xfrm: Add support for IPsec extended sequence numbers
[linux-2.6.git] / net / xfrm / xfrm_policy.c
1 /*
2  * xfrm_policy.c
3  *
4  * Changes:
5  *      Mitsuru KANDA @USAGI
6  *      Kazunori MIYAZAWA @USAGI
7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8  *              IPv6 support
9  *      Kazunori MIYAZAWA @USAGI
10  *      YOSHIFUJI Hideaki
11  *              Split up af-specific portion
12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
13  *
14  */
15
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
28 #include <net/dst.h>
29 #include <net/xfrm.h>
30 #include <net/ip.h>
31 #ifdef CONFIG_XFRM_STATISTICS
32 #include <net/snmp.h>
33 #endif
34
35 #include "xfrm_hash.h"
36
37 DEFINE_MUTEX(xfrm_cfg_mutex);
38 EXPORT_SYMBOL(xfrm_cfg_mutex);
39
40 static DEFINE_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_dst *xdst, int family);
54
55
56 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
57                                                 int dir);
58
59 static inline int
60 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
61 {
62         const struct flowi4 *fl4 = &fl->u.ip4;
63
64         return  addr_match(&fl4->daddr, &sel->daddr, sel->prefixlen_d) &&
65                 addr_match(&fl4->saddr, &sel->saddr, sel->prefixlen_s) &&
66                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
67                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
68                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
69                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
70 }
71
72 static inline int
73 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
74 {
75         const struct flowi6 *fl6 = &fl->u.ip6;
76
77         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
78                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
79                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
80                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
81                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
82                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
83 }
84
85 int xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
86                         unsigned short family)
87 {
88         switch (family) {
89         case AF_INET:
90                 return __xfrm4_selector_match(sel, fl);
91         case AF_INET6:
92                 return __xfrm6_selector_match(sel, fl);
93         }
94         return 0;
95 }
96
97 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
98                                                   const xfrm_address_t *saddr,
99                                                   const xfrm_address_t *daddr,
100                                                   int family)
101 {
102         struct xfrm_policy_afinfo *afinfo;
103         struct dst_entry *dst;
104
105         afinfo = xfrm_policy_get_afinfo(family);
106         if (unlikely(afinfo == NULL))
107                 return ERR_PTR(-EAFNOSUPPORT);
108
109         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
110
111         xfrm_policy_put_afinfo(afinfo);
112
113         return dst;
114 }
115
116 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
117                                                 xfrm_address_t *prev_saddr,
118                                                 xfrm_address_t *prev_daddr,
119                                                 int family)
120 {
121         struct net *net = xs_net(x);
122         xfrm_address_t *saddr = &x->props.saddr;
123         xfrm_address_t *daddr = &x->id.daddr;
124         struct dst_entry *dst;
125
126         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
127                 saddr = x->coaddr;
128                 daddr = prev_daddr;
129         }
130         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
131                 saddr = prev_saddr;
132                 daddr = x->coaddr;
133         }
134
135         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
136
137         if (!IS_ERR(dst)) {
138                 if (prev_saddr != saddr)
139                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
140                 if (prev_daddr != daddr)
141                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
142         }
143
144         return dst;
145 }
146
147 static inline unsigned long make_jiffies(long secs)
148 {
149         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
150                 return MAX_SCHEDULE_TIMEOUT-1;
151         else
152                 return secs*HZ;
153 }
154
155 static void xfrm_policy_timer(unsigned long data)
156 {
157         struct xfrm_policy *xp = (struct xfrm_policy*)data;
158         unsigned long now = get_seconds();
159         long next = LONG_MAX;
160         int warn = 0;
161         int dir;
162
163         read_lock(&xp->lock);
164
165         if (unlikely(xp->walk.dead))
166                 goto out;
167
168         dir = xfrm_policy_id2dir(xp->index);
169
170         if (xp->lft.hard_add_expires_seconds) {
171                 long tmo = xp->lft.hard_add_expires_seconds +
172                         xp->curlft.add_time - now;
173                 if (tmo <= 0)
174                         goto expired;
175                 if (tmo < next)
176                         next = tmo;
177         }
178         if (xp->lft.hard_use_expires_seconds) {
179                 long tmo = xp->lft.hard_use_expires_seconds +
180                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
181                 if (tmo <= 0)
182                         goto expired;
183                 if (tmo < next)
184                         next = tmo;
185         }
186         if (xp->lft.soft_add_expires_seconds) {
187                 long tmo = xp->lft.soft_add_expires_seconds +
188                         xp->curlft.add_time - now;
189                 if (tmo <= 0) {
190                         warn = 1;
191                         tmo = XFRM_KM_TIMEOUT;
192                 }
193                 if (tmo < next)
194                         next = tmo;
195         }
196         if (xp->lft.soft_use_expires_seconds) {
197                 long tmo = xp->lft.soft_use_expires_seconds +
198                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
199                 if (tmo <= 0) {
200                         warn = 1;
201                         tmo = XFRM_KM_TIMEOUT;
202                 }
203                 if (tmo < next)
204                         next = tmo;
205         }
206
207         if (warn)
208                 km_policy_expired(xp, dir, 0, 0);
209         if (next != LONG_MAX &&
210             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
211                 xfrm_pol_hold(xp);
212
213 out:
214         read_unlock(&xp->lock);
215         xfrm_pol_put(xp);
216         return;
217
218 expired:
219         read_unlock(&xp->lock);
220         if (!xfrm_policy_delete(xp, dir))
221                 km_policy_expired(xp, dir, 1, 0);
222         xfrm_pol_put(xp);
223 }
224
225 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
226 {
227         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
228
229         if (unlikely(pol->walk.dead))
230                 flo = NULL;
231         else
232                 xfrm_pol_hold(pol);
233
234         return flo;
235 }
236
237 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
238 {
239         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
240
241         return !pol->walk.dead;
242 }
243
244 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
245 {
246         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
247 }
248
249 static const struct flow_cache_ops xfrm_policy_fc_ops = {
250         .get = xfrm_policy_flo_get,
251         .check = xfrm_policy_flo_check,
252         .delete = xfrm_policy_flo_delete,
253 };
254
255 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
256  * SPD calls.
257  */
258
259 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
260 {
261         struct xfrm_policy *policy;
262
263         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
264
265         if (policy) {
266                 write_pnet(&policy->xp_net, net);
267                 INIT_LIST_HEAD(&policy->walk.all);
268                 INIT_HLIST_NODE(&policy->bydst);
269                 INIT_HLIST_NODE(&policy->byidx);
270                 rwlock_init(&policy->lock);
271                 atomic_set(&policy->refcnt, 1);
272                 setup_timer(&policy->timer, xfrm_policy_timer,
273                                 (unsigned long)policy);
274                 policy->flo.ops = &xfrm_policy_fc_ops;
275         }
276         return policy;
277 }
278 EXPORT_SYMBOL(xfrm_policy_alloc);
279
280 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
281
282 void xfrm_policy_destroy(struct xfrm_policy *policy)
283 {
284         BUG_ON(!policy->walk.dead);
285
286         if (del_timer(&policy->timer))
287                 BUG();
288
289         security_xfrm_policy_free(policy->security);
290         kfree(policy);
291 }
292 EXPORT_SYMBOL(xfrm_policy_destroy);
293
294 /* Rule must be locked. Release descentant resources, announce
295  * entry dead. The rule must be unlinked from lists to the moment.
296  */
297
298 static void xfrm_policy_kill(struct xfrm_policy *policy)
299 {
300         policy->walk.dead = 1;
301
302         atomic_inc(&policy->genid);
303
304         if (del_timer(&policy->timer))
305                 xfrm_pol_put(policy);
306
307         xfrm_pol_put(policy);
308 }
309
310 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
311
312 static inline unsigned int idx_hash(struct net *net, u32 index)
313 {
314         return __idx_hash(index, net->xfrm.policy_idx_hmask);
315 }
316
317 static struct hlist_head *policy_hash_bysel(struct net *net,
318                                             const struct xfrm_selector *sel,
319                                             unsigned short family, int dir)
320 {
321         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
322         unsigned int hash = __sel_hash(sel, family, hmask);
323
324         return (hash == hmask + 1 ?
