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_policy_afinfo_lock);
49 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
52 static struct kmem_cache *xfrm_dst_cache __read_mostly;
53 static __read_mostly seqcount_t xfrm_policy_hash_generation;
55 static void xfrm_init_pmtu(struct dst_entry *dst);
56 static int stale_bundle(struct dst_entry *dst);
57 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
58 static void xfrm_policy_queue_process(unsigned long arg);
60 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
61 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
64 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
66 return refcount_inc_not_zero(&policy->refcnt);
70 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
72 const struct flowi4 *fl4 = &fl->u.ip4;
74 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
75 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
76 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
77 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
78 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
79 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
83 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
85 const struct flowi6 *fl6 = &fl->u.ip6;
87 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
88 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
89 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
90 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
91 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
92 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
95 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
96 unsigned short family)
100 return __xfrm4_selector_match(sel, fl);
102 return __xfrm6_selector_match(sel, fl);
107 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
109 const struct xfrm_policy_afinfo *afinfo;
111 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
114 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
115 if (unlikely(!afinfo))
120 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
121 const xfrm_address_t *saddr,
122 const xfrm_address_t *daddr,
123 int family, u32 mark)
125 const struct xfrm_policy_afinfo *afinfo;
126 struct dst_entry *dst;
128 afinfo = xfrm_policy_get_afinfo(family);
129 if (unlikely(afinfo == NULL))
130 return ERR_PTR(-EAFNOSUPPORT);
132 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
138 EXPORT_SYMBOL(__xfrm_dst_lookup);
140 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
142 xfrm_address_t *prev_saddr,
143 xfrm_address_t *prev_daddr,
144 int family, u32 mark)
146 struct net *net = xs_net(x);
147 xfrm_address_t *saddr = &x->props.saddr;
148 xfrm_address_t *daddr = &x->id.daddr;
149 struct dst_entry *dst;
151 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
155 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
160 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
163 if (prev_saddr != saddr)
164 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
165 if (prev_daddr != daddr)
166 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
172 static inline unsigned long make_jiffies(long secs)
174 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
175 return MAX_SCHEDULE_TIMEOUT-1;
180 static void xfrm_policy_timer(unsigned long data)
182 struct xfrm_policy *xp = (struct xfrm_policy *)data;
183 unsigned long now = get_seconds();
184 long next = LONG_MAX;
188 read_lock(&xp->lock);
190 if (unlikely(xp->walk.dead))
193 dir = xfrm_policy_id2dir(xp->index);
195 if (xp->lft.hard_add_expires_seconds) {
196 long tmo = xp->lft.hard_add_expires_seconds +
197 xp->curlft.add_time - now;
203 if (xp->lft.hard_use_expires_seconds) {
204 long tmo = xp->lft.hard_use_expires_seconds +
205 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
211 if (xp->lft.soft_add_expires_seconds) {
212 long tmo = xp->lft.soft_add_expires_seconds +
213 xp->curlft.add_time - now;
216 tmo = XFRM_KM_TIMEOUT;
221 if (xp->lft.soft_use_expires_seconds) {
222 long tmo = xp->lft.soft_use_expires_seconds +
223 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
226 tmo = XFRM_KM_TIMEOUT;
233 km_policy_expired(xp, dir, 0, 0);
234 if (next != LONG_MAX &&
235 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
239 read_unlock(&xp->lock);
244 read_unlock(&xp->lock);
245 if (!xfrm_policy_delete(xp, dir))
246 km_policy_expired(xp, dir, 1, 0);
250 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
254 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
256 struct xfrm_policy *policy;
258 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
261 write_pnet(&policy->xp_net, net);
262 INIT_LIST_HEAD(&policy->walk.all);
263 INIT_HLIST_NODE(&policy->bydst);
264 INIT_HLIST_NODE(&policy->byidx);
265 rwlock_init(&policy->lock);
266 refcount_set(&policy->refcnt, 1);
267 skb_queue_head_init(&policy->polq.hold_queue);
268 setup_timer(&policy->timer, xfrm_policy_timer,
269 (unsigned long)policy);
270 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
271 (unsigned long)policy);
275 EXPORT_SYMBOL(xfrm_policy_alloc);
277 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
279 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
281 security_xfrm_policy_free(policy->security);
285 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
287 void xfrm_policy_destroy(struct xfrm_policy *policy)
289 BUG_ON(!policy->walk.dead);
291 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
294 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
296 EXPORT_SYMBOL(xfrm_policy_destroy);
298 /* Rule must be locked. Release descendant resources, announce
299 * entry dead. The rule must be unlinked from lists to the moment.
302 static void xfrm_policy_kill(struct xfrm_policy *policy)
304 write_lock_bh(&policy->lock);
305 policy->walk.dead = 1;
306 write_unlock_bh(&policy->lock);
308 atomic_inc(&policy->genid);
310 if (del_timer(&policy->polq.hold_timer))
311 xfrm_pol_put(policy);
312 skb_queue_purge(&policy->polq.hold_queue);
314 if (del_timer(&policy->timer))
315 xfrm_pol_put(policy);
317 xfrm_pol_put(policy);
320 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
322 static inline unsigned int idx_hash(struct net *net, u32 index)
324 return __idx_hash(index, net->xfrm.policy_idx_hmask);
327 /* calculate policy hash thresholds */
328 static void __get_hash_thresh(struct net *net,
329 unsigned short family, int dir,
330 u8 *dbits, u8 *sbits)
334 *dbits = net->xfrm.policy_bydst[dir].dbits4;
335 *sbits = net->xfrm.policy_bydst[dir].sbits4;
339 *dbits = net->xfrm.policy_bydst[dir].dbits6;
340 *sbits = net->xfrm.policy_bydst[dir].sbits6;
349 static struct hlist_head *policy_hash_bysel(struct net *net,
350 const struct xfrm_selector *sel,
351 unsigned short family, int dir)
353 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
358 __get_hash_thresh(net, family, dir, &dbits, &sbits);
359 hash = __sel_hash(sel, family, hmask, dbits, sbits);
361 if (hash == hmask + 1)
362 return &net->xfrm.policy_inexact[dir];
364 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
365 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
368 static struct hlist_head *policy_hash_direct(struct net *net,
369 const xfrm_address_t *daddr,
370 const xfrm_address_t *saddr,
371 unsigned short family, int dir)
373 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
378 __get_hash_thresh(net, family, dir, &dbits, &sbits);
379 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
381 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
382 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
385 static void xfrm_dst_hash_transfer(struct net *net,
386 struct hlist_head *list,
387 struct hlist_head *ndsttable,
388 unsigned int nhashmask,
391 struct hlist_node *tmp, *entry0 = NULL;
392 struct xfrm_policy *pol;
398 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
401 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
402 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
403 pol->family, nhashmask, dbits, sbits);
405 hlist_del_rcu(&pol->bydst);
406 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
411 hlist_del_rcu(&pol->bydst);
412 hlist_add_behind_rcu(&pol->bydst, entry0);
414 entry0 = &pol->bydst;
416 if (!hlist_empty(list)) {
422 static void xfrm_idx_hash_transfer(struct hlist_head *list,
423 struct hlist_head *nidxtable,
424 unsigned int nhashmask)
426 struct hlist_node *tmp;
427 struct xfrm_policy *pol;
429 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
432 h = __idx_hash(pol->index, nhashmask);
433 hlist_add_head(&pol->byidx, nidxtable+h);
437 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
439 return ((old_hmask + 1) << 1) - 1;
442 static void xfrm_bydst_resize(struct net *net, int dir)
