6 * Kazunori MIYAZAWA @USAGI
7 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
9 * Kazunori MIYAZAWA @USAGI
11 * Split up af-specific portion
12 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
17 #include <linux/config.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/workqueue.h>
23 #include <linux/notifier.h>
24 #include <linux/netdevice.h>
25 #include <linux/module.h>
29 DECLARE_MUTEX(xfrm_cfg_sem);
30 EXPORT_SYMBOL(xfrm_cfg_sem);
32 static DEFINE_RWLOCK(xfrm_policy_lock);
34 struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2];
35 EXPORT_SYMBOL(xfrm_policy_list);
37 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
38 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
40 static kmem_cache_t *xfrm_dst_cache;
42 static struct work_struct xfrm_policy_gc_work;
43 static struct list_head xfrm_policy_gc_list =
44 LIST_HEAD_INIT(xfrm_policy_gc_list);
45 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
47 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
48 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
50 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
52 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
53 struct xfrm_type_map *typemap;
56 if (unlikely(afinfo == NULL))
58 typemap = afinfo->type_map;
60 write_lock(&typemap->lock);
61 if (likely(typemap->map[type->proto] == NULL))
62 typemap->map[type->proto] = type;
65 write_unlock(&typemap->lock);
66 xfrm_policy_put_afinfo(afinfo);
69 EXPORT_SYMBOL(xfrm_register_type);
71 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
73 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
74 struct xfrm_type_map *typemap;
77 if (unlikely(afinfo == NULL))
79 typemap = afinfo->type_map;
81 write_lock(&typemap->lock);
82 if (unlikely(typemap->map[type->proto] != type))
85 typemap->map[type->proto] = NULL;
86 write_unlock(&typemap->lock);
87 xfrm_policy_put_afinfo(afinfo);
90 EXPORT_SYMBOL(xfrm_unregister_type);
92 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
94 struct xfrm_policy_afinfo *afinfo;
95 struct xfrm_type_map *typemap;
96 struct xfrm_type *type;
97 int modload_attempted = 0;
100 afinfo = xfrm_policy_get_afinfo(family);
101 if (unlikely(afinfo == NULL))
103 typemap = afinfo->type_map;
105 read_lock(&typemap->lock);
106 type = typemap->map[proto];
107 if (unlikely(type && !try_module_get(type->owner)))
109 read_unlock(&typemap->lock);
110 if (!type && !modload_attempted) {
111 xfrm_policy_put_afinfo(afinfo);
112 request_module("xfrm-type-%d-%d",
113 (int) family, (int) proto);
114 modload_attempted = 1;
118 xfrm_policy_put_afinfo(afinfo);
121 EXPORT_SYMBOL(xfrm_get_type);
123 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
124 unsigned short family)
126 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
129 if (unlikely(afinfo == NULL))
130 return -EAFNOSUPPORT;
132 if (likely(afinfo->dst_lookup != NULL))
133 err = afinfo->dst_lookup(dst, fl);
136 xfrm_policy_put_afinfo(afinfo);
139 EXPORT_SYMBOL(xfrm_dst_lookup);
141 void xfrm_put_type(struct xfrm_type *type)
143 module_put(type->owner);
146 static inline unsigned long make_jiffies(long secs)
148 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
149 return MAX_SCHEDULE_TIMEOUT-1;
154 static void xfrm_policy_timer(unsigned long data)
156 struct xfrm_policy *xp = (struct xfrm_policy*)data;
157 unsigned long now = (unsigned long)xtime.tv_sec;
158 long next = LONG_MAX;
162 read_lock(&xp->lock);
169 if (xp->lft.hard_add_expires_seconds) {
170 long tmo = xp->lft.hard_add_expires_seconds +
171 xp->curlft.add_time - now;
177 if (xp->lft.hard_use_expires_seconds) {
178 long tmo = xp->lft.hard_use_expires_seconds +
179 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
185 if (xp->lft.soft_add_expires_seconds) {
186 long tmo = xp->lft.soft_add_expires_seconds +
187 xp->curlft.add_time - now;
190 tmo = XFRM_KM_TIMEOUT;
195 if (xp->lft.soft_use_expires_seconds) {
196 long tmo = xp->lft.soft_use_expires_seconds +
197 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
200 tmo = XFRM_KM_TIMEOUT;
207 km_policy_expired(xp, dir, 0);
208 if (next != LONG_MAX &&
209 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
213 read_unlock(&xp->lock);
