1 // SPDX-License-Identifier: GPL-2.0-only
7 * Kazunori MIYAZAWA @USAGI
8 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
10 * Kazunori MIYAZAWA @USAGI
12 * Split up af-specific portion
13 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
17 #include <linux/err.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/workqueue.h>
23 #include <linux/notifier.h>
24 #include <linux/netdevice.h>
25 #include <linux/netfilter.h>
26 #include <linux/module.h>
27 #include <linux/cache.h>
28 #include <linux/cpu.h>
29 #include <linux/audit.h>
30 #include <linux/rhashtable.h>
31 #include <linux/if_tunnel.h>
36 #if IS_ENABLED(CONFIG_IPV6_MIP6)
39 #ifdef CONFIG_XFRM_STATISTICS
42 #ifdef CONFIG_XFRM_ESPINTCP
43 #include <net/espintcp.h>
46 #include "xfrm_hash.h"
48 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
49 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
50 #define XFRM_MAX_QUEUE_LEN 100
53 struct dst_entry *dst_orig;
57 /* prefixes smaller than this are stored in lists, not trees. */
58 #define INEXACT_PREFIXLEN_IPV4 16
59 #define INEXACT_PREFIXLEN_IPV6 48
61 struct xfrm_pol_inexact_node {
71 /* the policies matching this node, can be empty list */
72 struct hlist_head hhead;
75 /* xfrm inexact policy search tree:
76 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
78 * +---- root_d: sorted by daddr:prefix
80 * | xfrm_pol_inexact_node
82 * | +- root: sorted by saddr/prefix
84 * | | xfrm_pol_inexact_node
88 * | | + hhead: saddr:daddr policies
90 * | +- coarse policies and all any:daddr policies
92 * +---- root_s: sorted by saddr:prefix
94 * | xfrm_pol_inexact_node
98 * | + hhead: saddr:any policies
100 * +---- coarse policies and all any:any policies
102 * Lookups return four candidate lists:
103 * 1. any:any list from top-level xfrm_pol_inexact_bin
104 * 2. any:daddr list from daddr tree
105 * 3. saddr:daddr list from 2nd level daddr tree
106 * 4. saddr:any list from saddr tree
108 * This result set then needs to be searched for the policy with
109 * the lowest priority. If two results have same prio, youngest one wins.
112 struct xfrm_pol_inexact_key {
119 struct xfrm_pol_inexact_bin {
120 struct xfrm_pol_inexact_key k;
121 struct rhash_head head;
122 /* list containing '*:*' policies */
123 struct hlist_head hhead;
125 seqcount_spinlock_t count;
126 /* tree sorted by daddr/prefix */
127 struct rb_root root_d;
129 /* tree sorted by saddr/prefix */
130 struct rb_root root_s;
132 /* slow path below */
133 struct list_head inexact_bins;
137 enum xfrm_pol_inexact_candidate_type {
146 struct xfrm_pol_inexact_candidates {
147 struct hlist_head *res[XFRM_POL_CAND_MAX];
150 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
151 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
153 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
154 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
157 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
158 static __read_mostly seqcount_mutex_t xfrm_policy_hash_generation;
160 static struct rhashtable xfrm_policy_inexact_table;
161 static const struct rhashtable_params xfrm_pol_inexact_params;
163 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
164 static int stale_bundle(struct dst_entry *dst);
165 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
166 static void xfrm_policy_queue_process(struct timer_list *t);
168 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
169 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
172 static struct xfrm_pol_inexact_bin *
173 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
176 static struct xfrm_pol_inexact_bin *
177 xfrm_policy_inexact_lookup_rcu(struct net *net,
178 u8 type, u16 family, u8 dir, u32 if_id);
179 static struct xfrm_policy *
180 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
182 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
183 struct xfrm_policy *policy);
186 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
187 struct xfrm_pol_inexact_bin *b,
188 const xfrm_address_t *saddr,
189 const xfrm_address_t *daddr);
191 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
193 return refcount_inc_not_zero(&policy->refcnt);
197 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
199 const struct flowi4 *fl4 = &fl->u.ip4;
201 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
202 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
203 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
204 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
205 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
206 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
210 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
212 const struct flowi6 *fl6 = &fl->u.ip6;
214 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
215 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
216 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
217 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
218 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
219 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
222 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
223 unsigned short family)
227 return __xfrm4_selector_match(sel, fl);
229 return __xfrm6_selector_match(sel, fl);
234 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
236 const struct xfrm_policy_afinfo *afinfo;
238 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
241 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
242 if (unlikely(!afinfo))
247 /* Called with rcu_read_lock(). */
248 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
250 return rcu_dereference(xfrm_if_cb);
253 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
254 const xfrm_address_t *saddr,
255 const xfrm_address_t *daddr,
256 int family, u32 mark)
258 const struct xfrm_policy_afinfo *afinfo;
259 struct dst_entry *dst;
261 afinfo = xfrm_policy_get_afinfo(family);
262 if (unlikely(afinfo == NULL))
263 return ERR_PTR(-EAFNOSUPPORT);
265 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
271 EXPORT_SYMBOL(__xfrm_dst_lookup);
273 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
275 xfrm_address_t *prev_saddr,
276 xfrm_address_t *prev_daddr,
277 int family, u32 mark)
279 struct net *net = xs_net(x);
280 xfrm_address_t *saddr = &x->props.saddr;
281 xfrm_address_t *daddr = &x->id.daddr;
282 struct dst_entry *dst;
284 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
288 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
293 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
296 if (prev_saddr != saddr)
297 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
298 if (prev_daddr != daddr)
299 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
305 static inline unsigned long make_jiffies(long secs)
307 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
308 return MAX_SCHEDULE_TIMEOUT-1;
313 static void xfrm_policy_timer(struct timer_list *t)
315 struct xfrm_policy *xp = from_timer(xp, t, timer);
316 time64_t now = ktime_get_real_seconds();
317 time64_t next = TIME64_MAX;
321 read_lock(&xp->lock);
323 if (unlikely(xp->walk.dead))
326 dir = xfrm_policy_id2dir(xp->index);
328 if (xp->lft.hard_add_expires_seconds) {
329 time64_t tmo = xp->lft.hard_add_expires_seconds +
330 xp->curlft.add_time - now;
336 if (xp->lft.hard_use_expires_seconds) {
337 time64_t tmo = xp->lft.hard_use_expires_seconds +
338 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
344 if (xp->lft.soft_add_expires_seconds) {
345 time64_t tmo = xp->lft.soft_add_expires_seconds +
346 xp->curlft.add_time - now;
349 tmo = XFRM_KM_TIMEOUT;
354 if (xp->lft.soft_use_expires_seconds) {
355 time64_t tmo = xp->lft.soft_use_expires_seconds +
356 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
359 tmo = XFRM_KM_TIMEOUT;
366 km_policy_expired(xp, dir, 0, 0);
367 if (next != TIME64_MAX &&
368 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
372 read_unlock(&xp->lock);
377 read_unlock(&xp->lock);
378 if (!xfrm_policy_delete(xp, dir))
379 km_policy_expired(xp, dir, 1, 0);
383 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
387 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
389 struct xfrm_policy *policy;
391 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
394 write_pnet(&policy->xp_net, net);
395 INIT_LIST_HEAD(&policy->walk.all);
396 INIT_HLIST_NODE(&policy->bydst_inexact_list);
397 INIT_HLIST_NODE(&policy->bydst);
398 INIT_HLIST_NODE(&policy->byidx);
399 rwlock_init(&policy->lock);
400 refcount_set(&policy->refcnt, 1);
401 skb_queue_head_init(&policy->polq.hold_queue);
402 timer_setup(&policy->timer, xfrm_policy_timer, 0);
403 timer_setup(&policy->polq.hold_timer,
404 xfrm_policy_queue_process, 0);
408 EXPORT_SYMBOL(xfrm_policy_alloc);
410 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
412 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
414 security_xfrm_policy_free(policy->security);
418 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
420 void xfrm_policy_destroy(struct xfrm_policy *policy)
422 BUG_ON(!policy->walk.dead);
424 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
427 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
429 EXPORT_SYMBOL(xfrm_policy_destroy);
431 /* Rule must be locked. Release descendant resources, announce
432 * entry dead. The rule must be unlinked from lists to the moment.