444 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
445 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
446 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
447 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
448 struct hlist_head *odst;
454 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
455 write_seqcount_begin(&xfrm_policy_hash_generation);
457 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
458 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
460 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
461 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
463 for (i = hmask; i >= 0; i--)
464 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
466 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
467 net->xfrm.policy_bydst[dir].hmask = nhashmask;
469 write_seqcount_end(&xfrm_policy_hash_generation);
470 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
474 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
477 static void xfrm_byidx_resize(struct net *net, int total)
479 unsigned int hmask = net->xfrm.policy_idx_hmask;
480 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
481 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
482 struct hlist_head *oidx = net->xfrm.policy_byidx;
483 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
489 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
491 for (i = hmask; i >= 0; i--)
492 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
494 net->xfrm.policy_byidx = nidx;
495 net->xfrm.policy_idx_hmask = nhashmask;
497 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
499 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
502 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
504 unsigned int cnt = net->xfrm.policy_count[dir];
505 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
510 if ((hmask + 1) < xfrm_policy_hashmax &&
517 static inline int xfrm_byidx_should_resize(struct net *net, int total)
519 unsigned int hmask = net->xfrm.policy_idx_hmask;
521 if ((hmask + 1) < xfrm_policy_hashmax &&
528 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
530 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
531 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
532 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
533 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
534 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
535 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
536 si->spdhcnt = net->xfrm.policy_idx_hmask;
537 si->spdhmcnt = xfrm_policy_hashmax;
539 EXPORT_SYMBOL(xfrm_spd_getinfo);
541 static DEFINE_MUTEX(hash_resize_mutex);
542 static void xfrm_hash_resize(struct work_struct *work)
544 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
547 mutex_lock(&hash_resize_mutex);
550 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
551 if (xfrm_bydst_should_resize(net, dir, &total))
552 xfrm_bydst_resize(net, dir);
554 if (xfrm_byidx_should_resize(net, total))
555 xfrm_byidx_resize(net, total);
557 mutex_unlock(&hash_resize_mutex);
560 static void xfrm_hash_rebuild(struct work_struct *work)
562 struct net *net = container_of(work, struct net,
563 xfrm.policy_hthresh.work);
565 struct xfrm_policy *pol;
566 struct xfrm_policy *policy;
567 struct hlist_head *chain;
568 struct hlist_head *odst;
569 struct hlist_node *newpos;
573 u8 lbits4, rbits4, lbits6, rbits6;
575 mutex_lock(&hash_resize_mutex);
577 /* read selector prefixlen thresholds */
579 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
581 lbits4 = net->xfrm.policy_hthresh.lbits4;
582 rbits4 = net->xfrm.policy_hthresh.rbits4;
583 lbits6 = net->xfrm.policy_hthresh.lbits6;
584 rbits6 = net->xfrm.policy_hthresh.rbits6;
585 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
587 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
589 /* reset the bydst and inexact table in all directions */
590 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
591 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
592 hmask = net->xfrm.policy_bydst[dir].hmask;
593 odst = net->xfrm.policy_bydst[dir].table;
594 for (i = hmask; i >= 0; i--)
595 INIT_HLIST_HEAD(odst + i);
596 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
597 /* dir out => dst = remote, src = local */
598 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
599 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
600 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
601 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
603 /* dir in/fwd => dst = local, src = remote */
604 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
605 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
606 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
607 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
611 /* re-insert all policies by order of creation */
612 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
613 if (policy->walk.dead ||
614 xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
615 /* skip socket policies */
619 chain = policy_hash_bysel(net, &policy->selector,
621 xfrm_policy_id2dir(policy->index));
622 hlist_for_each_entry(pol, chain, bydst) {
623 if (policy->priority >= pol->priority)
624 newpos = &pol->bydst;
629 hlist_add_behind_rcu(&policy->bydst, newpos);
631 hlist_add_head_rcu(&policy->bydst, chain);
634 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
636 mutex_unlock(&hash_resize_mutex);
639 void xfrm_policy_hash_rebuild(struct net *net)
641 schedule_work(&net->xfrm.policy_hthresh.work);
643 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
645 /* Generate new index... KAME seems to generate them ordered by cost
646 * of an absolute inpredictability of ordering of rules. This will not pass. */
647 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
649 static u32 idx_generator;
652 struct hlist_head *list;
653 struct xfrm_policy *p;
658 idx = (idx_generator | dir);
667 list = net->xfrm.policy_byidx + idx_hash(net, idx);
669 hlist_for_each_entry(p, list, byidx) {
670 if (p->index == idx) {
680 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
682 u32 *p1 = (u32 *) s1;
683 u32 *p2 = (u32 *) s2;
684 int len = sizeof(struct xfrm_selector) / sizeof(u32);
687 for (i = 0; i < len; i++) {
695 static void xfrm_policy_requeue(struct xfrm_policy *old,
696 struct xfrm_policy *new)
698 struct xfrm_policy_queue *pq = &old->polq;
699 struct sk_buff_head list;
701 if (skb_queue_empty(&pq->hold_queue))
704 __skb_queue_head_init(&list);
706 spin_lock_bh(&pq->hold_queue.lock);
707 skb_queue_splice_init(&pq->hold_queue, &list);
708 if (del_timer(&pq->hold_timer))
710 spin_unlock_bh(&pq->hold_queue.lock);
714 spin_lock_bh(&pq->hold_queue.lock);
715 skb_queue_splice(&list, &pq->hold_queue);
716 pq->timeout = XFRM_QUEUE_TMO_MIN;
717 if (!mod_timer(&pq->hold_timer, jiffies))
719 spin_unlock_bh(&pq->hold_queue.lock);
722 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
723 struct xfrm_policy *pol)
725 if (policy->mark.v == pol->mark.v &&
726 policy->priority == pol->priority)
732 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
734 struct net *net = xp_net(policy);
735 struct xfrm_policy *pol;
736 struct xfrm_policy *delpol;
737 struct hlist_head *chain;
738 struct hlist_node *newpos;
740 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
741 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
744 hlist_for_each_entry(pol, chain, bydst) {
745 if (pol->type == policy->type &&
746 !selector_cmp(&pol->selector, &policy->selector) &&
747 xfrm_policy_mark_match(policy, pol) &&
748 xfrm_sec_ctx_match(pol->security, policy->security) &&
751 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
755 if (policy->priority > pol->priority)
757 } else if (policy->priority >= pol->priority) {
758 newpos = &pol->bydst;
765 hlist_add_behind_rcu(&policy->bydst, newpos);
767 hlist_add_head_rcu(&policy->bydst, chain);
768 __xfrm_policy_link(policy, dir);
770 /* After previous checking, family can either be AF_INET or AF_INET6 */
771 if (policy->family == AF_INET)
772 rt_genid_bump_ipv4(net);
774 rt_genid_bump_ipv6(net);
777 xfrm_policy_requeue(delpol, policy);
778 __xfrm_policy_unlink(delpol, dir);
780 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
781 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
782 policy->curlft.add_time = get_seconds();
783 policy->curlft.use_time = 0;
784 if (!mod_timer(&policy->timer, jiffies + HZ))
785 xfrm_pol_hold(policy);
786 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
789 xfrm_policy_kill(delpol);
790 else if (xfrm_bydst_should_resize(net, dir, NULL))
791 schedule_work(&net->xfrm.policy_hash_work);
795 EXPORT_SYMBOL(xfrm_policy_insert);
797 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
