6 * Kazunori MIYAZAWA @USAGI
7 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
9 * Kazunori MIYAZAWA @USAGI
11 * Split up af-specific portion
12 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
32 #ifdef CONFIG_XFRM_STATISTICS
36 #include "xfrm_hash.h"
38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
40 #define XFRM_MAX_QUEUE_LEN 100
42 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
43 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
46 static struct kmem_cache *xfrm_dst_cache __read_mostly;
48 static void xfrm_init_pmtu(struct dst_entry *dst);
49 static int stale_bundle(struct dst_entry *dst);
50 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
51 static void xfrm_policy_queue_process(unsigned long arg);
53 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
57 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
59 const struct flowi4 *fl4 = &fl->u.ip4;
61 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
62 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
63 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
64 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
65 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
66 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
70 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
72 const struct flowi6 *fl6 = &fl->u.ip6;
74 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
75 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
76 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
77 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
78 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
79 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
82 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
83 unsigned short family)
87 return __xfrm4_selector_match(sel, fl);
89 return __xfrm6_selector_match(sel, fl);
94 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
96 struct xfrm_policy_afinfo *afinfo;
98 if (unlikely(family >= NPROTO))
101 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
102 if (unlikely(!afinfo))
107 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
112 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
113 const xfrm_address_t *saddr,
114 const xfrm_address_t *daddr,
117 struct xfrm_policy_afinfo *afinfo;
118 struct dst_entry *dst;
120 afinfo = xfrm_policy_get_afinfo(family);
121 if (unlikely(afinfo == NULL))
122 return ERR_PTR(-EAFNOSUPPORT);
124 dst = afinfo->dst_lookup(net, tos, saddr, daddr);
126 xfrm_policy_put_afinfo(afinfo);
131 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
132 xfrm_address_t *prev_saddr,
133 xfrm_address_t *prev_daddr,
136 struct net *net = xs_net(x);
137 xfrm_address_t *saddr = &x->props.saddr;
138 xfrm_address_t *daddr = &x->id.daddr;
139 struct dst_entry *dst;
141 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
145 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
150 dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
153 if (prev_saddr != saddr)
154 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
155 if (prev_daddr != daddr)
156 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
162 static inline unsigned long make_jiffies(long secs)
164 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
165 return MAX_SCHEDULE_TIMEOUT-1;
170 static void xfrm_policy_timer(unsigned long data)
172 struct xfrm_policy *xp = (struct xfrm_policy *)data;
173 unsigned long now = get_seconds();
174 long next = LONG_MAX;
178 read_lock(&xp->lock);
180 if (unlikely(xp->walk.dead))
183 dir = xfrm_policy_id2dir(xp->index);
185 if (xp->lft.hard_add_expires_seconds) {
186 long tmo = xp->lft.hard_add_expires_seconds +
187 xp->curlft.add_time - now;
193 if (xp->lft.hard_use_expires_seconds) {
194 long tmo = xp->lft.hard_use_expires_seconds +
195 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
201 if (xp->lft.soft_add_expires_seconds) {
202 long tmo = xp->lft.soft_add_expires_seconds +
203 xp->curlft.add_time - now;
206 tmo = XFRM_KM_TIMEOUT;
211 if (xp->lft.soft_use_expires_seconds) {
212 long tmo = xp->lft.soft_use_expires_seconds +
213 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
216 tmo = XFRM_KM_TIMEOUT;
223 km_policy_expired(xp, dir, 0, 0);
224 if (next != LONG_MAX &&
225 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
229 read_unlock(&xp->lock);
234 read_unlock(&xp->lock);
235 if (!xfrm_policy_delete(xp, dir))
236 km_policy_expired(xp, dir, 1, 0);
240 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
242 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
244 if (unlikely(pol->walk.dead))
252 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
254 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
256 return !pol->walk.dead;
259 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
261 xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
264 static const struct flow_cache_ops xfrm_policy_fc_ops = {
265 .get = xfrm_policy_flo_get,
266 .check = xfrm_policy_flo_check,
267 .delete = xfrm_policy_flo_delete,
270 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
274 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
276 struct xfrm_policy *policy;
278 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
281 write_pnet(&policy->xp_net, net);
282 INIT_LIST_HEAD(&policy->walk.all);
283 INIT_HLIST_NODE(&policy->bydst);
284 INIT_HLIST_NODE(&policy->byidx);
285 rwlock_init(&policy->lock);
286 atomic_set(&policy->refcnt, 1);
287 skb_queue_head_init(&policy->polq.hold_queue);
288 setup_timer(&policy->timer, xfrm_policy_timer,
289 (unsigned long)policy);
290 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
291 (unsigned long)policy);
292 policy->flo.ops = &xfrm_policy_fc_ops;
296 EXPORT_SYMBOL(xfrm_policy_alloc);
298 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
300 void xfrm_policy_destroy(struct xfrm_policy *policy)
302 BUG_ON(!policy->walk.dead);
304 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
307 security_xfrm_policy_free(policy->security);
310 EXPORT_SYMBOL(xfrm_policy_destroy);
312 static void xfrm_queue_purge(struct sk_buff_head *list)
316 while ((skb = skb_dequeue(list)) != NULL)
320 /* Rule must be locked. Release descentant resources, announce
321 * entry dead. The rule must be unlinked from lists to the moment.
324 static void xfrm_policy_kill(struct xfrm_policy *policy)
326 policy->walk.dead = 1;
328 atomic_inc(&policy->genid);
330 if (del_timer(&policy->polq.hold_timer))
331 xfrm_pol_put(policy);
332 xfrm_queue_purge(&policy->polq.hold_queue);
334 if (del_timer(&policy->timer))
335 xfrm_pol_put(policy);
337 xfrm_pol_put(policy);
340 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
342 static inline unsigned int idx_hash(struct net *net, u32 index)
344 return __idx_hash(index, net->xfrm.policy_idx_hmask);
347 static struct hlist_head *policy_hash_bysel(struct net *net,
348 const struct xfrm_selector *sel,
349 unsigned short family, int dir)
351 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
352 unsigned int hash = __sel_hash(sel, family, hmask);
354 return (hash == hmask + 1 ?
