq->timeout = 0;
1841         xfrm_queue_purge(&pq->hold_queue);
1842         xfrm_pol_put(pol);
1843 }
1844
1845 static int xdst_queue_output(struct sk_buff *skb)
1846 {
1847         unsigned long sched_next;
1848         struct dst_entry *dst = skb_dst(skb);
1849         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1850         struct xfrm_policy *pol = xdst->pols[0];
1851         struct xfrm_policy_queue *pq = &pol->polq;
1852         const struct sk_buff *fclone = skb + 1;
1853
1854         if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
1855                      fclone->fclone == SKB_FCLONE_CLONE)) {
1856                 kfree_skb(skb);
1857                 return 0;
1858         }
1859
1860         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1861                 kfree_skb(skb);
1862                 return -EAGAIN;
1863         }
1864
1865         skb_dst_force(skb);
1866
1867         spin_lock_bh(&pq->hold_queue.lock);
1868
1869         if (!pq->timeout)
1870                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1871
1872         sched_next = jiffies + pq->timeout;
1873
1874         if (del_timer(&pq->hold_timer)) {
1875                 if (time_before(pq->hold_timer.expires, sched_next))
1876                         sched_next = pq->hold_timer.expires;
1877                 xfrm_pol_put(pol);
1878         }
1879
1880         __skb_queue_tail(&pq->hold_queue, skb);
1881         if (!mod_timer(&pq->hold_timer, sched_next))
1882                 xfrm_pol_hold(pol);
1883
1884         spin_unlock_bh(&pq->hold_queue.lock);
1885
1886         return 0;
1887 }
1888
1889 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1890                                                  struct dst_entry *dst,
1891                                                  const struct flowi *fl,
1892                                                  int num_xfrms,
1893                                                  u16 family)
1894 {
1895         int err;
1896         struct net_device *dev;
1897         struct dst_entry *dst1;
1898         struct xfrm_dst *xdst;
1899
1900         xdst = xfrm_alloc_dst(net, family);
1901         if (IS_ERR(xdst))
1902                 return xdst;
1903
1904         if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0)
1905                 return xdst;
1906
1907         dst1 = &xdst->u.dst;
1908         dst_hold(dst);
1909         xdst->route = dst;
1910
1911         dst_copy_metrics(dst1, dst);
1912
1913         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1914         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1915         dst1->lastuse = jiffies;
1916
1917         dst1->input = dst_discard;
1918         dst1->output = xdst_queue_output;
1919
1920         dst_hold(dst);
1921         dst1->child = dst;
1922         dst1->path = dst;
1923
1924         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1925
1926         err = -ENODEV;
1927         dev = dst->dev;
1928         if (!dev)
1929                 goto free_dst;
1930
1931         err = xfrm_fill_dst(xdst, dev, fl);
1932         if (err)
1933                 goto free_dst;
1934
1935 out:
1936         return xdst;
1937
1938 free_dst:
1939         dst_release(dst1);
1940         xdst = ERR_PTR(err);
1941         goto out;
1942 }
1943
1944 static struct flow_cache_object *
1945 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1946                    struct flow_cache_object *oldflo, void *ctx)
1947 {
1948         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1949         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1950         struct xfrm_dst *xdst, *new_xdst;
1951         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1952
1953         /* Check if the policies from old bundle are usable */
1954         xdst = NULL;
1955         if (oldflo) {
1956                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1957                 num_pols = xdst->num_pols;
1958                 num_xfrms = xdst->num_xfrms;
1959                 pol_dead = 0;
1960                 for (i = 0; i < num_pols; i++) {
1961                         pols[i] = xdst->pols[i];
1962                         pol_dead |= pols[i]->walk.dead;
1963                 }
1964                 if (pol_dead) {
1965                         dst_free(&xdst->u.dst);
1966                         xdst = NULL;
1967                         num_pols = 0;
1968                         num_xfrms = 0;
1969                         oldflo = NULL;
1970                 }
1971         }
1972
1973         /* Resolve policies to use if we couldn't get them from
1974          * previous cache entry */
1975         if (xdst == NULL) {
1976                 num_pols = 1;
1977                 pols[0] = __xfrm_policy_lookup(net, fl, family,
1978                                                flow_to_policy_dir(dir));
1979                 err = xfrm_expand_policies(fl, family, pols,
1980                                            &num_pols, &num_xfrms);
1981                 if (err < 0)
1982                         goto inc_error;
1983                 if (num_pols == 0)
1984                         return NULL;
1985                 if (num_xfrms <= 0)
1986                         goto make_dummy_bundle;
1987         }
1988
1989         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1990         if (IS_ERR(new_xdst)) {
1991                 err = PTR_ERR(new_xdst);
1992                 if (err != -EAGAIN)
1993                         goto error;
1994                 if (oldflo == NULL)
1995                         goto make_dummy_bundle;
1996                 dst_hold(&xdst->u.dst);
1997                 return oldflo;
1998         } else if (new_xdst == NULL) {
1999                 num_xfrms = 0;
2000                 if (oldflo == NULL)
2001                         goto make_dummy_bundle;
2002                 xdst->num_xfrms = 0;
2003                 dst_hold(&xdst->u.dst);
2004                 return oldflo;
2005         }
2006
2007         /* Kill the previous bundle */
2008         if (xdst) {
2009                 /* The policies were stolen for newly generated bundle */
2010                 xdst->num_pols = 0;
2011                 dst_free(&xdst->u.dst);
2012         }
2013
2014         /* Flow cache does not have reference, it dst_free()'s,
2015          * but we do need to return one reference for original caller */
