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/cpu.h>
28 #include <linux/audit.h>
33 #ifdef CONFIG_XFRM_STATISTICS
37 #include "xfrm_hash.h"
39 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
40 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
41 #define XFRM_MAX_QUEUE_LEN 100
44 struct dst_entry *dst_orig;
48 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
49 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
51 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
52 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
55 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
56 static __read_mostly seqcount_t xfrm_policy_hash_generation;
58 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
59 static int stale_bundle(struct dst_entry *dst);
60 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
61 static void xfrm_policy_queue_process(struct timer_list *t);
63 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
64 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
67 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
69 return refcount_inc_not_zero(&policy->refcnt);
73 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
75 const struct flowi4 *fl4 = &fl->u.ip4;
77 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
78 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
79 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
80 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
81 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
82 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
86 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
88 const struct flowi6 *fl6 = &fl->u.ip6;
90 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
91 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
92 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
93 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
94 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
95 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
98 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
99 unsigned short family)
103 return __xfrm4_selector_match(sel, fl);
105 return __xfrm6_selector_match(sel, fl);
110 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
112 const struct xfrm_policy_afinfo *afinfo;
114 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
117 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
118 if (unlikely(!afinfo))
123 /* Called with rcu_read_lock(). */
124 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
126 return rcu_dereference(xfrm_if_cb);
129 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
130 const xfrm_address_t *saddr,
131 const xfrm_address_t *daddr,
132 int family, u32 mark)
134 const struct xfrm_policy_afinfo *afinfo;
135 struct dst_entry *dst;
137 afinfo = xfrm_policy_get_afinfo(family);
138 if (unlikely(afinfo == NULL))
139 return ERR_PTR(-EAFNOSUPPORT);
141 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
147 EXPORT_SYMBOL(__xfrm_dst_lookup);
149 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
151 xfrm_address_t *prev_saddr,
152 xfrm_address_t *prev_daddr,
153 int family, u32 mark)
155 struct net *net = xs_net(x);
156 xfrm_address_t *saddr = &x->props.saddr;
157 xfrm_address_t *daddr = &x->id.daddr;
158 struct dst_entry *dst;
160 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
164 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
169 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
172 if (prev_saddr != saddr)
173 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
174 if (prev_daddr != daddr)
175 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
181 static inline unsigned long make_jiffies(long secs)
183 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
184 return MAX_SCHEDULE_TIMEOUT-1;
189 static void xfrm_policy_timer(struct timer_list *t)
191 struct xfrm_policy *xp = from_timer(xp, t, timer);
192 time64_t now = ktime_get_real_seconds();
193 time64_t next = TIME64_MAX;
197 read_lock(&xp->lock);
199 if (unlikely(xp->walk.dead))
202 dir = xfrm_policy_id2dir(xp->index);
204 if (xp->lft.hard_add_expires_seconds) {
205 time64_t tmo = xp->lft.hard_add_expires_seconds +
206 xp->curlft.add_time - now;
212 if (xp->lft.hard_use_expires_seconds) {
213 time64_t tmo = xp->lft.hard_use_expires_seconds +
214 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
220 if (xp->lft.soft_add_expires_seconds) {
221 time64_t tmo = xp->lft.soft_add_expires_seconds +
222 xp->curlft.add_time - now;
225 tmo = XFRM_KM_TIMEOUT;
230 if (xp->lft.soft_use_expires_seconds) {
231 time64_t tmo = xp->lft.soft_use_expires_seconds +
232 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
235 tmo = XFRM_KM_TIMEOUT;
242 km_policy_expired(xp, dir, 0, 0);
243 if (next != TIME64_MAX &&
244 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
248 read_unlock(&xp->lock);
253 read_unlock(&xp->lock);
254 if (!xfrm_policy_delete(xp, dir))
255 km_policy_expired(xp, dir, 1, 0);
259 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
263 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
265 struct xfrm_policy *policy;
267 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
270 write_pnet(&policy->xp_net, net);
271 INIT_LIST_HEAD(&policy->walk.all);
272 INIT_HLIST_NODE(&policy->bydst);
273 INIT_HLIST_NODE(&policy->byidx);
274 rwlock_init(&policy->lock);
275 refcount_set(&policy->refcnt, 1);
276 skb_queue_head_init(&policy->polq.hold_queue);
277 timer_setup(&policy->timer, xfrm_policy_timer, 0);
278 timer_setup(&policy->polq.hold_timer,
279 xfrm_policy_queue_process, 0);
283 EXPORT_SYMBOL(xfrm_policy_alloc);
285 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
287 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
289 security_xfrm_policy_free(policy->security);
293 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
295 void xfrm_policy_destroy(struct xfrm_policy *policy)
297 BUG_ON(!policy->walk.dead);
299 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
302 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
304 EXPORT_SYMBOL(xfrm_policy_destroy);
306 /* Rule must be locked. Release descendant resources, announce
307 * entry dead. The rule must be unlinked from lists to the moment.
310 static void xfrm_policy_kill(struct xfrm_policy *policy)
312 policy->walk.dead = 1;
314 atomic_inc(&policy->genid);
316 if (del_timer(&policy->polq.hold_timer))
317 xfrm_pol_put(policy);
318 skb_queue_purge(&policy->polq.hold_queue);
320 if (del_timer(&policy->timer))
321 xfrm_pol_put(policy);
323 xfrm_pol_put(policy);
326 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
328 static inline unsigned int idx_hash(struct net *net, u32 index)
330 return __idx_hash(index, net->xfrm.policy_idx_hmask);
333 /* calculate policy hash thresholds */
334 static void __get_hash_thresh(struct net *net,
335 unsigned short family, int dir,
336 u8 *dbits, u8 *sbits)
340 *dbits = net->xfrm.policy_bydst[dir].dbits4;
341 *sbits = net->xfrm.policy_bydst[dir].sbits4;
345 *dbits = net->xfrm.policy_bydst[dir].dbits6;
346 *sbits = net->xfrm.policy_bydst[dir].sbits6;
355 static struct hlist_head *policy_hash_bysel(struct net *net,
356 const struct xfrm_selector *sel,
357 unsigned short family, int dir)
359 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
364 __get_hash_thresh(net, family, dir, &dbits, &sbits);
365 hash = __sel_hash(sel, family, hmask, dbits, sbits);
367 if (hash == hmask + 1)
368 return &net->xfrm.policy_inexact[dir];
370 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
371 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
374 static struct hlist_head *policy_hash_direct(struct net *net,
375 const xfrm_address_t *daddr,
376 const xfrm_address_t *saddr,
377 unsigned short family, int dir)
379 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
384 __get_hash_thresh(net, family, dir, &dbits, &sbits);
385 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
387 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
388 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
391 static void xfrm_dst_hash_transfer(struct net *net,
392 struct hlist_head *list,
393 struct hlist_head *ndsttable,
394 unsigned int nhashmask,
397 struct hlist_node *tmp, *entry0 = NULL;
398 struct xfrm_policy *pol;
404 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
407 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
408 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
409 pol->family, nhashmask, dbits, sbits);
411 hlist_del_rcu(&pol->bydst);
412 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
417 hlist_del_rcu(&pol->bydst);
418 hlist_add_behind_rcu(&pol->bydst, entry0);
420 entry0 = &pol->bydst;
422 if (!hlist_empty(list)) {
428 static void xfrm_idx_hash_transfer(struct hlist_head *list,
429 struct hlist_head *nidxtable,
430 unsigned int nhashmask)
432 struct hlist_node *tmp;
433 struct xfrm_policy *pol;
435 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
438 h = __idx_hash(pol->index, nhashmask);