355 &net->xfrm.policy_inexact[dir] :
356 net->xfrm.policy_bydst[dir].table + hash);
359 static struct hlist_head *policy_hash_direct(struct net *net,
360 const xfrm_address_t *daddr,
361 const xfrm_address_t *saddr,
362 unsigned short family, int dir)
364 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
365 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
367 return net->xfrm.policy_bydst[dir].table + hash;
370 static void xfrm_dst_hash_transfer(struct hlist_head *list,
371 struct hlist_head *ndsttable,
372 unsigned int nhashmask)
374 struct hlist_node *tmp, *entry0 = NULL;
375 struct xfrm_policy *pol;
379 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
382 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
383 pol->family, nhashmask);
385 hlist_del(&pol->bydst);
386 hlist_add_head(&pol->bydst, ndsttable+h);
391 hlist_del(&pol->bydst);
392 hlist_add_behind(&pol->bydst, entry0);
394 entry0 = &pol->bydst;
396 if (!hlist_empty(list)) {
402 static void xfrm_idx_hash_transfer(struct hlist_head *list,
403 struct hlist_head *nidxtable,
404 unsigned int nhashmask)
406 struct hlist_node *tmp;
407 struct xfrm_policy *pol;
409 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
412 h = __idx_hash(pol->index, nhashmask);
413 hlist_add_head(&pol->byidx, nidxtable+h);
417 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
419 return ((old_hmask + 1) << 1) - 1;
422 static void xfrm_bydst_resize(struct net *net, int dir)
424 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
425 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
426 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
427 struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
428 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
434 write_lock_bh(&net->xfrm.xfrm_policy_lock);
436 for (i = hmask; i >= 0; i--)
437 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
439 net->xfrm.policy_bydst[dir].table = ndst;
440 net->xfrm.policy_bydst[dir].hmask = nhashmask;
442 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
444 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
447 static void xfrm_byidx_resize(struct net *net, int total)
449 unsigned int hmask = net->xfrm.policy_idx_hmask;
450 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
451 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
452 struct hlist_head *oidx = net->xfrm.policy_byidx;
453 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
459 write_lock_bh(&net->xfrm.xfrm_policy_lock);
461 for (i = hmask; i >= 0; i--)
462 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
464 net->xfrm.policy_byidx = nidx;
465 net->xfrm.policy_idx_hmask = nhashmask;
467 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
469 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
472 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
474 unsigned int cnt = net->xfrm.policy_count[dir];
475 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
480 if ((hmask + 1) < xfrm_policy_hashmax &&
487 static inline int xfrm_byidx_should_resize(struct net *net, int total)
489 unsigned int hmask = net->xfrm.policy_idx_hmask;
491 if ((hmask + 1) < xfrm_policy_hashmax &&
498 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
500 read_lock_bh(&net->xfrm.xfrm_policy_lock);
501 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
502 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
503 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
504 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
505 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
506 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
507 si->spdhcnt = net->xfrm.policy_idx_hmask;
508 si->spdhmcnt = xfrm_policy_hashmax;
509 read_unlock_bh(&net->xfrm.xfrm_policy_lock);
511 EXPORT_SYMBOL(xfrm_spd_getinfo);
513 static DEFINE_MUTEX(hash_resize_mutex);
514 static void xfrm_hash_resize(struct work_struct *work)
516 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
519 mutex_lock(&hash_resize_mutex);
522 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
523 if (xfrm_bydst_should_resize(net, dir, &total))
524 xfrm_bydst_resize(net, dir);
526 if (xfrm_byidx_should_resize(net, total))
527 xfrm_byidx_resize(net, total);
529 mutex_unlock(&hash_resize_mutex);
532 /* Generate new index... KAME seems to generate them ordered by cost
533 * of an absolute inpredictability of ordering of rules. This will not pass. */
534 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
536 static u32 idx_generator;
539 struct hlist_head *list;
540 struct xfrm_policy *p;
545 idx = (idx_generator | dir);
554 list = net->xfrm.policy_byidx + idx_hash(net, idx);
556 hlist_for_each_entry(p, list, byidx) {
557 if (p->index == idx) {
567 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
569 u32 *p1 = (u32 *) s1;
570 u32 *p2 = (u32 *) s2;
571 int len = sizeof(struct xfrm_selector) / sizeof(u32);
574 for (i = 0; i < len; i++) {
582 static void xfrm_policy_requeue(struct xfrm_policy *old,
583 struct xfrm_policy *new)
585 struct xfrm_policy_queue *pq = &old->polq;
586 struct sk_buff_head list;
588 __skb_queue_head_init(&list);
590 spin_lock_bh(&pq->hold_queue.lock);
591 skb_queue_splice_init(&pq->hold_queue, &list);
592 if (del_timer(&pq->hold_timer))
594 spin_unlock_bh(&pq->hold_queue.lock);
596 if (skb_queue_empty(&list))
601 spin_lock_bh(&pq->hold_queue.lock);
602 skb_queue_splice(&list, &pq->hold_queue);
603 pq->timeout = XFRM_QUEUE_TMO_MIN;
604 if (!mod_timer(&pq->hold_timer, jiffies))
606 spin_unlock_bh(&pq->hold_queue.lock);
609 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
610 struct xfrm_policy *pol)
612 u32 mark = policy->mark.v & policy->mark.m;
614 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
617 if ((mark & pol->mark.m) == pol->mark.v &&
618 policy->priority == pol->priority)
624 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
626 struct net *net = xp_net(policy);
627 struct xfrm_policy *pol;
628 struct xfrm_policy *delpol;
629 struct hlist_head *chain;
630 struct hlist_node *newpos;
632 write_lock_bh(&net->xfrm.xfrm_policy_lock);
633 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
636 hlist_for_each_entry(pol, chain, bydst) {
637 if (pol->type == policy->type &&
638 !selector_cmp(&pol->selector, &policy->selector) &&
639 xfrm_policy_mark_match(policy, pol) &&
640 xfrm_sec_ctx_match(pol->security, policy->security) &&
643 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
647 if (policy->priority > pol->priority)
649 } else if (policy->priority >= pol->priority) {
650 newpos = &pol->bydst;
657 hlist_add_behind(&policy->bydst, newpos);
659 hlist_add_head(&policy->bydst, chain);
660 xfrm_pol_hold(policy);
661 net->xfrm.policy_count[dir]++;
662 atomic_inc(&net->xfrm.flow_cache_genid);
664 /* After previous checking, family can either be AF_INET or AF_INET6 */
665 if (policy->family == AF_INET)
666 rt_genid_bump_ipv4(net);
668 rt_genid_bump_ipv6(net);
671 xfrm_policy_requeue(delpol, policy);
672 __xfrm_policy_unlink(delpol, dir);
674 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
675 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
676 policy->curlft.add_time = get_seconds();
677 policy->curlft.use_time = 0;
678 if (!mod_timer(&policy->timer, jiffies + HZ))
679 xfrm_pol_hold(policy);
680 list_add(&policy->walk.all, &net->xfrm.policy_all);
681 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
684 xfrm_policy_kill(delpol);
685 else if (xfrm_bydst_should_resize(net, dir, NULL))
686 schedule_work(&net->xfrm.policy_hash_work);
690 EXPORT_SYMBOL(xfrm_policy_insert);
692 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
693 int dir, struct xfrm_selector *sel,
694 struct xfrm_sec_ctx *ctx, int delete,
697 struct xfrm_policy *pol, *ret;
698 struct hlist_head *chain;
701 write_lock_bh(&net->xfrm.xfrm_policy_lock);
702 chain = policy_hash_bysel(net, sel, sel->family, dir);
704 hlist_for_each_entry(pol, chain, bydst) {
705 if (pol->type == type &&
706 (mark & pol->mark.m) == pol->mark.v &&
707 !selector_cmp(sel, &pol->selector) &&
708 xfrm_sec_ctx_match(ctx, pol->security)) {
711 *err = security_xfrm_policy_delete(
714 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
717 __xfrm_policy_unlink(pol, dir);
723 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
726 xfrm_policy_kill(ret);
729 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
731 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
732 int dir, u32 id, int delete, int *err)
734 struct xfrm_policy *pol, *ret;
735 struct hlist_head *chain;
738 if (xfrm_policy_id2dir(id) != dir)
