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 const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
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 refcount_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 const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
108 const struct xfrm_policy_afinfo *afinfo;
110 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
113 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
114 if (unlikely(!afinfo))
119 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
120 const xfrm_address_t *saddr,
121 const xfrm_address_t *daddr,
124 const struct xfrm_policy_afinfo *afinfo;
125 struct dst_entry *dst;
127 afinfo = xfrm_policy_get_afinfo(family);
128 if (unlikely(afinfo == NULL))
129 return ERR_PTR(-EAFNOSUPPORT);
131 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr);
137 EXPORT_SYMBOL(__xfrm_dst_lookup);
139 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
141 xfrm_address_t *prev_saddr,
142 xfrm_address_t *prev_daddr,
145 struct net *net = xs_net(x);
146 xfrm_address_t *saddr = &x->props.saddr;
147 xfrm_address_t *daddr = &x->id.daddr;
148 struct dst_entry *dst;
150 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
154 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
159 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family);
162 if (prev_saddr != saddr)
163 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
164 if (prev_daddr != daddr)
165 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
171 static inline unsigned long make_jiffies(long secs)
173 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
174 return MAX_SCHEDULE_TIMEOUT-1;
179 static void xfrm_policy_timer(unsigned long data)
181 struct xfrm_policy *xp = (struct xfrm_policy *)data;
182 unsigned long now = get_seconds();
183 long next = LONG_MAX;
187 read_lock(&xp->lock);
189 if (unlikely(xp->walk.dead))
192 dir = xfrm_policy_id2dir(xp->index);
194 if (xp->lft.hard_add_expires_seconds) {
195 long tmo = xp->lft.hard_add_expires_seconds +
196 xp->curlft.add_time - now;
202 if (xp->lft.hard_use_expires_seconds) {
203 long tmo = xp->lft.hard_use_expires_seconds +
204 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
210 if (xp->lft.soft_add_expires_seconds) {
211 long tmo = xp->lft.soft_add_expires_seconds +
212 xp->curlft.add_time - now;
215 tmo = XFRM_KM_TIMEOUT;
220 if (xp->lft.soft_use_expires_seconds) {
221 long tmo = xp->lft.soft_use_expires_seconds +
222 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
225 tmo = XFRM_KM_TIMEOUT;
232 km_policy_expired(xp, dir, 0, 0);
233 if (next != LONG_MAX &&
234 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
238 read_unlock(&xp->lock);
243 read_unlock(&xp->lock);
244 if (!xfrm_policy_delete(xp, dir))
245 km_policy_expired(xp, dir, 1, 0);
249 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
251 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
253 if (unlikely(pol->walk.dead))
261 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
263 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
265 return !pol->walk.dead;
268 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
270 xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
273 static const struct flow_cache_ops xfrm_policy_fc_ops = {
274 .get = xfrm_policy_flo_get,
275 .check = xfrm_policy_flo_check,
276 .delete = xfrm_policy_flo_delete,
279 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
283 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
285 struct xfrm_policy *policy;
287 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
290 write_pnet(&policy->xp_net, net);
291 INIT_LIST_HEAD(&policy->walk.all);
292 INIT_HLIST_NODE(&policy->bydst);
293 INIT_HLIST_NODE(&policy->byidx);
294 rwlock_init(&policy->lock);
295 refcount_set(&policy->refcnt, 1);
296 skb_queue_head_init(&policy->polq.hold_queue);
297 setup_timer(&policy->timer, xfrm_policy_timer,
298 (unsigned long)policy);
299 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
300 (unsigned long)policy);
301 policy->flo.ops = &xfrm_policy_fc_ops;
305 EXPORT_SYMBOL(xfrm_policy_alloc);
307 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
309 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
311 security_xfrm_policy_free(policy->security);
315 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
317 void xfrm_policy_destroy(struct xfrm_policy *policy)
319 BUG_ON(!policy->walk.dead);
321 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
324 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
326 EXPORT_SYMBOL(xfrm_policy_destroy);
328 /* Rule must be locked. Release descendant resources, announce
329 * entry dead. The rule must be unlinked from lists to the moment.
332 static void xfrm_policy_kill(struct xfrm_policy *policy)
334 policy->walk.dead = 1;
336 atomic_inc(&policy->genid);
338 if (del_timer(&policy->polq.hold_timer))
339 xfrm_pol_put(policy);
340 skb_queue_purge(&policy->polq.hold_queue);
342 if (del_timer(&policy->timer))
343 xfrm_pol_put(policy);
345 xfrm_pol_put(policy);
348 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
350 static inline unsigned int idx_hash(struct net *net, u32 index)
352 return __idx_hash(index, net->xfrm.policy_idx_hmask);
355 /* calculate policy hash thresholds */
356 static void __get_hash_thresh(struct net *net,
357 unsigned short family, int dir,
358 u8 *dbits, u8 *sbits)
362 *dbits = net->xfrm.policy_bydst[dir].dbits4;
363 *sbits = net->xfrm.policy_bydst[dir].sbits4;
367 *dbits = net->xfrm.policy_bydst[dir].dbits6;
368 *sbits = net->xfrm.policy_bydst[dir].sbits6;
377 static struct hlist_head *policy_hash_bysel(struct net *net,
378 const struct xfrm_selector *sel,
379 unsigned short family, int dir)
381 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
386 __get_hash_thresh(net, family, dir, &dbits, &sbits);
387 hash = __sel_hash(sel, family, hmask, dbits, sbits);
389 if (hash == hmask + 1)
390 return &net->xfrm.policy_inexact[dir];
392 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
393 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
396 static struct hlist_head *policy_hash_direct(struct net *net,
397 const xfrm_address_t *daddr,
398 const xfrm_address_t *saddr,
399 unsigned short family, int dir)
401 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
406 __get_hash_thresh(net, family, dir, &dbits, &sbits);
407 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
409 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
410 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
413 static void xfrm_dst_hash_transfer(struct net *net,
414 struct hlist_head *list,
415 struct hlist_head *ndsttable,
416 unsigned int nhashmask,
419 struct hlist_node *tmp, *entry0 = NULL;
420 struct xfrm_policy *pol;
426 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
429 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
430 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
431 pol->family, nhashmask, dbits, sbits);
433 hlist_del_rcu(&pol->bydst);
434 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
439 hlist_del_rcu(&pol->bydst);
440 hlist_add_behind_rcu(&pol->bydst, entry0);
442 entry0 = &pol->bydst;
444 if (!hlist_empty(list)) {
450 static void xfrm_idx_hash_transfer(struct hlist_head *list,
451 struct hlist_head *nidxtable,
452 unsigned int nhashmask)
454 struct hlist_node *tmp;
455 struct xfrm_policy *pol;
457 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
460 h = __idx_hash(pol->index, nhashmask);
461 hlist_add_head(&pol->byidx, nidxtable+h);
465 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
467 return ((old_hmask + 1) << 1) - 1;
470 static void xfrm_bydst_resize(struct net *net, int dir)
472 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
473 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
474 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
475 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
476 struct hlist_head *odst;
482 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
483 write_seqcount_begin(&xfrm_policy_hash_generation);
485 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
486 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
488 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
489 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
491 for (i = hmask; i >= 0; i--)
492 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
494 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
495 net->xfrm.policy_bydst[dir].hmask = nhashmask;
497 write_seqcount_end(&xfrm_policy_hash_generation);
498 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
502 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
505 static void xfrm_byidx_resize(struct net *net, int total)
507 unsigned int hmask = net->xfrm.policy_idx_hmask;
