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>
27 #include <linux/audit.h>
32 #ifdef CONFIG_XFRM_STATISTICS
36 #include "xfrm_hash.h"
38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
40 #define XFRM_MAX_QUEUE_LEN 100
43 struct dst_entry *dst_orig;
47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
48 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
51 static struct kmem_cache *xfrm_dst_cache __read_mostly;
52 static __read_mostly seqcount_t xfrm_policy_hash_generation;
54 static void xfrm_init_pmtu(struct dst_entry *dst);
55 static int stale_bundle(struct dst_entry *dst);
56 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
57 static void xfrm_policy_queue_process(unsigned long arg);
59 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
63 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
65 return atomic_inc_not_zero(&policy->refcnt);
69 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
71 const struct flowi4 *fl4 = &fl->u.ip4;
73 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
74 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
75 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
76 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
77 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
78 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
82 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
84 const struct flowi6 *fl6 = &fl->u.ip6;
86 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
87 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
88 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
89 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
90 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
91 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
94 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
95 unsigned short family)
99 return __xfrm4_selector_match(sel, fl);
101 return __xfrm6_selector_match(sel, fl);
106 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
108 struct xfrm_policy_afinfo *afinfo;
110 if (unlikely(family >= NPROTO))
113 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
114 if (unlikely(!afinfo))
119 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
124 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net,
126 const xfrm_address_t *saddr,
127 const xfrm_address_t *daddr,
130 struct xfrm_policy_afinfo *afinfo;
131 struct dst_entry *dst;
133 afinfo = xfrm_policy_get_afinfo(family);
134 if (unlikely(afinfo == NULL))
135 return ERR_PTR(-EAFNOSUPPORT);
137 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr);
139 xfrm_policy_put_afinfo(afinfo);
144 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
146 xfrm_address_t *prev_saddr,
147 xfrm_address_t *prev_daddr,
150 struct net *net = xs_net(x);
151 xfrm_address_t *saddr = &x->props.saddr;
152 xfrm_address_t *daddr = &x->id.daddr;
153 struct dst_entry *dst;
155 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
159 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
164 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family);
167 if (prev_saddr != saddr)
168 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
169 if (prev_daddr != daddr)
170 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
176 static inline unsigned long make_jiffies(long secs)
178 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
179 return MAX_SCHEDULE_TIMEOUT-1;
184 static void xfrm_policy_timer(unsigned long data)
186 struct xfrm_policy *xp = (struct xfrm_policy *)data;
187 unsigned long now = get_seconds();
188 long next = LONG_MAX;
192 read_lock(&xp->lock);
194 if (unlikely(xp->walk.dead))
197 dir = xfrm_policy_id2dir(xp->index);
199 if (xp->lft.hard_add_expires_seconds) {
200 long tmo = xp->lft.hard_add_expires_seconds +
201 xp->curlft.add_time - now;
207 if (xp->lft.hard_use_expires_seconds) {
208 long tmo = xp->lft.hard_use_expires_seconds +
209 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
215 if (xp->lft.soft_add_expires_seconds) {
216 long tmo = xp->lft.soft_add_expires_seconds +
217 xp->curlft.add_time - now;
220 tmo = XFRM_KM_TIMEOUT;
225 if (xp->lft.soft_use_expires_seconds) {
226 long tmo = xp->lft.soft_use_expires_seconds +
227 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
230 tmo = XFRM_KM_TIMEOUT;
237 km_policy_expired(xp, dir, 0, 0);
238 if (next != LONG_MAX &&
239 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
243 read_unlock(&xp->lock);
248 read_unlock(&xp->lock);
249 if (!xfrm_policy_delete(xp, dir))
250 km_policy_expired(xp, dir, 1, 0);
254 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
256 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
258 if (unlikely(pol->walk.dead))
266 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
268 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
270 return !pol->walk.dead;
273 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
275 xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
278 static const struct flow_cache_ops xfrm_policy_fc_ops = {
279 .get = xfrm_policy_flo_get,
280 .check = xfrm_policy_flo_check,
281 .delete = xfrm_policy_flo_delete,
284 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
288 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
290 struct xfrm_policy *policy;
292 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
295 write_pnet(&policy->xp_net, net);
296 INIT_LIST_HEAD(&policy->walk.all);
297 INIT_HLIST_NODE(&policy->bydst);
298 INIT_HLIST_NODE(&policy->byidx);
299 rwlock_init(&policy->lock);
300 atomic_set(&policy->refcnt, 1);
301 skb_queue_head_init(&policy->polq.hold_queue);
302 setup_timer(&policy->timer, xfrm_policy_timer,
303 (unsigned long)policy);
304 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
305 (unsigned long)policy);
306 policy->flo.ops = &xfrm_policy_fc_ops;
310 EXPORT_SYMBOL(xfrm_policy_alloc);
312 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
314 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
316 security_xfrm_policy_free(policy->security);
320 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
322 void xfrm_policy_destroy(struct xfrm_policy *policy)
324 BUG_ON(!policy->walk.dead);
326 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
329 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
331 EXPORT_SYMBOL(xfrm_policy_destroy);
333 /* Rule must be locked. Release descentant resources, announce
334 * entry dead. The rule must be unlinked from lists to the moment.
337 static void xfrm_policy_kill(struct xfrm_policy *policy)
339 write_lock_bh(&policy->lock);
340 policy->walk.dead = 1;
341 write_unlock_bh(&policy->lock);
343 atomic_inc(&policy->genid);
345 if (del_timer(&policy->polq.hold_timer))
346 xfrm_pol_put(policy);
347 skb_queue_purge(&policy->polq.hold_queue);
349 if (del_timer(&policy->timer))
350 xfrm_pol_put(policy);
352 xfrm_pol_put(policy);
355 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
357 static inline unsigned int idx_hash(struct net *net, u32 index)
359 return __idx_hash(index, net->xfrm.policy_idx_hmask);
362 /* calculate policy hash thresholds */
363 static void __get_hash_thresh(struct net *net,
364 unsigned short family, int dir,
365 u8 *dbits, u8 *sbits)
369 *dbits = net->xfrm.policy_bydst[dir].dbits4;
370 *sbits = net->xfrm.policy_bydst[dir].sbits4;
374 *dbits = net->xfrm.policy_bydst[dir].dbits6;
375 *sbits = net->xfrm.policy_bydst[dir].sbits6;
384 static struct hlist_head *policy_hash_bysel(struct net *net,
385 const struct xfrm_selector *sel,
386 unsigned short family, int dir)
388 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
393 __get_hash_thresh(net, family, dir, &dbits, &sbits);
394 hash = __sel_hash(sel, family, hmask, dbits, sbits);
396 if (hash == hmask + 1)
397 return &net->xfrm.policy_inexact[dir];
399 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
400 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
403 static struct hlist_head *policy_hash_direct(struct net *net,
404 const xfrm_address_t *daddr,
405 const xfrm_address_t *saddr,
406 unsigned short family, int dir)
408 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
413 __get_hash_thresh(net, family, dir, &dbits, &sbits);
414 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
416 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
417 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
420 static void xfrm_dst_hash_transfer(struct net *net,
421 struct hlist_head *list,
422 struct hlist_head *ndsttable,
423 unsigned int nhashmask,
426 struct hlist_node *tmp, *entry0 = NULL;
427 struct xfrm_policy *pol;
433 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
436 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
437 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
438 pol->family, nhashmask, dbits, sbits);
440 hlist_del_rcu(&pol->bydst);
441 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
446 hlist_del_rcu(&pol->bydst);
447 hlist_add_behind_rcu(&pol->bydst, entry0);
449 entry0 = &pol->bydst;
451 if (!hlist_empty(list)) {
457 static void xfrm_idx_hash_transfer(struct hlist_head *list,
458 struct hlist_head *nidxtable,
459 unsigned int nhashmask)
461 struct hlist_node *tmp;
462 struct xfrm_policy *pol;
464 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
467 h = __idx_hash(pol->index, nhashmask);
468 hlist_add_head(&pol->byidx, nidxtable+h);
472 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
