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>
34 #include <net/inet_ecn.h>
38 #if IS_ENABLED(CONFIG_IPV6_MIP6)
41 #ifdef CONFIG_XFRM_STATISTICS
45 #include "xfrm_hash.h"
47 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
48 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
49 #define XFRM_MAX_QUEUE_LEN 100
52 struct dst_entry *dst_orig;
56 /* prefixes smaller than this are stored in lists, not trees. */
57 #define INEXACT_PREFIXLEN_IPV4 16
58 #define INEXACT_PREFIXLEN_IPV6 48
60 struct xfrm_pol_inexact_node {
70 /* the policies matching this node, can be empty list */
71 struct hlist_head hhead;
74 /* xfrm inexact policy search tree:
75 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
77 * +---- root_d: sorted by daddr:prefix
79 * | xfrm_pol_inexact_node
81 * | +- root: sorted by saddr/prefix
83 * | | xfrm_pol_inexact_node
87 * | | + hhead: saddr:daddr policies
89 * | +- coarse policies and all any:daddr policies
91 * +---- root_s: sorted by saddr:prefix
93 * | xfrm_pol_inexact_node
97 * | + hhead: saddr:any policies
99 * +---- coarse policies and all any:any policies
101 * Lookups return four candidate lists:
102 * 1. any:any list from top-level xfrm_pol_inexact_bin
103 * 2. any:daddr list from daddr tree
104 * 3. saddr:daddr list from 2nd level daddr tree
105 * 4. saddr:any list from saddr tree
107 * This result set then needs to be searched for the policy with
108 * the lowest priority. If two results have same prio, youngest one wins.
111 struct xfrm_pol_inexact_key {
118 struct xfrm_pol_inexact_bin {
119 struct xfrm_pol_inexact_key k;
120 struct rhash_head head;
121 /* list containing '*:*' policies */
122 struct hlist_head hhead;
125 /* tree sorted by daddr/prefix */
126 struct rb_root root_d;
128 /* tree sorted by saddr/prefix */
129 struct rb_root root_s;
131 /* slow path below */
132 struct list_head inexact_bins;
136 enum xfrm_pol_inexact_candidate_type {
145 struct xfrm_pol_inexact_candidates {
146 struct hlist_head *res[XFRM_POL_CAND_MAX];
149 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
150 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
152 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
153 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
156 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
157 static __read_mostly seqcount_t xfrm_policy_hash_generation;
159 static struct rhashtable xfrm_policy_inexact_table;
160 static const struct rhashtable_params xfrm_pol_inexact_params;
162 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
163 static int stale_bundle(struct dst_entry *dst);
164 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
165 static void xfrm_policy_queue_process(struct timer_list *t);
167 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
168 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
171 static struct xfrm_pol_inexact_bin *
172 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
175 static struct xfrm_pol_inexact_bin *
176 xfrm_policy_inexact_lookup_rcu(struct net *net,
177 u8 type, u16 family, u8 dir, u32 if_id);
178 static struct xfrm_policy *
179 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
181 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
182 struct xfrm_policy *policy);
185 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
186 struct xfrm_pol_inexact_bin *b,
187 const xfrm_address_t *saddr,
188 const xfrm_address_t *daddr);
190 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
192 return refcount_inc_not_zero(&policy->refcnt);
196 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
198 const struct flowi4 *fl4 = &fl->u.ip4;
200 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
201 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
202 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
203 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
204 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
205 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
209 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
211 const struct flowi6 *fl6 = &fl->u.ip6;
213 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
214 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
215 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
216 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
217 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
218 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
221 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
222 unsigned short family)
226 return __xfrm4_selector_match(sel, fl);
228 return __xfrm6_selector_match(sel, fl);
233 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
235 const struct xfrm_policy_afinfo *afinfo;
237 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
240 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
241 if (unlikely(!afinfo))
246 /* Called with rcu_read_lock(). */
247 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
249 return rcu_dereference(xfrm_if_cb);
252 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
253 const xfrm_address_t *saddr,
254 const xfrm_address_t *daddr,
255 int family, u32 mark)
257 const struct xfrm_policy_afinfo *afinfo;
258 struct dst_entry *dst;
260 afinfo = xfrm_policy_get_afinfo(family);
261 if (unlikely(afinfo == NULL))
262 return ERR_PTR(-EAFNOSUPPORT);
264 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
270 EXPORT_SYMBOL(__xfrm_dst_lookup);
272 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
274 xfrm_address_t *prev_saddr,
275 xfrm_address_t *prev_daddr,
276 int family, u32 mark)
278 struct net *net = xs_net(x);
279 xfrm_address_t *saddr = &x->props.saddr;
280 xfrm_address_t *daddr = &x->id.daddr;
281 struct dst_entry *dst;
283 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
287 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
292 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
295 if (prev_saddr != saddr)
296 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
297 if (prev_daddr != daddr)
298 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
304 static inline unsigned long make_jiffies(long secs)
306 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
307 return MAX_SCHEDULE_TIMEOUT-1;
312 static void xfrm_policy_timer(struct timer_list *t)
314 struct xfrm_policy *xp = from_timer(xp, t, timer);
315 time64_t now = ktime_get_real_seconds();
316 time64_t next = TIME64_MAX;
320 read_lock(&xp->lock);
322 if (unlikely(xp->walk.dead))
325 dir = xfrm_policy_id2dir(xp->index);
327 if (xp->lft.hard_add_expires_seconds) {
328 time64_t tmo = xp->lft.hard_add_expires_seconds +
329 xp->curlft.add_time - now;
335 if (xp->lft.hard_use_expires_seconds) {
336 time64_t tmo = xp->lft.hard_use_expires_seconds +
337 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
343 if (xp->lft.soft_add_expires_seconds) {
344 time64_t tmo = xp->lft.soft_add_expires_seconds +
345 xp->curlft.add_time - now;
348 tmo = XFRM_KM_TIMEOUT;
353 if (xp->lft.soft_use_expires_seconds) {
354 time64_t tmo = xp->lft.soft_use_expires_seconds +
355 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
358 tmo = XFRM_KM_TIMEOUT;
365 km_policy_expired(xp, dir, 0, 0);
366 if (next != TIME64_MAX &&
367 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
371 read_unlock(&xp->lock);
376 read_unlock(&xp->lock);
377 if (!xfrm_policy_delete(xp, dir))
378 km_policy_expired(xp, dir, 1, 0);
382 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
386 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
388 struct xfrm_policy *policy;
390 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
393 write_pnet(&policy->xp_net, net);
394 INIT_LIST_HEAD(&policy->walk.all);
395 INIT_HLIST_NODE(&policy->bydst_inexact_list);
396 INIT_HLIST_NODE(&policy->bydst);
397 INIT_HLIST_NODE(&policy->byidx);
398 rwlock_init(&policy->lock);
399 refcount_set(&policy->refcnt, 1);
400 skb_queue_head_init(&policy->polq.hold_queue);
401 timer_setup(&policy->timer, xfrm_policy_timer, 0);
402 timer_setup(&policy->polq.hold_timer,
403 xfrm_policy_queue_process, 0);
407 EXPORT_SYMBOL(xfrm_policy_alloc);
409 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
411 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
413 security_xfrm_policy_free(policy->security);
417 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
419 void xfrm_policy_destroy(struct xfrm_policy *policy)
421 BUG_ON(!policy->walk.dead);
423 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
426 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
428 EXPORT_SYMBOL(xfrm_policy_destroy);
430 /* Rule must be locked. Release descendant resources, announce
431 * entry dead. The rule must be unlinked from lists to the moment.