508 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
509 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
510 struct hlist_head *oidx = net->xfrm.policy_byidx;
511 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
517 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
519 for (i = hmask; i >= 0; i--)
520 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
522 net->xfrm.policy_byidx = nidx;
523 net->xfrm.policy_idx_hmask = nhashmask;
525 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
527 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
530 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
532 unsigned int cnt = net->xfrm.policy_count[dir];
533 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
538 if ((hmask + 1) < xfrm_policy_hashmax &&
545 static inline int xfrm_byidx_should_resize(struct net *net, int total)
547 unsigned int hmask = net->xfrm.policy_idx_hmask;
549 if ((hmask + 1) < xfrm_policy_hashmax &&
556 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
558 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
559 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
560 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
561 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
562 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
563 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
564 si->spdhcnt = net->xfrm.policy_idx_hmask;
565 si->spdhmcnt = xfrm_policy_hashmax;
567 EXPORT_SYMBOL(xfrm_spd_getinfo);
569 static DEFINE_MUTEX(hash_resize_mutex);
570 static void xfrm_hash_resize(struct work_struct *work)
572 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
575 mutex_lock(&hash_resize_mutex);
578 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
579 if (xfrm_bydst_should_resize(net, dir, &total))
580 xfrm_bydst_resize(net, dir);
582 if (xfrm_byidx_should_resize(net, total))
583 xfrm_byidx_resize(net, total);
585 mutex_unlock(&hash_resize_mutex);
588 static void xfrm_hash_rebuild(struct work_struct *work)
590 struct net *net = container_of(work, struct net,
591 xfrm.policy_hthresh.work);
593 struct xfrm_policy *pol;
594 struct xfrm_policy *policy;
595 struct hlist_head *chain;
596 struct hlist_head *odst;
597 struct hlist_node *newpos;
601 u8 lbits4, rbits4, lbits6, rbits6;
603 mutex_lock(&hash_resize_mutex);
605 /* read selector prefixlen thresholds */
607 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
609 lbits4 = net->xfrm.policy_hthresh.lbits4;
610 rbits4 = net->xfrm.policy_hthresh.rbits4;
611 lbits6 = net->xfrm.policy_hthresh.lbits6;
612 rbits6 = net->xfrm.policy_hthresh.rbits6;
613 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
615 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
617 /* reset the bydst and inexact table in all directions */
618 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
619 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
620 hmask = net->xfrm.policy_bydst[dir].hmask;
621 odst = net->xfrm.policy_bydst[dir].table;
622 for (i = hmask; i >= 0; i--)
623 INIT_HLIST_HEAD(odst + i);
624 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
625 /* dir out => dst = remote, src = local */
626 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
627 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
628 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
629 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
631 /* dir in/fwd => dst = local, src = remote */
632 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
633 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
634 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
635 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
639 /* re-insert all policies by order of creation */
640 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
641 if (xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
642 /* skip socket policies */
646 chain = policy_hash_bysel(net, &policy->selector,
648 xfrm_policy_id2dir(policy->index));
649 hlist_for_each_entry(pol, chain, bydst) {
650 if (policy->priority >= pol->priority)
651 newpos = &pol->bydst;
656 hlist_add_behind(&policy->bydst, newpos);
658 hlist_add_head(&policy->bydst, chain);
661 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
663 mutex_unlock(&hash_resize_mutex);
666 void xfrm_policy_hash_rebuild(struct net *net)
668 schedule_work(&net->xfrm.policy_hthresh.work);
670 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
672 /* Generate new index... KAME seems to generate them ordered by cost
673 * of an absolute inpredictability of ordering of rules. This will not pass. */
674 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
676 static u32 idx_generator;
679 struct hlist_head *list;
680 struct xfrm_policy *p;
685 idx = (idx_generator | dir);
694 list = net->xfrm.policy_byidx + idx_hash(net, idx);
696 hlist_for_each_entry(p, list, byidx) {
697 if (p->index == idx) {
707 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
709 u32 *p1 = (u32 *) s1;
710 u32 *p2 = (u32 *) s2;
711 int len = sizeof(struct xfrm_selector) / sizeof(u32);
714 for (i = 0; i < len; i++) {
722 static void xfrm_policy_requeue(struct xfrm_policy *old,
723 struct xfrm_policy *new)
725 struct xfrm_policy_queue *pq = &old->polq;
726 struct sk_buff_head list;
728 if (skb_queue_empty(&pq->hold_queue))
731 __skb_queue_head_init(&list);
733 spin_lock_bh(&pq->hold_queue.lock);
734 skb_queue_splice_init(&pq->hold_queue, &list);
735 if (del_timer(&pq->hold_timer))
737 spin_unlock_bh(&pq->hold_queue.lock);
741 spin_lock_bh(&pq->hold_queue.lock);
742 skb_queue_splice(&list, &pq->hold_queue);
743 pq->timeout = XFRM_QUEUE_TMO_MIN;
744 if (!mod_timer(&pq->hold_timer, jiffies))
746 spin_unlock_bh(&pq->hold_queue.lock);
749 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
750 struct xfrm_policy *pol)
752 u32 mark = policy->mark.v & policy->mark.m;
754 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
757 if ((mark & pol->mark.m) == pol->mark.v &&
758 policy->priority == pol->priority)
764 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
766 struct net *net = xp_net(policy);
767 struct xfrm_policy *pol;
768 struct xfrm_policy *delpol;
769 struct hlist_head *chain;
770 struct hlist_node *newpos;
772 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
773 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
776 hlist_for_each_entry(pol, chain, bydst) {
777 if (pol->type == policy->type &&
778 !selector_cmp(&pol->selector, &policy->selector) &&
779 xfrm_policy_mark_match(policy, pol) &&
780 xfrm_sec_ctx_match(pol->security, policy->security) &&
783 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
787 if (policy->priority > pol->priority)
789 } else if (policy->priority >= pol->priority) {
790 newpos = &pol->bydst;
797 hlist_add_behind(&policy->bydst, newpos);
799 hlist_add_head(&policy->bydst, chain);
800 __xfrm_policy_link(policy, dir);
801 atomic_inc(&net->xfrm.flow_cache_genid);
803 /* After previous checking, family can either be AF_INET or AF_INET6 */
804 if (policy->family == AF_INET)
805 rt_genid_bump_ipv4(net);
807 rt_genid_bump_ipv6(net);
810 xfrm_policy_requeue(delpol, policy);
811 __xfrm_policy_unlink(delpol, dir);
813 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
814 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
815 policy->curlft.add_time = get_seconds();
816 policy->curlft.use_time = 0;
817 if (!mod_timer(&policy->timer, jiffies + HZ))
818 xfrm_pol_hold(policy);
819 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
822 xfrm_policy_kill(delpol);
823 else if (xfrm_bydst_should_resize(net, dir, NULL))
824 schedule_work(&net->xfrm.policy_hash_work);
828 EXPORT_SYMBOL(xfrm_policy_insert);
830 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
831 int dir, struct xfrm_selector *sel,
832 struct xfrm_sec_ctx *ctx, int delete,
835 struct xfrm_policy *pol, *ret;
836 struct hlist_head *chain;
839 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
840 chain = policy_hash_bysel(net, sel, sel->family, dir);
842 hlist_for_each_entry(pol, chain, bydst) {
843 if (pol->type == type &&
844 (mark & pol->mark.m) == pol->mark.v &&
845 !selector_cmp(sel, &pol->selector) &&
846 xfrm_sec_ctx_match(ctx, pol->security)) {
849 *err = security_xfrm_policy_delete(
852 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
855 __xfrm_policy_unlink(pol, dir);
861 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
864 xfrm_policy_kill(ret);
867 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
869 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
870 int dir, u32 id, int delete, int *err)
872 struct xfrm_policy *pol, *ret;
873 struct hlist_head *chain;
876 if (xfrm_policy_id2dir(id) != dir)
880 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
881 chain = net->xfrm.policy_byidx + idx_hash(net, id);
