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