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);
1909 static struct xfrm_pol_inexact_node *
1910 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1911 seqcount_spinlock_t *count,
1912 const xfrm_address_t *addr, u16 family)
1914 const struct rb_node *parent;
1918 seq = read_seqcount_begin(count);
1920 parent = rcu_dereference_raw(r->rb_node);
1922 struct xfrm_pol_inexact_node *node;
1925 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1927 delta = xfrm_policy_addr_delta(addr, &node->addr,
1928 node->prefixlen, family);
1930 parent = rcu_dereference_raw(parent->rb_left);
1932 } else if (delta > 0) {
1933 parent = rcu_dereference_raw(parent->rb_right);
1940 if (read_seqcount_retry(count, seq))
1947 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1948 struct xfrm_pol_inexact_bin *b,
1949 const xfrm_address_t *saddr,
1950 const xfrm_address_t *daddr)
1952 struct xfrm_pol_inexact_node *n;
1958 family = b->k.family;
1959 memset(cand, 0, sizeof(*cand));
1960 cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1962 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1965 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1966 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1969 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1972 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1975 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1980 static struct xfrm_pol_inexact_bin *
1981 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1984 struct xfrm_pol_inexact_key k = {
1991 write_pnet(&k.net, net);
1993 return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1994 xfrm_pol_inexact_params);
1997 static struct xfrm_pol_inexact_bin *
1998 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2001 struct xfrm_pol_inexact_bin *bin;
2003 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2006 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2012 static struct xfrm_policy *
2013 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2014 struct xfrm_policy *prefer,
2015 const struct flowi *fl,
2016 u8 type, u16 family, int dir, u32 if_id)
2018 u32 priority = prefer ? prefer->priority : ~0u;
2019 struct xfrm_policy *pol;
2024 hlist_for_each_entry_rcu(pol, chain, bydst) {
2027 if (pol->priority > priority)
2030 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2033 return ERR_PTR(err);
2039 /* matches. Is it older than *prefer? */
2040 if (pol->priority == priority &&
2041 prefer->pos < pol->pos)
2051 static struct xfrm_policy *
2052 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2053 struct xfrm_policy *prefer,
2054 const struct flowi *fl,
2055 u8 type, u16 family, int dir, u32 if_id)
2057 struct xfrm_policy *tmp;
2060 for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2061 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2063 fl, type, family, dir,
2076 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2077 const struct flowi *fl,
2081 struct xfrm_pol_inexact_candidates cand;
2082 const xfrm_address_t *daddr, *saddr;
2083 struct xfrm_pol_inexact_bin *bin;
2084 struct xfrm_policy *pol, *ret;
2085 struct hlist_head *chain;
2086 unsigned int sequence;
2089 daddr = xfrm_flowi_daddr(fl, family);
2090 saddr = xfrm_flowi_saddr(fl, family);
2091 if (unlikely(!daddr || !saddr))
2097 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2098 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2099 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2102 hlist_for_each_entry_rcu(pol, chain, bydst) {
2103 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2116 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2117 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2121 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2122 family, dir, if_id);
2130 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2133 if (ret && !xfrm_pol_hold_rcu(ret))
2141 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2142 const struct flowi *fl,
2143 u16 family, u8 dir, u32 if_id)
2145 #ifdef CONFIG_XFRM_SUB_POLICY
2146 struct xfrm_policy *pol;
2148 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2153 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2157 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2158 const struct flowi *fl,
2159 u16 family, u32 if_id)
2161 struct xfrm_policy *pol;
2165 pol = rcu_dereference(sk->sk_policy[dir]);
2170 if (pol->family != family) {
2175 match = xfrm_selector_match(&pol->selector, fl, family);
2177 if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2178 pol->if_id != if_id) {
2182 err = security_xfrm_policy_lookup(pol->security,
2185 if (!xfrm_pol_hold_rcu(pol))
2187 } else if (err == -ESRCH) {
2200 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2202 struct net *net = xp_net(pol);
2204 list_add(&pol->walk.all, &net->xfrm.policy_all);
2205 net->xfrm.policy_count[dir]++;
2209 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2212 struct net *net = xp_net(pol);
2214 if (list_empty(&pol->walk.all))
2217 /* Socket policies are not hashed. */
2218 if (!hlist_unhashed(&pol->bydst)) {
2219 hlist_del_rcu(&pol->bydst);
2220 hlist_del_init(&pol->bydst_inexact_list);
2221 hlist_del(&pol->byidx);
2224 list_del_init(&pol->walk.all);
2225 net->xfrm.policy_count[dir]--;
2230 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2232 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2235 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2237 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2240 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2242 struct net *net = xp_net(pol);
2244 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2245 pol = __xfrm_policy_unlink(pol, dir);
2246 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2248 xfrm_policy_kill(pol);
2253 EXPORT_SYMBOL(xfrm_policy_delete);
2255 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2257 struct net *net = sock_net(sk);
2258 struct xfrm_policy *old_pol;
2260 #ifdef CONFIG_XFRM_SUB_POLICY
2261 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2265 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2266 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2267 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2269 pol->curlft.add_time = ktime_get_real_seconds();
2270 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2271 xfrm_sk_policy_link(pol, dir);
2273 rcu_assign_pointer(sk->sk_policy[dir], pol);
2276 xfrm_policy_requeue(old_pol, pol);
2278 /* Unlinking succeeds always. This is the only function
2279 * allowed to delete or replace socket policy.
