6 * Kazunori MIYAZAWA @USAGI
7 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
9 * Kazunori MIYAZAWA @USAGI
11 * Split up af-specific portion
12 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/cpu.h>
28 #include <linux/audit.h>
29 #include <linux/rhashtable.h>
34 #ifdef CONFIG_XFRM_STATISTICS
38 #include "xfrm_hash.h"
40 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
41 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
42 #define XFRM_MAX_QUEUE_LEN 100
45 struct dst_entry *dst_orig;
49 /* prefixes smaller than this are stored in lists, not trees. */
50 #define INEXACT_PREFIXLEN_IPV4 16
51 #define INEXACT_PREFIXLEN_IPV6 48
53 struct xfrm_pol_inexact_node {
63 /* the policies matching this node, can be empty list */
64 struct hlist_head hhead;
67 /* xfrm inexact policy search tree:
68 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
70 * +---- root_d: sorted by daddr:prefix
72 * | xfrm_pol_inexact_node
74 * | +- root: sorted by saddr/prefix
76 * | | xfrm_pol_inexact_node
80 * | | + hhead: saddr:daddr policies
82 * | +- coarse policies and all any:daddr policies
84 * +---- root_s: sorted by saddr:prefix
86 * | xfrm_pol_inexact_node
90 * | + hhead: saddr:any policies
92 * +---- coarse policies and all any:any policies
94 * Lookups return four candidate lists:
95 * 1. any:any list from top-level xfrm_pol_inexact_bin
96 * 2. any:daddr list from daddr tree
97 * 3. saddr:daddr list from 2nd level daddr tree
98 * 4. saddr:any list from saddr tree
100 * This result set then needs to be searched for the policy with
101 * the lowest priority. If two results have same prio, youngest one wins.
104 struct xfrm_pol_inexact_key {
111 struct xfrm_pol_inexact_bin {
112 struct xfrm_pol_inexact_key k;
113 struct rhash_head head;
114 /* list containing '*:*' policies */
115 struct hlist_head hhead;
118 /* tree sorted by daddr/prefix */
119 struct rb_root root_d;
121 /* tree sorted by saddr/prefix */
122 struct rb_root root_s;
124 /* slow path below */
125 struct list_head inexact_bins;
129 enum xfrm_pol_inexact_candidate_type {
138 struct xfrm_pol_inexact_candidates {
139 struct hlist_head *res[XFRM_POL_CAND_MAX];
142 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
143 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
145 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
146 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
149 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
150 static __read_mostly seqcount_t xfrm_policy_hash_generation;
152 static struct rhashtable xfrm_policy_inexact_table;
153 static const struct rhashtable_params xfrm_pol_inexact_params;
155 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
156 static int stale_bundle(struct dst_entry *dst);
157 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
158 static void xfrm_policy_queue_process(struct timer_list *t);
160 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
161 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
164 static struct xfrm_pol_inexact_bin *
165 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
168 static struct xfrm_pol_inexact_bin *
169 xfrm_policy_inexact_lookup_rcu(struct net *net,
170 u8 type, u16 family, u8 dir, u32 if_id);
171 static struct xfrm_policy *
172 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
174 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
175 struct xfrm_policy *policy);
178 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
179 struct xfrm_pol_inexact_bin *b,
180 const xfrm_address_t *saddr,
181 const xfrm_address_t *daddr);
183 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
185 return refcount_inc_not_zero(&policy->refcnt);
189 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
191 const struct flowi4 *fl4 = &fl->u.ip4;
193 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
194 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
195 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
196 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
197 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
198 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
202 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
204 const struct flowi6 *fl6 = &fl->u.ip6;
206 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
207 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
208 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
209 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
210 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
211 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
214 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
215 unsigned short family)
219 return __xfrm4_selector_match(sel, fl);
221 return __xfrm6_selector_match(sel, fl);
226 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
228 const struct xfrm_policy_afinfo *afinfo;
230 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
233 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
234 if (unlikely(!afinfo))
239 /* Called with rcu_read_lock(). */
240 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
242 return rcu_dereference(xfrm_if_cb);
245 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
246 const xfrm_address_t *saddr,
247 const xfrm_address_t *daddr,
248 int family, u32 mark)
250 const struct xfrm_policy_afinfo *afinfo;
251 struct dst_entry *dst;
253 afinfo = xfrm_policy_get_afinfo(family);
254 if (unlikely(afinfo == NULL))
255 return ERR_PTR(-EAFNOSUPPORT);
257 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
263 EXPORT_SYMBOL(__xfrm_dst_lookup);
265 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
267 xfrm_address_t *prev_saddr,
268 xfrm_address_t *prev_daddr,
269 int family, u32 mark)
271 struct net *net = xs_net(x);
272 xfrm_address_t *saddr = &x->props.saddr;
273 xfrm_address_t *daddr = &x->id.daddr;
274 struct dst_entry *dst;
276 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
280 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
285 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
288 if (prev_saddr != saddr)
289 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
290 if (prev_daddr != daddr)
291 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
297 static inline unsigned long make_jiffies(long secs)
299 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
300 return MAX_SCHEDULE_TIMEOUT-1;
305 static void xfrm_policy_timer(struct timer_list *t)
307 struct xfrm_policy *xp = from_timer(xp, t, timer);
308 time64_t now = ktime_get_real_seconds();
309 time64_t next = TIME64_MAX;
313 read_lock(&xp->lock);
315 if (unlikely(xp->walk.dead))
318 dir = xfrm_policy_id2dir(xp->index);
320 if (xp->lft.hard_add_expires_seconds) {
321 time64_t tmo = xp->lft.hard_add_expires_seconds +
322 xp->curlft.add_time - now;
328 if (xp->lft.hard_use_expires_seconds) {
329 time64_t tmo = xp->lft.hard_use_expires_seconds +
330 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
336 if (xp->lft.soft_add_expires_seconds) {
337 time64_t tmo = xp->lft.soft_add_expires_seconds +
338 xp->curlft.add_time - now;
341 tmo = XFRM_KM_TIMEOUT;
346 if (xp->lft.soft_use_expires_seconds) {
347 time64_t tmo = xp->lft.soft_use_expires_seconds +
348 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
351 tmo = XFRM_KM_TIMEOUT;
358 km_policy_expired(xp, dir, 0, 0);
359 if (next != TIME64_MAX &&
360 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
364 read_unlock(&xp->lock);
369 read_unlock(&xp->lock);
370 if (!xfrm_policy_delete(xp, dir))
371 km_policy_expired(xp, dir, 1, 0);
375 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
379 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
381 struct xfrm_policy *policy;
383 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
386 write_pnet(&policy->xp_net, net);
387 INIT_LIST_HEAD(&policy->walk.all);
388 INIT_HLIST_NODE(&policy->bydst_inexact_list);
389 INIT_HLIST_NODE(&policy->bydst);
390 INIT_HLIST_NODE(&policy->byidx);
391 rwlock_init(&policy->lock);
392 refcount_set(&policy->refcnt, 1);
393 skb_queue_head_init(&policy->polq.hold_queue);
394 timer_setup(&policy->timer, xfrm_policy_timer, 0);
395 timer_setup(&policy->polq.hold_timer,
396 xfrm_policy_queue_process, 0);
400 EXPORT_SYMBOL(xfrm_policy_alloc);
402 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
404 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
406 security_xfrm_policy_free(policy->security);
410 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
412 void xfrm_policy_destroy(struct xfrm_policy *policy)
414 BUG_ON(!policy->walk.dead);
416 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
419 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
421 EXPORT_SYMBOL(xfrm_policy_destroy);
423 /* Rule must be locked. Release descendant resources, announce
424 * entry dead. The rule must be unlinked from lists to the moment.
