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 DEFINE_MUTEX(xfrm_cfg_mutex);
43 EXPORT_SYMBOL(xfrm_cfg_mutex);
45 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
46 static struct dst_entry *xfrm_policy_sk_bundles;
47 static DEFINE_RWLOCK(xfrm_policy_lock);
49 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
50 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
53 static struct kmem_cache *xfrm_dst_cache __read_mostly;
55 static void xfrm_init_pmtu(struct dst_entry *dst);
56 static int stale_bundle(struct dst_entry *dst);
57 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
58 static void xfrm_policy_queue_process(unsigned long arg);
60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
64 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
66 const struct flowi4 *fl4 = &fl->u.ip4;
68 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
69 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
70 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
71 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
72 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
73 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
77 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
79 const struct flowi6 *fl6 = &fl->u.ip6;
81 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
82 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
83 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
84 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
85 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
86 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
89 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
90 unsigned short family)
94 return __xfrm4_selector_match(sel, fl);
96 return __xfrm6_selector_match(sel, fl);
101 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
103 struct xfrm_policy_afinfo *afinfo;
105 if (unlikely(family >= NPROTO))
108 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
109 if (unlikely(!afinfo))
114 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
119 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
120 const xfrm_address_t *saddr,
121 const xfrm_address_t *daddr,
124 struct xfrm_policy_afinfo *afinfo;
125 struct dst_entry *dst;
127 afinfo = xfrm_policy_get_afinfo(family);
128 if (unlikely(afinfo == NULL))
129 return ERR_PTR(-EAFNOSUPPORT);
131 dst = afinfo->dst_lookup(net, tos, saddr, daddr);
133 xfrm_policy_put_afinfo(afinfo);
138 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
139 xfrm_address_t *prev_saddr,
140 xfrm_address_t *prev_daddr,
143 struct net *net = xs_net(x);
144 xfrm_address_t *saddr = &x->props.saddr;
145 xfrm_address_t *daddr = &x->id.daddr;
146 struct dst_entry *dst;
148 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
152 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
157 dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
160 if (prev_saddr != saddr)
161 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
162 if (prev_daddr != daddr)
163 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
169 static inline unsigned long make_jiffies(long secs)
171 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
172 return MAX_SCHEDULE_TIMEOUT-1;
177 static void xfrm_policy_timer(unsigned long data)
179 struct xfrm_policy *xp = (struct xfrm_policy*)data;
180 unsigned long now = get_seconds();
181 long next = LONG_MAX;
185 read_lock(&xp->lock);
187 if (unlikely(xp->walk.dead))
190 dir = xfrm_policy_id2dir(xp->index);
192 if (xp->lft.hard_add_expires_seconds) {
193 long tmo = xp->lft.hard_add_expires_seconds +
194 xp->curlft.add_time - now;
200 if (xp->lft.hard_use_expires_seconds) {
201 long tmo = xp->lft.hard_use_expires_seconds +
202 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
208 if (xp->lft.soft_add_expires_seconds) {
209 long tmo = xp->lft.soft_add_expires_seconds +
210 xp->curlft.add_time - now;
213 tmo = XFRM_KM_TIMEOUT;
218 if (xp->lft.soft_use_expires_seconds) {
219 long tmo = xp->lft.soft_use_expires_seconds +
220 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
223 tmo = XFRM_KM_TIMEOUT;
230 km_policy_expired(xp, dir, 0, 0);
231 if (next != LONG_MAX &&
232 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
236 read_unlock(&xp->lock);
241 read_unlock(&xp->lock);
242 if (!xfrm_policy_delete(xp, dir))
243 km_policy_expired(xp, dir, 1, 0);
247 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
249 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
251 if (unlikely(pol->walk.dead))
259 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
261 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
263 return !pol->walk.dead;
266 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
268 xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
271 static const struct flow_cache_ops xfrm_policy_fc_ops = {
272 .get = xfrm_policy_flo_get,
273 .check = xfrm_policy_flo_check,
274 .delete = xfrm_policy_flo_delete,
277 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
281 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
283 struct xfrm_policy *policy;
285 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
288 write_pnet(&policy->xp_net, net);
289 INIT_LIST_HEAD(&policy->walk.all);
290 INIT_HLIST_NODE(&policy->bydst);
291 INIT_HLIST_NODE(&policy->byidx);
292 rwlock_init(&policy->lock);
293 atomic_set(&policy->refcnt, 1);
294 skb_queue_head_init(&policy->polq.hold_queue);
295 setup_timer(&policy->timer, xfrm_policy_timer,
296 (unsigned long)policy);
297 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
298 (unsigned long)policy);
299 policy->flo.ops = &xfrm_policy_fc_ops;
303 EXPORT_SYMBOL(xfrm_policy_alloc);
305 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
307 void xfrm_policy_destroy(struct xfrm_policy *policy)
309 BUG_ON(!policy->walk.dead);
311 if (del_timer(&policy->timer))
314 security_xfrm_policy_free(policy->security);
317 EXPORT_SYMBOL(xfrm_policy_destroy);
319 static void xfrm_queue_purge(struct sk_buff_head *list)
323 while ((skb = skb_dequeue(list)) != NULL) {
329 /* Rule must be locked. Release descentant resources, announce
330 * entry dead. The rule must be unlinked from lists to the moment.
333 static void xfrm_policy_kill(struct xfrm_policy *policy)
335 policy->walk.dead = 1;
337 atomic_inc(&policy->genid);
339 del_timer(&policy->polq.hold_timer);
340 xfrm_queue_purge(&policy->polq.hold_queue);
342 if (del_timer(&policy->timer))
343 xfrm_pol_put(policy);
345 xfrm_pol_put(policy);
348 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
350 static inline unsigned int idx_hash(struct net *net, u32 index)
352 return __idx_hash(index, net->xfrm.policy_idx_hmask);
355 static struct hlist_head *policy_hash_bysel(struct net *net,
356 const struct xfrm_selector *sel,
357 unsigned short family, int dir)
359 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
360 unsigned int hash = __sel_hash(sel, family, hmask);
362 return (hash == hmask + 1 ?
