[PATCH] skge: use NAPI for transmit complete
[linux-2.6/verdex.git] / net / xfrm / xfrm_policy.c
blob3da67ca2c3ce9890b9774e4d2757d3f33928c493
1 /*
2 * xfrm_policy.c
4 * Changes:
5 * Mitsuru KANDA @USAGI
6 * Kazunori MIYAZAWA @USAGI
7 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8 * IPv6 support
9 * Kazunori MIYAZAWA @USAGI
10 * YOSHIFUJI Hideaki
11 * Split up af-specific portion
12 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
16 #include <linux/slab.h>
17 #include <linux/kmod.h>
18 #include <linux/list.h>
19 #include <linux/spinlock.h>
20 #include <linux/workqueue.h>
21 #include <linux/notifier.h>
22 #include <linux/netdevice.h>
23 #include <linux/netfilter.h>
24 #include <linux/module.h>
25 #include <net/xfrm.h>
26 #include <net/ip.h>
28 DEFINE_MUTEX(xfrm_cfg_mutex);
29 EXPORT_SYMBOL(xfrm_cfg_mutex);
31 static DEFINE_RWLOCK(xfrm_policy_lock);
33 struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2];
34 EXPORT_SYMBOL(xfrm_policy_list);
36 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
37 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
39 static kmem_cache_t *xfrm_dst_cache __read_mostly;
41 static struct work_struct xfrm_policy_gc_work;
42 static struct list_head xfrm_policy_gc_list =
43 LIST_HEAD_INIT(xfrm_policy_gc_list);
44 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
46 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
47 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
48 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
49 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
51 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
53 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
54 struct xfrm_type **typemap;
55 int err = 0;
57 if (unlikely(afinfo == NULL))
58 return -EAFNOSUPPORT;
59 typemap = afinfo->type_map;
61 if (likely(typemap[type->proto] == NULL))
62 typemap[type->proto] = type;
63 else
64 err = -EEXIST;
65 xfrm_policy_unlock_afinfo(afinfo);
66 return err;
68 EXPORT_SYMBOL(xfrm_register_type);
70 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
72 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
73 struct xfrm_type **typemap;
74 int err = 0;
76 if (unlikely(afinfo == NULL))
77 return -EAFNOSUPPORT;
78 typemap = afinfo->type_map;
80 if (unlikely(typemap[type->proto] != type))
81 err = -ENOENT;
82 else
83 typemap[type->proto] = NULL;
84 xfrm_policy_unlock_afinfo(afinfo);
85 return err;
87 EXPORT_SYMBOL(xfrm_unregister_type);
89 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
91 struct xfrm_policy_afinfo *afinfo;
92 struct xfrm_type **typemap;
93 struct xfrm_type *type;
94 int modload_attempted = 0;
96 retry:
97 afinfo = xfrm_policy_get_afinfo(family);
98 if (unlikely(afinfo == NULL))
99 return NULL;
100 typemap = afinfo->type_map;
102 type = typemap[proto];
103 if (unlikely(type && !try_module_get(type->owner)))
104 type = NULL;
105 if (!type && !modload_attempted) {
106 xfrm_policy_put_afinfo(afinfo);
107 request_module("xfrm-type-%d-%d",
108 (int) family, (int) proto);
109 modload_attempted = 1;
110 goto retry;
113 xfrm_policy_put_afinfo(afinfo);
114 return type;
117 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
118 unsigned short family)
120 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
121 int err = 0;
123 if (unlikely(afinfo == NULL))
124 return -EAFNOSUPPORT;
126 if (likely(afinfo->dst_lookup != NULL))
127 err = afinfo->dst_lookup(dst, fl);
128 else
129 err = -EINVAL;
130 xfrm_policy_put_afinfo(afinfo);
131 return err;
133 EXPORT_SYMBOL(xfrm_dst_lookup);
135 void xfrm_put_type(struct xfrm_type *type)
137 module_put(type->owner);
140 int xfrm_register_mode(struct xfrm_mode *mode, int family)
142 struct xfrm_policy_afinfo *afinfo;
143 struct xfrm_mode **modemap;
144 int err;
146 if (unlikely(mode->encap >= XFRM_MODE_MAX))
147 return -EINVAL;
149 afinfo = xfrm_policy_lock_afinfo(family);
150 if (unlikely(afinfo == NULL))
151 return -EAFNOSUPPORT;
153 err = -EEXIST;
154 modemap = afinfo->mode_map;
155 if (likely(modemap[mode->encap] == NULL)) {
156 modemap[mode->encap] = mode;
157 err = 0;
160 xfrm_policy_unlock_afinfo(afinfo);
161 return err;
163 EXPORT_SYMBOL(xfrm_register_mode);
165 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
167 struct xfrm_policy_afinfo *afinfo;
168 struct xfrm_mode **modemap;
169 int err;
171 if (unlikely(mode->encap >= XFRM_MODE_MAX))
172 return -EINVAL;
174 afinfo = xfrm_policy_lock_afinfo(family);
175 if (unlikely(afinfo == NULL))
176 return -EAFNOSUPPORT;