427 static void xfrm_policy_kill(struct xfrm_policy *policy)
429 policy->walk.dead = 1;
431 atomic_inc(&policy->genid);
433 if (del_timer(&policy->polq.hold_timer))
434 xfrm_pol_put(policy);
435 skb_queue_purge(&policy->polq.hold_queue);
437 if (del_timer(&policy->timer))
438 xfrm_pol_put(policy);
440 xfrm_pol_put(policy);
443 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
445 static inline unsigned int idx_hash(struct net *net, u32 index)
447 return __idx_hash(index, net->xfrm.policy_idx_hmask);
450 /* calculate policy hash thresholds */
451 static void __get_hash_thresh(struct net *net,
452 unsigned short family, int dir,
453 u8 *dbits, u8 *sbits)
457 *dbits = net->xfrm.policy_bydst[dir].dbits4;
458 *sbits = net->xfrm.policy_bydst[dir].sbits4;
462 *dbits = net->xfrm.policy_bydst[dir].dbits6;
463 *sbits = net->xfrm.policy_bydst[dir].sbits6;
472 static struct hlist_head *policy_hash_bysel(struct net *net,
473 const struct xfrm_selector *sel,
474 unsigned short family, int dir)
476 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
481 __get_hash_thresh(net, family, dir, &dbits, &sbits);
482 hash = __sel_hash(sel, family, hmask, dbits, sbits);
484 if (hash == hmask + 1)
487 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
488 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
491 static struct hlist_head *policy_hash_direct(struct net *net,
492 const xfrm_address_t *daddr,
493 const xfrm_address_t *saddr,
494 unsigned short family, int dir)
496 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
501 __get_hash_thresh(net, family, dir, &dbits, &sbits);
502 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
504 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
505 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
508 static void xfrm_dst_hash_transfer(struct net *net,
509 struct hlist_head *list,
510 struct hlist_head *ndsttable,
511 unsigned int nhashmask,
514 struct hlist_node *tmp, *entry0 = NULL;
515 struct xfrm_policy *pol;
521 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
524 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
525 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
526 pol->family, nhashmask, dbits, sbits);
528 hlist_del_rcu(&pol->bydst);
529 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
534 hlist_del_rcu(&pol->bydst);
535 hlist_add_behind_rcu(&pol->bydst, entry0);
537 entry0 = &pol->bydst;
539 if (!hlist_empty(list)) {
545 static void xfrm_idx_hash_transfer(struct hlist_head *list,
546 struct hlist_head *nidxtable,
547 unsigned int nhashmask)
549 struct hlist_node *tmp;
550 struct xfrm_policy *pol;
552 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
555 h = __idx_hash(pol->index, nhashmask);
556 hlist_add_head(&pol->byidx, nidxtable+h);
560 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
562 return ((old_hmask + 1) << 1) - 1;
565 static void xfrm_bydst_resize(struct net *net, int dir)
567 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
568 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
569 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
570 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
571 struct hlist_head *odst;
577 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
578 write_seqcount_begin(&xfrm_policy_hash_generation);
580 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
581 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
583 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
584 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
586 for (i = hmask; i >= 0; i--)
587 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
589 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
590 net->xfrm.policy_bydst[dir].hmask = nhashmask;
592 write_seqcount_end(&xfrm_policy_hash_generation);
593 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
597 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
600 static void xfrm_byidx_resize(struct net *net, int total)
602 unsigned int hmask = net->xfrm.policy_idx_hmask;
603 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
604 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
605 struct hlist_head *oidx = net->xfrm.policy_byidx;
606 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
612 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
614 for (i = hmask; i >= 0; i--)
615 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
617 net->xfrm.policy_byidx = nidx;
618 net->xfrm.policy_idx_hmask = nhashmask;
620 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
622 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
625 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
627 unsigned int cnt = net->xfrm.policy_count[dir];
628 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
633 if ((hmask + 1) < xfrm_policy_hashmax &&
640 static inline int xfrm_byidx_should_resize(struct net *net, int total)
642 unsigned int hmask = net->xfrm.policy_idx_hmask;
644 if ((hmask + 1) < xfrm_policy_hashmax &&
651 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
653 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
654 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
655 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
656 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
657 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
658 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
659 si->spdhcnt = net->xfrm.policy_idx_hmask;
660 si->spdhmcnt = xfrm_policy_hashmax;
662 EXPORT_SYMBOL(xfrm_spd_getinfo);
664 static DEFINE_MUTEX(hash_resize_mutex);
665 static void xfrm_hash_resize(struct work_struct *work)
667 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
670 mutex_lock(&hash_resize_mutex);
673 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
674 if (xfrm_bydst_should_resize(net, dir, &total))
675 xfrm_bydst_resize(net, dir);
677 if (xfrm_byidx_should_resize(net, total))
678 xfrm_byidx_resize(net, total);
680 mutex_unlock(&hash_resize_mutex);
683 /* Make sure *pol can be inserted into fastbin.
684 * Useful to check that later insert requests will be sucessful
685 * (provided xfrm_policy_lock is held throughout).
687 static struct xfrm_pol_inexact_bin *
688 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
690 struct xfrm_pol_inexact_bin *bin, *prev;
691 struct xfrm_pol_inexact_key k = {
692 .family = pol->family,
697 struct net *net = xp_net(pol);
699 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
701 write_pnet(&k.net, net);
702 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
703 xfrm_pol_inexact_params);
707 bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
712 INIT_HLIST_HEAD(&bin->hhead);
713 bin->root_d = RB_ROOT;
714 bin->root_s = RB_ROOT;
715 seqcount_init(&bin->count);
717 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
719 xfrm_pol_inexact_params);
721 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
727 return IS_ERR(prev) ? NULL : prev;
730 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
731 int family, u8 prefixlen)
733 if (xfrm_addr_any(addr, family))
736 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
739 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
746 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
748 const xfrm_address_t *addr;
749 bool saddr_any, daddr_any;
752 addr = &policy->selector.saddr;
753 prefixlen = policy->selector.prefixlen_s;
755 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
758 addr = &policy->selector.daddr;
759 prefixlen = policy->selector.prefixlen_d;
760 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
763 return saddr_any && daddr_any;
766 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
767 const xfrm_address_t *addr, u8 prefixlen)
770 node->prefixlen = prefixlen;
773 static struct xfrm_pol_inexact_node *
774 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
776 struct xfrm_pol_inexact_node *node;
778 node = kzalloc(sizeof(*node), GFP_ATOMIC);
780 xfrm_pol_inexact_node_init(node, addr, prefixlen);
785 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
786 const xfrm_address_t *b,
787 u8 prefixlen, u16 family)
789 unsigned int pdw, pbi;
794 if (sizeof(long) == 4 && prefixlen == 0)
795 return ntohl(a->a4) - ntohl(b->a4);
796 return (ntohl(a->a4) & ((~0UL << (32 - prefixlen)))) -
797 (ntohl(b->a4) & ((~0UL << (32 - prefixlen))));
799 pdw = prefixlen >> 5;
800 pbi = prefixlen & 0x1f;
803 delta = memcmp(a->a6, b->a6, pdw << 2);
808 u32 mask = ~0u << (32 - pbi);
810 delta = (ntohl(a->a6[pdw]) & mask) -
811 (ntohl(b->a6[pdw]) & mask);
821 static void xfrm_policy_inexact_list_reinsert(struct net *net,
822 struct xfrm_pol_inexact_node *n,
825 unsigned int matched_s, matched_d;
826 struct xfrm_policy *policy, *p;
831 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
832 struct hlist_node *newpos = NULL;
833 bool matches_s, matches_d;
835 if (!policy->bydst_reinsert)
838 WARN_ON_ONCE(policy->family != family);
840 policy->bydst_reinsert = false;
841 hlist_for_each_entry(p, &n->hhead, bydst) {
842 if (policy->priority > p->priority)
844 else if (policy->priority == p->priority &&
845 policy->pos > p->pos)
852 hlist_add_behind_rcu(&policy->bydst, newpos);
854 hlist_add_head_rcu(&policy->bydst, &n->hhead);
856 /* paranoia checks follow.
857 * Check that the reinserted policy matches at least
858 * saddr or daddr for current node prefix.
860 * Matching both is fine, matching saddr in one policy
861 * (but not daddr) and then matching only daddr in another
864 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
868 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
872 if (matches_s && matches_d)
875 WARN_ON_ONCE(!matches_s && !matches_d);
880 WARN_ON_ONCE(matched_s && matched_d);
884 static void xfrm_policy_inexact_node_reinsert(struct net *net,
885 struct xfrm_pol_inexact_node *n,
889 struct xfrm_pol_inexact_node *node;
890 struct rb_node **p, *parent;
892 /* we should not have another subtree here */
893 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
902 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
904 prefixlen = min(node->prefixlen, n->prefixlen);
906 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
909 p = &parent->rb_left;
910 } else if (delta > 0) {
911 p = &parent->rb_right;
913 struct xfrm_policy *tmp;
915 hlist_for_each_entry(tmp, &n->hhead, bydst) {
916 tmp->bydst_reinsert = true;
917 hlist_del_rcu(&tmp->bydst);
920 xfrm_policy_inexact_list_reinsert(net, node, family);
922 if (node->prefixlen == n->prefixlen) {
930 n->prefixlen = prefixlen;
935 rb_link_node_rcu(&n->node, parent, p);
936 rb_insert_color(&n->node, new);
939 /* merge nodes v and n */
940 static void xfrm_policy_inexact_node_merge(struct net *net,
941 struct xfrm_pol_inexact_node *v,
942 struct xfrm_pol_inexact_node *n,
945 struct xfrm_pol_inexact_node *node;
946 struct xfrm_policy *tmp;
947 struct rb_node *rnode;
949 /* To-be-merged node v has a subtree.
951 * Dismantle it and insert its nodes to n->root.
953 while ((rnode = rb_first(&v->root)) != NULL) {
954 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
955 rb_erase(&node->node, &v->root);
956 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
960 hlist_for_each_entry(tmp, &v->hhead, bydst) {
961 tmp->bydst_reinsert = true;
962 hlist_del_rcu(&tmp->bydst);
965 xfrm_policy_inexact_list_reinsert(net, n, family);
968 static struct xfrm_pol_inexact_node *
969 xfrm_policy_inexact_insert_node(struct net *net,
970 struct rb_root *root,
971 xfrm_address_t *addr,
972 u16 family, u8 prefixlen, u8 dir)
974 struct xfrm_pol_inexact_node *cached = NULL;
975 struct rb_node **p, *parent = NULL;
976 struct xfrm_pol_inexact_node *node;
983 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
985 delta = xfrm_policy_addr_delta(addr, &node->addr,
988 if (delta == 0 && prefixlen >= node->prefixlen) {
989 WARN_ON_ONCE(cached); /* ipsec policies got lost */
994 p = &parent->rb_left;
996 p = &parent->rb_right;
998 if (prefixlen < node->prefixlen) {
999 delta = xfrm_policy_addr_delta(addr, &node->addr,
1005 /* This node is a subnet of the new prefix. It needs
1006 * to be removed and re-inserted with the smaller
1007 * prefix and all nodes that are now also covered
1008 * by the reduced prefixlen.
1010 rb_erase(&node->node, root);
1013 xfrm_pol_inexact_node_init(node, addr,
1017 /* This node also falls within the new
1018 * prefixlen. Merge the to-be-reinserted
1019 * node and this one.