798 int dir, struct xfrm_selector *sel,
799 struct xfrm_sec_ctx *ctx, int delete,
802 struct xfrm_policy *pol, *ret;
803 struct hlist_head *chain;
806 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
807 chain = policy_hash_bysel(net, sel, sel->family, dir);
809 hlist_for_each_entry(pol, chain, bydst) {
810 if (pol->type == type &&
811 (mark & pol->mark.m) == pol->mark.v &&
812 !selector_cmp(sel, &pol->selector) &&
813 xfrm_sec_ctx_match(ctx, pol->security)) {
816 *err = security_xfrm_policy_delete(
819 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
822 __xfrm_policy_unlink(pol, dir);
828 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
831 xfrm_policy_kill(ret);
834 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
836 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
837 int dir, u32 id, int delete, int *err)
839 struct xfrm_policy *pol, *ret;
840 struct hlist_head *chain;
843 if (xfrm_policy_id2dir(id) != dir)
847 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
848 chain = net->xfrm.policy_byidx + idx_hash(net, id);
850 hlist_for_each_entry(pol, chain, byidx) {
851 if (pol->type == type && pol->index == id &&
852 (mark & pol->mark.m) == pol->mark.v) {
855 *err = security_xfrm_policy_delete(
858 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
861 __xfrm_policy_unlink(pol, dir);
867 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
870 xfrm_policy_kill(ret);
873 EXPORT_SYMBOL(xfrm_policy_byid);
875 #ifdef CONFIG_SECURITY_NETWORK_XFRM
877 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
881 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
882 struct xfrm_policy *pol;
885 hlist_for_each_entry(pol,
886 &net->xfrm.policy_inexact[dir], bydst) {
887 if (pol->type != type)
889 err = security_xfrm_policy_delete(pol->security);
891 xfrm_audit_policy_delete(pol, 0, task_valid);
895 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
896 hlist_for_each_entry(pol,
897 net->xfrm.policy_bydst[dir].table + i,
899 if (pol->type != type)
901 err = security_xfrm_policy_delete(
904 xfrm_audit_policy_delete(pol, 0,
915 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
921 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
923 int dir, err = 0, cnt = 0;
925 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
927 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
931 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
932 struct xfrm_policy *pol;
936 hlist_for_each_entry(pol,
937 &net->xfrm.policy_inexact[dir], bydst) {
938 if (pol->type != type)
940 __xfrm_policy_unlink(pol, dir);
941 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
944 xfrm_audit_policy_delete(pol, 1, task_valid);
946 xfrm_policy_kill(pol);
948 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
952 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
954 hlist_for_each_entry(pol,
955 net->xfrm.policy_bydst[dir].table + i,
957 if (pol->type != type)
959 __xfrm_policy_unlink(pol, dir);
960 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
963 xfrm_audit_policy_delete(pol, 1, task_valid);
964 xfrm_policy_kill(pol);
966 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
975 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
978 EXPORT_SYMBOL(xfrm_policy_flush);
980 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
981 int (*func)(struct xfrm_policy *, int, int, void*),
984 struct xfrm_policy *pol;
985 struct xfrm_policy_walk_entry *x;
988 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
989 walk->type != XFRM_POLICY_TYPE_ANY)
992 if (list_empty(&walk->walk.all) && walk->seq != 0)
995 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
996 if (list_empty(&walk->walk.all))
997 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
999 x = list_first_entry(&walk->walk.all,
1000 struct xfrm_policy_walk_entry, all);
1002 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1005 pol = container_of(x, struct xfrm_policy, walk);
1006 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1007 walk->type != pol->type)
1009 error = func(pol, xfrm_policy_id2dir(pol->index),
1012 list_move_tail(&walk->walk.all, &x->all);
1017 if (walk->seq == 0) {
1021 list_del_init(&walk->walk.all);
1023 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1026 EXPORT_SYMBOL(xfrm_policy_walk);
1028 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1030 INIT_LIST_HEAD(&walk->walk.all);
1031 walk->walk.dead = 1;
1035 EXPORT_SYMBOL(xfrm_policy_walk_init);
1037 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1039 if (list_empty(&walk->walk.all))
1042 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1043 list_del(&walk->walk.all);
1044 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1046 EXPORT_SYMBOL(xfrm_policy_walk_done);
1049 * Find policy to apply to this flow.
1051 * Returns 0 if policy found, else an -errno.
1053 static int xfrm_policy_match(const struct xfrm_policy *pol,
1054 const struct flowi *fl,
1055 u8 type, u16 family, int dir)
1057 const struct xfrm_selector *sel = &pol->selector;
1061 if (pol->family != family ||
1062 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1066 match = xfrm_selector_match(sel, fl, family);
1068 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1074 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1075 const struct flowi *fl,
1079 struct xfrm_policy *pol, *ret;
1080 const xfrm_address_t *daddr, *saddr;
1081 struct hlist_head *chain;
1082 unsigned int sequence;
1085 daddr = xfrm_flowi_daddr(fl, family);
1086 saddr = xfrm_flowi_saddr(fl, family);
1087 if (unlikely(!daddr || !saddr))
1093 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1094 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1095 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1099 hlist_for_each_entry_rcu(pol, chain, bydst) {
1100 err = xfrm_policy_match(pol, fl, type, family, dir);
1110 priority = ret->priority;
1114 chain = &net->xfrm.policy_inexact[dir];
1115 hlist_for_each_entry_rcu(pol, chain, bydst) {
1116 if ((pol->priority >= priority) && ret)
1119 err = xfrm_policy_match(pol, fl, type, family, dir);
1133 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1136 if (ret && !xfrm_pol_hold_rcu(ret))
1144 static struct xfrm_policy *
1145 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1147 #ifdef CONFIG_XFRM_SUB_POLICY
1148 struct xfrm_policy *pol;
1150 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1154 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1157 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1158 const struct flowi *fl, u16 family)
1160 struct xfrm_policy *pol;
1164 pol = rcu_dereference(sk->sk_policy[dir]);
1169 if (pol->family != family) {
1174 match = xfrm_selector_match(&pol->selector, fl, family);
1176 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1180 err = security_xfrm_policy_lookup(pol->security,
1184 if (!xfrm_pol_hold_rcu(pol))
1186 } else if (err == -ESRCH) {
1199 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1201 struct net *net = xp_net(pol);
1203 list_add(&pol->walk.all, &net->xfrm.policy_all);
1204 net->xfrm.policy_count[dir]++;
1208 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1211 struct net *net = xp_net(pol);
1213 if (list_empty(&pol->walk.all))
1216 /* Socket policies are not hashed. */
1217 if (!hlist_unhashed(&pol->bydst)) {
1218 hlist_del_rcu(&pol->bydst);
1219 hlist_del(&pol->byidx);
1222 list_del_init(&pol->walk.all);
1223 net->xfrm.policy_count[dir]--;
1228 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1230 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1233 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1235 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1238 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1240 struct net *net = xp_net(pol);
1242 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1243 pol = __xfrm_policy_unlink(pol, dir);
1244 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1246 xfrm_policy_kill(pol);
1251 EXPORT_SYMBOL(xfrm_policy_delete);
1253 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1255 struct net *net = sock_net(sk);
1256 struct xfrm_policy *old_pol;
1258 #ifdef CONFIG_XFRM_SUB_POLICY
1259 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1263 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1264 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1265 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1267 pol->curlft.add_time = get_seconds();
1268 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1269 xfrm_sk_policy_link(pol, dir);
1271 rcu_assign_pointer(sk->sk_policy[dir], pol);
1274 xfrm_policy_requeue(old_pol, pol);
1276 /* Unlinking succeeds always. This is the only function
1277 * allowed to delete or replace socket policy.
