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 DEFINE_MUTEX(xfrm_cfg_mutex
);
30 EXPORT_SYMBOL(xfrm_cfg_mutex
);
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
);
49 static struct xfrm_policy_afinfo
*xfrm_policy_lock_afinfo(unsigned int family
);
50 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo
*afinfo
);
52 int xfrm_register_type(struct xfrm_type
*type
, unsigned short family
)
54 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_lock_afinfo(family
);
55 struct xfrm_type
**typemap
;
58 if (unlikely(afinfo
== NULL
))
60 typemap
= afinfo
->type_map
;
62 if (likely(typemap
[type
->proto
] == NULL
))
63 typemap
[type
->proto
] = type
;
66 xfrm_policy_unlock_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_lock_afinfo(family
);
74 struct xfrm_type
**typemap
;
77 if (unlikely(afinfo
== NULL
))
79 typemap
= afinfo
->type_map
;
81 if (unlikely(typemap
[type
->proto
] != type
))
84 typemap
[type
->proto
] = NULL
;
85 xfrm_policy_unlock_afinfo(afinfo
);
88 EXPORT_SYMBOL(xfrm_unregister_type
);
90 struct xfrm_type
*xfrm_get_type(u8 proto
, unsigned short family
)
92 struct xfrm_policy_afinfo
*afinfo
;
93 struct xfrm_type
**typemap
;
94 struct xfrm_type
*type
;
95 int modload_attempted
= 0;
98 afinfo
= xfrm_policy_get_afinfo(family
);
99 if (unlikely(afinfo
== NULL
))
101 typemap
= afinfo
->type_map
;
103 type
= typemap
[proto
];
104 if (unlikely(type
&& !try_module_get(type
->owner
)))
106 if (!type
&& !modload_attempted
) {
107 xfrm_policy_put_afinfo(afinfo
);
108 request_module("xfrm-type-%d-%d",
109 (int) family
, (int) proto
);
110 modload_attempted
= 1;
114 xfrm_policy_put_afinfo(afinfo
);
118 int xfrm_dst_lookup(struct xfrm_dst
**dst
, struct flowi
*fl
,
119 unsigned short family
)
121 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
124 if (unlikely(afinfo
== NULL
))
125 return -EAFNOSUPPORT
;
127 if (likely(afinfo
->dst_lookup
!= NULL
))
128 err
= afinfo
->dst_lookup(dst
, fl
);
131 xfrm_policy_put_afinfo(afinfo
);
134 EXPORT_SYMBOL(xfrm_dst_lookup
);
136 void xfrm_put_type(struct xfrm_type
*type
)
138 module_put(type
->owner
);
141 int xfrm_register_mode(struct xfrm_mode
*mode
, int family
)
143 struct xfrm_policy_afinfo
*afinfo
;
144 struct xfrm_mode
**modemap
;
147 if (unlikely(mode
->encap
>= XFRM_MODE_MAX
))
150 afinfo
= xfrm_policy_lock_afinfo(family
);
151 if (unlikely(afinfo
== NULL
))
152 return -EAFNOSUPPORT
;
155 modemap
= afinfo
->mode_map
;
156 if (likely(modemap
[mode
->encap
] == NULL
)) {
157 modemap
[mode
->encap
] = mode
;
161 xfrm_policy_unlock_afinfo(afinfo
);
164 EXPORT_SYMBOL(xfrm_register_mode
);
166 int xfrm_unregister_mode(struct xfrm_mode
*mode
, int family
)
168 struct xfrm_policy_afinfo
*afinfo
;
169 struct xfrm_mode
**modemap
;
172 if (unlikely(mode
->encap
>= XFRM_MODE_MAX
))
175 afinfo
= xfrm_policy_lock_afinfo(family
);
176 if (unlikely(afinfo
== NULL
))
177 return -EAFNOSUPPORT
;
180 modemap
= afinfo
->mode_map
;
181 if (likely(modemap
[mode
->encap
] == mode
)) {
182 modemap
[mode
->encap
] = NULL
;
186 xfrm_policy_unlock_afinfo(afinfo
);
189 EXPORT_SYMBOL(xfrm_unregister_mode
);
191 struct xfrm_mode
*xfrm_get_mode(unsigned int encap
, int family
)
193 struct xfrm_policy_afinfo
*afinfo
;
194 struct xfrm_mode
*mode
;
195 int modload_attempted
= 0;
197 if (unlikely(encap
>= XFRM_MODE_MAX
))
201 afinfo
= xfrm_policy_get_afinfo(family
);
202 if (unlikely(afinfo
== NULL
))
205 mode