218 read_unlock(&xp->lock);
219 km_policy_expired(xp, dir, 1);
220 xfrm_policy_delete(xp, dir);
225 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
229 struct xfrm_policy *xfrm_policy_alloc(int gfp)
231 struct xfrm_policy *policy;
233 policy = kmalloc(sizeof(struct xfrm_policy), gfp);
236 memset(policy, 0, sizeof(struct xfrm_policy));
237 atomic_set(&policy->refcnt, 1);
238 rwlock_init(&policy->lock);
239 init_timer(&policy->timer);
240 policy->timer.data = (unsigned long)policy;
241 policy->timer.function = xfrm_policy_timer;
245 EXPORT_SYMBOL(xfrm_policy_alloc);
247 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
249 void __xfrm_policy_destroy(struct xfrm_policy *policy)
257 if (del_timer(&policy->timer))
262 EXPORT_SYMBOL(__xfrm_policy_destroy);
264 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
266 struct dst_entry *dst;
268 while ((dst = policy->bundles) != NULL) {
269 policy->bundles = dst->next;
273 if (del_timer(&policy->timer))
274 atomic_dec(&policy->refcnt);
276 if (atomic_read(&policy->refcnt) > 1)
279 xfrm_pol_put(policy);
282 static void xfrm_policy_gc_task(void *data)
284 struct xfrm_policy *policy;
285 struct list_head *entry, *tmp;
286 struct list_head gc_list = LIST_HEAD_INIT(gc_list);
288 spin_lock_bh(&xfrm_policy_gc_lock);
289 list_splice_init(&xfrm_policy_gc_list, &gc_list);
290 spin_unlock_bh(&xfrm_policy_gc_lock);
292 list_for_each_safe(entry, tmp, &gc_list) {
293 policy = list_entry(entry, struct xfrm_policy, list);
294 xfrm_policy_gc_kill(policy);
298 /* Rule must be locked. Release descentant resources, announce
299 * entry dead. The rule must be unlinked from lists to the moment.
302 static void xfrm_policy_kill(struct xfrm_policy *policy)
306 write_lock_bh(&policy->lock);
309 write_unlock_bh(&policy->lock);
311 if (unlikely(dead)) {
316 spin_lock(&xfrm_policy_gc_lock);
317 list_add(&policy->list, &xfrm_policy_gc_list);
318 spin_unlock(&xfrm_policy_gc_lock);
320 schedule_work(&xfrm_policy_gc_work);
323 /* Generate new index... KAME seems to generate them ordered by cost
324 * of an absolute inpredictability of ordering of rules. This will not pass. */
325 static u32 xfrm_gen_index(int dir)
328 struct xfrm_policy *p;
329 static u32 idx_generator;
332 idx = (idx_generator | dir);
336 for (p = xfrm_policy_list[dir]; p; p = p->next) {
345 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
347 struct xfrm_policy *pol, **p;
348 struct xfrm_policy *delpol = NULL;
349 struct xfrm_policy **newpos = NULL;
351 write_lock_bh(&xfrm_policy_lock);
352 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
353 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0) {
355 write_unlock_bh(&xfrm_policy_lock);
360 if (policy->priority > pol->priority)
362 } else if (policy->priority >= pol->priority) {
374 xfrm_pol_hold(policy);
377 atomic_inc(&flow_cache_genid);
378 policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
379 policy->curlft.add_time = (unsigned long)xtime.tv_sec;
380 policy->curlft.use_time = 0;
381 if (!mod_timer(&policy->timer, jiffies + HZ))
382 xfrm_pol_hold(policy);
383 write_unlock_bh(&xfrm_policy_lock);
386 xfrm_policy_kill(delpol);
390 EXPORT_SYMBOL(xfrm_policy_insert);
392 struct xfrm_policy *xfrm_policy_bysel(int dir, struct xfrm_selector *sel,
395 struct xfrm_policy *pol, **p;
397 write_lock_bh(&xfrm_policy_lock);
398 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
399 if (memcmp(sel, &pol->selector, sizeof(*sel)) == 0) {
406 write_unlock_bh(&xfrm_policy_lock);
409 atomic_inc(&flow_cache_genid);
410 xfrm_policy_kill(pol);
414 EXPORT_SYMBOL(xfrm_policy_bysel);
416 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
418 struct xfrm_policy *pol, **p;
420 write_lock_bh(&xfrm_policy_lock);
421 for (p = &xfrm_policy_list[id & 7]; (pol=*p)!=NULL; p = &pol->next) {
422 if (pol->index == id) {
429 write_unlock_bh(&xfrm_policy_lock);
432 atomic_inc(&flow_cache_genid);
433 xfrm_policy_kill(pol);
437 EXPORT_SYMBOL(xfrm_policy_byid);
439 void xfrm_policy_flush(void)
441 struct xfrm_policy *xp;
444 write_lock_bh(&xfrm_policy_lock);
445 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
446 while ((xp = xfrm_policy_list[dir]) != NULL) {