2281 xfrm_sk_policy_unlink(old_pol, dir);
2283 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2286 xfrm_policy_kill(old_pol);
2291 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2293 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2294 struct net *net = xp_net(old);
2297 newp->selector = old->selector;
2298 if (security_xfrm_policy_clone(old->security,
2301 return NULL; /* ENOMEM */
2303 newp->lft = old->lft;
2304 newp->curlft = old->curlft;
2305 newp->mark = old->mark;
2306 newp->if_id = old->if_id;
2307 newp->action = old->action;
2308 newp->flags = old->flags;
2309 newp->xfrm_nr = old->xfrm_nr;
2310 newp->index = old->index;
2311 newp->type = old->type;
2312 newp->family = old->family;
2313 memcpy(newp->xfrm_vec, old->xfrm_vec,
2314 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2315 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2316 xfrm_sk_policy_link(newp, dir);
2317 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2323 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2325 const struct xfrm_policy *p;
2326 struct xfrm_policy *np;
2330 for (i = 0; i < 2; i++) {
2331 p = rcu_dereference(osk->sk_policy[i]);
2333 np = clone_policy(p, i);
2334 if (unlikely(!np)) {
2338 rcu_assign_pointer(sk->sk_policy[i], np);
2346 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2347 xfrm_address_t *remote, unsigned short family, u32 mark)
2350 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2352 if (unlikely(afinfo == NULL))
2354 err = afinfo->get_saddr(net, oif, local, remote, mark);
2359 /* Resolve list of templates for the flow, given policy. */
2362 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2363 struct xfrm_state **xfrm, unsigned short family)
2365 struct net *net = xp_net(policy);
2368 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2369 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2372 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2373 struct xfrm_state *x;
2374 xfrm_address_t *remote = daddr;
2375 xfrm_address_t *local = saddr;
2376 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2378 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2379 tmpl->mode == XFRM_MODE_BEET) {
2380 remote = &tmpl->id.daddr;
2381 local = &tmpl->saddr;
2382 if (xfrm_addr_any(local, tmpl->encap_family)) {
2383 error = xfrm_get_saddr(net, fl->flowi_oif,
2385 tmpl->encap_family, 0);
2392 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2393 family, policy->if_id);
2395 if (x && x->km.state == XFRM_STATE_VALID) {
2402 error = (x->km.state == XFRM_STATE_ERROR ?
2405 } else if (error == -ESRCH) {
2409 if (!tmpl->optional)
2415 for (nx--; nx >= 0; nx--)
2416 xfrm_state_put(xfrm[nx]);
2421 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2422 struct xfrm_state **xfrm, unsigned short family)
2424 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2425 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2431 for (i = 0; i < npols; i++) {
2432 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2437 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2445 /* found states are sorted for outbound processing */
2447 xfrm_state_sort(xfrm, tpp, cnx, family);
2452 for (cnx--; cnx >= 0; cnx--)
2453 xfrm_state_put(tpp[cnx]);
2458 static int xfrm_get_tos(const struct flowi *fl, int family)
2460 if (family == AF_INET)
2461 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2466 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2468 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2469 struct dst_ops *dst_ops;
2470 struct xfrm_dst *xdst;
2473 return ERR_PTR(-EINVAL);
2477 dst_ops = &net->xfrm.xfrm4_dst_ops;
2479 #if IS_ENABLED(CONFIG_IPV6)
2481 dst_ops = &net->xfrm.xfrm6_dst_ops;
2487 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2490 struct dst_entry *dst = &xdst->u.dst;
2492 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2494 xdst = ERR_PTR(-ENOBUFS);
2501 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2504 if (dst->ops->family == AF_INET6) {
2505 struct rt6_info *rt = (struct rt6_info *)dst;
2506 path->path_cookie = rt6_get_cookie(rt);
2507 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2511 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2512 const struct flowi *fl)
2514 const struct xfrm_policy_afinfo *afinfo =
2515 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2521 err = afinfo->fill_dst(xdst, dev, fl);
2529 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2530 * all the metrics... Shortly, bundle a bundle.
2533 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2534 struct xfrm_state **xfrm,
2535 struct xfrm_dst **bundle,
2537 const struct flowi *fl,
2538 struct dst_entry *dst)
2540 const struct xfrm_state_afinfo *afinfo;
2541 const struct xfrm_mode *inner_mode;
2542 struct net *net = xp_net(policy);
2543 unsigned long now = jiffies;
2544 struct net_device *dev;
2545 struct xfrm_dst *xdst_prev = NULL;
2546 struct xfrm_dst *xdst0 = NULL;
2550 int nfheader_len = 0;
2551 int trailer_len = 0;
2553 int family = policy->selector.family;
2554 xfrm_address_t saddr, daddr;
2556 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2558 tos = xfrm_get_tos(fl, family);
2562 for (; i < nx; i++) {
2563 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2564 struct dst_entry *dst1 = &xdst->u.dst;
2566 err = PTR_ERR(xdst);
2576 /* Ref count is taken during xfrm_alloc_dst()
2577 * No need to do dst_clone() on dst1
2579 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2581 if (xfrm[i]->sel.family == AF_UNSPEC) {
2582 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2583 xfrm_af2proto(family));
2585 err = -EAFNOSUPPORT;
2590 inner_mode = &xfrm[i]->inner_mode;
2593 dst_copy_metrics(dst1, dst);
2595 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2598 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2599 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2601 family = xfrm[i]->props.family;
2602 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2603 &saddr, &daddr, family, mark);
2610 dst1->xfrm = xfrm[i];
2611 xdst->xfrm_genid = xfrm[i]->genid;
2613 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2614 dst1->lastuse = now;
2616 dst1->input = dst_discard;
2619 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2621 dst1->output = afinfo->output;
2623 dst1->output = dst_discard_out;
2628 header_len += xfrm[i]->props.header_len;
2629 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2630 nfheader_len += xfrm[i]->props.header_len;
2631 trailer_len += xfrm[i]->props.trailer_len;
2634 xfrm_dst_set_child(xdst_prev, dst);
2642 xfrm_init_path(xdst0, dst, nfheader_len);
2643 xfrm_init_pmtu(bundle, nx);
2645 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2646 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2647 err = xfrm_fill_dst(xdst_prev, dev, fl);
2651 xdst_prev->u.dst.header_len = header_len;
