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 __read_mostly;
55 static __read_mostly seqcount_t xfrm_policy_hash_generation;
57 static void xfrm_init_pmtu(struct dst_entry *dst);
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(unsigned long arg);
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(unsigned long data)
184 struct xfrm_policy *xp = (struct xfrm_policy *)data;
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 setup_timer(&policy->timer, xfrm_policy_timer,
271 (unsigned long)policy);
272 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
273 (unsigned long)policy);
277 EXPORT_SYMBOL(xfrm_policy_alloc);
279 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
281 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
283 security_xfrm_policy_free(policy->security);
287 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
289 void xfrm_policy_destroy(struct xfrm_policy *policy)
291 BUG_ON(!policy->walk.dead);
293 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
296 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
298 EXPORT_SYMBOL(xfrm_policy_destroy);
300 /* Rule must be locked. Release descendant resources, announce
301 * entry dead. The rule must be unlinked from lists to the moment.
304 static void xfrm_policy_kill(struct xfrm_policy *policy)
306 policy->walk.dead = 1;
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 (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 xfrm_policy_cache_flush();
981 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
984 EXPORT_SYMBOL(xfrm_policy_flush);
986 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
987 int (*func)(struct xfrm_policy *, int, int, void*),
990 struct xfrm_policy *pol;
991 struct xfrm_policy_walk_entry *x;
994 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
995 walk->type != XFRM_POLICY_TYPE_ANY)
998 if (list_empty(&walk->walk.all) && walk->seq != 0)
1001 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1002 if (list_empty(&walk->walk.all))
1003 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1005 x = list_first_entry(&walk->walk.all,
1006 struct xfrm_policy_walk_entry, all);
1008 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1011 pol = container_of(x, struct xfrm_policy, walk);
1012 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1013 walk->type != pol->type)
1015 error = func(pol, xfrm_policy_id2dir(pol->index),
1018 list_move_tail(&walk->walk.all, &x->all);
1023 if (walk->seq == 0) {
1027 list_del_init(&walk->walk.all);
1029 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1032 EXPORT_SYMBOL(xfrm_policy_walk);
1034 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1036 INIT_LIST_HEAD(&walk->walk.all);
1037 walk->walk.dead = 1;
1041 EXPORT_SYMBOL(xfrm_policy_walk_init);
1043 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1045 if (list_empty(&walk->walk.all))
1048 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1049 list_del(&walk->walk.all);
1050 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1052 EXPORT_SYMBOL(xfrm_policy_walk_done);
1055 * Find policy to apply to this flow.
1057 * Returns 0 if policy found, else an -errno.
1059 static int xfrm_policy_match(const struct xfrm_policy *pol,
1060 const struct flowi *fl,
1061 u8 type, u16 family, int dir)
1063 const struct xfrm_selector *sel = &pol->selector;
1067 if (pol->family != family ||
1068 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1072 match = xfrm_selector_match(sel, fl, family);
1074 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1080 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1081 const struct flowi *fl,
1085 struct xfrm_policy *pol, *ret;
1086 const xfrm_address_t *daddr, *saddr;
1087 struct hlist_head *chain;
1088 unsigned int sequence;
1091 daddr = xfrm_flowi_daddr(fl, family);
1092 saddr = xfrm_flowi_saddr(fl, family);
1093 if (unlikely(!daddr || !saddr))
1099 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1100 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1101 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1105 hlist_for_each_entry_rcu(pol, chain, bydst) {
1106 err = xfrm_policy_match(pol, fl, type, family, dir);
1116 priority = ret->priority;
1120 chain = &net->xfrm.policy_inexact[dir];
1121 hlist_for_each_entry_rcu(pol, chain, bydst) {
1122 if ((pol->priority >= priority) && ret)
1125 err = xfrm_policy_match(pol, fl, type, family, dir);
1139 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1142 if (ret && !xfrm_pol_hold_rcu(ret))
1150 static struct xfrm_policy *
1151 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1153 #ifdef CONFIG_XFRM_SUB_POLICY
1154 struct xfrm_policy *pol;
1156 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1160 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1163 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1164 const struct flowi *fl, u16 family)
1166 struct xfrm_policy *pol;
1170 pol = rcu_dereference(sk->sk_policy[dir]);
1172 bool 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 = xp_net(pol);
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 memcpy(newp->xfrm_vec, old->xfrm_vec,
1310 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1311 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1312 xfrm_sk_policy_link(newp, dir);
1313 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1319 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1321 const struct xfrm_policy *p;
1322 struct xfrm_policy *np;
1326 for (i = 0; i < 2; i++) {
1327 p = rcu_dereference(osk->sk_policy[i]);
1329 np = clone_policy(p, i);
1330 if (unlikely(!np)) {
1334 rcu_assign_pointer(sk->sk_policy[i], np);
1342 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1343 xfrm_address_t *remote, unsigned short family, u32 mark)
1346 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1348 if (unlikely(afinfo == NULL))
1350 err = afinfo->get_saddr(net, oif, local, remote, mark);
1355 /* Resolve list of templates for the flow, given policy. */
1358 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1359 struct xfrm_state **xfrm, unsigned short family)
1361 struct net *net = xp_net(policy);
1364 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1365 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1368 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1369 struct xfrm_state *x;
1370 xfrm_address_t *remote = daddr;
1371 xfrm_address_t *local = saddr;
1372 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1374 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1375 tmpl->mode == XFRM_MODE_BEET) {
1376 remote = &tmpl->id.daddr;
1377 local = &tmpl->saddr;
1378 if (xfrm_addr_any(local, tmpl->encap_family)) {
1379 error = xfrm_get_saddr(net, fl->flowi_oif,
1381 tmpl->encap_family, 0);
1388 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1390 if (x && x->km.state == XFRM_STATE_VALID) {
1397 error = (x->km.state == XFRM_STATE_ERROR ?
