net: file_operations should be const
[linux-2.6.git] / net / xfrm / xfrm_policy.c
1 /*
2  * xfrm_policy.c
3  *
4  * Changes:
5  *      Mitsuru KANDA @USAGI
6  *      Kazunori MIYAZAWA @USAGI
7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8  *              IPv6 support
9  *      Kazunori MIYAZAWA @USAGI
10  *      YOSHIFUJI Hideaki
11  *              Split up af-specific portion
12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
13  *
14  */
15
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
28 #include <net/dst.h>
29 #include <net/xfrm.h>
30 #include <net/ip.h>
31 #ifdef CONFIG_XFRM_STATISTICS
32 #include <net/snmp.h>
33 #endif
34
35 #include "xfrm_hash.h"
36
37 DEFINE_MUTEX(xfrm_cfg_mutex);
38 EXPORT_SYMBOL(xfrm_cfg_mutex);
39
40 static DEFINE_RWLOCK(xfrm_policy_lock);
41
42 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
43 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
44
45 static struct kmem_cache *xfrm_dst_cache __read_mostly;
46
47 static HLIST_HEAD(xfrm_policy_gc_list);
48 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
49
50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
52 static void xfrm_init_pmtu(struct dst_entry *dst);
53
54 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
55                                                 int dir);
56
57 static inline int
58 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
59 {
60         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
61                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
62                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
63                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
64                 (fl->proto == sel->proto || !sel->proto) &&
65                 (fl->oif == sel->ifindex || !sel->ifindex);
66 }
67
68 static inline int
69 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
70 {
71         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
72                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
73                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
74                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
75                 (fl->proto == sel->proto || !sel->proto) &&
76                 (fl->oif == sel->ifindex || !sel->ifindex);
77 }
78
79 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
80                     unsigned short family)
81 {
82         switch (family) {
83         case AF_INET:
84                 return __xfrm4_selector_match(sel, fl);
85         case AF_INET6:
86                 return __xfrm6_selector_match(sel, fl);
87         }
88         return 0;
89 }
90
91 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
92                                                   xfrm_address_t *saddr,
93                                                   xfrm_address_t *daddr,
94                                                   int family)
95 {
96         struct xfrm_policy_afinfo *afinfo;
97         struct dst_entry *dst;
98
99         afinfo = xfrm_policy_get_afinfo(family);
100         if (unlikely(afinfo == NULL))
101                 return ERR_PTR(-EAFNOSUPPORT);
102
103         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
104
105         xfrm_policy_put_afinfo(afinfo);
106
107         return dst;
108 }
109
110 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
111                                                 xfrm_address_t *prev_saddr,
112                                                 xfrm_address_t *prev_daddr,
113                                                 int family)
114 {
115         struct net *net = xs_net(x);
116         xfrm_address_t *saddr = &x->props.saddr;
117         xfrm_address_t *daddr = &x->id.daddr;
118         struct dst_entry *dst;
119
120         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
121                 saddr = x->coaddr;
122                 daddr = prev_daddr;
123         }
124         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
125                 saddr = prev_saddr;
126                 daddr = x->coaddr;
127         }
128
129         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
130
131         if (!IS_ERR(dst)) {
132                 if (prev_saddr != saddr)
133                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
134                 if (prev_daddr != daddr)
135                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
136         }
137
138         return dst;
139 }
140
141 static inline unsigned long make_jiffies(long secs)
142 {
143         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
144                 return MAX_SCHEDULE_TIMEOUT-1;
145         else
146                 return secs*HZ;
147 }
148
149 static void xfrm_policy_timer(unsigned long data)
150 {
151         struct xfrm_policy *xp = (struct xfrm_policy*)data;
152         unsigned long now = get_seconds();
153         long next = LONG_MAX;
154         int warn = 0;
155         int dir;
156
157         read_lock(&xp->lock);
158
159         if (xp->walk.dead)
160                 goto out;
161
162         dir = xfrm_policy_id2dir(xp->index);
163
164         if (xp->lft.hard_add_expires_seconds) {
165                 long tmo = xp->lft.hard_add_expires_seconds +
166                         xp->curlft.add_time - now;
167                 if (tmo <= 0)
168                         goto expired;
169                 if (tmo < next)
170                         next = tmo;
171         }
172         if (xp->lft.hard_use_expires_seconds) {
173                 long tmo = xp->lft.hard_use_expires_seconds +
174                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
175                 if (tmo <= 0)
176                         goto expired;
177                 if (tmo < next)
178                         next = tmo;
179         }
180         if (xp->lft.soft_add_expires_seconds) {
181                 long tmo = xp->lft.soft_add_expires_seconds +
182                         xp->curlft.add_time - now;
183                 if (tmo <= 0) {
184                         warn = 1;
185                         tmo = XFRM_KM_TIMEOUT;
186                 }
187                 if (tmo < next)
188                         next = tmo;
189         }
190         if (xp->lft.soft_use_expires_seconds) {
191                 long tmo = xp->lft.soft_use_expires_seconds +
192                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
193                 if (tmo <= 0) {
194                         warn = 1;
195                         tmo = XFRM_KM_TIMEOUT;
196                 }
197                 if (tmo < next)
198                         next = tmo;
199         }
200
201         if (warn)
202                 km_policy_expired(xp, dir, 0, 0);
203         if (next != LONG_MAX &&
204             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
205                 xfrm_pol_hold(xp);
206
207 out:
208         read_unlock(&xp->lock);
209         xfrm_pol_put(xp);
210         return;
211
212 expired:
213         read_unlock(&xp->lock);
214         if (!xfrm_policy_delete(xp, dir))
215                 km_policy_expired(xp, dir, 1, 0);
216         xfrm_pol_put(xp);
217 }
218
219
220 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
221  * SPD calls.
222  */
223
224 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
225 {
226         struct xfrm_policy *policy;
227
228         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
229
230         if (policy) {
231                 write_pnet(&policy->xp_net, net);
232                 INIT_LIST_HEAD(&policy->walk.all);
233                 INIT_HLIST_NODE(&policy->bydst);
234                 INIT_HLIST_NODE(&policy->byidx);
235                 rwlock_init(&policy->lock);
236                 atomic_set(&policy->refcnt, 1);
237                 setup_timer(&policy->timer, xfrm_policy_timer,
238                                 (unsigned long)policy);
239         }
240         return policy;
241 }
242 EXPORT_SYMBOL(xfrm_policy_alloc);
243
244 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
245
246 void xfrm_policy_destroy(struct xfrm_policy *policy)
247 {
248         BUG_ON(!policy->walk.dead);
249
250         BUG_ON(policy->bundles);
251
252         if (del_timer(&policy->timer))
253                 BUG();
254
255         security_xfrm_policy_free(policy->security);
256         kfree(policy);
257 }
258 EXPORT_SYMBOL(xfrm_policy_destroy);
259
260 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
261 {
262         struct dst_entry *dst;
263
264         while ((dst = policy->bundles) != NULL) {
265                 policy->bundles = dst->next;
266                 dst_free(dst);
267         }
268
269         if (del_timer(&policy->timer))
270                 atomic_dec(&policy->refcnt);
271
272         if (atomic_read(&policy->refcnt) > 1)
273                 flow_cache_flush();
274
275         xfrm_pol_put(policy);
276 }
277
278 static void xfrm_policy_gc_task(struct work_struct *work)
279 {
280         struct xfrm_policy *policy;
281         struct hlist_node *entry, *tmp;
282         struct hlist_head gc_list;
283
284         spin_lock_bh(&xfrm_policy_gc_lock);
285         gc_list.first = xfrm_policy_gc_list.first;
286         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
287         spin_unlock_bh(&xfrm_policy_gc_lock);
288
289         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
290                 xfrm_policy_gc_kill(policy);
291 }
292 static DECLARE_WORK(xfrm_policy_gc_work, xfrm_policy_gc_task);
293
294 /* Rule must be locked. Release descentant resources, announce
295  * entry dead. The rule must be unlinked from lists to the moment.
296  */
297
298 static void xfrm_policy_kill(struct xfrm_policy *policy)
299 {
300         int dead;
301
302         write_lock_bh(&policy->lock);
303         dead = policy->walk.dead;
304         policy->walk.dead = 1;
305         write_unlock_bh(&policy->lock);
306
307         if (unlikely(dead)) {
308                 WARN_ON(1);
309                 return;
310         }
311
312         spin_lock_bh(&xfrm_policy_gc_lock);
313         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
314         spin_unlock_bh(&xfrm_policy_gc_lock);
315
316         schedule_work(&xfrm_policy_gc_work);
317 }
318
319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
320
321 static inline unsigned int idx_hash(struct net *net, u32 index)
322 {
323         return __idx_hash(index, net->xfrm.policy_idx_hmask);
324 }
325
326 static struct hlist_head *policy_hash_bysel(struct net *net, struct xfrm_selector *sel, unsigned short family, int dir)
327 {
328         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
329         unsigned int hash = __sel_hash(sel, family, hmask);
330
331         return (hash == hmask + 1 ?
