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_PER_CPU(struct xfrm_dst *, xfrm_last_dst);
49 static struct work_struct *xfrm_pcpu_work __read_mostly;
50 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
51 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
54 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
55 static __read_mostly seqcount_t xfrm_policy_hash_generation;
57 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
58 static int stale_bundle(struct dst_entry *dst);
59 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
60 static void xfrm_policy_queue_process(struct timer_list *t);
62 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
63 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
66 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
68 return refcount_inc_not_zero(&policy->refcnt);
72 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
74 const struct flowi4 *fl4 = &fl->u.ip4;
76 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
77 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
78 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
79 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
80 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
81 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
85 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
87 const struct flowi6 *fl6 = &fl->u.ip6;
89 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
90 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
91 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
92 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
93 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
94 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
97 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
98 unsigned short family)
102 return __xfrm4_selector_match(sel, fl);
104 return __xfrm6_selector_match(sel, fl);
109 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
111 const struct xfrm_policy_afinfo *afinfo;
113 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
116 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
117 if (unlikely(!afinfo))
122 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
123 const xfrm_address_t *saddr,
124 const xfrm_address_t *daddr,
125 int family, u32 mark)
127 const struct xfrm_policy_afinfo *afinfo;
128 struct dst_entry *dst;
130 afinfo = xfrm_policy_get_afinfo(family);
131 if (unlikely(afinfo == NULL))
132 return ERR_PTR(-EAFNOSUPPORT);
134 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
140 EXPORT_SYMBOL(__xfrm_dst_lookup);
142 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
144 xfrm_address_t *prev_saddr,
145 xfrm_address_t *prev_daddr,
146 int family, u32 mark)
148 struct net *net = xs_net(x);
149 xfrm_address_t *saddr = &x->props.saddr;
150 xfrm_address_t *daddr = &x->id.daddr;
151 struct dst_entry *dst;
153 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
157 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
162 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
165 if (prev_saddr != saddr)
166 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
167 if (prev_daddr != daddr)
168 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
174 static inline unsigned long make_jiffies(long secs)
176 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
177 return MAX_SCHEDULE_TIMEOUT-1;
182 static void xfrm_policy_timer(struct timer_list *t)
184 struct xfrm_policy *xp = from_timer(xp, t, timer);
185 unsigned long now = get_seconds();
186 long next = LONG_MAX;
190 read_lock(&xp->lock);
192 if (unlikely(xp->walk.dead))
195 dir = xfrm_policy_id2dir(xp->index);
197 if (xp->lft.hard_add_expires_seconds) {
198 long tmo = xp->lft.hard_add_expires_seconds +
199 xp->curlft.add_time - now;
205 if (xp->lft.hard_use_expires_seconds) {
206 long tmo = xp->lft.hard_use_expires_seconds +
207 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
213 if (xp->lft.soft_add_expires_seconds) {
214 long tmo = xp->lft.soft_add_expires_seconds +
215 xp->curlft.add_time - now;
218 tmo = XFRM_KM_TIMEOUT;
223 if (xp->lft.soft_use_expires_seconds) {
224 long tmo = xp->lft.soft_use_expires_seconds +
225 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
228 tmo = XFRM_KM_TIMEOUT;
235 km_policy_expired(xp, dir, 0, 0);
236 if (next != LONG_MAX &&
237 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
241 read_unlock(&xp->lock);
246 read_unlock(&xp->lock);
247 if (!xfrm_policy_delete(xp, dir))
248 km_policy_expired(xp, dir, 1, 0);
252 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
256 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
258 struct xfrm_policy *policy;
260 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
263 write_pnet(&policy->xp_net, net);
264 INIT_LIST_HEAD(&policy->walk.all);
265 INIT_HLIST_NODE(&policy->bydst);
266 INIT_HLIST_NODE(&policy->byidx);
267 rwlock_init(&policy->lock);
268 refcount_set(&policy->refcnt, 1);
269 skb_queue_head_init(&policy->polq.hold_queue);
270 timer_setup(&policy->timer, xfrm_policy_timer, 0);
271 timer_setup(&policy->polq.hold_timer,
272 xfrm_policy_queue_process, 0);
276 EXPORT_SYMBOL(xfrm_policy_alloc);
278 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
280 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
282 security_xfrm_policy_free(policy->security);
286 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
288 void xfrm_policy_destroy(struct xfrm_policy *policy)
290 BUG_ON(!policy->walk.dead);
292 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
295 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
297 EXPORT_SYMBOL(xfrm_policy_destroy);
299 /* Rule must be locked. Release descendant resources, announce
300 * entry dead. The rule must be unlinked from lists to the moment.
303 static void xfrm_policy_kill(struct xfrm_policy *policy)
305 policy->walk.dead = 1;
307 atomic_inc(&policy->genid);
309 if (del_timer(&policy->polq.hold_timer))
310 xfrm_pol_put(policy);
311 skb_queue_purge(&policy->polq.hold_queue);
313 if (del_timer(&policy->timer))
314 xfrm_pol_put(policy);
316 xfrm_pol_put(policy);
319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
321 static inline unsigned int idx_hash(struct net *net, u32 index)
323 return __idx_hash(index, net->xfrm.policy_idx_hmask);
326 /* calculate policy hash thresholds */
327 static void __get_hash_thresh(struct net *net,
328 unsigned short family, int dir,
329 u8 *dbits, u8 *sbits)
333 *dbits = net->xfrm.policy_bydst[dir].dbits4;
334 *sbits = net->xfrm.policy_bydst[dir].sbits4;
338 *dbits = net->xfrm.policy_bydst[dir].dbits6;
339 *sbits = net->xfrm.policy_bydst[dir].sbits6;
348 static struct hlist_head *policy_hash_bysel(struct net *net,
349 const struct xfrm_selector *sel,
350 unsigned short family, int dir)
352 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
357 __get_hash_thresh(net, family, dir, &dbits, &sbits);
358 hash = __sel_hash(sel, family, hmask, dbits, sbits);
360 if (hash == hmask + 1)
361 return &net->xfrm.policy_inexact[dir];
363 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
364 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
367 static struct hlist_head *policy_hash_direct(struct net *net,
368 const xfrm_address_t *daddr,
369 const xfrm_address_t *saddr,
370 unsigned short family, int dir)
372 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
377 __get_hash_thresh(net, family, dir, &dbits, &sbits);
378 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
380 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
381 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
384 static void xfrm_dst_hash_transfer(struct net *net,
385 struct hlist_head *list,
386 struct hlist_head *ndsttable,
387 unsigned int nhashmask,
390 struct hlist_node *tmp, *entry0 = NULL;
391 struct xfrm_policy *pol;
397 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
400 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
401 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
402 pol->family, nhashmask, dbits, sbits);
404 hlist_del_rcu(&pol->bydst);
405 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
410 hlist_del_rcu(&pol->bydst);
411 hlist_add_behind_rcu(&pol->bydst, entry0);
413 entry0 = &pol->bydst;
415 if (!hlist_empty(list)) {
421 static void xfrm_idx_hash_transfer(struct hlist_head *list,
422 struct hlist_head *nidxtable,
423 unsigned int nhashmask)
425 struct hlist_node *tmp;
426 struct xfrm_policy *pol;
428 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
431 h = __idx_hash(pol->index, nhashmask);
432 hlist_add_head(&pol->byidx, nidxtable+h);
436 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
438 return ((old_hmask + 1) << 1) - 1;
441 static void xfrm_bydst_resize(struct net *net, int dir)