1021 xfrm_policy_inexact_node_merge(net, node,
1023 kfree_rcu(node, rcu);
1034 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1039 rb_link_node_rcu(&node->node, parent, p);
1040 rb_insert_color(&node->node, root);
1045 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1047 struct xfrm_pol_inexact_node *node;
1048 struct rb_node *rn = rb_first(r);
1051 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1053 xfrm_policy_inexact_gc_tree(&node->root, rm);
1056 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1061 rb_erase(&node->node, r);
1062 kfree_rcu(node, rcu);
1066 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1068 write_seqcount_begin(&b->count);
1069 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1070 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1071 write_seqcount_end(&b->count);
1073 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1074 !hlist_empty(&b->hhead)) {
1075 WARN_ON_ONCE(net_exit);
1079 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1080 xfrm_pol_inexact_params) == 0) {
1081 list_del(&b->inexact_bins);
1086 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1088 struct net *net = read_pnet(&b->k.net);
1090 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1091 __xfrm_policy_inexact_prune_bin(b, false);
1092 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1095 static void __xfrm_policy_inexact_flush(struct net *net)
1097 struct xfrm_pol_inexact_bin *bin, *t;
1099 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1101 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1102 __xfrm_policy_inexact_prune_bin(bin, false);
1105 static struct hlist_head *
1106 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1107 struct xfrm_policy *policy, u8 dir)
1109 struct xfrm_pol_inexact_node *n;
1112 net = xp_net(policy);
1113 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1115 if (xfrm_policy_inexact_insert_use_any_list(policy))
1118 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1120 policy->selector.prefixlen_d)) {
1121 write_seqcount_begin(&bin->count);
1122 n = xfrm_policy_inexact_insert_node(net,
1124 &policy->selector.saddr,
1126 policy->selector.prefixlen_s,
1128 write_seqcount_end(&bin->count);
1135 /* daddr is fixed */
1136 write_seqcount_begin(&bin->count);
1137 n = xfrm_policy_inexact_insert_node(net,
1139 &policy->selector.daddr,
1141 policy->selector.prefixlen_d, dir);
1142 write_seqcount_end(&bin->count);
1146 /* saddr is wildcard */
1147 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1149 policy->selector.prefixlen_s))
1152 write_seqcount_begin(&bin->count);
1153 n = xfrm_policy_inexact_insert_node(net,
1155 &policy->selector.saddr,
1157 policy->selector.prefixlen_s, dir);
1158 write_seqcount_end(&bin->count);
1165 static struct xfrm_policy *
1166 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1168 struct xfrm_pol_inexact_bin *bin;
1169 struct xfrm_policy *delpol;
1170 struct hlist_head *chain;
1173 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1175 return ERR_PTR(-ENOMEM);
1177 net = xp_net(policy);
1178 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1180 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1182 __xfrm_policy_inexact_prune_bin(bin, false);
1183 return ERR_PTR(-ENOMEM);
1186 delpol = xfrm_policy_insert_list(chain, policy, excl);
1187 if (delpol && excl) {
1188 __xfrm_policy_inexact_prune_bin(bin, false);
1189 return ERR_PTR(-EEXIST);
1192 chain = &net->xfrm.policy_inexact[dir];
1193 xfrm_policy_insert_inexact_list(chain, policy);
1196 __xfrm_policy_inexact_prune_bin(bin, false);
1201 static void xfrm_hash_rebuild(struct work_struct *work)
1203 struct net *net = container_of(work, struct net,
1204 xfrm.policy_hthresh.work);
1206 struct xfrm_policy *pol;
1207 struct xfrm_policy *policy;
1208 struct hlist_head *chain;
1209 struct hlist_head *odst;
1210 struct hlist_node *newpos;
1214 u8 lbits4, rbits4, lbits6, rbits6;
1216 mutex_lock(&hash_resize_mutex);
1218 /* read selector prefixlen thresholds */
1220 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1222 lbits4 = net->xfrm.policy_hthresh.lbits4;
1223 rbits4 = net->xfrm.policy_hthresh.rbits4;
1224 lbits6 = net->xfrm.policy_hthresh.lbits6;
1225 rbits6 = net->xfrm.policy_hthresh.rbits6;
1226 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1228 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1229 write_seqcount_begin(&xfrm_policy_hash_generation);
1231 /* make sure that we can insert the indirect policies again before
1232 * we start with destructive action.
1234 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1235 struct xfrm_pol_inexact_bin *bin;
1238 dir = xfrm_policy_id2dir(policy->index);
1239 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1242 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1243 if (policy->family == AF_INET) {
1251 if (policy->family == AF_INET) {
1260 if (policy->selector.prefixlen_d < dbits ||
1261 policy->selector.prefixlen_s < sbits)
1264 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1268 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1272 /* reset the bydst and inexact table in all directions */
1273 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1274 struct hlist_node *n;
1276 hlist_for_each_entry_safe(policy, n,
1277 &net->xfrm.policy_inexact[dir],
1279 hlist_del_init(&policy->bydst_inexact_list);
1281 hmask = net->xfrm.policy_bydst[dir].hmask;
1282 odst = net->xfrm.policy_bydst[dir].table;
1283 for (i = hmask; i >= 0; i--)
1284 INIT_HLIST_HEAD(odst + i);
1285 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1286 /* dir out => dst = remote, src = local */
1287 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1288 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1289 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1290 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1292 /* dir in/fwd => dst = local, src = remote */
1293 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1294 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1295 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1296 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1300 /* re-insert all policies by order of creation */
1301 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1302 if (policy->walk.dead)
1304 dir = xfrm_policy_id2dir(policy->index);
1305 if (dir >= XFRM_POLICY_MAX) {
1306 /* skip socket policies */
1310 chain = policy_hash_bysel(net, &policy->selector,
1311 policy->family, dir);
1313 hlist_del_rcu(&policy->bydst);
1316 void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1318 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1322 hlist_for_each_entry(pol, chain, bydst) {
1323 if (policy->priority >= pol->priority)
1324 newpos = &pol->bydst;
1329 hlist_add_behind_rcu(&policy->bydst, newpos);
1331 hlist_add_head_rcu(&policy->bydst, chain);
1335 __xfrm_policy_inexact_flush(net);
1336 write_seqcount_end(&xfrm_policy_hash_generation);
1337 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1339 mutex_unlock(&hash_resize_mutex);
1342 void xfrm_policy_hash_rebuild(struct net *net)
1344 schedule_work(&net->xfrm.policy_hthresh.work);
1346 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1348 /* Generate new index... KAME seems to generate them ordered by cost
1349 * of an absolute inpredictability of ordering of rules. This will not pass. */
1350 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1352 static u32 idx_generator;
1355 struct hlist_head *list;
1356 struct xfrm_policy *p;
1361 idx = (idx_generator | dir);
1370 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1372 hlist_for_each_entry(p, list, byidx) {
1373 if (p->index == idx) {
1383 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1385 u32 *p1 = (u32 *) s1;
1386 u32 *p2 = (u32 *) s2;
1387 int len = sizeof(struct xfrm_selector) / sizeof(u32);
1390 for (i = 0; i < len; i++) {
1398 static void xfrm_policy_requeue(struct xfrm_policy *old,
1399 struct xfrm_policy *new)
1401 struct xfrm_policy_queue *pq = &old->polq;
1402 struct sk_buff_head list;
1404 if (skb_queue_empty(&pq->hold_queue))
1407 __skb_queue_head_init(&list);
1409 spin_lock_bh(&pq->hold_queue.lock);
1410 skb_queue_splice_init(&pq->hold_queue, &list);
1411 if (del_timer(&pq->hold_timer))
1413 spin_unlock_bh(&pq->hold_queue.lock);
1417 spin_lock_bh(&pq->hold_queue.lock);
1418 skb_queue_splice(&list, &pq->hold_queue);
1419 pq->timeout = XFRM_QUEUE_TMO_MIN;
1420 if (!mod_timer(&pq->hold_timer, jiffies))
1422 spin_unlock_bh(&pq->hold_queue.lock);
1425 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
1426 struct xfrm_policy *pol)
1428 u32 mark = policy->mark.v & policy->mark.m;
1430 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
1433 if ((mark & pol->mark.m) == pol->mark.v &&
1434 policy->priority == pol->priority)
1440 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1442 const struct xfrm_pol_inexact_key *k = data;
1443 u32 a = k->type << 24 | k->dir << 16 | k->family;
1445 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1449 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1451 const struct xfrm_pol_inexact_bin *b = data;
1453 return xfrm_pol_bin_key(&b->k, 0, seed);
1456 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1459 const struct xfrm_pol_inexact_key *key = arg->key;
1460 const struct xfrm_pol_inexact_bin *b = ptr;
1463 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1466 ret = b->k.dir ^ key->dir;
1470 ret = b->k.type ^ key->type;
1474 ret = b->k.family ^ key->family;
1478 return b->k.if_id ^ key->if_id;
1481 static const struct rhashtable_params xfrm_pol_inexact_params = {
1482 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head),
1483 .hashfn = xfrm_pol_bin_key,
1484 .obj_hashfn = xfrm_pol_bin_obj,
1485 .obj_cmpfn = xfrm_pol_bin_cmp,
1486 .automatic_shrinking = true,
1489 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1490 struct xfrm_policy *policy)
1492 struct xfrm_policy *pol, *delpol = NULL;
1493 struct hlist_node *newpos = NULL;
1496 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1497 if (pol->type == policy->type &&
1498 pol->if_id == policy->if_id &&
1499 !selector_cmp(&pol->selector, &policy->selector) &&
1500 xfrm_policy_mark_match(policy, pol) &&
1501 xfrm_sec_ctx_match(pol->security, policy->security) &&
1504 if (policy->priority > pol->priority)
1506 } else if (policy->priority >= pol->priority) {
1507 newpos = &pol->bydst_inexact_list;
1515 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1517 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1519 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1525 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1526 struct xfrm_policy *policy,
1529 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1531 hlist_for_each_entry(pol, chain, bydst) {
1532 if (pol->type == policy->type &&
1533 pol->if_id == policy->if_id &&
1534 !selector_cmp(&pol->selector, &policy->selector) &&
1535 xfrm_policy_mark_match(policy, pol) &&
1536 xfrm_sec_ctx_match(pol->security, policy->security) &&
1539 return ERR_PTR(-EEXIST);
1541 if (policy->priority > pol->priority)
1543 } else if (policy->priority >= pol->priority) {
1552 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1554 hlist_add_head_rcu(&policy->bydst, chain);
1559 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1561 struct net *net = xp_net(policy);
1562 struct xfrm_policy *delpol;
1563 struct hlist_head *chain;
1565 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1566 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1568 delpol = xfrm_policy_insert_list(chain, policy, excl);
1570 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1572 if (IS_ERR(delpol)) {
1573 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1574 return PTR_ERR(delpol);
1577 __xfrm_policy_link(policy, dir);
1579 /* After previous checking, family can either be AF_INET or AF_INET6 */
1580 if (policy->family == AF_INET)
1581 rt_genid_bump_ipv4(net);
1583 rt_genid_bump_ipv6(net);
1586 xfrm_policy_requeue(delpol, policy);
1587 __xfrm_policy_unlink(delpol, dir);
1589 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1590 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1591 policy->curlft.add_time = ktime_get_real_seconds();
1592 policy->curlft.use_time = 0;
1593 if (!mod_timer(&policy->timer, jiffies + HZ))
1594 xfrm_pol_hold(policy);
