xfrm: Const'ify address args to hash helpers.
[linux-3.10.git] / net / xfrm / xfrm_policy.c
1 /*
2  * xfrm_policy.c
3  *
4  * Changes:
5  *      Mitsuru KANDA @USAGI
6  *      Kazunori MIYAZAWA @USAGI
7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8  *              IPv6 support
9  *      Kazunori MIYAZAWA @USAGI
10  *      YOSHIFUJI Hideaki
11  *              Split up af-specific portion
12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
13  *
14  */
15
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
28 #include <net/dst.h>
29 #include <net/xfrm.h>
30 #include <net/ip.h>
31 #ifdef CONFIG_XFRM_STATISTICS
32 #include <net/snmp.h>
33 #endif
34
35 #include "xfrm_hash.h"
36
37 DEFINE_MUTEX(xfrm_cfg_mutex);
38 EXPORT_SYMBOL(xfrm_cfg_mutex);
39
40 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
41 static struct dst_entry *xfrm_policy_sk_bundles;
42 static DEFINE_RWLOCK(xfrm_policy_lock);
43
44 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
45 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
46
47 static struct kmem_cache *xfrm_dst_cache __read_mostly;
48
49 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
50 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
51 static void xfrm_init_pmtu(struct dst_entry *dst);
52 static int stale_bundle(struct dst_entry *dst);
53 static int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *xdst,
54                           const struct flowi *fl, int family);
55
56
57 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
58                                                 int dir);
59
60 static inline int
61 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
62 {
63         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
64                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
65                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
66                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
67                 (fl->proto == sel->proto || !sel->proto) &&
68                 (fl->oif == sel->ifindex || !sel->ifindex);
69 }
70
71 static inline int
72 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
73 {
74         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
75                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
76                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
77                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
78                 (fl->proto == sel->proto || !sel->proto) &&
79                 (fl->oif == sel->ifindex || !sel->ifindex);
80 }
81
82 int xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
83                         unsigned short family)
84 {
85         switch (family) {
86         case AF_INET:
87                 return __xfrm4_selector_match(sel, fl);
88         case AF_INET6:
89                 return __xfrm6_selector_match(sel, fl);
90         }
91         return 0;
92 }
93
94 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
95                                                   const xfrm_address_t *saddr,
96                                                   const xfrm_address_t *daddr,
97                                                   int family)
98 {
99         struct xfrm_policy_afinfo *afinfo;
100         struct dst_entry *dst;
101
102         afinfo = xfrm_policy_get_afinfo(family);
103         if (unlikely(afinfo == NULL))
104                 return ERR_PTR(-EAFNOSUPPORT);
105
106         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
107
108         xfrm_policy_put_afinfo(afinfo);
109
110         return dst;
111 }
112
113 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
114                                                 xfrm_address_t *prev_saddr,
115                                                 xfrm_address_t *prev_daddr,
116                                                 int family)
117 {
118         struct net *net = xs_net(x);
119         xfrm_address_t *saddr = &x->props.saddr;
120         xfrm_address_t *daddr = &x->id.daddr;
121         struct dst_entry *dst;
122
123         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
124                 saddr = x->coaddr;
125                 daddr = prev_daddr;
126         }
127         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
128                 saddr = prev_saddr;
129                 daddr = x->coaddr;
130         }
131
132         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
133
134         if (!IS_ERR(dst)) {
135                 if (prev_saddr != saddr)
136                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
137                 if (prev_daddr != daddr)
138                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
139         }
140
141         return dst;
142 }
143
144 static inline unsigned long make_jiffies(long secs)
145 {
146         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
147                 return MAX_SCHEDULE_TIMEOUT-1;
148         else
149                 return secs*HZ;
150 }
151
152 static void xfrm_policy_timer(unsigned long data)
153 {
154         struct xfrm_policy *xp = (struct xfrm_policy*)data;
155         unsigned long now = get_seconds();
156         long next = LONG_MAX;
157         int warn = 0;
158         int dir;
159
160         read_lock(&xp->lock);
161
162         if (unlikely(xp->walk.dead))
163                 goto out;
164
165         dir = xfrm_policy_id2dir(xp->index);
166
167         if (xp->lft.hard_add_expires_seconds) {
168                 long tmo = xp->lft.hard_add_expires_seconds +
169                         xp->curlft.add_time - now;
170                 if (tmo <= 0)
171                         goto expired;
172                 if (tmo < next)
173                         next = tmo;
174         }
175         if (xp->lft.hard_use_expires_seconds) {
176                 long tmo = xp->lft.hard_use_expires_seconds +
177                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
178                 if (tmo <= 0)
179                         goto expired;
180                 if (tmo < next)
181                         next = tmo;
182         }
183         if (xp->lft.soft_add_expires_seconds) {
184                 long tmo = xp->lft.soft_add_expires_seconds +
185                         xp->curlft.add_time - now;
186                 if (tmo <= 0) {
187                         warn = 1;
188                         tmo = XFRM_KM_TIMEOUT;
189                 }
190                 if (tmo < next)
191                         next = tmo;
192         }
193         if (xp->lft.soft_use_expires_seconds) {
194                 long tmo = xp->lft.soft_use_expires_seconds +
195                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
196                 if (tmo <= 0) {
197                         warn = 1;
198                         tmo = XFRM_KM_TIMEOUT;
199                 }
200                 if (tmo < next)
201                         next = tmo;
202         }
203
204         if (warn)
205                 km_policy_expired(xp, dir, 0, 0);
206         if (next != LONG_MAX &&
207             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
208                 xfrm_pol_hold(xp);
209
210 out:
211         read_unlock(&xp->lock);
212         xfrm_pol_put(xp);
213         return;
214
215 expired:
216         read_unlock(&xp->lock);
217         if (!xfrm_policy_delete(xp, dir))
218                 km_policy_expired(xp, dir, 1, 0);
219         xfrm_pol_put(xp);
220 }
221
222 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
223 {
224         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
225
226         if (unlikely(pol->walk.dead))
227                 flo = NULL;
228         else
229                 xfrm_pol_hold(pol);
230
231         return flo;
232 }
233
234 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
235 {
236         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
237
238         return !pol->walk.dead;
239 }
240
241 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
242 {
243         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
244 }
245
246 static const struct flow_cache_ops xfrm_policy_fc_ops = {
247         .get = xfrm_policy_flo_get,
248         .check = xfrm_policy_flo_check,
249         .delete = xfrm_policy_flo_delete,
250 };
251
252 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
253  * SPD calls.
254  */
255
256 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
257 {
258         struct xfrm_policy *policy;
259
260         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
261
262         if (policy) {
263                 write_pnet(&policy->xp_net, net);
264                 INIT_LIST_HEAD(&policy->walk.all);
265                 INIT_HLIST_NODE(&policy->bydst);
266                 INIT_HLIST_NODE(&policy->byidx);
267                 rwlock_init(&policy->lock);
268                 atomic_set(&policy->refcnt, 1);
269                 setup_timer(&policy->timer, xfrm_policy_timer,
270                                 (unsigned long)policy);
271                 policy->flo.ops = &xfrm_policy_fc_ops;
272         }
273         return policy;
274 }
275 EXPORT_SYMBOL(xfrm_policy_alloc);
276
277 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
278
279 void xfrm_policy_destroy(struct xfrm_policy *policy)
280 {
281         BUG_ON(!policy->walk.dead);
282
283         if (del_timer(&policy->timer))
284                 BUG();
285
286         security_xfrm_policy_free(policy->security);
287         kfree(policy);
288 }
289 EXPORT_SYMBOL(xfrm_policy_destroy);
290
291 /* Rule must be locked. Release descentant resources, announce
292  * entry dead. The rule must be unlinked from lists to the moment.
293  */
294
295 static void xfrm_policy_kill(struct xfrm_policy *policy)
296 {
297         policy->walk.dead = 1;
298
299         atomic_inc(&policy->genid);
300
301         if (del_timer(&policy->timer))
302                 xfrm_pol_put(policy);
303
304         xfrm_pol_put(policy);
305 }
306
307 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
308
309 static inline unsigned int idx_hash(struct net *net, u32 index)
310 {
311         return __idx_hash(index, net->xfrm.policy_idx_hmask);
312 }
313
314 static struct hlist_head *policy_hash_bysel(struct net *net,
315                                             const struct xfrm_selector *sel,
316                                             unsigned short family, int dir)
317 {
318         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
319         unsigned int hash = __sel_hash(sel, family, hmask);
320
321         return (hash == hmask + 1 ?