2016         dst_hold(&new_xdst->u.dst);
2017         return &new_xdst->flo;
2018
2019 make_dummy_bundle:
2020         /* We found policies, but there's no bundles to instantiate:
2021          * either because the policy blocks, has no transformations or
2022          * we could not build template (no xfrm_states).*/
2023         xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
2024         if (IS_ERR(xdst)) {
2025                 xfrm_pols_put(pols, num_pols);
2026                 return ERR_CAST(xdst);
2027         }
2028         xdst->num_pols = num_pols;
2029         xdst->num_xfrms = num_xfrms;
2030         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2031
2032         dst_hold(&xdst->u.dst);
2033         return &xdst->flo;
2034
2035 inc_error:
2036         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2037 error:
2038         if (xdst != NULL)
2039                 dst_free(&xdst->u.dst);
2040         else
2041                 xfrm_pols_put(pols, num_pols);
2042         return ERR_PTR(err);
2043 }
2044
2045 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2046                                         struct dst_entry *dst_orig)
2047 {
2048         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2049         struct dst_entry *ret;
2050
2051         if (!afinfo) {
2052                 dst_release(dst_orig);
2053                 return ERR_PTR(-EINVAL);
2054         } else {
2055                 ret = afinfo->blackhole_route(net, dst_orig);
2056         }
2057         xfrm_policy_put_afinfo(afinfo);
2058
2059         return ret;
2060 }
2061
2062 /* Main function: finds/creates a bundle for given flow.
2063  *
2064  * At the moment we eat a raw IP route. Mostly to speed up lookups
2065  * on interfaces with disabled IPsec.
2066  */
2067 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2068                               const struct flowi *fl,
2069                               struct sock *sk, int flags)
2070 {
2071         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2072         struct flow_cache_object *flo;
2073         struct xfrm_dst *xdst;
2074         struct dst_entry *dst, *route;
2075         u16 family = dst_orig->ops->family;
2076         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2077         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2078
2079         dst = NULL;
2080         xdst = NULL;
2081         route = NULL;
2082
2083         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2084                 num_pols = 1;
2085                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2086                 err = xfrm_expand_policies(fl, family, pols,
2087                                            &num_pols, &num_xfrms);
2088                 if (err < 0)
2089                         goto dropdst;
2090
2091                 if (num_pols) {
2092                         if (num_xfrms <= 0) {
2093                                 drop_pols = num_pols;
2094                                 goto no_transform;
2095                         }
2096
2097                         xdst = xfrm_resolve_and_create_bundle(
2098                                         pols, num_pols, fl,
2099                                         family, dst_orig);
2100                         if (IS_ERR(xdst)) {
2101                                 xfrm_pols_put(pols, num_pols);
2102                                 err = PTR_ERR(xdst);
2103                                 goto dropdst;
2104                         } else if (xdst == NULL) {
2105                                 num_xfrms = 0;
2106                                 drop_pols = num_pols;
2107                                 goto no_transform;
2108                         }
2109
2110                         route = xdst->route;
2111                 }
2112         }
2113
2114         if (xdst == NULL) {
2115                 /* To accelerate a bit...  */
2116                 if ((dst_orig->flags & DST_NOXFRM) ||
2117                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2118                         goto nopol;
2119
2120                 flo = flow_cache_lookup(net, fl, family, dir,
2121                                         xfrm_bundle_lookup, dst_orig);
2122                 if (flo == NULL)
2123                         goto nopol;
2124                 if (IS_ERR(flo)) {
2125                         err = PTR_ERR(flo);
2126                         goto dropdst;
2127                 }
2128                 xdst = container_of(flo, struct xfrm_dst, flo);
2129
2130                 num_pols = xdst->num_pols;
2131                 num_xfrms = xdst->num_xfrms;
2132                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2133                 route = xdst->route;
2134         }
2135
2136         dst = &xdst->u.dst;
2137         if (route == NULL && num_xfrms > 0) {
2138                 /* The only case when xfrm_bundle_lookup() returns a
2139                  * bundle with null route, is when the template could
2140                  * not be resolved. It means policies are there, but
2141                  * bundle could not be created, since we don't yet
2142                  * have the xfrm_state's. We need to wait for KM to
2143                  * negotiate new SA's or bail out with error.*/
2144                 if (net->xfrm.sysctl_larval_drop) {
2145                         dst_release(dst);
2146                         xfrm_pols_put(pols, drop_pols);
2147                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2148
2149                         return make_blackhole(net, family, dst_orig);
2150                 }
2151
2152                 err = -EAGAIN;
2153
2154                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2155                 goto error;
2156         }
2157
2158 no_transform:
2159         if (num_pols == 0)
2160                 goto nopol;
2161
2162         if ((flags & XFRM_LOOKUP_ICMP) &&
2163             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2164                 err = -ENOENT;
2165                 goto error;
2166         }
2167
2168         for (i = 0; i < num_pols; i++)
2169                 pols[i]->curlft.use_time = get_seconds();
2170
2171         if (num_xfrms < 0) {
2172                 /* Prohibit the flow */
2173                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2174                 err = -EPERM;
2175                 goto error;