474 return ((old_hmask + 1) << 1) - 1;
477 static void xfrm_bydst_resize(struct net *net, int dir)
479 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
480 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
481 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
482 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
483 struct hlist_head *odst;
489 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
490 write_seqcount_begin(&xfrm_policy_hash_generation);
492 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
493 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
495 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
496 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
498 for (i = hmask; i >= 0; i--)
499 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
501 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
502 net->xfrm.policy_bydst[dir].hmask = nhashmask;
504 write_seqcount_end(&xfrm_policy_hash_generation);
505 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
509 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
512 static void xfrm_byidx_resize(struct net *net, int total)
514 unsigned int hmask = net->xfrm.policy_idx_hmask;
515 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
516 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
517 struct hlist_head *oidx = net->xfrm.policy_byidx;
518 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
524 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
526 for (i = hmask; i >= 0; i--)
527 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
529 net->xfrm.policy_byidx = nidx;
530 net->xfrm.policy_idx_hmask = nhashmask;
532 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
534 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
537 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
539 unsigned int cnt = net->xfrm.policy_count[dir];
540 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
545 if ((hmask + 1) < xfrm_policy_hashmax &&
552 static inline int xfrm_byidx_should_resize(struct net *net, int total)
554 unsigned int hmask = net->xfrm.policy_idx_hmask;
556 if ((hmask + 1) < xfrm_policy_hashmax &&
563 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
565 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
566 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
567 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
568 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
569 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
570 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
571 si->spdhcnt = net->xfrm.policy_idx_hmask;
572 si->spdhmcnt = xfrm_policy_hashmax;
574 EXPORT_SYMBOL(xfrm_spd_getinfo);
576 static DEFINE_MUTEX(hash_resize_mutex);
577 static void xfrm_hash_resize(struct work_struct *work)
579 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
582 mutex_lock(&hash_resize_mutex);
585 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
586 if (xfrm_bydst_should_resize(net, dir, &total))
587 xfrm_bydst_resize(net, dir);
589 if (xfrm_byidx_should_resize(net, total))
590 xfrm_byidx_resize(net, total);
592 mutex_unlock(&hash_resize_mutex);
595 static void xfrm_hash_rebuild(struct work_struct *work)
597 struct net *net = container_of(work, struct net,
598 xfrm.policy_hthresh.work);
600 struct xfrm_policy *pol;
601 struct xfrm_policy *policy;
602 struct hlist_head *chain;
603 struct hlist_head *odst;
604 struct hlist_node *newpos;
608 u8 lbits4, rbits4, lbits6, rbits6;
610 mutex_lock(&hash_resize_mutex);
612 /* read selector prefixlen thresholds */
614 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
616 lbits4 = net->xfrm.policy_hthresh.lbits4;
617 rbits4 = net->xfrm.policy_hthresh.rbits4;
618 lbits6 = net->xfrm.policy_hthresh.lbits6;
619 rbits6 = net->xfrm.policy_hthresh.rbits6;
620 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
622 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
624 /* reset the bydst and inexact table in all directions */
625 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
626 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
627 hmask = net->xfrm.policy_bydst[dir].hmask;
628 odst = net->xfrm.policy_bydst[dir].table;
629 for (i = hmask; i >= 0; i--)
630 INIT_HLIST_HEAD(odst + i);
631 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
632 /* dir out => dst = remote, src = local */
633 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
634 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
635 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
636 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
638 /* dir in/fwd => dst = local, src = remote */
639 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
640 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
641 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
642 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
646 /* re-insert all policies by order of creation */
647 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
648 if (policy->walk.dead ||
649 xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
650 /* skip socket policies */
654 chain = policy_hash_bysel(net, &policy->selector,
656 xfrm_policy_id2dir(policy->index));
657 hlist_for_each_entry(pol, chain, bydst) {
658 if (policy->priority >= pol->priority)
659 newpos = &pol->bydst;
664 hlist_add_behind_rcu(&policy->bydst, newpos);
666 hlist_add_head_rcu(&policy->bydst, chain);
669 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
671 mutex_unlock(&hash_resize_mutex);
674 void xfrm_policy_hash_rebuild(struct net *net)
676 schedule_work(&net->xfrm.policy_hthresh.work);
678 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
680 /* Generate new index... KAME seems to generate them ordered by cost
681 * of an absolute inpredictability of ordering of rules. This will not pass. */
682 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
684 static u32 idx_generator;
687 struct hlist_head *list;
688 struct xfrm_policy *p;
693 idx = (idx_generator | dir);
702 list = net->xfrm.policy_byidx + idx_hash(net, idx);
704 hlist_for_each_entry(p, list, byidx) {
705 if (p->index == idx) {
715 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
717 u32 *p1 = (u32 *) s1;
718 u32 *p2 = (u32 *) s2;
719 int len = sizeof(struct xfrm_selector) / sizeof(u32);
722 for (i = 0; i < len; i++) {
730 static void xfrm_policy_requeue(struct xfrm_policy *old,
731 struct xfrm_policy *new)
733 struct xfrm_policy_queue *pq = &old->polq;
734 struct sk_buff_head list;
736 if (skb_queue_empty(&pq->hold_queue))
739 __skb_queue_head_init(&list);
741 spin_lock_bh(&pq->hold_queue.lock);
742 skb_queue_splice_init(&pq->hold_queue, &list);
743 if (del_timer(&pq->hold_timer))
745 spin_unlock_bh(&pq->hold_queue.lock);
749 spin_lock_bh(&pq->hold_queue.lock);
750 skb_queue_splice(&list, &pq->hold_queue);
751 pq->timeout = XFRM_QUEUE_TMO_MIN;
752 if (!mod_timer(&pq->hold_timer, jiffies))
754 spin_unlock_bh(&pq->hold_queue.lock);
757 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
758 struct xfrm_policy *pol)
760 if (policy->mark.v == pol->mark.v &&
761 policy->priority == pol->priority)
767 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
769 struct net *net = xp_net(policy);
770 struct xfrm_policy *pol;
771 struct xfrm_policy *delpol;
772 struct hlist_head *chain;
773 struct hlist_node *newpos;
775 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
776 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
779 hlist_for_each_entry(pol, chain, bydst) {
780 if (pol->type == policy->type &&
781 !selector_cmp(&pol->selector, &policy->selector) &&
782 xfrm_policy_mark_match(policy, pol) &&
783 xfrm_sec_ctx_match(pol->security, policy->security) &&
786 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
790 if (policy->priority > pol->priority)
792 } else if (policy->priority >= pol->priority) {
793 newpos = &pol->bydst;
800 hlist_add_behind_rcu(&policy->bydst, newpos);
802 hlist_add_head_rcu(&policy->bydst, chain);
803 __xfrm_policy_link(policy, dir);
804 atomic_inc(&net->xfrm.flow_cache_genid);
806 /* After previous checking, family can either be AF_INET or AF_INET6 */
807 if (policy->family == AF_INET)
808 rt_genid_bump_ipv4(net);
810 rt_genid_bump_ipv6(net);
813 xfrm_policy_requeue(delpol, policy);
814 __xfrm_policy_unlink(delpol, dir);
816 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
817 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
818 policy->curlft.add_time = get_seconds();
819 policy->curlft.use_time = 0;
820 if (!mod_timer(&policy->timer, jiffies + HZ))
821 xfrm_pol_hold(policy);
822 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
825 xfrm_policy_kill(delpol);
826 else if (xfrm_bydst_should_resize(net, dir, NULL))
827 schedule_work(&net->xfrm.policy_hash_work);
831 EXPORT_SYMBOL(xfrm_policy_insert);
833 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
834 int dir, struct xfrm_selector *sel,
835 struct xfrm_sec_ctx *ctx, int delete,
838 struct xfrm_policy *pol, *ret;
839 struct hlist_head *chain;
842 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
843 chain = policy_hash_bysel(net, sel, sel->family, dir);
845 hlist_for_each_entry(pol, chain, bydst) {
846 if (pol->type == type &&
847 (mark & pol->mark.m) == pol->mark.v &&
848 !selector_cmp(sel, &pol->selector) &&
849 xfrm_sec_ctx_match(ctx, pol->security)) {
852 *err = security_xfrm_policy_delete(
