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 __read_mostly;
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);
122 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
123 unsigned short family)
125 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
128 if (unlikely(afinfo == NULL))
129 return -EAFNOSUPPORT;
131 if (likely(afinfo->dst_lookup != NULL))
132 err = afinfo->dst_lookup(dst, fl);
135 xfrm_policy_put_afinfo(afinfo);
138 EXPORT_SYMBOL(xfrm_dst_lookup);
140 void xfrm_put_type(struct xfrm_type *type)
142 module_put(type->owner);
145 static inline unsigned long make_jiffies(long secs)
147 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
148 return MAX_SCHEDULE_TIMEOUT-1;
153 static void xfrm_policy_timer(unsigned long data)
155 struct xfrm_policy *xp = (struct xfrm_policy*)data;
156 unsigned long now = (unsigned long)xtime.tv_sec;
157 long next = LONG_MAX;
161 read_lock(&xp->lock);
166 dir = xfrm_policy_id2dir(xp->index);
168 if (xp->lft.hard_add_expires_seconds) {
169 long tmo = xp->lft.hard_add_expires_seconds +
170 xp->curlft.add_time - now;
176 if (xp->lft.hard_use_expires_seconds) {
177 long tmo = xp->lft.hard_use_expires_seconds +
178 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
184 if (xp->lft.soft_add_expires_seconds) {
185 long tmo = xp->lft.soft_add_expires_seconds +
186 xp->curlft.add_time - now;
189 tmo = XFRM_KM_TIMEOUT;
194 if (xp->lft.soft_use_expires_seconds) {
195 long tmo = xp->lft.soft_use_expires_seconds +
196 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
199 tmo = XFRM_KM_TIMEOUT;
206 km_policy_expired(xp, dir, 0);
207 if (next != LONG_MAX &&
208 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
212 read_unlock(&xp->lock);
217 read_unlock(&xp->lock);
218 if (!xfrm_policy_delete(xp, dir))
219 km_policy_expired(xp, dir, 1);
224 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
228 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
230 struct xfrm_policy *policy;
232 policy = kmalloc(sizeof(struct xfrm_policy), gfp);
235 memset(policy, 0, sizeof(struct xfrm_policy));
236 atomic_set(&policy->refcnt, 1);
237 rwlock_init(&policy->lock);
238 init_timer(&policy->timer);
239 policy->timer.data = (unsigned long)policy;
240 policy->timer.function = xfrm_policy_timer;
244 EXPORT_SYMBOL(xfrm_policy_alloc);
246 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
248 void __xfrm_policy_destroy(struct xfrm_policy *policy)
256 if (del_timer(&policy->timer))
261 EXPORT_SYMBOL(__xfrm_policy_destroy);
263 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
265 struct dst_entry *dst;
267 while ((dst = policy->bundles) != NULL) {
268 policy->bundles = dst->next;
272 if (del_timer(&policy->timer))
273 atomic_dec(&policy->refcnt);
275 if (atomic_read(&policy->refcnt) > 1)
278 xfrm_pol_put(policy);
281 static void xfrm_policy_gc_task(void *data)
283 struct xfrm_policy *policy;
284 struct list_head *entry, *tmp;
285 struct list_head gc_list = LIST_HEAD_INIT(gc_list);
287 spin_lock_bh(&xfrm_policy_gc_lock);
288 list_splice_init(&xfrm_policy_gc_list, &gc_list);
289 spin_unlock_bh(&xfrm_policy_gc_lock);
291 list_for_each_safe(entry, tmp, &gc_list) {
292 policy = list_entry(entry, struct xfrm_policy, list);
293 xfrm_policy_gc_kill(policy);
297 /* Rule must be locked. Release descentant resources, announce
298 * entry dead. The rule must be unlinked from lists to the moment.