435 static void xfrm_policy_kill(struct xfrm_policy *policy)
437 write_lock_bh(&policy->lock);
438 policy->walk.dead = 1;
439 write_unlock_bh(&policy->lock);
441 atomic_inc(&policy->genid);
443 if (del_timer(&policy->polq.hold_timer))
444 xfrm_pol_put(policy);
445 skb_queue_purge(&policy->polq.hold_queue);
447 if (del_timer(&policy->timer))
448 xfrm_pol_put(policy);
450 xfrm_pol_put(policy);
453 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
455 static inline unsigned int idx_hash(struct net *net, u32 index)
457 return __idx_hash(index, net->xfrm.policy_idx_hmask);
460 /* calculate policy hash thresholds */
461 static void __get_hash_thresh(struct net *net,
462 unsigned short family, int dir,
463 u8 *dbits, u8 *sbits)
467 *dbits = net->xfrm.policy_bydst[dir].dbits4;
468 *sbits = net->xfrm.policy_bydst[dir].sbits4;
472 *dbits = net->xfrm.policy_bydst[dir].dbits6;
473 *sbits = net->xfrm.policy_bydst[dir].sbits6;
482 static struct hlist_head *policy_hash_bysel(struct net *net,
483 const struct xfrm_selector *sel,
484 unsigned short family, int dir)
486 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
491 __get_hash_thresh(net, family, dir, &dbits, &sbits);
492 hash = __sel_hash(sel, family, hmask, dbits, sbits);
494 if (hash == hmask + 1)
497 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
498 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
501 static struct hlist_head *policy_hash_direct(struct net *net,
502 const xfrm_address_t *daddr,
503 const xfrm_address_t *saddr,
504 unsigned short family, int dir)
506 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
511 __get_hash_thresh(net, family, dir, &dbits, &sbits);
512 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
514 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
515 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
518 static void xfrm_dst_hash_transfer(struct net *net,
519 struct hlist_head *list,
520 struct hlist_head *ndsttable,
521 unsigned int nhashmask,
524 struct hlist_node *tmp, *entry0 = NULL;
525 struct xfrm_policy *pol;
531 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
534 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
535 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
536 pol->family, nhashmask, dbits, sbits);
538 hlist_del_rcu(&pol->bydst);
539 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
544 hlist_del_rcu(&pol->bydst);
545 hlist_add_behind_rcu(&pol->bydst, entry0);
547 entry0 = &pol->bydst;
549 if (!hlist_empty(list)) {
555 static void xfrm_idx_hash_transfer(struct hlist_head *list,
556 struct hlist_head *nidxtable,
557 unsigned int nhashmask)
559 struct hlist_node *tmp;
560 struct xfrm_policy *pol;
562 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
565 h = __idx_hash(pol->index, nhashmask);
566 hlist_add_head(&pol->byidx, nidxtable+h);
570 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
572 return ((old_hmask + 1) << 1) - 1;
575 static void xfrm_bydst_resize(struct net *net, int dir)
577 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
578 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
579 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
580 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
581 struct hlist_head *odst;
587 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
588 write_seqcount_begin(&xfrm_policy_hash_generation);
590 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
591 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
593 for (i = hmask; i >= 0; i--)
594 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
596 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
597 net->xfrm.policy_bydst[dir].hmask = nhashmask;
599 write_seqcount_end(&xfrm_policy_hash_generation);
600 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
604 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
607 static void xfrm_byidx_resize(struct net *net, int total)
609 unsigned int hmask = net->xfrm.policy_idx_hmask;
610 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
611 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
612 struct hlist_head *oidx = net->xfrm.policy_byidx;
613 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
619 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
621 for (i = hmask; i >= 0; i--)
622 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
624 net->xfrm.policy_byidx = nidx;
625 net->xfrm.policy_idx_hmask = nhashmask;
627 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
629 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
632 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
634 unsigned int cnt = net->xfrm.policy_count[dir];
635 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
640 if ((hmask + 1) < xfrm_policy_hashmax &&
647 static inline int xfrm_byidx_should_resize(struct net *net, int total)
649 unsigned int hmask = net->xfrm.policy_idx_hmask;
651 if ((hmask + 1) < xfrm_policy_hashmax &&
658 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
660 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
661 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
662 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
663 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
664 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
665 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
666 si->spdhcnt = net->xfrm.policy_idx_hmask;
667 si->spdhmcnt = xfrm_policy_hashmax;
669 EXPORT_SYMBOL(xfrm_spd_getinfo);
671 static DEFINE_MUTEX(hash_resize_mutex);
672 static void xfrm_hash_resize(struct work_struct *work)
674 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
677 mutex_lock(&hash_resize_mutex);
680 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
681 if (xfrm_bydst_should_resize(net, dir, &total))
682 xfrm_bydst_resize(net, dir);
684 if (xfrm_byidx_should_resize(net, total))
685 xfrm_byidx_resize(net, total);
687 mutex_unlock(&hash_resize_mutex);
690 /* Make sure *pol can be inserted into fastbin.
691 * Useful to check that later insert requests will be successful
692 * (provided xfrm_policy_lock is held throughout).
694 static struct xfrm_pol_inexact_bin *
695 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
697 struct xfrm_pol_inexact_bin *bin, *prev;
698 struct xfrm_pol_inexact_key k = {
699 .family = pol->family,
704 struct net *net = xp_net(pol);
706 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
708 write_pnet(&k.net, net);
709 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
710 xfrm_pol_inexact_params);
714 bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
719 INIT_HLIST_HEAD(&bin->hhead);
720 bin->root_d = RB_ROOT;
721 bin->root_s = RB_ROOT;
722 seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
724 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
726 xfrm_pol_inexact_params);
728 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
734 return IS_ERR(prev) ? NULL : prev;
737 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
738 int family, u8 prefixlen)
740 if (xfrm_addr_any(addr, family))
743 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
746 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
753 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
755 const xfrm_address_t *addr;
756 bool saddr_any, daddr_any;
759 addr = &policy->selector.saddr;
760 prefixlen = policy->selector.prefixlen_s;
762 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
765 addr = &policy->selector.daddr;
766 prefixlen = policy->selector.prefixlen_d;
767 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
770 return saddr_any && daddr_any;
773 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
774 const xfrm_address_t *addr, u8 prefixlen)
777 node->prefixlen = prefixlen;
780 static struct xfrm_pol_inexact_node *
781 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
783 struct xfrm_pol_inexact_node *node;
785 node = kzalloc(sizeof(*node), GFP_ATOMIC);
787 xfrm_pol_inexact_node_init(node, addr, prefixlen);
792 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
793 const xfrm_address_t *b,
794 u8 prefixlen, u16 family)
797 unsigned int pdw, pbi;
804 mask = ~0U << (32 - prefixlen);
805 ma = ntohl(a->a4) & mask;
806 mb = ntohl(b->a4) & mask;
813 pdw = prefixlen >> 5;
814 pbi = prefixlen & 0x1f;
817 delta = memcmp(a->a6, b->a6, pdw << 2);
822 mask = ~0U << (32 - pbi);
823 ma = ntohl(a->a6[pdw]) & mask;
824 mb = ntohl(b->a6[pdw]) & mask;
838 static void xfrm_policy_inexact_list_reinsert(struct net *net,
839 struct xfrm_pol_inexact_node *n,
842 unsigned int matched_s, matched_d;
843 struct xfrm_policy *policy, *p;
848 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
849 struct hlist_node *newpos = NULL;
850 bool matches_s, matches_d;
852 if (!policy->bydst_reinsert)
855 WARN_ON_ONCE(policy->family != family);
857 policy->bydst_reinsert = false;
858 hlist_for_each_entry(p, &n->hhead, bydst) {
859 if (policy->priority > p->priority)
861 else if (policy->priority == p->priority &&
862 policy->pos > p->pos)
869 hlist_add_behind_rcu(&policy->bydst, newpos);
871 hlist_add_head_rcu(&policy->bydst, &n->hhead);
873 /* paranoia checks follow.
874 * Check that the reinserted policy matches at least
875 * saddr or daddr for current node prefix.
877 * Matching both is fine, matching saddr in one policy
878 * (but not daddr) and then matching only daddr in another
881 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
885 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
889 if (matches_s && matches_d)
892 WARN_ON_ONCE(!matches_s && !matches_d);
897 WARN_ON_ONCE(matched_s && matched_d);
901 static void xfrm_policy_inexact_node_reinsert(struct net *net,
902 struct xfrm_pol_inexact_node *n,
906 struct xfrm_pol_inexact_node *node;
907 struct rb_node **p, *parent;
909 /* we should not have another subtree here */
910 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
919 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
921 prefixlen = min(node->prefixlen, n->prefixlen);
923 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
926 p = &parent->rb_left;
927 } else if (delta > 0) {
928 p = &parent->rb_right;
930 bool same_prefixlen = node->prefixlen == n->prefixlen;
931 struct xfrm_policy *tmp;
933 hlist_for_each_entry(tmp, &n->hhead, bydst) {
934 tmp->bydst_reinsert = true;
935 hlist_del_rcu(&tmp->bydst);
938 node->prefixlen = prefixlen;
940 xfrm_policy_inexact_list_reinsert(net, node, family);
942 if (same_prefixlen) {
954 rb_link_node_rcu(&n->node, parent, p);
955 rb_insert_color(&n->node, new);
958 /* merge nodes v and n */
959 static void xfrm_policy_inexact_node_merge(struct net *net,
960 struct xfrm_pol_inexact_node *v,
961 struct xfrm_pol_inexact_node *n,
964 struct xfrm_pol_inexact_node *node;
965 struct xfrm_policy *tmp;
966 struct rb_node *rnode;
968 /* To-be-merged node v has a subtree.
970 * Dismantle it and insert its nodes to n->root.
972 while ((rnode = rb_first(&v->root)) != NULL) {
973 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
974 rb_erase(&node->node, &v->root);
975 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
979 hlist_for_each_entry(tmp, &v->hhead, bydst) {
980 tmp->bydst_reinsert = true;
981 hlist_del_rcu(&tmp->bydst);
984 xfrm_policy_inexact_list_reinsert(net, n, family);
987 static struct xfrm_pol_inexact_node *
988 xfrm_policy_inexact_insert_node(struct net *net,
989 struct rb_root *root,
990 xfrm_address_t *addr,
991 u16 family, u8 prefixlen, u8 dir)
993 struct xfrm_pol_inexact_node *cached = NULL;
994 struct rb_node **p, *parent = NULL;
995 struct xfrm_pol_inexact_node *node;
1002 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1004 delta = xfrm_policy_addr_delta(addr, &node->addr,
1007 if (delta == 0 && prefixlen >= node->prefixlen) {
1008 WARN_ON_ONCE(cached); /* ipsec policies got lost */
1013 p = &parent->rb_left;
1015 p = &parent->rb_right;
1017 if (prefixlen < node->prefixlen) {
1018 delta = xfrm_policy_addr_delta(addr, &node->addr,
1024 /* This node is a subnet of the new prefix. It needs
1025 * to be removed and re-inserted with the smaller
1026 * prefix and all nodes that are now also covered
1027 * by the reduced prefixlen.
1029 rb_erase(&node->node, root);
1032 xfrm_pol_inexact_node_init(node, addr,
1036 /* This node also falls within the new
1037 * prefixlen. Merge the to-be-reinserted
1038 * node and this one.
