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
16 #include <linux/config.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>
27 DECLARE_MUTEX(xfrm_cfg_sem);
29 static rwlock_t xfrm_policy_lock = RW_LOCK_UNLOCKED;
31 struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2];
33 static rwlock_t xfrm_policy_afinfo_lock = RW_LOCK_UNLOCKED;
34 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
36 kmem_cache_t *xfrm_dst_cache;
38 static struct work_struct xfrm_policy_gc_work;
39 static struct list_head xfrm_policy_gc_list =
40 LIST_HEAD_INIT(xfrm_policy_gc_list);
41 static spinlock_t xfrm_policy_gc_lock = SPIN_LOCK_UNLOCKED;
43 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
45 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
46 struct xfrm_type_map *typemap;
49 if (unlikely(afinfo == NULL))
51 typemap = afinfo->type_map;
53 write_lock(&typemap->lock);
54 if (likely(typemap->map[type->proto] == NULL))
55 typemap->map[type->proto] = type;
58 write_unlock(&typemap->lock);
59 xfrm_policy_put_afinfo(afinfo);
63 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
65 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
66 struct xfrm_type_map *typemap;
69 if (unlikely(afinfo == NULL))
71 typemap = afinfo->type_map;
73 write_lock(&typemap->lock);
74 if (unlikely(typemap->map[type->proto] != type))
77 typemap->map[type->proto] = NULL;
78 write_unlock(&typemap->lock);
79 xfrm_policy_put_afinfo(afinfo);
83 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
85 struct xfrm_policy_afinfo *afinfo;
86 struct xfrm_type_map *typemap;
87 struct xfrm_type *type;
88 int modload_attempted = 0;
91 afinfo = xfrm_policy_get_afinfo(family);
92 if (unlikely(afinfo == NULL))
94 typemap = afinfo->type_map;
96 read_lock(&typemap->lock);
97 type = typemap->map[proto];
98 if (unlikely(type && !try_module_get(type->owner)))
100 read_unlock(&typemap->lock);
101 if (!type && !modload_attempted) {
102 xfrm_policy_put_afinfo(afinfo);
103 request_module("xfrm-type-%d-%d",
104 (int) family, (int) proto);
105 modload_attempted = 1;
109 xfrm_policy_put_afinfo(afinfo);
113 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
114 unsigned short family)
116 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
119 if (unlikely(afinfo == NULL))
120 return -EAFNOSUPPORT;
122 if (likely(afinfo->dst_lookup != NULL))
123 err = afinfo->dst_lookup(dst, fl);
126 xfrm_policy_put_afinfo(afinfo);
130 void xfrm_put_type(struct xfrm_type *type)
132 module_put(type->owner);
135 static inline unsigned long make_jiffies(long secs)
137 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
138 return MAX_SCHEDULE_TIMEOUT-1;
143 static void xfrm_policy_timer(unsigned long data)
145 struct xfrm_policy *xp = (struct xfrm_policy*)data;
146 unsigned long now = (unsigned long)xtime.tv_sec;
147 long next = LONG_MAX;
156 if (xp->lft.hard_add_expires_seconds) {
157 long tmo = xp->lft.hard_add_expires_seconds +
158 xp->curlft.add_time - now;
164 if (xp->lft.hard_use_expires_seconds) {
165 long tmo = xp->lft.hard_use_expires_seconds +
166 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
172 if (xp->lft.soft_add_expires_seconds) {
173 long tmo = xp->lft.soft_add_expires_seconds +
174 xp->curlft.add_time - now;
177 tmo = XFRM_KM_TIMEOUT;
182 if (xp->lft.soft_use_expires_seconds) {
183 long tmo = xp->lft.soft_use_expires_seconds +
184 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
187 tmo = XFRM_KM_TIMEOUT;
194 km_policy_expired(xp, dir, 0);
195 if (next != LONG_MAX &&
196 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
204 km_policy_expired(xp, dir, 1);
205 xfrm_policy_delete(xp, dir);
210 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
214 struct xfrm_policy *xfrm_policy_alloc(int gfp)
216 struct xfrm_policy *policy;
218 policy = kmalloc(sizeof(struct xfrm_policy), gfp);
