1 // SPDX-License-Identifier: GPL-2.0-only
7 * Kazunori MIYAZAWA @USAGI
8 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
10 * Kazunori MIYAZAWA @USAGI
12 * Split up af-specific portion
13 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
17 #include <linux/err.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/workqueue.h>
23 #include <linux/notifier.h>
24 #include <linux/netdevice.h>
25 #include <linux/netfilter.h>
26 #include <linux/module.h>
27 #include <linux/cache.h>
28 #include <linux/cpu.h>
29 #include <linux/audit.h>
30 #include <linux/rhashtable.h>
31 #include <linux/if_tunnel.h>
34 #include <net/inet_ecn.h>
38 #if IS_ENABLED(CONFIG_IPV6_MIP6)
41 #ifdef CONFIG_XFRM_STATISTICS
44 #ifdef CONFIG_XFRM_ESPINTCP
45 #include <net/espintcp.h>
48 #include "xfrm_hash.h"
50 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
51 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
52 #define XFRM_MAX_QUEUE_LEN 100
55 struct dst_entry *dst_orig;
59 /* prefixes smaller than this are stored in lists, not trees. */
60 #define INEXACT_PREFIXLEN_IPV4 16
61 #define INEXACT_PREFIXLEN_IPV6 48
63 struct xfrm_pol_inexact_node {
73 /* the policies matching this node, can be empty list */
74 struct hlist_head hhead;
77 /* xfrm inexact policy search tree:
78 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
80 * +---- root_d: sorted by daddr:prefix
82 * | xfrm_pol_inexact_node
84 * | +- root: sorted by saddr/prefix
86 * | | xfrm_pol_inexact_node
90 * | | + hhead: saddr:daddr policies
92 * | +- coarse policies and all any:daddr policies
94 * +---- root_s: sorted by saddr:prefix
96 * | xfrm_pol_inexact_node
100 * | + hhead: saddr:any policies
102 * +---- coarse policies and all any:any policies
104 * Lookups return four candidate lists:
105 * 1. any:any list from top-level xfrm_pol_inexact_bin
106 * 2. any:daddr list from daddr tree
107 * 3. saddr:daddr list from 2nd level daddr tree
108 * 4. saddr:any list from saddr tree
110 * This result set then needs to be searched for the policy with
111 * the lowest priority. If two results have same prio, youngest one wins.
114 struct xfrm_pol_inexact_key {
121 struct xfrm_pol_inexact_bin {
122 struct xfrm_pol_inexact_key k;
123 struct rhash_head head;
124 /* list containing '*:*' policies */
125 struct hlist_head hhead;
127 seqcount_spinlock_t count;
128 /* tree sorted by daddr/prefix */
129 struct rb_root root_d;
131 /* tree sorted by saddr/prefix */
132 struct rb_root root_s;
134 /* slow path below */
135 struct list_head inexact_bins;
139 enum xfrm_pol_inexact_candidate_type {
148 struct xfrm_pol_inexact_candidates {
149 struct hlist_head *res[XFRM_POL_CAND_MAX];
152 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
153 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
155 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
156 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
159 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
161 static struct rhashtable xfrm_policy_inexact_table;
162 static const struct rhashtable_params xfrm_pol_inexact_params;
164 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
165 static int stale_bundle(struct dst_entry *dst);
166 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
167 static void xfrm_policy_queue_process(struct timer_list *t);
169 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
170 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
173 static struct xfrm_pol_inexact_bin *
174 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
177 static struct xfrm_pol_inexact_bin *
178 xfrm_policy_inexact_lookup_rcu(struct net *net,
179 u8 type, u16 family, u8 dir, u32 if_id);
180 static struct xfrm_policy *
181 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
183 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
184 struct xfrm_policy *policy);
187 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
188 struct xfrm_pol_inexact_bin *b,
189 const xfrm_address_t *saddr,
190 const xfrm_address_t *daddr);
192 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
194 return refcount_inc_not_zero(&policy->refcnt);
198 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
200 const struct flowi4 *fl4 = &fl->u.ip4;
202 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
203 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
204 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
205 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
206 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
207 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
211 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
213 const struct flowi6 *fl6 = &fl->u.ip6;
215 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
216 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
217 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
218 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
219 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
220 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
223 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
224 unsigned short family)
228 return __xfrm4_selector_match(sel, fl);
230 return __xfrm6_selector_match(sel, fl);
235 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
237 const struct xfrm_policy_afinfo *afinfo;
239 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
242 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
243 if (unlikely(!afinfo))
248 /* Called with rcu_read_lock(). */
249 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
251 return rcu_dereference(xfrm_if_cb);
254 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
255 const xfrm_address_t *saddr,
256 const xfrm_address_t *daddr,
257 int family, u32 mark)
259 const struct xfrm_policy_afinfo *afinfo;
260 struct dst_entry *dst;
262 afinfo = xfrm_policy_get_afinfo(family);
263 if (unlikely(afinfo == NULL))
264 return ERR_PTR(-EAFNOSUPPORT);
266 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
272 EXPORT_SYMBOL(__xfrm_dst_lookup);
274 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
276 xfrm_address_t *prev_saddr,
277 xfrm_address_t *prev_daddr,
278 int family, u32 mark)
280 struct net *net = xs_net(x);
281 xfrm_address_t *saddr = &x->props.saddr;
282 xfrm_address_t *daddr = &x->id.daddr;
283 struct dst_entry *dst;
285 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
289 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
294 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
297 if (prev_saddr != saddr)
298 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
299 if (prev_daddr != daddr)
300 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
306 static inline unsigned long make_jiffies(long secs)
308 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
309 return MAX_SCHEDULE_TIMEOUT-1;
314 static void xfrm_policy_timer(struct timer_list *t)
316 struct xfrm_policy *xp = from_timer(xp, t, timer);
317 time64_t now = ktime_get_real_seconds();
318 time64_t next = TIME64_MAX;
322 read_lock(&xp->lock);
324 if (unlikely(xp->walk.dead))
327 dir = xfrm_policy_id2dir(xp->index);
329 if (xp->lft.hard_add_expires_seconds) {
330 time64_t tmo = xp->lft.hard_add_expires_seconds +
331 xp->curlft.add_time - now;
337 if (xp->lft.hard_use_expires_seconds) {
338 time64_t tmo = xp->lft.hard_use_expires_seconds +
339 (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
345 if (xp->lft.soft_add_expires_seconds) {
346 time64_t tmo = xp->lft.soft_add_expires_seconds +
347 xp->curlft.add_time - now;
350 tmo = XFRM_KM_TIMEOUT;
355 if (xp->lft.soft_use_expires_seconds) {
356 time64_t tmo = xp->lft.soft_use_expires_seconds +
357 (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
360 tmo = XFRM_KM_TIMEOUT;
367 km_policy_expired(xp, dir, 0, 0);
368 if (next != TIME64_MAX &&
369 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
373 read_unlock(&xp->lock);
378 read_unlock(&xp->lock);
379 if (!xfrm_policy_delete(xp, dir))
380 km_policy_expired(xp, dir, 1, 0);
384 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
388 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
390 struct xfrm_policy *policy;
392 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
395 write_pnet(&policy->xp_net, net);
396 INIT_LIST_HEAD(&policy->walk.all);
397 INIT_HLIST_NODE(&policy->bydst_inexact_list);
398 INIT_HLIST_NODE(&policy->bydst);
399 INIT_HLIST_NODE(&policy->byidx);
400 rwlock_init(&policy->lock);
401 refcount_set(&policy->refcnt, 1);
402 skb_queue_head_init(&policy->polq.hold_queue);
403 timer_setup(&policy->timer, xfrm_policy_timer, 0);
404 timer_setup(&policy->polq.hold_timer,
405 xfrm_policy_queue_process, 0);
409 EXPORT_SYMBOL(xfrm_policy_alloc);
411 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
413 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
415 security_xfrm_policy_free(policy->security);
419 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
421 void xfrm_policy_destroy(struct xfrm_policy *policy)
423 BUG_ON(!policy->walk.dead);
425 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
428 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
430 EXPORT_SYMBOL(xfrm_policy_destroy);
432 /* Rule must be locked. Release descendant resources, announce
433 * entry dead. The rule must be unlinked from lists to the moment.