443 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
444 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
445 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
446 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
447 struct hlist_head *odst;
453 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
454 write_seqcount_begin(&xfrm_policy_hash_generation);
456 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
457 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
459 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
460 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
462 for (i = hmask; i >= 0; i--)
463 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
465 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
466 net->xfrm.policy_bydst[dir].hmask = nhashmask;
468 write_seqcount_end(&xfrm_policy_hash_generation);
469 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
473 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
476 static void xfrm_byidx_resize(struct net *net, int total)
478 unsigned int hmask = net->xfrm.policy_idx_hmask;
479 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
480 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
481 struct hlist_head *oidx = net->xfrm.policy_byidx;
482 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
488 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
490 for (i = hmask; i >= 0; i--)
491 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
493 net->xfrm.policy_byidx = nidx;
494 net->xfrm.policy_idx_hmask = nhashmask;
496 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
498 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
501 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
503 unsigned int cnt = net->xfrm.policy_count[dir];
504 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
509 if ((hmask + 1) < xfrm_policy_hashmax &&
516 static inline int xfrm_byidx_should_resize(struct net *net, int total)
518 unsigned int hmask = net->xfrm.policy_idx_hmask;
520 if ((hmask + 1) < xfrm_policy_hashmax &&
527 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
529 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
530 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
531 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
532 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
533 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
534 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
535 si->spdhcnt = net->xfrm.policy_idx_hmask;
536 si->spdhmcnt = xfrm_policy_hashmax;
538 EXPORT_SYMBOL(xfrm_spd_getinfo);
540 static DEFINE_MUTEX(hash_resize_mutex);
541 static void xfrm_hash_resize(struct work_struct *work)
543 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
546 mutex_lock(&hash_resize_mutex);
549 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
550 if (xfrm_bydst_should_resize(net, dir, &total))
551 xfrm_bydst_resize(net, dir);
553 if (xfrm_byidx_should_resize(net, total))
554 xfrm_byidx_resize(net, total);
556 mutex_unlock(&hash_resize_mutex);
559 static void xfrm_hash_rebuild(struct work_struct *work)
561 struct net *net = container_of(work, struct net,
562 xfrm.policy_hthresh.work);
564 struct xfrm_policy *pol;
565 struct xfrm_policy *policy;
566 struct hlist_head *chain;
567 struct hlist_head *odst;
568 struct hlist_node *newpos;
572 u8 lbits4, rbits4, lbits6, rbits6;
574 mutex_lock(&hash_resize_mutex);
576 /* read selector prefixlen thresholds */
578 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
580 lbits4 = net->xfrm.policy_hthresh.lbits4;
581 rbits4 = net->xfrm.policy_hthresh.rbits4;
582 lbits6 = net->xfrm.policy_hthresh.lbits6;
583 rbits6 = net->xfrm.policy_hthresh.rbits6;
584 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
586 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
588 /* reset the bydst and inexact table in all directions */
589 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
590 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
591 hmask = net->xfrm.policy_bydst[dir].hmask;
592 odst = net->xfrm.policy_bydst[dir].table;
593 for (i = hmask; i >= 0; i--)
594 INIT_HLIST_HEAD(odst + i);
595 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
596 /* dir out => dst = remote, src = local */
597 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
598 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
599 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
600 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
602 /* dir in/fwd => dst = local, src = remote */
603 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
604 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
605 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
606 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
610 /* re-insert all policies by order of creation */
611 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
612 if (policy->walk.dead ||
613 xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
614 /* skip socket policies */
618 chain = policy_hash_bysel(net, &policy->selector,
620 xfrm_policy_id2dir(policy->index));
621 hlist_for_each_entry(pol, chain, bydst) {
622 if (policy->priority >= pol->priority)
623 newpos = &pol->bydst;
628 hlist_add_behind(&policy->bydst, newpos);
630 hlist_add_head(&policy->bydst, chain);
633 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
635 mutex_unlock(&hash_resize_mutex);
638 void xfrm_policy_hash_rebuild(struct net *net)
640 schedule_work(&net->xfrm.policy_hthresh.work);
642 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
644 /* Generate new index... KAME seems to generate them ordered by cost
645 * of an absolute inpredictability of ordering of rules. This will not pass. */
646 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
648 static u32 idx_generator;
651 struct hlist_head *list;
652 struct xfrm_policy *p;
657 idx = (idx_generator | dir);
666 list = net->xfrm.policy_byidx + idx_hash(net, idx);
668 hlist_for_each_entry(p, list, byidx) {
669 if (p->index == idx) {
679 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
681 u32 *p1 = (u32 *) s1;
682 u32 *p2 = (u32 *) s2;
683 int len = sizeof(struct xfrm_selector) / sizeof(u32);
686 for (i = 0; i < len; i++) {
694 static void xfrm_policy_requeue(struct xfrm_policy *old,
695 struct xfrm_policy *new)
697 struct xfrm_policy_queue *pq = &old->polq;
698 struct sk_buff_head list;
700 if (skb_queue_empty(&pq->hold_queue))
703 __skb_queue_head_init(&list);
705 spin_lock_bh(&pq->hold_queue.lock);
706 skb_queue_splice_init(&pq->hold_queue, &list);
707 if (del_timer(&pq->hold_timer))
709 spin_unlock_bh(&pq->hold_queue.lock);
713 spin_lock_bh(&pq->hold_queue.lock);
714 skb_queue_splice(&list, &pq->hold_queue);
715 pq->timeout = XFRM_QUEUE_TMO_MIN;
716 if (!mod_timer(&pq->hold_timer, jiffies))
718 spin_unlock_bh(&pq->hold_queue.lock);
721 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
722 struct xfrm_policy *pol)
724 u32 mark = policy->mark.v & policy->mark.m;
726 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
729 if ((mark & pol->mark.m) == pol->mark.v &&
730 policy->priority == pol->priority)
736 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
738 struct net *net = xp_net(policy);
739 struct xfrm_policy *pol;
740 struct xfrm_policy *delpol;
741 struct hlist_head *chain;
742 struct hlist_node *newpos;
744 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
745 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
748 hlist_for_each_entry(pol, chain, bydst) {
749 if (pol->type == policy->type &&
750 !selector_cmp(&pol->selector, &policy->selector) &&
751 xfrm_policy_mark_match(policy, pol) &&
752 xfrm_sec_ctx_match(pol->security, policy->security) &&
755 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
759 if (policy->priority > pol->priority)
761 } else if (policy->priority >= pol->priority) {
762 newpos = &pol->bydst;
769 hlist_add_behind(&policy->bydst, newpos);
771 hlist_add_head(&policy->bydst, chain);
772 __xfrm_policy_link(policy, dir);
774 /* After previous checking, family can either be AF_INET or AF_INET6 */
775 if (policy->family == AF_INET)
776 rt_genid_bump_ipv4(net);
778 rt_genid_bump_ipv6(net);
781 xfrm_policy_requeue(delpol, policy);
782 __xfrm_policy_unlink(delpol, dir);
784 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
785 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
786 policy->curlft.add_time = get_seconds();
787 policy->curlft.use_time = 0;
788 if (!mod_timer(&policy->timer, jiffies + HZ))
789 xfrm_pol_hold(policy);
790 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
793 xfrm_policy_kill(delpol);
794 else if (xfrm_bydst_should_resize(net, dir, NULL))
795 schedule_work(&net->xfrm.policy_hash_work);
799 EXPORT_SYMBOL(xfrm_policy_insert);