883 hlist_for_each_entry(pol, chain, byidx) {
884 if (pol->type == type && pol->index == id &&
885 (mark & pol->mark.m) == pol->mark.v) {
888 *err = security_xfrm_policy_delete(
891 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
894 __xfrm_policy_unlink(pol, dir);
900 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
903 xfrm_policy_kill(ret);
906 EXPORT_SYMBOL(xfrm_policy_byid);
908 #ifdef CONFIG_SECURITY_NETWORK_XFRM
910 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
914 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
915 struct xfrm_policy *pol;
918 hlist_for_each_entry(pol,
919 &net->xfrm.policy_inexact[dir], bydst) {
920 if (pol->type != type)
922 err = security_xfrm_policy_delete(pol->security);
924 xfrm_audit_policy_delete(pol, 0, task_valid);
928 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
929 hlist_for_each_entry(pol,
930 net->xfrm.policy_bydst[dir].table + i,
932 if (pol->type != type)
934 err = security_xfrm_policy_delete(
937 xfrm_audit_policy_delete(pol, 0,
948 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
954 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
956 int dir, err = 0, cnt = 0;
958 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
960 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
964 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
965 struct xfrm_policy *pol;
969 hlist_for_each_entry(pol,
970 &net->xfrm.policy_inexact[dir], bydst) {
971 if (pol->type != type)
973 __xfrm_policy_unlink(pol, dir);
974 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
977 xfrm_audit_policy_delete(pol, 1, task_valid);
979 xfrm_policy_kill(pol);
981 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
985 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
987 hlist_for_each_entry(pol,
988 net->xfrm.policy_bydst[dir].table + i,
990 if (pol->type != type)
992 __xfrm_policy_unlink(pol, dir);
993 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
996 xfrm_audit_policy_delete(pol, 1, task_valid);
997 xfrm_policy_kill(pol);
999 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1008 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1011 EXPORT_SYMBOL(xfrm_policy_flush);
1013 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1014 int (*func)(struct xfrm_policy *, int, int, void*),
1017 struct xfrm_policy *pol;
1018 struct xfrm_policy_walk_entry *x;
1021 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1022 walk->type != XFRM_POLICY_TYPE_ANY)
1025 if (list_empty(&walk->walk.all) && walk->seq != 0)
1028 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1029 if (list_empty(&walk->walk.all))
1030 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1032 x = list_first_entry(&walk->walk.all,
1033 struct xfrm_policy_walk_entry, all);
1035 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1038 pol = container_of(x, struct xfrm_policy, walk);
1039 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1040 walk->type != pol->type)
1042 error = func(pol, xfrm_policy_id2dir(pol->index),
1045 list_move_tail(&walk->walk.all, &x->all);
1050 if (walk->seq == 0) {
1054 list_del_init(&walk->walk.all);
1056 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1059 EXPORT_SYMBOL(xfrm_policy_walk);
1061 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1063 INIT_LIST_HEAD(&walk->walk.all);
1064 walk->walk.dead = 1;
1068 EXPORT_SYMBOL(xfrm_policy_walk_init);
1070 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1072 if (list_empty(&walk->walk.all))
1075 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1076 list_del(&walk->walk.all);
1077 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1079 EXPORT_SYMBOL(xfrm_policy_walk_done);
1082 * Find policy to apply to this flow.
1084 * Returns 0 if policy found, else an -errno.
1086 static int xfrm_policy_match(const struct xfrm_policy *pol,
1087 const struct flowi *fl,
1088 u8 type, u16 family, int dir)
1090 const struct xfrm_selector *sel = &pol->selector;
1094 if (pol->family != family ||
1095 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1099 match = xfrm_selector_match(sel, fl, family);
1101 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1107 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1108 const struct flowi *fl,
1112 struct xfrm_policy *pol, *ret;
1113 const xfrm_address_t *daddr, *saddr;
1114 struct hlist_head *chain;
1115 unsigned int sequence;
1118 daddr = xfrm_flowi_daddr(fl, family);
1119 saddr = xfrm_flowi_saddr(fl, family);
1120 if (unlikely(!daddr || !saddr))
1126 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1127 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1128 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1132 hlist_for_each_entry_rcu(pol, chain, bydst) {
1133 err = xfrm_policy_match(pol, fl, type, family, dir);
1143 priority = ret->priority;
1147 chain = &net->xfrm.policy_inexact[dir];
1148 hlist_for_each_entry_rcu(pol, chain, bydst) {
1149 if ((pol->priority >= priority) && ret)
1152 err = xfrm_policy_match(pol, fl, type, family, dir);
1166 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1169 if (ret && !xfrm_pol_hold_rcu(ret))
1177 static struct xfrm_policy *
1178 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1180 #ifdef CONFIG_XFRM_SUB_POLICY
1181 struct xfrm_policy *pol;
1183 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1187 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1190 static int flow_to_policy_dir(int dir)
1192 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1193 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1194 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1200 return XFRM_POLICY_IN;
1202 return XFRM_POLICY_OUT;
1204 return XFRM_POLICY_FWD;
1208 static struct flow_cache_object *
1209 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1210 u8 dir, struct flow_cache_object *old_obj, void *ctx)
1212 struct xfrm_policy *pol;
1215 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1217 pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1218 if (IS_ERR_OR_NULL(pol))
1219 return ERR_CAST(pol);
1221 /* Resolver returns two references:
1222 * one for cache and one for caller of flow_cache_lookup() */
1228 static inline int policy_to_flow_dir(int dir)
1230 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1231 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1232 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1236 case XFRM_POLICY_IN:
1238 case XFRM_POLICY_OUT:
1239 return FLOW_DIR_OUT;
1240 case XFRM_POLICY_FWD:
1241 return FLOW_DIR_FWD;
1245 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1246 const struct flowi *fl, u16 family)
1248 struct xfrm_policy *pol;
1252 pol = rcu_dereference(sk->sk_policy[dir]);
1254 bool match = xfrm_selector_match(&pol->selector, fl, family);
1258 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1262 err = security_xfrm_policy_lookup(pol->security,
1264 policy_to_flow_dir(dir));
1266 if (!xfrm_pol_hold_rcu(pol))
1268 } else if (err == -ESRCH) {
1281 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1283 struct net *net = xp_net(pol);
1285 list_add(&pol->walk.all, &net->xfrm.policy_all);
1286 net->xfrm.policy_count[dir]++;
1290 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1293 struct net *net = xp_net(pol);
1295 if (list_empty(&pol->walk.all))
1298 /* Socket policies are not hashed. */
1299 if (!hlist_unhashed(&pol->bydst)) {
1300 hlist_del_rcu(&pol->bydst);
1301 hlist_del(&pol->byidx);
1304 list_del_init(&pol->walk.all);
1305 net->xfrm.policy_count[dir]--;
1310 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1312 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1315 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1317 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1320 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1322 struct net *net = xp_net(pol);
1324 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1325 pol = __xfrm_policy_unlink(pol, dir);
1326 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1328 xfrm_policy_kill(pol);
1333 EXPORT_SYMBOL(xfrm_policy_delete);
1335 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1337 struct net *net = xp_net(pol);
1338 struct xfrm_policy *old_pol;
1340 #ifdef CONFIG_XFRM_SUB_POLICY
1341 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1345 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1346 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1347 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1349 pol->curlft.add_time = get_seconds();
1350 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1351 xfrm_sk_policy_link(pol, dir);
1353 rcu_assign_pointer(sk->sk_policy[dir], pol);
1356 xfrm_policy_requeue(old_pol, pol);
1358 /* Unlinking succeeds always. This is the only function
1359 * allowed to delete or replace socket policy.