= afinfo
->mode_map
[encap
];
206 if (unlikely(mode
&& !try_module_get(mode
->owner
)))
208 if (!mode
&& !modload_attempted
) {
209 xfrm_policy_put_afinfo(afinfo
);
210 request_module("xfrm-mode-%d-%d", family
, encap
);
211 modload_attempted
= 1;
215 xfrm_policy_put_afinfo(afinfo
);
219 void xfrm_put_mode(struct xfrm_mode
*mode
)
221 module_put(mode
->owner
);
224 static inline unsigned long make_jiffies(long secs
)
226 if (secs
>= (MAX_SCHEDULE_TIMEOUT
-1)/HZ
)
227 return MAX_SCHEDULE_TIMEOUT
-1;
232 static void xfrm_policy_timer(unsigned long data
)
234 struct xfrm_policy
*xp
= (struct xfrm_policy
*)data
;
235 unsigned long now
= (unsigned long)xtime
.tv_sec
;
236 long next
= LONG_MAX
;
240 read_lock(&xp
->lock
);
245 dir
= xfrm_policy_id2dir(xp
->index
);
247 if (xp
->lft
.hard_add_expires_seconds
) {
248 long tmo
= xp
->lft
.hard_add_expires_seconds
+
249 xp
->curlft
.add_time
- now
;
255 if (xp
->lft
.hard_use_expires_seconds
) {
256 long tmo
= xp
->lft
.hard_use_expires_seconds
+
257 (xp
->curlft
.use_time
? : xp
->curlft
.add_time
) - now
;
263 if (xp
->lft
.soft_add_expires_seconds
) {
264 long tmo
= xp
->lft
.soft_add_expires_seconds
+
265 xp
->curlft
.add_time
- now
;
268 tmo
= XFRM_KM_TIMEOUT
;
273 if (xp
->lft
.soft_use_expires_seconds
) {
274 long tmo
= xp
->lft
.soft_use_expires_seconds
+
275 (xp
->curlft
.use_time
? : xp
->curlft
.add_time
) - now
;
278 tmo
= XFRM_KM_TIMEOUT
;
285 km_policy_expired(xp
, dir
, 0, 0);
286 if (next
!= LONG_MAX
&&
287 !mod_timer(&xp
->timer
, jiffies
+ make_jiffies(next
)))
291 read_unlock(&xp
->lock
);
296 read_unlock(&xp
->lock
);
297 if (!xfrm_policy_delete(xp
, dir
))
298 km_policy_expired(xp
, dir
, 1, 0);
303 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
307 struct xfrm_policy
*xfrm_policy_alloc(gfp_t gfp
)
309 struct xfrm_policy
*policy
;
311 policy
= kmalloc(sizeof(struct xfrm_policy
), gfp
);
314 memset(policy
, 0, sizeof(struct xfrm_policy
));
315 atomic_set(&policy
->refcnt
, 1);
316 rwlock_init(&policy
->lock
);
317 init_timer(&policy
->timer
);
318 policy
->timer
.data
= (unsigned long)policy
;
319 policy
->timer
.function
= xfrm_policy_timer
;
323 EXPORT_SYMBOL(xfrm_policy_alloc
);
325 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
327 void __xfrm_policy_destroy(struct xfrm_policy
*policy
)
329 BUG_ON(!policy
->dead
);
331 BUG_ON(policy
->bundles
);
333 if (del_timer(&policy
->timer
))
336 security_xfrm_policy_free(policy
);
339 EXPORT_SYMBOL(__xfrm_policy_destroy
);
341 static void xfrm_policy_gc_kill(struct xfrm_policy
*policy
)
343 struct dst_entry
*dst
;
345 while ((dst
= policy
->bundles
) != NULL
) {
346 policy
->bundles
= dst
->next
;
350 if (del_timer(&policy
->timer
))
351 atomic_dec(&policy
->refcnt
);
353 if (atomic_read(&policy
->refcnt
) > 1)
356 xfrm_pol_put(policy
);
359 static void xfrm_policy_gc_task(void *data
)
361 struct xfrm_policy
*policy
;
362 struct list_head
*entry
, *tmp
;
363 struct list_head gc_list
= LIST_HEAD_INIT(gc_list
);
365 spin_lock_bh(&xfrm_policy_gc_lock
);
366 list_splice_init(&xfrm_policy_gc_list
, &gc_list
);
367 spin_unlock_bh(&xfrm_policy_gc_lock
);
369 list_for_each_safe(entry
, tmp
, &gc_list
) {
370 policy
= list_entry(entry
, struct xfrm_policy
, list
);
371 xfrm_policy_gc_kill(policy
);
375 /* Rule must be locked. Release descentant resources, announce
376 * entry dead. The rule must be unlinked from lists to the moment.