1400 } else if (error == -ESRCH) {
1404 if (!tmpl->optional)
1410 for (nx--; nx >= 0; nx--)
1411 xfrm_state_put(xfrm[nx]);
1416 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1417 struct xfrm_state **xfrm, unsigned short family)
1419 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1420 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1426 for (i = 0; i < npols; i++) {
1427 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1432 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1440 /* found states are sorted for outbound processing */
1442 xfrm_state_sort(xfrm, tpp, cnx, family);
1447 for (cnx--; cnx >= 0; cnx--)
1448 xfrm_state_put(tpp[cnx]);
1453 static int xfrm_get_tos(const struct flowi *fl, int family)
1455 const struct xfrm_policy_afinfo *afinfo;
1458 afinfo = xfrm_policy_get_afinfo(family);
1459 tos = afinfo ? afinfo->get_tos(fl) : 0;
1466 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1468 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1469 struct dst_ops *dst_ops;
1470 struct xfrm_dst *xdst;
1473 return ERR_PTR(-EINVAL);
1477 dst_ops = &net->xfrm.xfrm4_dst_ops;
1479 #if IS_ENABLED(CONFIG_IPV6)
1481 dst_ops = &net->xfrm.xfrm6_dst_ops;
1487 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1490 struct dst_entry *dst = &xdst->u.dst;
1492 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1494 xdst = ERR_PTR(-ENOBUFS);
1501 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1504 const struct xfrm_policy_afinfo *afinfo =
1505 xfrm_policy_get_afinfo(dst->ops->family);
1511 err = afinfo->init_path(path, dst, nfheader_len);
1518 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1519 const struct flowi *fl)
1521 const struct xfrm_policy_afinfo *afinfo =
1522 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1528 err = afinfo->fill_dst(xdst, dev, fl);
1536 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1537 * all the metrics... Shortly, bundle a bundle.
1540 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1541 struct xfrm_state **xfrm, int nx,
1542 const struct flowi *fl,
1543 struct dst_entry *dst)
1545 struct net *net = xp_net(policy);
1546 unsigned long now = jiffies;
1547 struct net_device *dev;
1548 struct xfrm_mode *inner_mode;
1549 struct dst_entry *dst_prev = NULL;
1550 struct dst_entry *dst0 = NULL;
1554 int nfheader_len = 0;
1555 int trailer_len = 0;
1557 int family = policy->selector.family;
1558 xfrm_address_t saddr, daddr;
1560 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1562 tos = xfrm_get_tos(fl, family);
1566 for (; i < nx; i++) {
1567 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1568 struct dst_entry *dst1 = &xdst->u.dst;
1570 err = PTR_ERR(xdst);
1579 /* Ref count is taken during xfrm_alloc_dst()
1580 * No need to do dst_clone() on dst1
1582 dst_prev->child = dst1;
1584 if (xfrm[i]->sel.family == AF_UNSPEC) {
1585 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1586 xfrm_af2proto(family));
1588 err = -EAFNOSUPPORT;
1593 inner_mode = xfrm[i]->inner_mode;
1596 dst_copy_metrics(dst1, dst);
1598 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1599 family = xfrm[i]->props.family;
1600 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1601 &saddr, &daddr, family,
1602 xfrm[i]->props.output_mark);
1609 dst1->xfrm = xfrm[i];
1610 xdst->xfrm_genid = xfrm[i]->genid;
1612 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1613 dst1->flags |= DST_HOST;
1614 dst1->lastuse = now;
1616 dst1->input = dst_discard;
1617 dst1->output = inner_mode->afinfo->output;
1619 dst1->next = dst_prev;
1622 header_len += xfrm[i]->props.header_len;
1623 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1624 nfheader_len += xfrm[i]->props.header_len;
1625 trailer_len += xfrm[i]->props.trailer_len;
1628 dst_prev->child = dst;
1636 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1637 xfrm_init_pmtu(dst_prev);
1639 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1640 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1642 err = xfrm_fill_dst(xdst, dev, fl);
1646 dst_prev->header_len = header_len;
1647 dst_prev->trailer_len = trailer_len;
1648 header_len -= xdst->u.dst.xfrm->props.header_len;
1649 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1657 xfrm_state_put(xfrm[i]);
1660 dst_release_immediate(dst0);
1661 dst0 = ERR_PTR(err);
1665 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1666 struct xfrm_policy **pols,
1667 int *num_pols, int *num_xfrms)
1671 if (*num_pols == 0 || !pols[0]) {
1676 if (IS_ERR(pols[0]))
1677 return PTR_ERR(pols[0]);
1679 *num_xfrms = pols[0]->xfrm_nr;
1681 #ifdef CONFIG_XFRM_SUB_POLICY
1682 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1683 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1684 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1685 XFRM_POLICY_TYPE_MAIN,
1689 if (IS_ERR(pols[1])) {
1690 xfrm_pols_put(pols, *num_pols);
1691 return PTR_ERR(pols[1]);
1694 (*num_xfrms) += pols[1]->xfrm_nr;
1698 for (i = 0; i < *num_pols; i++) {