1361 xfrm_sk_policy_unlink(old_pol, dir);
1363 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1366 xfrm_policy_kill(old_pol);
1371 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1373 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1374 struct net *net = xp_net(old);
1377 newp->selector = old->selector;
1378 if (security_xfrm_policy_clone(old->security,
1381 return NULL; /* ENOMEM */
1383 newp->lft = old->lft;
1384 newp->curlft = old->curlft;
1385 newp->mark = old->mark;
1386 newp->action = old->action;
1387 newp->flags = old->flags;
1388 newp->xfrm_nr = old->xfrm_nr;
1389 newp->index = old->index;
1390 newp->type = old->type;
1391 memcpy(newp->xfrm_vec, old->xfrm_vec,
1392 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1393 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1394 xfrm_sk_policy_link(newp, dir);
1395 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1401 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1403 const struct xfrm_policy *p;
1404 struct xfrm_policy *np;
1408 for (i = 0; i < 2; i++) {
1409 p = rcu_dereference(osk->sk_policy[i]);
1411 np = clone_policy(p, i);
1412 if (unlikely(!np)) {
1416 rcu_assign_pointer(sk->sk_policy[i], np);
1424 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1425 xfrm_address_t *remote, unsigned short family)
1428 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1430 if (unlikely(afinfo == NULL))
1432 err = afinfo->get_saddr(net, oif, local, remote);
1437 /* Resolve list of templates for the flow, given policy. */
1440 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1441 struct xfrm_state **xfrm, unsigned short family)
1443 struct net *net = xp_net(policy);
1446 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1447 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1450 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1451 struct xfrm_state *x;
1452 xfrm_address_t *remote = daddr;
1453 xfrm_address_t *local = saddr;
1454 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1456 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1457 tmpl->mode == XFRM_MODE_BEET) {
1458 remote = &tmpl->id.daddr;
1459 local = &tmpl->saddr;
1460 if (xfrm_addr_any(local, tmpl->encap_family)) {
1461 error = xfrm_get_saddr(net, fl->flowi_oif,
1463 tmpl->encap_family);
1470 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1472 if (x && x->km.state == XFRM_STATE_VALID) {
1479 error = (x->km.state == XFRM_STATE_ERROR ?
1482 } else if (error == -ESRCH) {
1486 if (!tmpl->optional)
1492 for (nx--; nx >= 0; nx--)
1493 xfrm_state_put(xfrm[nx]);
1498 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1499 struct xfrm_state **xfrm, unsigned short family)
1501 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1502 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1508 for (i = 0; i < npols; i++) {
1509 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1514 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1522 /* found states are sorted for outbound processing */
1524 xfrm_state_sort(xfrm, tpp, cnx, family);
1529 for (cnx--; cnx >= 0; cnx--)
1530 xfrm_state_put(tpp[cnx]);
1535 static int xfrm_get_tos(const struct flowi *fl, int family)
1537 const struct xfrm_policy_afinfo *afinfo;
1540 afinfo = xfrm_policy_get_afinfo(family);
1541 tos = afinfo ? afinfo->get_tos(fl) : 0;
1548 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1550 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1551 struct dst_entry *dst = &xdst->u.dst;
1553 if (xdst->route == NULL) {
1554 /* Dummy bundle - if it has xfrms we were not
1555 * able to build bundle as template resolution failed.
1556 * It means we need to try again resolving. */
1557 if (xdst->num_xfrms > 0)
1559 } else if (dst->flags & DST_XFRM_QUEUE) {
1563 if (stale_bundle(dst))
1571 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1573 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1574 struct dst_entry *dst = &xdst->u.dst;
1578 if (stale_bundle(dst))
1584 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1586 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1587 struct dst_entry *dst = &xdst->u.dst;
1589 /* Mark DST_OBSOLETE_DEAD to fail the next xfrm_dst_check() */
1590 dst->obsolete = DST_OBSOLETE_DEAD;
1591 dst_release_immediate(dst);
1594 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1595 .get = xfrm_bundle_flo_get,
1596 .check = xfrm_bundle_flo_check,
1597 .delete = xfrm_bundle_flo_delete,
1600 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1602 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1603 struct dst_ops *dst_ops;
1604 struct xfrm_dst *xdst;
1607 return ERR_PTR(-EINVAL);
1611 dst_ops = &net->xfrm.xfrm4_dst_ops;
1613 #if IS_ENABLED(CONFIG_IPV6)
1615 dst_ops = &net->xfrm.xfrm6_dst_ops;
1621 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1624 struct dst_entry *dst = &xdst->u.dst;
1626 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1627 xdst->flo.ops = &xfrm_bundle_fc_ops;
1629 xdst = ERR_PTR(-ENOBUFS);
1636 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1639 const struct xfrm_policy_afinfo *afinfo =
1640 xfrm_policy_get_afinfo(dst->ops->family);
1646 err = afinfo->init_path(path, dst, nfheader_len);
1653 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1654 const struct flowi *fl)
1656 const struct xfrm_policy_afinfo *afinfo =
1657 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1663 err = afinfo->fill_dst(xdst, dev, fl);
1671 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1672 * all the metrics... Shortly, bundle a bundle.
1675 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1676 struct xfrm_state **xfrm, int nx,
1677 const struct flowi *fl,
1678 struct dst_entry *dst)
1680 struct net *net = xp_net(policy);
1681 unsigned long now = jiffies;
1682 struct net_device *dev;
1683 struct xfrm_mode *inner_mode;
1684 struct dst_entry *dst_prev = NULL;
1685 struct dst_entry *dst0 = NULL;
1689 int nfheader_len = 0;
1690 int trailer_len = 0;
1692 int family = policy->selector.family;
1693 xfrm_address_t saddr, daddr;
1695 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1697 tos = xfrm_get_tos(fl, family);
1701 for (; i < nx; i++) {
1702 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1703 struct dst_entry *dst1 = &xdst->u.dst;
1705 err = PTR_ERR(xdst);
1711 if (xfrm[i]->sel.family == AF_UNSPEC) {
1712 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1713 xfrm_af2proto(family));
1715 err = -EAFNOSUPPORT;
1720 inner_mode = xfrm[i]->inner_mode;
1725 /* Ref count is taken during xfrm_alloc_dst()
1726 * No need to do dst_clone() on dst1
1728 dst_prev->child = dst1;
1731 dst_copy_metrics(dst1, dst);
1733 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1734 family = xfrm[i]->props.family;
1735 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1736 &saddr, &daddr, family);
1743 dst1->xfrm = xfrm[i];
1744 xdst->xfrm_genid = xfrm[i]->genid;
1746 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1747 dst1->flags |= DST_HOST;
1748 dst1->lastuse = now;
1750 dst1->input = dst_discard;
1751 dst1->output = inner_mode->afinfo->output;
1753 dst1->next = dst_prev;
1756 header_len += xfrm[i]->props.header_len;
1757 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1758 nfheader_len += xfrm[i]->props.header_len;
1759 trailer_len += xfrm[i]->props.trailer_len;
1762 dst_prev->child = dst;
1770 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1771 xfrm_init_pmtu(dst_prev);
1773 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1774 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1776 err = xfrm_fill_dst(xdst, dev, fl);
1780 dst_prev->header_len = header_len;
1781 dst_prev->trailer_len = trailer_len;