178 err = -ENOENT;
179 modemap = afinfo->mode_map;
180 if (likely(modemap[mode->encap] == mode)) {
181 modemap[mode->encap] = NULL;
182 err = 0;
185 xfrm_policy_unlock_afinfo(afinfo);
186 return err;
188 EXPORT_SYMBOL(xfrm_unregister_mode);
190 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
192 struct xfrm_policy_afinfo *afinfo;
193 struct xfrm_mode *mode;
194 int modload_attempted = 0;
196 if (unlikely(encap >= XFRM_MODE_MAX))
197 return NULL;
199 retry:
200 afinfo = xfrm_policy_get_afinfo(family);
201 if (unlikely(afinfo == NULL))
202 return NULL;
204 mode = afinfo->mode_map[encap];
205 if (unlikely(mode && !try_module_get(mode->owner)))
206 mode = NULL;
207 if (!mode && !modload_attempted) {
208 xfrm_policy_put_afinfo(afinfo);
209 request_module("xfrm-mode-%d-%d", family, encap);
210 modload_attempted = 1;
211 goto retry;
214 xfrm_policy_put_afinfo(afinfo);
215 return mode;
218 void xfrm_put_mode(struct xfrm_mode *mode)
220 module_put(mode->owner);
223 static inline unsigned long make_jiffies(long secs)
225 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
226 return MAX_SCHEDULE_TIMEOUT-1;
227 else
228 return secs*HZ;
231 static void xfrm_policy_timer(unsigned long data)
233 struct xfrm_policy *xp = (struct xfrm_policy*)data;
234 unsigned long now = (unsigned long)xtime.tv_sec;
235 long next = LONG_MAX;
236 int warn = 0;
237 int dir;
239 read_lock(&xp->lock);
241 if (xp->dead)
242 goto out;
244 dir = xfrm_policy_id2dir(xp->index);
246 if (xp->lft.hard_add_expires_seconds) {
247 long tmo = xp->lft.hard_add_expires_seconds +
248 xp->curlft.add_time - now;
249 if (tmo <= 0)
250 goto expired;
251 if (tmo < next)
252 next = tmo;
254 if (xp->lft.hard_use_expires_seconds) {
255 long tmo = xp->lft.hard_use_expires_seconds +
256 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
257 if (tmo <= 0)
258 goto expired;
259 if (tmo < next)
260 next = tmo;
262 if (xp->lft.soft_add_expires_seconds) {
263 long tmo = xp->lft.soft_add_expires_seconds +
264 xp->curlft.add_time - now;
265 if (tmo <= 0) {
266 warn = 1;
267 tmo = XFRM_KM_TIMEOUT;
269 if (tmo < next)
270 next = tmo;
272 if (xp->lft.soft_use_expires_seconds) {
273 long tmo = xp->lft.soft_use_expires_seconds +
274 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
275 if (tmo <= 0) {
276 warn = 1;
277 tmo = XFRM_KM_TIMEOUT;
279 if (tmo < next)
280 next = tmo;
283 if (warn)
284 km_policy_expired(xp, dir, 0, 0);
285 if (next != LONG_MAX &&
286 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
287 xfrm_pol_hold(xp);
289 out:
290 read_unlock(&xp->lock);
291 xfrm_pol_put(xp);
292 return;
294 expired:
295 read_unlock(&xp->lock);
296 if (!xfrm_policy_delete(xp, dir))
297 km_policy_expired(xp, dir, 1, 0);
298 xfrm_pol_put(xp);
302 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
303 * SPD calls.
306 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
308 struct xfrm_policy *policy;
310 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
312 if (policy) {
313 atomic_set(&policy->refcnt, 1);
314 rwlock_init(&policy->lock);
315 init_timer(&policy->timer);
316 policy->timer.data = (unsigned long)policy;
317 policy->timer.function = xfrm_policy_timer;
319 return policy;
321 EXPORT_SYMBOL(xfrm_policy_alloc);
323 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
325 void __xfrm_policy_destroy(struct xfrm_policy *policy)
327 BUG_ON(!policy->dead);
329 BUG_ON(policy->bundles);
331 if (del_timer(&policy->timer))
332 BUG();
334 security_xfrm_policy_free(policy);
335 kfree(policy);
337 EXPORT_SYMBOL(__xfrm_policy_destroy);
339 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
341 struct dst_entry *dst;
343 while ((dst = policy->bundles) != NULL) {
344 policy->bundles = dst->next;
345 dst_free(dst);
348 if (del_timer(&policy->timer))
349 atomic_dec(&policy->refcnt);
351 if (atomic_read(&policy->refcnt) > 1)
352 flow_cache_flush();
354 xfrm_pol_put(policy);
357 static void xfrm_policy_gc_task(void *data)
359 struct xfrm_policy *policy;
360 struct list_head *entry, *tmp;
361 struct list_head gc_list = LIST_HEAD_INIT(gc_list);
363 spin_lock_bh(&xfrm_policy_gc_lock);
364 list_splice_init(&xfrm_policy_gc_list, &gc_list);
365 spin_unlock_bh(&xfrm_policy_gc_lock);
367 list_for_each_safe(entry, tmp, &gc_list) {
368 policy = list_entry(entry, struct xfrm_policy, list);
369 xfrm_policy_gc_kill(policy);
373 /* Rule must be locked. Release descentant resources, announce
374 * entry dead. The rule must be unlinked from lists to the moment.