439 hlist_add_head(&pol->byidx, nidxtable+h);
443 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
445 return ((old_hmask + 1) << 1) - 1;
448 static void xfrm_bydst_resize(struct net *net, int dir)
450 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
451 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
452 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
453 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
454 struct hlist_head *odst;
460 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
461 write_seqcount_begin(&xfrm_policy_hash_generation);
463 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
464 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
466 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
467 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
469 for (i = hmask; i >= 0; i--)
470 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
472 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
473 net->xfrm.policy_bydst[dir].hmask = nhashmask;
475 write_seqcount_end(&xfrm_policy_hash_generation);
476 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
480 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
483 static void xfrm_byidx_resize(struct net *net, int total)
485 unsigned int hmask = net->xfrm.policy_idx_hmask;
486 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
487 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
488 struct hlist_head *oidx = net->xfrm.policy_byidx;
489 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
495 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
497 for (i = hmask; i >= 0; i--)
498 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
500 net->xfrm.policy_byidx = nidx;
501 net->xfrm.policy_idx_hmask = nhashmask;
503 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
505 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
508 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
510 unsigned int cnt = net->xfrm.policy_count[dir];
511 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
516 if ((hmask + 1) < xfrm_policy_hashmax &&
523 static inline int xfrm_byidx_should_resize(struct net *net, int total)
525 unsigned int hmask = net->xfrm.policy_idx_hmask;
527 if ((hmask + 1) < xfrm_policy_hashmax &&
534 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
536 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
537 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
538 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
539 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
540 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
541 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
542 si->spdhcnt = net->xfrm.policy_idx_hmask;
543 si->spdhmcnt = xfrm_policy_hashmax;
545 EXPORT_SYMBOL(xfrm_spd_getinfo);
547 static DEFINE_MUTEX(hash_resize_mutex);
548 static void xfrm_hash_resize(struct work_struct *work)
550 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
553 mutex_lock(&hash_resize_mutex);
556 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
557 if (xfrm_bydst_should_resize(net, dir, &total))
558 xfrm_bydst_resize(net, dir);
560 if (xfrm_byidx_should_resize(net, total))
561 xfrm_byidx_resize(net, total);
563 mutex_unlock(&hash_resize_mutex);
566 static void xfrm_hash_rebuild(struct work_struct *work)
568 struct net *net = container_of(work, struct net,
569 xfrm.policy_hthresh.work);
571 struct xfrm_policy *pol;
572 struct xfrm_policy *policy;
573 struct hlist_head *chain;
574 struct hlist_head *odst;
575 struct hlist_node *newpos;
579 u8 lbits4, rbits4, lbits6, rbits6;
581 mutex_lock(&hash_resize_mutex);
583 /* read selector prefixlen thresholds */
585 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
587 lbits4 = net->xfrm.policy_hthresh.lbits4;
588 rbits4 = net->xfrm.policy_hthresh.rbits4;
589 lbits6 = net->xfrm.policy_hthresh.lbits6;
590 rbits6 = net->xfrm.policy_hthresh.rbits6;
591 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
593 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
595 /* reset the bydst and inexact table in all directions */
596 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
597 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
598 hmask = net->xfrm.policy_bydst[dir].hmask;
599 odst = net->xfrm.policy_bydst[dir].table;
600 for (i = hmask; i >= 0; i--)
601 INIT_HLIST_HEAD(odst + i);
602 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
603 /* dir out => dst = remote, src = local */
604 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
605 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
606 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
607 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
609 /* dir in/fwd => dst = local, src = remote */
610 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
611 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
612 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
613 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
617 /* re-insert all policies by order of creation */
618 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
619 if (policy->walk.dead ||
620 xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
621 /* skip socket policies */
625 chain = policy_hash_bysel(net, &policy->selector,
627 xfrm_policy_id2dir(policy->index));
628 hlist_for_each_entry(pol, chain, bydst) {
629 if (policy->priority >= pol->priority)
630 newpos = &pol->bydst;
635 hlist_add_behind(&policy->bydst, newpos);
637 hlist_add_head(&policy->bydst, chain);
640 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
642 mutex_unlock(&hash_resize_mutex);
645 void xfrm_policy_hash_rebuild(struct net *net)
647 schedule_work(&net->xfrm.policy_hthresh.work);
649 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
651 /* Generate new index... KAME seems to generate them ordered by cost
652 * of an absolute inpredictability of ordering of rules. This will not pass. */
653 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
655 static u32 idx_generator;
658 struct hlist_head *list;
659 struct xfrm_policy *p;
664 idx = (idx_generator | dir);
673 list = net->xfrm.policy_byidx + idx_hash(net, idx);
675 hlist_for_each_entry(p, list, byidx) {
676 if (p->index == idx) {
686 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
688 u32 *p1 = (u32 *) s1;
689 u32 *p2 = (u32 *) s2;
690 int len = sizeof(struct xfrm_selector) / sizeof(u32);
693 for (i = 0; i < len; i++) {
701 static void xfrm_policy_requeue(struct xfrm_policy *old,
702 struct xfrm_policy *new)
704 struct xfrm_policy_queue *pq = &old->polq;
705 struct sk_buff_head list;
707 if (skb_queue_empty(&pq->hold_queue))
710 __skb_queue_head_init(&list);
712 spin_lock_bh(&pq->hold_queue.lock);
713 skb_queue_splice_init(&pq->hold_queue, &list);
714 if (del_timer(&pq->hold_timer))
716 spin_unlock_bh(&pq->hold_queue.lock);
720 spin_lock_bh(&pq->hold_queue.lock);
721 skb_queue_splice(&list, &pq->hold_queue);
722 pq->timeout = XFRM_QUEUE_TMO_MIN;
723 if (!mod_timer(&pq->hold_timer, jiffies))
725 spin_unlock_bh(&pq->hold_queue.lock);
728 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
729 struct xfrm_policy *pol)
731 u32 mark = policy->mark.v & policy->mark.m;
733 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
736 if ((mark & pol->mark.m) == pol->mark.v &&
737 policy->priority == pol->priority)
743 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
745 struct net *net = xp_net(policy);
746 struct xfrm_policy *pol;
747 struct xfrm_policy *delpol;
748 struct hlist_head *chain;
749 struct hlist_node *newpos;
751 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
752 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
755 hlist_for_each_entry(pol, chain, bydst) {
756 if (pol->type == policy->type &&
757 pol->if_id == policy->if_id &&
758 !selector_cmp(&pol->selector, &policy->selector) &&
759 xfrm_policy_mark_match(policy, pol) &&
760 xfrm_sec_ctx_match(pol->security, policy->security) &&
763 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
767 if (policy->priority > pol->priority)
769 } else if (policy->priority >= pol->priority) {
770 newpos = &pol->bydst;
777 hlist_add_behind(&policy->bydst, newpos);
779 hlist_add_head(&policy->bydst, chain);
780 __xfrm_policy_link(policy, dir);
782 /* After previous checking, family can either be AF_INET or AF_INET6 */
783 if (policy->family == AF_INET)
784 rt_genid_bump_ipv4(net);
786 rt_genid_bump_ipv6(net);
789 xfrm_policy_requeue(delpol, policy);
790 __xfrm_policy_unlink(delpol, dir);
792 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
793 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
794 policy->curlft.add_time = ktime_get_real_seconds();
795 policy->curlft.use_time = 0;
796 if (!mod_timer(&policy->timer, jiffies + HZ))
797 xfrm_pol_hold(policy);
798 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
801 xfrm_policy_kill(delpol);
802 else if (xfrm_bydst_should_resize(net, dir, NULL))
803 schedule_work(&net->xfrm.policy_hash_work);
807 EXPORT_SYMBOL(xfrm_policy_insert);