1782 header_len -= xdst->u.dst.xfrm->props.header_len;
1783 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1791 xfrm_state_put(xfrm[i]);
1794 dst_release_immediate(dst0);
1795 dst0 = ERR_PTR(err);
1799 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1800 struct xfrm_policy **pols,
1801 int *num_pols, int *num_xfrms)
1805 if (*num_pols == 0 || !pols[0]) {
1810 if (IS_ERR(pols[0]))
1811 return PTR_ERR(pols[0]);
1813 *num_xfrms = pols[0]->xfrm_nr;
1815 #ifdef CONFIG_XFRM_SUB_POLICY
1816 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1817 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1818 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1819 XFRM_POLICY_TYPE_MAIN,
1823 if (IS_ERR(pols[1])) {
1824 xfrm_pols_put(pols, *num_pols);
1825 return PTR_ERR(pols[1]);
1828 (*num_xfrms) += pols[1]->xfrm_nr;
1832 for (i = 0; i < *num_pols; i++) {
1833 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1843 static struct xfrm_dst *
1844 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1845 const struct flowi *fl, u16 family,
1846 struct dst_entry *dst_orig)
1848 struct net *net = xp_net(pols[0]);
1849 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1850 struct dst_entry *dst;
1851 struct xfrm_dst *xdst;
1854 /* Try to instantiate a bundle */
1855 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1857 if (err != 0 && err != -EAGAIN)
1858 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1859 return ERR_PTR(err);
1862 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1864 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1865 return ERR_CAST(dst);
1868 xdst = (struct xfrm_dst *)dst;
1869 xdst->num_xfrms = err;
1870 xdst->num_pols = num_pols;
1871 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1872 xdst->policy_genid = atomic_read(&pols[0]->genid);
1877 static void xfrm_policy_queue_process(unsigned long arg)
1879 struct sk_buff *skb;
1881 struct dst_entry *dst;
1882 struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1883 struct net *net = xp_net(pol);
1884 struct xfrm_policy_queue *pq = &pol->polq;
1886 struct sk_buff_head list;
1888 spin_lock(&pq->hold_queue.lock);
1889 skb = skb_peek(&pq->hold_queue);
1891 spin_unlock(&pq->hold_queue.lock);
1896 xfrm_decode_session(skb, &fl, dst->ops->family);
1897 spin_unlock(&pq->hold_queue.lock);
1899 dst_hold(dst->path);
1900 dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1904 if (dst->flags & DST_XFRM_QUEUE) {
1907 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1910 pq->timeout = pq->timeout << 1;
1911 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1918 __skb_queue_head_init(&list);
1920 spin_lock(&pq->hold_queue.lock);
1922 skb_queue_splice_init(&pq->hold_queue, &list);
1923 spin_unlock(&pq->hold_queue.lock);
1925 while (!skb_queue_empty(&list)) {
1926 skb = __skb_dequeue(&list);
1928 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1929 dst_hold(skb_dst(skb)->path);
1930 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1938 skb_dst_set(skb, dst);
1940 dst_output(net, skb->sk, skb);
1949 skb_queue_purge(&pq->hold_queue);
1953 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1955 unsigned long sched_next;
1956 struct dst_entry *dst = skb_dst(skb);
1957 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1958 struct xfrm_policy *pol = xdst->pols[0];
1959 struct xfrm_policy_queue *pq = &pol->polq;
1961 if (unlikely(skb_fclone_busy(sk, skb))) {
1966 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1973 spin_lock_bh(&pq->hold_queue.lock);
1976 pq->timeout = XFRM_QUEUE_TMO_MIN;
1978 sched_next = jiffies + pq->timeout;
1980 if (del_timer(&pq->hold_timer)) {
1981 if (time_before(pq->hold_timer.expires, sched_next))
1982 sched_next = pq->hold_timer.expires;
1986 __skb_queue_tail(&pq->hold_queue, skb);
1987 if (!mod_timer(&pq->hold_timer, sched_next))
1990 spin_unlock_bh(&pq->hold_queue.lock);
1995 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1996 struct xfrm_flo *xflo,
1997 const struct flowi *fl,
2002 struct net_device *dev;
2003 struct dst_entry *dst;
2004 struct dst_entry *dst1;
2005 struct xfrm_dst *xdst;
2007 xdst = xfrm_alloc_dst(net, family);
2011 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2012 net->xfrm.sysctl_larval_drop ||
2016 dst = xflo->dst_orig;
2017 dst1 = &xdst->u.dst;
2021 dst_copy_metrics(dst1, dst);
2023 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2024 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2025 dst1->lastuse = jiffies;
2027 dst1->input = dst_discard;
2028 dst1->output = xdst_queue_output;
2034 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2041 err = xfrm_fill_dst(xdst, dev, fl);
2050 xdst = ERR_PTR(err);
2054 static struct flow_cache_object *
2055 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2056 struct flow_cache_object *oldflo, void *ctx)
2058 struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2059 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2060 struct xfrm_dst *xdst, *new_xdst;
2061 int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2063 /* Check if the policies from old bundle are usable */
2066 xdst = container_of(oldflo, struct xfrm_dst, flo);
2067 num_pols = xdst->num_pols;
2068 num_xfrms = xdst->num_xfrms;
2070 for (i = 0; i < num_pols; i++) {
2071 pols[i] = xdst->pols[i];
2072 pol_dead |= pols[i]->walk.dead;
2075 /* Mark DST_OBSOLETE_DEAD to fail the next
2078 xdst->u.dst.obsolete = DST_OBSOLETE_DEAD;
2079 dst_release_immediate(&xdst->u.dst);
2087 /* Resolve policies to use if we couldn't get them from
2088 * previous cache entry */
2091 pols[0] = __xfrm_policy_lookup(net, fl, family,
2092 flow_to_policy_dir(dir));
2093 err = xfrm_expand_policies(fl, family, pols,
2094 &num_pols, &num_xfrms);
2100 goto make_dummy_bundle;
2103 new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2105 if (IS_ERR(new_xdst)) {
2106 err = PTR_ERR(new_xdst);
2110 goto make_dummy_bundle;
2111 dst_hold(&xdst->u.dst);
2113 } else if (new_xdst == NULL) {
2116 goto make_dummy_bundle;
2117 xdst->num_xfrms = 0;
2118 dst_hold(&xdst->u.dst);
2122 /* Kill the previous bundle */
2124 /* The policies were stolen for newly generated bundle */
2126 /* Mark DST_OBSOLETE_DEAD to fail the next xfrm_dst_check() */
2127 xdst->u.dst.obsolete = DST_OBSOLETE_DEAD;
2128 dst_release_immediate(&xdst->u.dst);
2131 /* We do need to return one reference for original caller */
2132 dst_hold(&new_xdst->u.dst);
2133 return &new_xdst->flo;
2136 /* We found policies, but there's no bundles to instantiate:
2137 * either because the policy blocks, has no transformations or
2138 * we could not build template (no xfrm_states).*/
2139 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2141 xfrm_pols_put(pols, num_pols);
2142 return ERR_CAST(xdst);
2144 xdst->num_pols = num_pols;
2145 xdst->num_xfrms = num_xfrms;
2146 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2148 dst_hold(&xdst->u.dst);
2152 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2155 /* Mark DST_OBSOLETE_DEAD to fail the next xfrm_dst_check() */
2156 xdst->u.dst.obsolete = DST_OBSOLETE_DEAD;
2157 dst_release_immediate(&xdst->u.dst);
2159 xfrm_pols_put(pols, num_pols);
2160 return ERR_PTR(err);
2163 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2164 struct dst_entry *dst_orig)
2166 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2167 struct dst_entry *ret;
2170 dst_release(dst_orig);
2171 return ERR_PTR(-EINVAL);
2173 ret = afinfo->blackhole_route(net, dst_orig);
2180 /* Main function: finds/creates a bundle for given flow.
2182 * At the moment we eat a raw IP route. Mostly to speed up lookups
2183 * on interfaces with disabled IPsec.