301 static void xfrm_policy_kill(struct xfrm_policy *policy)
305 write_lock_bh(&policy->lock);
308 write_unlock_bh(&policy->lock);
310 if (unlikely(dead)) {
315 spin_lock(&xfrm_policy_gc_lock);
316 list_add(&policy->list, &xfrm_policy_gc_list);
317 spin_unlock(&xfrm_policy_gc_lock);
319 schedule_work(&xfrm_policy_gc_work);
322 /* Generate new index... KAME seems to generate them ordered by cost
323 * of an absolute inpredictability of ordering of rules. This will not pass. */
324 static u32 xfrm_gen_index(int dir)
327 struct xfrm_policy *p;
328 static u32 idx_generator;
331 idx = (idx_generator | dir);
335 for (p = xfrm_policy_list[dir]; p; p = p->next) {
344 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
346 struct xfrm_policy *pol, **p;
347 struct xfrm_policy *delpol = NULL;
348 struct xfrm_policy **newpos = NULL;
350 write_lock_bh(&xfrm_policy_lock);
351 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
352 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0) {
354 write_unlock_bh(&xfrm_policy_lock);
359 if (policy->priority > pol->priority)
361 } else if (policy->priority >= pol->priority) {
373 xfrm_pol_hold(policy);
376 atomic_inc(&flow_cache_genid);
377 policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
378 policy->curlft.add_time = (unsigned long)xtime.tv_sec;
379 policy->curlft.use_time = 0;
380 if (!mod_timer(&policy->timer, jiffies + HZ))
381 xfrm_pol_hold(policy);
382 write_unlock_bh(&xfrm_policy_lock);
385 xfrm_policy_kill(delpol);
389 EXPORT_SYMBOL(xfrm_policy_insert);
391 struct xfrm_policy *xfrm_policy_bysel(int dir, struct xfrm_selector *sel,
394 struct xfrm_policy *pol, **p;
396 write_lock_bh(&xfrm_policy_lock);
397 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
398 if (memcmp(sel, &pol->selector, sizeof(*sel)) == 0) {
405 write_unlock_bh(&xfrm_policy_lock);
408 atomic_inc(&flow_cache_genid);
409 xfrm_policy_kill(pol);
413 EXPORT_SYMBOL(xfrm_policy_bysel);
415 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
417 struct xfrm_policy *pol, **p;
419 write_lock_bh(&xfrm_policy_lock);
420 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
421 if (pol->index == id) {
428 write_unlock_bh(&xfrm_policy_lock);
431 atomic_inc(&flow_cache_genid);
432 xfrm_policy_kill(pol);
436 EXPORT_SYMBOL(xfrm_policy_byid);
438 void xfrm_policy_flush(void)
440 struct xfrm_policy *xp;
443 write_lock_bh(&xfrm_policy_lock);
444 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
445 while ((xp = xfrm_policy_list[dir]) != NULL) {
446 xfrm_policy_list[dir] = xp->next;
447 write_unlock_bh(&xfrm_policy_lock);
449 xfrm_policy_kill(xp);
451 write_lock_bh(&xfrm_policy_lock);
454 atomic_inc(&flow_cache_genid);
455 write_unlock_bh(&xfrm_policy_lock);
457 EXPORT_SYMBOL(xfrm_policy_flush);
459 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
462 struct xfrm_policy *xp;
467 read_lock_bh(&xfrm_policy_lock);
468 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
469 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
478 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
479 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
480 error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
487 read_unlock_bh(&xfrm_policy_lock);
490 EXPORT_SYMBOL(xfrm_policy_walk);
492 /* Find policy to apply to this flow. */
494 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
495 void **objp, atomic_t **obj_refp)
497 struct xfrm_policy *pol;
499 read_lock_bh(&xfrm_policy_lock);
500 for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
501 struct xfrm_selector *sel = &pol->selector;
504 if (pol->family != family)
507 match = xfrm_selector_match(sel, fl, family);
513 read_unlock_bh(&xfrm_policy_lock);
514 if ((*objp = (void *) pol) != NULL)
515 *obj_refp = &pol->refcnt;
518 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
520 struct xfrm_policy *pol;
522 read_lock_bh(&xfrm_policy_lock);
523 if ((pol = sk->sk_policy[dir]) != NULL) {
524 int match = xfrm_selector_match(&pol->selector, fl,