1595 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1598 xfrm_policy_kill(delpol);
1599 else if (xfrm_bydst_should_resize(net, dir, NULL))
1600 schedule_work(&net->xfrm.policy_hash_work);
1604 EXPORT_SYMBOL(xfrm_policy_insert);
1606 static struct xfrm_policy *
1607 __xfrm_policy_bysel_ctx(struct hlist_head *chain, u32 mark, u32 if_id,
1609 struct xfrm_selector *sel,
1610 struct xfrm_sec_ctx *ctx)
1612 struct xfrm_policy *pol;
1617 hlist_for_each_entry(pol, chain, bydst) {
1618 if (pol->type == type &&
1619 pol->if_id == if_id &&
1620 (mark & pol->mark.m) == pol->mark.v &&
1621 !selector_cmp(sel, &pol->selector) &&
1622 xfrm_sec_ctx_match(ctx, pol->security))
1629 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u32 if_id,
1631 struct xfrm_selector *sel,
1632 struct xfrm_sec_ctx *ctx, int delete,
1635 struct xfrm_pol_inexact_bin *bin = NULL;
1636 struct xfrm_policy *pol, *ret = NULL;
1637 struct hlist_head *chain;
1640 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1641 chain = policy_hash_bysel(net, sel, sel->family, dir);
1643 struct xfrm_pol_inexact_candidates cand;
1646 bin = xfrm_policy_inexact_lookup(net, type,
1647 sel->family, dir, if_id);
1649 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1653 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1656 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1661 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1662 struct xfrm_policy *tmp;
1664 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1670 if (!pol || tmp->pos < pol->pos)
1674 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1681 *err = security_xfrm_policy_delete(pol->security);
1683 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1686 __xfrm_policy_unlink(pol, dir);
1690 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1693 xfrm_policy_kill(ret);
1695 xfrm_policy_inexact_prune_bin(bin);
1698 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1700 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u32 if_id,
1701 u8 type, int dir, u32 id, int delete,
1704 struct xfrm_policy *pol, *ret;
1705 struct hlist_head *chain;
1708 if (xfrm_policy_id2dir(id) != dir)
1712 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1713 chain = net->xfrm.policy_byidx + idx_hash(net, id);
1715 hlist_for_each_entry(pol, chain, byidx) {
1716 if (pol->type == type && pol->index == id &&
1717 pol->if_id == if_id &&
1718 (mark & pol->mark.m) == pol->mark.v) {
1721 *err = security_xfrm_policy_delete(
1724 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1727 __xfrm_policy_unlink(pol, dir);
1733 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1736 xfrm_policy_kill(ret);
1739 EXPORT_SYMBOL(xfrm_policy_byid);
1741 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1743 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1745 struct xfrm_policy *pol;
1748 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1749 if (pol->walk.dead ||
1750 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1754 err = security_xfrm_policy_delete(pol->security);
1756 xfrm_audit_policy_delete(pol, 0, task_valid);
1764 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1770 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1772 int dir, err = 0, cnt = 0;
1773 struct xfrm_policy *pol;
1775 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1777 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1782 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1783 dir = xfrm_policy_id2dir(pol->index);
1784 if (pol->walk.dead ||
1785 dir >= XFRM_POLICY_MAX ||
1789 __xfrm_policy_unlink(pol, dir);
1790 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1792 xfrm_audit_policy_delete(pol, 1, task_valid);
1793 xfrm_policy_kill(pol);
1794 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1798 __xfrm_policy_inexact_flush(net);
1802 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1805 EXPORT_SYMBOL(xfrm_policy_flush);
1807 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1808 int (*func)(struct xfrm_policy *, int, int, void*),
1811 struct xfrm_policy *pol;
1812 struct xfrm_policy_walk_entry *x;
1815 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1816 walk->type != XFRM_POLICY_TYPE_ANY)
1819 if (list_empty(&walk->walk.all) && walk->seq != 0)
1822 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1823 if (list_empty(&walk->walk.all))
1824 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1826 x = list_first_entry(&walk->walk.all,
1827 struct xfrm_policy_walk_entry, all);
1829 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1832 pol = container_of(x, struct xfrm_policy, walk);
1833 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1834 walk->type != pol->type)
1836 error = func(pol, xfrm_policy_id2dir(pol->index),
1839 list_move_tail(&walk->walk.all, &x->all);
1844 if (walk->seq == 0) {
1848 list_del_init(&walk->walk.all);
1850 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1853 EXPORT_SYMBOL(xfrm_policy_walk);
1855 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1857 INIT_LIST_HEAD(&walk->walk.all);
1858 walk->walk.dead = 1;
1862 EXPORT_SYMBOL(xfrm_policy_walk_init);
1864 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1866 if (list_empty(&walk->walk.all))
1869 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1870 list_del(&walk->walk.all);
1871 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1873 EXPORT_SYMBOL(xfrm_policy_walk_done);
1876 * Find policy to apply to this flow.
1878 * Returns 0 if policy found, else an -errno.
1880 static int xfrm_policy_match(const struct xfrm_policy *pol,
1881 const struct flowi *fl,
1882 u8 type, u16 family, int dir, u32 if_id)
1884 const struct xfrm_selector *sel = &pol->selector;
1888 if (pol->family != family ||
1889 pol->if_id != if_id ||
1890 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1894 match = xfrm_selector_match(sel, fl, family);
1896 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1901 static struct xfrm_pol_inexact_node *
1902 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1904 const xfrm_address_t *addr, u16 family)
1906 const struct rb_node *parent;
1910 seq = read_seqcount_begin(count);
1912 parent = rcu_dereference_raw(r->rb_node);
1914 struct xfrm_pol_inexact_node *node;
1917 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1919 delta = xfrm_policy_addr_delta(addr, &node->addr,
1920 node->prefixlen, family);
1922 parent = rcu_dereference_raw(parent->rb_left);
1924 } else if (delta > 0) {
1925 parent = rcu_dereference_raw(parent->rb_right);
1932 if (read_seqcount_retry(count, seq))
1939 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1940 struct xfrm_pol_inexact_bin *b,
1941 const xfrm_address_t *saddr,
1942 const xfrm_address_t *daddr)
1944 struct xfrm_pol_inexact_node *n;
1950 family = b->k.family;
1951 memset(cand, 0, sizeof(*cand));
1952 cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1954 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1957 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1958 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1961 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1964 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1967 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1972 static struct xfrm_pol_inexact_bin *
1973 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1976 struct xfrm_pol_inexact_key k = {
1983 write_pnet(&k.net, net);
1985 return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1986 xfrm_pol_inexact_params);
1989 static struct xfrm_pol_inexact_bin *
1990 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1993 struct xfrm_pol_inexact_bin *bin;
1995 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1998 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2004 static struct xfrm_policy *
2005 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2006 struct xfrm_policy *prefer,
2007 const struct flowi *fl,
2008 u8 type, u16 family, int dir, u32 if_id)
2010 u32 priority = prefer ? prefer->priority : ~0u;
2011 struct xfrm_policy *pol;
2016 hlist_for_each_entry_rcu(pol, chain, bydst) {
2019 if (pol->priority > priority)
2022 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2025 return ERR_PTR(err);
2031 /* matches. Is it older than *prefer? */
2032 if (pol->priority == priority &&
2033 prefer->pos < pol->pos)
2043 static struct xfrm_policy *
2044 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2045 struct xfrm_policy *prefer,
2046 const struct flowi *fl,
2047 u8 type, u16 family, int dir, u32 if_id)
2049 struct xfrm_policy *tmp;
2052 for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2053 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2055 fl, type, family, dir,
2068 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2069 const struct flowi *fl,
2073 struct xfrm_pol_inexact_candidates cand;
2074 const xfrm_address_t *daddr, *saddr;
2075 struct xfrm_pol_inexact_bin *bin;
2076 struct xfrm_policy *pol, *ret;
2077 struct hlist_head *chain;
2078 unsigned int sequence;
2081 daddr = xfrm_flowi_daddr(fl, family);
2082 saddr = xfrm_flowi_saddr(fl, family);
2083 if (unlikely(!daddr || !saddr))
2089 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2090 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2091 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2094 hlist_for_each_entry_rcu(pol, chain, bydst) {
2095 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2108 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2109 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2113 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2114 family, dir, if_id);
2122 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2125 if (ret && !xfrm_pol_hold_rcu(ret))
2133 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2134 const struct flowi *fl,
2135 u16 family, u8 dir, u32 if_id)
2137 #ifdef CONFIG_XFRM_SUB_POLICY
2138 struct xfrm_policy *pol;
2140 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2145 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2149 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2150 const struct flowi *fl,
2151 u16 family, u32 if_id)
2153 struct xfrm_policy *pol;
2157 pol = rcu_dereference(sk->sk_policy[dir]);
2162 if (pol->family != family) {
2167 match = xfrm_selector_match(&pol->selector, fl, family);
2169 if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2170 pol->if_id != if_id) {
2174 err = security_xfrm_policy_lookup(pol->security,
2178 if (!xfrm_pol_hold_rcu(pol))
2180 } else if (err == -ESRCH) {
2193 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2195 struct net *net = xp_net(pol);
2197 list_add(&pol->walk.all, &net->xfrm.policy_all);
2198 net->xfrm.policy_count[dir]++;
2202 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2205 struct net *net = xp_net(pol);
2207 if (list_empty(&pol->walk.all))
2210 /* Socket policies are not hashed. */
2211 if (!hlist_unhashed(&pol->bydst)) {
2212 hlist_del_rcu(&pol->bydst);
2213 hlist_del_init(&pol->bydst_inexact_list);
2214 hlist_del(&pol->byidx);
2217 list_del_init(&pol->walk.all);
2218 net->xfrm.policy_count[dir]--;
2223 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2225 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2228 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2230 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2233 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2235 struct net *net = xp_net(pol);
2237 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2238 pol = __xfrm_policy_unlink(pol, dir);
2239 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2241 xfrm_policy_kill(pol);
2246 EXPORT_SYMBOL(xfrm_policy_delete);
2248 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2250 struct net *net = sock_net(sk);
2251 struct xfrm_policy *old_pol;
2253 #ifdef CONFIG_XFRM_SUB_POLICY
2254 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2258 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2259 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2260 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2262 pol->curlft.add_time = ktime_get_real_seconds();
2263 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2264 xfrm_sk_policy_link(pol, dir);
2266 rcu_assign_pointer(sk->sk_policy[dir], pol);
2269 xfrm_policy_requeue(old_pol, pol);
2271 /* Unlinking succeeds always. This is the only function
2272 * allowed to delete or replace socket policy.