434 static void xfrm_policy_kill(struct xfrm_policy *policy)
436 write_lock_bh(&policy->lock);
437 policy->walk.dead = 1;
438 write_unlock_bh(&policy->lock);
440 atomic_inc(&policy->genid);
442 if (del_timer(&policy->polq.hold_timer))
443 xfrm_pol_put(policy);
444 skb_queue_purge(&policy->polq.hold_queue);
446 if (del_timer(&policy->timer))
447 xfrm_pol_put(policy);
449 xfrm_pol_put(policy);
452 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
454 static inline unsigned int idx_hash(struct net *net, u32 index)
456 return __idx_hash(index, net->xfrm.policy_idx_hmask);
459 /* calculate policy hash thresholds */
460 static void __get_hash_thresh(struct net *net,
461 unsigned short family, int dir,
462 u8 *dbits, u8 *sbits)
466 *dbits = net->xfrm.policy_bydst[dir].dbits4;
467 *sbits = net->xfrm.policy_bydst[dir].sbits4;
471 *dbits = net->xfrm.policy_bydst[dir].dbits6;
472 *sbits = net->xfrm.policy_bydst[dir].sbits6;
481 static struct hlist_head *policy_hash_bysel(struct net *net,
482 const struct xfrm_selector *sel,
483 unsigned short family, int dir)
485 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
490 __get_hash_thresh(net, family, dir, &dbits, &sbits);
491 hash = __sel_hash(sel, family, hmask, dbits, sbits);
493 if (hash == hmask + 1)
496 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
497 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
500 static struct hlist_head *policy_hash_direct(struct net *net,
501 const xfrm_address_t *daddr,
502 const xfrm_address_t *saddr,
503 unsigned short family, int dir)
505 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
510 __get_hash_thresh(net, family, dir, &dbits, &sbits);
511 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
513 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
514 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
517 static void xfrm_dst_hash_transfer(struct net *net,
518 struct hlist_head *list,
519 struct hlist_head *ndsttable,
520 unsigned int nhashmask,
523 struct hlist_node *tmp, *entry0 = NULL;
524 struct xfrm_policy *pol;
530 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
533 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
534 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
535 pol->family, nhashmask, dbits, sbits);
537 hlist_del_rcu(&pol->bydst);
538 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
543 hlist_del_rcu(&pol->bydst);
544 hlist_add_behind_rcu(&pol->bydst, entry0);
546 entry0 = &pol->bydst;
548 if (!hlist_empty(list)) {
554 static void xfrm_idx_hash_transfer(struct hlist_head *list,
555 struct hlist_head *nidxtable,
556 unsigned int nhashmask)
558 struct hlist_node *tmp;
559 struct xfrm_policy *pol;
561 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
564 h = __idx_hash(pol->index, nhashmask);
565 hlist_add_head(&pol->byidx, nidxtable+h);
569 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
571 return ((old_hmask + 1) << 1) - 1;
574 static void xfrm_bydst_resize(struct net *net, int dir)
576 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
577 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
578 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
579 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
580 struct hlist_head *odst;
586 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
587 write_seqcount_begin(&xfrm_policy_hash_generation);
589 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
590 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
592 for (i = hmask; i >= 0; i--)
593 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
595 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
596 net->xfrm.policy_bydst[dir].hmask = nhashmask;
598 write_seqcount_end(&xfrm_policy_hash_generation);
599 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
603 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
606 static void xfrm_byidx_resize(struct net *net, int total)
608 unsigned int hmask = net->xfrm.policy_idx_hmask;
609 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
610 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
611 struct hlist_head *oidx = net->xfrm.policy_byidx;
612 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
618 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
620 for (i = hmask; i >= 0; i--)
621 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
623 net->xfrm.policy_byidx = nidx;
624 net->xfrm.policy_idx_hmask = nhashmask;
626 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
628 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
631 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
633 unsigned int cnt = net->xfrm.policy_count[dir];
634 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
639 if ((hmask + 1) < xfrm_policy_hashmax &&
646 static inline int xfrm_byidx_should_resize(struct net *net, int total)
648 unsigned int hmask = net->xfrm.policy_idx_hmask;
650 if ((hmask + 1) < xfrm_policy_hashmax &&
657 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
659 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
660 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
661 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
662 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
663 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
664 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
665 si->spdhcnt = net->xfrm.policy_idx_hmask;
666 si->spdhmcnt = xfrm_policy_hashmax;
668 EXPORT_SYMBOL(xfrm_spd_getinfo);
670 static DEFINE_MUTEX(hash_resize_mutex);
671 static void xfrm_hash_resize(struct work_struct *work)
673 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
676 mutex_lock(&hash_resize_mutex);
679 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
680 if (xfrm_bydst_should_resize(net, dir, &total))
681 xfrm_bydst_resize(net, dir);
683 if (xfrm_byidx_should_resize(net, total))
684 xfrm_byidx_resize(net, total);
686 mutex_unlock(&hash_resize_mutex);
689 /* Make sure *pol can be inserted into fastbin.
690 * Useful to check that later insert requests will be sucessful
691 * (provided xfrm_policy_lock is held throughout).
693 static struct xfrm_pol_inexact_bin *
694 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
696 struct xfrm_pol_inexact_bin *bin, *prev;
697 struct xfrm_pol_inexact_key k = {
698 .family = pol->family,
703 struct net *net = xp_net(pol);
705 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
707 write_pnet(&k.net, net);
708 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
709 xfrm_pol_inexact_params);
713 bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
718 INIT_HLIST_HEAD(&bin->hhead);
719 bin->root_d = RB_ROOT;
720 bin->root_s = RB_ROOT;
721 seqcount_init(&bin->count);
723 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
725 xfrm_pol_inexact_params);
727 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
733 return IS_ERR(prev) ? NULL : prev;
736 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
737 int family, u8 prefixlen)
739 if (xfrm_addr_any(addr, family))
742 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
745 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
752 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
754 const xfrm_address_t *addr;
755 bool saddr_any, daddr_any;
758 addr = &policy->selector.saddr;
759 prefixlen = policy->selector.prefixlen_s;
761 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
764 addr = &policy->selector.daddr;
765 prefixlen = policy->selector.prefixlen_d;
766 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
769 return saddr_any && daddr_any;
772 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
773 const xfrm_address_t *addr, u8 prefixlen)
776 node->prefixlen = prefixlen;
779 static struct xfrm_pol_inexact_node *
780 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
782 struct xfrm_pol_inexact_node *node;
784 node = kzalloc(sizeof(*node), GFP_ATOMIC);
786 xfrm_pol_inexact_node_init(node, addr, prefixlen);
791 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
792 const xfrm_address_t *b,
793 u8 prefixlen, u16 family)
796 unsigned int pdw, pbi;
803 mask = ~0U << (32 - prefixlen);
804 ma = ntohl(a->a4) & mask;
805 mb = ntohl(b->a4) & mask;
812 pdw = prefixlen >> 5;
813 pbi = prefixlen & 0x1f;
816 delta = memcmp(a->a6, b->a6, pdw << 2);
821 mask = ~0U << (32 - pbi);
822 ma = ntohl(a->a6[pdw]) & mask;
823 mb = ntohl(b->a6[pdw]) & mask;
837 static void xfrm_policy_inexact_list_reinsert(struct net *net,
838 struct xfrm_pol_inexact_node *n,
841 unsigned int matched_s, matched_d;
842 struct xfrm_policy *policy, *p;
847 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
848 struct hlist_node *newpos = NULL;
849 bool matches_s, matches_d;
851 if (!policy->bydst_reinsert)
854 WARN_ON_ONCE(policy->family != family);
856 policy->bydst_reinsert = false;
857 hlist_for_each_entry(p, &n->hhead, bydst) {
858 if (policy->priority > p->priority)
860 else if (policy->priority == p->priority &&
861 policy->pos > p->pos)
868 hlist_add_behind_rcu(&policy->bydst, newpos);
870 hlist_add_head_rcu(&policy->bydst, &n->hhead);
872 /* paranoia checks follow.
873 * Check that the reinserted policy matches at least
874 * saddr or daddr for current node prefix.
876 * Matching both is fine, matching saddr in one policy
877 * (but not daddr) and then matching only daddr in another
880 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
884 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
888 if (matches_s && matches_d)
891 WARN_ON_ONCE(!matches_s && !matches_d);
896 WARN_ON_ONCE(matched_s && matched_d);
900 static void xfrm_policy_inexact_node_reinsert(struct net *net,
901 struct xfrm_pol_inexact_node *n,
905 struct xfrm_pol_inexact_node *node;
906 struct rb_node **p, *parent;
908 /* we should not have another subtree here */
909 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
918 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
920 prefixlen = min(node->prefixlen, n->prefixlen);
922 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
925 p = &parent->rb_left;
926 } else if (delta > 0) {
927 p = &parent->rb_right;
929 bool same_prefixlen = node->prefixlen == n->prefixlen;
930 struct xfrm_policy *tmp;
932 hlist_for_each_entry(tmp, &n->hhead, bydst) {
933 tmp->bydst_reinsert = true;
934 hlist_del_rcu(&tmp->bydst);
937 node->prefixlen = prefixlen;
939 xfrm_policy_inexact_list_reinsert(net, node, family);
941 if (same_prefixlen) {
953 rb_link_node_rcu(&n->node, parent, p);
954 rb_insert_color(&n->node, new);
957 /* merge nodes v and n */
958 static void xfrm_policy_inexact_node_merge(struct net *net,
959 struct xfrm_pol_inexact_node *v,
960 struct xfrm_pol_inexact_node *n,
963 struct xfrm_pol_inexact_node *node;
964 struct xfrm_policy *tmp;
965 struct rb_node *rnode;
967 /* To-be-merged node v has a subtree.
969 * Dismantle it and insert its nodes to n->root.
971 while ((rnode = rb_first(&v->root)) != NULL) {
972 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
973 rb_erase(&node->node, &v->root);
974 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
978 hlist_for_each_entry(tmp, &v->hhead, bydst) {
979 tmp->bydst_reinsert = true;
980 hlist_del_rcu(&tmp->bydst);
983 xfrm_policy_inexact_list_reinsert(net, n, family);
986 static struct xfrm_pol_inexact_node *
987 xfrm_policy_inexact_insert_node(struct net *net,
988 struct rb_root *root,
989 xfrm_address_t *addr,
990 u16 family, u8 prefixlen, u8 dir)
992 struct xfrm_pol_inexact_node *cached = NULL;
993 struct rb_node **p, *parent = NULL;
994 struct xfrm_pol_inexact_node *node;
1001 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1003 delta = xfrm_policy_addr_delta(addr, &node->addr,
1006 if (delta == 0 && prefixlen >= node->prefixlen) {
1007 WARN_ON_ONCE(cached); /* ipsec policies got lost */
1012 p = &parent->rb_left;
1014 p = &parent->rb_right;
1016 if (prefixlen < node->prefixlen) {
1017 delta = xfrm_policy_addr_delta(addr, &node->addr,
1023 /* This node is a subnet of the new prefix. It needs
1024 * to be removed and re-inserted with the smaller
1025 * prefix and all nodes that are now also covered
1026 * by the reduced prefixlen.
1028 rb_erase(&node->node, root);
1031 xfrm_pol_inexact_node_init(node, addr,
1035 /* This node also falls within the new
1036 * prefixlen. Merge the to-be-reinserted
1037 * node and this one.
