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/audit.h>
31 #ifdef CONFIG_XFRM_STATISTICS
35 #include "xfrm_hash.h"
37 int sysctl_xfrm_larval_drop __read_mostly = 1;
39 #ifdef CONFIG_XFRM_STATISTICS
40 DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
41 EXPORT_SYMBOL(xfrm_statistics);
44 DEFINE_MUTEX(xfrm_cfg_mutex);
45 EXPORT_SYMBOL(xfrm_cfg_mutex);
47 static DEFINE_RWLOCK(xfrm_policy_lock);
49 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
50 EXPORT_SYMBOL(xfrm_policy_count);
52 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
53 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
55 static struct kmem_cache *xfrm_dst_cache __read_mostly;
57 static HLIST_HEAD(xfrm_policy_gc_list);
58 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
60 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
61 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
62 static void xfrm_init_pmtu(struct dst_entry *dst);
65 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
67 return addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
68 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
69 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
70 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
71 (fl->proto == sel->proto || !sel->proto) &&
72 (fl->oif == sel->ifindex || !sel->ifindex);
76 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
78 return addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
79 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
80 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
81 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
82 (fl->proto == sel->proto || !sel->proto) &&
83 (fl->oif == sel->ifindex || !sel->ifindex);
86 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
87 unsigned short family)
91 return __xfrm4_selector_match(sel, fl);
93 return __xfrm6_selector_match(sel, fl);
98 static inline struct dst_entry *__xfrm_dst_lookup(int tos,
99 xfrm_address_t *saddr,
100 xfrm_address_t *daddr,
103 struct xfrm_policy_afinfo *afinfo;
104 struct dst_entry *dst;
106 afinfo = xfrm_policy_get_afinfo(family);
107 if (unlikely(afinfo == NULL))
108 return ERR_PTR(-EAFNOSUPPORT);
110 dst = afinfo->dst_lookup(tos, saddr, daddr);
112 xfrm_policy_put_afinfo(afinfo);
117 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
118 xfrm_address_t *prev_saddr,
119 xfrm_address_t *prev_daddr,
122 xfrm_address_t *saddr = &x->props.saddr;
123 xfrm_address_t *daddr = &x->id.daddr;
124 struct dst_entry *dst;
126 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
130 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
135 dst = __xfrm_dst_lookup(tos, saddr, daddr, family);
138 if (prev_saddr != saddr)
139 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
140 if (prev_daddr != daddr)
141 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
147 static inline unsigned long make_jiffies(long secs)
149 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
150 return MAX_SCHEDULE_TIMEOUT-1;
155 static void xfrm_policy_timer(unsigned long data)
157 struct xfrm_policy *xp = (struct xfrm_policy*)data;
158 unsigned long now = get_seconds();
159 long next = LONG_MAX;
163 read_lock(&xp->lock);
168 dir = xfrm_policy_id2dir(xp->index);
170 if (xp->lft.hard_add_expires_seconds) {
171 long tmo = xp->lft.hard_add_expires_seconds +
172 xp->curlft.add_time - now;
178 if (xp->lft.hard_use_expires_seconds) {
179 long tmo = xp->lft.hard_use_expires_seconds +
180 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
186 if (xp->lft.soft_add_expires_seconds) {
187 long tmo = xp->lft.soft_add_expires_seconds +
188 xp->curlft.add_time - now;
191 tmo = XFRM_KM_TIMEOUT;