325                 &net->xfrm.policy_inexact[dir] :
326                 net->xfrm.policy_bydst[dir].table + hash);
327 }
328
329 static struct hlist_head *policy_hash_direct(struct net *net,
330                                              const xfrm_address_t *daddr,
331                                              const xfrm_address_t *saddr,
332                                              unsigned short family, int dir)
333 {
334         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
335         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
336
337         return net->xfrm.policy_bydst[dir].table + hash;
338 }
339
340 static void xfrm_dst_hash_transfer(struct hlist_head *list,
341                                    struct hlist_head *ndsttable,
342                                    unsigned int nhashmask)
343 {
344         struct hlist_node *entry, *tmp, *entry0 = NULL;
345         struct xfrm_policy *pol;
346         unsigned int h0 = 0;
347
348 redo:
349         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
350                 unsigned int h;
351
352                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
353                                 pol->family, nhashmask);
354                 if (!entry0) {
355                         hlist_del(entry);
356                         hlist_add_head(&pol->bydst, ndsttable+h);
357                         h0 = h;
358                 } else {
359                         if (h != h0)
360                                 continue;
361                         hlist_del(entry);
362                         hlist_add_after(entry0, &pol->bydst);
363                 }
364                 entry0 = entry;
365         }
366         if (!hlist_empty(list)) {
367                 entry0 = NULL;
368                 goto redo;
369         }
370 }
371
372 static void xfrm_idx_hash_transfer(struct hlist_head *list,
373                                    struct hlist_head *nidxtable,
374                                    unsigned int nhashmask)
375 {
376         struct hlist_node *entry, *tmp;
377         struct xfrm_policy *pol;
378
379         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
380                 unsigned int h;
381
382                 h = __idx_hash(pol->index, nhashmask);
383                 hlist_add_head(&pol->byidx, nidxtable+h);
384         }
385 }
386
387 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
388 {
389         return ((old_hmask + 1) << 1) - 1;
390 }
391
392 static void xfrm_bydst_resize(struct net *net, int dir)
393 {
394         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
395         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
396         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
397         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
398         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
399         int i;
400
401         if (!ndst)
402                 return;
403
404         write_lock_bh(&xfrm_policy_lock);
405
406         for (i = hmask; i >= 0; i--)
407                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
408
409         net->xfrm.policy_bydst[dir].table = ndst;
410         net->xfrm.policy_bydst[dir].hmask = nhashmask;
411
412         write_unlock_bh(&xfrm_policy_lock);
413
414         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
415 }
416
417 static void xfrm_byidx_resize(struct net *net, int total)
418 {
419         unsigned int hmask = net->xfrm.policy_idx_hmask;
420         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
421         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
422         struct hlist_head *oidx = net->xfrm.policy_byidx;
423         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
424         int i;
425
426         if (!nidx)
427                 return;
428
429         write_lock_bh(&xfrm_policy_lock);
430
431         for (i = hmask; i >= 0; i--)
432                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
433
434         net->xfrm.policy_byidx = nidx;
435         net->xfrm.policy_idx_hmask = nhashmask;
436
437         write_unlock_bh(&xfrm_policy_lock);
438
439         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
440 }
441
442 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
443 {
444         unsigned int cnt = net->xfrm.policy_count[dir];
445         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
446
447         if (total)
448                 *total += cnt;
449
450         if ((hmask + 1) < xfrm_policy_hashmax &&
451             cnt > hmask)
452                 return 1;
453
454         return 0;
455 }
456
457 static inline int xfrm_byidx_should_resize(struct net *net, int total)
458 {
459         unsigned int hmask = net->xfrm.policy_idx_hmask;
460
461         if ((hmask + 1) < xfrm_policy_hashmax &&
462             total > hmask)
463                 return 1;
464
465         return 0;
466 }
467
468 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
469 {
470         read_lock_bh(&xfrm_policy_lock);
471         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
472         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
473         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
474         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
475         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
476         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
477         si->spdhcnt = net->xfrm.policy_idx_hmask;
478         si->spdhmcnt = xfrm_policy_hashmax;
479         read_unlock_bh(&xfrm_policy_lock);
480 }
481 EXPORT_SYMBOL(xfrm_spd_getinfo);
482
483 static DEFINE_MUTEX(hash_resize_mutex);
484 static void xfrm_hash_resize(struct work_struct *work)
485 {
486         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
487         int dir, total;
488
489         mutex_lock(&hash_resize_mutex);
490
491         total = 0;
492         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
493                 if (xfrm_bydst_should_resize(net, dir, &total))
494                         xfrm_bydst_resize(net, dir);
495         }
496         if (xfrm_byidx_should_resize(net, total))
497                 xfrm_byidx_resize(net, total);
498
499         mutex_unlock(&hash_resize_mutex);
500 }
501
502 /* Generate new index... KAME seems to generate them ordered by cost
503  * of an absolute inpredictability of ordering of rules. This will not pass. */
504 static u32 xfrm_gen_index(struct net *net, int dir)
505 {
506         static u32 idx_generator;
507
508         for (;;) {
509                 struct hlist_node *entry;
510                 struct hlist_head *list;
511                 struct xfrm_policy *p;
512                 u32 idx;
513                 int found;
514
515                 idx = (idx_generator | dir);
516                 idx_generator += 8;
517                 if (idx == 0)
518                         idx = 8;
519                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
520                 found = 0;
521                 hlist_for_each_entry(p, entry, list, byidx) {
522                         if (p->index == idx) {
523                                 found = 1;
524                                 break;
525                         }
526                 }
527                 if (!found)
528                         return idx;
529         }
530 }
531
532 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
533 {
534         u32 *p1 = (u32 *) s1;
535         u32 *p2 = (u32 *) s2;
536         int len = sizeof(struct xfrm_selector) / sizeof(u32);
537         int i;
538
539         for (i = 0; i < len; i++) {
540                 if (p1[i] != p2[i])
541                         return 1;
542         }
543
544         return 0;
545 }
546
547 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
548 {
549         struct net *net = xp_net(policy);
550         struct xfrm_policy *pol;
551         struct xfrm_policy *delpol;
552         struct hlist_head *chain;
553         struct hlist_node *entry, *newpos;
554         u32 mark = policy->mark.v & policy->mark.m;
555
556         write_lock_bh(&xfrm_policy_lock);
557         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
558         delpol = NULL;
559         newpos = NULL;
560         hlist_for_each_entry(pol, entry, chain, bydst) {
561                 if (pol->type == policy->type &&
562                     !selector_cmp(&pol->selector, &policy->selector) &&
563                     (mark & pol->mark.m) == pol->mark.v &&
564                     xfrm_sec_ctx_match(pol->security, policy->security) &&
565                     !WARN_ON(delpol)) {
566                         if (excl) {
567                                 write_unlock_bh(&xfrm_policy_lock);
568                                 return -EEXIST;
569                         }
570                         delpol = pol;
571                         if (policy->priority > pol->priority)
572                                 continue;
573                 } else if (policy->priority >= pol->priority) {
574                         newpos = &pol->bydst;
575                         continue;
576                 }
577                 if (delpol)
578                         break;
579         }
580         if (newpos)
581                 hlist_add_after(newpos, &policy->bydst);
582         else
583                 hlist_add_head(&policy->bydst, chain);
584         xfrm_pol_hold(policy);
585         net->xfrm.policy_count[dir]++;
586         atomic_inc(&flow_cache_genid);
587         if (delpol)
588                 __xfrm_policy_unlink(delpol, dir);
589         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
590         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
591         policy->curlft.add_time = get_seconds();
592         policy->curlft.use_time = 0;
593         if (!mod_timer(&policy->timer, jiffies + HZ))
594                 xfrm_pol_hold(policy);
595         list_add(&policy->walk.all, &net->xfrm.policy_all);
596         write_unlock_bh(&xfrm_policy_lock);
597
598         if (delpol)
599                 xfrm_policy_kill(delpol);
600         else if (xfrm_bydst_should_resize(net, dir, NULL))
601                 schedule_work(&net->xfrm.policy_hash_work);
602
603         return 0;
604 }
605 EXPORT_SYMBOL(xfrm_policy_insert);
606
607 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
608                                           int dir, struct xfrm_selector *sel,
609                                           struct xfrm_sec_ctx *ctx, int delete,
610                                           int *err)
611 {
612         struct xfrm_policy *pol, *ret;
613         struct hlist_head *chain;
614         struct hlist_node *entry;
615
616         *err = 0;
617         write_lock_bh(&xfrm_policy_lock);
618         chain = policy_hash_bysel(net, sel, sel->family, dir);
619         ret = NULL;
620         hlist_for_each_entry(pol, entry, chain, bydst) {
621                 if (pol->type == type &&
622                     (mark & pol->mark.m) == pol->mark.v &&
623                     !selector_cmp(sel, &pol->selector) &&
624                     xfrm_sec_ctx_match(ctx, pol->security)) {
625                         xfrm_pol_hold(pol);
626                         if (delete) {
627                                 *err = security_xfrm_policy_delete(
628                                                                 pol->security);
629                                 if (*err) {
630                                         write_unlock_bh(&xfrm_policy_lock);
631                                         return pol;
632                                 }
633                                 __xfrm_policy_unlink(pol, dir);
634                         }
635                         ret = pol;
636                         break;
637                 }
638         }
639         write_unlock_bh(&xfrm_policy_lock);
640
641         if (ret && delete)
642                 xfrm_policy_kill(ret);
643         return ret;
644 }
645 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
646
647 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
648                                      int dir, u32 id, int delete, int *err)
649 {
650         struct xfrm_policy *pol, *ret;
651         struct hlist_head *chain;
652         struct hlist_node *entry;
653
654         *err = -ENOENT;
655         if (xfrm_policy_id2dir(id) != dir)
656                 return NULL;
657
658         *err = 0;
659         write_lock_bh(&xfrm_policy_lock);
660         chain = net->xfrm.policy_byidx + idx_hash(net, id);
661         ret = NULL;
662         hlist_for_each_entry(pol, entry, chain, byidx) {
663                 if (pol->type == type && pol->index == id &&
664                     (mark & pol->mark.m) == pol->mark.v) {
665                         xfrm_pol_hold(pol);
666                         if (delete) {
667                                 *err = security_xfrm_policy_delete(
668                                                                 pol->security);
669                                 if (*err) {
670                                         write_unlock_bh(&xfrm_policy_lock);
671                                         return pol;
672                                 }
673                                 __xfrm_policy_unlink(pol, dir);
674                         }
675                         ret = pol;
676                         break;
677                 }
678         }
679         write_unlock_bh(&xfrm_policy_lock);
680
681         if (ret && delete)
682                 xfrm_policy_kill(ret);
683         return ret;
684 }
685 EXPORT_SYMBOL(xfrm_policy_byid);
686
687 #ifdef CONFIG_SECURITY_NETWORK_XFRM
688 static inline int
689 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
690 {
691         int dir, err = 0;
692
693         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
694                 struct xfrm_policy *pol;
695                 struct hlist_node *entry;
696                 int i;
697
698                 hlist_for_each_entry(pol, entry,
699                                      &net->xfrm.policy_inexact[dir], bydst) {
700                         if (pol->type != type)
701                                 continue;
702                         err = security_xfrm_policy_delete(pol->security);
703                         if (err) {
704                                 xfrm_audit_policy_delete(pol, 0,
705                                                          audit_info->loginuid,
706                                                          audit_info->sessionid,
707                                                          audit_info->secid);
708                                 return err;
709                         }
710                 }
711                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
712                         hlist_for_each_entry(pol, entry,
713                                              net->xfrm.policy_bydst[dir].table + i,
714                                              bydst) {
715                                 if (pol->type != type)
716                                         continue;
717                                 err = security_xfrm_policy_delete(
718                                                                 pol->security);
719                                 if (err) {
720                                         xfrm_audit_policy_delete(pol, 0,
721                                                         audit_info->loginuid,
722                                                         audit_info->sessionid,
723                                                         audit_info->secid);
724                                         return err;
725                                 }
726                         }
727                 }
728         }
729         return err;
730 }
731 #else
732 static inline int
733 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
734 {
735         return 0;
736 }
737 #endif
738
739 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
740 {
741         int dir, err = 0, cnt = 0;
742
743         write_lock_bh(&xfrm_policy_lock);
744
745         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
746         if (err)
747                 goto out;
748
749         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
750                 struct xfrm_policy *pol;
751                 struct hlist_node *entry;
752                 int i;
753
754         again1:
755                 hlist_for_each_entry(pol, entry,
756                                      &net->xfrm.policy_inexact[dir], bydst) {
757                         if (pol->type != type)
758                                 continue;
759                         __xfrm_policy_unlink(pol, dir);
760                         write_unlock_bh(&xfrm_policy_lock);
761                         cnt++;
762
763                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
764                                                  audit_info->sessionid,
765                                                  audit_info->secid);
766
767                         xfrm_policy_kill(pol);
768
769                         write_lock_bh(&xfrm_policy_lock);
770                         goto again1;
771                 }
772
773                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
774         again2:
775                         hlist_for_each_entry(pol, entry,
776                                              net->xfrm.policy_bydst[dir].table + i,
777                                              bydst) {
778                                 if (pol->type != type)
779                                         continue;
780                                 __xfrm_policy_unlink(pol, dir);
781                                 write_unlock_bh(&xfrm_policy_lock);
782                                 cnt++;
783
784                                 xfrm_audit_policy_delete(pol, 1,
785                                                          audit_info->loginuid,
786                                                          audit_info->sessionid,
787                                                          audit_info->secid);
788                                 xfrm_policy_kill(pol);
789
790                                 write_lock_bh(&xfrm_policy_lock);
791                                 goto again2;
792                         }
793                 }
794
795         }
796         if (!cnt)
797                 err = -ESRCH;
798 out:
799         write_unlock_bh(&xfrm_policy_lock);
800         return err;
801 }
802 EXPORT_SYMBOL(xfrm_policy_flush);
803
804 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
805                      int (*func)(struct xfrm_policy *, int, int, void*),
806                      void *data)
807 {
808         struct xfrm_policy *pol;
809         struct xfrm_policy_walk_entry *x;
810         int error = 0;
811
812         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
813             walk->type != XFRM_POLICY_TYPE_ANY)
814                 return -EINVAL;
815
816         if (list_empty(&walk->walk.all) && walk->seq != 0)
817                 return 0;
818
819         write_lock_bh(&xfrm_policy_lock);
820         if (list_empty(&walk->walk.all))
821                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
822         else
823                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
824         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
825                 if (x->dead)
826                         continue;
827                 pol = container_of(x, struct xfrm_policy, walk);
828                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
829                     walk->type != pol->type)
830                         continue;
831                 error = func(pol, xfrm_policy_id2dir(pol->index),
832                              walk->seq, data);
833                 if (error) {
834                         list_move_tail(&walk->walk.all, &x->all);
835                         goto out;
836                 }
837                 walk->seq++;
838         }
839         if (walk->seq == 0) {
840                 error = -ENOENT;
841                 goto out;
842         }
843         list_del_init(&walk->walk.all);
844 out:
845         write_unlock_bh(&xfrm_policy_lock);
846         return error;
847 }
848 EXPORT_SYMBOL(xfrm_policy_walk);
849
850 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
851 {
852         INIT_LIST_HEAD(&walk->walk.all);
853         walk->walk.dead = 1;
854         walk->type = type;
855         walk->seq = 0;
856 }
857 EXPORT_SYMBOL(xfrm_policy_walk_init);
858
859 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
860 {
861         if (list_empty(&walk->walk.all))
862                 return;
863
864         write_lock_bh(&xfrm_policy_lock);
865         list_del(&walk->walk.all);
866         write_unlock_bh(&xfrm_policy_lock);
867 }
868 EXPORT_SYMBOL(xfrm_policy_walk_done);
869
870 /*
871  * Find policy to apply to this flow.