379 static void xfrm_policy_kill(struct xfrm_policy
*policy
)
383 write_lock_bh(&policy
->lock
);
386 write_unlock_bh(&policy
->lock
);
388 if (unlikely(dead
)) {
393 spin_lock(&xfrm_policy_gc_lock
);
394 list_add(&policy
->list
, &xfrm_policy_gc_list
);
395 spin_unlock(&xfrm_policy_gc_lock
);
397 schedule_work(&xfrm_policy_gc_work
);
400 /* Generate new index... KAME seems to generate them ordered by cost
401 * of an absolute inpredictability of ordering of rules. This will not pass. */
402 static u32
xfrm_gen_index(int dir
)
405 struct xfrm_policy
*p
;
406 static u32 idx_generator
;
409 idx
= (idx_generator
| dir
);
413 for (p
= xfrm_policy_list
[dir
]; p
; p
= p
->next
) {
422 int xfrm_policy_insert(int dir
, struct xfrm_policy
*policy
, int excl
)
424 struct xfrm_policy
*pol
, **p
;
425 struct xfrm_policy
*delpol
= NULL
;
426 struct xfrm_policy
**newpos
= NULL
;
427 struct dst_entry
*gc_list
;
429 write_lock_bh(&xfrm_policy_lock
);
430 for (p
= &xfrm_policy_list
[dir
]; (pol
=*p
)!=NULL
;) {
431 if (!delpol
&& memcmp(&policy
->selector
, &pol
->selector
, sizeof(pol
->selector
)) == 0 &&
432 xfrm_sec_ctx_match(pol
->security
, policy
->security
)) {
434 write_unlock_bh(&xfrm_policy_lock
);
439 if (policy
->priority
> pol
->priority
)
441 } else if (policy
->priority
>= pol
->priority
) {
453 xfrm_pol_hold(policy
);
456 atomic_inc(&flow_cache_genid
);
457 policy
->index
= delpol
? delpol
->index
: xfrm_gen_index(dir
);
458 policy
->curlft
.add_time
= (unsigned long)xtime
.tv_sec
;
459 policy
->curlft
.use_time
= 0;
460 if (!mod_timer(&policy
->timer
, jiffies
+ HZ
))
461 xfrm_pol_hold(policy
);
462 write_unlock_bh(&xfrm_policy_lock
);
465 xfrm_policy_kill(delpol
);
467 read_lock_bh(&xfrm_policy_lock
);
469 for (policy
= policy
->next
; policy
; policy
= policy
->next
) {
470 struct dst_entry
*dst
;
472 write_lock(&policy
->lock
);
473 dst
= policy
->bundles
;
475 struct dst_entry
*tail
= dst
;
478 tail
->next
= gc_list
;
481 policy
->bundles
= NULL
;
483 write_unlock(&policy
->lock
);
485 read_unlock_bh(&xfrm_policy_lock
);
488 struct dst_entry
*dst
= gc_list
;
496 EXPORT_SYMBOL(xfrm_policy_insert
);
498 struct xfrm_policy
*xfrm_policy_bysel_ctx(int dir
, struct xfrm_selector
*sel
,
499 struct xfrm_sec_ctx
*ctx
, int delete)
501 struct xfrm_policy
*pol
, **p
;
503 write_lock_bh(&xfrm_policy_lock
);
504 for (p
= &xfrm_policy_list
[dir
]; (pol
=*p
)!=NULL
; p
= &pol
->next
) {
505 if ((memcmp(sel
, &pol
->selector
, sizeof(*sel
)) == 0) &&
506 (xfrm_sec_ctx_match(ctx
, pol
->security
))) {
513 write_unlock_bh(&xfrm_policy_lock
);
516 atomic_inc(&flow_cache_genid
);
517 xfrm_policy_kill(pol
);
521 EXPORT_SYMBOL(xfrm_policy_bysel_ctx
);
523 struct xfrm_policy
*xfrm_policy_byid(int dir
, u32 id
, int delete)
525 struct xfrm_policy
*pol
, **p
;
527 write_lock_bh(&xfrm_policy_lock
);
528 for (p
= &xfrm_policy_list
[dir
]; (pol
=*p
)!=NULL
; p
= &pol
->next
) {
529 if (pol
->index
== id
) {
536 write_unlock_bh(&xfrm_policy_lock
);
539 atomic_inc(&flow_cache_genid
);
540 xfrm_policy_kill(pol
);
544 EXPORT_SYMBOL(xfrm_policy_byid
);
546 void xfrm_policy_flush(void)
548 struct xfrm_policy
*xp
;
551 write_lock_bh(&xfrm_policy_lock
);
552 for (dir
= 0; dir
< XFRM_POLICY_MAX
; dir
++) {
553 while ((xp
= xfrm_policy_list
[dir
]) != NULL
) {
554 xfrm_policy_list
[dir
] = xp
->next
;
555 write_unlock_bh(&xfrm_policy_lock
);
557 xfrm_policy_kill(xp
);
559 write_lock_bh(&xfrm_policy_lock
);
562 atomic_inc(&flow_cache_genid
);
563 write_unlock_bh(&xfrm_policy_lock
);
565 EXPORT_SYMBOL(xfrm_policy_flush
);
567 int xfrm_policy_walk(int (*func
)(struct xfrm_policy
*, int, int, void*),
570 struct xfrm_policy
*xp
;
575 read_lock_bh(&xfrm_policy_lock
);
576 for (dir
= 0; dir
< 2*XFRM_POLICY_MAX
; dir
++) {
577 for (xp
= xfrm_policy_list
[dir
]; xp
; xp
= xp
->next
)
586 for (dir
= 0; dir
< 2*XFRM_POLICY_MAX
; dir
++) {
587 for (xp
= xfrm_policy_list
[dir
]; xp
; xp
= xp
->next
) {
588 error
= func(xp
, dir
%XFRM_POLICY_MAX
, --count
, data
);
595 read_unlock_bh(&xfrm_policy_lock
);
598 EXPORT_SYMBOL(xfrm_policy_walk
);
600 /* Find policy to apply to this flow. */
602 static void xfrm_policy_lookup(struct flowi
