1 // SPDX-License-Identifier: GPL-2.0-only
7 * Kazunori MIYAZAWA @USAGI
8 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
10 * Kazunori MIYAZAWA @USAGI
12 * Split up af-specific portion
13 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
17 #include <linux/err.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/workqueue.h>
23 #include <linux/notifier.h>
24 #include <linux/netdevice.h>
25 #include <linux/netfilter.h>
26 #include <linux/module.h>
27 #include <linux/cache.h>
28 #include <linux/cpu.h>
29 #include <linux/audit.h>
30 #include <linux/rhashtable.h>
31 #include <linux/if_tunnel.h>
36 #if IS_ENABLED(CONFIG_IPV6_MIP6)
39 #ifdef CONFIG_XFRM_STATISTICS
43 #include "xfrm_hash.h"
45 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
46 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
47 #define XFRM_MAX_QUEUE_LEN 100
50 struct dst_entry *dst_orig;
54 /* prefixes smaller than this are stored in lists, not trees. */
55 #define INEXACT_PREFIXLEN_IPV4 16
56 #define INEXACT_PREFIXLEN_IPV6 48
58 struct xfrm_pol_inexact_node {
68 /* the policies matching this node, can be empty list */
69 struct hlist_head hhead;
72 /* xfrm inexact policy search tree:
73 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
75 * +---- root_d: sorted by daddr:prefix
77 * | xfrm_pol_inexact_node
79 * | +- root: sorted by saddr/prefix
81 * | | xfrm_pol_inexact_node
85 * | | + hhead: saddr:daddr policies
87 * | +- coarse policies and all any:daddr policies
89 * +---- root_s: sorted by saddr:prefix
91 * | xfrm_pol_inexact_node
95 * | + hhead: saddr:any policies
97 * +---- coarse policies and all any:any policies
99 * Lookups return four candidate lists:
100 * 1. any:any list from top-level xfrm_pol_inexact_bin
101 * 2. any:daddr list from daddr tree
102 * 3. saddr:daddr list from 2nd level daddr tree
103 * 4. saddr:any list from saddr tree
105 * This result set then needs to be searched for the policy with
106 * the lowest priority. If two results have same prio, youngest one wins.
109 struct xfrm_pol_inexact_key {
116 struct xfrm_pol_inexact_bin {
117 struct xfrm_pol_inexact_key k;
118 struct rhash_head head;
119 /* list containing '*:*' policies */
120 struct hlist_head hhead;
123 /* tree sorted by daddr/prefix */
124 struct rb_root root_d;
126 /* tree sorted by saddr/prefix */
127 struct rb_root root_s;
129 /* slow path below */
130 struct list_head inexact_bins;
134 enum xfrm_pol_inexact_candidate_type {
143 struct xfrm_pol_inexact_candidates {
144 struct hlist_head *res[XFRM_POL_CAND_MAX];
147 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
148 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
150 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
151 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
154 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
155 static __read_mostly seqcount_t xfrm_policy_hash_generation;
157 static struct rhashtable xfrm_policy_inexact_table;
158 static const struct rhashtable_params xfrm_pol_inexact_params;
160 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
161 static int stale_bundle(struct dst_entry *dst);
162 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
163 static void xfrm_policy_queue_process(struct timer_list *t);
165 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
166 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
169 static struct xfrm_pol_inexact_bin *
170 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
173 static struct xfrm_pol_inexact_bin *
174 xfrm_policy_inexact_lookup_rcu(struct net *net,
175 u8 type, u16 family, u8 dir, u32 if_id);
176 static struct xfrm_policy *
177 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
179 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
180 struct xfrm_policy *policy);
183 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
184 struct xfrm_pol_inexact_bin *b,
185 const xfrm_address_t *saddr,
186 const xfrm_address_t *daddr);
188 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
190 return refcount_inc_not_zero(&policy->refcnt);
194 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
196 const struct flowi4 *fl4 = &fl->u.ip4;
198 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
199 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
200 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
201 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
202 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
203 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
207 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
209 const struct flowi6 *fl6 = &fl->u.ip6;
211 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
212 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
213 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
214 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
215 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
216 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
219 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
220 unsigned short family)
224 return __xfrm4_selector_match(sel, fl);
226 return __xfrm6_selector_match(sel, fl);
231 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
233 const struct xfrm_policy_afinfo *afinfo;
235 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
238 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
239 if (unlikely(!afinfo))
244 /* Called with rcu_read_lock(). */
245 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
247 return rcu_dereference(xfrm_if_cb);
250 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
251 const xfrm_address_t *saddr,
252 const xfrm_address_t *daddr,
253 int family, u32 mark)
255 const struct xfrm_policy_afinfo *afinfo;
256 struct dst_entry *dst;
258 afinfo = xfrm_policy_get_afinfo(family);
259 if (unlikely(afinfo == NULL))
260 return ERR_PTR(-EAFNOSUPPORT);
262 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
268 EXPORT_SYMBOL(__xfrm_dst_lookup);
270 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
272 xfrm_address_t *prev_saddr,
273 xfrm_address_t *prev_daddr,
274 int family, u32 mark)
276 struct net *net = xs_net(x);
277 xfrm_address_t *saddr = &x->props.saddr;
278 xfrm_address_t *daddr = &x->id.daddr;
279 struct dst_entry *dst;
281 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
285 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
290 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
293 if (prev_saddr != saddr)
294 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
295 if (prev_daddr != daddr)
296 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
302 static inline unsigned long make_jiffies(long secs)
304 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
305 return MAX_SCHEDULE_TIMEOUT-1;
310 static void xfrm_policy_timer(struct timer_list *t)
312 struct xfrm_policy *xp = from_timer(xp, t, timer);
313 time64_t now = ktime_get_real_seconds();
314 time64_t next = TIME64_MAX;
318 read_lock(&xp->lock);
320 if (unlikely(xp->walk.dead))
323 dir = xfrm_policy_id2dir(xp->index);
325 if (xp->lft.hard_add_expires_seconds) {
326 time64_t tmo = xp->lft.hard_add_expires_seconds +
327 xp->curlft.add_time - now;
333 if (xp->lft.hard_use_expires_seconds) {
334 time64_t tmo = xp->lft.hard_use_expires_seconds +
335 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
341 if (xp->lft.soft_add_expires_seconds) {
342 time64_t tmo = xp->lft.soft_add_expires_seconds +
343 xp->curlft.add_time - now;
346 tmo = XFRM_KM_TIMEOUT;
351 if (xp->lft.soft_use_expires_seconds) {
352 time64_t tmo = xp->lft.soft_use_expires_seconds +
353 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
356 tmo = XFRM_KM_TIMEOUT;
363 km_policy_expired(xp, dir, 0, 0);
364 if (next != TIME64_MAX &&
365 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
369 read_unlock(&xp->lock);
374 read_unlock(&xp->lock);
375 if (!xfrm_policy_delete(xp, dir))
376 km_policy_expired(xp, dir, 1, 0);
380 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
384 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
386 struct xfrm_policy *policy;
388 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
391 write_pnet(&policy->xp_net, net);
392 INIT_LIST_HEAD(&policy->walk.all);
393 INIT_HLIST_NODE(&policy->bydst_inexact_list);
394 INIT_HLIST_NODE(&policy->bydst);
395 INIT_HLIST_NODE(&policy->byidx);
396 rwlock_init(&policy->lock);
397 refcount_set(&policy->refcnt, 1);
398 skb_queue_head_init(&policy->polq.hold_queue);
399 timer_setup(&policy->timer, xfrm_policy_timer, 0);
400 timer_setup(&policy->polq.hold_timer,
401 xfrm_policy_queue_process, 0);
405 EXPORT_SYMBOL(xfrm_policy_alloc);
407 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
409 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
411 security_xfrm_policy_free(policy->security);
415 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
417 void xfrm_policy_destroy(struct xfrm_policy *policy)
419 BUG_ON(!policy->walk.dead);
421 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
424 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
426 EXPORT_SYMBOL(xfrm_policy_destroy);
428 /* Rule must be locked. Release descendant resources, announce
429 * entry dead. The rule must be unlinked from lists to the moment.