2176         } else if (num_xfrms > 0) {
2177                 /* Flow transformed */
2178                 dst_release(dst_orig);
2179         } else {
2180                 /* Flow passes untransformed */
2181                 dst_release(dst);
2182                 dst = dst_orig;
2183         }
2184 ok:
2185         xfrm_pols_put(pols, drop_pols);
2186         if (dst && dst->xfrm &&
2187             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2188                 dst->flags |= DST_XFRM_TUNNEL;
2189         return dst;
2190
2191 nopol:
2192         if (!(flags & XFRM_LOOKUP_ICMP)) {
2193                 dst = dst_orig;
2194                 goto ok;
2195         }
2196         err = -ENOENT;
2197 error:
2198         dst_release(dst);
2199 dropdst:
2200         dst_release(dst_orig);
2201         xfrm_pols_put(pols, drop_pols);
2202         return ERR_PTR(err);
2203 }
2204 EXPORT_SYMBOL(xfrm_lookup);
2205
2206 static inline int
2207 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2208 {
2209         struct xfrm_state *x;
2210
2211         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2212                 return 0;
2213         x = skb->sp->xvec[idx];
2214         if (!x->type->reject)
2215                 return 0;
2216         return x->type->reject(x, skb, fl);
2217 }
2218
2219 /* When skb is transformed back to its "native" form, we have to
2220  * check policy restrictions. At the moment we make this in maximally
2221  * stupid way. Shame on me. :-) Of course, connected sockets must
2222  * have policy cached at them.
2223  */
2224
2225 static inline int
2226 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2227               unsigned short family)
2228 {
2229         if (xfrm_state_kern(x))
2230                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2231         return  x->id.proto == tmpl->id.proto &&
2232                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2233                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2234                 x->props.mode == tmpl->mode &&
2235                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2236                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2237                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2238                   xfrm_state_addr_cmp(tmpl, x, family));
2239 }
2240
2241 /*
2242  * 0 or more than 0 is returned when validation is succeeded (either bypass
2243  * because of optional transport mode, or next index of the mathced secpath
2244  * state with the template.
2245  * -1 is returned when no matching template is found.
2246  * Otherwise "-2 - errored_index" is returned.
2247  */
2248 static inline int
2249 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2250                unsigned short family)
2251 {
2252         int idx = start;
2253
2254         if (tmpl->optional) {
2255                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2256                         return start;
2257         } else
2258                 start = -1;
2259         for (; idx < sp->len; idx++) {
2260                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2261                         return ++idx;
2262                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2263                         if (start == -1)
2264                                 start = -2-idx;
2265                         break;
2266                 }
2267         }
2268         return start;
2269 }
2270
2271 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2272                           unsigned int family, int reverse)
2273 {
2274         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2275         int err;
2276
2277         if (unlikely(afinfo == NULL))
2278                 return -EAFNOSUPPORT;
2279
2280         afinfo->decode_session(skb, fl, reverse);
2281         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2282         xfrm_policy_put_afinfo(afinfo);
2283         return err;
2284 }
2285 EXPORT_SYMBOL(__xfrm_decode_session);
2286
2287 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2288 {
2289         for (; k < sp->len; k++) {
2290                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2291                         *idxp = k;
2292                         return 1;
2293                 }
2294         }
2295
2296         return 0;
2297 }
2298
2299 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2300                         unsigned short family)
2301 {
2302         struct net *net = dev_net(skb->dev);
2303         struct xfrm_policy *pol;
2304         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2305         int npols = 0;
2306         int xfrm_nr;
2307         int pi;
2308         int reverse;
2309         struct flowi fl;
2310         u8 fl_dir;
2311         int xerr_idx = -1;
2312
2313         reverse = dir & ~XFRM_POLICY_MASK;
2314         dir &= XFRM_POLICY_MASK;
2315         fl_dir = policy_to_flow_dir(dir);
2316
2317         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2318                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2319                 return 0;
2320         }
2321
2322         nf_nat_decode_session(skb, &fl, family);
2323
2324         /* First, check used SA against their selectors. */
2325         if (skb->sp) {
2326                 int i;
2327
2328                 for (i = skb->sp->len-1; i >= 0; i--) {
2329                         struct xfrm_state *x = skb->sp->xvec[i];
2330                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2331                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2332                                 return 0;
2333                         }
2334                 }
2335         }
2336
2337         pol = NULL;
2338         if (sk && sk->sk_policy[dir]) {
2339                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2340                 if (IS_ERR(pol)) {
2341                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2342                         return 0;
2343                 }
2344         }
2345
2346         if (!pol) {
2347                 struct flow_cache_object *flo;
2348
2349                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2350                                         xfrm_policy_lookup, NULL);
2351                 if (IS_ERR_OR_NULL(flo))
2352                         pol = ERR_CAST(flo);
2353                 else