872  *
873  * Returns 0 if policy found, else an -errno.
874  */
875 static int xfrm_policy_match(const struct xfrm_policy *pol,
876                              const struct flowi *fl,
877                              u8 type, u16 family, int dir)
878 {
879         const struct xfrm_selector *sel = &pol->selector;
880         int match, ret = -ESRCH;
881
882         if (pol->family != family ||
883             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
884             pol->type != type)
885                 return ret;
886
887         match = xfrm_selector_match(sel, fl, family);
888         if (match)
889                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
890                                                   dir);
891
892         return ret;
893 }
894
895 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
896                                                      const struct flowi *fl,
897                                                      u16 family, u8 dir)
898 {
899         int err;
900         struct xfrm_policy *pol, *ret;
901         const xfrm_address_t *daddr, *saddr;
902         struct hlist_node *entry;
903         struct hlist_head *chain;
904         u32 priority = ~0U;
905
906         daddr = xfrm_flowi_daddr(fl, family);
907         saddr = xfrm_flowi_saddr(fl, family);
908         if (unlikely(!daddr || !saddr))
909                 return NULL;
910
911         read_lock_bh(&xfrm_policy_lock);
912         chain = policy_hash_direct(net, daddr, saddr, family, dir);
913         ret = NULL;
914         hlist_for_each_entry(pol, entry, chain, bydst) {
915                 err = xfrm_policy_match(pol, fl, type, family, dir);
916                 if (err) {
917                         if (err == -ESRCH)
918                                 continue;
919                         else {
920                                 ret = ERR_PTR(err);
921                                 goto fail;
922                         }
923                 } else {
924                         ret = pol;
925                         priority = ret->priority;
926                         break;
927                 }
928         }
929         chain = &net->xfrm.policy_inexact[dir];
930         hlist_for_each_entry(pol, entry, chain, bydst) {
931                 err = xfrm_policy_match(pol, fl, type, family, dir);
932                 if (err) {
933                         if (err == -ESRCH)
934                                 continue;
935                         else {
936                                 ret = ERR_PTR(err);
937                                 goto fail;
938                         }
939                 } else if (pol->priority < priority) {
940                         ret = pol;
941                         break;
942                 }
943         }
944         if (ret)
945                 xfrm_pol_hold(ret);
946 fail:
947         read_unlock_bh(&xfrm_policy_lock);
948
949         return ret;
950 }
951
952 static struct xfrm_policy *
953 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
954 {
955 #ifdef CONFIG_XFRM_SUB_POLICY
956         struct xfrm_policy *pol;
957
958         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
959         if (pol != NULL)
960                 return pol;
961 #endif
962         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
963 }
964
965 static struct flow_cache_object *
966 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
967                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
968 {
969         struct xfrm_policy *pol;
970
971         if (old_obj)
972                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
973
974         pol = __xfrm_policy_lookup(net, fl, family, dir);
975         if (IS_ERR_OR_NULL(pol))
976                 return ERR_CAST(pol);
977
978         /* Resolver returns two references:
979          * one for cache and one for caller of flow_cache_lookup() */
980         xfrm_pol_hold(pol);
981
982         return &pol->flo;
983 }
984
985 static inline int policy_to_flow_dir(int dir)
986 {
987         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
988             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
989             XFRM_POLICY_FWD == FLOW_DIR_FWD)
990                 return dir;
991         switch (dir) {
992         default:
993         case XFRM_POLICY_IN:
994                 return FLOW_DIR_IN;
995         case XFRM_POLICY_OUT:
996                 return FLOW_DIR_OUT;
997         case XFRM_POLICY_FWD:
998                 return FLOW_DIR_FWD;
999         }
1000 }
1001
1002 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1003                                                  const struct flowi *fl)
1004 {
1005         struct xfrm_policy *pol;
1006
1007         read_lock_bh(&xfrm_policy_lock);
1008         if ((pol = sk->sk_policy[dir]) != NULL) {
1009                 int match = xfrm_selector_match(&pol->selector, fl,
1010                                                 sk->sk_family);
1011                 int err = 0;
1012
1013                 if (match) {
1014                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1015                                 pol = NULL;
1016                                 goto out;
1017                         }
1018                         err = security_xfrm_policy_lookup(pol->security,
1019                                                       fl->flowi_secid,
1020                                                       policy_to_flow_dir(dir));
1021                         if (!err)
1022                                 xfrm_pol_hold(pol);
1023                         else if (err == -ESRCH)
1024                                 pol = NULL;
1025                         else
1026                                 pol = ERR_PTR(err);
1027                 } else
1028                         pol = NULL;
1029         }
1030 out:
1031         read_unlock_bh(&xfrm_policy_lock);
1032         return pol;
1033 }
1034
1035 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1036 {
1037         struct net *net = xp_net(pol);
1038         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1039                                                      pol->family, dir);
1040
1041         list_add(&pol->walk.all, &net->xfrm.policy_all);
1042         hlist_add_head(&pol->bydst, chain);
1043         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1044         net->xfrm.policy_count[dir]++;
1045         xfrm_pol_hold(pol);
1046
1047         if (xfrm_bydst_should_resize(net, dir, NULL))
1048                 schedule_work(&net->xfrm.policy_hash_work);
1049 }
1050
1051 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1052                                                 int dir)
1053 {
1054         struct net *net = xp_net(pol);
1055
1056         if (hlist_unhashed(&pol->bydst))
1057                 return NULL;
1058
1059         hlist_del(&pol->bydst);
1060         hlist_del(&pol->byidx);
1061         list_del(&pol->walk.all);
1062         net->xfrm.policy_count[dir]--;
1063
1064         return pol;
1065 }
1066
1067 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1068 {
1069         write_lock_bh(&xfrm_policy_lock);
1070         pol = __xfrm_policy_unlink(pol, dir);
1071         write_unlock_bh(&xfrm_policy_lock);
1072         if (pol) {
1073                 xfrm_policy_kill(pol);
1074                 return 0;
1075         }
1076         return -ENOENT;
1077 }
1078 EXPORT_SYMBOL(xfrm_policy_delete);
1079
1080 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1081 {
1082         struct net *net = xp_net(pol);
1083         struct xfrm_policy *old_pol;
1084
1085 #ifdef CONFIG_XFRM_SUB_POLICY
1086         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1087                 return -EINVAL;
1088 #endif
1089
1090         write_lock_bh(&xfrm_policy_lock);
1091         old_pol = sk->sk_policy[dir];
1092         sk->sk_policy[dir] = pol;
1093         if (pol) {
1094                 pol->curlft.add_time = get_seconds();
1095                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1096                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1097         }
1098         if (old_pol)
1099                 /* Unlinking succeeds always. This is the only function
1100                  * allowed to delete or replace socket policy.