1039 xfrm_policy_inexact_node_merge(net, node,
1041 kfree_rcu(node, rcu);
1052 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1057 rb_link_node_rcu(&node->node, parent, p);
1058 rb_insert_color(&node->node, root);
1063 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1065 struct xfrm_pol_inexact_node *node;
1066 struct rb_node *rn = rb_first(r);
1069 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1071 xfrm_policy_inexact_gc_tree(&node->root, rm);
1074 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1079 rb_erase(&node->node, r);
1080 kfree_rcu(node, rcu);
1084 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1086 write_seqcount_begin(&b->count);
1087 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1088 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1089 write_seqcount_end(&b->count);
1091 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1092 !hlist_empty(&b->hhead)) {
1093 WARN_ON_ONCE(net_exit);
1097 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1098 xfrm_pol_inexact_params) == 0) {
1099 list_del(&b->inexact_bins);
1104 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1106 struct net *net = read_pnet(&b->k.net);
1108 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1109 __xfrm_policy_inexact_prune_bin(b, false);
1110 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1113 static void __xfrm_policy_inexact_flush(struct net *net)
1115 struct xfrm_pol_inexact_bin *bin, *t;
1117 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1119 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1120 __xfrm_policy_inexact_prune_bin(bin, false);
1123 static struct hlist_head *
1124 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1125 struct xfrm_policy *policy, u8 dir)
1127 struct xfrm_pol_inexact_node *n;
1130 net = xp_net(policy);
1131 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1133 if (xfrm_policy_inexact_insert_use_any_list(policy))
1136 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1138 policy->selector.prefixlen_d)) {
1139 write_seqcount_begin(&bin->count);
1140 n = xfrm_policy_inexact_insert_node(net,
1142 &policy->selector.saddr,
1144 policy->selector.prefixlen_s,
1146 write_seqcount_end(&bin->count);
1153 /* daddr is fixed */
1154 write_seqcount_begin(&bin->count);
1155 n = xfrm_policy_inexact_insert_node(net,
1157 &policy->selector.daddr,
1159 policy->selector.prefixlen_d, dir);
1160 write_seqcount_end(&bin->count);
1164 /* saddr is wildcard */
1165 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1167 policy->selector.prefixlen_s))
1170 write_seqcount_begin(&bin->count);
1171 n = xfrm_policy_inexact_insert_node(net,
1173 &policy->selector.saddr,
1175 policy->selector.prefixlen_s, dir);
1176 write_seqcount_end(&bin->count);
1183 static struct xfrm_policy *
1184 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1186 struct xfrm_pol_inexact_bin *bin;
1187 struct xfrm_policy *delpol;
1188 struct hlist_head *chain;
1191 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1193 return ERR_PTR(-ENOMEM);
1195 net = xp_net(policy);
1196 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1198 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1200 __xfrm_policy_inexact_prune_bin(bin, false);
1201 return ERR_PTR(-ENOMEM);
1204 delpol = xfrm_policy_insert_list(chain, policy, excl);
1205 if (delpol && excl) {
1206 __xfrm_policy_inexact_prune_bin(bin, false);
1207 return ERR_PTR(-EEXIST);
1210 chain = &net->xfrm.policy_inexact[dir];
1211 xfrm_policy_insert_inexact_list(chain, policy);
1214 __xfrm_policy_inexact_prune_bin(bin, false);
1219 static void xfrm_hash_rebuild(struct work_struct *work)
1221 struct net *net = container_of(work, struct net,
1222 xfrm.policy_hthresh.work);
1224 struct xfrm_policy *pol;
1225 struct xfrm_policy *policy;
1226 struct hlist_head *chain;
1227 struct hlist_head *odst;
1228 struct hlist_node *newpos;
1232 u8 lbits4, rbits4, lbits6, rbits6;
1234 mutex_lock(&hash_resize_mutex);
1236 /* read selector prefixlen thresholds */
1238 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1240 lbits4 = net->xfrm.policy_hthresh.lbits4;
1241 rbits4 = net->xfrm.policy_hthresh.rbits4;
1242 lbits6 = net->xfrm.policy_hthresh.lbits6;
1243 rbits6 = net->xfrm.policy_hthresh.rbits6;
1244 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1246 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1247 write_seqcount_begin(&xfrm_policy_hash_generation);
1249 /* make sure that we can insert the indirect policies again before
1250 * we start with destructive action.
1252 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1253 struct xfrm_pol_inexact_bin *bin;
1256 dir = xfrm_policy_id2dir(policy->index);
1257 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1260 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1261 if (policy->family == AF_INET) {
1269 if (policy->family == AF_INET) {
1278 if (policy->selector.prefixlen_d < dbits ||
1279 policy->selector.prefixlen_s < sbits)
1282 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1286 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1290 /* reset the bydst and inexact table in all directions */
1291 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1292 struct hlist_node *n;
1294 hlist_for_each_entry_safe(policy, n,
1295 &net->xfrm.policy_inexact[dir],
1296 bydst_inexact_list) {
1297 hlist_del_rcu(&policy->bydst);
1298 hlist_del_init(&policy->bydst_inexact_list);
1301 hmask = net->xfrm.policy_bydst[dir].hmask;
1302 odst = net->xfrm.policy_bydst[dir].table;
1303 for (i = hmask; i >= 0; i--) {
1304 hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1305 hlist_del_rcu(&policy->bydst);
1307 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1308 /* dir out => dst = remote, src = local */
1309 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1310 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1311 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1312 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1314 /* dir in/fwd => dst = local, src = remote */
1315 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1316 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1317 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1318 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1322 /* re-insert all policies by order of creation */
1323 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1324 if (policy->walk.dead)
1326 dir = xfrm_policy_id2dir(policy->index);
1327 if (dir >= XFRM_POLICY_MAX) {
1328 /* skip socket policies */
1332 chain = policy_hash_bysel(net, &policy->selector,
1333 policy->family, dir);
1336 void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1338 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1342 hlist_for_each_entry(pol, chain, bydst) {
1343 if (policy->priority >= pol->priority)
1344 newpos = &pol->bydst;
1349 hlist_add_behind_rcu(&policy->bydst, newpos);
1351 hlist_add_head_rcu(&policy->bydst, chain);
1355 __xfrm_policy_inexact_flush(net);
1356 write_seqcount_end(&xfrm_policy_hash_generation);
1357 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1359 mutex_unlock(&hash_resize_mutex);
1362 void xfrm_policy_hash_rebuild(struct net *net)
1364 schedule_work(&net->xfrm.policy_hthresh.work);
1366 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1368 /* Generate new index... KAME seems to generate them ordered by cost
1369 * of an absolute inpredictability of ordering of rules. This will not pass. */
1370 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1372 static u32 idx_generator;
1375 struct hlist_head *list;
1376 struct xfrm_policy *p;
1381 idx = (idx_generator | dir);
1390 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1392 hlist_for_each_entry(p, list, byidx) {
1393 if (p->index == idx) {
1403 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1405 u32 *p1 = (u32 *) s1;
1406 u32 *p2 = (u32 *) s2;
1407 int len = sizeof(struct xfrm_selector) / sizeof(u32);
1410 for (i = 0; i < len; i++) {
1418 static void xfrm_policy_requeue(struct xfrm_policy *old,
1419 struct xfrm_policy *new)
1421 struct xfrm_policy_queue *pq = &old->polq;
1422 struct sk_buff_head list;
1424 if (skb_queue_empty(&pq->hold_queue))
1427 __skb_queue_head_init(&list);
1429 spin_lock_bh(&pq->hold_queue.lock);
1430 skb_queue_splice_init(&pq->hold_queue, &list);
1431 if (del_timer(&pq->hold_timer))
1433 spin_unlock_bh(&pq->hold_queue.lock);
1437 spin_lock_bh(&pq->hold_queue.lock);
1438 skb_queue_splice(&list, &pq->hold_queue);
1439 pq->timeout = XFRM_QUEUE_TMO_MIN;
1440 if (!mod_timer(&pq->hold_timer, jiffies))
1442 spin_unlock_bh(&pq->hold_queue.lock);
1445 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1446 struct xfrm_policy *pol)
1448 return mark->v == pol->mark.v && mark->m == pol->mark.m;
1451 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1453 const struct xfrm_pol_inexact_key *k = data;
1454 u32 a = k->type << 24 | k->dir << 16 | k->family;
1456 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1460 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1462 const struct xfrm_pol_inexact_bin *b = data;
1464 return xfrm_pol_bin_key(&b->k, 0, seed);
1467 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1470 const struct xfrm_pol_inexact_key *key = arg->key;
1471 const struct xfrm_pol_inexact_bin *b = ptr;
1474 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1477 ret = b->k.dir ^ key->dir;
1481 ret = b->k.type ^ key->type;
1485 ret = b->k.family ^ key->family;
1489 return b->k.if_id ^ key->if_id;
1492 static const struct rhashtable_params xfrm_pol_inexact_params = {
1493 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head),
1494 .hashfn = xfrm_pol_bin_key,
1495 .obj_hashfn = xfrm_pol_bin_obj,
1496 .obj_cmpfn = xfrm_pol_bin_cmp,
1497 .automatic_shrinking = true,
1500 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1501 struct xfrm_policy *policy)
1503 struct xfrm_policy *pol, *delpol = NULL;
1504 struct hlist_node *newpos = NULL;
1507 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1508 if (pol->type == policy->type &&
1509 pol->if_id == policy->if_id &&
1510 !selector_cmp(&pol->selector, &policy->selector) &&
1511 xfrm_policy_mark_match(&policy->mark, pol) &&
1512 xfrm_sec_ctx_match(pol->security, policy->security) &&
1515 if (policy->priority > pol->priority)
1517 } else if (policy->priority >= pol->priority) {
1518 newpos = &pol->bydst_inexact_list;
1526 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1528 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1530 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1536 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1537 struct xfrm_policy *policy,
1540 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1542 hlist_for_each_entry(pol, chain, bydst) {
1543 if (pol->type == policy->type &&
1544 pol->if_id == policy->if_id &&
1545 !selector_cmp(&pol->selector, &policy->selector) &&
1546 xfrm_policy_mark_match(&policy->mark, pol) &&
1547 xfrm_sec_ctx_match(pol->security, policy->security) &&
1550 return ERR_PTR(-EEXIST);
1552 if (policy->priority > pol->priority)
1554 } else if (policy->priority >= pol->priority) {
1563 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1565 hlist_add_head_rcu(&policy->bydst, chain);
1570 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1572 struct net *net = xp_net(policy);
1573 struct xfrm_policy *delpol;
1574 struct hlist_head *chain;