1699 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1709 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1711 this_cpu_write(xfrm_last_dst, xdst);
1713 dst_release(&old->u.dst);
1716 static void __xfrm_pcpu_work_fn(void)
1718 struct xfrm_dst *old;
1720 old = this_cpu_read(xfrm_last_dst);
1721 if (old && !xfrm_bundle_ok(old))
1722 xfrm_last_dst_update(NULL, old);
1725 static void xfrm_pcpu_work_fn(struct work_struct *work)
1729 __xfrm_pcpu_work_fn();
1734 void xfrm_policy_cache_flush(void)
1736 struct xfrm_dst *old;
1742 for_each_possible_cpu(cpu) {
1743 old = per_cpu(xfrm_last_dst, cpu);
1744 if (old && !xfrm_bundle_ok(old)) {
1745 if (smp_processor_id() == cpu) {
1746 __xfrm_pcpu_work_fn();
1762 for_each_possible_cpu(cpu) {
1763 bool bundle_release;
1766 old = per_cpu(xfrm_last_dst, cpu);
1767 bundle_release = old && !xfrm_bundle_ok(old);
1770 if (!bundle_release)
1773 if (cpu_online(cpu)) {
1774 schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1779 old = per_cpu(xfrm_last_dst, cpu);
1780 if (old && !xfrm_bundle_ok(old)) {
1781 per_cpu(xfrm_last_dst, cpu) = NULL;
1782 dst_release(&old->u.dst);
1790 static bool xfrm_pol_dead(struct xfrm_dst *xdst)
1792 unsigned int num_pols = xdst->num_pols;
1793 unsigned int pol_dead = 0, i;
1795 for (i = 0; i < num_pols; i++)
1796 pol_dead |= xdst->pols[i]->walk.dead;
1798 /* Mark DST_OBSOLETE_DEAD to fail the next xfrm_dst_check() */
1800 xdst->u.dst.obsolete = DST_OBSOLETE_DEAD;
1805 static struct xfrm_dst *
1806 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1807 const struct flowi *fl, u16 family,
1808 struct dst_entry *dst_orig)
1810 struct net *net = xp_net(pols[0]);
1811 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1812 struct xfrm_dst *xdst, *old;
1813 struct dst_entry *dst;
1816 xdst = this_cpu_read(xfrm_last_dst);
1818 xdst->u.dst.dev == dst_orig->dev &&
1819 xdst->num_pols == num_pols &&
1820 !xfrm_pol_dead(xdst) &&
1821 memcmp(xdst->pols, pols,
1822 sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1823 xfrm_bundle_ok(xdst)) {
1824 dst_hold(&xdst->u.dst);
1829 /* Try to instantiate a bundle */
1830 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1832 if (err != 0 && err != -EAGAIN)
1833 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1834 return ERR_PTR(err);
1837 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1839 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1840 return ERR_CAST(dst);
1843 xdst = (struct xfrm_dst *)dst;
1844 xdst->num_xfrms = err;
1845 xdst->num_pols = num_pols;
1846 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1847 xdst->policy_genid = atomic_read(&pols[0]->genid);
1849 atomic_set(&xdst->u.dst.__refcnt, 2);
1850 xfrm_last_dst_update(xdst, old);
1855 static void xfrm_policy_queue_process(unsigned long arg)
1857 struct sk_buff *skb;
1859 struct dst_entry *dst;
1860 struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1861 struct net *net = xp_net(pol);
1862 struct xfrm_policy_queue *pq = &pol->polq;
1864 struct sk_buff_head list;
1866 spin_lock(&pq->hold_queue.lock);
1867 skb = skb_peek(&pq->hold_queue);
1869 spin_unlock(&pq->hold_queue.lock);
1874 xfrm_decode_session(skb, &fl, dst->ops->family);
1875 spin_unlock(&pq->hold_queue.lock);
1877 dst_hold(dst->path);
1878 dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1882 if (dst->flags & DST_XFRM_QUEUE) {
1885 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1888 pq->timeout = pq->timeout << 1;
1889 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1896 __skb_queue_head_init(&list);
1898 spin_lock(&pq->hold_queue.lock);
1900 skb_queue_splice_init(&pq->hold_queue, &list);
1901 spin_unlock(&pq->hold_queue.lock);
1903 while (!skb_queue_empty(&list)) {
1904 skb = __skb_dequeue(&list);
1906 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1907 dst_hold(skb_dst(skb)->path);
1908 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1916 skb_dst_set(skb, dst);
1918 dst_output(net, skb->sk, skb);
1927 skb_queue_purge(&pq->hold_queue);
1931 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1933 unsigned long sched_next;
1934 struct dst_entry *dst = skb_dst(skb);
1935 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1936 struct xfrm_policy *pol = xdst->pols[0];
1937 struct xfrm_policy_queue *pq = &pol->polq;
1939 if (unlikely(skb_fclone_busy(sk, skb))) {
1944 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1951 spin_lock_bh(&pq->hold_queue.lock);
1954 pq->timeout = XFRM_QUEUE_TMO_MIN;
1956 sched_next = jiffies + pq->timeout;
1958 if (del_timer(&pq->hold_timer)) {
1959 if (time_before(pq->hold_timer.expires, sched_next))
1960 sched_next = pq->hold_timer.expires;
1964 __skb_queue_tail(&pq->hold_queue, skb);
1965 if (!mod_timer(&pq->hold_timer, sched_next))
1968 spin_unlock_bh(&pq->hold_queue.lock);
1973 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1974 struct xfrm_flo *xflo,
1975 const struct flowi *fl,
1980 struct net_device *dev;
1981 struct dst_entry *dst;
1982 struct dst_entry *dst1;