436 static void xfrm_policy_kill(struct xfrm_policy *policy)
438 write_lock_bh(&policy->lock);
439 policy->walk.dead = 1;
440 write_unlock_bh(&policy->lock);
442 atomic_inc(&policy->genid);
444 if (del_timer(&policy->polq.hold_timer))
445 xfrm_pol_put(policy);
446 skb_queue_purge(&policy->polq.hold_queue);
448 if (del_timer(&policy->timer))
449 xfrm_pol_put(policy);
451 xfrm_pol_put(policy);
454 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
456 static inline unsigned int idx_hash(struct net *net, u32 index)
458 return __idx_hash(index, net->xfrm.policy_idx_hmask);
461 /* calculate policy hash thresholds */
462 static void __get_hash_thresh(struct net *net,
463 unsigned short family, int dir,
464 u8 *dbits, u8 *sbits)
468 *dbits = net->xfrm.policy_bydst[dir].dbits4;
469 *sbits = net->xfrm.policy_bydst[dir].sbits4;
473 *dbits = net->xfrm.policy_bydst[dir].dbits6;
474 *sbits = net->xfrm.policy_bydst[dir].sbits6;
483 static struct hlist_head *policy_hash_bysel(struct net *net,
484 const struct xfrm_selector *sel,
485 unsigned short family, int dir)
487 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
492 __get_hash_thresh(net, family, dir, &dbits, &sbits);
493 hash = __sel_hash(sel, family, hmask, dbits, sbits);
495 if (hash == hmask + 1)
498 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
499 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
502 static struct hlist_head *policy_hash_direct(struct net *net,
503 const xfrm_address_t *daddr,
504 const xfrm_address_t *saddr,
505 unsigned short family, int dir)
507 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
512 __get_hash_thresh(net, family, dir, &dbits, &sbits);
513 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
515 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
516 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
519 static void xfrm_dst_hash_transfer(struct net *net,
520 struct hlist_head *list,
521 struct hlist_head *ndsttable,
522 unsigned int nhashmask,
525 struct hlist_node *tmp, *entry0 = NULL;
526 struct xfrm_policy *pol;
532 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
535 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
536 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
537 pol->family, nhashmask, dbits, sbits);
539 hlist_del_rcu(&pol->bydst);
540 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
545 hlist_del_rcu(&pol->bydst);
546 hlist_add_behind_rcu(&pol->bydst, entry0);
548 entry0 = &pol->bydst;
550 if (!hlist_empty(list)) {
556 static void xfrm_idx_hash_transfer(struct hlist_head *list,
557 struct hlist_head *nidxtable,
558 unsigned int nhashmask)
560 struct hlist_node *tmp;
561 struct xfrm_policy *pol;
563 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
566 h = __idx_hash(pol->index, nhashmask);
567 hlist_add_head(&pol->byidx, nidxtable+h);
571 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
573 return ((old_hmask + 1) << 1) - 1;
576 static void xfrm_bydst_resize(struct net *net, int dir)
578 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
579 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
580 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
581 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
582 struct hlist_head *odst;
588 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
589 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
591 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
592 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
594 for (i = hmask; i >= 0; i--)
595 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
597 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
598 net->xfrm.policy_bydst[dir].hmask = nhashmask;
600 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
601 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
605 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
608 static void xfrm_byidx_resize(struct net *net, int total)
610 unsigned int hmask = net->xfrm.policy_idx_hmask;
611 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
612 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
613 struct hlist_head *oidx = net->xfrm.policy_byidx;
614 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
620 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
622 for (i = hmask; i >= 0; i--)
623 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
625 net->xfrm.policy_byidx = nidx;
626 net->xfrm.policy_idx_hmask = nhashmask;
628 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
630 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
633 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
635 unsigned int cnt = net->xfrm.policy_count[dir];
636 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
641 if ((hmask + 1) < xfrm_policy_hashmax &&
648 static inline int xfrm_byidx_should_resize(struct net *net, int total)
650 unsigned int hmask = net->xfrm.policy_idx_hmask;
652 if ((hmask + 1) < xfrm_policy_hashmax &&
659 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
661 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
662 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
663 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
664 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
665 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
666 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
667 si->spdhcnt = net->xfrm.policy_idx_hmask;
668 si->spdhmcnt = xfrm_policy_hashmax;
670 EXPORT_SYMBOL(xfrm_spd_getinfo);
672 static DEFINE_MUTEX(hash_resize_mutex);
673 static void xfrm_hash_resize(struct work_struct *work)
675 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
678 mutex_lock(&hash_resize_mutex);
681 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
682 if (xfrm_bydst_should_resize(net, dir, &total))
683 xfrm_bydst_resize(net, dir);
685 if (xfrm_byidx_should_resize(net, total))
686 xfrm_byidx_resize(net, total);
688 mutex_unlock(&hash_resize_mutex);
691 /* Make sure *pol can be inserted into fastbin.
692 * Useful to check that later insert requests will be successful
693 * (provided xfrm_policy_lock is held throughout).
695 static struct xfrm_pol_inexact_bin *
696 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
698 struct xfrm_pol_inexact_bin *bin, *prev;
699 struct xfrm_pol_inexact_key k = {
700 .family = pol->family,
705 struct net *net = xp_net(pol);
707 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
709 write_pnet(&k.net, net);
710 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
711 xfrm_pol_inexact_params);
715 bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
720 INIT_HLIST_HEAD(&bin->hhead);
721 bin->root_d = RB_ROOT;
722 bin->root_s = RB_ROOT;
723 seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
725 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
727 xfrm_pol_inexact_params);
729 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
735 return IS_ERR(prev) ? NULL : prev;
738 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
739 int family, u8 prefixlen)
741 if (xfrm_addr_any(addr, family))
744 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
747 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
754 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
756 const xfrm_address_t *addr;
757 bool saddr_any, daddr_any;
760 addr = &policy->selector.saddr;
761 prefixlen = policy->selector.prefixlen_s;
763 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
766 addr = &policy->selector.daddr;
767 prefixlen = policy->selector.prefixlen_d;
768 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
771 return saddr_any && daddr_any;
774 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
775 const xfrm_address_t *addr, u8 prefixlen)
778 node->prefixlen = prefixlen;
781 static struct xfrm_pol_inexact_node *
782 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
784 struct xfrm_pol_inexact_node *node;
786 node = kzalloc(sizeof(*node), GFP_ATOMIC);
788 xfrm_pol_inexact_node_init(node, addr, prefixlen);
793 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
794 const xfrm_address_t *b,
795 u8 prefixlen, u16 family)
798 unsigned int pdw, pbi;
805 mask = ~0U << (32 - prefixlen);
806 ma = ntohl(a->a4) & mask;
807 mb = ntohl(b->a4) & mask;
814 pdw = prefixlen >> 5;
815 pbi = prefixlen & 0x1f;
818 delta = memcmp(a->a6, b->a6, pdw << 2);
823 mask = ~0U << (32 - pbi);
824 ma = ntohl(a->a6[pdw]) & mask;
825 mb = ntohl(b->a6[pdw]) & mask;
839 static void xfrm_policy_inexact_list_reinsert(struct net *net,
840 struct xfrm_pol_inexact_node *n,
843 unsigned int matched_s, matched_d;
844 struct xfrm_policy *policy, *p;
849 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
850 struct hlist_node *newpos = NULL;
851 bool matches_s, matches_d;
853 if (policy->walk.dead || !policy->bydst_reinsert)
856 WARN_ON_ONCE(policy->family != family);
858 policy->bydst_reinsert = false;
859 hlist_for_each_entry(p, &n->hhead, bydst) {
860 if (policy->priority > p->priority)
862 else if (policy->priority == p->priority &&
863 policy->pos > p->pos)
870 hlist_add_behind_rcu(&policy->bydst, newpos);
872 hlist_add_head_rcu(&policy->bydst, &n->hhead);
874 /* paranoia checks follow.
875 * Check that the reinserted policy matches at least
876 * saddr or daddr for current node prefix.
878 * Matching both is fine, matching saddr in one policy
879 * (but not daddr) and then matching only daddr in another
882 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
886 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
890 if (matches_s && matches_d)
893 WARN_ON_ONCE(!matches_s && !matches_d);
898 WARN_ON_ONCE(matched_s && matched_d);
902 static void xfrm_policy_inexact_node_reinsert(struct net *net,
903 struct xfrm_pol_inexact_node *n,
907 struct xfrm_pol_inexact_node *node;
908 struct rb_node **p, *parent;
910 /* we should not have another subtree here */
911 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
920 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
922 prefixlen = min(node->prefixlen, n->prefixlen);
924 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
927 p = &parent->rb_left;
928 } else if (delta > 0) {
929 p = &parent->rb_right;
931 bool same_prefixlen = node->prefixlen == n->prefixlen;
932 struct xfrm_policy *tmp;
934 hlist_for_each_entry(tmp, &n->hhead, bydst) {
935 tmp->bydst_reinsert = true;
936 hlist_del_rcu(&tmp->bydst);
939 node->prefixlen = prefixlen;
941 xfrm_policy_inexact_list_reinsert(net, node, family);
943 if (same_prefixlen) {
955 rb_link_node_rcu(&n->node, parent, p);
956 rb_insert_color(&n->node, new);
959 /* merge nodes v and n */
960 static void xfrm_policy_inexact_node_merge(struct net *net,
961 struct xfrm_pol_inexact_node *v,
962 struct xfrm_pol_inexact_node *n,
965 struct xfrm_pol_inexact_node *node;
966 struct xfrm_policy *tmp;
967 struct rb_node *rnode;
969 /* To-be-merged node v has a subtree.
971 * Dismantle it and insert its nodes to n->root.
973 while ((rnode = rb_first(&v->root)) != NULL) {
974 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
975 rb_erase(&node->node, &v->root);
976 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
980 hlist_for_each_entry(tmp, &v->hhead, bydst) {
981 tmp->bydst_reinsert = true;
982 hlist_del_rcu(&tmp->bydst);
985 xfrm_policy_inexact_list_reinsert(net, n, family);
988 static struct xfrm_pol_inexact_node *
989 xfrm_policy_inexact_insert_node(struct net *net,
990 struct rb_root *root,
991 xfrm_address_t *addr,
992 u16 family, u8 prefixlen, u8 dir)
994 struct xfrm_pol_inexact_node *cached = NULL;
995 struct rb_node **p, *parent = NULL;
996 struct xfrm_pol_inexact_node *node;
1003 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1005 delta = xfrm_policy_addr_delta(addr, &node->addr,
1008 if (delta == 0 && prefixlen >= node->prefixlen) {
1009 WARN_ON_ONCE(cached); /* ipsec policies got lost */
1014 p = &parent->rb_left;
1016 p = &parent->rb_right;
1018 if (prefixlen < node->prefixlen) {
1019 delta = xfrm_policy_addr_delta(addr, &node->addr,
1025 /* This node is a subnet of the new prefix. It needs
1026 * to be removed and re-inserted with the smaller
1027 * prefix and all nodes that are now also covered
1028 * by the reduced prefixlen.
1030 rb_erase(&node->node, root);
1033 xfrm_pol_inexact_node_init(node, addr,
1037 /* This node also falls within the new
1038 * prefixlen. Merge the to-be-reinserted
1039 * node and this one.