196 if (xp->lft.soft_use_expires_seconds) {
197 long tmo = xp->lft.soft_use_expires_seconds +
198 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
201 tmo = XFRM_KM_TIMEOUT;
208 km_policy_expired(xp, dir, 0, 0);
209 if (next != LONG_MAX &&
210 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
214 read_unlock(&xp->lock);
219 read_unlock(&xp->lock);
220 if (!xfrm_policy_delete(xp, dir))
221 km_policy_expired(xp, dir, 1, 0);
226 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
230 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
232 struct xfrm_policy *policy;
234 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
237 write_pnet(&policy->xp_net, net);
238 INIT_LIST_HEAD(&policy->walk.all);
239 INIT_HLIST_NODE(&policy->bydst);
240 INIT_HLIST_NODE(&policy->byidx);
241 rwlock_init(&policy->lock);
242 atomic_set(&policy->refcnt, 1);
243 setup_timer(&policy->timer, xfrm_policy_timer,
244 (unsigned long)policy);
248 EXPORT_SYMBOL(xfrm_policy_alloc);
250 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
252 void xfrm_policy_destroy(struct xfrm_policy *policy)
254 BUG_ON(!policy->walk.dead);
256 BUG_ON(policy->bundles);
258 if (del_timer(&policy->timer))
261 security_xfrm_policy_free(policy->security);
264 EXPORT_SYMBOL(xfrm_policy_destroy);
266 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
268 struct dst_entry *dst;
270 while ((dst = policy->bundles) != NULL) {
271 policy->bundles = dst->next;
275 if (del_timer(&policy->timer))
276 atomic_dec(&policy->refcnt);
278 if (atomic_read(&policy->refcnt) > 1)
281 xfrm_pol_put(policy);
284 static void xfrm_policy_gc_task(struct work_struct *work)
286 struct xfrm_policy *policy;
287 struct hlist_node *entry, *tmp;
288 struct hlist_head gc_list;
290 spin_lock_bh(&xfrm_policy_gc_lock);
291 gc_list.first = xfrm_policy_gc_list.first;
292 INIT_HLIST_HEAD(&xfrm_policy_gc_list);
293 spin_unlock_bh(&xfrm_policy_gc_lock);
295 hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
296 xfrm_policy_gc_kill(policy);
298 static DECLARE_WORK(xfrm_policy_gc_work, xfrm_policy_gc_task);
300 /* Rule must be locked. Release descentant resources, announce
301 * entry dead. The rule must be unlinked from lists to the moment.
304 static void xfrm_policy_kill(struct xfrm_policy *policy)
308 write_lock_bh(&policy->lock);
309 dead = policy->walk.dead;
310 policy->walk.dead = 1;
311 write_unlock_bh(&policy->lock);
313 if (unlikely(dead)) {
318 spin_lock_bh(&xfrm_policy_gc_lock);
319 hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
320 spin_unlock_bh(&xfrm_policy_gc_lock);
322 schedule_work(&xfrm_policy_gc_work);
325 struct xfrm_policy_hash {
326 struct hlist_head *table;
330 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
331 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
332 static unsigned int xfrm_idx_hmask __read_mostly;
333 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
335 static inline unsigned int idx_hash(u32 index)
337 return __idx_hash(index, xfrm_idx_hmask);
340 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
342 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
343 unsigned int hash = __sel_hash(sel, family, hmask);
345 return (hash == hmask + 1 ?
346 &xfrm_policy_inexact[dir] :
347 xfrm_policy_bydst[dir].table + hash);
350 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
352 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
353 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
355 return xfrm_policy_bydst[dir].table + hash;
358 static void xfrm_dst_hash_transfer(struct hlist_head *list,
359 struct hlist_head *ndsttable,
360 unsigned int nhashmask)
362 struct hlist_node *entry, *tmp, *entry0 = NULL;
363 struct xfrm_policy *pol;