447 xfrm_policy_list[dir] = xp->next;
448 write_unlock_bh(&xfrm_policy_lock);
450 xfrm_policy_kill(xp);
452 write_lock_bh(&xfrm_policy_lock);
455 atomic_inc(&flow_cache_genid);
456 write_unlock_bh(&xfrm_policy_lock);
458 EXPORT_SYMBOL(xfrm_policy_flush);
460 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
463 struct xfrm_policy *xp;
468 read_lock_bh(&xfrm_policy_lock);
469 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
470 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
479 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
480 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
481 error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
488 read_unlock_bh(&xfrm_policy_lock);
491 EXPORT_SYMBOL(xfrm_policy_walk);
493 /* Find policy to apply to this flow. */
495 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
496 void **objp, atomic_t **obj_refp)
498 struct xfrm_policy *pol;
500 read_lock_bh(&xfrm_policy_lock);
501 for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
502 struct xfrm_selector *sel = &pol->selector;
505 if (pol->family != family)
508 match = xfrm_selector_match(sel, fl, family);
514 read_unlock_bh(&xfrm_policy_lock);
515 if ((*objp = (void *) pol) != NULL)
516 *obj_refp = &pol->refcnt;
519 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
521 struct xfrm_policy *pol;
523 read_lock_bh(&xfrm_policy_lock);
524 if ((pol = sk->sk_policy[dir]) != NULL) {
525 int match = xfrm_selector_match(&pol->selector, fl,
532 read_unlock_bh(&xfrm_policy_lock);
536 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
538 pol->next = xfrm_policy_list[dir];
539 xfrm_policy_list[dir] = pol;
543 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
546 struct xfrm_policy **polp;
548 for (polp = &xfrm_policy_list[dir];
549 *polp != NULL; polp = &(*polp)->next) {
558 void xfrm_policy_delete(struct xfrm_policy *pol, int dir)
560 write_lock_bh(&xfrm_policy_lock);
561 pol = __xfrm_policy_unlink(pol, dir);
562 write_unlock_bh(&xfrm_policy_lock);
564 if (dir < XFRM_POLICY_MAX)
565 atomic_inc(&flow_cache_genid);
566 xfrm_policy_kill(pol);
570 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
572 struct xfrm_policy *old_pol;
574 write_lock_bh(&xfrm_policy_lock);
575 old_pol = sk->sk_policy[dir];
576 sk->sk_policy[dir] = pol;
578 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
579 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
580 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
583 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
584 write_unlock_bh(&xfrm_policy_lock);
587 xfrm_policy_kill(old_pol);
592 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
594 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
597 newp->selector = old->selector;
598 newp->lft = old->lft;
599 newp->curlft = old->curlft;
600 newp->action = old->action;
601 newp->flags = old->flags;
602 newp->xfrm_nr = old->xfrm_nr;
603 newp->index = old->index;
604 memcpy(newp->xfrm_vec, old->xfrm_vec,
605 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
606 write_lock_bh(&xfrm_policy_lock);
607 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
608 write_unlock_bh(&xfrm_policy_lock);
614 int __xfrm_sk_clone_policy(struct sock *sk)
616 struct xfrm_policy *p0 = sk->sk_policy[0],
617 *p1 = sk->sk_policy[1];
619 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
620 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
622 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
627 /* Resolve list of templates for the flow, given policy. */
630 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
631 struct xfrm_state **xfrm,
632 unsigned short family)
636 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
637 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
639 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
640 struct xfrm_state *x;
641 xfrm_address_t *remote = daddr;
642 xfrm_address_t *local = saddr;
643 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
646 remote = &tmpl->id.daddr;
647 local = &tmpl->saddr;
650 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
652 if (x && x->km.state == XFRM_STATE_VALID) {
659 error = (x->km.state == XFRM_STATE_ERROR ?