1041 xfrm_policy_inexact_node_merge(net, node,
1043 kfree_rcu(node, rcu);
1054 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1059 rb_link_node_rcu(&node->node, parent, p);
1060 rb_insert_color(&node->node, root);
1065 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1067 struct xfrm_pol_inexact_node *node;
1068 struct rb_node *rn = rb_first(r);
1071 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1073 xfrm_policy_inexact_gc_tree(&node->root, rm);
1076 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1081 rb_erase(&node->node, r);
1082 kfree_rcu(node, rcu);
1086 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1088 write_seqcount_begin(&b->count);
1089 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1090 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1091 write_seqcount_end(&b->count);
1093 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1094 !hlist_empty(&b->hhead)) {
1095 WARN_ON_ONCE(net_exit);
1099 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1100 xfrm_pol_inexact_params) == 0) {
1101 list_del(&b->inexact_bins);
1106 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1108 struct net *net = read_pnet(&b->k.net);
1110 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1111 __xfrm_policy_inexact_prune_bin(b, false);
1112 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1115 static void __xfrm_policy_inexact_flush(struct net *net)
1117 struct xfrm_pol_inexact_bin *bin, *t;
1119 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1121 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1122 __xfrm_policy_inexact_prune_bin(bin, false);
1125 static struct hlist_head *
1126 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1127 struct xfrm_policy *policy, u8 dir)
1129 struct xfrm_pol_inexact_node *n;
1132 net = xp_net(policy);
1133 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1135 if (xfrm_policy_inexact_insert_use_any_list(policy))
1138 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1140 policy->selector.prefixlen_d)) {
1141 write_seqcount_begin(&bin->count);
1142 n = xfrm_policy_inexact_insert_node(net,
1144 &policy->selector.saddr,
1146 policy->selector.prefixlen_s,
1148 write_seqcount_end(&bin->count);
1155 /* daddr is fixed */
1156 write_seqcount_begin(&bin->count);
1157 n = xfrm_policy_inexact_insert_node(net,
1159 &policy->selector.daddr,
1161 policy->selector.prefixlen_d, dir);
1162 write_seqcount_end(&bin->count);
1166 /* saddr is wildcard */
1167 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1169 policy->selector.prefixlen_s))
1172 write_seqcount_begin(&bin->count);
1173 n = xfrm_policy_inexact_insert_node(net,
1175 &policy->selector.saddr,
1177 policy->selector.prefixlen_s, dir);
1178 write_seqcount_end(&bin->count);
1185 static struct xfrm_policy *
1186 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1188 struct xfrm_pol_inexact_bin *bin;
1189 struct xfrm_policy *delpol;
1190 struct hlist_head *chain;
1193 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1195 return ERR_PTR(-ENOMEM);
1197 net = xp_net(policy);
1198 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1200 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1202 __xfrm_policy_inexact_prune_bin(bin, false);
1203 return ERR_PTR(-ENOMEM);
1206 delpol = xfrm_policy_insert_list(chain, policy, excl);
1207 if (delpol && excl) {
1208 __xfrm_policy_inexact_prune_bin(bin, false);
1209 return ERR_PTR(-EEXIST);
1212 chain = &net->xfrm.policy_inexact[dir];
1213 xfrm_policy_insert_inexact_list(chain, policy);
1216 __xfrm_policy_inexact_prune_bin(bin, false);
1221 static void xfrm_hash_rebuild(struct work_struct *work)
1223 struct net *net = container_of(work, struct net,
1224 xfrm.policy_hthresh.work);
1226 struct xfrm_policy *pol;
1227 struct xfrm_policy *policy;
1228 struct hlist_head *chain;
1229 struct hlist_head *odst;
1230 struct hlist_node *newpos;
1234 u8 lbits4, rbits4, lbits6, rbits6;
1236 mutex_lock(&hash_resize_mutex);
1238 /* read selector prefixlen thresholds */
1240 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1242 lbits4 = net->xfrm.policy_hthresh.lbits4;
1243 rbits4 = net->xfrm.policy_hthresh.rbits4;
1244 lbits6 = net->xfrm.policy_hthresh.lbits6;
1245 rbits6 = net->xfrm.policy_hthresh.rbits6;
1246 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1248 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1249 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1251 /* make sure that we can insert the indirect policies again before
1252 * we start with destructive action.
1254 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1255 struct xfrm_pol_inexact_bin *bin;
1258 if (policy->walk.dead)
1261 dir = xfrm_policy_id2dir(policy->index);
1262 if (dir >= XFRM_POLICY_MAX)
1265 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1266 if (policy->family == AF_INET) {
1274 if (policy->family == AF_INET) {
1283 if (policy->selector.prefixlen_d < dbits ||
1284 policy->selector.prefixlen_s < sbits)
1287 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1291 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1295 /* reset the bydst and inexact table in all directions */
1296 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1297 struct hlist_node *n;
1299 hlist_for_each_entry_safe(policy, n,
1300 &net->xfrm.policy_inexact[dir],
1301 bydst_inexact_list) {
1302 hlist_del_rcu(&policy->bydst);
1303 hlist_del_init(&policy->bydst_inexact_list);
1306 hmask = net->xfrm.policy_bydst[dir].hmask;
1307 odst = net->xfrm.policy_bydst[dir].table;
1308 for (i = hmask; i >= 0; i--) {
1309 hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1310 hlist_del_rcu(&policy->bydst);
1312 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1313 /* dir out => dst = remote, src = local */
1314 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1315 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1316 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1317 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1319 /* dir in/fwd => dst = local, src = remote */
1320 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1321 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1322 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1323 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1327 /* re-insert all policies by order of creation */
1328 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1329 if (policy->walk.dead)
1331 dir = xfrm_policy_id2dir(policy->index);
1332 if (dir >= XFRM_POLICY_MAX) {
1333 /* skip socket policies */
1337 chain = policy_hash_bysel(net, &policy->selector,
1338 policy->family, dir);
1341 void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1343 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1347 hlist_for_each_entry(pol, chain, bydst) {
1348 if (policy->priority >= pol->priority)
1349 newpos = &pol->bydst;
1354 hlist_add_behind_rcu(&policy->bydst, newpos);
1356 hlist_add_head_rcu(&policy->bydst, chain);
1360 __xfrm_policy_inexact_flush(net);
1361 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1362 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1364 mutex_unlock(&hash_resize_mutex);
1367 void xfrm_policy_hash_rebuild(struct net *net)
1369 schedule_work(&net->xfrm.policy_hthresh.work);
1371 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1373 /* Generate new index... KAME seems to generate them ordered by cost
1374 * of an absolute inpredictability of ordering of rules. This will not pass. */
1375 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1378 struct hlist_head *list;
1379 struct xfrm_policy *p;
1384 idx = (net->xfrm.idx_generator | dir);
1385 net->xfrm.idx_generator += 8;
1393 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1395 hlist_for_each_entry(p, list, byidx) {
1396 if (p->index == idx) {
1406 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1408 u32 *p1 = (u32 *) s1;
1409 u32 *p2 = (u32 *) s2;
1410 int len = sizeof(struct xfrm_selector) / sizeof(u32);
1413 for (i = 0; i < len; i++) {
1421 static void xfrm_policy_requeue(struct xfrm_policy *old,
1422 struct xfrm_policy *new)
1424 struct xfrm_policy_queue *pq = &old->polq;
1425 struct sk_buff_head list;
1427 if (skb_queue_empty(&pq->hold_queue))
1430 __skb_queue_head_init(&list);
1432 spin_lock_bh(&pq->hold_queue.lock);
1433 skb_queue_splice_init(&pq->hold_queue, &list);
1434 if (del_timer(&pq->hold_timer))
1436 spin_unlock_bh(&pq->hold_queue.lock);
1440 spin_lock_bh(&pq->hold_queue.lock);
1441 skb_queue_splice(&list, &pq->hold_queue);
1442 pq->timeout = XFRM_QUEUE_TMO_MIN;
1443 if (!mod_timer(&pq->hold_timer, jiffies))
1445 spin_unlock_bh(&pq->hold_queue.lock);
1448 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1449 struct xfrm_policy *pol)
1451 return mark->v == pol->mark.v && mark->m == pol->mark.m;
1454 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1456 const struct xfrm_pol_inexact_key *k = data;
1457 u32 a = k->type << 24 | k->dir << 16 | k->family;
1459 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1463 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1465 const struct xfrm_pol_inexact_bin *b = data;
1467 return xfrm_pol_bin_key(&b->k, 0, seed);
1470 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1473 const struct xfrm_pol_inexact_key *key = arg->key;
1474 const struct xfrm_pol_inexact_bin *b = ptr;
1477 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1480 ret = b->k.dir ^ key->dir;
1484 ret = b->k.type ^ key->type;
1488 ret = b->k.family ^ key->family;
1492 return b->k.if_id ^ key->if_id;
1495 static const struct rhashtable_params xfrm_pol_inexact_params = {
1496 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head),
1497 .hashfn = xfrm_pol_bin_key,
1498 .obj_hashfn = xfrm_pol_bin_obj,
1499 .obj_cmpfn = xfrm_pol_bin_cmp,
1500 .automatic_shrinking = true,
1503 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1504 struct xfrm_policy *policy)
1506 struct xfrm_policy *pol, *delpol = NULL;
1507 struct hlist_node *newpos = NULL;
1510 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1511 if (pol->type == policy->type &&
1512 pol->if_id == policy->if_id &&
1513 !selector_cmp(&pol->selector, &policy->selector) &&
1514 xfrm_policy_mark_match(&policy->mark, pol) &&
1515 xfrm_sec_ctx_match(pol->security, policy->security) &&
1518 if (policy->priority > pol->priority)
1520 } else if (policy->priority >= pol->priority) {
1521 newpos = &pol->bydst_inexact_list;
1529 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1531 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1533 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1539 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1540 struct xfrm_policy *policy,
1543 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1545 hlist_for_each_entry(pol, chain, bydst) {
1546 if (pol->type == policy->type &&
1547 pol->if_id == policy->if_id &&
1548 !selector_cmp(&pol->selector, &policy->selector) &&
1549 xfrm_policy_mark_match(&policy->mark, pol) &&
1550 xfrm_sec_ctx_match(pol->security, policy->security) &&
1553 return ERR_PTR(-EEXIST);
1555 if (policy->priority > pol->priority)
1557 } else if (policy->priority >= pol->priority) {
1566 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1568 hlist_add_head_rcu(&policy->bydst, chain);
1573 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1575 struct net *net = xp_net(policy);
1576 struct xfrm_policy *delpol;
1577 struct hlist_head *chain;
1579 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1580 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1582 delpol = xfrm_policy_insert_list(chain, policy, excl);
1584 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1586 if (IS_ERR(delpol)) {
1587 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1588 return PTR_ERR(delpol);
1591 __xfrm_policy_link(policy, dir);
1593 /* After previous checking, family can either be AF_INET or AF_INET6 */
1594 if (policy->family == AF_INET)
1595 rt_genid_bump_ipv4(net);
1597 rt_genid_bump_ipv6(net);
1600 xfrm_policy_requeue(delpol, policy);
1601 __xfrm_policy_unlink(delpol, dir);
1603 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1604 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1605 policy->curlft.add_time = ktime_get_real_seconds();
1606 policy->curlft.use_time = 0;
1607 if (!mod_timer(&policy->timer, jiffies + HZ))
1608 xfrm_pol_hold(policy);