432 static void xfrm_policy_kill(struct xfrm_policy *policy)
434 policy->walk.dead = 1;
436 atomic_inc(&policy->genid);
438 if (del_timer(&policy->polq.hold_timer))
439 xfrm_pol_put(policy);
440 skb_queue_purge(&policy->polq.hold_queue);
442 if (del_timer(&policy->timer))
443 xfrm_pol_put(policy);
445 xfrm_pol_put(policy);
448 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
450 static inline unsigned int idx_hash(struct net *net, u32 index)
452 return __idx_hash(index, net->xfrm.policy_idx_hmask);
455 /* calculate policy hash thresholds */
456 static void __get_hash_thresh(struct net *net,
457 unsigned short family, int dir,
458 u8 *dbits, u8 *sbits)
462 *dbits = net->xfrm.policy_bydst[dir].dbits4;
463 *sbits = net->xfrm.policy_bydst[dir].sbits4;
467 *dbits = net->xfrm.policy_bydst[dir].dbits6;
468 *sbits = net->xfrm.policy_bydst[dir].sbits6;
477 static struct hlist_head *policy_hash_bysel(struct net *net,
478 const struct xfrm_selector *sel,
479 unsigned short family, int dir)
481 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
486 __get_hash_thresh(net, family, dir, &dbits, &sbits);
487 hash = __sel_hash(sel, family, hmask, dbits, sbits);
489 if (hash == hmask + 1)
492 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
493 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
496 static struct hlist_head *policy_hash_direct(struct net *net,
497 const xfrm_address_t *daddr,
498 const xfrm_address_t *saddr,
499 unsigned short family, int dir)
501 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
506 __get_hash_thresh(net, family, dir, &dbits, &sbits);
507 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
509 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
510 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
513 static void xfrm_dst_hash_transfer(struct net *net,
514 struct hlist_head *list,
515 struct hlist_head *ndsttable,
516 unsigned int nhashmask,
519 struct hlist_node *tmp, *entry0 = NULL;
520 struct xfrm_policy *pol;
526 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
529 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
530 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
531 pol->family, nhashmask, dbits, sbits);
533 hlist_del_rcu(&pol->bydst);
534 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
539 hlist_del_rcu(&pol->bydst);
540 hlist_add_behind_rcu(&pol->bydst, entry0);
542 entry0 = &pol->bydst;
544 if (!hlist_empty(list)) {
550 static void xfrm_idx_hash_transfer(struct hlist_head *list,
551 struct hlist_head *nidxtable,
552 unsigned int nhashmask)
554 struct hlist_node *tmp;
555 struct xfrm_policy *pol;
557 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
560 h = __idx_hash(pol->index, nhashmask);
561 hlist_add_head(&pol->byidx, nidxtable+h);
565 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
567 return ((old_hmask + 1) << 1) - 1;
570 static void xfrm_bydst_resize(struct net *net, int dir)
572 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
573 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
574 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
575 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
576 struct hlist_head *odst;
582 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
583 write_seqcount_begin(&xfrm_policy_hash_generation);
585 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
586 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
588 for (i = hmask; i >= 0; i--)
589 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
591 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
592 net->xfrm.policy_bydst[dir].hmask = nhashmask;
594 write_seqcount_end(&xfrm_policy_hash_generation);
595 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
599 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
602 static void xfrm_byidx_resize(struct net *net, int total)
604 unsigned int hmask = net->xfrm.policy_idx_hmask;
605 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
606 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
607 struct hlist_head *oidx = net->xfrm.policy_byidx;
608 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
614 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
616 for (i = hmask; i >= 0; i--)
617 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
619 net->xfrm.policy_byidx = nidx;
620 net->xfrm.policy_idx_hmask = nhashmask;
622 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
624 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
627 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
629 unsigned int cnt = net->xfrm.policy_count[dir];
630 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
635 if ((hmask + 1) < xfrm_policy_hashmax &&
642 static inline int xfrm_byidx_should_resize(struct net *net, int total)
644 unsigned int hmask = net->xfrm.policy_idx_hmask;
646 if ((hmask + 1) < xfrm_policy_hashmax &&
653 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
655 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
656 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
657 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
658 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
659 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
660 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
661 si->spdhcnt = net->xfrm.policy_idx_hmask;
662 si->spdhmcnt = xfrm_policy_hashmax;
664 EXPORT_SYMBOL(xfrm_spd_getinfo);
666 static DEFINE_MUTEX(hash_resize_mutex);
667 static void xfrm_hash_resize(struct work_struct *work)
669 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
672 mutex_lock(&hash_resize_mutex);
675 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
676 if (xfrm_bydst_should_resize(net, dir, &total))
677 xfrm_bydst_resize(net, dir);
679 if (xfrm_byidx_should_resize(net, total))
680 xfrm_byidx_resize(net, total);
682 mutex_unlock(&hash_resize_mutex);
685 /* Make sure *pol can be inserted into fastbin.
686 * Useful to check that later insert requests will be sucessful
687 * (provided xfrm_policy_lock is held throughout).
689 static struct xfrm_pol_inexact_bin *
690 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
692 struct xfrm_pol_inexact_bin *bin, *prev;
693 struct xfrm_pol_inexact_key k = {
694 .family = pol->family,
699 struct net *net = xp_net(pol);
701 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
703 write_pnet(&k.net, net);
704 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
705 xfrm_pol_inexact_params);
709 bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
714 INIT_HLIST_HEAD(&bin->hhead);
715 bin->root_d = RB_ROOT;
716 bin->root_s = RB_ROOT;
717 seqcount_init(&bin->count);
719 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
721 xfrm_pol_inexact_params);
723 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
729 return IS_ERR(prev) ? NULL : prev;
732 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
733 int family, u8 prefixlen)
735 if (xfrm_addr_any(addr, family))
738 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
741 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
748 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
750 const xfrm_address_t *addr;
751 bool saddr_any, daddr_any;
754 addr = &policy->selector.saddr;
755 prefixlen = policy->selector.prefixlen_s;
757 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
760 addr = &policy->selector.daddr;
761 prefixlen = policy->selector.prefixlen_d;
762 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
765 return saddr_any && daddr_any;
768 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
769 const xfrm_address_t *addr, u8 prefixlen)
772 node->prefixlen = prefixlen;
775 static struct xfrm_pol_inexact_node *
776 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
778 struct xfrm_pol_inexact_node *node;
780 node = kzalloc(sizeof(*node), GFP_ATOMIC);
782 xfrm_pol_inexact_node_init(node, addr, prefixlen);
787 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
788 const xfrm_address_t *b,
789 u8 prefixlen, u16 family)
791 unsigned int pdw, pbi;
796 if (sizeof(long) == 4 && prefixlen == 0)
797 return ntohl(a->a4) - ntohl(b->a4);
798 return (ntohl(a->a4) & ((~0UL << (32 - prefixlen)))) -
799 (ntohl(b->a4) & ((~0UL << (32 - prefixlen))));
801 pdw = prefixlen >> 5;
802 pbi = prefixlen & 0x1f;
805 delta = memcmp(a->a6, b->a6, pdw << 2);
810 u32 mask = ~0u << (32 - pbi);
812 delta = (ntohl(a->a6[pdw]) & mask) -
813 (ntohl(b->a6[pdw]) & mask);
823 static void xfrm_policy_inexact_list_reinsert(struct net *net,
824 struct xfrm_pol_inexact_node *n,
827 unsigned int matched_s, matched_d;
828 struct xfrm_policy *policy, *p;
833 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
834 struct hlist_node *newpos = NULL;
835 bool matches_s, matches_d;
837 if (!policy->bydst_reinsert)
840 WARN_ON_ONCE(policy->family != family);
842 policy->bydst_reinsert = false;
843 hlist_for_each_entry(p, &n->hhead, bydst) {
844 if (policy->priority > p->priority)
846 else if (policy->priority == p->priority &&
847 policy->pos > p->pos)
854 hlist_add_behind_rcu(&policy->bydst, newpos);
856 hlist_add_head_rcu(&policy->bydst, &n->hhead);
858 /* paranoia checks follow.
859 * Check that the reinserted policy matches at least
860 * saddr or daddr for current node prefix.
862 * Matching both is fine, matching saddr in one policy
863 * (but not daddr) and then matching only daddr in another
866 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
870 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
874 if (matches_s && matches_d)
877 WARN_ON_ONCE(!matches_s && !matches_d);
882 WARN_ON_ONCE(matched_s && matched_d);
886 static void xfrm_policy_inexact_node_reinsert(struct net *net,
887 struct xfrm_pol_inexact_node *n,
891 struct xfrm_pol_inexact_node *node;
892 struct rb_node **p, *parent;
894 /* we should not have another subtree here */
895 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
904 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
906 prefixlen = min(node->prefixlen, n->prefixlen);
908 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
911 p = &parent->rb_left;
912 } else if (delta > 0) {
913 p = &parent->rb_right;
915 bool same_prefixlen = node->prefixlen == n->prefixlen;
916 struct xfrm_policy *tmp;
918 hlist_for_each_entry(tmp, &n->hhead, bydst) {
919 tmp->bydst_reinsert = true;
920 hlist_del_rcu(&tmp->bydst);
923 node->prefixlen = prefixlen;
925 xfrm_policy_inexact_list_reinsert(net, node, family);
927 if (same_prefixlen) {
939 rb_link_node_rcu(&n->node, parent, p);
940 rb_insert_color(&n->node, new);
943 /* merge nodes v and n */
944 static void xfrm_policy_inexact_node_merge(struct net *net,
945 struct xfrm_pol_inexact_node *v,
946 struct xfrm_pol_inexact_node *n,
949 struct xfrm_pol_inexact_node *node;
950 struct xfrm_policy *tmp;
951 struct rb_node *rnode;
953 /* To-be-merged node v has a subtree.
955 * Dismantle it and insert its nodes to n->root.
957 while ((rnode = rb_first(&v->root)) != NULL) {
958 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
959 rb_erase(&node->node, &v->root);
960 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
964 hlist_for_each_entry(tmp, &v->hhead, bydst) {
965 tmp->bydst_reinsert = true;
966 hlist_del_rcu(&tmp->bydst);
969 xfrm_policy_inexact_list_reinsert(net, n, family);
972 static struct xfrm_pol_inexact_node *
973 xfrm_policy_inexact_insert_node(struct net *net,
974 struct rb_root *root,
975 xfrm_address_t *addr,
976 u16 family, u8 prefixlen, u8 dir)
978 struct xfrm_pol_inexact_node *cached = NULL;
979 struct rb_node **p, *parent = NULL;
980 struct xfrm_pol_inexact_node *node;
987 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
989 delta = xfrm_policy_addr_delta(addr, &node->addr,
992 if (delta == 0 && prefixlen >= node->prefixlen) {
993 WARN_ON_ONCE(cached); /* ipsec policies got lost */
998 p = &parent->rb_left;
1000 p = &parent->rb_right;
1002 if (prefixlen < node->prefixlen) {
1003 delta = xfrm_policy_addr_delta(addr, &node->addr,
1009 /* This node is a subnet of the new prefix. It needs
1010 * to be removed and re-inserted with the smaller
1011 * prefix and all nodes that are now also covered
1012 * by the reduced prefixlen.
1014 rb_erase(&node->node, root);
1017 xfrm_pol_inexact_node_init(node, addr,
1021 /* This node also falls within the new
1022 * prefixlen. Merge the to-be-reinserted
1023 * node and this one.