2652 xdst_prev->u.dst.trailer_len = trailer_len;
2653 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2654 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2657 return &xdst0->u.dst;
2661 xfrm_state_put(xfrm[i]);
2664 dst_release_immediate(&xdst0->u.dst);
2666 return ERR_PTR(err);
2669 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2670 struct xfrm_policy **pols,
2671 int *num_pols, int *num_xfrms)
2675 if (*num_pols == 0 || !pols[0]) {
2680 if (IS_ERR(pols[0]))
2681 return PTR_ERR(pols[0]);
2683 *num_xfrms = pols[0]->xfrm_nr;
2685 #ifdef CONFIG_XFRM_SUB_POLICY
2686 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2687 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2688 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2689 XFRM_POLICY_TYPE_MAIN,
2694 if (IS_ERR(pols[1])) {
2695 xfrm_pols_put(pols, *num_pols);
2696 return PTR_ERR(pols[1]);
2699 (*num_xfrms) += pols[1]->xfrm_nr;
2703 for (i = 0; i < *num_pols; i++) {
2704 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2714 static struct xfrm_dst *
2715 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2716 const struct flowi *fl, u16 family,
2717 struct dst_entry *dst_orig)
2719 struct net *net = xp_net(pols[0]);
2720 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2721 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2722 struct xfrm_dst *xdst;
2723 struct dst_entry *dst;
2726 /* Try to instantiate a bundle */
2727 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2733 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2734 return ERR_PTR(err);
2737 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2739 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2740 return ERR_CAST(dst);
2743 xdst = (struct xfrm_dst *)dst;
2744 xdst->num_xfrms = err;
2745 xdst->num_pols = num_pols;
2746 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2747 xdst->policy_genid = atomic_read(&pols[0]->genid);
2752 static void xfrm_policy_queue_process(struct timer_list *t)
2754 struct sk_buff *skb;
2756 struct dst_entry *dst;
2757 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2758 struct net *net = xp_net(pol);
2759 struct xfrm_policy_queue *pq = &pol->polq;
2761 struct sk_buff_head list;
2764 spin_lock(&pq->hold_queue.lock);
2765 skb = skb_peek(&pq->hold_queue);
2767 spin_unlock(&pq->hold_queue.lock);
2773 /* Fixup the mark to support VTI. */
2774 skb_mark = skb->mark;
2775 skb->mark = pol->mark.v;
2776 xfrm_decode_session(skb, &fl, dst->ops->family);
2777 skb->mark = skb_mark;
2778 spin_unlock(&pq->hold_queue.lock);
2780 dst_hold(xfrm_dst_path(dst));
2781 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2785 if (dst->flags & DST_XFRM_QUEUE) {
2788 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2791 pq->timeout = pq->timeout << 1;
2792 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2799 __skb_queue_head_init(&list);
2801 spin_lock(&pq->hold_queue.lock);
2803 skb_queue_splice_init(&pq->hold_queue, &list);
2804 spin_unlock(&pq->hold_queue.lock);
2806 while (!skb_queue_empty(&list)) {
2807 skb = __skb_dequeue(&list);
2809 /* Fixup the mark to support VTI. */
2810 skb_mark = skb->mark;
2811 skb->mark = pol->mark.v;
2812 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2813 skb->mark = skb_mark;
2815 dst_hold(xfrm_dst_path(skb_dst(skb)));
2816 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2824 skb_dst_set(skb, dst);
2826 dst_output(net, skb->sk, skb);
2835 skb_queue_purge(&pq->hold_queue);
2839 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2841 unsigned long sched_next;
2842 struct dst_entry *dst = skb_dst(skb);
2843 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2844 struct xfrm_policy *pol = xdst->pols[0];
2845 struct xfrm_policy_queue *pq = &pol->polq;
2847 if (unlikely(skb_fclone_busy(sk, skb))) {
2852 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2859 spin_lock_bh(&pq->hold_queue.lock);
2862 pq->timeout = XFRM_QUEUE_TMO_MIN;
2864 sched_next = jiffies + pq->timeout;
2866 if (del_timer(&pq->hold_timer)) {
2867 if (time_before(pq->hold_timer.expires, sched_next))
2868 sched_next = pq->hold_timer.expires;
2872 __skb_queue_tail(&pq->hold_queue, skb);
2873 if (!mod_timer(&pq->hold_timer, sched_next))
2876 spin_unlock_bh(&pq->hold_queue.lock);
2881 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2882 struct xfrm_flo *xflo,
2883 const struct flowi *fl,
2888 struct net_device *dev;
2889 struct dst_entry *dst;
2890 struct dst_entry *dst1;
2891 struct xfrm_dst *xdst;
2893 xdst = xfrm_alloc_dst(net, family);
2897 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2898 net->xfrm.sysctl_larval_drop ||
2902 dst = xflo->dst_orig;
2903 dst1 = &xdst->u.dst;
2907 dst_copy_metrics(dst1, dst);
2909 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2910 dst1->flags |= DST_XFRM_QUEUE;
2911 dst1->lastuse = jiffies;
2913 dst1->input = dst_discard;
2914 dst1->output = xdst_queue_output;
2917 xfrm_dst_set_child(xdst, dst);
2920 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2927 err = xfrm_fill_dst(xdst, dev, fl);
2936 xdst = ERR_PTR(err);
2940 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2941 const struct flowi *fl,
2943 struct xfrm_flo *xflo, u32 if_id)
2945 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2946 int num_pols = 0, num_xfrms = 0, err;
2947 struct xfrm_dst *xdst;
2949 /* Resolve policies to use if we couldn't get them from
2950 * previous cache entry */
2952 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2953 err = xfrm_expand_policies(fl, family, pols,
2954 &num_pols, &num_xfrms);
2960 goto make_dummy_bundle;
2962 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2965 err = PTR_ERR(xdst);
2966 if (err == -EREMOTE) {
2967 xfrm_pols_put(pols, num_pols);
2973 goto make_dummy_bundle;
2974 } else if (xdst == NULL) {
2976 goto make_dummy_bundle;
2982 /* We found policies, but there's no bundles to instantiate:
2983 * either because the policy blocks, has no transformations or
2984 * we could not build template (no xfrm_states).*/
2985 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2987 xfrm_pols_put(pols, num_pols);
2988 return ERR_CAST(xdst);
2990 xdst->num_pols = num_pols;
2991 xdst->num_xfrms = num_xfrms;
2992 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2997 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2999 xfrm_pols_put(pols, num_pols);
3000 return ERR_PTR(err);
3003 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3004 struct dst_entry *dst_orig)
3006 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3007 struct dst_entry *ret;
3010 dst_release(dst_orig);
3011 return ERR_PTR(-EINVAL);
3013 ret = afinfo->blackhole_route(net, dst_orig);
3020 /* Finds/creates a bundle for given flow and if_id
3022 * At the moment we eat a raw IP route. Mostly to speed up lookups
3023 * on interfaces with disabled IPsec.