531 read_unlock_bh(&xfrm_policy_lock);
535 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
537 pol->next = xfrm_policy_list[dir];
538 xfrm_policy_list[dir] = pol;
542 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
545 struct xfrm_policy **polp;
547 for (polp = &xfrm_policy_list[dir];
548 *polp != NULL; polp = &(*polp)->next) {
557 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
559 write_lock_bh(&xfrm_policy_lock);
560 pol = __xfrm_policy_unlink(pol, dir);
561 write_unlock_bh(&xfrm_policy_lock);
563 if (dir < XFRM_POLICY_MAX)
564 atomic_inc(&flow_cache_genid);
565 xfrm_policy_kill(pol);
571 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
573 struct xfrm_policy *old_pol;
575 write_lock_bh(&xfrm_policy_lock);
576 old_pol = sk->sk_policy[dir];
577 sk->sk_policy[dir] = pol;
579 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
580 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
581 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
584 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
585 write_unlock_bh(&xfrm_policy_lock);
588 xfrm_policy_kill(old_pol);
593 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
595 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
598 newp->selector = old->selector;
599 newp->lft = old->lft;
600 newp->curlft = old->curlft;
601 newp->action = old->action;
602 newp->flags = old->flags;
603 newp->xfrm_nr = old->xfrm_nr;
604 newp->index = old->index;
605 memcpy(newp->xfrm_vec, old->xfrm_vec,
606 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
607 write_lock_bh(&xfrm_policy_lock);
608 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
609 write_unlock_bh(&xfrm_policy_lock);
615 int __xfrm_sk_clone_policy(struct sock *sk)
617 struct xfrm_policy *p0 = sk->sk_policy[0],
618 *p1 = sk->sk_policy[1];
620 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
621 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
623 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
628 /* Resolve list of templates for the flow, given policy. */
631 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
632 struct xfrm_state **xfrm,
633 unsigned short family)
637 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
638 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
640 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
641 struct xfrm_state *x;
642 xfrm_address_t *remote = daddr;
643 xfrm_address_t *local = saddr;
644 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
647 remote = &tmpl->id.daddr;
648 local = &tmpl->saddr;
651 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
653 if (x && x->km.state == XFRM_STATE_VALID) {
660 error = (x->km.state == XFRM_STATE_ERROR ?
671 for (nx--; nx>=0; nx--)
672 xfrm_state_put(xfrm[nx]);
676 /* Check that the bundle accepts the flow and its components are
680 static struct dst_entry *
681 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
684 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
685 if (unlikely(afinfo == NULL))
686 return ERR_PTR(-EINVAL);
687 x = afinfo->find_bundle(fl, policy);
688 xfrm_policy_put_afinfo(afinfo);
692 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
693 * all the metrics... Shortly, bundle a bundle.
697 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
698 struct flowi *fl, struct dst_entry **dst_p,
699 unsigned short family)
702 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
703 if (unlikely(afinfo == NULL))
705 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
706 xfrm_policy_put_afinfo(afinfo);
710 static inline int policy_to_flow_dir(int dir)
712 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
713 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
714 XFRM_POLICY_FWD == FLOW_DIR_FWD)
720 case XFRM_POLICY_OUT:
722 case XFRM_POLICY_FWD:
727 static int stale_bundle(struct dst_entry *dst);
729 /* Main function: finds/creates a bundle for given flow.
731 * At the moment we eat a raw IP route. Mostly to speed up lookups
732 * on interfaces with disabled IPsec.