332                 &net->xfrm.policy_inexact[dir] :
333                 net->xfrm.policy_bydst[dir].table + hash);
334 }
335
336 static struct hlist_head *policy_hash_direct(struct net *net, xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
337 {
338         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
339         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
340
341         return net->xfrm.policy_bydst[dir].table + hash;
342 }
343
344 static void xfrm_dst_hash_transfer(struct hlist_head *list,
345                                    struct hlist_head *ndsttable,
346                                    unsigned int nhashmask)
347 {
348         struct hlist_node *entry, *tmp, *entry0 = NULL;
349         struct xfrm_policy *pol;
350         unsigned int h0 = 0;
351
352 redo:
353         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
354                 unsigned int h;
355
356                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
357                                 pol->family, nhashmask);
358                 if (!entry0) {
359                         hlist_del(entry);
360                         hlist_add_head(&pol->bydst, ndsttable+h);
361                         h0 = h;
362                 } else {
363                         if (h != h0)
364                                 continue;
365                         hlist_del(entry);
366                         hlist_add_after(entry0, &pol->bydst);
367                 }
368                 entry0 = entry;
369         }
370         if (!hlist_empty(list)) {
371                 entry0 = NULL;
372                 goto redo;
373         }
374 }
375
376 static void xfrm_idx_hash_transfer(struct hlist_head *list,
377                                    struct hlist_head *nidxtable,
378                                    unsigned int nhashmask)
379 {
380         struct hlist_node *entry, *tmp;
381         struct xfrm_policy *pol;
382
383         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
384                 unsigned int h;
385
386                 h = __idx_hash(pol->index, nhashmask);
387                 hlist_add_head(&pol->byidx, nidxtable+h);
388         }
389 }
390
391 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
392 {
393         return ((old_hmask + 1) << 1) - 1;
394 }
395
396 static void xfrm_bydst_resize(struct net *net, int dir)
397 {
398         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
399         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
400         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
401         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
402         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
403         int i;
404
405         if (!ndst)
406                 return;
407
408         write_lock_bh(&xfrm_policy_lock);
409
410         for (i = hmask; i >= 0; i--)
411                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
412
413         net->xfrm.policy_bydst[dir].table = ndst;
414         net->xfrm.policy_bydst[dir].hmask = nhashmask;
415
416         write_unlock_bh(&xfrm_policy_lock);
417
418         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
419 }
420
421 static void xfrm_byidx_resize(struct net *net, int total)
422 {
423         unsigned int hmask = net->xfrm.policy_idx_hmask;
424         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
425         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
426         struct hlist_head *oidx = net->xfrm.policy_byidx;
427         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
428         int i;
429
430         if (!nidx)
431                 return;
432
433         write_lock_bh(&xfrm_policy_lock);
434
435         for (i = hmask; i >= 0; i--)
436                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
437
438         net->xfrm.policy_byidx = nidx;
439         net->xfrm.policy_idx_hmask = nhashmask;
440
441         write_unlock_bh(&xfrm_policy_lock);
442
443         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
444 }
445
446 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
447 {
448         unsigned int cnt = net->xfrm.policy_count[dir];
449         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
450
451         if (total)
452                 *total += cnt;
453
454         if ((hmask + 1) < xfrm_policy_hashmax &&
455             cnt > hmask)
456                 return 1;
457
458         return 0;
459 }
460
461 static inline int xfrm_byidx_should_resize(struct net *net, int total)
462 {
463         unsigned int hmask = net->xfrm.policy_idx_hmask;
464
465         if ((hmask + 1) < xfrm_policy_hashmax &&
466             total > hmask)
467                 return 1;
468
469         return 0;
470 }
471
472 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
473 {
474         read_lock_bh(&xfrm_policy_lock);
475         si->incnt = init_net.xfrm.policy_count[XFRM_POLICY_IN];
476         si->outcnt = init_net.xfrm.policy_count[XFRM_POLICY_OUT];
477         si->fwdcnt = init_net.xfrm.policy_count[XFRM_POLICY_FWD];
478         si->inscnt = init_net.xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
479         si->outscnt = init_net.xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
480         si->fwdscnt = init_net.xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
481         si->spdhcnt = init_net.xfrm.policy_idx_hmask;
482         si->spdhmcnt = xfrm_policy_hashmax;
483         read_unlock_bh(&xfrm_policy_lock);
484 }
485 EXPORT_SYMBOL(xfrm_spd_getinfo);
486
487 static DEFINE_MUTEX(hash_resize_mutex);
488 static void xfrm_hash_resize(struct work_struct *work)
489 {
490         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
491         int dir, total;
492
493         mutex_lock(&hash_resize_mutex);
494
495         total = 0;
496         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
497                 if (xfrm_bydst_should_resize(net, dir, &total))
498                         xfrm_bydst_resize(net, dir);
499         }
500         if (xfrm_byidx_should_resize(net, total))
501                 xfrm_byidx_resize(net, total);
502
503         mutex_unlock(&hash_resize_mutex);
504 }
505
506 /* Generate new index... KAME seems to generate them ordered by cost
507  * of an absolute inpredictability of ordering of rules. This will not pass. */
508 static u32 xfrm_gen_index(struct net *net, int dir)
509 {
510         static u32 idx_generator;
511
512         for (;;) {
513                 struct hlist_node *entry;
514                 struct hlist_head *list;
515                 struct xfrm_policy *p;
516                 u32 idx;
517                 int found;
518
519                 idx = (idx_generator | dir);
520                 idx_generator += 8;
521                 if (idx == 0)
522                         idx = 8;
523                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
524                 found = 0;
525                 hlist_for_each_entry(p, entry, list, byidx) {
526                         if (p->index == idx) {
527                                 found = 1;
528                                 break;
529                         }
530                 }
531                 if (!found)
532                         return idx;
533         }
534 }
535
536 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
537 {
538         u32 *p1 = (u32 *) s1;
539         u32 *p2 = (u32 *) s2;
540         int len = sizeof(struct xfrm_selector) / sizeof(u32);
541         int i;
542
543         for (i = 0; i < len; i++) {
544                 if (p1[i] != p2[i])
545                         return 1;
546         }
547
548         return 0;
549 }
550
551 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
552 {
553         struct net *net = xp_net(policy);
554         struct xfrm_policy *pol;
555         struct xfrm_policy *delpol;
556         struct hlist_head *chain;
557         struct hlist_node *entry, *newpos;
558         struct dst_entry *gc_list;
559
560         write_lock_bh(&xfrm_policy_lock);
561         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
562         delpol = NULL;
563         newpos = NULL;
564         hlist_for_each_entry(pol, entry, chain, bydst) {
565                 if (pol->type == policy->type &&
566                     !selector_cmp(&pol->selector, &policy->selector) &&
567                     xfrm_sec_ctx_match(pol->security, policy->security) &&
568                     !WARN_ON(delpol)) {
569                         if (excl) {
570                                 write_unlock_bh(&xfrm_policy_lock);
571                                 return -EEXIST;
572                         }
573                         delpol = pol;
574                         if (policy->priority > pol->priority)
575                                 continue;
576                 } else if (policy->priority >= pol->priority) {
577                         newpos = &pol->bydst;
578                         continue;
579                 }
580                 if (delpol)
581                         break;
582         }
583         if (newpos)
584                 hlist_add_after(newpos, &policy->bydst);
585         else
586                 hlist_add_head(&policy->bydst, chain);
587         xfrm_pol_hold(policy);
588         net->xfrm.policy_count[dir]++;
589         atomic_inc(&flow_cache_genid);
590         if (delpol)
591                 __xfrm_policy_unlink(delpol, dir);
592         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
593         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
594         policy->curlft.add_time = get_seconds();
595         policy->curlft.use_time = 0;
596         if (!mod_timer(&policy->timer, jiffies + HZ))
597                 xfrm_pol_hold(policy);
598         list_add(&policy->walk.all, &net->xfrm.policy_all);
599         write_unlock_bh(&xfrm_policy_lock);
600
601         if (delpol)
602                 xfrm_policy_kill(delpol);
603         else if (xfrm_bydst_should_resize(net, dir, NULL))
604                 schedule_work(&net->xfrm.policy_hash_work);
605
606         read_lock_bh(&xfrm_policy_lock);
607         gc_list = NULL;
608         entry = &policy->bydst;
609         hlist_for_each_entry_continue(policy, entry, bydst) {
610                 struct dst_entry *dst;
611
612                 write_lock(&policy->lock);
613                 dst = policy->bundles;
614                 if (dst) {
615                         struct dst_entry *tail = dst;
616                         while (tail->next)
617                                 tail = tail->next;
618                         tail->next = gc_list;
619                         gc_list = dst;
620
621                         policy->bundles = NULL;
622                 }
623                 write_unlock(&policy->lock);
624         }
625         read_unlock_bh(&xfrm_policy_lock);
626
627         while (gc_list) {
628                 struct dst_entry *dst = gc_list;
629
630                 gc_list = dst->next;
631                 dst_free(dst);
632         }
633
634         return 0;
635 }
636 EXPORT_SYMBOL(xfrm_policy_insert);
637
638 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u8 type, int dir,
639                                           struct xfrm_selector *sel,
640                                           struct xfrm_sec_ctx *ctx, int delete,
641                                           int *err)
642 {
643         struct xfrm_policy *pol, *ret;
644         struct hlist_head *chain;
645         struct hlist_node *entry;
646
647         *err = 0;
648         write_lock_bh(&xfrm_policy_lock);
649         chain = policy_hash_bysel(net, sel, sel->family, dir);
650         ret = NULL;
651         hlist_for_each_entry(pol, entry, chain, bydst) {
652                 if (pol->type == type &&
653                     !selector_cmp(sel, &pol->selector) &&
654                     xfrm_sec_ctx_match(ctx, pol->security)) {
655                         xfrm_pol_hold(pol);
656                         if (delete) {
657                                 *err = security_xfrm_policy_delete(
658                                                                 pol->security);
659                                 if (*err) {
660                                         write_unlock_bh(&xfrm_policy_lock);
661                                         return pol;
662                                 }
663                                 __xfrm_policy_unlink(pol, dir);
664                         }
665                         ret = pol;
666                         break;
667                 }
668         }
669         write_unlock_bh(&xfrm_policy_lock);
