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>
24 #include <linux/netfilter.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
)
250 BUG_ON(!policy
->dead
);
252 BUG_ON(policy
->bundles
);
254 if (del_timer(&policy
->timer
))
257 security_xfrm_policy_free(policy
);
260 EXPORT_SYMBOL(__xfrm_policy_destroy
);
262 static void xfrm_policy_gc_kill(struct xfrm_policy
*policy
)
264 struct dst_entry
*dst
;
266 while ((dst
= policy
->bundles
) != NULL
) {
267 policy
->bundles
= dst
->next
;
271 if (del_timer(&policy
->timer
))
272 atomic_dec(&policy
->refcnt
);
274 if (atomic_read(&policy
->refcnt
) > 1)
277 xfrm_pol_put(policy
);
280 static void xfrm_policy_gc_task(void *data
)
282 struct xfrm_policy
*policy
;
283 struct list_head
*entry
, *tmp
;
284 struct list_head gc_list
= LIST_HEAD_INIT(gc_list
);
286 spin_lock_bh(&xfrm_policy_gc_lock
);
287 list_splice_init(&xfrm_policy_gc_list
, &gc_list
);
288 spin_unlock_bh(&xfrm_policy_gc_lock
);
290 list_for_each_safe(entry
, tmp
, &gc_list
) {
291 policy
= list_entry(entry
, struct xfrm_policy
, list
);
292 xfrm_policy_gc_kill(policy
);
296 /* Rule must be locked. Release descentant resources, announce
297 * entry dead. The rule must be unlinked from lists to the moment.
300 static void xfrm_policy_kill(struct xfrm_policy
*policy
)
304 write_lock_bh(&policy
->lock
);
307 write_unlock_bh(&policy
->lock
);
309 if (unlikely(dead
)) {
314 spin_lock(&xfrm_policy_gc_lock
);
315 list_add(&policy
->list
, &xfrm_policy_gc_list
);
316 spin_unlock(&xfrm_policy_gc_lock
);
318 schedule_work(&xfrm_policy_gc_work
);
321 /* Generate new index... KAME seems to generate them ordered by cost
322 * of an absolute inpredictability of ordering of rules. This will not pass. */
323 static u32
xfrm_gen_index(int dir
)
326 struct xfrm_policy
*p
;
327 static u32 idx_generator
;
330 idx
= (idx_generator
| dir
);
334 for (p
= xfrm_policy_list
[dir
]; p
; p
= p
->next
) {
343 int xfrm_policy_insert(int dir
, struct xfrm_policy
*policy
, int excl
)
345 struct xfrm_policy
*pol
, **p
;
346 struct xfrm_policy
*delpol
= NULL
;
347 struct xfrm_policy
**newpos
= NULL
;
348 struct dst_entry
*gc_list
;
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 &&
353 xfrm_sec_ctx_match(pol
->security
, policy
->security
)) {
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
);
388 read_lock_bh(&xfrm_policy_lock
);
390 for (policy
= policy
->next
; policy
; policy
= policy
->next
) {
391 struct dst_entry
*dst
;
393 write_lock(&policy
->lock
);
394 dst
= policy
->bundles
;
396 struct dst_entry
*tail
= dst
;
399 tail
->next
= gc_list
;
402 policy
->bundles
= NULL
;
404 write_unlock(&policy
->lock
);
406 read_unlock_bh(&xfrm_policy_lock
);
409 struct dst_entry
*dst
= gc_list
;
417 EXPORT_SYMBOL(xfrm_policy_insert
);