734 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
735 struct sock *sk, int flags)
737 struct xfrm_policy *policy;
738 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
739 struct dst_entry *dst, *dst_orig = *dst_p;
743 u16 family = dst_orig->ops->family;
745 genid = atomic_read(&flow_cache_genid);
747 if (sk && sk->sk_policy[1])
748 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
751 /* To accelerate a bit... */
752 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
755 policy = flow_cache_lookup(fl, family,
756 policy_to_flow_dir(XFRM_POLICY_OUT),
763 policy->curlft.use_time = (unsigned long)xtime.tv_sec;
765 switch (policy->action) {
766 case XFRM_POLICY_BLOCK:
767 /* Prohibit the flow */
771 case XFRM_POLICY_ALLOW:
772 if (policy->xfrm_nr == 0) {
773 /* Flow passes not transformed. */
774 xfrm_pol_put(policy);
778 /* Try to find matching bundle.
780 * LATER: help from flow cache. It is optional, this
781 * is required only for output policy.
783 dst = xfrm_find_bundle(fl, policy, family);
792 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
794 if (unlikely(nx<0)) {
796 if (err == -EAGAIN && flags) {
797 DECLARE_WAITQUEUE(wait, current);
799 add_wait_queue(&km_waitq, &wait);
800 set_current_state(TASK_INTERRUPTIBLE);
802 set_current_state(TASK_RUNNING);
803 remove_wait_queue(&km_waitq, &wait);
805 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
807 if (nx == -EAGAIN && signal_pending(current)) {
812 genid != atomic_read(&flow_cache_genid)) {
813 xfrm_pol_put(policy);
822 /* Flow passes not transformed. */
823 xfrm_pol_put(policy);
828 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
833 xfrm_state_put(xfrm[i]);
837 write_lock_bh(&policy->lock);
838 if (unlikely(policy->dead || stale_bundle(dst))) {
839 /* Wow! While we worked on resolving, this
840 * policy has gone. Retry. It is not paranoia,
841 * we just cannot enlist new bundle to dead object.
842 * We can't enlist stable bundles either.
844 write_unlock_bh(&policy->lock);
846 xfrm_pol_put(policy);
851 dst->next = policy->bundles;
852 policy->bundles = dst;
854 write_unlock_bh(&policy->lock);
857 dst_release(dst_orig);
858 xfrm_pol_put(policy);
862 dst_release(dst_orig);
863 xfrm_pol_put(policy);
867 EXPORT_SYMBOL(xfrm_lookup);
869 /* When skb is transformed back to its "native" form, we have to
870 * check policy restrictions. At the moment we make this in maximally
871 * stupid way. Shame on me. :-) Of course, connected sockets must
872 * have policy cached at them.
876 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
877 unsigned short family)
879 if (xfrm_state_kern(x))
880 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
881 return x->id.proto == tmpl->id.proto &&
882 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
883 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
884 x->props.mode == tmpl->mode &&
885 (tmpl->aalgos & (1<<x->props.aalgo)) &&
886 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
890 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
891 unsigned short family)
895 if (tmpl->optional) {
900 for (; idx < sp->len; idx++) {
901 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
903 if (sp->x[idx].xvec->props.mode)
910 _decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
912 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
914 if (unlikely(afinfo == NULL))
915 return -EAFNOSUPPORT;
917 afinfo->decode_session(skb, fl);
918 xfrm_policy_put_afinfo(afinfo);
922 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
924 for (; k < sp->len; k++) {
925 if (sp->x[k].xvec->props.mode)
932 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
933 unsigned short family)
935 struct xfrm_policy *pol;
938 if (_decode_session(skb, &fl, family) < 0)
941 /* First, check used SA against their selectors. */
945 for (i=skb->sp->len-1; i>=0; i--) {
946 struct sec_decap_state *xvec = &(skb->sp->x[i]);
947 if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
950 /* If there is a post_input processor, try running it */
951 if (xvec->xvec->type->post_input &&
952 (xvec->xvec->type->post_input)(xvec->xvec,
960 if (sk && sk->sk_policy[dir])
961 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
964 pol = flow_cache_lookup(&fl, family,
965 policy_to_flow_dir(dir),
969 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
971 pol->curlft.use_time = (unsigned long)xtime.tv_sec;
973 if (pol->action == XFRM_POLICY_ALLOW) {
975 static struct sec_path dummy;
978 if ((sp = skb->sp) == NULL)
981 /* For each tunnel xfrm, find the first matching tmpl.
