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/err.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>
26 #include <linux/cache.h>
31 #include "xfrm_hash.h"
33 int sysctl_xfrm_larval_drop __read_mostly;
35 DEFINE_MUTEX(xfrm_cfg_mutex);
36 EXPORT_SYMBOL(xfrm_cfg_mutex);
38 static DEFINE_RWLOCK(xfrm_policy_lock);
40 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
41 EXPORT_SYMBOL(xfrm_policy_count);
43 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
44 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
46 static struct kmem_cache *xfrm_dst_cache __read_mostly;
48 static struct work_struct xfrm_policy_gc_work;
49 static HLIST_HEAD(xfrm_policy_gc_list);
50 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
52 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
53 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
54 static void xfrm_init_pmtu(struct dst_entry *dst);
57 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
59 return addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
60 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
61 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
62 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
63 (fl->proto == sel->proto || !sel->proto) &&
64 (fl->oif == sel->ifindex || !sel->ifindex);
68 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
70 return addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
71 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
72 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
73 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
74 (fl->proto == sel->proto || !sel->proto) &&
75 (fl->oif == sel->ifindex || !sel->ifindex);
78 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
79 unsigned short family)
83 return __xfrm4_selector_match(sel, fl);
85 return __xfrm6_selector_match(sel, fl);
90 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
93 xfrm_address_t *saddr = &x->props.saddr;
94 xfrm_address_t *daddr = &x->id.daddr;
95 struct xfrm_policy_afinfo *afinfo;
96 struct dst_entry *dst;
98 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR)
100 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR)
103 afinfo = xfrm_policy_get_afinfo(family);
104 if (unlikely(afinfo == NULL))
105 return ERR_PTR(-EAFNOSUPPORT);
107 dst = afinfo->dst_lookup(tos, saddr, daddr);
108 xfrm_policy_put_afinfo(afinfo);
112 static inline unsigned long make_jiffies(long secs)
114 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
115 return MAX_SCHEDULE_TIMEOUT-1;
120 static void xfrm_policy_timer(unsigned long data)
122 struct xfrm_policy *xp = (struct xfrm_policy*)data;
123 unsigned long now = get_seconds();
124 long next = LONG_MAX;
128 read_lock(&xp->lock);
133 dir = xfrm_policy_id2dir(xp->index);
135 if (xp->lft.hard_add_expires_seconds) {
136 long tmo = xp->lft.hard_add_expires_seconds +
137 xp->curlft.add_time - now;
143 if (xp->lft.hard_use_expires_seconds) {
144 long tmo = xp->lft.hard_use_expires_seconds +
145 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
151 if (xp->lft.soft_add_expires_seconds) {
152 long tmo = xp->lft.soft_add_expires_seconds +
153 xp->curlft.add_time - now;
156 tmo = XFRM_KM_TIMEOUT;
161 if (xp->lft.soft_use_expires_seconds) {
162 long tmo = xp->lft.soft_use_expires_seconds +
163 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
166 tmo = XFRM_KM_TIMEOUT;
173 km_policy_expired(xp, dir, 0, 0);
174 if (next != LONG_MAX &&
175 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
179 read_unlock(&xp->lock);
184 read_unlock(&xp->lock);
185 if (!xfrm_policy_delete(xp, dir))
186 km_policy_expired(xp, dir, 1, 0);
191 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
195 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
197 struct xfrm_policy *policy;
199 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
202 INIT_HLIST_NODE(&policy->bydst);
203 INIT_HLIST_NODE(&policy->byidx);
204 rwlock_init(&policy->lock);
205 atomic_set(&policy->refcnt, 1);
206 setup_timer(&policy->timer, xfrm_policy_timer,
207 (unsigned long)policy);
211 EXPORT_SYMBOL(xfrm_policy_alloc);
213 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
215 void __xfrm_policy_destroy(struct xfrm_policy *policy)
217 BUG_ON(!policy->dead);
219 BUG_ON(policy->bundles);
221 if (del_timer(&policy->timer))
224 security_xfrm_policy_free(policy);
227 EXPORT_SYMBOL(__xfrm_policy_destroy);
229 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
231 struct dst_entry *dst;
233 while ((dst = policy->bundles) != NULL) {
234 policy->bundles = dst->next;
238 if (del_timer(&policy->timer))
239 atomic_dec(&policy->refcnt);
241 if (atomic_read(&policy->refcnt) > 1)
244 xfrm_pol_put(policy);
247 static void xfrm_policy_gc_task(struct work_struct *work)
249 struct xfrm_policy *policy;
250 struct hlist_node *entry, *tmp;
251 struct hlist_head gc_list;
253 spin_lock_bh(&xfrm_policy_gc_lock);
254 gc_list.first = xfrm_policy_gc_list.first;
255 INIT_HLIST_HEAD(&xfrm_policy_gc_list);
256 spin_unlock_bh(&xfrm_policy_gc_lock);
258 hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
259 xfrm_policy_gc_kill(policy);
262 /* Rule must be locked. Release descentant resources, announce
263 * entry dead. The rule must be unlinked from lists to the moment.
266 static void xfrm_policy_kill(struct xfrm_policy *policy)
270 write_lock_bh(&policy->lock);
273 write_unlock_bh(&policy->lock);
275 if (unlikely(dead)) {
280 spin_lock(&xfrm_policy_gc_lock);
281 hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
282 spin_unlock(&xfrm_policy_gc_lock);
284 schedule_work(&xfrm_policy_gc_work);
287 struct xfrm_policy_hash {
288 struct hlist_head *table;
292 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
293 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
294 static struct hlist_head *xfrm_policy_byidx __read_mostly;
295 static unsigned int xfrm_idx_hmask __read_mostly;
296 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
298 static inline unsigned int idx_hash(u32 index)
300 return __idx_hash(index, xfrm_idx_hmask);
303 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
305 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
306 unsigned int hash = __sel_hash(sel, family, hmask);
308 return (hash == hmask + 1 ?