221 memset(policy, 0, sizeof(struct xfrm_policy));
222 atomic_set(&policy->refcnt, 1);
223 policy->lock = RW_LOCK_UNLOCKED;
224 init_timer(&policy->timer);
225 policy->timer.data = (unsigned long)policy;
226 policy->timer.function = xfrm_policy_timer;
231 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
233 void __xfrm_policy_destroy(struct xfrm_policy *policy)
241 if (del_timer(&policy->timer))
247 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
249 struct dst_entry *dst;
251 while ((dst = policy->bundles) != NULL) {
252 policy->bundles = dst->next;
256 if (del_timer(&policy->timer))
257 atomic_dec(&policy->refcnt);
259 if (atomic_read(&policy->refcnt) > 1)
262 xfrm_pol_put(policy);
265 static void xfrm_policy_gc_task(void *data)
267 struct xfrm_policy *policy;
268 struct list_head *entry, *tmp;
269 struct list_head gc_list = LIST_HEAD_INIT(gc_list);
271 spin_lock_bh(&xfrm_policy_gc_lock);
272 list_splice_init(&xfrm_policy_gc_list, &gc_list);
273 spin_unlock_bh(&xfrm_policy_gc_lock);
275 list_for_each_safe(entry, tmp, &gc_list) {
276 policy = list_entry(entry, struct xfrm_policy, list);
277 xfrm_policy_gc_kill(policy);
281 /* Rule must be locked. Release descentant resources, announce
282 * entry dead. The rule must be unlinked from lists to the moment.
285 void xfrm_policy_kill(struct xfrm_policy *policy)
287 write_lock_bh(&policy->lock);
293 spin_lock(&xfrm_policy_gc_lock);
294 list_add(&policy->list, &xfrm_policy_gc_list);
295 spin_unlock(&xfrm_policy_gc_lock);
296 schedule_work(&xfrm_policy_gc_work);
299 write_unlock_bh(&policy->lock);
302 /* Generate new index... KAME seems to generate them ordered by cost
303 * of an absolute inpredictability of ordering of rules. This will not pass. */
304 static u32 xfrm_gen_index(int dir)
307 struct xfrm_policy *p;
308 static u32 idx_generator;
311 idx = (idx_generator | dir);
315 for (p = xfrm_policy_list[dir]; p; p = p->next) {
324 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
326 struct xfrm_policy *pol, **p;
327 struct xfrm_policy *delpol = NULL;
328 struct xfrm_policy **newpos = NULL;
330 write_lock_bh(&xfrm_policy_lock);
331 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
332 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0) {
334 write_unlock_bh(&xfrm_policy_lock);
339 if (policy->priority > pol->priority)
341 } else if (policy->priority >= pol->priority)
350 xfrm_pol_hold(policy);
353 atomic_inc(&flow_cache_genid);
354 policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
355 policy->curlft.add_time = (unsigned long)xtime.tv_sec;
356 policy->curlft.use_time = 0;
357 if (!mod_timer(&policy->timer, jiffies + HZ))
358 xfrm_pol_hold(policy);
359 write_unlock_bh(&xfrm_policy_lock);
362 xfrm_policy_kill(delpol);
367 struct xfrm_policy *xfrm_policy_bysel(int dir, struct xfrm_selector *sel,
370 struct xfrm_policy *pol, **p;
372 write_lock_bh(&xfrm_policy_lock);
373 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
374 if (memcmp(sel, &pol->selector, sizeof(*sel)) == 0) {
381 write_unlock_bh(&xfrm_policy_lock);
384 atomic_inc(&flow_cache_genid);
385 xfrm_policy_kill(pol);
390 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
392 struct xfrm_policy *pol, **p;
394 write_lock_bh(&xfrm_policy_lock);
395 for (p = &xfrm_policy_list[id & 7]; (pol=*p)!=NULL; p = &pol->next) {
396 if (pol->index == id) {
403 write_unlock_bh(&xfrm_policy_lock);
406 atomic_inc(&flow_cache_genid);
407 xfrm_policy_kill(pol);
412 void xfrm_policy_flush(void)
414 struct xfrm_policy *xp;
417 write_lock_bh(&xfrm_policy_lock);
418 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
419 while ((xp = xfrm_policy_list[dir]) != NULL) {
420 xfrm_policy_list[dir] = xp->next;
421 write_unlock_bh(&xfrm_policy_lock);
423 xfrm_policy_kill(xp);
425 write_lock_bh(&xfrm_policy_lock);
428 atomic_inc(&flow_cache_genid);
429 write_unlock_bh(&xfrm_policy_lock);