1040 xfrm_policy_inexact_node_merge(net, node,
1042 kfree_rcu(node, rcu);
1053 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1058 rb_link_node_rcu(&node->node, parent, p);
1059 rb_insert_color(&node->node, root);
1064 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1066 struct xfrm_pol_inexact_node *node;
1067 struct rb_node *rn = rb_first(r);
1070 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1072 xfrm_policy_inexact_gc_tree(&node->root, rm);
1075 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1080 rb_erase(&node->node, r);
1081 kfree_rcu(node, rcu);
1085 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1087 write_seqcount_begin(&b->count);
1088 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1089 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1090 write_seqcount_end(&b->count);
1092 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1093 !hlist_empty(&b->hhead)) {
1094 WARN_ON_ONCE(net_exit);
1098 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1099 xfrm_pol_inexact_params) == 0) {
1100 list_del(&b->inexact_bins);
1105 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1107 struct net *net = read_pnet(&b->k.net);
1109 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1110 __xfrm_policy_inexact_prune_bin(b, false);
1111 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1114 static void __xfrm_policy_inexact_flush(struct net *net)
1116 struct xfrm_pol_inexact_bin *bin, *t;
1118 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1120 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1121 __xfrm_policy_inexact_prune_bin(bin, false);
1124 static struct hlist_head *
1125 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1126 struct xfrm_policy *policy, u8 dir)
1128 struct xfrm_pol_inexact_node *n;
1131 net = xp_net(policy);
1132 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1134 if (xfrm_policy_inexact_insert_use_any_list(policy))
1137 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1139 policy->selector.prefixlen_d)) {
1140 write_seqcount_begin(&bin->count);
1141 n = xfrm_policy_inexact_insert_node(net,
1143 &policy->selector.saddr,
1145 policy->selector.prefixlen_s,
1147 write_seqcount_end(&bin->count);
1154 /* daddr is fixed */
1155 write_seqcount_begin(&bin->count);
1156 n = xfrm_policy_inexact_insert_node(net,
1158 &policy->selector.daddr,
1160 policy->selector.prefixlen_d, dir);
1161 write_seqcount_end(&bin->count);
1165 /* saddr is wildcard */
1166 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1168 policy->selector.prefixlen_s))
1171 write_seqcount_begin(&bin->count);
1172 n = xfrm_policy_inexact_insert_node(net,
1174 &policy->selector.saddr,
1176 policy->selector.prefixlen_s, dir);
1177 write_seqcount_end(&bin->count);
1184 static struct xfrm_policy *
1185 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1187 struct xfrm_pol_inexact_bin *bin;
1188 struct xfrm_policy *delpol;
1189 struct hlist_head *chain;
1192 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1194 return ERR_PTR(-ENOMEM);
1196 net = xp_net(policy);
1197 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1199 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1201 __xfrm_policy_inexact_prune_bin(bin, false);
1202 return ERR_PTR(-ENOMEM);
1205 delpol = xfrm_policy_insert_list(chain, policy, excl);
1206 if (delpol && excl) {
1207 __xfrm_policy_inexact_prune_bin(bin, false);
1208 return ERR_PTR(-EEXIST);
1211 chain = &net->xfrm.policy_inexact[dir];
1212 xfrm_policy_insert_inexact_list(chain, policy);
1215 __xfrm_policy_inexact_prune_bin(bin, false);
1220 static void xfrm_hash_rebuild(struct work_struct *work)
1222 struct net *net = container_of(work, struct net,
1223 xfrm.policy_hthresh.work);
1225 struct xfrm_policy *pol;
1226 struct xfrm_policy *policy;
1227 struct hlist_head *chain;
1228 struct hlist_head *odst;
1229 struct hlist_node *newpos;
1233 u8 lbits4, rbits4, lbits6, rbits6;
1235 mutex_lock(&hash_resize_mutex);
1237 /* read selector prefixlen thresholds */
1239 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1241 lbits4 = net->xfrm.policy_hthresh.lbits4;
1242 rbits4 = net->xfrm.policy_hthresh.rbits4;
1243 lbits6 = net->xfrm.policy_hthresh.lbits6;
1244 rbits6 = net->xfrm.policy_hthresh.rbits6;
1245 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1247 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1248 write_seqcount_begin(&xfrm_policy_hash_generation);
1250 /* make sure that we can insert the indirect policies again before
1251 * we start with destructive action.
1253 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1254 struct xfrm_pol_inexact_bin *bin;
1257 dir = xfrm_policy_id2dir(policy->index);
1258 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1261 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1262 if (policy->family == AF_INET) {
1270 if (policy->family == AF_INET) {
1279 if (policy->selector.prefixlen_d < dbits ||
1280 policy->selector.prefixlen_s < sbits)
1283 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1287 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1291 /* reset the bydst and inexact table in all directions */
1292 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1293 struct hlist_node *n;
1295 hlist_for_each_entry_safe(policy, n,
1296 &net->xfrm.policy_inexact[dir],
1297 bydst_inexact_list) {
1298 hlist_del_rcu(&policy->bydst);
1299 hlist_del_init(&policy->bydst_inexact_list);
1302 hmask = net->xfrm.policy_bydst[dir].hmask;
1303 odst = net->xfrm.policy_bydst[dir].table;
1304 for (i = hmask; i >= 0; i--) {
1305 hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1306 hlist_del_rcu(&policy->bydst);
1308 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1309 /* dir out => dst = remote, src = local */
1310 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1311 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1312 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1313 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1315 /* dir in/fwd => dst = local, src = remote */
1316 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1317 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1318 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1319 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1323 /* re-insert all policies by order of creation */
1324 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1325 if (policy->walk.dead)
1327 dir = xfrm_policy_id2dir(policy->index);
1328 if (dir >= XFRM_POLICY_MAX) {
1329 /* skip socket policies */
1333 chain = policy_hash_bysel(net, &policy->selector,
1334 policy->family, dir);
1337 void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1339 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1343 hlist_for_each_entry(pol, chain, bydst) {
1344 if (policy->priority >= pol->priority)
1345 newpos = &pol->bydst;
1350 hlist_add_behind_rcu(&policy->bydst, newpos);
1352 hlist_add_head_rcu(&policy->bydst, chain);
1356 __xfrm_policy_inexact_flush(net);
1357 write_seqcount_end(&xfrm_policy_hash_generation);
1358 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1360 mutex_unlock(&hash_resize_mutex);
1363 void xfrm_policy_hash_rebuild(struct net *net)
1365 schedule_work(&net->xfrm.policy_hthresh.work);
1367 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1369 /* Generate new index... KAME seems to generate them ordered by cost
1370 * of an absolute inpredictability of ordering of rules. This will not pass. */
1371 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1373 static u32 idx_generator;
1376 struct hlist_head *list;
1377 struct xfrm_policy *p;
1382 idx = (idx_generator | dir);
1391 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1393 hlist_for_each_entry(p, list, byidx) {
1394 if (p->index == idx) {
1404 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1406 u32 *p1 = (u32 *) s1;
1407 u32 *p2 = (u32 *) s2;
1408 int len = sizeof(struct xfrm_selector) / sizeof(u32);
1411 for (i = 0; i < len; i++) {
1419 static void xfrm_policy_requeue(struct xfrm_policy *old,
1420 struct xfrm_policy *new)
1422 struct xfrm_policy_queue *pq = &old->polq;
1423 struct sk_buff_head list;
1425 if (skb_queue_empty(&pq->hold_queue))
1428 __skb_queue_head_init(&list);
1430 spin_lock_bh(&pq->hold_queue.lock);
1431 skb_queue_splice_init(&pq->hold_queue, &list);
1432 if (del_timer(&pq->hold_timer))
1434 spin_unlock_bh(&pq->hold_queue.lock);
1438 spin_lock_bh(&pq->hold_queue.lock);
1439 skb_queue_splice(&list, &pq->hold_queue);
1440 pq->timeout = XFRM_QUEUE_TMO_MIN;
1441 if (!mod_timer(&pq->hold_timer, jiffies))
1443 spin_unlock_bh(&pq->hold_queue.lock);
1446 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1447 struct xfrm_policy *pol)
1449 return mark->v == pol->mark.v && mark->m == pol->mark.m;
1452 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1454 const struct xfrm_pol_inexact_key *k = data;
1455 u32 a = k->type << 24 | k->dir << 16 | k->family;
1457 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1461 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1463 const struct xfrm_pol_inexact_bin *b = data;
1465 return xfrm_pol_bin_key(&b->k, 0, seed);
1468 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1471 const struct xfrm_pol_inexact_key *key = arg->key;
1472 const struct xfrm_pol_inexact_bin *b = ptr;
1475 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1478 ret = b->k.dir ^ key->dir;
1482 ret = b->k.type ^ key->type;
1486 ret = b->k.family ^ key->family;
1490 return b->k.if_id ^ key->if_id;
1493 static const struct rhashtable_params xfrm_pol_inexact_params = {
1494 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head),
1495 .hashfn = xfrm_pol_bin_key,
1496 .obj_hashfn = xfrm_pol_bin_obj,
1497 .obj_cmpfn = xfrm_pol_bin_cmp,
1498 .automatic_shrinking = true,
1501 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1502 struct xfrm_policy *policy)
1504 struct xfrm_policy *pol, *delpol = NULL;
1505 struct hlist_node *newpos = NULL;
1508 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1509 if (pol->type == policy->type &&
1510 pol->if_id == policy->if_id &&
1511 !selector_cmp(&pol->selector, &policy->selector) &&
1512 xfrm_policy_mark_match(&policy->mark, pol) &&
1513 xfrm_sec_ctx_match(pol->security, policy->security) &&
1516 if (policy->priority > pol->priority)
1518 } else if (policy->priority >= pol->priority) {
1519 newpos = &pol->bydst_inexact_list;
1527 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1529 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1531 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1537 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1538 struct xfrm_policy *policy,
1541 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1543 hlist_for_each_entry(pol, chain, bydst) {
1544 if (pol->type == policy->type &&
1545 pol->if_id == policy->if_id &&
1546 !selector_cmp(&pol->selector, &policy->selector) &&
1547 xfrm_policy_mark_match(&policy->mark, pol) &&
1548 xfrm_sec_ctx_match(pol->security, policy->security) &&
1551 return ERR_PTR(-EEXIST);
1553 if (policy->priority > pol->priority)
1555 } else if (policy->priority >= pol->priority) {
1564 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1566 hlist_add_head_rcu(&policy->bydst, chain);
1571 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1573 struct net *net = xp_net(policy);
1574 struct xfrm_policy *delpol;
1575 struct hlist_head *chain;
1577 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1578 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1580 delpol = xfrm_policy_insert_list(chain, policy, excl);
1582 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1584 if (IS_ERR(delpol)) {
1585 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1586 return PTR_ERR(delpol);
1589 __xfrm_policy_link(policy, dir);
1591 /* After previous checking, family can either be AF_INET or AF_INET6 */
1592 if (policy->family == AF_INET)
1593 rt_genid_bump_ipv4(net);
1595 rt_genid_bump_ipv6(net);
1598 xfrm_policy_requeue(delpol, policy);
1599 __xfrm_policy_unlink(delpol, dir);
1601 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1602 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1603 policy->curlft.add_time = ktime_get_real_seconds();
1604 policy->curlft.use_time = 0;
1605 if (!mod_timer(&policy->timer, jiffies + HZ))
1606 xfrm_pol_hold(policy);
1607 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1610 xfrm_policy_kill(delpol);