309 &xfrm_policy_inexact[dir] :
310 xfrm_policy_bydst[dir].table + hash);
313 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
315 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
316 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
318 return xfrm_policy_bydst[dir].table + hash;
321 static void xfrm_dst_hash_transfer(struct hlist_head *list,
322 struct hlist_head *ndsttable,
323 unsigned int nhashmask)
325 struct hlist_node *entry, *tmp;
326 struct xfrm_policy *pol;
328 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
331 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
332 pol->family, nhashmask);
333 hlist_add_head(&pol->bydst, ndsttable+h);
337 static void xfrm_idx_hash_transfer(struct hlist_head *list,
338 struct hlist_head *nidxtable,
339 unsigned int nhashmask)
341 struct hlist_node *entry, *tmp;
342 struct xfrm_policy *pol;
344 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
347 h = __idx_hash(pol->index, nhashmask);
348 hlist_add_head(&pol->byidx, nidxtable+h);
352 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
354 return ((old_hmask + 1) << 1) - 1;
357 static void xfrm_bydst_resize(int dir)
359 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
360 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
361 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
362 struct hlist_head *odst = xfrm_policy_bydst[dir].table;
363 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
369 write_lock_bh(&xfrm_policy_lock);
371 for (i = hmask; i >= 0; i--)
372 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
374 xfrm_policy_bydst[dir].table = ndst;
375 xfrm_policy_bydst[dir].hmask = nhashmask;
377 write_unlock_bh(&xfrm_policy_lock);
379 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
382 static void xfrm_byidx_resize(int total)
384 unsigned int hmask = xfrm_idx_hmask;
385 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
386 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
387 struct hlist_head *oidx = xfrm_policy_byidx;
388 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
394 write_lock_bh(&xfrm_policy_lock);
396 for (i = hmask; i >= 0; i--)
397 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
399 xfrm_policy_byidx = nidx;
400 xfrm_idx_hmask = nhashmask;
402 write_unlock_bh(&xfrm_policy_lock);
404 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
407 static inline int xfrm_bydst_should_resize(int dir, int *total)
409 unsigned int cnt = xfrm_policy_count[dir];
410 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
415 if ((hmask + 1) < xfrm_policy_hashmax &&
422 static inline int xfrm_byidx_should_resize(int total)
424 unsigned int hmask = xfrm_idx_hmask;
426 if ((hmask + 1) < xfrm_policy_hashmax &&
433 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
435 read_lock_bh(&xfrm_policy_lock);
436 si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
437 si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
438 si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
439 si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
440 si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
441 si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
442 si->spdhcnt = xfrm_idx_hmask;
443 si->spdhmcnt = xfrm_policy_hashmax;
444 read_unlock_bh(&xfrm_policy_lock);
446 EXPORT_SYMBOL(xfrm_spd_getinfo);
448 static DEFINE_MUTEX(hash_resize_mutex);
449 static void xfrm_hash_resize(struct work_struct *__unused)
453 mutex_lock(&hash_resize_mutex);
456 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
457 if (xfrm_bydst_should_resize(dir, &total))
458 xfrm_bydst_resize(dir);
460 if (xfrm_byidx_should_resize(total))
461 xfrm_byidx_resize(total);
463 mutex_unlock(&hash_resize_mutex);
466 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
468 /* Generate new index... KAME seems to generate them ordered by cost
469 * of an absolute inpredictability of ordering of rules. This will not pass. */
470 static u32 xfrm_gen_index(u8 type, int dir)
472 static u32 idx_generator;
475 struct hlist_node *entry;
476 struct hlist_head *list;
477 struct xfrm_policy *p;
481 idx = (idx_generator | dir);
485 list = xfrm_policy_byidx + idx_hash(idx);
487 hlist_for_each_entry(p, entry, list, byidx) {
488 if (p->index == idx) {
498 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
500 u32 *p1 = (u32 *) s1;
501 u32 *p2 = (u32 *) s2;
502 int len = sizeof(struct xfrm_selector) / sizeof(u32);
505 for (i = 0; i < len; i++) {
513 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
515 struct xfrm_policy *pol;
516 struct xfrm_policy *delpol;
517 struct hlist_head *chain;
518 struct hlist_node *entry, *newpos;
519 struct dst_entry *gc_list;
521 write_lock_bh(&xfrm_policy_lock);
522 chain = policy_hash_bysel(&policy->selector, policy->family, dir);
525 hlist_for_each_entry(pol, entry, chain, bydst) {
526 if (pol->type == policy->type &&
527 !selector_cmp(&pol->selector, &policy->selector) &&
528 xfrm_sec_ctx_match(pol->security, policy->security) &&
531 write_unlock_bh(&xfrm_policy_lock);
535 if (policy->priority > pol->priority)
537 } else if (policy->priority >= pol->priority) {
538 newpos = &pol->bydst;
545 hlist_add_after(newpos, &policy->bydst);
547 hlist_add_head(&policy->bydst, chain);
548 xfrm_pol_hold(policy);
549 xfrm_policy_count[dir]++;