2274 xfrm_sk_policy_unlink(old_pol, dir);
2276 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2279 xfrm_policy_kill(old_pol);
2284 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2286 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2287 struct net *net = xp_net(old);
2290 newp->selector = old->selector;
2291 if (security_xfrm_policy_clone(old->security,
2294 return NULL; /* ENOMEM */
2296 newp->lft = old->lft;
2297 newp->curlft = old->curlft;
2298 newp->mark = old->mark;
2299 newp->if_id = old->if_id;
2300 newp->action = old->action;
2301 newp->flags = old->flags;
2302 newp->xfrm_nr = old->xfrm_nr;
2303 newp->index = old->index;
2304 newp->type = old->type;
2305 newp->family = old->family;
2306 memcpy(newp->xfrm_vec, old->xfrm_vec,
2307 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2308 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2309 xfrm_sk_policy_link(newp, dir);
2310 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2316 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2318 const struct xfrm_policy *p;
2319 struct xfrm_policy *np;
2323 for (i = 0; i < 2; i++) {
2324 p = rcu_dereference(osk->sk_policy[i]);
2326 np = clone_policy(p, i);
2327 if (unlikely(!np)) {
2331 rcu_assign_pointer(sk->sk_policy[i], np);
2339 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2340 xfrm_address_t *remote, unsigned short family, u32 mark)
2343 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2345 if (unlikely(afinfo == NULL))
2347 err = afinfo->get_saddr(net, oif, local, remote, mark);
2352 /* Resolve list of templates for the flow, given policy. */
2355 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2356 struct xfrm_state **xfrm, unsigned short family)
2358 struct net *net = xp_net(policy);
2361 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2362 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2365 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2366 struct xfrm_state *x;
2367 xfrm_address_t *remote = daddr;
2368 xfrm_address_t *local = saddr;
2369 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2371 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2372 tmpl->mode == XFRM_MODE_BEET) {
2373 remote = &tmpl->id.daddr;
2374 local = &tmpl->saddr;
2375 if (xfrm_addr_any(local, tmpl->encap_family)) {
2376 error = xfrm_get_saddr(net, fl->flowi_oif,
2378 tmpl->encap_family, 0);
2385 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2386 family, policy->if_id);
2388 if (x && x->km.state == XFRM_STATE_VALID) {
2395 error = (x->km.state == XFRM_STATE_ERROR ?
2398 } else if (error == -ESRCH) {
2402 if (!tmpl->optional)
2408 for (nx--; nx >= 0; nx--)
2409 xfrm_state_put(xfrm[nx]);
2414 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2415 struct xfrm_state **xfrm, unsigned short family)
2417 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2418 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2424 for (i = 0; i < npols; i++) {
2425 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2430 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2438 /* found states are sorted for outbound processing */
2440 xfrm_state_sort(xfrm, tpp, cnx, family);
2445 for (cnx--; cnx >= 0; cnx--)
2446 xfrm_state_put(tpp[cnx]);
2451 static int xfrm_get_tos(const struct flowi *fl, int family)
2453 const struct xfrm_policy_afinfo *afinfo;
2456 afinfo = xfrm_policy_get_afinfo(family);
2460 tos = afinfo->get_tos(fl);
2467 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2469 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2470 struct dst_ops *dst_ops;
2471 struct xfrm_dst *xdst;
2474 return ERR_PTR(-EINVAL);
2478 dst_ops = &net->xfrm.xfrm4_dst_ops;
2480 #if IS_ENABLED(CONFIG_IPV6)
2482 dst_ops = &net->xfrm.xfrm6_dst_ops;
2488 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2491 struct dst_entry *dst = &xdst->u.dst;
2493 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2495 xdst = ERR_PTR(-ENOBUFS);
2502 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2505 const struct xfrm_policy_afinfo *afinfo =
2506 xfrm_policy_get_afinfo(dst->ops->family);
2512 err = afinfo->init_path(path, dst, nfheader_len);
2519 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2520 const struct flowi *fl)
2522 const struct xfrm_policy_afinfo *afinfo =
2523 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2529 err = afinfo->fill_dst(xdst, dev, fl);
2537 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2538 * all the metrics... Shortly, bundle a bundle.
2541 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2542 struct xfrm_state **xfrm,
2543 struct xfrm_dst **bundle,
2545 const struct flowi *fl,
2546 struct dst_entry *dst)
2548 struct net *net = xp_net(policy);
2549 unsigned long now = jiffies;
2550 struct net_device *dev;
2551 struct xfrm_mode *inner_mode;
2552 struct xfrm_dst *xdst_prev = NULL;
2553 struct xfrm_dst *xdst0 = NULL;
2557 int nfheader_len = 0;
2558 int trailer_len = 0;
2560 int family = policy->selector.family;
2561 xfrm_address_t saddr, daddr;
2563 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2565 tos = xfrm_get_tos(fl, family);
2569 for (; i < nx; i++) {
2570 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2571 struct dst_entry *dst1 = &xdst->u.dst;
2573 err = PTR_ERR(xdst);
2583 /* Ref count is taken during xfrm_alloc_dst()
2584 * No need to do dst_clone() on dst1
2586 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2588 if (xfrm[i]->sel.family == AF_UNSPEC) {
2589 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2590 xfrm_af2proto(family));
2592 err = -EAFNOSUPPORT;
2597 inner_mode = xfrm[i]->inner_mode;
2600 dst_copy_metrics(dst1, dst);
2602 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2605 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2606 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2608 family = xfrm[i]->props.family;
2609 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2610 &saddr, &daddr, family, mark);
2617 dst1->xfrm = xfrm[i];
2618 xdst->xfrm_genid = xfrm[i]->genid;
2620 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2621 dst1->flags |= DST_HOST;
2622 dst1->lastuse = now;
2624 dst1->input = dst_discard;
2625 dst1->output = inner_mode->afinfo->output;
2629 header_len += xfrm[i]->props.header_len;
2630 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2631 nfheader_len += xfrm[i]->props.header_len;
2632 trailer_len += xfrm[i]->props.trailer_len;
2635 xfrm_dst_set_child(xdst_prev, dst);
2643 xfrm_init_path(xdst0, dst, nfheader_len);
2644 xfrm_init_pmtu(bundle, nx);
2646 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2647 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2648 err = xfrm_fill_dst(xdst_prev, dev, fl);
2652 xdst_prev->u.dst.header_len = header_len;
2653 xdst_prev->u.dst.trailer_len = trailer_len;
2654 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2655 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2658 return &xdst0->u.dst;
2662 xfrm_state_put(xfrm[i]);
2665 dst_release_immediate(&xdst0->u.dst);
2667 return ERR_PTR(err);
2670 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2671 struct xfrm_policy **pols,
2672 int *num_pols, int *num_xfrms)
2676 if (*num_pols == 0 || !pols[0]) {
2681 if (IS_ERR(pols[0]))
2682 return PTR_ERR(pols[0]);
2684 *num_xfrms = pols[0]->xfrm_nr;
2686 #ifdef CONFIG_XFRM_SUB_POLICY
2687 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2688 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2689 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2690 XFRM_POLICY_TYPE_MAIN,
2695 if (IS_ERR(pols[1])) {
2696 xfrm_pols_put(pols, *num_pols);
2697 return PTR_ERR(pols[1]);
2700 (*num_xfrms) += pols[1]->xfrm_nr;
2704 for (i = 0; i < *num_pols; i++) {
2705 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2715 static struct xfrm_dst *
2716 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2717 const struct flowi *fl, u16 family,
2718 struct dst_entry *dst_orig)
2720 struct net *net = xp_net(pols[0]);
2721 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2722 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2723 struct xfrm_dst *xdst;
2724 struct dst_entry *dst;
2727 /* Try to instantiate a bundle */
2728 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2734 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2735 return ERR_PTR(err);
2738 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2740 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2741 return ERR_CAST(dst);
2744 xdst = (struct xfrm_dst *)dst;
2745 xdst->num_xfrms = err;
2746 xdst->num_pols = num_pols;
2747 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2748 xdst->policy_genid = atomic_read(&pols[0]->genid);
2753 static void xfrm_policy_queue_process(struct timer_list *t)
2755 struct sk_buff *skb;
2757 struct dst_entry *dst;
2758 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2759 struct net *net = xp_net(pol);
2760 struct xfrm_policy_queue *pq = &pol->polq;
2762 struct sk_buff_head list;
2764 spin_lock(&pq->hold_queue.lock);
2765 skb = skb_peek(&pq->hold_queue);
2767 spin_unlock(&pq->hold_queue.lock);
2772 xfrm_decode_session(skb, &fl, dst->ops->family);
2773 spin_unlock(&pq->hold_queue.lock);