1576 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1577 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1579 delpol = xfrm_policy_insert_list(chain, policy, excl);
1581 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1583 if (IS_ERR(delpol)) {
1584 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1585 return PTR_ERR(delpol);
1588 __xfrm_policy_link(policy, dir);
1590 /* After previous checking, family can either be AF_INET or AF_INET6 */
1591 if (policy->family == AF_INET)
1592 rt_genid_bump_ipv4(net);
1594 rt_genid_bump_ipv6(net);
1597 xfrm_policy_requeue(delpol, policy);
1598 __xfrm_policy_unlink(delpol, dir);
1600 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1601 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1602 policy->curlft.add_time = ktime_get_real_seconds();
1603 policy->curlft.use_time = 0;
1604 if (!mod_timer(&policy->timer, jiffies + HZ))
1605 xfrm_pol_hold(policy);
1606 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1609 xfrm_policy_kill(delpol);
1610 else if (xfrm_bydst_should_resize(net, dir, NULL))
1611 schedule_work(&net->xfrm.policy_hash_work);
1615 EXPORT_SYMBOL(xfrm_policy_insert);
1617 static struct xfrm_policy *
1618 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1619 u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1620 struct xfrm_sec_ctx *ctx)
1622 struct xfrm_policy *pol;
1627 hlist_for_each_entry(pol, chain, bydst) {
1628 if (pol->type == type &&
1629 pol->if_id == if_id &&
1630 xfrm_policy_mark_match(mark, pol) &&
1631 !selector_cmp(sel, &pol->selector) &&
1632 xfrm_sec_ctx_match(ctx, pol->security))
1639 struct xfrm_policy *
1640 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1641 u8 type, int dir, struct xfrm_selector *sel,
1642 struct xfrm_sec_ctx *ctx, int delete, int *err)
1644 struct xfrm_pol_inexact_bin *bin = NULL;
1645 struct xfrm_policy *pol, *ret = NULL;
1646 struct hlist_head *chain;
1649 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1650 chain = policy_hash_bysel(net, sel, sel->family, dir);
1652 struct xfrm_pol_inexact_candidates cand;
1655 bin = xfrm_policy_inexact_lookup(net, type,
1656 sel->family, dir, if_id);
1658 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1662 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1665 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1670 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1671 struct xfrm_policy *tmp;
1673 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1679 if (!pol || tmp->pos < pol->pos)
1683 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1690 *err = security_xfrm_policy_delete(pol->security);
1692 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1695 __xfrm_policy_unlink(pol, dir);
1699 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1702 xfrm_policy_kill(ret);
1704 xfrm_policy_inexact_prune_bin(bin);
1707 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1709 struct xfrm_policy *
1710 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1711 u8 type, int dir, u32 id, int delete, int *err)
1713 struct xfrm_policy *pol, *ret;
1714 struct hlist_head *chain;
1717 if (xfrm_policy_id2dir(id) != dir)
1721 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1722 chain = net->xfrm.policy_byidx + idx_hash(net, id);
1724 hlist_for_each_entry(pol, chain, byidx) {
1725 if (pol->type == type && pol->index == id &&
1726 pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1729 *err = security_xfrm_policy_delete(
1732 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1735 __xfrm_policy_unlink(pol, dir);
1741 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1744 xfrm_policy_kill(ret);
1747 EXPORT_SYMBOL(xfrm_policy_byid);
1749 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1751 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1753 struct xfrm_policy *pol;
1756 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1757 if (pol->walk.dead ||
1758 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1762 err = security_xfrm_policy_delete(pol->security);
1764 xfrm_audit_policy_delete(pol, 0, task_valid);
1772 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1778 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1780 int dir, err = 0, cnt = 0;
1781 struct xfrm_policy *pol;
1783 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1785 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1790 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1791 dir = xfrm_policy_id2dir(pol->index);
1792 if (pol->walk.dead ||
1793 dir >= XFRM_POLICY_MAX ||
1797 __xfrm_policy_unlink(pol, dir);
1798 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1800 xfrm_audit_policy_delete(pol, 1, task_valid);
1801 xfrm_policy_kill(pol);
1802 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1806 __xfrm_policy_inexact_flush(net);
1810 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1813 EXPORT_SYMBOL(xfrm_policy_flush);
1815 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1816 int (*func)(struct xfrm_policy *, int, int, void*),
1819 struct xfrm_policy *pol;
1820 struct xfrm_policy_walk_entry *x;
1823 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1824 walk->type != XFRM_POLICY_TYPE_ANY)
1827 if (list_empty(&walk->walk.all) && walk->seq != 0)
1830 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1831 if (list_empty(&walk->walk.all))
1832 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1834 x = list_first_entry(&walk->walk.all,
1835 struct xfrm_policy_walk_entry, all);
1837 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1840 pol = container_of(x, struct xfrm_policy, walk);
1841 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1842 walk->type != pol->type)
1844 error = func(pol, xfrm_policy_id2dir(pol->index),
1847 list_move_tail(&walk->walk.all, &x->all);
1852 if (walk->seq == 0) {
1856 list_del_init(&walk->walk.all);
1858 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1861 EXPORT_SYMBOL(xfrm_policy_walk);
1863 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1865 INIT_LIST_HEAD(&walk->walk.all);
1866 walk->walk.dead = 1;
1870 EXPORT_SYMBOL(xfrm_policy_walk_init);
1872 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1874 if (list_empty(&walk->walk.all))
1877 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1878 list_del(&walk->walk.all);
1879 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1881 EXPORT_SYMBOL(xfrm_policy_walk_done);
1884 * Find policy to apply to this flow.
1886 * Returns 0 if policy found, else an -errno.
1888 static int xfrm_policy_match(const struct xfrm_policy *pol,
1889 const struct flowi *fl,
1890 u8 type, u16 family, int dir, u32 if_id)
1892 const struct xfrm_selector *sel = &pol->selector;
1896 if (pol->family != family ||
1897 pol->if_id != if_id ||
1898 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1902 match = xfrm_selector_match(sel, fl, family);
1904 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,
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,
2186 if (!xfrm_pol_hold_rcu(pol))
2188 } else if (err == -ESRCH) {
2201 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2203 struct net *net = xp_net(pol);
2205 list_add(&pol->walk.all, &net->xfrm.policy_all);
2206 net->xfrm.policy_count[dir]++;
2210 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2213 struct net *net = xp_net(pol);
2215 if (list_empty(&pol->walk.all))
2218 /* Socket policies are not hashed. */
2219 if (!hlist_unhashed(&pol->bydst)) {
2220 hlist_del_rcu(&pol->bydst);
2221 hlist_del_init(&pol->bydst_inexact_list);
2222 hlist_del(&pol->byidx);
2225 list_del_init(&pol->walk.all);
2226 net->xfrm.policy_count[dir]--;
2231 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2233 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2236 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2238 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2241 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2243 struct net *net = xp_net(pol);
2245 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2246 pol = __xfrm_policy_unlink(pol, dir);
2247 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2249 xfrm_policy_kill(pol);
2254 EXPORT_SYMBOL(xfrm_policy_delete);
2256 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2258 struct net *net = sock_net(sk);
2259 struct xfrm_policy *old_pol;
2261 #ifdef CONFIG_XFRM_SUB_POLICY
2262 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2266 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2267 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2268 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2270 pol->curlft.add_time = ktime_get_real_seconds();
2271 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2272 xfrm_sk_policy_link(pol, dir);
2274 rcu_assign_pointer(sk->sk_policy[dir], pol);
2277 xfrm_policy_requeue(old_pol, pol);
2279 /* Unlinking succeeds always. This is the only function
2280 * allowed to delete or replace socket policy.
2282 xfrm_sk_policy_unlink(old_pol, dir);
2284 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2287 xfrm_policy_kill(old_pol);
2292 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2294 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2295 struct net *net = xp_net(old);
2298 newp->selector = old->selector;
2299 if (security_xfrm_policy_clone(old->security,
2302 return NULL; /* ENOMEM */
2304 newp->lft = old->lft;
2305 newp->curlft = old->curlft;
2306 newp->mark = old->mark;
2307 newp->if_id = old->if_id;
2308 newp->action = old->action;
2309 newp->flags = old->flags;
2310 newp->xfrm_nr = old->xfrm_nr;
2311 newp->index = old->index;
2312 newp->type = old->type;
2313 newp->family = old->family;
2314 memcpy(newp->xfrm_vec, old->xfrm_vec,
2315 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2316 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2317 xfrm_sk_policy_link(newp, dir);
2318 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2324 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2326 const struct xfrm_policy *p;
2327 struct xfrm_policy *np;
2331 for (i = 0; i < 2; i++) {
2332 p = rcu_dereference(osk->sk_policy[i]);
2334 np = clone_policy(p, i);
2335 if (unlikely(!np)) {
2339 rcu_assign_pointer(sk->sk_policy[i], np);
2347 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2348 xfrm_address_t *remote, unsigned short family, u32 mark)
2351 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2353 if (unlikely(afinfo == NULL))
2355 err = afinfo->get_saddr(net, oif, local, remote, mark);
2360 /* Resolve list of templates for the flow, given policy. */
2363 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2364 struct xfrm_state **xfrm, unsigned short family)
2366 struct net *net = xp_net(policy);
2369 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2370 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2373 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2374 struct xfrm_state *x;
2375 xfrm_address_t *remote = daddr;
2376 xfrm_address_t *local = saddr;
2377 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2379 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2380 tmpl->mode == XFRM_MODE_BEET) {
2381 remote = &tmpl->id.daddr;
2382 local = &tmpl->saddr;
2383 if (xfrm_addr_any(local, tmpl->encap_family)) {
2384 error = xfrm_get_saddr(net, fl->flowi_oif,
2386 tmpl->encap_family, 0);
2393 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2394 family, policy->if_id);
2396 if (x && x->km.state == XFRM_STATE_VALID) {
2403 error = (x->km.state == XFRM_STATE_ERROR ?