742 write_lock_bh(&net->xfrm.xfrm_policy_lock);
743 chain = net->xfrm.policy_byidx + idx_hash(net, id);
745 hlist_for_each_entry(pol, chain, byidx) {
746 if (pol->type == type && pol->index == id &&
747 (mark & pol->mark.m) == pol->mark.v) {
750 *err = security_xfrm_policy_delete(
753 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
756 __xfrm_policy_unlink(pol, dir);
762 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
765 xfrm_policy_kill(ret);
768 EXPORT_SYMBOL(xfrm_policy_byid);
770 #ifdef CONFIG_SECURITY_NETWORK_XFRM
772 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
776 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
777 struct xfrm_policy *pol;
780 hlist_for_each_entry(pol,
781 &net->xfrm.policy_inexact[dir], bydst) {
782 if (pol->type != type)
784 err = security_xfrm_policy_delete(pol->security);
786 xfrm_audit_policy_delete(pol, 0, task_valid);
790 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
791 hlist_for_each_entry(pol,
792 net->xfrm.policy_bydst[dir].table + i,
794 if (pol->type != type)
796 err = security_xfrm_policy_delete(
799 xfrm_audit_policy_delete(pol, 0,
810 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
816 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
818 int dir, err = 0, cnt = 0;
820 write_lock_bh(&net->xfrm.xfrm_policy_lock);
822 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
826 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
827 struct xfrm_policy *pol;
831 hlist_for_each_entry(pol,
832 &net->xfrm.policy_inexact[dir], bydst) {
833 if (pol->type != type)
835 __xfrm_policy_unlink(pol, dir);
836 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
839 xfrm_audit_policy_delete(pol, 1, task_valid);
841 xfrm_policy_kill(pol);
843 write_lock_bh(&net->xfrm.xfrm_policy_lock);
847 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
849 hlist_for_each_entry(pol,
850 net->xfrm.policy_bydst[dir].table + i,
852 if (pol->type != type)
854 __xfrm_policy_unlink(pol, dir);
855 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
858 xfrm_audit_policy_delete(pol, 1, task_valid);
859 xfrm_policy_kill(pol);
861 write_lock_bh(&net->xfrm.xfrm_policy_lock);
870 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
873 EXPORT_SYMBOL(xfrm_policy_flush);
875 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
876 int (*func)(struct xfrm_policy *, int, int, void*),
879 struct xfrm_policy *pol;
880 struct xfrm_policy_walk_entry *x;
883 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
884 walk->type != XFRM_POLICY_TYPE_ANY)
887 if (list_empty(&walk->walk.all) && walk->seq != 0)
890 write_lock_bh(&net->xfrm.xfrm_policy_lock);
891 if (list_empty(&walk->walk.all))
892 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
894 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
895 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
898 pol = container_of(x, struct xfrm_policy, walk);
899 if (walk->type != XFRM_POLICY_TYPE_ANY &&
900 walk->type != pol->type)
902 error = func(pol, xfrm_policy_id2dir(pol->index),
905 list_move_tail(&walk->walk.all, &x->all);
910 if (walk->seq == 0) {
914 list_del_init(&walk->walk.all);
916 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
919 EXPORT_SYMBOL(xfrm_policy_walk);
921 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
923 INIT_LIST_HEAD(&walk->walk.all);
928 EXPORT_SYMBOL(xfrm_policy_walk_init);
930 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
932 if (list_empty(&walk->walk.all))
935 write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
936 list_del(&walk->walk.all);
937 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
939 EXPORT_SYMBOL(xfrm_policy_walk_done);
942 * Find policy to apply to this flow.
944 * Returns 0 if policy found, else an -errno.
946 static int xfrm_policy_match(const struct xfrm_policy *pol,
947 const struct flowi *fl,
948 u8 type, u16 family, int dir)
950 const struct xfrm_selector *sel = &pol->selector;
954 if (pol->family != family ||
955 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
959 match = xfrm_selector_match(sel, fl, family);
961 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
967 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
968 const struct flowi *fl,
972 struct xfrm_policy *pol, *ret;
973 const xfrm_address_t *daddr, *saddr;
974 struct hlist_head *chain;
977 daddr = xfrm_flowi_daddr(fl, family);
978 saddr = xfrm_flowi_saddr(fl, family);
979 if (unlikely(!daddr || !saddr))
982 read_lock_bh(&net->xfrm.xfrm_policy_lock);
983 chain = policy_hash_direct(net, daddr, saddr, family, dir);
985 hlist_for_each_entry(pol, chain, bydst) {
986 err = xfrm_policy_match(pol, fl, type, family, dir);
996 priority = ret->priority;
1000 chain = &net->xfrm.policy_inexact[dir];
1001 hlist_for_each_entry(pol, chain, bydst) {
1002 err = xfrm_policy_match(pol, fl, type, family, dir);
1010 } else if (pol->priority < priority) {
1018 read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1023 static struct xfrm_policy *
1024 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1026 #ifdef CONFIG_XFRM_SUB_POLICY
1027 struct xfrm_policy *pol;
1029 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1033 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1036 static int flow_to_policy_dir(int dir)
1038 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1039 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1040 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1046 return XFRM_POLICY_IN;
1048 return XFRM_POLICY_OUT;
1050 return XFRM_POLICY_FWD;
1054 static struct flow_cache_object *
1055 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1056 u8 dir, struct flow_cache_object *old_obj, void *ctx)
1058 struct xfrm_policy *pol;
1061 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1063 pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1064 if (IS_ERR_OR_NULL(pol))
1065 return ERR_CAST(pol);
1067 /* Resolver returns two references:
1068 * one for cache and one for caller of flow_cache_lookup() */
1074 static inline int policy_to_flow_dir(int dir)
1076 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1077 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1078 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1082 case XFRM_POLICY_IN:
1084 case XFRM_POLICY_OUT:
1085 return FLOW_DIR_OUT;
1086 case XFRM_POLICY_FWD:
1087 return FLOW_DIR_FWD;
1091 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1092 const struct flowi *fl)
1094 struct xfrm_policy *pol;
1095 struct net *net = sock_net(sk);
1097 read_lock_bh(&net->xfrm.xfrm_policy_lock);
1098 if ((pol = sk->sk_policy[dir]) != NULL) {
1099 bool match = xfrm_selector_match(&pol->selector, fl,
1104 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1108 err = security_xfrm_policy_lookup(pol->security,
1110 policy_to_flow_dir(dir));
1113 else if (err == -ESRCH)
1121 read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1125 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1127 struct net *net = xp_net(pol);
1128 struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1131 list_add(&pol->walk.all, &net->xfrm.policy_all);
1132 hlist_add_head(&pol->bydst, chain);
1133 hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1134 net->xfrm.policy_count[dir]++;
1137 if (xfrm_bydst_should_resize(net, dir, NULL))
1138 schedule_work(&net->xfrm.policy_hash_work);
1141 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1144 struct net *net = xp_net(pol);
1146 if (hlist_unhashed(&pol->bydst))
1149 hlist_del_init(&pol->bydst);
1150 hlist_del(&pol->byidx);
1151 list_del(&pol->walk.all);
1152 net->xfrm.policy_count[dir]--;
1157 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1159 struct net *net = xp_net(pol);
1161 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1162 pol = __xfrm_policy_unlink(pol, dir);
1163 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1165 xfrm_policy_kill(pol);
1170 EXPORT_SYMBOL(xfrm_policy_delete);
1172 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1174 struct net *net = xp_net(pol);
1175 struct xfrm_policy *old_pol;
1177 #ifdef CONFIG_XFRM_SUB_POLICY
1178 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1182 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1183 old_pol = sk->sk_policy[dir];
1184 sk->sk_policy[dir] = pol;
1186 pol->curlft.add_time = get_seconds();
1187 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1188 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1192 xfrm_policy_requeue(old_pol, pol);
1194 /* Unlinking succeeds always. This is the only function
1195 * allowed to delete or replace socket policy.