2185 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2186 const struct flowi *fl,
2187 const struct sock *sk, int flags)
2189 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2190 struct flow_cache_object *flo;
2191 struct xfrm_dst *xdst;
2192 struct dst_entry *dst, *route;
2193 u16 family = dst_orig->ops->family;
2194 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2195 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2201 sk = sk_const_to_full_sk(sk);
2202 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2204 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2205 err = xfrm_expand_policies(fl, family, pols,
2206 &num_pols, &num_xfrms);
2211 if (num_xfrms <= 0) {
2212 drop_pols = num_pols;
2216 xdst = xfrm_resolve_and_create_bundle(
2220 xfrm_pols_put(pols, num_pols);
2221 err = PTR_ERR(xdst);
2223 } else if (xdst == NULL) {
2225 drop_pols = num_pols;
2229 dst_hold(&xdst->u.dst);
2230 route = xdst->route;
2235 struct xfrm_flo xflo;
2237 xflo.dst_orig = dst_orig;
2240 /* To accelerate a bit... */
2241 if ((dst_orig->flags & DST_NOXFRM) ||
2242 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2245 flo = flow_cache_lookup(net, fl, family, dir,
2246 xfrm_bundle_lookup, &xflo);
2253 xdst = container_of(flo, struct xfrm_dst, flo);
2255 num_pols = xdst->num_pols;
2256 num_xfrms = xdst->num_xfrms;
2257 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2258 route = xdst->route;
2262 if (route == NULL && num_xfrms > 0) {
2263 /* The only case when xfrm_bundle_lookup() returns a
2264 * bundle with null route, is when the template could
2265 * not be resolved. It means policies are there, but
2266 * bundle could not be created, since we don't yet
2267 * have the xfrm_state's. We need to wait for KM to
2268 * negotiate new SA's or bail out with error.*/
2269 if (net->xfrm.sysctl_larval_drop) {
2270 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2277 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2285 if ((flags & XFRM_LOOKUP_ICMP) &&
2286 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2291 for (i = 0; i < num_pols; i++)
2292 pols[i]->curlft.use_time = get_seconds();
2294 if (num_xfrms < 0) {
2295 /* Prohibit the flow */
2296 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2299 } else if (num_xfrms > 0) {
2300 /* Flow transformed */
2301 dst_release(dst_orig);
2303 /* Flow passes untransformed */
2308 xfrm_pols_put(pols, drop_pols);
2309 if (dst && dst->xfrm &&
2310 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2311 dst->flags |= DST_XFRM_TUNNEL;
2315 if (!(flags & XFRM_LOOKUP_ICMP)) {
2323 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2324 dst_release(dst_orig);
2325 xfrm_pols_put(pols, drop_pols);
2326 return ERR_PTR(err);
2328 EXPORT_SYMBOL(xfrm_lookup);
2330 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2331 * Otherwise we may send out blackholed packets.
2333 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2334 const struct flowi *fl,
2335 const struct sock *sk, int flags)
2337 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2338 flags | XFRM_LOOKUP_QUEUE |
2339 XFRM_LOOKUP_KEEP_DST_REF);
2341 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2342 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2346 EXPORT_SYMBOL(xfrm_lookup_route);
2349 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2351 struct xfrm_state *x;
2353 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2355 x = skb->sp->xvec[idx];
2356 if (!x->type->reject)
2358 return x->type->reject(x, skb, fl);
2361 /* When skb is transformed back to its "native" form, we have to
2362 * check policy restrictions. At the moment we make this in maximally
2363 * stupid way. Shame on me. :-) Of course, connected sockets must
2364 * have policy cached at them.
2368 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2369 unsigned short family)
2371 if (xfrm_state_kern(x))
2372 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2373 return x->id.proto == tmpl->id.proto &&
2374 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2375 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2376 x->props.mode == tmpl->mode &&
2377 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2378 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2379 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2380 xfrm_state_addr_cmp(tmpl, x, family));
2384 * 0 or more than 0 is returned when validation is succeeded (either bypass
2385 * because of optional transport mode, or next index of the mathced secpath
2386 * state with the template.
2387 * -1 is returned when no matching template is found.
2388 * Otherwise "-2 - errored_index" is returned.
2391 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2392 unsigned short family)
2396 if (tmpl->optional) {
2397 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2401 for (; idx < sp->len; idx++) {
2402 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2404 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2413 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2414 unsigned int family, int reverse)
2416 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2419 if (unlikely(afinfo == NULL))
2420 return -EAFNOSUPPORT;
2422 afinfo->decode_session(skb, fl, reverse);
2423 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2427 EXPORT_SYMBOL(__xfrm_decode_session);
2429 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2431 for (; k < sp->len; k++) {
2432 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2441 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2442 unsigned short family)
2444 struct net *net = dev_net(skb->dev);
2445 struct xfrm_policy *pol;
2446 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2455 reverse = dir & ~XFRM_POLICY_MASK;
2456 dir &= XFRM_POLICY_MASK;
2457 fl_dir = policy_to_flow_dir(dir);
2459 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2460 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2464 nf_nat_decode_session(skb, &fl, family);
2466 /* First, check used SA against their selectors. */
2470 for (i = skb->sp->len-1; i >= 0; i--) {
2471 struct xfrm_state *x = skb->sp->xvec[i];
2472 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2473 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2480 sk = sk_to_full_sk(sk);
2481 if (sk && sk->sk_policy[dir]) {
2482 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2484 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2490 struct flow_cache_object *flo;
2492 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2493 xfrm_policy_lookup, NULL);
2494 if (IS_ERR_OR_NULL(flo))
2495 pol = ERR_CAST(flo);
2497 pol = container_of(flo, struct xfrm_policy, flo);
2501 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2506 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2507 xfrm_secpath_reject(xerr_idx, skb, &fl);
2508 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2514 pol->curlft.use_time = get_seconds();
2518 #ifdef CONFIG_XFRM_SUB_POLICY
2519 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2520 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2524 if (IS_ERR(pols[1])) {
2525 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2528 pols[1]->curlft.use_time = get_seconds();
2534 if (pol->action == XFRM_POLICY_ALLOW) {
2535 struct sec_path *sp;
2536 static struct sec_path dummy;
2537 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2538 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2539 struct xfrm_tmpl **tpp = tp;
2543 if ((sp = skb->sp) == NULL)
2546 for (pi = 0; pi < npols; pi++) {
2547 if (pols[pi] != pol &&
2548 pols[pi]->action != XFRM_POLICY_ALLOW) {
2549 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2552 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2553 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2556 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2557 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2561 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2565 /* For each tunnel xfrm, find the first matching tmpl.
2566 * For each tmpl before that, find corresponding xfrm.
2567 * Order is _important_. Later we will implement
2568 * some barriers, but at the moment barriers
2569 * are implied between each two transformations.
2571 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2572 k = xfrm_policy_ok(tpp[i], sp, k, family);
2575 /* "-2 - errored_index" returned */
2577 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2582 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2583 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2587 xfrm_pols_put(pols, npols);
2590 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2593 xfrm_secpath_reject(xerr_idx, skb, &fl);
2595 xfrm_pols_put(pols, npols);
2598 EXPORT_SYMBOL(__xfrm_policy_check);
2600 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2602 struct net *net = dev_net(skb->dev);
2604 struct dst_entry *dst;
2607 if (xfrm_decode_session(skb, &fl, family) < 0) {
2608 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2614 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2619 skb_dst_set(skb, dst);
2622 EXPORT_SYMBOL(__xfrm_route_forward);
2624 /* Optimize later using cookies and generation ids. */
2626 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2628 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2629 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2630 * get validated by dst_ops->check on every use. We do this
2631 * because when a normal route referenced by an XFRM dst is
2632 * obsoleted we do not go looking around for all parent
2633 * referencing XFRM dsts so that we can invalidate them. It
2634 * is just too much work. Instead we make the checks here on
2635 * every use. For example:
2637 * XFRM dst A --> IPv4 dst X
2639 * X is the "xdst->route" of A (X is also the "dst->path" of A
2640 * in this example). If X is marked obsolete, "A" will not
2641 * notice. That's what we are validating here via the
2642 * stale_bundle() check.