363 &net->xfrm.policy_inexact[dir] :
364 net->xfrm.policy_bydst[dir].table + hash);
367 static struct hlist_head *policy_hash_direct(struct net *net,
368 const xfrm_address_t *daddr,
369 const xfrm_address_t *saddr,
370 unsigned short family, int dir)
372 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
373 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
375 return net->xfrm.policy_bydst[dir].table + hash;
378 static void xfrm_dst_hash_transfer(struct hlist_head *list,
379 struct hlist_head *ndsttable,
380 unsigned int nhashmask)
382 struct hlist_node *tmp, *entry0 = NULL;
383 struct xfrm_policy *pol;
387 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
390 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
391 pol->family, nhashmask);
393 hlist_del(&pol->bydst);
394 hlist_add_head(&pol->bydst, ndsttable+h);
399 hlist_del(&pol->bydst);
400 hlist_add_after(entry0, &pol->bydst);
402 entry0 = &pol->bydst;
404 if (!hlist_empty(list)) {
410 static void xfrm_idx_hash_transfer(struct hlist_head *list,
411 struct hlist_head *nidxtable,
412 unsigned int nhashmask)
414 struct hlist_node *tmp;
415 struct xfrm_policy *pol;
417 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
420 h = __idx_hash(pol->index, nhashmask);
421 hlist_add_head(&pol->byidx, nidxtable+h);
425 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
427 return ((old_hmask + 1) << 1) - 1;
430 static void xfrm_bydst_resize(struct net *net, int dir)
432 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
433 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
434 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
435 struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
436 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
442 write_lock_bh(&xfrm_policy_lock);
444 for (i = hmask; i >= 0; i--)
445 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
447 net->xfrm.policy_bydst[dir].table = ndst;
448 net->xfrm.policy_bydst[dir].hmask = nhashmask;
450 write_unlock_bh(&xfrm_policy_lock);
452 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
455 static void xfrm_byidx_resize(struct net *net, int total)
457 unsigned int hmask = net->xfrm.policy_idx_hmask;
458 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
459 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
460 struct hlist_head *oidx = net->xfrm.policy_byidx;
461 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
467 write_lock_bh(&xfrm_policy_lock);
469 for (i = hmask; i >= 0; i--)
470 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
472 net->xfrm.policy_byidx = nidx;
473 net->xfrm.policy_idx_hmask = nhashmask;
475 write_unlock_bh(&xfrm_policy_lock);
477 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
480 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
482 unsigned int cnt = net->xfrm.policy_count[dir];
483 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
488 if ((hmask + 1) < xfrm_policy_hashmax &&
495 static inline int xfrm_byidx_should_resize(struct net *net, int total)
497 unsigned int hmask = net->xfrm.policy_idx_hmask;
499 if ((hmask + 1) < xfrm_policy_hashmax &&
506 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
508 read_lock_bh(&xfrm_policy_lock);
509 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
510 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
511 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
512 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
513 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
514 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
515 si->spdhcnt = net->xfrm.policy_idx_hmask;
516 si->spdhmcnt = xfrm_policy_hashmax;
517 read_unlock_bh(&xfrm_policy_lock);
519 EXPORT_SYMBOL(xfrm_spd_getinfo);
521 static DEFINE_MUTEX(hash_resize_mutex);
522 static void xfrm_hash_resize(struct work_struct *work)
524 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
527 mutex_lock(&hash_resize_mutex);
530 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
531 if (xfrm_bydst_should_resize(net, dir, &total))
532 xfrm_bydst_resize(net, dir);
534 if (xfrm_byidx_should_resize(net, total))
535 xfrm_byidx_resize(net, total);
537 mutex_unlock(&hash_resize_mutex);
540 /* Generate new index... KAME seems to generate them ordered by cost
541 * of an absolute inpredictability of ordering of rules. This will not pass. */
542 static u32 xfrm_gen_index(struct net *net, int dir)
544 static u32 idx_generator;
547 struct hlist_head *list;
548 struct xfrm_policy *p;
552 idx = (idx_generator | dir);
556 list = net->xfrm.policy_byidx + idx_hash(net, idx);
558 hlist_for_each_entry(p, list, byidx) {
559 if (p->index == idx) {
569 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
571 u32 *p1 = (u32 *) s1;
572 u32 *p2 = (u32 *) s2;
573 int len = sizeof(struct xfrm_selector) / sizeof(u32);
576 for (i = 0; i < len; i++) {
584 static void xfrm_policy_requeue(struct xfrm_policy *old,
585 struct xfrm_policy *new)
587 struct xfrm_policy_queue *pq = &old->polq;
588 struct sk_buff_head list;
590 __skb_queue_head_init(&list);
592 spin_lock_bh(&pq->hold_queue.lock);
593 skb_queue_splice_init(&pq->hold_queue, &list);
594 del_timer(&pq->hold_timer);
595 spin_unlock_bh(&pq->hold_queue.lock);
597 if (skb_queue_empty(&list))
602 spin_lock_bh(&pq->hold_queue.lock);
603 skb_queue_splice(&list, &pq->hold_queue);
604 pq->timeout = XFRM_QUEUE_TMO_MIN;
605 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(&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(&xfrm_policy_lock);
647 if (policy->priority > pol->priority)
649 } else if (policy->priority >= pol->priority) {
650 newpos = &pol->bydst;
657 hlist_add_after(newpos, &policy->bydst);
659 hlist_add_head(&policy->bydst, chain);
660 xfrm_pol_hold(policy);
661 net->xfrm.policy_count[dir]++;
662 atomic_inc(&flow_cache_genid);
665 xfrm_policy_requeue(delpol, policy);
666 __xfrm_policy_unlink(delpol, dir);
668 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
669 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
670 policy->curlft.add_time = get_seconds();
671 policy->curlft.use_time = 0;
672 if (!mod_timer(&policy->timer, jiffies + HZ))
673 xfrm_pol_hold(policy);
674 list_add(&policy->walk.all, &net->xfrm.policy_all);
675 write_unlock_bh(&xfrm_policy_lock);
678 xfrm_policy_kill(delpol);
679 else if (xfrm_bydst_should_resize(net, dir, NULL))
680 schedule_work(&net->xfrm.policy_hash_work);
684 EXPORT_SYMBOL(xfrm_policy_insert);
686 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
687 int dir, struct xfrm_selector *sel,
688 struct xfrm_sec_ctx *ctx, int delete,
691 struct xfrm_policy *pol, *ret;
692 struct hlist_head *chain;
695 write_lock_bh(&xfrm_policy_lock);
696 chain = policy_hash_bysel(net, sel, sel->family, dir);
698 hlist_for_each_entry(pol, chain, bydst) {
699 if (pol->type == type &&
700 (mark & pol->mark.m) == pol->mark.v &&
701 !selector_cmp(sel, &pol->selector) &&
702 xfrm_sec_ctx_match(ctx, pol->security)) {
705 *err = security_xfrm_policy_delete(
708 write_unlock_bh(&xfrm_policy_lock);
711 __xfrm_policy_unlink(pol, dir);
717 write_unlock_bh(&xfrm_policy_lock);
720 xfrm_policy_kill(ret);
723 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
725 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
726 int dir, u32 id, int delete, int *err)
728 struct xfrm_policy *pol, *ret;
729 struct hlist_head *chain;
732 if (xfrm_policy_id2dir(id) != dir)
736 write_lock_bh(&xfrm_policy_lock);
737 chain = net->xfrm.policy_byidx + idx_hash(net, id);
739 hlist_for_each_entry(pol, chain, byidx) {
740 if (pol->type == type && pol->index == id &&
741 (mark & pol->mark.m) == pol->mark.v) {
744 *err = security_xfrm_policy_delete(
747 write_unlock_bh(&xfrm_policy_lock);
750 __xfrm_policy_unlink(pol, dir);