367 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
370 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
371 pol->family, nhashmask);
374 hlist_add_head(&pol->bydst, ndsttable+h);
380 hlist_add_after(entry0, &pol->bydst);
384 if (!hlist_empty(list)) {
390 static void xfrm_idx_hash_transfer(struct hlist_head *list,
391 struct hlist_head *nidxtable,
392 unsigned int nhashmask)
394 struct hlist_node *entry, *tmp;
395 struct xfrm_policy *pol;
397 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
400 h = __idx_hash(pol->index, nhashmask);
401 hlist_add_head(&pol->byidx, nidxtable+h);
405 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
407 return ((old_hmask + 1) << 1) - 1;
410 static void xfrm_bydst_resize(int dir)
412 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
413 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
414 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
415 struct hlist_head *odst = xfrm_policy_bydst[dir].table;
416 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
422 write_lock_bh(&xfrm_policy_lock);
424 for (i = hmask; i >= 0; i--)
425 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
427 xfrm_policy_bydst[dir].table = ndst;
428 xfrm_policy_bydst[dir].hmask = nhashmask;
430 write_unlock_bh(&xfrm_policy_lock);
432 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
435 static void xfrm_byidx_resize(int total)
437 unsigned int hmask = xfrm_idx_hmask;
438 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
439 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
440 struct hlist_head *oidx = init_net.xfrm.policy_byidx;
441 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
447 write_lock_bh(&xfrm_policy_lock);
449 for (i = hmask; i >= 0; i--)
450 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
452 init_net.xfrm.policy_byidx = nidx;
453 xfrm_idx_hmask = nhashmask;
455 write_unlock_bh(&xfrm_policy_lock);
457 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
460 static inline int xfrm_bydst_should_resize(int dir, int *total)
462 unsigned int cnt = xfrm_policy_count[dir];
463 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
468 if ((hmask + 1) < xfrm_policy_hashmax &&
475 static inline int xfrm_byidx_should_resize(int total)
477 unsigned int hmask = xfrm_idx_hmask;
479 if ((hmask + 1) < xfrm_policy_hashmax &&
486 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
488 read_lock_bh(&xfrm_policy_lock);
489 si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
490 si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
491 si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
492 si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
493 si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
494 si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
495 si->spdhcnt = xfrm_idx_hmask;
496 si->spdhmcnt = xfrm_policy_hashmax;
497 read_unlock_bh(&xfrm_policy_lock);
499 EXPORT_SYMBOL(xfrm_spd_getinfo);
501 static DEFINE_MUTEX(hash_resize_mutex);
502 static void xfrm_hash_resize(struct work_struct *__unused)
506 mutex_lock(&hash_resize_mutex);
509 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
510 if (xfrm_bydst_should_resize(dir, &total))
511 xfrm_bydst_resize(dir);
513 if (xfrm_byidx_should_resize(total))
514 xfrm_byidx_resize(total);
516 mutex_unlock(&hash_resize_mutex);
519 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
521 /* Generate new index... KAME seems to generate them ordered by cost
522 * of an absolute inpredictability of ordering of rules. This will not pass. */
523 static u32 xfrm_gen_index(int dir)
525 static u32 idx_generator;
528 struct hlist_node *entry;
529 struct hlist_head *list;
530 struct xfrm_policy *p;