2406 } else if (error == -ESRCH) {
2410 if (!tmpl->optional)
2416 for (nx--; nx >= 0; nx--)
2417 xfrm_state_put(xfrm[nx]);
2422 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2423 struct xfrm_state **xfrm, unsigned short family)
2425 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2426 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2432 for (i = 0; i < npols; i++) {
2433 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2438 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2446 /* found states are sorted for outbound processing */
2448 xfrm_state_sort(xfrm, tpp, cnx, family);
2453 for (cnx--; cnx >= 0; cnx--)
2454 xfrm_state_put(tpp[cnx]);
2459 static int xfrm_get_tos(const struct flowi *fl, int family)
2461 if (family == AF_INET)
2462 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2467 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2469 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2470 struct dst_ops *dst_ops;
2471 struct xfrm_dst *xdst;
2474 return ERR_PTR(-EINVAL);
2478 dst_ops = &net->xfrm.xfrm4_dst_ops;
2480 #if IS_ENABLED(CONFIG_IPV6)
2482 dst_ops = &net->xfrm.xfrm6_dst_ops;
2488 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2491 struct dst_entry *dst = &xdst->u.dst;
2493 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2495 xdst = ERR_PTR(-ENOBUFS);
2502 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2505 if (dst->ops->family == AF_INET6) {
2506 struct rt6_info *rt = (struct rt6_info *)dst;
2507 path->path_cookie = rt6_get_cookie(rt);
2508 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2512 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2513 const struct flowi *fl)
2515 const struct xfrm_policy_afinfo *afinfo =
2516 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2522 err = afinfo->fill_dst(xdst, dev, fl);
2530 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2531 * all the metrics... Shortly, bundle a bundle.
2534 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2535 struct xfrm_state **xfrm,
2536 struct xfrm_dst **bundle,
2538 const struct flowi *fl,
2539 struct dst_entry *dst)
2541 const struct xfrm_state_afinfo *afinfo;
2542 const struct xfrm_mode *inner_mode;
2543 struct net *net = xp_net(policy);
2544 unsigned long now = jiffies;
2545 struct net_device *dev;
2546 struct xfrm_dst *xdst_prev = NULL;
2547 struct xfrm_dst *xdst0 = NULL;
2551 int nfheader_len = 0;
2552 int trailer_len = 0;
2554 int family = policy->selector.family;
2555 xfrm_address_t saddr, daddr;
2557 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2559 tos = xfrm_get_tos(fl, family);
2563 for (; i < nx; i++) {
2564 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2565 struct dst_entry *dst1 = &xdst->u.dst;
2567 err = PTR_ERR(xdst);
2577 /* Ref count is taken during xfrm_alloc_dst()
2578 * No need to do dst_clone() on dst1
2580 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2582 if (xfrm[i]->sel.family == AF_UNSPEC) {
2583 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2584 xfrm_af2proto(family));
2586 err = -EAFNOSUPPORT;
2591 inner_mode = &xfrm[i]->inner_mode;
2594 dst_copy_metrics(dst1, dst);
2596 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2599 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2600 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2602 family = xfrm[i]->props.family;
2603 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2604 &saddr, &daddr, family, mark);
2611 dst1->xfrm = xfrm[i];
2612 xdst->xfrm_genid = xfrm[i]->genid;
2614 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2615 dst1->flags |= DST_HOST;
2616 dst1->lastuse = now;
2618 dst1->input = dst_discard;
2621 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2623 dst1->output = afinfo->output;
2625 dst1->output = dst_discard_out;
2630 header_len += xfrm[i]->props.header_len;
2631 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2632 nfheader_len += xfrm[i]->props.header_len;
2633 trailer_len += xfrm[i]->props.trailer_len;
2636 xfrm_dst_set_child(xdst_prev, dst);
2644 xfrm_init_path(xdst0, dst, nfheader_len);
2645 xfrm_init_pmtu(bundle, nx);
2647 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2648 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2649 err = xfrm_fill_dst(xdst_prev, dev, fl);
2653 xdst_prev->u.dst.header_len = header_len;
2654 xdst_prev->u.dst.trailer_len = trailer_len;
2655 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2656 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2659 return &xdst0->u.dst;
2663 xfrm_state_put(xfrm[i]);
2666 dst_release_immediate(&xdst0->u.dst);
2668 return ERR_PTR(err);
2671 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2672 struct xfrm_policy **pols,
2673 int *num_pols, int *num_xfrms)
2677 if (*num_pols == 0 || !pols[0]) {
2682 if (IS_ERR(pols[0])) {
2684 return PTR_ERR(pols[0]);
2687 *num_xfrms = pols[0]->xfrm_nr;
2689 #ifdef CONFIG_XFRM_SUB_POLICY
2690 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2691 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2692 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2693 XFRM_POLICY_TYPE_MAIN,
2698 if (IS_ERR(pols[1])) {
2699 xfrm_pols_put(pols, *num_pols);
2701 return PTR_ERR(pols[1]);
2704 (*num_xfrms) += pols[1]->xfrm_nr;
2708 for (i = 0; i < *num_pols; i++) {
2709 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2719 static struct xfrm_dst *
2720 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2721 const struct flowi *fl, u16 family,
2722 struct dst_entry *dst_orig)
2724 struct net *net = xp_net(pols[0]);
2725 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2726 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2727 struct xfrm_dst *xdst;
2728 struct dst_entry *dst;
2731 /* Try to instantiate a bundle */
2732 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2738 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2739 return ERR_PTR(err);
2742 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2744 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2745 return ERR_CAST(dst);
2748 xdst = (struct xfrm_dst *)dst;
2749 xdst->num_xfrms = err;
2750 xdst->num_pols = num_pols;
2751 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2752 xdst->policy_genid = atomic_read(&pols[0]->genid);
2757 static void xfrm_policy_queue_process(struct timer_list *t)
2759 struct sk_buff *skb;
2761 struct dst_entry *dst;
2762 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2763 struct net *net = xp_net(pol);
2764 struct xfrm_policy_queue *pq = &pol->polq;
2766 struct sk_buff_head list;
2768 spin_lock(&pq->hold_queue.lock);
2769 skb = skb_peek(&pq->hold_queue);
2771 spin_unlock(&pq->hold_queue.lock);
2776 xfrm_decode_session(skb, &fl, dst->ops->family);
2777 spin_unlock(&pq->hold_queue.lock);
2779 dst_hold(xfrm_dst_path(dst));
2780 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2784 if (dst->flags & DST_XFRM_QUEUE) {
2787 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2790 pq->timeout = pq->timeout << 1;
2791 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2798 __skb_queue_head_init(&list);
2800 spin_lock(&pq->hold_queue.lock);
2802 skb_queue_splice_init(&pq->hold_queue, &list);
2803 spin_unlock(&pq->hold_queue.lock);
2805 while (!skb_queue_empty(&list)) {
2806 skb = __skb_dequeue(&list);
2808 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2809 dst_hold(xfrm_dst_path(skb_dst(skb)));
2810 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2818 skb_dst_set(skb, dst);
2820 dst_output(net, skb->sk, skb);
2829 skb_queue_purge(&pq->hold_queue);
2833 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2835 unsigned long sched_next;
2836 struct dst_entry *dst = skb_dst(skb);
2837 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2838 struct xfrm_policy *pol = xdst->pols[0];
2839 struct xfrm_policy_queue *pq = &pol->polq;
2841 if (unlikely(skb_fclone_busy(sk, skb))) {
2846 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2853 spin_lock_bh(&pq->hold_queue.lock);
2856 pq->timeout = XFRM_QUEUE_TMO_MIN;
2858 sched_next = jiffies + pq->timeout;
2860 if (del_timer(&pq->hold_timer)) {
2861 if (time_before(pq->hold_timer.expires, sched_next))
2862 sched_next = pq->hold_timer.expires;
2866 __skb_queue_tail(&pq->hold_queue, skb);
2867 if (!mod_timer(&pq->hold_timer, sched_next))
2870 spin_unlock_bh(&pq->hold_queue.lock);
2875 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2876 struct xfrm_flo *xflo,
2877 const struct flowi *fl,
2882 struct net_device *dev;
2883 struct dst_entry *dst;
2884 struct dst_entry *dst1;
2885 struct xfrm_dst *xdst;
2887 xdst = xfrm_alloc_dst(net, family);
2891 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2892 net->xfrm.sysctl_larval_drop ||
2896 dst = xflo->dst_orig;
2897 dst1 = &xdst->u.dst;
2901 dst_copy_metrics(dst1, dst);
2903 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2904 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2905 dst1->lastuse = jiffies;
2907 dst1->input = dst_discard;
2908 dst1->output = xdst_queue_output;
2911 xfrm_dst_set_child(xdst, dst);
2914 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2921 err = xfrm_fill_dst(xdst, dev, fl);
2930 xdst = ERR_PTR(err);
2934 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2935 const struct flowi *fl,
2937 struct xfrm_flo *xflo, u32 if_id)
2939 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2940 int num_pols = 0, num_xfrms = 0, err;
2941 struct xfrm_dst *xdst;
2943 /* Resolve policies to use if we couldn't get them from
2944 * previous cache entry */
2946 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2947 err = xfrm_expand_policies(fl, family, pols,
2948 &num_pols, &num_xfrms);
2954 goto make_dummy_bundle;
2956 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2959 err = PTR_ERR(xdst);
2960 if (err == -EREMOTE) {
2961 xfrm_pols_put(pols, num_pols);
2967 goto make_dummy_bundle;
2968 } else if (xdst == NULL) {
2970 goto make_dummy_bundle;
2976 /* We found policies, but there's no bundles to instantiate:
2977 * either because the policy blocks, has no transformations or
2978 * we could not build template (no xfrm_states).*/
2979 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2981 xfrm_pols_put(pols, num_pols);
2982 return ERR_CAST(xdst);
2984 xdst->num_pols = num_pols;
2985 xdst->num_xfrms = num_xfrms;
2986 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2991 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2993 xfrm_pols_put(pols, num_pols);
2994 return ERR_PTR(err);
2997 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2998 struct dst_entry *dst_orig)
3000 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3001 struct dst_entry *ret;
3004 dst_release(dst_orig);
3005 return ERR_PTR(-EINVAL);
3007 ret = afinfo->blackhole_route(net, dst_orig);
3014 /* Finds/creates a bundle for given flow and if_id
3016 * At the moment we eat a raw IP route. Mostly to speed up lookups
3017 * on interfaces with disabled IPsec.