1197 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1199 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1202 xfrm_policy_kill(old_pol);
1207 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1209 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1210 struct net *net = xp_net(old);
1213 newp->selector = old->selector;
1214 if (security_xfrm_policy_clone(old->security,
1217 return NULL; /* ENOMEM */
1219 newp->lft = old->lft;
1220 newp->curlft = old->curlft;
1221 newp->mark = old->mark;
1222 newp->action = old->action;
1223 newp->flags = old->flags;
1224 newp->xfrm_nr = old->xfrm_nr;
1225 newp->index = old->index;
1226 newp->type = old->type;
1227 memcpy(newp->xfrm_vec, old->xfrm_vec,
1228 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1229 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1230 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1231 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1237 int __xfrm_sk_clone_policy(struct sock *sk)
1239 struct xfrm_policy *p0 = sk->sk_policy[0],
1240 *p1 = sk->sk_policy[1];
1242 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1243 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1245 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1251 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1252 unsigned short family)
1255 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1257 if (unlikely(afinfo == NULL))
1259 err = afinfo->get_saddr(net, local, remote);
1260 xfrm_policy_put_afinfo(afinfo);
1264 /* Resolve list of templates for the flow, given policy. */
1267 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1268 struct xfrm_state **xfrm, unsigned short family)
1270 struct net *net = xp_net(policy);
1273 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1274 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1277 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1278 struct xfrm_state *x;
1279 xfrm_address_t *remote = daddr;
1280 xfrm_address_t *local = saddr;
1281 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1283 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1284 tmpl->mode == XFRM_MODE_BEET) {
1285 remote = &tmpl->id.daddr;
1286 local = &tmpl->saddr;
1287 if (xfrm_addr_any(local, tmpl->encap_family)) {
1288 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1295 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1297 if (x && x->km.state == XFRM_STATE_VALID) {
1304 error = (x->km.state == XFRM_STATE_ERROR ?
1307 } else if (error == -ESRCH) {
1311 if (!tmpl->optional)
1317 for (nx--; nx >= 0; nx--)
1318 xfrm_state_put(xfrm[nx]);
1323 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1324 struct xfrm_state **xfrm, unsigned short family)
1326 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1327 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1333 for (i = 0; i < npols; i++) {
1334 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1339 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1347 /* found states are sorted for outbound processing */
1349 xfrm_state_sort(xfrm, tpp, cnx, family);
1354 for (cnx--; cnx >= 0; cnx--)
1355 xfrm_state_put(tpp[cnx]);
1360 /* Check that the bundle accepts the flow and its components are
1364 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1366 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1372 tos = afinfo->get_tos(fl);
1374 xfrm_policy_put_afinfo(afinfo);
1379 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1381 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1382 struct dst_entry *dst = &xdst->u.dst;
1384 if (xdst->route == NULL) {
1385 /* Dummy bundle - if it has xfrms we were not
1386 * able to build bundle as template resolution failed.
1387 * It means we need to try again resolving. */
1388 if (xdst->num_xfrms > 0)
1390 } else if (dst->flags & DST_XFRM_QUEUE) {
1394 if (stale_bundle(dst))
1402 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1404 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1405 struct dst_entry *dst = &xdst->u.dst;
1409 if (stale_bundle(dst))
1415 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1417 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1418 struct dst_entry *dst = &xdst->u.dst;
1423 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1424 .get = xfrm_bundle_flo_get,
1425 .check = xfrm_bundle_flo_check,
1426 .delete = xfrm_bundle_flo_delete,
1429 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1431 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1432 struct dst_ops *dst_ops;
1433 struct xfrm_dst *xdst;
1436 return ERR_PTR(-EINVAL);
1440 dst_ops = &net->xfrm.xfrm4_dst_ops;
1442 #if IS_ENABLED(CONFIG_IPV6)
1444 dst_ops = &net->xfrm.xfrm6_dst_ops;
1450 xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1453 struct dst_entry *dst = &xdst->u.dst;
1455 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1456 xdst->flo.ops = &xfrm_bundle_fc_ops;
1457 if (afinfo->init_dst)
1458 afinfo->init_dst(net, xdst);
1460 xdst = ERR_PTR(-ENOBUFS);
1462 xfrm_policy_put_afinfo(afinfo);
1467 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1470 struct xfrm_policy_afinfo *afinfo =
1471 xfrm_policy_get_afinfo(dst->ops->family);
1477 err = afinfo->init_path(path, dst, nfheader_len);
1479 xfrm_policy_put_afinfo(afinfo);
1484 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1485 const struct flowi *fl)
1487 struct xfrm_policy_afinfo *afinfo =
1488 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1494 err = afinfo->fill_dst(xdst, dev, fl);
1496 xfrm_policy_put_afinfo(afinfo);
1502 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1503 * all the metrics... Shortly, bundle a bundle.
1506 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1507 struct xfrm_state **xfrm, int nx,
1508 const struct flowi *fl,
1509 struct dst_entry *dst)
1511 struct net *net = xp_net(policy);
1512 unsigned long now = jiffies;
1513 struct net_device *dev;
1514 struct xfrm_mode *inner_mode;
1515 struct dst_entry *dst_prev = NULL;
1516 struct dst_entry *dst0 = NULL;
1520 int nfheader_len = 0;
1521 int trailer_len = 0;
1523 int family = policy->selector.family;
1524 xfrm_address_t saddr, daddr;
1526 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1528 tos = xfrm_get_tos(fl, family);
1535 for (; i < nx; i++) {
1536 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1537 struct dst_entry *dst1 = &xdst->u.dst;
1539 err = PTR_ERR(xdst);
1545 if (xfrm[i]->sel.family == AF_UNSPEC) {
1546 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1547 xfrm_af2proto(family));
1549 err = -EAFNOSUPPORT;
1554 inner_mode = xfrm[i]->inner_mode;
1559 dst_prev->child = dst_clone(dst1);
1560 dst1->flags |= DST_NOHASH;
1564 dst_copy_metrics(dst1, dst);
1566 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1567 family = xfrm[i]->props.family;
1568 dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1576 dst1->xfrm = xfrm[i];
1577 xdst->xfrm_genid = xfrm[i]->genid;
1579 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1580 dst1->flags |= DST_HOST;
1581 dst1->lastuse = now;
1583 dst1->input = dst_discard;
1584 dst1->output = inner_mode->afinfo->output;
1586 dst1->next = dst_prev;
1589 header_len += xfrm[i]->props.header_len;
1590 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1591 nfheader_len += xfrm[i]->props.header_len;
1592 trailer_len += xfrm[i]->props.trailer_len;
1595 dst_prev->child = dst;
1603 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1604 xfrm_init_pmtu(dst_prev);
