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
))
259 security_xfrm_policy_free(policy
);
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
;
350 struct dst_entry
*gc_list
;
352 write_lock_bh(&xfrm_policy_lock
);
353 for (p
= &xfrm_policy_list
[dir
]; (pol
=*p
)!=NULL
;) {
354 if (!delpol
&& memcmp(&policy
->selector
, &pol
->selector
, sizeof(pol
->selector
)) == 0 &&
355 xfrm_sec_ctx_match(pol
->security
, policy
->security
)) {
357 write_unlock_bh(&xfrm_policy_lock
);
362 if (policy
->priority
> pol
->priority
)
364 } else if (policy
->priority
>= pol
->priority
) {
376 xfrm_pol_hold(policy
);
379 atomic_inc(&flow_cache_genid
);
380 policy
->index
= delpol
? delpol
->index
: xfrm_gen_index(dir
);
381 policy
->curlft
.add_time
= (unsigned long)xtime
.tv_sec
;
382 policy
->curlft
.use_time
= 0;
383 if (!mod_timer(&policy
->timer
, jiffies
+ HZ
))
384 xfrm_pol_hold(policy
);
385 write_unlock_bh(&xfrm_policy_lock
);
388 xfrm_policy_kill(delpol
);
390 read_lock_bh(&xfrm_policy_lock
);
392 for (policy
= policy
->next
; policy
; policy
= policy
->next
) {
393 struct dst_entry
*dst
;
395 write_lock(&policy
->lock
);
396 dst
= policy
->bundles
;
398 struct dst_entry
*tail
= dst
;
401 tail
->next
= gc_list
;
404 policy
->bundles
= NULL
;
406 write_unlock(&policy
->lock
);
408 read_unlock_bh(&xfrm_policy_lock
);
411 struct dst_entry
*dst
= gc_list
;
419 EXPORT_SYMBOL(xfrm_policy_insert
);
421 struct xfrm_policy
*xfrm_policy_bysel_ctx(int dir
, struct xfrm_selector
*sel
,
422 struct xfrm_sec_ctx
*ctx
, int delete)
424 struct xfrm_policy
*pol
, **p
;
426 write_lock_bh(&xfrm_policy_lock
);
427 for (p
= &xfrm_policy_list
[dir
]; (pol
=*p
)!=NULL
; p
= &pol
->next
) {
428 if ((memcmp(sel
, &pol
->selector
, sizeof(*sel
)) == 0) &&
429 (xfrm_sec_ctx_match(ctx
, pol
->security
))) {
436 write_unlock_bh(&xfrm_policy_lock
);
439 atomic_inc(&flow_cache_genid
);
440 xfrm_policy_kill(pol
);
444 EXPORT_SYMBOL(xfrm_policy_bysel_ctx
);
446 struct xfrm_policy
*xfrm_policy_byid(int dir
, u32 id
, int delete)
448 struct xfrm_policy
*pol
, **p
;
450 write_lock_bh(&xfrm_policy_lock
);
451 for (p
= &xfrm_policy_list
[dir
]; (pol
=*p
)!=NULL
; p
= &pol
->next
) {
452 if (pol
->index
== id
) {
459 write_unlock_bh(&xfrm_policy_lock
);
462 atomic_inc(&flow_cache_genid
);
463 xfrm_policy_kill(pol
);
467 EXPORT_SYMBOL(xfrm_policy_byid
);
469 void xfrm_policy_flush(void)
471 struct xfrm_policy
*xp
;
474 write_lock_bh(&xfrm_policy_lock
);
475 for (dir
= 0; dir
< XFRM_POLICY_MAX
; dir
++) {
476 while ((xp
= xfrm_policy_list
[dir
]) != NULL
) {
477 xfrm_policy_list
[dir
] = xp
->next
;
478 write_unlock_bh(&xfrm_policy_lock
);
480 xfrm_policy_kill(xp
);
482 write_lock_bh(&xfrm_policy_lock
);
485 atomic_inc(&flow_cache_genid
);
486 write_unlock_bh(&xfrm_policy_lock
);
488 EXPORT_SYMBOL(xfrm_policy_flush
);
490 int xfrm_policy_walk(int (*func