3019 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3022 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3023 struct dst_entry *dst_orig,
3024 const struct flowi *fl,
3025 const struct sock *sk,
3026 int flags, u32 if_id)
3028 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3029 struct xfrm_dst *xdst;
3030 struct dst_entry *dst, *route;
3031 u16 family = dst_orig->ops->family;
3032 u8 dir = XFRM_POLICY_OUT;
3033 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3039 sk = sk_const_to_full_sk(sk);
3040 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3042 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3044 err = xfrm_expand_policies(fl, family, pols,
3045 &num_pols, &num_xfrms);
3050 if (num_xfrms <= 0) {
3051 drop_pols = num_pols;
3055 xdst = xfrm_resolve_and_create_bundle(
3060 xfrm_pols_put(pols, num_pols);
3061 err = PTR_ERR(xdst);
3062 if (err == -EREMOTE)
3066 } else if (xdst == NULL) {
3068 drop_pols = num_pols;
3072 route = xdst->route;
3077 struct xfrm_flo xflo;
3079 xflo.dst_orig = dst_orig;
3082 /* To accelerate a bit... */
3083 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3084 !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3087 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3091 err = PTR_ERR(xdst);
3095 num_pols = xdst->num_pols;
3096 num_xfrms = xdst->num_xfrms;
3097 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3098 route = xdst->route;
3102 if (route == NULL && num_xfrms > 0) {
3103 /* The only case when xfrm_bundle_lookup() returns a
3104 * bundle with null route, is when the template could
3105 * not be resolved. It means policies are there, but
3106 * bundle could not be created, since we don't yet
3107 * have the xfrm_state's. We need to wait for KM to
3108 * negotiate new SA's or bail out with error.*/
3109 if (net->xfrm.sysctl_larval_drop) {
3110 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3117 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3125 if ((flags & XFRM_LOOKUP_ICMP) &&
3126 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3131 for (i = 0; i < num_pols; i++)
3132 pols[i]->curlft.use_time = ktime_get_real_seconds();
3134 if (num_xfrms < 0) {
3135 /* Prohibit the flow */
3136 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3139 } else if (num_xfrms > 0) {
3140 /* Flow transformed */
3141 dst_release(dst_orig);
3143 /* Flow passes untransformed */
3148 xfrm_pols_put(pols, drop_pols);
3149 if (dst && dst->xfrm &&
3150 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3151 dst->flags |= DST_XFRM_TUNNEL;
3155 if (!(flags & XFRM_LOOKUP_ICMP)) {
3163 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3164 dst_release(dst_orig);
3165 xfrm_pols_put(pols, drop_pols);
3166 return ERR_PTR(err);
3168 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3170 /* Main function: finds/creates a bundle for given flow.
3172 * At the moment we eat a raw IP route. Mostly to speed up lookups
3173 * on interfaces with disabled IPsec.
3175 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3176 const struct flowi *fl, const struct sock *sk,
3179 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3181 EXPORT_SYMBOL(xfrm_lookup);
3183 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3184 * Otherwise we may send out blackholed packets.
3186 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3187 const struct flowi *fl,
3188 const struct sock *sk, int flags)
3190 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3191 flags | XFRM_LOOKUP_QUEUE |
3192 XFRM_LOOKUP_KEEP_DST_REF);
3194 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
3195 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3198 dst_release(dst_orig);
3202 EXPORT_SYMBOL(xfrm_lookup_route);
3205 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3207 struct sec_path *sp = skb_sec_path(skb);
3208 struct xfrm_state *x;
3210 if (!sp || idx < 0 || idx >= sp->len)
3213 if (!x->type->reject)
3215 return x->type->reject(x, skb, fl);
3218 /* When skb is transformed back to its "native" form, we have to
3219 * check policy restrictions. At the moment we make this in maximally
3220 * stupid way. Shame on me. :-) Of course, connected sockets must
3221 * have policy cached at them.
3225 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3226 unsigned short family, u32 if_id)
3228 if (xfrm_state_kern(x))
3229 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3230 return x->id.proto == tmpl->id.proto &&
3231 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3232 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3233 x->props.mode == tmpl->mode &&
3234 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3235 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3236 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3237 xfrm_state_addr_cmp(tmpl, x, family)) &&
3238 (if_id == 0 || if_id == x->if_id);
3242 * 0 or more than 0 is returned when validation is succeeded (either bypass
3243 * because of optional transport mode, or next index of the mathced secpath
3244 * state with the template.
3245 * -1 is returned when no matching template is found.
3246 * Otherwise "-2 - errored_index" is returned.
3249 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3250 unsigned short family, u32 if_id)
3254 if (tmpl->optional) {
3255 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3259 for (; idx < sp->len; idx++) {
3260 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3262 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3272 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3274 const struct iphdr *iph = ip_hdr(skb);
3276 u8 *xprth = skb_network_header(skb) + ihl * 4;
3277 struct flowi4 *fl4 = &fl->u.ip4;
3280 if (skb_dst(skb) && skb_dst(skb)->dev)
3281 oif = skb_dst(skb)->dev->ifindex;
3283 memset(fl4, 0, sizeof(struct flowi4));
3284 fl4->flowi4_mark = skb->mark;
3285 fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3287 fl4->flowi4_proto = iph->protocol;
3288 fl4->daddr = reverse ? iph->saddr : iph->daddr;
3289 fl4->saddr = reverse ? iph->daddr : iph->saddr;
3290 fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3292 if (!ip_is_fragment(iph)) {
3293 switch (iph->protocol) {
3295 case IPPROTO_UDPLITE:
3299 if (xprth + 4 < skb->data ||
3300 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3303 xprth = skb_network_header(skb) + ihl * 4;
3304 ports = (__be16 *)xprth;
3306 fl4->fl4_sport = ports[!!reverse];
3307 fl4->fl4_dport = ports[!reverse];
3311 if (xprth + 2 < skb->data ||
3312 pskb_may_pull(skb, xprth + 2 - skb->data)) {
3315 xprth = skb_network_header(skb) + ihl * 4;
3318 fl4->fl4_icmp_type = icmp[0];
3319 fl4->fl4_icmp_code = icmp[1];
3323 if (xprth + 4 < skb->data ||
3324 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3327 xprth = skb_network_header(skb) + ihl * 4;
3328 ehdr = (__be32 *)xprth;
3330 fl4->fl4_ipsec_spi = ehdr[0];
3334 if (xprth + 8 < skb->data ||
3335 pskb_may_pull(skb, xprth + 8 - skb->data)) {
3338 xprth = skb_network_header(skb) + ihl * 4;
3339 ah_hdr = (__be32 *)xprth;
3341 fl4->fl4_ipsec_spi = ah_hdr[1];
3345 if (xprth + 4 < skb->data ||
3346 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3349 xprth = skb_network_header(skb) + ihl * 4;
3350 ipcomp_hdr = (__be16 *)xprth;
3352 fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
3356 if (xprth + 12 < skb->data ||
3357 pskb_may_pull(skb, xprth + 12 - skb->data)) {
3361 xprth = skb_network_header(skb) + ihl * 4;
3362 greflags = (__be16 *)xprth;
3363 gre_hdr = (__be32 *)xprth;
3365 if (greflags[0] & GRE_KEY) {
3366 if (greflags[0] & GRE_CSUM)
3368 fl4->fl4_gre_key = gre_hdr[1];
3373 fl4->fl4_ipsec_spi = 0;
3379 #if IS_ENABLED(CONFIG_IPV6)
3381 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3383 struct flowi6 *fl6 = &fl->u.ip6;
3385 const struct ipv6hdr *hdr = ipv6_hdr(skb);
3386 u32 offset = sizeof(*hdr);
3387 struct ipv6_opt_hdr *exthdr;
3388 const unsigned char *nh = skb_network_header(skb);
3389 u16 nhoff = IP6CB(skb)->nhoff;
3394 nhoff = offsetof(struct ipv6hdr, nexthdr);
3396 nexthdr = nh[nhoff];
3398 if (skb_dst(skb) && skb_dst(skb)->dev)
3399 oif = skb_dst(skb)->dev->ifindex;
3401 memset(fl6, 0, sizeof(struct flowi6));
3402 fl6->flowi6_mark = skb->mark;
3403 fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3405 fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3406 fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3408 while (nh + offset + sizeof(*exthdr) < skb->data ||
3409 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3410 nh = skb_network_header(skb);
3411 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3414 case NEXTHDR_FRAGMENT:
3417 case NEXTHDR_ROUTING:
3420 offset += ipv6_optlen(exthdr);
3421 nexthdr = exthdr->nexthdr;
3422 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3425 case IPPROTO_UDPLITE:
3429 if (!onlyproto && (nh + offset + 4 < skb->data ||
3430 pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3433 nh = skb_network_header(skb);
3434 ports = (__be16 *)(nh + offset);
3435 fl6->fl6_sport = ports[!!reverse];
3436 fl6->fl6_dport = ports[!reverse];
3438 fl6->flowi6_proto = nexthdr;
3440 case IPPROTO_ICMPV6:
3441 if (!onlyproto && (nh + offset + 2 < skb->data ||
3442 pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3445 nh = skb_network_header(skb);
3446 icmp = (u8 *)(nh + offset);
3447 fl6->fl6_icmp_type = icmp[0];
3448 fl6->fl6_icmp_code = icmp[1];
3450 fl6->flowi6_proto = nexthdr;
3454 (nh + offset + 12 < skb->data ||
3455 pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3456 struct gre_base_hdr *gre_hdr;
3459 nh = skb_network_header(skb);
3460 gre_hdr = (struct gre_base_hdr *)(nh + offset);
3461 gre_key = (__be32 *)(gre_hdr + 1);
3463 if (gre_hdr->flags & GRE_KEY) {
3464 if (gre_hdr->flags & GRE_CSUM)
3466 fl6->fl6_gre_key = *gre_key;
3469 fl6->flowi6_proto = nexthdr;
3472 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3474 offset += ipv6_optlen(exthdr);
3475 if (!onlyproto && (nh + offset + 3 < skb->data ||
3476 pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3479 nh = skb_network_header(skb);
3480 mh = (struct ip6_mh *)(nh + offset);
3481 fl6->fl6_mh_type = mh->ip6mh_type;
3483 fl6->flowi6_proto = nexthdr;
3486 /* XXX Why are there these headers? */
3491 fl6->fl6_ipsec_spi = 0;
3492 fl6->flowi6_proto = nexthdr;
3499 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3500 unsigned int family, int reverse)
3504 decode_session4(skb, fl, reverse);
3506 #if IS_ENABLED(CONFIG_IPV6)
3508 decode_session6(skb, fl, reverse);
3512 return -EAFNOSUPPORT;
3515 return security_xfrm_decode_session(skb, &fl->flowi_secid);
3517 EXPORT_SYMBOL(__xfrm_decode_session);
3519 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3521 for (; k < sp->len; k++) {
3522 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3531 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3532 unsigned short family)
3534 struct net *net = dev_net(skb->dev);
3535 struct xfrm_policy *pol;
3536 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3543 const struct xfrm_if_cb *ifcb;
3544 struct sec_path *sp;
3549 ifcb = xfrm_if_get_cb();
3552 xi = ifcb->decode_session(skb, family);
3554 if_id = xi->p.if_id;
3560 reverse = dir & ~XFRM_POLICY_MASK;
3561 dir &= XFRM_POLICY_MASK;
3563 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3564 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3568 nf_nat_decode_session(skb, &fl, family);
3570 /* First, check used SA against their selectors. */
3571 sp = skb_sec_path(skb);
3575 for (i = sp->len - 1; i >= 0; i--) {
3576 struct xfrm_state *x = sp->xvec[i];
3577 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3578 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3585 sk = sk_to_full_sk(sk);
3586 if (sk && sk->sk_policy[dir]) {
3587 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3589 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3595 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3598 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3603 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3604 xfrm_secpath_reject(xerr_idx, skb, &fl);
3605 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3611 pol->curlft.use_time = ktime_get_real_seconds();
3615 #ifdef CONFIG_XFRM_SUB_POLICY
3616 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3617 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3619 XFRM_POLICY_IN, if_id);
3621 if (IS_ERR(pols[1])) {
3622 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3623 xfrm_pol_put(pols[0]);
3626 pols[1]->curlft.use_time = ktime_get_real_seconds();
3632 if (pol->action == XFRM_POLICY_ALLOW) {
3633 static struct sec_path dummy;
3634 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3635 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3636 struct xfrm_tmpl **tpp = tp;
3640 sp = skb_sec_path(skb);
3644 for (pi = 0; pi < npols; pi++) {
3645 if (pols[pi] != pol &&
3646 pols[pi]->action != XFRM_POLICY_ALLOW) {
3647 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3650 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3651 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3654 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3655 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3659 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3663 /* For each tunnel xfrm, find the first matching tmpl.