377 static void xfrm_policy_kill(struct xfrm_policy *policy)
379 int dead;
381 write_lock_bh(&policy->lock);
382 dead = policy->dead;
383 policy->dead = 1;
384 write_unlock_bh(&policy->lock);
386 if (unlikely(dead)) {
387 WARN_ON(1);
388 return;
391 spin_lock(&xfrm_policy_gc_lock);
392 list_add(&policy->list, &xfrm_policy_gc_list);
393 spin_unlock(&xfrm_policy_gc_lock);
395 schedule_work(&xfrm_policy_gc_work);
398 /* Generate new index... KAME seems to generate them ordered by cost
399 * of an absolute inpredictability of ordering of rules. This will not pass. */
400 static u32 xfrm_gen_index(int dir)
402 u32 idx;
403 struct xfrm_policy *p;
404 static u32 idx_generator;
406 for (;;) {
407 idx = (idx_generator | dir);
408 idx_generator += 8;
409 if (idx == 0)
410 idx = 8;
411 for (p = xfrm_policy_list[dir]; p; p = p->next) {
412 if (p->index == idx)
413 break;
415 if (!p)
416 return idx;
420 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
422 struct xfrm_policy *pol, **p;
423 struct xfrm_policy *delpol = NULL;
424 struct xfrm_policy **newpos = NULL;
425 struct dst_entry *gc_list;
427 write_lock_bh(&xfrm_policy_lock);
428 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
429 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0 &&
430 xfrm_sec_ctx_match(pol->security, policy->security)) {
431 if (excl) {
432 write_unlock_bh(&xfrm_policy_lock);
433 return -EEXIST;
435 *p = pol->next;
436 delpol = pol;
437 if (policy->priority > pol->priority)
438 continue;
439 } else if (policy->priority >= pol->priority) {
440 p = &pol->next;
441 continue;
443 if (!newpos)
444 newpos = p;
445 if (delpol)
446 break;
447 p = &pol->next;
449 if (newpos)
450 p = newpos;
451 xfrm_pol_hold(policy);
452 policy->next = *p;
453 *p = policy;
454 atomic_inc(&flow_cache_genid);
455 policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
456 policy->curlft.add_time = (unsigned long)xtime.tv_sec;
457 policy->curlft.use_time = 0;
458 if (!mod_timer(&policy->timer, jiffies + HZ))
459 xfrm_pol_hold(policy);
460 write_unlock_bh(&xfrm_policy_lock);
462 if (delpol)
463 xfrm_policy_kill(delpol);
465 read_lock_bh(&xfrm_policy_lock);
466 gc_list = NULL;
467 for (policy = policy->next; policy; policy = policy->next) {
468 struct dst_entry *dst;
470 write_lock(&policy->lock);
471 dst = policy->bundles;
472 if (dst) {
473 struct dst_entry *tail = dst;
474 while (tail->next)
475 tail = tail->next;
476 tail->next = gc_list;
477 gc_list = dst;
479 policy->bundles = NULL;
481 write_unlock(&policy->lock);
483 read_unlock_bh(&xfrm_policy_lock);
485 while (gc_list) {
486 struct dst_entry *dst = gc_list;
488 gc_list = dst->next;
489 dst_free(dst);
492 return 0;
494 EXPORT_SYMBOL(xfrm_policy_insert);
496 struct xfrm_policy *xfrm_policy_bysel_ctx(int dir, struct xfrm_selector *sel,
497 struct xfrm_sec_ctx *ctx, int delete)
499 struct xfrm_policy *pol, **p;
501 write_lock_bh(&xfrm_policy_lock);
502 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
503 if ((memcmp(sel, &pol->selector, sizeof(*sel)) == 0) &&
504 (xfrm_sec_ctx_match(ctx, pol->security))) {
505 xfrm_pol_hold(pol);
506 if (delete)
507 *p = pol->next;
508 break;
511 write_unlock_bh(&xfrm_policy_lock);
513 if (pol && delete) {
514 atomic_inc(&flow_cache_genid);
515 xfrm_policy_kill(pol);
517 return pol;
519 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
521 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
523 struct xfrm_policy *pol, **p;
525 write_lock_bh(&xfrm_policy_lock);
526 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
527 if (pol->index == id) {
528 xfrm_pol_hold(pol);
529 if (delete)
530 *p = pol->next;
531 break;
534 write_unlock_bh(&xfrm_policy_lock);
536 if (pol && delete) {
537 atomic_inc(&flow_cache_genid);
538 xfrm_policy_kill(pol);
540 return pol;
542 EXPORT_SYMBOL(xfrm_policy_byid);
544 void xfrm_policy_flush(void)
546 struct xfrm_policy *xp;
547 int dir;
549 write_lock_bh(&xfrm_policy_lock);
550 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
551 while ((xp = xfrm_policy_list[dir]) != NULL) {
552 xfrm_policy_list[dir] = xp->next;
553 write_unlock_bh(&xfrm_policy_lock);
555 xfrm_policy_kill(xp);