809 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u32 if_id,
811 struct xfrm_selector *sel,
812 struct xfrm_sec_ctx *ctx, int delete,
815 struct xfrm_policy *pol, *ret;
816 struct hlist_head *chain;
819 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
820 chain = policy_hash_bysel(net, sel, sel->family, dir);
822 hlist_for_each_entry(pol, chain, bydst) {
823 if (pol->type == type &&
824 pol->if_id == if_id &&
825 (mark & pol->mark.m) == pol->mark.v &&
826 !selector_cmp(sel, &pol->selector) &&
827 xfrm_sec_ctx_match(ctx, pol->security)) {
830 *err = security_xfrm_policy_delete(
833 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
836 __xfrm_policy_unlink(pol, dir);
842 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
845 xfrm_policy_kill(ret);
848 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
850 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u32 if_id,
851 u8 type, int dir, u32 id, int delete,
854 struct xfrm_policy *pol, *ret;
855 struct hlist_head *chain;
858 if (xfrm_policy_id2dir(id) != dir)
862 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
863 chain = net->xfrm.policy_byidx + idx_hash(net, id);
865 hlist_for_each_entry(pol, chain, byidx) {
866 if (pol->type == type && pol->index == id &&
867 pol->if_id == if_id &&
868 (mark & pol->mark.m) == pol->mark.v) {
871 *err = security_xfrm_policy_delete(
874 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
877 __xfrm_policy_unlink(pol, dir);
883 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
886 xfrm_policy_kill(ret);
889 EXPORT_SYMBOL(xfrm_policy_byid);
891 #ifdef CONFIG_SECURITY_NETWORK_XFRM
893 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
897 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
898 struct xfrm_policy *pol;
901 hlist_for_each_entry(pol,
902 &net->xfrm.policy_inexact[dir], bydst) {
903 if (pol->type != type)
905 err = security_xfrm_policy_delete(pol->security);
907 xfrm_audit_policy_delete(pol, 0, task_valid);
911 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
912 hlist_for_each_entry(pol,
913 net->xfrm.policy_bydst[dir].table + i,
915 if (pol->type != type)
917 err = security_xfrm_policy_delete(
920 xfrm_audit_policy_delete(pol, 0,
931 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
937 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
939 int dir, err = 0, cnt = 0;
941 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
943 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
947 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
948 struct xfrm_policy *pol;
952 hlist_for_each_entry(pol,
953 &net->xfrm.policy_inexact[dir], bydst) {
954 if (pol->type != type)
956 __xfrm_policy_unlink(pol, dir);
957 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
960 xfrm_audit_policy_delete(pol, 1, task_valid);
962 xfrm_policy_kill(pol);
964 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
968 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
970 hlist_for_each_entry(pol,
971 net->xfrm.policy_bydst[dir].table + i,
973 if (pol->type != type)
975 __xfrm_policy_unlink(pol, dir);
976 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
979 xfrm_audit_policy_delete(pol, 1, task_valid);
980 xfrm_policy_kill(pol);
982 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
991 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
994 EXPORT_SYMBOL(xfrm_policy_flush);
996 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
997 int (*func)(struct xfrm_policy *, int, int, void*),
1000 struct xfrm_policy *pol;
1001 struct xfrm_policy_walk_entry *x;
1004 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1005 walk->type != XFRM_POLICY_TYPE_ANY)
1008 if (list_empty(&walk->walk.all) && walk->seq != 0)
1011 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1012 if (list_empty(&walk->walk.all))
1013 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1015 x = list_first_entry(&walk->walk.all,
1016 struct xfrm_policy_walk_entry, all);
1018 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1021 pol = container_of(x, struct xfrm_policy, walk);
1022 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1023 walk->type != pol->type)
1025 error = func(pol, xfrm_policy_id2dir(pol->index),
1028 list_move_tail(&walk->walk.all, &x->all);
1033 if (walk->seq == 0) {
1037 list_del_init(&walk->walk.all);
1039 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1042 EXPORT_SYMBOL(xfrm_policy_walk);
1044 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1046 INIT_LIST_HEAD(&walk->walk.all);
1047 walk->walk.dead = 1;
1051 EXPORT_SYMBOL(xfrm_policy_walk_init);
1053 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1055 if (list_empty(&walk->walk.all))
1058 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1059 list_del(&walk->walk.all);
1060 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1062 EXPORT_SYMBOL(xfrm_policy_walk_done);
1065 * Find policy to apply to this flow.
1067 * Returns 0 if policy found, else an -errno.
1069 static int xfrm_policy_match(const struct xfrm_policy *pol,
1070 const struct flowi *fl,
1071 u8 type, u16 family, int dir, u32 if_id)
1073 const struct xfrm_selector *sel = &pol->selector;
1077 if (pol->family != family ||
1078 pol->if_id != if_id ||
1079 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1083 match = xfrm_selector_match(sel, fl, family);
1085 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1091 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1092 const struct flowi *fl,
1097 struct xfrm_policy *pol, *ret;
1098 const xfrm_address_t *daddr, *saddr;
1099 struct hlist_head *chain;
1100 unsigned int sequence;
1103 daddr = xfrm_flowi_daddr(fl, family);
1104 saddr = xfrm_flowi_saddr(fl, family);
1105 if (unlikely(!daddr || !saddr))
1111 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1112 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1113 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1117 hlist_for_each_entry_rcu(pol, chain, bydst) {
1118 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
1128 priority = ret->priority;
1132 chain = &net->xfrm.policy_inexact[dir];
1133 hlist_for_each_entry_rcu(pol, chain, bydst) {
1134 if ((pol->priority >= priority) && ret)
1137 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
1151 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1154 if (ret && !xfrm_pol_hold_rcu(ret))
1162 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
1163 const struct flowi *fl,
1164 u16 family, u8 dir, u32 if_id)
1166 #ifdef CONFIG_XFRM_SUB_POLICY
1167 struct xfrm_policy *pol;
1169 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
1174 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
1178 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1179 const struct flowi *fl,
1180 u16 family, u32 if_id)
1182 struct xfrm_policy *pol;
1186 pol = rcu_dereference(sk->sk_policy[dir]);
1191 if (pol->family != family) {
1196 match = xfrm_selector_match(&pol->selector, fl, family);
1198 if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
1199 pol->if_id != if_id) {
1203 err = security_xfrm_policy_lookup(pol->security,
1207 if (!xfrm_pol_hold_rcu(pol))
1209 } else if (err == -ESRCH) {
1222 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1224 struct net *net = xp_net(pol);
1226 list_add(&pol->walk.all, &net->xfrm.policy_all);
1227 net->xfrm.policy_count[dir]++;
1231 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1234 struct net *net = xp_net(pol);
1236 if (list_empty(&pol->walk.all))
1239 /* Socket policies are not hashed. */
1240 if (!hlist_unhashed(&pol->bydst)) {
1241 hlist_del_rcu(&pol->bydst);
1242 hlist_del(&pol->byidx);
1245 list_del_init(&pol->walk.all);
1246 net->xfrm.policy_count[dir]--;
1251 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1253 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1256 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1258 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1261 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1263 struct net *net = xp_net(pol);
1265 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1266 pol = __xfrm_policy_unlink(pol, dir);
1267 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1269 xfrm_policy_kill(pol);
1274 EXPORT_SYMBOL(xfrm_policy_delete);
1276 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1278 struct net *net = sock_net(sk);
1279 struct xfrm_policy *old_pol;
1281 #ifdef CONFIG_XFRM_SUB_POLICY
1282 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1286 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1287 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1288 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1290 pol->curlft.add_time = ktime_get_real_seconds();
1291 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1292 xfrm_sk_policy_link(pol, dir);
1294 rcu_assign_pointer(sk->sk_policy[dir], pol);
1297 xfrm_policy_requeue(old_pol, pol);
1299 /* Unlinking succeeds always. This is the only function
1300 * allowed to delete or replace socket policy.