2354                         pol = container_of(flo, struct xfrm_policy, flo);
2355         }
2356
2357         if (IS_ERR(pol)) {
2358                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2359                 return 0;
2360         }
2361
2362         if (!pol) {
2363                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2364                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2365                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2366                         return 0;
2367                 }
2368                 return 1;
2369         }
2370
2371         pol->curlft.use_time = get_seconds();
2372
2373         pols[0] = pol;
2374         npols++;
2375 #ifdef CONFIG_XFRM_SUB_POLICY
2376         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2377                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2378                                                     &fl, family,
2379                                                     XFRM_POLICY_IN);
2380                 if (pols[1]) {
2381                         if (IS_ERR(pols[1])) {
2382                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2383                                 return 0;
2384                         }
2385                         pols[1]->curlft.use_time = get_seconds();
2386                         npols++;
2387                 }
2388         }
2389 #endif
2390
2391         if (pol->action == XFRM_POLICY_ALLOW) {
2392                 struct sec_path *sp;
2393                 static struct sec_path dummy;
2394                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2395                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2396                 struct xfrm_tmpl **tpp = tp;
2397                 int ti = 0;
2398                 int i, k;
2399
2400                 if ((sp = skb->sp) == NULL)
2401                         sp = &dummy;
2402
2403                 for (pi = 0; pi < npols; pi++) {
2404                         if (pols[pi] != pol &&
2405                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2406                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2407                                 goto reject;
2408                         }
2409                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2410                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2411                                 goto reject_error;
2412                         }
2413                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2414                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2415                 }
2416                 xfrm_nr = ti;
2417                 if (npols > 1) {
2418                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2419                         tpp = stp;
2420                 }
2421
2422                 /* For each tunnel xfrm, find the first matching tmpl.
2423                  * For each tmpl before that, find corresponding xfrm.
2424                  * Order is _important_. Later we will implement
2425                  * some barriers, but at the moment barriers
2426                  * are implied between each two transformations.
2427                  */
2428                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2429                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2430                         if (k < 0) {
2431                                 if (k < -1)
2432                                         /* "-2 - errored_index" returned */
2433                                         xerr_idx = -(2+k);
2434                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2435                                 goto reject;
2436                         }
2437                 }
2438
2439                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2440                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2441                         goto reject;
2442                 }
2443
2444                 xfrm_pols_put(pols, npols);
2445                 return 1;
2446         }
2447         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2448
2449 reject:
2450         xfrm_secpath_reject(xerr_idx, skb, &fl);
2451 reject_error:
2452         xfrm_pols_put(pols, npols);
2453         return 0;
2454 }
2455 EXPORT_SYMBOL(__xfrm_policy_check);
2456
2457 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2458 {
2459         struct net *net = dev_net(skb->dev);
2460         struct flowi fl;
2461         struct dst_entry *dst;
2462         int res = 1;
2463
2464         if (xfrm_decode_session(skb, &fl, family) < 0) {
2465                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2466                 return 0;
2467         }
2468
2469         skb_dst_force(skb);
2470
2471         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2472         if (IS_ERR(dst)) {
2473                 res = 0;
2474                 dst = NULL;
2475         }
2476         skb_dst_set(skb, dst);
2477         return res;
2478 }
2479 EXPORT_SYMBOL(__xfrm_route_forward);
2480
2481 /* Optimize later using cookies and generation ids. */
2482
2483 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2484 {
2485         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2486          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2487          * get validated by dst_ops->check on every use.  We do this
2488          * because when a normal route referenced by an XFRM dst is
2489          * obsoleted we do not go looking around for all parent
2490          * referencing XFRM dsts so that we can invalidate them.  It
2491          * is just too much work.  Instead we make the checks here on
2492          * every use.  For example:
2493          *
2494          *      XFRM dst A --> IPv4 dst X
2495          *
2496          * X is the "xdst->route" of A (X is also the "dst->path" of A
2497          * in this example).  If X is marked obsolete, "A" will not
2498          * notice.  That's what we are validating here via the
2499          * stale_bundle() check.
2500          *
2501          * When a policy's bundle is pruned, we dst_free() the XFRM
2502          * dst which causes it's ->obsolete field to be set to
2503          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2504          * this, we want to force a new route lookup.