2775 dst_hold(xfrm_dst_path(dst));
2776 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2780 if (dst->flags & DST_XFRM_QUEUE) {
2783 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2786 pq->timeout = pq->timeout << 1;
2787 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2794 __skb_queue_head_init(&list);
2796 spin_lock(&pq->hold_queue.lock);
2798 skb_queue_splice_init(&pq->hold_queue, &list);
2799 spin_unlock(&pq->hold_queue.lock);
2801 while (!skb_queue_empty(&list)) {
2802 skb = __skb_dequeue(&list);
2804 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2805 dst_hold(xfrm_dst_path(skb_dst(skb)));
2806 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2814 skb_dst_set(skb, dst);
2816 dst_output(net, skb->sk, skb);
2825 skb_queue_purge(&pq->hold_queue);
2829 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2831 unsigned long sched_next;
2832 struct dst_entry *dst = skb_dst(skb);
2833 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2834 struct xfrm_policy *pol = xdst->pols[0];
2835 struct xfrm_policy_queue *pq = &pol->polq;
2837 if (unlikely(skb_fclone_busy(sk, skb))) {
2842 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2849 spin_lock_bh(&pq->hold_queue.lock);
2852 pq->timeout = XFRM_QUEUE_TMO_MIN;
2854 sched_next = jiffies + pq->timeout;
2856 if (del_timer(&pq->hold_timer)) {
2857 if (time_before(pq->hold_timer.expires, sched_next))
2858 sched_next = pq->hold_timer.expires;
2862 __skb_queue_tail(&pq->hold_queue, skb);
2863 if (!mod_timer(&pq->hold_timer, sched_next))
2866 spin_unlock_bh(&pq->hold_queue.lock);
2871 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2872 struct xfrm_flo *xflo,
2873 const struct flowi *fl,
2878 struct net_device *dev;
2879 struct dst_entry *dst;
2880 struct dst_entry *dst1;
2881 struct xfrm_dst *xdst;
2883 xdst = xfrm_alloc_dst(net, family);
2887 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2888 net->xfrm.sysctl_larval_drop ||
2892 dst = xflo->dst_orig;
2893 dst1 = &xdst->u.dst;
2897 dst_copy_metrics(dst1, dst);
2899 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2900 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2901 dst1->lastuse = jiffies;
2903 dst1->input = dst_discard;
2904 dst1->output = xdst_queue_output;
2907 xfrm_dst_set_child(xdst, dst);
2910 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2917 err = xfrm_fill_dst(xdst, dev, fl);
2926 xdst = ERR_PTR(err);
2930 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2931 const struct flowi *fl,
2933 struct xfrm_flo *xflo, u32 if_id)
2935 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2936 int num_pols = 0, num_xfrms = 0, err;
2937 struct xfrm_dst *xdst;
2939 /* Resolve policies to use if we couldn't get them from
2940 * previous cache entry */
2942 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2943 err = xfrm_expand_policies(fl, family, pols,
2944 &num_pols, &num_xfrms);
2950 goto make_dummy_bundle;
2952 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2955 err = PTR_ERR(xdst);
2956 if (err == -EREMOTE) {
2957 xfrm_pols_put(pols, num_pols);
2963 goto make_dummy_bundle;
2964 } else if (xdst == NULL) {
2966 goto make_dummy_bundle;
2972 /* We found policies, but there's no bundles to instantiate:
2973 * either because the policy blocks, has no transformations or
2974 * we could not build template (no xfrm_states).*/
2975 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2977 xfrm_pols_put(pols, num_pols);
2978 return ERR_CAST(xdst);
2980 xdst->num_pols = num_pols;
2981 xdst->num_xfrms = num_xfrms;
2982 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2987 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2989 xfrm_pols_put(pols, num_pols);
2990 return ERR_PTR(err);
2993 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2994 struct dst_entry *dst_orig)
2996 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2997 struct dst_entry *ret;
3000 dst_release(dst_orig);
3001 return ERR_PTR(-EINVAL);
3003 ret = afinfo->blackhole_route(net, dst_orig);
3010 /* Finds/creates a bundle for given flow and if_id
3012 * At the moment we eat a raw IP route. Mostly to speed up lookups
3013 * on interfaces with disabled IPsec.
3015 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3018 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3019 struct dst_entry *dst_orig,
3020 const struct flowi *fl,
3021 const struct sock *sk,
3022 int flags, u32 if_id)
3024 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3025 struct xfrm_dst *xdst;
3026 struct dst_entry *dst, *route;
3027 u16 family = dst_orig->ops->family;
3028 u8 dir = XFRM_POLICY_OUT;
3029 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3035 sk = sk_const_to_full_sk(sk);
3036 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3038 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3040 err = xfrm_expand_policies(fl, family, pols,
3041 &num_pols, &num_xfrms);
3046 if (num_xfrms <= 0) {
3047 drop_pols = num_pols;
3051 xdst = xfrm_resolve_and_create_bundle(
3056 xfrm_pols_put(pols, num_pols);
3057 err = PTR_ERR(xdst);
3058 if (err == -EREMOTE)
3062 } else if (xdst == NULL) {
3064 drop_pols = num_pols;
3068 route = xdst->route;
3073 struct xfrm_flo xflo;
3075 xflo.dst_orig = dst_orig;
3078 /* To accelerate a bit... */
3079 if ((dst_orig->flags & DST_NOXFRM) ||
3080 !net->xfrm.policy_count[XFRM_POLICY_OUT])
3083 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3087 err = PTR_ERR(xdst);
3091 num_pols = xdst->num_pols;
3092 num_xfrms = xdst->num_xfrms;
3093 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3094 route = xdst->route;
3098 if (route == NULL && num_xfrms > 0) {
3099 /* The only case when xfrm_bundle_lookup() returns a
3100 * bundle with null route, is when the template could
3101 * not be resolved. It means policies are there, but
3102 * bundle could not be created, since we don't yet
3103 * have the xfrm_state's. We need to wait for KM to
3104 * negotiate new SA's or bail out with error.*/
3105 if (net->xfrm.sysctl_larval_drop) {
3106 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3113 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3121 if ((flags & XFRM_LOOKUP_ICMP) &&
3122 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3127 for (i = 0; i < num_pols; i++)
3128 pols[i]->curlft.use_time = ktime_get_real_seconds();
3130 if (num_xfrms < 0) {
3131 /* Prohibit the flow */
3132 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3135 } else if (num_xfrms > 0) {
3136 /* Flow transformed */
3137 dst_release(dst_orig);
3139 /* Flow passes untransformed */
3144 xfrm_pols_put(pols, drop_pols);
3145 if (dst && dst->xfrm &&
3146 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3147 dst->flags |= DST_XFRM_TUNNEL;
3151 if (!(flags & XFRM_LOOKUP_ICMP)) {
3159 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3160 dst_release(dst_orig);
3161 xfrm_pols_put(pols, drop_pols);
3162 return ERR_PTR(err);
3164 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3166 /* Main function: finds/creates a bundle for given flow.
3168 * At the moment we eat a raw IP route. Mostly to speed up lookups
3169 * on interfaces with disabled IPsec.
3171 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3172 const struct flowi *fl, const struct sock *sk,
3175 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3177 EXPORT_SYMBOL(xfrm_lookup);
3179 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3180 * Otherwise we may send out blackholed packets.
3182 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3183 const struct flowi *fl,
3184 const struct sock *sk, int flags)
3186 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3187 flags | XFRM_LOOKUP_QUEUE |
3188 XFRM_LOOKUP_KEEP_DST_REF);
3190 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
3191 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3194 dst_release(dst_orig);
3198 EXPORT_SYMBOL(xfrm_lookup_route);
3201 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3203 struct sec_path *sp = skb_sec_path(skb);
3204 struct xfrm_state *x;
3206 if (!sp || idx < 0 || idx >= sp->len)
3209 if (!x->type->reject)
3211 return x->type->reject(x, skb, fl);
3214 /* When skb is transformed back to its "native" form, we have to
3215 * check policy restrictions. At the moment we make this in maximally
3216 * stupid way. Shame on me. :-) Of course, connected sockets must
3217 * have policy cached at them.
3221 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3222 unsigned short family)
3224 if (xfrm_state_kern(x))
3225 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3226 return x->id.proto == tmpl->id.proto &&
3227 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3228 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3229 x->props.mode == tmpl->mode &&
3230 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3231 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3232 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3233 xfrm_state_addr_cmp(tmpl, x, family));
3237 * 0 or more than 0 is returned when validation is succeeded (either bypass
3238 * because of optional transport mode, or next index of the mathced secpath
3239 * state with the template.
3240 * -1 is returned when no matching template is found.