1025 xfrm_policy_inexact_node_merge(net, node,
1027 kfree_rcu(node, rcu);
1038 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1043 rb_link_node_rcu(&node->node, parent, p);
1044 rb_insert_color(&node->node, root);
1049 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1051 struct xfrm_pol_inexact_node *node;
1052 struct rb_node *rn = rb_first(r);
1055 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1057 xfrm_policy_inexact_gc_tree(&node->root, rm);
1060 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1065 rb_erase(&node->node, r);
1066 kfree_rcu(node, rcu);
1070 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1072 write_seqcount_begin(&b->count);
1073 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1074 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1075 write_seqcount_end(&b->count);
1077 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1078 !hlist_empty(&b->hhead)) {
1079 WARN_ON_ONCE(net_exit);
1083 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1084 xfrm_pol_inexact_params) == 0) {
1085 list_del(&b->inexact_bins);
1090 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1092 struct net *net = read_pnet(&b->k.net);
1094 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1095 __xfrm_policy_inexact_prune_bin(b, false);
1096 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1099 static void __xfrm_policy_inexact_flush(struct net *net)
1101 struct xfrm_pol_inexact_bin *bin, *t;
1103 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1105 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1106 __xfrm_policy_inexact_prune_bin(bin, false);
1109 static struct hlist_head *
1110 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1111 struct xfrm_policy *policy, u8 dir)
1113 struct xfrm_pol_inexact_node *n;
1116 net = xp_net(policy);
1117 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1119 if (xfrm_policy_inexact_insert_use_any_list(policy))
1122 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1124 policy->selector.prefixlen_d)) {
1125 write_seqcount_begin(&bin->count);
1126 n = xfrm_policy_inexact_insert_node(net,
1128 &policy->selector.saddr,
1130 policy->selector.prefixlen_s,
1132 write_seqcount_end(&bin->count);
1139 /* daddr is fixed */
1140 write_seqcount_begin(&bin->count);
1141 n = xfrm_policy_inexact_insert_node(net,
1143 &policy->selector.daddr,
1145 policy->selector.prefixlen_d, dir);
1146 write_seqcount_end(&bin->count);
1150 /* saddr is wildcard */
1151 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1153 policy->selector.prefixlen_s))
1156 write_seqcount_begin(&bin->count);
1157 n = xfrm_policy_inexact_insert_node(net,
1159 &policy->selector.saddr,
1161 policy->selector.prefixlen_s, dir);
1162 write_seqcount_end(&bin->count);
1169 static struct xfrm_policy *
1170 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1172 struct xfrm_pol_inexact_bin *bin;
1173 struct xfrm_policy *delpol;
1174 struct hlist_head *chain;
1177 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1179 return ERR_PTR(-ENOMEM);
1181 net = xp_net(policy);
1182 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1184 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1186 __xfrm_policy_inexact_prune_bin(bin, false);
1187 return ERR_PTR(-ENOMEM);
1190 delpol = xfrm_policy_insert_list(chain, policy, excl);
1191 if (delpol && excl) {
1192 __xfrm_policy_inexact_prune_bin(bin, false);
1193 return ERR_PTR(-EEXIST);
1196 chain = &net->xfrm.policy_inexact[dir];
1197 xfrm_policy_insert_inexact_list(chain, policy);
1200 __xfrm_policy_inexact_prune_bin(bin, false);
1205 static void xfrm_hash_rebuild(struct work_struct *work)
1207 struct net *net = container_of(work, struct net,
1208 xfrm.policy_hthresh.work);
1210 struct xfrm_policy *pol;
1211 struct xfrm_policy *policy;
1212 struct hlist_head *chain;
1213 struct hlist_head *odst;
1214 struct hlist_node *newpos;
1218 u8 lbits4, rbits4, lbits6, rbits6;
1220 mutex_lock(&hash_resize_mutex);
1222 /* read selector prefixlen thresholds */
1224 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1226 lbits4 = net->xfrm.policy_hthresh.lbits4;
1227 rbits4 = net->xfrm.policy_hthresh.rbits4;
1228 lbits6 = net->xfrm.policy_hthresh.lbits6;
1229 rbits6 = net->xfrm.policy_hthresh.rbits6;
1230 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1232 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1233 write_seqcount_begin(&xfrm_policy_hash_generation);
1235 /* make sure that we can insert the indirect policies again before
1236 * we start with destructive action.
1238 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1239 struct xfrm_pol_inexact_bin *bin;
1242 dir = xfrm_policy_id2dir(policy->index);
1243 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1246 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1247 if (policy->family == AF_INET) {
1255 if (policy->family == AF_INET) {
1264 if (policy->selector.prefixlen_d < dbits ||
1265 policy->selector.prefixlen_s < sbits)
1268 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1272 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1276 /* reset the bydst and inexact table in all directions */
1277 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1278 struct hlist_node *n;
1280 hlist_for_each_entry_safe(policy, n,
1281 &net->xfrm.policy_inexact[dir],
1282 bydst_inexact_list) {
1283 hlist_del_rcu(&policy->bydst);
1284 hlist_del_init(&policy->bydst_inexact_list);
1287 hmask = net->xfrm.policy_bydst[dir].hmask;
1288 odst = net->xfrm.policy_bydst[dir].table;
1289 for (i = hmask; i >= 0; i--) {
1290 hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1291 hlist_del_rcu(&policy->bydst);
1293 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1294 /* dir out => dst = remote, src = local */
1295 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1296 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1297 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1298 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1300 /* dir in/fwd => dst = local, src = remote */
1301 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1302 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1303 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1304 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1308 /* re-insert all policies by order of creation */
1309 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1310 if (policy->walk.dead)
1312 dir = xfrm_policy_id2dir(policy->index);
1313 if (dir >= XFRM_POLICY_MAX) {
1314 /* skip socket policies */
1318 chain = policy_hash_bysel(net, &policy->selector,
1319 policy->family, dir);
1322 void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1324 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1328 hlist_for_each_entry(pol, chain, bydst) {
1329 if (policy->priority >= pol->priority)
1330 newpos = &pol->bydst;
1335 hlist_add_behind_rcu(&policy->bydst, newpos);
1337 hlist_add_head_rcu(&policy->bydst, chain);
1341 __xfrm_policy_inexact_flush(net);
1342 write_seqcount_end(&xfrm_policy_hash_generation);
1343 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1345 mutex_unlock(&hash_resize_mutex);
1348 void xfrm_policy_hash_rebuild(struct net *net)
1350 schedule_work(&net->xfrm.policy_hthresh.work);
1352 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1354 /* Generate new index... KAME seems to generate them ordered by cost
1355 * of an absolute inpredictability of ordering of rules. This will not pass. */
1356 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1358 static u32 idx_generator;
1361 struct hlist_head *list;
1362 struct xfrm_policy *p;
1367 idx = (idx_generator | dir);
1376 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1378 hlist_for_each_entry(p, list, byidx) {
1379 if (p->index == idx) {
1389 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1391 u32 *p1 = (u32 *) s1;
1392 u32 *p2 = (u32 *) s2;
1393 int len = sizeof(struct xfrm_selector) / sizeof(u32);
1396 for (i = 0; i < len; i++) {
1404 static void xfrm_policy_requeue(struct xfrm_policy *old,
1405 struct xfrm_policy *new)
1407 struct xfrm_policy_queue *pq = &old->polq;
1408 struct sk_buff_head list;
1410 if (skb_queue_empty(&pq->hold_queue))
1413 __skb_queue_head_init(&list);
1415 spin_lock_bh(&pq->hold_queue.lock);
1416 skb_queue_splice_init(&pq->hold_queue, &list);
1417 if (del_timer(&pq->hold_timer))
1419 spin_unlock_bh(&pq->hold_queue.lock);
1423 spin_lock_bh(&pq->hold_queue.lock);
1424 skb_queue_splice(&list, &pq->hold_queue);
1425 pq->timeout = XFRM_QUEUE_TMO_MIN;
1426 if (!mod_timer(&pq->hold_timer, jiffies))
1428 spin_unlock_bh(&pq->hold_queue.lock);
1431 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
1432 struct xfrm_policy *pol)
1434 u32 mark = policy->mark.v & policy->mark.m;
1436 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
1439 if ((mark & pol->mark.m) == pol->mark.v &&
1440 policy->priority == pol->priority)
1446 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1448 const struct xfrm_pol_inexact_key *k = data;
1449 u32 a = k->type << 24 | k->dir << 16 | k->family;
1451 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1455 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1457 const struct xfrm_pol_inexact_bin *b = data;
1459 return xfrm_pol_bin_key(&b->k, 0, seed);
1462 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1465 const struct xfrm_pol_inexact_key *key = arg->key;
1466 const struct xfrm_pol_inexact_bin *b = ptr;
1469 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1472 ret = b->k.dir ^ key->dir;
1476 ret = b->k.type ^ key->type;
1480 ret = b->k.family ^ key->family;
1484 return b->k.if_id ^ key->if_id;
1487 static const struct rhashtable_params xfrm_pol_inexact_params = {
1488 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head),
1489 .hashfn = xfrm_pol_bin_key,
1490 .obj_hashfn = xfrm_pol_bin_obj,
1491 .obj_cmpfn = xfrm_pol_bin_cmp,
1492 .automatic_shrinking = true,
1495 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1496 struct xfrm_policy *policy)
1498 struct xfrm_policy *pol, *delpol = NULL;
1499 struct hlist_node *newpos = NULL;
1502 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1503 if (pol->type == policy->type &&
1504 pol->if_id == policy->if_id &&
1505 !selector_cmp(&pol->selector, &policy->selector) &&
1506 xfrm_policy_mark_match(policy, pol) &&
1507 xfrm_sec_ctx_match(pol->security, policy->security) &&
1510 if (policy->priority > pol->priority)
1512 } else if (policy->priority >= pol->priority) {
1513 newpos = &pol->bydst_inexact_list;
1521 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1523 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1525 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1531 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1532 struct xfrm_policy *policy,
1535 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1537 hlist_for_each_entry(pol, chain, bydst) {
1538 if (pol->type == policy->type &&
1539 pol->if_id == policy->if_id &&
1540 !selector_cmp(&pol->selector, &policy->selector) &&
1541 xfrm_policy_mark_match(policy, pol) &&
1542 xfrm_sec_ctx_match(pol->security, policy->security) &&
1545 return ERR_PTR(-EEXIST);
1547 if (policy->priority > pol->priority)
1549 } else if (policy->priority >= pol->priority) {
1558 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1560 hlist_add_head_rcu(&policy->bydst, chain);
1565 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1567 struct net *net = xp_net(policy);
1568 struct xfrm_policy *delpol;
1569 struct hlist_head *chain;
1571 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1572 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1574 delpol = xfrm_policy_insert_list(chain, policy, excl);
1576 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1578 if (IS_ERR(delpol)) {
1579 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1580 return PTR_ERR(delpol);
1583 __xfrm_policy_link(policy, dir);
1585 /* After previous checking, family can either be AF_INET or AF_INET6 */
1586 if (policy->family == AF_INET)
1587 rt_genid_bump_ipv4(net);
1589 rt_genid_bump_ipv6(net);
1592 xfrm_policy_requeue(delpol, policy);
1593 __xfrm_policy_unlink(delpol, dir);
1595 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1596 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1597 policy->curlft.add_time = ktime_get_real_seconds();
1598 policy->curlft.use_time = 0;
1599 if (!mod_timer(&policy->timer, jiffies + HZ))