855 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
858 __xfrm_policy_unlink(pol, dir);
864 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
867 xfrm_policy_kill(ret);
870 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
872 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
873 int dir, u32 id, int delete, int *err)
875 struct xfrm_policy *pol, *ret;
876 struct hlist_head *chain;
879 if (xfrm_policy_id2dir(id) != dir)
883 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
884 chain = net->xfrm.policy_byidx + idx_hash(net, id);
886 hlist_for_each_entry(pol, chain, byidx) {
887 if (pol->type == type && pol->index == id &&
888 (mark & pol->mark.m) == pol->mark.v) {
891 *err = security_xfrm_policy_delete(
894 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
897 __xfrm_policy_unlink(pol, dir);
903 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
906 xfrm_policy_kill(ret);
909 EXPORT_SYMBOL(xfrm_policy_byid);
911 #ifdef CONFIG_SECURITY_NETWORK_XFRM
913 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
917 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
918 struct xfrm_policy *pol;
921 hlist_for_each_entry(pol,
922 &net->xfrm.policy_inexact[dir], bydst) {
923 if (pol->type != type)
925 err = security_xfrm_policy_delete(pol->security);
927 xfrm_audit_policy_delete(pol, 0, task_valid);
931 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
932 hlist_for_each_entry(pol,
933 net->xfrm.policy_bydst[dir].table + i,
935 if (pol->type != type)
937 err = security_xfrm_policy_delete(
940 xfrm_audit_policy_delete(pol, 0,
951 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
957 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
959 int dir, err = 0, cnt = 0;
961 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
963 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
967 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
968 struct xfrm_policy *pol;
972 hlist_for_each_entry(pol,
973 &net->xfrm.policy_inexact[dir], bydst) {
974 if (pol->type != type)
976 __xfrm_policy_unlink(pol, dir);
977 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
980 xfrm_audit_policy_delete(pol, 1, task_valid);
982 xfrm_policy_kill(pol);
984 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
988 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
990 hlist_for_each_entry(pol,
991 net->xfrm.policy_bydst[dir].table + i,
993 if (pol->type != type)
995 __xfrm_policy_unlink(pol, dir);
996 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
999 xfrm_audit_policy_delete(pol, 1, task_valid);
1000 xfrm_policy_kill(pol);
1002 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1011 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1014 EXPORT_SYMBOL(xfrm_policy_flush);
1016 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1017 int (*func)(struct xfrm_policy *, int, int, void*),
1020 struct xfrm_policy *pol;
1021 struct xfrm_policy_walk_entry *x;
1024 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1025 walk->type != XFRM_POLICY_TYPE_ANY)
1028 if (list_empty(&walk->walk.all) && walk->seq != 0)
1031 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1032 if (list_empty(&walk->walk.all))
1033 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1035 x = list_first_entry(&walk->walk.all,
1036 struct xfrm_policy_walk_entry, all);
1038 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1041 pol = container_of(x, struct xfrm_policy, walk);
1042 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1043 walk->type != pol->type)
1045 error = func(pol, xfrm_policy_id2dir(pol->index),
1048 list_move_tail(&walk->walk.all, &x->all);
1053 if (walk->seq == 0) {
1057 list_del_init(&walk->walk.all);
1059 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1062 EXPORT_SYMBOL(xfrm_policy_walk);
1064 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1066 INIT_LIST_HEAD(&walk->walk.all);
1067 walk->walk.dead = 1;
1071 EXPORT_SYMBOL(xfrm_policy_walk_init);
1073 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1075 if (list_empty(&walk->walk.all))
1078 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1079 list_del(&walk->walk.all);
1080 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1082 EXPORT_SYMBOL(xfrm_policy_walk_done);
1085 * Find policy to apply to this flow.
1087 * Returns 0 if policy found, else an -errno.
1089 static int xfrm_policy_match(const struct xfrm_policy *pol,
1090 const struct flowi *fl,
1091 u8 type, u16 family, int dir)
1093 const struct xfrm_selector *sel = &pol->selector;
1097 if (pol->family != family ||
1098 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1102 match = xfrm_selector_match(sel, fl, family);
1104 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1110 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1111 const struct flowi *fl,
1115 struct xfrm_policy *pol, *ret;
1116 const xfrm_address_t *daddr, *saddr;
1117 struct hlist_head *chain;
1118 unsigned int sequence;
1121 daddr = xfrm_flowi_daddr(fl, family);
1122 saddr = xfrm_flowi_saddr(fl, family);
1123 if (unlikely(!daddr || !saddr))
1129 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1130 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1131 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1135 hlist_for_each_entry_rcu(pol, chain, bydst) {
1136 err = xfrm_policy_match(pol, fl, type, family, dir);
1146 priority = ret->priority;
1150 chain = &net->xfrm.policy_inexact[dir];
1151 hlist_for_each_entry_rcu(pol, chain, bydst) {
1152 if ((pol->priority >= priority) && ret)
1155 err = xfrm_policy_match(pol, fl, type, family, dir);
1169 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1172 if (ret && !xfrm_pol_hold_rcu(ret))
1180 static struct xfrm_policy *
1181 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1183 #ifdef CONFIG_XFRM_SUB_POLICY
1184 struct xfrm_policy *pol;
1186 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1190 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1193 static int flow_to_policy_dir(int dir)
1195 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1196 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1197 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1203 return XFRM_POLICY_IN;
1205 return XFRM_POLICY_OUT;
1207 return XFRM_POLICY_FWD;
1211 static struct flow_cache_object *
1212 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1213 u8 dir, struct flow_cache_object *old_obj, void *ctx)
1215 struct xfrm_policy *pol;
1218 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1220 pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1221 if (IS_ERR_OR_NULL(pol))
1222 return ERR_CAST(pol);
1224 /* Resolver returns two references:
1225 * one for cache and one for caller of flow_cache_lookup() */
1231 static inline int policy_to_flow_dir(int dir)
1233 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1234 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1235 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1239 case XFRM_POLICY_IN:
1241 case XFRM_POLICY_OUT:
1242 return FLOW_DIR_OUT;
1243 case XFRM_POLICY_FWD:
1244 return FLOW_DIR_FWD;
1248 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1249 const struct flowi *fl, u16 family)
1251 struct xfrm_policy *pol;
1255 pol = rcu_dereference(sk->sk_policy[dir]);
1260 if (pol->family != family) {
1265 match = xfrm_selector_match(&pol->selector, fl, family);
1267 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1271 err = security_xfrm_policy_lookup(pol->security,
1273 policy_to_flow_dir(dir));
1275 if (!xfrm_pol_hold_rcu(pol))
1277 } else if (err == -ESRCH) {
1290 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1292 struct net *net = xp_net(pol);
1294 list_add(&pol->walk.all, &net->xfrm.policy_all);
1295 net->xfrm.policy_count[dir]++;
1299 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1302 struct net *net = xp_net(pol);
1304 if (list_empty(&pol->walk.all))
1307 /* Socket policies are not hashed. */
1308 if (!hlist_unhashed(&pol->bydst)) {
1309 hlist_del_rcu(&pol->bydst);
1310 hlist_del(&pol->byidx);
1313 list_del_init(&pol->walk.all);
1314 net->xfrm.policy_count[dir]--;
1319 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1321 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1324 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1326 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1329 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1331 struct net *net = xp_net(pol);
1333 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1334 pol = __xfrm_policy_unlink(pol, dir);
1335 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1337 xfrm_policy_kill(pol);
1342 EXPORT_SYMBOL(xfrm_policy_delete);
1344 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1346 struct net *net = sock_net(sk);
1347 struct xfrm_policy *old_pol;
1349 #ifdef CONFIG_XFRM_SUB_POLICY
1350 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1354 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1355 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1356 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1358 pol->curlft.add_time = get_seconds();
1359 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1360 xfrm_sk_policy_link(pol, dir);
1362 rcu_assign_pointer(sk->sk_policy[dir], pol);
1365 xfrm_policy_requeue(old_pol, pol);
1367 /* Unlinking succeeds always. This is the only function
1368 * allowed to delete or replace socket policy.