557 write_lock_bh(&xfrm_policy_lock);
560 atomic_inc(&flow_cache_genid);
561 write_unlock_bh(&xfrm_policy_lock);
563 EXPORT_SYMBOL(xfrm_policy_flush);
565 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
566 void *data)
568 struct xfrm_policy *xp;
569 int dir;
570 int count = 0;
571 int error = 0;
573 read_lock_bh(&xfrm_policy_lock);
574 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
575 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
576 count++;
579 if (count == 0) {
580 error = -ENOENT;
581 goto out;
584 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
585 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
586 error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
587 if (error)
588 goto out;
592 out:
593 read_unlock_bh(&xfrm_policy_lock);
594 return error;
596 EXPORT_SYMBOL(xfrm_policy_walk);
598 /* Find policy to apply to this flow. */
600 static void xfrm_policy_lookup(struct flowi *fl, u32 sk_sid, u16 family, u8 dir,
601 void **objp, atomic_t **obj_refp)
603 struct xfrm_policy *pol;
605 read_lock_bh(&xfrm_policy_lock);
606 for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
607 struct xfrm_selector *sel = &pol->selector;
608 int match;
610 if (pol->family != family)
611 continue;
613 match = xfrm_selector_match(sel, fl, family);
615 if (match) {
616 if (!security_xfrm_policy_lookup(pol, sk_sid, dir)) {
617 xfrm_pol_hold(pol);
618 break;
622 read_unlock_bh(&xfrm_policy_lock);
623 if ((*objp = (void *) pol) != NULL)
624 *obj_refp = &pol->refcnt;
627 static inline int policy_to_flow_dir(int dir)
629 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
630 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
631 XFRM_POLICY_FWD == FLOW_DIR_FWD)
632 return dir;
633 switch (dir) {
634 default:
635 case XFRM_POLICY_IN:
636 return FLOW_DIR_IN;
637 case XFRM_POLICY_OUT:
638 return FLOW_DIR_OUT;
639 case XFRM_POLICY_FWD:
640 return FLOW_DIR_FWD;
644 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl, u32 sk_sid)
646 struct xfrm_policy *pol;
648 read_lock_bh(&xfrm_policy_lock);
649 if ((pol = sk->sk_policy[dir]) != NULL) {
650 int match = xfrm_selector_match(&pol->selector, fl,
651 sk->sk_family);
652 int err = 0;
654 if (match)
655 err = security_xfrm_policy_lookup(pol, sk_sid, policy_to_flow_dir(dir));
657 if (match && !err)
658 xfrm_pol_hold(pol);
659 else
660 pol = NULL;
662 read_unlock_bh(&xfrm_policy_lock);
663 return pol;
666 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
668 pol->next = xfrm_policy_list[dir];
669 xfrm_policy_list[dir] = pol;
670 xfrm_pol_hold(pol);
673 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
674 int dir)
676 struct xfrm_policy **polp;
678 for (polp = &xfrm_policy_list[dir];
679 *polp != NULL; polp = &(*polp)->next) {
680 if (*polp == pol) {
681 *polp = pol->next;
682 return pol;
685 return NULL;
688 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
690 write_lock_bh(&xfrm_policy_lock);
691 pol = __xfrm_policy_unlink(pol, dir);
692 write_unlock_bh(&xfrm_policy_lock);
693 if (pol) {
694 if (dir < XFRM_POLICY_MAX)
695 atomic_inc(&flow_cache_genid);
696 xfrm_policy_kill(pol);
697 return 0;
699 return -ENOENT;
701 EXPORT_SYMBOL(xfrm_policy_delete);
703 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
705 struct xfrm_policy *old_pol;
707 write_lock_bh(&xfrm_policy_lock);
708 old_pol = sk->sk_policy[dir];
709 sk->sk_policy[dir] = pol;
710 if (pol) {
711 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
712 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
713 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
715 if (old_pol)
716 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
717 write_unlock_bh(&xfrm_policy_lock);
719 if (old_pol) {
720 xfrm_policy_kill(old_pol);
722 return 0;
725 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
727 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
729 if (newp) {
730 newp->selector = old->selector;
731 if (security_xfrm_policy_clone(old, newp)) {
732 kfree(newp);
733 return NULL; /* ENOMEM */
735 newp->lft = old->lft;
736 newp->curlft = old->curlft;
737 newp->action = old->action;
738 newp->flags = old->flags;
739 newp->xfrm_nr = old->xfrm_nr;