756 write_unlock_bh(&xfrm_policy_lock);
759 xfrm_policy_kill(ret);
762 EXPORT_SYMBOL(xfrm_policy_byid);
764 #ifdef CONFIG_SECURITY_NETWORK_XFRM
766 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
770 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
771 struct xfrm_policy *pol;
774 hlist_for_each_entry(pol,
775 &net->xfrm.policy_inexact[dir], bydst) {
776 if (pol->type != type)
778 err = security_xfrm_policy_delete(pol->security);
780 xfrm_audit_policy_delete(pol, 0,
781 audit_info->loginuid,
782 audit_info->sessionid,
787 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
788 hlist_for_each_entry(pol,
789 net->xfrm.policy_bydst[dir].table + i,
791 if (pol->type != type)
793 err = security_xfrm_policy_delete(
796 xfrm_audit_policy_delete(pol, 0,
797 audit_info->loginuid,
798 audit_info->sessionid,
809 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
815 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
817 int dir, err = 0, cnt = 0;
819 write_lock_bh(&xfrm_policy_lock);
821 err = xfrm_policy_flush_secctx_check(net, type, audit_info);
825 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
826 struct xfrm_policy *pol;
830 hlist_for_each_entry(pol,
831 &net->xfrm.policy_inexact[dir], bydst) {
832 if (pol->type != type)
834 __xfrm_policy_unlink(pol, dir);
835 write_unlock_bh(&xfrm_policy_lock);
838 xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
839 audit_info->sessionid,
842 xfrm_policy_kill(pol);
844 write_lock_bh(&xfrm_policy_lock);
848 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
850 hlist_for_each_entry(pol,
851 net->xfrm.policy_bydst[dir].table + i,
853 if (pol->type != type)
855 __xfrm_policy_unlink(pol, dir);
856 write_unlock_bh(&xfrm_policy_lock);
859 xfrm_audit_policy_delete(pol, 1,
860 audit_info->loginuid,
861 audit_info->sessionid,
863 xfrm_policy_kill(pol);
865 write_lock_bh(&xfrm_policy_lock);
874 write_unlock_bh(&xfrm_policy_lock);
877 EXPORT_SYMBOL(xfrm_policy_flush);
879 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
880 int (*func)(struct xfrm_policy *, int, int, void*),
883 struct xfrm_policy *pol;
884 struct xfrm_policy_walk_entry *x;
887 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
888 walk->type != XFRM_POLICY_TYPE_ANY)
891 if (list_empty(&walk->walk.all) && walk->seq != 0)
894 write_lock_bh(&xfrm_policy_lock);
895 if (list_empty(&walk->walk.all))
896 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
898 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
899 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
902 pol = container_of(x, struct xfrm_policy, walk);
903 if (walk->type != XFRM_POLICY_TYPE_ANY &&
904 walk->type != pol->type)
906 error = func(pol, xfrm_policy_id2dir(pol->index),
909 list_move_tail(&walk->walk.all, &x->all);
914 if (walk->seq == 0) {
918 list_del_init(&walk->walk.all);
920 write_unlock_bh(&xfrm_policy_lock);
923 EXPORT_SYMBOL(xfrm_policy_walk);
925 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
927 INIT_LIST_HEAD(&walk->walk.all);
932 EXPORT_SYMBOL(xfrm_policy_walk_init);
934 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
936 if (list_empty(&walk->walk.all))
939 write_lock_bh(&xfrm_policy_lock);
940 list_del(&walk->walk.all);
941 write_unlock_bh(&xfrm_policy_lock);
943 EXPORT_SYMBOL(xfrm_policy_walk_done);
946 * Find policy to apply to this flow.
948 * Returns 0 if policy found, else an -errno.
950 static int xfrm_policy_match(const struct xfrm_policy *pol,
951 const struct flowi *fl,
952 u8 type, u16 family, int dir)
954 const struct xfrm_selector *sel = &pol->selector;
958 if (pol->family != family ||
959 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
963 match = xfrm_selector_match(sel, fl, family);
965 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
971 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
972 const struct flowi *fl,
976 struct xfrm_policy *pol, *ret;
977 const xfrm_address_t *daddr, *saddr;
978 struct hlist_head *chain;
981 daddr = xfrm_flowi_daddr(fl, family);
982 saddr = xfrm_flowi_saddr(fl, family);
983 if (unlikely(!daddr || !saddr))
986 read_lock_bh(&xfrm_policy_lock);
987 chain = policy_hash_direct(net, daddr, saddr, family, dir);
989 hlist_for_each_entry(pol, chain, bydst) {
990 err = xfrm_policy_match(pol, fl, type, family, dir);
1000 priority = ret->priority;
1004 chain = &net->xfrm.policy_inexact[dir];
1005 hlist_for_each_entry(pol, chain, bydst) {
1006 err = xfrm_policy_match(pol, fl, type, family, dir);
1014 } else if (pol->priority < priority) {
1022 read_unlock_bh(&xfrm_policy_lock);
1027 static struct xfrm_policy *
1028 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1030 #ifdef CONFIG_XFRM_SUB_POLICY
1031 struct xfrm_policy *pol;
1033 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1037 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1040 static struct flow_cache_object *
1041 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1042 u8 dir, struct flow_cache_object *old_obj, void *ctx)
1044 struct xfrm_policy *pol;
1047 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1049 pol = __xfrm_policy_lookup(net, fl, family, dir);
1050 if (IS_ERR_OR_NULL(pol))
1051 return ERR_CAST(pol);
1053 /* Resolver returns two references:
1054 * one for cache and one for caller of flow_cache_lookup() */
1060 static inline int policy_to_flow_dir(int dir)
1062 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1063 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1064 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1068 case XFRM_POLICY_IN:
1070 case XFRM_POLICY_OUT:
1071 return FLOW_DIR_OUT;
1072 case XFRM_POLICY_FWD:
1073 return FLOW_DIR_FWD;
1077 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1078 const struct flowi *fl)
1080 struct xfrm_policy *pol;
1082 read_lock_bh(&xfrm_policy_lock);
1083 if ((pol = sk->sk_policy[dir]) != NULL) {
1084 bool match = xfrm_selector_match(&pol->selector, fl,
1089 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1093 err = security_xfrm_policy_lookup(pol->security,
1095 policy_to_flow_dir(dir));
1098 else if (err == -ESRCH)
1106 read_unlock_bh(&xfrm_policy_lock);
1110 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1112 struct net *net = xp_net(pol);
1113 struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1116 list_add(&pol->walk.all, &net->xfrm.policy_all);
1117 hlist_add_head(&pol->bydst, chain);
1118 hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1119 net->xfrm.policy_count[dir]++;
1122 if (xfrm_bydst_should_resize(net, dir, NULL))
1123 schedule_work(&net->xfrm.policy_hash_work);
1126 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1129 struct net *net = xp_net(pol);
1131 if (hlist_unhashed(&pol->bydst))
1134 hlist_del(&pol->bydst);
1135 hlist_del(&pol->byidx);
1136 list_del(&pol->walk.all);
1137 net->xfrm.policy_count[dir]--;
1142 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1144 write_lock_bh(&xfrm_policy_lock);
1145 pol = __xfrm_policy_unlink(pol, dir);
1146 write_unlock_bh(&xfrm_policy_lock);
1148 xfrm_policy_kill(pol);
1153 EXPORT_SYMBOL(xfrm_policy_delete);
1155 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1157 struct net *net = xp_net(pol);
1158 struct xfrm_policy *old_pol;
1160 #ifdef CONFIG_XFRM_SUB_POLICY
1161 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1165 write_lock_bh(&xfrm_policy_lock);
1166 old_pol = sk->sk_policy[dir];
1167 sk->sk_policy[dir] = pol;
1169 pol->curlft.add_time = get_seconds();
1170 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1171 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1175 xfrm_policy_requeue(old_pol, pol);
1177 /* Unlinking succeeds always. This is the only function
1178 * allowed to delete or replace socket policy.