1611 else if (xfrm_bydst_should_resize(net, dir, NULL))
1612 schedule_work(&net->xfrm.policy_hash_work);
1616 EXPORT_SYMBOL(xfrm_policy_insert);
1618 static struct xfrm_policy *
1619 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1620 u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1621 struct xfrm_sec_ctx *ctx)
1623 struct xfrm_policy *pol;
1628 hlist_for_each_entry(pol, chain, bydst) {
1629 if (pol->type == type &&
1630 pol->if_id == if_id &&
1631 xfrm_policy_mark_match(mark, pol) &&
1632 !selector_cmp(sel, &pol->selector) &&
1633 xfrm_sec_ctx_match(ctx, pol->security))
1640 struct xfrm_policy *
1641 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1642 u8 type, int dir, struct xfrm_selector *sel,
1643 struct xfrm_sec_ctx *ctx, int delete, int *err)
1645 struct xfrm_pol_inexact_bin *bin = NULL;
1646 struct xfrm_policy *pol, *ret = NULL;
1647 struct hlist_head *chain;
1650 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1651 chain = policy_hash_bysel(net, sel, sel->family, dir);
1653 struct xfrm_pol_inexact_candidates cand;
1656 bin = xfrm_policy_inexact_lookup(net, type,
1657 sel->family, dir, if_id);
1659 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1663 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1666 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1671 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1672 struct xfrm_policy *tmp;
1674 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1680 if (!pol || tmp->pos < pol->pos)
1684 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1691 *err = security_xfrm_policy_delete(pol->security);
1693 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1696 __xfrm_policy_unlink(pol, dir);
1700 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1703 xfrm_policy_kill(ret);
1705 xfrm_policy_inexact_prune_bin(bin);
1708 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1710 struct xfrm_policy *
1711 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1712 u8 type, int dir, u32 id, int delete, int *err)
1714 struct xfrm_policy *pol, *ret;
1715 struct hlist_head *chain;
1718 if (xfrm_policy_id2dir(id) != dir)
1722 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1723 chain = net->xfrm.policy_byidx + idx_hash(net, id);
1725 hlist_for_each_entry(pol, chain, byidx) {
1726 if (pol->type == type && pol->index == id &&
1727 pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1730 *err = security_xfrm_policy_delete(
1733 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1736 __xfrm_policy_unlink(pol, dir);
1742 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1745 xfrm_policy_kill(ret);
1748 EXPORT_SYMBOL(xfrm_policy_byid);
1750 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1752 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1754 struct xfrm_policy *pol;
1757 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1758 if (pol->walk.dead ||
1759 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1763 err = security_xfrm_policy_delete(pol->security);
1765 xfrm_audit_policy_delete(pol, 0, task_valid);
1773 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1779 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1781 int dir, err = 0, cnt = 0;
1782 struct xfrm_policy *pol;
1784 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1786 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1791 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1792 dir = xfrm_policy_id2dir(pol->index);
1793 if (pol->walk.dead ||
1794 dir >= XFRM_POLICY_MAX ||
1798 __xfrm_policy_unlink(pol, dir);
1799 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1801 xfrm_audit_policy_delete(pol, 1, task_valid);
1802 xfrm_policy_kill(pol);
1803 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1807 __xfrm_policy_inexact_flush(net);
1811 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1814 EXPORT_SYMBOL(xfrm_policy_flush);
1816 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1817 int (*func)(struct xfrm_policy *, int, int, void*),
1820 struct xfrm_policy *pol;
1821 struct xfrm_policy_walk_entry *x;
1824 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1825 walk->type != XFRM_POLICY_TYPE_ANY)
1828 if (list_empty(&walk->walk.all) && walk->seq != 0)
1831 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1832 if (list_empty(&walk->walk.all))
1833 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1835 x = list_first_entry(&walk->walk.all,
1836 struct xfrm_policy_walk_entry, all);
1838 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1841 pol = container_of(x, struct xfrm_policy, walk);
1842 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1843 walk->type != pol->type)
1845 error = func(pol, xfrm_policy_id2dir(pol->index),
1848 list_move_tail(&walk->walk.all, &x->all);
1853 if (walk->seq == 0) {
1857 list_del_init(&walk->walk.all);
1859 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1862 EXPORT_SYMBOL(xfrm_policy_walk);
1864 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1866 INIT_LIST_HEAD(&walk->walk.all);
1867 walk->walk.dead = 1;
1871 EXPORT_SYMBOL(xfrm_policy_walk_init);
1873 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1875 if (list_empty(&walk->walk.all))
1878 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1879 list_del(&walk->walk.all);
1880 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1882 EXPORT_SYMBOL(xfrm_policy_walk_done);
1885 * Find policy to apply to this flow.
1887 * Returns 0 if policy found, else an -errno.
1889 static int xfrm_policy_match(const struct xfrm_policy *pol,
1890 const struct flowi *fl,
1891 u8 type, u16 family, int dir, u32 if_id)
1893 const struct xfrm_selector *sel = &pol->selector;
1897 if (pol->family != family ||
1898 pol->if_id != if_id ||
1899 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1903 match = xfrm_selector_match(sel, fl, family);
1905 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1910 static struct xfrm_pol_inexact_node *
1911 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1912 seqcount_spinlock_t *count,
1913 const xfrm_address_t *addr, u16 family)
1915 const struct rb_node *parent;
1919 seq = read_seqcount_begin(count);
1921 parent = rcu_dereference_raw(r->rb_node);
1923 struct xfrm_pol_inexact_node *node;
1926 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1928 delta = xfrm_policy_addr_delta(addr, &node->addr,
1929 node->prefixlen, family);
1931 parent = rcu_dereference_raw(parent->rb_left);
1933 } else if (delta > 0) {
1934 parent = rcu_dereference_raw(parent->rb_right);
1941 if (read_seqcount_retry(count, seq))
1948 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1949 struct xfrm_pol_inexact_bin *b,
1950 const xfrm_address_t *saddr,
1951 const xfrm_address_t *daddr)
1953 struct xfrm_pol_inexact_node *n;
1959 family = b->k.family;
1960 memset(cand, 0, sizeof(*cand));
1961 cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1963 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1966 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1967 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1970 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1973 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1976 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1981 static struct xfrm_pol_inexact_bin *
1982 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1985 struct xfrm_pol_inexact_key k = {
1992 write_pnet(&k.net, net);
1994 return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1995 xfrm_pol_inexact_params);
1998 static struct xfrm_pol_inexact_bin *
1999 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2002 struct xfrm_pol_inexact_bin *bin;
2004 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2007 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2013 static struct xfrm_policy *
2014 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2015 struct xfrm_policy *prefer,
2016 const struct flowi *fl,
2017 u8 type, u16 family, int dir, u32 if_id)
2019 u32 priority = prefer ? prefer->priority : ~0u;
2020 struct xfrm_policy *pol;
2025 hlist_for_each_entry_rcu(pol, chain, bydst) {
2028 if (pol->priority > priority)
2031 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2034 return ERR_PTR(err);
2040 /* matches. Is it older than *prefer? */
2041 if (pol->priority == priority &&
2042 prefer->pos < pol->pos)
2052 static struct xfrm_policy *
2053 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2054 struct xfrm_policy *prefer,
2055 const struct flowi *fl,
2056 u8 type, u16 family, int dir, u32 if_id)
2058 struct xfrm_policy *tmp;
2061 for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2062 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2064 fl, type, family, dir,
2077 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2078 const struct flowi *fl,
2082 struct xfrm_pol_inexact_candidates cand;
2083 const xfrm_address_t *daddr, *saddr;
2084 struct xfrm_pol_inexact_bin *bin;
2085 struct xfrm_policy *pol, *ret;
2086 struct hlist_head *chain;
2087 unsigned int sequence;
2090 daddr = xfrm_flowi_daddr(fl, family);
2091 saddr = xfrm_flowi_saddr(fl, family);
2092 if (unlikely(!daddr || !saddr))
2098 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2099 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2100 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2103 hlist_for_each_entry_rcu(pol, chain, bydst) {
2104 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2117 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2118 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2122 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2123 family, dir, if_id);
2131 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2134 if (ret && !xfrm_pol_hold_rcu(ret))
2142 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2143 const struct flowi *fl,
2144 u16 family, u8 dir, u32 if_id)
2146 #ifdef CONFIG_XFRM_SUB_POLICY
2147 struct xfrm_policy *pol;
2149 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2154 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2158 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2159 const struct flowi *fl,
2160 u16 family, u32 if_id)
2162 struct xfrm_policy *pol;
2166 pol = rcu_dereference(sk->sk_policy[dir]);
2171 if (pol->family != family) {
2176 match = xfrm_selector_match(&pol->selector, fl, family);
2178 if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2179 pol->if_id != if_id) {
2183 err = security_xfrm_policy_lookup(pol->security,
2187 if (!xfrm_pol_hold_rcu(pol))
2189 } else if (err == -ESRCH) {
2202 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2204 struct net *net = xp_net(pol);
2206 list_add(&pol->walk.all, &net->xfrm.policy_all);
2207 net->xfrm.policy_count[dir]++;
2211 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2214 struct net *net = xp_net(pol);
2216 if (list_empty(&pol->walk.all))
2219 /* Socket policies are not hashed. */
2220 if (!hlist_unhashed(&pol->bydst)) {
2221 hlist_del_rcu(&pol->bydst);
2222 hlist_del_init(&pol->bydst_inexact_list);
2223 hlist_del(&pol->byidx);
2226 list_del_init(&pol->walk.all);
2227 net->xfrm.policy_count[dir]--;
2232 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2234 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2237 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2239 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2242 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2244 struct net *net = xp_net(pol);
2246 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2247 pol = __xfrm_policy_unlink(pol, dir);
2248 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2250 xfrm_policy_kill(pol);
2255 EXPORT_SYMBOL(xfrm_policy_delete);
2257 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2259 struct net *net = sock_net(sk);
2260 struct xfrm_policy *old_pol;
2262 #ifdef CONFIG_XFRM_SUB_POLICY
2263 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2267 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2268 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2269 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2271 pol->curlft.add_time = ktime_get_real_seconds();
2272 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2273 xfrm_sk_policy_link(pol, dir);
2275 rcu_assign_pointer(sk->sk_policy[dir], pol);
2278 xfrm_policy_requeue(old_pol, pol);
2280 /* Unlinking succeeds always. This is the only function
2281 * allowed to delete or replace socket policy.