1279 xfrm_sk_policy_unlink(old_pol, dir);
1281 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1284 xfrm_policy_kill(old_pol);
1289 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1291 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1292 struct net *net = xp_net(old);
1295 newp->selector = old->selector;
1296 if (security_xfrm_policy_clone(old->security,
1299 return NULL; /* ENOMEM */
1301 newp->lft = old->lft;
1302 newp->curlft = old->curlft;
1303 newp->mark = old->mark;
1304 newp->action = old->action;
1305 newp->flags = old->flags;
1306 newp->xfrm_nr = old->xfrm_nr;
1307 newp->index = old->index;
1308 newp->type = old->type;
1309 newp->family = old->family;
1310 memcpy(newp->xfrm_vec, old->xfrm_vec,
1311 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1312 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1313 xfrm_sk_policy_link(newp, dir);
1314 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1320 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1322 const struct xfrm_policy *p;
1323 struct xfrm_policy *np;
1327 for (i = 0; i < 2; i++) {
1328 p = rcu_dereference(osk->sk_policy[i]);
1330 np = clone_policy(p, i);
1331 if (unlikely(!np)) {
1335 rcu_assign_pointer(sk->sk_policy[i], np);
1343 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1344 xfrm_address_t *remote, unsigned short family, u32 mark)
1347 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1349 if (unlikely(afinfo == NULL))
1351 err = afinfo->get_saddr(net, oif, local, remote, mark);
1356 /* Resolve list of templates for the flow, given policy. */
1359 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1360 struct xfrm_state **xfrm, unsigned short family)
1362 struct net *net = xp_net(policy);
1365 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1366 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1369 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1370 struct xfrm_state *x;
1371 xfrm_address_t *remote = daddr;
1372 xfrm_address_t *local = saddr;
1373 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1375 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1376 tmpl->mode == XFRM_MODE_BEET) {
1377 remote = &tmpl->id.daddr;
1378 local = &tmpl->saddr;
1379 if (xfrm_addr_any(local, tmpl->encap_family)) {
1380 error = xfrm_get_saddr(net, fl->flowi_oif,
1382 tmpl->encap_family, 0);
1389 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1391 if (x && x->km.state == XFRM_STATE_VALID) {
1398 error = (x->km.state == XFRM_STATE_ERROR ?
1401 } else if (error == -ESRCH) {
1405 if (!tmpl->optional)
1411 for (nx--; nx >= 0; nx--)
1412 xfrm_state_put(xfrm[nx]);
1417 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1418 struct xfrm_state **xfrm, unsigned short family)
1420 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1421 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1427 for (i = 0; i < npols; i++) {
1428 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1433 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1441 /* found states are sorted for outbound processing */
1443 xfrm_state_sort(xfrm, tpp, cnx, family);
1448 for (cnx--; cnx >= 0; cnx--)
1449 xfrm_state_put(tpp[cnx]);
1454 static int xfrm_get_tos(const struct flowi *fl, int family)
1456 const struct xfrm_policy_afinfo *afinfo;
1459 afinfo = xfrm_policy_get_afinfo(family);
1463 tos = afinfo->get_tos(fl);
1470 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1472 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1473 struct dst_ops *dst_ops;
1474 struct xfrm_dst *xdst;
1477 return ERR_PTR(-EINVAL);
1481 dst_ops = &net->xfrm.xfrm4_dst_ops;
1483 #if IS_ENABLED(CONFIG_IPV6)
1485 dst_ops = &net->xfrm.xfrm6_dst_ops;
1491 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1494 struct dst_entry *dst = &xdst->u.dst;
1496 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1498 xdst = ERR_PTR(-ENOBUFS);
1505 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1508 const struct xfrm_policy_afinfo *afinfo =
1509 xfrm_policy_get_afinfo(dst->ops->family);
1515 err = afinfo->init_path(path, dst, nfheader_len);
1522 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1523 const struct flowi *fl)
1525 const struct xfrm_policy_afinfo *afinfo =
1526 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1532 err = afinfo->fill_dst(xdst, dev, fl);
1540 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1541 * all the metrics... Shortly, bundle a bundle.