550 atomic_inc(&flow_cache_genid);
552 hlist_del(&delpol->bydst);
553 hlist_del(&delpol->byidx);
554 xfrm_policy_count[dir]--;
556 policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
557 hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
558 policy->curlft.add_time = get_seconds();
559 policy->curlft.use_time = 0;
560 if (!mod_timer(&policy->timer, jiffies + HZ))
561 xfrm_pol_hold(policy);
562 write_unlock_bh(&xfrm_policy_lock);
565 xfrm_policy_kill(delpol);
566 else if (xfrm_bydst_should_resize(dir, NULL))
567 schedule_work(&xfrm_hash_work);
569 read_lock_bh(&xfrm_policy_lock);
571 entry = &policy->bydst;
572 hlist_for_each_entry_continue(policy, entry, bydst) {
573 struct dst_entry *dst;
575 write_lock(&policy->lock);
576 dst = policy->bundles;
578 struct dst_entry *tail = dst;
581 tail->next = gc_list;
584 policy->bundles = NULL;
586 write_unlock(&policy->lock);
588 read_unlock_bh(&xfrm_policy_lock);
591 struct dst_entry *dst = gc_list;
599 EXPORT_SYMBOL(xfrm_policy_insert);
601 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
602 struct xfrm_selector *sel,
603 struct xfrm_sec_ctx *ctx, int delete,
606 struct xfrm_policy *pol, *ret;
607 struct hlist_head *chain;
608 struct hlist_node *entry;
611 write_lock_bh(&xfrm_policy_lock);
612 chain = policy_hash_bysel(sel, sel->family, dir);
614 hlist_for_each_entry(pol, entry, chain, bydst) {
615 if (pol->type == type &&
616 !selector_cmp(sel, &pol->selector) &&
617 xfrm_sec_ctx_match(ctx, pol->security)) {
620 *err = security_xfrm_policy_delete(pol);
622 write_unlock_bh(&xfrm_policy_lock);
625 hlist_del(&pol->bydst);
626 hlist_del(&pol->byidx);
627 xfrm_policy_count[dir]--;
633 write_unlock_bh(&xfrm_policy_lock);
636 atomic_inc(&flow_cache_genid);
637 xfrm_policy_kill(ret);
641 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
643 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
646 struct xfrm_policy *pol, *ret;
647 struct hlist_head *chain;
648 struct hlist_node *entry;
651 if (xfrm_policy_id2dir(id) != dir)
655 write_lock_bh(&xfrm_policy_lock);
656 chain = xfrm_policy_byidx + idx_hash(id);
658 hlist_for_each_entry(pol, entry, chain, byidx) {
659 if (pol->type == type && pol->index == id) {
662 *err = security_xfrm_policy_delete(pol);
664 write_unlock_bh(&xfrm_policy_lock);
667 hlist_del(&pol->bydst);
668 hlist_del(&pol->byidx);
669 xfrm_policy_count[dir]--;
675 write_unlock_bh(&xfrm_policy_lock);
678 atomic_inc(&flow_cache_genid);
679 xfrm_policy_kill(ret);
683 EXPORT_SYMBOL(xfrm_policy_byid);
685 #ifdef CONFIG_SECURITY_NETWORK_XFRM
687 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
691 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
692 struct xfrm_policy *pol;
693 struct hlist_node *entry;
696 hlist_for_each_entry(pol, entry,
697 &xfrm_policy_inexact[dir], bydst) {
698 if (pol->type != type)
700 err = security_xfrm_policy_delete(pol);
702 xfrm_audit_policy_delete(pol, 0,
703 audit_info->loginuid,
708 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
709 hlist_for_each_entry(pol, entry,
710 xfrm_policy_bydst[dir].table + i,
712 if (pol->type != type)
714 err = security_xfrm_policy_delete(pol);
716 xfrm_audit_policy_delete(pol, 0,
717 audit_info->loginuid,
728 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
734 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
738 write_lock_bh(&xfrm_policy_lock);
740 err = xfrm_policy_flush_secctx_check(type, audit_info);
744 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
745 struct xfrm_policy *pol;
746 struct hlist_node *entry;
751 hlist_for_each_entry(pol, entry,
752 &xfrm_policy_inexact[dir], bydst) {
753 if (pol->type != type)
755 hlist_del(&pol->bydst);
756 hlist_del(&pol->byidx);
757 write_unlock_bh(&xfrm_policy_lock);
759 xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
762 xfrm_policy_kill(pol);
765 write_lock_bh(&xfrm_policy_lock);
769 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
771 hlist_for_each_entry(pol, entry,
772 xfrm_policy_bydst[dir].table + i,
774 if (pol->type != type)
776 hlist_del(&pol->bydst);
777 hlist_del(&pol->byidx);
778 write_unlock_bh(&xfrm_policy_lock);
780 xfrm_audit_policy_delete(pol, 1,
781 audit_info->loginuid,
783 xfrm_policy_kill(pol);
786 write_lock_bh(&xfrm_policy_lock);
791 xfrm_policy_count[dir] -= killed;
793 atomic_inc(&flow_cache_genid);
795 write_unlock_bh(&xfrm_policy_lock);
798 EXPORT_SYMBOL(xfrm_policy_flush);
800 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
803 struct xfrm_policy *pol, *last = NULL;
804 struct hlist_node *entry;
805 int dir, last_dir = 0, count, error;
807 read_lock_bh(&xfrm_policy_lock);
810 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
811 struct hlist_head *table = xfrm_policy_bydst[dir].table;
814 hlist_for_each_entry(pol, entry,
815 &xfrm_policy_inexact[dir], bydst) {
816 if (pol->type != type)
819 error = func(last, last_dir % XFRM_POLICY_MAX,
828 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
829 hlist_for_each_entry(pol, entry, table + i, bydst) {
830 if (pol->type != type)
833 error = func(last, last_dir % XFRM_POLICY_MAX,
848 error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
850 read_unlock_bh(&xfrm_policy_lock);
853 EXPORT_SYMBOL(xfrm_policy_walk);
856 * Find policy to apply to this flow.
858 * Returns 0 if policy found, else an -errno.