1302 xfrm_sk_policy_unlink(old_pol, dir);
1304 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1307 xfrm_policy_kill(old_pol);
1312 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1314 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1315 struct net *net = xp_net(old);
1318 newp->selector = old->selector;
1319 if (security_xfrm_policy_clone(old->security,
1322 return NULL; /* ENOMEM */
1324 newp->lft = old->lft;
1325 newp->curlft = old->curlft;
1326 newp->mark = old->mark;
1327 newp->if_id = old->if_id;
1328 newp->action = old->action;
1329 newp->flags = old->flags;
1330 newp->xfrm_nr = old->xfrm_nr;
1331 newp->index = old->index;
1332 newp->type = old->type;
1333 newp->family = old->family;
1334 memcpy(newp->xfrm_vec, old->xfrm_vec,
1335 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1336 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1337 xfrm_sk_policy_link(newp, dir);
1338 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1344 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1346 const struct xfrm_policy *p;
1347 struct xfrm_policy *np;
1351 for (i = 0; i < 2; i++) {
1352 p = rcu_dereference(osk->sk_policy[i]);
1354 np = clone_policy(p, i);
1355 if (unlikely(!np)) {
1359 rcu_assign_pointer(sk->sk_policy[i], np);
1367 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1368 xfrm_address_t *remote, unsigned short family, u32 mark)
1371 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1373 if (unlikely(afinfo == NULL))
1375 err = afinfo->get_saddr(net, oif, local, remote, mark);
1380 /* Resolve list of templates for the flow, given policy. */
1383 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1384 struct xfrm_state **xfrm, unsigned short family)
1386 struct net *net = xp_net(policy);
1389 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1390 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1393 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1394 struct xfrm_state *x;
1395 xfrm_address_t *remote = daddr;
1396 xfrm_address_t *local = saddr;
1397 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1399 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1400 tmpl->mode == XFRM_MODE_BEET) {
1401 remote = &tmpl->id.daddr;
1402 local = &tmpl->saddr;
1403 if (xfrm_addr_any(local, tmpl->encap_family)) {
1404 error = xfrm_get_saddr(net, fl->flowi_oif,
1406 tmpl->encap_family, 0);
1413 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
1414 family, policy->if_id);
1416 if (x && x->km.state == XFRM_STATE_VALID) {
1423 error = (x->km.state == XFRM_STATE_ERROR ?
1426 } else if (error == -ESRCH) {
1430 if (!tmpl->optional)
1436 for (nx--; nx >= 0; nx--)
1437 xfrm_state_put(xfrm[nx]);
1442 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1443 struct xfrm_state **xfrm, unsigned short family)
1445 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1446 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1452 for (i = 0; i < npols; i++) {
1453 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1458 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1466 /* found states are sorted for outbound processing */
1468 xfrm_state_sort(xfrm, tpp, cnx, family);
1473 for (cnx--; cnx >= 0; cnx--)
1474 xfrm_state_put(tpp[cnx]);
1479 static int xfrm_get_tos(const struct flowi *fl, int family)
1481 const struct xfrm_policy_afinfo *afinfo;
1484 afinfo = xfrm_policy_get_afinfo(family);
1488 tos = afinfo->get_tos(fl);
1495 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1497 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1498 struct dst_ops *dst_ops;
1499 struct xfrm_dst *xdst;
1502 return ERR_PTR(-EINVAL);
1506 dst_ops = &net->xfrm.xfrm4_dst_ops;
1508 #if IS_ENABLED(CONFIG_IPV6)
1510 dst_ops = &net->xfrm.xfrm6_dst_ops;
1516 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1519 struct dst_entry *dst = &xdst->u.dst;
1521 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1523 xdst = ERR_PTR(-ENOBUFS);
1530 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1533 const struct xfrm_policy_afinfo *afinfo =
1534 xfrm_policy_get_afinfo(dst->ops->family);
1540 err = afinfo->init_path(path, dst, nfheader_len);
1547 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1548 const struct flowi *fl)
1550 const struct xfrm_policy_afinfo *afinfo =
1551 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1557 err = afinfo->fill_dst(xdst, dev, fl);
1565 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1566 * all the metrics... Shortly, bundle a bundle.
1569 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1570 struct xfrm_state **xfrm,
1571 struct xfrm_dst **bundle,
1573 const struct flowi *fl,
1574 struct dst_entry *dst)
1576 struct net *net = xp_net(policy);
1577 unsigned long now = jiffies;
1578 struct net_device *dev;
1579 struct xfrm_mode *inner_mode;
1580 struct xfrm_dst *xdst_prev = NULL;
1581 struct xfrm_dst *xdst0 = NULL;
1585 int nfheader_len = 0;
1586 int trailer_len = 0;
1588 int family = policy->selector.family;
1589 xfrm_address_t saddr, daddr;
1591 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1593 tos = xfrm_get_tos(fl, family);
1597 for (; i < nx; i++) {
1598 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1599 struct dst_entry *dst1 = &xdst->u.dst;
1601 err = PTR_ERR(xdst);
1611 /* Ref count is taken during xfrm_alloc_dst()
1612 * No need to do dst_clone() on dst1
1614 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
1616 if (xfrm[i]->sel.family == AF_UNSPEC) {
1617 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1618 xfrm_af2proto(family));
1620 err = -EAFNOSUPPORT;
1625 inner_mode = xfrm[i]->inner_mode;
1628 dst_copy_metrics(dst1, dst);
1630 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1631 __u32 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
1633 family = xfrm[i]->props.family;
1634 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1635 &saddr, &daddr, family, mark);
1642 dst1->xfrm = xfrm[i];
1643 xdst->xfrm_genid = xfrm[i]->genid;
1645 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1646 dst1->flags |= DST_HOST;
1647 dst1->lastuse = now;
1649 dst1->input = dst_discard;
1650 dst1->output = inner_mode->afinfo->output;
1654 header_len += xfrm[i]->props.header_len;
1655 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1656 nfheader_len += xfrm[i]->props.header_len;
1657 trailer_len += xfrm[i]->props.trailer_len;
1660 xfrm_dst_set_child(xdst_prev, dst);
1668 xfrm_init_path(xdst0, dst, nfheader_len);
1669 xfrm_init_pmtu(bundle, nx);
1671 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
1672 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
1673 err = xfrm_fill_dst(xdst_prev, dev, fl);