1544 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1545 struct xfrm_state **xfrm, int nx,
1546 const struct flowi *fl,
1547 struct dst_entry *dst)
1549 struct net *net = xp_net(policy);
1550 unsigned long now = jiffies;
1551 struct net_device *dev;
1552 struct xfrm_mode *inner_mode;
1553 struct dst_entry *dst_prev = NULL;
1554 struct dst_entry *dst0 = NULL;
1558 int nfheader_len = 0;
1559 int trailer_len = 0;
1561 int family = policy->selector.family;
1562 xfrm_address_t saddr, daddr;
1564 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1566 tos = xfrm_get_tos(fl, family);
1570 for (; i < nx; i++) {
1571 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1572 struct dst_entry *dst1 = &xdst->u.dst;
1574 err = PTR_ERR(xdst);
1583 /* Ref count is taken during xfrm_alloc_dst()
1584 * No need to do dst_clone() on dst1
1586 dst_prev->child = dst1;
1588 if (xfrm[i]->sel.family == AF_UNSPEC) {
1589 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1590 xfrm_af2proto(family));
1592 err = -EAFNOSUPPORT;
1597 inner_mode = xfrm[i]->inner_mode;
1600 dst_copy_metrics(dst1, dst);
1602 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1603 family = xfrm[i]->props.family;
1604 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1605 &saddr, &daddr, family,
1606 xfrm[i]->props.output_mark);
1613 dst1->xfrm = xfrm[i];
1614 xdst->xfrm_genid = xfrm[i]->genid;
1616 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1617 dst1->flags |= DST_HOST;
1618 dst1->lastuse = now;
1620 dst1->input = dst_discard;
1621 dst1->output = inner_mode->afinfo->output;
1623 dst1->next = dst_prev;
1626 header_len += xfrm[i]->props.header_len;
1627 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1628 nfheader_len += xfrm[i]->props.header_len;
1629 trailer_len += xfrm[i]->props.trailer_len;
1632 dst_prev->child = dst;
1640 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1641 xfrm_init_pmtu(dst_prev);
1643 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1644 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1646 err = xfrm_fill_dst(xdst, dev, fl);
1650 dst_prev->header_len = header_len;
1651 dst_prev->trailer_len = trailer_len;
1652 header_len -= xdst->u.dst.xfrm->props.header_len;
1653 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1661 xfrm_state_put(xfrm[i]);
1664 dst_release_immediate(dst0);
1665 dst0 = ERR_PTR(err);
1669 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1670 struct xfrm_policy **pols,
1671 int *num_pols, int *num_xfrms)
1675 if (*num_pols == 0 || !pols[0]) {
1680 if (IS_ERR(pols[0]))
1681 return PTR_ERR(pols[0]);
1683 *num_xfrms = pols[0]->xfrm_nr;
1685 #ifdef CONFIG_XFRM_SUB_POLICY
1686 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1687 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1688 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1689 XFRM_POLICY_TYPE_MAIN,
1693 if (IS_ERR(pols[1])) {
1694 xfrm_pols_put(pols, *num_pols);
1695 return PTR_ERR(pols[1]);
1698 (*num_xfrms) += pols[1]->xfrm_nr;
1702 for (i = 0; i < *num_pols; i++) {
1703 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1713 static struct xfrm_dst *
1714 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1715 const struct flowi *fl, u16 family,
1716 struct dst_entry *dst_orig)
1718 struct net *net = xp_net(pols[0]);
1719 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1720 struct xfrm_dst *xdst;
1721 struct dst_entry *dst;
1724 /* Try to instantiate a bundle */
1725 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1731 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1732 return ERR_PTR(err);
1735 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1737 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1738 return ERR_CAST(dst);
1741 xdst = (struct xfrm_dst *)dst;
1742 xdst->num_xfrms = err;
1743 xdst->num_pols = num_pols;
1744 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1745 xdst->policy_genid = atomic_read(&pols[0]->genid);
1750 static void xfrm_policy_queue_process(unsigned long arg)
1752 struct sk_buff *skb;
1754 struct dst_entry *dst;
1755 struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1756 struct net *net = xp_net(pol);
1757 struct xfrm_policy_queue *pq = &pol->polq;
1759 struct sk_buff_head list;
1761 spin_lock(&pq->hold_queue.lock);
1762 skb = skb_peek(&pq->hold_queue);
1764 spin_unlock(&pq->hold_queue.lock);
1769 xfrm_decode_session(skb, &fl, dst->ops->family);
1770 spin_unlock(&pq->hold_queue.lock);
1772 dst_hold(dst->path);
1773 dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1777 if (dst->flags & DST_XFRM_QUEUE) {
1780 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1783 pq->timeout = pq->timeout << 1;
1784 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1791 __skb_queue_head_init(&list);
1793 spin_lock(&pq->hold_queue.lock);
1795 skb_queue_splice_init(&pq->hold_queue, &list);
1796 spin_unlock(&pq->hold_queue.lock);
1798 while (!skb_queue_empty(&list)) {
1799 skb = __skb_dequeue(&list);
1801 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1802 dst_hold(skb_dst(skb)->path);
1803 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1811 skb_dst_set(skb, dst);
1813 dst_output(net, skb->sk, skb);
1822 skb_queue_purge(&pq->hold_queue);
1826 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1828 unsigned long sched_next;
1829 struct dst_entry *dst = skb_dst(skb);
1830 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1831 struct xfrm_policy *pol = xdst->pols[0];
1832 struct xfrm_policy_queue *pq = &pol->polq;
1834 if (unlikely(skb_fclone_busy(sk, skb))) {
1839 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1846 spin_lock_bh(&pq->hold_queue.lock);
1849 pq->timeout = XFRM_QUEUE_TMO_MIN;
1851 sched_next = jiffies + pq->timeout;
1853 if (del_timer(&pq->hold_timer)) {
1854 if (time_before(pq->hold_timer.expires, sched_next))
1855 sched_next = pq->hold_timer.expires;
1859 __skb_queue_tail(&pq->hold_queue, skb);
1860 if (!mod_timer(&pq->hold_timer, sched_next))
1863 spin_unlock_bh(&pq->hold_queue.lock);
1868 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1869 struct xfrm_flo *xflo,
1870 const struct flowi *fl,
1875 struct net_device *dev;
1876 struct dst_entry *dst;
1877 struct dst_entry *dst1;
1878 struct xfrm_dst *xdst;
1880 xdst = xfrm_alloc_dst(net, family);
1884 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
1885 net->xfrm.sysctl_larval_drop ||
1889 dst = xflo->dst_orig;
1890 dst1 = &xdst->u.dst;
1894 dst_copy_metrics(dst1, dst);
1896 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1897 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1898 dst1->lastuse = jiffies;
1900 dst1->input = dst_discard;
1901 dst1->output = xdst_queue_output;
1907 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1914 err = xfrm_fill_dst(xdst, dev, fl);
1923 xdst = ERR_PTR(err);
1927 static struct xfrm_dst *
1928 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
1930 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1931 int num_pols = 0, num_xfrms = 0, err;
1932 struct xfrm_dst *xdst;
1934 /* Resolve policies to use if we couldn't get them from
1935 * previous cache entry */
1937 pols[0] = xfrm_policy_lookup(net, fl, family, dir);
1938 err = xfrm_expand_policies(fl, family, pols,
1939 &num_pols, &num_xfrms);
1945 goto make_dummy_bundle;
1947 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
1950 err = PTR_ERR(xdst);
1953 goto make_dummy_bundle;
1954 } else if (xdst == NULL) {
1956 goto make_dummy_bundle;
1962 /* We found policies, but there's no bundles to instantiate:
1963 * either because the policy blocks, has no transformations or
1964 * we could not build template (no xfrm_states).*/
1965 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
1967 xfrm_pols_put(pols, num_pols);
1968 return ERR_CAST(xdst);
1970 xdst->num_pols = num_pols;
1971 xdst->num_xfrms = num_xfrms;
1972 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1977 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1979 xfrm_pols_put(pols, num_pols);
1980 return ERR_PTR(err);
1983 static struct dst_entry *make_blackhole(struct net *net, u16 family,
1984 struct dst_entry *dst_orig)
1986 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1987 struct dst_entry *ret;
1990 dst_release(dst_orig);
1991 return ERR_PTR(-EINVAL);
1993 ret = afinfo->blackhole_route(net, dst_orig);
2000 /* Main function: finds/creates a bundle for given flow.
2002 * At the moment we eat a raw IP route. Mostly to speed up lookups
2003 * on interfaces with disabled IPsec.