1370 xfrm_sk_policy_unlink(old_pol, dir);
1372 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1375 xfrm_policy_kill(old_pol);
1380 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1382 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1383 struct net *net = xp_net(old);
1386 newp->selector = old->selector;
1387 if (security_xfrm_policy_clone(old->security,
1390 return NULL; /* ENOMEM */
1392 newp->lft = old->lft;
1393 newp->curlft = old->curlft;
1394 newp->mark = old->mark;
1395 newp->action = old->action;
1396 newp->flags = old->flags;
1397 newp->xfrm_nr = old->xfrm_nr;
1398 newp->index = old->index;
1399 newp->type = old->type;
1400 newp->family = old->family;
1401 memcpy(newp->xfrm_vec, old->xfrm_vec,
1402 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1403 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1404 xfrm_sk_policy_link(newp, dir);
1405 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1411 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1413 const struct xfrm_policy *p;
1414 struct xfrm_policy *np;
1418 for (i = 0; i < 2; i++) {
1419 p = rcu_dereference(osk->sk_policy[i]);
1421 np = clone_policy(p, i);
1422 if (unlikely(!np)) {
1426 rcu_assign_pointer(sk->sk_policy[i], np);
1434 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1435 xfrm_address_t *remote, unsigned short family)
1438 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1440 if (unlikely(afinfo == NULL))
1442 err = afinfo->get_saddr(net, oif, local, remote);
1443 xfrm_policy_put_afinfo(afinfo);
1447 /* Resolve list of templates for the flow, given policy. */
1450 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1451 struct xfrm_state **xfrm, unsigned short family)
1453 struct net *net = xp_net(policy);
1456 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1457 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1460 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1461 struct xfrm_state *x;
1462 xfrm_address_t *remote = daddr;
1463 xfrm_address_t *local = saddr;
1464 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1466 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1467 tmpl->mode == XFRM_MODE_BEET) {
1468 remote = &tmpl->id.daddr;
1469 local = &tmpl->saddr;
1470 if (xfrm_addr_any(local, tmpl->encap_family)) {
1471 error = xfrm_get_saddr(net, fl->flowi_oif,
1473 tmpl->encap_family);
1480 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1482 if (x && x->km.state == XFRM_STATE_VALID) {
1489 error = (x->km.state == XFRM_STATE_ERROR ?
1492 } else if (error == -ESRCH) {
1496 if (!tmpl->optional)
1502 for (nx--; nx >= 0; nx--)
1503 xfrm_state_put(xfrm[nx]);
1508 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1509 struct xfrm_state **xfrm, unsigned short family)
1511 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1512 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1518 for (i = 0; i < npols; i++) {
1519 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1524 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1532 /* found states are sorted for outbound processing */
1534 xfrm_state_sort(xfrm, tpp, cnx, family);
1539 for (cnx--; cnx >= 0; cnx--)
1540 xfrm_state_put(tpp[cnx]);
1545 /* Check that the bundle accepts the flow and its components are
1549 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1551 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1557 tos = afinfo->get_tos(fl);
1559 xfrm_policy_put_afinfo(afinfo);
1564 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1566 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1567 struct dst_entry *dst = &xdst->u.dst;
1569 if (xdst->route == NULL) {
1570 /* Dummy bundle - if it has xfrms we were not
1571 * able to build bundle as template resolution failed.
1572 * It means we need to try again resolving. */
1573 if (xdst->num_xfrms > 0)
1575 } else if (dst->flags & DST_XFRM_QUEUE) {
1579 if (stale_bundle(dst))
1587 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1589 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1590 struct dst_entry *dst = &xdst->u.dst;
1594 if (stale_bundle(dst))
1600 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1602 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1603 struct dst_entry *dst = &xdst->u.dst;
1608 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1609 .get = xfrm_bundle_flo_get,
1610 .check = xfrm_bundle_flo_check,
1611 .delete = xfrm_bundle_flo_delete,
1614 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1616 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1617 struct dst_ops *dst_ops;
1618 struct xfrm_dst *xdst;
1621 return ERR_PTR(-EINVAL);
1625 dst_ops = &net->xfrm.xfrm4_dst_ops;
1627 #if IS_ENABLED(CONFIG_IPV6)
1629 dst_ops = &net->xfrm.xfrm6_dst_ops;
1635 xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1638 struct dst_entry *dst = &xdst->u.dst;
1640 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1641 xdst->flo.ops = &xfrm_bundle_fc_ops;
1643 xdst = ERR_PTR(-ENOBUFS);
1645 xfrm_policy_put_afinfo(afinfo);
1650 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1653 struct xfrm_policy_afinfo *afinfo =
1654 xfrm_policy_get_afinfo(dst->ops->family);
1660 err = afinfo->init_path(path, dst, nfheader_len);
1662 xfrm_policy_put_afinfo(afinfo);
1667 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1668 const struct flowi *fl)
1670 struct xfrm_policy_afinfo *afinfo =
1671 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1677 err = afinfo->fill_dst(xdst, dev, fl);
1679 xfrm_policy_put_afinfo(afinfo);
1685 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1686 * all the metrics... Shortly, bundle a bundle.