1600 xfrm_pol_hold(policy);
1601 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1604 xfrm_policy_kill(delpol);
1605 else if (xfrm_bydst_should_resize(net, dir, NULL))
1606 schedule_work(&net->xfrm.policy_hash_work);
1610 EXPORT_SYMBOL(xfrm_policy_insert);
1612 static struct xfrm_policy *
1613 __xfrm_policy_bysel_ctx(struct hlist_head *chain, u32 mark, u32 if_id,
1615 struct xfrm_selector *sel,
1616 struct xfrm_sec_ctx *ctx)
1618 struct xfrm_policy *pol;
1623 hlist_for_each_entry(pol, chain, bydst) {
1624 if (pol->type == type &&
1625 pol->if_id == if_id &&
1626 (mark & pol->mark.m) == pol->mark.v &&
1627 !selector_cmp(sel, &pol->selector) &&
1628 xfrm_sec_ctx_match(ctx, pol->security))
1635 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u32 if_id,
1637 struct xfrm_selector *sel,
1638 struct xfrm_sec_ctx *ctx, int delete,
1641 struct xfrm_pol_inexact_bin *bin = NULL;
1642 struct xfrm_policy *pol, *ret = NULL;
1643 struct hlist_head *chain;
1646 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1647 chain = policy_hash_bysel(net, sel, sel->family, dir);
1649 struct xfrm_pol_inexact_candidates cand;
1652 bin = xfrm_policy_inexact_lookup(net, type,
1653 sel->family, dir, if_id);
1655 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1659 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1662 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1667 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1668 struct xfrm_policy *tmp;
1670 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1676 if (!pol || tmp->pos < pol->pos)
1680 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1687 *err = security_xfrm_policy_delete(pol->security);
1689 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1692 __xfrm_policy_unlink(pol, dir);
1696 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1699 xfrm_policy_kill(ret);
1701 xfrm_policy_inexact_prune_bin(bin);
1704 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1706 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u32 if_id,
1707 u8 type, int dir, u32 id, int delete,
1710 struct xfrm_policy *pol, *ret;
1711 struct hlist_head *chain;
1714 if (xfrm_policy_id2dir(id) != dir)
1718 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1719 chain = net->xfrm.policy_byidx + idx_hash(net, id);
1721 hlist_for_each_entry(pol, chain, byidx) {
1722 if (pol->type == type && pol->index == id &&
1723 pol->if_id == if_id &&
1724 (mark & pol->mark.m) == pol->mark.v) {
1727 *err = security_xfrm_policy_delete(
1730 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1733 __xfrm_policy_unlink(pol, dir);
1739 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1742 xfrm_policy_kill(ret);
1745 EXPORT_SYMBOL(xfrm_policy_byid);
1747 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1749 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1751 struct xfrm_policy *pol;
1754 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1755 if (pol->walk.dead ||
1756 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1760 err = security_xfrm_policy_delete(pol->security);
1762 xfrm_audit_policy_delete(pol, 0, task_valid);
1770 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1776 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1778 int dir, err = 0, cnt = 0;
1779 struct xfrm_policy *pol;
1781 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1783 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1788 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1789 dir = xfrm_policy_id2dir(pol->index);
1790 if (pol->walk.dead ||
1791 dir >= XFRM_POLICY_MAX ||
1795 __xfrm_policy_unlink(pol, dir);
1796 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1798 xfrm_audit_policy_delete(pol, 1, task_valid);
1799 xfrm_policy_kill(pol);
1800 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1804 __xfrm_policy_inexact_flush(net);
1808 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1811 EXPORT_SYMBOL(xfrm_policy_flush);
1813 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1814 int (*func)(struct xfrm_policy *, int, int, void*),
1817 struct xfrm_policy *pol;
1818 struct xfrm_policy_walk_entry *x;
1821 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1822 walk->type != XFRM_POLICY_TYPE_ANY)
1825 if (list_empty(&walk->walk.all) && walk->seq != 0)
1828 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1829 if (list_empty(&walk->walk.all))
1830 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1832 x = list_first_entry(&walk->walk.all,
1833 struct xfrm_policy_walk_entry, all);
1835 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1838 pol = container_of(x, struct xfrm_policy, walk);
1839 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1840 walk->type != pol->type)
1842 error = func(pol, xfrm_policy_id2dir(pol->index),
1845 list_move_tail(&walk->walk.all, &x->all);
1850 if (walk->seq == 0) {
1854 list_del_init(&walk->walk.all);
1856 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1859 EXPORT_SYMBOL(xfrm_policy_walk);
1861 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1863 INIT_LIST_HEAD(&walk->walk.all);
1864 walk->walk.dead = 1;
1868 EXPORT_SYMBOL(xfrm_policy_walk_init);
1870 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1872 if (list_empty(&walk->walk.all))
1875 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1876 list_del(&walk->walk.all);
1877 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1879 EXPORT_SYMBOL(xfrm_policy_walk_done);
1882 * Find policy to apply to this flow.
1884 * Returns 0 if policy found, else an -errno.
1886 static int xfrm_policy_match(const struct xfrm_policy *pol,
1887 const struct flowi *fl,
1888 u8 type, u16 family, int dir, u32 if_id)
1890 const struct xfrm_selector *sel = &pol->selector;
1894 if (pol->family != family ||
1895 pol->if_id != if_id ||
1896 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1900 match = xfrm_selector_match(sel, fl, family);
1902 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1907 static struct xfrm_pol_inexact_node *
1908 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1910 const xfrm_address_t *addr, u16 family)
1912 const struct rb_node *parent;
1916 seq = read_seqcount_begin(count);
1918 parent = rcu_dereference_raw(r->rb_node);
1920 struct xfrm_pol_inexact_node *node;
1923 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1925 delta = xfrm_policy_addr_delta(addr, &node->addr,
1926 node->prefixlen, family);
1928 parent = rcu_dereference_raw(parent->rb_left);
1930 } else if (delta > 0) {
1931 parent = rcu_dereference_raw(parent->rb_right);
1938 if (read_seqcount_retry(count, seq))
1945 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1946 struct xfrm_pol_inexact_bin *b,
1947 const xfrm_address_t *saddr,
1948 const xfrm_address_t *daddr)
1950 struct xfrm_pol_inexact_node *n;
1956 family = b->k.family;
1957 memset(cand, 0, sizeof(*cand));
1958 cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1960 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1963 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1964 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1967 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1970 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1973 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1978 static struct xfrm_pol_inexact_bin *
1979 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1982 struct xfrm_pol_inexact_key k = {
1989 write_pnet(&k.net, net);
1991 return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1992 xfrm_pol_inexact_params);
1995 static struct xfrm_pol_inexact_bin *
1996 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1999 struct xfrm_pol_inexact_bin *bin;
2001 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2004 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2010 static struct xfrm_policy *
2011 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2012 struct xfrm_policy *prefer,
2013 const struct flowi *fl,
2014 u8 type, u16 family, int dir, u32 if_id)
2016 u32 priority = prefer ? prefer->priority : ~0u;
2017 struct xfrm_policy *pol;
2022 hlist_for_each_entry_rcu(pol, chain, bydst) {
2025 if (pol->priority > priority)
2028 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2031 return ERR_PTR(err);
2037 /* matches. Is it older than *prefer? */
2038 if (pol->priority == priority &&
2039 prefer->pos < pol->pos)
2049 static struct xfrm_policy *
2050 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2051 struct xfrm_policy *prefer,
2052 const struct flowi *fl,
2053 u8 type, u16 family, int dir, u32 if_id)
2055 struct xfrm_policy *tmp;
2058 for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2059 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2061 fl, type, family, dir,
2074 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2075 const struct flowi *fl,
2079 struct xfrm_pol_inexact_candidates cand;
2080 const xfrm_address_t *daddr, *saddr;
2081 struct xfrm_pol_inexact_bin *bin;
2082 struct xfrm_policy *pol, *ret;
2083 struct hlist_head *chain;
2084 unsigned int sequence;
2087 daddr = xfrm_flowi_daddr(fl, family);
2088 saddr = xfrm_flowi_saddr(fl, family);
2089 if (unlikely(!daddr || !saddr))
2095 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2096 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2097 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2100 hlist_for_each_entry_rcu(pol, chain, bydst) {
2101 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2114 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2115 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2119 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2120 family, dir, if_id);
2128 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2131 if (ret && !xfrm_pol_hold_rcu(ret))
2139 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2140 const struct flowi *fl,
2141 u16 family, u8 dir, u32 if_id)
2143 #ifdef CONFIG_XFRM_SUB_POLICY
2144 struct xfrm_policy *pol;
2146 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2151 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2155 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2156 const struct flowi *fl,
2157 u16 family, u32 if_id)
2159 struct xfrm_policy *pol;
2163 pol = rcu_dereference(sk->sk_policy[dir]);
2168 if (pol->family != family) {
2173 match = xfrm_selector_match(&pol->selector, fl, family);
2175 if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2176 pol->if_id != if_id) {
2180 err = security_xfrm_policy_lookup(pol->security,
2184 if (!xfrm_pol_hold_rcu(pol))
2186 } else if (err == -ESRCH) {
2199 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2201 struct net *net = xp_net(pol);
2203 list_add(&pol->walk.all, &net->xfrm.policy_all);
2204 net->xfrm.policy_count[dir]++;
2208 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2211 struct net *net = xp_net(pol);
2213 if (list_empty(&pol->walk.all))
2216 /* Socket policies are not hashed. */
2217 if (!hlist_unhashed(&pol->bydst)) {
2218 hlist_del_rcu(&pol->bydst);
2219 hlist_del_init(&pol->bydst_inexact_list);
2220 hlist_del(&pol->byidx);
2223 list_del_init(&pol->walk.all);
2224 net->xfrm.policy_count[dir]--;
2229 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2231 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2234 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2236 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2239 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2241 struct net *net = xp_net(pol);
2243 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2244 pol = __xfrm_policy_unlink(pol, dir);
2245 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2247 xfrm_policy_kill(pol);
2252 EXPORT_SYMBOL(xfrm_policy_delete);
2254 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2256 struct net *net = sock_net(sk);
2257 struct xfrm_policy *old_pol;
2259 #ifdef CONFIG_XFRM_SUB_POLICY
2260 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2264 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2265 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2266 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2268 pol->curlft.add_time = ktime_get_real_seconds();
2269 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2270 xfrm_sk_policy_link(pol, dir);
2272 rcu_assign_pointer(sk->sk_policy[dir], pol);
2275 xfrm_policy_requeue(old_pol, pol);
2277 /* Unlinking succeeds always. This is the only function
2278 * allowed to delete or replace socket policy.