1180 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1182 write_unlock_bh(&xfrm_policy_lock);
1185 xfrm_policy_kill(old_pol);
1190 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1192 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1195 newp->selector = old->selector;
1196 if (security_xfrm_policy_clone(old->security,
1199 return NULL; /* ENOMEM */
1201 newp->lft = old->lft;
1202 newp->curlft = old->curlft;
1203 newp->mark = old->mark;
1204 newp->action = old->action;
1205 newp->flags = old->flags;
1206 newp->xfrm_nr = old->xfrm_nr;
1207 newp->index = old->index;
1208 newp->type = old->type;
1209 memcpy(newp->xfrm_vec, old->xfrm_vec,
1210 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1211 write_lock_bh(&xfrm_policy_lock);
1212 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1213 write_unlock_bh(&xfrm_policy_lock);
1219 int __xfrm_sk_clone_policy(struct sock *sk)
1221 struct xfrm_policy *p0 = sk->sk_policy[0],
1222 *p1 = sk->sk_policy[1];
1224 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1225 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1227 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1233 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1234 unsigned short family)
1237 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1239 if (unlikely(afinfo == NULL))
1241 err = afinfo->get_saddr(net, local, remote);
1242 xfrm_policy_put_afinfo(afinfo);
1246 /* Resolve list of templates for the flow, given policy. */
1249 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1250 struct xfrm_state **xfrm, unsigned short family)
1252 struct net *net = xp_net(policy);
1255 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1256 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1259 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1260 struct xfrm_state *x;
1261 xfrm_address_t *remote = daddr;
1262 xfrm_address_t *local = saddr;
1263 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1265 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1266 tmpl->mode == XFRM_MODE_BEET) {
1267 remote = &tmpl->id.daddr;
1268 local = &tmpl->saddr;
1269 if (xfrm_addr_any(local, tmpl->encap_family)) {
1270 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1277 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1279 if (x && x->km.state == XFRM_STATE_VALID) {
1286 error = (x->km.state == XFRM_STATE_ERROR ?
1290 else if (error == -ESRCH)
1293 if (!tmpl->optional)
1299 for (nx--; nx>=0; nx--)
1300 xfrm_state_put(xfrm[nx]);
1305 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1306 struct xfrm_state **xfrm, unsigned short family)
1308 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1309 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1315 for (i = 0; i < npols; i++) {
1316 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1321 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1329 /* found states are sorted for outbound processing */
1331 xfrm_state_sort(xfrm, tpp, cnx, family);
1336 for (cnx--; cnx>=0; cnx--)
1337 xfrm_state_put(tpp[cnx]);
1342 /* Check that the bundle accepts the flow and its components are
1346 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1348 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1354 tos = afinfo->get_tos(fl);
1356 xfrm_policy_put_afinfo(afinfo);
1361 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1363 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1364 struct dst_entry *dst = &xdst->u.dst;
1366 if (xdst->route == NULL) {
1367 /* Dummy bundle - if it has xfrms we were not
1368 * able to build bundle as template resolution failed.
1369 * It means we need to try again resolving. */
1370 if (xdst->num_xfrms > 0)
1372 } else if (dst->flags & DST_XFRM_QUEUE) {
1376 if (stale_bundle(dst))
1384 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1386 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1387 struct dst_entry *dst = &xdst->u.dst;
1391 if (stale_bundle(dst))
1397 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1399 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1400 struct dst_entry *dst = &xdst->u.dst;
1405 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1406 .get = xfrm_bundle_flo_get,
1407 .check = xfrm_bundle_flo_check,
1408 .delete = xfrm_bundle_flo_delete,
1411 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1413 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1414 struct dst_ops *dst_ops;
1415 struct xfrm_dst *xdst;
1418 return ERR_PTR(-EINVAL);
1422 dst_ops = &net->xfrm.xfrm4_dst_ops;
1424 #if IS_ENABLED(CONFIG_IPV6)
1426 dst_ops = &net->xfrm.xfrm6_dst_ops;
1432 xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1435 struct dst_entry *dst = &xdst->u.dst;
1437 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1438 xdst->flo.ops = &xfrm_bundle_fc_ops;
1439 if (afinfo->init_dst)
1440 afinfo->init_dst(net, xdst);
1442 xdst = ERR_PTR(-ENOBUFS);
1444 xfrm_policy_put_afinfo(afinfo);
1449 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1452 struct xfrm_policy_afinfo *afinfo =
1453 xfrm_policy_get_afinfo(dst->ops->family);
1459 err = afinfo->init_path(path, dst, nfheader_len);
1461 xfrm_policy_put_afinfo(afinfo);
1466 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1467 const struct flowi *fl)
1469 struct xfrm_policy_afinfo *afinfo =
1470 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1476 err = afinfo->fill_dst(xdst, dev, fl);
1478 xfrm_policy_put_afinfo(afinfo);
1484 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1485 * all the metrics... Shortly, bundle a bundle.
1488 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1489 struct xfrm_state **xfrm, int nx,
1490 const struct flowi *fl,
1491 struct dst_entry *dst)
1493 struct net *net = xp_net(policy);
1494 unsigned long now = jiffies;
1495 struct net_device *dev;
1496 struct xfrm_mode *inner_mode;
1497 struct dst_entry *dst_prev = NULL;
1498 struct dst_entry *dst0 = NULL;
1502 int nfheader_len = 0;
1503 int trailer_len = 0;
1505 int family = policy->selector.family;
1506 xfrm_address_t saddr, daddr;
1508 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1510 tos = xfrm_get_tos(fl, family);
1517 for (; i < nx; i++) {
1518 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1519 struct dst_entry *dst1 = &xdst->u.dst;
1521 err = PTR_ERR(xdst);
1527 if (xfrm[i]->sel.family == AF_UNSPEC) {
1528 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1529 xfrm_af2proto(family));
1531 err = -EAFNOSUPPORT;
1536 inner_mode = xfrm[i]->inner_mode;
1541 dst_prev->child = dst_clone(dst1);
1542 dst1->flags |= DST_NOHASH;
1546 dst_copy_metrics(dst1, dst);
1548 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1549 family = xfrm[i]->props.family;
1550 dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1558 dst1->xfrm = xfrm[i];
1559 xdst->xfrm_genid = xfrm[i]->genid;
1561 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1562 dst1->flags |= DST_HOST;
1563 dst1->lastuse = now;
1565 dst1->input = dst_discard;
1566 dst1->output = inner_mode->afinfo->output;
1568 dst1->next = dst_prev;
1571 header_len += xfrm[i]->props.header_len;
1572 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1573 nfheader_len += xfrm[i]->props.header_len;
1574 trailer_len += xfrm[i]->props.trailer_len;
1577 dst_prev->child = dst;
1585 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1586 xfrm_init_pmtu(dst_prev);
1588 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1589 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1591 err = xfrm_fill_dst(xdst, dev, fl);
1595 dst_prev->header_len = header_len;
1596 dst_prev->trailer_len = trailer_len;
1597 header_len -= xdst->u.dst.xfrm->props.header_len;
1598 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1606 xfrm_state_put(xfrm[i]);