2005 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2006 const struct flowi *fl,
2007 const struct sock *sk, int flags)
2009 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2010 struct xfrm_dst *xdst;
2011 struct dst_entry *dst, *route;
2012 u16 family = dst_orig->ops->family;
2013 u8 dir = XFRM_POLICY_OUT;
2014 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2020 sk = sk_const_to_full_sk(sk);
2021 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2023 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2024 err = xfrm_expand_policies(fl, family, pols,
2025 &num_pols, &num_xfrms);
2030 if (num_xfrms <= 0) {
2031 drop_pols = num_pols;
2035 xdst = xfrm_resolve_and_create_bundle(
2040 xfrm_pols_put(pols, num_pols);
2041 err = PTR_ERR(xdst);
2043 } else if (xdst == NULL) {
2045 drop_pols = num_pols;
2049 route = xdst->route;
2054 struct xfrm_flo xflo;
2056 xflo.dst_orig = dst_orig;
2059 /* To accelerate a bit... */
2060 if ((dst_orig->flags & DST_NOXFRM) ||
2061 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2064 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2068 err = PTR_ERR(xdst);
2072 num_pols = xdst->num_pols;
2073 num_xfrms = xdst->num_xfrms;
2074 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2075 route = xdst->route;
2079 if (route == NULL && num_xfrms > 0) {
2080 /* The only case when xfrm_bundle_lookup() returns a
2081 * bundle with null route, is when the template could
2082 * not be resolved. It means policies are there, but
2083 * bundle could not be created, since we don't yet
2084 * have the xfrm_state's. We need to wait for KM to
2085 * negotiate new SA's or bail out with error.*/
2086 if (net->xfrm.sysctl_larval_drop) {
2087 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2094 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2102 if ((flags & XFRM_LOOKUP_ICMP) &&
2103 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2108 for (i = 0; i < num_pols; i++)
2109 pols[i]->curlft.use_time = get_seconds();
2111 if (num_xfrms < 0) {
2112 /* Prohibit the flow */
2113 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2116 } else if (num_xfrms > 0) {
2117 /* Flow transformed */
2118 dst_release(dst_orig);
2120 /* Flow passes untransformed */
2125 xfrm_pols_put(pols, drop_pols);
2126 if (dst && dst->xfrm &&
2127 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2128 dst->flags |= DST_XFRM_TUNNEL;
2132 if (!(flags & XFRM_LOOKUP_ICMP)) {
2140 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2141 dst_release(dst_orig);
2142 xfrm_pols_put(pols, drop_pols);
2143 return ERR_PTR(err);
2145 EXPORT_SYMBOL(xfrm_lookup);
2147 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2148 * Otherwise we may send out blackholed packets.
2150 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2151 const struct flowi *fl,
2152 const struct sock *sk, int flags)
2154 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2155 flags | XFRM_LOOKUP_QUEUE |
2156 XFRM_LOOKUP_KEEP_DST_REF);
2158 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2159 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2162 dst_release(dst_orig);
2166 EXPORT_SYMBOL(xfrm_lookup_route);
2169 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2171 struct xfrm_state *x;
2173 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2175 x = skb->sp->xvec[idx];
2176 if (!x->type->reject)
2178 return x->type->reject(x, skb, fl);
2181 /* When skb is transformed back to its "native" form, we have to
2182 * check policy restrictions. At the moment we make this in maximally
2183 * stupid way. Shame on me. :-) Of course, connected sockets must
2184 * have policy cached at them.
2188 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2189 unsigned short family)
2191 if (xfrm_state_kern(x))
2192 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2193 return x->id.proto == tmpl->id.proto &&
2194 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2195 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2196 x->props.mode == tmpl->mode &&
2197 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2198 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2199 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2200 xfrm_state_addr_cmp(tmpl, x, family));
2204 * 0 or more than 0 is returned when validation is succeeded (either bypass
2205 * because of optional transport mode, or next index of the mathced secpath
2206 * state with the template.
2207 * -1 is returned when no matching template is found.
2208 * Otherwise "-2 - errored_index" is returned.
2211 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2212 unsigned short family)
2216 if (tmpl->optional) {
2217 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2221 for (; idx < sp->len; idx++) {
2222 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2224 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2233 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2234 unsigned int family, int reverse)
2236 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2239 if (unlikely(afinfo == NULL))
2240 return -EAFNOSUPPORT;
2242 afinfo->decode_session(skb, fl, reverse);
2243 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2247 EXPORT_SYMBOL(__xfrm_decode_session);
2249 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2251 for (; k < sp->len; k++) {
2252 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2261 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2262 unsigned short family)
2264 struct net *net = dev_net(skb->dev);
2265 struct xfrm_policy *pol;
2266 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2274 reverse = dir & ~XFRM_POLICY_MASK;
2275 dir &= XFRM_POLICY_MASK;
2277 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2278 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2282 nf_nat_decode_session(skb, &fl, family);
2284 /* First, check used SA against their selectors. */
2288 for (i = skb->sp->len-1; i >= 0; i--) {
2289 struct xfrm_state *x = skb->sp->xvec[i];
2290 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2291 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2298 sk = sk_to_full_sk(sk);
2299 if (sk && sk->sk_policy[dir]) {
2300 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2302 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2308 pol = xfrm_policy_lookup(net, &fl, family, dir);
2311 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2316 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2317 xfrm_secpath_reject(xerr_idx, skb, &fl);
2318 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2324 pol->curlft.use_time = get_seconds();
2328 #ifdef CONFIG_XFRM_SUB_POLICY
2329 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2330 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2334 if (IS_ERR(pols[1])) {
2335 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2338 pols[1]->curlft.use_time = get_seconds();
2344 if (pol->action == XFRM_POLICY_ALLOW) {
2345 struct sec_path *sp;
2346 static struct sec_path dummy;
2347 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2348 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2349 struct xfrm_tmpl **tpp = tp;
2353 if ((sp = skb->sp) == NULL)
2356 for (pi = 0; pi < npols; pi++) {
2357 if (pols[pi] != pol &&
2358 pols[pi]->action != XFRM_POLICY_ALLOW) {
2359 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2362 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2363 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2366 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2367 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2371 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2375 /* For each tunnel xfrm, find the first matching tmpl.
2376 * For each tmpl before that, find corresponding xfrm.
2377 * Order is _important_. Later we will implement
2378 * some barriers, but at the moment barriers
2379 * are implied between each two transformations.
2381 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2382 k = xfrm_policy_ok(tpp[i], sp, k, family);
2385 /* "-2 - errored_index" returned */
2387 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2392 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2393 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2397 xfrm_pols_put(pols, npols);
2400 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2403 xfrm_secpath_reject(xerr_idx, skb, &fl);
2405 xfrm_pols_put(pols, npols);
2408 EXPORT_SYMBOL(__xfrm_policy_check);
2410 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2412 struct net *net = dev_net(skb->dev);
2414 struct dst_entry *dst;
2417 if (xfrm_decode_session(skb, &fl, family) < 0) {
2418 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2423 if (!skb_dst(skb)) {
2424 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2428 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2433 skb_dst_set(skb, dst);
2436 EXPORT_SYMBOL(__xfrm_route_forward);
2438 /* Optimize later using cookies and generation ids. */
2440 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2442 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2443 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2444 * get validated by dst_ops->check on every use. We do this
2445 * because when a normal route referenced by an XFRM dst is
2446 * obsoleted we do not go looking around for all parent
2447 * referencing XFRM dsts so that we can invalidate them. It
2448 * is just too much work. Instead we make the checks here on
2449 * every use. For example:
2451 * XFRM dst A --> IPv4 dst X
2453 * X is the "xdst->route" of A (X is also the "dst->path" of A
2454 * in this example). If X is marked obsolete, "A" will not
2455 * notice. That's what we are validating here via the
2456 * stale_bundle() check.