1101                  */
1102                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1103         write_unlock_bh(&xfrm_policy_lock);
1104
1105         if (old_pol) {
1106                 xfrm_policy_kill(old_pol);
1107         }
1108         return 0;
1109 }
1110
1111 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1112 {
1113         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1114
1115         if (newp) {
1116                 newp->selector = old->selector;
1117                 if (security_xfrm_policy_clone(old->security,
1118                                                &newp->security)) {
1119                         kfree(newp);
1120                         return NULL;  /* ENOMEM */
1121                 }
1122                 newp->lft = old->lft;
1123                 newp->curlft = old->curlft;
1124                 newp->mark = old->mark;
1125                 newp->action = old->action;
1126                 newp->flags = old->flags;
1127                 newp->xfrm_nr = old->xfrm_nr;
1128                 newp->index = old->index;
1129                 newp->type = old->type;
1130                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1131                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1132                 write_lock_bh(&xfrm_policy_lock);
1133                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1134                 write_unlock_bh(&xfrm_policy_lock);
1135                 xfrm_pol_put(newp);
1136         }
1137         return newp;
1138 }
1139
1140 int __xfrm_sk_clone_policy(struct sock *sk)
1141 {
1142         struct xfrm_policy *p0 = sk->sk_policy[0],
1143                            *p1 = sk->sk_policy[1];
1144
1145         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1146         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1147                 return -ENOMEM;
1148         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1149                 return -ENOMEM;
1150         return 0;
1151 }
1152
1153 static int
1154 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1155                unsigned short family)
1156 {
1157         int err;
1158         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1159
1160         if (unlikely(afinfo == NULL))
1161                 return -EINVAL;
1162         err = afinfo->get_saddr(net, local, remote);
1163         xfrm_policy_put_afinfo(afinfo);
1164         return err;
1165 }
1166
1167 /* Resolve list of templates for the flow, given policy. */
1168
1169 static int
1170 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1171                       struct xfrm_state **xfrm, unsigned short family)
1172 {
1173         struct net *net = xp_net(policy);
1174         int nx;
1175         int i, error;
1176         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1177         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1178         xfrm_address_t tmp;
1179
1180         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1181                 struct xfrm_state *x;
1182                 xfrm_address_t *remote = daddr;
1183                 xfrm_address_t *local  = saddr;
1184                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1185
1186                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1187                     tmpl->mode == XFRM_MODE_BEET) {
1188                         remote = &tmpl->id.daddr;
1189                         local = &tmpl->saddr;
1190                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1191                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1192                                 if (error)
1193                                         goto fail;
1194                                 local = &tmp;
1195                         }
1196                 }
1197
1198                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1199
1200                 if (x && x->km.state == XFRM_STATE_VALID) {
1201                         xfrm[nx++] = x;
1202                         daddr = remote;
1203                         saddr = local;
1204                         continue;
1205                 }
1206                 if (x) {
1207                         error = (x->km.state == XFRM_STATE_ERROR ?
1208                                  -EINVAL : -EAGAIN);
1209                         xfrm_state_put(x);
1210                 }
1211                 else if (error == -ESRCH)
1212                         error = -EAGAIN;
1213
1214                 if (!tmpl->optional)
1215                         goto fail;
1216         }
1217         return nx;
1218
1219 fail:
1220         for (nx--; nx>=0; nx--)
1221                 xfrm_state_put(xfrm[nx]);
1222         return error;
1223 }
1224
1225 static int
1226 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1227                   struct xfrm_state **xfrm, unsigned short family)
1228 {
1229         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1230         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1231         int cnx = 0;
1232         int error;
1233         int ret;
1234         int i;
1235
1236         for (i = 0; i < npols; i++) {
1237                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1238                         error = -ENOBUFS;
1239                         goto fail;
1240                 }
1241
1242                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1243                 if (ret < 0) {
1244                         error = ret;
1245                         goto fail;
1246                 } else
1247                         cnx += ret;
1248         }
1249
1250         /* found states are sorted for outbound processing */
1251         if (npols > 1)
1252                 xfrm_state_sort(xfrm, tpp, cnx, family);
1253
1254         return cnx;
1255
1256  fail:
1257         for (cnx--; cnx>=0; cnx--)
1258                 xfrm_state_put(tpp[cnx]);
1259         return error;
1260
1261 }
1262
1263 /* Check that the bundle accepts the flow and its components are
1264  * still valid.
1265  */
1266
1267 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1268 {
1269         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1270         int tos;
1271
1272         if (!afinfo)
1273                 return -EINVAL;
1274
1275         tos = afinfo->get_tos(fl);
1276
1277         xfrm_policy_put_afinfo(afinfo);
1278
1279         return tos;
1280 }
1281
1282 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1283 {
1284         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1285         struct dst_entry *dst = &xdst->u.dst;
1286
1287         if (xdst->route == NULL) {
1288                 /* Dummy bundle - if it has xfrms we were not
1289                  * able to build bundle as template resolution failed.
1290                  * It means we need to try again resolving. */
1291                 if (xdst->num_xfrms > 0)
1292                         return NULL;
1293         } else {
1294                 /* Real bundle */
1295                 if (stale_bundle(dst))
1296                         return NULL;
1297         }
1298
1299         dst_hold(dst);
1300         return flo;
1301 }
1302
1303 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1304 {
1305         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1306         struct dst_entry *dst = &xdst->u.dst;
1307
1308         if (!xdst->route)
1309                 return 0;
1310         if (stale_bundle(dst))
1311                 return 0;
1312
1313         return 1;
1314 }
1315
1316 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1317 {
1318         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1319         struct dst_entry *dst = &xdst->u.dst;
1320
1321         dst_free(dst);
1322 }
1323
1324 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1325         .get = xfrm_bundle_flo_get,
1326         .check = xfrm_bundle_flo_check,
1327         .delete = xfrm_bundle_flo_delete,
1328 };
1329
1330 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1331 {
1332         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1333         struct dst_ops *dst_ops;
1334         struct xfrm_dst *xdst;
1335
1336         if (!afinfo)
1337                 return ERR_PTR(-EINVAL);
1338
1339         switch (family) {
1340         case AF_INET:
1341                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1342                 break;
1343 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1344         case AF_INET6:
1345                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1346                 break;
1347 #endif
1348         default:
1349                 BUG();
1350         }
1351         xdst = dst_alloc(dst_ops, 0);
1352         xfrm_policy_put_afinfo(afinfo);
1353
1354         if (likely(xdst))
1355                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1356         else
1357                 xdst = ERR_PTR(-ENOBUFS);
1358
1359         return xdst;
1360 }
1361
1362 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1363                                  int nfheader_len)
1364 {
1365         struct xfrm_policy_afinfo *afinfo =
1366                 xfrm_policy_get_afinfo(dst->ops->family);
1367         int err;
1368
1369         if (!afinfo)
1370                 return -EINVAL;
1371
1372         err = afinfo->init_path(path, dst, nfheader_len);
1373
1374         xfrm_policy_put_afinfo(afinfo);
1375
1376         return err;
1377 }
1378
1379 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1380                                 const struct flowi *fl)
1381 {
1382         struct xfrm_policy_afinfo *afinfo =
1383                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1384         int err;
1385
1386         if (!afinfo)
1387                 return -EINVAL;
1388
1389         err = afinfo->fill_dst(xdst, dev, fl);
1390
1391         xfrm_policy_put_afinfo(afinfo);
1392
1393         return err;
1394 }
1395
1396
1397 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1398  * all the metrics... Shortly, bundle a bundle.
1399  */
1400
1401 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1402                                             struct xfrm_state **xfrm, int nx,
1403                                             const struct flowi *fl,
1404                                             struct dst_entry *dst)
1405 {
1406         struct net *net = xp_net(policy);
1407         unsigned long now = jiffies;
1408         struct net_device *dev;
1409         struct dst_entry *dst_prev = NULL;
1410         struct dst_entry *dst0 = NULL;
1411         int i = 0;
1412         int err;
1413         int header_len = 0;
1414         int nfheader_len = 0;
1415         int trailer_len = 0;
1416         int tos;
1417         int family = policy->selector.family;
1418         xfrm_address_t saddr, daddr;
1419
1420         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1421
1422         tos = xfrm_get_tos(fl, family);
1423         err = tos;
1424         if (tos < 0)
1425                 goto put_states;
1426
1427         dst_hold(dst);
1428
1429         for (; i < nx; i++) {
1430                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1431                 struct dst_entry *dst1 = &xdst->u.dst;
1432
1433                 err = PTR_ERR(xdst);
1434                 if (IS_ERR(xdst)) {
1435                         dst_release(dst);
1436                         goto put_states;
1437                 }
1438
1439                 if (!dst_prev)
1440                         dst0 = dst1;
1441                 else {
1442                         dst_prev->child = dst_clone(dst1);
1443                         dst1->flags |= DST_NOHASH;
1444                 }
1445
1446                 xdst->route = dst;
1447                 dst_copy_metrics(dst1, dst);
1448
1449                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1450                         family = xfrm[i]->props.family;
1451                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1452                                               family);
1453                         err = PTR_ERR(dst);
1454                         if (IS_ERR(dst))
1455                                 goto put_states;
1456                 } else
1457                         dst_hold(dst);
1458
1459                 dst1->xfrm = xfrm[i];
1460                 xdst->xfrm_genid = xfrm[i]->genid;
1461
1462                 dst1->obsolete = -1;
1463                 dst1->flags |= DST_HOST;
1464                 dst1->lastuse = now;
1465
1466                 dst1->input = dst_discard;
1467                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1468
1469                 dst1->next = dst_prev;
1470                 dst_prev = dst1;
1471
1472                 header_len += xfrm[i]->props.header_len;
1473                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1474                         nfheader_len += xfrm[i]->props.header_len;
1475                 trailer_len += xfrm[i]->props.trailer_len;
1476         }
1477
1478         dst_prev->child = dst;
1479         dst0->path = dst;
1480
1481         err = -ENODEV;
1482         dev = dst->dev;
1483         if (!dev)
1484                 goto free_dst;
1485
1486         /* Copy neighbour for reachability confirmation */
1487         dst0->neighbour = neigh_clone(dst->neighbour);
1488
1489         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1490         xfrm_init_pmtu(dst_prev);
1491
1492         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1493                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1494
1495                 err = xfrm_fill_dst(xdst, dev, fl);
1496                 if (err)
1497                         goto free_dst;
1498
1499                 dst_prev->header_len = header_len;
1500                 dst_prev->trailer_len = trailer_len;
1501                 header_len -= xdst->u.dst.xfrm->props.header_len;
1502                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1503         }
1504
1505 out:
1506         return dst0;
1507
1508 put_states:
1509         for (; i < nx; i++)
1510                 xfrm_state_put(xfrm[i]);
1511 free_dst:
1512         if (dst0)
1513                 dst_free(dst0);
1514         dst0 = ERR_PTR(err);
1515         goto out;
1516 }
1517
1518 static int inline
1519 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1520 {
1521         if (!*target) {
1522                 *target = kmalloc(size, GFP_ATOMIC);
1523                 if (!*target)
1524                         return -ENOMEM;
1525         }
1526         memcpy(*target, src, size);
1527         return 0;
1528 }
1529
1530 static int inline
1531 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1532 {
1533 #ifdef CONFIG_XFRM_SUB_POLICY
1534         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1535         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1536                                    sel, sizeof(*sel));
1537 #else
1538         return 0;
1539 #endif
1540 }
1541
1542 static int inline
1543 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1544 {
1545 #ifdef CONFIG_XFRM_SUB_POLICY
1546         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1547         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1548 #else
1549         return 0;
1550 #endif
1551 }
1552