534 idx = (idx_generator | dir);
538 list = init_net.xfrm.policy_byidx + idx_hash(idx);
540 hlist_for_each_entry(p, entry, list, byidx) {
541 if (p->index == idx) {
551 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
553 u32 *p1 = (u32 *) s1;
554 u32 *p2 = (u32 *) s2;
555 int len = sizeof(struct xfrm_selector) / sizeof(u32);
558 for (i = 0; i < len; i++) {
566 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
568 struct xfrm_policy *pol;
569 struct xfrm_policy *delpol;
570 struct hlist_head *chain;
571 struct hlist_node *entry, *newpos;
572 struct dst_entry *gc_list;
574 write_lock_bh(&xfrm_policy_lock);
575 chain = policy_hash_bysel(&policy->selector, policy->family, dir);
578 hlist_for_each_entry(pol, entry, chain, bydst) {
579 if (pol->type == policy->type &&
580 !selector_cmp(&pol->selector, &policy->selector) &&
581 xfrm_sec_ctx_match(pol->security, policy->security) &&
584 write_unlock_bh(&xfrm_policy_lock);
588 if (policy->priority > pol->priority)
590 } else if (policy->priority >= pol->priority) {
591 newpos = &pol->bydst;
598 hlist_add_after(newpos, &policy->bydst);
600 hlist_add_head(&policy->bydst, chain);
601 xfrm_pol_hold(policy);
602 xfrm_policy_count[dir]++;
603 atomic_inc(&flow_cache_genid);
605 hlist_del(&delpol->bydst);
606 hlist_del(&delpol->byidx);
607 list_del(&delpol->walk.all);
608 xfrm_policy_count[dir]--;
610 policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
611 hlist_add_head(&policy->byidx, init_net.xfrm.policy_byidx+idx_hash(policy->index));
612 policy->curlft.add_time = get_seconds();
613 policy->curlft.use_time = 0;
614 if (!mod_timer(&policy->timer, jiffies + HZ))
615 xfrm_pol_hold(policy);
616 list_add(&policy->walk.all, &init_net.xfrm.policy_all);
617 write_unlock_bh(&xfrm_policy_lock);
620 xfrm_policy_kill(delpol);
621 else if (xfrm_bydst_should_resize(dir, NULL))
622 schedule_work(&xfrm_hash_work);
624 read_lock_bh(&xfrm_policy_lock);
626 entry = &policy->bydst;
627 hlist_for_each_entry_continue(policy, entry, bydst) {
628 struct dst_entry *dst;
630 write_lock(&policy->lock);
631 dst = policy->bundles;
633 struct dst_entry *tail = dst;
636 tail->next = gc_list;
639 policy->bundles = NULL;
641 write_unlock(&policy->lock);
643 read_unlock_bh(&xfrm_policy_lock);
646 struct dst_entry *dst = gc_list;
654 EXPORT_SYMBOL(xfrm_policy_insert);
656 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
657 struct xfrm_selector *sel,
658 struct xfrm_sec_ctx *ctx, int delete,
661 struct xfrm_policy *pol, *ret;
662 struct hlist_head *chain;
663 struct hlist_node *entry;
666 write_lock_bh(&xfrm_policy_lock);
667 chain = policy_hash_bysel(sel, sel->family, dir);
669 hlist_for_each_entry(pol, entry, chain, bydst) {
670 if (pol->type == type &&
671 !selector_cmp(sel, &pol->selector) &&
672 xfrm_sec_ctx_match(ctx, pol->security)) {
675 *err = security_xfrm_policy_delete(
678 write_unlock_bh(&xfrm_policy_lock);
681 hlist_del(&pol->bydst);
682 hlist_del(&pol->byidx);
683 list_del(&pol->walk.all);
684 xfrm_policy_count[dir]--;
690 write_unlock_bh(&xfrm_policy_lock);
693 atomic_inc(&flow_cache_genid);
694 xfrm_policy_kill(ret);
698 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
700 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
703 struct xfrm_policy *pol, *ret;
704 struct hlist_head *chain;
705 struct hlist_node *entry;
708 if (xfrm_policy_id2dir(id) != dir)
712 write_lock_bh(&xfrm_policy_lock);
713 chain = init_net.xfrm.policy_byidx + idx_hash(id);
715 hlist_for_each_entry(pol, entry, chain, byidx) {
716 if (pol->type == type && pol->index == id) {
719 *err = security_xfrm_policy_delete(
722 write_unlock_bh(&xfrm_policy_lock);
725 hlist_del(&pol->bydst);
726 hlist_del(&pol->byidx);