419 struct xfrm_policy
*xfrm_policy_bysel_ctx(int dir
, struct xfrm_selector
*sel
,
420 struct xfrm_sec_ctx
*ctx
, int delete)
422 struct xfrm_policy
*pol
, **p
;
424 write_lock_bh(&xfrm_policy_lock
);
425 for (p
= &xfrm_policy_list
[dir
]; (pol
=*p
)!=NULL
; p
= &pol
->next
) {
426 if ((memcmp(sel
, &pol
->selector
, sizeof(*sel
)) == 0) &&
427 (xfrm_sec_ctx_match(ctx
, pol
->security
))) {
434 write_unlock_bh(&xfrm_policy_lock
);
437 atomic_inc(&flow_cache_genid
);
438 xfrm_policy_kill(pol
);
442 EXPORT_SYMBOL(xfrm_policy_bysel_ctx
);
444 struct xfrm_policy
*xfrm_policy_byid(int dir
, u32 id
, int delete)
446 struct xfrm_policy
*pol
, **p
;
448 write_lock_bh(&xfrm_policy_lock
);
449 for (p
= &xfrm_policy_list
[dir
]; (pol
=*p
)!=NULL
; p
= &pol
->next
) {
450 if (pol
->index
== id
) {
457 write_unlock_bh(&xfrm_policy_lock
);
460 atomic_inc(&flow_cache_genid
);
461 xfrm_policy_kill(pol
);
465 EXPORT_SYMBOL(xfrm_policy_byid
);
467 void xfrm_policy_flush(void)
469 struct xfrm_policy
*xp
;
472 write_lock_bh(&xfrm_policy_lock
);
473 for (dir
= 0; dir
< XFRM_POLICY_MAX
; dir
++) {
474 while ((xp
= xfrm_policy_list
[dir
]) != NULL
) {
475 xfrm_policy_list
[dir
] = xp
->next
;
476 write_unlock_bh(&xfrm_policy_lock
);
478 xfrm_policy_kill(xp
);
480 write_lock_bh(&xfrm_policy_lock
);
483 atomic_inc(&flow_cache_genid
);
484 write_unlock_bh(&xfrm_policy_lock
);
486 EXPORT_SYMBOL(xfrm_policy_flush
);
488 int xfrm_policy_walk(int (*func
)(struct xfrm_policy
*, int, int, void*),
491 struct xfrm_policy
*xp
;
496 read_lock_bh(&xfrm_policy_lock
);
497 for (dir
= 0; dir
< 2*XFRM_POLICY_MAX
; dir
++) {
498 for (xp
= xfrm_policy_list
[dir
]; xp
; xp
= xp
->next
)
507 for (dir
= 0; dir
< 2*XFRM_POLICY_MAX
; dir
++) {
508 for (xp
= xfrm_policy_list
[dir
]; xp
; xp
= xp
->next
) {
509 error
= func(xp
, dir
%XFRM_POLICY_MAX
, --count
, data
);
516 read_unlock_bh(&xfrm_policy_lock
);
519 EXPORT_SYMBOL(xfrm_policy_walk
);
521 /* Find policy to apply to this flow. */
523 static void xfrm_policy_lookup(struct flowi
*fl
, u32 sk_sid
, u16 family
, u8 dir
,
524 void **objp
, atomic_t
**obj_refp
)
526 struct xfrm_policy
*pol
;
528 read_lock_bh(&xfrm_policy_lock
);
529 for (pol
= xfrm_policy_list
[dir
]; pol
; pol
= pol
->next
) {
530 struct xfrm_selector
*sel
= &pol
->selector
;
533 if (pol
->family
!= family
)
536 match
= xfrm_selector_match(sel
, fl
, family
);
539 if (!security_xfrm_policy_lookup(pol
, sk_sid
, dir
)) {
545 read_unlock_bh(&xfrm_policy_lock
);
546 if ((*objp
= (void *) pol
) != NULL
)
547 *obj_refp
= &pol
->refcnt
;
550 static inline int policy_to_flow_dir(int dir
)
552 if (XFRM_POLICY_IN
== FLOW_DIR_IN
&&
553 XFRM_POLICY_OUT
== FLOW_DIR_OUT
&&
554 XFRM_POLICY_FWD
== FLOW_DIR_FWD
)
560 case XFRM_POLICY_OUT
:
562 case XFRM_POLICY_FWD
:
567 static struct xfrm_policy
*xfrm_sk_policy_lookup(struct sock
*sk
, int dir
, struct flowi
*fl
, u32 sk_sid
)
569 struct xfrm_policy
*pol
;
571 read_lock_bh(&xfrm_policy_lock
);
572 if ((pol
= sk
->sk_policy
[dir
]) != NULL
) {
573 int match
= xfrm_selector_match(&pol
->selector
, fl
,
578 err
= security_xfrm_policy_lookup(pol
, sk_sid
, policy_to_flow_dir(dir
));
585 read_unlock_bh(&xfrm_policy_lock
);
589 static void __xfrm_policy_link(struct xfrm_policy
*pol
, int dir
)
591 pol
->next
= xfrm_policy_list
[dir
];
592 xfrm_policy_list
[dir
] = pol
;
596 static struct xfrm_policy
*__xfrm_policy_unlink(struct xfrm_policy
*pol
,
599 struct xfrm_policy
**polp
;
601 for (polp
= &xfrm_policy_list
[dir
];
602 *polp
!= NULL
; polp
= &(*polp
)->next
) {
611 int xfrm_policy_delete(struct xfrm_policy
*pol
, int dir
)
613 write_lock_bh(&xfrm_policy_lock
);
614 pol
= __xfrm_policy_unlink(pol
, dir
);
615 write_unlock_bh(&xfrm_policy_lock
);
617 if (dir
< XFRM_POLICY_MAX
)
618 atomic_inc(&flow_cache_genid
);
619 xfrm_policy_kill(pol
);
625 int xfrm_sk_policy_insert(struct sock
*sk
, int dir
, struct xfrm_policy
*pol
)
627 struct xfrm_policy
*old_pol
;
629 write_lock_bh(&xfrm_policy_lock
);
630 old_pol
= sk
->sk_policy
[dir
];
631 sk
->sk_policy
[dir
] = pol
;
633 pol
->curlft
.add_time
= (unsigned long)xtime
.tv_sec
;
634 pol
->index
= xfrm_gen_index(XFRM_POLICY_MAX
+dir
);
635 __xfrm_policy_link(pol
, XFRM_POLICY_MAX
+dir
);
638 __xfrm_policy_unlink(old_pol
, XFRM_POLICY_MAX
+dir
);
639 write_unlock_bh(&xfrm_policy_lock
);
642 xfrm_policy_kill(old_pol
);
647 static struct xfrm_policy
*clone_policy(struct xfrm_policy
*old
, int dir
)
649 struct xfrm_policy
*newp
= xfrm_policy_alloc(GFP_ATOMIC
);
652 newp
->selector
= old
->selector
;
653 if (security_xfrm_policy_clone(old
, newp
)) {
655 return NULL
; /* ENOMEM */
657 newp
->lft
= old
->lft
;
658 newp
->curlft
= old
->curlft
;
659 newp
->action
= old
->action
;
660 newp
->flags
= old
->flags
;
661 newp
->xfrm_nr
= old
->xfrm_nr
;
662 newp
->index
= old
->index
;
663 memcpy(newp
->xfrm_vec
, old
->xfrm_vec
,
664 newp
->xfrm_nr
*sizeof(struct xfrm_tmpl
));
665 write_lock_bh(&xfrm_policy_lock
);
666 __xfrm_policy_link(newp
, XFRM_POLICY_MAX
+dir
);
667 write_unlock_bh(&xfrm_policy_lock
);
673 int __xfrm_sk_clone_policy(struct sock
*sk
)
675 struct xfrm_policy
*p0
= sk
->sk_policy
[0],
676 *p1
= sk
->sk_policy
[1];
678 sk
->sk_policy
[0] = sk
->sk_policy
[1] = NULL
;
679 if (p0
&& (sk
->sk_policy
[0] = clone_policy(p0
, 0)) == NULL
)
681 if (p1
&& (sk
->sk_policy
[1] = clone_policy(p1
, 1)) == NULL
)
686 /* Resolve list of templates for the flow, given policy. */
689 xfrm_tmpl_resolve(struct xfrm_policy
*policy
, struct flowi
*fl
,
690 struct xfrm_state
**xfrm
,
691 unsigned short family
)