1553 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1554                                 struct xfrm_policy **pols,
1555                                 int *num_pols, int *num_xfrms)
1556 {
1557         int i;
1558
1559         if (*num_pols == 0 || !pols[0]) {
1560                 *num_pols = 0;
1561                 *num_xfrms = 0;
1562                 return 0;
1563         }
1564         if (IS_ERR(pols[0]))
1565                 return PTR_ERR(pols[0]);
1566
1567         *num_xfrms = pols[0]->xfrm_nr;
1568
1569 #ifdef CONFIG_XFRM_SUB_POLICY
1570         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1571             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1572                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1573                                                     XFRM_POLICY_TYPE_MAIN,
1574                                                     fl, family,
1575                                                     XFRM_POLICY_OUT);
1576                 if (pols[1]) {
1577                         if (IS_ERR(pols[1])) {
1578                                 xfrm_pols_put(pols, *num_pols);
1579                                 return PTR_ERR(pols[1]);
1580                         }
1581                         (*num_pols) ++;
1582                         (*num_xfrms) += pols[1]->xfrm_nr;
1583                 }
1584         }
1585 #endif
1586         for (i = 0; i < *num_pols; i++) {
1587                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1588                         *num_xfrms = -1;
1589                         break;
1590                 }
1591         }
1592
1593         return 0;
1594
1595 }
1596
1597 static struct xfrm_dst *
1598 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1599                                const struct flowi *fl, u16 family,
1600                                struct dst_entry *dst_orig)
1601 {
1602         struct net *net = xp_net(pols[0]);
1603         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1604         struct dst_entry *dst;
1605         struct xfrm_dst *xdst;
1606         int err;
1607
1608         /* Try to instantiate a bundle */
1609         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1610         if (err <= 0) {
1611                 if (err != 0 && err != -EAGAIN)
1612                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1613                 return ERR_PTR(err);
1614         }
1615
1616         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1617         if (IS_ERR(dst)) {
1618                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1619                 return ERR_CAST(dst);
1620         }
1621
1622         xdst = (struct xfrm_dst *)dst;
1623         xdst->num_xfrms = err;
1624         if (num_pols > 1)
1625                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1626         else
1627                 err = xfrm_dst_update_origin(dst, fl);
1628         if (unlikely(err)) {
1629                 dst_free(dst);
1630                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1631                 return ERR_PTR(err);
1632         }
1633
1634         xdst->num_pols = num_pols;
1635         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1636         xdst->policy_genid = atomic_read(&pols[0]->genid);
1637
1638         return xdst;
1639 }
1640
1641 static struct flow_cache_object *
1642 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1643                    struct flow_cache_object *oldflo, void *ctx)
1644 {
1645         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1646         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1647         struct xfrm_dst *xdst, *new_xdst;
1648         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1649
1650         /* Check if the policies from old bundle are usable */
1651         xdst = NULL;
1652         if (oldflo) {
1653                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1654                 num_pols = xdst->num_pols;
1655                 num_xfrms = xdst->num_xfrms;
1656                 pol_dead = 0;
1657                 for (i = 0; i < num_pols; i++) {
1658                         pols[i] = xdst->pols[i];
1659                         pol_dead |= pols[i]->walk.dead;
1660                 }
1661                 if (pol_dead) {
1662                         dst_free(&xdst->u.dst);
1663                         xdst = NULL;
1664                         num_pols = 0;
1665                         num_xfrms = 0;
1666                         oldflo = NULL;
1667                 }
1668         }
1669
1670         /* Resolve policies to use if we couldn't get them from
1671          * previous cache entry */
1672         if (xdst == NULL) {
1673                 num_pols = 1;
1674                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1675                 err = xfrm_expand_policies(fl, family, pols,
1676                                            &num_pols, &num_xfrms);
1677                 if (err < 0)
1678                         goto inc_error;
1679                 if (num_pols == 0)
1680                         return NULL;
1681                 if (num_xfrms <= 0)
1682                         goto make_dummy_bundle;
1683         }
1684
1685         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1686         if (IS_ERR(new_xdst)) {
1687                 err = PTR_ERR(new_xdst);
1688                 if (err != -EAGAIN)
1689                         goto error;
1690                 if (oldflo == NULL)
1691                         goto make_dummy_bundle;
1692                 dst_hold(&xdst->u.dst);
1693                 return oldflo;
1694         } else if (new_xdst == NULL) {
1695                 num_xfrms = 0;
1696                 if (oldflo == NULL)
1697                         goto make_dummy_bundle;
1698                 xdst->num_xfrms = 0;
1699                 dst_hold(&xdst->u.dst);
1700                 return oldflo;
1701         }
1702
1703         /* Kill the previous bundle */
1704         if (xdst) {
1705                 /* The policies were stolen for newly generated bundle */
1706                 xdst->num_pols = 0;
1707                 dst_free(&xdst->u.dst);
1708         }
1709
1710         /* Flow cache does not have reference, it dst_free()'s,
1711          * but we do need to return one reference for original caller */
1712         dst_hold(&new_xdst->u.dst);
1713         return &new_xdst->flo;
1714
1715 make_dummy_bundle:
1716         /* We found policies, but there's no bundles to instantiate:
1717          * either because the policy blocks, has no transformations or
1718          * we could not build template (no xfrm_states).*/
1719         xdst = xfrm_alloc_dst(net, family);
1720         if (IS_ERR(xdst)) {
1721                 xfrm_pols_put(pols, num_pols);
1722                 return ERR_CAST(xdst);
1723         }
1724         xdst->num_pols = num_pols;
1725         xdst->num_xfrms = num_xfrms;
1726         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1727
1728         dst_hold(&xdst->u.dst);
1729         return &xdst->flo;
1730
1731 inc_error:
1732         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1733 error:
1734         if (xdst != NULL)
1735                 dst_free(&xdst->u.dst);
1736         else
1737                 xfrm_pols_put(pols, num_pols);
1738         return ERR_PTR(err);
1739 }
1740
1741 static struct dst_entry *make_blackhole(struct net *net, u16 family,
1742                                         struct dst_entry *dst_orig)
1743 {
1744         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1745         struct dst_entry *ret;
1746
1747         if (!afinfo) {
1748                 dst_release(dst_orig);
1749                 ret = ERR_PTR(-EINVAL);
1750         } else {
1751                 ret = afinfo->blackhole_route(net, dst_orig);
1752         }
1753         xfrm_policy_put_afinfo(afinfo);
1754
1755         return ret;
1756 }
1757
1758 /* Main function: finds/creates a bundle for given flow.
1759  *
1760  * At the moment we eat a raw IP route. Mostly to speed up lookups
1761  * on interfaces with disabled IPsec.
1762  */
1763 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
1764                               const struct flowi *fl,
1765                               struct sock *sk, int flags)
1766 {
1767         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1768         struct flow_cache_object *flo;
1769         struct xfrm_dst *xdst;
1770         struct dst_entry *dst, *route;
1771         u16 family = dst_orig->ops->family;
1772         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1773         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1774
1775 restart:
1776         dst = NULL;
1777         xdst = NULL;
1778         route = NULL;
1779
1780         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1781                 num_pols = 1;
1782                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1783                 err = xfrm_expand_policies(fl, family, pols,
1784                                            &num_pols, &num_xfrms);
1785                 if (err < 0)
1786                         goto dropdst;
1787
1788                 if (num_pols) {
1789                         if (num_xfrms <= 0) {
1790                                 drop_pols = num_pols;
1791                                 goto no_transform;
1792                         }
1793
1794                         xdst = xfrm_resolve_and_create_bundle(
1795                                         pols, num_pols, fl,
1796                                         family, dst_orig);
1797                         if (IS_ERR(xdst)) {
1798                                 xfrm_pols_put(pols, num_pols);
1799                                 err = PTR_ERR(xdst);
1800                                 goto dropdst;
1801                         } else if (xdst == NULL) {
1802                                 num_xfrms = 0;
1803                                 drop_pols = num_pols;
1804                                 goto no_transform;
1805                         }
1806
1807                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1808                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1809                         xfrm_policy_sk_bundles = &xdst->u.dst;
1810                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1811
1812                         route = xdst->route;
1813                 }
1814         }
1815
1816         if (xdst == NULL) {
1817                 /* To accelerate a bit...  */
1818                 if ((dst_orig->flags & DST_NOXFRM) ||
1819                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1820                         goto nopol;
1821
1822                 flo = flow_cache_lookup(net, fl, family, dir,
1823                                         xfrm_bundle_lookup, dst_orig);
1824                 if (flo == NULL)
1825                         goto nopol;
1826                 if (IS_ERR(flo)) {
1827                         err = PTR_ERR(flo);
1828                         goto dropdst;
1829                 }
1830                 xdst = container_of(flo, struct xfrm_dst, flo);
1831
1832                 num_pols = xdst->num_pols;
1833                 num_xfrms = xdst->num_xfrms;
1834                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1835                 route = xdst->route;
1836         }
1837
1838         dst = &xdst->u.dst;
1839         if (route == NULL && num_xfrms > 0) {
1840                 /* The only case when xfrm_bundle_lookup() returns a
1841                  * bundle with null route, is when the template could
1842                  * not be resolved. It means policies are there, but
1843                  * bundle could not be created, since we don't yet
1844                  * have the xfrm_state's. We need to wait for KM to
1845                  * negotiate new SA's or bail out with error.*/
1846                 if (net->xfrm.sysctl_larval_drop) {
1847                         /* EREMOTE tells the caller to generate
1848                          * a one-shot blackhole route. */
1849                         dst_release(dst);
1850                         xfrm_pols_put(pols, drop_pols);
1851                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1852
1853                         return make_blackhole(net, family, dst_orig);
1854                 }
1855                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
1856                         DECLARE_WAITQUEUE(wait, current);
1857
1858                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1859                         set_current_state(TASK_INTERRUPTIBLE);
1860                         schedule();
1861                         set_current_state(TASK_RUNNING);
1862                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1863
1864                         if (!signal_pending(current)) {
1865                                 dst_release(dst);
1866                                 goto restart;
1867                         }
1868
1869                         err = -ERESTART;
1870                 } else
1871                         err = -EAGAIN;
1872
1873                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1874                 goto error;
1875         }
1876
1877 no_transform:
1878         if (num_pols == 0)
1879                 goto nopol;
1880
1881         if ((flags & XFRM_LOOKUP_ICMP) &&
1882             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1883                 err = -ENOENT;
1884                 goto error;
1885         }
1886
1887         for (i = 0; i < num_pols; i++)
1888                 pols[i]->curlft.use_time = get_seconds();
1889
1890         if (num_xfrms < 0) {
1891                 /* Prohibit the flow */
1892                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1893                 err = -EPERM;
1894                 goto error;
1895         } else if (num_xfrms > 0) {
1896                 /* Flow transformed */
1897                 dst_release(dst_orig);
1898         } else {
1899                 /* Flow passes untransformed */
1900                 dst_release(dst);
1901                 dst = dst_orig;
1902         }
1903 ok:
1904         xfrm_pols_put(pols, drop_pols);
1905         return dst;
1906
1907 nopol:
1908         if (!(flags & XFRM_LOOKUP_ICMP)) {
1909                 dst = dst_orig;
1910                 goto ok;
1911         }
1912         err = -ENOENT;
1913 error:
1914         dst_release(dst);
1915 dropdst:
1916         dst_release(dst_orig);
1917         xfrm_pols_put(pols, drop_pols);
1918         return ERR_PTR(err);
1919 }
1920 EXPORT_SYMBOL(xfrm_lookup);
1921
1922 static inline int
1923 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
1924 {
1925         struct xfrm_state *x;
1926
1927         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1928                 return 0;
1929         x = skb->sp->xvec[idx];
1930         if (!x->type->reject)
1931                 return 0;
1932         return x->type->reject(x, skb, fl);
1933 }
1934
1935 /* When skb is transformed back to its "native" form, we have to
1936  * check policy restrictions. At the moment we make this in maximally
1937  * stupid way. Shame on me. :-) Of course, connected sockets must
1938  * have policy cached at them.