1689 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1690 struct xfrm_state **xfrm, int nx,
1691 const struct flowi *fl,
1692 struct dst_entry *dst)
1694 struct net *net = xp_net(policy);
1695 unsigned long now = jiffies;
1696 struct net_device *dev;
1697 struct xfrm_mode *inner_mode;
1698 struct dst_entry *dst_prev = NULL;
1699 struct dst_entry *dst0 = NULL;
1703 int nfheader_len = 0;
1704 int trailer_len = 0;
1706 int family = policy->selector.family;
1707 xfrm_address_t saddr, daddr;
1709 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1711 tos = xfrm_get_tos(fl, family);
1718 for (; i < nx; i++) {
1719 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1720 struct dst_entry *dst1 = &xdst->u.dst;
1722 err = PTR_ERR(xdst);
1728 if (xfrm[i]->sel.family == AF_UNSPEC) {
1729 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1730 xfrm_af2proto(family));
1732 err = -EAFNOSUPPORT;
1737 inner_mode = xfrm[i]->inner_mode;
1742 dst_prev->child = dst_clone(dst1);
1743 dst1->flags |= DST_NOHASH;
1747 dst_copy_metrics(dst1, dst);
1749 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1750 family = xfrm[i]->props.family;
1751 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1752 &saddr, &daddr, family);
1759 dst1->xfrm = xfrm[i];
1760 xdst->xfrm_genid = xfrm[i]->genid;
1762 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1763 dst1->flags |= DST_HOST;
1764 dst1->lastuse = now;
1766 dst1->input = dst_discard;
1767 dst1->output = inner_mode->afinfo->output;
1769 dst1->next = dst_prev;
1772 header_len += xfrm[i]->props.header_len;
1773 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1774 nfheader_len += xfrm[i]->props.header_len;
1775 trailer_len += xfrm[i]->props.trailer_len;
1778 dst_prev->child = dst;
1786 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1787 xfrm_init_pmtu(dst_prev);
1789 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1790 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1792 err = xfrm_fill_dst(xdst, dev, fl);
1796 dst_prev->header_len = header_len;
1797 dst_prev->trailer_len = trailer_len;
1798 header_len -= xdst->u.dst.xfrm->props.header_len;
1799 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1807 xfrm_state_put(xfrm[i]);
1811 dst0 = ERR_PTR(err);
1815 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1816 struct xfrm_policy **pols,
1817 int *num_pols, int *num_xfrms)
1821 if (*num_pols == 0 || !pols[0]) {
1826 if (IS_ERR(pols[0]))
1827 return PTR_ERR(pols[0]);
1829 *num_xfrms = pols[0]->xfrm_nr;
1831 #ifdef CONFIG_XFRM_SUB_POLICY
1832 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1833 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1834 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1835 XFRM_POLICY_TYPE_MAIN,
1839 if (IS_ERR(pols[1])) {
1840 xfrm_pols_put(pols, *num_pols);
1841 return PTR_ERR(pols[1]);
1844 (*num_xfrms) += pols[1]->xfrm_nr;
1848 for (i = 0; i < *num_pols; i++) {
1849 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1859 static struct xfrm_dst *
1860 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1861 const struct flowi *fl, u16 family,
1862 struct dst_entry *dst_orig)
1864 struct net *net = xp_net(pols[0]);
1865 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1866 struct dst_entry *dst;
1867 struct xfrm_dst *xdst;
1870 /* Try to instantiate a bundle */
1871 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1877 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1878 return ERR_PTR(err);
1881 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1883 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1884 return ERR_CAST(dst);
1887 xdst = (struct xfrm_dst *)dst;
1888 xdst->num_xfrms = err;
1889 xdst->num_pols = num_pols;
1890 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1891 xdst->policy_genid = atomic_read(&pols[0]->genid);
1896 static void xfrm_policy_queue_process(unsigned long arg)
1898 struct sk_buff *skb;
1900 struct dst_entry *dst;
1901 struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1902 struct net *net = xp_net(pol);
1903 struct xfrm_policy_queue *pq = &pol->polq;
1905 struct sk_buff_head list;
1907 spin_lock(&pq->hold_queue.lock);
1908 skb = skb_peek(&pq->hold_queue);
1910 spin_unlock(&pq->hold_queue.lock);
1915 xfrm_decode_session(skb, &fl, dst->ops->family);
1916 spin_unlock(&pq->hold_queue.lock);
1918 dst_hold(dst->path);
1919 dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1923 if (dst->flags & DST_XFRM_QUEUE) {
1926 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1929 pq->timeout = pq->timeout << 1;
1930 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1937 __skb_queue_head_init(&list);
1939 spin_lock(&pq->hold_queue.lock);
1941 skb_queue_splice_init(&pq->hold_queue, &list);
1942 spin_unlock(&pq->hold_queue.lock);
1944 while (!skb_queue_empty(&list)) {
1945 skb = __skb_dequeue(&list);
1947 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1948 dst_hold(skb_dst(skb)->path);
1949 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1957 skb_dst_set(skb, dst);
1959 dst_output(net, skb->sk, skb);
1968 skb_queue_purge(&pq->hold_queue);
1972 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1974 unsigned long sched_next;
1975 struct dst_entry *dst = skb_dst(skb);
1976 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1977 struct xfrm_policy *pol = xdst->pols[0];
1978 struct xfrm_policy_queue *pq = &pol->polq;
1980 if (unlikely(skb_fclone_busy(sk, skb))) {
1985 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1992 spin_lock_bh(&pq->hold_queue.lock);
1995 pq->timeout = XFRM_QUEUE_TMO_MIN;
1997 sched_next = jiffies + pq->timeout;
1999 if (del_timer(&pq->hold_timer)) {
2000 if (time_before(pq->hold_timer.expires, sched_next))
2001 sched_next = pq->hold_timer.expires;
2005 __skb_queue_tail(&pq->hold_queue, skb);
2006 if (!mod_timer(&pq->hold_timer, sched_next))
2009 spin_unlock_bh(&pq->hold_queue.lock);
2014 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2015 struct xfrm_flo *xflo,
2016 const struct flowi *fl,
2021 struct net_device *dev;
2022 struct dst_entry *dst;
2023 struct dst_entry *dst1;
2024 struct xfrm_dst *xdst;
2026 xdst = xfrm_alloc_dst(net, family);
2030 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2031 net->xfrm.sysctl_larval_drop ||
2035 dst = xflo->dst_orig;
2036 dst1 = &xdst->u.dst;
2040 dst_copy_metrics(dst1, dst);
2042 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2043 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2044 dst1->lastuse = jiffies;
2046 dst1->input = dst_discard;
2047 dst1->output = xdst_queue_output;
2053 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2060 err = xfrm_fill_dst(xdst, dev, fl);
2069 xdst = ERR_PTR(err);
2073 static struct flow_cache_object *
2074 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2075 struct flow_cache_object *oldflo, void *ctx)
2077 struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2078 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2079 struct xfrm_dst *xdst, *new_xdst;
2080 int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2082 /* Check if the policies from old bundle are usable */
2085 xdst = container_of(oldflo, struct xfrm_dst, flo);
2086 num_pols = xdst->num_pols;
2087 num_xfrms = xdst->num_xfrms;
2089 for (i = 0; i < num_pols; i++) {
2090 pols[i] = xdst->pols[i];
2091 pol_dead |= pols[i]->walk.dead;
2094 dst_free(&xdst->u.dst);
2102 /* Resolve policies to use if we couldn't get them from
2103 * previous cache entry */
2106 pols[0] = __xfrm_policy_lookup(net, fl, family,
2107 flow_to_policy_dir(dir));
2108 err = xfrm_expand_policies(fl, family, pols,
2109 &num_pols, &num_xfrms);
2115 goto make_dummy_bundle;
2118 new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2120 if (IS_ERR(new_xdst)) {
2121 err = PTR_ERR(new_xdst);
2125 goto make_dummy_bundle;
2126 dst_hold(&xdst->u.dst);
2128 } else if (new_xdst == NULL) {
2131 goto make_dummy_bundle;
2132 xdst->num_xfrms = 0;
2133 dst_hold(&xdst->u.dst);
2137 /* Kill the previous bundle */
2139 /* The policies were stolen for newly generated bundle */
2141 dst_free(&xdst->u.dst);
2144 /* Flow cache does not have reference, it dst_free()'s,
2145 * but we do need to return one reference for original caller */
2146 dst_hold(&new_xdst->u.dst);
2147 return &new_xdst->flo;
2150 /* We found policies, but there's no bundles to instantiate:
2151 * either because the policy blocks, has no transformations or
2152 * we could not build template (no xfrm_states).*/
2153 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2155 xfrm_pols_put(pols, num_pols);
2156 return ERR_CAST(xdst);
2158 xdst->num_pols = num_pols;
2159 xdst->num_xfrms = num_xfrms;
2160 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2162 dst_hold(&xdst->u.dst);
2166 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2169 dst_free(&xdst->u.dst);
2171 xfrm_pols_put(pols, num_pols);
2172 return ERR_PTR(err);
2175 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2176 struct dst_entry *dst_orig)
2178 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2179 struct dst_entry *ret;
2182 dst_release(dst_orig);
2183 return ERR_PTR(-EINVAL);
2185 ret = afinfo->blackhole_route(net, dst_orig);
2187 xfrm_policy_put_afinfo(afinfo);
2192 /* Main function: finds/creates a bundle for given flow.