1609 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1612 xfrm_policy_kill(delpol);
1613 else if (xfrm_bydst_should_resize(net, dir, NULL))
1614 schedule_work(&net->xfrm.policy_hash_work);
1618 EXPORT_SYMBOL(xfrm_policy_insert);
1620 static struct xfrm_policy *
1621 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1622 u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1623 struct xfrm_sec_ctx *ctx)
1625 struct xfrm_policy *pol;
1630 hlist_for_each_entry(pol, chain, bydst) {
1631 if (pol->type == type &&
1632 pol->if_id == if_id &&
1633 xfrm_policy_mark_match(mark, pol) &&
1634 !selector_cmp(sel, &pol->selector) &&
1635 xfrm_sec_ctx_match(ctx, pol->security))
1642 struct xfrm_policy *
1643 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1644 u8 type, int dir, struct xfrm_selector *sel,
1645 struct xfrm_sec_ctx *ctx, int delete, int *err)
1647 struct xfrm_pol_inexact_bin *bin = NULL;
1648 struct xfrm_policy *pol, *ret = NULL;
1649 struct hlist_head *chain;
1652 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1653 chain = policy_hash_bysel(net, sel, sel->family, dir);
1655 struct xfrm_pol_inexact_candidates cand;
1658 bin = xfrm_policy_inexact_lookup(net, type,
1659 sel->family, dir, if_id);
1661 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1665 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1668 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1673 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1674 struct xfrm_policy *tmp;
1676 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1682 if (!pol || tmp->pos < pol->pos)
1686 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1693 *err = security_xfrm_policy_delete(pol->security);
1695 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1698 __xfrm_policy_unlink(pol, dir);
1702 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1705 xfrm_policy_kill(ret);
1707 xfrm_policy_inexact_prune_bin(bin);
1710 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1712 struct xfrm_policy *
1713 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1714 u8 type, int dir, u32 id, int delete, int *err)
1716 struct xfrm_policy *pol, *ret;
1717 struct hlist_head *chain;
1720 if (xfrm_policy_id2dir(id) != dir)
1724 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1725 chain = net->xfrm.policy_byidx + idx_hash(net, id);
1727 hlist_for_each_entry(pol, chain, byidx) {
1728 if (pol->type == type && pol->index == id &&
1729 pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1732 *err = security_xfrm_policy_delete(
1735 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1738 __xfrm_policy_unlink(pol, dir);
1744 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1747 xfrm_policy_kill(ret);
1750 EXPORT_SYMBOL(xfrm_policy_byid);
1752 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1754 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1756 struct xfrm_policy *pol;
1759 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1760 if (pol->walk.dead ||
1761 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1765 err = security_xfrm_policy_delete(pol->security);
1767 xfrm_audit_policy_delete(pol, 0, task_valid);
1775 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1781 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1783 int dir, err = 0, cnt = 0;
1784 struct xfrm_policy *pol;
1786 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1788 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1793 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1797 dir = xfrm_policy_id2dir(pol->index);
1798 if (dir >= XFRM_POLICY_MAX ||
1802 __xfrm_policy_unlink(pol, dir);
1803 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1805 xfrm_audit_policy_delete(pol, 1, task_valid);
1806 xfrm_policy_kill(pol);
1807 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1811 __xfrm_policy_inexact_flush(net);
1815 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1818 EXPORT_SYMBOL(xfrm_policy_flush);
1820 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1821 int (*func)(struct xfrm_policy *, int, int, void*),
1824 struct xfrm_policy *pol;
1825 struct xfrm_policy_walk_entry *x;
1828 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1829 walk->type != XFRM_POLICY_TYPE_ANY)
1832 if (list_empty(&walk->walk.all) && walk->seq != 0)
1835 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1836 if (list_empty(&walk->walk.all))
1837 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1839 x = list_first_entry(&walk->walk.all,
1840 struct xfrm_policy_walk_entry, all);
1842 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1845 pol = container_of(x, struct xfrm_policy, walk);
1846 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1847 walk->type != pol->type)
1849 error = func(pol, xfrm_policy_id2dir(pol->index),
1852 list_move_tail(&walk->walk.all, &x->all);
1857 if (walk->seq == 0) {
1861 list_del_init(&walk->walk.all);
1863 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1866 EXPORT_SYMBOL(xfrm_policy_walk);
1868 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1870 INIT_LIST_HEAD(&walk->walk.all);
1871 walk->walk.dead = 1;
1875 EXPORT_SYMBOL(xfrm_policy_walk_init);
1877 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1879 if (list_empty(&walk->walk.all))
1882 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1883 list_del(&walk->walk.all);
1884 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1886 EXPORT_SYMBOL(xfrm_policy_walk_done);
1889 * Find policy to apply to this flow.
1891 * Returns 0 if policy found, else an -errno.
1893 static int xfrm_policy_match(const struct xfrm_policy *pol,
1894 const struct flowi *fl,
1895 u8 type, u16 family, u32 if_id)
1897 const struct xfrm_selector *sel = &pol->selector;
1901 if (pol->family != family ||
1902 pol->if_id != if_id ||
1903 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1907 match = xfrm_selector_match(sel, fl, family);
1909 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1913 static struct xfrm_pol_inexact_node *
1914 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1915 seqcount_spinlock_t *count,
1916 const xfrm_address_t *addr, u16 family)
1918 const struct rb_node *parent;
1922 seq = read_seqcount_begin(count);
1924 parent = rcu_dereference_raw(r->rb_node);
1926 struct xfrm_pol_inexact_node *node;
1929 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1931 delta = xfrm_policy_addr_delta(addr, &node->addr,
1932 node->prefixlen, family);
1934 parent = rcu_dereference_raw(parent->rb_left);
1936 } else if (delta > 0) {
1937 parent = rcu_dereference_raw(parent->rb_right);
1944 if (read_seqcount_retry(count, seq))
1951 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1952 struct xfrm_pol_inexact_bin *b,
1953 const xfrm_address_t *saddr,
1954 const xfrm_address_t *daddr)
1956 struct xfrm_pol_inexact_node *n;
1962 family = b->k.family;
1963 memset(cand, 0, sizeof(*cand));
1964 cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1966 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1969 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1970 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1973 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1976 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1979 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1984 static struct xfrm_pol_inexact_bin *
1985 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1988 struct xfrm_pol_inexact_key k = {
1995 write_pnet(&k.net, net);
1997 return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1998 xfrm_pol_inexact_params);
2001 static struct xfrm_pol_inexact_bin *
2002 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2005 struct xfrm_pol_inexact_bin *bin;
2007 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2010 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2016 static struct xfrm_policy *
2017 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2018 struct xfrm_policy *prefer,
2019 const struct flowi *fl,
2020 u8 type, u16 family, u32 if_id)
2022 u32 priority = prefer ? prefer->priority : ~0u;
2023 struct xfrm_policy *pol;
2028 hlist_for_each_entry_rcu(pol, chain, bydst) {
2031 if (pol->priority > priority)
2034 err = xfrm_policy_match(pol, fl, type, family, if_id);
2037 return ERR_PTR(err);
2043 /* matches. Is it older than *prefer? */
2044 if (pol->priority == priority &&
2045 prefer->pos < pol->pos)
2055 static struct xfrm_policy *
2056 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2057 struct xfrm_policy *prefer,
2058 const struct flowi *fl,
2059 u8 type, u16 family, u32 if_id)
2061 struct xfrm_policy *tmp;
2064 for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2065 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2067 fl, type, family, if_id);
2079 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2080 const struct flowi *fl,
2084 struct xfrm_pol_inexact_candidates cand;
2085 const xfrm_address_t *daddr, *saddr;
2086 struct xfrm_pol_inexact_bin *bin;
2087 struct xfrm_policy *pol, *ret;
2088 struct hlist_head *chain;
2089 unsigned int sequence;
2092 daddr = xfrm_flowi_daddr(fl, family);
2093 saddr = xfrm_flowi_saddr(fl, family);
2094 if (unlikely(!daddr || !saddr))
2100 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2101 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2102 } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2105 hlist_for_each_entry_rcu(pol, chain, bydst) {
2106 err = xfrm_policy_match(pol, fl, type, family, if_id);
2119 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2120 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2124 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2133 if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2136 if (ret && !xfrm_pol_hold_rcu(ret))
2144 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2145 const struct flowi *fl,
2146 u16 family, u8 dir, u32 if_id)
2148 #ifdef CONFIG_XFRM_SUB_POLICY
2149 struct xfrm_policy *pol;
2151 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2156 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2160 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2161 const struct flowi *fl,
2162 u16 family, u32 if_id)
2164 struct xfrm_policy *pol;
2168 pol = rcu_dereference(sk->sk_policy[dir]);
2173 if (pol->family != family) {
2178 match = xfrm_selector_match(&pol->selector, fl, family);
2180 if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v ||
2181 pol->if_id != if_id) {
2185 err = security_xfrm_policy_lookup(pol->security,
2188 if (!xfrm_pol_hold_rcu(pol))
2190 } else if (err == -ESRCH) {
2203 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2205 struct net *net = xp_net(pol);
2207 list_add(&pol->walk.all, &net->xfrm.policy_all);
2208 net->xfrm.policy_count[dir]++;
2212 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2215 struct net *net = xp_net(pol);
2217 if (list_empty(&pol->walk.all))
2220 /* Socket policies are not hashed. */
2221 if (!hlist_unhashed(&pol->bydst)) {
2222 hlist_del_rcu(&pol->bydst);
2223 hlist_del_init(&pol->bydst_inexact_list);
2224 hlist_del(&pol->byidx);
2227 list_del_init(&pol->walk.all);
2228 net->xfrm.policy_count[dir]--;
2233 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2235 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2238 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2240 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2243 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2245 struct net *net = xp_net(pol);
2247 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2248 pol = __xfrm_policy_unlink(pol, dir);
2249 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2251 xfrm_policy_kill(pol);
2256 EXPORT_SYMBOL(xfrm_policy_delete);
2258 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2260 struct net *net = sock_net(sk);
2261 struct xfrm_policy *old_pol;
2263 #ifdef CONFIG_XFRM_SUB_POLICY
2264 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2268 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2269 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2270 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2272 pol->curlft.add_time = ktime_get_real_seconds();
2273 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2274 xfrm_sk_policy_link(pol, dir);
2276 rcu_assign_pointer(sk->sk_policy[dir], pol);
2279 xfrm_policy_requeue(old_pol, pol);
2281 /* Unlinking succeeds always. This is the only function
2282 * allowed to delete or replace socket policy.