*fl
, u32 sk_sid
, u16 family
, u8 dir
,
603 void **objp
, atomic_t
**obj_refp
)
605 struct xfrm_policy
*pol
;
607 read_lock_bh(&xfrm_policy_lock
);
608 for (pol
= xfrm_policy_list
[dir
]; pol
; pol
= pol
->next
) {
609 struct xfrm_selector
*sel
= &pol
->selector
;
612 if (pol
->family
!= family
)
615 match
= xfrm_selector_match(sel
, fl
, family
);
618 if (!security_xfrm_policy_lookup(pol
, sk_sid
, dir
)) {
624 read_unlock_bh(&xfrm_policy_lock
);
625 if ((*objp
= (void *) pol
) != NULL
)
626 *obj_refp
= &pol
->refcnt
;
629 static inline int policy_to_flow_dir(int dir
)
631 if (XFRM_POLICY_IN
== FLOW_DIR_IN
&&
632 XFRM_POLICY_OUT
== FLOW_DIR_OUT
&&
633 XFRM_POLICY_FWD
== FLOW_DIR_FWD
)
639 case XFRM_POLICY_OUT
:
641 case XFRM_POLICY_FWD
:
646 static struct xfrm_policy
*xfrm_sk_policy_lookup(struct sock
*sk
, int dir
, struct flowi
*fl
, u32 sk_sid
)
648 struct xfrm_policy
*pol
;
650 read_lock_bh(&xfrm_policy_lock
);
651 if ((pol
= sk
->sk_policy
[dir
]) != NULL
) {
652 int match
= xfrm_selector_match(&pol
->selector
, fl
,
657 err
= security_xfrm_policy_lookup(pol
, sk_sid
, policy_to_flow_dir(dir
));
664 read_unlock_bh(&xfrm_policy_lock
);
668 static void __xfrm_policy_link(struct xfrm_policy
*pol
, int dir
)
670 pol
->next
= xfrm_policy_list
[dir
];
671 xfrm_policy_list
[dir
] = pol
;
675 static struct xfrm_policy
*__xfrm_policy_unlink(struct xfrm_policy
*pol
,
678 struct xfrm_policy
**polp
;
680 for (polp
= &xfrm_policy_list
[dir
];
681 *polp
!= NULL
; polp
= &(*polp
)->next
) {
690 int xfrm_policy_delete(struct xfrm_policy
*pol
, int dir
)
692 write_lock_bh(&xfrm_policy_lock
);
693 pol
= __xfrm_policy_unlink(pol
, dir
);
694 write_unlock_bh(&xfrm_policy_lock
);
696 if (dir
< XFRM_POLICY_MAX
)
697 atomic_inc(&flow_cache_genid
);
698 xfrm_policy_kill(pol
);
703 EXPORT_SYMBOL(xfrm_policy_delete
);
705 int xfrm_sk_policy_insert(struct sock
*sk
, int dir
, struct xfrm_policy
*pol
)
707 struct xfrm_policy
*old_pol
;
709 write_lock_bh(&xfrm_policy_lock
);
710 old_pol
= sk
->sk_policy
[dir
];
711 sk
->sk_policy
[dir
] = pol
;
713 pol
->curlft
.add_time
= (unsigned long)xtime
.tv_sec
;
714 pol
->index
= xfrm_gen_index(XFRM_POLICY_MAX
+dir
);
715 __xfrm_policy_link(pol
, XFRM_POLICY_MAX
+dir
);
718 __xfrm_policy_unlink(old_pol
, XFRM_POLICY_MAX
+dir
);
719 write_unlock_bh(&xfrm_policy_lock
);
722 xfrm_policy_kill(old_pol
);
727 static struct xfrm_policy
*clone_policy(struct xfrm_policy
*old
, int dir
)
729 struct xfrm_policy
*newp
= xfrm_policy_alloc(GFP_ATOMIC
);
732 newp
->selector
= old
->selector
;
733 if (security_xfrm_policy_clone(old
, newp
)) {
735 return NULL
; /* ENOMEM */
737 newp
->lft
= old
->lft
;
738 newp
->curlft
= old
->curlft
;
739 newp
->action
= old
->action
;
740 newp
->flags
= old
->flags
;
741 newp
->xfrm_nr
= old
->xfrm_nr
;
742 newp
->index
= old
->index
;
743 memcpy(newp
->xfrm_vec
, old
->xfrm_vec
,
744 newp
->xfrm_nr
*sizeof(struct xfrm_tmpl
));
745 write_lock_bh(&xfrm_policy_lock
);
746 __xfrm_policy_link(newp
, XFRM_POLICY_MAX
+dir
);
747 write_unlock_bh(&xfrm_policy_lock
);
753 int __xfrm_sk_clone_policy(struct sock
*sk
)
755 struct xfrm_policy
*p0
= sk
->sk_policy
[0],
756 *p1
= sk
->sk_policy
[1];
758 sk
->sk_policy
[0] = sk
->sk_policy
[1] = NULL
;
759 if (p0
&& (sk
->sk_policy
[0] = clone_policy(p0
, 0)) == NULL
)
761 if (p1
&& (sk
->sk_policy
[1] = clone_policy(p1
, 1)) == NULL
)
766 /* Resolve list of templates for the flow, given policy. */
769 xfrm_tmpl_resolve(struct xfrm_policy
*policy
, struct flowi
*fl
,
770 struct xfrm_state
**xfrm
,
771 unsigned short family
)
775 xfrm_address_t
*daddr
= xfrm_flowi_daddr(fl
, family
);
776 xfrm_address_t
*saddr
= xfrm_flowi_saddr(fl
, family
);
778 for (nx
=0, i
= 0; i
< policy
->xfrm_nr
; i
++) {
779 struct xfrm_state
*x
;
780 xfrm_address_t
*remote
= daddr
;
781 xfrm_address_t
*local
= saddr
;
782 struct xfrm_tmpl
*tmpl
= &policy
->xfrm_vec
[i
];
785 remote
= &tmpl
->id
.daddr
;
786 local
= &tmpl
->saddr
;
789 x
= xfrm_state_find(remote
, local
, fl
, tmpl
, policy
, &error
, family
);
791 if (x
&& x
->km
.state
== XFRM_STATE_VALID
) {
798 error
= (x
->km
.state
== XFRM_STATE_ERROR
?