322                 &net->xfrm.policy_inexact[dir] :
323                 net->xfrm.policy_bydst[dir].table + hash);
324 }
325
326 static struct hlist_head *policy_hash_direct(struct net *net,
327                                              const xfrm_address_t *daddr,
328                                              const xfrm_address_t *saddr,
329                                              unsigned short family, int dir)
330 {
331         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
332         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
333
334         return net->xfrm.policy_bydst[dir].table + hash;
335 }
336
337 static void xfrm_dst_hash_transfer(struct hlist_head *list,
338                                    struct hlist_head *ndsttable,
339                                    unsigned int nhashmask)
340 {
341         struct hlist_node *entry, *tmp, *entry0 = NULL;
342         struct xfrm_policy *pol;
343         unsigned int h0 = 0;
344
345 redo:
346         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
347                 unsigned int h;
348
349                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
350                                 pol->family, nhashmask);
351                 if (!entry0) {
352                         hlist_del(entry);
353                         hlist_add_head(&pol->bydst, ndsttable+h);
354                         h0 = h;
355                 } else {
356                         if (h != h0)
357                                 continue;
358                         hlist_del(entry);
359                         hlist_add_after(entry0, &pol->bydst);
360                 }
361                 entry0 = entry;
362         }
363         if (!hlist_empty(list)) {
364                 entry0 = NULL;
365                 goto redo;
366         }
367 }
368
369 static void xfrm_idx_hash_transfer(struct hlist_head *list,
370                                    struct hlist_head *nidxtable,
371                                    unsigned int nhashmask)
372 {
373         struct hlist_node *entry, *tmp;
374         struct xfrm_policy *pol;
375
376         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
377                 unsigned int h;
378
379                 h = __idx_hash(pol->index, nhashmask);
380                 hlist_add_head(&pol->byidx, nidxtable+h);
381         }
382 }
383
384 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
385 {
386         return ((old_hmask + 1) << 1) - 1;
387 }
388
389 static void xfrm_bydst_resize(struct net *net, int dir)
390 {
391         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
392         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
393         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
394         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
395         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
396         int i;
397
398         if (!ndst)
399                 return;
400
401         write_lock_bh(&xfrm_policy_lock);
402
403         for (i = hmask; i >= 0; i--)
404                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
405
406         net->xfrm.policy_bydst[dir].table = ndst;
407         net->xfrm.policy_bydst[dir].hmask = nhashmask;
408
409         write_unlock_bh(&xfrm_policy_lock);
410
411         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
412 }
413
414 static void xfrm_byidx_resize(struct net *net, int total)
415 {
416         unsigned int hmask = net->xfrm.policy_idx_hmask;
417         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
418         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
419         struct hlist_head *oidx = net->xfrm.policy_byidx;
420         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
421         int i;
422
423         if (!nidx)
424                 return;
425
426         write_lock_bh(&xfrm_policy_lock);
427
428         for (i = hmask; i >= 0; i--)
429                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
430
431         net->xfrm.policy_byidx = nidx;
432         net->xfrm.policy_idx_hmask = nhashmask;
433
434         write_unlock_bh(&xfrm_policy_lock);
435
436         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
437 }
438
439 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
440 {
441         unsigned int cnt = net->xfrm.policy_count[dir];
442         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
443
444         if (total)
445                 *total += cnt;
446
447         if ((hmask + 1) < xfrm_policy_hashmax &&
448             cnt > hmask)
449                 return 1;
450
451         return 0;
452 }
453
454 static inline int xfrm_byidx_should_resize(struct net *net, int total)
455 {
456         unsigned int hmask = net->xfrm.policy_idx_hmask;
457
458         if ((hmask + 1) < xfrm_policy_hashmax &&
459             total > hmask)
460                 return 1;
461
462         return 0;
463 }
464
465 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
466 {
467         read_lock_bh(&xfrm_policy_lock);
468         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
469         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
470         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
471         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
472         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
473         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
474         si->spdhcnt = net->xfrm.policy_idx_hmask;
475         si->spdhmcnt = xfrm_policy_hashmax;
476         read_unlock_bh(&xfrm_policy_lock);
477 }
478 EXPORT_SYMBOL(xfrm_spd_getinfo);
479
480 static DEFINE_MUTEX(hash_resize_mutex);
481 static void xfrm_hash_resize(struct work_struct *work)
482 {
483         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
484         int dir, total;
485
486         mutex_lock(&hash_resize_mutex);
487
488         total = 0;
489         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
490                 if (xfrm_bydst_should_resize(net, dir, &total))
491                         xfrm_bydst_resize(net, dir);
492         }
493         if (xfrm_byidx_should_resize(net, total))
494                 xfrm_byidx_resize(net, total);
495
496         mutex_unlock(&hash_resize_mutex);
497 }
498
499 /* Generate new index... KAME seems to generate them ordered by cost
500  * of an absolute inpredictability of ordering of rules. This will not pass. */
501 static u32 xfrm_gen_index(struct net *net, int dir)
502 {
503         static u32 idx_generator;
504
505         for (;;) {
506                 struct hlist_node *entry;
507                 struct hlist_head *list;
508                 struct xfrm_policy *p;
509                 u32 idx;
510                 int found;
511
512                 idx = (idx_generator | dir);
513                 idx_generator += 8;
514                 if (idx == 0)
515                         idx = 8;
516                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
517                 found = 0;
518                 hlist_for_each_entry(p, entry, list, byidx) {
519                         if (p->index == idx) {
520                                 found = 1;
521                                 break;
522                         }
523                 }
524                 if (!found)
525                         return idx;
526         }
527 }
528
529 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
530 {
531         u32 *p1 = (u32 *) s1;
532         u32 *p2 = (u32 *) s2;
533         int len = sizeof(struct xfrm_selector) / sizeof(u32);
534         int i;
535
536         for (i = 0; i < len; i++) {
537                 if (p1[i] != p2[i])
538                         return 1;
539         }
540
541         return 0;
542 }
543
544 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
545 {
546         struct net *net = xp_net(policy);
547         struct xfrm_policy *pol;
548         struct xfrm_policy *delpol;
549         struct hlist_head *chain;
550         struct hlist_node *entry, *newpos;
551         u32 mark = policy->mark.v & policy->mark.m;
552
553         write_lock_bh(&xfrm_policy_lock);
554         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
555         delpol = NULL;
556         newpos = NULL;
557         hlist_for_each_entry(pol, entry, chain, bydst) {
558                 if (pol->type == policy->type &&
559                     !selector_cmp(&pol->selector, &policy->selector) &&
560                     (mark & pol->mark.m) == pol->mark.v &&
561                     xfrm_sec_ctx_match(pol->security, policy->security) &&
562                     !WARN_ON(delpol)) {
563                         if (excl) {
564                                 write_unlock_bh(&xfrm_policy_lock);
565                                 return -EEXIST;
566                         }
567                         delpol = pol;
568                         if (policy->priority > pol->priority)
569                                 continue;
570                 } else if (policy->priority >= pol->priority) {
571                         newpos = &pol->bydst;
572                         continue;
573                 }
574                 if (delpol)
575                         break;
576         }
577         if (newpos)
578                 hlist_add_after(newpos, &policy->bydst);
579         else
580                 hlist_add_head(&policy->bydst, chain);
581         xfrm_pol_hold(policy);
582         net->xfrm.policy_count[dir]++;
583         atomic_inc(&flow_cache_genid);
584         if (delpol)
585                 __xfrm_policy_unlink(delpol, dir);
586         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
587         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
588         policy->curlft.add_time = get_seconds();
589         policy->curlft.use_time = 0;
590         if (!mod_timer(&policy->timer, jiffies + HZ))
591                 xfrm_pol_hold(policy);
592         list_add(&policy->walk.all, &net->xfrm.policy_all);
593         write_unlock_bh(&xfrm_policy_lock);
594
595         if (delpol)
596                 xfrm_policy_kill(delpol);
597         else if (xfrm_bydst_should_resize(net, dir, NULL))
598                 schedule_work(&net->xfrm.policy_hash_work);
599
600         return 0;
601 }
602 EXPORT_SYMBOL(xfrm_policy_insert);
603
604 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
605                                           int dir, struct xfrm_selector *sel,
606                                           struct xfrm_sec_ctx *ctx, int delete,
607                                           int *err)
608 {
609         struct xfrm_policy *pol, *ret;
610         struct hlist_head *chain;
611         struct hlist_node *entry;
612
613         *err = 0;
614         write_lock_bh(&xfrm_policy_lock);
615         chain = policy_hash_bysel(net, sel, sel->family, dir);
616         ret = NULL;
617         hlist_for_each_entry(pol, entry, chain, bydst) {
618                 if (pol->type == type &&
619                     (mark & pol->mark.m) == pol->mark.v &&
620                     !selector_cmp(sel, &pol->selector) &&
621                     xfrm_sec_ctx_match(ctx, pol->security)) {
622                         xfrm_pol_hold(pol);
623                         if (delete) {
624                                 *err = security_xfrm_policy_delete(
625                                                                 pol->security);
626                                 if (*err) {
627                                         write_unlock_bh(&xfrm_policy_lock);
628                                         return pol;
629                                 }
630                                 __xfrm_policy_unlink(pol, dir);
631                         }
632                         ret = pol;
633                         break;
634                 }
635         }
636         write_unlock_bh(&xfrm_policy_lock);
637
638         if (ret && delete)
639                 xfrm_policy_kill(ret);
640         return ret;
641 }
642 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
643
644 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
645                                      int dir, u32 id, int delete, int *err)
646 {
647         struct xfrm_policy *pol, *ret;
648         struct hlist_head *chain;
649         struct hlist_node *entry;
650
651         *err = -ENOENT;
652         if (xfrm_policy_id2dir(id) != dir)
653                 return NULL;
654
655         *err = 0;
656         write_lock_bh(&xfrm_policy_lock);
657         chain = net->xfrm.policy_byidx + idx_hash(net, id);
658         ret = NULL;
659         hlist_for_each_entry(pol, entry, chain, byidx) {
660                 if (pol->type == type && pol->index == id &&
661                     (mark & pol->mark.m) == pol->mark.v) {
662                         xfrm_pol_hold(pol);
663                         if (delete) {
664                                 *err = security_xfrm_policy_delete(
665                                                                 pol->security);
666                                 if (*err) {
667                                         write_unlock_bh(&xfrm_policy_lock);
668                                         return pol;
669                                 }
670                                 __xfrm_policy_unlink(pol, dir);
671                         }
672                         ret = pol;
673                         break;
674                 }
675         }
676         write_unlock_bh(&xfrm_policy_lock);
677
678         if (ret && delete)
679                 xfrm_policy_kill(ret);
680         return ret;
681 }
682 EXPORT_SYMBOL(xfrm_policy_byid);
683
684 #ifdef CONFIG_SECURITY_NETWORK_XFRM
685 static inline int
686 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
687 {
688         int dir, err = 0;
689
690         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
691                 struct xfrm_policy *pol;
692                 struct hlist_node *entry;
693                 int i;
694
695                 hlist_for_each_entry(pol, entry,
696                                      &net->xfrm.policy_inexact[dir], bydst) {
697                         if (pol->type != type)
698                                 continue;
699                         err = security_xfrm_policy_delete(pol->security);
700                         if (err) {
701                                 xfrm_audit_policy_delete(pol, 0,
702                                                          audit_info->loginuid,
703                                                          audit_info->sessionid,
704                                                          audit_info->secid);