1939  */
1940
1941 static inline int
1942 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
1943               unsigned short family)
1944 {
1945         if (xfrm_state_kern(x))
1946                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1947         return  x->id.proto == tmpl->id.proto &&
1948                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1949                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1950                 x->props.mode == tmpl->mode &&
1951                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1952                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1953                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1954                   xfrm_state_addr_cmp(tmpl, x, family));
1955 }
1956
1957 /*
1958  * 0 or more than 0 is returned when validation is succeeded (either bypass
1959  * because of optional transport mode, or next index of the mathced secpath
1960  * state with the template.
1961  * -1 is returned when no matching template is found.
1962  * Otherwise "-2 - errored_index" is returned.
1963  */
1964 static inline int
1965 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
1966                unsigned short family)
1967 {
1968         int idx = start;
1969
1970         if (tmpl->optional) {
1971                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1972                         return start;
1973         } else
1974                 start = -1;
1975         for (; idx < sp->len; idx++) {
1976                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1977                         return ++idx;
1978                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1979                         if (start == -1)
1980                                 start = -2-idx;
1981                         break;
1982                 }
1983         }
1984         return start;
1985 }
1986
1987 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1988                           unsigned int family, int reverse)
1989 {
1990         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1991         int err;
1992
1993         if (unlikely(afinfo == NULL))
1994                 return -EAFNOSUPPORT;
1995
1996         afinfo->decode_session(skb, fl, reverse);
1997         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
1998         xfrm_policy_put_afinfo(afinfo);
1999         return err;
2000 }
2001 EXPORT_SYMBOL(__xfrm_decode_session);
2002
2003 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2004 {
2005         for (; k < sp->len; k++) {
2006                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2007                         *idxp = k;
2008                         return 1;
2009                 }
2010         }
2011
2012         return 0;
2013 }
2014
2015 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2016                         unsigned short family)
2017 {
2018         struct net *net = dev_net(skb->dev);
2019         struct xfrm_policy *pol;
2020         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2021         int npols = 0;
2022         int xfrm_nr;
2023         int pi;
2024         int reverse;
2025         struct flowi fl;
2026         u8 fl_dir;
2027         int xerr_idx = -1;
2028
2029         reverse = dir & ~XFRM_POLICY_MASK;
2030         dir &= XFRM_POLICY_MASK;
2031         fl_dir = policy_to_flow_dir(dir);
2032
2033         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2034                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2035                 return 0;
2036         }
2037
2038         nf_nat_decode_session(skb, &fl, family);
2039
2040         /* First, check used SA against their selectors. */
2041         if (skb->sp) {
2042                 int i;
2043
2044                 for (i=skb->sp->len-1; i>=0; i--) {
2045                         struct xfrm_state *x = skb->sp->xvec[i];
2046                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2047                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2048                                 return 0;
2049                         }
2050                 }
2051         }
2052
2053         pol = NULL;
2054         if (sk && sk->sk_policy[dir]) {
2055                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2056                 if (IS_ERR(pol)) {
2057                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2058                         return 0;
2059                 }
2060         }
2061
2062         if (!pol) {
2063                 struct flow_cache_object *flo;
2064
2065                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2066                                         xfrm_policy_lookup, NULL);
2067                 if (IS_ERR_OR_NULL(flo))
2068                         pol = ERR_CAST(flo);
2069                 else
2070                         pol = container_of(flo, struct xfrm_policy, flo);
2071         }
2072
2073         if (IS_ERR(pol)) {
2074                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2075                 return 0;
2076         }
2077
2078         if (!pol) {
2079                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2080                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2081                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2082                         return 0;
2083                 }
2084                 return 1;
2085         }
2086
2087         pol->curlft.use_time = get_seconds();
2088
2089         pols[0] = pol;
2090         npols ++;
2091 #ifdef CONFIG_XFRM_SUB_POLICY
2092         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2093                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2094                                                     &fl, family,
2095                                                     XFRM_POLICY_IN);
2096                 if (pols[1]) {
2097                         if (IS_ERR(pols[1])) {
2098                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2099                                 return 0;
2100                         }
2101                         pols[1]->curlft.use_time = get_seconds();
2102                         npols ++;
2103                 }
2104         }
2105 #endif
2106
2107         if (pol->action == XFRM_POLICY_ALLOW) {
2108                 struct sec_path *sp;
2109                 static struct sec_path dummy;
2110                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2111                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2112                 struct xfrm_tmpl **tpp = tp;
2113                 int ti = 0;
2114                 int i, k;
2115
2116                 if ((sp = skb->sp) == NULL)
2117                         sp = &dummy;
2118
2119                 for (pi = 0; pi < npols; pi++) {
2120                         if (pols[pi] != pol &&
2121                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2122                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2123                                 goto reject;
2124                         }
2125                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2126                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2127                                 goto reject_error;
2128                         }
2129                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2130                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2131                 }
2132                 xfrm_nr = ti;
2133                 if (npols > 1) {
2134                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2135                         tpp = stp;
2136                 }
2137
2138                 /* For each tunnel xfrm, find the first matching tmpl.
2139                  * For each tmpl before that, find corresponding xfrm.
2140                  * Order is _important_. Later we will implement
2141                  * some barriers, but at the moment barriers
2142                  * are implied between each two transformations.
2143                  */
2144                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2145                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2146                         if (k < 0) {
2147                                 if (k < -1)
2148                                         /* "-2 - errored_index" returned */
2149                                         xerr_idx = -(2+k);
2150                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2151                                 goto reject;
2152                         }
2153                 }
2154
2155                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2156                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2157                         goto reject;
2158                 }
2159
2160                 xfrm_pols_put(pols, npols);
2161                 return 1;
2162         }
2163         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2164
2165 reject:
2166         xfrm_secpath_reject(xerr_idx, skb, &fl);
2167 reject_error:
2168         xfrm_pols_put(pols, npols);
2169         return 0;
2170 }
2171 EXPORT_SYMBOL(__xfrm_policy_check);
2172
2173 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2174 {
2175         struct net *net = dev_net(skb->dev);
2176         struct flowi fl;
2177         struct dst_entry *dst;
2178         int res = 0;
2179
2180         if (xfrm_decode_session(skb, &fl, family) < 0) {
2181                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2182                 return 0;
2183         }
2184
2185         skb_dst_force(skb);
2186
2187         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2188         if (IS_ERR(dst)) {
2189                 res = 1;
2190                 dst = NULL;
2191         }
2192         skb_dst_set(skb, dst);
2193         return res;
2194 }
2195 EXPORT_SYMBOL(__xfrm_route_forward);
2196
2197 /* Optimize later using cookies and generation ids. */
2198
2199 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2200 {
2201         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2202          * to "-1" to force all XFRM destinations to get validated by
2203          * dst_ops->check on every use.  We do this because when a
2204          * normal route referenced by an XFRM dst is obsoleted we do
2205          * not go looking around for all parent referencing XFRM dsts
2206          * so that we can invalidate them.  It is just too much work.
2207          * Instead we make the checks here on every use.  For example:
2208          *
2209          *      XFRM dst A --> IPv4 dst X
2210          *
2211          * X is the "xdst->route" of A (X is also the "dst->path" of A
2212          * in this example).  If X is marked obsolete, "A" will not
2213          * notice.  That's what we are validating here via the
2214          * stale_bundle() check.
2215          *
2216          * When a policy's bundle is pruned, we dst_free() the XFRM
2217          * dst which causes it's ->obsolete field to be set to a
2218          * positive non-zero integer.  If an XFRM dst has been pruned
2219          * like this, we want to force a new route lookup.