)(struct xfrm_policy
*, int, int, void*),
493 struct xfrm_policy
*xp
;
498 read_lock_bh(&xfrm_policy_lock
);
499 for (dir
= 0; dir
< 2*XFRM_POLICY_MAX
; dir
++) {
500 for (xp
= xfrm_policy_list
[dir
]; xp
; xp
= xp
->next
)
509 for (dir
= 0; dir
< 2*XFRM_POLICY_MAX
; dir
++) {
510 for (xp
= xfrm_policy_list
[dir
]; xp
; xp
= xp
->next
) {
511 error
= func(xp
, dir
%XFRM_POLICY_MAX
, --count
, data
);
518 read_unlock_bh(&xfrm_policy_lock
);
521 EXPORT_SYMBOL(xfrm_policy_walk
);
523 /* Find policy to apply to this flow. */
525 static void xfrm_policy_lookup(struct flowi
*fl
, u32 sk_sid
, u16 family
, u8 dir
,
526 void **objp
, atomic_t
**obj_refp
)
528 struct xfrm_policy
*pol
;
530 read_lock_bh(&xfrm_policy_lock
);
531 for (pol
= xfrm_policy_list
[dir
]; pol
; pol
= pol
->next
) {
532 struct xfrm_selector
*sel
= &pol
->selector
;
535 if (pol
->family
!= family
)
538 match
= xfrm_selector_match(sel
, fl
, family
);
541 if (!security_xfrm_policy_lookup(pol
, sk_sid
, dir
)) {
547 read_unlock_bh(&xfrm_policy_lock
);
548 if ((*objp
= (void *) pol
) != NULL
)
549 *obj_refp
= &pol
->refcnt
;
552 static inline int policy_to_flow_dir(int dir
)
554 if (XFRM_POLICY_IN
== FLOW_DIR_IN
&&
555 XFRM_POLICY_OUT
== FLOW_DIR_OUT
&&
556 XFRM_POLICY_FWD
== FLOW_DIR_FWD
)
562 case XFRM_POLICY_OUT
:
564 case XFRM_POLICY_FWD
:
569 static struct xfrm_policy
*xfrm_sk_policy_lookup(struct sock
*sk
, int dir
, struct flowi
*fl
, u32 sk_sid
)
571 struct xfrm_policy
*pol
;
573 read_lock_bh(&xfrm_policy_lock
);
574 if ((pol
= sk
->sk_policy
[dir
]) != NULL
) {
575 int match
= xfrm_selector_match(&pol
->selector
, fl
,
580 err
= security_xfrm_policy_lookup(pol
, sk_sid
, policy_to_flow_dir(dir
));
587 read_unlock_bh(&xfrm_policy_lock
);
591 static void __xfrm_policy_link(struct xfrm_policy
*pol
, int dir
)
593 pol
->next
= xfrm_policy_list
[dir
];
594 xfrm_policy_list
[dir
] = pol
;
598 static struct xfrm_policy
*__xfrm_policy_unlink(struct xfrm_policy
*pol
,
601 struct xfrm_policy
**polp
;
603 for (polp
= &xfrm_policy_list
[dir
];
604 *polp
!= NULL
; polp
= &(*polp
)->next
) {
613 int xfrm_policy_delete(struct xfrm_policy
*pol
, int dir
)
615 write_lock_bh(&xfrm_policy_lock
);
616 pol
= __xfrm_policy_unlink(pol
, dir
);
617 write_unlock_bh(&xfrm_policy_lock
);
619 if (dir
< XFRM_POLICY_MAX
)
620 atomic_inc(&flow_cache_genid
);
621 xfrm_policy_kill(pol
);
627 int xfrm_sk_policy_insert(struct sock
*sk
, int dir
, struct xfrm_policy
*pol
)
629 struct xfrm_policy
*old_pol
;
631 write_lock_bh(&xfrm_policy_lock
);
632 old_pol
= sk
->sk_policy
[dir
];
633 sk
->sk_policy
[dir
] = pol
;
635 pol
->curlft
.add_time
= (unsigned long)xtime
.tv_sec
;
636 pol
->index
= xfrm_gen_index(XFRM_POLICY_MAX
+dir
);
637 __xfrm_policy_link(pol
, XFRM_POLICY_MAX
+dir
);
640 __xfrm_policy_unlink(old_pol
, XFRM_POLICY_MAX
+dir
);
641 write_unlock_bh(&xfrm_policy_lock