860 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
861 u8 type, u16 family, int dir)
863 struct xfrm_selector *sel = &pol->selector;
864 int match, ret = -ESRCH;
866 if (pol->family != family ||
870 match = xfrm_selector_match(sel, fl, family);
872 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
877 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
881 struct xfrm_policy *pol, *ret;
882 xfrm_address_t *daddr, *saddr;
883 struct hlist_node *entry;
884 struct hlist_head *chain;
887 daddr = xfrm_flowi_daddr(fl, family);
888 saddr = xfrm_flowi_saddr(fl, family);
889 if (unlikely(!daddr || !saddr))
892 read_lock_bh(&xfrm_policy_lock);
893 chain = policy_hash_direct(daddr, saddr, family, dir);
895 hlist_for_each_entry(pol, entry, chain, bydst) {
896 err = xfrm_policy_match(pol, fl, type, family, dir);
906 priority = ret->priority;
910 chain = &xfrm_policy_inexact[dir];
911 hlist_for_each_entry(pol, entry, chain, bydst) {
912 err = xfrm_policy_match(pol, fl, type, family, dir);
920 } else if (pol->priority < priority) {
928 read_unlock_bh(&xfrm_policy_lock);
933 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
934 void **objp, atomic_t **obj_refp)
936 struct xfrm_policy *pol;
939 #ifdef CONFIG_XFRM_SUB_POLICY
940 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
948 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
953 #ifdef CONFIG_XFRM_SUB_POLICY
956 if ((*objp = (void *) pol) != NULL)
957 *obj_refp = &pol->refcnt;
961 static inline int policy_to_flow_dir(int dir)
963 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
964 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
965 XFRM_POLICY_FWD == FLOW_DIR_FWD)
971 case XFRM_POLICY_OUT:
973 case XFRM_POLICY_FWD:
978 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
980 struct xfrm_policy *pol;
982 read_lock_bh(&xfrm_policy_lock);
983 if ((pol = sk->sk_policy[dir]) != NULL) {
984 int match = xfrm_selector_match(&pol->selector, fl,
989 err = security_xfrm_policy_lookup(pol, fl->secid,
990 policy_to_flow_dir(dir));
993 else if (err == -ESRCH)
1000 read_unlock_bh(&xfrm_policy_lock);
1004 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1006 struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1009 hlist_add_head(&pol->bydst, chain);
1010 hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1011 xfrm_policy_count[dir]++;
1014 if (xfrm_bydst_should_resize(dir, NULL))
1015 schedule_work(&xfrm_hash_work);
1018 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1021 if (hlist_unhashed(&pol->bydst))
1024 hlist_del(&pol->bydst);
1025 hlist_del(&pol->byidx);
1026 xfrm_policy_count[dir]--;
1031 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1033 write_lock_bh(&xfrm_policy_lock);
1034 pol = __xfrm_policy_unlink(pol, dir);
1035 write_unlock_bh(&xfrm_policy_lock);
1037 if (dir < XFRM_POLICY_MAX)
1038 atomic_inc(&flow_cache_genid);
1039 xfrm_policy_kill(pol);
1044 EXPORT_SYMBOL(xfrm_policy_delete);
1046 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1048 struct xfrm_policy *old_pol;
1050 #ifdef CONFIG_XFRM_SUB_POLICY
1051 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1055 write_lock_bh(&xfrm_policy_lock);
1056 old_pol = sk->sk_policy[dir];
1057 sk->sk_policy[dir] = pol;
1059 pol->curlft.add_time = get_seconds();
1060 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1061 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1064 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1065 write_unlock_bh(&xfrm_policy_lock);
1068 xfrm_policy_kill(old_pol);
1073 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1075 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1078 newp->selector = old->selector;
1079 if (security_xfrm_policy_clone(old, newp)) {
1081 return NULL; /* ENOMEM */
1083 newp->lft = old->lft;
1084 newp->curlft = old->curlft;
1085 newp->action = old->action;
1086 newp->flags = old->flags;
1087 newp->xfrm_nr = old->xfrm_nr;
1088 newp->index = old->index;
1089 newp->type = old->type;
1090 memcpy(newp->xfrm_vec, old->xfrm_vec,
1091 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1092 write_lock_bh(&xfrm_policy_lock);
1093 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1094 write_unlock_bh(&xfrm_policy_lock);
1100 int __xfrm_sk_clone_policy(struct sock *sk)
1102 struct xfrm_policy *p0 = sk->sk_policy[0],
1103 *p1 = sk->sk_policy[1];
1105 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1106 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1108 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1114 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1115 unsigned short family)
1118 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1120 if (unlikely(afinfo == NULL))
1122 err = afinfo->get_saddr(local, remote);
1123 xfrm_policy_put_afinfo(afinfo);
1127 /* Resolve list of templates for the flow, given policy. */
1130 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1131 struct xfrm_state **xfrm,
1132 unsigned short family)
1136 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1137 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1140 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1141 struct xfrm_state *x;
1142 xfrm_address_t *remote = daddr;
1143 xfrm_address_t *local = saddr;
1144 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1146 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1147 tmpl->mode == XFRM_MODE_BEET) {
1148 remote = &tmpl->id.daddr;
1149 local = &tmpl->saddr;
1150 family = tmpl->encap_family;
1151 if (xfrm_addr_any(local, family)) {
1152 error = xfrm_get_saddr(&tmp, remote, family);
1159 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1161 if (x && x->km.state == XFRM_STATE_VALID) {
1168 error = (x->km.state == XFRM_STATE_ERROR ?
1173 if (!tmpl->optional)
1179 for (nx--; nx>=0; nx--)
1180 xfrm_state_put(xfrm[nx]);
1185 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1186 struct xfrm_state **xfrm,
1187 unsigned short family)
1189 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1190 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1196 for (i = 0; i < npols; i++) {
1197 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1202 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1210 /* found states are sorted for outbound processing */
1212 xfrm_state_sort(xfrm, tpp, cnx, family);
1217 for (cnx--; cnx>=0; cnx--)
1218 xfrm_state_put(tpp[cnx]);
1223 /* Check that the bundle accepts the flow and its components are
1227 static struct dst_entry *
1228 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1230 struct dst_entry *x;
1231 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1232 if (unlikely(afinfo == NULL))
1233 return ERR_PTR(-EINVAL);
1234 x = afinfo->find_bundle(fl, policy);
1235 xfrm_policy_put_afinfo(afinfo);
1239 static inline int xfrm_get_tos(struct flowi *fl, int family)
1241 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1247 tos = afinfo->get_tos(fl);
1249 xfrm_policy_put_afinfo(afinfo);
1254 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1256 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1257 struct xfrm_dst *xdst;
1260 return ERR_PTR(-EINVAL);
1262 xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1264 xfrm_policy_put_afinfo(afinfo);
1269 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1272 struct xfrm_policy_afinfo *afinfo =
1273 xfrm_policy_get_afinfo(dst->ops->family);
1279 err = afinfo->init_path(path, dst, nfheader_len);
1281 xfrm_policy_put_afinfo(afinfo);
1286 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1288 struct xfrm_policy_afinfo *afinfo =
1289 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1295 err = afinfo->fill_dst(xdst, dev);
1297 xfrm_policy_put_afinfo(afinfo);
1302 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1303 * all the metrics... Shortly, bundle a bundle.