3025 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3028 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3029 struct dst_entry *dst_orig,
3030 const struct flowi *fl,
3031 const struct sock *sk,
3032 int flags, u32 if_id)
3034 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3035 struct xfrm_dst *xdst;
3036 struct dst_entry *dst, *route;
3037 u16 family = dst_orig->ops->family;
3038 u8 dir = XFRM_POLICY_OUT;
3039 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3045 sk = sk_const_to_full_sk(sk);
3046 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3048 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3050 err = xfrm_expand_policies(fl, family, pols,
3051 &num_pols, &num_xfrms);
3056 if (num_xfrms <= 0) {
3057 drop_pols = num_pols;
3061 xdst = xfrm_resolve_and_create_bundle(
3066 xfrm_pols_put(pols, num_pols);
3067 err = PTR_ERR(xdst);
3068 if (err == -EREMOTE)
3072 } else if (xdst == NULL) {
3074 drop_pols = num_pols;
3078 route = xdst->route;
3083 struct xfrm_flo xflo;
3085 xflo.dst_orig = dst_orig;
3088 /* To accelerate a bit... */
3089 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3090 !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3093 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3097 err = PTR_ERR(xdst);
3101 num_pols = xdst->num_pols;
3102 num_xfrms = xdst->num_xfrms;
3103 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3104 route = xdst->route;
3108 if (route == NULL && num_xfrms > 0) {
3109 /* The only case when xfrm_bundle_lookup() returns a
3110 * bundle with null route, is when the template could
3111 * not be resolved. It means policies are there, but
3112 * bundle could not be created, since we don't yet
3113 * have the xfrm_state's. We need to wait for KM to
3114 * negotiate new SA's or bail out with error.*/
3115 if (net->xfrm.sysctl_larval_drop) {
3116 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3123 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3131 if ((flags & XFRM_LOOKUP_ICMP) &&
3132 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3137 for (i = 0; i < num_pols; i++)
3138 pols[i]->curlft.use_time = ktime_get_real_seconds();
3140 if (num_xfrms < 0) {
3141 /* Prohibit the flow */
3142 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3145 } else if (num_xfrms > 0) {
3146 /* Flow transformed */
3147 dst_release(dst_orig);
3149 /* Flow passes untransformed */
3154 xfrm_pols_put(pols, drop_pols);
3155 if (dst && dst->xfrm &&
3156 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3157 dst->flags |= DST_XFRM_TUNNEL;
3161 if (!(flags & XFRM_LOOKUP_ICMP)) {
3169 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3170 dst_release(dst_orig);
3171 xfrm_pols_put(pols, drop_pols);
3172 return ERR_PTR(err);
3174 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3176 /* Main function: finds/creates a bundle for given flow.
3178 * At the moment we eat a raw IP route. Mostly to speed up lookups
3179 * on interfaces with disabled IPsec.
3181 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3182 const struct flowi *fl, const struct sock *sk,
3185 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3187 EXPORT_SYMBOL(xfrm_lookup);
3189 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3190 * Otherwise we may send out blackholed packets.
3192 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3193 const struct flowi *fl,
3194 const struct sock *sk, int flags)
3196 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3197 flags | XFRM_LOOKUP_QUEUE |
3198 XFRM_LOOKUP_KEEP_DST_REF);
3200 if (PTR_ERR(dst) == -EREMOTE)
3201 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3204 dst_release(dst_orig);
3208 EXPORT_SYMBOL(xfrm_lookup_route);
3211 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3213 struct sec_path *sp = skb_sec_path(skb);
3214 struct xfrm_state *x;
3216 if (!sp || idx < 0 || idx >= sp->len)
3219 if (!x->type->reject)
3221 return x->type->reject(x, skb, fl);
3224 /* When skb is transformed back to its "native" form, we have to
3225 * check policy restrictions. At the moment we make this in maximally
3226 * stupid way. Shame on me. :-) Of course, connected sockets must
3227 * have policy cached at them.
3231 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3232 unsigned short family)
3234 if (xfrm_state_kern(x))
3235 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3236 return x->id.proto == tmpl->id.proto &&
3237 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3238 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3239 x->props.mode == tmpl->mode &&
3240 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3241 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3242 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3243 xfrm_state_addr_cmp(tmpl, x, family));
3247 * 0 or more than 0 is returned when validation is succeeded (either bypass
3248 * because of optional transport mode, or next index of the mathced secpath
3249 * state with the template.
3250 * -1 is returned when no matching template is found.
3251 * Otherwise "-2 - errored_index" is returned.