670
671         if (ret && delete) {
672                 atomic_inc(&flow_cache_genid);
673                 xfrm_policy_kill(ret);
674         }
675         return ret;
676 }
677 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
678
679 struct xfrm_policy *xfrm_policy_byid(struct net *net, u8 type, int dir, u32 id,
680                                      int delete, int *err)
681 {
682         struct xfrm_policy *pol, *ret;
683         struct hlist_head *chain;
684         struct hlist_node *entry;
685
686         *err = -ENOENT;
687         if (xfrm_policy_id2dir(id) != dir)
688                 return NULL;
689
690         *err = 0;
691         write_lock_bh(&xfrm_policy_lock);
692         chain = net->xfrm.policy_byidx + idx_hash(net, id);
693         ret = NULL;
694         hlist_for_each_entry(pol, entry, chain, byidx) {
695                 if (pol->type == type && pol->index == id) {
696                         xfrm_pol_hold(pol);
697                         if (delete) {
698                                 *err = security_xfrm_policy_delete(
699                                                                 pol->security);
700                                 if (*err) {
701                                         write_unlock_bh(&xfrm_policy_lock);
702                                         return pol;
703                                 }
704                                 __xfrm_policy_unlink(pol, dir);
705                         }
706                         ret = pol;
707                         break;
708                 }
709         }
710         write_unlock_bh(&xfrm_policy_lock);
711
712         if (ret && delete) {
713                 atomic_inc(&flow_cache_genid);
714                 xfrm_policy_kill(ret);
715         }
716         return ret;
717 }
718 EXPORT_SYMBOL(xfrm_policy_byid);
719
720 #ifdef CONFIG_SECURITY_NETWORK_XFRM
721 static inline int
722 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
723 {
724         int dir, err = 0;
725
726         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
727                 struct xfrm_policy *pol;
728                 struct hlist_node *entry;
729                 int i;
730
731                 hlist_for_each_entry(pol, entry,
732                                      &net->xfrm.policy_inexact[dir], bydst) {
733                         if (pol->type != type)
734                                 continue;
735                         err = security_xfrm_policy_delete(pol->security);
736                         if (err) {
737                                 xfrm_audit_policy_delete(pol, 0,
738                                                          audit_info->loginuid,
739                                                          audit_info->sessionid,
740                                                          audit_info->secid);
741                                 return err;
742                         }
743                 }
744                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
745                         hlist_for_each_entry(pol, entry,
746                                              net->xfrm.policy_bydst[dir].table + i,
747                                              bydst) {
748                                 if (pol->type != type)
749                                         continue;
750                                 err = security_xfrm_policy_delete(
751                                                                 pol->security);
752                                 if (err) {
753                                         xfrm_audit_policy_delete(pol, 0,
754                                                         audit_info->loginuid,
755                                                         audit_info->sessionid,
756                                                         audit_info->secid);
757                                         return err;
758                                 }
759                         }
760                 }
761         }
762         return err;
763 }
764 #else
765 static inline int
766 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
767 {
768         return 0;
769 }
770 #endif
771
772 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
773 {
774         int dir, err = 0;
775
776         write_lock_bh(&xfrm_policy_lock);
777
778         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
779         if (err)
780                 goto out;
781
782         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
783                 struct xfrm_policy *pol;
784                 struct hlist_node *entry;
785                 int i;
786
787         again1:
788                 hlist_for_each_entry(pol, entry,
789                                      &net->xfrm.policy_inexact[dir], bydst) {
790                         if (pol->type != type)
791                                 continue;
792                         __xfrm_policy_unlink(pol, dir);
793                         write_unlock_bh(&xfrm_policy_lock);
794
795                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
796                                                  audit_info->sessionid,
797                                                  audit_info->secid);
798
799                         xfrm_policy_kill(pol);
800
801                         write_lock_bh(&xfrm_policy_lock);
802                         goto again1;
803                 }
804
805                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
806         again2:
807                         hlist_for_each_entry(pol, entry,
808                                              net->xfrm.policy_bydst[dir].table + i,
809                                              bydst) {
810                                 if (pol->type != type)
811                                         continue;
812                                 __xfrm_policy_unlink(pol, dir);
813                                 write_unlock_bh(&xfrm_policy_lock);
814
815                                 xfrm_audit_policy_delete(pol, 1,
816                                                          audit_info->loginuid,
817                                                          audit_info->sessionid,
818                                                          audit_info->secid);
819                                 xfrm_policy_kill(pol);
820
821                                 write_lock_bh(&xfrm_policy_lock);
822                                 goto again2;
823                         }
824                 }
825
826         }
827         atomic_inc(&flow_cache_genid);
828 out:
829         write_unlock_bh(&xfrm_policy_lock);
830         return err;
831 }
832 EXPORT_SYMBOL(xfrm_policy_flush);
833
834 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
835                      int (*func)(struct xfrm_policy *, int, int, void*),
836                      void *data)
837 {
838         struct xfrm_policy *pol;
839         struct xfrm_policy_walk_entry *x;
840         int error = 0;
841
842         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
843             walk->type != XFRM_POLICY_TYPE_ANY)
844                 return -EINVAL;
845
846         if (list_empty(&walk->walk.all) && walk->seq != 0)
847                 return 0;
848
849         write_lock_bh(&xfrm_policy_lock);
850         if (list_empty(&walk->walk.all))
851                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
852         else
853                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
854         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
855                 if (x->dead)
856                         continue;
857                 pol = container_of(x, struct xfrm_policy, walk);
858                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
859                     walk->type != pol->type)
860                         continue;
861                 error = func(pol, xfrm_policy_id2dir(pol->index),
862                              walk->seq, data);
863                 if (error) {
864                         list_move_tail(&walk->walk.all, &x->all);
865                         goto out;
866                 }
867                 walk->seq++;
868         }
869         if (walk->seq == 0) {
870                 error = -ENOENT;
871                 goto out;
872         }
873         list_del_init(&walk->walk.all);
874 out:
875         write_unlock_bh(&xfrm_policy_lock);
876         return error;
877 }
878 EXPORT_SYMBOL(xfrm_policy_walk);
879
880 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
881 {
882         INIT_LIST_HEAD(&walk->walk.all);
883         walk->walk.dead = 1;
884         walk->type = type;
885         walk->seq = 0;
886 }
887 EXPORT_SYMBOL(xfrm_policy_walk_init);
888
889 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
890 {
891         if (list_empty(&walk->walk.all))
892                 return;
893
894         write_lock_bh(&xfrm_policy_lock);
895         list_del(&walk->walk.all);
896         write_unlock_bh(&xfrm_policy_lock);
897 }
898 EXPORT_SYMBOL(xfrm_policy_walk_done);
899
900 /*
901  * Find policy to apply to this flow.
902  *
903  * Returns 0 if policy found, else an -errno.
904  */
905 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
906                              u8 type, u16 family, int dir)
907 {
908         struct xfrm_selector *sel = &pol->selector;
909         int match, ret = -ESRCH;
910
911         if (pol->family != family ||
912             pol->type != type)
913                 return ret;
914
915         match = xfrm_selector_match(sel, fl, family);
916         if (match)
917                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
918                                                   dir);
919
920         return ret;
921 }
922
923 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
924                                                      struct flowi *fl,
925                                                      u16 family, u8 dir)
926 {
927         int err;
928         struct xfrm_policy *pol, *ret;
929         xfrm_address_t *daddr, *saddr;
930         struct hlist_node *entry;
931         struct hlist_head *chain;
932         u32 priority = ~0U;
933
934         daddr = xfrm_flowi_daddr(fl, family);
935         saddr = xfrm_flowi_saddr(fl, family);
936         if (unlikely(!daddr || !saddr))
937                 return NULL;
938
939         read_lock_bh(&xfrm_policy_lock);
940         chain = policy_hash_direct(net, daddr, saddr, family, dir);
941         ret = NULL;
942         hlist_for_each_entry(pol, entry, chain, bydst) {
943                 err = xfrm_policy_match(pol, fl, type, family, dir);
944                 if (err) {
945                         if (err == -ESRCH)
946                                 continue;
947                         else {
948                                 ret = ERR_PTR(err);
949                                 goto fail;
950                         }
951                 } else {
952                         ret = pol;
953                         priority = ret->priority;
954                         break;
955                 }
956         }
957         chain = &net->xfrm.policy_inexact[dir];
958         hlist_for_each_entry(pol, entry, chain, bydst) {
959                 err = xfrm_policy_match(pol, fl, type, family, dir);
960                 if (err) {
961                         if (err == -ESRCH)
962                                 continue;
963                         else {
964                                 ret = ERR_PTR(err);
965                                 goto fail;
966                         }
967                 } else if (pol->priority < priority) {
968                         ret = pol;
969                         break;
970                 }
971         }
972         if (ret)
973                 xfrm_pol_hold(ret);
974 fail:
975         read_unlock_bh(&xfrm_policy_lock);
976
977         return ret;
978 }
979
980 static int xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
981                               u8 dir, void **objp, atomic_t **obj_refp)
982 {
983         struct xfrm_policy *pol;
984         int err = 0;
985
986 #ifdef CONFIG_XFRM_SUB_POLICY
987         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
988         if (IS_ERR(pol)) {
989                 err = PTR_ERR(pol);
990                 pol = NULL;
991         }
992         if (pol || err)
993                 goto end;
994 #endif
995         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
996         if (IS_ERR(pol)) {
997                 err = PTR_ERR(pol);