670 for (nx--; nx>=0; nx--)
671 xfrm_state_put(xfrm[nx]);
675 /* Check that the bundle accepts the flow and its components are
679 static struct dst_entry *
680 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
683 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
684 if (unlikely(afinfo == NULL))
685 return ERR_PTR(-EINVAL);
686 x = afinfo->find_bundle(fl, policy);
687 xfrm_policy_put_afinfo(afinfo);
691 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
692 * all the metrics... Shortly, bundle a bundle.
696 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
697 struct flowi *fl, struct dst_entry **dst_p,
698 unsigned short family)
701 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
702 if (unlikely(afinfo == NULL))
704 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
705 xfrm_policy_put_afinfo(afinfo);
709 static inline int policy_to_flow_dir(int dir)
711 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
712 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
713 XFRM_POLICY_FWD == FLOW_DIR_FWD)
719 case XFRM_POLICY_OUT:
721 case XFRM_POLICY_FWD:
726 static int stale_bundle(struct dst_entry *dst);
728 /* Main function: finds/creates a bundle for given flow.
730 * At the moment we eat a raw IP route. Mostly to speed up lookups
731 * on interfaces with disabled IPsec.
733 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
734 struct sock *sk, int flags)
736 struct xfrm_policy *policy;
737 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
738 struct dst_entry *dst, *dst_orig = *dst_p;
742 u16 family = dst_orig->ops->family;
744 genid = atomic_read(&flow_cache_genid);
746 if (sk && sk->sk_policy[1])
747 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
750 /* To accelerate a bit... */
751 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
754 policy = flow_cache_lookup(fl, family,
755 policy_to_flow_dir(XFRM_POLICY_OUT),
762 policy->curlft.use_time = (unsigned long)xtime.tv_sec;
764 switch (policy->action) {
765 case XFRM_POLICY_BLOCK:
766 /* Prohibit the flow */
767 xfrm_pol_put(policy);
770 case XFRM_POLICY_ALLOW:
771 if (policy->xfrm_nr == 0) {
772 /* Flow passes not transformed. */
773 xfrm_pol_put(policy);
777 /* Try to find matching bundle.
779 * LATER: help from flow cache. It is optional, this
780 * is required only for output policy.
782 dst = xfrm_find_bundle(fl, policy, family);
784 xfrm_pol_put(policy);
791 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
793 if (unlikely(nx<0)) {
795 if (err == -EAGAIN && flags) {
796 DECLARE_WAITQUEUE(wait, current);
798 add_wait_queue(&km_waitq, &wait);
799 set_current_state(TASK_INTERRUPTIBLE);
801 set_current_state(TASK_RUNNING);
802 remove_wait_queue(&km_waitq, &wait);
804 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
806 if (nx == -EAGAIN && signal_pending(current)) {
811 genid != atomic_read(&flow_cache_genid)) {
812 xfrm_pol_put(policy);
821 /* Flow passes not transformed. */
822 xfrm_pol_put(policy);
827 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
832 xfrm_state_put(xfrm[i]);
836 write_lock_bh(&policy->lock);
837 if (unlikely(policy->dead || stale_bundle(dst))) {
838 /* Wow! While we worked on resolving, this
839 * policy has gone. Retry. It is not paranoia,
840 * we just cannot enlist new bundle to dead object.
841 * We can't enlist stable bundles either.
843 write_unlock_bh(&policy->lock);
845 xfrm_pol_put(policy);
850 dst->next = policy->bundles;
851 policy->bundles = dst;
853 write_unlock_bh(&policy->lock);
856 dst_release(dst_orig);
857 xfrm_pol_put(policy);
861 dst_release(dst_orig);
862 xfrm_pol_put(policy);
866 EXPORT_SYMBOL(xfrm_lookup);
868 /* When skb is transformed back to its "native" form, we have to
869 * check policy restrictions. At the moment we make this in maximally
870 * stupid way. Shame on me. :-) Of course, connected sockets must
871 * have policy cached at them.