801 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
802 int dir, struct xfrm_selector *sel,
803 struct xfrm_sec_ctx *ctx, int delete,
806 struct xfrm_policy *pol, *ret;
807 struct hlist_head *chain;
810 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
811 chain = policy_hash_bysel(net, sel, sel->family, dir);
813 hlist_for_each_entry(pol, chain, bydst) {
814 if (pol->type == type &&
815 (mark & pol->mark.m) == pol->mark.v &&
816 !selector_cmp(sel, &pol->selector) &&
817 xfrm_sec_ctx_match(ctx, pol->security)) {
820 *err = security_xfrm_policy_delete(
823 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
826 __xfrm_policy_unlink(pol, dir);
832 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
835 xfrm_policy_kill(ret);
838 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
840 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
841 int dir, u32 id, int delete, int *err)
843 struct xfrm_policy *pol, *ret;
844 struct hlist_head *chain;
847 if (xfrm_policy_id2dir(id) != dir)
851 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
852 chain = net->xfrm.policy_byidx + idx_hash(net, id);
854 hlist_for_each_entry(pol, chain, byidx) {
855 if (pol->type == type && pol->index == id &&
856 (mark & pol->mark.m) == pol->mark.v) {
859 *err = security_xfrm_policy_delete(
862 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
865 __xfrm_policy_unlink(pol, dir);
871 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
874 xfrm_policy_kill(ret);
877 EXPORT_SYMBOL(xfrm_policy_byid);
879 #ifdef CONFIG_SECURITY_NETWORK_XFRM
881 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
885 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
886 struct xfrm_policy *pol;
889 hlist_for_each_entry(pol,
890 &net->xfrm.policy_inexact[dir], bydst) {
891 if (pol->type != type)
893 err = security_xfrm_policy_delete(pol->security);
895 xfrm_audit_policy_delete(pol, 0, task_valid);
899 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
900 hlist_for_each_entry(pol,
901 net->xfrm.policy_bydst[dir].table + i,
903 if (pol->type != type)
905 err = security_xfrm_policy_delete(
908 xfrm_audit_policy_delete(pol, 0,
919 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
925 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
927 int dir, err = 0, cnt = 0;
929 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
931 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
935 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
936 struct xfrm_policy *pol;
940 hlist_for_each_entry(pol,
941 &net->xfrm.policy_inexact[dir], bydst) {
942 if (pol->type != type)
944 __xfrm_policy_unlink(pol, dir);
945 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
948 xfrm_audit_policy_delete(pol, 1, task_valid);
950 xfrm_policy_kill(pol);
952 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
956 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
958 hlist_for_each_entry(pol,
959 net->xfrm.policy_bydst[dir].table + i,
961 if (pol->type != type)
963 __xfrm_policy_unlink(pol, dir);
964 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
967 xfrm_audit_policy_delete(pol, 1, task_valid);
968 xfrm_policy_kill(pol);
970 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
979 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
982 EXPORT_SYMBOL(xfrm_policy_flush);
984 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
985 int (*func)(struct xfrm_policy *, int, int, void*),
988 struct xfrm_policy *pol;
989 struct xfrm_policy_walk_entry *x;
992 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
993 walk->type != XFRM_POLICY_TYPE_ANY)
996 if (list_empty(&walk->walk.all) && walk->seq != 0)
999 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1000 if (list_empty(&walk->walk.all))
1001 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1003 x = list_first_entry(&walk->walk.all,
1004 struct xfrm_policy_walk_entry, all);
1006 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1009 pol = container_of(x, struct xfrm_policy, walk);
1010 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1011 walk->type != pol->type)
1013 error = func(pol, xfrm_policy_id2dir(pol->index),
1016 list_move_tail(&walk->walk.all, &x->all);
1021 if (walk->seq == 0) {
1025 list_del_init(&walk->walk.all);
1027 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1030 EXPORT_SYMBOL(xfrm_policy_walk);
1032 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1034 INIT_LIST_HEAD(&walk->walk.all);
1035 walk->walk.dead = 1;
1039 EXPORT_SYMBOL(xfrm_policy_walk_init);
1041 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1043 if (list_empty(&walk->walk.all))
1046 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1047 list_del(&walk->walk.all);
1048 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1050 EXPORT_SYMBOL(xfrm_policy_walk_done);
1053 * Find policy to apply to this flow.
1055 * Returns 0 if policy found, else an -errno.
1057 static int xfrm_policy_match(const struct xfrm_policy *pol,
1058 const struct flowi *fl,
1059 u8 type, u16 family, int dir)
1061 const struct xfrm_selector *sel = &pol->selector;
1065 if (pol->family != family ||
1066 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1070 match = xfrm_selector_match(sel, fl, family);
1072 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1078 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1079 const struct flowi *fl,
1083 struct xfrm_policy *pol, *ret;
1084 const xfrm_address_t *daddr, *saddr;
1085 struct hlist_head *chain;
1086 unsigned int sequence;
1089 daddr = xfrm_flowi_daddr(fl, family);
1090 saddr = xfrm_flowi_saddr(fl, family);
1091 if (unlikely(!daddr || !saddr))
1097 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1098 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1099 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1103 hlist_for_each_entry_rcu(pol, chain, bydst) {
1104 err = xfrm_policy_match(pol, fl, type, family, dir);
1114 priority = ret->priority;
1118 chain = &net->xfrm.policy_inexact[dir];
1119 hlist_for_each_entry_rcu(pol, chain, bydst) {
1120 if ((pol->priority >= priority) && ret)
1123 err = xfrm_policy_match(pol, fl, type, family, dir);
1137 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1140 if (ret && !xfrm_pol_hold_rcu(ret))
1148 static struct xfrm_policy *
1149 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1151 #ifdef CONFIG_XFRM_SUB_POLICY
1152 struct xfrm_policy *pol;
1154 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1158 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1161 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1162 const struct flowi *fl, u16 family)
1164 struct xfrm_policy *pol;
1168 pol = rcu_dereference(sk->sk_policy[dir]);
1173 if (pol->family != family) {
1178 match = xfrm_selector_match(&pol->selector, fl, family);
1180 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1184 err = security_xfrm_policy_lookup(pol->security,
1188 if (!xfrm_pol_hold_rcu(pol))
1190 } else if (err == -ESRCH) {
1203 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1205 struct net *net = xp_net(pol);
1207 list_add(&pol->walk.all, &net->xfrm.policy_all);
1208 net->xfrm.policy_count[dir]++;
1212 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1215 struct net *net = xp_net(pol);
1217 if (list_empty(&pol->walk.all))
1220 /* Socket policies are not hashed. */
1221 if (!hlist_unhashed(&pol->bydst)) {
1222 hlist_del_rcu(&pol->bydst);
1223 hlist_del(&pol->byidx);
1226 list_del_init(&pol->walk.all);
1227 net->xfrm.policy_count[dir]--;
1232 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1234 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1237 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1239 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1242 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1244 struct net *net = xp_net(pol);
1246 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1247 pol = __xfrm_policy_unlink(pol, dir);
1248 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1250 xfrm_policy_kill(pol);
1255 EXPORT_SYMBOL(xfrm_policy_delete);
1257 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1259 struct net *net = sock_net(sk);
1260 struct xfrm_policy *old_pol;
1262 #ifdef CONFIG_XFRM_SUB_POLICY
1263 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1267 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1268 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1269 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1271 pol->curlft.add_time = get_seconds();
1272 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1273 xfrm_sk_policy_link(pol, dir);
1275 rcu_assign_pointer(sk->sk_policy[dir], pol);
1278 xfrm_policy_requeue(old_pol, pol);
1280 /* Unlinking succeeds always. This is the only function
1281 * allowed to delete or replace socket policy.