3254 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3255 unsigned short family)
3259 if (tmpl->optional) {
3260 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3264 for (; idx < sp->len; idx++) {
3265 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3267 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3277 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3279 const struct iphdr *iph = ip_hdr(skb);
3281 u8 *xprth = skb_network_header(skb) + ihl * 4;
3282 struct flowi4 *fl4 = &fl->u.ip4;
3285 if (skb_dst(skb) && skb_dst(skb)->dev)
3286 oif = skb_dst(skb)->dev->ifindex;
3288 memset(fl4, 0, sizeof(struct flowi4));
3289 fl4->flowi4_mark = skb->mark;
3290 fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3292 fl4->flowi4_proto = iph->protocol;
3293 fl4->daddr = reverse ? iph->saddr : iph->daddr;
3294 fl4->saddr = reverse ? iph->daddr : iph->saddr;
3295 fl4->flowi4_tos = iph->tos;
3297 if (!ip_is_fragment(iph)) {
3298 switch (iph->protocol) {
3300 case IPPROTO_UDPLITE:
3304 if (xprth + 4 < skb->data ||
3305 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3308 xprth = skb_network_header(skb) + ihl * 4;
3309 ports = (__be16 *)xprth;
3311 fl4->fl4_sport = ports[!!reverse];
3312 fl4->fl4_dport = ports[!reverse];
3316 if (xprth + 2 < skb->data ||
3317 pskb_may_pull(skb, xprth + 2 - skb->data)) {
3320 xprth = skb_network_header(skb) + ihl * 4;
3323 fl4->fl4_icmp_type = icmp[0];
3324 fl4->fl4_icmp_code = icmp[1];
3328 if (xprth + 12 < skb->data ||
3329 pskb_may_pull(skb, xprth + 12 - skb->data)) {
3333 xprth = skb_network_header(skb) + ihl * 4;
3334 greflags = (__be16 *)xprth;
3335 gre_hdr = (__be32 *)xprth;
3337 if (greflags[0] & GRE_KEY) {
3338 if (greflags[0] & GRE_CSUM)
3340 fl4->fl4_gre_key = gre_hdr[1];
3350 #if IS_ENABLED(CONFIG_IPV6)
3352 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3354 struct flowi6 *fl6 = &fl->u.ip6;
3356 const struct ipv6hdr *hdr = ipv6_hdr(skb);
3357 u32 offset = sizeof(*hdr);
3358 struct ipv6_opt_hdr *exthdr;
3359 const unsigned char *nh = skb_network_header(skb);
3360 u16 nhoff = IP6CB(skb)->nhoff;
3365 nhoff = offsetof(struct ipv6hdr, nexthdr);
3367 nexthdr = nh[nhoff];
3369 if (skb_dst(skb) && skb_dst(skb)->dev)
3370 oif = skb_dst(skb)->dev->ifindex;
3372 memset(fl6, 0, sizeof(struct flowi6));
3373 fl6->flowi6_mark = skb->mark;
3374 fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3376 fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3377 fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3379 while (nh + offset + sizeof(*exthdr) < skb->data ||
3380 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3381 nh = skb_network_header(skb);
3382 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3385 case NEXTHDR_FRAGMENT:
3388 case NEXTHDR_ROUTING:
3391 offset += ipv6_optlen(exthdr);
3392 nexthdr = exthdr->nexthdr;
3393 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3396 case IPPROTO_UDPLITE:
3400 if (!onlyproto && (nh + offset + 4 < skb->data ||
3401 pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3404 nh = skb_network_header(skb);
3405 ports = (__be16 *)(nh + offset);
3406 fl6->fl6_sport = ports[!!reverse];
3407 fl6->fl6_dport = ports[!reverse];
3409 fl6->flowi6_proto = nexthdr;
3411 case IPPROTO_ICMPV6:
3412 if (!onlyproto && (nh + offset + 2 < skb->data ||
3413 pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3416 nh = skb_network_header(skb);
3417 icmp = (u8 *)(nh + offset);
3418 fl6->fl6_icmp_type = icmp[0];
3419 fl6->fl6_icmp_code = icmp[1];
3421 fl6->flowi6_proto = nexthdr;
3423 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3425 offset += ipv6_optlen(exthdr);
3426 if (!onlyproto && (nh + offset + 3 < skb->data ||
3427 pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3430 nh = skb_network_header(skb);
3431 mh = (struct ip6_mh *)(nh + offset);
3432 fl6->fl6_mh_type = mh->ip6mh_type;
3434 fl6->flowi6_proto = nexthdr;
3438 fl6->flowi6_proto = nexthdr;
3445 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3446 unsigned int family, int reverse)
3450 decode_session4(skb, fl, reverse);
3452 #if IS_ENABLED(CONFIG_IPV6)
3454 decode_session6(skb, fl, reverse);
3458 return -EAFNOSUPPORT;
3461 return security_xfrm_decode_session(skb, &fl->flowi_secid);
3463 EXPORT_SYMBOL(__xfrm_decode_session);
3465 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3467 for (; k < sp->len; k++) {
3468 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3477 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3478 unsigned short family)
3480 struct net *net = dev_net(skb->dev);
3481 struct xfrm_policy *pol;
3482 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3489 const struct xfrm_if_cb *ifcb;
3490 struct sec_path *sp;
3495 ifcb = xfrm_if_get_cb();
3498 xi = ifcb->decode_session(skb, family);
3500 if_id = xi->p.if_id;
3506 reverse = dir & ~XFRM_POLICY_MASK;
3507 dir &= XFRM_POLICY_MASK;
3509 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3510 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3514 nf_nat_decode_session(skb, &fl, family);
3516 /* First, check used SA against their selectors. */
3517 sp = skb_sec_path(skb);
3521 for (i = sp->len - 1; i >= 0; i--) {
3522 struct xfrm_state *x = sp->xvec[i];
3523 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3524 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3531 sk = sk_to_full_sk(sk);
3532 if (sk && sk->sk_policy[dir]) {
3533 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3535 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3541 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3544 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3549 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3550 xfrm_secpath_reject(xerr_idx, skb, &fl);
3551 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3557 pol->curlft.use_time = ktime_get_real_seconds();
3561 #ifdef CONFIG_XFRM_SUB_POLICY
3562 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3563 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3565 XFRM_POLICY_IN, if_id);
3567 if (IS_ERR(pols[1])) {
3568 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3571 pols[1]->curlft.use_time = ktime_get_real_seconds();
3577 if (pol->action == XFRM_POLICY_ALLOW) {
3578 static struct sec_path dummy;
3579 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3580 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3581 struct xfrm_tmpl **tpp = tp;
3585 sp = skb_sec_path(skb);
3589 for (pi = 0; pi < npols; pi++) {
3590 if (pols[pi] != pol &&
3591 pols[pi]->action != XFRM_POLICY_ALLOW) {
3592 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3595 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3596 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3599 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3600 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3604 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3608 /* For each tunnel xfrm, find the first matching tmpl.