705                                 return err;
706                         }
707                 }
708                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
709                         hlist_for_each_entry(pol, entry,
710                                              net->xfrm.policy_bydst[dir].table + i,
711                                              bydst) {
712                                 if (pol->type != type)
713                                         continue;
714                                 err = security_xfrm_policy_delete(
715                                                                 pol->security);
716                                 if (err) {
717                                         xfrm_audit_policy_delete(pol, 0,
718                                                         audit_info->loginuid,
719                                                         audit_info->sessionid,
720                                                         audit_info->secid);
721                                         return err;
722                                 }
723                         }
724                 }
725         }
726         return err;
727 }
728 #else
729 static inline int
730 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
731 {
732         return 0;
733 }
734 #endif
735
736 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
737 {
738         int dir, err = 0, cnt = 0;
739
740         write_lock_bh(&xfrm_policy_lock);
741
742         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
743         if (err)
744                 goto out;
745
746         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
747                 struct xfrm_policy *pol;
748                 struct hlist_node *entry;
749                 int i;
750
751         again1:
752                 hlist_for_each_entry(pol, entry,
753                                      &net->xfrm.policy_inexact[dir], bydst) {
754                         if (pol->type != type)
755                                 continue;
756                         __xfrm_policy_unlink(pol, dir);
757                         write_unlock_bh(&xfrm_policy_lock);
758                         cnt++;
759
760                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
761                                                  audit_info->sessionid,
762                                                  audit_info->secid);
763
764                         xfrm_policy_kill(pol);
765
766                         write_lock_bh(&xfrm_policy_lock);
767                         goto again1;
768                 }
769
770                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
771         again2:
772                         hlist_for_each_entry(pol, entry,
773                                              net->xfrm.policy_bydst[dir].table + i,
774                                              bydst) {
775                                 if (pol->type != type)
776                                         continue;
777                                 __xfrm_policy_unlink(pol, dir);
778                                 write_unlock_bh(&xfrm_policy_lock);
779                                 cnt++;
780
781                                 xfrm_audit_policy_delete(pol, 1,
782                                                          audit_info->loginuid,
783                                                          audit_info->sessionid,
784                                                          audit_info->secid);
785                                 xfrm_policy_kill(pol);
786
787                                 write_lock_bh(&xfrm_policy_lock);
788                                 goto again2;
789                         }
790                 }
791
792         }
793         if (!cnt)
794                 err = -ESRCH;
795 out:
796         write_unlock_bh(&xfrm_policy_lock);
797         return err;
798 }
799 EXPORT_SYMBOL(xfrm_policy_flush);
800
801 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
802                      int (*func)(struct xfrm_policy *, int, int, void*),
803                      void *data)
804 {
805         struct xfrm_policy *pol;
806         struct xfrm_policy_walk_entry *x;
807         int error = 0;
808
809         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
810             walk->type != XFRM_POLICY_TYPE_ANY)
811                 return -EINVAL;
812
813         if (list_empty(&walk->walk.all) && walk->seq != 0)
814                 return 0;
815
816         write_lock_bh(&xfrm_policy_lock);
817         if (list_empty(&walk->walk.all))
818                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
819         else
820                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
821         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
822                 if (x->dead)
823                         continue;
824                 pol = container_of(x, struct xfrm_policy, walk);
825                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
826                     walk->type != pol->type)
827                         continue;
828                 error = func(pol, xfrm_policy_id2dir(pol->index),
829                              walk->seq, data);
830                 if (error) {
831                         list_move_tail(&walk->walk.all, &x->all);
832                         goto out;
833                 }
834                 walk->seq++;
835         }
836         if (walk->seq == 0) {
837                 error = -ENOENT;
838                 goto out;
839         }
840         list_del_init(&walk->walk.all);
841 out:
842         write_unlock_bh(&xfrm_policy_lock);
843         return error;
844 }
845 EXPORT_SYMBOL(xfrm_policy_walk);
846
847 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
848 {
849         INIT_LIST_HEAD(&walk->walk.all);
850         walk->walk.dead = 1;
851         walk->type = type;
852         walk->seq = 0;
853 }
854 EXPORT_SYMBOL(xfrm_policy_walk_init);
855
856 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
857 {
858         if (list_empty(&walk->walk.all))
859                 return;
860
861         write_lock_bh(&xfrm_policy_lock);
862         list_del(&walk->walk.all);
863         write_unlock_bh(&xfrm_policy_lock);
864 }
865 EXPORT_SYMBOL(xfrm_policy_walk_done);
866
867 /*
868  * Find policy to apply to this flow.
869  *
870  * Returns 0 if policy found, else an -errno.
871  */
872 static int xfrm_policy_match(struct xfrm_policy *pol, const struct flowi *fl,
873                              u8 type, u16 family, int dir)
874 {
875         struct xfrm_selector *sel = &pol->selector;
876         int match, ret = -ESRCH;
877
878         if (pol->family != family ||
879             (fl->mark & pol->mark.m) != pol->mark.v ||
880             pol->type != type)
881                 return ret;
882
883         match = xfrm_selector_match(sel, fl, family);
884         if (match)
885                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
886                                                   dir);
887
888         return ret;
889 }
890
891 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
892                                                      const struct flowi *fl,
893                                                      u16 family, u8 dir)
894 {
895         int err;
896         struct xfrm_policy *pol, *ret;
897         xfrm_address_t *daddr, *saddr;
898         struct hlist_node *entry;
899         struct hlist_head *chain;
900         u32 priority = ~0U;
901
902         daddr = xfrm_flowi_daddr(fl, family);
903         saddr = xfrm_flowi_saddr(fl, family);
904         if (unlikely(!daddr || !saddr))
905                 return NULL;
906
907         read_lock_bh(&xfrm_policy_lock);
908         chain = policy_hash_direct(net, daddr, saddr, family, dir);
909         ret = NULL;
910         hlist_for_each_entry(pol, entry, chain, bydst) {
911                 err = xfrm_policy_match(pol, fl, type, family, dir);
912                 if (err) {
913                         if (err == -ESRCH)
914                                 continue;
915                         else {
916                                 ret = ERR_PTR(err);
917                                 goto fail;
918                         }
919                 } else {
920                         ret = pol;
921                         priority = ret->priority;
922                         break;
923                 }
924         }
925         chain = &net->xfrm.policy_inexact[dir];
926         hlist_for_each_entry(pol, entry, chain, bydst) {
927                 err = xfrm_policy_match(pol, fl, type, family, dir);
928                 if (err) {
929                         if (err == -ESRCH)
930                                 continue;
931                         else {
932                                 ret = ERR_PTR(err);
933                                 goto fail;
934                         }
935                 } else if (pol->priority < priority) {
936                         ret = pol;
937                         break;
938                 }
939         }
940         if (ret)
941                 xfrm_pol_hold(ret);
942 fail:
943         read_unlock_bh(&xfrm_policy_lock);
944
945         return ret;
946 }
947
948 static struct xfrm_policy *
949 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
950 {
951 #ifdef CONFIG_XFRM_SUB_POLICY
952         struct xfrm_policy *pol;
953
954         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
955         if (pol != NULL)
956                 return pol;
957 #endif
958         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
959 }
960
961 static struct flow_cache_object *
962 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
963                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
964 {
965         struct xfrm_policy *pol;
966
967         if (old_obj)
968                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
969
970         pol = __xfrm_policy_lookup(net, fl, family, dir);
971         if (IS_ERR_OR_NULL(pol))
972                 return ERR_CAST(pol);
973
974         /* Resolver returns two references:
975          * one for cache and one for caller of flow_cache_lookup() */
976         xfrm_pol_hold(pol);
977
978         return &pol->flo;
979 }
980
981 static inline int policy_to_flow_dir(int dir)
982 {
983         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
984             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
985             XFRM_POLICY_FWD == FLOW_DIR_FWD)
986                 return dir;
987         switch (dir) {
988         default:
989         case XFRM_POLICY_IN:
990                 return FLOW_DIR_IN;
991         case XFRM_POLICY_OUT:
992                 return FLOW_DIR_OUT;
993         case XFRM_POLICY_FWD:
994                 return FLOW_DIR_FWD;
995         }
996 }
997
998 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
999                                                  const struct flowi *fl)
1000 {
1001         struct xfrm_policy *pol;
1002
1003         read_lock_bh(&xfrm_policy_lock);
1004         if ((pol = sk->sk_policy[dir]) != NULL) {
1005                 int match = xfrm_selector_match(&pol->selector, fl,
1006                                                 sk->sk_family);
1007                 int err = 0;
1008
1009                 if (match) {
1010                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1011                                 pol = NULL;
1012                                 goto out;
1013                         }
1014                         err = security_xfrm_policy_lookup(pol->security,
1015                                                       fl->secid,
1016                                                       policy_to_flow_dir(dir));
1017                         if (!err)
1018                                 xfrm_pol_hold(pol);
1019                         else if (err == -ESRCH)
1020                                 pol = NULL;
1021                         else
1022                                 pol = ERR_PTR(err);
1023                 } else
1024                         pol = NULL;
1025         }
1026 out:
1027         read_unlock_bh(&xfrm_policy_lock);
1028         return pol;
1029 }
1030
1031 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1032 {
1033         struct net *net = xp_net(pol);
1034         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1035                                                      pol->family, dir);
1036
1037         list_add(&pol->walk.all, &net->xfrm.policy_all);
1038         hlist_add_head(&pol->bydst, chain);
1039         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1040         net->xfrm.policy_count[dir]++;
1041         xfrm_pol_hold(pol);
1042
1043         if (xfrm_bydst_should_resize(net, dir, NULL))
1044                 schedule_work(&net->xfrm.policy_hash_work);
1045 }
1046
1047 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1048                                                 int dir)
1049 {
1050         struct net *net = xp_net(pol);
1051
1052         if (hlist_unhashed(&pol->bydst))
1053                 return NULL;
1054
1055         hlist_del(&pol->bydst);
1056         hlist_del(&pol->byidx);
1057         list_del(&pol->walk.all);
1058         net->xfrm.policy_count[dir]--;
1059
1060         return pol;
1061 }
1062
1063 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1064 {
1065         write_lock_bh(&xfrm_policy_lock);
1066         pol = __xfrm_policy_unlink(pol, dir);
1067         write_unlock_bh(&xfrm_policy_lock);
1068         if (pol) {
1069                 xfrm_policy_kill(pol);
1070                 return 0;
1071         }
1072         return -ENOENT;
1073 }
1074 EXPORT_SYMBOL(xfrm_policy_delete);
1075
1076 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1077 {
1078         struct net *net = xp_net(pol);
1079         struct xfrm_policy *old_pol;
1080
1081 #ifdef CONFIG_XFRM_SUB_POLICY
1082         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1083                 return -EINVAL;
1084 #endif
1085
1086         write_lock_bh(&xfrm_policy_lock);
1087         old_pol = sk->sk_policy[dir];
1088         sk->sk_policy[dir] = pol;
1089         if (pol) {
1090                 pol->curlft.add_time = get_seconds();
1091                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1092                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1093         }
1094         if (old_pol)
1095                 /* Unlinking succeeds always. This is the only function
1096                  * allowed to delete or replace socket policy.