1983 struct xfrm_dst *xdst;
1985 xdst = xfrm_alloc_dst(net, family);
1989 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
1990 net->xfrm.sysctl_larval_drop ||
1994 dst = xflo->dst_orig;
1995 dst1 = &xdst->u.dst;
1999 dst_copy_metrics(dst1, dst);
2001 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2002 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2003 dst1->lastuse = jiffies;
2005 dst1->input = dst_discard;
2006 dst1->output = xdst_queue_output;
2012 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2019 err = xfrm_fill_dst(xdst, dev, fl);
2028 xdst = ERR_PTR(err);
2032 static struct xfrm_dst *
2033 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2035 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2036 int num_pols = 0, num_xfrms = 0, err;
2037 struct xfrm_dst *xdst;
2039 /* Resolve policies to use if we couldn't get them from
2040 * previous cache entry */
2042 pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2043 err = xfrm_expand_policies(fl, family, pols,
2044 &num_pols, &num_xfrms);
2050 goto make_dummy_bundle;
2052 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2055 err = PTR_ERR(xdst);
2058 goto make_dummy_bundle;
2059 } else if (xdst == NULL) {
2061 goto make_dummy_bundle;
2067 /* We found policies, but there's no bundles to instantiate:
2068 * either because the policy blocks, has no transformations or
2069 * we could not build template (no xfrm_states).*/
2070 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2072 xfrm_pols_put(pols, num_pols);
2073 return ERR_CAST(xdst);
2075 xdst->num_pols = num_pols;
2076 xdst->num_xfrms = num_xfrms;
2077 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2082 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2084 xfrm_pols_put(pols, num_pols);
2085 return ERR_PTR(err);
2088 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2089 struct dst_entry *dst_orig)
2091 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2092 struct dst_entry *ret;
2095 dst_release(dst_orig);
2096 return ERR_PTR(-EINVAL);
2098 ret = afinfo->blackhole_route(net, dst_orig);
2105 /* Main function: finds/creates a bundle for given flow.
2107 * At the moment we eat a raw IP route. Mostly to speed up lookups
2108 * on interfaces with disabled IPsec.
2110 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2111 const struct flowi *fl,
2112 const struct sock *sk, int flags)
2114 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2115 struct xfrm_dst *xdst;
2116 struct dst_entry *dst, *route;
2117 u16 family = dst_orig->ops->family;
2118 u8 dir = XFRM_POLICY_OUT;
2119 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2125 sk = sk_const_to_full_sk(sk);
2126 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2128 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2129 err = xfrm_expand_policies(fl, family, pols,
2130 &num_pols, &num_xfrms);
2135 if (num_xfrms <= 0) {
2136 drop_pols = num_pols;
2140 xdst = xfrm_resolve_and_create_bundle(
2144 xfrm_pols_put(pols, num_pols);
2145 err = PTR_ERR(xdst);
2147 } else if (xdst == NULL) {
2149 drop_pols = num_pols;
2153 route = xdst->route;
2158 struct xfrm_flo xflo;
2160 xflo.dst_orig = dst_orig;
2163 /* To accelerate a bit... */
2164 if ((dst_orig->flags & DST_NOXFRM) ||
2165 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2168 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2172 err = PTR_ERR(xdst);
2176 num_pols = xdst->num_pols;
2177 num_xfrms = xdst->num_xfrms;
2178 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2179 route = xdst->route;
2183 if (route == NULL && num_xfrms > 0) {
2184 /* The only case when xfrm_bundle_lookup() returns a
2185 * bundle with null route, is when the template could
2186 * not be resolved. It means policies are there, but
2187 * bundle could not be created, since we don't yet
2188 * have the xfrm_state's. We need to wait for KM to
2189 * negotiate new SA's or bail out with error.*/
2190 if (net->xfrm.sysctl_larval_drop) {
2191 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2198 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2206 if ((flags & XFRM_LOOKUP_ICMP) &&
2207 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2212 for (i = 0; i < num_pols; i++)
2213 pols[i]->curlft.use_time = get_seconds();
2215 if (num_xfrms < 0) {
2216 /* Prohibit the flow */
2217 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2220 } else if (num_xfrms > 0) {
2221 /* Flow transformed */
2222 dst_release(dst_orig);
2224 /* Flow passes untransformed */
2229 xfrm_pols_put(pols, drop_pols);
2230 if (dst && dst->xfrm &&
2231 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2232 dst->flags |= DST_XFRM_TUNNEL;
2236 if (!(flags & XFRM_LOOKUP_ICMP)) {
2244 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2245 dst_release(dst_orig);
2246 xfrm_pols_put(pols, drop_pols);
2247 return ERR_PTR(err);
2249 EXPORT_SYMBOL(xfrm_lookup);
2251 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2252 * Otherwise we may send out blackholed packets.