740 newp->index = old->index;
741 memcpy(newp->xfrm_vec, old->xfrm_vec,
742 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
743 write_lock_bh(&xfrm_policy_lock);
744 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
745 write_unlock_bh(&xfrm_policy_lock);
746 xfrm_pol_put(newp);
748 return newp;
751 int __xfrm_sk_clone_policy(struct sock *sk)
753 struct xfrm_policy *p0 = sk->sk_policy[0],
754 *p1 = sk->sk_policy[1];
756 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
757 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
758 return -ENOMEM;
759 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
760 return -ENOMEM;
761 return 0;
764 /* Resolve list of templates for the flow, given policy. */
766 static int
767 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
768 struct xfrm_state **xfrm,
769 unsigned short family)
771 int nx;
772 int i, error;
773 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
774 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
776 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
777 struct xfrm_state *x;
778 xfrm_address_t *remote = daddr;
779 xfrm_address_t *local = saddr;
780 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
782 if (tmpl->mode) {
783 remote = &tmpl->id.daddr;
784 local = &tmpl->saddr;
787 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
789 if (x && x->km.state == XFRM_STATE_VALID) {
790 xfrm[nx++] = x;
791 daddr = remote;
792 saddr = local;
793 continue;
795 if (x) {
796 error = (x->km.state == XFRM_STATE_ERROR ?
797 -EINVAL : -EAGAIN);
798 xfrm_state_put(x);
801 if (!tmpl->optional)
802 goto fail;
804 return nx;
806 fail:
807 for (nx--; nx>=0; nx--)
808 xfrm_state_put(xfrm[nx]);
809 return error;
812 /* Check that the bundle accepts the flow and its components are
813 * still valid.
816 static struct dst_entry *
817 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
819 struct dst_entry *x;
820 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
821 if (unlikely(afinfo == NULL))
822 return ERR_PTR(-EINVAL);
823 x = afinfo->find_bundle(fl, policy);
824 xfrm_policy_put_afinfo(afinfo);
825 return x;
828 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
829 * all the metrics... Shortly, bundle a bundle.
832 static int
833 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
834 struct flowi *fl, struct dst_entry **dst_p,
835 unsigned short family)
837 int err;
838 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
839 if (unlikely(afinfo == NULL))
840 return -EINVAL;
841 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
842 xfrm_policy_put_afinfo(afinfo);
843 return err;
847 static int stale_bundle(struct dst_entry *dst);
849 /* Main function: finds/creates a bundle for given flow.
851 * At the moment we eat a raw IP route. Mostly to speed up lookups
852 * on interfaces with disabled IPsec.
854 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
855 struct sock *sk, int flags)
857 struct xfrm_policy *policy;
858 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
859 struct dst_entry *dst, *dst_orig = *dst_p;
860 int nx = 0;
861 int err;
862 u32 genid;
863 u16 family;
864 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
865 u32 sk_sid = security_sk_sid(sk, fl, dir);
866 restart:
867 genid = atomic_read(&flow_cache_genid);
868 policy = NULL;
869 if (sk && sk->sk_policy[1])
870 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, sk_sid);
872 if (!policy) {
873 /* To accelerate a bit... */
874 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
875 return 0;
877 policy = flow_cache_lookup(fl, sk_sid, dst_orig->ops->family,
878 dir, xfrm_policy_lookup);
881 if (!policy)
882 return 0;
884 family = dst_orig->ops->family;
885 policy->curlft.use_time = (unsigned long)xtime.tv_sec;
887 switch (policy->action) {
888 case XFRM_POLICY_BLOCK:
889 /* Prohibit the flow */
890 err = -EPERM;
891 goto error;
893 case XFRM_POLICY_ALLOW:
894 if (policy->xfrm_nr == 0) {
895 /* Flow passes not transformed. */
896 xfrm_pol_put(policy);
897 return 0;
900 /* Try to find matching bundle.
902 * LATER: help from flow cache. It is optional, this
903 * is required only for output policy.