1306 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1307 struct xfrm_state **xfrm, int nx,
1309 struct dst_entry *dst)
1311 unsigned long now = jiffies;
1312 struct net_device *dev;
1313 struct dst_entry *dst_prev = NULL;
1314 struct dst_entry *dst0 = NULL;
1318 int nfheader_len = 0;
1319 int trailer_len = 0;
1321 int family = policy->selector.family;
1323 tos = xfrm_get_tos(fl, family);
1330 for (; i < nx; i++) {
1331 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1332 struct dst_entry *dst1 = &xdst->u.dst;
1334 err = PTR_ERR(xdst);
1343 dst_prev->child = dst_clone(dst1);
1344 dst1->flags |= DST_NOHASH;
1348 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1350 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1351 family = xfrm[i]->props.family;
1352 dst = xfrm_dst_lookup(xfrm[i], tos, family);
1359 dst1->xfrm = xfrm[i];
1360 xdst->genid = xfrm[i]->genid;
1362 dst1->obsolete = -1;
1363 dst1->flags |= DST_HOST;
1364 dst1->lastuse = now;
1366 dst1->input = dst_discard;
1367 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1369 dst1->next = dst_prev;
1372 header_len += xfrm[i]->props.header_len;
1373 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1374 nfheader_len += xfrm[i]->props.header_len;
1375 trailer_len += xfrm[i]->props.trailer_len;
1378 dst_prev->child = dst;
1386 /* Copy neighbout for reachability confirmation */
1387 dst0->neighbour = neigh_clone(dst->neighbour);
1389 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1390 xfrm_init_pmtu(dst_prev);
1392 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1393 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1395 err = xfrm_fill_dst(xdst, dev);
1399 dst_prev->header_len = header_len;
1400 dst_prev->trailer_len = trailer_len;
1401 header_len -= xdst->u.dst.xfrm->props.header_len;
1402 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1410 xfrm_state_put(xfrm[i]);
1414 dst0 = ERR_PTR(err);
1419 xfrm_dst_alloc_copy(void **target, void *src, int size)
1422 *target = kmalloc(size, GFP_ATOMIC);
1426 memcpy(*target, src, size);
1431 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1433 #ifdef CONFIG_XFRM_SUB_POLICY
1434 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1435 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1443 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1445 #ifdef CONFIG_XFRM_SUB_POLICY
1446 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1447 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1453 static int stale_bundle(struct dst_entry *dst);
1455 /* Main function: finds/creates a bundle for given flow.
1457 * At the moment we eat a raw IP route. Mostly to speed up lookups
1458 * on interfaces with disabled IPsec.
1460 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1461 struct sock *sk, int flags)
1463 struct xfrm_policy *policy;
1464 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1469 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1470 struct dst_entry *dst, *dst_orig = *dst_p;
1475 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1478 genid = atomic_read(&flow_cache_genid);
1480 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1486 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1487 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1488 err = PTR_ERR(policy);
1494 /* To accelerate a bit... */
1495 if ((dst_orig->flags & DST_NOXFRM) ||
1496 !xfrm_policy_count[XFRM_POLICY_OUT])
1499 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1500 dir, xfrm_policy_lookup);
1501 err = PTR_ERR(policy);
1509 family = dst_orig->ops->family;
1512 xfrm_nr += pols[0]->xfrm_nr;
1515 if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1518 policy->curlft.use_time = get_seconds();
1520 switch (policy->action) {
1522 case XFRM_POLICY_BLOCK:
1523 /* Prohibit the flow */
1527 case XFRM_POLICY_ALLOW:
1528 #ifndef CONFIG_XFRM_SUB_POLICY
1529 if (policy->xfrm_nr == 0) {
1530 /* Flow passes not transformed. */
1531 xfrm_pol_put(policy);
1536 /* Try to find matching bundle.
1538 * LATER: help from flow cache. It is optional, this
1539 * is required only for output policy.
1541 dst = xfrm_find_bundle(fl, policy, family);
1550 #ifdef CONFIG_XFRM_SUB_POLICY
1551 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1552 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1556 if (IS_ERR(pols[1])) {
1557 err = PTR_ERR(pols[1]);
1560 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1565 xfrm_nr += pols[1]->xfrm_nr;
1570 * Because neither flowi nor bundle information knows about
1571 * transformation template size. On more than one policy usage
1572 * we can realize whether all of them is bypass or not after
1573 * they are searched. See above not-transformed bypass
1574 * is surrounded by non-sub policy configuration, too.
1577 /* Flow passes not transformed. */
1578 xfrm_pols_put(pols, npols);
1583 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1585 if (unlikely(nx<0)) {
1587 if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1588 /* EREMOTE tells the caller to generate
1589 * a one-shot blackhole route.
1591 xfrm_pol_put(policy);
1594 if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1595 DECLARE_WAITQUEUE(wait, current);
1597 add_wait_queue(&km_waitq, &wait);
1598 set_current_state(TASK_INTERRUPTIBLE);
1600 set_current_state(TASK_RUNNING);
1601 remove_wait_queue(&km_waitq, &wait);
1603 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1605 if (nx == -EAGAIN && signal_pending(current)) {
1609 if (nx == -EAGAIN ||
1610 genid != atomic_read(&flow_cache_genid)) {
1611 xfrm_pols_put(pols, npols);
1620 /* Flow passes not transformed. */
1621 xfrm_pols_put(pols, npols);
1625 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1630 for (pi = 0; pi < npols; pi++) {
1631 read_lock_bh(&pols[pi]->lock);
1632 pol_dead |= pols[pi]->dead;
1633 read_unlock_bh(&pols[pi]->lock);
1636 write_lock_bh(&policy->lock);
1637 if (unlikely(pol_dead || stale_bundle(dst))) {
1638 /* Wow! While we worked on resolving, this
1639 * policy has gone. Retry. It is not paranoia,
1640 * we just cannot enlist new bundle to dead object.
1641 * We can't enlist stable bundles either.
1643 write_unlock_bh(&policy->lock);
1647 err = -EHOSTUNREACH;
1652 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1654 err = xfrm_dst_update_origin(dst, fl);
1655 if (unlikely(err)) {
1656 write_unlock_bh(&policy->lock);
1662 dst->next = policy->bundles;
1663 policy->bundles = dst;
1665 write_unlock_bh(&policy->lock);
1668 dst_release(dst_orig);
1669 xfrm_pols_put(pols, npols);
1673 xfrm_pols_put(pols, npols);
1675 dst_release(dst_orig);
1681 if (flags & XFRM_LOOKUP_ICMP)
1685 EXPORT_SYMBOL(__xfrm_lookup);
1687 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1688 struct sock *sk, int flags)
1690 int err = __xfrm_lookup(dst_p, fl, sk, flags);
1692 if (err == -EREMOTE) {
1693 dst_release(*dst_p);
1700 EXPORT_SYMBOL(xfrm_lookup);
1703 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1705 struct xfrm_state *x;
1707 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1709 x = skb->sp->xvec[idx];
1710 if (!x->type->reject)
1712 return x->type->reject(x, skb, fl);
1715 /* When skb is transformed back to its "native" form, we have to
1716 * check policy restrictions. At the moment we make this in maximally
1717 * stupid way. Shame on me. :-) Of course, connected sockets must
1718 * have policy cached at them.