809 for (nx
--; nx
>=0; nx
--)
810 xfrm_state_put(xfrm
[nx
]);
814 /* Check that the bundle accepts the flow and its components are
818 static struct dst_entry
*
819 xfrm_find_bundle(struct flowi
*fl
, struct xfrm_policy
*policy
, unsigned short family
)
822 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
823 if (unlikely(afinfo
== NULL
))
824 return ERR_PTR(-EINVAL
);
825 x
= afinfo
->find_bundle(fl
, policy
);
826 xfrm_policy_put_afinfo(afinfo
);
830 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
831 * all the metrics... Shortly, bundle a bundle.
835 xfrm_bundle_create(struct xfrm_policy
*policy
, struct xfrm_state
**xfrm
, int nx
,
836 struct flowi
*fl
, struct dst_entry
**dst_p
,
837 unsigned short family
)
840 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
841 if (unlikely(afinfo
== NULL
))
843 err
= afinfo
->bundle_create(policy
, xfrm
, nx
, fl
, dst_p
);
844 xfrm_policy_put_afinfo(afinfo
);
849 static int stale_bundle(struct dst_entry
*dst
);
851 /* Main function: finds/creates a bundle for given flow.
853 * At the moment we eat a raw IP route. Mostly to speed up lookups
854 * on interfaces with disabled IPsec.
856 int xfrm_lookup(struct dst_entry
**dst_p
, struct flowi
*fl
,
857 struct sock
*sk
, int flags
)
859 struct xfrm_policy
*policy
;
860 struct xfrm_state
*xfrm
[XFRM_MAX_DEPTH
];
861 struct dst_entry
*dst
, *dst_orig
= *dst_p
;
866 u8 dir
= policy_to_flow_dir(XFRM_POLICY_OUT
);
867 u32 sk_sid
= security_sk_sid(sk
, fl
, dir
);
869 genid
= atomic_read(&flow_cache_genid
);
871 if (sk
&& sk
->sk_policy
[1])
872 policy
= xfrm_sk_policy_lookup(sk
, XFRM_POLICY_OUT
, fl
, sk_sid
);
875 /* To accelerate a bit... */
876 if ((dst_orig
->flags
& DST_NOXFRM
) || !xfrm_policy_list
[XFRM_POLICY_OUT
])
879 policy
= flow_cache_lookup(fl
, sk_sid
, dst_orig
->ops
->family
,
880 dir
, xfrm_policy_lookup
);
886 family
= dst_orig
->ops
->family
;
887 policy
->curlft
.use_time
= (unsigned long)xtime
.tv_sec
;
889 switch (policy
->action
) {
890 case XFRM_POLICY_BLOCK
:
891 /* Prohibit the flow */
895 case XFRM_POLICY_ALLOW
:
896 if (policy
->xfrm_nr
== 0) {
897 /* Flow passes not transformed. */
898 xfrm_pol_put(policy
);
902 /* Try to find matching bundle.
904 * LATER: help from flow cache. It is optional, this
905 * is required only for output policy.