2505          */
2506         if (dst->obsolete < 0 && !stale_bundle(dst))
2507                 return dst;
2508
2509         return NULL;
2510 }
2511
2512 static int stale_bundle(struct dst_entry *dst)
2513 {
2514         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2515 }
2516
2517 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2518 {
2519         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2520                 dst->dev = dev_net(dev)->loopback_dev;
2521                 dev_hold(dst->dev);
2522                 dev_put(dev);
2523         }
2524 }
2525 EXPORT_SYMBOL(xfrm_dst_ifdown);
2526
2527 static void xfrm_link_failure(struct sk_buff *skb)
2528 {
2529         /* Impossible. Such dst must be popped before reaches point of failure. */
2530 }
2531
2532 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2533 {
2534         if (dst) {
2535                 if (dst->obsolete) {
2536                         dst_release(dst);
2537                         dst = NULL;
2538                 }
2539         }
2540         return dst;
2541 }
2542
2543 void xfrm_garbage_collect(struct net *net)
2544 {
2545         flow_cache_flush(net);
2546 }
2547 EXPORT_SYMBOL(xfrm_garbage_collect);
2548
2549 static void xfrm_garbage_collect_deferred(struct net *net)
2550 {
2551         flow_cache_flush_deferred(net);
2552 }
2553
2554 static void xfrm_init_pmtu(struct dst_entry *dst)
2555 {
2556         do {
2557                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2558                 u32 pmtu, route_mtu_cached;
2559
2560                 pmtu = dst_mtu(dst->child);
2561                 xdst->child_mtu_cached = pmtu;
2562
2563                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2564
2565                 route_mtu_cached = dst_mtu(xdst->route);
2566                 xdst->route_mtu_cached = route_mtu_cached;
2567
2568                 if (pmtu > route_mtu_cached)
2569                         pmtu = route_mtu_cached;
2570
2571                 dst_metric_set(dst, RTAX_MTU, pmtu);
2572         } while ((dst = dst->next));
2573 }
2574
2575 /* Check that the bundle accepts the flow and its components are
2576  * still valid.
2577  */
2578
2579 static int xfrm_bundle_ok(struct xfrm_dst *first)
2580 {
2581         struct dst_entry *dst = &first->u.dst;
2582         struct xfrm_dst *last;
2583         u32 mtu;
2584
2585         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2586             (dst->dev && !netif_running(dst->dev)))
2587                 return 0;
2588
2589         if (dst->flags & DST_XFRM_QUEUE)
2590                 return 1;
2591
2592         last = NULL;
2593
2594         do {
2595                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2596
2597                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2598                         return 0;
2599                 if (xdst->xfrm_genid != dst->xfrm->genid)
2600                         return 0;
2601                 if (xdst->num_pols > 0 &&
2602                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2603                         return 0;
2604
2605                 mtu = dst_mtu(dst->child);
2606                 if (xdst->child_mtu_cached != mtu) {
2607                         last = xdst;
2608                         xdst->child_mtu_cached = mtu;
2609                 }
2610
2611                 if (!dst_check(xdst->route, xdst->route_cookie))
2612                         return 0;
2613                 mtu = dst_mtu(xdst->route);
2614                 if (xdst->route_mtu_cached != mtu) {
2615                         last = xdst;
2616                         xdst->route_mtu_cached = mtu;
2617                 }
2618
2619                 dst = dst->child;
2620         } while (dst->xfrm);
2621
2622         if (likely(!last))
2623                 return 1;
2624
2625         mtu = last->child_mtu_cached;
2626         for (;;) {
2627                 dst = &last->u.dst;
2628
2629                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2630                 if (mtu > last->route_mtu_cached)
2631                         mtu = last->route_mtu_cached;
2632                 dst_metric_set(dst, RTAX_MTU, mtu);
2633
2634                 if (last == first)
2635                         break;
2636
2637                 last = (struct xfrm_dst *)last->u.dst.next;
2638                 last->child_mtu_cached = mtu;
2639         }
2640
2641         return 1;
2642 }
2643
2644 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2645 {
2646         return dst_metric_advmss(dst->path);
2647 }
2648
2649 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2650 {
2651         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2652
2653         return mtu ? : dst_mtu(dst->path);
2654 }
2655
2656 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2657                                            struct sk_buff *skb,
2658                                            const void *daddr)
2659 {
2660         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2661 }
2662
2663 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2664 {
2665         struct net *net;
2666         int err = 0;
2667         if (unlikely(afinfo == NULL))
2668                 return -EINVAL;
2669         if (unlikely(afinfo->family >= NPROTO))
2670                 return -EAFNOSUPPORT;
2671         spin_lock(&xfrm_policy_afinfo_lock);
2672         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2673                 err = -ENOBUFS;
2674         else {
2675                 struct dst_ops *dst_ops = afinfo->dst_ops;
2676                 if (likely(dst_ops->kmem_cachep == NULL))
2677                         dst_ops->kmem_cachep = xfrm_dst_cache;
2678                 if (likely(dst_ops->check == NULL))
2679                         dst_ops->check = xfrm_dst_check;
2680                 if (likely(dst_ops->default_advmss == NULL))
2681                         dst_ops->default_advmss = xfrm_default_advmss;
2682                 if (likely(dst_ops->mtu == NULL))
2683                         dst_ops->mtu = xfrm_mtu;
2684                 if (likely(dst_ops->negative_advice == NULL))
2685                         dst_ops->negative_advice = xfrm_negative_advice;
2686                 if (likely(dst_ops->link_failure == NULL))
2687                         dst_ops->link_failure = xfrm_link_failure;
2688                 if (likely(dst_ops->neigh_lookup == NULL))
2689                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2690                 if (likely(afinfo->garbage_collect == NULL))
2691                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2692                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2693         }