875 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
876 unsigned short family)
878 if (xfrm_state_kern(x))
879 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
880 return x->id.proto == tmpl->id.proto &&
881 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
882 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
883 x->props.mode == tmpl->mode &&
884 (tmpl->aalgos & (1<<x->props.aalgo)) &&
885 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
889 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
890 unsigned short family)
894 if (tmpl->optional) {
899 for (; idx < sp->len; idx++) {
900 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
902 if (sp->x[idx].xvec->props.mode)
909 _decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
911 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
913 if (unlikely(afinfo == NULL))
914 return -EAFNOSUPPORT;
916 afinfo->decode_session(skb, fl);
917 xfrm_policy_put_afinfo(afinfo);
921 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
923 for (; k < sp->len; k++) {
924 if (sp->x[k].xvec->props.mode)
931 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
932 unsigned short family)
934 struct xfrm_policy *pol;
937 if (_decode_session(skb, &fl, family) < 0)
940 /* First, check used SA against their selectors. */
944 for (i=skb->sp->len-1; i>=0; i--) {
945 struct sec_decap_state *xvec = &(skb->sp->x[i]);
946 if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
949 /* If there is a post_input processor, try running it */
950 if (xvec->xvec->type->post_input &&
951 (xvec->xvec->type->post_input)(xvec->xvec,
959 if (sk && sk->sk_policy[dir])
960 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
963 pol = flow_cache_lookup(&fl, family,
964 policy_to_flow_dir(dir),
968 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
970 pol->curlft.use_time = (unsigned long)xtime.tv_sec;
972 if (pol->action == XFRM_POLICY_ALLOW) {
974 static struct sec_path dummy;
977 if ((sp = skb->sp) == NULL)
980 /* For each tunnel xfrm, find the first matching tmpl.
981 * For each tmpl before that, find corresponding xfrm.
982 * Order is _important_. Later we will implement
983 * some barriers, but at the moment barriers
984 * are implied between each two transformations.
986 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
987 k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
992 if (secpath_has_tunnel(sp, k))
1003 EXPORT_SYMBOL(__xfrm_policy_check);
1005 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1009 if (_decode_session(skb, &fl, family) < 0)
1012 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1014 EXPORT_SYMBOL(__xfrm_route_forward);
1016 /* Optimize later using cookies and generation ids. */
1018 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1020 if (!stale_bundle(dst))
1026 static int stale_bundle(struct dst_entry *dst)
1028 return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1031 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1033 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1034 dst->dev = &loopback_dev;
1035 dev_hold(&loopback_dev);
1039 EXPORT_SYMBOL(xfrm_dst_ifdown);
1041 static void xfrm_link_failure(struct sk_buff *skb)
1043 /* Impossible. Such dst must be popped before reaches point of failure. */
1047 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1050 if (dst->obsolete) {
1058 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1061 struct xfrm_policy *pol;
1062 struct dst_entry *dst, **dstp, *gc_list = NULL;
1064 read_lock_bh(&xfrm_policy_lock);
1065 for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1066 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1067 write_lock(&pol->lock);
1068 dstp = &pol->bundles;
1069 while ((dst=*dstp) != NULL) {
1072 dst->next = gc_list;
1078 write_unlock(&pol->lock);
1081 read_unlock_bh(&xfrm_policy_lock);
1085 gc_list = dst->next;
1090 static int unused_bundle(struct dst_entry *dst)
1092 return !atomic_read(&dst->__refcnt);
1095 static void __xfrm_garbage_collect(void)
1097 xfrm_prune_bundles(unused_bundle);
1100 int xfrm_flush_bundles(void)
1102 xfrm_prune_bundles(stale_bundle);
1106 void xfrm_init_pmtu(struct dst_entry *dst)
1109 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1110 u32 pmtu, route_mtu_cached;
1112 pmtu = dst_mtu(dst->child);
1113 xdst->child_mtu_cached = pmtu;
1115 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1117 route_mtu_cached = dst_mtu(xdst->route);
1118 xdst->route_mtu_cached = route_mtu_cached;
1120 if (pmtu > route_mtu_cached)
1121 pmtu = route_mtu_cached;
1123 dst->metrics[RTAX_MTU-1] = pmtu;
1124 } while ((dst = dst->next));
1127 EXPORT_SYMBOL(xfrm_init_pmtu);
1129 /* Check that the bundle accepts the flow and its components are
1133 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1135 struct dst_entry *dst = &first->u.dst;
1136 struct xfrm_dst *last;
1139 if (!dst_check(dst->path, 0) ||
1140 (dst->dev && !netif_running(dst->dev)))
1146 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1148 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1150 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1153 mtu = dst_mtu(dst->child);
1154 if (xdst->child_mtu_cached != mtu) {
1156 xdst->child_mtu_cached = mtu;
1159 if (!dst_check(xdst->route, 0))
1161 mtu = dst_mtu(xdst->route);
1162 if (xdst->route_mtu_cached != mtu) {
1164 xdst->route_mtu_cached = mtu;
1168 } while (dst->xfrm);
1173 mtu = last->child_mtu_cached;
1177 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1178 if (mtu > last->route_mtu_cached)
1179 mtu = last->route_mtu_cached;
1180 dst->metrics[RTAX_MTU-1] = mtu;
1185 last = last->u.next;
1186 last->child_mtu_cached = mtu;
1192 EXPORT_SYMBOL(xfrm_bundle_ok);
1194 /* Well... that's _TASK_. We need to scan through transformation
1195 * list and figure out what mss tcp should generate in order to
1196 * final datagram fit to mtu. Mama mia... :-)
1198 * Apparently, some easy way exists, but we used to choose the most
1199 * bizarre ones. :-) So, raising Kalashnikov... tra-ta-ta.