907 dst
= xfrm_find_bundle(fl
, policy
, family
);
916 nx
= xfrm_tmpl_resolve(policy
, fl
, xfrm
, family
);
918 if (unlikely(nx
<0)) {
920 if (err
== -EAGAIN
&& flags
) {
921 DECLARE_WAITQUEUE(wait
, current
);
923 add_wait_queue(&km_waitq
, &wait
);
924 set_current_state(TASK_INTERRUPTIBLE
);
926 set_current_state(TASK_RUNNING
);
927 remove_wait_queue(&km_waitq
, &wait
);
929 nx
= xfrm_tmpl_resolve(policy
, fl
, xfrm
, family
);
931 if (nx
== -EAGAIN
&& signal_pending(current
)) {
936 genid
!= atomic_read(&flow_cache_genid
)) {
937 xfrm_pol_put(policy
);
946 /* Flow passes not transformed. */
947 xfrm_pol_put(policy
);
952 err
= xfrm_bundle_create(policy
, xfrm
, nx
, fl
, &dst
, family
);
957 xfrm_state_put(xfrm
[i
]);
961 write_lock_bh(&policy
->lock
);
962 if (unlikely(policy
->dead
|| stale_bundle(dst
))) {
963 /* Wow! While we worked on resolving, this
964 * policy has gone. Retry. It is not paranoia,
965 * we just cannot enlist new bundle to dead object.
966 * We can't enlist stable bundles either.
968 write_unlock_bh(&policy
->lock
);
975 dst
->next
= policy
->bundles
;
976 policy
->bundles
= dst
;
978 write_unlock_bh(&policy
->lock
);
981 dst_release(dst_orig
);
982 xfrm_pol_put(policy
);
986 dst_release(dst_orig
);
987 xfrm_pol_put(policy
);
991 EXPORT_SYMBOL(xfrm_lookup
);
993 /* When skb is transformed back to its "native" form, we have to
994 * check policy restrictions. At the moment we make this in maximally
995 * stupid way. Shame on me. :-) Of course, connected sockets must
996 * have policy cached at them.
1000 xfrm_state_ok(struct xfrm_tmpl
*tmpl
, struct xfrm_state
*x
,
1001 unsigned short family
)
1003 if (xfrm_state_kern(x
))
1004 return tmpl
->optional
&& !xfrm_state_addr_cmp(tmpl
, x
, family
);
1005 return x
->id
.proto
== tmpl
->id
.proto
&&
1006 (x
->id
.spi
== tmpl
->id
.spi
|| !tmpl
->id
.spi
) &&
1007 (x
->props
.reqid
== tmpl
->reqid
|| !tmpl
->reqid
) &&
1008 x
->props
.mode
== tmpl
->mode
&&
1009 (tmpl
->aalgos
& (1<<x
->props
.aalgo
)) &&
1010 !(x
->props
.mode
&& xfrm_state_addr_cmp(tmpl
, x
, family
));
1014 xfrm_policy_ok(struct xfrm_tmpl
*tmpl
, struct sec_path
*sp
, int start
,
1015 unsigned short family
)
1019 if (tmpl
->optional
) {
1024 for (; idx
< sp
->len
; idx
++) {
1025 if (xfrm_state_ok(tmpl
, sp
->xvec
[idx
], family
))
1027 if (sp
->xvec
[idx
]->props
.mode
)
1034 xfrm_decode_session(struct sk_buff
*skb
, struct flowi
*fl
, unsigned short family
)
1036 struct xfrm_policy_afinfo
*afinfo
= xfrm_policy_get_afinfo(family
);
1038 if (unlikely(afinfo
== NULL
))
1039 return -EAFNOSUPPORT
;
1041 afinfo
->decode_session(skb
, fl
);
1042 xfrm_policy_put_afinfo(afinfo
);
1045 EXPORT_SYMBOL(xfrm_decode_session
);
1047 static inline int secpath_has_tunnel(struct sec_path
*sp
, int k
)
1049 for (; k
< sp
->len
; k
++) {
1050 if (sp
->xvec
[k
]->props
.mode
)
1057 int __xfrm_policy_check(struct sock
*sk
, int dir
, struct sk_buff
*skb
,
1058 unsigned short family
)
1060 struct xfrm_policy
*pol
;
1062 u8 fl_dir
= policy_to_flow_dir(dir
);
1065 if (xfrm_decode_session(skb
, &fl
, family
) < 0)
1067 nf_nat_decode_session(skb
, &fl
, family
);
1069 sk_sid
= security_sk_sid(sk
, &fl
, fl_dir
);
1071 /* First, check used SA against their selectors. */
1075 for (i
=skb
->sp
->len
-1; i
>=0; i
--) {
1076 struct xfrm_state
*x
= skb
->sp
->xvec
[i
];
1077 if (!xfrm_selector_match(&x
->sel
, &fl
, family
))
1083 if (sk
&& sk
->sk_policy
[dir
])
1084 pol
= xfrm_sk_policy_lookup(sk
, dir
, &fl
, sk_sid
);
1087 pol
= flow_cache_lookup(&fl
, sk_sid
, family
, fl_dir
,
1088 xfrm_policy_lookup
);
1091 return !skb
->sp
|| !secpath_has_tunnel(skb
->sp
, 0);
1093 pol
->curlft
.use_time
= (unsigned long)xtime
.tv_sec
;
1095 if (pol
->action
== XFRM_POLICY_ALLOW
) {
1096 struct sec_path
*sp
;
1097 static struct sec_path dummy
;
1100 if ((sp
= skb
->sp
) == NULL
)
1103 /* For each tunnel xfrm, find the first matching tmpl.
1104 * For each tmpl before that, find corresponding xfrm.