);
644 xfrm_policy_kill(old_pol
);
649 static struct xfrm_policy
*clone_policy(struct xfrm_policy
*old
, int dir
)
651 struct xfrm_policy
*newp
= xfrm_policy_alloc(GFP_ATOMIC
);
654 newp
->selector
= old
->selector
;
655 if (security_xfrm_policy_clone(old
, newp
)) {
657 return NULL
; /* ENOMEM */
659 newp
->lft
= old
->lft
;
660 newp
->curlft
= old
->curlft
;
661 newp
->action
= old
->action
;
662 newp
->flags
= old
->flags
;
663 newp
->xfrm_nr
= old
->xfrm_nr
;
664 newp
->index
= old
->index
;
665 memcpy(newp
->xfrm_vec
, old
->xfrm_vec
,
666 newp
->xfrm_nr
*sizeof(struct xfrm_tmpl
));
667 write_lock_bh(&xfrm_policy_lock
);
668 __xfrm_policy_link(newp
, XFRM_POLICY_MAX
+dir
);
669 write_unlock_bh(&xfrm_policy_lock
);
675 int __xfrm_sk_clone_policy(struct sock
*sk
)
677 struct xfrm_policy
*p0
= sk
->sk_policy
[0],
678 *p1
= sk
->sk_policy
[1];
680 sk
->sk_policy
[0] = sk
->sk_policy
[1] = NULL
;
681 if (p0
&& (sk
->sk_policy
[0] = clone_policy(p0
, 0)) == NULL
)
683 if (p1
&& (sk
->sk_policy
[1] = clone_policy(p1
, 1)) == NULL
)
688 /* Resolve list of templates for the flow, given policy. */
691 xfrm_tmpl_resolve(struct xfrm_policy
*policy
, struct flowi
*fl
,
692 struct xfrm_state
**xfrm
,
693 unsigned short family
)
697 xfrm_address_t
*daddr
= xfrm_flowi_daddr(fl
, family
);
698 xfrm_address_t
*saddr
= xfrm_flowi_saddr(fl
, family
);
700 for (nx
=0, i
= 0; i
< policy
->xfrm_nr
; i
++) {
701 struct xfrm_state
*x
;
702 xfrm_address_t
*remote
= daddr
;
703 xfrm_address_t
*local
= saddr
;
704 struct xfrm_tmpl
*tmpl
= &policy
->xfrm_vec
[i
];
707 remote
= &tmpl
->id
.daddr
;
708 local
= &tmpl
->saddr
;
711 x
= xfrm_state_find(remote
, local
, fl
, tmpl
, policy
, &error
, family
);
713 if (x
&& x
->km
.state
== XFRM_STATE_VALID
) {
720 error
= (x
->km
.state
== XFRM_STATE_ERROR
?
731 for (nx
--; nx
>=0; nx
--)
732 xfrm_state_put(xfrm
[nx
]);
736 /* Check that the bundle accepts the flow and its components are
740 static struct dst_entry
*
741 xfrm_find_bundle(struct flowi
*fl
, struct xfrm_policy
*policy
, unsigned short family
)
744 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
745 if (unlikely(afinfo
== NULL
))
746 return ERR_PTR(-EINVAL
);
747 x
= afinfo
->find_bundle(fl
, policy
);
748 xfrm_policy_put_afinfo(afinfo
);
752 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
753 * all the metrics... Shortly, bundle a bundle.
757 xfrm_bundle_create(struct xfrm_policy
*policy
, struct xfrm_state
**xfrm
, int nx
,
758 struct flowi
*fl
, struct dst_entry
**dst_p
,
759 unsigned short family
)
762 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
763 if (unlikely(afinfo
== NULL
))
765 err
= afinfo
->bundle_create(policy
, xfrm
, nx
, fl
, dst_p
);
766 xfrm_policy_put_afinfo(afinfo
);
771 static int stale_bundle(struct dst_entry
*dst
);
773 /* Main function: finds/creates a bundle for given flow.