1677 xdst_prev->u.dst.header_len = header_len;
1678 xdst_prev->u.dst.trailer_len = trailer_len;
1679 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
1680 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
1683 return &xdst0->u.dst;
1687 xfrm_state_put(xfrm[i]);
1690 dst_release_immediate(&xdst0->u.dst);
1692 return ERR_PTR(err);
1695 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1696 struct xfrm_policy **pols,
1697 int *num_pols, int *num_xfrms)
1701 if (*num_pols == 0 || !pols[0]) {
1706 if (IS_ERR(pols[0]))
1707 return PTR_ERR(pols[0]);
1709 *num_xfrms = pols[0]->xfrm_nr;
1711 #ifdef CONFIG_XFRM_SUB_POLICY
1712 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1713 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1714 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1715 XFRM_POLICY_TYPE_MAIN,
1720 if (IS_ERR(pols[1])) {
1721 xfrm_pols_put(pols, *num_pols);
1722 return PTR_ERR(pols[1]);
1725 (*num_xfrms) += pols[1]->xfrm_nr;
1729 for (i = 0; i < *num_pols; i++) {
1730 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1740 static struct xfrm_dst *
1741 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1742 const struct flowi *fl, u16 family,
1743 struct dst_entry *dst_orig)
1745 struct net *net = xp_net(pols[0]);
1746 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1747 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
1748 struct xfrm_dst *xdst;
1749 struct dst_entry *dst;
1752 /* Try to instantiate a bundle */
1753 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1759 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1760 return ERR_PTR(err);
1763 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
1765 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1766 return ERR_CAST(dst);
1769 xdst = (struct xfrm_dst *)dst;
1770 xdst->num_xfrms = err;
1771 xdst->num_pols = num_pols;
1772 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1773 xdst->policy_genid = atomic_read(&pols[0]->genid);
1778 static void xfrm_policy_queue_process(struct timer_list *t)
1780 struct sk_buff *skb;
1782 struct dst_entry *dst;
1783 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1784 struct net *net = xp_net(pol);
1785 struct xfrm_policy_queue *pq = &pol->polq;
1787 struct sk_buff_head list;
1789 spin_lock(&pq->hold_queue.lock);
1790 skb = skb_peek(&pq->hold_queue);
1792 spin_unlock(&pq->hold_queue.lock);
1797 xfrm_decode_session(skb, &fl, dst->ops->family);
1798 spin_unlock(&pq->hold_queue.lock);
1800 dst_hold(xfrm_dst_path(dst));
1801 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
1805 if (dst->flags & DST_XFRM_QUEUE) {
1808 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1811 pq->timeout = pq->timeout << 1;
1812 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1819 __skb_queue_head_init(&list);
1821 spin_lock(&pq->hold_queue.lock);
1823 skb_queue_splice_init(&pq->hold_queue, &list);
1824 spin_unlock(&pq->hold_queue.lock);
1826 while (!skb_queue_empty(&list)) {
1827 skb = __skb_dequeue(&list);
1829 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1830 dst_hold(xfrm_dst_path(skb_dst(skb)));
1831 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
1839 skb_dst_set(skb, dst);
1841 dst_output(net, skb->sk, skb);
1850 skb_queue_purge(&pq->hold_queue);
1854 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1856 unsigned long sched_next;
1857 struct dst_entry *dst = skb_dst(skb);
1858 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1859 struct xfrm_policy *pol = xdst->pols[0];
1860 struct xfrm_policy_queue *pq = &pol->polq;
1862 if (unlikely(skb_fclone_busy(sk, skb))) {
1867 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1874 spin_lock_bh(&pq->hold_queue.lock);
1877 pq->timeout = XFRM_QUEUE_TMO_MIN;
1879 sched_next = jiffies + pq->timeout;
1881 if (del_timer(&pq->hold_timer)) {
1882 if (time_before(pq->hold_timer.expires, sched_next))
1883 sched_next = pq->hold_timer.expires;
1887 __skb_queue_tail(&pq->hold_queue, skb);
1888 if (!mod_timer(&pq->hold_timer, sched_next))
1891 spin_unlock_bh(&pq->hold_queue.lock);
1896 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1897 struct xfrm_flo *xflo,
1898 const struct flowi *fl,
1903 struct net_device *dev;
1904 struct dst_entry *dst;
1905 struct dst_entry *dst1;
1906 struct xfrm_dst *xdst;
1908 xdst = xfrm_alloc_dst(net, family);
1912 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
1913 net->xfrm.sysctl_larval_drop ||
1917 dst = xflo->dst_orig;
1918 dst1 = &xdst->u.dst;
1922 dst_copy_metrics(dst1, dst);
1924 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1925 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1926 dst1->lastuse = jiffies;
1928 dst1->input = dst_discard;
1929 dst1->output = xdst_queue_output;
1932 xfrm_dst_set_child(xdst, dst);
1935 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1942 err = xfrm_fill_dst(xdst, dev, fl);
1951 xdst = ERR_PTR(err);
1955 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
1956 const struct flowi *fl,
1958 struct xfrm_flo *xflo, u32 if_id)
1960 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1961 int num_pols = 0, num_xfrms = 0, err;
1962 struct xfrm_dst *xdst;
1964 /* Resolve policies to use if we couldn't get them from
1965 * previous cache entry */
1967 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
1968 err = xfrm_expand_policies(fl, family, pols,
1969 &num_pols, &num_xfrms);
1975 goto make_dummy_bundle;
1977 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
1980 err = PTR_ERR(xdst);
1981 if (err == -EREMOTE) {
1982 xfrm_pols_put(pols, num_pols);
1988 goto make_dummy_bundle;
1989 } else if (xdst == NULL) {
1991 goto make_dummy_bundle;
1997 /* We found policies, but there's no bundles to instantiate:
1998 * either because the policy blocks, has no transformations or
1999 * we could not build template (no xfrm_states).*/
2000 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2002 xfrm_pols_put(pols, num_pols);
2003 return ERR_CAST(xdst);
2005 xdst->num_pols = num_pols;
2006 xdst->num_xfrms = num_xfrms;
2007 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2012 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2014 xfrm_pols_put(pols, num_pols);
2015 return ERR_PTR(err);
2018 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2019 struct dst_entry *dst_orig)
2021 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2022 struct dst_entry *ret;
2025 dst_release(dst_orig);
2026 return ERR_PTR(-EINVAL);
2028 ret = afinfo->blackhole_route(net, dst_orig);
2035 /* Finds/creates a bundle for given flow and if_id
2037 * At the moment we eat a raw IP route. Mostly to speed up lookups
2038 * on interfaces with disabled IPsec.