695 xfrm_address_t
*daddr
= xfrm_flowi_daddr(fl
, family
);
696 xfrm_address_t
*saddr
= xfrm_flowi_saddr(fl
, family
);
698 for (nx
=0, i
= 0; i
< policy
->xfrm_nr
; i
++) {
699 struct xfrm_state
*x
;
700 xfrm_address_t
*remote
= daddr
;
701 xfrm_address_t
*local
= saddr
;
702 struct xfrm_tmpl
*tmpl
= &policy
->xfrm_vec
[i
];
705 remote
= &tmpl
->id
.daddr
;
706 local
= &tmpl
->saddr
;
709 x
= xfrm_state_find(remote
, local
, fl
, tmpl
, policy
, &error
, family
);
711 if (x
&& x
->km
.state
== XFRM_STATE_VALID
) {
718 error
= (x
->km
.state
== XFRM_STATE_ERROR
?
729 for (nx
--; nx
>=0; nx
--)
730 xfrm_state_put(xfrm
[nx
]);
734 /* Check that the bundle accepts the flow and its components are
738 static struct dst_entry
*
739 xfrm_find_bundle(struct flowi
*fl
, struct xfrm_policy
*policy
, unsigned short family
)
742 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
743 if (unlikely(afinfo
== NULL
))
744 return ERR_PTR(-EINVAL
);
745 x
= afinfo
->find_bundle(fl
, policy
);
746 xfrm_policy_put_afinfo(afinfo
);
750 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
751 * all the metrics... Shortly, bundle a bundle.
755 xfrm_bundle_create(struct xfrm_policy
*policy
, struct xfrm_state
**xfrm
, int nx
,
756 struct flowi
*fl
, struct dst_entry
**dst_p
,
757 unsigned short family
)
760 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
761 if (unlikely(afinfo
== NULL
))
763 err
= afinfo
->bundle_create(policy
, xfrm
, nx
, fl
, dst_p
);
764 xfrm_policy_put_afinfo(afinfo
);
769 static int stale_bundle(struct dst_entry
*dst
);
771 /* Main function: finds/creates a bundle for given flow.
773 * At the moment we eat a raw IP route. Mostly to speed up lookups
774 * on interfaces with disabled IPsec.
776 int xfrm_lookup(struct dst_entry
**dst_p
, struct flowi
*fl
,
777 struct sock
*sk
, int flags
)
779 struct xfrm_policy
*policy
;
780 struct xfrm_state
*xfrm
[XFRM_MAX_DEPTH
];
781 struct dst_entry
*dst
, *dst_orig
= *dst_p
;
786 u8 dir
= policy_to_flow_dir(XFRM_POLICY_OUT
);
787 u32 sk_sid
= security_sk_sid(sk
, fl
, dir
);
789 genid
= atomic_read(&flow_cache_genid
);
791 if (sk
&& sk
->sk_policy
[1])
792 policy
= xfrm_sk_policy_lookup(sk
, XFRM_POLICY_OUT
, fl
, sk_sid
);
795 /* To accelerate a bit... */
796 if ((dst_orig
->flags
& DST_NOXFRM
) || !xfrm_policy_list
[XFRM_POLICY_OUT
])
799 policy
= flow_cache_lookup(fl
, sk_sid
, dst_orig
->ops
->family
,
800 dir
, xfrm_policy_lookup
);
806 family
= dst_orig
->ops
->family
;
807 policy
->curlft
.use_time
= (unsigned long)xtime
.tv_sec
;
809 switch (policy
->action
) {
810 case XFRM_POLICY_BLOCK
:
811 /* Prohibit the flow */
815 case XFRM_POLICY_ALLOW
:
816 if (policy
->xfrm_nr
== 0) {
817 /* Flow passes not transformed. */
818 xfrm_pol_put(policy
);
822 /* Try to find matching bundle.