2280 xfrm_sk_policy_unlink(old_pol, dir);
2282 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2285 xfrm_policy_kill(old_pol);
2290 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2292 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2293 struct net *net = xp_net(old);
2296 newp->selector = old->selector;
2297 if (security_xfrm_policy_clone(old->security,
2300 return NULL; /* ENOMEM */
2302 newp->lft = old->lft;
2303 newp->curlft = old->curlft;
2304 newp->mark = old->mark;
2305 newp->if_id = old->if_id;
2306 newp->action = old->action;
2307 newp->flags = old->flags;
2308 newp->xfrm_nr = old->xfrm_nr;
2309 newp->index = old->index;
2310 newp->type = old->type;
2311 newp->family = old->family;
2312 memcpy(newp->xfrm_vec, old->xfrm_vec,
2313 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2314 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2315 xfrm_sk_policy_link(newp, dir);
2316 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2322 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2324 const struct xfrm_policy *p;
2325 struct xfrm_policy *np;
2329 for (i = 0; i < 2; i++) {
2330 p = rcu_dereference(osk->sk_policy[i]);
2332 np = clone_policy(p, i);
2333 if (unlikely(!np)) {
2337 rcu_assign_pointer(sk->sk_policy[i], np);
2345 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2346 xfrm_address_t *remote, unsigned short family, u32 mark)
2349 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2351 if (unlikely(afinfo == NULL))
2353 err = afinfo->get_saddr(net, oif, local, remote, mark);
2358 /* Resolve list of templates for the flow, given policy. */
2361 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2362 struct xfrm_state **xfrm, unsigned short family)
2364 struct net *net = xp_net(policy);
2367 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2368 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2371 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2372 struct xfrm_state *x;
2373 xfrm_address_t *remote = daddr;
2374 xfrm_address_t *local = saddr;
2375 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2377 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2378 tmpl->mode == XFRM_MODE_BEET) {
2379 remote = &tmpl->id.daddr;
2380 local = &tmpl->saddr;
2381 if (xfrm_addr_any(local, tmpl->encap_family)) {
2382 error = xfrm_get_saddr(net, fl->flowi_oif,
2384 tmpl->encap_family, 0);
2391 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2392 family, policy->if_id);
2394 if (x && x->km.state == XFRM_STATE_VALID) {
2401 error = (x->km.state == XFRM_STATE_ERROR ?
2404 } else if (error == -ESRCH) {
2408 if (!tmpl->optional)
2414 for (nx--; nx >= 0; nx--)
2415 xfrm_state_put(xfrm[nx]);
2420 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2421 struct xfrm_state **xfrm, unsigned short family)
2423 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2424 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2430 for (i = 0; i < npols; i++) {
2431 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2436 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2444 /* found states are sorted for outbound processing */
2446 xfrm_state_sort(xfrm, tpp, cnx, family);
2451 for (cnx--; cnx >= 0; cnx--)
2452 xfrm_state_put(tpp[cnx]);
2457 static int xfrm_get_tos(const struct flowi *fl, int family)
2459 if (family == AF_INET)
2460 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2465 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2467 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2468 struct dst_ops *dst_ops;
2469 struct xfrm_dst *xdst;
2472 return ERR_PTR(-EINVAL);
2476 dst_ops = &net->xfrm.xfrm4_dst_ops;
2478 #if IS_ENABLED(CONFIG_IPV6)
2480 dst_ops = &net->xfrm.xfrm6_dst_ops;
2486 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2489 struct dst_entry *dst = &xdst->u.dst;
2491 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2493 xdst = ERR_PTR(-ENOBUFS);
2500 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2503 if (dst->ops->family == AF_INET6) {
2504 struct rt6_info *rt = (struct rt6_info *)dst;
2505 path->path_cookie = rt6_get_cookie(rt);
2506 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2510 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2511 const struct flowi *fl)
2513 const struct xfrm_policy_afinfo *afinfo =
2514 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2520 err = afinfo->fill_dst(xdst, dev, fl);
2528 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2529 * all the metrics... Shortly, bundle a bundle.
2532 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2533 struct xfrm_state **xfrm,
2534 struct xfrm_dst **bundle,
2536 const struct flowi *fl,
2537 struct dst_entry *dst)
2539 const struct xfrm_state_afinfo *afinfo;
2540 const struct xfrm_mode *inner_mode;
2541 struct net *net = xp_net(policy);
2542 unsigned long now = jiffies;
2543 struct net_device *dev;
2544 struct xfrm_dst *xdst_prev = NULL;
2545 struct xfrm_dst *xdst0 = NULL;
2549 int nfheader_len = 0;
2550 int trailer_len = 0;
2552 int family = policy->selector.family;
2553 xfrm_address_t saddr, daddr;
2555 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2557 tos = xfrm_get_tos(fl, family);
2561 for (; i < nx; i++) {
2562 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2563 struct dst_entry *dst1 = &xdst->u.dst;
2565 err = PTR_ERR(xdst);
2575 /* Ref count is taken during xfrm_alloc_dst()
2576 * No need to do dst_clone() on dst1
2578 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2580 if (xfrm[i]->sel.family == AF_UNSPEC) {
2581 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2582 xfrm_af2proto(family));
2584 err = -EAFNOSUPPORT;
2589 inner_mode = &xfrm[i]->inner_mode;
2592 dst_copy_metrics(dst1, dst);
2594 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2597 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2598 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2600 family = xfrm[i]->props.family;
2601 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2602 &saddr, &daddr, family, mark);
2609 dst1->xfrm = xfrm[i];
2610 xdst->xfrm_genid = xfrm[i]->genid;
2612 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2613 dst1->flags |= DST_HOST;
2614 dst1->lastuse = now;
2616 dst1->input = dst_discard;
2619 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2621 dst1->output = afinfo->output;
2623 dst1->output = dst_discard_out;
2628 header_len += xfrm[i]->props.header_len;
2629 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2630 nfheader_len += xfrm[i]->props.header_len;
2631 trailer_len += xfrm[i]->props.trailer_len;
2634 xfrm_dst_set_child(xdst_prev, dst);
2642 xfrm_init_path(xdst0, dst, nfheader_len);
2643 xfrm_init_pmtu(bundle, nx);
2645 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2646 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2647 err = xfrm_fill_dst(xdst_prev, dev, fl);
2651 xdst_prev->u.dst.header_len = header_len;
2652 xdst_prev->u.dst.trailer_len = trailer_len;
2653 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2654 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2657 return &xdst0->u.dst;
2661 xfrm_state_put(xfrm[i]);
2664 dst_release_immediate(&xdst0->u.dst);
2666 return ERR_PTR(err);
2669 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2670 struct xfrm_policy **pols,
2671 int *num_pols, int *num_xfrms)
2675 if (*num_pols == 0 || !pols[0]) {
2680 if (IS_ERR(pols[0]))
2681 return PTR_ERR(pols[0]);
2683 *num_xfrms = pols[0]->xfrm_nr;
2685 #ifdef CONFIG_XFRM_SUB_POLICY
2686 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2687 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2688 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2689 XFRM_POLICY_TYPE_MAIN,
2694 if (IS_ERR(pols[1])) {
2695 xfrm_pols_put(pols, *num_pols);
2696 return PTR_ERR(pols[1]);
2699 (*num_xfrms) += pols[1]->xfrm_nr;
2703 for (i = 0; i < *num_pols; i++) {
2704 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2714 static struct xfrm_dst *
2715 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2716 const struct flowi *fl, u16 family,
2717 struct dst_entry *dst_orig)
2719 struct net *net = xp_net(pols[0]);
2720 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2721 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2722 struct xfrm_dst *xdst;
2723 struct dst_entry *dst;
2726 /* Try to instantiate a bundle */
2727 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2733 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2734 return ERR_PTR(err);
2737 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2739 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2740 return ERR_CAST(dst);
2743 xdst = (struct xfrm_dst *)dst;
2744 xdst->num_xfrms = err;
2745 xdst->num_pols = num_pols;
2746 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2747 xdst->policy_genid = atomic_read(&pols[0]->genid);
2752 static void xfrm_policy_queue_process(struct timer_list *t)
2754 struct sk_buff *skb;
2756 struct dst_entry *dst;
2757 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2758 struct net *net = xp_net(pol);
2759 struct xfrm_policy_queue *pq = &pol->polq;
2761 struct sk_buff_head list;
2763 spin_lock(&pq->hold_queue.lock);
2764 skb = skb_peek(&pq->hold_queue);
2766 spin_unlock(&pq->hold_queue.lock);
2771 xfrm_decode_session(skb, &fl, dst->ops->family);
2772 spin_unlock(&pq->hold_queue.lock);
2774 dst_hold(xfrm_dst_path(dst));
2775 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2779 if (dst->flags & DST_XFRM_QUEUE) {
2782 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2785 pq->timeout = pq->timeout << 1;
2786 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2793 __skb_queue_head_init(&list);
2795 spin_lock(&pq->hold_queue.lock);
2797 skb_queue_splice_init(&pq->hold_queue, &list);
2798 spin_unlock(&pq->hold_queue.lock);
2800 while (!skb_queue_empty(&list)) {
2801 skb = __skb_dequeue(&list);
2803 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2804 dst_hold(xfrm_dst_path(skb_dst(skb)));
2805 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2813 skb_dst_set(skb, dst);
2815 dst_output(net, skb->sk, skb);
2824 skb_queue_purge(&pq->hold_queue);
2828 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2830 unsigned long sched_next;
2831 struct dst_entry *dst = skb_dst(skb);
2832 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2833 struct xfrm_policy *pol = xdst->pols[0];
2834 struct xfrm_policy_queue *pq = &pol->polq;
2836 if (unlikely(skb_fclone_busy(sk, skb))) {
2841 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2848 spin_lock_bh(&pq->hold_queue.lock);
2851 pq->timeout = XFRM_QUEUE_TMO_MIN;
2853 sched_next = jiffies + pq->timeout;
2855 if (del_timer(&pq->hold_timer)) {
2856 if (time_before(pq->hold_timer.expires, sched_next))
2857 sched_next = pq->hold_timer.expires;
2861 __skb_queue_tail(&pq->hold_queue, skb);
2862 if (!mod_timer(&pq->hold_timer, sched_next))
2865 spin_unlock_bh(&pq->hold_queue.lock);
2870 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2871 struct xfrm_flo *xflo,
2872 const struct flowi *fl,
2877 struct net_device *dev;
2878 struct dst_entry *dst;
2879 struct dst_entry *dst1;
2880 struct xfrm_dst *xdst;
2882 xdst = xfrm_alloc_dst(net, family);
2886 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2887 net->xfrm.sysctl_larval_drop ||
2891 dst = xflo->dst_orig;
2892 dst1 = &xdst->u.dst;
2896 dst_copy_metrics(dst1, dst);
2898 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2899 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2900 dst1->lastuse = jiffies;
2902 dst1->input = dst_discard;
2903 dst1->output = xdst_queue_output;
2906 xfrm_dst_set_child(xdst, dst);
2909 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2916 err = xfrm_fill_dst(xdst, dev, fl);
2925 xdst = ERR_PTR(err);
2929 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2930 const struct flowi *fl,
2932 struct xfrm_flo *xflo, u32 if_id)
2934 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2935 int num_pols = 0, num_xfrms = 0, err;
2936 struct xfrm_dst *xdst;
2938 /* Resolve policies to use if we couldn't get them from
2939 * previous cache entry */
2941 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2942 err = xfrm_expand_policies(fl, family, pols,
2943 &num_pols, &num_xfrms);
2949 goto make_dummy_bundle;
2951 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2954 err = PTR_ERR(xdst);
2955 if (err == -EREMOTE) {
2956 xfrm_pols_put(pols, num_pols);
2962 goto make_dummy_bundle;
2963 } else if (xdst == NULL) {
2965 goto make_dummy_bundle;
2971 /* We found policies, but there's no bundles to instantiate:
2972 * either because the policy blocks, has no transformations or
2973 * we could not build template (no xfrm_states).*/
2974 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2976 xfrm_pols_put(pols, num_pols);
2977 return ERR_CAST(xdst);
2979 xdst->num_pols = num_pols;
2980 xdst->num_xfrms = num_xfrms;
2981 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2986 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2988 xfrm_pols_put(pols, num_pols);
2989 return ERR_PTR(err);
2992 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2993 struct dst_entry *dst_orig)
2995 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2996 struct dst_entry *ret;
2999 dst_release(dst_orig);
3000 return ERR_PTR(-EINVAL);
3002 ret = afinfo->blackhole_route(net, dst_orig);
3009 /* Finds/creates a bundle for given flow and if_id
3011 * At the moment we eat a raw IP route. Mostly to speed up lookups
3012 * on interfaces with disabled IPsec.