1105 * Order is _important_. Later we will implement
1106 * some barriers, but at the moment barriers
1107 * are implied between each two transformations.
1109 for (i
= pol
->xfrm_nr
-1, k
= 0; i
>= 0; i
--) {
1110 k
= xfrm_policy_ok(pol
->xfrm_vec
+i
, sp
, k
, family
);
1115 if (secpath_has_tunnel(sp
, k
))
1126 EXPORT_SYMBOL(__xfrm_policy_check
);
1128 int __xfrm_route_forward(struct sk_buff
*skb
, unsigned short family
)
1132 if (xfrm_decode_session(skb
, &fl
, family
) < 0)
1135 return xfrm_lookup(&skb
->dst
, &fl
, NULL
, 0) == 0;
1137 EXPORT_SYMBOL(__xfrm_route_forward
);
1139 static struct dst_entry
*xfrm_dst_check(struct dst_entry
*dst
, u32 cookie
)
1141 /* If it is marked obsolete, which is how we even get here,
1142 * then we have purged it from the policy bundle list and we
1143 * did that for a good reason.
1148 static int stale_bundle(struct dst_entry
*dst
)
1150 return !xfrm_bundle_ok((struct xfrm_dst
*)dst
, NULL
, AF_UNSPEC
);
1153 void xfrm_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
)
1155 while ((dst
= dst
->child
) && dst
->xfrm
&& dst
->dev
== dev
) {
1156 dst
->dev
= &loopback_dev
;
1157 dev_hold(&loopback_dev
);
1161 EXPORT_SYMBOL(xfrm_dst_ifdown
);
1163 static void xfrm_link_failure(struct sk_buff
*skb
)
1165 /* Impossible. Such dst must be popped before reaches point of failure. */
1169 static struct dst_entry
*xfrm_negative_advice(struct dst_entry
*dst
)
1172 if (dst
->obsolete
) {
1180 static void xfrm_prune_bundles(int (*func
)(struct dst_entry
*))
1183 struct xfrm_policy
*pol
;
1184 struct dst_entry
*dst
, **dstp
, *gc_list
= NULL
;
1186 read_lock_bh(&xfrm_policy_lock
);
1187 for (i
=0; i
<2*XFRM_POLICY_MAX
; i
++) {
1188 for (pol
= xfrm_policy_list
[i
]; pol
; pol
= pol
->next
) {
1189 write_lock(&pol
->lock
);
1190 dstp
= &pol
->bundles
;
1191 while ((dst
=*dstp
) != NULL
) {
1194 dst
->next
= gc_list
;
1200 write_unlock(&pol
->lock
);
1203 read_unlock_bh(&xfrm_policy_lock
);
1207 gc_list
= dst
->next
;
1212 static int unused_bundle(struct dst_entry
*dst
)
1214 return !atomic_read(&dst
->__refcnt
);
1217 static void __xfrm_garbage_collect(void)
1219 xfrm_prune_bundles(unused_bundle
);
1222 int xfrm_flush_bundles(void)
1224 xfrm_prune_bundles(stale_bundle
);
1228 static int always_true(struct dst_entry
*dst
)
1233 void xfrm_flush_all_bundles(void)
1235 xfrm_prune_bundles(always_true
);
1238 void xfrm_init_pmtu(struct dst_entry
*dst
)
1241 struct xfrm_dst
*xdst
= (struct xfrm_dst
*)dst
;
1242 u32 pmtu
, route_mtu_cached
;
1244 pmtu
= dst_mtu(dst
->child
);
1245 xdst
->child_mtu_cached
= pmtu
;
1247 pmtu
= xfrm_state_mtu(dst
->xfrm
, pmtu
);
1249 route_mtu_cached
= dst_mtu(xdst
->route
);
1250 xdst
->route_mtu_cached
= route_mtu_cached
;
1252 if (pmtu
> route_mtu_cached
)
1253 pmtu
= route_mtu_cached
;
1255 dst
->metrics
[RTAX_MTU
-1] = pmtu
;
1256 } while ((dst
= dst
->next
));
1259 EXPORT_SYMBOL(xfrm_init_pmtu
);
1261 /* Check that the bundle accepts the flow and its components are
1265 int xfrm_bundle_ok(struct xfrm_dst
*first
, struct flowi
*fl
, int family
)
1267 struct dst_entry
*dst
= &first
->u
.dst
;
1268 struct xfrm_dst
*last
;
1271 if (!dst_check(dst
->path
, ((struct xfrm_dst
*)dst
)->path_cookie
) ||
1272 (dst
->dev
&& !netif_running(dst
->dev
)))
1278 struct xfrm_dst
*xdst
= (struct xfrm_dst
*)dst
;
1280 if (fl
&& !xfrm_selector_match(&dst
->xfrm
->sel
, fl
, family
))
1282 if (dst
->xfrm
->km
.state
!= XFRM_STATE_VALID
)
1285 mtu
= dst_mtu(dst
->child
);
1286 if (xdst
->child_mtu_cached
!= mtu
) {
1288 xdst
->child_mtu_cached
= mtu
;
1291 if (!dst_check(xdst
->route
, xdst
->route_cookie
))
1293 mtu