2284 xfrm_sk_policy_unlink(old_pol, dir);
2286 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2289 xfrm_policy_kill(old_pol);
2294 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2296 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2297 struct net *net = xp_net(old);
2300 newp->selector = old->selector;
2301 if (security_xfrm_policy_clone(old->security,
2304 return NULL; /* ENOMEM */
2306 newp->lft = old->lft;
2307 newp->curlft = old->curlft;
2308 newp->mark = old->mark;
2309 newp->if_id = old->if_id;
2310 newp->action = old->action;
2311 newp->flags = old->flags;
2312 newp->xfrm_nr = old->xfrm_nr;
2313 newp->index = old->index;
2314 newp->type = old->type;
2315 newp->family = old->family;
2316 memcpy(newp->xfrm_vec, old->xfrm_vec,
2317 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2318 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2319 xfrm_sk_policy_link(newp, dir);
2320 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2326 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2328 const struct xfrm_policy *p;
2329 struct xfrm_policy *np;
2333 for (i = 0; i < 2; i++) {
2334 p = rcu_dereference(osk->sk_policy[i]);
2336 np = clone_policy(p, i);
2337 if (unlikely(!np)) {
2341 rcu_assign_pointer(sk->sk_policy[i], np);
2349 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2350 xfrm_address_t *remote, unsigned short family, u32 mark)
2353 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2355 if (unlikely(afinfo == NULL))
2357 err = afinfo->get_saddr(net, oif, local, remote, mark);
2362 /* Resolve list of templates for the flow, given policy. */
2365 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2366 struct xfrm_state **xfrm, unsigned short family)
2368 struct net *net = xp_net(policy);
2371 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2372 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2375 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2376 struct xfrm_state *x;
2377 xfrm_address_t *remote = daddr;
2378 xfrm_address_t *local = saddr;
2379 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2381 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2382 tmpl->mode == XFRM_MODE_BEET) {
2383 remote = &tmpl->id.daddr;
2384 local = &tmpl->saddr;
2385 if (xfrm_addr_any(local, tmpl->encap_family)) {
2386 error = xfrm_get_saddr(net, fl->flowi_oif,
2388 tmpl->encap_family, 0);
2395 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2396 family, policy->if_id);
2398 if (x && x->km.state == XFRM_STATE_VALID) {
2405 error = (x->km.state == XFRM_STATE_ERROR ?
2408 } else if (error == -ESRCH) {
2412 if (!tmpl->optional)
2418 for (nx--; nx >= 0; nx--)
2419 xfrm_state_put(xfrm[nx]);
2424 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2425 struct xfrm_state **xfrm, unsigned short family)
2427 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2428 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2434 for (i = 0; i < npols; i++) {
2435 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2440 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2448 /* found states are sorted for outbound processing */
2450 xfrm_state_sort(xfrm, tpp, cnx, family);
2455 for (cnx--; cnx >= 0; cnx--)
2456 xfrm_state_put(tpp[cnx]);
2461 static int xfrm_get_tos(const struct flowi *fl, int family)
2463 if (family == AF_INET)
2464 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2469 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2471 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2472 struct dst_ops *dst_ops;
2473 struct xfrm_dst *xdst;
2476 return ERR_PTR(-EINVAL);
2480 dst_ops = &net->xfrm.xfrm4_dst_ops;
2482 #if IS_ENABLED(CONFIG_IPV6)
2484 dst_ops = &net->xfrm.xfrm6_dst_ops;
2490 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2493 memset_after(xdst, 0, u.dst);
2495 xdst = ERR_PTR(-ENOBUFS);
2502 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2505 if (dst->ops->family == AF_INET6) {
2506 struct rt6_info *rt = (struct rt6_info *)dst;
2507 path->path_cookie = rt6_get_cookie(rt);
2508 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2512 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2513 const struct flowi *fl)
2515 const struct xfrm_policy_afinfo *afinfo =
2516 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2522 err = afinfo->fill_dst(xdst, dev, fl);
2530 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2531 * all the metrics... Shortly, bundle a bundle.
2534 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2535 struct xfrm_state **xfrm,
2536 struct xfrm_dst **bundle,
2538 const struct flowi *fl,
2539 struct dst_entry *dst)
2541 const struct xfrm_state_afinfo *afinfo;
2542 const struct xfrm_mode *inner_mode;
2543 struct net *net = xp_net(policy);
2544 unsigned long now = jiffies;
2545 struct net_device *dev;
2546 struct xfrm_dst *xdst_prev = NULL;
2547 struct xfrm_dst *xdst0 = NULL;
2551 int nfheader_len = 0;
2552 int trailer_len = 0;
2554 int family = policy->selector.family;
2555 xfrm_address_t saddr, daddr;
2557 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2559 tos = xfrm_get_tos(fl, family);
2563 for (; i < nx; i++) {
2564 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2565 struct dst_entry *dst1 = &xdst->u.dst;
2567 err = PTR_ERR(xdst);
2577 /* Ref count is taken during xfrm_alloc_dst()
2578 * No need to do dst_clone() on dst1
2580 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2582 if (xfrm[i]->sel.family == AF_UNSPEC) {
2583 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2584 xfrm_af2proto(family));
2586 err = -EAFNOSUPPORT;
2591 inner_mode = &xfrm[i]->inner_mode;
2594 dst_copy_metrics(dst1, dst);
2596 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2600 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2601 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2603 family = xfrm[i]->props.family;
2604 oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2605 dst = xfrm_dst_lookup(xfrm[i], tos, oif,
2606 &saddr, &daddr, family, mark);
2613 dst1->xfrm = xfrm[i];
2614 xdst->xfrm_genid = xfrm[i]->genid;
2616 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2617 dst1->lastuse = now;
2619 dst1->input = dst_discard;
2622 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2624 dst1->output = afinfo->output;
2626 dst1->output = dst_discard_out;
2631 header_len += xfrm[i]->props.header_len;
2632 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2633 nfheader_len += xfrm[i]->props.header_len;
2634 trailer_len += xfrm[i]->props.trailer_len;
2637 xfrm_dst_set_child(xdst_prev, dst);
2645 xfrm_init_path(xdst0, dst, nfheader_len);
2646 xfrm_init_pmtu(bundle, nx);
2648 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2649 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2650 err = xfrm_fill_dst(xdst_prev, dev, fl);
2654 xdst_prev->u.dst.header_len = header_len;
2655 xdst_prev->u.dst.trailer_len = trailer_len;
2656 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2657 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2660 return &xdst0->u.dst;
2664 xfrm_state_put(xfrm[i]);
2667 dst_release_immediate(&xdst0->u.dst);
2669 return ERR_PTR(err);
2672 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2673 struct xfrm_policy **pols,
2674 int *num_pols, int *num_xfrms)
2678 if (*num_pols == 0 || !pols[0]) {
2683 if (IS_ERR(pols[0])) {
2685 return PTR_ERR(pols[0]);
2688 *num_xfrms = pols[0]->xfrm_nr;
2690 #ifdef CONFIG_XFRM_SUB_POLICY
2691 if (pols[0]->action == XFRM_POLICY_ALLOW &&
2692 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2693 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2694 XFRM_POLICY_TYPE_MAIN,
2699 if (IS_ERR(pols[1])) {
2700 xfrm_pols_put(pols, *num_pols);
2702 return PTR_ERR(pols[1]);
2705 (*num_xfrms) += pols[1]->xfrm_nr;
2709 for (i = 0; i < *num_pols; i++) {
2710 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2720 static struct xfrm_dst *
2721 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2722 const struct flowi *fl, u16 family,
2723 struct dst_entry *dst_orig)
2725 struct net *net = xp_net(pols[0]);
2726 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2727 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2728 struct xfrm_dst *xdst;
2729 struct dst_entry *dst;
2732 /* Try to instantiate a bundle */
2733 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2739 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2740 return ERR_PTR(err);
2743 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2745 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2746 return ERR_CAST(dst);
2749 xdst = (struct xfrm_dst *)dst;
2750 xdst->num_xfrms = err;
2751 xdst->num_pols = num_pols;
2752 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2753 xdst->policy_genid = atomic_read(&pols[0]->genid);
2758 static void xfrm_policy_queue_process(struct timer_list *t)
2760 struct sk_buff *skb;
2762 struct dst_entry *dst;
2763 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2764 struct net *net = xp_net(pol);
2765 struct xfrm_policy_queue *pq = &pol->polq;
2767 struct sk_buff_head list;
2770 spin_lock(&pq->hold_queue.lock);
2771 skb = skb_peek(&pq->hold_queue);
2773 spin_unlock(&pq->hold_queue.lock);
2779 /* Fixup the mark to support VTI. */
2780 skb_mark = skb->mark;
2781 skb->mark = pol->mark.v;
2782 xfrm_decode_session(skb, &fl, dst->ops->family);
2783 skb->mark = skb_mark;
2784 spin_unlock(&pq->hold_queue.lock);
2786 dst_hold(xfrm_dst_path(dst));
2787 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2791 if (dst->flags & DST_XFRM_QUEUE) {
2794 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2797 pq->timeout = pq->timeout << 1;
2798 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2805 __skb_queue_head_init(&list);
2807 spin_lock(&pq->hold_queue.lock);
2809 skb_queue_splice_init(&pq->hold_queue, &list);
2810 spin_unlock(&pq->hold_queue.lock);
2812 while (!skb_queue_empty(&list)) {
2813 skb = __skb_dequeue(&list);
2815 /* Fixup the mark to support VTI. */
2816 skb_mark = skb->mark;
2817 skb->mark = pol->mark.v;
2818 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2819 skb->mark = skb_mark;
2821 dst_hold(xfrm_dst_path(skb_dst(skb)));
2822 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2830 skb_dst_set(skb, dst);
2832 dst_output(net, skb->sk, skb);
2841 skb_queue_purge(&pq->hold_queue);
2845 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2847 unsigned long sched_next;
2848 struct dst_entry *dst = skb_dst(skb);
2849 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2850 struct xfrm_policy *pol = xdst->pols[0];
2851 struct xfrm_policy_queue *pq = &pol->polq;
2853 if (unlikely(skb_fclone_busy(sk, skb))) {
2858 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2865 spin_lock_bh(&pq->hold_queue.lock);
2868 pq->timeout = XFRM_QUEUE_TMO_MIN;
2870 sched_next = jiffies + pq->timeout;
2872 if (del_timer(&pq->hold_timer)) {
2873 if (time_before(pq->hold_timer.expires, sched_next))
2874 sched_next = pq->hold_timer.expires;
2878 __skb_queue_tail(&pq->hold_queue, skb);
2879 if (!mod_timer(&pq->hold_timer, sched_next))
2882 spin_unlock_bh(&pq->hold_queue.lock);
2887 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2888 struct xfrm_flo *xflo,
2889 const struct flowi *fl,
2894 struct net_device *dev;
2895 struct dst_entry *dst;
2896 struct dst_entry *dst1;
2897 struct xfrm_dst *xdst;
2899 xdst = xfrm_alloc_dst(net, family);
2903 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2904 net->xfrm.sysctl_larval_drop ||
2908 dst = xflo->dst_orig;
2909 dst1 = &xdst->u.dst;
2913 dst_copy_metrics(dst1, dst);
2915 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2916 dst1->flags |= DST_XFRM_QUEUE;
2917 dst1->lastuse = jiffies;
2919 dst1->input = dst_discard;
2920 dst1->output = xdst_queue_output;
2923 xfrm_dst_set_child(xdst, dst);
2926 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2933 err = xfrm_fill_dst(xdst, dev, fl);
2942 xdst = ERR_PTR(err);
2946 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2947 const struct flowi *fl,
2949 struct xfrm_flo *xflo, u32 if_id)
2951 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2952 int num_pols = 0, num_xfrms = 0, err;
2953 struct xfrm_dst *xdst;
2955 /* Resolve policies to use if we couldn't get them from
2956 * previous cache entry */
2958 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2959 err = xfrm_expand_policies(fl, family, pols,
2960 &num_pols, &num_xfrms);
2966 goto make_dummy_bundle;
2968 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2971 err = PTR_ERR(xdst);
2972 if (err == -EREMOTE) {
2973 xfrm_pols_put(pols, num_pols);
2979 goto make_dummy_bundle;
2980 } else if (xdst == NULL) {
2982 goto make_dummy_bundle;
2988 /* We found policies, but there's no bundles to instantiate:
2989 * either because the policy blocks, has no transformations or
2990 * we could not build template (no xfrm_states).*/
2991 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2993 xfrm_pols_put(pols, num_pols);
2994 return ERR_CAST(xdst);
2996 xdst->num_pols = num_pols;
2997 xdst->num_xfrms = num_xfrms;
2998 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
3003 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3005 xfrm_pols_put(pols, num_pols);
3006 return ERR_PTR(err);
3009 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3010 struct dst_entry *dst_orig)
3012 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3013 struct dst_entry *ret;
3016 dst_release(dst_orig);
3017 return ERR_PTR(-EINVAL);
3019 ret = afinfo->blackhole_route(net, dst_orig);
3026 /* Finds/creates a bundle for given flow and if_id
3028 * At the moment we eat a raw IP route. Mostly to speed up lookups
3029 * on interfaces with disabled IPsec.