982 * For each tmpl before that, find corresponding xfrm.
983 * Order is _important_. Later we will implement
984 * some barriers, but at the moment barriers
985 * are implied between each two transformations.
987 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
988 k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
993 if (secpath_has_tunnel(sp, k))
1004 EXPORT_SYMBOL(__xfrm_policy_check);
1006 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1010 if (_decode_session(skb, &fl, family) < 0)
1013 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1015 EXPORT_SYMBOL(__xfrm_route_forward);
1017 /* Optimize later using cookies and generation ids. */
1019 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1021 if (!stale_bundle(dst))
1027 static int stale_bundle(struct dst_entry *dst)
1029 return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1032 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1034 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1035 dst->dev = &loopback_dev;
1036 dev_hold(&loopback_dev);
1040 EXPORT_SYMBOL(xfrm_dst_ifdown);
1042 static void xfrm_link_failure(struct sk_buff *skb)
1044 /* Impossible. Such dst must be popped before reaches point of failure. */
1048 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1051 if (dst->obsolete) {
1059 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1062 struct xfrm_policy *pol;
1063 struct dst_entry *dst, **dstp, *gc_list = NULL;
1065 read_lock_bh(&xfrm_policy_lock);
1066 for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1067 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1068 write_lock(&pol->lock);
1069 dstp = &pol->bundles;
1070 while ((dst=*dstp) != NULL) {
1073 dst->next = gc_list;
1079 write_unlock(&pol->lock);
1082 read_unlock_bh(&xfrm_policy_lock);
1086 gc_list = dst->next;
1091 static int unused_bundle(struct dst_entry *dst)
1093 return !atomic_read(&dst->__refcnt);
1096 static void __xfrm_garbage_collect(void)
1098 xfrm_prune_bundles(unused_bundle);
1101 int xfrm_flush_bundles(void)
1103 xfrm_prune_bundles(stale_bundle);
1107 void xfrm_init_pmtu(struct dst_entry *dst)
1110 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1111 u32 pmtu, route_mtu_cached;
1113 pmtu = dst_mtu(dst->child);
1114 xdst->child_mtu_cached = pmtu;
1116 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1118 route_mtu_cached = dst_mtu(xdst->route);
1119 xdst->route_mtu_cached = route_mtu_cached;
1121 if (pmtu > route_mtu_cached)
1122 pmtu = route_mtu_cached;
1124 dst->metrics[RTAX_MTU-1] = pmtu;
1125 } while ((dst = dst->next));
1128 EXPORT_SYMBOL(xfrm_init_pmtu);
1130 /* Check that the bundle accepts the flow and its components are
1134 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1136 struct dst_entry *dst = &first->u.dst;
1137 struct xfrm_dst *last;
1140 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1141 (dst->dev && !netif_running(dst->dev)))
1147 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1149 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1151 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1154 mtu = dst_mtu(dst->child);
1155 if (xdst->child_mtu_cached != mtu) {
1157 xdst->child_mtu_cached = mtu;
1160 if (!dst_check(xdst->route, xdst->route_cookie))
1162 mtu = dst_mtu(xdst->route);
1163 if (xdst->route_mtu_cached != mtu) {
1165 xdst->route_mtu_cached = mtu;
1169 } while (dst->xfrm);
1174 mtu = last->child_mtu_cached;
1178 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1179 if (mtu > last->route_mtu_cached)
1180 mtu = last->route_mtu_cached;
1181 dst->metrics[RTAX_MTU-1] = mtu;
1186 last = last->u.next;
1187 last->child_mtu_cached = mtu;
1193 EXPORT_SYMBOL(xfrm_bundle_ok);
1195 /* Well... that's _TASK_. We need to scan through transformation
1196 * list and figure out what mss tcp should generate in order to
1197 * final datagram fit to mtu. Mama mia... :-)
1199 * Apparently, some easy way exists, but we used to choose the most
1200 * bizarre ones. :-) So, raising Kalashnikov... tra-ta-ta.