1610 dst0 = ERR_PTR(err);
1615 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1618 *target = kmalloc(size, GFP_ATOMIC);
1622 memcpy(*target, src, size);
1627 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1629 #ifdef CONFIG_XFRM_SUB_POLICY
1630 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1631 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1639 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1641 #ifdef CONFIG_XFRM_SUB_POLICY
1642 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1643 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1649 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1650 struct xfrm_policy **pols,
1651 int *num_pols, int *num_xfrms)
1655 if (*num_pols == 0 || !pols[0]) {
1660 if (IS_ERR(pols[0]))
1661 return PTR_ERR(pols[0]);
1663 *num_xfrms = pols[0]->xfrm_nr;
1665 #ifdef CONFIG_XFRM_SUB_POLICY
1666 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1667 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1668 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1669 XFRM_POLICY_TYPE_MAIN,
1673 if (IS_ERR(pols[1])) {
1674 xfrm_pols_put(pols, *num_pols);
1675 return PTR_ERR(pols[1]);
1678 (*num_xfrms) += pols[1]->xfrm_nr;
1682 for (i = 0; i < *num_pols; i++) {
1683 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1693 static struct xfrm_dst *
1694 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1695 const struct flowi *fl, u16 family,
1696 struct dst_entry *dst_orig)
1698 struct net *net = xp_net(pols[0]);
1699 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1700 struct dst_entry *dst;
1701 struct xfrm_dst *xdst;
1704 /* Try to instantiate a bundle */
1705 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1707 if (err != 0 && err != -EAGAIN)
1708 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1709 return ERR_PTR(err);
1712 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1714 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1715 return ERR_CAST(dst);
1718 xdst = (struct xfrm_dst *)dst;
1719 xdst->num_xfrms = err;
1721 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1723 err = xfrm_dst_update_origin(dst, fl);
1724 if (unlikely(err)) {
1726 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1727 return ERR_PTR(err);
1730 xdst->num_pols = num_pols;
1731 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1732 xdst->policy_genid = atomic_read(&pols[0]->genid);
1737 static void xfrm_policy_queue_process(unsigned long arg)
1740 struct sk_buff *skb;
1742 struct dst_entry *dst;
1743 struct net_device *dev;
1744 struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1745 struct xfrm_policy_queue *pq = &pol->polq;
1747 struct sk_buff_head list;
1749 spin_lock(&pq->hold_queue.lock);
1750 skb = skb_peek(&pq->hold_queue);
1753 xfrm_decode_session(skb, &fl, dst->ops->family);
1754 spin_unlock(&pq->hold_queue.lock);
1756 dst_hold(dst->path);
1757 dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1762 if (dst->flags & DST_XFRM_QUEUE) {
1765 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1768 pq->timeout = pq->timeout << 1;
1769 mod_timer(&pq->hold_timer, jiffies + pq->timeout);
1775 __skb_queue_head_init(&list);
1777 spin_lock(&pq->hold_queue.lock);
1779 skb_queue_splice_init(&pq->hold_queue, &list);
1780 spin_unlock(&pq->hold_queue.lock);
1782 while (!skb_queue_empty(&list)) {
1783 skb = __skb_dequeue(&list);
1785 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1786 dst_hold(skb_dst(skb)->path);
1787 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1797 skb_dst_set(skb, dst);
1800 err = dst_output(skb);
1808 xfrm_queue_purge(&pq->hold_queue);
1811 static int xdst_queue_output(struct sk_buff *skb)
1813 unsigned long sched_next;
1814 struct dst_entry *dst = skb_dst(skb);
1815 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1816 struct xfrm_policy_queue *pq = &xdst->pols[0]->polq;
1818 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1826 spin_lock_bh(&pq->hold_queue.lock);
1829 pq->timeout = XFRM_QUEUE_TMO_MIN;
1831 sched_next = jiffies + pq->timeout;
1833 if (del_timer(&pq->hold_timer)) {
1834 if (time_before(pq->hold_timer.expires, sched_next))
1835 sched_next = pq->hold_timer.expires;
1838 __skb_queue_tail(&pq->hold_queue, skb);
1839 mod_timer(&pq->hold_timer, sched_next);
1841 spin_unlock_bh(&pq->hold_queue.lock);
1846 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1847 struct dst_entry *dst,
1848 const struct flowi *fl,
1853 struct net_device *dev;
1854 struct dst_entry *dst1;
1855 struct xfrm_dst *xdst;
1857 xdst = xfrm_alloc_dst(net, family);
1861 if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0 ||
1862 (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP))
1865 dst1 = &xdst->u.dst;
1869 dst_copy_metrics(dst1, dst);
1871 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1872 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1873 dst1->lastuse = jiffies;
1875 dst1->input = dst_discard;
1876 dst1->output = xdst_queue_output;
1882 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1889 err = xfrm_fill_dst(xdst, dev, fl);
1898 xdst = ERR_PTR(err);
1902 static struct flow_cache_object *
1903 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1904 struct flow_cache_object *oldflo, void *ctx)
1906 struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1907 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1908 struct xfrm_dst *xdst, *new_xdst;
1909 int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1911 /* Check if the policies from old bundle are usable */
1914 xdst = container_of(oldflo, struct xfrm_dst, flo);
1915 num_pols = xdst->num_pols;
1916 num_xfrms = xdst->num_xfrms;
1918 for (i = 0; i < num_pols; i++) {
1919 pols[i] = xdst->pols[i];
1920 pol_dead |= pols[i]->walk.dead;
1923 dst_free(&xdst->u.dst);
1931 /* Resolve policies to use if we couldn't get them from
1932 * previous cache entry */
1935 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1936 err = xfrm_expand_policies(fl, family, pols,
1937 &num_pols, &num_xfrms);
1943 goto make_dummy_bundle;
1946 new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1947 if (IS_ERR(new_xdst)) {
1948 err = PTR_ERR(new_xdst);
1952 goto make_dummy_bundle;
1953 dst_hold(&xdst->u.dst);
1955 } else if (new_xdst == NULL) {
1958 goto make_dummy_bundle;
1959 xdst->num_xfrms = 0;
1960 dst_hold(&xdst->u.dst);
1964 /* Kill the previous bundle */
1966 /* The policies were stolen for newly generated bundle */
1968 dst_free(&xdst->u.dst);
1971 /* Flow cache does not have reference, it dst_free()'s,
1972 * but we do need to return one reference for original caller */
1973 dst_hold(&new_xdst->u.dst);
1974 return &new_xdst->flo;
1977 /* We found policies, but there's no bundles to instantiate:
1978 * either because the policy blocks, has no transformations or
1979 * we could not build template (no xfrm_states).*/
1980 xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
1982 xfrm_pols_put(pols, num_pols);
1983 return ERR_CAST(xdst);
1985 xdst->num_pols = num_pols;
1986 xdst->num_xfrms = num_xfrms;
1987 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1989 dst_hold(&xdst->u.dst);
1993 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1996 dst_free(&xdst->u.dst);
1998 xfrm_pols_put(pols, num_pols);
1999 return ERR_PTR(err);
2002 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2003 struct dst_entry *dst_orig)
2005 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2006 struct dst_entry *ret;
2009 dst_release(dst_orig);
2010 return ERR_PTR(-EINVAL);
2012 ret = afinfo->blackhole_route(net, dst_orig);
2014 xfrm_policy_put_afinfo(afinfo);
2019 /* Main function: finds/creates a bundle for given flow.
2021 * At the moment we eat a raw IP route. Mostly to speed up lookups
2022 * on interfaces with disabled IPsec.