432 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
435 struct xfrm_policy *xp;
440 read_lock_bh(&xfrm_policy_lock);
441 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
442 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
451 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
452 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
453 error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
460 read_unlock_bh(&xfrm_policy_lock);
465 /* Find policy to apply to this flow. */
467 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
468 void **objp, atomic_t **obj_refp)
470 struct xfrm_policy *pol;
472 read_lock_bh(&xfrm_policy_lock);
473 for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
474 struct xfrm_selector *sel = &pol->selector;
477 if (pol->family != family)
480 match = xfrm_selector_match(sel, fl, family);
486 read_unlock_bh(&xfrm_policy_lock);
487 if ((*objp = (void *) pol) != NULL)
488 *obj_refp = &pol->refcnt;
491 struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
493 struct xfrm_policy *pol;
495 read_lock_bh(&xfrm_policy_lock);
496 if ((pol = sk->sk_policy[dir]) != NULL) {
497 int match = xfrm_selector_match(&pol->selector, fl,
504 read_unlock_bh(&xfrm_policy_lock);
508 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
510 pol->next = xfrm_policy_list[dir];
511 xfrm_policy_list[dir] = pol;
515 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
518 struct xfrm_policy **polp;
520 for (polp = &xfrm_policy_list[dir];
521 *polp != NULL; polp = &(*polp)->next) {
530 void xfrm_policy_delete(struct xfrm_policy *pol, int dir)
532 write_lock_bh(&xfrm_policy_lock);
533 pol = __xfrm_policy_unlink(pol, dir);
534 write_unlock_bh(&xfrm_policy_lock);
536 xfrm_policy_kill(pol);
539 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
541 struct xfrm_policy *old_pol;
543 write_lock_bh(&xfrm_policy_lock);
544 old_pol = sk->sk_policy[dir];
545 sk->sk_policy[dir] = pol;
547 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
548 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
549 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
552 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
553 write_unlock_bh(&xfrm_policy_lock);
556 xfrm_policy_kill(old_pol);
561 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
563 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
566 newp->selector = old->selector;
567 newp->lft = old->lft;
568 newp->curlft = old->curlft;
569 newp->action = old->action;
570 newp->flags = old->flags;
571 newp->xfrm_nr = old->xfrm_nr;
572 newp->index = old->index;
573 memcpy(newp->xfrm_vec, old->xfrm_vec,
574 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
575 write_lock_bh(&xfrm_policy_lock);
576 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
577 write_unlock_bh(&xfrm_policy_lock);
583 int __xfrm_sk_clone_policy(struct sock *sk)
585 struct xfrm_policy *p0 = sk->sk_policy[0],
586 *p1 = sk->sk_policy[1];
588 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
589 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
591 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
596 /* Resolve list of templates for the flow, given policy. */
599 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
600 struct xfrm_state **xfrm,
601 unsigned short family)
605 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
606 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
608 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
609 struct xfrm_state *x;
610 xfrm_address_t *remote = daddr;
611 xfrm_address_t *local = saddr;
612 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
615 remote = &tmpl->id.daddr;
616 local = &tmpl->saddr;
619 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
621 if (x && x->km.state == XFRM_STATE_VALID) {
628 error = (x->km.state == XFRM_STATE_ERROR ?