2194 * At the moment we eat a raw IP route. Mostly to speed up lookups
2195 * on interfaces with disabled IPsec.
2197 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2198 const struct flowi *fl,
2199 const struct sock *sk, int flags)
2201 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2202 struct flow_cache_object *flo;
2203 struct xfrm_dst *xdst;
2204 struct dst_entry *dst, *route;
2205 u16 family = dst_orig->ops->family;
2206 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2207 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2213 sk = sk_const_to_full_sk(sk);
2214 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2216 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2217 err = xfrm_expand_policies(fl, family, pols,
2218 &num_pols, &num_xfrms);
2223 if (num_xfrms <= 0) {
2224 drop_pols = num_pols;
2228 xdst = xfrm_resolve_and_create_bundle(
2232 xfrm_pols_put(pols, num_pols);
2233 err = PTR_ERR(xdst);
2235 } else if (xdst == NULL) {
2237 drop_pols = num_pols;
2241 dst_hold(&xdst->u.dst);
2242 xdst->u.dst.flags |= DST_NOCACHE;
2243 route = xdst->route;
2248 struct xfrm_flo xflo;
2250 xflo.dst_orig = dst_orig;
2253 /* To accelerate a bit... */
2254 if ((dst_orig->flags & DST_NOXFRM) ||
2255 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2258 flo = flow_cache_lookup(net, fl, family, dir,
2259 xfrm_bundle_lookup, &xflo);
2266 xdst = container_of(flo, struct xfrm_dst, flo);
2268 num_pols = xdst->num_pols;
2269 num_xfrms = xdst->num_xfrms;
2270 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2271 route = xdst->route;
2275 if (route == NULL && num_xfrms > 0) {
2276 /* The only case when xfrm_bundle_lookup() returns a
2277 * bundle with null route, is when the template could
2278 * not be resolved. It means policies are there, but
2279 * bundle could not be created, since we don't yet
2280 * have the xfrm_state's. We need to wait for KM to
2281 * negotiate new SA's or bail out with error.*/
2282 if (net->xfrm.sysctl_larval_drop) {
2283 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2290 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2298 if ((flags & XFRM_LOOKUP_ICMP) &&
2299 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2304 for (i = 0; i < num_pols; i++)
2305 pols[i]->curlft.use_time = get_seconds();
2307 if (num_xfrms < 0) {
2308 /* Prohibit the flow */
2309 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2312 } else if (num_xfrms > 0) {
2313 /* Flow transformed */
2314 dst_release(dst_orig);
2316 /* Flow passes untransformed */
2321 xfrm_pols_put(pols, drop_pols);
2322 if (dst && dst->xfrm &&
2323 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2324 dst->flags |= DST_XFRM_TUNNEL;
2328 if (!(flags & XFRM_LOOKUP_ICMP)) {
2336 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2337 dst_release(dst_orig);
2338 xfrm_pols_put(pols, drop_pols);
2339 return ERR_PTR(err);
2341 EXPORT_SYMBOL(xfrm_lookup);
2343 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2344 * Otherwise we may send out blackholed packets.
2346 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2347 const struct flowi *fl,
2348 const struct sock *sk, int flags)
2350 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2351 flags | XFRM_LOOKUP_QUEUE |
2352 XFRM_LOOKUP_KEEP_DST_REF);
2354 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2355 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2358 dst_release(dst_orig);
2362 EXPORT_SYMBOL(xfrm_lookup_route);
2365 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2367 struct xfrm_state *x;
2369 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2371 x = skb->sp->xvec[idx];
2372 if (!x->type->reject)
2374 return x->type->reject(x, skb, fl);
2377 /* When skb is transformed back to its "native" form, we have to
2378 * check policy restrictions. At the moment we make this in maximally
2379 * stupid way. Shame on me. :-) Of course, connected sockets must
2380 * have policy cached at them.
2384 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2385 unsigned short family)
2387 if (xfrm_state_kern(x))
2388 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2389 return x->id.proto == tmpl->id.proto &&
2390 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2391 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2392 x->props.mode == tmpl->mode &&
2393 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2394 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2395 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2396 xfrm_state_addr_cmp(tmpl, x, family));
2400 * 0 or more than 0 is returned when validation is succeeded (either bypass
2401 * because of optional transport mode, or next index of the mathced secpath
2402 * state with the template.
2403 * -1 is returned when no matching template is found.
2404 * Otherwise "-2 - errored_index" is returned.
2407 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2408 unsigned short family)
2412 if (tmpl->optional) {
2413 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2417 for (; idx < sp->len; idx++) {
2418 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2420 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2429 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2430 unsigned int family, int reverse)
2432 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2435 if (unlikely(afinfo == NULL))
2436 return -EAFNOSUPPORT;
2438 afinfo->decode_session(skb, fl, reverse);
2439 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2440 xfrm_policy_put_afinfo(afinfo);
2443 EXPORT_SYMBOL(__xfrm_decode_session);
2445 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2447 for (; k < sp->len; k++) {
2448 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2457 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2458 unsigned short family)
2460 struct net *net = dev_net(skb->dev);
2461 struct xfrm_policy *pol;
2462 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2471 reverse = dir & ~XFRM_POLICY_MASK;
2472 dir &= XFRM_POLICY_MASK;
2473 fl_dir = policy_to_flow_dir(dir);
2475 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2476 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2480 nf_nat_decode_session(skb, &fl, family);
2482 /* First, check used SA against their selectors. */
2486 for (i = skb->sp->len-1; i >= 0; i--) {
2487 struct xfrm_state *x = skb->sp->xvec[i];
2488 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2489 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2496 sk = sk_to_full_sk(sk);
2497 if (sk && sk->sk_policy[dir]) {
2498 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2500 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2506 struct flow_cache_object *flo;
2508 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2509 xfrm_policy_lookup, NULL);
2510 if (IS_ERR_OR_NULL(flo))
2511 pol = ERR_CAST(flo);
2513 pol = container_of(flo, struct xfrm_policy, flo);
2517 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2522 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2523 xfrm_secpath_reject(xerr_idx, skb, &fl);
2524 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2530 pol->curlft.use_time = get_seconds();
2534 #ifdef CONFIG_XFRM_SUB_POLICY
2535 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2536 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2540 if (IS_ERR(pols[1])) {
2541 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2544 pols[1]->curlft.use_time = get_seconds();
2550 if (pol->action == XFRM_POLICY_ALLOW) {
2551 struct sec_path *sp;
2552 static struct sec_path dummy;
2553 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2554 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2555 struct xfrm_tmpl **tpp = tp;
2559 if ((sp = skb->sp) == NULL)
2562 for (pi = 0; pi < npols; pi++) {
2563 if (pols[pi] != pol &&
2564 pols[pi]->action != XFRM_POLICY_ALLOW) {
2565 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2568 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2569 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2572 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2573 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2577 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2581 /* For each tunnel xfrm, find the first matching tmpl.
2582 * For each tmpl before that, find corresponding xfrm.
2583 * Order is _important_. Later we will implement
2584 * some barriers, but at the moment barriers
2585 * are implied between each two transformations.