1097                  */
1098                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1099         write_unlock_bh(&xfrm_policy_lock);
1100
1101         if (old_pol) {
1102                 xfrm_policy_kill(old_pol);
1103         }
1104         return 0;
1105 }
1106
1107 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1108 {
1109         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1110
1111         if (newp) {
1112                 newp->selector = old->selector;
1113                 if (security_xfrm_policy_clone(old->security,
1114                                                &newp->security)) {
1115                         kfree(newp);
1116                         return NULL;  /* ENOMEM */
1117                 }
1118                 newp->lft = old->lft;
1119                 newp->curlft = old->curlft;
1120                 newp->mark = old->mark;
1121                 newp->action = old->action;
1122                 newp->flags = old->flags;
1123                 newp->xfrm_nr = old->xfrm_nr;
1124                 newp->index = old->index;
1125                 newp->type = old->type;
1126                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1127                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1128                 write_lock_bh(&xfrm_policy_lock);
1129                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1130                 write_unlock_bh(&xfrm_policy_lock);
1131                 xfrm_pol_put(newp);
1132         }
1133         return newp;
1134 }
1135
1136 int __xfrm_sk_clone_policy(struct sock *sk)
1137 {
1138         struct xfrm_policy *p0 = sk->sk_policy[0],
1139                            *p1 = sk->sk_policy[1];
1140
1141         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1142         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1143                 return -ENOMEM;
1144         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1145                 return -ENOMEM;
1146         return 0;
1147 }
1148
1149 static int
1150 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1151                unsigned short family)
1152 {
1153         int err;
1154         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1155
1156         if (unlikely(afinfo == NULL))
1157                 return -EINVAL;
1158         err = afinfo->get_saddr(net, local, remote);
1159         xfrm_policy_put_afinfo(afinfo);
1160         return err;
1161 }
1162
1163 /* Resolve list of templates for the flow, given policy. */
1164
1165 static int
1166 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1167                       struct xfrm_state **xfrm, unsigned short family)
1168 {
1169         struct net *net = xp_net(policy);
1170         int nx;
1171         int i, error;
1172         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1173         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1174         xfrm_address_t tmp;
1175
1176         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1177                 struct xfrm_state *x;
1178                 xfrm_address_t *remote = daddr;
1179                 xfrm_address_t *local  = saddr;
1180                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1181
1182                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1183                     tmpl->mode == XFRM_MODE_BEET) {
1184                         remote = &tmpl->id.daddr;
1185                         local = &tmpl->saddr;
1186                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1187                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1188                                 if (error)
1189                                         goto fail;
1190                                 local = &tmp;
1191                         }
1192                 }
1193
1194                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1195
1196                 if (x && x->km.state == XFRM_STATE_VALID) {
1197                         xfrm[nx++] = x;
1198                         daddr = remote;
1199                         saddr = local;
1200                         continue;
1201                 }
1202                 if (x) {
1203                         error = (x->km.state == XFRM_STATE_ERROR ?
1204                                  -EINVAL : -EAGAIN);
1205                         xfrm_state_put(x);
1206                 }
1207                 else if (error == -ESRCH)
1208                         error = -EAGAIN;
1209
1210                 if (!tmpl->optional)
1211                         goto fail;
1212         }
1213         return nx;
1214
1215 fail:
1216         for (nx--; nx>=0; nx--)
1217                 xfrm_state_put(xfrm[nx]);
1218         return error;
1219 }
1220
1221 static int
1222 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1223                   struct xfrm_state **xfrm, unsigned short family)
1224 {
1225         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1226         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1227         int cnx = 0;
1228         int error;
1229         int ret;
1230         int i;
1231
1232         for (i = 0; i < npols; i++) {
1233                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1234                         error = -ENOBUFS;
1235                         goto fail;
1236                 }
1237
1238                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1239                 if (ret < 0) {
1240                         error = ret;
1241                         goto fail;
1242                 } else
1243                         cnx += ret;
1244         }
1245
1246         /* found states are sorted for outbound processing */
1247         if (npols > 1)
1248                 xfrm_state_sort(xfrm, tpp, cnx, family);
1249
1250         return cnx;
1251
1252  fail:
1253         for (cnx--; cnx>=0; cnx--)
1254                 xfrm_state_put(tpp[cnx]);
1255         return error;
1256
1257 }
1258
1259 /* Check that the bundle accepts the flow and its components are
1260  * still valid.
1261  */
1262
1263 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1264 {
1265         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1266         int tos;
1267
1268         if (!afinfo)
1269                 return -EINVAL;
1270
1271         tos = afinfo->get_tos(fl);
1272
1273         xfrm_policy_put_afinfo(afinfo);
1274
1275         return tos;
1276 }
1277
1278 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1279 {
1280         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1281         struct dst_entry *dst = &xdst->u.dst;
1282
1283         if (xdst->route == NULL) {
1284                 /* Dummy bundle - if it has xfrms we were not
1285                  * able to build bundle as template resolution failed.
1286                  * It means we need to try again resolving. */
1287                 if (xdst->num_xfrms > 0)
1288                         return NULL;
1289         } else {
1290                 /* Real bundle */
1291                 if (stale_bundle(dst))
1292                         return NULL;
1293         }
1294
1295         dst_hold(dst);
1296         return flo;
1297 }
1298
1299 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1300 {
1301         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1302         struct dst_entry *dst = &xdst->u.dst;
1303
1304         if (!xdst->route)
1305                 return 0;
1306         if (stale_bundle(dst))
1307                 return 0;
1308
1309         return 1;
1310 }
1311
1312 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1313 {
1314         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1315         struct dst_entry *dst = &xdst->u.dst;
1316
1317         dst_free(dst);
1318 }
1319
1320 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1321         .get = xfrm_bundle_flo_get,
1322         .check = xfrm_bundle_flo_check,
1323         .delete = xfrm_bundle_flo_delete,
1324 };
1325
1326 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1327 {
1328         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1329         struct dst_ops *dst_ops;
1330         struct xfrm_dst *xdst;
1331
1332         if (!afinfo)
1333                 return ERR_PTR(-EINVAL);
1334
1335         switch (family) {
1336         case AF_INET:
1337                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1338                 break;
1339 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1340         case AF_INET6:
1341                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1342                 break;
1343 #endif
1344         default:
1345                 BUG();
1346         }
1347         xdst = dst_alloc(dst_ops, 0);
1348         xfrm_policy_put_afinfo(afinfo);
1349
1350         if (likely(xdst))
1351                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1352         else
1353                 xdst = ERR_PTR(-ENOBUFS);
1354
1355         return xdst;
1356 }
1357
1358 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1359                                  int nfheader_len)
1360 {
1361         struct xfrm_policy_afinfo *afinfo =
1362                 xfrm_policy_get_afinfo(dst->ops->family);
1363         int err;
1364
1365         if (!afinfo)
1366                 return -EINVAL;
1367
1368         err = afinfo->init_path(path, dst, nfheader_len);
1369
1370         xfrm_policy_put_afinfo(afinfo);
1371
1372         return err;
1373 }
1374
1375 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1376                                 const struct flowi *fl)
1377 {
1378         struct xfrm_policy_afinfo *afinfo =
1379                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1380         int err;
1381
1382         if (!afinfo)
1383                 return -EINVAL;
1384
1385         err = afinfo->fill_dst(xdst, dev, fl);
1386
1387         xfrm_policy_put_afinfo(afinfo);
1388
1389         return err;
1390 }
1391
1392
1393 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1394  * all the metrics... Shortly, bundle a bundle.
1395  */
1396
1397 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1398                                             struct xfrm_state **xfrm, int nx,
1399                                             const struct flowi *fl,
1400                                             struct dst_entry *dst)
1401 {
1402         struct net *net = xp_net(policy);
1403         unsigned long now = jiffies;
1404         struct net_device *dev;
1405         struct dst_entry *dst_prev = NULL;
1406         struct dst_entry *dst0 = NULL;
1407         int i = 0;
1408         int err;
1409         int header_len = 0;
1410         int nfheader_len = 0;
1411         int trailer_len = 0;
1412         int tos;
1413         int family = policy->selector.family;
1414         xfrm_address_t saddr, daddr;
1415
1416         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1417
1418         tos = xfrm_get_tos(fl, family);
1419         err = tos;
1420         if (tos < 0)
1421                 goto put_states;
1422
1423         dst_hold(dst);
1424
1425         for (; i < nx; i++) {
1426                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1427                 struct dst_entry *dst1 = &xdst->u.dst;
1428
1429                 err = PTR_ERR(xdst);
1430                 if (IS_ERR(xdst)) {
1431                         dst_release(dst);
1432                         goto put_states;
1433                 }
1434
1435                 if (!dst_prev)
1436                         dst0 = dst1;
1437                 else {
1438                         dst_prev->child = dst_clone(dst1);
1439                         dst1->flags |= DST_NOHASH;
1440                 }
1441
1442                 xdst->route = dst;
1443                 dst_copy_metrics(dst1, dst);
1444
1445                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1446                         family = xfrm[i]->props.family;
1447                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1448                                               family);
1449                         err = PTR_ERR(dst);
1450                         if (IS_ERR(dst))
1451                                 goto put_states;
1452                 } else
1453                         dst_hold(dst);
1454
1455                 dst1->xfrm = xfrm[i];
1456                 xdst->xfrm_genid = xfrm[i]->genid;
1457
1458                 dst1->obsolete = -1;
1459                 dst1->flags |= DST_HOST;
1460                 dst1->lastuse = now;
1461
1462                 dst1->input = dst_discard;
1463                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1464
1465                 dst1->next = dst_prev;
1466                 dst_prev = dst1;
1467
1468                 header_len += xfrm[i]->props.header_len;
1469                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1470                         nfheader_len += xfrm[i]->props.header_len;
1471                 trailer_len += xfrm[i]->props.trailer_len;
1472         }
1473
1474         dst_prev->child = dst;
1475         dst0->path = dst;
1476
1477         err = -ENODEV;
1478         dev = dst->dev;
1479         if (!dev)
1480                 goto free_dst;
1481
1482         /* Copy neighbour for reachability confirmation */
1483         dst0->neighbour = neigh_clone(dst->neighbour);
1484
1485         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1486         xfrm_init_pmtu(dst_prev);
1487
1488         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1489                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1490
1491                 err = xfrm_fill_dst(xdst, dev, fl);
1492                 if (err)
1493                         goto free_dst;
1494
1495                 dst_prev->header_len = header_len;
1496                 dst_prev->trailer_len = trailer_len;
1497                 header_len -= xdst->u.dst.xfrm->props.header_len;
1498                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1499         }
1500
1501 out:
1502         return dst0;
1503
1504 put_states:
1505         for (; i < nx; i++)
1506                 xfrm_state_put(xfrm[i]);
1507 free_dst:
1508         if (dst0)
1509                 dst_free(dst0);
1510         dst0 = ERR_PTR(err);
1511         goto out;
1512 }
1513
1514 static int inline
1515 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1516 {
1517         if (!*target) {
1518                 *target = kmalloc(size, GFP_ATOMIC);
1519                 if (!*target)
1520                         return -ENOMEM;
1521         }
1522         memcpy(*target, src, size);
1523         return 0;
1524 }
1525
1526 static int inline
1527 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1528 {
1529 #ifdef CONFIG_XFRM_SUB_POLICY
1530         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1531         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1532                                    sel, sizeof(*sel));
1533 #else
1534         return 0;
1535 #endif
1536 }
1537
1538 static int inline
1539 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1540 {
1541 #ifdef CONFIG_XFRM_SUB_POLICY
1542         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1543         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1544 #else