905 dst = xfrm_find_bundle(fl, policy, family);
906 if (IS_ERR(dst)) {
907 err = PTR_ERR(dst);
908 goto error;
911 if (dst)
912 break;
914 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
916 if (unlikely(nx<0)) {
917 err = nx;
918 if (err == -EAGAIN && flags) {
919 DECLARE_WAITQUEUE(wait, current);
921 add_wait_queue(&km_waitq, &wait);
922 set_current_state(TASK_INTERRUPTIBLE);
923 schedule();
924 set_current_state(TASK_RUNNING);
925 remove_wait_queue(&km_waitq, &wait);
927 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
929 if (nx == -EAGAIN && signal_pending(current)) {
930 err = -ERESTART;
931 goto error;
933 if (nx == -EAGAIN ||
934 genid != atomic_read(&flow_cache_genid)) {
935 xfrm_pol_put(policy);
936 goto restart;
938 err = nx;
940 if (err < 0)
941 goto error;
943 if (nx == 0) {
944 /* Flow passes not transformed. */
945 xfrm_pol_put(policy);
946 return 0;
949 dst = dst_orig;
950 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
952 if (unlikely(err)) {
953 int i;
954 for (i=0; i<nx; i++)
955 xfrm_state_put(xfrm[i]);
956 goto error;
959 write_lock_bh(&policy->lock);
960 if (unlikely(policy->dead || stale_bundle(dst))) {
961 /* Wow! While we worked on resolving, this
962 * policy has gone. Retry. It is not paranoia,
963 * we just cannot enlist new bundle to dead object.
964 * We can't enlist stable bundles either.
966 write_unlock_bh(&policy->lock);
967 if (dst)
968 dst_free(dst);
970 err = -EHOSTUNREACH;
971 goto error;
973 dst->next = policy->bundles;
974 policy->bundles = dst;
975 dst_hold(dst);
976 write_unlock_bh(&policy->lock);
978 *dst_p = dst;
979 dst_release(dst_orig);
980 xfrm_pol_put(policy);
981 return 0;
983 error:
984 dst_release(dst_orig);
985 xfrm_pol_put(policy);
986 *dst_p = NULL;
987 return err;
989 EXPORT_SYMBOL(xfrm_lookup);
991 /* When skb is transformed back to its "native" form, we have to
992 * check policy restrictions. At the moment we make this in maximally
993 * stupid way. Shame on me. :-) Of course, connected sockets must
994 * have policy cached at them.
997 static inline int
998 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
999 unsigned short family)
1001 if (xfrm_state_kern(x))
1002 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1003 return x->id.proto == tmpl->id.proto &&
1004 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1005 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1006 x->props.mode == tmpl->mode &&
1007 (tmpl->aalgos & (1<<x->props.aalgo)) &&
1008 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
1011 static inline int
1012 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1013 unsigned short family)
1015 int idx = start;
1017 if (tmpl->optional) {
1018 if (!tmpl->mode)
1019 return start;
1020 } else
1021 start = -1;
1022 for (; idx < sp->len; idx++) {
1023 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1024 return ++idx;
1025 if (sp->xvec[idx]->props.mode)
1026 break;
1028 return start;
1032 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1034 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1036 if (unlikely(afinfo == NULL))
1037 return -EAFNOSUPPORT;
1039 afinfo->decode_session(skb, fl);
1040 xfrm_policy_put_afinfo(afinfo);
1041 return 0;
1043 EXPORT_SYMBOL(xfrm_decode_session);
1045 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
1047 for (; k < sp->len; k++) {
1048 if (sp->xvec[k]->props.mode)
1049 return 1;
1052 return 0;
1055 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1056 unsigned short family)
1058 struct xfrm_policy *pol;
1059 struct flowi fl;
1060 u8 fl_dir = policy_to_flow_dir(dir);
1061 u32 sk_sid;
1063 if (xfrm_decode_session(skb, &fl, family) < 0)
1064 return 0;
1065 nf_nat_decode_session(skb, &fl, family);
1067 sk_sid = security_sk_sid(sk, &fl, fl_dir);
1069 /* First, check used SA against their selectors. */
1070 if (skb->sp) {
1071 int i;
1073 for (i=skb->sp->len-1; i>=0; i--) {
1074 struct xfrm_state *x = skb->sp->xvec[i];
1075 if (!xfrm_selector_match(&x->sel, &fl, family))
1076 return 0;
1080 pol = NULL;
1081 if (sk && sk->sk_policy[dir])
1082 pol = xfrm_sk_policy_lookup(sk, dir, &fl, sk_sid);
1084 if (!pol)
1085 pol = flow_cache_lookup(&fl, sk_sid, family, fl_dir,
1086 xfrm_policy_lookup);
1088 if (!pol)
1089 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
1091 pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1093 if (pol->action == XFRM_POLICY_ALLOW) {
1094 struct sec_path *sp;
1095 static struct sec_path dummy;
1096 int i, k;
1098 if ((sp = skb->sp) == NULL)
1099 sp = &dummy;
1101 /* For each tunnel xfrm, find the first matching tmpl.
1102 * For each tmpl before that, find corresponding xfrm.