2220          */
2221         if (dst->obsolete < 0 && !stale_bundle(dst))
2222                 return dst;
2223
2224         return NULL;
2225 }
2226
2227 static int stale_bundle(struct dst_entry *dst)
2228 {
2229         return !xfrm_bundle_ok((struct xfrm_dst *)dst, AF_UNSPEC);
2230 }
2231
2232 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2233 {
2234         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2235                 dst->dev = dev_net(dev)->loopback_dev;
2236                 dev_hold(dst->dev);
2237                 dev_put(dev);
2238         }
2239 }
2240 EXPORT_SYMBOL(xfrm_dst_ifdown);
2241
2242 static void xfrm_link_failure(struct sk_buff *skb)
2243 {
2244         /* Impossible. Such dst must be popped before reaches point of failure. */
2245 }
2246
2247 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2248 {
2249         if (dst) {
2250                 if (dst->obsolete) {
2251                         dst_release(dst);
2252                         dst = NULL;
2253                 }
2254         }
2255         return dst;
2256 }
2257
2258 static void __xfrm_garbage_collect(struct net *net)
2259 {
2260         struct dst_entry *head, *next;
2261
2262         flow_cache_flush();
2263
2264         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2265         head = xfrm_policy_sk_bundles;
2266         xfrm_policy_sk_bundles = NULL;
2267         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2268
2269         while (head) {
2270                 next = head->next;
2271                 dst_free(head);
2272                 head = next;
2273         }
2274 }
2275
2276 static void xfrm_init_pmtu(struct dst_entry *dst)
2277 {
2278         do {
2279                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2280                 u32 pmtu, route_mtu_cached;
2281
2282                 pmtu = dst_mtu(dst->child);
2283                 xdst->child_mtu_cached = pmtu;
2284
2285                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2286
2287                 route_mtu_cached = dst_mtu(xdst->route);
2288                 xdst->route_mtu_cached = route_mtu_cached;
2289
2290                 if (pmtu > route_mtu_cached)
2291                         pmtu = route_mtu_cached;
2292
2293                 dst_metric_set(dst, RTAX_MTU, pmtu);
2294         } while ((dst = dst->next));
2295 }
2296
2297 /* Check that the bundle accepts the flow and its components are
2298  * still valid.
2299  */
2300
2301 static int xfrm_bundle_ok(struct xfrm_dst *first, int family)
2302 {
2303         struct dst_entry *dst = &first->u.dst;
2304         struct xfrm_dst *last;
2305         u32 mtu;
2306
2307         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2308             (dst->dev && !netif_running(dst->dev)))
2309                 return 0;
2310
2311         last = NULL;
2312
2313         do {
2314                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2315
2316                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2317                         return 0;
2318                 if (xdst->xfrm_genid != dst->xfrm->genid)
2319                         return 0;
2320                 if (xdst->num_pols > 0 &&
2321                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2322                         return 0;
2323
2324                 mtu = dst_mtu(dst->child);
2325                 if (xdst->child_mtu_cached != mtu) {
2326                         last = xdst;
2327                         xdst->child_mtu_cached = mtu;
2328                 }
2329
2330                 if (!dst_check(xdst->route, xdst->route_cookie))
2331                         return 0;
2332                 mtu = dst_mtu(xdst->route);
2333                 if (xdst->route_mtu_cached != mtu) {
2334                         last = xdst;
2335                         xdst->route_mtu_cached = mtu;
2336                 }
2337
2338                 dst = dst->child;
2339         } while (dst->xfrm);
2340
2341         if (likely(!last))
2342                 return 1;
2343
2344         mtu = last->child_mtu_cached;
2345         for (;;) {
2346                 dst = &last->u.dst;
2347
2348                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2349                 if (mtu > last->route_mtu_cached)
2350                         mtu = last->route_mtu_cached;
2351                 dst_metric_set(dst, RTAX_MTU, mtu);
2352
2353                 if (last == first)
2354                         break;
2355
2356                 last = (struct xfrm_dst *)last->u.dst.next;
2357                 last->child_mtu_cached = mtu;
2358         }
2359
2360         return 1;
2361 }
2362
2363 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2364 {
2365         return dst_metric_advmss(dst->path);
2366 }
2367
2368 static unsigned int xfrm_default_mtu(const struct dst_entry *dst)
2369 {
2370         return dst_mtu(dst->path);
2371 }
2372
2373 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2374 {
2375         struct net *net;
2376         int err = 0;
2377         if (unlikely(afinfo == NULL))
2378                 return -EINVAL;
2379         if (unlikely(afinfo->family >= NPROTO))
2380                 return -EAFNOSUPPORT;
2381         write_lock_bh(&xfrm_policy_afinfo_lock);
2382         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2383                 err = -ENOBUFS;
2384         else {
2385                 struct dst_ops *dst_ops = afinfo->dst_ops;
2386                 if (likely(dst_ops->kmem_cachep == NULL))
2387                         dst_ops->kmem_cachep = xfrm_dst_cache;
2388                 if (likely(dst_ops->check == NULL))
2389                         dst_ops->check = xfrm_dst_check;
2390                 if (likely(dst_ops->default_advmss == NULL))
2391                         dst_ops->default_advmss = xfrm_default_advmss;
2392                 if (likely(dst_ops->default_mtu == NULL))
2393                         dst_ops->default_mtu = xfrm_default_mtu;
2394                 if (likely(dst_ops->negative_advice == NULL))
2395                         dst_ops->negative_advice = xfrm_negative_advice;
2396                 if (likely(dst_ops->link_failure == NULL))
2397                         dst_ops->link_failure = xfrm_link_failure;
2398                 if (likely(afinfo->garbage_collect == NULL))
2399                         afinfo->garbage_collect = __xfrm_garbage_collect;
2400                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2401         }
2402         write_unlock_bh(&xfrm_policy_afinfo_lock);
2403
2404         rtnl_lock();
2405         for_each_net(net) {
2406                 struct dst_ops *xfrm_dst_ops;
2407
2408                 switch (afinfo->family) {
2409                 case AF_INET:
2410                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2411                         break;
2412 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2413                 case AF_INET6:
2414                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2415                         break;
2416 #endif
2417                 default:
2418                         BUG();
2419                 }
2420                 *xfrm_dst_ops = *afinfo->dst_ops;
2421         }
2422         rtnl_unlock();
2423
2424         return err;
2425 }
2426 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2427
2428 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2429 {
2430         int err = 0;
2431         if (unlikely(afinfo == NULL))
2432                 return -EINVAL;
2433         if (unlikely(afinfo->family >= NPROTO))
2434                 return -EAFNOSUPPORT;
2435         write_lock_bh(&xfrm_policy_afinfo_lock);
2436         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2437                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2438                         err = -EINVAL;
2439                 else {
2440                         struct dst_ops *dst_ops = afinfo->dst_ops;
2441                         xfrm_policy_afinfo[afinfo->family] = NULL;
2442                         dst_ops->kmem_cachep = NULL;
2443                         dst_ops->check = NULL;
2444                         dst_ops->negative_advice = NULL;
2445                         dst_ops->link_failure = NULL;
2446                         afinfo->garbage_collect = NULL;
2447                 }
2448         }
2449         write_unlock_bh(&xfrm_policy_afinfo_lock);
2450         return err;
2451 }
2452 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2453
2454 static void __net_init xfrm_dst_ops_init(struct net *net)
2455 {
2456         struct xfrm_policy_afinfo *afinfo;
2457
2458         read_lock_bh(&xfrm_policy_afinfo_lock);
2459         afinfo = xfrm_policy_afinfo[AF_INET];
2460         if (afinfo)
2461                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2462 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2463         afinfo = xfrm_policy_afinfo[AF_INET6];
2464         if (afinfo)
2465                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2466 #endif
2467         read_unlock_bh(&xfrm_policy_afinfo_lock);
2468 }
2469
2470 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2471 {
2472         struct xfrm_policy_afinfo *afinfo;
2473         if (unlikely(family >= NPROTO))
2474                 return NULL;
2475         read_lock(&xfrm_policy_afinfo_lock);
2476         afinfo = xfrm_policy_afinfo[family];
2477         if (unlikely(!afinfo))
2478                 read_unlock(&xfrm_policy_afinfo_lock);
2479         return afinfo;
2480 }
2481
2482 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2483 {
2484         read_unlock(&xfrm_policy_afinfo_lock);
2485 }
2486
2487 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2488 {
2489         struct net_device *dev = ptr;
2490
2491         switch (event) {
2492         case NETDEV_DOWN:
2493                 __xfrm_garbage_collect(dev_net(dev));
2494         }
2495         return NOTIFY_DONE;
2496 }
2497
2498 static struct notifier_block xfrm_dev_notifier = {
2499         .notifier_call  = xfrm_dev_event,
2500 };
2501
2502 #ifdef CONFIG_XFRM_STATISTICS
2503 static int __net_init xfrm_statistics_init(struct net *net)
2504 {
2505         int rv;
2506
2507         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2508                           sizeof(struct linux_xfrm_mib),
2509                           __alignof__(struct linux_xfrm_mib)) < 0)
2510                 return -ENOMEM;
2511         rv = xfrm_proc_init(net);
2512         if (rv < 0)
2513                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2514         return rv;
2515 }
2516
2517 static void xfrm_statistics_fini(struct net *net)
2518 {
2519         xfrm_proc_fini(net);
2520         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2521 }
2522 #else
2523 static int __net_init xfrm_statistics_init(struct net *net)
2524 {
2525         return 0;
2526 }
2527
2528 static void xfrm_statistics_fini(struct net *net)
2529 {
2530 }
2531 #endif
2532
2533 static int __net_init xfrm_policy_init(struct net *net)
2534 {
2535         unsigned int hmask, sz;
2536         int dir;
2537
2538         if (net_eq(net, &init_net))
2539                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2540                                            sizeof(struct xfrm_dst),
2541                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2542                                            NULL);
2543
2544         hmask = 8 - 1;
2545         sz = (hmask+1) * sizeof(struct hlist_head);
2546
2547         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2548         if (!net->xfrm.policy_byidx)
2549                 goto out_byidx;
2550         net->xfrm.policy_idx_hmask = hmask;
2551
2552         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2553                 struct xfrm_policy_hash *htab;
2554
2555                 net->xfrm.policy_count[dir] = 0;
2556                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2557
2558                 htab = &net->xfrm.policy_bydst[dir];
2559                 htab->table = xfrm_hash_alloc(sz);
2560                 if (!htab->table)
2561                         goto out_bydst;
2562                 htab->hmask = hmask;
2563         }
2564
2565         INIT_LIST_HEAD(&net->xfrm.policy_all);
2566         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2567         if (net_eq(net, &init_net))
2568                 register_netdevice_notifier(&xfrm_dev_notifier);
2569         return 0;
2570
2571 out_bydst:
2572         for (dir--; dir >= 0; dir--) {
2573                 struct xfrm_policy_hash *htab;
2574
2575                 htab = &net->xfrm.policy_bydst[dir];
2576                 xfrm_hash_free(htab->table, sz);
2577         }
2578         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2579 out_byidx:
2580         return -ENOMEM;
2581 }
2582
2583 static void xfrm_policy_fini(struct net *net)
2584 {
2585         struct xfrm_audit audit_info;
2586         unsigned int sz;
2587         int dir;
2588
2589         flush_work(&net->xfrm.policy_hash_work);
2590 #ifdef CONFIG_XFRM_SUB_POLICY
2591         audit_info.loginuid = -1;
2592         audit_info.sessionid = -1;
2593         audit_info.secid = 0;
2594         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2595 #endif
2596         audit_info.loginuid = -1;
2597         audit_info.sessionid = -1;
2598         audit_info.secid = 0;
2599         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2600
2601         WARN_ON(!list_empty(&net->xfrm.policy_all));
2602
2603         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2604                 struct xfrm_policy_hash *htab;
2605
2606                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2607