1545         return 0;
1546 #endif
1547 }
1548
1549 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1550                                 struct xfrm_policy **pols,
1551                                 int *num_pols, int *num_xfrms)
1552 {
1553         int i;
1554
1555         if (*num_pols == 0 || !pols[0]) {
1556                 *num_pols = 0;
1557                 *num_xfrms = 0;
1558                 return 0;
1559         }
1560         if (IS_ERR(pols[0]))
1561                 return PTR_ERR(pols[0]);
1562
1563         *num_xfrms = pols[0]->xfrm_nr;
1564
1565 #ifdef CONFIG_XFRM_SUB_POLICY
1566         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1567             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1568                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1569                                                     XFRM_POLICY_TYPE_MAIN,
1570                                                     fl, family,
1571                                                     XFRM_POLICY_OUT);
1572                 if (pols[1]) {
1573                         if (IS_ERR(pols[1])) {
1574                                 xfrm_pols_put(pols, *num_pols);
1575                                 return PTR_ERR(pols[1]);
1576                         }
1577                         (*num_pols) ++;
1578                         (*num_xfrms) += pols[1]->xfrm_nr;
1579                 }
1580         }
1581 #endif
1582         for (i = 0; i < *num_pols; i++) {
1583                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1584                         *num_xfrms = -1;
1585                         break;
1586                 }
1587         }
1588
1589         return 0;
1590
1591 }
1592
1593 static struct xfrm_dst *
1594 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1595                                const struct flowi *fl, u16 family,
1596                                struct dst_entry *dst_orig)
1597 {
1598         struct net *net = xp_net(pols[0]);
1599         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1600         struct dst_entry *dst;
1601         struct xfrm_dst *xdst;
1602         int err;
1603
1604         /* Try to instantiate a bundle */
1605         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1606         if (err <= 0) {
1607                 if (err != 0 && err != -EAGAIN)
1608                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1609                 return ERR_PTR(err);
1610         }
1611
1612         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1613         if (IS_ERR(dst)) {
1614                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1615                 return ERR_CAST(dst);
1616         }
1617
1618         xdst = (struct xfrm_dst *)dst;
1619         xdst->num_xfrms = err;
1620         if (num_pols > 1)
1621                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1622         else
1623                 err = xfrm_dst_update_origin(dst, fl);
1624         if (unlikely(err)) {
1625                 dst_free(dst);
1626                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1627                 return ERR_PTR(err);
1628         }
1629
1630         xdst->num_pols = num_pols;
1631         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1632         xdst->policy_genid = atomic_read(&pols[0]->genid);
1633
1634         return xdst;
1635 }
1636
1637 static struct flow_cache_object *
1638 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1639                    struct flow_cache_object *oldflo, void *ctx)
1640 {
1641         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1642         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1643         struct xfrm_dst *xdst, *new_xdst;
1644         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1645
1646         /* Check if the policies from old bundle are usable */
1647         xdst = NULL;
1648         if (oldflo) {
1649                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1650                 num_pols = xdst->num_pols;
1651                 num_xfrms = xdst->num_xfrms;
1652                 pol_dead = 0;
1653                 for (i = 0; i < num_pols; i++) {
1654                         pols[i] = xdst->pols[i];
1655                         pol_dead |= pols[i]->walk.dead;
1656                 }
1657                 if (pol_dead) {
1658                         dst_free(&xdst->u.dst);
1659                         xdst = NULL;
1660                         num_pols = 0;
1661                         num_xfrms = 0;
1662                         oldflo = NULL;
1663                 }
1664         }
1665
1666         /* Resolve policies to use if we couldn't get them from
1667          * previous cache entry */
1668         if (xdst == NULL) {
1669                 num_pols = 1;
1670                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1671                 err = xfrm_expand_policies(fl, family, pols,
1672                                            &num_pols, &num_xfrms);
1673                 if (err < 0)
1674                         goto inc_error;
1675                 if (num_pols == 0)
1676                         return NULL;
1677                 if (num_xfrms <= 0)
1678                         goto make_dummy_bundle;
1679         }
1680
1681         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1682         if (IS_ERR(new_xdst)) {
1683                 err = PTR_ERR(new_xdst);
1684                 if (err != -EAGAIN)
1685                         goto error;
1686                 if (oldflo == NULL)
1687                         goto make_dummy_bundle;
1688                 dst_hold(&xdst->u.dst);
1689                 return oldflo;
1690         } else if (new_xdst == NULL) {
1691                 num_xfrms = 0;
1692                 if (oldflo == NULL)
1693                         goto make_dummy_bundle;
1694                 xdst->num_xfrms = 0;
1695                 dst_hold(&xdst->u.dst);
1696                 return oldflo;
1697         }
1698
1699         /* Kill the previous bundle */
1700         if (xdst) {
1701                 /* The policies were stolen for newly generated bundle */
1702                 xdst->num_pols = 0;
1703                 dst_free(&xdst->u.dst);
1704         }
1705
1706         /* Flow cache does not have reference, it dst_free()'s,
1707          * but we do need to return one reference for original caller */
1708         dst_hold(&new_xdst->u.dst);
1709         return &new_xdst->flo;
1710
1711 make_dummy_bundle:
1712         /* We found policies, but there's no bundles to instantiate:
1713          * either because the policy blocks, has no transformations or
1714          * we could not build template (no xfrm_states).*/
1715         xdst = xfrm_alloc_dst(net, family);
1716         if (IS_ERR(xdst)) {
1717                 xfrm_pols_put(pols, num_pols);
1718                 return ERR_CAST(xdst);
1719         }
1720         xdst->num_pols = num_pols;
1721         xdst->num_xfrms = num_xfrms;
1722         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1723
1724         dst_hold(&xdst->u.dst);
1725         return &xdst->flo;
1726
1727 inc_error:
1728         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1729 error:
1730         if (xdst != NULL)
1731                 dst_free(&xdst->u.dst);
1732         else
1733                 xfrm_pols_put(pols, num_pols);
1734         return ERR_PTR(err);
1735 }
1736
1737 /* Main function: finds/creates a bundle for given flow.
1738  *
1739  * At the moment we eat a raw IP route. Mostly to speed up lookups
1740  * on interfaces with disabled IPsec.
1741  */
1742 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p,
1743                   const struct flowi *fl,
1744                   struct sock *sk, int flags)
1745 {
1746         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1747         struct flow_cache_object *flo;
1748         struct xfrm_dst *xdst;
1749         struct dst_entry *dst, *dst_orig = *dst_p, *route;
1750         u16 family = dst_orig->ops->family;
1751         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1752         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1753
1754 restart:
1755         dst = NULL;
1756         xdst = NULL;
1757         route = NULL;
1758
1759         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1760                 num_pols = 1;
1761                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1762                 err = xfrm_expand_policies(fl, family, pols,
1763                                            &num_pols, &num_xfrms);
1764                 if (err < 0)
1765                         goto dropdst;
1766
1767                 if (num_pols) {
1768                         if (num_xfrms <= 0) {
1769                                 drop_pols = num_pols;
1770                                 goto no_transform;
1771                         }
1772
1773                         xdst = xfrm_resolve_and_create_bundle(
1774                                         pols, num_pols, fl,
1775                                         family, dst_orig);
1776                         if (IS_ERR(xdst)) {
1777                                 xfrm_pols_put(pols, num_pols);
1778                                 err = PTR_ERR(xdst);
1779                                 goto dropdst;
1780                         } else if (xdst == NULL) {
1781                                 num_xfrms = 0;
1782                                 drop_pols = num_pols;
1783                                 goto no_transform;
1784                         }
1785
1786                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1787                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1788                         xfrm_policy_sk_bundles = &xdst->u.dst;
1789                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1790
1791                         route = xdst->route;
1792                 }
1793         }
1794
1795         if (xdst == NULL) {
1796                 /* To accelerate a bit...  */
1797                 if ((dst_orig->flags & DST_NOXFRM) ||
1798                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1799                         goto nopol;
1800
1801                 flo = flow_cache_lookup(net, fl, family, dir,
1802                                         xfrm_bundle_lookup, dst_orig);
1803                 if (flo == NULL)
1804                         goto nopol;
1805                 if (IS_ERR(flo)) {
1806                         err = PTR_ERR(flo);
1807                         goto dropdst;
1808                 }
1809                 xdst = container_of(flo, struct xfrm_dst, flo);
1810
1811                 num_pols = xdst->num_pols;
1812                 num_xfrms = xdst->num_xfrms;
1813                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1814                 route = xdst->route;
1815         }
1816
1817         dst = &xdst->u.dst;
1818         if (route == NULL && num_xfrms > 0) {
1819                 /* The only case when xfrm_bundle_lookup() returns a
1820                  * bundle with null route, is when the template could
1821                  * not be resolved. It means policies are there, but
1822                  * bundle could not be created, since we don't yet
1823                  * have the xfrm_state's. We need to wait for KM to
1824                  * negotiate new SA's or bail out with error.*/
1825                 if (net->xfrm.sysctl_larval_drop) {
1826                         /* EREMOTE tells the caller to generate
1827                          * a one-shot blackhole route. */
1828                         dst_release(dst);
1829                         xfrm_pols_put(pols, drop_pols);
1830                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1831                         return -EREMOTE;
1832                 }
1833                 if (flags & XFRM_LOOKUP_WAIT) {
1834                         DECLARE_WAITQUEUE(wait, current);
1835
1836                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1837                         set_current_state(TASK_INTERRUPTIBLE);
1838                         schedule();
1839                         set_current_state(TASK_RUNNING);
1840                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1841
1842                         if (!signal_pending(current)) {
1843                                 dst_release(dst);
1844                                 goto restart;
1845                         }
1846
1847                         err = -ERESTART;
1848                 } else
1849                         err = -EAGAIN;
1850
1851                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1852                 goto error;
1853         }
1854
1855 no_transform:
1856         if (num_pols == 0)
1857                 goto nopol;
1858
1859         if ((flags & XFRM_LOOKUP_ICMP) &&
1860             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1861                 err = -ENOENT;
1862                 goto error;
1863         }
1864
1865         for (i = 0; i < num_pols; i++)
1866                 pols[i]->curlft.use_time = get_seconds();
1867
1868         if (num_xfrms < 0) {
1869                 /* Prohibit the flow */
1870                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1871                 err = -EPERM;
1872                 goto error;
1873         } else if (num_xfrms > 0) {
1874                 /* Flow transformed */
1875                 *dst_p = dst;
1876                 dst_release(dst_orig);
1877         } else {
1878                 /* Flow passes untransformed */
1879                 dst_release(dst);
1880         }
1881 ok:
1882         xfrm_pols_put(pols, drop_pols);
1883         return 0;
1884
1885 nopol:
1886         if (!(flags & XFRM_LOOKUP_ICMP))
1887                 goto ok;
1888         err = -ENOENT;
1889 error:
1890         dst_release(dst);
1891 dropdst:
1892         dst_release(dst_orig);
1893         *dst_p = NULL;
1894         xfrm_pols_put(pols, drop_pols);
1895         return err;
1896 }
1897 EXPORT_SYMBOL(__xfrm_lookup);
1898
1899 int xfrm_lookup(struct net *net, struct dst_entry **dst_p,
1900                 const struct flowi *fl,
1901                 struct sock *sk, int flags)
1902 {
1903         int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1904
1905         if (err == -EREMOTE) {
1906                 dst_release(*dst_p);
1907                 *dst_p = NULL;
1908                 err = -EAGAIN;
1909         }
1910
1911         return err;
1912 }
1913 EXPORT_SYMBOL(xfrm_lookup);
1914
1915 static inline int
1916 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
1917 {
1918         struct xfrm_state *x;
1919
1920         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1921                 return 0;
1922         x = skb->sp->xvec[idx];
1923         if (!x->type->reject)
1924                 return 0;
1925         return x->type->reject(x, skb, fl);
1926 }
1927
1928 /* When skb is transformed back to its "native" form, we have to
1929  * check policy restrictions. At the moment we make this in maximally
1930  * stupid way. Shame on me. :-) Of course, connected sockets must
1931  * have policy cached at them.