2283 xfrm_sk_policy_unlink(old_pol, dir);
2285 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2288 xfrm_policy_kill(old_pol);
2293 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2295 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2296 struct net *net = xp_net(old);
2299 newp->selector = old->selector;
2300 if (security_xfrm_policy_clone(old->security,
2303 return NULL; /* ENOMEM */
2305 newp->lft = old->lft;
2306 newp->curlft = old->curlft;
2307 newp->mark = old->mark;
2308 newp->if_id = old->if_id;
2309 newp->action = old->action;
2310 newp->flags = old->flags;
2311 newp->xfrm_nr = old->xfrm_nr;
2312 newp->index = old->index;
2313 newp->type = old->type;
2314 newp->family = old->family;
2315 memcpy(newp->xfrm_vec, old->xfrm_vec,
2316 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2317 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2318 xfrm_sk_policy_link(newp, dir);
2319 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2325 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2327 const struct xfrm_policy *p;
2328 struct xfrm_policy *np;
2332 for (i = 0; i < 2; i++) {
2333 p = rcu_dereference(osk->sk_policy[i]);
2335 np = clone_policy(p, i);
2336 if (unlikely(!np)) {
2340 rcu_assign_pointer(sk->sk_policy[i], np);
2348 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2349 xfrm_address_t *remote, unsigned short family, u32 mark)
2352 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2354 if (unlikely(afinfo == NULL))
2356 err = afinfo->get_saddr(net, oif, local, remote, mark);
2361 /* Resolve list of templates for the flow, given policy. */
2364 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2365 struct xfrm_state **xfrm, unsigned short family)
2367 struct net *net = xp_net(policy);
2370 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2371 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2374 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2375 struct xfrm_state *x;
2376 xfrm_address_t *remote = daddr;
2377 xfrm_address_t *local = saddr;
2378 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2380 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2381 tmpl->mode == XFRM_MODE_BEET) {
2382 remote = &tmpl->id.daddr;
2383 local = &tmpl->saddr;
2384 if (xfrm_addr_any(local, tmpl->encap_family)) {
2385 error = xfrm_get_saddr(net, fl->flowi_oif,
2387 tmpl->encap_family, 0);
2394 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2395 family, policy->if_id);
2397 if (x && x->km.state == XFRM_STATE_VALID) {
2404 error = (x->km.state == XFRM_STATE_ERROR ?
2407 } else if (error == -ESRCH) {
2411 if (!tmpl->optional)
2417 for (nx--; nx >= 0; nx--)
2418 xfrm_state_put(xfrm[nx]);
2423 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2424 struct xfrm_state **xfrm, unsigned short family)
2426 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2427 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2433 for (i = 0; i < npols; i++) {
2434 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2439 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2447 /* found states are sorted for outbound processing */
2449 xfrm_state_sort(xfrm, tpp, cnx, family);
2454 for (cnx--; cnx >= 0; cnx--)
2455 xfrm_state_put(tpp[cnx]);
2460 static int xfrm_get_tos(const struct flowi *fl, int family)
2462 if (family == AF_INET)
2463 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2468 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2470 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2471 struct dst_ops *dst_ops;
2472 struct xfrm_dst *xdst;
2475 return ERR_PTR(-EINVAL);
2479 dst_ops = &net->xfrm.xfrm4_dst_ops;
2481 #if IS_ENABLED(CONFIG_IPV6)
2483 dst_ops = &net->xfrm.xfrm6_dst_ops;
2489 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2492 struct dst_entry *dst = &xdst->u.dst;
2494 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2496 xdst = ERR_PTR(-ENOBUFS);
2503 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2506 if (dst->ops->family == AF_INET6) {
2507 struct rt6_info *rt = (struct rt6_info *)dst;
2508 path->path_cookie = rt6_get_cookie(rt);
2509 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2513 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2514 const struct flowi *fl)
2516 const struct xfrm_policy_afinfo *afinfo =
2517 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2523 err = afinfo->fill_dst(xdst, dev, fl);
2531 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2532 * all the metrics... Shortly, bundle a bundle.
2535 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2536 struct xfrm_state **xfrm,
2537 struct xfrm_dst **bundle,
2539 const struct flowi *fl,
2540 struct dst_entry *dst)
2542 const struct xfrm_state_afinfo *afinfo;
2543 const struct xfrm_mode *inner_mode;
2544 struct net *net = xp_net(policy);
2545 unsigned long now = jiffies;
2546 struct net_device *dev;
2547 struct xfrm_dst *xdst_prev = NULL;
2548 struct xfrm_dst *xdst0 = NULL;
2552 int nfheader_len = 0;
2553 int trailer_len = 0;
2555 int family = policy->selector.family;
2556 xfrm_address_t saddr, daddr;
2558 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2560 tos = xfrm_get_tos(fl, family);
2564 for (; i < nx; i++) {
2565 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2566 struct dst_entry *dst1 = &xdst->u.dst;
2568 err = PTR_ERR(xdst);
2578 /* Ref count is taken during xfrm_alloc_dst()
2579 * No need to do dst_clone() on dst1
2581 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2583 if (xfrm[i]->sel.family == AF_UNSPEC) {
2584 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2585 xfrm_af2proto(family));
2587 err = -EAFNOSUPPORT;
2592 inner_mode = &xfrm[i]->inner_mode;
2595 dst_copy_metrics(dst1, dst);
2597 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2600 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2601 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2603 family = xfrm[i]->props.family;
2604 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2605 &saddr, &daddr, family, mark);
2612 dst1->xfrm = xfrm[i];
2613 xdst->xfrm_genid = xfrm[i]->genid;
2615 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2616 dst1->lastuse = now;
2618 dst1->input = dst_discard;
2621 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2623 dst1->output = afinfo->output;
2625 dst1->output = dst_discard_out;
2630 header_len += xfrm[i]->props.header_len;
2631 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2632 nfheader_len += xfrm[i]->props.header_len;
2633 trailer_len += xfrm[i]->props.trailer_len;
2636 xfrm_dst_set_child(xdst_prev, dst);
2644 xfrm_init_path(xdst0, dst, nfheader_len);
2645 xfrm_init_pmtu(bundle, nx);
2647 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2648 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2649 err = xfrm_fill_dst(xdst_prev, dev, fl);
2653 xdst_prev->u.dst.header_len = header_len;
2654 xdst_prev->u.dst.trailer_len = trailer_len;
2655 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2656 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2659 return &xdst0->u.dst;
2663 xfrm_state_put(xfrm[i]);
2666 dst_release_immediate(&xdst0->u.dst);
2668 return ERR_PTR(err);
2671 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2672 struct xfrm_policy **pols,
2673 int *num_pols, int *num_xfrms)
2677 if (*num_pols == 0 || !pols[0]) {
2682 if (IS_ERR(pols[0]))
2683 return PTR_ERR(pols[0]);
2685 *num_xfrms = pols[0]->xfrm_nr;
2687 #ifdef CONFIG_XFRM_SUB_POLICY
2688 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2689 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2690 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2691 XFRM_POLICY_TYPE_MAIN,
2696 if (IS_ERR(pols[1])) {
2697 xfrm_pols_put(pols, *num_pols);
2698 return PTR_ERR(pols[1]);
2701 (*num_xfrms) += pols[1]->xfrm_nr;
2705 for (i = 0; i < *num_pols; i++) {
2706 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2716 static struct xfrm_dst *
2717 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2718 const struct flowi *fl, u16 family,
2719 struct dst_entry *dst_orig)
2721 struct net *net = xp_net(pols[0]);
2722 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2723 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2724 struct xfrm_dst *xdst;
2725 struct dst_entry *dst;
2728 /* Try to instantiate a bundle */
2729 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2735 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2736 return ERR_PTR(err);
2739 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2741 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2742 return ERR_CAST(dst);
2745 xdst = (struct xfrm_dst *)dst;
2746 xdst->num_xfrms = err;
2747 xdst->num_pols = num_pols;
2748 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2749 xdst->policy_genid = atomic_read(&pols[0]->genid);
2754 static void xfrm_policy_queue_process(struct timer_list *t)
2756 struct sk_buff *skb;
2758 struct dst_entry *dst;
2759 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2760 struct net *net = xp_net(pol);
2761 struct xfrm_policy_queue *pq = &pol->polq;
2763 struct sk_buff_head list;
2766 spin_lock(&pq->hold_queue.lock);
2767 skb = skb_peek(&pq->hold_queue);
2769 spin_unlock(&pq->hold_queue.lock);
2775 /* Fixup the mark to support VTI. */
2776 skb_mark = skb->mark;
2777 skb->mark = pol->mark.v;
2778 xfrm_decode_session(skb, &fl, dst->ops->family);
2779 skb->mark = skb_mark;
2780 spin_unlock(&pq->hold_queue.lock);
2782 dst_hold(xfrm_dst_path(dst));
2783 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2787 if (dst->flags & DST_XFRM_QUEUE) {
2790 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2793 pq->timeout = pq->timeout << 1;
2794 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2801 __skb_queue_head_init(&list);
2803 spin_lock(&pq->hold_queue.lock);
2805 skb_queue_splice_init(&pq->hold_queue, &list);
2806 spin_unlock(&pq->hold_queue.lock);
2808 while (!skb_queue_empty(&list)) {
2809 skb = __skb_dequeue(&list);
2811 /* Fixup the mark to support VTI. */
2812 skb_mark = skb->mark;
2813 skb->mark = pol->mark.v;
2814 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2815 skb->mark = skb_mark;
2817 dst_hold(xfrm_dst_path(skb_dst(skb)));
2818 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2826 skb_dst_set(skb, dst);
2828 dst_output(net, skb->sk, skb);
2837 skb_queue_purge(&pq->hold_queue);
2841 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2843 unsigned long sched_next;
2844 struct dst_entry *dst = skb_dst(skb);
2845 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2846 struct xfrm_policy *pol = xdst->pols[0];
2847 struct xfrm_policy_queue *pq = &pol->polq;
2849 if (unlikely(skb_fclone_busy(sk, skb))) {
2854 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2861 spin_lock_bh(&pq->hold_queue.lock);
2864 pq->timeout = XFRM_QUEUE_TMO_MIN;
2866 sched_next = jiffies + pq->timeout;
2868 if (del_timer(&pq->hold_timer)) {
2869 if (time_before(pq->hold_timer.expires, sched_next))
2870 sched_next = pq->hold_timer.expires;
2874 __skb_queue_tail(&pq->hold_queue, skb);
2875 if (!mod_timer(&pq->hold_timer, sched_next))
2878 spin_unlock_bh(&pq->hold_queue.lock);
2883 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2884 struct xfrm_flo *xflo,
2885 const struct flowi *fl,
2890 struct net_device *dev;
2891 struct dst_entry *dst;
2892 struct dst_entry *dst1;
2893 struct xfrm_dst *xdst;
2895 xdst = xfrm_alloc_dst(net, family);
2899 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2900 net->xfrm.sysctl_larval_drop ||
2904 dst = xflo->dst_orig;
2905 dst1 = &xdst->u.dst;
2909 dst_copy_metrics(dst1, dst);
2911 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2912 dst1->flags |= DST_XFRM_QUEUE;
2913 dst1->lastuse = jiffies;
2915 dst1->input = dst_discard;
2916 dst1->output = xdst_queue_output;
2919 xfrm_dst_set_child(xdst, dst);
2922 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2929 err = xfrm_fill_dst(xdst, dev, fl);
2938 xdst = ERR_PTR(err);
2942 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2943 const struct flowi *fl,
2945 struct xfrm_flo *xflo, u32 if_id)
2947 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2948 int num_pols = 0, num_xfrms = 0, err;
2949 struct xfrm_dst *xdst;
2951 /* Resolve policies to use if we couldn't get them from
2952 * previous cache entry */
2954 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2955 err = xfrm_expand_policies(fl, family, pols,
2956 &num_pols, &num_xfrms);
2962 goto make_dummy_bundle;
2964 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2967 err = PTR_ERR(xdst);
2968 if (err == -EREMOTE) {
2969 xfrm_pols_put(pols, num_pols);
2975 goto make_dummy_bundle;
2976 } else if (xdst == NULL) {
2978 goto make_dummy_bundle;
2984 /* We found policies, but there's no bundles to instantiate:
2985 * either because the policy blocks, has no transformations or
2986 * we could not build template (no xfrm_states).*/
2987 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2989 xfrm_pols_put(pols, num_pols);
2990 return ERR_CAST(xdst);
2992 xdst->num_pols = num_pols;
2993 xdst->num_xfrms = num_xfrms;
2994 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2999 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3001 xfrm_pols_put(pols, num_pols);
3002 return ERR_PTR(err);
3005 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3006 struct dst_entry *dst_orig)
3008 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3009 struct dst_entry *ret;
3012 dst_release(dst_orig);
3013 return ERR_PTR(-EINVAL);
3015 ret = afinfo->blackhole_route(net, dst_orig);
3022 /* Finds/creates a bundle for given flow and if_id
3024 * At the moment we eat a raw IP route. Mostly to speed up lookups
3025 * on interfaces with disabled IPsec.