639 for (nx--; nx>=0; nx--)
640 xfrm_state_put(xfrm[nx]);
644 /* Check that the bundle accepts the flow and its components are
648 static struct dst_entry *
649 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
652 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
653 if (unlikely(afinfo == NULL))
654 return ERR_PTR(-EINVAL);
655 x = afinfo->find_bundle(fl, policy);
656 xfrm_policy_put_afinfo(afinfo);
660 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
661 * all the metrics... Shortly, bundle a bundle.
665 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
666 struct flowi *fl, struct dst_entry **dst_p,
667 unsigned short family)
670 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
671 if (unlikely(afinfo == NULL))
673 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
674 xfrm_policy_put_afinfo(afinfo);
678 static inline int policy_to_flow_dir(int dir)
680 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
681 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
682 XFRM_POLICY_FWD == FLOW_DIR_FWD)
688 case XFRM_POLICY_OUT:
690 case XFRM_POLICY_FWD:
695 static int stale_bundle(struct dst_entry *dst);
697 /* Main function: finds/creates a bundle for given flow.
699 * At the moment we eat a raw IP route. Mostly to speed up lookups
700 * on interfaces with disabled IPsec.
702 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
703 struct sock *sk, int flags)
705 struct xfrm_policy *policy;
706 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
707 struct rtable *rt = (struct rtable*)*dst_p;
708 struct dst_entry *dst;
712 u16 family = (*dst_p)->ops->family;
717 fl->fl4_src = rt->rt_src;
719 fl->fl4_dst = rt->rt_dst;
721 /* Still not clear... */
727 genid = atomic_read(&flow_cache_genid);
729 if (sk && sk->sk_policy[1])
730 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
733 /* To accelerate a bit... */
734 if ((rt->u.dst.flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
737 policy = flow_cache_lookup(fl, family,
738 policy_to_flow_dir(XFRM_POLICY_OUT),
745 policy->curlft.use_time = (unsigned long)xtime.tv_sec;
747 switch (policy->action) {
748 case XFRM_POLICY_BLOCK:
749 /* Prohibit the flow */
750 xfrm_pol_put(policy);
753 case XFRM_POLICY_ALLOW:
754 if (policy->xfrm_nr == 0) {
755 /* Flow passes not transformed. */
756 xfrm_pol_put(policy);
760 /* Try to find matching bundle.
762 * LATER: help from flow cache. It is optional, this
763 * is required only for output policy.
765 dst = xfrm_find_bundle(fl, policy, family);
767 xfrm_pol_put(policy);
774 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
776 if (unlikely(nx<0)) {
778 if (err == -EAGAIN && flags) {
779 DECLARE_WAITQUEUE(wait, current);
781 add_wait_queue(&km_waitq, &wait);
782 set_current_state(TASK_INTERRUPTIBLE);
784 set_current_state(TASK_RUNNING);
785 remove_wait_queue(&km_waitq, &wait);
787 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
789 if (nx == -EAGAIN && signal_pending(current)) {
794 genid != atomic_read(&flow_cache_genid)) {
795 xfrm_pol_put(policy);
804 /* Flow passes not transformed. */
805 xfrm_pol_put(policy);
810 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
815 xfrm_state_put(xfrm[i]);
819 write_lock_bh(&policy->lock);
820 if (unlikely(policy->dead || stale_bundle(dst))) {
821 /* Wow! While we worked on resolving, this
822 * policy has gone. Retry. It is not paranoia,
823 * we just cannot enlist new bundle to dead object.
824 * We can't enlist stable bundles either.
826 write_unlock_bh(&policy->lock);
828 xfrm_pol_put(policy);
833 dst->next = policy->bundles;
834 policy->bundles = dst;
836 write_unlock_bh(&policy->lock);
840 xfrm_pol_put(policy);
845 xfrm_pol_put(policy);
850 /* When skb is transformed back to its "native" form, we have to
851 * check policy restrictions. At the moment we make this in maximally
852 * stupid way. Shame on me. :-) Of course, connected sockets must
853 * have policy cached at them.