1606 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1607 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1609 err = xfrm_fill_dst(xdst, dev, fl);
1613 dst_prev->header_len = header_len;
1614 dst_prev->trailer_len = trailer_len;
1615 header_len -= xdst->u.dst.xfrm->props.header_len;
1616 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1624 xfrm_state_put(xfrm[i]);
1628 dst0 = ERR_PTR(err);
1632 #ifdef CONFIG_XFRM_SUB_POLICY
1633 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1636 *target = kmalloc(size, GFP_ATOMIC);
1641 memcpy(*target, src, size);
1646 static int xfrm_dst_update_parent(struct dst_entry *dst,
1647 const struct xfrm_selector *sel)
1649 #ifdef CONFIG_XFRM_SUB_POLICY
1650 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1651 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1658 static int xfrm_dst_update_origin(struct dst_entry *dst,
1659 const struct flowi *fl)
1661 #ifdef CONFIG_XFRM_SUB_POLICY
1662 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1663 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
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 dst_entry *dst;
1721 struct xfrm_dst *xdst;
1724 /* Try to instantiate a bundle */
1725 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1727 if (err != 0 && err != -EAGAIN)
1728 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1729 return ERR_PTR(err);
1732 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1734 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1735 return ERR_CAST(dst);
1738 xdst = (struct xfrm_dst *)dst;
1739 xdst->num_xfrms = err;
1741 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1743 err = xfrm_dst_update_origin(dst, fl);
1744 if (unlikely(err)) {
1746 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1747 return ERR_PTR(err);
1750 xdst->num_pols = num_pols;
1751 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1752 xdst->policy_genid = atomic_read(&pols[0]->genid);
1757 static void xfrm_policy_queue_process(unsigned long arg)
1760 struct sk_buff *skb;
1762 struct dst_entry *dst;
1763 struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1764 struct xfrm_policy_queue *pq = &pol->polq;
1766 struct sk_buff_head list;
1768 spin_lock(&pq->hold_queue.lock);
1769 skb = skb_peek(&pq->hold_queue);
1771 spin_unlock(&pq->hold_queue.lock);
1776 xfrm_decode_session(skb, &fl, dst->ops->family);
1777 spin_unlock(&pq->hold_queue.lock);
1779 dst_hold(dst->path);
1780 dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1785 if (dst->flags & DST_XFRM_QUEUE) {
1788 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1791 pq->timeout = pq->timeout << 1;
1792 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1799 __skb_queue_head_init(&list);
1801 spin_lock(&pq->hold_queue.lock);
1803 skb_queue_splice_init(&pq->hold_queue, &list);
1804 spin_unlock(&pq->hold_queue.lock);
1806 while (!skb_queue_empty(&list)) {
1807 skb = __skb_dequeue(&list);
1809 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1810 dst_hold(skb_dst(skb)->path);
1811 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1820 skb_dst_set(skb, dst);
1822 err = dst_output(skb);
1831 xfrm_queue_purge(&pq->hold_queue);
1835 static int xdst_queue_output(struct sock *sk, struct sk_buff *skb)
1837 unsigned long sched_next;
1838 struct dst_entry *dst = skb_dst(skb);
1839 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1840 struct xfrm_policy *pol = xdst->pols[0];
1841 struct xfrm_policy_queue *pq = &pol->polq;
1842 const struct sk_buff *fclone = skb + 1;
1844 if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
1845 fclone->fclone == SKB_FCLONE_CLONE)) {
1850 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1857 spin_lock_bh(&pq->hold_queue.lock);
1860 pq->timeout = XFRM_QUEUE_TMO_MIN;
1862 sched_next = jiffies + pq->timeout;
1864 if (del_timer(&pq->hold_timer)) {
1865 if (time_before(pq->hold_timer.expires, sched_next))
1866 sched_next = pq->hold_timer.expires;
1870 __skb_queue_tail(&pq->hold_queue, skb);
1871 if (!mod_timer(&pq->hold_timer, sched_next))
1874 spin_unlock_bh(&pq->hold_queue.lock);
1879 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1880 struct dst_entry *dst,
1881 const struct flowi *fl,
1886 struct net_device *dev;
1887 struct dst_entry *dst1;
1888 struct xfrm_dst *xdst;
1890 xdst = xfrm_alloc_dst(net, family);
1894 if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0)
1897 dst1 = &xdst->u.dst;
1901 dst_copy_metrics(dst1, dst);
1903 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1904 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1905 dst1->lastuse = jiffies;
1907 dst1->input = dst_discard;
1908 dst1->output = xdst_queue_output;
1914 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1921 err = xfrm_fill_dst(xdst, dev, fl);
1930 xdst = ERR_PTR(err);
1934 static struct flow_cache_object *
1935 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1936 struct flow_cache_object *oldflo, void *ctx)
1938 struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1939 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1940 struct xfrm_dst *xdst, *new_xdst;
1941 int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1943 /* Check if the policies from old bundle are usable */
1946 xdst = container_of(oldflo, struct xfrm_dst, flo);
1947 num_pols = xdst->num_pols;
1948 num_xfrms = xdst->num_xfrms;
1950 for (i = 0; i < num_pols; i++) {
1951 pols[i] = xdst->pols[i];
1952 pol_dead |= pols[i]->walk.dead;
1955 dst_free(&xdst->u.dst);
1963 /* Resolve policies to use if we couldn't get them from
1964 * previous cache entry */
1967 pols[0] = __xfrm_policy_lookup(net, fl, family,
1968 flow_to_policy_dir(dir));
1969 err = xfrm_expand_policies(fl, family, pols,
1970 &num_pols, &num_xfrms);
1976 goto make_dummy_bundle;
1979 new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1980 if (IS_ERR(new_xdst)) {
1981 err = PTR_ERR(new_xdst);
1985 goto make_dummy_bundle;
1986 dst_hold(&xdst->u.dst);
1988 } else if (new_xdst == NULL) {
1991 goto make_dummy_bundle;
1992 xdst->num_xfrms = 0;
1993 dst_hold(&xdst->u.dst);
1997 /* Kill the previous bundle */
1999 /* The policies were stolen for newly generated bundle */
2001 dst_free(&xdst->u.dst);
2004 /* Flow cache does not have reference, it dst_free()'s,
2005 * but we do need to return one reference for original caller */
2006 dst_hold(&new_xdst->u.dst);
2007 return &new_xdst->flo;
2010 /* We found policies, but there's no bundles to instantiate:
2011 * either because the policy blocks, has no transformations or
2012 * we could not build template (no xfrm_states).*/
2013 xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
2015 xfrm_pols_put(pols, num_pols);
2016 return ERR_CAST(xdst);
2018 xdst->num_pols = num_pols;
2019 xdst->num_xfrms = num_xfrms;
2020 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2022 dst_hold(&xdst->u.dst);
2026 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2029 dst_free(&xdst->u.dst);
2031 xfrm_pols_put(pols, num_pols);
2032 return ERR_PTR(err);
2035 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2036 struct dst_entry *dst_orig)
2038 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2039 struct dst_entry *ret;
2042 dst_release(dst_orig);
2043 return ERR_PTR(-EINVAL);
2045 ret = afinfo->blackhole_route(net, dst_orig);
2047 xfrm_policy_put_afinfo(afinfo);
2052 /* Main function: finds/creates a bundle for given flow.
2054 * At the moment we eat a raw IP route. Mostly to speed up lookups
2055 * on interfaces with disabled IPsec.