2040 * xfrm_lookup uses an if_id of 0 by default, and is provided for
2043 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
2044 struct dst_entry *dst_orig,
2045 const struct flowi *fl,
2046 const struct sock *sk,
2047 int flags, u32 if_id)
2049 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2050 struct xfrm_dst *xdst;
2051 struct dst_entry *dst, *route;
2052 u16 family = dst_orig->ops->family;
2053 u8 dir = XFRM_POLICY_OUT;
2054 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2060 sk = sk_const_to_full_sk(sk);
2061 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2063 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
2065 err = xfrm_expand_policies(fl, family, pols,
2066 &num_pols, &num_xfrms);
2071 if (num_xfrms <= 0) {
2072 drop_pols = num_pols;
2076 xdst = xfrm_resolve_and_create_bundle(
2081 xfrm_pols_put(pols, num_pols);
2082 err = PTR_ERR(xdst);
2083 if (err == -EREMOTE)
2087 } else if (xdst == NULL) {
2089 drop_pols = num_pols;
2093 route = xdst->route;
2098 struct xfrm_flo xflo;
2100 xflo.dst_orig = dst_orig;
2103 /* To accelerate a bit... */
2104 if ((dst_orig->flags & DST_NOXFRM) ||
2105 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2108 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
2112 err = PTR_ERR(xdst);
2116 num_pols = xdst->num_pols;
2117 num_xfrms = xdst->num_xfrms;
2118 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2119 route = xdst->route;
2123 if (route == NULL && num_xfrms > 0) {
2124 /* The only case when xfrm_bundle_lookup() returns a
2125 * bundle with null route, is when the template could
2126 * not be resolved. It means policies are there, but
2127 * bundle could not be created, since we don't yet
2128 * have the xfrm_state's. We need to wait for KM to
2129 * negotiate new SA's or bail out with error.*/
2130 if (net->xfrm.sysctl_larval_drop) {
2131 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2138 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2146 if ((flags & XFRM_LOOKUP_ICMP) &&
2147 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2152 for (i = 0; i < num_pols; i++)
2153 pols[i]->curlft.use_time = ktime_get_real_seconds();
2155 if (num_xfrms < 0) {
2156 /* Prohibit the flow */
2157 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2160 } else if (num_xfrms > 0) {
2161 /* Flow transformed */
2162 dst_release(dst_orig);
2164 /* Flow passes untransformed */
2169 xfrm_pols_put(pols, drop_pols);
2170 if (dst && dst->xfrm &&
2171 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2172 dst->flags |= DST_XFRM_TUNNEL;
2176 if (!(flags & XFRM_LOOKUP_ICMP)) {
2184 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2185 dst_release(dst_orig);
2186 xfrm_pols_put(pols, drop_pols);
2187 return ERR_PTR(err);
2189 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
2191 /* Main function: finds/creates a bundle for given flow.
2193 * At the moment we eat a raw IP route. Mostly to speed up lookups
2194 * on interfaces with disabled IPsec.
2196 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2197 const struct flowi *fl, const struct sock *sk,
2200 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
2202 EXPORT_SYMBOL(xfrm_lookup);
2204 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2205 * Otherwise we may send out blackholed packets.
2207 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2208 const struct flowi *fl,
2209 const struct sock *sk, int flags)
2211 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2212 flags | XFRM_LOOKUP_QUEUE |
2213 XFRM_LOOKUP_KEEP_DST_REF);
2215 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2216 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2219 dst_release(dst_orig);
2223 EXPORT_SYMBOL(xfrm_lookup_route);
2226 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2228 struct xfrm_state *x;
2230 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2232 x = skb->sp->xvec[idx];
2233 if (!x->type->reject)
2235 return x->type->reject(x, skb, fl);
2238 /* When skb is transformed back to its "native" form, we have to
2239 * check policy restrictions. At the moment we make this in maximally
2240 * stupid way. Shame on me. :-) Of course, connected sockets must
2241 * have policy cached at them.
2245 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2246 unsigned short family)
2248 if (xfrm_state_kern(x))
2249 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2250 return x->id.proto == tmpl->id.proto &&
2251 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2252 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2253 x->props.mode == tmpl->mode &&
2254 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2255 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2256 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2257 xfrm_state_addr_cmp(tmpl, x, family));
2261 * 0 or more than 0 is returned when validation is succeeded (either bypass
2262 * because of optional transport mode, or next index of the mathced secpath
2263 * state with the template.
2264 * -1 is returned when no matching template is found.
2265 * Otherwise "-2 - errored_index" is returned.
2268 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2269 unsigned short family)
2273 if (tmpl->optional) {
2274 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2278 for (; idx < sp->len; idx++) {
2279 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2281 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2290 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2291 unsigned int family, int reverse)
2293 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2296 if (unlikely(afinfo == NULL))
2297 return -EAFNOSUPPORT;
2299 afinfo->decode_session(skb, fl, reverse);
2301 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2305 EXPORT_SYMBOL(__xfrm_decode_session);
2307 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2309 for (; k < sp->len; k++) {
2310 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2319 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2320 unsigned short family)
2322 struct net *net = dev_net(skb->dev);
2323 struct xfrm_policy *pol;
2324 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2331 const struct xfrm_if_cb *ifcb;
2336 ifcb = xfrm_if_get_cb();
2339 xi = ifcb->decode_session(skb);
2341 if_id = xi->p.if_id;
2345 reverse = dir & ~XFRM_POLICY_MASK;
2346 dir &= XFRM_POLICY_MASK;
2348 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2349 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2353 nf_nat_decode_session(skb, &fl, family);
2355 /* First, check used SA against their selectors. */
2359 for (i = skb->sp->len-1; i >= 0; i--) {
2360 struct xfrm_state *x = skb->sp->xvec[i];
2361 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2362 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2369 sk = sk_to_full_sk(sk);
2370 if (sk && sk->sk_policy[dir]) {
2371 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
2373 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2379 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
2382 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2387 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2388 xfrm_secpath_reject(xerr_idx, skb, &fl);
2389 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2395 pol->curlft.use_time = ktime_get_real_seconds();
2399 #ifdef CONFIG_XFRM_SUB_POLICY
2400 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2401 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2403 XFRM_POLICY_IN, if_id);
2405 if (IS_ERR(pols[1])) {
2406 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2409 pols[1]->curlft.use_time = ktime_get_real_seconds();
2415 if (pol->action == XFRM_POLICY_ALLOW) {
2416 struct sec_path *sp;
2417 static struct sec_path dummy;
2418 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2419 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2420 struct xfrm_tmpl **tpp = tp;
2424 if ((sp = skb->sp) == NULL)
2427 for (pi = 0; pi < npols; pi++) {
2428 if (pols[pi] != pol &&
2429 pols[pi]->action != XFRM_POLICY_ALLOW) {
2430 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2433 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2434 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2437 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2438 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2442 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2446 /* For each tunnel xfrm, find the first matching tmpl.
2447 * For each tmpl before that, find corresponding xfrm.
2448 * Order is _important_. Later we will implement
2449 * some barriers, but at the moment barriers
2450 * are implied between each two transformations.
2452 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2453 k = xfrm_policy_ok(tpp[i], sp, k, family);
2456 /* "-2 - errored_index" returned */
2458 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2463 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2464 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2468 xfrm_pols_put(pols, npols);
2471 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2474 xfrm_secpath_reject(xerr_idx, skb, &fl);
2476 xfrm_pols_put(pols, npols);
2479 EXPORT_SYMBOL(__xfrm_policy_check);
2481 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2483 struct net *net = dev_net(skb->dev);
2485 struct dst_entry *dst;
2488 if (xfrm_decode_session(skb, &fl, family) < 0) {
2489 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2494 if (!skb_dst(skb)) {
2495 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2499 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2504 skb_dst_set(skb, dst);
2507 EXPORT_SYMBOL(__xfrm_route_forward);
2509 /* Optimize later using cookies and generation ids. */
2511 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2513 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2514 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2515 * get validated by dst_ops->check on every use. We do this
2516 * because when a normal route referenced by an XFRM dst is
2517 * obsoleted we do not go looking around for all parent
2518 * referencing XFRM dsts so that we can invalidate them. It
2519 * is just too much work. Instead we make the checks here on
2520 * every use. For example:
2522 * XFRM dst A --> IPv4 dst X
2524 * X is the "xdst->route" of A (X is also the "dst->path" of A
2525 * in this example). If X is marked obsolete, "A" will not
2526 * notice. That's what we are validating here via the
2527 * stale_bundle() check.
2529 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2531 * This will force stale_bundle() to fail on any xdst bundle with
2532 * this dst linked in it.