= dst_mtu(xdst
->route
);
1294 if (xdst
->route_mtu_cached
!= mtu
) {
1296 xdst
->route_mtu_cached
= mtu
;
1300 } while (dst
->xfrm
);
1305 mtu
= last
->child_mtu_cached
;
1309 mtu
= xfrm_state_mtu(dst
->xfrm
, mtu
);
1310 if (mtu
> last
->route_mtu_cached
)
1311 mtu
= last
->route_mtu_cached
;
1312 dst
->metrics
[RTAX_MTU
-1] = mtu
;
1317 last
= last
->u
.next
;
1318 last
->child_mtu_cached
= mtu
;
1324 EXPORT_SYMBOL(xfrm_bundle_ok
);
1326 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1329 if (unlikely(afinfo
== NULL
))
1331 if (unlikely(afinfo
->family
>= NPROTO
))
1332 return -EAFNOSUPPORT
;
1333 write_lock_bh(&xfrm_policy_afinfo_lock
);
1334 if (unlikely(xfrm_policy_afinfo
[afinfo
->family
] != NULL
))
1337 struct dst_ops
*dst_ops
= afinfo
->dst_ops
;
1338 if (likely(dst_ops
->kmem_cachep
== NULL
))
1339 dst_ops
->kmem_cachep
= xfrm_dst_cache
;
1340 if (likely(dst_ops
->check
== NULL
))
1341 dst_ops
->check
= xfrm_dst_check
;
1342 if (likely(dst_ops
->negative_advice
== NULL
))
1343 dst_ops
->negative_advice
= xfrm_negative_advice
;
1344 if (likely(dst_ops
->link_failure
== NULL
))
1345 dst_ops
->link_failure
= xfrm_link_failure
;
1346 if (likely(afinfo
->garbage_collect
== NULL
))
1347 afinfo
->garbage_collect
= __xfrm_garbage_collect
;
1348 xfrm_policy_afinfo
[afinfo
->family
] = afinfo
;
1350 write_unlock_bh(&xfrm_policy_afinfo_lock
);
1353 EXPORT_SYMBOL(xfrm_policy_register_afinfo
);
1355 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1358 if (unlikely(afinfo
== NULL
))
1360 if (unlikely(afinfo
->family
>= NPROTO
))
1361 return -EAFNOSUPPORT
;
1362 write_lock_bh(&xfrm_policy_afinfo_lock
);
1363 if (likely(xfrm_policy_afinfo
[afinfo
->family
] != NULL
)) {
1364 if (unlikely(xfrm_policy_afinfo
[afinfo
->family
] != afinfo
))
1367 struct dst_ops
*dst_ops
= afinfo
->dst_ops
;
1368 xfrm_policy_afinfo
[afinfo
->family
] = NULL
;
1369 dst_ops
->kmem_cachep
= NULL
;
1370 dst_ops
->check
= NULL
;
1371 dst_ops
->negative_advice
= NULL
;
1372 dst_ops
->link_failure
= NULL
;
1373 afinfo
->garbage_collect
= NULL
;
1376 write_unlock_bh(&xfrm_policy_afinfo_lock
);
1379 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo
);
1381 static struct xfrm_policy_afinfo
*xfrm_policy_get_afinfo(unsigned short family
)
1383 struct xfrm_policy_afinfo
*afinfo
;
1384 if (unlikely(family
>= NPROTO
))
1386 read_lock(&xfrm_policy_afinfo_lock
);
1387 afinfo
= xfrm_policy_afinfo
[family
];
1388 if (unlikely(!afinfo
))
1389 read_unlock(&xfrm_policy_afinfo_lock
);
1393 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1395 read_unlock(&xfrm_policy_afinfo_lock
);
1398 static struct xfrm_policy_afinfo
*xfrm_policy_lock_afinfo(unsigned int family
)
1400 struct xfrm_policy_afinfo
*afinfo
;
1401 if (unlikely(family
>= NPROTO
))
1403 write_lock_bh(&xfrm_policy_afinfo_lock
);
1404 afinfo
= xfrm_policy_afinfo
[family
];
1405 if (unlikely(!afinfo
))
1406 write_unlock_bh(&xfrm_policy_afinfo_lock
);
1410 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo
*afinfo
)
1412 write_unlock_bh(&xfrm_policy_afinfo_lock
);
1415 static int xfrm_dev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
1419 xfrm_flush_bundles();
1424 static struct notifier_block xfrm_dev_notifier
= {
1430 static void __init
xfrm_policy_init(void)
1432 xfrm_dst_cache
= kmem_cache_create("xfrm_dst_cache",
1433 sizeof(struct xfrm_dst
),
1434 0, SLAB_HWCACHE_ALIGN
,
1436 if (!xfrm_dst_cache
)
1437 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1439 INIT_WORK(&xfrm_policy_gc_work
, xfrm_policy_gc_task
, NULL
);
1440 register_netdevice_notifier(&xfrm_dev_notifier
);
1443 void __init
xfrm_init(void)