2694         spin_unlock(&xfrm_policy_afinfo_lock);
2695
2696         rtnl_lock();
2697         for_each_net(net) {
2698                 struct dst_ops *xfrm_dst_ops;
2699
2700                 switch (afinfo->family) {
2701                 case AF_INET:
2702                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2703                         break;
2704 #if IS_ENABLED(CONFIG_IPV6)
2705                 case AF_INET6:
2706                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2707                         break;
2708 #endif
2709                 default:
2710                         BUG();
2711                 }
2712                 *xfrm_dst_ops = *afinfo->dst_ops;
2713         }
2714         rtnl_unlock();
2715
2716         return err;
2717 }
2718 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2719
2720 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2721 {
2722         int err = 0;
2723         if (unlikely(afinfo == NULL))
2724                 return -EINVAL;
2725         if (unlikely(afinfo->family >= NPROTO))
2726                 return -EAFNOSUPPORT;
2727         spin_lock(&xfrm_policy_afinfo_lock);
2728         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2729                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2730                         err = -EINVAL;
2731                 else
2732                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2733                                          NULL);
2734         }
2735         spin_unlock(&xfrm_policy_afinfo_lock);
2736         if (!err) {
2737                 struct dst_ops *dst_ops = afinfo->dst_ops;
2738
2739                 synchronize_rcu();
2740
2741                 dst_ops->kmem_cachep = NULL;
2742                 dst_ops->check = NULL;
2743                 dst_ops->negative_advice = NULL;
2744                 dst_ops->link_failure = NULL;
2745                 afinfo->garbage_collect = NULL;
2746         }
2747         return err;
2748 }
2749 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2750
2751 static void __net_init xfrm_dst_ops_init(struct net *net)
2752 {
2753         struct xfrm_policy_afinfo *afinfo;
2754
2755         rcu_read_lock();
2756         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2757         if (afinfo)
2758                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2759 #if IS_ENABLED(CONFIG_IPV6)
2760         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2761         if (afinfo)
2762                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2763 #endif
2764         rcu_read_unlock();
2765 }
2766
2767 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2768 {
2769         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2770
2771         switch (event) {
2772         case NETDEV_DOWN:
2773                 xfrm_garbage_collect(dev_net(dev));
2774         }
2775         return NOTIFY_DONE;
2776 }
2777
2778 static struct notifier_block xfrm_dev_notifier = {
2779         .notifier_call  = xfrm_dev_event,
2780 };
2781
2782 #ifdef CONFIG_XFRM_STATISTICS
2783 static int __net_init xfrm_statistics_init(struct net *net)
2784 {
2785         int rv;
2786
2787         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2788                           sizeof(struct linux_xfrm_mib),
2789                           __alignof__(struct linux_xfrm_mib)) < 0)
2790                 return -ENOMEM;
2791         rv = xfrm_proc_init(net);
2792         if (rv < 0)
2793                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2794         return rv;
2795 }
2796
2797 static void xfrm_statistics_fini(struct net *net)
2798 {
2799         xfrm_proc_fini(net);
2800         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2801 }
2802 #else
2803 static int __net_init xfrm_statistics_init(struct net *net)
2804 {
2805         return 0;
2806 }
2807
2808 static void xfrm_statistics_fini(struct net *net)
2809 {
2810 }
2811 #endif
2812
2813 static int __net_init xfrm_policy_init(struct net *net)
2814 {
2815         unsigned int hmask, sz;
2816         int dir;
2817
2818         if (net_eq(net, &init_net))
2819                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2820                                            sizeof(struct xfrm_dst),
2821                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2822                                            NULL);
2823
2824         hmask = 8 - 1;
2825         sz = (hmask+1) * sizeof(struct hlist_head);
2826
2827         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2828         if (!net->xfrm.policy_byidx)
2829                 goto out_byidx;
2830         net->xfrm.policy_idx_hmask = hmask;
2831
2832         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2833                 struct xfrm_policy_hash *htab;
2834
2835                 net->xfrm.policy_count[dir] = 0;
2836                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2837
2838                 htab = &net->xfrm.policy_bydst[dir];
2839                 htab->table = xfrm_hash_alloc(sz);
2840                 if (!htab->table)
2841                         goto out_bydst;
2842                 htab->hmask = hmask;
2843         }
2844
2845         INIT_LIST_HEAD(&net->xfrm.policy_all);
2846         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2847         if (net_eq(net, &init_net))
2848                 register_netdevice_notifier(&xfrm_dev_notifier);
2849         return 0;
2850
2851 out_bydst:
2852         for (dir--; dir >= 0; dir--) {
2853                 struct xfrm_policy_hash *htab;
2854
2855                 htab = &net->xfrm.policy_bydst[dir];
2856                 xfrm_hash_free(htab->table, sz);
2857         }
2858         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2859 out_byidx:
2860         return -ENOMEM;
2861 }
2862
2863 static void xfrm_policy_fini(struct net *net)
2864 {
2865         struct xfrm_audit audit_info;
2866         unsigned int sz;
2867         int dir;
2868
2869         flush_work(&net->xfrm.policy_hash_work);
2870 #ifdef CONFIG_XFRM_SUB_POLICY
2871         audit_info.loginuid = INVALID_UID;
2872         audit_info.sessionid = (unsigned int)-1;
2873         audit_info.secid = 0;
2874         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2875 #endif
2876         audit_info.loginuid = INVALID_UID;
2877         audit_info.sessionid = (unsigned int)-1;
2878         audit_info.secid = 0;
2879         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2880
2881         WARN_ON(!list_empty(&net->xfrm.policy_all));
2882
2883         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2884                 struct xfrm_policy_hash *htab;
2885