3031 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3034 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3035 struct dst_entry *dst_orig,
3036 const struct flowi *fl,
3037 const struct sock *sk,
3038 int flags, u32 if_id)
3040 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3041 struct xfrm_dst *xdst;
3042 struct dst_entry *dst, *route;
3043 u16 family = dst_orig->ops->family;
3044 u8 dir = XFRM_POLICY_OUT;
3045 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3051 sk = sk_const_to_full_sk(sk);
3052 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3054 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3056 err = xfrm_expand_policies(fl, family, pols,
3057 &num_pols, &num_xfrms);
3062 if (num_xfrms <= 0) {
3063 drop_pols = num_pols;
3067 xdst = xfrm_resolve_and_create_bundle(
3072 xfrm_pols_put(pols, num_pols);
3073 err = PTR_ERR(xdst);
3074 if (err == -EREMOTE)
3078 } else if (xdst == NULL) {
3080 drop_pols = num_pols;
3084 route = xdst->route;
3089 struct xfrm_flo xflo;
3091 xflo.dst_orig = dst_orig;
3094 /* To accelerate a bit... */
3095 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3096 !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3099 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3103 err = PTR_ERR(xdst);
3107 num_pols = xdst->num_pols;
3108 num_xfrms = xdst->num_xfrms;
3109 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3110 route = xdst->route;
3114 if (route == NULL && num_xfrms > 0) {
3115 /* The only case when xfrm_bundle_lookup() returns a
3116 * bundle with null route, is when the template could
3117 * not be resolved. It means policies are there, but
3118 * bundle could not be created, since we don't yet
3119 * have the xfrm_state's. We need to wait for KM to
3120 * negotiate new SA's or bail out with error.*/
3121 if (net->xfrm.sysctl_larval_drop) {
3122 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3129 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3137 if ((flags & XFRM_LOOKUP_ICMP) &&
3138 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3143 for (i = 0; i < num_pols; i++)
3144 WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds());
3146 if (num_xfrms < 0) {
3147 /* Prohibit the flow */
3148 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3151 } else if (num_xfrms > 0) {
3152 /* Flow transformed */
3153 dst_release(dst_orig);
3155 /* Flow passes untransformed */
3160 xfrm_pols_put(pols, drop_pols);
3161 if (dst && dst->xfrm &&
3162 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3163 dst->flags |= DST_XFRM_TUNNEL;
3167 if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3168 net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3172 if (!(flags & XFRM_LOOKUP_ICMP)) {
3180 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3181 dst_release(dst_orig);
3182 xfrm_pols_put(pols, drop_pols);
3183 return ERR_PTR(err);
3185 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3187 /* Main function: finds/creates a bundle for given flow.
3189 * At the moment we eat a raw IP route. Mostly to speed up lookups
3190 * on interfaces with disabled IPsec.
3192 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3193 const struct flowi *fl, const struct sock *sk,
3196 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3198 EXPORT_SYMBOL(xfrm_lookup);
3200 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3201 * Otherwise we may send out blackholed packets.
3203 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3204 const struct flowi *fl,
3205 const struct sock *sk, int flags)
3207 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3208 flags | XFRM_LOOKUP_QUEUE |
3209 XFRM_LOOKUP_KEEP_DST_REF);
3211 if (PTR_ERR(dst) == -EREMOTE)
3212 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3215 dst_release(dst_orig);
3219 EXPORT_SYMBOL(xfrm_lookup_route);
3222 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3224 struct sec_path *sp = skb_sec_path(skb);
3225 struct xfrm_state *x;
3227 if (!sp || idx < 0 || idx >= sp->len)
3230 if (!x->type->reject)
3232 return x->type->reject(x, skb, fl);
3235 /* When skb is transformed back to its "native" form, we have to
3236 * check policy restrictions. At the moment we make this in maximally
3237 * stupid way. Shame on me. :-) Of course, connected sockets must
3238 * have policy cached at them.
3242 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3243 unsigned short family, u32 if_id)
3245 if (xfrm_state_kern(x))
3246 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3247 return x->id.proto == tmpl->id.proto &&
3248 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3249 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3250 x->props.mode == tmpl->mode &&
3251 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3252 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3253 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3254 xfrm_state_addr_cmp(tmpl, x, family)) &&
3255 (if_id == 0 || if_id == x->if_id);
3259 * 0 or more than 0 is returned when validation is succeeded (either bypass
3260 * because of optional transport mode, or next index of the matched secpath
3261 * state with the template.
3262 * -1 is returned when no matching template is found.
3263 * Otherwise "-2 - errored_index" is returned.
3266 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3267 unsigned short family, u32 if_id)
3271 if (tmpl->optional) {
3272 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3276 for (; idx < sp->len; idx++) {
3277 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3279 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3280 if (idx < sp->verified_cnt) {
3281 /* Secpath entry previously verified, consider optional and
3282 * continue searching
3296 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3298 const struct iphdr *iph = ip_hdr(skb);
3300 u8 *xprth = skb_network_header(skb) + ihl * 4;
3301 struct flowi4 *fl4 = &fl->u.ip4;
3304 if (skb_dst(skb) && skb_dst(skb)->dev)
3305 oif = skb_dst(skb)->dev->ifindex;
3307 memset(fl4, 0, sizeof(struct flowi4));
3308 fl4->flowi4_mark = skb->mark;
3309 fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3311 fl4->flowi4_proto = iph->protocol;
3312 fl4->daddr = reverse ? iph->saddr : iph->daddr;
3313 fl4->saddr = reverse ? iph->daddr : iph->saddr;
3314 fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3316 if (!ip_is_fragment(iph)) {
3317 switch (iph->protocol) {
3319 case IPPROTO_UDPLITE:
3323 if (xprth + 4 < skb->data ||
3324 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3327 xprth = skb_network_header(skb) + ihl * 4;
3328 ports = (__be16 *)xprth;
3330 fl4->fl4_sport = ports[!!reverse];
3331 fl4->fl4_dport = ports[!reverse];
3335 if (xprth + 2 < skb->data ||
3336 pskb_may_pull(skb, xprth + 2 - skb->data)) {
3339 xprth = skb_network_header(skb) + ihl * 4;
3342 fl4->fl4_icmp_type = icmp[0];
3343 fl4->fl4_icmp_code = icmp[1];
3347 if (xprth + 12 < skb->data ||
3348 pskb_may_pull(skb, xprth + 12 - skb->data)) {
3352 xprth = skb_network_header(skb) + ihl * 4;
3353 greflags = (__be16 *)xprth;
3354 gre_hdr = (__be32 *)xprth;
3356 if (greflags[0] & GRE_KEY) {
3357 if (greflags[0] & GRE_CSUM)
3359 fl4->fl4_gre_key = gre_hdr[1];
3369 #if IS_ENABLED(CONFIG_IPV6)
3371 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3373 struct flowi6 *fl6 = &fl->u.ip6;
3375 const struct ipv6hdr *hdr = ipv6_hdr(skb);
3376 u32 offset = sizeof(*hdr);
3377 struct ipv6_opt_hdr *exthdr;
3378 const unsigned char *nh = skb_network_header(skb);
3379 u16 nhoff = IP6CB(skb)->nhoff;
3384 nhoff = offsetof(struct ipv6hdr, nexthdr);
3386 nexthdr = nh[nhoff];
3388 if (skb_dst(skb) && skb_dst(skb)->dev)
3389 oif = skb_dst(skb)->dev->ifindex;
3391 memset(fl6, 0, sizeof(struct flowi6));
3392 fl6->flowi6_mark = skb->mark;
3393 fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3395 fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3396 fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3398 while (nh + offset + sizeof(*exthdr) < skb->data ||
3399 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3400 nh = skb_network_header(skb);
3401 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3404 case NEXTHDR_FRAGMENT:
3407 case NEXTHDR_ROUTING:
3410 offset += ipv6_optlen(exthdr);
3411 nexthdr = exthdr->nexthdr;
3414 case IPPROTO_UDPLITE:
3418 if (!onlyproto && (nh + offset + 4 < skb->data ||
3419 pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3422 nh = skb_network_header(skb);
3423 ports = (__be16 *)(nh + offset);
3424 fl6->fl6_sport = ports[!!reverse];
3425 fl6->fl6_dport = ports[!reverse];
3427 fl6->flowi6_proto = nexthdr;
3429 case IPPROTO_ICMPV6:
3430 if (!onlyproto && (nh + offset + 2 < skb->data ||
3431 pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3434 nh = skb_network_header(skb);
3435 icmp = (u8 *)(nh + offset);
3436 fl6->fl6_icmp_type = icmp[0];
3437 fl6->fl6_icmp_code = icmp[1];
3439 fl6->flowi6_proto = nexthdr;
3443 (nh + offset + 12 < skb->data ||
3444 pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3445 struct gre_base_hdr *gre_hdr;
3448 nh = skb_network_header(skb);
3449 gre_hdr = (struct gre_base_hdr *)(nh + offset);
3450 gre_key = (__be32 *)(gre_hdr + 1);
3452 if (gre_hdr->flags & GRE_KEY) {
3453 if (gre_hdr->flags & GRE_CSUM)
3455 fl6->fl6_gre_key = *gre_key;
3458 fl6->flowi6_proto = nexthdr;
3461 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3463 offset += ipv6_optlen(exthdr);
3464 if (!onlyproto && (nh + offset + 3 < skb->data ||
3465 pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3468 nh = skb_network_header(skb);
3469 mh = (struct ip6_mh *)(nh + offset);
3470 fl6->fl6_mh_type = mh->ip6mh_type;
3472 fl6->flowi6_proto = nexthdr;
3476 fl6->flowi6_proto = nexthdr;
3483 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3484 unsigned int family, int reverse)
3488 decode_session4(skb, fl, reverse);
3490 #if IS_ENABLED(CONFIG_IPV6)
3492 decode_session6(skb, fl, reverse);
3496 return -EAFNOSUPPORT;
3499 return security_xfrm_decode_session(skb, &fl->flowi_secid);
3501 EXPORT_SYMBOL(__xfrm_decode_session);
3503 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3505 for (; k < sp->len; k++) {
3506 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3515 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3516 unsigned short family)
3518 struct net *net = dev_net(skb->dev);
3519 struct xfrm_policy *pol;
3520 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3527 const struct xfrm_if_cb *ifcb;
3528 struct sec_path *sp;
3532 ifcb = xfrm_if_get_cb();
3535 struct xfrm_if_decode_session_result r;
3537 if (ifcb->decode_session(skb, family, &r)) {