998                 pol = NULL;
999         }
1000 #ifdef CONFIG_XFRM_SUB_POLICY
1001 end:
1002 #endif
1003         if ((*objp = (void *) pol) != NULL)
1004                 *obj_refp = &pol->refcnt;
1005         return err;
1006 }
1007
1008 static inline int policy_to_flow_dir(int dir)
1009 {
1010         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1011             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1012             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1013                 return dir;
1014         switch (dir) {
1015         default:
1016         case XFRM_POLICY_IN:
1017                 return FLOW_DIR_IN;
1018         case XFRM_POLICY_OUT:
1019                 return FLOW_DIR_OUT;
1020         case XFRM_POLICY_FWD:
1021                 return FLOW_DIR_FWD;
1022         }
1023 }
1024
1025 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1026 {
1027         struct xfrm_policy *pol;
1028
1029         read_lock_bh(&xfrm_policy_lock);
1030         if ((pol = sk->sk_policy[dir]) != NULL) {
1031                 int match = xfrm_selector_match(&pol->selector, fl,
1032                                                 sk->sk_family);
1033                 int err = 0;
1034
1035                 if (match) {
1036                         err = security_xfrm_policy_lookup(pol->security,
1037                                                       fl->secid,
1038                                                       policy_to_flow_dir(dir));
1039                         if (!err)
1040                                 xfrm_pol_hold(pol);
1041                         else if (err == -ESRCH)
1042                                 pol = NULL;
1043                         else
1044                                 pol = ERR_PTR(err);
1045                 } else
1046                         pol = NULL;
1047         }
1048         read_unlock_bh(&xfrm_policy_lock);
1049         return pol;
1050 }
1051
1052 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1053 {
1054         struct net *net = xp_net(pol);
1055         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1056                                                      pol->family, dir);
1057
1058         list_add(&pol->walk.all, &net->xfrm.policy_all);
1059         hlist_add_head(&pol->bydst, chain);
1060         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1061         net->xfrm.policy_count[dir]++;
1062         xfrm_pol_hold(pol);
1063
1064         if (xfrm_bydst_should_resize(net, dir, NULL))
1065                 schedule_work(&net->xfrm.policy_hash_work);
1066 }
1067
1068 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1069                                                 int dir)
1070 {
1071         struct net *net = xp_net(pol);
1072
1073         if (hlist_unhashed(&pol->bydst))
1074                 return NULL;
1075
1076         hlist_del(&pol->bydst);
1077         hlist_del(&pol->byidx);
1078         list_del(&pol->walk.all);
1079         net->xfrm.policy_count[dir]--;
1080
1081         return pol;
1082 }
1083
1084 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1085 {
1086         write_lock_bh(&xfrm_policy_lock);
1087         pol = __xfrm_policy_unlink(pol, dir);
1088         write_unlock_bh(&xfrm_policy_lock);
1089         if (pol) {
1090                 if (dir < XFRM_POLICY_MAX)
1091                         atomic_inc(&flow_cache_genid);
1092                 xfrm_policy_kill(pol);
1093                 return 0;
1094         }
1095         return -ENOENT;
1096 }
1097 EXPORT_SYMBOL(xfrm_policy_delete);
1098
1099 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1100 {
1101         struct net *net = xp_net(pol);
1102         struct xfrm_policy *old_pol;
1103
1104 #ifdef CONFIG_XFRM_SUB_POLICY
1105         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1106                 return -EINVAL;
1107 #endif
1108
1109         write_lock_bh(&xfrm_policy_lock);
1110         old_pol = sk->sk_policy[dir];
1111         sk->sk_policy[dir] = pol;
1112         if (pol) {
1113                 pol->curlft.add_time = get_seconds();
1114                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1115                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1116         }
1117         if (old_pol)
1118                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1119         write_unlock_bh(&xfrm_policy_lock);
1120
1121         if (old_pol) {
1122                 xfrm_policy_kill(old_pol);
1123         }
1124         return 0;
1125 }
1126
1127 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1128 {
1129         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1130
1131         if (newp) {
1132                 newp->selector = old->selector;
1133                 if (security_xfrm_policy_clone(old->security,
1134                                                &newp->security)) {
1135                         kfree(newp);
1136                         return NULL;  /* ENOMEM */
1137                 }
1138                 newp->lft = old->lft;
1139                 newp->curlft = old->curlft;
1140                 newp->action = old->action;
1141                 newp->flags = old->flags;
1142                 newp->xfrm_nr = old->xfrm_nr;
1143                 newp->index = old->index;
1144                 newp->type = old->type;
1145                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1146                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1147                 write_lock_bh(&xfrm_policy_lock);
1148                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1149                 write_unlock_bh(&xfrm_policy_lock);
1150                 xfrm_pol_put(newp);
1151         }
1152         return newp;
1153 }
1154
1155 int __xfrm_sk_clone_policy(struct sock *sk)
1156 {
1157         struct xfrm_policy *p0 = sk->sk_policy[0],
1158                            *p1 = sk->sk_policy[1];
1159
1160         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1161         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1162                 return -ENOMEM;
1163         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1164                 return -ENOMEM;
1165         return 0;
1166 }
1167
1168 static int
1169 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1170                unsigned short family)
1171 {
1172         int err;
1173         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1174
1175         if (unlikely(afinfo == NULL))
1176                 return -EINVAL;
1177         err = afinfo->get_saddr(net, local, remote);
1178         xfrm_policy_put_afinfo(afinfo);
1179         return err;
1180 }
1181
1182 /* Resolve list of templates for the flow, given policy. */
1183
1184 static int
1185 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1186                       struct xfrm_state **xfrm,
1187                       unsigned short family)
1188 {
1189         struct net *net = xp_net(policy);
1190         int nx;
1191         int i, error;
1192         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1193         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1194         xfrm_address_t tmp;
1195
1196         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1197                 struct xfrm_state *x;
1198                 xfrm_address_t *remote = daddr;
1199                 xfrm_address_t *local  = saddr;
1200                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1201
1202                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1203                     tmpl->mode == XFRM_MODE_BEET) {
1204                         remote = &tmpl->id.daddr;
1205                         local = &tmpl->saddr;
1206                         family = tmpl->encap_family;
1207                         if (xfrm_addr_any(local, family)) {
1208                                 error = xfrm_get_saddr(net, &tmp, remote, family);
1209                                 if (error)
1210                                         goto fail;
1211                                 local = &tmp;
1212                         }
1213                 }
1214
1215                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1216
1217                 if (x && x->km.state == XFRM_STATE_VALID) {
1218                         xfrm[nx++] = x;
1219                         daddr = remote;
1220                         saddr = local;
1221                         continue;
1222                 }
1223                 if (x) {
1224                         error = (x->km.state == XFRM_STATE_ERROR ?
1225                                  -EINVAL : -EAGAIN);
1226                         xfrm_state_put(x);
1227                 }
1228                 else if (error == -ESRCH)
1229                         error = -EAGAIN;
1230
1231                 if (!tmpl->optional)
1232                         goto fail;
1233         }
1234         return nx;
1235
1236 fail:
1237         for (nx--; nx>=0; nx--)
1238                 xfrm_state_put(xfrm[nx]);
1239         return error;
1240 }
1241
1242 static int
1243 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1244                   struct xfrm_state **xfrm,
1245                   unsigned short family)
1246 {
1247         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1248         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1249         int cnx = 0;
1250         int error;
1251         int ret;
1252         int i;
1253
1254         for (i = 0; i < npols; i++) {
1255                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1256                         error = -ENOBUFS;
1257                         goto fail;
1258                 }
1259
1260                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1261                 if (ret < 0) {
1262                         error = ret;
1263                         goto fail;
1264                 } else
1265                         cnx += ret;
1266         }
1267
1268         /* found states are sorted for outbound processing */
1269         if (npols > 1)
1270                 xfrm_state_sort(xfrm, tpp, cnx, family);
1271
1272         return cnx;
1273
1274  fail:
1275         for (cnx--; cnx>=0; cnx--)
1276                 xfrm_state_put(tpp[cnx]);
1277         return error;
1278
1279 }
1280
1281 /* Check that the bundle accepts the flow and its components are
1282  * still valid.
1283  */
1284
1285 static struct dst_entry *
1286 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1287 {
1288         struct dst_entry *x;
1289         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1290         if (unlikely(afinfo == NULL))
1291                 return ERR_PTR(-EINVAL);
1292         x = afinfo->find_bundle(fl, policy);
1293         xfrm_policy_put_afinfo(afinfo);
1294         return x;
1295 }
1296
1297 static inline int xfrm_get_tos(struct flowi *fl, int family)
1298 {
1299         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1300         int tos;
1301
1302         if (!afinfo)
1303                 return -EINVAL;
1304
1305         tos = afinfo->get_tos(fl);
1306
1307         xfrm_policy_put_afinfo(afinfo);
1308
1309         return tos;
1310 }
1311
1312 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1313 {
1314         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1315         struct xfrm_dst *xdst;
1316
1317         if (!afinfo)
1318                 return ERR_PTR(-EINVAL);
1319
1320         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1321
1322         xfrm_policy_put_afinfo(afinfo);
1323
1324         return xdst;
1325 }
1326
1327 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1328                                  int nfheader_len)
1329 {
1330         struct xfrm_policy_afinfo *afinfo =
1331                 xfrm_policy_get_afinfo(dst->ops->family);
1332         int err;
1333
1334         if (!afinfo)
1335                 return -EINVAL;
1336
1337         err = afinfo->init_path(path, dst, nfheader_len);
1338
1339         xfrm_policy_put_afinfo(afinfo);
1340
1341         return err;
1342 }
1343
1344 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1345 {
1346         struct xfrm_policy_afinfo *afinfo =
1347                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1348         int err;
1349
1350         if (!afinfo)
1351                 return -EINVAL;
1352
1353         err = afinfo->fill_dst(xdst, dev);
1354
1355         xfrm_policy_put_afinfo(afinfo);
1356
1357         return err;
1358 }
1359
1360 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1361  * all the metrics... Shortly, bundle a bundle.