3027 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3030 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3031 struct dst_entry *dst_orig,
3032 const struct flowi *fl,
3033 const struct sock *sk,
3034 int flags, u32 if_id)
3036 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3037 struct xfrm_dst *xdst;
3038 struct dst_entry *dst, *route;
3039 u16 family = dst_orig->ops->family;
3040 u8 dir = XFRM_POLICY_OUT;
3041 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3047 sk = sk_const_to_full_sk(sk);
3048 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3050 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3052 err = xfrm_expand_policies(fl, family, pols,
3053 &num_pols, &num_xfrms);
3058 if (num_xfrms <= 0) {
3059 drop_pols = num_pols;
3063 xdst = xfrm_resolve_and_create_bundle(
3068 xfrm_pols_put(pols, num_pols);
3069 err = PTR_ERR(xdst);
3070 if (err == -EREMOTE)
3074 } else if (xdst == NULL) {
3076 drop_pols = num_pols;
3080 route = xdst->route;
3085 struct xfrm_flo xflo;
3087 xflo.dst_orig = dst_orig;
3090 /* To accelerate a bit... */
3091 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3092 !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3095 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3099 err = PTR_ERR(xdst);
3103 num_pols = xdst->num_pols;
3104 num_xfrms = xdst->num_xfrms;
3105 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3106 route = xdst->route;
3110 if (route == NULL && num_xfrms > 0) {
3111 /* The only case when xfrm_bundle_lookup() returns a
3112 * bundle with null route, is when the template could
3113 * not be resolved. It means policies are there, but
3114 * bundle could not be created, since we don't yet
3115 * have the xfrm_state's. We need to wait for KM to
3116 * negotiate new SA's or bail out with error.*/
3117 if (net->xfrm.sysctl_larval_drop) {
3118 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3125 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3133 if ((flags & XFRM_LOOKUP_ICMP) &&
3134 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3139 for (i = 0; i < num_pols; i++)
3140 pols[i]->curlft.use_time = ktime_get_real_seconds();
3142 if (num_xfrms < 0) {
3143 /* Prohibit the flow */
3144 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3147 } else if (num_xfrms > 0) {
3148 /* Flow transformed */
3149 dst_release(dst_orig);
3151 /* Flow passes untransformed */
3156 xfrm_pols_put(pols, drop_pols);
3157 if (dst && dst->xfrm &&
3158 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3159 dst->flags |= DST_XFRM_TUNNEL;
3163 if (!(flags & XFRM_LOOKUP_ICMP)) {
3171 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3172 dst_release(dst_orig);
3173 xfrm_pols_put(pols, drop_pols);
3174 return ERR_PTR(err);
3176 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3178 /* Main function: finds/creates a bundle for given flow.
3180 * At the moment we eat a raw IP route. Mostly to speed up lookups
3181 * on interfaces with disabled IPsec.
3183 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3184 const struct flowi *fl, const struct sock *sk,
3187 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3189 EXPORT_SYMBOL(xfrm_lookup);
3191 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3192 * Otherwise we may send out blackholed packets.
3194 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3195 const struct flowi *fl,
3196 const struct sock *sk, int flags)
3198 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3199 flags | XFRM_LOOKUP_QUEUE |
3200 XFRM_LOOKUP_KEEP_DST_REF);
3202 if (PTR_ERR(dst) == -EREMOTE)
3203 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3206 dst_release(dst_orig);
3210 EXPORT_SYMBOL(xfrm_lookup_route);
3213 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3215 struct sec_path *sp = skb_sec_path(skb);
3216 struct xfrm_state *x;
3218 if (!sp || idx < 0 || idx >= sp->len)
3221 if (!x->type->reject)
3223 return x->type->reject(x, skb, fl);
3226 /* When skb is transformed back to its "native" form, we have to
3227 * check policy restrictions. At the moment we make this in maximally
3228 * stupid way. Shame on me. :-) Of course, connected sockets must
3229 * have policy cached at them.
3233 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3234 unsigned short family)
3236 if (xfrm_state_kern(x))
3237 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3238 return x->id.proto == tmpl->id.proto &&
3239 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3240 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3241 x->props.mode == tmpl->mode &&
3242 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3243 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3244 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3245 xfrm_state_addr_cmp(tmpl, x, family));
3249 * 0 or more than 0 is returned when validation is succeeded (either bypass
3250 * because of optional transport mode, or next index of the mathced secpath
3251 * state with the template.
3252 * -1 is returned when no matching template is found.
3253 * Otherwise "-2 - errored_index" is returned.