1103 * Order is _important_. Later we will implement
1104 * some barriers, but at the moment barriers
1105 * are implied between each two transformations.
1107 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
1108 k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
1109 if (k < 0)
1110 goto reject;
1113 if (secpath_has_tunnel(sp, k))
1114 goto reject;
1116 xfrm_pol_put(pol);
1117 return 1;
1120 reject:
1121 xfrm_pol_put(pol);
1122 return 0;
1124 EXPORT_SYMBOL(__xfrm_policy_check);
1126 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1128 struct flowi fl;
1130 if (xfrm_decode_session(skb, &fl, family) < 0)
1131 return 0;
1133 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1135 EXPORT_SYMBOL(__xfrm_route_forward);
1137 /* Optimize later using cookies and generation ids. */
1139 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1141 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1142 * to "-1" to force all XFRM destinations to get validated by
1143 * dst_ops->check on every use. We do this because when a
1144 * normal route referenced by an XFRM dst is obsoleted we do
1145 * not go looking around for all parent referencing XFRM dsts
1146 * so that we can invalidate them. It is just too much work.
1147 * Instead we make the checks here on every use. For example:
1149 * XFRM dst A --> IPv4 dst X
1151 * X is the "xdst->route" of A (X is also the "dst->path" of A
1152 * in this example). If X is marked obsolete, "A" will not
1153 * notice. That's what we are validating here via the
1154 * stale_bundle() check.
1156 * When a policy's bundle is pruned, we dst_free() the XFRM
1157 * dst which causes it's ->obsolete field to be set to a
1158 * positive non-zero integer. If an XFRM dst has been pruned
1159 * like this, we want to force a new route lookup.
1161 if (dst->obsolete < 0 && !stale_bundle(dst))
1162 return dst;
1164 return NULL;
1167 static int stale_bundle(struct dst_entry *dst)
1169 return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1172 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1174 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1175 dst->dev = &loopback_dev;
1176 dev_hold(&loopback_dev);
1177 dev_put(dev);
1180 EXPORT_SYMBOL(xfrm_dst_ifdown);
1182 static void xfrm_link_failure(struct sk_buff *skb)
1184 /* Impossible. Such dst must be popped before reaches point of failure. */
1185 return;
1188 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1190 if (dst) {
1191 if (dst->obsolete) {
1192 dst_release(dst);
1193 dst = NULL;
1196 return dst;
1199 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1201 int i;
1202 struct xfrm_policy *pol;
1203 struct dst_entry *dst, **dstp, *gc_list = NULL;
1205 read_lock_bh(&xfrm_policy_lock);
1206 for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1207 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1208 write_lock(&pol->lock);
1209 dstp = &pol->bundles;
1210 while ((dst=*dstp) != NULL) {
1211 if (func(dst)) {
1212 *dstp = dst->next;
1213 dst->next = gc_list;
1214 gc_list = dst;
1215 } else {
1216 dstp = &dst->next;
1219 write_unlock(&pol->lock);
1222 read_unlock_bh(&xfrm_policy_lock);
1224 while (gc_list) {
1225 dst = gc_list;
1226 gc_list = dst->next;
1227 dst_free(dst);
1231 static int unused_bundle(struct dst_entry *dst)
1233 return !atomic_read(&dst->__refcnt);
1236 static void __xfrm_garbage_collect(void)
1238 xfrm_prune_bundles(unused_bundle);
1241 int xfrm_flush_bundles(void)
1243 xfrm_prune_bundles(stale_bundle);
1244 return 0;
1247 static int always_true(struct dst_entry *dst)
1249 return 1;
1252 void xfrm_flush_all_bundles(void)
1254 xfrm_prune_bundles(always_true);
1257 void xfrm_init_pmtu(struct dst_entry *dst)
1259 do {
1260 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1261 u32 pmtu, route_mtu_cached;
1263 pmtu = dst_mtu(dst->child);
1264 xdst->child_mtu_cached = pmtu;
1266 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1268 route_mtu_cached = dst_mtu(xdst->route);
1269 xdst->route_mtu_cached = route_mtu_cached;
1271 if (pmtu > route_mtu_cached)
1272 pmtu = route_mtu_cached;
1274 dst->metrics[RTAX_MTU-1] = pmtu;
1275 } while ((dst = dst->next));
1278 EXPORT_SYMBOL(xfrm_init_pmtu);
1280 /* Check that the bundle accepts the flow and its components are
1281 * still valid.