1202 * Consider this function as something like dark humour. :-)
1204 static int xfrm_get_mss(struct dst_entry *dst, u32 mtu)
1206 int res = mtu - dst->header_len;
1209 struct dst_entry *d = dst;
1213 struct xfrm_state *x = d->xfrm;
1215 spin_lock_bh(&x->lock);
1216 if (x->km.state == XFRM_STATE_VALID &&
1217 x->type && x->type->get_max_size)
1218 m = x->type->get_max_size(d->xfrm, m);
1220 m += x->props.header_len;
1221 spin_unlock_bh(&x->lock);
1223 } while ((d = d->child) != NULL);
1232 return res + dst->header_len;
1235 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1238 if (unlikely(afinfo == NULL))
1240 if (unlikely(afinfo->family >= NPROTO))
1241 return -EAFNOSUPPORT;
1242 write_lock(&xfrm_policy_afinfo_lock);
1243 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1246 struct dst_ops *dst_ops = afinfo->dst_ops;
1247 if (likely(dst_ops->kmem_cachep == NULL))
1248 dst_ops->kmem_cachep = xfrm_dst_cache;
1249 if (likely(dst_ops->check == NULL))
1250 dst_ops->check = xfrm_dst_check;
1251 if (likely(dst_ops->negative_advice == NULL))
1252 dst_ops->negative_advice = xfrm_negative_advice;
1253 if (likely(dst_ops->link_failure == NULL))
1254 dst_ops->link_failure = xfrm_link_failure;
1255 if (likely(dst_ops->get_mss == NULL))
1256 dst_ops->get_mss = xfrm_get_mss;
1257 if (likely(afinfo->garbage_collect == NULL))
1258 afinfo->garbage_collect = __xfrm_garbage_collect;
1259 xfrm_policy_afinfo[afinfo->family] = afinfo;
1261 write_unlock(&xfrm_policy_afinfo_lock);
1264 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1266 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1269 if (unlikely(afinfo == NULL))
1271 if (unlikely(afinfo->family >= NPROTO))
1272 return -EAFNOSUPPORT;
1273 write_lock(&xfrm_policy_afinfo_lock);
1274 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1275 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1278 struct dst_ops *dst_ops = afinfo->dst_ops;
1279 xfrm_policy_afinfo[afinfo->family] = NULL;
1280 dst_ops->kmem_cachep = NULL;
1281 dst_ops->check = NULL;
1282 dst_ops->negative_advice = NULL;
1283 dst_ops->link_failure = NULL;
1284 dst_ops->get_mss = NULL;
1285 afinfo->garbage_collect = NULL;
1288 write_unlock(&xfrm_policy_afinfo_lock);
1291 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1293 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1295 struct xfrm_policy_afinfo *afinfo;
1296 if (unlikely(family >= NPROTO))
1298 read_lock(&xfrm_policy_afinfo_lock);
1299 afinfo = xfrm_policy_afinfo[family];
1300 if (likely(afinfo != NULL))
1301 read_lock(&afinfo->lock);
1302 read_unlock(&xfrm_policy_afinfo_lock);
1306 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1308 if (unlikely(afinfo == NULL))
1310 read_unlock(&afinfo->lock);
1313 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1317 xfrm_flush_bundles();
1322 static struct notifier_block xfrm_dev_notifier = {
1328 static void __init xfrm_policy_init(void)
1330 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1331 sizeof(struct xfrm_dst),
1332 0, SLAB_HWCACHE_ALIGN,
1334 if (!xfrm_dst_cache)
1335 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1337 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1338 register_netdevice_notifier(&xfrm_dev_notifier);
1341 void __init xfrm_init(void)