2587 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2588 k = xfrm_policy_ok(tpp[i], sp, k, family);
2591 /* "-2 - errored_index" returned */
2593 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2598 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2599 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2603 xfrm_pols_put(pols, npols);
2606 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2609 xfrm_secpath_reject(xerr_idx, skb, &fl);
2611 xfrm_pols_put(pols, npols);
2614 EXPORT_SYMBOL(__xfrm_policy_check);
2616 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2618 struct net *net = dev_net(skb->dev);
2620 struct dst_entry *dst;
2623 if (xfrm_decode_session(skb, &fl, family) < 0) {
2624 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2630 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2635 skb_dst_set(skb, dst);
2638 EXPORT_SYMBOL(__xfrm_route_forward);
2640 /* Optimize later using cookies and generation ids. */
2642 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2644 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2645 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2646 * get validated by dst_ops->check on every use. We do this
2647 * because when a normal route referenced by an XFRM dst is
2648 * obsoleted we do not go looking around for all parent
2649 * referencing XFRM dsts so that we can invalidate them. It
2650 * is just too much work. Instead we make the checks here on
2651 * every use. For example:
2653 * XFRM dst A --> IPv4 dst X
2655 * X is the "xdst->route" of A (X is also the "dst->path" of A
2656 * in this example). If X is marked obsolete, "A" will not
2657 * notice. That's what we are validating here via the
2658 * stale_bundle() check.
2660 * When a policy's bundle is pruned, we dst_free() the XFRM
2661 * dst which causes it's ->obsolete field to be set to
2662 * DST_OBSOLETE_DEAD. If an XFRM dst has been pruned like
2663 * this, we want to force a new route lookup.
2665 if (dst->obsolete < 0 && !stale_bundle(dst))
2671 static int stale_bundle(struct dst_entry *dst)
2673 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2676 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2678 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2679 dst->dev = dev_net(dev)->loopback_dev;
2684 EXPORT_SYMBOL(xfrm_dst_ifdown);
2686 static void xfrm_link_failure(struct sk_buff *skb)
2688 /* Impossible. Such dst must be popped before reaches point of failure. */
2691 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2694 if (dst->obsolete) {
2702 void xfrm_garbage_collect(struct net *net)
2704 flow_cache_flush(net);
2706 EXPORT_SYMBOL(xfrm_garbage_collect);
2708 static void xfrm_garbage_collect_deferred(struct net *net)
2710 flow_cache_flush_deferred(net);
2713 static void xfrm_init_pmtu(struct dst_entry *dst)
2716 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2717 u32 pmtu, route_mtu_cached;
2719 pmtu = dst_mtu(dst->child);
2720 xdst->child_mtu_cached = pmtu;
2722 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2724 route_mtu_cached = dst_mtu(xdst->route);
2725 xdst->route_mtu_cached = route_mtu_cached;
2727 if (pmtu > route_mtu_cached)
2728 pmtu = route_mtu_cached;
2730 dst_metric_set(dst, RTAX_MTU, pmtu);
2731 } while ((dst = dst->next));
2734 /* Check that the bundle accepts the flow and its components are
2738 static int xfrm_bundle_ok(struct xfrm_dst *first)
2740 struct dst_entry *dst = &first->u.dst;
2741 struct xfrm_dst *last;
2744 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2745 (dst->dev && !netif_running(dst->dev)))
2748 if (dst->flags & DST_XFRM_QUEUE)
2754 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2756 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2758 if (xdst->xfrm_genid != dst->xfrm->genid)
2760 if (xdst->num_pols > 0 &&
2761 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2764 mtu = dst_mtu(dst->child);
2765 if (xdst->child_mtu_cached != mtu) {
2767 xdst->child_mtu_cached = mtu;
2770 if (!dst_check(xdst->route, xdst->route_cookie))
2772 mtu = dst_mtu(xdst->route);
2773 if (xdst->route_mtu_cached != mtu) {
2775 xdst->route_mtu_cached = mtu;
2779 } while (dst->xfrm);
2784 mtu = last->child_mtu_cached;
2788 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2789 if (mtu > last->route_mtu_cached)
2790 mtu = last->route_mtu_cached;
2791 dst_metric_set(dst, RTAX_MTU, mtu);
2796 last = (struct xfrm_dst *)last->u.dst.next;
2797 last->child_mtu_cached = mtu;
2803 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2805 return dst_metric_advmss(dst->path);
2808 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2810 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2812 return mtu ? : dst_mtu(dst->path);
2815 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2816 struct sk_buff *skb,
2819 return dst->path->ops->neigh_lookup(dst, skb, daddr);
2822 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2825 if (unlikely(afinfo == NULL))
2827 if (unlikely(afinfo->family >= NPROTO))
2828 return -EAFNOSUPPORT;
2829 spin_lock(&xfrm_policy_afinfo_lock);
2830 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2833 struct dst_ops *dst_ops = afinfo->dst_ops;
2834 if (likely(dst_ops->kmem_cachep == NULL))
2835 dst_ops->kmem_cachep = xfrm_dst_cache;
2836 if (likely(dst_ops->check == NULL))
2837 dst_ops->check = xfrm_dst_check;
2838 if (likely(dst_ops->default_advmss == NULL))
2839 dst_ops->default_advmss = xfrm_default_advmss;
2840 if (likely(dst_ops->mtu == NULL))
2841 dst_ops->mtu = xfrm_mtu;
2842 if (likely(dst_ops->negative_advice == NULL))
2843 dst_ops->negative_advice = xfrm_negative_advice;
2844 if (likely(dst_ops->link_failure == NULL))
2845 dst_ops->link_failure = xfrm_link_failure;
2846 if (likely(dst_ops->neigh_lookup == NULL))
2847 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2848 if (likely(afinfo->garbage_collect == NULL))
2849 afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2850 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2852 spin_unlock(&xfrm_policy_afinfo_lock);
2856 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2858 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2861 if (unlikely(afinfo == NULL))
2863 if (unlikely(afinfo->family >= NPROTO))
2864 return -EAFNOSUPPORT;
2865 spin_lock(&xfrm_policy_afinfo_lock);
2866 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2867 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2870 RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2873 spin_unlock(&xfrm_policy_afinfo_lock);
2875 struct dst_ops *dst_ops = afinfo->dst_ops;
2879 dst_ops->kmem_cachep = NULL;
2880 dst_ops->check = NULL;
2881 dst_ops->negative_advice = NULL;
2882 dst_ops->link_failure = NULL;
2883 afinfo->garbage_collect = NULL;
2887 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2889 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2891 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2895 xfrm_garbage_collect(dev_net(dev));
2900 static struct notifier_block xfrm_dev_notifier = {
2901 .notifier_call = xfrm_dev_event,
2904 #ifdef CONFIG_XFRM_STATISTICS
2905 static int __net_init xfrm_statistics_init(struct net *net)
2908 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2909 if (!net->mib.xfrm_statistics)
2911 rv = xfrm_proc_init(net);
2913 free_percpu(net->mib.xfrm_statistics);
2917 static void xfrm_statistics_fini(struct net *net)
2919 xfrm_proc_fini(net);
2920 free_percpu(net->mib.xfrm_statistics);
2923 static int __net_init xfrm_statistics_init(struct net *net)
2928 static void xfrm_statistics_fini(struct net *net)
2933 static int __net_init xfrm_policy_init(struct net *net)
2935 unsigned int hmask, sz;
2938 if (net_eq(net, &init_net))
2939 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2940 sizeof(struct xfrm_dst),
2941 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2945 sz = (hmask+1) * sizeof(struct hlist_head);
2947 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2948 if (!net->xfrm.policy_byidx)
2950 net->xfrm.policy_idx_hmask = hmask;
2952 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2953 struct xfrm_policy_hash *htab;
2955 net->xfrm.policy_count[dir] = 0;
2956 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2957 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2959 htab = &net->xfrm.policy_bydst[dir];
2960 htab->table = xfrm_hash_alloc(sz);
2963 htab->hmask = hmask;
2969 net->xfrm.policy_hthresh.lbits4 = 32;
2970 net->xfrm.policy_hthresh.rbits4 = 32;
2971 net->xfrm.policy_hthresh.lbits6 = 128;
2972 net->xfrm.policy_hthresh.rbits6 = 128;
2974 seqlock_init(&net->xfrm.policy_hthresh.lock);
2976 INIT_LIST_HEAD(&net->xfrm.policy_all);