3544 reverse = dir & ~XFRM_POLICY_MASK;
3545 dir &= XFRM_POLICY_MASK;
3547 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3548 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3552 nf_nat_decode_session(skb, &fl, family);
3554 /* First, check used SA against their selectors. */
3555 sp = skb_sec_path(skb);
3559 for (i = sp->len - 1; i >= 0; i--) {
3560 struct xfrm_state *x = sp->xvec[i];
3561 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3562 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3569 sk = sk_to_full_sk(sk);
3570 if (sk && sk->sk_policy[dir]) {
3571 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3573 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3579 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3582 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3587 if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3588 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3592 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3593 xfrm_secpath_reject(xerr_idx, skb, &fl);
3594 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3600 /* This lockless write can happen from different cpus. */
3601 WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3605 #ifdef CONFIG_XFRM_SUB_POLICY
3606 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3607 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3609 XFRM_POLICY_IN, if_id);
3611 if (IS_ERR(pols[1])) {
3612 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3613 xfrm_pol_put(pols[0]);
3616 /* This write can happen from different cpus. */
3617 WRITE_ONCE(pols[1]->curlft.use_time,
3618 ktime_get_real_seconds());
3624 if (pol->action == XFRM_POLICY_ALLOW) {
3625 static struct sec_path dummy;
3626 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3627 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3628 struct xfrm_tmpl **tpp = tp;
3632 sp = skb_sec_path(skb);
3636 for (pi = 0; pi < npols; pi++) {
3637 if (pols[pi] != pol &&
3638 pols[pi]->action != XFRM_POLICY_ALLOW) {
3639 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3642 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3643 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3646 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3647 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3652 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3656 /* For each tunnel xfrm, find the first matching tmpl.
3657 * For each tmpl before that, find corresponding xfrm.
3658 * Order is _important_. Later we will implement
3659 * some barriers, but at the moment barriers
3660 * are implied between each two transformations.
3661 * Upon success, marks secpath entries as having been
3662 * verified to allow them to be skipped in future policy
3663 * checks (e.g. nested tunnels).
3665 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3666 k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3669 /* "-2 - errored_index" returned */
3671 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3676 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3677 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3681 xfrm_pols_put(pols, npols);
3682 sp->verified_cnt = k;
3686 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3689 xfrm_secpath_reject(xerr_idx, skb, &fl);
3691 xfrm_pols_put(pols, npols);
3694 EXPORT_SYMBOL(__xfrm_policy_check);
3696 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3698 struct net *net = dev_net(skb->dev);
3700 struct dst_entry *dst;
3703 if (xfrm_decode_session(skb, &fl, family) < 0) {
3704 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3709 if (!skb_dst(skb)) {
3710 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3714 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3719 skb_dst_set(skb, dst);
3722 EXPORT_SYMBOL(__xfrm_route_forward);
3724 /* Optimize later using cookies and generation ids. */
3726 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3728 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3729 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3730 * get validated by dst_ops->check on every use. We do this
3731 * because when a normal route referenced by an XFRM dst is
3732 * obsoleted we do not go looking around for all parent
3733 * referencing XFRM dsts so that we can invalidate them. It
3734 * is just too much work. Instead we make the checks here on
3735 * every use. For example:
3737 * XFRM dst A --> IPv4 dst X
3739 * X is the "xdst->route" of A (X is also the "dst->path" of A
3740 * in this example). If X is marked obsolete, "A" will not
3741 * notice. That's what we are validating here via the
3742 * stale_bundle() check.
3744 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3746 * This will force stale_bundle() to fail on any xdst bundle with
3747 * this dst linked in it.
3749 if (dst->obsolete < 0 && !stale_bundle(dst))
3755 static int stale_bundle(struct dst_entry *dst)
3757 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3760 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3762 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3763 dst->dev = blackhole_netdev;
3768 EXPORT_SYMBOL(xfrm_dst_ifdown);
3770 static void xfrm_link_failure(struct sk_buff *skb)
3772 /* Impossible. Such dst must be popped before reaches point of failure. */
3775 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3778 if (dst->obsolete) {
3786 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3789 struct xfrm_dst *xdst = bundle[nr];
3790 u32 pmtu, route_mtu_cached;
3791 struct dst_entry *dst;
3794 pmtu = dst_mtu(xfrm_dst_child(dst));
3795 xdst->child_mtu_cached = pmtu;
3797 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3799 route_mtu_cached = dst_mtu(xdst->route);
3800 xdst->route_mtu_cached = route_mtu_cached;
3802 if (pmtu > route_mtu_cached)
3803 pmtu = route_mtu_cached;
3805 dst_metric_set(dst, RTAX_MTU, pmtu);
3809 /* Check that the bundle accepts the flow and its components are
3813 static int xfrm_bundle_ok(struct xfrm_dst *first)
3815 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3816 struct dst_entry *dst = &first->u.dst;
3817 struct xfrm_dst *xdst;
3821 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3822 (dst->dev && !netif_running(dst->dev)))
3825 if (dst->flags & DST_XFRM_QUEUE)
3828 start_from = nr = 0;
3830 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3832 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3834 if (xdst->xfrm_genid != dst->xfrm->genid)
3836 if (xdst->num_pols > 0 &&
3837 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3840 bundle[nr++] = xdst;
3842 mtu = dst_mtu(xfrm_dst_child(dst));
3843 if (xdst->child_mtu_cached != mtu) {
3845 xdst->child_mtu_cached = mtu;
3848 if (!dst_check(xdst->route, xdst->route_cookie))
3850 mtu = dst_mtu(xdst->route);
3851 if (xdst->route_mtu_cached != mtu) {
3853 xdst->route_mtu_cached = mtu;
3856 dst = xfrm_dst_child(dst);
3857 } while (dst->xfrm);
3859 if (likely(!start_from))
3862 xdst = bundle[start_from - 1];
3863 mtu = xdst->child_mtu_cached;
3864 while (start_from--) {
3867 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3868 if (mtu > xdst->route_mtu_cached)
3869 mtu = xdst->route_mtu_cached;
3870 dst_metric_set(dst, RTAX_MTU, mtu);
3874 xdst = bundle[start_from - 1];
3875 xdst->child_mtu_cached = mtu;
3881 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3883 return dst_metric_advmss(xfrm_dst_path(dst));
3886 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3888 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3890 return mtu ? : dst_mtu(xfrm_dst_path(dst));
3893 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3897 const struct xfrm_state *xfrm = dst->xfrm;
3899 dst = xfrm_dst_child(dst);
3901 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3903 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3904 daddr = xfrm->coaddr;
3905 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3906 daddr = &xfrm->id.daddr;
3911 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3912 struct sk_buff *skb,
3915 const struct dst_entry *path = xfrm_dst_path(dst);
3918 daddr = xfrm_get_dst_nexthop(dst, daddr);
3919 return path->ops->neigh_lookup(path, skb, daddr);
3922 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3924 const struct dst_entry *path = xfrm_dst_path(dst);
3926 daddr = xfrm_get_dst_nexthop(dst, daddr);
3927 path->ops->confirm_neigh(path, daddr);
3930 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3934 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3935 return -EAFNOSUPPORT;
3937 spin_lock(&xfrm_policy_afinfo_lock);
3938 if (unlikely(xfrm_policy_afinfo[family] != NULL))
3941 struct dst_ops *dst_ops = afinfo->dst_ops;
3942 if (likely(dst_ops->kmem_cachep == NULL))
3943 dst_ops->kmem_cachep = xfrm_dst_cache;
3944 if (likely(dst_ops->check == NULL))
3945 dst_ops->check = xfrm_dst_check;
3946 if (likely(dst_ops->default_advmss == NULL))
3947 dst_ops->default_advmss = xfrm_default_advmss;
3948 if (likely(dst_ops->mtu == NULL))
3949 dst_ops->mtu = xfrm_mtu;
3950 if (likely(dst_ops->negative_advice == NULL))
3951 dst_ops->negative_advice = xfrm_negative_advice;
3952 if (likely(dst_ops->link_failure == NULL))
3953 dst_ops->link_failure = xfrm_link_failure;
3954 if (likely(dst_ops->neigh_lookup == NULL))
3955 dst_ops->neigh_lookup = xfrm_neigh_lookup;
3956 if (likely(!dst_ops->confirm_neigh))
3957 dst_ops->confirm_neigh = xfrm_confirm_neigh;
3958 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3960 spin_unlock(&xfrm_policy_afinfo_lock);
3964 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3966 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3968 struct dst_ops *dst_ops = afinfo->dst_ops;
3971 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3972 if (xfrm_policy_afinfo[i] != afinfo)
3974 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3980 dst_ops->kmem_cachep = NULL;
3981 dst_ops->check = NULL;
3982 dst_ops->negative_advice = NULL;
3983 dst_ops->link_failure = NULL;
3985 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3987 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3989 spin_lock(&xfrm_if_cb_lock);
3990 rcu_assign_pointer(xfrm_if_cb, ifcb);
3991 spin_unlock(&xfrm_if_cb_lock);
3993 EXPORT_SYMBOL(xfrm_if_register_cb);
3995 void xfrm_if_unregister_cb(void)
3997 RCU_INIT_POINTER(xfrm_if_cb, NULL);
4000 EXPORT_SYMBOL(xfrm_if_unregister_cb);
4002 #ifdef CONFIG_XFRM_STATISTICS
4003 static int __net_init xfrm_statistics_init(struct net *net)
4006 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4007 if (!net->mib.xfrm_statistics)
4009 rv = xfrm_proc_init(net);
4011 free_percpu(net->mib.xfrm_statistics);