2057 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2058 const struct flowi *fl,
2059 struct sock *sk, int flags)
2061 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2062 struct flow_cache_object *flo;
2063 struct xfrm_dst *xdst;
2064 struct dst_entry *dst, *route;
2065 u16 family = dst_orig->ops->family;
2066 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2067 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2073 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2075 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2076 err = xfrm_expand_policies(fl, family, pols,
2077 &num_pols, &num_xfrms);
2082 if (num_xfrms <= 0) {
2083 drop_pols = num_pols;
2087 xdst = xfrm_resolve_and_create_bundle(
2091 xfrm_pols_put(pols, num_pols);
2092 err = PTR_ERR(xdst);
2094 } else if (xdst == NULL) {
2096 drop_pols = num_pols;
2100 dst_hold(&xdst->u.dst);
2101 xdst->u.dst.flags |= DST_NOCACHE;
2102 route = xdst->route;
2107 /* To accelerate a bit... */
2108 if ((dst_orig->flags & DST_NOXFRM) ||
2109 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2112 flo = flow_cache_lookup(net, fl, family, dir,
2113 xfrm_bundle_lookup, dst_orig);
2120 xdst = container_of(flo, struct xfrm_dst, flo);
2122 num_pols = xdst->num_pols;
2123 num_xfrms = xdst->num_xfrms;
2124 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2125 route = xdst->route;
2129 if (route == NULL && num_xfrms > 0) {
2130 /* The only case when xfrm_bundle_lookup() returns a
2131 * bundle with null route, is when the template could
2132 * not be resolved. It means policies are there, but
2133 * bundle could not be created, since we don't yet
2134 * have the xfrm_state's. We need to wait for KM to
2135 * negotiate new SA's or bail out with error.*/
2136 if (net->xfrm.sysctl_larval_drop) {
2138 xfrm_pols_put(pols, drop_pols);
2139 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2141 return make_blackhole(net, family, dst_orig);
2146 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2154 if ((flags & XFRM_LOOKUP_ICMP) &&
2155 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2160 for (i = 0; i < num_pols; i++)
2161 pols[i]->curlft.use_time = get_seconds();
2163 if (num_xfrms < 0) {
2164 /* Prohibit the flow */
2165 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2168 } else if (num_xfrms > 0) {
2169 /* Flow transformed */
2170 dst_release(dst_orig);
2172 /* Flow passes untransformed */
2177 xfrm_pols_put(pols, drop_pols);
2178 if (dst && dst->xfrm &&
2179 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2180 dst->flags |= DST_XFRM_TUNNEL;
2184 if (!(flags & XFRM_LOOKUP_ICMP)) {
2192 dst_release(dst_orig);
2193 xfrm_pols_put(pols, drop_pols);
2194 return ERR_PTR(err);
2196 EXPORT_SYMBOL(xfrm_lookup);
2199 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2201 struct xfrm_state *x;
2203 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2205 x = skb->sp->xvec[idx];
2206 if (!x->type->reject)
2208 return x->type->reject(x, skb, fl);
2211 /* When skb is transformed back to its "native" form, we have to
2212 * check policy restrictions. At the moment we make this in maximally
2213 * stupid way. Shame on me. :-) Of course, connected sockets must
2214 * have policy cached at them.
2218 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2219 unsigned short family)
2221 if (xfrm_state_kern(x))
2222 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2223 return x->id.proto == tmpl->id.proto &&
2224 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2225 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2226 x->props.mode == tmpl->mode &&
2227 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2228 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2229 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2230 xfrm_state_addr_cmp(tmpl, x, family));
2234 * 0 or more than 0 is returned when validation is succeeded (either bypass
2235 * because of optional transport mode, or next index of the mathced secpath
2236 * state with the template.
2237 * -1 is returned when no matching template is found.
2238 * Otherwise "-2 - errored_index" is returned.
2241 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2242 unsigned short family)
2246 if (tmpl->optional) {
2247 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2251 for (; idx < sp->len; idx++) {
2252 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2254 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2263 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2264 unsigned int family, int reverse)
2266 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2269 if (unlikely(afinfo == NULL))
2270 return -EAFNOSUPPORT;
2272 afinfo->decode_session(skb, fl, reverse);
2273 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2274 xfrm_policy_put_afinfo(afinfo);
2277 EXPORT_SYMBOL(__xfrm_decode_session);
2279 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2281 for (; k < sp->len; k++) {
2282 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2291 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2292 unsigned short family)
2294 struct net *net = dev_net(skb->dev);
2295 struct xfrm_policy *pol;
2296 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2305 reverse = dir & ~XFRM_POLICY_MASK;
2306 dir &= XFRM_POLICY_MASK;
2307 fl_dir = policy_to_flow_dir(dir);
2309 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2310 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2314 nf_nat_decode_session(skb, &fl, family);
2316 /* First, check used SA against their selectors. */
2320 for (i = skb->sp->len-1; i >= 0; i--) {
2321 struct xfrm_state *x = skb->sp->xvec[i];
2322 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2323 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2330 if (sk && sk->sk_policy[dir]) {
2331 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2333 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2339 struct flow_cache_object *flo;
2341 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2342 xfrm_policy_lookup, NULL);
2343 if (IS_ERR_OR_NULL(flo))
2344 pol = ERR_CAST(flo);
2346 pol = container_of(flo, struct xfrm_policy, flo);
2350 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2355 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2356 xfrm_secpath_reject(xerr_idx, skb, &fl);
2357 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2363 pol->curlft.use_time = get_seconds();
2367 #ifdef CONFIG_XFRM_SUB_POLICY
2368 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2369 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2373 if (IS_ERR(pols[1])) {
2374 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2377 pols[1]->curlft.use_time = get_seconds();
2383 if (pol->action == XFRM_POLICY_ALLOW) {
2384 struct sec_path *sp;
2385 static struct sec_path dummy;
2386 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2387 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2388 struct xfrm_tmpl **tpp = tp;
2392 if ((sp = skb->sp) == NULL)
2395 for (pi = 0; pi < npols; pi++) {
2396 if (pols[pi] != pol &&
2397 pols[pi]->action != XFRM_POLICY_ALLOW) {
2398 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2401 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2402 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2405 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2406 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2410 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2414 /* For each tunnel xfrm, find the first matching tmpl.
2415 * For each tmpl before that, find corresponding xfrm.
2416 * Order is _important_. Later we will implement
2417 * some barriers, but at the moment barriers
2418 * are implied between each two transformations.
2420 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2421 k = xfrm_policy_ok(tpp[i], sp, k, family);
2424 /* "-2 - errored_index" returned */
2426 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2431 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2432 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2436 xfrm_pols_put(pols, npols);
2439 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2442 xfrm_secpath_reject(xerr_idx, skb, &fl);
2444 xfrm_pols_put(pols, npols);
2447 EXPORT_SYMBOL(__xfrm_policy_check);
2449 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2451 struct net *net = dev_net(skb->dev);
2453 struct dst_entry *dst;
2456 if (xfrm_decode_session(skb, &fl, family) < 0) {
2457 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2463 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2468 skb_dst_set(skb, dst);
2471 EXPORT_SYMBOL(__xfrm_route_forward);
2473 /* Optimize later using cookies and generation ids. */
2475 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2477 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2478 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2479 * get validated by dst_ops->check on every use. We do this
2480 * because when a normal route referenced by an XFRM dst is
2481 * obsoleted we do not go looking around for all parent
2482 * referencing XFRM dsts so that we can invalidate them. It
2483 * is just too much work. Instead we make the checks here on
2484 * every use. For example:
2486 * XFRM dst A --> IPv4 dst X
2488 * X is the "xdst->route" of A (X is also the "dst->path" of A
2489 * in this example). If X is marked obsolete, "A" will not
2490 * notice. That's what we are validating here via the
2491 * stale_bundle() check.