1284 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1286 struct dst_entry *dst = &first->u.dst;
1287 struct xfrm_dst *last;
1288 u32 mtu;
1290 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1291 (dst->dev && !netif_running(dst->dev)))
1292 return 0;
1294 last = NULL;
1296 do {
1297 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1299 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1300 return 0;
1301 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1302 return 0;
1304 mtu = dst_mtu(dst->child);
1305 if (xdst->child_mtu_cached != mtu) {
1306 last = xdst;
1307 xdst->child_mtu_cached = mtu;
1310 if (!dst_check(xdst->route, xdst->route_cookie))
1311 return 0;
1312 mtu = dst_mtu(xdst->route);
1313 if (xdst->route_mtu_cached != mtu) {
1314 last = xdst;
1315 xdst->route_mtu_cached = mtu;
1318 dst = dst->child;
1319 } while (dst->xfrm);
1321 if (likely(!last))
1322 return 1;
1324 mtu = last->child_mtu_cached;
1325 for (;;) {
1326 dst = &last->u.dst;
1328 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1329 if (mtu > last->route_mtu_cached)
1330 mtu = last->route_mtu_cached;
1331 dst->metrics[RTAX_MTU-1] = mtu;
1333 if (last == first)
1334 break;
1336 last = last->u.next;
1337 last->child_mtu_cached = mtu;
1340 return 1;
1343 EXPORT_SYMBOL(xfrm_bundle_ok);
1345 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1347 int err = 0;
1348 if (unlikely(afinfo == NULL))
1349 return -EINVAL;
1350 if (unlikely(afinfo->family >= NPROTO))
1351 return -EAFNOSUPPORT;
1352 write_lock_bh(&xfrm_policy_afinfo_lock);
1353 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1354 err = -ENOBUFS;
1355 else {
1356 struct dst_ops *dst_ops = afinfo->dst_ops;
1357 if (likely(dst_ops->kmem_cachep == NULL))
1358 dst_ops->kmem_cachep = xfrm_dst_cache;
1359 if (likely(dst_ops->check == NULL))
1360 dst_ops->check = xfrm_dst_check;
1361 if (likely(dst_ops->negative_advice == NULL))
1362 dst_ops->negative_advice = xfrm_negative_advice;
1363 if (likely(dst_ops->link_failure == NULL))
1364 dst_ops->link_failure = xfrm_link_failure;
1365 if (likely(afinfo->garbage_collect == NULL))
1366 afinfo->garbage_collect = __xfrm_garbage_collect;
1367 xfrm_policy_afinfo[afinfo->family] = afinfo;
1369 write_unlock_bh(&xfrm_policy_afinfo_lock);
1370 return err;
1372 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1374 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1376 int err = 0;
1377 if (unlikely(afinfo == NULL))
1378 return -EINVAL;
1379 if (unlikely(afinfo->family >= NPROTO))
1380 return -EAFNOSUPPORT;
1381 write_lock_bh(&xfrm_policy_afinfo_lock);
1382 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1383 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1384 err = -EINVAL;
1385 else {
1386 struct dst_ops *dst_ops = afinfo->dst_ops;
1387 xfrm_policy_afinfo[afinfo->family] = NULL;
1388 dst_ops->kmem_cachep = NULL;
1389 dst_ops->check = NULL;
1390 dst_ops->negative_advice = NULL;
1391 dst_ops->link_failure = NULL;
1392 afinfo->garbage_collect = NULL;
1395 write_unlock_bh(&xfrm_policy_afinfo_lock);
1396 return err;
1398 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1400 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1402 struct xfrm_policy_afinfo *afinfo;
1403 if (unlikely(family >= NPROTO))
1404 return NULL;
1405 read_lock(&xfrm_policy_afinfo_lock);
1406 afinfo = xfrm_policy_afinfo[family];
1407 if (unlikely(!afinfo))
1408 read_unlock(&xfrm_policy_afinfo_lock);
1409 return afinfo;
1412 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1414 read_unlock(&xfrm_policy_afinfo_lock);
1417 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
1419 struct xfrm_policy_afinfo *afinfo;
1420 if (unlikely(family >= NPROTO))
1421 return NULL;
1422 write_lock_bh(&xfrm_policy_afinfo_lock);
1423 afinfo = xfrm_policy_afinfo[family];
1424 if (unlikely(!afinfo))
1425 write_unlock_bh(&xfrm_policy_afinfo_lock);
1426 return afinfo;
1429 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
1431 write_unlock_bh(&xfrm_policy_afinfo_lock);
1434 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1436 switch (event) {
1437 case NETDEV_DOWN:
1438 xfrm_flush_bundles();
1440 return NOTIFY_DONE;
1443 static struct notifier_block xfrm_dev_notifier = {
1444 xfrm_dev_event,
1445 NULL,
1449 static void __init xfrm_policy_init(void)
1451 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1452 sizeof(struct xfrm_dst),
1453 0, SLAB_HWCACHE_ALIGN,
1454 NULL, NULL);
1455 if (!xfrm_dst_cache)
1456 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1458 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1459 register_netdevice_notifier(&xfrm_dev_notifier);
1462 void __init xfrm_init(void)
1464 xfrm_state_init();
1465 xfrm_policy_init();
1466 xfrm_input_init();