1722 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1723 unsigned short family)
1725 if (xfrm_state_kern(x))
1726 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1727 return x->id.proto == tmpl->id.proto &&
1728 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1729 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1730 x->props.mode == tmpl->mode &&
1731 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1732 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1733 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1734 xfrm_state_addr_cmp(tmpl, x, family));
1738 * 0 or more than 0 is returned when validation is succeeded (either bypass
1739 * because of optional transport mode, or next index of the mathced secpath
1740 * state with the template.
1741 * -1 is returned when no matching template is found.
1742 * Otherwise "-2 - errored_index" is returned.
1745 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1746 unsigned short family)
1750 if (tmpl->optional) {
1751 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1755 for (; idx < sp->len; idx++) {
1756 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1758 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1767 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1768 unsigned int family, int reverse)
1770 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1773 if (unlikely(afinfo == NULL))
1774 return -EAFNOSUPPORT;
1776 afinfo->decode_session(skb, fl, reverse);
1777 err = security_xfrm_decode_session(skb, &fl->secid);
1778 xfrm_policy_put_afinfo(afinfo);
1781 EXPORT_SYMBOL(__xfrm_decode_session);
1783 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1785 for (; k < sp->len; k++) {
1786 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1795 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1796 unsigned short family)
1798 struct xfrm_policy *pol;
1799 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1808 reverse = dir & ~XFRM_POLICY_MASK;
1809 dir &= XFRM_POLICY_MASK;
1810 fl_dir = policy_to_flow_dir(dir);
1812 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0)
1814 nf_nat_decode_session(skb, &fl, family);
1816 /* First, check used SA against their selectors. */
1820 for (i=skb->sp->len-1; i>=0; i--) {
1821 struct xfrm_state *x = skb->sp->xvec[i];
1822 if (!xfrm_selector_match(&x->sel, &fl, family))
1828 if (sk && sk->sk_policy[dir]) {
1829 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1835 pol = flow_cache_lookup(&fl, family, fl_dir,
1836 xfrm_policy_lookup);
1842 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1843 xfrm_secpath_reject(xerr_idx, skb, &fl);
1849 pol->curlft.use_time = get_seconds();
1853 #ifdef CONFIG_XFRM_SUB_POLICY
1854 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1855 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1859 if (IS_ERR(pols[1]))
1861 pols[1]->curlft.use_time = get_seconds();
1867 if (pol->action == XFRM_POLICY_ALLOW) {
1868 struct sec_path *sp;
1869 static struct sec_path dummy;
1870 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1871 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1872 struct xfrm_tmpl **tpp = tp;
1876 if ((sp = skb->sp) == NULL)
1879 for (pi = 0; pi < npols; pi++) {
1880 if (pols[pi] != pol &&
1881 pols[pi]->action != XFRM_POLICY_ALLOW)
1883 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1885 for (i = 0; i < pols[pi]->xfrm_nr; i++)
1886 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1890 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1894 /* For each tunnel xfrm, find the first matching tmpl.
1895 * For each tmpl before that, find corresponding xfrm.
1896 * Order is _important_. Later we will implement
1897 * some barriers, but at the moment barriers
1898 * are implied between each two transformations.
1900 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1901 k = xfrm_policy_ok(tpp[i], sp, k, family);
1904 /* "-2 - errored_index" returned */
1910 if (secpath_has_nontransport(sp, k, &xerr_idx))
1913 xfrm_pols_put(pols, npols);
1918 xfrm_secpath_reject(xerr_idx, skb, &fl);
1920 xfrm_pols_put(pols, npols);
1923 EXPORT_SYMBOL(__xfrm_policy_check);
1925 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1929 if (xfrm_decode_session(skb, &fl, family) < 0)
1932 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1934 EXPORT_SYMBOL(__xfrm_route_forward);
1936 /* Optimize later using cookies and generation ids. */
1938 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1940 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1941 * to "-1" to force all XFRM destinations to get validated by
1942 * dst_ops->check on every use. We do this because when a
1943 * normal route referenced by an XFRM dst is obsoleted we do
1944 * not go looking around for all parent referencing XFRM dsts
1945 * so that we can invalidate them. It is just too much work.
1946 * Instead we make the checks here on every use. For example:
1948 * XFRM dst A --> IPv4 dst X
1950 * X is the "xdst->route" of A (X is also the "dst->path" of A
1951 * in this example). If X is marked obsolete, "A" will not
1952 * notice. That's what we are validating here via the
1953 * stale_bundle() check.
1955 * When a policy's bundle is pruned, we dst_free() the XFRM
1956 * dst which causes it's ->obsolete field to be set to a
1957 * positive non-zero integer. If an XFRM dst has been pruned
1958 * like this, we want to force a new route lookup.