2644 * When an xdst is removed from flow cache, DST_OBSOLETE_DEAD will
2646 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2648 * Both will force stable_bundle() to fail on any xdst bundle with
2649 * this dst linked in it.
2651 if (dst->obsolete < 0 && !stale_bundle(dst))
2657 static int stale_bundle(struct dst_entry *dst)
2659 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2662 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2664 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2665 dst->dev = dev_net(dev)->loopback_dev;
2670 EXPORT_SYMBOL(xfrm_dst_ifdown);
2672 static void xfrm_link_failure(struct sk_buff *skb)
2674 /* Impossible. Such dst must be popped before reaches point of failure. */
2677 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2680 if (dst->obsolete) {
2688 void xfrm_garbage_collect(struct net *net)
2690 flow_cache_flush(net);
2692 EXPORT_SYMBOL(xfrm_garbage_collect);
2694 void xfrm_garbage_collect_deferred(struct net *net)
2696 flow_cache_flush_deferred(net);
2698 EXPORT_SYMBOL(xfrm_garbage_collect_deferred);
2700 static void xfrm_init_pmtu(struct dst_entry *dst)
2703 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2704 u32 pmtu, route_mtu_cached;
2706 pmtu = dst_mtu(dst->child);
2707 xdst->child_mtu_cached = pmtu;
2709 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2711 route_mtu_cached = dst_mtu(xdst->route);
2712 xdst->route_mtu_cached = route_mtu_cached;
2714 if (pmtu > route_mtu_cached)
2715 pmtu = route_mtu_cached;
2717 dst_metric_set(dst, RTAX_MTU, pmtu);
2718 } while ((dst = dst->next));
2721 /* Check that the bundle accepts the flow and its components are
2725 static int xfrm_bundle_ok(struct xfrm_dst *first)
2727 struct dst_entry *dst = &first->u.dst;
2728 struct xfrm_dst *last;
2731 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2732 (dst->dev && !netif_running(dst->dev)))
2735 if (dst->flags & DST_XFRM_QUEUE)
2741 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2743 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2745 if (xdst->xfrm_genid != dst->xfrm->genid)
2747 if (xdst->num_pols > 0 &&
2748 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2751 mtu = dst_mtu(dst->child);
2752 if (xdst->child_mtu_cached != mtu) {
2754 xdst->child_mtu_cached = mtu;
2757 if (!dst_check(xdst->route, xdst->route_cookie))
2759 mtu = dst_mtu(xdst->route);
2760 if (xdst->route_mtu_cached != mtu) {
2762 xdst->route_mtu_cached = mtu;
2766 } while (dst->xfrm);
2771 mtu = last->child_mtu_cached;
2775 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2776 if (mtu > last->route_mtu_cached)
2777 mtu = last->route_mtu_cached;
2778 dst_metric_set(dst, RTAX_MTU, mtu);
2783 last = (struct xfrm_dst *)last->u.dst.next;
2784 last->child_mtu_cached = mtu;
2790 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2792 return dst_metric_advmss(dst->path);
2795 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2797 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2799 return mtu ? : dst_mtu(dst->path);
2802 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2805 const struct dst_entry *path = dst->path;
2807 for (; dst != path; dst = dst->child) {
2808 const struct xfrm_state *xfrm = dst->xfrm;
2810 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2812 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2813 daddr = xfrm->coaddr;
2814 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2815 daddr = &xfrm->id.daddr;
2820 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2821 struct sk_buff *skb,
2824 const struct dst_entry *path = dst->path;
2827 daddr = xfrm_get_dst_nexthop(dst, daddr);
2828 return path->ops->neigh_lookup(path, skb, daddr);
2831 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2833 const struct dst_entry *path = dst->path;
2835 daddr = xfrm_get_dst_nexthop(dst, daddr);
2836 path->ops->confirm_neigh(path, daddr);
2839 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2843 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2844 return -EAFNOSUPPORT;
2846 spin_lock(&xfrm_policy_afinfo_lock);
2847 if (unlikely(xfrm_policy_afinfo[family] != NULL))
2850 struct dst_ops *dst_ops = afinfo->dst_ops;
2851 if (likely(dst_ops->kmem_cachep == NULL))
2852 dst_ops->kmem_cachep = xfrm_dst_cache;
2853 if (likely(dst_ops->check == NULL))
2854 dst_ops->check = xfrm_dst_check;
2855 if (likely(dst_ops->default_advmss == NULL))
2856 dst_ops->default_advmss = xfrm_default_advmss;
2857 if (likely(dst_ops->mtu == NULL))
2858 dst_ops->mtu = xfrm_mtu;
2859 if (likely(dst_ops->negative_advice == NULL))
2860 dst_ops->negative_advice = xfrm_negative_advice;
2861 if (likely(dst_ops->link_failure == NULL))
2862 dst_ops->link_failure = xfrm_link_failure;
2863 if (likely(dst_ops->neigh_lookup == NULL))
2864 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2865 if (likely(!dst_ops->confirm_neigh))
2866 dst_ops->confirm_neigh = xfrm_confirm_neigh;
2867 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2869 spin_unlock(&xfrm_policy_afinfo_lock);
2873 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2875 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2877 struct dst_ops *dst_ops = afinfo->dst_ops;
2880 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2881 if (xfrm_policy_afinfo[i] != afinfo)
2883 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2889 dst_ops->kmem_cachep = NULL;
2890 dst_ops->check = NULL;
2891 dst_ops->negative_advice = NULL;
2892 dst_ops->link_failure = NULL;
2894 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2896 #ifdef CONFIG_XFRM_STATISTICS
2897 static int __net_init xfrm_statistics_init(struct net *net)
2900 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2901 if (!net->mib.xfrm_statistics)
2903 rv = xfrm_proc_init(net);
2905 free_percpu(net->mib.xfrm_statistics);
2909 static void xfrm_statistics_fini(struct net *net)
2911 xfrm_proc_fini(net);
2912 free_percpu(net->mib.xfrm_statistics);
2915 static int __net_init xfrm_statistics_init(struct net *net)
2920 static void xfrm_statistics_fini(struct net *net)
2925 static int __net_init xfrm_policy_init(struct net *net)
2927 unsigned int hmask, sz;
2930 if (net_eq(net, &init_net))
2931 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2932 sizeof(struct xfrm_dst),
2933 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2937 sz = (hmask+1) * sizeof(struct hlist_head);
2939 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2940 if (!net->xfrm.policy_byidx)
2942 net->xfrm.policy_idx_hmask = hmask;
2944 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2945 struct xfrm_policy_hash *htab;
2947 net->xfrm.policy_count[dir] = 0;
2948 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2949 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2951 htab = &net->xfrm.policy_bydst[dir];
2952 htab->table = xfrm_hash_alloc(sz);
2955 htab->hmask = hmask;
2961 net->xfrm.policy_hthresh.lbits4 = 32;
2962 net->xfrm.policy_hthresh.rbits4 = 32;
2963 net->xfrm.policy_hthresh.lbits6 = 128;
2964 net->xfrm.policy_hthresh.rbits6 = 128;
2966 seqlock_init(&net->xfrm.policy_hthresh.lock);
2968 INIT_LIST_HEAD(&net->xfrm.policy_all);
2969 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2970 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2971 if (net_eq(net, &init_net))
2976 for (dir--; dir >= 0; dir--) {
2977 struct xfrm_policy_hash *htab;
2979 htab = &net->xfrm.policy_bydst[dir];
2980 xfrm_hash_free(htab->table, sz);
2982 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2987 static void xfrm_policy_fini(struct net *net)
2992 flush_work(&net->xfrm.policy_hash_work);
2993 #ifdef CONFIG_XFRM_SUB_POLICY
2994 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2996 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2998 WARN_ON(!list_empty(&net->xfrm.policy_all));
3000 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3001 struct xfrm_policy_hash *htab;