2886                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2887
2888                 htab = &net->xfrm.policy_bydst[dir];
2889                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2890                 WARN_ON(!hlist_empty(htab->table));
2891                 xfrm_hash_free(htab->table, sz);
2892         }
2893
2894         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2895         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2896         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2897 }
2898
2899 static int __net_init xfrm_net_init(struct net *net)
2900 {
2901         int rv;
2902
2903         rv = xfrm_statistics_init(net);
2904         if (rv < 0)
2905                 goto out_statistics;
2906         rv = xfrm_state_init(net);
2907         if (rv < 0)
2908                 goto out_state;
2909         rv = xfrm_policy_init(net);
2910         if (rv < 0)
2911                 goto out_policy;
2912         xfrm_dst_ops_init(net);
2913         rv = xfrm_sysctl_init(net);
2914         if (rv < 0)
2915                 goto out_sysctl;
2916         rv = flow_cache_init(net);
2917         if (rv < 0)
2918                 goto out;
2919
2920         /* Initialize the per-net locks here */
2921         spin_lock_init(&net->xfrm.xfrm_state_lock);
2922         rwlock_init(&net->xfrm.xfrm_policy_lock);
2923         mutex_init(&net->xfrm.xfrm_cfg_mutex);
2924
2925         return 0;
2926
2927 out:
2928         xfrm_sysctl_fini(net);
2929 out_sysctl:
2930         xfrm_policy_fini(net);
2931 out_policy:
2932         xfrm_state_fini(net);
2933 out_state:
2934         xfrm_statistics_fini(net);
2935 out_statistics:
2936         return rv;
2937 }
2938
2939 static void __net_exit xfrm_net_exit(struct net *net)
2940 {
2941         flow_cache_fini(net);
2942         xfrm_sysctl_fini(net);
2943         xfrm_policy_fini(net);
2944         xfrm_state_fini(net);
2945         xfrm_statistics_fini(net);
2946 }
2947
2948 static struct pernet_operations __net_initdata xfrm_net_ops = {
2949         .init = xfrm_net_init,
2950         .exit = xfrm_net_exit,
2951 };
2952
2953 void __init xfrm_init(void)
2954 {
2955         register_pernet_subsys(&xfrm_net_ops);
2956         xfrm_input_init();
2957 }
2958
2959 #ifdef CONFIG_AUDITSYSCALL
2960 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2961                                          struct audit_buffer *audit_buf)
2962 {
2963         struct xfrm_sec_ctx *ctx = xp->security;
2964         struct xfrm_selector *sel = &xp->selector;
2965
2966         if (ctx)
2967                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2968                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2969
2970         switch (sel->family) {
2971         case AF_INET:
2972                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2973                 if (sel->prefixlen_s != 32)
2974                         audit_log_format(audit_buf, " src_prefixlen=%d",
2975                                          sel->prefixlen_s);
2976                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2977                 if (sel->prefixlen_d != 32)
2978                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2979                                          sel->prefixlen_d);
2980                 break;
2981         case AF_INET6:
2982                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2983                 if (sel->prefixlen_s != 128)
2984                         audit_log_format(audit_buf, " src_prefixlen=%d",
2985                                          sel->prefixlen_s);
2986                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2987                 if (sel->prefixlen_d != 128)
2988                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2989                                          sel->prefixlen_d);
2990                 break;
2991         }
2992 }
2993
2994 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2995                            kuid_t auid, unsigned int sessionid, u32 secid)
2996 {
2997         struct audit_buffer *audit_buf;
2998
2999         audit_buf = xfrm_audit_start("SPD-add");
3000         if (audit_buf == NULL)
3001                 return;
3002         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3003         audit_log_format(audit_buf, " res=%u", result);
3004         xfrm_audit_common_policyinfo(xp, audit_buf);
3005         audit_log_end(audit_buf);
3006 }
3007 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3008
3009 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3010                               kuid_t auid, unsigned int sessionid, u32 secid)
3011 {
3012         struct audit_buffer *audit_buf;
3013
3014         audit_buf = xfrm_audit_start("SPD-delete");
3015         if (audit_buf == NULL)
3016                 return;
3017         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3018         audit_log_format(audit_buf, " res=%u", result);
3019         xfrm_audit_common_policyinfo(xp, audit_buf);
3020         audit_log_end(audit_buf);
3021 }
3022 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3023 #endif
3024
3025 #ifdef CONFIG_XFRM_MIGRATE
3026 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3027                                         const struct xfrm_selector *sel_tgt)
3028 {
3029         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3030                 if (sel_tgt->family == sel_cmp->family &&
3031                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3032                                     sel_cmp->family) &&
3033                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3034                                     sel_cmp->family) &&
3035                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3036                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3037                         return true;
3038                 }
3039         } else {
3040                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3041                         return true;
3042                 }
3043         }
3044         return false;
3045 }
3046
3047 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3048                                                     u8 dir, u8 type, struct net *net)
3049 {
3050         struct xfrm_policy *pol, *ret = NULL;
3051         struct hlist_head *chain;
3052         u32 priority = ~0U;
3053
3054         read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3055         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3056         hlist_for_each_entry(pol, chain, bydst) {
3057                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3058                     pol->type == type) {