3664 * For each tmpl before that, find corresponding xfrm.
3665 * Order is _important_. Later we will implement
3666 * some barriers, but at the moment barriers
3667 * are implied between each two transformations.
3669 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3670 k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3673 /* "-2 - errored_index" returned */
3675 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3680 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3681 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3685 xfrm_pols_put(pols, npols);
3688 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3691 xfrm_secpath_reject(xerr_idx, skb, &fl);
3693 xfrm_pols_put(pols, npols);
3696 EXPORT_SYMBOL(__xfrm_policy_check);
3698 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3700 struct net *net = dev_net(skb->dev);
3702 struct dst_entry *dst;
3705 if (xfrm_decode_session(skb, &fl, family) < 0) {
3706 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3711 if (!skb_dst(skb)) {
3712 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3716 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3721 skb_dst_set(skb, dst);
3724 EXPORT_SYMBOL(__xfrm_route_forward);
3726 /* Optimize later using cookies and generation ids. */
3728 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3730 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3731 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3732 * get validated by dst_ops->check on every use. We do this
3733 * because when a normal route referenced by an XFRM dst is
3734 * obsoleted we do not go looking around for all parent
3735 * referencing XFRM dsts so that we can invalidate them. It
3736 * is just too much work. Instead we make the checks here on
3737 * every use. For example:
3739 * XFRM dst A --> IPv4 dst X
3741 * X is the "xdst->route" of A (X is also the "dst->path" of A
3742 * in this example). If X is marked obsolete, "A" will not
3743 * notice. That's what we are validating here via the
3744 * stale_bundle() check.
3746 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3748 * This will force stale_bundle() to fail on any xdst bundle with
3749 * this dst linked in it.
3751 if (dst->obsolete < 0 && !stale_bundle(dst))
3757 static int stale_bundle(struct dst_entry *dst)
3759 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3762 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3764 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3765 dst->dev = dev_net(dev)->loopback_dev;
3770 EXPORT_SYMBOL(xfrm_dst_ifdown);
3772 static void xfrm_link_failure(struct sk_buff *skb)
3774 /* Impossible. Such dst must be popped before reaches point of failure. */
3777 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3780 if (dst->obsolete) {
3788 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3791 struct xfrm_dst *xdst = bundle[nr];
3792 u32 pmtu, route_mtu_cached;
3793 struct dst_entry *dst;
3796 pmtu = dst_mtu(xfrm_dst_child(dst));
3797 xdst->child_mtu_cached = pmtu;
3799 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3801 route_mtu_cached = dst_mtu(xdst->route);
3802 xdst->route_mtu_cached = route_mtu_cached;
3804 if (pmtu > route_mtu_cached)
3805 pmtu = route_mtu_cached;
3807 dst_metric_set(dst, RTAX_MTU, pmtu);
3811 /* Check that the bundle accepts the flow and its components are
3815 static int xfrm_bundle_ok(struct xfrm_dst *first)
3817 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3818 struct dst_entry *dst = &first->u.dst;
3819 struct xfrm_dst *xdst;
3823 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3824 (dst->dev && !netif_running(dst->dev)))
3827 if (dst->flags & DST_XFRM_QUEUE)
3830 start_from = nr = 0;
3832 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3834 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3836 if (xdst->xfrm_genid != dst->xfrm->genid)
3838 if (xdst->num_pols > 0 &&
3839 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3842 bundle[nr++] = xdst;
3844 mtu = dst_mtu(xfrm_dst_child(dst));
3845 if (xdst->child_mtu_cached != mtu) {
3847 xdst->child_mtu_cached = mtu;
3850 if (!dst_check(xdst->route, xdst->route_cookie))
3852 mtu = dst_mtu(xdst->route);
3853 if (xdst->route_mtu_cached != mtu) {
3855 xdst->route_mtu_cached = mtu;
3858 dst = xfrm_dst_child(dst);
3859 } while (dst->xfrm);
3861 if (likely(!start_from))
3864 xdst = bundle[start_from - 1];
3865 mtu = xdst->child_mtu_cached;
3866 while (start_from--) {
3869 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3870 if (mtu > xdst->route_mtu_cached)
3871 mtu = xdst->route_mtu_cached;
3872 dst_metric_set(dst, RTAX_MTU, mtu);
3876 xdst = bundle[start_from - 1];
3877 xdst->child_mtu_cached = mtu;
3883 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3885 return dst_metric_advmss(xfrm_dst_path(dst));
3888 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3890 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3892 return mtu ? : dst_mtu(xfrm_dst_path(dst));
3895 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3899 const struct xfrm_state *xfrm = dst->xfrm;
3901 dst = xfrm_dst_child(dst);
3903 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3905 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3906 daddr = xfrm->coaddr;
3907 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3908 daddr = &xfrm->id.daddr;
3913 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3914 struct sk_buff *skb,
3917 const struct dst_entry *path = xfrm_dst_path(dst);
3920 daddr = xfrm_get_dst_nexthop(dst, daddr);
3921 return path->ops->neigh_lookup(path, skb, daddr);
3924 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3926 const struct dst_entry *path = xfrm_dst_path(dst);
3928 daddr = xfrm_get_dst_nexthop(dst, daddr);
3929 path->ops->confirm_neigh(path, daddr);
3932 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3936 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3937 return -EAFNOSUPPORT;
3939 spin_lock(&xfrm_policy_afinfo_lock);
3940 if (unlikely(xfrm_policy_afinfo[family] != NULL))
3943 struct dst_ops *dst_ops = afinfo->dst_ops;
3944 if (likely(dst_ops->kmem_cachep == NULL))
3945 dst_ops->kmem_cachep = xfrm_dst_cache;
3946 if (likely(dst_ops->check == NULL))
3947 dst_ops->check = xfrm_dst_check;
3948 if (likely(dst_ops->default_advmss == NULL))
3949 dst_ops->default_advmss = xfrm_default_advmss;
3950 if (likely(dst_ops->mtu == NULL))
3951 dst_ops->mtu = xfrm_mtu;
3952 if (likely(dst_ops->negative_advice == NULL))
3953 dst_ops->negative_advice = xfrm_negative_advice;
3954 if (likely(dst_ops->link_failure == NULL))
3955 dst_ops->link_failure = xfrm_link_failure;
3956 if (likely(dst_ops->neigh_lookup == NULL))
3957 dst_ops->neigh_lookup = xfrm_neigh_lookup;
3958 if (likely(!dst_ops->confirm_neigh))
3959 dst_ops->confirm_neigh = xfrm_confirm_neigh;
3960 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3962 spin_unlock(&xfrm_policy_afinfo_lock);
3966 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3968 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3970 struct dst_ops *dst_ops = afinfo->dst_ops;
3973 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3974 if (xfrm_policy_afinfo[i] != afinfo)
3976 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3982 dst_ops->kmem_cachep = NULL;
3983 dst_ops->check = NULL;
3984 dst_ops->negative_advice = NULL;
3985 dst_ops->link_failure = NULL;
3987 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3989 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3991 spin_lock(&xfrm_if_cb_lock);
3992 rcu_assign_pointer(xfrm_if_cb, ifcb);
3993 spin_unlock(&xfrm_if_cb_lock);
3995 EXPORT_SYMBOL(xfrm_if_register_cb);
3997 void xfrm_if_unregister_cb(void)
3999 RCU_INIT_POINTER(xfrm_if_cb, NULL);
4002 EXPORT_SYMBOL(xfrm_if_unregister_cb);
4004 #ifdef CONFIG_XFRM_STATISTICS
4005 static int __net_init xfrm_statistics_init(struct net *net)
4008 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4009 if (!net->mib.xfrm_statistics)
4011 rv = xfrm_proc_init(net);
4013 free_percpu(net->mib.xfrm_statistics);
4017 static void xfrm_statistics_fini(struct net *net)
4019 xfrm_proc_fini(net);
4020 free_percpu(net->mib.xfrm_statistics);
4023 static int __net_init xfrm_statistics_init(struct net *net)
4028 static void xfrm_statistics_fini(struct net *net)
4033 static int __net_init xfrm_policy_init(struct net *net)
4035 unsigned int hmask, sz;
4038 if (net_eq(net, &init_net)) {
4039 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4040 sizeof(struct xfrm_dst),
4041 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4043 err = rhashtable_init(&xfrm_policy_inexact_table,
4044 &xfrm_pol_inexact_params);
4049 sz = (hmask+1) * sizeof(struct hlist_head);
4051 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4052 if (!net->xfrm.policy_byidx)
4054 net->xfrm.policy_idx_hmask = hmask;
4056 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4057 struct xfrm_policy_hash *htab;
4059 net->xfrm.policy_count[dir] = 0;
4060 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4061 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4063 htab = &net->xfrm.policy_bydst[dir];
4064 htab->table = xfrm_hash_alloc(sz);
4067 htab->hmask = hmask;
4073 net->xfrm.policy_hthresh.lbits4 = 32;