1201 * Consider this function as something like dark humour. :-)
1203 static int xfrm_get_mss(struct dst_entry *dst, u32 mtu)
1205 int res = mtu - dst->header_len;
1208 struct dst_entry *d = dst;
1212 struct xfrm_state *x = d->xfrm;
1214 spin_lock_bh(&x->lock);
1215 if (x->km.state == XFRM_STATE_VALID &&
1216 x->type && x->type->get_max_size)
1217 m = x->type->get_max_size(d->xfrm, m);
1219 m += x->props.header_len;
1220 spin_unlock_bh(&x->lock);
1222 } while ((d = d->child) != NULL);
1231 return res + dst->header_len;
1234 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1237 if (unlikely(afinfo == NULL))
1239 if (unlikely(afinfo->family >= NPROTO))
1240 return -EAFNOSUPPORT;
1241 write_lock(&xfrm_policy_afinfo_lock);
1242 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1245 struct dst_ops *dst_ops = afinfo->dst_ops;
1246 if (likely(dst_ops->kmem_cachep == NULL))
1247 dst_ops->kmem_cachep = xfrm_dst_cache;
1248 if (likely(dst_ops->check == NULL))
1249 dst_ops->check = xfrm_dst_check;
1250 if (likely(dst_ops->negative_advice == NULL))
1251 dst_ops->negative_advice = xfrm_negative_advice;
1252 if (likely(dst_ops->link_failure == NULL))
1253 dst_ops->link_failure = xfrm_link_failure;
1254 if (likely(dst_ops->get_mss == NULL))
1255 dst_ops->get_mss = xfrm_get_mss;
1256 if (likely(afinfo->garbage_collect == NULL))
1257 afinfo->garbage_collect = __xfrm_garbage_collect;
1258 xfrm_policy_afinfo[afinfo->family] = afinfo;
1260 write_unlock(&xfrm_policy_afinfo_lock);
1263 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1265 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1268 if (unlikely(afinfo == NULL))
1270 if (unlikely(afinfo->family >= NPROTO))
1271 return -EAFNOSUPPORT;
1272 write_lock(&xfrm_policy_afinfo_lock);
1273 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1274 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1277 struct dst_ops *dst_ops = afinfo->dst_ops;
1278 xfrm_policy_afinfo[afinfo->family] = NULL;
1279 dst_ops->kmem_cachep = NULL;
1280 dst_ops->check = NULL;
1281 dst_ops->negative_advice = NULL;
1282 dst_ops->link_failure = NULL;
1283 dst_ops->get_mss = NULL;
1284 afinfo->garbage_collect = NULL;
1287 write_unlock(&xfrm_policy_afinfo_lock);
1290 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1292 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1294 struct xfrm_policy_afinfo *afinfo;
1295 if (unlikely(family >= NPROTO))
1297 read_lock(&xfrm_policy_afinfo_lock);
1298 afinfo = xfrm_policy_afinfo[family];
1299 if (likely(afinfo != NULL))
1300 read_lock(&afinfo->lock);
1301 read_unlock(&xfrm_policy_afinfo_lock);
1305 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1307 if (unlikely(afinfo == NULL))
1309 read_unlock(&afinfo->lock);
1312 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1316 xfrm_flush_bundles();
1321 static struct notifier_block xfrm_dev_notifier = {
1327 static void __init xfrm_policy_init(void)
1329 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1330 sizeof(struct xfrm_dst),
1331 0, SLAB_HWCACHE_ALIGN,
1333 if (!xfrm_dst_cache)
1334 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1336 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1337 register_netdevice_notifier(&xfrm_dev_notifier);
1340 void __init xfrm_init(void)