3059                         ret = pol;
3060                         priority = ret->priority;
3061                         break;
3062                 }
3063         }
3064         chain = &net->xfrm.policy_inexact[dir];
3065         hlist_for_each_entry(pol, chain, bydst) {
3066                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3067                     pol->type == type &&
3068                     pol->priority < priority) {
3069                         ret = pol;
3070                         break;
3071                 }
3072         }
3073
3074         if (ret)
3075                 xfrm_pol_hold(ret);
3076
3077         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3078
3079         return ret;
3080 }
3081
3082 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3083 {
3084         int match = 0;
3085
3086         if (t->mode == m->mode && t->id.proto == m->proto &&
3087             (m->reqid == 0 || t->reqid == m->reqid)) {
3088                 switch (t->mode) {
3089                 case XFRM_MODE_TUNNEL:
3090                 case XFRM_MODE_BEET:
3091                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3092                                             m->old_family) &&
3093                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3094                                             m->old_family)) {
3095                                 match = 1;
3096                         }
3097                         break;
3098                 case XFRM_MODE_TRANSPORT:
3099                         /* in case of transport mode, template does not store
3100                            any IP addresses, hence we just compare mode and
3101                            protocol */
3102                         match = 1;
3103                         break;
3104                 default:
3105                         break;
3106                 }
3107         }
3108         return match;
3109 }
3110
3111 /* update endpoint address(es) of template(s) */
3112 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3113                                struct xfrm_migrate *m, int num_migrate)
3114 {
3115         struct xfrm_migrate *mp;
3116         int i, j, n = 0;
3117
3118         write_lock_bh(&pol->lock);
3119         if (unlikely(pol->walk.dead)) {
3120                 /* target policy has been deleted */
3121                 write_unlock_bh(&pol->lock);
3122                 return -ENOENT;
3123         }
3124
3125         for (i = 0; i < pol->xfrm_nr; i++) {
3126                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3127                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3128                                 continue;
3129                         n++;
3130                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3131                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3132                                 continue;
3133                         /* update endpoints */
3134                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3135                                sizeof(pol->xfrm_vec[i].id.daddr));
3136                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3137                                sizeof(pol->xfrm_vec[i].saddr));
3138                         pol->xfrm_vec[i].encap_family = mp->new_family;
3139                         /* flush bundles */
3140                         atomic_inc(&pol->genid);
3141                 }
3142         }
3143
3144         write_unlock_bh(&pol->lock);
3145
3146         if (!n)
3147                 return -ENODATA;
3148
3149         return 0;
3150 }
3151
3152 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3153 {
3154         int i, j;
3155
3156         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3157                 return -EINVAL;
3158
3159         for (i = 0; i < num_migrate; i++) {
3160                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3161                                     m[i].old_family) &&
3162                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3163                                     m[i].old_family))
3164                         return -EINVAL;
3165                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3166                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3167                         return -EINVAL;
3168
3169                 /* check if there is any duplicated entry */
3170                 for (j = i + 1; j < num_migrate; j++) {
3171                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3172                                     sizeof(m[i].old_daddr)) &&
3173                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3174                                     sizeof(m[i].old_saddr)) &&
3175                             m[i].proto == m[j].proto &&
3176                             m[i].mode == m[j].mode &&
3177                             m[i].reqid == m[j].reqid &&
3178                             m[i].old_family == m[j].old_family)
3179                                 return -EINVAL;
3180                 }
3181         }
3182
3183         return 0;
3184 }
3185
3186 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3187                  struct xfrm_migrate *m, int num_migrate,
3188                  struct xfrm_kmaddress *k, struct net *net)
3189 {
3190         int i, err, nx_cur = 0, nx_new = 0;
3191         struct xfrm_policy *pol = NULL;
3192         struct xfrm_state *x, *xc;
3193         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3194         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3195         struct xfrm_migrate *mp;
3196
3197         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3198                 goto out;
3199
3200         /* Stage 1 - find policy */
3201         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3202                 err = -ENOENT;
3203                 goto out;
3204         }
3205
3206         /* Stage 2 - find and update state(s) */
3207         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3208                 if ((x = xfrm_migrate_state_find(mp, net))) {
3209                         x_cur[nx_cur] = x;
3210                         nx_cur++;
3211                         if ((xc = xfrm_state_migrate(x, mp))) {
3212                                 x_new[nx_new] = xc;
3213                                 nx_new++;
3214                         } else {
3215                                 err = -ENODATA;
3216                                 goto restore_state;
3217                         }
3218                 }
3219         }
3220
3221         /* Stage 3 - update policy */
3222