2493 * When a policy's bundle is pruned, we dst_free() the XFRM
2494 * dst which causes it's ->obsolete field to be set to
2495 * DST_OBSOLETE_DEAD. If an XFRM dst has been pruned like
2496 * this, we want to force a new route lookup.
2498 if (dst->obsolete < 0 && !stale_bundle(dst))
2504 static int stale_bundle(struct dst_entry *dst)
2506 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2509 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2511 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2512 dst->dev = dev_net(dev)->loopback_dev;
2517 EXPORT_SYMBOL(xfrm_dst_ifdown);
2519 static void xfrm_link_failure(struct sk_buff *skb)
2521 /* Impossible. Such dst must be popped before reaches point of failure. */
2524 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2527 if (dst->obsolete) {
2535 void xfrm_garbage_collect(struct net *net)
2537 flow_cache_flush(net);
2539 EXPORT_SYMBOL(xfrm_garbage_collect);
2541 static void xfrm_garbage_collect_deferred(struct net *net)
2543 flow_cache_flush_deferred(net);
2546 static void xfrm_init_pmtu(struct dst_entry *dst)
2549 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2550 u32 pmtu, route_mtu_cached;
2552 pmtu = dst_mtu(dst->child);
2553 xdst->child_mtu_cached = pmtu;
2555 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2557 route_mtu_cached = dst_mtu(xdst->route);
2558 xdst->route_mtu_cached = route_mtu_cached;
2560 if (pmtu > route_mtu_cached)
2561 pmtu = route_mtu_cached;
2563 dst_metric_set(dst, RTAX_MTU, pmtu);
2564 } while ((dst = dst->next));
2567 /* Check that the bundle accepts the flow and its components are
2571 static int xfrm_bundle_ok(struct xfrm_dst *first)
2573 struct dst_entry *dst = &first->u.dst;
2574 struct xfrm_dst *last;
2577 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2578 (dst->dev && !netif_running(dst->dev)))
2581 if (dst->flags & DST_XFRM_QUEUE)
2587 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2589 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2591 if (xdst->xfrm_genid != dst->xfrm->genid)
2593 if (xdst->num_pols > 0 &&
2594 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2597 mtu = dst_mtu(dst->child);
2598 if (xdst->child_mtu_cached != mtu) {
2600 xdst->child_mtu_cached = mtu;
2603 if (!dst_check(xdst->route, xdst->route_cookie))
2605 mtu = dst_mtu(xdst->route);
2606 if (xdst->route_mtu_cached != mtu) {
2608 xdst->route_mtu_cached = mtu;
2612 } while (dst->xfrm);
2617 mtu = last->child_mtu_cached;
2621 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2622 if (mtu > last->route_mtu_cached)
2623 mtu = last->route_mtu_cached;
2624 dst_metric_set(dst, RTAX_MTU, mtu);
2629 last = (struct xfrm_dst *)last->u.dst.next;
2630 last->child_mtu_cached = mtu;
2636 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2638 return dst_metric_advmss(dst->path);
2641 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2643 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2645 return mtu ? : dst_mtu(dst->path);
2648 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2649 struct sk_buff *skb,
2652 return dst->path->ops->neigh_lookup(dst, skb, daddr);
2655 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2659 if (unlikely(afinfo == NULL))
2661 if (unlikely(afinfo->family >= NPROTO))
2662 return -EAFNOSUPPORT;
2663 spin_lock(&xfrm_policy_afinfo_lock);
2664 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2667 struct dst_ops *dst_ops = afinfo->dst_ops;
2668 if (likely(dst_ops->kmem_cachep == NULL))
2669 dst_ops->kmem_cachep = xfrm_dst_cache;
2670 if (likely(dst_ops->check == NULL))
2671 dst_ops->check = xfrm_dst_check;
2672 if (likely(dst_ops->default_advmss == NULL))
2673 dst_ops->default_advmss = xfrm_default_advmss;
2674 if (likely(dst_ops->mtu == NULL))
2675 dst_ops->mtu = xfrm_mtu;
2676 if (likely(dst_ops->negative_advice == NULL))
2677 dst_ops->negative_advice = xfrm_negative_advice;
2678 if (likely(dst_ops->link_failure == NULL))
2679 dst_ops->link_failure = xfrm_link_failure;
2680 if (likely(dst_ops->neigh_lookup == NULL))
2681 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2682 if (likely(afinfo->garbage_collect == NULL))
2683 afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2684 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2686 spin_unlock(&xfrm_policy_afinfo_lock);
2690 struct dst_ops *xfrm_dst_ops;
2692 switch (afinfo->family) {
2694 xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2696 #if IS_ENABLED(CONFIG_IPV6)
2698 xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2704 *xfrm_dst_ops = *afinfo->dst_ops;
2710 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2712 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2715 if (unlikely(afinfo == NULL))
2717 if (unlikely(afinfo->family >= NPROTO))
2718 return -EAFNOSUPPORT;
2719 spin_lock(&xfrm_policy_afinfo_lock);
2720 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2721 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2724 RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2727 spin_unlock(&xfrm_policy_afinfo_lock);
2729 struct dst_ops *dst_ops = afinfo->dst_ops;
2733 dst_ops->kmem_cachep = NULL;
2734 dst_ops->check = NULL;
2735 dst_ops->negative_advice = NULL;
2736 dst_ops->link_failure = NULL;
2737 afinfo->garbage_collect = NULL;
2741 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2743 static void __net_init xfrm_dst_ops_init(struct net *net)
2745 struct xfrm_policy_afinfo *afinfo;
2748 afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2750 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2751 #if IS_ENABLED(CONFIG_IPV6)
2752 afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2754 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2759 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2761 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2765 xfrm_garbage_collect(dev_net(dev));
2770 static struct notifier_block xfrm_dev_notifier = {
2771 .notifier_call = xfrm_dev_event,
2774 #ifdef CONFIG_XFRM_STATISTICS
2775 static int __net_init xfrm_statistics_init(struct net *net)
2778 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2779 if (!net->mib.xfrm_statistics)
2781 rv = xfrm_proc_init(net);
2783 free_percpu(net->mib.xfrm_statistics);
2787 static void xfrm_statistics_fini(struct net *net)
2789 xfrm_proc_fini(net);
2790 free_percpu(net->mib.xfrm_statistics);
2793 static int __net_init xfrm_statistics_init(struct net *net)
2798 static void xfrm_statistics_fini(struct net *net)
2803 static int __net_init xfrm_policy_init(struct net *net)
2805 unsigned int hmask, sz;
2808 if (net_eq(net, &init_net))
2809 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2810 sizeof(struct xfrm_dst),
2811 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2815 sz = (hmask+1) * sizeof(struct hlist_head);
2817 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2818 if (!net->xfrm.policy_byidx)
2820 net->xfrm.policy_idx_hmask = hmask;
2822 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2823 struct xfrm_policy_hash *htab;
2825 net->xfrm.policy_count[dir] = 0;
2826 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2828 htab = &net->xfrm.policy_bydst[dir];
2829 htab->table = xfrm_hash_alloc(sz);
2832 htab->hmask = hmask;
2835 INIT_LIST_HEAD(&net->xfrm.policy_all);
2836 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2837 if (net_eq(net, &init_net))
2838 register_netdevice_notifier(&xfrm_dev_notifier);
2842 for (dir--; dir >= 0; dir--) {
2843 struct xfrm_policy_hash *htab;
2845 htab = &net->xfrm.policy_bydst[dir];
2846 xfrm_hash_free(htab->table, sz);
2848 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2853 static void xfrm_policy_fini(struct net *net)
2858 flush_work(&net->xfrm.policy_hash_work);
2859 #ifdef CONFIG_XFRM_SUB_POLICY
2860 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2862 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2864 WARN_ON(!list_empty(&net->xfrm.policy_all));