727 list_del(&pol->walk.all);
728 xfrm_policy_count[dir]--;
734 write_unlock_bh(&xfrm_policy_lock);
737 atomic_inc(&flow_cache_genid);
738 xfrm_policy_kill(ret);
742 EXPORT_SYMBOL(xfrm_policy_byid);
744 #ifdef CONFIG_SECURITY_NETWORK_XFRM
746 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
750 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
751 struct xfrm_policy *pol;
752 struct hlist_node *entry;
755 hlist_for_each_entry(pol, entry,
756 &xfrm_policy_inexact[dir], bydst) {
757 if (pol->type != type)
759 err = security_xfrm_policy_delete(pol->security);
761 xfrm_audit_policy_delete(pol, 0,
762 audit_info->loginuid,
763 audit_info->sessionid,
768 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
769 hlist_for_each_entry(pol, entry,
770 xfrm_policy_bydst[dir].table + i,
772 if (pol->type != type)
774 err = security_xfrm_policy_delete(
777 xfrm_audit_policy_delete(pol, 0,
778 audit_info->loginuid,
779 audit_info->sessionid,
790 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
796 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
800 write_lock_bh(&xfrm_policy_lock);
802 err = xfrm_policy_flush_secctx_check(type, audit_info);
806 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
807 struct xfrm_policy *pol;
808 struct hlist_node *entry;
813 hlist_for_each_entry(pol, entry,
814 &xfrm_policy_inexact[dir], bydst) {
815 if (pol->type != type)
817 hlist_del(&pol->bydst);
818 hlist_del(&pol->byidx);
819 write_unlock_bh(&xfrm_policy_lock);
821 xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
822 audit_info->sessionid,
825 xfrm_policy_kill(pol);
828 write_lock_bh(&xfrm_policy_lock);
832 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
834 hlist_for_each_entry(pol, entry,
835 xfrm_policy_bydst[dir].table + i,
837 if (pol->type != type)
839 hlist_del(&pol->bydst);
840 hlist_del(&pol->byidx);
841 list_del(&pol->walk.all);
842 write_unlock_bh(&xfrm_policy_lock);
844 xfrm_audit_policy_delete(pol, 1,
845 audit_info->loginuid,
846 audit_info->sessionid,
848 xfrm_policy_kill(pol);
851 write_lock_bh(&xfrm_policy_lock);
856 xfrm_policy_count[dir] -= killed;
858 atomic_inc(&flow_cache_genid);
860 write_unlock_bh(&xfrm_policy_lock);
863 EXPORT_SYMBOL(xfrm_policy_flush);
865 int xfrm_policy_walk(struct xfrm_policy_walk *walk,
866 int (*func)(struct xfrm_policy *, int, int, void*),
869 struct xfrm_policy *pol;
870 struct xfrm_policy_walk_entry *x;
873 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
874 walk->type != XFRM_POLICY_TYPE_ANY)
877 if (list_empty(&walk->walk.all) && walk->seq != 0)
880 write_lock_bh(&xfrm_policy_lock);
881 if (list_empty(&walk->walk.all))
882 x = list_first_entry(&init_net.xfrm.policy_all, struct xfrm_policy_walk_entry, all);
884 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
885 list_for_each_entry_from(x, &init_net.xfrm.policy_all, all) {
888 pol = container_of(x, struct xfrm_policy, walk);
889 if (walk->type != XFRM_POLICY_TYPE_ANY &&
890 walk->type != pol->type)
892 error = func(pol, xfrm_policy_id2dir(pol->index),
895 list_move_tail(&walk->walk.all, &x->all);
900 if (walk->seq == 0) {
904 list_del_init(&walk->walk.all);
906 write_unlock_bh(&xfrm_policy_lock);
909 EXPORT_SYMBOL(xfrm_policy_walk);
911 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
913 INIT_LIST_HEAD(&walk->walk.all);
918 EXPORT_SYMBOL(xfrm_policy_walk_init);
920 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
922 if (list_empty(&walk->walk.all))
925 write_lock_bh(&xfrm_policy_lock);
926 list_del(&walk->walk.all);
927 write_unlock_bh(&xfrm_policy_lock);
929 EXPORT_SYMBOL(xfrm_policy_walk_done);
932 * Find policy to apply to this flow.
934 * Returns 0 if policy found, else an -errno.