3014 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3017 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3018 struct dst_entry *dst_orig,
3019 const struct flowi *fl,
3020 const struct sock *sk,
3021 int flags, u32 if_id)
3023 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3024 struct xfrm_dst *xdst;
3025 struct dst_entry *dst, *route;
3026 u16 family = dst_orig->ops->family;
3027 u8 dir = XFRM_POLICY_OUT;
3028 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3034 sk = sk_const_to_full_sk(sk);
3035 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3037 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3039 err = xfrm_expand_policies(fl, family, pols,
3040 &num_pols, &num_xfrms);
3045 if (num_xfrms <= 0) {
3046 drop_pols = num_pols;
3050 xdst = xfrm_resolve_and_create_bundle(
3055 xfrm_pols_put(pols, num_pols);
3056 err = PTR_ERR(xdst);
3057 if (err == -EREMOTE)
3061 } else if (xdst == NULL) {
3063 drop_pols = num_pols;
3067 route = xdst->route;
3072 struct xfrm_flo xflo;
3074 xflo.dst_orig = dst_orig;
3077 /* To accelerate a bit... */
3078 if ((dst_orig->flags & DST_NOXFRM) ||
3079 !net->xfrm.policy_count[XFRM_POLICY_OUT])
3082 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3086 err = PTR_ERR(xdst);
3090 num_pols = xdst->num_pols;
3091 num_xfrms = xdst->num_xfrms;
3092 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3093 route = xdst->route;
3097 if (route == NULL && num_xfrms > 0) {
3098 /* The only case when xfrm_bundle_lookup() returns a
3099 * bundle with null route, is when the template could
3100 * not be resolved. It means policies are there, but
3101 * bundle could not be created, since we don't yet
3102 * have the xfrm_state's. We need to wait for KM to
3103 * negotiate new SA's or bail out with error.*/
3104 if (net->xfrm.sysctl_larval_drop) {
3105 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3112 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3120 if ((flags & XFRM_LOOKUP_ICMP) &&
3121 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3126 for (i = 0; i < num_pols; i++)
3127 pols[i]->curlft.use_time = ktime_get_real_seconds();
3129 if (num_xfrms < 0) {
3130 /* Prohibit the flow */
3131 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3134 } else if (num_xfrms > 0) {
3135 /* Flow transformed */
3136 dst_release(dst_orig);
3138 /* Flow passes untransformed */
3143 xfrm_pols_put(pols, drop_pols);
3144 if (dst && dst->xfrm &&
3145 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3146 dst->flags |= DST_XFRM_TUNNEL;
3150 if (!(flags & XFRM_LOOKUP_ICMP)) {
3158 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3159 dst_release(dst_orig);
3160 xfrm_pols_put(pols, drop_pols);
3161 return ERR_PTR(err);
3163 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3165 /* Main function: finds/creates a bundle for given flow.
3167 * At the moment we eat a raw IP route. Mostly to speed up lookups
3168 * on interfaces with disabled IPsec.
3170 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3171 const struct flowi *fl, const struct sock *sk,
3174 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3176 EXPORT_SYMBOL(xfrm_lookup);
3178 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3179 * Otherwise we may send out blackholed packets.
3181 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3182 const struct flowi *fl,
3183 const struct sock *sk, int flags)
3185 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3186 flags | XFRM_LOOKUP_QUEUE |
3187 XFRM_LOOKUP_KEEP_DST_REF);
3189 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
3190 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3193 dst_release(dst_orig);
3197 EXPORT_SYMBOL(xfrm_lookup_route);
3200 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3202 struct sec_path *sp = skb_sec_path(skb);
3203 struct xfrm_state *x;
3205 if (!sp || idx < 0 || idx >= sp->len)
3208 if (!x->type->reject)
3210 return x->type->reject(x, skb, fl);
3213 /* When skb is transformed back to its "native" form, we have to
3214 * check policy restrictions. At the moment we make this in maximally
3215 * stupid way. Shame on me. :-) Of course, connected sockets must
3216 * have policy cached at them.
3220 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3221 unsigned short family)
3223 if (xfrm_state_kern(x))
3224 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3225 return x->id.proto == tmpl->id.proto &&
3226 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3227 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3228 x->props.mode == tmpl->mode &&
3229 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3230 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3231 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3232 xfrm_state_addr_cmp(tmpl, x, family));
3236 * 0 or more than 0 is returned when validation is succeeded (either bypass
3237 * because of optional transport mode, or next index of the mathced secpath
3238 * state with the template.
3239 * -1 is returned when no matching template is found.
3240 * Otherwise "-2 - errored_index" is returned.
3243 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3244 unsigned short family)
3248 if (tmpl->optional) {
3249 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3253 for (; idx < sp->len; idx++) {
3254 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3256 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3266 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3268 const struct iphdr *iph = ip_hdr(skb);
3270 u8 *xprth = skb_network_header(skb) + ihl * 4;
3271 struct flowi4 *fl4 = &fl->u.ip4;
3274 if (skb_dst(skb) && skb_dst(skb)->dev)
3275 oif = skb_dst(skb)->dev->ifindex;
3277 memset(fl4, 0, sizeof(struct flowi4));
3278 fl4->flowi4_mark = skb->mark;
3279 fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3281 fl4->flowi4_proto = iph->protocol;
3282 fl4->daddr = reverse ? iph->saddr : iph->daddr;
3283 fl4->saddr = reverse ? iph->daddr : iph->saddr;
3284 fl4->flowi4_tos = iph->tos;
3286 if (!ip_is_fragment(iph)) {
3287 switch (iph->protocol) {
3289 case IPPROTO_UDPLITE:
3293 if (xprth + 4 < skb->data ||
3294 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3297 xprth = skb_network_header(skb) + ihl * 4;
3298 ports = (__be16 *)xprth;
3300 fl4->fl4_sport = ports[!!reverse];
3301 fl4->fl4_dport = ports[!reverse];
3305 if (xprth + 2 < skb->data ||
3306 pskb_may_pull(skb, xprth + 2 - skb->data)) {
3309 xprth = skb_network_header(skb) + ihl * 4;
3312 fl4->fl4_icmp_type = icmp[0];
3313 fl4->fl4_icmp_code = icmp[1];
3317 if (xprth + 4 < skb->data ||
3318 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3321 xprth = skb_network_header(skb) + ihl * 4;
3322 ehdr = (__be32 *)xprth;
3324 fl4->fl4_ipsec_spi = ehdr[0];
3328 if (xprth + 8 < skb->data ||
3329 pskb_may_pull(skb, xprth + 8 - skb->data)) {
3332 xprth = skb_network_header(skb) + ihl * 4;
3333 ah_hdr = (__be32 *)xprth;
3335 fl4->fl4_ipsec_spi = ah_hdr[1];
3339 if (xprth + 4 < skb->data ||
3340 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3343 xprth = skb_network_header(skb) + ihl * 4;
3344 ipcomp_hdr = (__be16 *)xprth;
3346 fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
3350 if (xprth + 12 < skb->data ||
3351 pskb_may_pull(skb, xprth + 12 - skb->data)) {
3355 xprth = skb_network_header(skb) + ihl * 4;
3356 greflags = (__be16 *)xprth;
3357 gre_hdr = (__be32 *)xprth;
3359 if (greflags[0] & GRE_KEY) {
3360 if (greflags[0] & GRE_CSUM)
3362 fl4->fl4_gre_key = gre_hdr[1];
3367 fl4->fl4_ipsec_spi = 0;
3373 #if IS_ENABLED(CONFIG_IPV6)
3375 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3377 struct flowi6 *fl6 = &fl->u.ip6;
3379 const struct ipv6hdr *hdr = ipv6_hdr(skb);
3380 u32 offset = sizeof(*hdr);
3381 struct ipv6_opt_hdr *exthdr;
3382 const unsigned char *nh = skb_network_header(skb);
3383 u16 nhoff = IP6CB(skb)->nhoff;
3388 nhoff = offsetof(struct ipv6hdr, nexthdr);
3390 nexthdr = nh[nhoff];
3392 if (skb_dst(skb) && skb_dst(skb)->dev)
3393 oif = skb_dst(skb)->dev->ifindex;
3395 memset(fl6, 0, sizeof(struct flowi6));
3396 fl6->flowi6_mark = skb->mark;
3397 fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3399 fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3400 fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3402 while (nh + offset + sizeof(*exthdr) < skb->data ||
3403 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3404 nh = skb_network_header(skb);
3405 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3408 case NEXTHDR_FRAGMENT:
3411 case NEXTHDR_ROUTING:
3414 offset += ipv6_optlen(exthdr);
3415 nexthdr = exthdr->nexthdr;
3416 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3419 case IPPROTO_UDPLITE:
3423 if (!onlyproto && (nh + offset + 4 < skb->data ||
3424 pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3427 nh = skb_network_header(skb);
3428 ports = (__be16 *)(nh + offset);
3429 fl6->fl6_sport = ports[!!reverse];
3430 fl6->fl6_dport = ports[!reverse];
3432 fl6->flowi6_proto = nexthdr;
3434 case IPPROTO_ICMPV6:
3435 if (!onlyproto && (nh + offset + 2 < skb->data ||
3436 pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3439 nh = skb_network_header(skb);
3440 icmp = (u8 *)(nh + offset);
3441 fl6->fl6_icmp_type = icmp[0];
3442 fl6->fl6_icmp_code = icmp[1];
3444 fl6->flowi6_proto = nexthdr;
3446 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3448 offset += ipv6_optlen(exthdr);
3449 if (!onlyproto && (nh + offset + 3 < skb->data ||
3450 pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3453 nh = skb_network_header(skb);
3454 mh = (struct ip6_mh *)(nh + offset);
3455 fl6->fl6_mh_type = mh->ip6mh_type;
3457 fl6->flowi6_proto = nexthdr;
3460 /* XXX Why are there these headers? */
3465 fl6->fl6_ipsec_spi = 0;
3466 fl6->flowi6_proto = nexthdr;
3473 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3474 unsigned int family, int reverse)
3478 decode_session4(skb, fl, reverse);
3480 #if IS_ENABLED(CONFIG_IPV6)
3482 decode_session6(skb, fl, reverse);
3486 return -EAFNOSUPPORT;
3489 return security_xfrm_decode_session(skb, &fl->flowi_secid);
3491 EXPORT_SYMBOL(__xfrm_decode_session);
3493 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3495 for (; k < sp->len; k++) {
3496 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3505 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3506 unsigned short family)