3256 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3257 unsigned short family)
3261 if (tmpl->optional) {
3262 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3266 for (; idx < sp->len; idx++) {
3267 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3269 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3279 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3281 const struct iphdr *iph = ip_hdr(skb);
3283 u8 *xprth = skb_network_header(skb) + ihl * 4;
3284 struct flowi4 *fl4 = &fl->u.ip4;
3287 if (skb_dst(skb) && skb_dst(skb)->dev)
3288 oif = skb_dst(skb)->dev->ifindex;
3290 memset(fl4, 0, sizeof(struct flowi4));
3291 fl4->flowi4_mark = skb->mark;
3292 fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3294 fl4->flowi4_proto = iph->protocol;
3295 fl4->daddr = reverse ? iph->saddr : iph->daddr;
3296 fl4->saddr = reverse ? iph->daddr : iph->saddr;
3297 fl4->flowi4_tos = iph->tos;
3299 if (!ip_is_fragment(iph)) {
3300 switch (iph->protocol) {
3302 case IPPROTO_UDPLITE:
3306 if (xprth + 4 < skb->data ||
3307 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3310 xprth = skb_network_header(skb) + ihl * 4;
3311 ports = (__be16 *)xprth;
3313 fl4->fl4_sport = ports[!!reverse];
3314 fl4->fl4_dport = ports[!reverse];
3318 if (xprth + 2 < skb->data ||
3319 pskb_may_pull(skb, xprth + 2 - skb->data)) {
3322 xprth = skb_network_header(skb) + ihl * 4;
3325 fl4->fl4_icmp_type = icmp[0];
3326 fl4->fl4_icmp_code = icmp[1];
3330 if (xprth + 12 < skb->data ||
3331 pskb_may_pull(skb, xprth + 12 - skb->data)) {
3335 xprth = skb_network_header(skb) + ihl * 4;
3336 greflags = (__be16 *)xprth;
3337 gre_hdr = (__be32 *)xprth;
3339 if (greflags[0] & GRE_KEY) {
3340 if (greflags[0] & GRE_CSUM)
3342 fl4->fl4_gre_key = gre_hdr[1];
3352 #if IS_ENABLED(CONFIG_IPV6)
3354 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3356 struct flowi6 *fl6 = &fl->u.ip6;
3358 const struct ipv6hdr *hdr = ipv6_hdr(skb);
3359 u32 offset = sizeof(*hdr);
3360 struct ipv6_opt_hdr *exthdr;
3361 const unsigned char *nh = skb_network_header(skb);
3362 u16 nhoff = IP6CB(skb)->nhoff;
3367 nhoff = offsetof(struct ipv6hdr, nexthdr);
3369 nexthdr = nh[nhoff];
3371 if (skb_dst(skb) && skb_dst(skb)->dev)
3372 oif = skb_dst(skb)->dev->ifindex;
3374 memset(fl6, 0, sizeof(struct flowi6));
3375 fl6->flowi6_mark = skb->mark;
3376 fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3378 fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3379 fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3381 while (nh + offset + sizeof(*exthdr) < skb->data ||
3382 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3383 nh = skb_network_header(skb);
3384 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3387 case NEXTHDR_FRAGMENT:
3390 case NEXTHDR_ROUTING:
3393 offset += ipv6_optlen(exthdr);
3394 nexthdr = exthdr->nexthdr;
3395 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3398 case IPPROTO_UDPLITE:
3402 if (!onlyproto && (nh + offset + 4 < skb->data ||
3403 pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3406 nh = skb_network_header(skb);
3407 ports = (__be16 *)(nh + offset);
3408 fl6->fl6_sport = ports[!!reverse];
3409 fl6->fl6_dport = ports[!reverse];
3411 fl6->flowi6_proto = nexthdr;
3413 case IPPROTO_ICMPV6:
3414 if (!onlyproto && (nh + offset + 2 < skb->data ||
3415 pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3418 nh = skb_network_header(skb);
3419 icmp = (u8 *)(nh + offset);
3420 fl6->fl6_icmp_type = icmp[0];
3421 fl6->fl6_icmp_code = icmp[1];
3423 fl6->flowi6_proto = nexthdr;
3425 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3427 offset += ipv6_optlen(exthdr);
3428 if (!onlyproto && (nh + offset + 3 < skb->data ||
3429 pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3432 nh = skb_network_header(skb);
3433 mh = (struct ip6_mh *)(nh + offset);
3434 fl6->fl6_mh_type = mh->ip6mh_type;
3436 fl6->flowi6_proto = nexthdr;
3440 fl6->flowi6_proto = nexthdr;
3447 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3448 unsigned int family, int reverse)
3452 decode_session4(skb, fl, reverse);
3454 #if IS_ENABLED(CONFIG_IPV6)
3456 decode_session6(skb, fl, reverse);
3460 return -EAFNOSUPPORT;
3463 return security_xfrm_decode_session(skb, &fl->flowi_secid);
3465 EXPORT_SYMBOL(__xfrm_decode_session);
3467 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3469 for (; k < sp->len; k++) {
3470 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3479 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3480 unsigned short family)
3482 struct net *net = dev_net(skb->dev);
3483 struct xfrm_policy *pol;
3484 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3491 const struct xfrm_if_cb *ifcb;
3492 struct sec_path *sp;
3497 ifcb = xfrm_if_get_cb();
3500 xi = ifcb->decode_session(skb, family);
3502 if_id = xi->p.if_id;
3508 reverse = dir & ~XFRM_POLICY_MASK;
3509 dir &= XFRM_POLICY_MASK;
3511 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3512 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3516 nf_nat_decode_session(skb, &fl, family);
3518 /* First, check used SA against their selectors. */
3519 sp = skb_sec_path(skb);
3523 for (i = sp->len - 1; i >= 0; i--) {
3524 struct xfrm_state *x = sp->xvec[i];
3525 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3526 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3533 sk = sk_to_full_sk(sk);
3534 if (sk && sk->sk_policy[dir]) {
3535 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3537 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3543 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3546 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3551 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3552 xfrm_secpath_reject(xerr_idx, skb, &fl);
3553 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3559 pol->curlft.use_time = ktime_get_real_seconds();
3563 #ifdef CONFIG_XFRM_SUB_POLICY
3564 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3565 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3567 XFRM_POLICY_IN, if_id);
3569 if (IS_ERR(pols[1])) {
3570 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3573 pols[1]->curlft.use_time = ktime_get_real_seconds();
3579 if (pol->action == XFRM_POLICY_ALLOW) {
3580 static struct sec_path dummy;
3581 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3582 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3583 struct xfrm_tmpl **tpp = tp;
3587 sp = skb_sec_path(skb);
3591 for (pi = 0; pi < npols; pi++) {
3592 if (pols[pi] != pol &&
3593 pols[pi]->action != XFRM_POLICY_ALLOW) {
3594 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3597 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3598 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3601 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3602 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3606 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3610 /* For each tunnel xfrm, find the first matching tmpl.
3611 * For each tmpl before that, find corresponding xfrm.
3612 * Order is _important_. Later we will implement
3613 * some barriers, but at the moment barriers
3614 * are implied between each two transformations.
3616 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3617 k = xfrm_policy_ok(tpp[i], sp, k, family);
3620 /* "-2 - errored_index" returned */
3622 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3627 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3628 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3632 xfrm_pols_put(pols, npols);
3635 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3638 xfrm_secpath_reject(xerr_idx, skb, &fl);
3640 xfrm_pols_put(pols, npols);
3643 EXPORT_SYMBOL(__xfrm_policy_check);
3645 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3647 struct net *net = dev_net(skb->dev);
3649 struct dst_entry *dst;
3652 if (xfrm_decode_session(skb, &fl, family) < 0) {
3653 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3658 if (!skb_dst(skb)) {
3659 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3663 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3668 skb_dst_set(skb, dst);
3671 EXPORT_SYMBOL(__xfrm_route_forward);
3673 /* Optimize later using cookies and generation ids. */
3675 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3677 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3678 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3679 * get validated by dst_ops->check on every use. We do this
3680 * because when a normal route referenced by an XFRM dst is
3681 * obsoleted we do not go looking around for all parent
3682 * referencing XFRM dsts so that we can invalidate them. It
3683 * is just too much work. Instead we make the checks here on
3684 * every use. For example:
3686 * XFRM dst A --> IPv4 dst X
3688 * X is the "xdst->route" of A (X is also the "dst->path" of A
3689 * in this example). If X is marked obsolete, "A" will not
3690 * notice. That's what we are validating here via the
3691 * stale_bundle() check.
3693 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3695 * This will force stale_bundle() to fail on any xdst bundle with
3696 * this dst linked in it.
3698 if (dst->obsolete < 0 && !stale_bundle(dst))
3704 static int stale_bundle(struct dst_entry *dst)
3706 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3709 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3711 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3712 dst->dev = dev_net(dev)->loopback_dev;
3717 EXPORT_SYMBOL(xfrm_dst_ifdown);
3719 static void xfrm_link_failure(struct sk_buff *skb)
3721 /* Impossible. Such dst must be popped before reaches point of failure. */
3724 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3727 if (dst->obsolete) {
3735 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3738 struct xfrm_dst *xdst = bundle[nr];
3739 u32 pmtu, route_mtu_cached;
3740 struct dst_entry *dst;
3743 pmtu = dst_mtu(xfrm_dst_child(dst));
3744 xdst->child_mtu_cached = pmtu;
3746 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3748 route_mtu_cached = dst_mtu(xdst->route);
3749 xdst->route_mtu_cached = route_mtu_cached;
3751 if (pmtu > route_mtu_cached)
3752 pmtu = route_mtu_cached;
3754 dst_metric_set(dst, RTAX_MTU, pmtu);
3758 /* Check that the bundle accepts the flow and its components are
3762 static int xfrm_bundle_ok(struct xfrm_dst *first)
3764 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3765 struct dst_entry *dst = &first->u.dst;
3766 struct xfrm_dst *xdst;
3770 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3771 (dst->dev && !netif_running(dst->dev)))
3774 if (dst->flags & DST_XFRM_QUEUE)
3777 start_from = nr = 0;
3779 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3781 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3783 if (xdst->xfrm_genid != dst->xfrm->genid)
3785 if (xdst->num_pols > 0 &&
3786 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3789 bundle[nr++] = xdst;
3791 mtu = dst_mtu(xfrm_dst_child(dst));
3792 if (xdst->child_mtu_cached != mtu) {
3794 xdst->child_mtu_cached = mtu;
3797 if (!dst_check(xdst->route, xdst->route_cookie))
3799 mtu = dst_mtu(xdst->route);
3800 if (xdst->route_mtu_cached != mtu) {
3802 xdst->route_mtu_cached = mtu;
3805 dst = xfrm_dst_child(dst);
3806 } while (dst->xfrm);
3808 if (likely(!start_from))
3811 xdst = bundle[start_from - 1];
3812 mtu = xdst->child_mtu_cached;
3813 while (start_from--) {
3816 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3817 if (mtu > xdst->route_mtu_cached)
3818 mtu = xdst->route_mtu_cached;
3819 dst_metric_set(dst, RTAX_MTU, mtu);
3823 xdst = bundle[start_from - 1];
3824 xdst->child_mtu_cached = mtu;
3830 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3832 return dst_metric_advmss(xfrm_dst_path(dst));
3835 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3837 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3839 return mtu ? : dst_mtu(xfrm_dst_path(dst));
3842 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3846 const struct xfrm_state *xfrm = dst->xfrm;
3848 dst = xfrm_dst_child(dst);
3850 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3852 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3853 daddr = xfrm->coaddr;
3854 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3855 daddr = &xfrm->id.daddr;
3860 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3861 struct sk_buff *skb,
3864 const struct dst_entry *path = xfrm_dst_path(dst);
3867 daddr = xfrm_get_dst_nexthop(dst, daddr);
3868 return path->ops->neigh_lookup(path, skb, daddr);
3871 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3873 const struct dst_entry *path = xfrm_dst_path(dst);