1960 if (dst->obsolete < 0 && !stale_bundle(dst))
1966 static int stale_bundle(struct dst_entry *dst)
1968 return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1971 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1973 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1974 dst->dev = dev->nd_net->loopback_dev;
1979 EXPORT_SYMBOL(xfrm_dst_ifdown);
1981 static void xfrm_link_failure(struct sk_buff *skb)
1983 /* Impossible. Such dst must be popped before reaches point of failure. */
1987 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1990 if (dst->obsolete) {
1998 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2000 struct dst_entry *dst, **dstp;
2002 write_lock(&pol->lock);
2003 dstp = &pol->bundles;
2004 while ((dst=*dstp) != NULL) {
2007 dst->next = *gc_list_p;
2013 write_unlock(&pol->lock);
2016 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2018 struct dst_entry *gc_list = NULL;
2021 read_lock_bh(&xfrm_policy_lock);
2022 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2023 struct xfrm_policy *pol;
2024 struct hlist_node *entry;
2025 struct hlist_head *table;
2028 hlist_for_each_entry(pol, entry,
2029 &xfrm_policy_inexact[dir], bydst)
2030 prune_one_bundle(pol, func, &gc_list);
2032 table = xfrm_policy_bydst[dir].table;
2033 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2034 hlist_for_each_entry(pol, entry, table + i, bydst)
2035 prune_one_bundle(pol, func, &gc_list);
2038 read_unlock_bh(&xfrm_policy_lock);
2041 struct dst_entry *dst = gc_list;
2042 gc_list = dst->next;
2047 static int unused_bundle(struct dst_entry *dst)
2049 return !atomic_read(&dst->__refcnt);
2052 static void __xfrm_garbage_collect(void)
2054 xfrm_prune_bundles(unused_bundle);
2057 static int xfrm_flush_bundles(void)
2059 xfrm_prune_bundles(stale_bundle);
2063 static void xfrm_init_pmtu(struct dst_entry *dst)
2066 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2067 u32 pmtu, route_mtu_cached;
2069 pmtu = dst_mtu(dst->child);
2070 xdst->child_mtu_cached = pmtu;
2072 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2074 route_mtu_cached = dst_mtu(xdst->route);
2075 xdst->route_mtu_cached = route_mtu_cached;
2077 if (pmtu > route_mtu_cached)
2078 pmtu = route_mtu_cached;
2080 dst->metrics[RTAX_MTU-1] = pmtu;
2081 } while ((dst = dst->next));
2084 /* Check that the bundle accepts the flow and its components are
2088 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2089 struct flowi *fl, int family, int strict)
2091 struct dst_entry *dst = &first->u.dst;
2092 struct xfrm_dst *last;
2095 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2096 (dst->dev && !netif_running(dst->dev)))
2098 #ifdef CONFIG_XFRM_SUB_POLICY
2100 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2102 if (first->partner &&
2103 !xfrm_selector_match(first->partner, fl, family))
2111 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2113 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2116 !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2118 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2120 if (xdst->genid != dst->xfrm->genid)
2124 !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2125 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2128 mtu = dst_mtu(dst->child);
2129 if (xdst->child_mtu_cached != mtu) {
2131 xdst->child_mtu_cached = mtu;
2134 if (!dst_check(xdst->route, xdst->route_cookie))
2136 mtu = dst_mtu(xdst->route);
2137 if (xdst->route_mtu_cached != mtu) {
2139 xdst->route_mtu_cached = mtu;
2143 } while (dst->xfrm);
2148 mtu = last->child_mtu_cached;
2152 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2153 if (mtu > last->route_mtu_cached)
2154 mtu = last->route_mtu_cached;
2155 dst->metrics[RTAX_MTU-1] = mtu;
2160 last = (struct xfrm_dst *)last->u.dst.next;
2161 last->child_mtu_cached = mtu;
2167 EXPORT_SYMBOL(xfrm_bundle_ok);
2169 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2172 if (unlikely(afinfo == NULL))
2174 if (unlikely(afinfo->family >= NPROTO))
2175 return -EAFNOSUPPORT;
2176 write_lock_bh(&xfrm_policy_afinfo_lock);
2177 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2180 struct dst_ops *dst_ops = afinfo->dst_ops;
2181 if (likely(dst_ops->kmem_cachep == NULL))
2182 dst_ops->kmem_cachep = xfrm_dst_cache;
2183 if (likely(dst_ops->check == NULL))
2184 dst_ops->check = xfrm_dst_check;
2185 if (likely(dst_ops->negative_advice == NULL))
2186 dst_ops->negative_advice = xfrm_negative_advice;
2187 if (likely(dst_ops->link_failure == NULL))
2188 dst_ops->link_failure = xfrm_link_failure;
2189 if (likely(afinfo->garbage_collect == NULL))
2190 afinfo->garbage_collect = __xfrm_garbage_collect;
2191 xfrm_policy_afinfo[afinfo->family] = afinfo;
2193 write_unlock_bh(&xfrm_policy_afinfo_lock);
2196 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2198 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2201 if (unlikely(afinfo == NULL))
2203 if (unlikely(afinfo->family >= NPROTO))
2204 return -EAFNOSUPPORT;
2205 write_lock_bh(&xfrm_policy_afinfo_lock);
2206 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2207 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2210 struct dst_ops *dst_ops = afinfo->dst_ops;
2211 xfrm_policy_afinfo[afinfo->family] = NULL;
2212 dst_ops->kmem_cachep = NULL;
2213 dst_ops->check = NULL;
2214 dst_ops->negative_advice = NULL;
2215 dst_ops->link_failure = NULL;
2216 afinfo->garbage_collect = NULL;
2219 write_unlock_bh(&xfrm_policy_afinfo_lock);
2222 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2224 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2226 struct xfrm_policy_afinfo *afinfo;
2227 if (unlikely(family >= NPROTO))
2229 read_lock(&xfrm_policy_afinfo_lock);
2230 afinfo = xfrm_policy_afinfo[family];
2231 if (unlikely(!afinfo))
2232 read_unlock(&xfrm_policy_afinfo_lock);
2236 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2238 read_unlock(&xfrm_policy_afinfo_lock);
2241 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2243 struct net_device *dev = ptr;
2245 if (dev->nd_net != &init_net)
2250 xfrm_flush_bundles();
2255 static struct notifier_block xfrm_dev_notifier = {
2261 static void __init xfrm_policy_init(void)
2263 unsigned int hmask, sz;
2266 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2267 sizeof(struct xfrm_dst),
2268 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2272 sz = (hmask+1) * sizeof(struct hlist_head);
2274 xfrm_policy_byidx = xfrm_hash_alloc(sz);
2275 xfrm_idx_hmask = hmask;
2276 if (!xfrm_policy_byidx)
2277 panic("XFRM: failed to allocate byidx hash\n");
2279 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2280 struct xfrm_policy_hash *htab;
2282 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2284 htab = &xfrm_policy_bydst[dir];
2285 htab->table = xfrm_hash_alloc(sz);
2286 htab->hmask = hmask;
2288 panic("XFRM: failed to allocate bydst hash\n");