2977 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2978 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2979 if (net_eq(net, &init_net))
2980 register_netdevice_notifier(&xfrm_dev_notifier);
2984 for (dir--; dir >= 0; dir--) {
2985 struct xfrm_policy_hash *htab;
2987 htab = &net->xfrm.policy_bydst[dir];
2988 xfrm_hash_free(htab->table, sz);
2990 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2995 static void xfrm_policy_fini(struct net *net)
3000 flush_work(&net->xfrm.policy_hash_work);
3001 #ifdef CONFIG_XFRM_SUB_POLICY
3002 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3004 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3006 WARN_ON(!list_empty(&net->xfrm.policy_all));
3008 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3009 struct xfrm_policy_hash *htab;
3011 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3013 htab = &net->xfrm.policy_bydst[dir];
3014 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3015 WARN_ON(!hlist_empty(htab->table));
3016 xfrm_hash_free(htab->table, sz);
3019 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3020 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3021 xfrm_hash_free(net->xfrm.policy_byidx, sz);
3024 static int __net_init xfrm_net_init(struct net *net)
3028 /* Initialize the per-net locks here */
3029 spin_lock_init(&net->xfrm.xfrm_state_lock);
3030 spin_lock_init(&net->xfrm.xfrm_policy_lock);
3031 mutex_init(&net->xfrm.xfrm_cfg_mutex);
3033 rv = xfrm_statistics_init(net);
3035 goto out_statistics;
3036 rv = xfrm_state_init(net);
3039 rv = xfrm_policy_init(net);
3042 rv = xfrm_sysctl_init(net);
3045 rv = flow_cache_init(net);
3052 xfrm_sysctl_fini(net);
3054 xfrm_policy_fini(net);
3056 xfrm_state_fini(net);
3058 xfrm_statistics_fini(net);
3063 static void __net_exit xfrm_net_exit(struct net *net)
3065 flow_cache_fini(net);
3066 xfrm_sysctl_fini(net);
3067 xfrm_policy_fini(net);
3068 xfrm_state_fini(net);
3069 xfrm_statistics_fini(net);
3072 static struct pernet_operations __net_initdata xfrm_net_ops = {
3073 .init = xfrm_net_init,
3074 .exit = xfrm_net_exit,
3077 void __init xfrm_init(void)
3079 register_pernet_subsys(&xfrm_net_ops);
3080 seqcount_init(&xfrm_policy_hash_generation);
3084 #ifdef CONFIG_AUDITSYSCALL
3085 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3086 struct audit_buffer *audit_buf)
3088 struct xfrm_sec_ctx *ctx = xp->security;
3089 struct xfrm_selector *sel = &xp->selector;
3092 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3093 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3095 switch (sel->family) {
3097 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3098 if (sel->prefixlen_s != 32)
3099 audit_log_format(audit_buf, " src_prefixlen=%d",
3101 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3102 if (sel->prefixlen_d != 32)
3103 audit_log_format(audit_buf, " dst_prefixlen=%d",
3107 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3108 if (sel->prefixlen_s != 128)
3109 audit_log_format(audit_buf, " src_prefixlen=%d",
3111 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3112 if (sel->prefixlen_d != 128)
3113 audit_log_format(audit_buf, " dst_prefixlen=%d",
3119 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3121 struct audit_buffer *audit_buf;
3123 audit_buf = xfrm_audit_start("SPD-add");
3124 if (audit_buf == NULL)
3126 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3127 audit_log_format(audit_buf, " res=%u", result);
3128 xfrm_audit_common_policyinfo(xp, audit_buf);
3129 audit_log_end(audit_buf);
3131 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3133 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3136 struct audit_buffer *audit_buf;
3138 audit_buf = xfrm_audit_start("SPD-delete");
3139 if (audit_buf == NULL)
3141 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3142 audit_log_format(audit_buf, " res=%u", result);
3143 xfrm_audit_common_policyinfo(xp, audit_buf);
3144 audit_log_end(audit_buf);
3146 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3149 #ifdef CONFIG_XFRM_MIGRATE
3150 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3151 const struct xfrm_selector *sel_tgt)
3153 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3154 if (sel_tgt->family == sel_cmp->family &&
3155 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3157 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3159 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3160 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3164 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3171 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3172 u8 dir, u8 type, struct net *net)
3174 struct xfrm_policy *pol, *ret = NULL;
3175 struct hlist_head *chain;
3178 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3179 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3180 hlist_for_each_entry(pol, chain, bydst) {
3181 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3182 pol->type == type) {
3184 priority = ret->priority;
3188 chain = &net->xfrm.policy_inexact[dir];
3189 hlist_for_each_entry(pol, chain, bydst) {
3190 if ((pol->priority >= priority) && ret)
3193 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3194 pol->type == type) {
3202 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3207 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3211 if (t->mode == m->mode && t->id.proto == m->proto &&
3212 (m->reqid == 0 || t->reqid == m->reqid)) {
3214 case XFRM_MODE_TUNNEL:
3215 case XFRM_MODE_BEET:
3216 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3218 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3223 case XFRM_MODE_TRANSPORT:
3224 /* in case of transport mode, template does not store
3225 any IP addresses, hence we just compare mode and
3236 /* update endpoint address(es) of template(s) */
3237 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3238 struct xfrm_migrate *m, int num_migrate)
3240 struct xfrm_migrate *mp;
3243 write_lock_bh(&pol->lock);
3244 if (unlikely(pol->walk.dead)) {
3245 /* target policy has been deleted */
3246 write_unlock_bh(&pol->lock);
3250 for (i = 0; i < pol->xfrm_nr; i++) {
3251 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3252 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3255 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3256 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3258 /* update endpoints */
3259 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3260 sizeof(pol->xfrm_vec[i].id.daddr));
3261 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3262 sizeof(pol->xfrm_vec[i].saddr));
3263 pol->xfrm_vec[i].encap_family = mp->new_family;
3265 atomic_inc(&pol->genid);
3269 write_unlock_bh(&pol->lock);
3277 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3281 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3284 for (i = 0; i < num_migrate; i++) {
3285 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3287 xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3290 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3291 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3294 /* check if there is any duplicated entry */
3295 for (j = i + 1; j < num_migrate; j++) {
3296 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3297 sizeof(m[i].old_daddr)) &&
3298 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3299 sizeof(m[i].old_saddr)) &&
3300 m[i].proto == m[j].proto &&
3301 m[i].mode == m[j].mode &&
3302 m[i].reqid == m[j].reqid &&
3303 m[i].old_family == m[j].old_family)
3311 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3312 struct xfrm_migrate *m, int num_migrate,
3313 struct xfrm_kmaddress *k, struct net *net)
3315 int i, err, nx_cur = 0, nx_new = 0;
3316 struct xfrm_policy *pol = NULL;
3317 struct xfrm_state *x, *xc;
3318 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3319 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3320 struct xfrm_migrate *mp;
3322 /* Stage 0 - sanity checks */
3323 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3326 if (dir >= XFRM_POLICY_MAX) {
3331 /* Stage 1 - find policy */
3332 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3337 /* Stage 2 - find and update state(s) */
3338 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3339 if ((x = xfrm_migrate_state_find(mp, net))) {
3342 if ((xc = xfrm_state_migrate(x, mp))) {
3352 /* Stage 3 - update policy */
3353 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3356 /* Stage 4 - delete old state(s) */
3358 xfrm_states_put(x_cur, nx_cur);
3359 xfrm_states_delete(x_cur, nx_cur);
3362 /* Stage 5 - announce */
3363 km_migrate(sel, dir, type, m, num_migrate, k);
3375 xfrm_states_put(x_cur, nx_cur);
3377 xfrm_states_delete(x_new, nx_new);
3381 EXPORT_SYMBOL(xfrm_migrate);