3609 * For each tmpl before that, find corresponding xfrm.
3610 * Order is _important_. Later we will implement
3611 * some barriers, but at the moment barriers
3612 * are implied between each two transformations.
3614 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3615 k = xfrm_policy_ok(tpp[i], sp, k, family);
3618 /* "-2 - errored_index" returned */
3620 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3625 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3626 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3630 xfrm_pols_put(pols, npols);
3633 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3636 xfrm_secpath_reject(xerr_idx, skb, &fl);
3638 xfrm_pols_put(pols, npols);
3641 EXPORT_SYMBOL(__xfrm_policy_check);
3643 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3645 struct net *net = dev_net(skb->dev);
3647 struct dst_entry *dst;
3650 if (xfrm_decode_session(skb, &fl, family) < 0) {
3651 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3656 if (!skb_dst(skb)) {
3657 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3661 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3666 skb_dst_set(skb, dst);
3669 EXPORT_SYMBOL(__xfrm_route_forward);
3671 /* Optimize later using cookies and generation ids. */
3673 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3675 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3676 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3677 * get validated by dst_ops->check on every use. We do this
3678 * because when a normal route referenced by an XFRM dst is
3679 * obsoleted we do not go looking around for all parent
3680 * referencing XFRM dsts so that we can invalidate them. It
3681 * is just too much work. Instead we make the checks here on
3682 * every use. For example:
3684 * XFRM dst A --> IPv4 dst X
3686 * X is the "xdst->route" of A (X is also the "dst->path" of A
3687 * in this example). If X is marked obsolete, "A" will not
3688 * notice. That's what we are validating here via the
3689 * stale_bundle() check.
3691 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3693 * This will force stale_bundle() to fail on any xdst bundle with
3694 * this dst linked in it.
3696 if (dst->obsolete < 0 && !stale_bundle(dst))
3702 static int stale_bundle(struct dst_entry *dst)
3704 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3707 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3709 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3710 dst->dev = dev_net(dev)->loopback_dev;
3715 EXPORT_SYMBOL(xfrm_dst_ifdown);
3717 static void xfrm_link_failure(struct sk_buff *skb)
3719 /* Impossible. Such dst must be popped before reaches point of failure. */
3722 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3725 if (dst->obsolete) {
3733 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3736 struct xfrm_dst *xdst = bundle[nr];
3737 u32 pmtu, route_mtu_cached;
3738 struct dst_entry *dst;
3741 pmtu = dst_mtu(xfrm_dst_child(dst));
3742 xdst->child_mtu_cached = pmtu;
3744 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3746 route_mtu_cached = dst_mtu(xdst->route);
3747 xdst->route_mtu_cached = route_mtu_cached;
3749 if (pmtu > route_mtu_cached)
3750 pmtu = route_mtu_cached;
3752 dst_metric_set(dst, RTAX_MTU, pmtu);
3756 /* Check that the bundle accepts the flow and its components are
3760 static int xfrm_bundle_ok(struct xfrm_dst *first)
3762 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3763 struct dst_entry *dst = &first->u.dst;
3764 struct xfrm_dst *xdst;
3768 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3769 (dst->dev && !netif_running(dst->dev)))
3772 if (dst->flags & DST_XFRM_QUEUE)
3775 start_from = nr = 0;
3777 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3779 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3781 if (xdst->xfrm_genid != dst->xfrm->genid)
3783 if (xdst->num_pols > 0 &&
3784 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3787 bundle[nr++] = xdst;
3789 mtu = dst_mtu(xfrm_dst_child(dst));
3790 if (xdst->child_mtu_cached != mtu) {
3792 xdst->child_mtu_cached = mtu;
3795 if (!dst_check(xdst->route, xdst->route_cookie))
3797 mtu = dst_mtu(xdst->route);
3798 if (xdst->route_mtu_cached != mtu) {
3800 xdst->route_mtu_cached = mtu;
3803 dst = xfrm_dst_child(dst);
3804 } while (dst->xfrm);
3806 if (likely(!start_from))
3809 xdst = bundle[start_from - 1];
3810 mtu = xdst->child_mtu_cached;
3811 while (start_from--) {
3814 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3815 if (mtu > xdst->route_mtu_cached)
3816 mtu = xdst->route_mtu_cached;
3817 dst_metric_set(dst, RTAX_MTU, mtu);
3821 xdst = bundle[start_from - 1];
3822 xdst->child_mtu_cached = mtu;
3828 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3830 return dst_metric_advmss(xfrm_dst_path(dst));
3833 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3835 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3837 return mtu ? : dst_mtu(xfrm_dst_path(dst));
3840 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3844 const struct xfrm_state *xfrm = dst->xfrm;
3846 dst = xfrm_dst_child(dst);
3848 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3850 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3851 daddr = xfrm->coaddr;
3852 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3853 daddr = &xfrm->id.daddr;
3858 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3859 struct sk_buff *skb,
3862 const struct dst_entry *path = xfrm_dst_path(dst);
3865 daddr = xfrm_get_dst_nexthop(dst, daddr);
3866 return path->ops->neigh_lookup(path, skb, daddr);
3869 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3871 const struct dst_entry *path = xfrm_dst_path(dst);
3873 daddr = xfrm_get_dst_nexthop(dst, daddr);
3874 path->ops->confirm_neigh(path, daddr);
3877 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3881 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3882 return -EAFNOSUPPORT;
3884 spin_lock(&xfrm_policy_afinfo_lock);
3885 if (unlikely(xfrm_policy_afinfo[family] != NULL))
3888 struct dst_ops *dst_ops = afinfo->dst_ops;
3889 if (likely(dst_ops->kmem_cachep == NULL))