2291 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2292 register_netdevice_notifier(&xfrm_dev_notifier);
2295 void __init xfrm_init(void)
2302 #ifdef CONFIG_AUDITSYSCALL
2303 static inline void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2304 struct audit_buffer *audit_buf)
2306 struct xfrm_sec_ctx *ctx = xp->security;
2307 struct xfrm_selector *sel = &xp->selector;
2310 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2311 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2313 switch(sel->family) {
2315 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2316 NIPQUAD(sel->saddr.a4));
2317 if (sel->prefixlen_s != 32)
2318 audit_log_format(audit_buf, " src_prefixlen=%d",
2320 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2321 NIPQUAD(sel->daddr.a4));
2322 if (sel->prefixlen_d != 32)
2323 audit_log_format(audit_buf, " dst_prefixlen=%d",
2327 audit_log_format(audit_buf, " src=" NIP6_FMT,
2328 NIP6(*(struct in6_addr *)sel->saddr.a6));
2329 if (sel->prefixlen_s != 128)
2330 audit_log_format(audit_buf, " src_prefixlen=%d",
2332 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2333 NIP6(*(struct in6_addr *)sel->daddr.a6));
2334 if (sel->prefixlen_d != 128)
2335 audit_log_format(audit_buf, " dst_prefixlen=%d",
2342 xfrm_audit_policy_add(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2344 struct audit_buffer *audit_buf;
2345 extern int audit_enabled;
2347 if (audit_enabled == 0)
2349 audit_buf = xfrm_audit_start(auid, sid);
2350 if (audit_buf == NULL)
2352 audit_log_format(audit_buf, " op=SPD-add res=%u", result);
2353 xfrm_audit_common_policyinfo(xp, audit_buf);
2354 audit_log_end(audit_buf);
2356 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2359 xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2361 struct audit_buffer *audit_buf;
2362 extern int audit_enabled;
2364 if (audit_enabled == 0)
2366 audit_buf = xfrm_audit_start(auid, sid);
2367 if (audit_buf == NULL)
2369 audit_log_format(audit_buf, " op=SPD-delete res=%u", result);
2370 xfrm_audit_common_policyinfo(xp, audit_buf);
2371 audit_log_end(audit_buf);
2373 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2376 #ifdef CONFIG_XFRM_MIGRATE
2377 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2378 struct xfrm_selector *sel_tgt)
2380 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2381 if (sel_tgt->family == sel_cmp->family &&
2382 xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2383 sel_cmp->family) == 0 &&
2384 xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2385 sel_cmp->family) == 0 &&
2386 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2387 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2391 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2398 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2401 struct xfrm_policy *pol, *ret = NULL;
2402 struct hlist_node *entry;
2403 struct hlist_head *chain;
2406 read_lock_bh(&xfrm_policy_lock);
2407 chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2408 hlist_for_each_entry(pol, entry, chain, bydst) {
2409 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2410 pol->type == type) {
2412 priority = ret->priority;
2416 chain = &xfrm_policy_inexact[dir];
2417 hlist_for_each_entry(pol, entry, chain, bydst) {
2418 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2419 pol->type == type &&
2420 pol->priority < priority) {
2429 read_unlock_bh(&xfrm_policy_lock);
2434 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2438 if (t->mode == m->mode && t->id.proto == m->proto &&
2439 (m->reqid == 0 || t->reqid == m->reqid)) {
2441 case XFRM_MODE_TUNNEL:
2442 case XFRM_MODE_BEET:
2443 if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2444 m->old_family) == 0 &&
2445 xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2446 m->old_family) == 0) {
2450 case XFRM_MODE_TRANSPORT:
2451 /* in case of transport mode, template does not store
2452 any IP addresses, hence we just compare mode and
2463 /* update endpoint address(es) of template(s) */
2464 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2465 struct xfrm_migrate *m, int num_migrate)
2467 struct xfrm_migrate *mp;
2468 struct dst_entry *dst;
2471 write_lock_bh(&pol->lock);
2472 if (unlikely(pol->dead)) {
2473 /* target policy has been deleted */
2474 write_unlock_bh(&pol->lock);
2478 for (i = 0; i < pol->xfrm_nr; i++) {
2479 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2480 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2483 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2484 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2486 /* update endpoints */
2487 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2488 sizeof(pol->xfrm_vec[i].id.daddr));
2489 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2490 sizeof(pol->xfrm_vec[i].saddr));
2491 pol->xfrm_vec[i].encap_family = mp->new_family;
2493 while ((dst = pol->bundles) != NULL) {
2494 pol->bundles = dst->next;
2500 write_unlock_bh(&pol->lock);
2508 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2512 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2515 for (i = 0; i < num_migrate; i++) {
2516 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2517 m[i].old_family) == 0) &&
2518 (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2519 m[i].old_family) == 0))
2521 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2522 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2525 /* check if there is any duplicated entry */
2526 for (j = i + 1; j < num_migrate; j++) {
2527 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2528 sizeof(m[i].old_daddr)) &&
2529 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2530 sizeof(m[i].old_saddr)) &&
2531 m[i].proto == m[j].proto &&
2532 m[i].mode == m[j].mode &&
2533 m[i].reqid == m[j].reqid &&
2534 m[i].old_family == m[j].old_family)
2542 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2543 struct xfrm_migrate *m, int num_migrate)
2545 int i, err, nx_cur = 0, nx_new = 0;
2546 struct xfrm_policy *pol = NULL;
2547 struct xfrm_state *x, *xc;
2548 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2549 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2550 struct xfrm_migrate *mp;
2552 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2555 /* Stage 1 - find policy */
2556 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2561 /* Stage 2 - find and update state(s) */
2562 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2563 if ((x = xfrm_migrate_state_find(mp))) {
2566 if ((xc = xfrm_state_migrate(x, mp))) {
2576 /* Stage 3 - update policy */
2577 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2580 /* Stage 4 - delete old state(s) */
2582 xfrm_states_put(x_cur, nx_cur);
2583 xfrm_states_delete(x_cur, nx_cur);
2586 /* Stage 5 - announce */
2587 km_migrate(sel, dir, type, m, num_migrate);
2599 xfrm_states_put(x_cur, nx_cur);
2601 xfrm_states_delete(x_new, nx_new);
2605 EXPORT_SYMBOL(xfrm_migrate);