1283 xfrm_sk_policy_unlink(old_pol, dir);
1285 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1288 xfrm_policy_kill(old_pol);
1293 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1295 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1296 struct net *net = xp_net(old);
1299 newp->selector = old->selector;
1300 if (security_xfrm_policy_clone(old->security,
1303 return NULL; /* ENOMEM */
1305 newp->lft = old->lft;
1306 newp->curlft = old->curlft;
1307 newp->mark = old->mark;
1308 newp->action = old->action;
1309 newp->flags = old->flags;
1310 newp->xfrm_nr = old->xfrm_nr;
1311 newp->index = old->index;
1312 newp->type = old->type;
1313 newp->family = old->family;
1314 memcpy(newp->xfrm_vec, old->xfrm_vec,
1315 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1316 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1317 xfrm_sk_policy_link(newp, dir);
1318 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1324 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1326 const struct xfrm_policy *p;
1327 struct xfrm_policy *np;
1331 for (i = 0; i < 2; i++) {
1332 p = rcu_dereference(osk->sk_policy[i]);
1334 np = clone_policy(p, i);
1335 if (unlikely(!np)) {
1339 rcu_assign_pointer(sk->sk_policy[i], np);
1347 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1348 xfrm_address_t *remote, unsigned short family, u32 mark)
1351 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1353 if (unlikely(afinfo == NULL))
1355 err = afinfo->get_saddr(net, oif, local, remote, mark);
1360 /* Resolve list of templates for the flow, given policy. */
1363 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1364 struct xfrm_state **xfrm, unsigned short family)
1366 struct net *net = xp_net(policy);
1369 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1370 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1373 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1374 struct xfrm_state *x;
1375 xfrm_address_t *remote = daddr;
1376 xfrm_address_t *local = saddr;
1377 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1379 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1380 tmpl->mode == XFRM_MODE_BEET) {
1381 remote = &tmpl->id.daddr;
1382 local = &tmpl->saddr;
1383 if (xfrm_addr_any(local, tmpl->encap_family)) {
1384 error = xfrm_get_saddr(net, fl->flowi_oif,
1386 tmpl->encap_family, 0);
1393 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1395 if (x && x->km.state == XFRM_STATE_VALID) {
1402 error = (x->km.state == XFRM_STATE_ERROR ?
1405 } else if (error == -ESRCH) {
1409 if (!tmpl->optional)
1415 for (nx--; nx >= 0; nx--)
1416 xfrm_state_put(xfrm[nx]);
1421 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1422 struct xfrm_state **xfrm, unsigned short family)
1424 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1425 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1431 for (i = 0; i < npols; i++) {
1432 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1437 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1445 /* found states are sorted for outbound processing */
1447 xfrm_state_sort(xfrm, tpp, cnx, family);
1452 for (cnx--; cnx >= 0; cnx--)
1453 xfrm_state_put(tpp[cnx]);
1458 static int xfrm_get_tos(const struct flowi *fl, int family)
1460 const struct xfrm_policy_afinfo *afinfo;
1463 afinfo = xfrm_policy_get_afinfo(family);
1467 tos = afinfo->get_tos(fl);
1474 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1476 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1477 struct dst_ops *dst_ops;
1478 struct xfrm_dst *xdst;
1481 return ERR_PTR(-EINVAL);
1485 dst_ops = &net->xfrm.xfrm4_dst_ops;
1487 #if IS_ENABLED(CONFIG_IPV6)
1489 dst_ops = &net->xfrm.xfrm6_dst_ops;
1495 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1498 struct dst_entry *dst = &xdst->u.dst;
1500 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1502 xdst = ERR_PTR(-ENOBUFS);
1509 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1512 const struct xfrm_policy_afinfo *afinfo =
1513 xfrm_policy_get_afinfo(dst->ops->family);
1519 err = afinfo->init_path(path, dst, nfheader_len);
1526 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1527 const struct flowi *fl)
1529 const struct xfrm_policy_afinfo *afinfo =
1530 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1536 err = afinfo->fill_dst(xdst, dev, fl);
1544 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1545 * all the metrics... Shortly, bundle a bundle.
1548 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1549 struct xfrm_state **xfrm,
1550 struct xfrm_dst **bundle,
1552 const struct flowi *fl,
1553 struct dst_entry *dst)
1555 struct net *net = xp_net(policy);
1556 unsigned long now = jiffies;
1557 struct net_device *dev;
1558 struct xfrm_mode *inner_mode;
1559 struct xfrm_dst *xdst_prev = NULL;
1560 struct xfrm_dst *xdst0 = NULL;
1564 int nfheader_len = 0;
1565 int trailer_len = 0;
1567 int family = policy->selector.family;
1568 xfrm_address_t saddr, daddr;
1570 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1572 tos = xfrm_get_tos(fl, family);
1576 for (; i < nx; i++) {
1577 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1578 struct dst_entry *dst1 = &xdst->u.dst;
1580 err = PTR_ERR(xdst);
1590 /* Ref count is taken during xfrm_alloc_dst()
1591 * No need to do dst_clone() on dst1
1593 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
1595 if (xfrm[i]->sel.family == AF_UNSPEC) {
1596 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1597 xfrm_af2proto(family));
1599 err = -EAFNOSUPPORT;
1604 inner_mode = xfrm[i]->inner_mode;
1607 dst_copy_metrics(dst1, dst);
1609 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1610 family = xfrm[i]->props.family;
1611 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1612 &saddr, &daddr, family,
1613 xfrm[i]->props.output_mark);
1620 dst1->xfrm = xfrm[i];
1621 xdst->xfrm_genid = xfrm[i]->genid;
1623 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1624 dst1->flags |= DST_HOST;
1625 dst1->lastuse = now;
1627 dst1->input = dst_discard;
1628 dst1->output = inner_mode->afinfo->output;
1632 header_len += xfrm[i]->props.header_len;
1633 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1634 nfheader_len += xfrm[i]->props.header_len;
1635 trailer_len += xfrm[i]->props.trailer_len;
1638 xfrm_dst_set_child(xdst_prev, dst);
1646 xfrm_init_path(xdst0, dst, nfheader_len);
1647 xfrm_init_pmtu(bundle, nx);
1649 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
1650 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
1651 err = xfrm_fill_dst(xdst_prev, dev, fl);
1655 xdst_prev->u.dst.header_len = header_len;
1656 xdst_prev->u.dst.trailer_len = trailer_len;
1657 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
1658 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
1662 return &xdst0->u.dst;
1666 xfrm_state_put(xfrm[i]);
1669 dst_release_immediate(&xdst0->u.dst);
1670 xdst0 = ERR_PTR(err);
1674 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1675 struct xfrm_policy **pols,
1676 int *num_pols, int *num_xfrms)
1680 if (*num_pols == 0 || !pols[0]) {
1685 if (IS_ERR(pols[0]))
1686 return PTR_ERR(pols[0]);
1688 *num_xfrms = pols[0]->xfrm_nr;
1690 #ifdef CONFIG_XFRM_SUB_POLICY
1691 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1692 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1693 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1694 XFRM_POLICY_TYPE_MAIN,
1698 if (IS_ERR(pols[1])) {
1699 xfrm_pols_put(pols, *num_pols);