824 * LATER: help from flow cache. It is optional, this
825 * is required only for output policy.
827 dst
= xfrm_find_bundle(fl
, policy
, family
);
836 nx
= xfrm_tmpl_resolve(policy
, fl
, xfrm
, family
);
838 if (unlikely(nx
<0)) {
840 if (err
== -EAGAIN
&& flags
) {
841 DECLARE_WAITQUEUE(wait
, current
);
843 add_wait_queue(&km_waitq
, &wait
);
844 set_current_state(TASK_INTERRUPTIBLE
);
846 set_current_state(TASK_RUNNING
);
847 remove_wait_queue(&km_waitq
, &wait
);
849 nx
= xfrm_tmpl_resolve(policy
, fl
, xfrm
, family
);
851 if (nx
== -EAGAIN
&& signal_pending(current
)) {
856 genid
!= atomic_read(&flow_cache_genid
)) {
857 xfrm_pol_put(policy
);
866 /* Flow passes not transformed. */
867 xfrm_pol_put(policy
);
872 err
= xfrm_bundle_create(policy
, xfrm
, nx
, fl
, &dst
, family
);
877 xfrm_state_put(xfrm
[i
]);
881 write_lock_bh(&policy
->lock
);
882 if (unlikely(policy
->dead
|| stale_bundle(dst
))) {
883 /* Wow! While we worked on resolving, this
884 * policy has gone. Retry. It is not paranoia,
885 * we just cannot enlist new bundle to dead object.
886 * We can't enlist stable bundles either.
888 write_unlock_bh(&policy
->lock
);
895 dst
->next
= policy
->bundles
;
896 policy
->bundles
= dst
;
898 write_unlock_bh(&policy
->lock
);
901 dst_release(dst_orig
);
902 xfrm_pol_put(policy
);
906 dst_release(dst_orig
);
907 xfrm_pol_put(policy
);
911 EXPORT_SYMBOL(xfrm_lookup
);
913 /* When skb is transformed back to its "native" form, we have to
914 * check policy restrictions. At the moment we make this in maximally
915 * stupid way. Shame on me. :-) Of course, connected sockets must
916 * have policy cached at them.
920 xfrm_state_ok(struct xfrm_tmpl
*tmpl
, struct xfrm_state
*x
,
921 unsigned short family
)
923 if (xfrm_state_kern(x
))
924 return tmpl
->optional
&& !xfrm_state_addr_cmp(tmpl
, x
, family
);
925 return x
->id
.proto
== tmpl
->id
.proto
&&
926 (x
->id
.spi
== tmpl
->id
.spi
|| !tmpl
->id
.spi
) &&
927 (x
->props
.reqid
== tmpl
->reqid
|| !tmpl
->reqid
) &&
928 x
->props
.mode
== tmpl
->mode
&&
929 (tmpl
->aalgos
& (1<<x
->props
.aalgo
)) &&
930 !(x
->props
.mode
&& xfrm_state_addr_cmp(tmpl
, x
, family
));
934 xfrm_policy_ok(struct xfrm_tmpl
*tmpl
, struct sec_path
*sp
, int start
,
935 unsigned short family
)
939 if (tmpl
->optional
) {
944 for (; idx
< sp
->len
; idx
++) {
945 if (xfrm_state_ok(tmpl
, sp
->x
[idx
].xvec
, family
))
947 if (sp
->x
[idx
].xvec
->props
.mode
)
954 xfrm_decode_session(struct sk_buff
*skb
, struct flowi
*fl
, unsigned short family