2254 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2255 const struct flowi *fl,
2256 const struct sock *sk, int flags)
2258 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2259 flags | XFRM_LOOKUP_QUEUE |
2260 XFRM_LOOKUP_KEEP_DST_REF);
2262 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2263 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2267 EXPORT_SYMBOL(xfrm_lookup_route);
2270 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2272 struct xfrm_state *x;
2274 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2276 x = skb->sp->xvec[idx];
2277 if (!x->type->reject)
2279 return x->type->reject(x, skb, fl);
2282 /* When skb is transformed back to its "native" form, we have to
2283 * check policy restrictions. At the moment we make this in maximally
2284 * stupid way. Shame on me. :-) Of course, connected sockets must
2285 * have policy cached at them.
2289 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2290 unsigned short family)
2292 if (xfrm_state_kern(x))
2293 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2294 return x->id.proto == tmpl->id.proto &&
2295 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2296 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2297 x->props.mode == tmpl->mode &&
2298 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2299 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2300 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2301 xfrm_state_addr_cmp(tmpl, x, family));
2305 * 0 or more than 0 is returned when validation is succeeded (either bypass
2306 * because of optional transport mode, or next index of the mathced secpath
2307 * state with the template.
2308 * -1 is returned when no matching template is found.
2309 * Otherwise "-2 - errored_index" is returned.
2312 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2313 unsigned short family)
2317 if (tmpl->optional) {
2318 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2322 for (; idx < sp->len; idx++) {
2323 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2325 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2334 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2335 unsigned int family, int reverse)
2337 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2340 if (unlikely(afinfo == NULL))
2341 return -EAFNOSUPPORT;
2343 afinfo->decode_session(skb, fl, reverse);
2344 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2348 EXPORT_SYMBOL(__xfrm_decode_session);
2350 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2352 for (; k < sp->len; k++) {
2353 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2362 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2363 unsigned short family)
2365 struct net *net = dev_net(skb->dev);
2366 struct xfrm_policy *pol;
2367 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2375 reverse = dir & ~XFRM_POLICY_MASK;
2376 dir &= XFRM_POLICY_MASK;
2378 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2379 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2383 nf_nat_decode_session(skb, &fl, family);
2385 /* First, check used SA against their selectors. */
2389 for (i = skb->sp->len-1; i >= 0; i--) {
2390 struct xfrm_state *x = skb->sp->xvec[i];
2391 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2392 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2399 sk = sk_to_full_sk(sk);
2400 if (sk && sk->sk_policy[dir]) {
2401 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2403 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2409 pol = xfrm_policy_lookup(net, &fl, family, dir);
2412 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2417 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2418 xfrm_secpath_reject(xerr_idx, skb, &fl);
2419 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2425 pol->curlft.use_time = get_seconds();
2429 #ifdef CONFIG_XFRM_SUB_POLICY
2430 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2431 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2435 if (IS_ERR(pols[1])) {
2436 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2439 pols[1]->curlft.use_time = get_seconds();
2445 if (pol->action == XFRM_POLICY_ALLOW) {
2446 struct sec_path *sp;
2447 static struct sec_path dummy;
2448 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2449 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2450 struct xfrm_tmpl **tpp = tp;
2454 if ((sp = skb->sp) == NULL)
2457 for (pi = 0; pi < npols; pi++) {
2458 if (pols[pi] != pol &&
2459 pols[pi]->action != XFRM_POLICY_ALLOW) {
2460 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2463 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2464 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2467 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2468 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2472 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2476 /* For each tunnel xfrm, find the first matching tmpl.
2477 * For each tmpl before that, find corresponding xfrm.
2478 * Order is _important_. Later we will implement
2479 * some barriers, but at the moment barriers
2480 * are implied between each two transformations.