2608                 htab = &net->xfrm.policy_bydst[dir];
2609                 sz = (htab->hmask + 1);
2610                 WARN_ON(!hlist_empty(htab->table));
2611                 xfrm_hash_free(htab->table, sz);
2612         }
2613
2614         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2615         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2616         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2617 }
2618
2619 static int __net_init xfrm_net_init(struct net *net)
2620 {
2621         int rv;
2622
2623         rv = xfrm_statistics_init(net);
2624         if (rv < 0)
2625                 goto out_statistics;
2626         rv = xfrm_state_init(net);
2627         if (rv < 0)
2628                 goto out_state;
2629         rv = xfrm_policy_init(net);
2630         if (rv < 0)
2631                 goto out_policy;
2632         xfrm_dst_ops_init(net);
2633         rv = xfrm_sysctl_init(net);
2634         if (rv < 0)
2635                 goto out_sysctl;
2636         return 0;
2637
2638 out_sysctl:
2639         xfrm_policy_fini(net);
2640 out_policy:
2641         xfrm_state_fini(net);
2642 out_state:
2643         xfrm_statistics_fini(net);
2644 out_statistics:
2645         return rv;
2646 }
2647
2648 static void __net_exit xfrm_net_exit(struct net *net)
2649 {
2650         xfrm_sysctl_fini(net);
2651         xfrm_policy_fini(net);
2652         xfrm_state_fini(net);
2653         xfrm_statistics_fini(net);
2654 }
2655
2656 static struct pernet_operations __net_initdata xfrm_net_ops = {
2657         .init = xfrm_net_init,
2658         .exit = xfrm_net_exit,
2659 };
2660
2661 void __init xfrm_init(void)
2662 {
2663         register_pernet_subsys(&xfrm_net_ops);
2664         xfrm_input_init();
2665 }
2666
2667 #ifdef CONFIG_AUDITSYSCALL
2668 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2669                                          struct audit_buffer *audit_buf)
2670 {
2671         struct xfrm_sec_ctx *ctx = xp->security;
2672         struct xfrm_selector *sel = &xp->selector;
2673
2674         if (ctx)
2675                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2676                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2677
2678         switch(sel->family) {
2679         case AF_INET:
2680                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2681                 if (sel->prefixlen_s != 32)
2682                         audit_log_format(audit_buf, " src_prefixlen=%d",
2683                                          sel->prefixlen_s);
2684                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2685                 if (sel->prefixlen_d != 32)
2686                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2687                                          sel->prefixlen_d);
2688                 break;
2689         case AF_INET6:
2690                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2691                 if (sel->prefixlen_s != 128)
2692                         audit_log_format(audit_buf, " src_prefixlen=%d",
2693                                          sel->prefixlen_s);
2694                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2695                 if (sel->prefixlen_d != 128)
2696                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2697                                          sel->prefixlen_d);
2698                 break;
2699         }
2700 }
2701
2702 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2703                            uid_t auid, u32 sessionid, u32 secid)
2704 {
2705         struct audit_buffer *audit_buf;
2706
2707         audit_buf = xfrm_audit_start("SPD-add");
2708         if (audit_buf == NULL)
2709                 return;
2710         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2711         audit_log_format(audit_buf, " res=%u", result);
2712         xfrm_audit_common_policyinfo(xp, audit_buf);
2713         audit_log_end(audit_buf);
2714 }
2715 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2716
2717 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2718                               uid_t auid, u32 sessionid, u32 secid)
2719 {
2720         struct audit_buffer *audit_buf;
2721
2722         audit_buf = xfrm_audit_start("SPD-delete");
2723         if (audit_buf == NULL)
2724                 return;
2725         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2726         audit_log_format(audit_buf, " res=%u", result);
2727         xfrm_audit_common_policyinfo(xp, audit_buf);
2728         audit_log_end(audit_buf);
2729 }
2730 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2731 #endif
2732
2733 #ifdef CONFIG_XFRM_MIGRATE
2734 static int xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
2735                                        const struct xfrm_selector *sel_tgt)
2736 {
2737         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2738                 if (sel_tgt->family == sel_cmp->family &&
2739                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2740                                   sel_cmp->family) == 0 &&
2741                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2742                                   sel_cmp->family) == 0 &&
2743                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2744                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2745                         return 1;
2746                 }
2747         } else {
2748                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2749                         return 1;
2750                 }
2751         }
2752         return 0;
2753 }
2754
2755 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
2756                                                      u8 dir, u8 type)
2757 {
2758         struct xfrm_policy *pol, *ret = NULL;
2759         struct hlist_node *entry;
2760         struct hlist_head *chain;
2761         u32 priority = ~0U;
2762
2763         read_lock_bh(&xfrm_policy_lock);
2764         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2765         hlist_for_each_entry(pol, entry, chain, bydst) {
2766                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2767                     pol->type == type) {
2768                         ret = pol;
2769                         priority = ret->priority;
2770                         break;
2771                 }
2772         }
2773         chain = &init_net.xfrm.policy_inexact[dir];
2774         hlist_for_each_entry(pol, entry, chain, bydst) {
2775                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2776                     pol->type == type &&
2777                     pol->priority < priority) {
2778                         ret = pol;
2779                         break;
2780                 }
2781         }
2782
2783         if (ret)
2784                 xfrm_pol_hold(ret);
2785
2786         read_unlock_bh(&xfrm_policy_lock);
2787
2788         return ret;
2789 }
2790
2791 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2792 {
2793         int match = 0;
2794
2795         if (t->mode == m->mode && t->id.proto == m->proto &&
2796             (m->reqid == 0 || t->reqid == m->reqid)) {
2797                 switch (t->mode) {
2798                 case XFRM_MODE_TUNNEL:
2799                 case XFRM_MODE_BEET:
2800                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2801                                           m->old_family) == 0 &&
2802                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2803                                           m->old_family) == 0) {
2804                                 match = 1;
2805                         }
2806                         break;
2807                 case XFRM_MODE_TRANSPORT:
2808                         /* in case of transport mode, template does not store
2809                            any IP addresses, hence we just compare mode and
2810                            protocol */
2811                         match = 1;
2812                         break;
2813                 default:
2814                         break;
2815                 }
2816         }
2817         return match;
2818 }
2819
2820 /* update endpoint address(es) of template(s) */
2821 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2822                                struct xfrm_migrate *m, int num_migrate)
2823 {
2824         struct xfrm_migrate *mp;
2825         int i, j, n = 0;
2826
2827         write_lock_bh(&pol->lock);
2828         if (unlikely(pol->walk.dead)) {
2829                 /* target policy has been deleted */
2830                 write_unlock_bh(&pol->lock);
2831                 return -ENOENT;
2832         }
2833
2834         for (i = 0; i < pol->xfrm_nr; i++) {
2835                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2836                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2837                                 continue;
2838                         n++;
2839                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2840                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2841                                 continue;
2842                         /* update endpoints */
2843                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2844                                sizeof(pol->xfrm_vec[i].id.daddr));
2845                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2846                                sizeof(pol->xfrm_vec[i].saddr));
2847                         pol->xfrm_vec[i].encap_family = mp->new_family;
2848                         /* flush bundles */
2849                         atomic_inc(&pol->genid);
2850                 }
2851         }
2852
2853         write_unlock_bh(&pol->lock);
2854
2855         if (!n)
2856                 return -ENODATA;
2857
2858         return 0;
2859 }
2860
2861 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
2862 {
2863         int i, j;
2864
2865         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2866                 return -EINVAL;
2867
2868         for (i = 0; i < num_migrate; i++) {
2869                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2870                                    m[i].old_family) == 0) &&
2871                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2872                                    m[i].old_family) == 0))
2873                         return -EINVAL;
2874                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2875                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2876                         return -EINVAL;
2877
2878                 /* check if there is any duplicated entry */
2879                 for (j = i + 1; j < num_migrate; j++) {
2880                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2881                                     sizeof(m[i].old_daddr)) &&
2882                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2883                                     sizeof(m[i].old_saddr)) &&
2884                             m[i].proto == m[j].proto &&
2885                             m[i].mode == m[j].mode &&
2886                             m[i].reqid == m[j].reqid &&
2887                             m[i].old_family == m[j].old_family)
2888                                 return -EINVAL;
2889                 }
2890         }
2891
2892         return 0;
2893 }
2894
2895 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
2896                  struct xfrm_migrate *m, int num_migrate,
2897                  struct xfrm_kmaddress *k)
2898 {
2899         int i, err, nx_cur = 0, nx_new = 0;
2900         struct xfrm_policy *pol = NULL;
2901         struct xfrm_state *x, *xc;
2902         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2903         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2904         struct xfrm_migrate *mp;
2905
2906         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2907                 goto out;
2908
2909         /* Stage 1 - find policy */
2910         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2911                 err = -ENOENT;
2912                 goto out;
2913         }
2914
2915         /* Stage 2 - find and update state(s) */
2916         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2917                 if ((x = xfrm_migrate_state_find(mp))) {
2918                         x_cur[nx_cur] = x;
2919                         nx_cur++;
2920                         if ((xc = xfrm_state_migrate(x, mp))) {
2921                                 x_new[nx_new] = xc;
2922                                 nx_new++;
2923                         } else {
2924                                 err = -ENODATA;
2925                                 goto restore_state;
2926                         }
2927                 }
2928         }
2929
2930         /* Stage 3 - update policy */
2931         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2932                 goto restore_state;
2933
2934         /* Stage 4 - delete old state(s) */
2935         if (nx_cur) {
2936                 xfrm_states_put(x_cur, nx_cur);
2937                 xfrm_states_delete(x_cur, nx_cur);
2938         }
2939
2940         /* Stage 5 - announce */
2941         km_migrate(sel, dir, type, m, num_migrate, k);
2942
2943         xfrm_pol_put(pol);
2944
2945         return 0;
2946 out:
2947         return err;
2948
2949 restore_state:
2950         if (pol)
2951                 xfrm_pol_put(pol);
2952         if (nx_cur)
2953                 xfrm_states_put(x_cur, nx_cur);
2954         if (nx_new)
2955                 xfrm_states_delete(x_new, nx_new);
2956
2957         return err;
2958 }
2959 EXPORT_SYMBOL(xfrm_migrate);
2960 #endif