1932  */
1933
1934 static inline int
1935 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1936               unsigned short family)
1937 {
1938         if (xfrm_state_kern(x))
1939                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1940         return  x->id.proto == tmpl->id.proto &&
1941                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1942                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1943                 x->props.mode == tmpl->mode &&
1944                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1945                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1946                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1947                   xfrm_state_addr_cmp(tmpl, x, family));
1948 }
1949
1950 /*
1951  * 0 or more than 0 is returned when validation is succeeded (either bypass
1952  * because of optional transport mode, or next index of the mathced secpath
1953  * state with the template.
1954  * -1 is returned when no matching template is found.
1955  * Otherwise "-2 - errored_index" is returned.
1956  */
1957 static inline int
1958 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1959                unsigned short family)
1960 {
1961         int idx = start;
1962
1963         if (tmpl->optional) {
1964                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1965                         return start;
1966         } else
1967                 start = -1;
1968         for (; idx < sp->len; idx++) {
1969                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1970                         return ++idx;
1971                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1972                         if (start == -1)
1973                                 start = -2-idx;
1974                         break;
1975                 }
1976         }
1977         return start;
1978 }
1979
1980 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1981                           unsigned int family, int reverse)
1982 {
1983         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1984         int err;
1985
1986         if (unlikely(afinfo == NULL))
1987                 return -EAFNOSUPPORT;
1988
1989         afinfo->decode_session(skb, fl, reverse);
1990         err = security_xfrm_decode_session(skb, &fl->secid);
1991         xfrm_policy_put_afinfo(afinfo);
1992         return err;
1993 }
1994 EXPORT_SYMBOL(__xfrm_decode_session);
1995
1996 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1997 {
1998         for (; k < sp->len; k++) {
1999                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2000                         *idxp = k;
2001                         return 1;
2002                 }
2003         }
2004
2005         return 0;
2006 }
2007
2008 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2009                         unsigned short family)
2010 {
2011         struct net *net = dev_net(skb->dev);
2012         struct xfrm_policy *pol;
2013         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2014         int npols = 0;
2015         int xfrm_nr;
2016         int pi;
2017         int reverse;
2018         struct flowi fl;
2019         u8 fl_dir;
2020         int xerr_idx = -1;
2021
2022         reverse = dir & ~XFRM_POLICY_MASK;
2023         dir &= XFRM_POLICY_MASK;
2024         fl_dir = policy_to_flow_dir(dir);
2025
2026         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2027                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2028                 return 0;
2029         }
2030
2031         nf_nat_decode_session(skb, &fl, family);
2032
2033         /* First, check used SA against their selectors. */
2034         if (skb->sp) {
2035                 int i;
2036
2037                 for (i=skb->sp->len-1; i>=0; i--) {
2038                         struct xfrm_state *x = skb->sp->xvec[i];
2039                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2040                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2041                                 return 0;
2042                         }
2043                 }
2044         }
2045
2046         pol = NULL;
2047         if (sk && sk->sk_policy[dir]) {
2048                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2049                 if (IS_ERR(pol)) {
2050                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2051                         return 0;
2052                 }
2053         }
2054
2055         if (!pol) {
2056                 struct flow_cache_object *flo;
2057
2058                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2059                                         xfrm_policy_lookup, NULL);
2060                 if (IS_ERR_OR_NULL(flo))
2061                         pol = ERR_CAST(flo);
2062                 else
2063                         pol = container_of(flo, struct xfrm_policy, flo);
2064         }
2065
2066         if (IS_ERR(pol)) {
2067                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2068                 return 0;
2069         }
2070
2071         if (!pol) {
2072                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2073                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2074                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2075                         return 0;
2076                 }
2077                 return 1;
2078         }
2079
2080         pol->curlft.use_time = get_seconds();
2081
2082         pols[0] = pol;
2083         npols ++;
2084 #ifdef CONFIG_XFRM_SUB_POLICY
2085         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2086                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2087                                                     &fl, family,
2088                                                     XFRM_POLICY_IN);
2089                 if (pols[1]) {
2090                         if (IS_ERR(pols[1])) {
2091                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2092                                 return 0;
2093                         }
2094                         pols[1]->curlft.use_time = get_seconds();
2095                         npols ++;
2096                 }
2097         }
2098 #endif
2099
2100         if (pol->action == XFRM_POLICY_ALLOW) {
2101                 struct sec_path *sp;
2102                 static struct sec_path dummy;
2103                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2104                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2105                 struct xfrm_tmpl **tpp = tp;
2106                 int ti = 0;
2107                 int i, k;
2108
2109                 if ((sp = skb->sp) == NULL)
2110                         sp = &dummy;
2111
2112                 for (pi = 0; pi < npols; pi++) {
2113                         if (pols[pi] != pol &&
2114                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2115                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2116                                 goto reject;
2117                         }
2118                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2119                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2120                                 goto reject_error;
2121                         }
2122                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2123                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2124                 }
2125                 xfrm_nr = ti;
2126                 if (npols > 1) {
2127                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2128                         tpp = stp;
2129                 }
2130
2131                 /* For each tunnel xfrm, find the first matching tmpl.
2132                  * For each tmpl before that, find corresponding xfrm.
2133                  * Order is _important_. Later we will implement
2134                  * some barriers, but at the moment barriers
2135                  * are implied between each two transformations.
2136                  */
2137                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2138                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2139                         if (k < 0) {
2140                                 if (k < -1)
2141                                         /* "-2 - errored_index" returned */
2142                                         xerr_idx = -(2+k);
2143                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2144                                 goto reject;
2145                         }
2146                 }
2147
2148                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2149                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2150                         goto reject;
2151                 }
2152
2153                 xfrm_pols_put(pols, npols);
2154                 return 1;
2155         }
2156         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2157
2158 reject:
2159         xfrm_secpath_reject(xerr_idx, skb, &fl);
2160 reject_error:
2161         xfrm_pols_put(pols, npols);
2162         return 0;
2163 }
2164 EXPORT_SYMBOL(__xfrm_policy_check);
2165
2166 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2167 {
2168         struct net *net = dev_net(skb->dev);
2169         struct flowi fl;
2170         struct dst_entry *dst;
2171         int res;
2172
2173         if (xfrm_decode_session(skb, &fl, family) < 0) {
2174                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2175                 return 0;
2176         }
2177
2178         skb_dst_force(skb);
2179         dst = skb_dst(skb);
2180
2181         res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
2182         skb_dst_set(skb, dst);
2183         return res;
2184 }
2185 EXPORT_SYMBOL(__xfrm_route_forward);
2186
2187 /* Optimize later using cookies and generation ids. */
2188
2189 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2190 {
2191         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2192          * to "-1" to force all XFRM destinations to get validated by
2193          * dst_ops->check on every use.  We do this because when a
2194          * normal route referenced by an XFRM dst is obsoleted we do
2195          * not go looking around for all parent referencing XFRM dsts
2196          * so that we can invalidate them.  It is just too much work.
2197          * Instead we make the checks here on every use.  For example:
2198          *
2199          *      XFRM dst A --> IPv4 dst X
2200          *
2201          * X is the "xdst->route" of A (X is also the "dst->path" of A
2202          * in this example).  If X is marked obsolete, "A" will not
2203          * notice.  That's what we are validating here via the
2204          * stale_bundle() check.
2205          *
2206          * When a policy's bundle is pruned, we dst_free() the XFRM
2207          * dst which causes it's ->obsolete field to be set to a
2208          * positive non-zero integer.  If an XFRM dst has been pruned
2209          * like this, we want to force a new route lookup.
2210          */
2211         if (dst->obsolete < 0 && !stale_bundle(dst))
2212                 return dst;
2213
2214         return NULL;
2215 }
2216
2217 static int stale_bundle(struct dst_entry *dst)
2218 {
2219         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC);
2220 }
2221
2222 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2223 {
2224         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2225                 dst->dev = dev_net(dev)->loopback_dev;
2226                 dev_hold(dst->dev);
2227                 dev_put(dev);
2228         }
2229 }
2230 EXPORT_SYMBOL(xfrm_dst_ifdown);
2231
2232 static void xfrm_link_failure(struct sk_buff *skb)
2233 {
2234         /* Impossible. Such dst must be popped before reaches point of failure. */
2235 }
2236
2237 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2238 {
2239         if (dst) {
2240                 if (dst->obsolete) {
2241                         dst_release(dst);
2242                         dst = NULL;
2243                 }
2244         }
2245         return dst;
2246 }
2247
2248 static void __xfrm_garbage_collect(struct net *net)
2249 {
2250         struct dst_entry *head, *next;
2251
2252         flow_cache_flush();
2253
2254         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2255         head = xfrm_policy_sk_bundles;
2256         xfrm_policy_sk_bundles = NULL;
2257         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2258
2259         while (head) {
2260                 next = head->next;
2261                 dst_free(head);
2262                 head = next;
2263         }
2264 }
2265
2266 static void xfrm_init_pmtu(struct dst_entry *dst)
2267 {
2268         do {
2269                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2270                 u32 pmtu, route_mtu_cached;
2271
2272                 pmtu = dst_mtu(dst->child);
2273                 xdst->child_mtu_cached = pmtu;
2274
2275                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2276
2277                 route_mtu_cached = dst_mtu(xdst->route);
2278                 xdst->route_mtu_cached = route_mtu_cached;
2279
2280                 if (pmtu > route_mtu_cached)
2281                         pmtu = route_mtu_cached;
2282
2283                 dst_metric_set(dst, RTAX_MTU, pmtu);
2284         } while ((dst = dst->next));
2285 }
2286
2287 /* Check that the bundle accepts the flow and its components are
2288  * still valid.