2534 if (dst->obsolete < 0 && !stale_bundle(dst))
2540 static int stale_bundle(struct dst_entry *dst)
2542 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2545 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2547 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
2548 dst->dev = dev_net(dev)->loopback_dev;
2553 EXPORT_SYMBOL(xfrm_dst_ifdown);
2555 static void xfrm_link_failure(struct sk_buff *skb)
2557 /* Impossible. Such dst must be popped before reaches point of failure. */
2560 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2563 if (dst->obsolete) {
2571 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
2574 struct xfrm_dst *xdst = bundle[nr];
2575 u32 pmtu, route_mtu_cached;
2576 struct dst_entry *dst;
2579 pmtu = dst_mtu(xfrm_dst_child(dst));
2580 xdst->child_mtu_cached = pmtu;
2582 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2584 route_mtu_cached = dst_mtu(xdst->route);
2585 xdst->route_mtu_cached = route_mtu_cached;
2587 if (pmtu > route_mtu_cached)
2588 pmtu = route_mtu_cached;
2590 dst_metric_set(dst, RTAX_MTU, pmtu);
2594 /* Check that the bundle accepts the flow and its components are
2598 static int xfrm_bundle_ok(struct xfrm_dst *first)
2600 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2601 struct dst_entry *dst = &first->u.dst;
2602 struct xfrm_dst *xdst;
2606 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
2607 (dst->dev && !netif_running(dst->dev)))
2610 if (dst->flags & DST_XFRM_QUEUE)
2613 start_from = nr = 0;
2615 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2617 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2619 if (xdst->xfrm_genid != dst->xfrm->genid)
2621 if (xdst->num_pols > 0 &&
2622 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2625 bundle[nr++] = xdst;
2627 mtu = dst_mtu(xfrm_dst_child(dst));
2628 if (xdst->child_mtu_cached != mtu) {
2630 xdst->child_mtu_cached = mtu;
2633 if (!dst_check(xdst->route, xdst->route_cookie))
2635 mtu = dst_mtu(xdst->route);
2636 if (xdst->route_mtu_cached != mtu) {
2638 xdst->route_mtu_cached = mtu;
2641 dst = xfrm_dst_child(dst);
2642 } while (dst->xfrm);
2644 if (likely(!start_from))
2647 xdst = bundle[start_from - 1];
2648 mtu = xdst->child_mtu_cached;
2649 while (start_from--) {
2652 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2653 if (mtu > xdst->route_mtu_cached)
2654 mtu = xdst->route_mtu_cached;
2655 dst_metric_set(dst, RTAX_MTU, mtu);
2659 xdst = bundle[start_from - 1];
2660 xdst->child_mtu_cached = mtu;
2666 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2668 return dst_metric_advmss(xfrm_dst_path(dst));
2671 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2673 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2675 return mtu ? : dst_mtu(xfrm_dst_path(dst));
2678 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2682 const struct xfrm_state *xfrm = dst->xfrm;
2684 dst = xfrm_dst_child(dst);
2686 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2688 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2689 daddr = xfrm->coaddr;
2690 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2691 daddr = &xfrm->id.daddr;
2696 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2697 struct sk_buff *skb,
2700 const struct dst_entry *path = xfrm_dst_path(dst);
2703 daddr = xfrm_get_dst_nexthop(dst, daddr);
2704 return path->ops->neigh_lookup(path, skb, daddr);
2707 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2709 const struct dst_entry *path = xfrm_dst_path(dst);
2711 daddr = xfrm_get_dst_nexthop(dst, daddr);
2712 path->ops->confirm_neigh(path, daddr);
2715 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2719 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2720 return -EAFNOSUPPORT;
2722 spin_lock(&xfrm_policy_afinfo_lock);
2723 if (unlikely(xfrm_policy_afinfo[family] != NULL))
2726 struct dst_ops *dst_ops = afinfo->dst_ops;
2727 if (likely(dst_ops->kmem_cachep == NULL))
2728 dst_ops->kmem_cachep = xfrm_dst_cache;
2729 if (likely(dst_ops->check == NULL))
2730 dst_ops->check = xfrm_dst_check;
2731 if (likely(dst_ops->default_advmss == NULL))
2732 dst_ops->default_advmss = xfrm_default_advmss;
2733 if (likely(dst_ops->mtu == NULL))
2734 dst_ops->mtu = xfrm_mtu;
2735 if (likely(dst_ops->negative_advice == NULL))
2736 dst_ops->negative_advice = xfrm_negative_advice;
2737 if (likely(dst_ops->link_failure == NULL))
2738 dst_ops->link_failure = xfrm_link_failure;
2739 if (likely(dst_ops->neigh_lookup == NULL))
2740 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2741 if (likely(!dst_ops->confirm_neigh))
2742 dst_ops->confirm_neigh = xfrm_confirm_neigh;
2743 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2745 spin_unlock(&xfrm_policy_afinfo_lock);
2749 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2751 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2753 struct dst_ops *dst_ops = afinfo->dst_ops;
2756 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2757 if (xfrm_policy_afinfo[i] != afinfo)
2759 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2765 dst_ops->kmem_cachep = NULL;
2766 dst_ops->check = NULL;
2767 dst_ops->negative_advice = NULL;
2768 dst_ops->link_failure = NULL;
2770 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2772 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
2774 spin_lock(&xfrm_if_cb_lock);
2775 rcu_assign_pointer(xfrm_if_cb, ifcb);
2776 spin_unlock(&xfrm_if_cb_lock);
2778 EXPORT_SYMBOL(xfrm_if_register_cb);
2780 void xfrm_if_unregister_cb(void)
2782 RCU_INIT_POINTER(xfrm_if_cb, NULL);
2785 EXPORT_SYMBOL(xfrm_if_unregister_cb);
2787 #ifdef CONFIG_XFRM_STATISTICS
2788 static int __net_init xfrm_statistics_init(struct net *net)
2791 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2792 if (!net->mib.xfrm_statistics)
2794 rv = xfrm_proc_init(net);
2796 free_percpu(net->mib.xfrm_statistics);
2800 static void xfrm_statistics_fini(struct net *net)
2802 xfrm_proc_fini(net);
2803 free_percpu(net->mib.xfrm_statistics);
2806 static int __net_init xfrm_statistics_init(struct net *net)
2811 static void xfrm_statistics_fini(struct net *net)
2816 static int __net_init xfrm_policy_init(struct net *net)
2818 unsigned int hmask, sz;
2821 if (net_eq(net, &init_net))
2822 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2823 sizeof(struct xfrm_dst),
2824 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2828 sz = (hmask+1) * sizeof(struct hlist_head);
2830 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2831 if (!net->xfrm.policy_byidx)
2833 net->xfrm.policy_idx_hmask = hmask;
2835 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2836 struct xfrm_policy_hash *htab;
2838 net->xfrm.policy_count[dir] = 0;
2839 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2840 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2842 htab = &net->xfrm.policy_bydst[dir];
2843 htab->table = xfrm_hash_alloc(sz);
2846 htab->hmask = hmask;
2852 net->xfrm.policy_hthresh.lbits4 = 32;
2853 net->xfrm.policy_hthresh.rbits4 = 32;
2854 net->xfrm.policy_hthresh.lbits6 = 128;
2855 net->xfrm.policy_hthresh.rbits6 = 128;
2857 seqlock_init(&net->xfrm.policy_hthresh.lock);
2859 INIT_LIST_HEAD(&net->xfrm.policy_all);
2860 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2861 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2865 for (dir--; dir >= 0; dir--) {
2866 struct xfrm_policy_hash *htab;
2868 htab = &net->xfrm.policy_bydst[dir];
2869 xfrm_hash_free(htab->table, sz);
2871 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2876 static void xfrm_policy_fini(struct net *net)
2881 flush_work(&net->xfrm.policy_hash_work);
2882 #ifdef CONFIG_XFRM_SUB_POLICY
2883 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2885 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2887 WARN_ON(!list_empty(&net->xfrm.policy_all));
2889 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2890 struct xfrm_policy_hash *htab;