4015 static void xfrm_statistics_fini(struct net *net)
4017 xfrm_proc_fini(net);
4018 free_percpu(net->mib.xfrm_statistics);
4021 static int __net_init xfrm_statistics_init(struct net *net)
4026 static void xfrm_statistics_fini(struct net *net)
4031 static int __net_init xfrm_policy_init(struct net *net)
4033 unsigned int hmask, sz;
4036 if (net_eq(net, &init_net)) {
4037 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4038 sizeof(struct xfrm_dst),
4039 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4041 err = rhashtable_init(&xfrm_policy_inexact_table,
4042 &xfrm_pol_inexact_params);
4047 sz = (hmask+1) * sizeof(struct hlist_head);
4049 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4050 if (!net->xfrm.policy_byidx)
4052 net->xfrm.policy_idx_hmask = hmask;
4054 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4055 struct xfrm_policy_hash *htab;
4057 net->xfrm.policy_count[dir] = 0;
4058 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4059 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4061 htab = &net->xfrm.policy_bydst[dir];
4062 htab->table = xfrm_hash_alloc(sz);
4065 htab->hmask = hmask;
4071 net->xfrm.policy_hthresh.lbits4 = 32;
4072 net->xfrm.policy_hthresh.rbits4 = 32;
4073 net->xfrm.policy_hthresh.lbits6 = 128;
4074 net->xfrm.policy_hthresh.rbits6 = 128;
4076 seqlock_init(&net->xfrm.policy_hthresh.lock);
4078 INIT_LIST_HEAD(&net->xfrm.policy_all);
4079 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4080 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4081 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4085 for (dir--; dir >= 0; dir--) {
4086 struct xfrm_policy_hash *htab;
4088 htab = &net->xfrm.policy_bydst[dir];
4089 xfrm_hash_free(htab->table, sz);
4091 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4096 static void xfrm_policy_fini(struct net *net)
4098 struct xfrm_pol_inexact_bin *b, *t;
4102 flush_work(&net->xfrm.policy_hash_work);
4103 #ifdef CONFIG_XFRM_SUB_POLICY
4104 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4106 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4108 WARN_ON(!list_empty(&net->xfrm.policy_all));
4110 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4111 struct xfrm_policy_hash *htab;
4113 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4115 htab = &net->xfrm.policy_bydst[dir];
4116 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4117 WARN_ON(!hlist_empty(htab->table));
4118 xfrm_hash_free(htab->table, sz);
4121 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4122 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4123 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4125 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4126 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4127 __xfrm_policy_inexact_prune_bin(b, true);
4128 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4131 static int __net_init xfrm_net_init(struct net *net)
4135 /* Initialize the per-net locks here */
4136 spin_lock_init(&net->xfrm.xfrm_state_lock);
4137 spin_lock_init(&net->xfrm.xfrm_policy_lock);
4138 seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4139 mutex_init(&net->xfrm.xfrm_cfg_mutex);
4140 net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4141 net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4142 net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4144 rv = xfrm_statistics_init(net);
4146 goto out_statistics;
4147 rv = xfrm_state_init(net);
4150 rv = xfrm_policy_init(net);
4153 rv = xfrm_sysctl_init(net);
4160 xfrm_policy_fini(net);
4162 xfrm_state_fini(net);
4164 xfrm_statistics_fini(net);
4169 static void __net_exit xfrm_net_exit(struct net *net)
4171 xfrm_sysctl_fini(net);
4172 xfrm_policy_fini(net);
4173 xfrm_state_fini(net);
4174 xfrm_statistics_fini(net);
4177 static struct pernet_operations __net_initdata xfrm_net_ops = {
4178 .init = xfrm_net_init,
4179 .exit = xfrm_net_exit,
4182 void __init xfrm_init(void)
4184 register_pernet_subsys(&xfrm_net_ops);
4188 #ifdef CONFIG_XFRM_ESPINTCP
4193 #ifdef CONFIG_AUDITSYSCALL
4194 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4195 struct audit_buffer *audit_buf)
4197 struct xfrm_sec_ctx *ctx = xp->security;
4198 struct xfrm_selector *sel = &xp->selector;
4201 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4202 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4204 switch (sel->family) {
4206 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4207 if (sel->prefixlen_s != 32)
4208 audit_log_format(audit_buf, " src_prefixlen=%d",
4210 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4211 if (sel->prefixlen_d != 32)
4212 audit_log_format(audit_buf, " dst_prefixlen=%d",
4216 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4217 if (sel->prefixlen_s != 128)
4218 audit_log_format(audit_buf, " src_prefixlen=%d",
4220 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4221 if (sel->prefixlen_d != 128)
4222 audit_log_format(audit_buf, " dst_prefixlen=%d",
4228 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4230 struct audit_buffer *audit_buf;
4232 audit_buf = xfrm_audit_start("SPD-add");
4233 if (audit_buf == NULL)
4235 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4236 audit_log_format(audit_buf, " res=%u", result);
4237 xfrm_audit_common_policyinfo(xp, audit_buf);
4238 audit_log_end(audit_buf);
4240 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4242 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4245 struct audit_buffer *audit_buf;
4247 audit_buf = xfrm_audit_start("SPD-delete");
4248 if (audit_buf == NULL)
4250 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4251 audit_log_format(audit_buf, " res=%u", result);
4252 xfrm_audit_common_policyinfo(xp, audit_buf);
4253 audit_log_end(audit_buf);
4255 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4258 #ifdef CONFIG_XFRM_MIGRATE
4259 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4260 const struct xfrm_selector *sel_tgt)
4262 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4263 if (sel_tgt->family == sel_cmp->family &&
4264 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4266 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4268 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4269 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4273 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4280 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4281 u8 dir, u8 type, struct net *net, u32 if_id)
4283 struct xfrm_policy *pol, *ret = NULL;
4284 struct hlist_head *chain;
4287 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4288 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4289 hlist_for_each_entry(pol, chain, bydst) {
4290 if ((if_id == 0 || pol->if_id == if_id) &&
4291 xfrm_migrate_selector_match(sel, &pol->selector) &&
4292 pol->type == type) {
4294 priority = ret->priority;
4298 chain = &net->xfrm.policy_inexact[dir];
4299 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4300 if ((pol->priority >= priority) && ret)
4303 if ((if_id == 0 || pol->if_id == if_id) &&
4304 xfrm_migrate_selector_match(sel, &pol->selector) &&
4305 pol->type == type) {
4313 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4318 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4322 if (t->mode == m->mode && t->id.proto == m->proto &&
4323 (m->reqid == 0 || t->reqid == m->reqid)) {
4325 case XFRM_MODE_TUNNEL:
4326 case XFRM_MODE_BEET:
4327 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4329 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4334 case XFRM_MODE_TRANSPORT:
4335 /* in case of transport mode, template does not store
4336 any IP addresses, hence we just compare mode and
4347 /* update endpoint address(es) of template(s) */
4348 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4349 struct xfrm_migrate *m, int num_migrate)
4351 struct xfrm_migrate *mp;
4354 write_lock_bh(&pol->lock);
4355 if (unlikely(pol->walk.dead)) {
4356 /* target policy has been deleted */
4357 write_unlock_bh(&pol->lock);
4361 for (i = 0; i < pol->xfrm_nr; i++) {
4362 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4363 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4366 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4367 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4369 /* update endpoints */
4370 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4371 sizeof(pol->xfrm_vec[i].id.daddr));
4372 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4373 sizeof(pol->xfrm_vec[i].saddr));
4374 pol->xfrm_vec[i].encap_family = mp->new_family;
4376 atomic_inc(&pol->genid);
4380 write_unlock_bh(&pol->lock);
4388 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4392 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4395 for (i = 0; i < num_migrate; i++) {
4396 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4397 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4400 /* check if there is any duplicated entry */
4401 for (j = i + 1; j < num_migrate; j++) {
4402 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4403 sizeof(m[i].old_daddr)) &&
4404 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4405 sizeof(m[i].old_saddr)) &&
4406 m[i].proto == m[j].proto &&
4407 m[i].mode == m[j].mode &&
4408 m[i].reqid == m[j].reqid &&
4409 m[i].old_family == m[j].old_family)
4417 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4418 struct xfrm_migrate *m, int num_migrate,
4419 struct xfrm_kmaddress *k, struct net *net,
4420 struct xfrm_encap_tmpl *encap, u32 if_id)
4422 int i, err, nx_cur = 0, nx_new = 0;
4423 struct xfrm_policy *pol = NULL;
4424 struct xfrm_state *x, *xc;
4425 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4426 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4427 struct xfrm_migrate *mp;
4429 /* Stage 0 - sanity checks */
4430 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4433 if (dir >= XFRM_POLICY_MAX) {
4438 /* Stage 1 - find policy */
4439 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id)) == NULL) {
4444 /* Stage 2 - find and update state(s) */
4445 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4446 if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4449 xc = xfrm_state_migrate(x, mp, encap);
4460 /* Stage 3 - update policy */
4461 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4464 /* Stage 4 - delete old state(s) */
4466 xfrm_states_put(x_cur, nx_cur);
4467 xfrm_states_delete(x_cur, nx_cur);
4470 /* Stage 5 - announce */
4471 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4483 xfrm_states_put(x_cur, nx_cur);
4485 xfrm_states_delete(x_new, nx_new);
4489 EXPORT_SYMBOL(xfrm_migrate);