2024 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2025 const struct flowi *fl,
2026 struct sock *sk, int flags)
2028 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2029 struct flow_cache_object *flo;
2030 struct xfrm_dst *xdst;
2031 struct dst_entry *dst, *route;
2032 u16 family = dst_orig->ops->family;
2033 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2034 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2041 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2043 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2044 err = xfrm_expand_policies(fl, family, pols,
2045 &num_pols, &num_xfrms);
2050 if (num_xfrms <= 0) {
2051 drop_pols = num_pols;
2055 xdst = xfrm_resolve_and_create_bundle(
2059 xfrm_pols_put(pols, num_pols);
2060 err = PTR_ERR(xdst);
2062 } else if (xdst == NULL) {
2064 drop_pols = num_pols;
2068 dst_hold(&xdst->u.dst);
2070 spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2071 xdst->u.dst.next = xfrm_policy_sk_bundles;
2072 xfrm_policy_sk_bundles = &xdst->u.dst;
2073 spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2075 route = xdst->route;
2080 /* To accelerate a bit... */
2081 if ((dst_orig->flags & DST_NOXFRM) ||
2082 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2085 flo = flow_cache_lookup(net, fl, family, dir,
2086 xfrm_bundle_lookup, dst_orig);
2093 xdst = container_of(flo, struct xfrm_dst, flo);
2095 num_pols = xdst->num_pols;
2096 num_xfrms = xdst->num_xfrms;
2097 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
2098 route = xdst->route;
2102 if (route == NULL && num_xfrms > 0) {
2103 /* The only case when xfrm_bundle_lookup() returns a
2104 * bundle with null route, is when the template could
2105 * not be resolved. It means policies are there, but
2106 * bundle could not be created, since we don't yet
2107 * have the xfrm_state's. We need to wait for KM to
2108 * negotiate new SA's or bail out with error.*/
2109 if (net->xfrm.sysctl_larval_drop) {
2110 /* EREMOTE tells the caller to generate
2111 * a one-shot blackhole route. */
2113 xfrm_pols_put(pols, drop_pols);
2114 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2116 return make_blackhole(net, family, dst_orig);
2118 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
2119 DECLARE_WAITQUEUE(wait, current);
2121 add_wait_queue(&net->xfrm.km_waitq, &wait);
2122 set_current_state(TASK_INTERRUPTIBLE);
2124 set_current_state(TASK_RUNNING);
2125 remove_wait_queue(&net->xfrm.km_waitq, &wait);
2127 if (!signal_pending(current)) {
2136 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2144 if ((flags & XFRM_LOOKUP_ICMP) &&
2145 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2150 for (i = 0; i < num_pols; i++)
2151 pols[i]->curlft.use_time = get_seconds();
2153 if (num_xfrms < 0) {
2154 /* Prohibit the flow */
2155 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2158 } else if (num_xfrms > 0) {
2159 /* Flow transformed */
2160 dst_release(dst_orig);
2162 /* Flow passes untransformed */
2167 xfrm_pols_put(pols, drop_pols);
2168 if (dst && dst->xfrm &&
2169 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2170 dst->flags |= DST_XFRM_TUNNEL;
2174 if (!(flags & XFRM_LOOKUP_ICMP)) {
2182 dst_release(dst_orig);
2183 xfrm_pols_put(pols, drop_pols);
2184 return ERR_PTR(err);
2186 EXPORT_SYMBOL(xfrm_lookup);
2189 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2191 struct xfrm_state *x;
2193 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2195 x = skb->sp->xvec[idx];
2196 if (!x->type->reject)
2198 return x->type->reject(x, skb, fl);
2201 /* When skb is transformed back to its "native" form, we have to
2202 * check policy restrictions. At the moment we make this in maximally
2203 * stupid way. Shame on me. :-) Of course, connected sockets must
2204 * have policy cached at them.
2208 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2209 unsigned short family)
2211 if (xfrm_state_kern(x))
2212 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2213 return x->id.proto == tmpl->id.proto &&
2214 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2215 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2216 x->props.mode == tmpl->mode &&
2217 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2218 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2219 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2220 xfrm_state_addr_cmp(tmpl, x, family));
2224 * 0 or more than 0 is returned when validation is succeeded (either bypass
2225 * because of optional transport mode, or next index of the mathced secpath
2226 * state with the template.
2227 * -1 is returned when no matching template is found.
2228 * Otherwise "-2 - errored_index" is returned.
2231 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2232 unsigned short family)
2236 if (tmpl->optional) {
2237 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2241 for (; idx < sp->len; idx++) {
2242 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2244 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2253 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2254 unsigned int family, int reverse)
2256 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2259 if (unlikely(afinfo == NULL))
2260 return -EAFNOSUPPORT;
2262 afinfo->decode_session(skb, fl, reverse);
2263 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2264 xfrm_policy_put_afinfo(afinfo);
2267 EXPORT_SYMBOL(__xfrm_decode_session);
2269 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2271 for (; k < sp->len; k++) {
2272 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2281 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2282 unsigned short family)
2284 struct net *net = dev_net(skb->dev);
2285 struct xfrm_policy *pol;
2286 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2295 reverse = dir & ~XFRM_POLICY_MASK;
2296 dir &= XFRM_POLICY_MASK;
2297 fl_dir = policy_to_flow_dir(dir);
2299 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2300 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2304 nf_nat_decode_session(skb, &fl, family);
2306 /* First, check used SA against their selectors. */
2310 for (i=skb->sp->len-1; i>=0; i--) {
2311 struct xfrm_state *x = skb->sp->xvec[i];
2312 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2313 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2320 if (sk && sk->sk_policy[dir]) {
2321 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2323 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2329 struct flow_cache_object *flo;
2331 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2332 xfrm_policy_lookup, NULL);
2333 if (IS_ERR_OR_NULL(flo))
2334 pol = ERR_CAST(flo);
2336 pol = container_of(flo, struct xfrm_policy, flo);
2340 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2345 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2346 xfrm_secpath_reject(xerr_idx, skb, &fl);
2347 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2353 pol->curlft.use_time = get_seconds();
2357 #ifdef CONFIG_XFRM_SUB_POLICY
2358 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2359 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2363 if (IS_ERR(pols[1])) {
2364 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2367 pols[1]->curlft.use_time = get_seconds();
2373 if (pol->action == XFRM_POLICY_ALLOW) {
2374 struct sec_path *sp;
2375 static struct sec_path dummy;
2376 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2377 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2378 struct xfrm_tmpl **tpp = tp;
2382 if ((sp = skb->sp) == NULL)
2385 for (pi = 0; pi < npols; pi++) {
2386 if (pols[pi] != pol &&
2387 pols[pi]->action != XFRM_POLICY_ALLOW) {
2388 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2391 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2392 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2395 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2396 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2400 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2404 /* For each tunnel xfrm, find the first matching tmpl.
2405 * For each tmpl before that, find corresponding xfrm.
2406 * Order is _important_. Later we will implement
2407 * some barriers, but at the moment barriers
2408 * are implied between each two transformations.
2410 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2411 k = xfrm_policy_ok(tpp[i], sp, k, family);
2414 /* "-2 - errored_index" returned */
2416 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2421 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2422 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2426 xfrm_pols_put(pols, npols);
2429 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2432 xfrm_secpath_reject(xerr_idx, skb, &fl);
2434 xfrm_pols_put(pols, npols);
2437 EXPORT_SYMBOL(__xfrm_policy_check);
2439 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2441 struct net *net = dev_net(skb->dev);
2443 struct dst_entry *dst;
2446 if (xfrm_decode_session(skb, &fl, family) < 0) {
2447 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2453 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2458 skb_dst_set(skb, dst);
2461 EXPORT_SYMBOL(__xfrm_route_forward);
2463 /* Optimize later using cookies and generation ids. */
2465 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2467 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2468 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2469 * get validated by dst_ops->check on every use. We do this
2470 * because when a normal route referenced by an XFRM dst is
2471 * obsoleted we do not go looking around for all parent
2472 * referencing XFRM dsts so that we can invalidate them. It
2473 * is just too much work. Instead we make the checks here on
2474 * every use. For example:
2476 * XFRM dst A --> IPv4 dst X
2478 * X is the "xdst->route" of A (X is also the "dst->path" of A
2479 * in this example). If X is marked obsolete, "A" will not
2480 * notice. That's what we are validating here via the
2481 * stale_bundle() check.