2892 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2894 htab = &net->xfrm.policy_bydst[dir];
2895 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2896 WARN_ON(!hlist_empty(htab->table));
2897 xfrm_hash_free(htab->table, sz);
2900 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2901 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2902 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2905 static int __net_init xfrm_net_init(struct net *net)
2909 /* Initialize the per-net locks here */
2910 spin_lock_init(&net->xfrm.xfrm_state_lock);
2911 spin_lock_init(&net->xfrm.xfrm_policy_lock);
2912 mutex_init(&net->xfrm.xfrm_cfg_mutex);
2914 rv = xfrm_statistics_init(net);
2916 goto out_statistics;
2917 rv = xfrm_state_init(net);
2920 rv = xfrm_policy_init(net);
2923 rv = xfrm_sysctl_init(net);
2930 xfrm_policy_fini(net);
2932 xfrm_state_fini(net);
2934 xfrm_statistics_fini(net);
2939 static void __net_exit xfrm_net_exit(struct net *net)
2941 xfrm_sysctl_fini(net);
2942 xfrm_policy_fini(net);
2943 xfrm_state_fini(net);
2944 xfrm_statistics_fini(net);
2947 static struct pernet_operations __net_initdata xfrm_net_ops = {
2948 .init = xfrm_net_init,
2949 .exit = xfrm_net_exit,
2952 void __init xfrm_init(void)
2954 register_pernet_subsys(&xfrm_net_ops);
2956 seqcount_init(&xfrm_policy_hash_generation);
2959 RCU_INIT_POINTER(xfrm_if_cb, NULL);
2963 #ifdef CONFIG_AUDITSYSCALL
2964 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2965 struct audit_buffer *audit_buf)
2967 struct xfrm_sec_ctx *ctx = xp->security;
2968 struct xfrm_selector *sel = &xp->selector;
2971 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2972 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2974 switch (sel->family) {
2976 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2977 if (sel->prefixlen_s != 32)
2978 audit_log_format(audit_buf, " src_prefixlen=%d",
2980 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2981 if (sel->prefixlen_d != 32)
2982 audit_log_format(audit_buf, " dst_prefixlen=%d",
2986 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2987 if (sel->prefixlen_s != 128)
2988 audit_log_format(audit_buf, " src_prefixlen=%d",
2990 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2991 if (sel->prefixlen_d != 128)
2992 audit_log_format(audit_buf, " dst_prefixlen=%d",
2998 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3000 struct audit_buffer *audit_buf;
3002 audit_buf = xfrm_audit_start("SPD-add");
3003 if (audit_buf == NULL)
3005 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3006 audit_log_format(audit_buf, " res=%u", result);
3007 xfrm_audit_common_policyinfo(xp, audit_buf);
3008 audit_log_end(audit_buf);
3010 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3012 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3015 struct audit_buffer *audit_buf;
3017 audit_buf = xfrm_audit_start("SPD-delete");
3018 if (audit_buf == NULL)
3020 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3021 audit_log_format(audit_buf, " res=%u", result);
3022 xfrm_audit_common_policyinfo(xp, audit_buf);
3023 audit_log_end(audit_buf);
3025 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3028 #ifdef CONFIG_XFRM_MIGRATE
3029 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3030 const struct xfrm_selector *sel_tgt)
3032 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3033 if (sel_tgt->family == sel_cmp->family &&
3034 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3036 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3038 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3039 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3043 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3050 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3051 u8 dir, u8 type, struct net *net)
3053 struct xfrm_policy *pol, *ret = NULL;
3054 struct hlist_head *chain;
3057 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3058 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3059 hlist_for_each_entry(pol, chain, bydst) {
3060 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3061 pol->type == type) {
3063 priority = ret->priority;
3067 chain = &net->xfrm.policy_inexact[dir];
3068 hlist_for_each_entry(pol, chain, bydst) {
3069 if ((pol->priority >= priority) && ret)
3072 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3073 pol->type == type) {
3081 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3086 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3090 if (t->mode == m->mode && t->id.proto == m->proto &&
3091 (m->reqid == 0 || t->reqid == m->reqid)) {
3093 case XFRM_MODE_TUNNEL:
3094 case XFRM_MODE_BEET:
3095 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3097 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3102 case XFRM_MODE_TRANSPORT:
3103 /* in case of transport mode, template does not store
3104 any IP addresses, hence we just compare mode and
3115 /* update endpoint address(es) of template(s) */
3116 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3117 struct xfrm_migrate *m, int num_migrate)
3119 struct xfrm_migrate *mp;
3122 write_lock_bh(&pol->lock);
3123 if (unlikely(pol->walk.dead)) {
3124 /* target policy has been deleted */
3125 write_unlock_bh(&pol->lock);
3129 for (i = 0; i < pol->xfrm_nr; i++) {
3130 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3131 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3134 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3135 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3137 /* update endpoints */
3138 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3139 sizeof(pol->xfrm_vec[i].id.daddr));
3140 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3141 sizeof(pol->xfrm_vec[i].saddr));
3142 pol->xfrm_vec[i].encap_family = mp->new_family;
3144 atomic_inc(&pol->genid);
3148 write_unlock_bh(&pol->lock);
3156 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3160 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3163 for (i = 0; i < num_migrate; i++) {
3164 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3165 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3168 /* check if there is any duplicated entry */
3169 for (j = i + 1; j < num_migrate; j++) {
3170 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3171 sizeof(m[i].old_daddr)) &&
3172 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3173 sizeof(m[i].old_saddr)) &&
3174 m[i].proto == m[j].proto &&
3175 m[i].mode == m[j].mode &&
3176 m[i].reqid == m[j].reqid &&
3177 m[i].old_family == m[j].old_family)
3185 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3186 struct xfrm_migrate *m, int num_migrate,
3187 struct xfrm_kmaddress *k, struct net *net,
3188 struct xfrm_encap_tmpl *encap)
3190 int i, err, nx_cur = 0, nx_new = 0;
3191 struct xfrm_policy *pol = NULL;
3192 struct xfrm_state *x, *xc;
3193 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3194 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3195 struct xfrm_migrate *mp;
3197 /* Stage 0 - sanity checks */
3198 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3201 if (dir >= XFRM_POLICY_MAX) {
3206 /* Stage 1 - find policy */
3207 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3212 /* Stage 2 - find and update state(s) */
3213 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3214 if ((x = xfrm_migrate_state_find(mp, net))) {
3217 xc = xfrm_state_migrate(x, mp, encap);
3228 /* Stage 3 - update policy */
3229 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3232 /* Stage 4 - delete old state(s) */
3234 xfrm_states_put(x_cur, nx_cur);
3235 xfrm_states_delete(x_cur, nx_cur);
3238 /* Stage 5 - announce */
3239 km_migrate(sel, dir, type, m, num_migrate, k, encap);
3251 xfrm_states_put(x_cur, nx_cur);
3253 xfrm_states_delete(x_new, nx_new);
3257 EXPORT_SYMBOL(xfrm_migrate);