3890 dst_ops->kmem_cachep = xfrm_dst_cache;
3891 if (likely(dst_ops->check == NULL))
3892 dst_ops->check = xfrm_dst_check;
3893 if (likely(dst_ops->default_advmss == NULL))
3894 dst_ops->default_advmss = xfrm_default_advmss;
3895 if (likely(dst_ops->mtu == NULL))
3896 dst_ops->mtu = xfrm_mtu;
3897 if (likely(dst_ops->negative_advice == NULL))
3898 dst_ops->negative_advice = xfrm_negative_advice;
3899 if (likely(dst_ops->link_failure == NULL))
3900 dst_ops->link_failure = xfrm_link_failure;
3901 if (likely(dst_ops->neigh_lookup == NULL))
3902 dst_ops->neigh_lookup = xfrm_neigh_lookup;
3903 if (likely(!dst_ops->confirm_neigh))
3904 dst_ops->confirm_neigh = xfrm_confirm_neigh;
3905 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3907 spin_unlock(&xfrm_policy_afinfo_lock);
3911 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3913 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3915 struct dst_ops *dst_ops = afinfo->dst_ops;
3918 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3919 if (xfrm_policy_afinfo[i] != afinfo)
3921 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3927 dst_ops->kmem_cachep = NULL;
3928 dst_ops->check = NULL;
3929 dst_ops->negative_advice = NULL;
3930 dst_ops->link_failure = NULL;
3932 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3934 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3936 spin_lock(&xfrm_if_cb_lock);
3937 rcu_assign_pointer(xfrm_if_cb, ifcb);
3938 spin_unlock(&xfrm_if_cb_lock);
3940 EXPORT_SYMBOL(xfrm_if_register_cb);
3942 void xfrm_if_unregister_cb(void)
3944 RCU_INIT_POINTER(xfrm_if_cb, NULL);
3947 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3949 #ifdef CONFIG_XFRM_STATISTICS
3950 static int __net_init xfrm_statistics_init(struct net *net)
3953 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3954 if (!net->mib.xfrm_statistics)
3956 rv = xfrm_proc_init(net);
3958 free_percpu(net->mib.xfrm_statistics);
3962 static void xfrm_statistics_fini(struct net *net)
3964 xfrm_proc_fini(net);
3965 free_percpu(net->mib.xfrm_statistics);
3968 static int __net_init xfrm_statistics_init(struct net *net)
3973 static void xfrm_statistics_fini(struct net *net)
3978 static int __net_init xfrm_policy_init(struct net *net)
3980 unsigned int hmask, sz;
3983 if (net_eq(net, &init_net)) {
3984 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
3985 sizeof(struct xfrm_dst),
3986 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3988 err = rhashtable_init(&xfrm_policy_inexact_table,
3989 &xfrm_pol_inexact_params);
3994 sz = (hmask+1) * sizeof(struct hlist_head);
3996 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
3997 if (!net->xfrm.policy_byidx)
3999 net->xfrm.policy_idx_hmask = hmask;
4001 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4002 struct xfrm_policy_hash *htab;
4004 net->xfrm.policy_count[dir] = 0;
4005 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4006 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4008 htab = &net->xfrm.policy_bydst[dir];
4009 htab->table = xfrm_hash_alloc(sz);
4012 htab->hmask = hmask;
4018 net->xfrm.policy_hthresh.lbits4 = 32;
4019 net->xfrm.policy_hthresh.rbits4 = 32;
4020 net->xfrm.policy_hthresh.lbits6 = 128;
4021 net->xfrm.policy_hthresh.rbits6 = 128;
4023 seqlock_init(&net->xfrm.policy_hthresh.lock);
4025 INIT_LIST_HEAD(&net->xfrm.policy_all);
4026 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4027 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4028 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4032 for (dir--; dir >= 0; dir--) {
4033 struct xfrm_policy_hash *htab;
4035 htab = &net->xfrm.policy_bydst[dir];
4036 xfrm_hash_free(htab->table, sz);
4038 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4043 static void xfrm_policy_fini(struct net *net)
4045 struct xfrm_pol_inexact_bin *b, *t;
4049 flush_work(&net->xfrm.policy_hash_work);
4050 #ifdef CONFIG_XFRM_SUB_POLICY
4051 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4053 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4055 WARN_ON(!list_empty(&net->xfrm.policy_all));
4057 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4058 struct xfrm_policy_hash *htab;
4060 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4062 htab = &net->xfrm.policy_bydst[dir];
4063 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4064 WARN_ON(!hlist_empty(htab->table));
4065 xfrm_hash_free(htab->table, sz);
4068 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4069 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4070 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4072 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4073 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4074 __xfrm_policy_inexact_prune_bin(b, true);
4075 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4078 static int __net_init xfrm_net_init(struct net *net)
4082 /* Initialize the per-net locks here */
4083 spin_lock_init(&net->xfrm.xfrm_state_lock);
4084 spin_lock_init(&net->xfrm.xfrm_policy_lock);
4085 mutex_init(&net->xfrm.xfrm_cfg_mutex);
4087 rv = xfrm_statistics_init(net);
4089 goto out_statistics;
4090 rv = xfrm_state_init(net);
4093 rv = xfrm_policy_init(net);
4096 rv = xfrm_sysctl_init(net);
4103 xfrm_policy_fini(net);
4105 xfrm_state_fini(net);
4107 xfrm_statistics_fini(net);
4112 static void __net_exit xfrm_net_exit(struct net *net)
4114 xfrm_sysctl_fini(net);
4115 xfrm_policy_fini(net);
4116 xfrm_state_fini(net);
4117 xfrm_statistics_fini(net);
4120 static struct pernet_operations __net_initdata xfrm_net_ops = {
4121 .init = xfrm_net_init,
4122 .exit = xfrm_net_exit,
4125 void __init xfrm_init(void)
4127 register_pernet_subsys(&xfrm_net_ops);
4129 seqcount_mutex_init(&xfrm_policy_hash_generation, &hash_resize_mutex);
4132 #ifdef CONFIG_XFRM_ESPINTCP
4137 #ifdef CONFIG_AUDITSYSCALL
4138 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4139 struct audit_buffer *audit_buf)
4141 struct xfrm_sec_ctx *ctx = xp->security;
4142 struct xfrm_selector *sel = &xp->selector;
4145 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4146 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4148 switch (sel->family) {