775 * At the moment we eat a raw IP route. Mostly to speed up lookups
776 * on interfaces with disabled IPsec.
778 int xfrm_lookup(struct dst_entry
**dst_p
, struct flowi
*fl
,
779 struct sock
*sk
, int flags
)
781 struct xfrm_policy
*policy
;
782 struct xfrm_state
*xfrm
[XFRM_MAX_DEPTH
];
783 struct dst_entry
*dst
, *dst_orig
= *dst_p
;
787 u16 family
= dst_orig
->ops
->family
;
788 u8 dir
= policy_to_flow_dir(XFRM_POLICY_OUT
);
789 u32 sk_sid
= security_sk_sid(sk
, fl
, dir
);
791 genid
= atomic_read(&flow_cache_genid
);
793 if (sk
&& sk
->sk_policy
[1])
794 policy
= xfrm_sk_policy_lookup(sk
, XFRM_POLICY_OUT
, fl
, sk_sid
);
797 /* To accelerate a bit... */
798 if ((dst_orig
->flags
& DST_NOXFRM
) || !xfrm_policy_list
[XFRM_POLICY_OUT
])
801 policy
= flow_cache_lookup(fl
, sk_sid
, family
, dir
,
808 policy
->curlft
.use_time
= (unsigned long)xtime
.tv_sec
;
810 switch (policy
->action
) {
811 case XFRM_POLICY_BLOCK
:
812 /* Prohibit the flow */
816 case XFRM_POLICY_ALLOW
:
817 if (policy
->xfrm_nr
== 0) {
818 /* Flow passes not transformed. */
819 xfrm_pol_put(policy
);
823 /* Try to find matching bundle.
825 * LATER: help from flow cache. It is optional, this
826 * is required only for output policy.
828 dst
= xfrm_find_bundle(fl
, policy
, family
);
837 nx
= xfrm_tmpl_resolve(policy
, fl
, xfrm
, family
);
839 if (unlikely(nx
<0)) {
841 if (err
== -EAGAIN
&& flags
) {
842 DECLARE_WAITQUEUE(wait
, current
);
844 add_wait_queue(&km_waitq
, &wait
);
845 set_current_state(TASK_INTERRUPTIBLE
);
847 set_current_state(TASK_RUNNING
);
848 remove_wait_queue(&km_waitq
, &wait
);
850 nx
= xfrm_tmpl_resolve(policy
, fl
, xfrm
, family
);
852 if (nx
== -EAGAIN
&& signal_pending(current
)) {
857 genid
!= atomic_read(&flow_cache_genid
)) {
858 xfrm_pol_put(policy
);
867 /* Flow passes not transformed. */
868 xfrm_pol_put(policy
);
873 err
= xfrm_bundle_create(policy
, xfrm
, nx
, fl
, &dst
, family
);
878 xfrm_state_put(xfrm
[i
]);
882 write_lock_bh(&policy
->lock
);
883 if (unlikely(policy
->dead
|| stale_bundle(dst
))) {
884 /* Wow! While we worked on resolving, this
885 * policy has gone. Retry. It is not paranoia,
886 * we just cannot enlist new bundle to dead object.
887 * We can't enlist stable bundles either.
889 write_unlock_bh(&policy
->lock
);
891 xfrm_pol_put(policy
);
896 dst
->next
= policy
->bundles
;
897 policy
->bundles
= dst
;
899 write_unlock_bh(&policy
->lock
);
902 dst_release(dst_orig
);
903 xfrm_pol_put(policy
);
907 dst_release(dst_orig
);
908 xfrm_pol_put(policy
);
912 EXPORT_SYMBOL(xfrm_lookup
);
914 /* When skb is transformed back to its "native" form, we have to
915 * check policy restrictions. At the moment we make this in maximally
916 * stupid way. Shame on me. :-) Of course, connected sockets must
917 * have policy cached at them.