4074 net->xfrm.policy_hthresh.rbits4 = 32;
4075 net->xfrm.policy_hthresh.lbits6 = 128;
4076 net->xfrm.policy_hthresh.rbits6 = 128;
4078 seqlock_init(&net->xfrm.policy_hthresh.lock);
4080 INIT_LIST_HEAD(&net->xfrm.policy_all);
4081 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4082 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4083 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4087 for (dir--; dir >= 0; dir--) {
4088 struct xfrm_policy_hash *htab;
4090 htab = &net->xfrm.policy_bydst[dir];
4091 xfrm_hash_free(htab->table, sz);
4093 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4098 static void xfrm_policy_fini(struct net *net)
4100 struct xfrm_pol_inexact_bin *b, *t;
4104 flush_work(&net->xfrm.policy_hash_work);
4105 #ifdef CONFIG_XFRM_SUB_POLICY
4106 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4108 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4110 WARN_ON(!list_empty(&net->xfrm.policy_all));
4112 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4113 struct xfrm_policy_hash *htab;
4115 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4117 htab = &net->xfrm.policy_bydst[dir];
4118 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4119 WARN_ON(!hlist_empty(htab->table));
4120 xfrm_hash_free(htab->table, sz);
4123 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4124 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4125 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4127 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4128 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4129 __xfrm_policy_inexact_prune_bin(b, true);
4130 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4133 static int __net_init xfrm_net_init(struct net *net)
4137 /* Initialize the per-net locks here */
4138 spin_lock_init(&net->xfrm.xfrm_state_lock);
4139 spin_lock_init(&net->xfrm.xfrm_policy_lock);
4140 mutex_init(&net->xfrm.xfrm_cfg_mutex);
4142 rv = xfrm_statistics_init(net);
4144 goto out_statistics;
4145 rv = xfrm_state_init(net);
4148 rv = xfrm_policy_init(net);
4151 rv = xfrm_sysctl_init(net);
4158 xfrm_policy_fini(net);
4160 xfrm_state_fini(net);
4162 xfrm_statistics_fini(net);
4167 static void __net_exit xfrm_net_exit(struct net *net)
4169 xfrm_sysctl_fini(net);
4170 xfrm_policy_fini(net);
4171 xfrm_state_fini(net);
4172 xfrm_statistics_fini(net);
4175 static struct pernet_operations __net_initdata xfrm_net_ops = {
4176 .init = xfrm_net_init,
4177 .exit = xfrm_net_exit,
4180 void __init xfrm_init(void)
4182 register_pernet_subsys(&xfrm_net_ops);
4184 seqcount_init(&xfrm_policy_hash_generation);
4187 RCU_INIT_POINTER(xfrm_if_cb, NULL);
4191 #ifdef CONFIG_AUDITSYSCALL
4192 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4193 struct audit_buffer *audit_buf)
4195 struct xfrm_sec_ctx *ctx = xp->security;
4196 struct xfrm_selector *sel = &xp->selector;
4199 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4200 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4202 switch (sel->family) {
4204 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4205 if (sel->prefixlen_s != 32)
4206 audit_log_format(audit_buf, " src_prefixlen=%d",
4208 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4209 if (sel->prefixlen_d != 32)
4210 audit_log_format(audit_buf, " dst_prefixlen=%d",
4214 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4215 if (sel->prefixlen_s != 128)
4216 audit_log_format(audit_buf, " src_prefixlen=%d",
4218 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4219 if (sel->prefixlen_d != 128)
4220 audit_log_format(audit_buf, " dst_prefixlen=%d",
4226 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4228 struct audit_buffer *audit_buf;
4230 audit_buf = xfrm_audit_start("SPD-add");
4231 if (audit_buf == NULL)
4233 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4234 audit_log_format(audit_buf, " res=%u", result);
4235 xfrm_audit_common_policyinfo(xp, audit_buf);
4236 audit_log_end(audit_buf);
4238 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4240 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4243 struct audit_buffer *audit_buf;
4245 audit_buf = xfrm_audit_start("SPD-delete");
4246 if (audit_buf == NULL)
4248 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4249 audit_log_format(audit_buf, " res=%u", result);
4250 xfrm_audit_common_policyinfo(xp, audit_buf);
4251 audit_log_end(audit_buf);
4253 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4256 #ifdef CONFIG_XFRM_MIGRATE
4257 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4258 const struct xfrm_selector *sel_tgt)
4260 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4261 if (sel_tgt->family == sel_cmp->family &&
4262 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4264 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4266 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4267 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4271 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4278 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4279 u8 dir, u8 type, struct net *net, u32 if_id)
4281 struct xfrm_policy *pol, *ret = NULL;
4282 struct hlist_head *chain;
4285 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4286 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4287 hlist_for_each_entry(pol, chain, bydst) {
4288 if ((if_id == 0 || pol->if_id == if_id) &&
4289 xfrm_migrate_selector_match(sel, &pol->selector) &&
4290 pol->type == type) {
4292 priority = ret->priority;
4296 chain = &net->xfrm.policy_inexact[dir];
4297 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4298 if ((pol->priority >= priority) && ret)
4301 if ((if_id == 0 || pol->if_id == if_id) &&
4302 xfrm_migrate_selector_match(sel, &pol->selector) &&
4303 pol->type == type) {
4311 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4316 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4320 if (t->mode == m->mode && t->id.proto == m->proto &&
4321 (m->reqid == 0 || t->reqid == m->reqid)) {
4323 case XFRM_MODE_TUNNEL:
4324 case XFRM_MODE_BEET:
4325 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4327 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4332 case XFRM_MODE_TRANSPORT:
4333 /* in case of transport mode, template does not store
4334 any IP addresses, hence we just compare mode and
4345 /* update endpoint address(es) of template(s) */
4346 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4347 struct xfrm_migrate *m, int num_migrate)
4349 struct xfrm_migrate *mp;
4352 write_lock_bh(&pol->lock);
4353 if (unlikely(pol->walk.dead)) {
4354 /* target policy has been deleted */
4355 write_unlock_bh(&pol->lock);
4359 for (i = 0; i < pol->xfrm_nr; i++) {
4360 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4361 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4364 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4365 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4367 /* update endpoints */
4368 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4369 sizeof(pol->xfrm_vec[i].id.daddr));
4370 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4371 sizeof(pol->xfrm_vec[i].saddr));
4372 pol->xfrm_vec[i].encap_family = mp->new_family;
4374 atomic_inc(&pol->genid);
4378 write_unlock_bh(&pol->lock);
4386 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4390 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4393 for (i = 0; i < num_migrate; i++) {
4394 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4395 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4398 /* check if there is any duplicated entry */
4399 for (j = i + 1; j < num_migrate; j++) {
4400 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4401 sizeof(m[i].old_daddr)) &&
4402 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4403 sizeof(m[i].old_saddr)) &&
4404 m[i].proto == m[j].proto &&
4405 m[i].mode == m[j].mode &&
4406 m[i].reqid == m[j].reqid &&
4407 m[i].old_family == m[j].old_family)
4415 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4416 struct xfrm_migrate *m, int num_migrate,
4417 struct xfrm_kmaddress *k, struct net *net,
4418 struct xfrm_encap_tmpl *encap, u32 if_id)
4420 int i, err, nx_cur = 0, nx_new = 0;
4421 struct xfrm_policy *pol = NULL;
4422 struct xfrm_state *x, *xc;
4423 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4424 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4425 struct xfrm_migrate *mp;
4427 /* Stage 0 - sanity checks */
4428 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4431 if (dir >= XFRM_POLICY_MAX) {
4436 /* Stage 1 - find policy */
4437 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id)) == NULL) {
4442 /* Stage 2 - find and update state(s) */
4443 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4444 if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4447 xc = xfrm_state_migrate(x, mp, encap);
4458 /* Stage 3 - update policy */
4459 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4462 /* Stage 4 - delete old state(s) */
4464 xfrm_states_put(x_cur, nx_cur);
4465 xfrm_states_delete(x_cur, nx_cur);
4468 /* Stage 5 - announce */
4469 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4481 xfrm_states_put(x_cur, nx_cur);
4483 xfrm_states_delete(x_new, nx_new);
4487 EXPORT_SYMBOL(xfrm_migrate);