3241 * Otherwise "-2 - errored_index" is returned.
3244 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3245 unsigned short family)
3249 if (tmpl->optional) {
3250 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3254 for (; idx < sp->len; idx++) {
3255 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3257 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3266 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3267 unsigned int family, int reverse)
3269 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3272 if (unlikely(afinfo == NULL))
3273 return -EAFNOSUPPORT;
3275 afinfo->decode_session(skb, fl, reverse);
3277 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
3281 EXPORT_SYMBOL(__xfrm_decode_session);
3283 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3285 for (; k < sp->len; k++) {
3286 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3295 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3296 unsigned short family)
3298 struct net *net = dev_net(skb->dev);
3299 struct xfrm_policy *pol;
3300 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3307 const struct xfrm_if_cb *ifcb;
3308 struct sec_path *sp;
3313 ifcb = xfrm_if_get_cb();
3316 xi = ifcb->decode_session(skb);
3318 if_id = xi->p.if_id;
3324 reverse = dir & ~XFRM_POLICY_MASK;
3325 dir &= XFRM_POLICY_MASK;
3327 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3328 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3332 nf_nat_decode_session(skb, &fl, family);
3334 /* First, check used SA against their selectors. */
3335 sp = skb_sec_path(skb);
3339 for (i = sp->len - 1; i >= 0; i--) {
3340 struct xfrm_state *x = sp->xvec[i];
3341 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3342 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3349 sk = sk_to_full_sk(sk);
3350 if (sk && sk->sk_policy[dir]) {
3351 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3353 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3359 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3362 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3367 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3368 xfrm_secpath_reject(xerr_idx, skb, &fl);
3369 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3375 pol->curlft.use_time = ktime_get_real_seconds();
3379 #ifdef CONFIG_XFRM_SUB_POLICY
3380 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3381 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3383 XFRM_POLICY_IN, if_id);
3385 if (IS_ERR(pols[1])) {
3386 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3389 pols[1]->curlft.use_time = ktime_get_real_seconds();
3395 if (pol->action == XFRM_POLICY_ALLOW) {
3396 static struct sec_path dummy;
3397 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3398 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3399 struct xfrm_tmpl **tpp = tp;
3403 sp = skb_sec_path(skb);
3407 for (pi = 0; pi < npols; pi++) {
3408 if (pols[pi] != pol &&
3409 pols[pi]->action != XFRM_POLICY_ALLOW) {
3410 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3413 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3414 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3417 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3418 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3422 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
3426 /* For each tunnel xfrm, find the first matching tmpl.
3427 * For each tmpl before that, find corresponding xfrm.
3428 * Order is _important_. Later we will implement
3429 * some barriers, but at the moment barriers
3430 * are implied between each two transformations.
3432 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3433 k = xfrm_policy_ok(tpp[i], sp, k, family);
3436 /* "-2 - errored_index" returned */
3438 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3443 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3444 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3448 xfrm_pols_put(pols, npols);
3451 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3454 xfrm_secpath_reject(xerr_idx, skb, &fl);
3456 xfrm_pols_put(pols, npols);
3459 EXPORT_SYMBOL(__xfrm_policy_check);
3461 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3463 struct net *net = dev_net(skb->dev);
3465 struct dst_entry *dst;
3468 if (xfrm_decode_session(skb, &fl, family) < 0) {
3469 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3474 if (!skb_dst(skb)) {
3475 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3479 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3484 skb_dst_set(skb, dst);
3487 EXPORT_SYMBOL(__xfrm_route_forward);
3489 /* Optimize later using cookies and generation ids. */
3491 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3493 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3494 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3495 * get validated by dst_ops->check on every use. We do this
3496 * because when a normal route referenced by an XFRM dst is
3497 * obsoleted we do not go looking around for all parent
3498 * referencing XFRM dsts so that we can invalidate them. It
3499 * is just too much work. Instead we make the checks here on
3500 * every use. For example:
3502 * XFRM dst A --> IPv4 dst X
3504 * X is the "xdst->route" of A (X is also the "dst->path" of A
3505 * in this example). If X is marked obsolete, "A" will not
3506 * notice. That's what we are validating here via the
3507 * stale_bundle() check.
3509 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3511 * This will force stale_bundle() to fail on any xdst bundle with
3512 * this dst linked in it.
3514 if (dst->obsolete < 0 && !stale_bundle(dst))
3520 static int stale_bundle(struct dst_entry *dst)
3522 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3525 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3527 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3528 dst->dev = dev_net(dev)->loopback_dev;
3533 EXPORT_SYMBOL(xfrm_dst_ifdown);
3535 static void xfrm_link_failure(struct sk_buff *skb)
3537 /* Impossible. Such dst must be popped before reaches point of failure. */
3540 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3543 if (dst->obsolete) {
3551 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3554 struct xfrm_dst *xdst = bundle[nr];
3555 u32 pmtu, route_mtu_cached;
3556 struct dst_entry *dst;
3559 pmtu = dst_mtu(xfrm_dst_child(dst));
3560 xdst->child_mtu_cached = pmtu;
3562 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3564 route_mtu_cached = dst_mtu(xdst->route);
3565 xdst->route_mtu_cached = route_mtu_cached;
3567 if (pmtu > route_mtu_cached)
3568 pmtu = route_mtu_cached;
3570 dst_metric_set(dst, RTAX_MTU, pmtu);
3574 /* Check that the bundle accepts the flow and its components are
3578 static int xfrm_bundle_ok(struct xfrm_dst *first)
3580 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3581 struct dst_entry *dst = &first->u.dst;
3582 struct xfrm_dst *xdst;
3586 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3587 (dst->dev && !netif_running(dst->dev)))
3590 if (dst->flags & DST_XFRM_QUEUE)
3593 start_from = nr = 0;
3595 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3597 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3599 if (xdst->xfrm_genid != dst->xfrm->genid)
3601 if (xdst->num_pols > 0 &&
3602 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3605 bundle[nr++] = xdst;
3607 mtu = dst_mtu(xfrm_dst_child(dst));
3608 if (xdst->child_mtu_cached != mtu) {
3610 xdst->child_mtu_cached = mtu;
3613 if (!dst_check(xdst->route, xdst->route_cookie))
3615 mtu = dst_mtu(xdst->route);
3616 if (xdst->route_mtu_cached != mtu) {
3618 xdst->route_mtu_cached = mtu;
3621 dst = xfrm_dst_child(dst);
3622 } while (dst->xfrm);
3624 if (likely(!start_from))
3627 xdst = bundle[start_from - 1];
3628 mtu = xdst->child_mtu_cached;
3629 while (start_from--) {
3632 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3633 if (mtu > xdst->route_mtu_cached)
3634 mtu = xdst->route_mtu_cached;
3635 dst_metric_set(dst, RTAX_MTU, mtu);
3639 xdst = bundle[start_from - 1];
3640 xdst->child_mtu_cached = mtu;
3646 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3648 return dst_metric_advmss(xfrm_dst_path(dst));
3651 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3653 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3655 return mtu ? : dst_mtu(xfrm_dst_path(dst));
3658 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3662 const struct xfrm_state *xfrm = dst->xfrm;
3664 dst = xfrm_dst_child(dst);
3666 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3668 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3669 daddr = xfrm->coaddr;
3670 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3671 daddr = &xfrm->id.daddr;
3676 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3677 struct sk_buff *skb,
3680 const struct dst_entry *path = xfrm_dst_path(dst);
3683 daddr = xfrm_get_dst_nexthop(dst, daddr);
3684 return path->ops->neigh_lookup(path, skb, daddr);
3687 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3689 const struct dst_entry *path = xfrm_dst_path(dst);
3691 daddr = xfrm_get_dst_nexthop(dst, daddr);
3692 path->ops->confirm_neigh(path, daddr);
3695 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3699 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3700 return -EAFNOSUPPORT;
3702 spin_lock(&xfrm_policy_afinfo_lock);
3703 if (unlikely(xfrm_policy_afinfo[family] != NULL))
3706 struct dst_ops *dst_ops = afinfo->dst_ops;
3707 if (likely(dst_ops->kmem_cachep == NULL))
3708 dst_ops->kmem_cachep = xfrm_dst_cache;
3709 if (likely(dst_ops->check == NULL))
3710 dst_ops->check = xfrm_dst_check;
3711 if (likely(dst_ops->default_advmss == NULL))
3712 dst_ops->default_advmss = xfrm_default_advmss;
3713 if (likely(dst_ops->mtu == NULL))
3714 dst_ops->mtu = xfrm_mtu;
3715 if (likely(dst_ops->negative_advice == NULL))
3716 dst_ops->negative_advice = xfrm_negative_advice;
3717 if (likely(dst_ops->link_failure == NULL))
3718 dst_ops->link_failure = xfrm_link_failure;
3719 if (likely(dst_ops->neigh_lookup == NULL))
3720 dst_ops->neigh_lookup = xfrm_neigh_lookup;
3721 if (likely(!dst_ops->confirm_neigh))
3722 dst_ops->confirm_neigh = xfrm_confirm_neigh;
3723 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3725 spin_unlock(&xfrm_policy_afinfo_lock);
3729 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3731 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3733 struct dst_ops *dst_ops = afinfo->dst_ops;
3736 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3737 if (xfrm_policy_afinfo[i] != afinfo)
3739 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3745 dst_ops->kmem_cachep = NULL;
3746 dst_ops->check = NULL;
3747 dst_ops->negative_advice = NULL;
3748 dst_ops->link_failure = NULL;
3750 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3752 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3754 spin_lock(&xfrm_if_cb_lock);