921 xfrm_state_ok(struct xfrm_tmpl
*tmpl
, struct xfrm_state
*x
,
922 unsigned short family
)
924 if (xfrm_state_kern(x
))
925 return tmpl
->optional
&& !xfrm_state_addr_cmp(tmpl
, x
, family
);
926 return x
->id
.proto
== tmpl
->id
.proto
&&
927 (x
->id
.spi
== tmpl
->id
.spi
|| !tmpl
->id
.spi
) &&
928 (x
->props
.reqid
== tmpl
->reqid
|| !tmpl
->reqid
) &&
929 x
->props
.mode
== tmpl
->mode
&&
930 (tmpl
->aalgos
& (1<<x
->props
.aalgo
)) &&
931 !(x
->props
.mode
&& xfrm_state_addr_cmp(tmpl
, x
, family
));
935 xfrm_policy_ok(struct xfrm_tmpl
*tmpl
, struct sec_path
*sp
, int start
,
936 unsigned short family
)
940 if (tmpl
->optional
) {
945 for (; idx
< sp
->len
; idx
++) {
946 if (xfrm_state_ok(tmpl
, sp
->x
[idx
].xvec
, family
))
948 if (sp
->x
[idx
].xvec
->props
.mode
)
955 _decode_session(struct sk_buff
*skb
, struct flowi
*fl
, unsigned short family
)
957 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
959 if (unlikely(afinfo
== NULL
))
960 return -EAFNOSUPPORT
;
962 afinfo
->decode_session(skb
, fl
);
963 xfrm_policy_put_afinfo(afinfo
);
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 (_decode_session(skb
, &fl
, family
) < 0)
988 sk_sid
= security_sk_sid(sk
, &fl
, fl_dir
);
990 /* First, check used SA against their selectors. */
994 for (i
=skb
->sp
->len
-1; i
>=0; i
--) {
995 struct sec_decap_state
*xvec
= &(skb
->sp
->x
[i
]);
996 if (!xfrm_selector_match(&xvec
->xvec
->sel
, &fl
, family
))
999 /* If there is a post_input processor, try running it */
1000 if (xvec
->xvec
->type
->post_input
&&
1001 (xvec
->xvec
->type
->post_input
)(xvec
->xvec
,
1009 if (sk
&& sk
->sk_policy
[dir
])
1010 pol
= xfrm_sk_policy_lookup(sk
, dir
, &fl
, sk_sid
);
1013 pol
= flow_cache_lookup(&fl
, sk_sid
, family
, fl_dir
,
1014 xfrm_policy_lookup
);
1017 return !skb
->sp
|| !secpath_has_tunnel(skb
->sp
, 0);
1019 pol
->curlft
.use_time
= (unsigned long)xtime
.tv_sec
;
1021 if (pol
->action
== XFRM_POLICY_ALLOW
) {
1022 struct sec_path
*sp
;
1023 static struct sec_path dummy
;
1026 if ((sp
= skb
->sp
) == NULL
)
1029 /* For each tunnel xfrm, find the first matching tmpl.
1030 * For each tmpl before that, find corresponding xfrm.
1031 * Order is _important_. Later we will implement
1032 * some barriers, but at the moment barriers
1033 * are implied between each two transformations.
1035 for (i
= pol
->xfrm_nr
-1, k
= 0; i
>= 0; i
--) {
1036 k
= xfrm_policy_ok(pol
->xfrm_vec
+i
, sp
, k
, family
);
1041 if (secpath_has_tunnel(sp
, k
))
1052 EXPORT_SYMBOL(__xfrm_policy_check
);
1054 int __xfrm_route_forward(struct sk_buff
*skb
, unsigned short family
)
1058 if (_decode_session(skb
, &fl
, family
) < 0)
1061 return xfrm_lookup(&skb
->dst
, &fl
, NULL
, 0) == 0;
1063 EXPORT_SYMBOL(__xfrm_route_forward
);
1065 static struct dst_entry
*xfrm_dst_check(struct dst_entry
*dst
, u32 cookie
)
1067 /* If it is marked obsolete, which is how we even get here,
1068 * then we have purged it from the policy bundle list and we
1069 * did that for a good reason.