3508 struct net *net = dev_net(skb->dev);
3509 struct xfrm_policy *pol;
3510 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3517 const struct xfrm_if_cb *ifcb;
3518 struct sec_path *sp;
3523 ifcb = xfrm_if_get_cb();
3526 xi = ifcb->decode_session(skb, family);
3528 if_id = xi->p.if_id;
3534 reverse = dir & ~XFRM_POLICY_MASK;
3535 dir &= XFRM_POLICY_MASK;
3537 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3538 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3542 nf_nat_decode_session(skb, &fl, family);
3544 /* First, check used SA against their selectors. */
3545 sp = skb_sec_path(skb);
3549 for (i = sp->len - 1; i >= 0; i--) {
3550 struct xfrm_state *x = sp->xvec[i];
3551 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3552 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3559 sk = sk_to_full_sk(sk);
3560 if (sk && sk->sk_policy[dir]) {
3561 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3563 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3569 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3572 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3577 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3578 xfrm_secpath_reject(xerr_idx, skb, &fl);
3579 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3585 pol->curlft.use_time = ktime_get_real_seconds();
3589 #ifdef CONFIG_XFRM_SUB_POLICY
3590 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3591 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3593 XFRM_POLICY_IN, if_id);
3595 if (IS_ERR(pols[1])) {
3596 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3599 pols[1]->curlft.use_time = ktime_get_real_seconds();
3605 if (pol->action == XFRM_POLICY_ALLOW) {
3606 static struct sec_path dummy;
3607 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3608 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3609 struct xfrm_tmpl **tpp = tp;
3613 sp = skb_sec_path(skb);
3617 for (pi = 0; pi < npols; pi++) {
3618 if (pols[pi] != pol &&
3619 pols[pi]->action != XFRM_POLICY_ALLOW) {
3620 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3623 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3624 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3627 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3628 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3632 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3636 /* For each tunnel xfrm, find the first matching tmpl.
3637 * For each tmpl before that, find corresponding xfrm.
3638 * Order is _important_. Later we will implement
3639 * some barriers, but at the moment barriers
3640 * are implied between each two transformations.
3642 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3643 k = xfrm_policy_ok(tpp[i], sp, k, family);
3646 /* "-2 - errored_index" returned */
3648 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3653 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3654 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3658 xfrm_pols_put(pols, npols);
3661 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3664 xfrm_secpath_reject(xerr_idx, skb, &fl);
3666 xfrm_pols_put(pols, npols);
3669 EXPORT_SYMBOL(__xfrm_policy_check);
3671 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3673 struct net *net = dev_net(skb->dev);
3675 struct dst_entry *dst;
3678 if (xfrm_decode_session(skb, &fl, family) < 0) {
3679 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3684 if (!skb_dst(skb)) {
3685 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3689 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3694 skb_dst_set(skb, dst);
3697 EXPORT_SYMBOL(__xfrm_route_forward);
3699 /* Optimize later using cookies and generation ids. */
3701 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3703 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3704 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3705 * get validated by dst_ops->check on every use. We do this
3706 * because when a normal route referenced by an XFRM dst is
3707 * obsoleted we do not go looking around for all parent
3708 * referencing XFRM dsts so that we can invalidate them. It
3709 * is just too much work. Instead we make the checks here on
3710 * every use. For example:
3712 * XFRM dst A --> IPv4 dst X
3714 * X is the "xdst->route" of A (X is also the "dst->path" of A
3715 * in this example). If X is marked obsolete, "A" will not
3716 * notice. That's what we are validating here via the
3717 * stale_bundle() check.
3719 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3721 * This will force stale_bundle() to fail on any xdst bundle with
3722 * this dst linked in it.
3724 if (dst->obsolete < 0 && !stale_bundle(dst))
3730 static int stale_bundle(struct dst_entry *dst)
3732 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3735 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3737 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3738 dst->dev = dev_net(dev)->loopback_dev;
3743 EXPORT_SYMBOL(xfrm_dst_ifdown);
3745 static void xfrm_link_failure(struct sk_buff *skb)
3747 /* Impossible. Such dst must be popped before reaches point of failure. */
3750 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3753 if (dst->obsolete) {
3761 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3764 struct xfrm_dst *xdst = bundle[nr];
3765 u32 pmtu, route_mtu_cached;
3766 struct dst_entry *dst;
3769 pmtu = dst_mtu(xfrm_dst_child(dst));
3770 xdst->child_mtu_cached = pmtu;
3772 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3774 route_mtu_cached = dst_mtu(xdst->route);
3775 xdst->route_mtu_cached = route_mtu_cached;
3777 if (pmtu > route_mtu_cached)
3778 pmtu = route_mtu_cached;
3780 dst_metric_set(dst, RTAX_MTU, pmtu);
3784 /* Check that the bundle accepts the flow and its components are
3788 static int xfrm_bundle_ok(struct xfrm_dst *first)
3790 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3791 struct dst_entry *dst = &first->u.dst;
3792 struct xfrm_dst *xdst;
3796 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3797 (dst->dev && !netif_running(dst->dev)))
3800 if (dst->flags & DST_XFRM_QUEUE)
3803 start_from = nr = 0;
3805 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3807 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3809 if (xdst->xfrm_genid != dst->xfrm->genid)
3811 if (xdst->num_pols > 0 &&
3812 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3815 bundle[nr++] = xdst;
3817 mtu = dst_mtu(xfrm_dst_child(dst));
3818 if (xdst->child_mtu_cached != mtu) {
3820 xdst->child_mtu_cached = mtu;
3823 if (!dst_check(xdst->route, xdst->route_cookie))
3825 mtu = dst_mtu(xdst->route);
3826 if (xdst->route_mtu_cached != mtu) {
3828 xdst->route_mtu_cached = mtu;
3831 dst = xfrm_dst_child(dst);
3832 } while (dst->xfrm);
3834 if (likely(!start_from))
3837 xdst = bundle[start_from - 1];
3838 mtu = xdst->child_mtu_cached;
3839 while (start_from--) {
3842 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3843 if (mtu > xdst->route_mtu_cached)
3844 mtu = xdst->route_mtu_cached;
3845 dst_metric_set(dst, RTAX_MTU, mtu);
3849 xdst = bundle[start_from - 1];
3850 xdst->child_mtu_cached = mtu;
3856 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3858 return dst_metric_advmss(xfrm_dst_path(dst));
3861 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3863 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3865 return mtu ? : dst_mtu(xfrm_dst_path(dst));
3868 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3872 const struct xfrm_state *xfrm = dst->xfrm;
3874 dst = xfrm_dst_child(dst);
3876 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3878 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3879 daddr = xfrm->coaddr;
3880 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3881 daddr = &xfrm->id.daddr;
3886 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3887 struct sk_buff *skb,
3890 const struct dst_entry *path = xfrm_dst_path(dst);
3893 daddr = xfrm_get_dst_nexthop(dst, daddr);
3894 return path->ops->neigh_lookup(path, skb, daddr);
3897 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3899 const struct dst_entry *path = xfrm_dst_path(dst);
3901 daddr = xfrm_get_dst_nexthop(dst, daddr);
3902 path->ops->confirm_neigh(path, daddr);
3905 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3909 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3910 return -EAFNOSUPPORT;
3912 spin_lock(&xfrm_policy_afinfo_lock);
3913 if (unlikely(xfrm_policy_afinfo[family] != NULL))
3916 struct dst_ops *dst_ops = afinfo->dst_ops;
3917 if (likely(dst_ops->kmem_cachep == NULL))
3918 dst_ops->kmem_cachep = xfrm_dst_cache;
3919 if (likely(dst_ops->check == NULL))
3920 dst_ops->check = xfrm_dst_check;
3921 if (likely(dst_ops->default_advmss == NULL))
3922 dst_ops->default_advmss = xfrm_default_advmss;
3923 if (likely(dst_ops->mtu == NULL))
3924 dst_ops->mtu = xfrm_mtu;
3925 if (likely(dst_ops->negative_advice == NULL))
3926 dst_ops->negative_advice = xfrm_negative_advice;
3927 if (likely(dst_ops->link_failure == NULL))
3928 dst_ops->link_failure = xfrm_link_failure;
3929 if (likely(dst_ops->neigh_lookup == NULL))
3930 dst_ops->neigh_lookup = xfrm_neigh_lookup;
3931 if (likely(!dst_ops->confirm_neigh))
3932 dst_ops->confirm_neigh = xfrm_confirm_neigh;
3933 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3935 spin_unlock(&xfrm_policy_afinfo_lock);
3939 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3941 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3943 struct dst_ops *dst_ops = afinfo->dst_ops;
3946 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3947 if (xfrm_policy_afinfo[i] != afinfo)
3949 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3955 dst_ops->kmem_cachep = NULL;
3956 dst_ops->check = NULL;
3957 dst_ops->negative_advice = NULL;
3958 dst_ops->link_failure = NULL;
3960 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3962 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3964 spin_lock(&xfrm_if_cb_lock);
3965 rcu_assign_pointer(xfrm_if_cb, ifcb);
3966 spin_unlock(&xfrm_if_cb_lock);
3968 EXPORT_SYMBOL(xfrm_if_register_cb);
3970 void xfrm_if_unregister_cb(void)
3972 RCU_INIT_POINTER(xfrm_if_cb, NULL);
3975 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3977 #ifdef CONFIG_XFRM_STATISTICS
3978 static int __net_init xfrm_statistics_init(struct net *net)
3981 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3982 if (!net->mib.xfrm_statistics)