2458 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2460 * This will force stale_bundle() to fail on any xdst bundle with
2461 * this dst linked in it.
2463 if (dst->obsolete < 0 && !stale_bundle(dst))
2469 static int stale_bundle(struct dst_entry *dst)
2471 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2474 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2476 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2477 dst->dev = dev_net(dev)->loopback_dev;
2482 EXPORT_SYMBOL(xfrm_dst_ifdown);
2484 static void xfrm_link_failure(struct sk_buff *skb)
2486 /* Impossible. Such dst must be popped before reaches point of failure. */
2489 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2492 if (dst->obsolete) {
2500 static void xfrm_init_pmtu(struct dst_entry *dst)
2503 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2504 u32 pmtu, route_mtu_cached;
2506 pmtu = dst_mtu(dst->child);
2507 xdst->child_mtu_cached = pmtu;
2509 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2511 route_mtu_cached = dst_mtu(xdst->route);
2512 xdst->route_mtu_cached = route_mtu_cached;
2514 if (pmtu > route_mtu_cached)
2515 pmtu = route_mtu_cached;
2517 dst_metric_set(dst, RTAX_MTU, pmtu);
2518 } while ((dst = dst->next));
2521 /* Check that the bundle accepts the flow and its components are
2525 static int xfrm_bundle_ok(struct xfrm_dst *first)
2527 struct dst_entry *dst = &first->u.dst;
2528 struct xfrm_dst *last;
2531 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2532 (dst->dev && !netif_running(dst->dev)))
2535 if (dst->flags & DST_XFRM_QUEUE)
2541 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2543 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2545 if (xdst->xfrm_genid != dst->xfrm->genid)
2547 if (xdst->num_pols > 0 &&
2548 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2551 mtu = dst_mtu(dst->child);
2552 if (xdst->child_mtu_cached != mtu) {
2554 xdst->child_mtu_cached = mtu;
2557 if (!dst_check(xdst->route, xdst->route_cookie))
2559 mtu = dst_mtu(xdst->route);
2560 if (xdst->route_mtu_cached != mtu) {
2562 xdst->route_mtu_cached = mtu;
2566 } while (dst->xfrm);
2571 mtu = last->child_mtu_cached;
2575 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2576 if (mtu > last->route_mtu_cached)
2577 mtu = last->route_mtu_cached;
2578 dst_metric_set(dst, RTAX_MTU, mtu);
2583 last = (struct xfrm_dst *)last->u.dst.next;
2584 last->child_mtu_cached = mtu;
2590 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2592 return dst_metric_advmss(dst->path);
2595 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2597 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2599 return mtu ? : dst_mtu(dst->path);
2602 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2605 const struct dst_entry *path = dst->path;
2607 for (; dst != path; dst = dst->child) {
2608 const struct xfrm_state *xfrm = dst->xfrm;
2610 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2612 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2613 daddr = xfrm->coaddr;
2614 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2615 daddr = &xfrm->id.daddr;
2620 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2621 struct sk_buff *skb,
2624 const struct dst_entry *path = dst->path;
2627 daddr = xfrm_get_dst_nexthop(dst, daddr);
2628 return path->ops->neigh_lookup(path, skb, daddr);
2631 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2633 const struct dst_entry *path = dst->path;
2635 daddr = xfrm_get_dst_nexthop(dst, daddr);
2636 path->ops->confirm_neigh(path, daddr);
2639 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2643 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2644 return -EAFNOSUPPORT;
2646 spin_lock(&xfrm_policy_afinfo_lock);
2647 if (unlikely(xfrm_policy_afinfo[family] != NULL))
2650 struct dst_ops *dst_ops = afinfo->dst_ops;
2651 if (likely(dst_ops->kmem_cachep == NULL))
2652 dst_ops->kmem_cachep = xfrm_dst_cache;
2653 if (likely(dst_ops->check == NULL))
2654 dst_ops->check = xfrm_dst_check;
2655 if (likely(dst_ops->default_advmss == NULL))
2656 dst_ops->default_advmss = xfrm_default_advmss;
2657 if (likely(dst_ops->mtu == NULL))
2658 dst_ops->mtu = xfrm_mtu;
2659 if (likely(dst_ops->negative_advice == NULL))
2660 dst_ops->negative_advice = xfrm_negative_advice;
2661 if (likely(dst_ops->link_failure == NULL))
2662 dst_ops->link_failure = xfrm_link_failure;
2663 if (likely(dst_ops->neigh_lookup == NULL))
2664 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2665 if (likely(!dst_ops->confirm_neigh))
2666 dst_ops->confirm_neigh = xfrm_confirm_neigh;
2667 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2669 spin_unlock(&xfrm_policy_afinfo_lock);
2673 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2675 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2677 struct dst_ops *dst_ops = afinfo->dst_ops;
2680 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2681 if (xfrm_policy_afinfo[i] != afinfo)
2683 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2689 dst_ops->kmem_cachep = NULL;
2690 dst_ops->check = NULL;
2691 dst_ops->negative_advice = NULL;
2692 dst_ops->link_failure = NULL;
2694 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2696 #ifdef CONFIG_XFRM_STATISTICS
2697 static int __net_init xfrm_statistics_init(struct net *net)
2700 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2701 if (!net->mib.xfrm_statistics)
2703 rv = xfrm_proc_init(net);
2705 free_percpu(net->mib.xfrm_statistics);
2709 static void xfrm_statistics_fini(struct net *net)
2711 xfrm_proc_fini(net);
2712 free_percpu(net->mib.xfrm_statistics);
2715 static int __net_init xfrm_statistics_init(struct net *net)
2720 static void xfrm_statistics_fini(struct net *net)
2725 static int __net_init xfrm_policy_init(struct net *net)
2727 unsigned int hmask, sz;
2730 if (net_eq(net, &init_net))
2731 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2732 sizeof(struct xfrm_dst),
2733 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2737 sz = (hmask+1) * sizeof(struct hlist_head);
2739 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2740 if (!net->xfrm.policy_byidx)
2742 net->xfrm.policy_idx_hmask = hmask;
2744 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2745 struct xfrm_policy_hash *htab;
2747 net->xfrm.policy_count[dir] = 0;
2748 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2749 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2751 htab = &net->xfrm.policy_bydst[dir];
2752 htab->table = xfrm_hash_alloc(sz);
2755 htab->hmask = hmask;
2761 net->xfrm.policy_hthresh.lbits4 = 32;
2762 net->xfrm.policy_hthresh.rbits4 = 32;
2763 net->xfrm.policy_hthresh.lbits6 = 128;
2764 net->xfrm.policy_hthresh.rbits6 = 128;
2766 seqlock_init(&net->xfrm.policy_hthresh.lock);
2768 INIT_LIST_HEAD(&net->xfrm.policy_all);
2769 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2770 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2771 if (net_eq(net, &init_net))
2776 for (dir--; dir >= 0; dir--) {
2777 struct xfrm_policy_hash *htab;
2779 htab = &net->xfrm.policy_bydst[dir];
2780 xfrm_hash_free(htab->table, sz);
2782 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2787 static void xfrm_policy_fini(struct net *net)
2792 flush_work(&net->xfrm.policy_hash_work);
2793 #ifdef CONFIG_XFRM_SUB_POLICY
2794 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2796 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2798 WARN_ON(!list_empty(&net->xfrm.policy_all));
2800 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2801 struct xfrm_policy_hash *htab;
2803 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2805 htab = &net->xfrm.policy_bydst[dir];