1074 static int stale_bundle(struct dst_entry
*dst
)
1076 return !xfrm_bundle_ok((struct xfrm_dst
*)dst
, NULL
, AF_UNSPEC
);
1079 void xfrm_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
)
1081 while ((dst
= dst
->child
) && dst
->xfrm
&& dst
->dev
== dev
) {
1082 dst
->dev
= &loopback_dev
;
1083 dev_hold(&loopback_dev
);
1087 EXPORT_SYMBOL(xfrm_dst_ifdown
);
1089 static void xfrm_link_failure(struct sk_buff
*skb
)
1091 /* Impossible. Such dst must be popped before reaches point of failure. */
1095 static struct dst_entry
*xfrm_negative_advice(struct dst_entry
*dst
)
1098 if (dst
->obsolete
) {
1106 static void xfrm_prune_bundles(int (*func
)(struct dst_entry
*))
1109 struct xfrm_policy
*pol
;
1110 struct dst_entry
*dst
, **dstp
, *gc_list
= NULL
;
1112 read_lock_bh(&xfrm_policy_lock
);
1113 for (i
=0; i
<2*XFRM_POLICY_MAX
; i
++) {
1114 for (pol
= xfrm_policy_list
[i
]; pol
; pol
= pol
->next
) {
1115 write_lock(&pol
->lock
);
1116 dstp
= &pol
->bundles
;
1117 while ((dst
=*dstp
) != NULL
) {
1120 dst
->next
= gc_list
;
1126 write_unlock(&pol
->lock
);
1129 read_unlock_bh(&xfrm_policy_lock
);
1133 gc_list
= dst
->next
;
1138 static int unused_bundle(struct dst_entry
*dst
)
1140 return !atomic_read(&dst
->__refcnt
);
1143 static void __xfrm_garbage_collect(void)
1145 xfrm_prune_bundles(unused_bundle
);
1148 int xfrm_flush_bundles(void)
1150 xfrm_prune_bundles(stale_bundle
);
1154 static int always_true(struct dst_entry
*dst
)
1159 void xfrm_flush_all_bundles(void)
1161 xfrm_prune_bundles(always_true
);
1164 void xfrm_init_pmtu(struct dst_entry
*dst
)
1167 struct xfrm_dst
*xdst
= (struct xfrm_dst
*)dst
;
1168 u32 pmtu
, route_mtu_cached
;
1170 pmtu
= dst_mtu(dst
->child
);
1171 xdst
->child_mtu_cached
= pmtu
;
1173 pmtu
= xfrm_state_mtu(dst
->xfrm
, pmtu
);
1175 route_mtu_cached
= dst_mtu(xdst
->route
);
1176 xdst
->route_mtu_cached
= route_mtu_cached
;
1178 if (pmtu
> route_mtu_cached
)
1179 pmtu
= route_mtu_cached
;
1181 dst
->metrics
[RTAX_MTU
-1] = pmtu
;
1182 } while ((dst
= dst
->next
));
1185 EXPORT_SYMBOL(xfrm_init_pmtu
);
1187 /* Check that the bundle accepts the flow and its components are
1191 int xfrm_bundle_ok(struct xfrm_dst
*first
, struct flowi
*fl
, int family
)
1193 struct dst_entry
*dst
= &first
->u
.dst
;
1194 struct xfrm_dst
*last
;
1197 if (!dst_check(dst
->path
, ((struct xfrm_dst
*)dst
)->path_cookie
) ||
1198 (dst
->dev
&& !netif_running(dst
->dev
)))
1204 struct xfrm_dst
*xdst
= (struct xfrm_dst
*)dst
;
1206 if (fl
&& !xfrm_selector_match(&dst
->xfrm
->sel
, fl
, family
))
1208 if (dst
->xfrm
->km
.state
!= XFRM_STATE_VALID
)
1211 mtu
= dst_mtu(dst
->child
);
1212 if (xdst
->child_mtu_cached
!= mtu
) {
1214 xdst
->child_mtu_cached
= mtu
;
1217 if (!dst_check(xdst
->route
, xdst
->route_cookie
))
1219 mtu
= dst_mtu(xdst
->route
);
1220 if (xdst
->route_mtu_cached
!= mtu
) {
1222 xdst
->route_mtu_cached
= mtu
;
1226 } while (dst
->xfrm
);
1231 mtu
= last
->child_mtu_cached
;
1235 mtu
= xfrm_state_mtu(dst
->xfrm
, mtu