1700 return PTR_ERR(pols[1]);
1703 (*num_xfrms) += pols[1]->xfrm_nr;
1707 for (i = 0; i < *num_pols; i++) {
1708 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1718 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1720 this_cpu_write(xfrm_last_dst, xdst);
1722 dst_release(&old->u.dst);
1725 static void __xfrm_pcpu_work_fn(void)
1727 struct xfrm_dst *old;
1729 old = this_cpu_read(xfrm_last_dst);
1730 if (old && !xfrm_bundle_ok(old))
1731 xfrm_last_dst_update(NULL, old);
1734 static void xfrm_pcpu_work_fn(struct work_struct *work)
1738 __xfrm_pcpu_work_fn();
1743 void xfrm_policy_cache_flush(void)
1745 struct xfrm_dst *old;
1753 for_each_possible_cpu(cpu) {
1754 old = per_cpu(xfrm_last_dst, cpu);
1755 if (old && !xfrm_bundle_ok(old)) {
1756 if (smp_processor_id() == cpu) {
1757 __xfrm_pcpu_work_fn();
1773 for_each_possible_cpu(cpu) {
1774 bool bundle_release;
1777 old = per_cpu(xfrm_last_dst, cpu);
1778 bundle_release = old && !xfrm_bundle_ok(old);
1781 if (!bundle_release)
1784 if (cpu_online(cpu)) {
1785 schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1790 old = per_cpu(xfrm_last_dst, cpu);
1791 if (old && !xfrm_bundle_ok(old)) {
1792 per_cpu(xfrm_last_dst, cpu) = NULL;
1793 dst_release(&old->u.dst);
1801 static bool xfrm_xdst_can_reuse(struct xfrm_dst *xdst,
1802 struct xfrm_state * const xfrm[],
1805 const struct dst_entry *dst = &xdst->u.dst;
1808 if (xdst->num_xfrms != num)
1811 for (i = 0; i < num; i++) {
1812 if (!dst || dst->xfrm != xfrm[i])
1814 dst = xfrm_dst_child(dst);
1817 return xfrm_bundle_ok(xdst);
1820 static struct xfrm_dst *
1821 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1822 const struct flowi *fl, u16 family,
1823 struct dst_entry *dst_orig)
1825 struct net *net = xp_net(pols[0]);
1826 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1827 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
1828 struct xfrm_dst *xdst, *old;
1829 struct dst_entry *dst;
1832 /* Try to instantiate a bundle */
1833 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1835 if (err != 0 && err != -EAGAIN)
1836 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1837 return ERR_PTR(err);
1840 xdst = this_cpu_read(xfrm_last_dst);
1842 xdst->u.dst.dev == dst_orig->dev &&
1843 xdst->num_pols == num_pols &&
1844 memcmp(xdst->pols, pols,
1845 sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1846 xfrm_xdst_can_reuse(xdst, xfrm, err)) {
1847 dst_hold(&xdst->u.dst);
1848 xfrm_pols_put(pols, num_pols);
1850 xfrm_state_put(xfrm[--err]);
1856 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
1858 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1859 return ERR_CAST(dst);
1862 xdst = (struct xfrm_dst *)dst;
1863 xdst->num_xfrms = err;
1864 xdst->num_pols = num_pols;
1865 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1866 xdst->policy_genid = atomic_read(&pols[0]->genid);
1868 atomic_set(&xdst->u.dst.__refcnt, 2);
1869 xfrm_last_dst_update(xdst, old);
1874 static void xfrm_policy_queue_process(struct timer_list *t)
1876 struct sk_buff *skb;
1878 struct dst_entry *dst;
1879 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1880 struct net *net = xp_net(pol);
1881 struct xfrm_policy_queue *pq = &pol->polq;
1883 struct sk_buff_head list;
1885 spin_lock(&pq->hold_queue.lock);
1886 skb = skb_peek(&pq->hold_queue);
1888 spin_unlock(&pq->hold_queue.lock);
1893 xfrm_decode_session(skb, &fl, dst->ops->family);
1894 spin_unlock(&pq->hold_queue.lock);
1896 dst_hold(xfrm_dst_path(dst));
1897 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
1901 if (dst->flags & DST_XFRM_QUEUE) {
1904 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1907 pq->timeout = pq->timeout << 1;
1908 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1915 __skb_queue_head_init(&list);
1917 spin_lock(&pq->hold_queue.lock);
1919 skb_queue_splice_init(&pq->hold_queue, &list);
1920 spin_unlock(&pq->hold_queue.lock);
1922 while (!skb_queue_empty(&list)) {
1923 skb = __skb_dequeue(&list);
1925 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1926 dst_hold(xfrm_dst_path(skb_dst(skb)));
1927 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
1935 skb_dst_set(skb, dst);
1937 dst_output(net, skb->sk, skb);
1946 skb_queue_purge(&pq->hold_queue);
1950 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1952 unsigned long sched_next;
1953 struct dst_entry *dst = skb_dst(skb);
1954 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1955 struct xfrm_policy *pol = xdst->pols[0];
1956 struct xfrm_policy_queue *pq = &pol->polq;
1958 if (unlikely(skb_fclone_busy(sk, skb))) {
1963 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1970 spin_lock_bh(&pq->hold_queue.lock);
1973 pq->timeout = XFRM_QUEUE_TMO_MIN;
1975 sched_next = jiffies + pq->timeout;
1977 if (del_timer(&pq->hold_timer)) {
1978 if (time_before(pq->hold_timer.expires, sched_next))
1979 sched_next = pq->hold_timer.expires;
1983 __skb_queue_tail(&pq->hold_queue, skb);
1984 if (!mod_timer(&pq->hold_timer, sched_next))
1987 spin_unlock_bh(&pq->hold_queue.lock);
1992 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1993 struct xfrm_flo *xflo,
1994 const struct flowi *fl,
1999 struct net_device *dev;
2000 struct dst_entry *dst;
2001 struct dst_entry *dst1;
2002 struct xfrm_dst *xdst;
2004 xdst = xfrm_alloc_dst(net, family);
2008 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2009 net->xfrm.sysctl_larval_drop ||
2013 dst = xflo->dst_orig;
2014 dst1 = &xdst->u.dst;
2018 dst_copy_metrics(dst1, dst);
2020 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2021 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2022 dst1->lastuse = jiffies;
2024 dst1->input = dst_discard;
2025 dst1->output = xdst_queue_output;
2028 xfrm_dst_set_child(xdst, dst);
2031 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2038 err = xfrm_fill_dst(xdst, dev, fl);
2047 xdst = ERR_PTR(err);
2051 static struct xfrm_dst *
2052 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2054 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2055 int num_pols = 0, num_xfrms = 0, err;
2056 struct xfrm_dst *xdst;
2058 /* Resolve policies to use if we couldn't get them from
2059 * previous cache entry */
2061 pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2062 err = xfrm_expand_policies(fl, family, pols,
2063 &num_pols, &num_xfrms);
2069 goto make_dummy_bundle;
2072 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2077 err = PTR_ERR(xdst);
2080 goto make_dummy_bundle;
2081 } else if (xdst == NULL) {
2083 goto make_dummy_bundle;
2089 /* We found policies, but there's no bundles to instantiate:
2090 * either because the policy blocks, has no transformations or
2091 * we could not build template (no xfrm_states).*/
2092 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2094 xfrm_pols_put(pols, num_pols);
2095 return ERR_CAST(xdst);
2097 xdst->num_pols = num_pols;
2098 xdst->num_xfrms = num_xfrms;
2099 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2104 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2106 xfrm_pols_put(pols, num_pols);
2107 return ERR_PTR(err);
2110 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2111 struct dst_entry *dst_orig)
2113 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2114 struct dst_entry *ret;
2117 dst_release(dst_orig);
2118 return ERR_PTR(-EINVAL);
2120 ret = afinfo->blackhole_route(net, dst_orig);
2127 /* Main function: finds/creates a bundle for given flow.