1362  */
1363
1364 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1365                                             struct xfrm_state **xfrm, int nx,
1366                                             struct flowi *fl,
1367                                             struct dst_entry *dst)
1368 {
1369         unsigned long now = jiffies;
1370         struct net_device *dev;
1371         struct dst_entry *dst_prev = NULL;
1372         struct dst_entry *dst0 = NULL;
1373         int i = 0;
1374         int err;
1375         int header_len = 0;
1376         int nfheader_len = 0;
1377         int trailer_len = 0;
1378         int tos;
1379         int family = policy->selector.family;
1380         xfrm_address_t saddr, daddr;
1381
1382         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1383
1384         tos = xfrm_get_tos(fl, family);
1385         err = tos;
1386         if (tos < 0)
1387                 goto put_states;
1388
1389         dst_hold(dst);
1390
1391         for (; i < nx; i++) {
1392                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1393                 struct dst_entry *dst1 = &xdst->u.dst;
1394
1395                 err = PTR_ERR(xdst);
1396                 if (IS_ERR(xdst)) {
1397                         dst_release(dst);
1398                         goto put_states;
1399                 }
1400
1401                 if (!dst_prev)
1402                         dst0 = dst1;
1403                 else {
1404                         dst_prev->child = dst_clone(dst1);
1405                         dst1->flags |= DST_NOHASH;
1406                 }
1407
1408                 xdst->route = dst;
1409                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1410
1411                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1412                         family = xfrm[i]->props.family;
1413                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1414                                               family);
1415                         err = PTR_ERR(dst);
1416                         if (IS_ERR(dst))
1417                                 goto put_states;
1418                 } else
1419                         dst_hold(dst);
1420
1421                 dst1->xfrm = xfrm[i];
1422                 xdst->genid = xfrm[i]->genid;
1423
1424                 dst1->obsolete = -1;
1425                 dst1->flags |= DST_HOST;
1426                 dst1->lastuse = now;
1427
1428                 dst1->input = dst_discard;
1429                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1430
1431                 dst1->next = dst_prev;
1432                 dst_prev = dst1;
1433
1434                 header_len += xfrm[i]->props.header_len;
1435                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1436                         nfheader_len += xfrm[i]->props.header_len;
1437                 trailer_len += xfrm[i]->props.trailer_len;
1438         }
1439
1440         dst_prev->child = dst;
1441         dst0->path = dst;
1442
1443         err = -ENODEV;
1444         dev = dst->dev;
1445         if (!dev)
1446                 goto free_dst;
1447
1448         /* Copy neighbout for reachability confirmation */
1449         dst0->neighbour = neigh_clone(dst->neighbour);
1450
1451         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1452         xfrm_init_pmtu(dst_prev);
1453
1454         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1455                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1456
1457                 err = xfrm_fill_dst(xdst, dev);
1458                 if (err)
1459                         goto free_dst;
1460
1461                 dst_prev->header_len = header_len;
1462                 dst_prev->trailer_len = trailer_len;
1463                 header_len -= xdst->u.dst.xfrm->props.header_len;
1464                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1465         }
1466
1467 out:
1468         return dst0;
1469
1470 put_states:
1471         for (; i < nx; i++)
1472                 xfrm_state_put(xfrm[i]);
1473 free_dst:
1474         if (dst0)
1475                 dst_free(dst0);
1476         dst0 = ERR_PTR(err);
1477         goto out;
1478 }
1479
1480 static int inline
1481 xfrm_dst_alloc_copy(void **target, void *src, int size)
1482 {
1483         if (!*target) {
1484                 *target = kmalloc(size, GFP_ATOMIC);
1485                 if (!*target)
1486                         return -ENOMEM;
1487         }
1488         memcpy(*target, src, size);
1489         return 0;
1490 }
1491
1492 static int inline
1493 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1494 {
1495 #ifdef CONFIG_XFRM_SUB_POLICY
1496         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1497         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1498                                    sel, sizeof(*sel));
1499 #else
1500         return 0;
1501 #endif
1502 }
1503
1504 static int inline
1505 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1506 {
1507 #ifdef CONFIG_XFRM_SUB_POLICY
1508         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1509         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1510 #else
1511         return 0;
1512 #endif
1513 }
1514
1515 static int stale_bundle(struct dst_entry *dst);
1516
1517 /* Main function: finds/creates a bundle for given flow.
1518  *
1519  * At the moment we eat a raw IP route. Mostly to speed up lookups
1520  * on interfaces with disabled IPsec.
1521  */
1522 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1523                   struct sock *sk, int flags)
1524 {
1525         struct xfrm_policy *policy;
1526         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1527         int npols;
1528         int pol_dead;
1529         int xfrm_nr;
1530         int pi;
1531         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1532         struct dst_entry *dst, *dst_orig = *dst_p;
1533         int nx = 0;
1534         int err;
1535         u32 genid;
1536         u16 family;
1537         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1538
1539 restart:
1540         genid = atomic_read(&flow_cache_genid);
1541         policy = NULL;
1542         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1543                 pols[pi] = NULL;
1544         npols = 0;
1545         pol_dead = 0;
1546         xfrm_nr = 0;
1547
1548         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1549                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1550                 err = PTR_ERR(policy);
1551                 if (IS_ERR(policy)) {
1552                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1553                         goto dropdst;
1554                 }
1555         }
1556
1557         if (!policy) {
1558                 /* To accelerate a bit...  */
1559                 if ((dst_orig->flags & DST_NOXFRM) ||
1560                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1561                         goto nopol;
1562
1563                 policy = flow_cache_lookup(net, fl, dst_orig->ops->family,
1564                                            dir, xfrm_policy_lookup);
1565                 err = PTR_ERR(policy);
1566                 if (IS_ERR(policy)) {
1567                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1568                         goto dropdst;
1569                 }
1570         }
1571
1572         if (!policy)
1573                 goto nopol;
1574
1575         family = dst_orig->ops->family;
1576         pols[0] = policy;
1577         npols ++;
1578         xfrm_nr += pols[0]->xfrm_nr;
1579
1580         err = -ENOENT;
1581         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1582                 goto error;
1583
1584         policy->curlft.use_time = get_seconds();
1585
1586         switch (policy->action) {
1587         default:
1588         case XFRM_POLICY_BLOCK:
1589                 /* Prohibit the flow */
1590                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1591                 err = -EPERM;
1592                 goto error;
1593
1594         case XFRM_POLICY_ALLOW:
1595 #ifndef CONFIG_XFRM_SUB_POLICY
1596                 if (policy->xfrm_nr == 0) {
1597                         /* Flow passes not transformed. */
1598                         xfrm_pol_put(policy);
1599                         return 0;
1600                 }
1601 #endif
1602
1603                 /* Try to find matching bundle.
1604                  *
1605                  * LATER: help from flow cache. It is optional, this
1606                  * is required only for output policy.
1607                  */
1608                 dst = xfrm_find_bundle(fl, policy, family);
1609                 if (IS_ERR(dst)) {
1610                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1611                         err = PTR_ERR(dst);
1612                         goto error;
1613                 }
1614
1615                 if (dst)
1616                         break;
1617
1618 #ifdef CONFIG_XFRM_SUB_POLICY
1619                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1620                         pols[1] = xfrm_policy_lookup_bytype(net,
1621                                                             XFRM_POLICY_TYPE_MAIN,
1622                                                             fl, family,
1623                                                             XFRM_POLICY_OUT);
1624                         if (pols[1]) {
1625                                 if (IS_ERR(pols[1])) {
1626                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1627                                         err = PTR_ERR(pols[1]);
1628                                         goto error;
1629                                 }
1630                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1631                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1632                                         err = -EPERM;
1633                                         goto error;
1634                                 }
1635                                 npols ++;
1636                                 xfrm_nr += pols[1]->xfrm_nr;
1637                         }
1638                 }
1639
1640                 /*
1641                  * Because neither flowi nor bundle information knows about
1642                  * transformation template size. On more than one policy usage
1643                  * we can realize whether all of them is bypass or not after
1644                  * they are searched. See above not-transformed bypass
1645                  * is surrounded by non-sub policy configuration, too.
1646                  */
1647                 if (xfrm_nr == 0) {
1648                         /* Flow passes not transformed. */
1649                         xfrm_pols_put(pols, npols);
1650                         return 0;
1651                 }
1652
1653 #endif
1654                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1655
1656                 if (unlikely(nx<0)) {
1657                         err = nx;
1658                         if (err == -EAGAIN && net->xfrm.sysctl_larval_drop) {
1659                                 /* EREMOTE tells the caller to generate
1660                                  * a one-shot blackhole route.
1661                                  */
1662                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1663                                 xfrm_pol_put(policy);
1664                                 return -EREMOTE;
1665                         }
1666                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1667                                 DECLARE_WAITQUEUE(wait, current);
1668
1669                                 add_wait_queue(&net->xfrm.km_waitq, &wait);
1670                                 set_current_state(TASK_INTERRUPTIBLE);
1671                                 schedule();
1672                                 set_current_state(TASK_RUNNING);
1673                                 remove_wait_queue(&net->xfrm.km_waitq, &wait);
1674
1675                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1676
1677                                 if (nx == -EAGAIN && signal_pending(current)) {
1678                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1679                                         err = -ERESTART;
1680                                         goto error;
1681                                 }
1682                                 if (nx == -EAGAIN ||
1683                                     genid != atomic_read(&flow_cache_genid)) {
1684                                         xfrm_pols_put(pols, npols);
1685                                         goto restart;
1686                                 }
1687                                 err = nx;
1688                         }
1689                         if (err < 0) {
1690                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1691                                 goto error;
1692                         }
1693                 }
1694                 if (nx == 0) {
1695                         /* Flow passes not transformed. */
1696                         xfrm_pols_put(pols, npols);
1697                         return 0;
1698                 }
1699
1700                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1701                 err = PTR_ERR(dst);
1702                 if (IS_ERR(dst)) {
1703                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1704                         goto error;
1705                 }
1706
1707                 for (pi = 0; pi < npols; pi++) {
1708                         read_lock_bh(&pols[pi]->lock);
1709                         pol_dead |= pols[pi]->walk.dead;
1710                         read_unlock_bh(&pols[pi]->lock);
1711                 }
1712
1713                 write_lock_bh(&policy->lock);
1714                 if (unlikely(pol_dead || stale_bundle(dst))) {
1715                         /* Wow! While we worked on resolving, this
1716                          * policy has gone. Retry. It is not paranoia,
1717                          * we just cannot enlist new bundle to dead object.
1718                          * We can't enlist stable bundles either.
1719                          */
1720                         write_unlock_bh(&policy->lock);
1721                         dst_free(dst);
1722
1723                         if (pol_dead)
1724                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLDEAD);
1725                         else
1726                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1727                         err = -EHOSTUNREACH;
1728                         goto error;
1729                 }
1730
1731                 if (npols > 1)
1732                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1733                 else
1734                         err = xfrm_dst_update_origin(dst, fl);
1735                 if (unlikely(err)) {
1736                         write_unlock_bh(&policy->lock);
1737                         dst_free(dst);
1738                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1739                         goto error;
1740                 }
1741
1742                 dst->next = policy->bundles;
1743                 policy->bundles = dst;
1744                 dst_hold(dst);
1745                 write_unlock_bh(&policy->lock);
1746         }
1747         *dst_p = dst;
1748         dst_release(dst_orig);
1749         xfrm_pols_put(pols, npols);
1750         return 0;
1751
1752 error:
1753         xfrm_pols_put(pols, npols);
1754 dropdst:
1755         dst_release(dst_orig);
1756         *dst_p = NULL;
1757         return err;
1758
1759 nopol:
1760         err = -ENOENT;
1761         if (flags & XFRM_LOOKUP_ICMP)
1762                 goto dropdst;
1763         return 0;
1764 }
1765 EXPORT_SYMBOL(__xfrm_lookup);
1766
1767 int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1768                 struct sock *sk, int flags)
1769 {
1770         int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1771
1772         if (err == -EREMOTE) {
1773                 dst_release(*dst_p);
1774                 *dst_p = NULL;
1775                 err = -EAGAIN;
1776         }
1777
1778         return err;
1779 }
1780 EXPORT_SYMBOL(xfrm_lookup);
1781
1782 static inline int
1783 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1784 {
1785         struct xfrm_state *x;
1786
1787         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1788                 return 0;
1789         x = skb->sp->xvec[idx];
1790         if (!x->type->reject)
1791                 return 0;
1792         return x->type->reject(x, skb, fl);
1793 }
1794
1795 /* When skb is transformed back to its "native" form, we have to
1796  * check policy restrictions. At the moment we make this in maximally
1797  * stupid way. Shame on me. :-) Of course, connected sockets must
1798  * have policy cached at them.