3875 daddr = xfrm_get_dst_nexthop(dst, daddr);
3876 path->ops->confirm_neigh(path, daddr);
3879 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3883 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3884 return -EAFNOSUPPORT;
3886 spin_lock(&xfrm_policy_afinfo_lock);
3887 if (unlikely(xfrm_policy_afinfo[family] != NULL))
3890 struct dst_ops *dst_ops = afinfo->dst_ops;
3891 if (likely(dst_ops->kmem_cachep == NULL))
3892 dst_ops->kmem_cachep = xfrm_dst_cache;
3893 if (likely(dst_ops->check == NULL))
3894 dst_ops->check = xfrm_dst_check;
3895 if (likely(dst_ops->default_advmss == NULL))
3896 dst_ops->default_advmss = xfrm_default_advmss;
3897 if (likely(dst_ops->mtu == NULL))
3898 dst_ops->mtu = xfrm_mtu;
3899 if (likely(dst_ops->negative_advice == NULL))
3900 dst_ops->negative_advice = xfrm_negative_advice;
3901 if (likely(dst_ops->link_failure == NULL))
3902 dst_ops->link_failure = xfrm_link_failure;
3903 if (likely(dst_ops->neigh_lookup == NULL))
3904 dst_ops->neigh_lookup = xfrm_neigh_lookup;
3905 if (likely(!dst_ops->confirm_neigh))
3906 dst_ops->confirm_neigh = xfrm_confirm_neigh;
3907 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3909 spin_unlock(&xfrm_policy_afinfo_lock);
3913 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3915 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3917 struct dst_ops *dst_ops = afinfo->dst_ops;
3920 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3921 if (xfrm_policy_afinfo[i] != afinfo)
3923 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3929 dst_ops->kmem_cachep = NULL;
3930 dst_ops->check = NULL;
3931 dst_ops->negative_advice = NULL;
3932 dst_ops->link_failure = NULL;
3934 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3936 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3938 spin_lock(&xfrm_if_cb_lock);
3939 rcu_assign_pointer(xfrm_if_cb, ifcb);
3940 spin_unlock(&xfrm_if_cb_lock);
3942 EXPORT_SYMBOL(xfrm_if_register_cb);
3944 void xfrm_if_unregister_cb(void)
3946 RCU_INIT_POINTER(xfrm_if_cb, NULL);
3949 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3951 #ifdef CONFIG_XFRM_STATISTICS
3952 static int __net_init xfrm_statistics_init(struct net *net)
3955 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3956 if (!net->mib.xfrm_statistics)
3958 rv = xfrm_proc_init(net);
3960 free_percpu(net->mib.xfrm_statistics);
3964 static void xfrm_statistics_fini(struct net *net)
3966 xfrm_proc_fini(net);
3967 free_percpu(net->mib.xfrm_statistics);
3970 static int __net_init xfrm_statistics_init(struct net *net)
3975 static void xfrm_statistics_fini(struct net *net)
3980 static int __net_init xfrm_policy_init(struct net *net)
3982 unsigned int hmask, sz;
3985 if (net_eq(net, &init_net)) {
3986 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
3987 sizeof(struct xfrm_dst),
3988 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3990 err = rhashtable_init(&xfrm_policy_inexact_table,
3991 &xfrm_pol_inexact_params);
3996 sz = (hmask+1) * sizeof(struct hlist_head);
3998 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
3999 if (!net->xfrm.policy_byidx)
4001 net->xfrm.policy_idx_hmask = hmask;
4003 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4004 struct xfrm_policy_hash *htab;
4006 net->xfrm.policy_count[dir] = 0;
4007 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4008 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4010 htab = &net->xfrm.policy_bydst[dir];
4011 htab->table = xfrm_hash_alloc(sz);
4014 htab->hmask = hmask;
4020 net->xfrm.policy_hthresh.lbits4 = 32;
4021 net->xfrm.policy_hthresh.rbits4 = 32;
4022 net->xfrm.policy_hthresh.lbits6 = 128;
4023 net->xfrm.policy_hthresh.rbits6 = 128;
4025 seqlock_init(&net->xfrm.policy_hthresh.lock);
4027 INIT_LIST_HEAD(&net->xfrm.policy_all);
4028 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4029 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4030 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4034 for (dir--; dir >= 0; dir--) {
4035 struct xfrm_policy_hash *htab;
4037 htab = &net->xfrm.policy_bydst[dir];
4038 xfrm_hash_free(htab->table, sz);
4040 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4045 static void xfrm_policy_fini(struct net *net)
4047 struct xfrm_pol_inexact_bin *b, *t;
4051 flush_work(&net->xfrm.policy_hash_work);
4052 #ifdef CONFIG_XFRM_SUB_POLICY
4053 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4055 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4057 WARN_ON(!list_empty(&net->xfrm.policy_all));
4059 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4060 struct xfrm_policy_hash *htab;
4062 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4064 htab = &net->xfrm.policy_bydst[dir];
4065 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4066 WARN_ON(!hlist_empty(htab->table));
4067 xfrm_hash_free(htab->table, sz);
4070 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4071 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4072 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4074 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4075 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4076 __xfrm_policy_inexact_prune_bin(b, true);
4077 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4080 static int __net_init xfrm_net_init(struct net *net)
4084 /* Initialize the per-net locks here */
4085 spin_lock_init(&net->xfrm.xfrm_state_lock);
4086 spin_lock_init(&net->xfrm.xfrm_policy_lock);
4087 mutex_init(&net->xfrm.xfrm_cfg_mutex);
4089 rv = xfrm_statistics_init(net);
4091 goto out_statistics;
4092 rv = xfrm_state_init(net);
4095 rv = xfrm_policy_init(net);
4098 rv = xfrm_sysctl_init(net);
4105 xfrm_policy_fini(net);
4107 xfrm_state_fini(net);
4109 xfrm_statistics_fini(net);
4114 static void __net_exit xfrm_net_exit(struct net *net)
4116 xfrm_sysctl_fini(net);
4117 xfrm_policy_fini(net);
4118 xfrm_state_fini(net);
4119 xfrm_statistics_fini(net);
4122 static struct pernet_operations __net_initdata xfrm_net_ops = {
4123 .init = xfrm_net_init,
4124 .exit = xfrm_net_exit,
4127 void __init xfrm_init(void)
4129 register_pernet_subsys(&xfrm_net_ops);
4131 seqcount_mutex_init(&xfrm_policy_hash_generation, &hash_resize_mutex);
4134 #ifdef CONFIG_XFRM_ESPINTCP
4139 #ifdef CONFIG_AUDITSYSCALL
4140 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4141 struct audit_buffer *audit_buf)
4143 struct xfrm_sec_ctx *ctx = xp->security;
4144 struct xfrm_selector *sel = &xp->selector;
4147 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4148 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4150 switch (sel->family) {
4152 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4153 if (sel->prefixlen_s != 32)
4154 audit_log_format(audit_buf, " src_prefixlen=%d",
4156 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4157 if (sel->prefixlen_d != 32)
4158 audit_log_format(audit_buf, " dst_prefixlen=%d",
4162 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4163 if (sel->prefixlen_s != 128)
4164 audit_log_format(audit_buf, " src_prefixlen=%d",
4166 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4167 if (sel->prefixlen_d != 128)
4168 audit_log_format(audit_buf, " dst_prefixlen=%d",
4174 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4176 struct audit_buffer *audit_buf;
4178 audit_buf = xfrm_audit_start("SPD-add");
4179 if (audit_buf == NULL)
4181 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4182 audit_log_format(audit_buf, " res=%u", result);
4183 xfrm_audit_common_policyinfo(xp, audit_buf);
4184 audit_log_end(audit_buf);
4186 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4188 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4191 struct audit_buffer *audit_buf;
4193 audit_buf = xfrm_audit_start("SPD-delete");
4194 if (audit_buf == NULL)
4196 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4197 audit_log_format(audit_buf, " res=%u", result);
4198 xfrm_audit_common_policyinfo(xp, audit_buf);
4199 audit_log_end(audit_buf);
4201 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4204 #ifdef CONFIG_XFRM_MIGRATE
4205 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4206 const struct xfrm_selector *sel_tgt)
4208 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4209 if (sel_tgt->family == sel_cmp->family &&
4210 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4212 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4214 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4215 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4219 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4226 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4227 u8 dir, u8 type, struct net *net)
4229 struct xfrm_policy *pol, *ret = NULL;
4230 struct hlist_head *chain;
4233 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4234 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4235 hlist_for_each_entry(pol, chain, bydst) {
4236 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4237 pol->type == type) {
4239 priority = ret->priority;
4243 chain = &net->xfrm.policy_inexact[dir];
4244 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4245 if ((pol->priority >= priority) && ret)
4248 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4249 pol->type == type) {
4257 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4262 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4266 if (t->mode == m->mode && t->id.proto == m->proto &&
4267 (m->reqid == 0 || t->reqid == m->reqid)) {
4269 case XFRM_MODE_TUNNEL:
4270 case XFRM_MODE_BEET:
4271 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4273 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4278 case XFRM_MODE_TRANSPORT:
4279 /* in case of transport mode, template does not store
4280 any IP addresses, hence we just compare mode and
4291 /* update endpoint address(es) of template(s) */
4292 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4293 struct xfrm_migrate *m, int num_migrate)
4295 struct xfrm_migrate *mp;
4298 write_lock_bh(&pol->lock);
4299 if (unlikely(pol->walk.dead)) {
4300 /* target policy has been deleted */
4301 write_unlock_bh(&pol->lock);
4305 for (i = 0; i < pol->xfrm_nr; i++) {
4306 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4307 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4310 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4311 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4313 /* update endpoints */
4314 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4315 sizeof(pol->xfrm_vec[i].id.daddr));
4316 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4317 sizeof(pol->xfrm_vec[i].saddr));
4318 pol->xfrm_vec[i].encap_family = mp->new_family;
4320 atomic_inc(&pol->genid);
4324 write_unlock_bh(&pol->lock);
4332 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4336 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4339 for (i = 0; i < num_migrate; i++) {
4340 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4341 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4344 /* check if there is any duplicated entry */
4345 for (j = i + 1; j < num_migrate; j++) {
4346 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4347 sizeof(m[i].old_daddr)) &&
4348 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4349 sizeof(m[i].old_saddr)) &&
4350 m[i].proto == m[j].proto &&
4351 m[i].mode == m[j].mode &&
4352 m[i].reqid == m[j].reqid &&
4353 m[i].old_family == m[j].old_family)
4361 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4362 struct xfrm_migrate *m, int num_migrate,
4363 struct xfrm_kmaddress *k, struct net *net,
4364 struct xfrm_encap_tmpl *encap)
4366 int i, err, nx_cur = 0, nx_new = 0;
4367 struct xfrm_policy *pol = NULL;
4368 struct xfrm_state *x, *xc;
4369 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4370 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4371 struct xfrm_migrate *mp;
4373 /* Stage 0 - sanity checks */
4374 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4377 if (dir >= XFRM_POLICY_MAX) {
4382 /* Stage 1 - find policy */
4383 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4388 /* Stage 2 - find and update state(s) */
4389 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4390 if ((x = xfrm_migrate_state_find(mp, net))) {
4393 xc = xfrm_state_migrate(x, mp, encap);
4404 /* Stage 3 - update policy */
4405 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4408 /* Stage 4 - delete old state(s) */
4410 xfrm_states_put(x_cur, nx_cur);
4411 xfrm_states_delete(x_cur, nx_cur);
4414 /* Stage 5 - announce */
4415 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4427 xfrm_states_put(x_cur, nx_cur);
4429 xfrm_states_delete(x_new, nx_new);
4433 EXPORT_SYMBOL(xfrm_migrate);