2482 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2483 k = xfrm_policy_ok(tpp[i], sp, k, family);
2486 /* "-2 - errored_index" returned */
2488 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2493 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2494 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2498 xfrm_pols_put(pols, npols);
2501 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2504 xfrm_secpath_reject(xerr_idx, skb, &fl);
2506 xfrm_pols_put(pols, npols);
2509 EXPORT_SYMBOL(__xfrm_policy_check);
2511 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2513 struct net *net = dev_net(skb->dev);
2515 struct dst_entry *dst;
2518 if (xfrm_decode_session(skb, &fl, family) < 0) {
2519 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2525 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2530 skb_dst_set(skb, dst);
2533 EXPORT_SYMBOL(__xfrm_route_forward);
2535 /* Optimize later using cookies and generation ids. */
2537 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2539 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2540 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2541 * get validated by dst_ops->check on every use. We do this
2542 * because when a normal route referenced by an XFRM dst is
2543 * obsoleted we do not go looking around for all parent
2544 * referencing XFRM dsts so that we can invalidate them. It
2545 * is just too much work. Instead we make the checks here on
2546 * every use. For example:
2548 * XFRM dst A --> IPv4 dst X
2550 * X is the "xdst->route" of A (X is also the "dst->path" of A
2551 * in this example). If X is marked obsolete, "A" will not
2552 * notice. That's what we are validating here via the
2553 * stale_bundle() check.
2555 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2557 * This will force stale_bundle() to fail on any xdst bundle with
2558 * this dst linked in it.
2560 if (dst->obsolete < 0 && !stale_bundle(dst))
2566 static int stale_bundle(struct dst_entry *dst)
2568 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2571 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2573 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2574 dst->dev = dev_net(dev)->loopback_dev;
2579 EXPORT_SYMBOL(xfrm_dst_ifdown);
2581 static void xfrm_link_failure(struct sk_buff *skb)
2583 /* Impossible. Such dst must be popped before reaches point of failure. */
2586 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2589 if (dst->obsolete) {
2597 static void xfrm_init_pmtu(struct dst_entry *dst)
2600 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2601 u32 pmtu, route_mtu_cached;
2603 pmtu = dst_mtu(dst->child);
2604 xdst->child_mtu_cached = pmtu;
2606 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2608 route_mtu_cached = dst_mtu(xdst->route);
2609 xdst->route_mtu_cached = route_mtu_cached;
2611 if (pmtu > route_mtu_cached)
2612 pmtu = route_mtu_cached;
2614 dst_metric_set(dst, RTAX_MTU, pmtu);
2615 } while ((dst = dst->next));
2618 /* Check that the bundle accepts the flow and its components are
2622 static int xfrm_bundle_ok(struct xfrm_dst *first)
2624 struct dst_entry *dst = &first->u.dst;
2625 struct xfrm_dst *last;
2628 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2629 (dst->dev && !netif_running(dst->dev)))
2632 if (dst->flags & DST_XFRM_QUEUE)
2638 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2640 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2642 if (xdst->xfrm_genid != dst->xfrm->genid)
2644 if (xdst->num_pols > 0 &&
2645 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2648 mtu = dst_mtu(dst->child);
2649 if (xdst->child_mtu_cached != mtu) {
2651 xdst->child_mtu_cached = mtu;
2654 if (!dst_check(xdst->route, xdst->route_cookie))
2656 mtu = dst_mtu(xdst->route);
2657 if (xdst->route_mtu_cached != mtu) {
2659 xdst->route_mtu_cached = mtu;
2663 } while (dst->xfrm);
2668 mtu = last->child_mtu_cached;
2672 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2673 if (mtu > last->route_mtu_cached)
2674 mtu = last->route_mtu_cached;
2675 dst_metric_set(dst, RTAX_MTU, mtu);
2680 last = (struct xfrm_dst *)last->u.dst.next;
2681 last->child_mtu_cached = mtu;
2687 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2689 return dst_metric_advmss(dst->path);
2692 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2694 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2696 return mtu ? : dst_mtu(dst->path);
2699 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2702 const struct dst_entry *path = dst->path;
2704 for (; dst != path; dst = dst->child) {
2705 const struct xfrm_state *xfrm = dst->xfrm;
2707 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2709 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2710 daddr = xfrm->coaddr;
2711 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2712 daddr = &xfrm->id.daddr;
2717 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2718 struct sk_buff *skb,
2721 const struct dst_entry *path = dst->path;
2724 daddr = xfrm_get_dst_nexthop(dst, daddr);
2725 return path->ops->neigh_lookup(path, skb, daddr);
2728 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2730 const struct dst_entry *path = dst->path;
2732 daddr = xfrm_get_dst_nexthop(dst, daddr);
2733 path->ops->confirm_neigh(path, daddr);
2736 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2740 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2741 return -EAFNOSUPPORT;
2743 spin_lock(&xfrm_policy_afinfo_lock);
2744 if (unlikely(xfrm_policy_afinfo[family] != NULL))
2747 struct dst_ops *dst_ops = afinfo->dst_ops;
2748 if (likely(dst_ops->kmem_cachep == NULL))
2749 dst_ops->kmem_cachep = xfrm_dst_cache;
2750 if (likely(dst_ops->check == NULL))
2751 dst_ops->check = xfrm_dst_check;