)
956 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
958 if (unlikely(afinfo
== NULL
))
959 return -EAFNOSUPPORT
;
961 afinfo
->decode_session(skb
, fl
);
962 xfrm_policy_put_afinfo(afinfo
);
965 EXPORT_SYMBOL(xfrm_decode_session
);
967 static inline int secpath_has_tunnel(struct sec_path
*sp
, int k
)
969 for (; k
< sp
->len
; k
++) {
970 if (sp
->x
[k
].xvec
->props
.mode
)
977 int __xfrm_policy_check(struct sock
*sk
, int dir
, struct sk_buff
*skb
,
978 unsigned short family
)
980 struct xfrm_policy
*pol
;
982 u8 fl_dir
= policy_to_flow_dir(dir
);
985 if (xfrm_decode_session(skb
, &fl
, family
) < 0)
987 nf_nat_decode_session(skb
, &fl
, family
);
989 sk_sid
= security_sk_sid(sk
, &fl
, fl_dir
);
991 /* First, check used SA against their selectors. */
995 for (i
=skb
->sp
->len
-1; i
>=0; i
--) {
996 struct sec_decap_state
*xvec
= &(skb
->sp
->x
[i
]);
997 if (!xfrm_selector_match(&xvec
->xvec
->sel
, &fl
, family
))
1000 /* If there is a post_input processor, try running it */
1001 if (xvec
->xvec
->type
->post_input
&&
1002 (xvec
->xvec
->type
->post_input
)(xvec
->xvec
,
1010 if (sk
&& sk
->sk_policy
[dir
])
1011 pol
= xfrm_sk_policy_lookup(sk
, dir
, &fl
, sk_sid
);
1014 pol
= flow_cache_lookup(&fl
, sk_sid
, family
, fl_dir
,
1015 xfrm_policy_lookup
);
1018 return !skb
->sp
|| !secpath_has_tunnel(skb
->sp
, 0);
1020 pol
->curlft
.use_time
= (unsigned long)xtime
.tv_sec
;
1022 if (pol
->action
== XFRM_POLICY_ALLOW
) {
1023 struct sec_path
*sp
;
1024 static struct sec_path dummy
;
1027 if ((sp
= skb
->sp
) == NULL
)
1030 /* For each tunnel xfrm, find the first matching tmpl.
1031 * For each tmpl before that, find corresponding xfrm.
1032 * Order is _important_. Later we will implement
1033 * some barriers, but at the moment barriers
1034 * are implied between each two transformations.
1036 for (i
= pol
->xfrm_nr
-1, k
= 0; i
>= 0; i
--) {
1037 k
= xfrm_policy_ok(pol
->xfrm_vec
+i
, sp
, k
, family
);
1042 if (secpath_has_tunnel(sp
, k
))
1053 EXPORT_SYMBOL(__xfrm_policy_check
);
1055 int __xfrm_route_forward(struct sk_buff
*skb
, unsigned short family
)
1059 if (xfrm_decode_session(skb
, &fl
, family
) < 0)
1062 return xfrm_lookup(&skb
->dst
, &fl
, NULL
, 0) == 0;
1064 EXPORT_SYMBOL(__xfrm_route_forward
);
1066 static struct dst_entry
*xfrm_dst_check(struct dst_entry
*dst
, u32 cookie
)
1068 /* If it is marked obsolete, which is how we even get here,
1069 * then we have purged it from the policy bundle list and we
1070 * did that for a good reason.