2129 * At the moment we eat a raw IP route. Mostly to speed up lookups
2130 * on interfaces with disabled IPsec.
2132 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2133 const struct flowi *fl,
2134 const struct sock *sk, int flags)
2136 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2137 struct xfrm_dst *xdst;
2138 struct dst_entry *dst, *route;
2139 u16 family = dst_orig->ops->family;
2140 u8 dir = XFRM_POLICY_OUT;
2141 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2147 sk = sk_const_to_full_sk(sk);
2148 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2150 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2151 err = xfrm_expand_policies(fl, family, pols,
2152 &num_pols, &num_xfrms);
2157 if (num_xfrms <= 0) {
2158 drop_pols = num_pols;
2163 xdst = xfrm_resolve_and_create_bundle(
2169 xfrm_pols_put(pols, num_pols);
2170 err = PTR_ERR(xdst);
2172 } else if (xdst == NULL) {
2174 drop_pols = num_pols;
2178 route = xdst->route;
2183 struct xfrm_flo xflo;
2185 xflo.dst_orig = dst_orig;
2188 /* To accelerate a bit... */
2189 if ((dst_orig->flags & DST_NOXFRM) ||
2190 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2193 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2197 err = PTR_ERR(xdst);
2201 num_pols = xdst->num_pols;
2202 num_xfrms = xdst->num_xfrms;
2203 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2204 route = xdst->route;
2208 if (route == NULL && num_xfrms > 0) {
2209 /* The only case when xfrm_bundle_lookup() returns a
2210 * bundle with null route, is when the template could
2211 * not be resolved. It means policies are there, but
2212 * bundle could not be created, since we don't yet
2213 * have the xfrm_state's. We need to wait for KM to
2214 * negotiate new SA's or bail out with error.*/
2215 if (net->xfrm.sysctl_larval_drop) {
2216 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2223 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2231 if ((flags & XFRM_LOOKUP_ICMP) &&
2232 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2237 for (i = 0; i < num_pols; i++)
2238 pols[i]->curlft.use_time = get_seconds();
2240 if (num_xfrms < 0) {
2241 /* Prohibit the flow */
2242 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2245 } else if (num_xfrms > 0) {
2246 /* Flow transformed */
2247 dst_release(dst_orig);
2249 /* Flow passes untransformed */
2254 xfrm_pols_put(pols, drop_pols);
2255 if (dst && dst->xfrm &&
2256 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2257 dst->flags |= DST_XFRM_TUNNEL;
2261 if (!(flags & XFRM_LOOKUP_ICMP)) {
2269 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2270 dst_release(dst_orig);
2271 xfrm_pols_put(pols, drop_pols);
2272 return ERR_PTR(err);
2274 EXPORT_SYMBOL(xfrm_lookup);
2276 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2277 * Otherwise we may send out blackholed packets.
2279 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2280 const struct flowi *fl,
2281 const struct sock *sk, int flags)
2283 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2284 flags | XFRM_LOOKUP_QUEUE |
2285 XFRM_LOOKUP_KEEP_DST_REF);
2287 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2288 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2292 EXPORT_SYMBOL(xfrm_lookup_route);
2295 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2297 struct xfrm_state *x;
2299 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2301 x = skb->sp->xvec[idx];
2302 if (!x->type->reject)
2304 return x->type->reject(x, skb, fl);
2307 /* When skb is transformed back to its "native" form, we have to
2308 * check policy restrictions. At the moment we make this in maximally
2309 * stupid way. Shame on me. :-) Of course, connected sockets must
2310 * have policy cached at them.
2314 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2315 unsigned short family)
2317 if (xfrm_state_kern(x))
2318 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2319 return x->id.proto == tmpl->id.proto &&
2320 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2321 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2322 x->props.mode == tmpl->mode &&
2323 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2324 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2325 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2326 xfrm_state_addr_cmp(tmpl, x, family));
2330 * 0 or more than 0 is returned when validation is succeeded (either bypass
2331 * because of optional transport mode, or next index of the mathced secpath
2332 * state with the template.
2333 * -1 is returned when no matching template is found.
2334 * Otherwise "-2 - errored_index" is returned.
2337 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2338 unsigned short family)
2342 if (tmpl->optional) {
2343 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2347 for (; idx < sp->len; idx++) {
2348 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2350 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2359 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2360 unsigned int family, int reverse)
2362 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2365 if (unlikely(afinfo == NULL))
2366 return -EAFNOSUPPORT;
2368 afinfo->decode_session(skb, fl, reverse);
2369 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2373 EXPORT_SYMBOL(__xfrm_decode_session);
2375 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2377 for (; k < sp->len; k++) {
2378 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2387 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2388 unsigned short family)
2390 struct net *net = dev_net(skb->dev);
2391 struct xfrm_policy *pol;
2392 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2400 reverse = dir & ~XFRM_POLICY_MASK;
2401 dir &= XFRM_POLICY_MASK;
2403 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2404 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2408 nf_nat_decode_session(skb, &fl, family);
2410 /* First, check used SA against their selectors. */
2414 for (i = skb->sp->len-1; i >= 0; i--) {
2415 struct xfrm_state *x = skb->sp->xvec[i];
2416 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2417 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2424 sk = sk_to_full_sk(sk);
2425 if (sk && sk->sk_policy[dir]) {
2426 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2428 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2434 pol = xfrm_policy_lookup(net, &fl, family, dir);
2437 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2442 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2443 xfrm_secpath_reject(xerr_idx, skb, &fl);
2444 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2450 pol->curlft.use_time = get_seconds();
2454 #ifdef CONFIG_XFRM_SUB_POLICY
2455 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2456 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2460 if (IS_ERR(pols[1])) {
2461 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2464 pols[1]->curlft.use_time = get_seconds();
2470 if (pol->action == XFRM_POLICY_ALLOW) {
2471 struct sec_path *sp;
2472 static struct sec_path dummy;
2473 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2474 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2475 struct xfrm_tmpl **tpp = tp;
2479 if ((sp = skb->sp) == NULL)
2482 for (pi = 0; pi < npols; pi++) {
2483 if (pols[pi] != pol &&
2484 pols[pi]->action != XFRM_POLICY_ALLOW) {
2485 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2488 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2489 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2492 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2493 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2497 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2501 /* For each tunnel xfrm, find the first matching tmpl.
2502 * For each tmpl before that, find corresponding xfrm.
2503 * Order is _important_. Later we will implement
2504 * some barriers, but at the moment barriers
2505 * are implied between each two transformations.
2507 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2508 k = xfrm_policy_ok(tpp[i], sp, k, family);
2511 /* "-2 - errored_index" returned */
2513 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2518 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2519 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2523 xfrm_pols_put(pols, npols);
2526 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2529 xfrm_secpath_reject(xerr_idx, skb, &fl);
2531 xfrm_pols_put(pols, npols);
2534 EXPORT_SYMBOL(__xfrm_policy_check);
2536 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2538 struct net *net = dev_net(skb->dev);
2540 struct dst_entry *dst;
2543 if (xfrm_decode_session(skb, &fl, family) < 0) {
2544 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2550 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2555 skb_dst_set(skb, dst);
2558 EXPORT_SYMBOL(__xfrm_route_forward);
2560 /* Optimize later using cookies and generation ids. */
2562 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2564 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2565 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2566 * get validated by dst_ops->check on every use. We do this
2567 * because when a normal route referenced by an XFRM dst is
2568 * obsoleted we do not go looking around for all parent
2569 * referencing XFRM dsts so that we can invalidate them. It
2570 * is just too much work. Instead we make the checks here on
2571 * every use. For example:
2573 * XFRM dst A --> IPv4 dst X
2575 * X is the "xdst->route" of A (X is also the "dst->path" of A
2576 * in this example). If X is marked obsolete, "A" will not
2577 * notice. That's what we are validating here via the
2578 * stale_bundle() check.