3003 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3005 htab = &net->xfrm.policy_bydst[dir];
3006 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3007 WARN_ON(!hlist_empty(htab->table));
3008 xfrm_hash_free(htab->table, sz);
3011 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3012 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3013 xfrm_hash_free(net->xfrm.policy_byidx, sz);
3016 static int __net_init xfrm_net_init(struct net *net)
3020 /* Initialize the per-net locks here */
3021 spin_lock_init(&net->xfrm.xfrm_state_lock);
3022 spin_lock_init(&net->xfrm.xfrm_policy_lock);
3023 mutex_init(&net->xfrm.xfrm_cfg_mutex);
3025 rv = xfrm_statistics_init(net);
3027 goto out_statistics;
3028 rv = xfrm_state_init(net);
3031 rv = xfrm_policy_init(net);
3034 rv = xfrm_sysctl_init(net);
3037 rv = flow_cache_init(net);
3044 xfrm_sysctl_fini(net);
3046 xfrm_policy_fini(net);
3048 xfrm_state_fini(net);
3050 xfrm_statistics_fini(net);
3055 static void __net_exit xfrm_net_exit(struct net *net)
3057 flow_cache_fini(net);
3058 xfrm_sysctl_fini(net);
3059 xfrm_policy_fini(net);
3060 xfrm_state_fini(net);
3061 xfrm_statistics_fini(net);
3064 static struct pernet_operations __net_initdata xfrm_net_ops = {
3065 .init = xfrm_net_init,
3066 .exit = xfrm_net_exit,
3069 void __init xfrm_init(void)
3071 flow_cache_hp_init();
3072 register_pernet_subsys(&xfrm_net_ops);
3073 seqcount_init(&xfrm_policy_hash_generation);
3077 #ifdef CONFIG_AUDITSYSCALL
3078 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3079 struct audit_buffer *audit_buf)
3081 struct xfrm_sec_ctx *ctx = xp->security;
3082 struct xfrm_selector *sel = &xp->selector;
3085 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3086 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3088 switch (sel->family) {
3090 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3091 if (sel->prefixlen_s != 32)
3092 audit_log_format(audit_buf, " src_prefixlen=%d",
3094 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3095 if (sel->prefixlen_d != 32)
3096 audit_log_format(audit_buf, " dst_prefixlen=%d",
3100 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3101 if (sel->prefixlen_s != 128)
3102 audit_log_format(audit_buf, " src_prefixlen=%d",
3104 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3105 if (sel->prefixlen_d != 128)
3106 audit_log_format(audit_buf, " dst_prefixlen=%d",
3112 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3114 struct audit_buffer *audit_buf;
3116 audit_buf = xfrm_audit_start("SPD-add");
3117 if (audit_buf == NULL)
3119 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3120 audit_log_format(audit_buf, " res=%u", result);
3121 xfrm_audit_common_policyinfo(xp, audit_buf);
3122 audit_log_end(audit_buf);
3124 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3126 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3129 struct audit_buffer *audit_buf;
3131 audit_buf = xfrm_audit_start("SPD-delete");
3132 if (audit_buf == NULL)
3134 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3135 audit_log_format(audit_buf, " res=%u", result);
3136 xfrm_audit_common_policyinfo(xp, audit_buf);
3137 audit_log_end(audit_buf);
3139 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3142 #ifdef CONFIG_XFRM_MIGRATE
3143 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3144 const struct xfrm_selector *sel_tgt)
3146 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3147 if (sel_tgt->family == sel_cmp->family &&
3148 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3150 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3152 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3153 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3157 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3164 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3165 u8 dir, u8 type, struct net *net)
3167 struct xfrm_policy *pol, *ret = NULL;
3168 struct hlist_head *chain;
3171 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3172 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3173 hlist_for_each_entry(pol, chain, bydst) {
3174 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3175 pol->type == type) {
3177 priority = ret->priority;
3181 chain = &net->xfrm.policy_inexact[dir];
3182 hlist_for_each_entry(pol, chain, bydst) {
3183 if ((pol->priority >= priority) && ret)
3186 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3187 pol->type == type) {
3195 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3200 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3204 if (t->mode == m->mode && t->id.proto == m->proto &&
3205 (m->reqid == 0 || t->reqid == m->reqid)) {
3207 case XFRM_MODE_TUNNEL:
3208 case XFRM_MODE_BEET:
3209 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3211 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3216 case XFRM_MODE_TRANSPORT:
3217 /* in case of transport mode, template does not store
3218 any IP addresses, hence we just compare mode and
3229 /* update endpoint address(es) of template(s) */
3230 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3231 struct xfrm_migrate *m, int num_migrate)
3233 struct xfrm_migrate *mp;
3236 write_lock_bh(&pol->lock);
3237 if (unlikely(pol->walk.dead)) {
3238 /* target policy has been deleted */
3239 write_unlock_bh(&pol->lock);
3243 for (i = 0; i < pol->xfrm_nr; i++) {
3244 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3245 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3248 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3249 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3251 /* update endpoints */
3252 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3253 sizeof(pol->xfrm_vec[i].id.daddr));
3254 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3255 sizeof(pol->xfrm_vec[i].saddr));
3256 pol->xfrm_vec[i].encap_family = mp->new_family;
3258 atomic_inc(&pol->genid);
3262 write_unlock_bh(&pol->lock);
3270 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3274 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3277 for (i = 0; i < num_migrate; i++) {
3278 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3279 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3282 /* check if there is any duplicated entry */
3283 for (j = i + 1; j < num_migrate; j++) {
3284 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3285 sizeof(m[i].old_daddr)) &&
3286 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3287 sizeof(m[i].old_saddr)) &&
3288 m[i].proto == m[j].proto &&
3289 m[i].mode == m[j].mode &&
3290 m[i].reqid == m[j].reqid &&
3291 m[i].old_family == m[j].old_family)
3299 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3300 struct xfrm_migrate *m, int num_migrate,
3301 struct xfrm_kmaddress *k, struct net *net,
3302 struct xfrm_encap_tmpl *encap)
3304 int i, err, nx_cur = 0, nx_new = 0;
3305 struct xfrm_policy *pol = NULL;
3306 struct xfrm_state *x, *xc;
3307 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3308 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3309 struct xfrm_migrate *mp;
3311 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3314 /* Stage 1 - find policy */
3315 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3320 /* Stage 2 - find and update state(s) */
3321 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3322 if ((x = xfrm_migrate_state_find(mp, net))) {
3325 xc = xfrm_state_migrate(x, mp, encap);
3336 /* Stage 3 - update policy */
3337 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3340 /* Stage 4 - delete old state(s) */
3342 xfrm_states_put(x_cur, nx_cur);
3343 xfrm_states_delete(x_cur, nx_cur);
3346 /* Stage 5 - announce */
3347 km_migrate(sel, dir, type, m, num_migrate, k, encap);
3359 xfrm_states_put(x_cur, nx_cur);
3361 xfrm_states_delete(x_new, nx_new);
3365 EXPORT_SYMBOL(xfrm_migrate);