1075 static int stale_bundle(struct dst_entry
*dst
)
1077 return !xfrm_bundle_ok((struct xfrm_dst
*)dst
, NULL
, AF_UNSPEC
);
1080 void xfrm_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
)
1082 while ((dst
= dst
->child
) && dst
->xfrm
&& dst
->dev
== dev
) {
1083 dst
->dev
= &loopback_dev
;
1084 dev_hold(&loopback_dev
);
1088 EXPORT_SYMBOL(xfrm_dst_ifdown
);
1090 static void xfrm_link_failure(struct sk_buff
*skb
)
1092 /* Impossible. Such dst must be popped before reaches point of failure. */
1096 static struct dst_entry
*xfrm_negative_advice(struct dst_entry
*dst
)
1099 if (dst
->obsolete
) {
1107 static void xfrm_prune_bundles(int (*func
)(struct dst_entry
*))
1110 struct xfrm_policy
*pol
;
1111 struct dst_entry
*dst
, **dstp
, *gc_list
= NULL
;
1113 read_lock_bh(&xfrm_policy_lock
);
1114 for (i
=0; i
<2*XFRM_POLICY_MAX
; i
++) {
1115 for (pol
= xfrm_policy_list
[i
]; pol
; pol
= pol
->next
) {
1116 write_lock(&pol
->lock
);
1117 dstp
= &pol
->bundles
;
1118 while ((dst
=*dstp
) != NULL
) {
1121 dst
->next
= gc_list
;
1127 write_unlock(&pol
->lock
);
1130 read_unlock_bh(&xfrm_policy_lock
);
1134 gc_list
= dst
->next
;
1139 static int unused_bundle(struct dst_entry
*dst
)
1141 return !atomic_read(&dst
->__refcnt
);
1144 static void __xfrm_garbage_collect(void)
1146 xfrm_prune_bundles(unused_bundle
);
1149 int xfrm_flush_bundles(void)
1151 xfrm_prune_bundles(stale_bundle
);
1155 static int always_true(struct dst_entry
*dst
)
1160 void xfrm_flush_all_bundles(void)
1162 xfrm_prune_bundles(always_true
);
1165 void xfrm_init_pmtu(struct dst_entry
*dst
)
1168 struct xfrm_dst
*xdst
= (struct xfrm_dst
*)dst
;
1169 u32 pmtu
, route_mtu_cached
;
1171 pmtu
= dst_mtu(dst
->child
);
1172 xdst
->child_mtu_cached
= pmtu
;
1174 pmtu
= xfrm_state_mtu(dst
->xfrm
, pmtu
);
1176 route_mtu_cached
= dst_mtu(xdst
->route
);
1177 xdst
->route_mtu_cached
= route_mtu_cached
;
1179 if (pmtu
> route_mtu_cached
)
1180 pmtu
= route_mtu_cached
;
1182 dst
->metrics
[RTAX_MTU
-1] = pmtu
;
1183 } while ((dst
= dst
->next
));
1186 EXPORT_SYMBOL(xfrm_init_pmtu
);
1188 /* Check that the bundle accepts the flow and its components are
1192 int xfrm_bundle_ok(struct xfrm_dst
*first
, struct flowi
*fl
, int family
)
1194 struct dst_entry
*dst
= &first
->u
.dst
;
1195 struct xfrm_dst
*last
;
1198 if (!dst_check(dst
->path
, ((struct xfrm_dst
*)dst
)->path_cookie
) ||
1199 (dst
->dev
&& !netif_running(dst
->dev
)))
1205 struct xfrm_dst
*xdst
= (struct xfrm_dst
*)dst
;
1207 if (fl
&& !xfrm_selector_match(&dst
->xfrm
->sel
, fl
, family
))
1209 if (dst
->xfrm
->km
.state
!= XFRM_STATE_VALID
)
1212 mtu
= dst_mtu(dst
->child
);
1213 if (xdst
->child_mtu_cached
!= mtu
) {
1215 xdst
->child_mtu_cached
= mtu
;
1218 if (!dst_check(xdst
->route
, xdst
->route_cookie
))
1220 mtu
= dst_mtu(xdst
->route
);
1221 if (xdst
->route_mtu_cached
!= mtu
) {
1223 xdst
->route_mtu_cached
= mtu
;
1227 } while (dst
->xfrm
);
1232 mtu
= last
->child_mtu_cached
;
1236 mtu
= xfrm_state_mtu(dst
->xfrm
, mtu