2289  */
2290
2291 static int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2292                           const struct flowi *fl, int family)
2293 {
2294         struct dst_entry *dst = &first->u.dst;
2295         struct xfrm_dst *last;
2296         u32 mtu;
2297
2298         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2299             (dst->dev && !netif_running(dst->dev)))
2300                 return 0;
2301 #ifdef CONFIG_XFRM_SUB_POLICY
2302         if (fl) {
2303                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2304                         return 0;
2305                 if (first->partner &&
2306                     !xfrm_selector_match(first->partner, fl, family))
2307                         return 0;
2308         }
2309 #endif
2310
2311         last = NULL;
2312
2313         do {
2314                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2315
2316                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2317                         return 0;
2318                 if (fl && pol &&
2319                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2320                         return 0;
2321                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2322                         return 0;
2323                 if (xdst->xfrm_genid != dst->xfrm->genid)
2324                         return 0;
2325                 if (xdst->num_pols > 0 &&
2326                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2327                         return 0;
2328
2329                 mtu = dst_mtu(dst->child);
2330                 if (xdst->child_mtu_cached != mtu) {
2331                         last = xdst;
2332                         xdst->child_mtu_cached = mtu;
2333                 }
2334
2335                 if (!dst_check(xdst->route, xdst->route_cookie))
2336                         return 0;
2337                 mtu = dst_mtu(xdst->route);
2338                 if (xdst->route_mtu_cached != mtu) {
2339                         last = xdst;
2340                         xdst->route_mtu_cached = mtu;
2341                 }
2342
2343                 dst = dst->child;
2344         } while (dst->xfrm);
2345
2346         if (likely(!last))
2347                 return 1;
2348
2349         mtu = last->child_mtu_cached;
2350         for (;;) {
2351                 dst = &last->u.dst;
2352
2353                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2354                 if (mtu > last->route_mtu_cached)
2355                         mtu = last->route_mtu_cached;
2356                 dst_metric_set(dst, RTAX_MTU, mtu);
2357
2358                 if (last == first)
2359                         break;
2360
2361                 last = (struct xfrm_dst *)last->u.dst.next;
2362                 last->child_mtu_cached = mtu;
2363         }
2364
2365         return 1;
2366 }
2367
2368 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2369 {
2370         return dst_metric_advmss(dst->path);
2371 }
2372
2373 static unsigned int xfrm_default_mtu(const struct dst_entry *dst)
2374 {
2375         return dst_mtu(dst->path);
2376 }
2377
2378 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2379 {
2380         struct net *net;
2381         int err = 0;
2382         if (unlikely(afinfo == NULL))
2383                 return -EINVAL;
2384         if (unlikely(afinfo->family >= NPROTO))
2385                 return -EAFNOSUPPORT;
2386         write_lock_bh(&xfrm_policy_afinfo_lock);
2387         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2388                 err = -ENOBUFS;
2389         else {
2390                 struct dst_ops *dst_ops = afinfo->dst_ops;
2391                 if (likely(dst_ops->kmem_cachep == NULL))
2392                         dst_ops->kmem_cachep = xfrm_dst_cache;
2393                 if (likely(dst_ops->check == NULL))
2394                         dst_ops->check = xfrm_dst_check;
2395                 if (likely(dst_ops->default_advmss == NULL))
2396                         dst_ops->default_advmss = xfrm_default_advmss;
2397                 if (likely(dst_ops->default_mtu == NULL))
2398                         dst_ops->default_mtu = xfrm_default_mtu;
2399                 if (likely(dst_ops->negative_advice == NULL))
2400                         dst_ops->negative_advice = xfrm_negative_advice;
2401                 if (likely(dst_ops->link_failure == NULL))
2402                         dst_ops->link_failure = xfrm_link_failure;
2403                 if (likely(afinfo->garbage_collect == NULL))
2404                         afinfo->garbage_collect = __xfrm_garbage_collect;
2405                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2406         }
2407         write_unlock_bh(&xfrm_policy_afinfo_lock);
2408
2409         rtnl_lock();
2410         for_each_net(net) {
2411                 struct dst_ops *xfrm_dst_ops;
2412
2413                 switch (afinfo->family) {
2414                 case AF_INET:
2415                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2416                         break;
2417 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2418                 case AF_INET6:
2419                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2420                         break;
2421 #endif
2422                 default:
2423                         BUG();
2424                 }
2425                 *xfrm_dst_ops = *afinfo->dst_ops;
2426         }
2427         rtnl_unlock();
2428
2429         return err;
2430 }
2431 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2432
2433 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2434 {
2435         int err = 0;
2436         if (unlikely(afinfo == NULL))
2437                 return -EINVAL;
2438         if (unlikely(afinfo->family >= NPROTO))
2439                 return -EAFNOSUPPORT;
2440         write_lock_bh(&xfrm_policy_afinfo_lock);
2441         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2442                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2443                         err = -EINVAL;
2444                 else {
2445                         struct dst_ops *dst_ops = afinfo->dst_ops;
2446                         xfrm_policy_afinfo[afinfo->family] = NULL;
2447                         dst_ops->kmem_cachep = NULL;
2448                         dst_ops->check = NULL;
2449                         dst_ops->negative_advice = NULL;
2450                         dst_ops->link_failure = NULL;
2451                         afinfo->garbage_collect = NULL;
2452                 }
2453         }
2454         write_unlock_bh(&xfrm_policy_afinfo_lock);
2455         return err;
2456 }
2457 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2458
2459 static void __net_init xfrm_dst_ops_init(struct net *net)
2460 {
2461         struct xfrm_policy_afinfo *afinfo;
2462
2463         read_lock_bh(&xfrm_policy_afinfo_lock);
2464         afinfo = xfrm_policy_afinfo[AF_INET];
2465         if (afinfo)
2466                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2467 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2468         afinfo = xfrm_policy_afinfo[AF_INET6];
2469         if (afinfo)
2470                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2471 #endif
2472         read_unlock_bh(&xfrm_policy_afinfo_lock);
2473 }
2474
2475 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2476 {
2477         struct xfrm_policy_afinfo *afinfo;
2478         if (unlikely(family >= NPROTO))
2479                 return NULL;
2480         read_lock(&xfrm_policy_afinfo_lock);
2481         afinfo = xfrm_policy_afinfo[family];
2482         if (unlikely(!afinfo))
2483                 read_unlock(&xfrm_policy_afinfo_lock);
2484         return afinfo;
2485 }
2486
2487 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2488 {
2489         read_unlock(&xfrm_policy_afinfo_lock);
2490 }
2491
2492 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2493 {
2494         struct net_device *dev = ptr;
2495
2496         switch (event) {
2497         case NETDEV_DOWN:
2498                 __xfrm_garbage_collect(dev_net(dev));
2499         }
2500         return NOTIFY_DONE;
2501 }
2502
2503 static struct notifier_block xfrm_dev_notifier = {
2504         .notifier_call  = xfrm_dev_event,
2505 };
2506
2507 #ifdef CONFIG_XFRM_STATISTICS
2508 static int __net_init xfrm_statistics_init(struct net *net)
2509 {
2510         int rv;
2511
2512         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2513                           sizeof(struct linux_xfrm_mib),
2514                           __alignof__(struct linux_xfrm_mib)) < 0)
2515                 return -ENOMEM;
2516         rv = xfrm_proc_init(net);
2517         if (rv < 0)
2518                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2519         return rv;
2520 }
2521
2522 static void xfrm_statistics_fini(struct net *net)
2523 {
2524         xfrm_proc_fini(net);
2525         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2526 }
2527 #else
2528 static int __net_init xfrm_statistics_init(struct net *net)
2529 {
2530         return 0;
2531 }
2532
2533 static void xfrm_statistics_fini(struct net *net)
2534 {
2535 }
2536 #endif
2537
2538 static int __net_init xfrm_policy_init(struct net *net)
2539 {
2540         unsigned int hmask, sz;
2541         int dir;
2542
2543         if (net_eq(net, &init_net))
2544                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2545                                            sizeof(struct xfrm_dst),
2546                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2547                                            NULL);
2548
2549         hmask = 8 - 1;
2550         sz = (hmask+1) * sizeof(struct hlist_head);
2551
2552         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2553         if (!net->xfrm.policy_byidx)
2554                 goto out_byidx;
2555         net->xfrm.policy_idx_hmask = hmask;
2556
2557         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2558                 struct xfrm_policy_hash *htab;
2559
2560                 net->xfrm.policy_count[dir] = 0;
2561                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2562
2563                 htab = &net->xfrm.policy_bydst[dir];
2564                 htab->table = xfrm_hash_alloc(sz);
2565                 if (!htab->table)
2566                         goto out_bydst;
2567                 htab->hmask = hmask;
2568         }
2569
2570         INIT_LIST_HEAD(&net->xfrm.policy_all);
2571         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2572         if (net_eq(net, &init_net))
2573                 register_netdevice_notifier(&xfrm_dev_notifier);
2574         return 0;
2575
2576 out_bydst:
2577         for (dir--; dir >= 0; dir--) {
2578                 struct xfrm_policy_hash *htab;
2579
2580                 htab = &net->xfrm.policy_bydst[dir];
2581                 xfrm_hash_free(htab->table, sz);
2582         }
2583         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2584 out_byidx:
2585         return -ENOMEM;
2586 }
2587
2588 static void xfrm_policy_fini(struct net *net)
2589 {
2590         struct xfrm_audit audit_info;
2591         unsigned int sz;
2592         int dir;
2593
2594         flush_work(&net->xfrm.policy_hash_work);
2595 #ifdef CONFIG_XFRM_SUB_POLICY
2596         audit_info.loginuid = -1;
2597         audit_info.sessionid = -1;
2598         audit_info.secid = 0;
2599         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2600 #endif
2601         audit_info.loginuid = -1;
2602         audit_info.sessionid = -1;
2603         audit_info.secid = 0;
2604         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2605
2606         WARN_ON(!list_empty(&net->xfrm.policy_all));
2607
2608         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2609                 struct xfrm_policy_hash *htab;
2610
2611                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2612
2613                 htab = &net->xfrm.policy_bydst[dir];
2614                 sz = (htab->hmask + 1);
2615                 WARN_ON(!hlist_empty(htab->table));
2616                 xfrm_hash_free(htab->table, sz);
2617         }
2618
2619         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2620         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2621         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2622 }
2623
2624 static int __net_init xfrm_net_init(struct net *net)
2625 {
2626         int rv;
2627
2628         rv = xfrm_statistics_init(net);
2629         if (rv < 0)
2630                 goto out_statistics;
2631         rv = xfrm_state_init(net);
2632         if (rv < 0)
2633                 goto out_state;
2634         rv = xfrm_policy_init(net);
2635         if (rv < 0)
2636                 goto out_policy;
2637         xfrm_dst_ops_init(net);