2580 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2582 * This will force stale_bundle() to fail on any xdst bundle with
2583 * this dst linked in it.
2585 if (dst->obsolete < 0 && !stale_bundle(dst))
2591 static int stale_bundle(struct dst_entry *dst)
2593 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2596 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2598 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
2599 dst->dev = dev_net(dev)->loopback_dev;
2604 EXPORT_SYMBOL(xfrm_dst_ifdown);
2606 static void xfrm_link_failure(struct sk_buff *skb)
2608 /* Impossible. Such dst must be popped before reaches point of failure. */
2611 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2614 if (dst->obsolete) {
2622 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
2625 struct xfrm_dst *xdst = bundle[nr];
2626 u32 pmtu, route_mtu_cached;
2627 struct dst_entry *dst;
2630 pmtu = dst_mtu(xfrm_dst_child(dst));
2631 xdst->child_mtu_cached = pmtu;
2633 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2635 route_mtu_cached = dst_mtu(xdst->route);
2636 xdst->route_mtu_cached = route_mtu_cached;
2638 if (pmtu > route_mtu_cached)
2639 pmtu = route_mtu_cached;
2641 dst_metric_set(dst, RTAX_MTU, pmtu);
2645 /* Check that the bundle accepts the flow and its components are
2649 static int xfrm_bundle_ok(struct xfrm_dst *first)
2651 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2652 struct dst_entry *dst = &first->u.dst;
2653 struct xfrm_dst *xdst;
2657 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
2658 (dst->dev && !netif_running(dst->dev)))
2661 if (dst->flags & DST_XFRM_QUEUE)
2664 start_from = nr = 0;
2666 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2668 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2670 if (xdst->xfrm_genid != dst->xfrm->genid)
2672 if (xdst->num_pols > 0 &&
2673 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2676 bundle[nr++] = xdst;
2678 mtu = dst_mtu(xfrm_dst_child(dst));
2679 if (xdst->child_mtu_cached != mtu) {
2681 xdst->child_mtu_cached = mtu;
2684 if (!dst_check(xdst->route, xdst->route_cookie))
2686 mtu = dst_mtu(xdst->route);
2687 if (xdst->route_mtu_cached != mtu) {
2689 xdst->route_mtu_cached = mtu;
2692 dst = xfrm_dst_child(dst);
2693 } while (dst->xfrm);
2695 if (likely(!start_from))
2698 xdst = bundle[start_from - 1];
2699 mtu = xdst->child_mtu_cached;
2700 while (start_from--) {
2703 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2704 if (mtu > xdst->route_mtu_cached)
2705 mtu = xdst->route_mtu_cached;
2706 dst_metric_set(dst, RTAX_MTU, mtu);
2710 xdst = bundle[start_from - 1];
2711 xdst->child_mtu_cached = mtu;
2717 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2719 return dst_metric_advmss(xfrm_dst_path(dst));
2722 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2724 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2726 return mtu ? : dst_mtu(xfrm_dst_path(dst));
2729 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2733 const struct xfrm_state *xfrm = dst->xfrm;
2735 dst = xfrm_dst_child(dst);
2737 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2739 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2740 daddr = xfrm->coaddr;
2741 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2742 daddr = &xfrm->id.daddr;
2747 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2748 struct sk_buff *skb,
2751 const struct dst_entry *path = xfrm_dst_path(dst);
2754 daddr = xfrm_get_dst_nexthop(dst, daddr);
2755 return path->ops->neigh_lookup(path, skb, daddr);
2758 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2760 const struct dst_entry *path = xfrm_dst_path(dst);
2762 daddr = xfrm_get_dst_nexthop(dst, daddr);
2763 path->ops->confirm_neigh(path, daddr);
2766 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2770 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2771 return -EAFNOSUPPORT;
2773 spin_lock(&xfrm_policy_afinfo_lock);
2774 if (unlikely(xfrm_policy_afinfo[family] != NULL))
2777 struct dst_ops *dst_ops = afinfo->dst_ops;
2778 if (likely(dst_ops->kmem_cachep == NULL))
2779 dst_ops->kmem_cachep = xfrm_dst_cache;
2780 if (likely(dst_ops->check == NULL))
2781 dst_ops->check = xfrm_dst_check;
2782 if (likely(dst_ops->default_advmss == NULL))
2783 dst_ops->default_advmss = xfrm_default_advmss;
2784 if (likely(dst_ops->mtu == NULL))
2785 dst_ops->mtu = xfrm_mtu;
2786 if (likely(dst_ops->negative_advice == NULL))
2787 dst_ops->negative_advice = xfrm_negative_advice;
2788 if (likely(dst_ops->link_failure == NULL))
2789 dst_ops->link_failure = xfrm_link_failure;
2790 if (likely(dst_ops->neigh_lookup == NULL))
2791 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2792 if (likely(!dst_ops->confirm_neigh))
2793 dst_ops->confirm_neigh = xfrm_confirm_neigh;
2794 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2796 spin_unlock(&xfrm_policy_afinfo_lock);
2800 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2802 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2804 struct dst_ops *dst_ops = afinfo->dst_ops;
2807 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2808 if (xfrm_policy_afinfo[i] != afinfo)
2810 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2816 dst_ops->kmem_cachep = NULL;
2817 dst_ops->check = NULL;
2818 dst_ops->negative_advice = NULL;
2819 dst_ops->link_failure = NULL;
2821 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2823 #ifdef CONFIG_XFRM_STATISTICS
2824 static int __net_init xfrm_statistics_init(struct net *net)
2827 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2828 if (!net->mib.xfrm_statistics)
2830 rv = xfrm_proc_init(net);
2832 free_percpu(net->mib.xfrm_statistics);
2836 static void xfrm_statistics_fini(struct net *net)
2838 xfrm_proc_fini(net);
2839 free_percpu(net->mib.xfrm_statistics);
2842 static int __net_init xfrm_statistics_init(struct net *net)
2847 static void xfrm_statistics_fini(struct net *net)
2852 static int __net_init xfrm_policy_init(struct net *net)
2854 unsigned int hmask, sz;
2857 if (net_eq(net, &init_net))
2858 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2859 sizeof(struct xfrm_dst),
2860 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2864 sz = (hmask+1) * sizeof(struct hlist_head);
2866 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2867 if (!net->xfrm.policy_byidx)
2869 net->xfrm.policy_idx_hmask = hmask;
2871 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2872 struct xfrm_policy_hash *htab;
2874 net->xfrm.policy_count[dir] = 0;
2875 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2876 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2878 htab = &net->xfrm.policy_bydst[dir];
2879 htab->table = xfrm_hash_alloc(sz);
2882 htab->hmask = hmask;
2888 net->xfrm.policy_hthresh.lbits4 = 32;
2889 net->xfrm.policy_hthresh.rbits4 = 32;
2890 net->xfrm.policy_hthresh.lbits6 = 128;
2891 net->xfrm.policy_hthresh.rbits6 = 128;
2893 seqlock_init(&net->xfrm.policy_hthresh.lock);
2895 INIT_LIST_HEAD(&net->xfrm.policy_all);
2896 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2897 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2901 for (dir--; dir >= 0; dir--) {
2902 struct xfrm_policy_hash *htab;
2904 htab = &net->xfrm.policy_bydst[dir];
2905 xfrm_hash_free(htab->table, sz);
2907 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2912 static void xfrm_policy_fini(struct net *net)
2917 flush_work(&net->xfrm.policy_hash_work);
2918 #ifdef CONFIG_XFRM_SUB_POLICY
2919 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2921 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2923 WARN_ON(!list_empty(&net->xfrm.policy_all));
2925 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2926 struct xfrm_policy_hash *htab;
2928 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2930 htab = &net->xfrm.policy_bydst[dir];