1799  */
1800
1801 static inline int
1802 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1803               unsigned short family)
1804 {
1805         if (xfrm_state_kern(x))
1806                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1807         return  x->id.proto == tmpl->id.proto &&
1808                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1809                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1810                 x->props.mode == tmpl->mode &&
1811                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1812                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1813                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1814                   xfrm_state_addr_cmp(tmpl, x, family));
1815 }
1816
1817 /*
1818  * 0 or more than 0 is returned when validation is succeeded (either bypass
1819  * because of optional transport mode, or next index of the mathced secpath
1820  * state with the template.
1821  * -1 is returned when no matching template is found.
1822  * Otherwise "-2 - errored_index" is returned.
1823  */
1824 static inline int
1825 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1826                unsigned short family)
1827 {
1828         int idx = start;
1829
1830         if (tmpl->optional) {
1831                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1832                         return start;
1833         } else
1834                 start = -1;
1835         for (; idx < sp->len; idx++) {
1836                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1837                         return ++idx;
1838                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1839                         if (start == -1)
1840                                 start = -2-idx;
1841                         break;
1842                 }
1843         }
1844         return start;
1845 }
1846
1847 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1848                           unsigned int family, int reverse)
1849 {
1850         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1851         int err;
1852
1853         if (unlikely(afinfo == NULL))
1854                 return -EAFNOSUPPORT;
1855
1856         afinfo->decode_session(skb, fl, reverse);
1857         err = security_xfrm_decode_session(skb, &fl->secid);
1858         xfrm_policy_put_afinfo(afinfo);
1859         return err;
1860 }
1861 EXPORT_SYMBOL(__xfrm_decode_session);
1862
1863 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1864 {
1865         for (; k < sp->len; k++) {
1866                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1867                         *idxp = k;
1868                         return 1;
1869                 }
1870         }
1871
1872         return 0;
1873 }
1874
1875 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1876                         unsigned short family)
1877 {
1878         struct net *net = dev_net(skb->dev);
1879         struct xfrm_policy *pol;
1880         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1881         int npols = 0;
1882         int xfrm_nr;
1883         int pi;
1884         int reverse;
1885         struct flowi fl;
1886         u8 fl_dir;
1887         int xerr_idx = -1;
1888
1889         reverse = dir & ~XFRM_POLICY_MASK;
1890         dir &= XFRM_POLICY_MASK;
1891         fl_dir = policy_to_flow_dir(dir);
1892
1893         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1894                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
1895                 return 0;
1896         }
1897
1898         nf_nat_decode_session(skb, &fl, family);
1899
1900         /* First, check used SA against their selectors. */
1901         if (skb->sp) {
1902                 int i;
1903
1904                 for (i=skb->sp->len-1; i>=0; i--) {
1905                         struct xfrm_state *x = skb->sp->xvec[i];
1906                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1907                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
1908                                 return 0;
1909                         }
1910                 }
1911         }
1912
1913         pol = NULL;
1914         if (sk && sk->sk_policy[dir]) {
1915                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1916                 if (IS_ERR(pol)) {
1917                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1918                         return 0;
1919                 }
1920         }
1921
1922         if (!pol)
1923                 pol = flow_cache_lookup(net, &fl, family, fl_dir,
1924                                         xfrm_policy_lookup);
1925
1926         if (IS_ERR(pol)) {
1927                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1928                 return 0;
1929         }
1930
1931         if (!pol) {
1932                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1933                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1934                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
1935                         return 0;
1936                 }
1937                 return 1;
1938         }
1939
1940         pol->curlft.use_time = get_seconds();
1941
1942         pols[0] = pol;
1943         npols ++;
1944 #ifdef CONFIG_XFRM_SUB_POLICY
1945         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1946                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
1947                                                     &fl, family,
1948                                                     XFRM_POLICY_IN);
1949                 if (pols[1]) {
1950                         if (IS_ERR(pols[1])) {
1951                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1952                                 return 0;
1953                         }
1954                         pols[1]->curlft.use_time = get_seconds();
1955                         npols ++;
1956                 }
1957         }
1958 #endif
1959
1960         if (pol->action == XFRM_POLICY_ALLOW) {
1961                 struct sec_path *sp;
1962                 static struct sec_path dummy;
1963                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1964                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1965                 struct xfrm_tmpl **tpp = tp;
1966                 int ti = 0;
1967                 int i, k;
1968
1969                 if ((sp = skb->sp) == NULL)
1970                         sp = &dummy;
1971
1972                 for (pi = 0; pi < npols; pi++) {
1973                         if (pols[pi] != pol &&
1974                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1975                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
1976                                 goto reject;
1977                         }
1978                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1979                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
1980                                 goto reject_error;
1981                         }
1982                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1983                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1984                 }
1985                 xfrm_nr = ti;
1986                 if (npols > 1) {
1987                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1988                         tpp = stp;
1989                 }
1990
1991                 /* For each tunnel xfrm, find the first matching tmpl.
1992                  * For each tmpl before that, find corresponding xfrm.
1993                  * Order is _important_. Later we will implement
1994                  * some barriers, but at the moment barriers
1995                  * are implied between each two transformations.
1996                  */
1997                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1998                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1999                         if (k < 0) {
2000                                 if (k < -1)
2001                                         /* "-2 - errored_index" returned */
2002                                         xerr_idx = -(2+k);
2003                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2004                                 goto reject;
2005                         }
2006                 }
2007
2008                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2009                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2010                         goto reject;
2011                 }
2012
2013                 xfrm_pols_put(pols, npols);
2014                 return 1;
2015         }
2016         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2017
2018 reject:
2019         xfrm_secpath_reject(xerr_idx, skb, &fl);
2020 reject_error:
2021         xfrm_pols_put(pols, npols);
2022         return 0;
2023 }
2024 EXPORT_SYMBOL(__xfrm_policy_check);
2025
2026 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2027 {
2028         struct net *net = dev_net(skb->dev);
2029         struct flowi fl;
2030         struct dst_entry *dst;
2031         int res;
2032
2033         if (xfrm_decode_session(skb, &fl, family) < 0) {
2034                 /* XXX: we should have something like FWDHDRERROR here. */
2035                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2036                 return 0;
2037         }
2038
2039         dst = skb_dst(skb);
2040
2041         res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
2042         skb_dst_set(skb, dst);
2043         return res;
2044 }
2045 EXPORT_SYMBOL(__xfrm_route_forward);
2046
2047 /* Optimize later using cookies and generation ids. */
2048
2049 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2050 {
2051         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2052          * to "-1" to force all XFRM destinations to get validated by
2053          * dst_ops->check on every use.  We do this because when a
2054          * normal route referenced by an XFRM dst is obsoleted we do
2055          * not go looking around for all parent referencing XFRM dsts
2056          * so that we can invalidate them.  It is just too much work.
2057          * Instead we make the checks here on every use.  For example:
2058          *
2059          *      XFRM dst A --> IPv4 dst X
2060          *
2061          * X is the "xdst->route" of A (X is also the "dst->path" of A
2062          * in this example).  If X is marked obsolete, "A" will not
2063          * notice.  That's what we are validating here via the
2064          * stale_bundle() check.
2065          *
2066          * When a policy's bundle is pruned, we dst_free() the XFRM
2067          * dst which causes it's ->obsolete field to be set to a
2068          * positive non-zero integer.  If an XFRM dst has been pruned
2069          * like this, we want to force a new route lookup.
2070          */
2071         if (dst->obsolete < 0 && !stale_bundle(dst))
2072                 return dst;
2073
2074         return NULL;
2075 }
2076
2077 static int stale_bundle(struct dst_entry *dst)
2078 {
2079         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2080 }
2081
2082 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2083 {
2084         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2085                 dst->dev = dev_net(dev)->loopback_dev;
2086                 dev_hold(dst->dev);
2087                 dev_put(dev);
2088         }
2089 }
2090 EXPORT_SYMBOL(xfrm_dst_ifdown);
2091
2092 static void xfrm_link_failure(struct sk_buff *skb)
2093 {
2094         /* Impossible. Such dst must be popped before reaches point of failure. */
2095         return;
2096 }
2097
2098 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2099 {
2100         if (dst) {
2101                 if (dst->obsolete) {
2102                         dst_release(dst);
2103                         dst = NULL;
2104                 }
2105         }
2106         return dst;
2107 }
2108
2109 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2110 {
2111         struct dst_entry *dst, **dstp;
2112
2113         write_lock(&pol->lock);
2114         dstp = &pol->bundles;
2115         while ((dst=*dstp) != NULL) {
2116                 if (func(dst)) {
2117                         *dstp = dst->next;
2118                         dst->next = *gc_list_p;
2119                         *gc_list_p = dst;
2120                 } else {
2121                         dstp = &dst->next;
2122                 }
2123         }
2124         write_unlock(&pol->lock);
2125 }
2126
2127 static void xfrm_prune_bundles(struct net *net, int (*func)(struct dst_entry *))
2128 {
2129         struct dst_entry *gc_list = NULL;
2130         int dir;
2131
2132         read_lock_bh(&xfrm_policy_lock);
2133         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2134                 struct xfrm_policy *pol;
2135                 struct hlist_node *entry;
2136                 struct hlist_head *table;
2137                 int i;
2138
2139                 hlist_for_each_entry(pol, entry,
2140                                      &net->xfrm.policy_inexact[dir], bydst)
2141                         prune_one_bundle(pol, func, &gc_list);
2142
2143                 table = net->xfrm.policy_bydst[dir].table;
2144                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2145                         hlist_for_each_entry(pol, entry, table + i, bydst)
2146                                 prune_one_bundle(pol, func, &gc_list);
2147                 }
2148         }
2149         read_unlock_bh(&xfrm_policy_lock);
2150
2151         while (gc_list) {
2152                 struct dst_entry *dst = gc_list;
2153                 gc_list = dst->next;
2154                 dst_free(dst);
2155         }
2156 }
2157
2158 static int unused_bundle(struct dst_entry *dst)
2159 {
2160         return !atomic_read(&dst->__refcnt);
2161 }
2162
2163 static void __xfrm_garbage_collect(struct net *net)
2164 {
2165         xfrm_prune_bundles(net, unused_bundle);
2166 }
2167
2168 static int xfrm_flush_bundles(struct net *net)
2169 {
2170         xfrm_prune_bundles(net, stale_bundle);
2171         return 0;
2172 }
2173
2174 static void xfrm_init_pmtu(struct dst_entry *dst)
2175 {
2176         do {
2177                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2178                 u32 pmtu, route_mtu_cached;
2179
2180                 pmtu = dst_mtu(dst->child);
2181                 xdst->child_mtu_cached = pmtu;
2182
2183                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2184
2185                 route_mtu_cached = dst_mtu(xdst->route);
2186                 xdst->route_mtu_cached = route_mtu_cached;
2187
2188                 if (pmtu > route_mtu_cached)
2189                         pmtu = route_mtu_cached;
2190
2191                 dst->metrics[RTAX_MTU-1] = pmtu;
2192         } while ((dst = dst->next));
2193 }
2194
2195 /* Check that the bundle accepts the flow and its components are
2196  * still valid.