2752 if (likely(dst_ops->default_advmss == NULL))
2753 dst_ops->default_advmss = xfrm_default_advmss;
2754 if (likely(dst_ops->mtu == NULL))
2755 dst_ops->mtu = xfrm_mtu;
2756 if (likely(dst_ops->negative_advice == NULL))
2757 dst_ops->negative_advice = xfrm_negative_advice;
2758 if (likely(dst_ops->link_failure == NULL))
2759 dst_ops->link_failure = xfrm_link_failure;
2760 if (likely(dst_ops->neigh_lookup == NULL))
2761 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2762 if (likely(!dst_ops->confirm_neigh))
2763 dst_ops->confirm_neigh = xfrm_confirm_neigh;
2764 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2766 spin_unlock(&xfrm_policy_afinfo_lock);
2770 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2772 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2774 struct dst_ops *dst_ops = afinfo->dst_ops;
2777 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2778 if (xfrm_policy_afinfo[i] != afinfo)
2780 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2786 dst_ops->kmem_cachep = NULL;
2787 dst_ops->check = NULL;
2788 dst_ops->negative_advice = NULL;
2789 dst_ops->link_failure = NULL;
2791 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2793 #ifdef CONFIG_XFRM_STATISTICS
2794 static int __net_init xfrm_statistics_init(struct net *net)
2797 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2798 if (!net->mib.xfrm_statistics)
2800 rv = xfrm_proc_init(net);
2802 free_percpu(net->mib.xfrm_statistics);
2806 static void xfrm_statistics_fini(struct net *net)
2808 xfrm_proc_fini(net);
2809 free_percpu(net->mib.xfrm_statistics);
2812 static int __net_init xfrm_statistics_init(struct net *net)
2817 static void xfrm_statistics_fini(struct net *net)
2822 static int __net_init xfrm_policy_init(struct net *net)
2824 unsigned int hmask, sz;
2827 if (net_eq(net, &init_net))
2828 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2829 sizeof(struct xfrm_dst),
2830 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2834 sz = (hmask+1) * sizeof(struct hlist_head);
2836 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2837 if (!net->xfrm.policy_byidx)
2839 net->xfrm.policy_idx_hmask = hmask;
2841 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2842 struct xfrm_policy_hash *htab;
2844 net->xfrm.policy_count[dir] = 0;
2845 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2846 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2848 htab = &net->xfrm.policy_bydst[dir];
2849 htab->table = xfrm_hash_alloc(sz);
2852 htab->hmask = hmask;
2858 net->xfrm.policy_hthresh.lbits4 = 32;
2859 net->xfrm.policy_hthresh.rbits4 = 32;
2860 net->xfrm.policy_hthresh.lbits6 = 128;
2861 net->xfrm.policy_hthresh.rbits6 = 128;
2863 seqlock_init(&net->xfrm.policy_hthresh.lock);
2865 INIT_LIST_HEAD(&net->xfrm.policy_all);
2866 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2867 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2868 if (net_eq(net, &init_net))
2873 for (dir--; dir >= 0; dir--) {
2874 struct xfrm_policy_hash *htab;
2876 htab = &net->xfrm.policy_bydst[dir];
2877 xfrm_hash_free(htab->table, sz);
2879 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2884 static void xfrm_policy_fini(struct net *net)
2889 flush_work(&net->xfrm.policy_hash_work);
2890 #ifdef CONFIG_XFRM_SUB_POLICY
2891 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2893 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2895 WARN_ON(!list_empty(&net->xfrm.policy_all));
2897 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2898 struct xfrm_policy_hash *htab;
2900 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2902 htab = &net->xfrm.policy_bydst[dir];
2903 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2904 WARN_ON(!hlist_empty(htab->table));
2905 xfrm_hash_free(htab->table, sz);
2908 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2909 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2910 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2913 static int __net_init xfrm_net_init(struct net *net)
2917 /* Initialize the per-net locks here */
2918 spin_lock_init(&net->xfrm.xfrm_state_lock);
2919 spin_lock_init(&net->xfrm.xfrm_policy_lock);
2920 mutex_init(&net->xfrm.xfrm_cfg_mutex);
2922 rv = xfrm_statistics_init(net);
2924 goto out_statistics;
2925 rv = xfrm_state_init(net);
2928 rv = xfrm_policy_init(net);
2931 rv = xfrm_sysctl_init(net);
2938 xfrm_policy_fini(net);
2940 xfrm_state_fini(net);
2942 xfrm_statistics_fini(net);
2947 static void __net_exit xfrm_net_exit(struct net *net)
2949 xfrm_sysctl_fini(net);
2950 xfrm_policy_fini(net);
2951 xfrm_state_fini(net);
2952 xfrm_statistics_fini(net);
2955 static struct pernet_operations __net_initdata xfrm_net_ops = {
2956 .init = xfrm_net_init,
2957 .exit = xfrm_net_exit,
2960 void __init xfrm_init(void)
2964 xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2966 BUG_ON(!xfrm_pcpu_work);
2968 for (i = 0; i < NR_CPUS; i++)
2969 INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
2971 register_pernet_subsys(&xfrm_net_ops);
2972 seqcount_init(&xfrm_policy_hash_generation);
2976 #ifdef CONFIG_AUDITSYSCALL
2977 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2978 struct audit_buffer *audit_buf)
2980 struct xfrm_sec_ctx *ctx = xp->security;
2981 struct xfrm_selector *sel = &xp->selector;
2984 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2985 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2987 switch (sel->family) {
2989 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2990 if (sel->prefixlen_s != 32)
2991 audit_log_format(audit_buf, " src_prefixlen=%d",
2993 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2994 if (sel->prefixlen_d != 32)
2995 audit_log_format(audit_buf, " dst_prefixlen=%d",