);
1236 if (mtu
> last
->route_mtu_cached
)
1237 mtu
= last
->route_mtu_cached
;
1238 dst
->metrics
[RTAX_MTU
-1] = mtu
;
1243 last
= last
->u
.next
;
1244 last
->child_mtu_cached
= mtu
;
1250 EXPORT_SYMBOL(xfrm_bundle_ok
);
1252 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1255 if (unlikely(afinfo
== NULL
))
1257 if (unlikely(afinfo
->family
>= NPROTO
))
1258 return -EAFNOSUPPORT
;
1259 write_lock(&xfrm_policy_afinfo_lock
);
1260 if (unlikely(xfrm_policy_afinfo
[afinfo
->family
] != NULL
))
1263 struct dst_ops
*dst_ops
= afinfo
->dst_ops
;
1264 if (likely(dst_ops
->kmem_cachep
== NULL
))
1265 dst_ops
->kmem_cachep
= xfrm_dst_cache
;
1266 if (likely(dst_ops
->check
== NULL
))
1267 dst_ops
->check
= xfrm_dst_check
;
1268 if (likely(dst_ops
->negative_advice
== NULL
))
1269 dst_ops
->negative_advice
= xfrm_negative_advice
;
1270 if (likely(dst_ops
->link_failure
== NULL
))
1271 dst_ops
->link_failure
= xfrm_link_failure
;
1272 if (likely(afinfo
->garbage_collect
== NULL
))
1273 afinfo
->garbage_collect
= __xfrm_garbage_collect
;
1274 xfrm_policy_afinfo
[afinfo
->family
] = afinfo
;
1276 write_unlock(&xfrm_policy_afinfo_lock
);
1279 EXPORT_SYMBOL(xfrm_policy_register_afinfo
);
1281 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1284 if (unlikely(afinfo
== NULL
))
1286 if (unlikely(afinfo
->family
>= NPROTO
))
1287 return -EAFNOSUPPORT
;
1288 write_lock(&xfrm_policy_afinfo_lock
);
1289 if (likely(xfrm_policy_afinfo
[afinfo
->family
] != NULL
)) {
1290 if (unlikely(xfrm_policy_afinfo
[afinfo
->family
] != afinfo
))
1293 struct dst_ops
*dst_ops
= afinfo
->dst_ops
;
1294 xfrm_policy_afinfo
[afinfo
->family
] = NULL
;
1295 dst_ops
->kmem_cachep
= NULL
;
1296 dst_ops
->check
= NULL
;
1297 dst_ops
->negative_advice
= NULL
;
1298 dst_ops
->link_failure
= NULL
;
1299 afinfo
->garbage_collect
= NULL
;
1302 write_unlock(&xfrm_policy_afinfo_lock
);
1305 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo
);
1307 static struct xfrm_policy_afinfo
*xfrm_policy_get_afinfo(unsigned short family
)
1309 struct xfrm_policy_afinfo
*afinfo
;
1310 if (unlikely(family
>= NPROTO
))
1312 read_lock(&xfrm_policy_afinfo_lock
);
1313 afinfo
= xfrm_policy_afinfo
[family
];
1314 if (likely(afinfo
!= NULL
))
1315 read_lock(&afinfo
->lock
);
1316 read_unlock(&xfrm_policy_afinfo_lock
);
1320 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1322 if (unlikely(afinfo
== NULL
))
1324 read_unlock(&afinfo
->lock
);
1327 static int xfrm_dev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
1331 xfrm_flush_bundles();
1336 static struct notifier_block xfrm_dev_notifier
= {
1342 static void __init
xfrm_policy_init(void)
1344 xfrm_dst_cache
= kmem_cache_create("xfrm_dst_cache",
1345 sizeof(struct xfrm_dst
),
1346 0, SLAB_HWCACHE_ALIGN
,
1348 if (!xfrm_dst_cache
)
1349 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1351 INIT_WORK(&xfrm_policy_gc_work
, xfrm_policy_gc_task
, NULL
);
1352 register_netdevice_notifier(&xfrm_dev_notifier
);
1355 void __init
xfrm_init(void)