2197  */
2198
2199 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2200                 struct flowi *fl, int family, int strict)
2201 {
2202         struct dst_entry *dst = &first->u.dst;
2203         struct xfrm_dst *last;
2204         u32 mtu;
2205
2206         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2207             (dst->dev && !netif_running(dst->dev)))
2208                 return 0;
2209 #ifdef CONFIG_XFRM_SUB_POLICY
2210         if (fl) {
2211                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2212                         return 0;
2213                 if (first->partner &&
2214                     !xfrm_selector_match(first->partner, fl, family))
2215                         return 0;
2216         }
2217 #endif
2218
2219         last = NULL;
2220
2221         do {
2222                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2223
2224                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2225                         return 0;
2226                 if (fl && pol &&
2227                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2228                         return 0;
2229                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2230                         return 0;
2231                 if (xdst->genid != dst->xfrm->genid)
2232                         return 0;
2233
2234                 if (strict && fl &&
2235                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2236                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2237                         return 0;
2238
2239                 mtu = dst_mtu(dst->child);
2240                 if (xdst->child_mtu_cached != mtu) {
2241                         last = xdst;
2242                         xdst->child_mtu_cached = mtu;
2243                 }
2244
2245                 if (!dst_check(xdst->route, xdst->route_cookie))
2246                         return 0;
2247                 mtu = dst_mtu(xdst->route);
2248                 if (xdst->route_mtu_cached != mtu) {
2249                         last = xdst;
2250                         xdst->route_mtu_cached = mtu;
2251                 }
2252
2253                 dst = dst->child;
2254         } while (dst->xfrm);
2255
2256         if (likely(!last))
2257                 return 1;
2258
2259         mtu = last->child_mtu_cached;
2260         for (;;) {
2261                 dst = &last->u.dst;
2262
2263                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2264                 if (mtu > last->route_mtu_cached)
2265                         mtu = last->route_mtu_cached;
2266                 dst->metrics[RTAX_MTU-1] = mtu;
2267
2268                 if (last == first)
2269                         break;
2270
2271                 last = (struct xfrm_dst *)last->u.dst.next;
2272                 last->child_mtu_cached = mtu;
2273         }
2274
2275         return 1;
2276 }
2277
2278 EXPORT_SYMBOL(xfrm_bundle_ok);
2279
2280 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2281 {
2282         int err = 0;
2283         if (unlikely(afinfo == NULL))
2284                 return -EINVAL;
2285         if (unlikely(afinfo->family >= NPROTO))
2286                 return -EAFNOSUPPORT;
2287         write_lock_bh(&xfrm_policy_afinfo_lock);
2288         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2289                 err = -ENOBUFS;
2290         else {
2291                 struct dst_ops *dst_ops = afinfo->dst_ops;
2292                 if (likely(dst_ops->kmem_cachep == NULL))
2293                         dst_ops->kmem_cachep = xfrm_dst_cache;
2294                 if (likely(dst_ops->check == NULL))
2295                         dst_ops->check = xfrm_dst_check;
2296                 if (likely(dst_ops->negative_advice == NULL))
2297                         dst_ops->negative_advice = xfrm_negative_advice;
2298                 if (likely(dst_ops->link_failure == NULL))
2299                         dst_ops->link_failure = xfrm_link_failure;
2300                 if (likely(afinfo->garbage_collect == NULL))
2301                         afinfo->garbage_collect = __xfrm_garbage_collect;
2302                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2303         }
2304         write_unlock_bh(&xfrm_policy_afinfo_lock);
2305         return err;
2306 }
2307 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2308
2309 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2310 {
2311         int err = 0;
2312         if (unlikely(afinfo == NULL))
2313                 return -EINVAL;
2314         if (unlikely(afinfo->family >= NPROTO))
2315                 return -EAFNOSUPPORT;
2316         write_lock_bh(&xfrm_policy_afinfo_lock);
2317         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2318                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2319                         err = -EINVAL;
2320                 else {
2321                         struct dst_ops *dst_ops = afinfo->dst_ops;
2322                         xfrm_policy_afinfo[afinfo->family] = NULL;
2323                         dst_ops->kmem_cachep = NULL;
2324                         dst_ops->check = NULL;
2325                         dst_ops->negative_advice = NULL;
2326                         dst_ops->link_failure = NULL;
2327                         afinfo->garbage_collect = NULL;
2328                 }
2329         }
2330         write_unlock_bh(&xfrm_policy_afinfo_lock);
2331         return err;
2332 }
2333 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2334
2335 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2336 {
2337         struct xfrm_policy_afinfo *afinfo;
2338         if (unlikely(family >= NPROTO))
2339                 return NULL;
2340         read_lock(&xfrm_policy_afinfo_lock);
2341         afinfo = xfrm_policy_afinfo[family];
2342         if (unlikely(!afinfo))
2343                 read_unlock(&xfrm_policy_afinfo_lock);
2344         return afinfo;
2345 }
2346
2347 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2348 {
2349         read_unlock(&xfrm_policy_afinfo_lock);
2350 }
2351
2352 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2353 {
2354         struct net_device *dev = ptr;
2355
2356         switch (event) {
2357         case NETDEV_DOWN:
2358                 xfrm_flush_bundles(dev_net(dev));
2359         }
2360         return NOTIFY_DONE;
2361 }
2362
2363 static struct notifier_block xfrm_dev_notifier = {
2364         .notifier_call  = xfrm_dev_event,
2365 };
2366
2367 #ifdef CONFIG_XFRM_STATISTICS
2368 static int __net_init xfrm_statistics_init(struct net *net)
2369 {
2370         int rv;
2371
2372         if (snmp_mib_init((void **)net->mib.xfrm_statistics,
2373                           sizeof(struct linux_xfrm_mib)) < 0)
2374                 return -ENOMEM;
2375         rv = xfrm_proc_init(net);
2376         if (rv < 0)
2377                 snmp_mib_free((void **)net->mib.xfrm_statistics);
2378         return rv;
2379 }
2380
2381 static void xfrm_statistics_fini(struct net *net)
2382 {
2383         xfrm_proc_fini(net);
2384         snmp_mib_free((void **)net->mib.xfrm_statistics);
2385 }
2386 #else
2387 static int __net_init xfrm_statistics_init(struct net *net)
2388 {
2389         return 0;
2390 }
2391
2392 static void xfrm_statistics_fini(struct net *net)
2393 {
2394 }
2395 #endif
2396
2397 static int __net_init xfrm_policy_init(struct net *net)
2398 {
2399         unsigned int hmask, sz;
2400         int dir;
2401
2402         if (net_eq(net, &init_net))
2403                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2404                                            sizeof(struct xfrm_dst),
2405                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2406                                            NULL);
2407
2408         hmask = 8 - 1;
2409         sz = (hmask+1) * sizeof(struct hlist_head);
2410
2411         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2412         if (!net->xfrm.policy_byidx)
2413                 goto out_byidx;
2414         net->xfrm.policy_idx_hmask = hmask;
2415
2416         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2417                 struct xfrm_policy_hash *htab;
2418
2419                 net->xfrm.policy_count[dir] = 0;
2420                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2421
2422                 htab = &net->xfrm.policy_bydst[dir];
2423                 htab->table = xfrm_hash_alloc(sz);
2424                 if (!htab->table)
2425                         goto out_bydst;
2426                 htab->hmask = hmask;
2427         }
2428
2429         INIT_LIST_HEAD(&net->xfrm.policy_all);
2430         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2431         if (net_eq(net, &init_net))
2432                 register_netdevice_notifier(&xfrm_dev_notifier);
2433         return 0;
2434
2435 out_bydst:
2436         for (dir--; dir >= 0; dir--) {
2437                 struct xfrm_policy_hash *htab;
2438
2439                 htab = &net->xfrm.policy_bydst[dir];
2440                 xfrm_hash_free(htab->table, sz);
2441         }
2442         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2443 out_byidx:
2444         return -ENOMEM;
2445 }
2446
2447 static void xfrm_policy_fini(struct net *net)
2448 {
2449         struct xfrm_audit audit_info;
2450         unsigned int sz;
2451         int dir;
2452
2453         flush_work(&net->xfrm.policy_hash_work);
2454 #ifdef CONFIG_XFRM_SUB_POLICY
2455         audit_info.loginuid = -1;
2456         audit_info.sessionid = -1;
2457         audit_info.secid = 0;
2458         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2459 #endif
2460         audit_info.loginuid = -1;
2461         audit_info.sessionid = -1;
2462         audit_info.secid = 0;
2463         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2464         flush_work(&xfrm_policy_gc_work);
2465
2466         WARN_ON(!list_empty(&net->xfrm.policy_all));
2467
2468         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2469                 struct xfrm_policy_hash *htab;
2470
2471                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2472
2473                 htab = &net->xfrm.policy_bydst[dir];
2474                 sz = (htab->hmask + 1);
2475                 WARN_ON(!hlist_empty(htab->table));
2476                 xfrm_hash_free(htab->table, sz);
2477         }
2478
2479         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2480         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2481         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2482 }
2483
2484 static int __net_init xfrm_net_init(struct net *net)
2485 {
2486         int rv;
2487
2488         rv = xfrm_statistics_init(net);
2489         if (rv < 0)
2490                 goto out_statistics;
2491         rv = xfrm_state_init(net);
2492         if (rv < 0)
2493                 goto out_state;
2494         rv = xfrm_policy_init(net);
2495         if (rv < 0)
2496                 goto out_policy;
2497         rv = xfrm_sysctl_init(net);
2498         if (rv < 0)
2499                 goto out_sysctl;
2500         return 0;
2501
2502 out_sysctl:
2503         xfrm_policy_fini(net);
2504 out_policy:
2505         xfrm_state_fini(net);
2506 out_state:
2507         xfrm_statistics_fini(net);
2508 out_statistics:
2509         return rv;
2510 }
2511
2512 static void __net_exit xfrm_net_exit(struct net *net)
2513 {
2514         xfrm_sysctl_fini(net);
2515         xfrm_policy_fini(net);
2516         xfrm_state_fini(net);
2517         xfrm_statistics_fini(net);
2518 }
2519
2520 static struct pernet_operations __net_initdata xfrm_net_ops = {
2521         .init = xfrm_net_init,
2522         .exit = xfrm_net_exit,
2523 };
2524
2525 void __init xfrm_init(void)
2526 {
2527         register_pernet_subsys(&xfrm_net_ops);
2528         xfrm_input_init();
2529 }
2530
2531 #ifdef CONFIG_AUDITSYSCALL
2532 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2533                                          struct audit_buffer *audit_buf)