2866 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2867 struct xfrm_policy_hash *htab;
2869 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2871 htab = &net->xfrm.policy_bydst[dir];
2872 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2873 WARN_ON(!hlist_empty(htab->table));
2874 xfrm_hash_free(htab->table, sz);
2877 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2878 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2879 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2882 static int __net_init xfrm_net_init(struct net *net)
2886 rv = xfrm_statistics_init(net);
2888 goto out_statistics;
2889 rv = xfrm_state_init(net);
2892 rv = xfrm_policy_init(net);
2895 xfrm_dst_ops_init(net);
2896 rv = xfrm_sysctl_init(net);
2899 rv = flow_cache_init(net);
2903 /* Initialize the per-net locks here */
2904 spin_lock_init(&net->xfrm.xfrm_state_lock);
2905 rwlock_init(&net->xfrm.xfrm_policy_lock);
2906 mutex_init(&net->xfrm.xfrm_cfg_mutex);
2911 xfrm_sysctl_fini(net);
2913 xfrm_policy_fini(net);
2915 xfrm_state_fini(net);
2917 xfrm_statistics_fini(net);
2922 static void __net_exit xfrm_net_exit(struct net *net)
2924 flow_cache_fini(net);
2925 xfrm_sysctl_fini(net);
2926 xfrm_policy_fini(net);
2927 xfrm_state_fini(net);
2928 xfrm_statistics_fini(net);
2931 static struct pernet_operations __net_initdata xfrm_net_ops = {
2932 .init = xfrm_net_init,
2933 .exit = xfrm_net_exit,
2936 void __init xfrm_init(void)
2938 register_pernet_subsys(&xfrm_net_ops);
2942 #ifdef CONFIG_AUDITSYSCALL
2943 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2944 struct audit_buffer *audit_buf)
2946 struct xfrm_sec_ctx *ctx = xp->security;
2947 struct xfrm_selector *sel = &xp->selector;
2950 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2951 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2953 switch (sel->family) {
2955 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2956 if (sel->prefixlen_s != 32)
2957 audit_log_format(audit_buf, " src_prefixlen=%d",
2959 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2960 if (sel->prefixlen_d != 32)
2961 audit_log_format(audit_buf, " dst_prefixlen=%d",
2965 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2966 if (sel->prefixlen_s != 128)
2967 audit_log_format(audit_buf, " src_prefixlen=%d",
2969 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2970 if (sel->prefixlen_d != 128)
2971 audit_log_format(audit_buf, " dst_prefixlen=%d",
2977 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
2979 struct audit_buffer *audit_buf;
2981 audit_buf = xfrm_audit_start("SPD-add");
2982 if (audit_buf == NULL)
2984 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2985 audit_log_format(audit_buf, " res=%u", result);
2986 xfrm_audit_common_policyinfo(xp, audit_buf);
2987 audit_log_end(audit_buf);
2989 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2991 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2994 struct audit_buffer *audit_buf;
2996 audit_buf = xfrm_audit_start("SPD-delete");
2997 if (audit_buf == NULL)
2999 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3000 audit_log_format(audit_buf, " res=%u", result);
3001 xfrm_audit_common_policyinfo(xp, audit_buf);
3002 audit_log_end(audit_buf);
3004 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3007 #ifdef CONFIG_XFRM_MIGRATE
3008 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3009 const struct xfrm_selector *sel_tgt)
3011 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3012 if (sel_tgt->family == sel_cmp->family &&
3013 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3015 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3017 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3018 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3022 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3029 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3030 u8 dir, u8 type, struct net *net)
3032 struct xfrm_policy *pol, *ret = NULL;
3033 struct hlist_head *chain;
3036 read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3037 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3038 hlist_for_each_entry(pol, chain, bydst) {
3039 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3040 pol->type == type) {
3042 priority = ret->priority;
3046 chain = &net->xfrm.policy_inexact[dir];
3047 hlist_for_each_entry(pol, chain, bydst) {
3048 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3049 pol->type == type &&
3050 pol->priority < priority) {
3059 read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3064 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3068 if (t->mode == m->mode && t->id.proto == m->proto &&
3069 (m->reqid == 0 || t->reqid == m->reqid)) {
3071 case XFRM_MODE_TUNNEL:
3072 case XFRM_MODE_BEET:
3073 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3075 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3080 case XFRM_MODE_TRANSPORT:
3081 /* in case of transport mode, template does not store
3082 any IP addresses, hence we just compare mode and
3093 /* update endpoint address(es) of template(s) */
3094 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3095 struct xfrm_migrate *m, int num_migrate)
3097 struct xfrm_migrate *mp;
3100 write_lock_bh(&pol->lock);
3101 if (unlikely(pol->walk.dead)) {
3102 /* target policy has been deleted */
3103 write_unlock_bh(&pol->lock);
3107 for (i = 0; i < pol->xfrm_nr; i++) {
3108 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3109 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3112 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3113 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3115 /* update endpoints */
3116 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3117 sizeof(pol->xfrm_vec[i].id.daddr));
3118 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3119 sizeof(pol->xfrm_vec[i].saddr));
3120 pol->xfrm_vec[i].encap_family = mp->new_family;
3122 atomic_inc(&pol->genid);
3126 write_unlock_bh(&pol->lock);
3134 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3138 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3141 for (i = 0; i < num_migrate; i++) {
3142 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3144 xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3147 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3148 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3151 /* check if there is any duplicated entry */
3152 for (j = i + 1; j < num_migrate; j++) {
3153 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3154 sizeof(m[i].old_daddr)) &&
3155 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3156 sizeof(m[i].old_saddr)) &&
3157 m[i].proto == m[j].proto &&
3158 m[i].mode == m[j].mode &&
3159 m[i].reqid == m[j].reqid &&
3160 m[i].old_family == m[j].old_family)
3168 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3169 struct xfrm_migrate *m, int num_migrate,
3170 struct xfrm_kmaddress *k, struct net *net)
3172 int i, err, nx_cur = 0, nx_new = 0;
3173 struct xfrm_policy *pol = NULL;
3174 struct xfrm_state *x, *xc;
3175 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3176 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3177 struct xfrm_migrate *mp;
3179 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3182 /* Stage 1 - find policy */
3183 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3188 /* Stage 2 - find and update state(s) */
3189 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3190 if ((x = xfrm_migrate_state_find(mp, net))) {
3193 if ((xc = xfrm_state_migrate(x, mp))) {
3203 /* Stage 3 - update policy */
3204 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3207 /* Stage 4 - delete old state(s) */
3209 xfrm_states_put(x_cur, nx_cur);
3210 xfrm_states_delete(x_cur, nx_cur);
3213 /* Stage 5 - announce */
3214 km_migrate(sel, dir, type, m, num_migrate, k);
3226 xfrm_states_put(x_cur, nx_cur);
3228 xfrm_states_delete(x_new, nx_new);
3232 EXPORT_SYMBOL(xfrm_migrate);