4150 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4151 if (sel->prefixlen_s != 32)
4152 audit_log_format(audit_buf, " src_prefixlen=%d",
4154 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4155 if (sel->prefixlen_d != 32)
4156 audit_log_format(audit_buf, " dst_prefixlen=%d",
4160 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4161 if (sel->prefixlen_s != 128)
4162 audit_log_format(audit_buf, " src_prefixlen=%d",
4164 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4165 if (sel->prefixlen_d != 128)
4166 audit_log_format(audit_buf, " dst_prefixlen=%d",
4172 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4174 struct audit_buffer *audit_buf;
4176 audit_buf = xfrm_audit_start("SPD-add");
4177 if (audit_buf == NULL)
4179 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4180 audit_log_format(audit_buf, " res=%u", result);
4181 xfrm_audit_common_policyinfo(xp, audit_buf);
4182 audit_log_end(audit_buf);
4184 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4186 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4189 struct audit_buffer *audit_buf;
4191 audit_buf = xfrm_audit_start("SPD-delete");
4192 if (audit_buf == NULL)
4194 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4195 audit_log_format(audit_buf, " res=%u", result);
4196 xfrm_audit_common_policyinfo(xp, audit_buf);
4197 audit_log_end(audit_buf);
4199 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4202 #ifdef CONFIG_XFRM_MIGRATE
4203 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4204 const struct xfrm_selector *sel_tgt)
4206 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4207 if (sel_tgt->family == sel_cmp->family &&
4208 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4210 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4212 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4213 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4217 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4224 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4225 u8 dir, u8 type, struct net *net)
4227 struct xfrm_policy *pol, *ret = NULL;
4228 struct hlist_head *chain;
4231 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4232 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4233 hlist_for_each_entry(pol, chain, bydst) {
4234 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4235 pol->type == type) {
4237 priority = ret->priority;
4241 chain = &net->xfrm.policy_inexact[dir];
4242 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4243 if ((pol->priority >= priority) && ret)
4246 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4247 pol->type == type) {
4255 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4260 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4264 if (t->mode == m->mode && t->id.proto == m->proto &&
4265 (m->reqid == 0 || t->reqid == m->reqid)) {
4267 case XFRM_MODE_TUNNEL:
4268 case XFRM_MODE_BEET:
4269 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4271 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4276 case XFRM_MODE_TRANSPORT:
4277 /* in case of transport mode, template does not store
4278 any IP addresses, hence we just compare mode and
4289 /* update endpoint address(es) of template(s) */
4290 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4291 struct xfrm_migrate *m, int num_migrate)
4293 struct xfrm_migrate *mp;
4296 write_lock_bh(&pol->lock);
4297 if (unlikely(pol->walk.dead)) {
4298 /* target policy has been deleted */
4299 write_unlock_bh(&pol->lock);
4303 for (i = 0; i < pol->xfrm_nr; i++) {
4304 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4305 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4308 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4309 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4311 /* update endpoints */
4312 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4313 sizeof(pol->xfrm_vec[i].id.daddr));
4314 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4315 sizeof(pol->xfrm_vec[i].saddr));
4316 pol->xfrm_vec[i].encap_family = mp->new_family;
4318 atomic_inc(&pol->genid);
4322 write_unlock_bh(&pol->lock);
4330 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4334 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4337 for (i = 0; i < num_migrate; i++) {
4338 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4339 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4342 /* check if there is any duplicated entry */
4343 for (j = i + 1; j < num_migrate; j++) {
4344 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4345 sizeof(m[i].old_daddr)) &&
4346 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4347 sizeof(m[i].old_saddr)) &&
4348 m[i].proto == m[j].proto &&
4349 m[i].mode == m[j].mode &&
4350 m[i].reqid == m[j].reqid &&
4351 m[i].old_family == m[j].old_family)
4359 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4360 struct xfrm_migrate *m, int num_migrate,
4361 struct xfrm_kmaddress *k, struct net *net,
4362 struct xfrm_encap_tmpl *encap)
4364 int i, err, nx_cur = 0, nx_new = 0;
4365 struct xfrm_policy *pol = NULL;
4366 struct xfrm_state *x, *xc;
4367 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4368 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4369 struct xfrm_migrate *mp;
4371 /* Stage 0 - sanity checks */
4372 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4375 if (dir >= XFRM_POLICY_MAX) {
4380 /* Stage 1 - find policy */
4381 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4386 /* Stage 2 - find and update state(s) */
4387 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4388 if ((x = xfrm_migrate_state_find(mp, net))) {
4391 xc = xfrm_state_migrate(x, mp, encap);
4402 /* Stage 3 - update policy */
4403 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4406 /* Stage 4 - delete old state(s) */
4408 xfrm_states_put(x_cur, nx_cur);
4409 xfrm_states_delete(x_cur, nx_cur);
4412 /* Stage 5 - announce */
4413 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4425 xfrm_states_put(x_cur, nx_cur);
4427 xfrm_states_delete(x_new, nx_new);
4431 EXPORT_SYMBOL(xfrm_migrate);