2534 {
2535         struct xfrm_sec_ctx *ctx = xp->security;
2536         struct xfrm_selector *sel = &xp->selector;
2537
2538         if (ctx)
2539                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2540                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2541
2542         switch(sel->family) {
2543         case AF_INET:
2544                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2545                 if (sel->prefixlen_s != 32)
2546                         audit_log_format(audit_buf, " src_prefixlen=%d",
2547                                          sel->prefixlen_s);
2548                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2549                 if (sel->prefixlen_d != 32)
2550                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2551                                          sel->prefixlen_d);
2552                 break;
2553         case AF_INET6:
2554                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2555                 if (sel->prefixlen_s != 128)
2556                         audit_log_format(audit_buf, " src_prefixlen=%d",
2557                                          sel->prefixlen_s);
2558                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2559                 if (sel->prefixlen_d != 128)
2560                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2561                                          sel->prefixlen_d);
2562                 break;
2563         }
2564 }
2565
2566 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2567                            uid_t auid, u32 sessionid, u32 secid)
2568 {
2569         struct audit_buffer *audit_buf;
2570
2571         audit_buf = xfrm_audit_start("SPD-add");
2572         if (audit_buf == NULL)
2573                 return;
2574         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2575         audit_log_format(audit_buf, " res=%u", result);
2576         xfrm_audit_common_policyinfo(xp, audit_buf);
2577         audit_log_end(audit_buf);
2578 }
2579 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2580
2581 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2582                               uid_t auid, u32 sessionid, u32 secid)
2583 {
2584         struct audit_buffer *audit_buf;
2585
2586         audit_buf = xfrm_audit_start("SPD-delete");
2587         if (audit_buf == NULL)
2588                 return;
2589         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2590         audit_log_format(audit_buf, " res=%u", result);
2591         xfrm_audit_common_policyinfo(xp, audit_buf);
2592         audit_log_end(audit_buf);
2593 }
2594 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2595 #endif
2596
2597 #ifdef CONFIG_XFRM_MIGRATE
2598 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2599                                        struct xfrm_selector *sel_tgt)
2600 {
2601         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2602                 if (sel_tgt->family == sel_cmp->family &&
2603                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2604                                   sel_cmp->family) == 0 &&
2605                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2606                                   sel_cmp->family) == 0 &&
2607                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2608                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2609                         return 1;
2610                 }
2611         } else {
2612                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2613                         return 1;
2614                 }
2615         }
2616         return 0;
2617 }
2618
2619 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2620                                                      u8 dir, u8 type)
2621 {
2622         struct xfrm_policy *pol, *ret = NULL;
2623         struct hlist_node *entry;
2624         struct hlist_head *chain;
2625         u32 priority = ~0U;
2626
2627         read_lock_bh(&xfrm_policy_lock);
2628         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2629         hlist_for_each_entry(pol, entry, chain, bydst) {
2630                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2631                     pol->type == type) {
2632                         ret = pol;
2633                         priority = ret->priority;
2634                         break;
2635                 }
2636         }
2637         chain = &init_net.xfrm.policy_inexact[dir];
2638         hlist_for_each_entry(pol, entry, chain, bydst) {
2639                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2640                     pol->type == type &&
2641                     pol->priority < priority) {
2642                         ret = pol;
2643                         break;
2644                 }
2645         }
2646
2647         if (ret)
2648                 xfrm_pol_hold(ret);
2649
2650         read_unlock_bh(&xfrm_policy_lock);
2651
2652         return ret;
2653 }
2654
2655 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2656 {
2657         int match = 0;
2658
2659         if (t->mode == m->mode && t->id.proto == m->proto &&
2660             (m->reqid == 0 || t->reqid == m->reqid)) {
2661                 switch (t->mode) {
2662                 case XFRM_MODE_TUNNEL:
2663                 case XFRM_MODE_BEET:
2664                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2665                                           m->old_family) == 0 &&
2666                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2667                                           m->old_family) == 0) {
2668                                 match = 1;
2669                         }
2670                         break;
2671                 case XFRM_MODE_TRANSPORT:
2672                         /* in case of transport mode, template does not store
2673                            any IP addresses, hence we just compare mode and
2674                            protocol */
2675                         match = 1;
2676                         break;
2677                 default:
2678                         break;
2679                 }
2680         }
2681         return match;
2682 }
2683
2684 /* update endpoint address(es) of template(s) */
2685 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2686                                struct xfrm_migrate *m, int num_migrate)
2687 {
2688         struct xfrm_migrate *mp;
2689         struct dst_entry *dst;
2690         int i, j, n = 0;
2691
2692         write_lock_bh(&pol->lock);
2693         if (unlikely(pol->walk.dead)) {
2694                 /* target policy has been deleted */
2695                 write_unlock_bh(&pol->lock);
2696                 return -ENOENT;
2697         }
2698
2699         for (i = 0; i < pol->xfrm_nr; i++) {
2700                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2701                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2702                                 continue;
2703                         n++;
2704                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2705                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2706                                 continue;
2707                         /* update endpoints */
2708                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2709                                sizeof(pol->xfrm_vec[i].id.daddr));
2710                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2711                                sizeof(pol->xfrm_vec[i].saddr));
2712                         pol->xfrm_vec[i].encap_family = mp->new_family;
2713                         /* flush bundles */
2714                         while ((dst = pol->bundles) != NULL) {
2715                                 pol->bundles = dst->next;
2716                                 dst_free(dst);
2717                         }
2718                 }
2719         }
2720
2721         write_unlock_bh(&pol->lock);
2722
2723         if (!n)
2724                 return -ENODATA;
2725
2726         return 0;
2727 }
2728
2729 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2730 {
2731         int i, j;
2732
2733         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2734                 return -EINVAL;
2735
2736         for (i = 0; i < num_migrate; i++) {
2737                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2738                                    m[i].old_family) == 0) &&
2739                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2740                                    m[i].old_family) == 0))
2741                         return -EINVAL;
2742                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2743                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2744                         return -EINVAL;
2745
2746                 /* check if there is any duplicated entry */
2747                 for (j = i + 1; j < num_migrate; j++) {
2748                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2749                                     sizeof(m[i].old_daddr)) &&
2750                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2751                                     sizeof(m[i].old_saddr)) &&
2752                             m[i].proto == m[j].proto &&
2753                             m[i].mode == m[j].mode &&
2754                             m[i].reqid == m[j].reqid &&
2755                             m[i].old_family == m[j].old_family)
2756                                 return -EINVAL;
2757                 }
2758         }
2759
2760         return 0;
2761 }
2762
2763 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2764                  struct xfrm_migrate *m, int num_migrate,
2765                  struct xfrm_kmaddress *k)
2766 {
2767         int i, err, nx_cur = 0, nx_new = 0;
2768         struct xfrm_policy *pol = NULL;
2769         struct xfrm_state *x, *xc;
2770         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2771         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2772         struct xfrm_migrate *mp;
2773
2774         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2775                 goto out;
2776
2777         /* Stage 1 - find policy */
2778         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2779                 err = -ENOENT;
2780                 goto out;
2781         }
2782
2783         /* Stage 2 - find and update state(s) */
2784         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2785                 if ((x = xfrm_migrate_state_find(mp))) {
2786                         x_cur[nx_cur] = x;
2787                         nx_cur++;
2788                         if ((xc = xfrm_state_migrate(x, mp))) {
2789                                 x_new[nx_new] = xc;
2790                                 nx_new++;
2791                         } else {
2792                                 err = -ENODATA;
2793                                 goto restore_state;
2794                         }
2795                 }
2796         }
2797
2798         /* Stage 3 - update policy */
2799         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2800                 goto restore_state;
2801
2802         /* Stage 4 - delete old state(s) */
2803         if (nx_cur) {
2804                 xfrm_states_put(x_cur, nx_cur);
2805                 xfrm_states_delete(x_cur, nx_cur);
2806         }
2807
2808         /* Stage 5 - announce */
2809         km_migrate(sel, dir, type, m, num_migrate, k);
2810
2811         xfrm_pol_put(pol);
2812
2813         return 0;
2814 out:
2815         return err;
2816
2817 restore_state:
2818         if (pol)
2819                 xfrm_pol_put(pol);
2820         if (nx_cur)
2821                 xfrm_states_put(x_cur, nx_cur);
2822         if (nx_new)
2823                 xfrm_states_delete(x_new, nx_new);
2824
2825         return err;
2826 }
2827 EXPORT_SYMBOL(xfrm_migrate);
2828 #endif