);
1237 if (mtu
> last
->route_mtu_cached
)
1238 mtu
= last
->route_mtu_cached
;
1239 dst
->metrics
[RTAX_MTU
-1] = mtu
;
1244 last
= last
->u
.next
;
1245 last
->child_mtu_cached
= mtu
;
1251 EXPORT_SYMBOL(xfrm_bundle_ok
);
1253 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1256 if (unlikely(afinfo
== NULL
))
1258 if (unlikely(afinfo
->family
>= NPROTO
))
1259 return -EAFNOSUPPORT
;
1260 write_lock(&xfrm_policy_afinfo_lock
);
1261 if (unlikely(xfrm_policy_afinfo
[afinfo
->family
] != NULL
))
1264 struct dst_ops
*dst_ops
= afinfo
->dst_ops
;
1265 if (likely(dst_ops
->kmem_cachep
== NULL
))
1266 dst_ops
->kmem_cachep
= xfrm_dst_cache
;
1267 if (likely(dst_ops
->check
== NULL
))
1268 dst_ops
->check
= xfrm_dst_check
;
1269 if (likely(dst_ops
->negative_advice
== NULL
))
1270 dst_ops
->negative_advice
= xfrm_negative_advice
;
1271 if (likely(dst_ops
->link_failure
== NULL
))
1272 dst_ops
->link_failure
= xfrm_link_failure
;
1273 if (likely(afinfo
->garbage_collect
== NULL
))
1274 afinfo
->garbage_collect
= __xfrm_garbage_collect
;
1275 xfrm_policy_afinfo
[afinfo
->family
] = afinfo
;
1277 write_unlock(&xfrm_policy_afinfo_lock
);
1280 EXPORT_SYMBOL(xfrm_policy_register_afinfo
);
1282 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1285 if (unlikely(afinfo
== NULL
))
1287 if (unlikely(afinfo
->family
>= NPROTO
))
1288 return -EAFNOSUPPORT
;
1289 write_lock(&xfrm_policy_afinfo_lock
);
1290 if (likely(xfrm_policy_afinfo
[afinfo
->family
] != NULL
)) {
1291 if (unlikely(xfrm_policy_afinfo
[afinfo
->family
] != afinfo
))
1294 struct dst_ops
*dst_ops
= afinfo
->dst_ops
;
1295 xfrm_policy_afinfo
[afinfo
->family
] = NULL
;
1296 dst_ops
->kmem_cachep
= NULL
;
1297 dst_ops
->check
= NULL
;
1298 dst_ops
->negative_advice
= NULL
;
1299 dst_ops
->link_failure
= NULL
;
1300 afinfo
->garbage_collect
= NULL
;
1303 write_unlock(&xfrm_policy_afinfo_lock
);
1306 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo
);
1308 static struct xfrm_policy_afinfo
*xfrm_policy_get_afinfo(unsigned short family
)
1310 struct xfrm_policy_afinfo
*afinfo
;
1311 if (unlikely(family
>= NPROTO
))
1313 read_lock(&xfrm_policy_afinfo_lock
);
1314 afinfo
= xfrm_policy_afinfo
[family
];
1315 if (likely(afinfo
!= NULL
))
1316 read_lock(&afinfo
->lock
);
1317 read_unlock(&xfrm_policy_afinfo_lock
);
1321 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1323 if (unlikely(afinfo
== NULL
))
1325 read_unlock(&afinfo
->lock
);
1328 static int xfrm_dev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
1332 xfrm_flush_bundles();
1337 static struct notifier_block xfrm_dev_notifier
= {
1343 static void __init
xfrm_policy_init(void)
1345 xfrm_dst_cache
= kmem_cache_create("xfrm_dst_cache",
1346 sizeof(struct xfrm_dst
),
1347 0, SLAB_HWCACHE_ALIGN
,
1349 if (!xfrm_dst_cache
)
1350 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1352 INIT_WORK(&xfrm_policy_gc_work
, xfrm_policy_gc_task
, NULL
);
1353 register_netdevice_notifier(&xfrm_dev_notifier
);
1356 void __init
xfrm_init(void)