2483 * When a policy's bundle is pruned, we dst_free() the XFRM
2484 * dst which causes it's ->obsolete field to be set to
2485 * DST_OBSOLETE_DEAD. If an XFRM dst has been pruned like
2486 * this, we want to force a new route lookup.
2488 if (dst->obsolete < 0 && !stale_bundle(dst))
2494 static int stale_bundle(struct dst_entry *dst)
2496 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2499 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2501 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2502 dst->dev = dev_net(dev)->loopback_dev;
2507 EXPORT_SYMBOL(xfrm_dst_ifdown);
2509 static void xfrm_link_failure(struct sk_buff *skb)
2511 /* Impossible. Such dst must be popped before reaches point of failure. */
2514 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2517 if (dst->obsolete) {
2525 static void __xfrm_garbage_collect(struct net *net)
2527 struct dst_entry *head, *next;
2529 spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2530 head = xfrm_policy_sk_bundles;
2531 xfrm_policy_sk_bundles = NULL;
2532 spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2541 static void xfrm_garbage_collect(struct net *net)
2544 __xfrm_garbage_collect(net);
2547 static void xfrm_garbage_collect_deferred(struct net *net)
2549 flow_cache_flush_deferred();
2550 __xfrm_garbage_collect(net);
2553 static void xfrm_init_pmtu(struct dst_entry *dst)
2556 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2557 u32 pmtu, route_mtu_cached;
2559 pmtu = dst_mtu(dst->child);
2560 xdst->child_mtu_cached = pmtu;
2562 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2564 route_mtu_cached = dst_mtu(xdst->route);
2565 xdst->route_mtu_cached = route_mtu_cached;
2567 if (pmtu > route_mtu_cached)
2568 pmtu = route_mtu_cached;
2570 dst_metric_set(dst, RTAX_MTU, pmtu);
2571 } while ((dst = dst->next));
2574 /* Check that the bundle accepts the flow and its components are
2578 static int xfrm_bundle_ok(struct xfrm_dst *first)
2580 struct dst_entry *dst = &first->u.dst;
2581 struct xfrm_dst *last;
2584 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2585 (dst->dev && !netif_running(dst->dev)))
2588 if (dst->flags & DST_XFRM_QUEUE)
2594 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2596 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2598 if (xdst->xfrm_genid != dst->xfrm->genid)
2600 if (xdst->num_pols > 0 &&
2601 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2604 mtu = dst_mtu(dst->child);
2605 if (xdst->child_mtu_cached != mtu) {
2607 xdst->child_mtu_cached = mtu;
2610 if (!dst_check(xdst->route, xdst->route_cookie))
2612 mtu = dst_mtu(xdst->route);
2613 if (xdst->route_mtu_cached != mtu) {
2615 xdst->route_mtu_cached = mtu;
2619 } while (dst->xfrm);
2624 mtu = last->child_mtu_cached;
2628 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2629 if (mtu > last->route_mtu_cached)
2630 mtu = last->route_mtu_cached;
2631 dst_metric_set(dst, RTAX_MTU, mtu);
2636 last = (struct xfrm_dst *)last->u.dst.next;
2637 last->child_mtu_cached = mtu;
2643 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2645 return dst_metric_advmss(dst->path);
2648 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2650 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2652 return mtu ? : dst_mtu(dst->path);
2655 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2656 struct sk_buff *skb,
2659 return dst->path->ops->neigh_lookup(dst, skb, daddr);
2662 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2666 if (unlikely(afinfo == NULL))
2668 if (unlikely(afinfo->family >= NPROTO))
2669 return -EAFNOSUPPORT;
2670 spin_lock(&xfrm_policy_afinfo_lock);
2671 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2674 struct dst_ops *dst_ops = afinfo->dst_ops;
2675 if (likely(dst_ops->kmem_cachep == NULL))
2676 dst_ops->kmem_cachep = xfrm_dst_cache;
2677 if (likely(dst_ops->check == NULL))
2678 dst_ops->check = xfrm_dst_check;
2679 if (likely(dst_ops->default_advmss == NULL))
2680 dst_ops->default_advmss = xfrm_default_advmss;
2681 if (likely(dst_ops->mtu == NULL))
2682 dst_ops->mtu = xfrm_mtu;
2683 if (likely(dst_ops->negative_advice == NULL))
2684 dst_ops->negative_advice = xfrm_negative_advice;
2685 if (likely(dst_ops->link_failure == NULL))
2686 dst_ops->link_failure = xfrm_link_failure;
2687 if (likely(dst_ops->neigh_lookup == NULL))
2688 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2689 if (likely(afinfo->garbage_collect == NULL))
2690 afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2691 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2693 spin_unlock(&xfrm_policy_afinfo_lock);
2697 struct dst_ops *xfrm_dst_ops;
2699 switch (afinfo->family) {
2701 xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2703 #if IS_ENABLED(CONFIG_IPV6)
2705 xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2711 *xfrm_dst_ops = *afinfo->dst_ops;
2717 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2719 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2722 if (unlikely(afinfo == NULL))
2724 if (unlikely(afinfo->family >= NPROTO))
2725 return -EAFNOSUPPORT;
2726 spin_lock(&xfrm_policy_afinfo_lock);
2727 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2728 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2731 RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2734 spin_unlock(&xfrm_policy_afinfo_lock);
2736 struct dst_ops *dst_ops = afinfo->dst_ops;
2740 dst_ops->kmem_cachep = NULL;
2741 dst_ops->check = NULL;
2742 dst_ops->negative_advice = NULL;
2743 dst_ops->link_failure = NULL;
2744 afinfo->garbage_collect = NULL;
2748 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2750 static void __net_init xfrm_dst_ops_init(struct net *net)
2752 struct xfrm_policy_afinfo *afinfo;
2755 afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2757 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2758 #if IS_ENABLED(CONFIG_IPV6)
2759 afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2761 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2766 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2768 struct net_device *dev = ptr;
2772 xfrm_garbage_collect(dev_net(dev));
2777 static struct notifier_block xfrm_dev_notifier = {
2778 .notifier_call = xfrm_dev_event,
2781 #ifdef CONFIG_XFRM_STATISTICS
2782 static int __net_init xfrm_statistics_init(struct net *net)
2786 if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2787 sizeof(struct linux_xfrm_mib),
2788 __alignof__(struct linux_xfrm_mib)) < 0)
2790 rv = xfrm_proc_init(net);
2792 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2796 static void xfrm_statistics_fini(struct net *net)
2798 xfrm_proc_fini(net);
2799 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2802 static int __net_init xfrm_statistics_init(struct net *net)
2807 static void xfrm_statistics_fini(struct net *net)
2812 static int __net_init xfrm_policy_init(struct net *net)
2814 unsigned int hmask, sz;
2817 if (net_eq(net, &init_net))
2818 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2819 sizeof(struct xfrm_dst),
2820 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2824 sz = (hmask+1) * sizeof(struct hlist_head);
2826 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2827 if (!net->xfrm.policy_byidx)
2829 net->xfrm.policy_idx_hmask = hmask;
2831 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2832 struct xfrm_policy_hash *htab;
2834 net->xfrm.policy_count[dir] = 0;
2835 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2837 htab = &net->xfrm.policy_bydst[dir];
2838 htab->table = xfrm_hash_alloc(sz);
2841 htab->hmask = hmask;
2844 INIT_LIST_HEAD(&net->xfrm.policy_all);
2845 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2846 if (net_eq(net, &init_net))
2847 register_netdevice_notifier(&xfrm_dev_notifier);
2851 for (dir--; dir >= 0; dir--) {
2852 struct xfrm_policy_hash *htab;
2854 htab = &net->xfrm.policy_bydst[dir];
2855 xfrm_hash_free(htab->table, sz);
2857 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2862 static void xfrm_policy_fini(struct net *net)
2864 struct xfrm_audit audit_info;
2868 flush_work(&net->xfrm.policy_hash_work);