857 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
858 unsigned short family)
860 if (xfrm_state_kern(x))
861 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
862 return x->id.proto == tmpl->id.proto &&
863 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
864 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
865 x->props.mode == tmpl->mode &&
866 (tmpl->aalgos & (1<<x->props.aalgo)) &&
867 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
871 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
872 unsigned short family)
876 if (tmpl->optional) {
881 for (; idx < sp->len; idx++) {
882 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
884 if (sp->x[idx].xvec->props.mode)
891 _decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
893 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
895 if (unlikely(afinfo == NULL))
896 return -EAFNOSUPPORT;
898 afinfo->decode_session(skb, fl);
899 xfrm_policy_put_afinfo(afinfo);
903 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
904 unsigned short family)
906 struct xfrm_policy *pol;
909 if (_decode_session(skb, &fl, family) < 0)
912 /* First, check used SA against their selectors. */
916 for (i=skb->sp->len-1; i>=0; i--) {
917 struct sec_decap_state *xvec = &(skb->sp->x[i]);
918 if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
921 /* If there is a post_input processor, try running it */
922 if (xvec->xvec->type->post_input &&
923 (xvec->xvec->type->post_input)(xvec->xvec,
931 if (sk && sk->sk_policy[dir])
932 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
935 pol = flow_cache_lookup(&fl, family,
936 policy_to_flow_dir(dir),
942 pol->curlft.use_time = (unsigned long)xtime.tv_sec;
944 if (pol->action == XFRM_POLICY_ALLOW) {
946 static struct sec_path dummy;
949 if ((sp = skb->sp) == NULL)
952 /* For each tunnel xfrm, find the first matching tmpl.
953 * For each tmpl before that, find corresponding xfrm.
954 * Order is _important_. Later we will implement
955 * some barriers, but at the moment barriers
956 * are implied between each two transformations.
958 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
959 k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
964 for (; k < sp->len; k++) {
965 if (sp->x[k].xvec->props.mode)
978 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
982 if (_decode_session(skb, &fl, family) < 0)
985 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
988 /* Optimize later using cookies and generation ids. */
990 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
992 if (!stale_bundle(dst))
999 static int stale_bundle(struct dst_entry *dst)
1001 struct dst_entry *child = dst;
1004 if (child->obsolete > 0 ||
1005 (child->dev && !netif_running(child->dev)) ||
1006 (child->xfrm && child->xfrm->km.state != XFRM_STATE_VALID)) {
1009 child = child->child;
1015 static void xfrm_dst_destroy(struct dst_entry *dst)
1017 xfrm_state_put(dst->xfrm);
1021 static void xfrm_link_failure(struct sk_buff *skb)
1023 /* Impossible. Such dst must be popped before reaches point of failure. */
1027 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1030 if (dst->obsolete) {
1038 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1041 struct xfrm_policy *pol;
1042 struct dst_entry *dst, **dstp, *gc_list = NULL;
1044 read_lock_bh(&xfrm_policy_lock);
1045 for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1046 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1047 write_lock(&pol->lock);
1048 dstp = &pol->bundles;
1049 while ((dst=*dstp) != NULL) {
1052 dst->next = gc_list;
1058 write_unlock(&pol->lock);
1061 read_unlock_bh(&xfrm_policy_lock);
1065 gc_list = dst->next;
1070 static int unused_bundle(struct dst_entry *dst)
1072 return !atomic_read(&dst->__refcnt);
1075 static void __xfrm_garbage_collect(void)
1077 xfrm_prune_bundles(unused_bundle);
1080 int xfrm_flush_bundles(void)
1082 xfrm_prune_bundles(stale_bundle);
1086 /* Well... that's _TASK_. We need to scan through transformation
1087 * list and figure out what mss tcp should generate in order to
1088 * final datagram fit to mtu. Mama mia... :-)
1090 * Apparently, some easy way exists, but we used to choose the most
1091 * bizarre ones. :-) So, raising Kalashnikov... tra-ta-ta.