3755 rcu_assign_pointer(xfrm_if_cb, ifcb);
3756 spin_unlock(&xfrm_if_cb_lock);
3758 EXPORT_SYMBOL(xfrm_if_register_cb);
3760 void xfrm_if_unregister_cb(void)
3762 RCU_INIT_POINTER(xfrm_if_cb, NULL);
3765 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3767 #ifdef CONFIG_XFRM_STATISTICS
3768 static int __net_init xfrm_statistics_init(struct net *net)
3771 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3772 if (!net->mib.xfrm_statistics)
3774 rv = xfrm_proc_init(net);
3776 free_percpu(net->mib.xfrm_statistics);
3780 static void xfrm_statistics_fini(struct net *net)
3782 xfrm_proc_fini(net);
3783 free_percpu(net->mib.xfrm_statistics);
3786 static int __net_init xfrm_statistics_init(struct net *net)
3791 static void xfrm_statistics_fini(struct net *net)
3796 static int __net_init xfrm_policy_init(struct net *net)
3798 unsigned int hmask, sz;
3801 if (net_eq(net, &init_net)) {
3802 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
3803 sizeof(struct xfrm_dst),
3804 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3806 err = rhashtable_init(&xfrm_policy_inexact_table,
3807 &xfrm_pol_inexact_params);
3812 sz = (hmask+1) * sizeof(struct hlist_head);
3814 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
3815 if (!net->xfrm.policy_byidx)
3817 net->xfrm.policy_idx_hmask = hmask;
3819 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3820 struct xfrm_policy_hash *htab;
3822 net->xfrm.policy_count[dir] = 0;
3823 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
3824 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
3826 htab = &net->xfrm.policy_bydst[dir];
3827 htab->table = xfrm_hash_alloc(sz);
3830 htab->hmask = hmask;
3836 net->xfrm.policy_hthresh.lbits4 = 32;
3837 net->xfrm.policy_hthresh.rbits4 = 32;
3838 net->xfrm.policy_hthresh.lbits6 = 128;
3839 net->xfrm.policy_hthresh.rbits6 = 128;
3841 seqlock_init(&net->xfrm.policy_hthresh.lock);
3843 INIT_LIST_HEAD(&net->xfrm.policy_all);
3844 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
3845 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
3846 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
3850 for (dir--; dir >= 0; dir--) {
3851 struct xfrm_policy_hash *htab;
3853 htab = &net->xfrm.policy_bydst[dir];
3854 xfrm_hash_free(htab->table, sz);
3856 xfrm_hash_free(net->xfrm.policy_byidx, sz);
3861 static void xfrm_policy_fini(struct net *net)
3863 struct xfrm_pol_inexact_bin *b, *t;
3867 flush_work(&net->xfrm.policy_hash_work);
3868 #ifdef CONFIG_XFRM_SUB_POLICY
3869 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3871 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3873 WARN_ON(!list_empty(&net->xfrm.policy_all));
3875 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3876 struct xfrm_policy_hash *htab;
3878 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3880 htab = &net->xfrm.policy_bydst[dir];
3881 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3882 WARN_ON(!hlist_empty(htab->table));
3883 xfrm_hash_free(htab->table, sz);
3886 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3887 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3888 xfrm_hash_free(net->xfrm.policy_byidx, sz);
3890 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3891 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
3892 __xfrm_policy_inexact_prune_bin(b, true);
3893 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3896 static int __net_init xfrm_net_init(struct net *net)
3900 /* Initialize the per-net locks here */
3901 spin_lock_init(&net->xfrm.xfrm_state_lock);
3902 spin_lock_init(&net->xfrm.xfrm_policy_lock);
3903 mutex_init(&net->xfrm.xfrm_cfg_mutex);
3905 rv = xfrm_statistics_init(net);
3907 goto out_statistics;
3908 rv = xfrm_state_init(net);
3911 rv = xfrm_policy_init(net);
3914 rv = xfrm_sysctl_init(net);
3921 xfrm_policy_fini(net);
3923 xfrm_state_fini(net);
3925 xfrm_statistics_fini(net);
3930 static void __net_exit xfrm_net_exit(struct net *net)
3932 xfrm_sysctl_fini(net);
3933 xfrm_policy_fini(net);
3934 xfrm_state_fini(net);
3935 xfrm_statistics_fini(net);
3938 static struct pernet_operations __net_initdata xfrm_net_ops = {
3939 .init = xfrm_net_init,
3940 .exit = xfrm_net_exit,
3943 void __init xfrm_init(void)
3945 register_pernet_subsys(&xfrm_net_ops);
3947 seqcount_init(&xfrm_policy_hash_generation);
3950 RCU_INIT_POINTER(xfrm_if_cb, NULL);
3954 #ifdef CONFIG_AUDITSYSCALL
3955 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3956 struct audit_buffer *audit_buf)
3958 struct xfrm_sec_ctx *ctx = xp->security;
3959 struct xfrm_selector *sel = &xp->selector;
3962 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3963 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3965 switch (sel->family) {
3967 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3968 if (sel->prefixlen_s != 32)
3969 audit_log_format(audit_buf, " src_prefixlen=%d",
3971 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3972 if (sel->prefixlen_d != 32)
3973 audit_log_format(audit_buf, " dst_prefixlen=%d",
3977 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3978 if (sel->prefixlen_s != 128)
3979 audit_log_format(audit_buf, " src_prefixlen=%d",
3981 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3982 if (sel->prefixlen_d != 128)
3983 audit_log_format(audit_buf, " dst_prefixlen=%d",
3989 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3991 struct audit_buffer *audit_buf;
3993 audit_buf = xfrm_audit_start("SPD-add");
3994 if (audit_buf == NULL)
3996 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3997 audit_log_format(audit_buf, " res=%u", result);
3998 xfrm_audit_common_policyinfo(xp, audit_buf);
3999 audit_log_end(audit_buf);
4001 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4003 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4006 struct audit_buffer *audit_buf;
4008 audit_buf = xfrm_audit_start("SPD-delete");
4009 if (audit_buf == NULL)
4011 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4012 audit_log_format(audit_buf, " res=%u", result);
4013 xfrm_audit_common_policyinfo(xp, audit_buf);
4014 audit_log_end(audit_buf);
4016 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4019 #ifdef CONFIG_XFRM_MIGRATE
4020 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4021 const struct xfrm_selector *sel_tgt)
4023 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4024 if (sel_tgt->family == sel_cmp->family &&
4025 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4027 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4029 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4030 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4034 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4041 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4042 u8 dir, u8 type, struct net *net)
4044 struct xfrm_policy *pol, *ret = NULL;
4045 struct hlist_head *chain;
4048 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4049 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4050 hlist_for_each_entry(pol, chain, bydst) {
4051 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4052 pol->type == type) {
4054 priority = ret->priority;
4058 chain = &net->xfrm.policy_inexact[dir];
4059 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4060 if ((pol->priority >= priority) && ret)
4063 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4064 pol->type == type) {
4072 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4077 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4081 if (t->mode == m->mode && t->id.proto == m->proto &&
4082 (m->reqid == 0 || t->reqid == m->reqid)) {
4084 case XFRM_MODE_TUNNEL:
4085 case XFRM_MODE_BEET:
4086 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4088 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4093 case XFRM_MODE_TRANSPORT:
4094 /* in case of transport mode, template does not store
4095 any IP addresses, hence we just compare mode and
4106 /* update endpoint address(es) of template(s) */
4107 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4108 struct xfrm_migrate *m, int num_migrate)
4110 struct xfrm_migrate *mp;
4113 write_lock_bh(&pol->lock);
4114 if (unlikely(pol->walk.dead)) {
4115 /* target policy has been deleted */
4116 write_unlock_bh(&pol->lock);
4120 for (i = 0; i < pol->xfrm_nr; i++) {
4121 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4122 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4125 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4126 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4128 /* update endpoints */
4129 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4130 sizeof(pol->xfrm_vec[i].id.daddr));
4131 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4132 sizeof(pol->xfrm_vec[i].saddr));
4133 pol->xfrm_vec[i].encap_family = mp->new_family;
4135 atomic_inc(&pol->genid);
4139 write_unlock_bh(&pol->lock);
4147 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4151 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4154 for (i = 0; i < num_migrate; i++) {
4155 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4156 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4159 /* check if there is any duplicated entry */
4160 for (j = i + 1; j < num_migrate; j++) {
4161 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4162 sizeof(m[i].old_daddr)) &&
4163 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4164 sizeof(m[i].old_saddr)) &&
4165 m[i].proto == m[j].proto &&
4166 m[i].mode == m[j].mode &&
4167 m[i].reqid == m[j].reqid &&
4168 m[i].old_family == m[j].old_family)
4176 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4177 struct xfrm_migrate *m, int num_migrate,
4178 struct xfrm_kmaddress *k, struct net *net,
4179 struct xfrm_encap_tmpl *encap)
4181 int i, err, nx_cur = 0, nx_new = 0;
4182 struct xfrm_policy *pol = NULL;
4183 struct xfrm_state *x, *xc;
4184 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4185 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4186 struct xfrm_migrate *mp;
4188 /* Stage 0 - sanity checks */
4189 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4192 if (dir >= XFRM_POLICY_MAX) {
4197 /* Stage 1 - find policy */
4198 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4203 /* Stage 2 - find and update state(s) */
4204 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4205 if ((x = xfrm_migrate_state_find(mp, net))) {
4208 xc = xfrm_state_migrate(x, mp, encap);
4219 /* Stage 3 - update policy */
4220 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4223 /* Stage 4 - delete old state(s) */
4225 xfrm_states_put(x_cur, nx_cur);
4226 xfrm_states_delete(x_cur, nx_cur);
4229 /* Stage 5 - announce */
4230 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4242 xfrm_states_put(x_cur, nx_cur);
4244 xfrm_states_delete(x_new, nx_new);
4248 EXPORT_SYMBOL(xfrm_migrate);