2869 #ifdef CONFIG_XFRM_SUB_POLICY
2870 audit_info.loginuid = INVALID_UID;
2871 audit_info.sessionid = -1;
2872 audit_info.secid = 0;
2873 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2875 audit_info.loginuid = INVALID_UID;
2876 audit_info.sessionid = -1;
2877 audit_info.secid = 0;
2878 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2880 WARN_ON(!list_empty(&net->xfrm.policy_all));
2882 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2883 struct xfrm_policy_hash *htab;
2885 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2887 htab = &net->xfrm.policy_bydst[dir];
2888 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2889 WARN_ON(!hlist_empty(htab->table));
2890 xfrm_hash_free(htab->table, sz);
2893 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2894 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2895 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2898 static int __net_init xfrm_net_init(struct net *net)
2902 rv = xfrm_statistics_init(net);
2904 goto out_statistics;
2905 rv = xfrm_state_init(net);
2908 rv = xfrm_policy_init(net);
2911 xfrm_dst_ops_init(net);
2912 rv = xfrm_sysctl_init(net);
2918 xfrm_policy_fini(net);
2920 xfrm_state_fini(net);
2922 xfrm_statistics_fini(net);
2927 static void __net_exit xfrm_net_exit(struct net *net)
2929 xfrm_sysctl_fini(net);
2930 xfrm_policy_fini(net);
2931 xfrm_state_fini(net);
2932 xfrm_statistics_fini(net);
2935 static struct pernet_operations __net_initdata xfrm_net_ops = {
2936 .init = xfrm_net_init,
2937 .exit = xfrm_net_exit,
2940 void __init xfrm_init(void)
2942 register_pernet_subsys(&xfrm_net_ops);
2946 #ifdef CONFIG_AUDITSYSCALL
2947 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2948 struct audit_buffer *audit_buf)
2950 struct xfrm_sec_ctx *ctx = xp->security;
2951 struct xfrm_selector *sel = &xp->selector;
2954 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2955 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2957 switch(sel->family) {
2959 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2960 if (sel->prefixlen_s != 32)
2961 audit_log_format(audit_buf, " src_prefixlen=%d",
2963 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2964 if (sel->prefixlen_d != 32)
2965 audit_log_format(audit_buf, " dst_prefixlen=%d",
2969 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2970 if (sel->prefixlen_s != 128)
2971 audit_log_format(audit_buf, " src_prefixlen=%d",
2973 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2974 if (sel->prefixlen_d != 128)
2975 audit_log_format(audit_buf, " dst_prefixlen=%d",
2981 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2982 kuid_t auid, u32 sessionid, u32 secid)
2984 struct audit_buffer *audit_buf;
2986 audit_buf = xfrm_audit_start("SPD-add");
2987 if (audit_buf == NULL)
2989 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2990 audit_log_format(audit_buf, " res=%u", result);
2991 xfrm_audit_common_policyinfo(xp, audit_buf);
2992 audit_log_end(audit_buf);
2994 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2996 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2997 kuid_t auid, u32 sessionid, u32 secid)
2999 struct audit_buffer *audit_buf;
3001 audit_buf = xfrm_audit_start("SPD-delete");
3002 if (audit_buf == NULL)
3004 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3005 audit_log_format(audit_buf, " res=%u", result);
3006 xfrm_audit_common_policyinfo(xp, audit_buf);
3007 audit_log_end(audit_buf);
3009 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3012 #ifdef CONFIG_XFRM_MIGRATE
3013 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3014 const struct xfrm_selector *sel_tgt)
3016 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3017 if (sel_tgt->family == sel_cmp->family &&
3018 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3020 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3022 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3023 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3027 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3034 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3037 struct xfrm_policy *pol, *ret = NULL;
3038 struct hlist_head *chain;
3041 read_lock_bh(&xfrm_policy_lock);
3042 chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
3043 hlist_for_each_entry(pol, chain, bydst) {
3044 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3045 pol->type == type) {
3047 priority = ret->priority;
3051 chain = &init_net.xfrm.policy_inexact[dir];
3052 hlist_for_each_entry(pol, chain, bydst) {
3053 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3054 pol->type == type &&
3055 pol->priority < priority) {
3064 read_unlock_bh(&xfrm_policy_lock);
3069 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3073 if (t->mode == m->mode && t->id.proto == m->proto &&
3074 (m->reqid == 0 || t->reqid == m->reqid)) {
3076 case XFRM_MODE_TUNNEL:
3077 case XFRM_MODE_BEET:
3078 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3080 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3085 case XFRM_MODE_TRANSPORT:
3086 /* in case of transport mode, template does not store
3087 any IP addresses, hence we just compare mode and
3098 /* update endpoint address(es) of template(s) */
3099 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3100 struct xfrm_migrate *m, int num_migrate)
3102 struct xfrm_migrate *mp;
3105 write_lock_bh(&pol->lock);
3106 if (unlikely(pol->walk.dead)) {
3107 /* target policy has been deleted */
3108 write_unlock_bh(&pol->lock);
3112 for (i = 0; i < pol->xfrm_nr; i++) {
3113 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3114 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3117 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3118 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3120 /* update endpoints */
3121 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3122 sizeof(pol->xfrm_vec[i].id.daddr));
3123 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3124 sizeof(pol->xfrm_vec[i].saddr));
3125 pol->xfrm_vec[i].encap_family = mp->new_family;
3127 atomic_inc(&pol->genid);
3131 write_unlock_bh(&pol->lock);
3139 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3143 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3146 for (i = 0; i < num_migrate; i++) {
3147 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3149 xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3152 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3153 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3156 /* check if there is any duplicated entry */
3157 for (j = i + 1; j < num_migrate; j++) {
3158 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3159 sizeof(m[i].old_daddr)) &&
3160 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3161 sizeof(m[i].old_saddr)) &&
3162 m[i].proto == m[j].proto &&
3163 m[i].mode == m[j].mode &&
3164 m[i].reqid == m[j].reqid &&
3165 m[i].old_family == m[j].old_family)
3173 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3174 struct xfrm_migrate *m, int num_migrate,
3175 struct xfrm_kmaddress *k)
3177 int i, err, nx_cur = 0, nx_new = 0;
3178 struct xfrm_policy *pol = NULL;
3179 struct xfrm_state *x, *xc;
3180 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3181 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3182 struct xfrm_migrate *mp;
3184 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3187 /* Stage 1 - find policy */
3188 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
3193 /* Stage 2 - find and update state(s) */
3194 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3195 if ((x = xfrm_migrate_state_find(mp))) {
3198 if ((xc = xfrm_state_migrate(x, mp))) {
3208 /* Stage 3 - update policy */
3209 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3212 /* Stage 4 - delete old state(s) */
3214 xfrm_states_put(x_cur, nx_cur);
3215 xfrm_states_delete(x_cur, nx_cur);
3218 /* Stage 5 - announce */
3219 km_migrate(sel, dir, type, m, num_migrate, k);
3231 xfrm_states_put(x_cur, nx_cur);
3233 xfrm_states_delete(x_new, nx_new);
3237 EXPORT_SYMBOL(xfrm_migrate);