2931 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2932 WARN_ON(!hlist_empty(htab->table));
2933 xfrm_hash_free(htab->table, sz);
2936 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2937 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2938 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2941 static int __net_init xfrm_net_init(struct net *net)
2945 /* Initialize the per-net locks here */
2946 spin_lock_init(&net->xfrm.xfrm_state_lock);
2947 spin_lock_init(&net->xfrm.xfrm_policy_lock);
2948 mutex_init(&net->xfrm.xfrm_cfg_mutex);
2950 rv = xfrm_statistics_init(net);
2952 goto out_statistics;
2953 rv = xfrm_state_init(net);
2956 rv = xfrm_policy_init(net);
2959 rv = xfrm_sysctl_init(net);
2966 xfrm_policy_fini(net);
2968 xfrm_state_fini(net);
2970 xfrm_statistics_fini(net);
2975 static void __net_exit xfrm_net_exit(struct net *net)
2977 xfrm_sysctl_fini(net);
2978 xfrm_policy_fini(net);
2979 xfrm_state_fini(net);
2980 xfrm_statistics_fini(net);
2983 static struct pernet_operations __net_initdata xfrm_net_ops = {
2984 .init = xfrm_net_init,
2985 .exit = xfrm_net_exit,
2988 void __init xfrm_init(void)
2992 xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2994 BUG_ON(!xfrm_pcpu_work);
2996 for (i = 0; i < NR_CPUS; i++)
2997 INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
2999 register_pernet_subsys(&xfrm_net_ops);
3001 seqcount_init(&xfrm_policy_hash_generation);
3005 #ifdef CONFIG_AUDITSYSCALL
3006 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3007 struct audit_buffer *audit_buf)
3009 struct xfrm_sec_ctx *ctx = xp->security;
3010 struct xfrm_selector *sel = &xp->selector;
3013 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3014 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3016 switch (sel->family) {
3018 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3019 if (sel->prefixlen_s != 32)
3020 audit_log_format(audit_buf, " src_prefixlen=%d",
3022 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3023 if (sel->prefixlen_d != 32)
3024 audit_log_format(audit_buf, " dst_prefixlen=%d",
3028 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3029 if (sel->prefixlen_s != 128)
3030 audit_log_format(audit_buf, " src_prefixlen=%d",
3032 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3033 if (sel->prefixlen_d != 128)
3034 audit_log_format(audit_buf, " dst_prefixlen=%d",
3040 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3042 struct audit_buffer *audit_buf;
3044 audit_buf = xfrm_audit_start("SPD-add");
3045 if (audit_buf == NULL)
3047 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3048 audit_log_format(audit_buf, " res=%u", result);
3049 xfrm_audit_common_policyinfo(xp, audit_buf);
3050 audit_log_end(audit_buf);
3052 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3054 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3057 struct audit_buffer *audit_buf;
3059 audit_buf = xfrm_audit_start("SPD-delete");
3060 if (audit_buf == NULL)
3062 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3063 audit_log_format(audit_buf, " res=%u", result);
3064 xfrm_audit_common_policyinfo(xp, audit_buf);
3065 audit_log_end(audit_buf);
3067 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3070 #ifdef CONFIG_XFRM_MIGRATE
3071 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3072 const struct xfrm_selector *sel_tgt)
3074 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3075 if (sel_tgt->family == sel_cmp->family &&
3076 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3078 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3080 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3081 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3085 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3092 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3093 u8 dir, u8 type, struct net *net)
3095 struct xfrm_policy *pol, *ret = NULL;
3096 struct hlist_head *chain;
3099 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3100 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3101 hlist_for_each_entry(pol, chain, bydst) {
3102 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3103 pol->type == type) {
3105 priority = ret->priority;
3109 chain = &net->xfrm.policy_inexact[dir];
3110 hlist_for_each_entry(pol, chain, bydst) {
3111 if ((pol->priority >= priority) && ret)
3114 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3115 pol->type == type) {
3123 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3128 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3132 if (t->mode == m->mode && t->id.proto == m->proto &&
3133 (m->reqid == 0 || t->reqid == m->reqid)) {
3135 case XFRM_MODE_TUNNEL:
3136 case XFRM_MODE_BEET:
3137 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3139 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3144 case XFRM_MODE_TRANSPORT:
3145 /* in case of transport mode, template does not store
3146 any IP addresses, hence we just compare mode and
3157 /* update endpoint address(es) of template(s) */
3158 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3159 struct xfrm_migrate *m, int num_migrate)
3161 struct xfrm_migrate *mp;
3164 write_lock_bh(&pol->lock);
3165 if (unlikely(pol->walk.dead)) {
3166 /* target policy has been deleted */
3167 write_unlock_bh(&pol->lock);
3171 for (i = 0; i < pol->xfrm_nr; i++) {
3172 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3173 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3176 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3177 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3179 /* update endpoints */
3180 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3181 sizeof(pol->xfrm_vec[i].id.daddr));
3182 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3183 sizeof(pol->xfrm_vec[i].saddr));
3184 pol->xfrm_vec[i].encap_family = mp->new_family;
3186 atomic_inc(&pol->genid);
3190 write_unlock_bh(&pol->lock);
3198 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3202 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3205 for (i = 0; i < num_migrate; i++) {
3206 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3207 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3210 /* check if there is any duplicated entry */
3211 for (j = i + 1; j < num_migrate; j++) {
3212 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3213 sizeof(m[i].old_daddr)) &&
3214 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3215 sizeof(m[i].old_saddr)) &&
3216 m[i].proto == m[j].proto &&
3217 m[i].mode == m[j].mode &&
3218 m[i].reqid == m[j].reqid &&
3219 m[i].old_family == m[j].old_family)
3227 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3228 struct xfrm_migrate *m, int num_migrate,
3229 struct xfrm_kmaddress *k, struct net *net,
3230 struct xfrm_encap_tmpl *encap)
3232 int i, err, nx_cur = 0, nx_new = 0;
3233 struct xfrm_policy *pol = NULL;
3234 struct xfrm_state *x, *xc;
3235 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3236 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3237 struct xfrm_migrate *mp;
3239 /* Stage 0 - sanity checks */
3240 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3243 if (dir >= XFRM_POLICY_MAX) {
3248 /* Stage 1 - find policy */
3249 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3254 /* Stage 2 - find and update state(s) */
3255 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3256 if ((x = xfrm_migrate_state_find(mp, net))) {
3259 xc = xfrm_state_migrate(x, mp, encap);
3270 /* Stage 3 - update policy */
3271 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3274 /* Stage 4 - delete old state(s) */
3276 xfrm_states_put(x_cur, nx_cur);
3277 xfrm_states_delete(x_cur, nx_cur);
3280 /* Stage 5 - announce */
3281 km_migrate(sel, dir, type, m, num_migrate, k, encap);
3293 xfrm_states_put(x_cur, nx_cur);
3295 xfrm_states_delete(x_new, nx_new);
3299 EXPORT_SYMBOL(xfrm_migrate);