3984 rv = xfrm_proc_init(net);
3986 free_percpu(net->mib.xfrm_statistics);
3990 static void xfrm_statistics_fini(struct net *net)
3992 xfrm_proc_fini(net);
3993 free_percpu(net->mib.xfrm_statistics);
3996 static int __net_init xfrm_statistics_init(struct net *net)
4001 static void xfrm_statistics_fini(struct net *net)
4006 static int __net_init xfrm_policy_init(struct net *net)
4008 unsigned int hmask, sz;
4011 if (net_eq(net, &init_net)) {
4012 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4013 sizeof(struct xfrm_dst),
4014 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4016 err = rhashtable_init(&xfrm_policy_inexact_table,
4017 &xfrm_pol_inexact_params);
4022 sz = (hmask+1) * sizeof(struct hlist_head);
4024 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4025 if (!net->xfrm.policy_byidx)
4027 net->xfrm.policy_idx_hmask = hmask;
4029 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4030 struct xfrm_policy_hash *htab;
4032 net->xfrm.policy_count[dir] = 0;
4033 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4034 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4036 htab = &net->xfrm.policy_bydst[dir];
4037 htab->table = xfrm_hash_alloc(sz);
4040 htab->hmask = hmask;
4046 net->xfrm.policy_hthresh.lbits4 = 32;
4047 net->xfrm.policy_hthresh.rbits4 = 32;
4048 net->xfrm.policy_hthresh.lbits6 = 128;
4049 net->xfrm.policy_hthresh.rbits6 = 128;
4051 seqlock_init(&net->xfrm.policy_hthresh.lock);
4053 INIT_LIST_HEAD(&net->xfrm.policy_all);
4054 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4055 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4056 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4060 for (dir--; dir >= 0; dir--) {
4061 struct xfrm_policy_hash *htab;
4063 htab = &net->xfrm.policy_bydst[dir];
4064 xfrm_hash_free(htab->table, sz);
4066 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4071 static void xfrm_policy_fini(struct net *net)
4073 struct xfrm_pol_inexact_bin *b, *t;
4077 flush_work(&net->xfrm.policy_hash_work);
4078 #ifdef CONFIG_XFRM_SUB_POLICY
4079 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4081 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4083 WARN_ON(!list_empty(&net->xfrm.policy_all));
4085 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4086 struct xfrm_policy_hash *htab;
4088 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4090 htab = &net->xfrm.policy_bydst[dir];
4091 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4092 WARN_ON(!hlist_empty(htab->table));
4093 xfrm_hash_free(htab->table, sz);
4096 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4097 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4098 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4100 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4101 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4102 __xfrm_policy_inexact_prune_bin(b, true);
4103 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4106 static int __net_init xfrm_net_init(struct net *net)
4110 /* Initialize the per-net locks here */
4111 spin_lock_init(&net->xfrm.xfrm_state_lock);
4112 spin_lock_init(&net->xfrm.xfrm_policy_lock);
4113 mutex_init(&net->xfrm.xfrm_cfg_mutex);
4115 rv = xfrm_statistics_init(net);
4117 goto out_statistics;
4118 rv = xfrm_state_init(net);
4121 rv = xfrm_policy_init(net);
4124 rv = xfrm_sysctl_init(net);
4131 xfrm_policy_fini(net);
4133 xfrm_state_fini(net);
4135 xfrm_statistics_fini(net);
4140 static void __net_exit xfrm_net_exit(struct net *net)
4142 xfrm_sysctl_fini(net);
4143 xfrm_policy_fini(net);
4144 xfrm_state_fini(net);
4145 xfrm_statistics_fini(net);
4148 static struct pernet_operations __net_initdata xfrm_net_ops = {
4149 .init = xfrm_net_init,
4150 .exit = xfrm_net_exit,
4153 void __init xfrm_init(void)
4155 register_pernet_subsys(&xfrm_net_ops);
4157 seqcount_init(&xfrm_policy_hash_generation);
4160 RCU_INIT_POINTER(xfrm_if_cb, NULL);
4164 #ifdef CONFIG_AUDITSYSCALL
4165 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4166 struct audit_buffer *audit_buf)
4168 struct xfrm_sec_ctx *ctx = xp->security;
4169 struct xfrm_selector *sel = &xp->selector;
4172 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4173 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4175 switch (sel->family) {
4177 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4178 if (sel->prefixlen_s != 32)
4179 audit_log_format(audit_buf, " src_prefixlen=%d",
4181 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4182 if (sel->prefixlen_d != 32)
4183 audit_log_format(audit_buf, " dst_prefixlen=%d",
4187 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4188 if (sel->prefixlen_s != 128)
4189 audit_log_format(audit_buf, " src_prefixlen=%d",
4191 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4192 if (sel->prefixlen_d != 128)
4193 audit_log_format(audit_buf, " dst_prefixlen=%d",
4199 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4201 struct audit_buffer *audit_buf;
4203 audit_buf = xfrm_audit_start("SPD-add");
4204 if (audit_buf == NULL)
4206 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4207 audit_log_format(audit_buf, " res=%u", result);
4208 xfrm_audit_common_policyinfo(xp, audit_buf);
4209 audit_log_end(audit_buf);
4211 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4213 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4216 struct audit_buffer *audit_buf;
4218 audit_buf = xfrm_audit_start("SPD-delete");
4219 if (audit_buf == NULL)
4221 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4222 audit_log_format(audit_buf, " res=%u", result);
4223 xfrm_audit_common_policyinfo(xp, audit_buf);
4224 audit_log_end(audit_buf);
4226 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4229 #ifdef CONFIG_XFRM_MIGRATE
4230 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4231 const struct xfrm_selector *sel_tgt)
4233 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4234 if (sel_tgt->family == sel_cmp->family &&
4235 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4237 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4239 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4240 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4244 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4251 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4252 u8 dir, u8 type, struct net *net)
4254 struct xfrm_policy *pol, *ret = NULL;
4255 struct hlist_head *chain;
4258 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4259 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4260 hlist_for_each_entry(pol, chain, bydst) {
4261 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4262 pol->type == type) {
4264 priority = ret->priority;
4268 chain = &net->xfrm.policy_inexact[dir];
4269 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4270 if ((pol->priority >= priority) && ret)
4273 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4274 pol->type == type) {
4282 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4287 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4291 if (t->mode == m->mode && t->id.proto == m->proto &&
4292 (m->reqid == 0 || t->reqid == m->reqid)) {
4294 case XFRM_MODE_TUNNEL:
4295 case XFRM_MODE_BEET:
4296 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4298 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4303 case XFRM_MODE_TRANSPORT:
4304 /* in case of transport mode, template does not store
4305 any IP addresses, hence we just compare mode and
4316 /* update endpoint address(es) of template(s) */
4317 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4318 struct xfrm_migrate *m, int num_migrate)
4320 struct xfrm_migrate *mp;
4323 write_lock_bh(&pol->lock);
4324 if (unlikely(pol->walk.dead)) {
4325 /* target policy has been deleted */
4326 write_unlock_bh(&pol->lock);
4330 for (i = 0; i < pol->xfrm_nr; i++) {
4331 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4332 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4335 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4336 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4338 /* update endpoints */
4339 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4340 sizeof(pol->xfrm_vec[i].id.daddr));
4341 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4342 sizeof(pol->xfrm_vec[i].saddr));
4343 pol->xfrm_vec[i].encap_family = mp->new_family;
4345 atomic_inc(&pol->genid);
4349 write_unlock_bh(&pol->lock);
4357 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4361 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4364 for (i = 0; i < num_migrate; i++) {
4365 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4366 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4369 /* check if there is any duplicated entry */
4370 for (j = i + 1; j < num_migrate; j++) {
4371 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4372 sizeof(m[i].old_daddr)) &&
4373 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4374 sizeof(m[i].old_saddr)) &&
4375 m[i].proto == m[j].proto &&
4376 m[i].mode == m[j].mode &&
4377 m[i].reqid == m[j].reqid &&
4378 m[i].old_family == m[j].old_family)
4386 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4387 struct xfrm_migrate *m, int num_migrate,
4388 struct xfrm_kmaddress *k, struct net *net,
4389 struct xfrm_encap_tmpl *encap)
4391 int i, err, nx_cur = 0, nx_new = 0;
4392 struct xfrm_policy *pol = NULL;
4393 struct xfrm_state *x, *xc;
4394 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4395 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4396 struct xfrm_migrate *mp;
4398 /* Stage 0 - sanity checks */
4399 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4402 if (dir >= XFRM_POLICY_MAX) {
4407 /* Stage 1 - find policy */
4408 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4413 /* Stage 2 - find and update state(s) */
4414 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4415 if ((x = xfrm_migrate_state_find(mp, net))) {
4418 xc = xfrm_state_migrate(x, mp, encap);
4429 /* Stage 3 - update policy */
4430 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4433 /* Stage 4 - delete old state(s) */
4435 xfrm_states_put(x_cur, nx_cur);
4436 xfrm_states_delete(x_cur, nx_cur);
4439 /* Stage 5 - announce */
4440 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4452 xfrm_states_put(x_cur, nx_cur);
4454 xfrm_states_delete(x_new, nx_new);
4458 EXPORT_SYMBOL(xfrm_migrate);