2806 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2807 WARN_ON(!hlist_empty(htab->table));
2808 xfrm_hash_free(htab->table, sz);
2811 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2812 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2813 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2816 static int __net_init xfrm_net_init(struct net *net)
2820 /* Initialize the per-net locks here */
2821 spin_lock_init(&net->xfrm.xfrm_state_lock);
2822 spin_lock_init(&net->xfrm.xfrm_policy_lock);
2823 mutex_init(&net->xfrm.xfrm_cfg_mutex);
2825 rv = xfrm_statistics_init(net);
2827 goto out_statistics;
2828 rv = xfrm_state_init(net);
2831 rv = xfrm_policy_init(net);
2834 rv = xfrm_sysctl_init(net);
2841 xfrm_policy_fini(net);
2843 xfrm_state_fini(net);
2845 xfrm_statistics_fini(net);
2850 static void __net_exit xfrm_net_exit(struct net *net)
2852 xfrm_sysctl_fini(net);
2853 xfrm_policy_fini(net);
2854 xfrm_state_fini(net);
2855 xfrm_statistics_fini(net);
2858 static struct pernet_operations __net_initdata xfrm_net_ops = {
2859 .init = xfrm_net_init,
2860 .exit = xfrm_net_exit,
2863 void __init xfrm_init(void)
2865 register_pernet_subsys(&xfrm_net_ops);
2866 seqcount_init(&xfrm_policy_hash_generation);
2870 #ifdef CONFIG_AUDITSYSCALL
2871 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2872 struct audit_buffer *audit_buf)
2874 struct xfrm_sec_ctx *ctx = xp->security;
2875 struct xfrm_selector *sel = &xp->selector;
2878 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2879 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2881 switch (sel->family) {
2883 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2884 if (sel->prefixlen_s != 32)
2885 audit_log_format(audit_buf, " src_prefixlen=%d",
2887 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2888 if (sel->prefixlen_d != 32)
2889 audit_log_format(audit_buf, " dst_prefixlen=%d",
2893 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2894 if (sel->prefixlen_s != 128)
2895 audit_log_format(audit_buf, " src_prefixlen=%d",
2897 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2898 if (sel->prefixlen_d != 128)
2899 audit_log_format(audit_buf, " dst_prefixlen=%d",
2905 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
2907 struct audit_buffer *audit_buf;
2909 audit_buf = xfrm_audit_start("SPD-add");
2910 if (audit_buf == NULL)
2912 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2913 audit_log_format(audit_buf, " res=%u", result);
2914 xfrm_audit_common_policyinfo(xp, audit_buf);
2915 audit_log_end(audit_buf);
2917 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2919 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2922 struct audit_buffer *audit_buf;
2924 audit_buf = xfrm_audit_start("SPD-delete");
2925 if (audit_buf == NULL)
2927 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2928 audit_log_format(audit_buf, " res=%u", result);
2929 xfrm_audit_common_policyinfo(xp, audit_buf);
2930 audit_log_end(audit_buf);
2932 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2935 #ifdef CONFIG_XFRM_MIGRATE
2936 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
2937 const struct xfrm_selector *sel_tgt)
2939 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2940 if (sel_tgt->family == sel_cmp->family &&
2941 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
2943 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
2945 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2946 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2950 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2957 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
2958 u8 dir, u8 type, struct net *net)
2960 struct xfrm_policy *pol, *ret = NULL;
2961 struct hlist_head *chain;
2964 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2965 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
2966 hlist_for_each_entry(pol, chain, bydst) {
2967 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2968 pol->type == type) {
2970 priority = ret->priority;
2974 chain = &net->xfrm.policy_inexact[dir];
2975 hlist_for_each_entry(pol, chain, bydst) {
2976 if ((pol->priority >= priority) && ret)
2979 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2980 pol->type == type) {
2988 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2993 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2997 if (t->mode == m->mode && t->id.proto == m->proto &&
2998 (m->reqid == 0 || t->reqid == m->reqid)) {
3000 case XFRM_MODE_TUNNEL:
3001 case XFRM_MODE_BEET:
3002 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3004 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3009 case XFRM_MODE_TRANSPORT:
3010 /* in case of transport mode, template does not store
3011 any IP addresses, hence we just compare mode and
3022 /* update endpoint address(es) of template(s) */
3023 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3024 struct xfrm_migrate *m, int num_migrate)
3026 struct xfrm_migrate *mp;
3029 write_lock_bh(&pol->lock);
3030 if (unlikely(pol->walk.dead)) {
3031 /* target policy has been deleted */
3032 write_unlock_bh(&pol->lock);
3036 for (i = 0; i < pol->xfrm_nr; i++) {
3037 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3038 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3041 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3042 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3044 /* update endpoints */
3045 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3046 sizeof(pol->xfrm_vec[i].id.daddr));
3047 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3048 sizeof(pol->xfrm_vec[i].saddr));
3049 pol->xfrm_vec[i].encap_family = mp->new_family;
3051 atomic_inc(&pol->genid);
3055 write_unlock_bh(&pol->lock);
3063 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3067 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3070 for (i = 0; i < num_migrate; i++) {
3071 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3072 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3075 /* check if there is any duplicated entry */
3076 for (j = i + 1; j < num_migrate; j++) {
3077 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3078 sizeof(m[i].old_daddr)) &&
3079 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3080 sizeof(m[i].old_saddr)) &&
3081 m[i].proto == m[j].proto &&
3082 m[i].mode == m[j].mode &&
3083 m[i].reqid == m[j].reqid &&
3084 m[i].old_family == m[j].old_family)
3092 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3093 struct xfrm_migrate *m, int num_migrate,
3094 struct xfrm_kmaddress *k, struct net *net,
3095 struct xfrm_encap_tmpl *encap)
3097 int i, err, nx_cur = 0, nx_new = 0;
3098 struct xfrm_policy *pol = NULL;
3099 struct xfrm_state *x, *xc;
3100 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3101 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3102 struct xfrm_migrate *mp;
3104 /* Stage 0 - sanity checks */
3105 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3108 if (dir >= XFRM_POLICY_MAX) {
3113 /* Stage 1 - find policy */
3114 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3119 /* Stage 2 - find and update state(s) */
3120 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3121 if ((x = xfrm_migrate_state_find(mp, net))) {
3124 xc = xfrm_state_migrate(x, mp, encap);
3135 /* Stage 3 - update policy */
3136 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3139 /* Stage 4 - delete old state(s) */
3141 xfrm_states_put(x_cur, nx_cur);
3142 xfrm_states_delete(x_cur, nx_cur);
3145 /* Stage 5 - announce */
3146 km_migrate(sel, dir, type, m, num_migrate, k, encap);
3158 xfrm_states_put(x_cur, nx_cur);
3160 xfrm_states_delete(x_new, nx_new);
3164 EXPORT_SYMBOL(xfrm_migrate);