2999 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3000 if (sel->prefixlen_s != 128)
3001 audit_log_format(audit_buf, " src_prefixlen=%d",
3003 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3004 if (sel->prefixlen_d != 128)
3005 audit_log_format(audit_buf, " dst_prefixlen=%d",
3011 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3013 struct audit_buffer *audit_buf;
3015 audit_buf = xfrm_audit_start("SPD-add");
3016 if (audit_buf == NULL)
3018 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3019 audit_log_format(audit_buf, " res=%u", result);
3020 xfrm_audit_common_policyinfo(xp, audit_buf);
3021 audit_log_end(audit_buf);
3023 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3025 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3028 struct audit_buffer *audit_buf;
3030 audit_buf = xfrm_audit_start("SPD-delete");
3031 if (audit_buf == NULL)
3033 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3034 audit_log_format(audit_buf, " res=%u", result);
3035 xfrm_audit_common_policyinfo(xp, audit_buf);
3036 audit_log_end(audit_buf);
3038 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3041 #ifdef CONFIG_XFRM_MIGRATE
3042 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3043 const struct xfrm_selector *sel_tgt)
3045 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3046 if (sel_tgt->family == sel_cmp->family &&
3047 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3049 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3051 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3052 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3056 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3063 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3064 u8 dir, u8 type, struct net *net)
3066 struct xfrm_policy *pol, *ret = NULL;
3067 struct hlist_head *chain;
3070 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3071 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3072 hlist_for_each_entry(pol, chain, bydst) {
3073 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3074 pol->type == type) {
3076 priority = ret->priority;
3080 chain = &net->xfrm.policy_inexact[dir];
3081 hlist_for_each_entry(pol, chain, bydst) {
3082 if ((pol->priority >= priority) && ret)
3085 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3086 pol->type == type) {
3094 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3099 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3103 if (t->mode == m->mode && t->id.proto == m->proto &&
3104 (m->reqid == 0 || t->reqid == m->reqid)) {
3106 case XFRM_MODE_TUNNEL:
3107 case XFRM_MODE_BEET:
3108 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3110 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3115 case XFRM_MODE_TRANSPORT:
3116 /* in case of transport mode, template does not store
3117 any IP addresses, hence we just compare mode and
3128 /* update endpoint address(es) of template(s) */
3129 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3130 struct xfrm_migrate *m, int num_migrate)
3132 struct xfrm_migrate *mp;
3135 write_lock_bh(&pol->lock);
3136 if (unlikely(pol->walk.dead)) {
3137 /* target policy has been deleted */
3138 write_unlock_bh(&pol->lock);
3142 for (i = 0; i < pol->xfrm_nr; i++) {
3143 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3144 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3147 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3148 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3150 /* update endpoints */
3151 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3152 sizeof(pol->xfrm_vec[i].id.daddr));
3153 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3154 sizeof(pol->xfrm_vec[i].saddr));
3155 pol->xfrm_vec[i].encap_family = mp->new_family;
3157 atomic_inc(&pol->genid);
3161 write_unlock_bh(&pol->lock);
3169 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3173 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3176 for (i = 0; i < num_migrate; i++) {
3177 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3178 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3181 /* check if there is any duplicated entry */
3182 for (j = i + 1; j < num_migrate; j++) {
3183 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3184 sizeof(m[i].old_daddr)) &&
3185 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3186 sizeof(m[i].old_saddr)) &&
3187 m[i].proto == m[j].proto &&
3188 m[i].mode == m[j].mode &&
3189 m[i].reqid == m[j].reqid &&
3190 m[i].old_family == m[j].old_family)
3198 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3199 struct xfrm_migrate *m, int num_migrate,
3200 struct xfrm_kmaddress *k, struct net *net,
3201 struct xfrm_encap_tmpl *encap)
3203 int i, err, nx_cur = 0, nx_new = 0;
3204 struct xfrm_policy *pol = NULL;
3205 struct xfrm_state *x, *xc;
3206 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3207 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3208 struct xfrm_migrate *mp;
3210 /* Stage 0 - sanity checks */
3211 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3214 if (dir >= XFRM_POLICY_MAX) {
3219 /* Stage 1 - find policy */
3220 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3225 /* Stage 2 - find and update state(s) */
3226 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3227 if ((x = xfrm_migrate_state_find(mp, net))) {
3230 xc = xfrm_state_migrate(x, mp, encap);
3241 /* Stage 3 - update policy */
3242 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3245 /* Stage 4 - delete old state(s) */
3247 xfrm_states_put(x_cur, nx_cur);
3248 xfrm_states_delete(x_cur, nx_cur);
3251 /* Stage 5 - announce */
3252 km_migrate(sel, dir, type, m, num_migrate, k, encap);
3264 xfrm_states_put(x_cur, nx_cur);
3266 xfrm_states_delete(x_new, nx_new);
3270 EXPORT_SYMBOL(xfrm_migrate);