2638         rv = xfrm_sysctl_init(net);
2639         if (rv < 0)
2640                 goto out_sysctl;
2641         return 0;
2642
2643 out_sysctl:
2644         xfrm_policy_fini(net);
2645 out_policy:
2646         xfrm_state_fini(net);
2647 out_state:
2648         xfrm_statistics_fini(net);
2649 out_statistics:
2650         return rv;
2651 }
2652
2653 static void __net_exit xfrm_net_exit(struct net *net)
2654 {
2655         xfrm_sysctl_fini(net);
2656         xfrm_policy_fini(net);
2657         xfrm_state_fini(net);
2658         xfrm_statistics_fini(net);
2659 }
2660
2661 static struct pernet_operations __net_initdata xfrm_net_ops = {
2662         .init = xfrm_net_init,
2663         .exit = xfrm_net_exit,
2664 };
2665
2666 void __init xfrm_init(void)
2667 {
2668         register_pernet_subsys(&xfrm_net_ops);
2669         xfrm_input_init();
2670 }
2671
2672 #ifdef CONFIG_AUDITSYSCALL
2673 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2674                                          struct audit_buffer *audit_buf)
2675 {
2676         struct xfrm_sec_ctx *ctx = xp->security;
2677         struct xfrm_selector *sel = &xp->selector;
2678
2679         if (ctx)
2680                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2681                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2682
2683         switch(sel->family) {
2684         case AF_INET:
2685                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2686                 if (sel->prefixlen_s != 32)
2687                         audit_log_format(audit_buf, " src_prefixlen=%d",
2688                                          sel->prefixlen_s);
2689                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2690                 if (sel->prefixlen_d != 32)
2691                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2692                                          sel->prefixlen_d);
2693                 break;
2694         case AF_INET6:
2695                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2696                 if (sel->prefixlen_s != 128)
2697                         audit_log_format(audit_buf, " src_prefixlen=%d",
2698                                          sel->prefixlen_s);
2699                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2700                 if (sel->prefixlen_d != 128)
2701                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2702                                          sel->prefixlen_d);
2703                 break;
2704         }
2705 }
2706
2707 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2708                            uid_t auid, u32 sessionid, u32 secid)
2709 {
2710         struct audit_buffer *audit_buf;
2711
2712         audit_buf = xfrm_audit_start("SPD-add");
2713         if (audit_buf == NULL)
2714                 return;
2715         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2716         audit_log_format(audit_buf, " res=%u", result);
2717         xfrm_audit_common_policyinfo(xp, audit_buf);
2718         audit_log_end(audit_buf);
2719 }
2720 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2721
2722 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2723                               uid_t auid, u32 sessionid, u32 secid)
2724 {
2725         struct audit_buffer *audit_buf;
2726
2727         audit_buf = xfrm_audit_start("SPD-delete");
2728         if (audit_buf == NULL)
2729                 return;
2730         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2731         audit_log_format(audit_buf, " res=%u", result);
2732         xfrm_audit_common_policyinfo(xp, audit_buf);
2733         audit_log_end(audit_buf);
2734 }
2735 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2736 #endif
2737
2738 #ifdef CONFIG_XFRM_MIGRATE
2739 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2740                                        struct xfrm_selector *sel_tgt)
2741 {
2742         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2743                 if (sel_tgt->family == sel_cmp->family &&
2744                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2745                                   sel_cmp->family) == 0 &&
2746                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2747                                   sel_cmp->family) == 0 &&
2748                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2749                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2750                         return 1;
2751                 }
2752         } else {
2753                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2754                         return 1;
2755                 }
2756         }
2757         return 0;
2758 }
2759
2760 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2761                                                      u8 dir, u8 type)
2762 {
2763         struct xfrm_policy *pol, *ret = NULL;
2764         struct hlist_node *entry;
2765         struct hlist_head *chain;
2766         u32 priority = ~0U;
2767
2768         read_lock_bh(&xfrm_policy_lock);
2769         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2770         hlist_for_each_entry(pol, entry, chain, bydst) {
2771                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2772                     pol->type == type) {
2773                         ret = pol;
2774                         priority = ret->priority;
2775                         break;
2776                 }
2777         }
2778         chain = &init_net.xfrm.policy_inexact[dir];
2779         hlist_for_each_entry(pol, entry, chain, bydst) {
2780                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2781                     pol->type == type &&
2782                     pol->priority < priority) {
2783                         ret = pol;
2784                         break;
2785                 }
2786         }
2787
2788         if (ret)
2789                 xfrm_pol_hold(ret);
2790
2791         read_unlock_bh(&xfrm_policy_lock);
2792
2793         return ret;
2794 }
2795
2796 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2797 {
2798         int match = 0;
2799
2800         if (t->mode == m->mode && t->id.proto == m->proto &&
2801             (m->reqid == 0 || t->reqid == m->reqid)) {
2802                 switch (t->mode) {
2803                 case XFRM_MODE_TUNNEL:
2804                 case XFRM_MODE_BEET:
2805                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2806                                           m->old_family) == 0 &&
2807                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2808                                           m->old_family) == 0) {
2809                                 match = 1;
2810                         }
2811                         break;
2812                 case XFRM_MODE_TRANSPORT:
2813                         /* in case of transport mode, template does not store
2814                            any IP addresses, hence we just compare mode and
2815                            protocol */
2816                         match = 1;
2817                         break;
2818                 default:
2819                         break;
2820                 }
2821         }
2822         return match;
2823 }
2824
2825 /* update endpoint address(es) of template(s) */
2826 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2827                                struct xfrm_migrate *m, int num_migrate)
2828 {
2829         struct xfrm_migrate *mp;
2830         int i, j, n = 0;
2831
2832         write_lock_bh(&pol->lock);
2833         if (unlikely(pol->walk.dead)) {
2834                 /* target policy has been deleted */
2835                 write_unlock_bh(&pol->lock);
2836                 return -ENOENT;
2837         }
2838
2839         for (i = 0; i < pol->xfrm_nr; i++) {
2840                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2841                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2842                                 continue;
2843                         n++;
2844                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2845                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2846                                 continue;
2847                         /* update endpoints */
2848                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2849                                sizeof(pol->xfrm_vec[i].id.daddr));
2850                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2851                                sizeof(pol->xfrm_vec[i].saddr));
2852                         pol->xfrm_vec[i].encap_family = mp->new_family;
2853                         /* flush bundles */
2854                         atomic_inc(&pol->genid);
2855                 }
2856         }
2857
2858         write_unlock_bh(&pol->lock);
2859
2860         if (!n)
2861                 return -ENODATA;
2862
2863         return 0;
2864 }
2865
2866 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
2867 {
2868         int i, j;
2869
2870         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2871                 return -EINVAL;
2872
2873         for (i = 0; i < num_migrate; i++) {
2874                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2875                                    m[i].old_family) == 0) &&
2876                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2877                                    m[i].old_family) == 0))
2878                         return -EINVAL;
2879                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2880                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2881                         return -EINVAL;
2882
2883                 /* check if there is any duplicated entry */
2884                 for (j = i + 1; j < num_migrate; j++) {
2885                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2886                                     sizeof(m[i].old_daddr)) &&
2887                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2888                                     sizeof(m[i].old_saddr)) &&
2889                             m[i].proto == m[j].proto &&
2890                             m[i].mode == m[j].mode &&
2891                             m[i].reqid == m[j].reqid &&
2892                             m[i].old_family == m[j].old_family)
2893                                 return -EINVAL;
2894                 }
2895         }
2896
2897         return 0;
2898 }
2899
2900 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2901                  struct xfrm_migrate *m, int num_migrate,
2902                  struct xfrm_kmaddress *k)
2903 {
2904         int i, err, nx_cur = 0, nx_new = 0;
2905         struct xfrm_policy *pol = NULL;
2906         struct xfrm_state *x, *xc;
2907         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2908         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2909         struct xfrm_migrate *mp;
2910
2911         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2912                 goto out;
2913
2914         /* Stage 1 - find policy */
2915         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2916                 err = -ENOENT;
2917                 goto out;
2918         }
2919
2920         /* Stage 2 - find and update state(s) */
2921         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2922                 if ((x = xfrm_migrate_state_find(mp))) {
2923                         x_cur[nx_cur] = x;
2924                         nx_cur++;
2925                         if ((xc = xfrm_state_migrate(x, mp))) {
2926                                 x_new[nx_new] = xc;
2927                                 nx_new++;
2928                         } else {
2929                                 err = -ENODATA;
2930                                 goto restore_state;
2931                         }
2932                 }
2933         }
2934
2935         /* Stage 3 - update policy */
2936         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2937                 goto restore_state;
2938
2939         /* Stage 4 - delete old state(s) */
2940         if (nx_cur) {
2941                 xfrm_states_put(x_cur, nx_cur);
2942                 xfrm_states_delete(x_cur, nx_cur);
2943         }
2944
2945         /* Stage 5 - announce */
2946         km_migrate(sel, dir, type, m, num_migrate, k);
2947
2948         xfrm_pol_put(pol);
2949
2950         return 0;
2951 out:
2952         return err;
2953
2954 restore_state:
2955         if (pol)
2956                 xfrm_pol_put(pol);
2957         if (nx_cur)
2958                 xfrm_states_put(x_cur, nx_cur);
2959         if (nx_new)
2960                 xfrm_states_delete(x_new, nx_new);
2961
2962         return err;
2963 }
2964 EXPORT_SYMBOL(xfrm_migrate);
2965 #endif