1093 * Consider this function as something like dark humour. :-)
1095 static int xfrm_get_mss(struct dst_entry *dst, u32 mtu)
1097 int res = mtu - dst->header_len;
1100 struct dst_entry *d = dst;
1104 struct xfrm_state *x = d->xfrm;
1106 spin_lock_bh(&x->lock);
1107 if (x->km.state == XFRM_STATE_VALID &&
1108 x->type && x->type->get_max_size)
1109 m = x->type->get_max_size(d->xfrm, m);
1111 m += x->props.header_len;
1112 spin_unlock_bh(&x->lock);
1114 } while ((d = d->child) != NULL);
1123 return res + dst->header_len;
1126 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1129 if (unlikely(afinfo == NULL))
1131 if (unlikely(afinfo->family >= NPROTO))
1132 return -EAFNOSUPPORT;
1133 write_lock(&xfrm_policy_afinfo_lock);
1134 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1137 struct dst_ops *dst_ops = afinfo->dst_ops;
1138 if (likely(dst_ops->kmem_cachep == NULL))
1139 dst_ops->kmem_cachep = xfrm_dst_cache;
1140 if (likely(dst_ops->check == NULL))
1141 dst_ops->check = xfrm_dst_check;
1142 if (likely(dst_ops->destroy == NULL))
1143 dst_ops->destroy = xfrm_dst_destroy;
1144 if (likely(dst_ops->negative_advice == NULL))
1145 dst_ops->negative_advice = xfrm_negative_advice;
1146 if (likely(dst_ops->link_failure == NULL))
1147 dst_ops->link_failure = xfrm_link_failure;
1148 if (likely(dst_ops->get_mss == NULL))
1149 dst_ops->get_mss = xfrm_get_mss;
1150 if (likely(afinfo->garbage_collect == NULL))
1151 afinfo->garbage_collect = __xfrm_garbage_collect;
1152 xfrm_policy_afinfo[afinfo->family] = afinfo;
1154 write_unlock(&xfrm_policy_afinfo_lock);
1158 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1161 if (unlikely(afinfo == NULL))
1163 if (unlikely(afinfo->family >= NPROTO))
1164 return -EAFNOSUPPORT;
1165 write_lock(&xfrm_policy_afinfo_lock);
1166 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1167 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1170 struct dst_ops *dst_ops = afinfo->dst_ops;
1171 xfrm_policy_afinfo[afinfo->family] = NULL;
1172 dst_ops->kmem_cachep = NULL;
1173 dst_ops->check = NULL;
1174 dst_ops->destroy = NULL;
1175 dst_ops->negative_advice = NULL;
1176 dst_ops->link_failure = NULL;
1177 dst_ops->get_mss = NULL;
1178 afinfo->garbage_collect = NULL;
1181 write_unlock(&xfrm_policy_afinfo_lock);
1185 struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1187 struct xfrm_policy_afinfo *afinfo;
1188 if (unlikely(family >= NPROTO))
1190 read_lock(&xfrm_policy_afinfo_lock);
1191 afinfo = xfrm_policy_afinfo[family];
1192 if (likely(afinfo != NULL))
1193 read_lock(&afinfo->lock);
1194 read_unlock(&xfrm_policy_afinfo_lock);
1198 void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1200 if (unlikely(afinfo == NULL))
1202 read_unlock(&afinfo->lock);
1205 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1209 xfrm_flush_bundles();
1214 struct notifier_block xfrm_dev_notifier = {
1220 void __init xfrm_policy_init(void)
1222 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1223 sizeof(struct xfrm_dst),
1224 0, SLAB_HWCACHE_ALIGN,
1226 if (!xfrm_dst_cache)
1227 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1229 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1230 register_netdevice_notifier(&xfrm_dev_notifier);
1233 void __init xfrm_init(void)