936 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
937 u8 type, u16 family, int dir)
939 struct xfrm_selector *sel = &pol->selector;
940 int match, ret = -ESRCH;
942 if (pol->family != family ||
946 match = xfrm_selector_match(sel, fl, family);
948 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
954 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
958 struct xfrm_policy *pol, *ret;
959 xfrm_address_t *daddr, *saddr;
960 struct hlist_node *entry;
961 struct hlist_head *chain;
964 daddr = xfrm_flowi_daddr(fl, family);
965 saddr = xfrm_flowi_saddr(fl, family);
966 if (unlikely(!daddr || !saddr))
969 read_lock_bh(&xfrm_policy_lock);
970 chain = policy_hash_direct(daddr, saddr, family, dir);
972 hlist_for_each_entry(pol, entry, chain, bydst) {
973 err = xfrm_policy_match(pol, fl, type, family, dir);
983 priority = ret->priority;
987 chain = &xfrm_policy_inexact[dir];
988 hlist_for_each_entry(pol, entry, chain, bydst) {
989 err = xfrm_policy_match(pol, fl, type, family, dir);
997 } else if (pol->priority < priority) {
1005 read_unlock_bh(&xfrm_policy_lock);
1010 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1011 void **objp, atomic_t **obj_refp)
1013 struct xfrm_policy *pol;
1016 #ifdef CONFIG_XFRM_SUB_POLICY
1017 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1025 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1030 #ifdef CONFIG_XFRM_SUB_POLICY
1033 if ((*objp = (void *) pol) != NULL)
1034 *obj_refp = &pol->refcnt;
1038 static inline int policy_to_flow_dir(int dir)
1040 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1041 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1042 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1046 case XFRM_POLICY_IN:
1048 case XFRM_POLICY_OUT:
1049 return FLOW_DIR_OUT;
1050 case XFRM_POLICY_FWD:
1051 return FLOW_DIR_FWD;
1055 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1057 struct xfrm_policy *pol;
1059 read_lock_bh(&xfrm_policy_lock);
1060 if ((pol = sk->sk_policy[dir]) != NULL) {
1061 int match = xfrm_selector_match(&pol->selector, fl,
1066 err = security_xfrm_policy_lookup(pol->security,
1068 policy_to_flow_dir(dir));
1071 else if (err == -ESRCH)
1078 read_unlock_bh(&xfrm_policy_lock);
1082 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1084 struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1087 list_add(&pol->walk.all, &init_net.xfrm.policy_all);
1088 hlist_add_head(&pol->bydst, chain);
1089 hlist_add_head(&pol->byidx, init_net.xfrm.policy_byidx+idx_hash(pol->index));
1090 xfrm_policy_count[dir]++;
1093 if (xfrm_bydst_should_resize(dir, NULL))
1094 schedule_work(&xfrm_hash_work);
1097 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1100 if (hlist_unhashed(&pol->bydst))
1103 hlist_del(&pol->bydst);
1104 hlist_del(&pol->byidx);
1105 list_del(&pol->walk.all);
1106 xfrm_policy_count[dir]--;
1111 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1113 write_lock_bh(&xfrm_policy_lock);
1114 pol = __xfrm_policy_unlink(pol, dir);
1115 write_unlock_bh(&xfrm_policy_lock);
1117 if (dir < XFRM_POLICY_MAX)
1118 atomic_inc(&flow_cache_genid);
1119 xfrm_policy_kill(pol);
1124 EXPORT_SYMBOL(xfrm_policy_delete);
1126 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1128 struct xfrm_policy *old_pol;
1130 #ifdef CONFIG_XFRM_SUB_POLICY
1131 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1135 write_lock_bh(&xfrm_policy_lock);
1136 old_pol = sk->sk_policy[dir];
1137 sk->sk_policy[dir] = pol;
1139 pol->curlft.add_time = get_seconds();
1140 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
1141 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1144 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1145 write_unlock_bh(&xfrm_policy_lock);
1148 xfrm_policy_kill(old_pol);
1153 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1155 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1158 newp->selector = old->selector;
1159 if (security_xfrm_policy_clone(old->security,
1162 return NULL; /* ENOMEM */
1164 newp->lft = old->lft;
1165 newp->curlft = old->curlft;
1166 newp->action = old->action;
1167 newp->flags = old->flags;
1168 newp->xfrm_nr = old->xfrm_nr;
1169 newp->index = old->index;
1170 newp->type = old->type;
1171 memcpy(newp->xfrm_vec, old->xfrm_vec,
1172 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1173 write_lock_bh(&xfrm_policy_lock);
1174 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1175 write_unlock_bh(&xfrm_policy_lock);
1181 int __xfrm_sk_clone_policy(struct sock *sk)
1183 struct xfrm_policy *p0 = sk->sk_policy[0],
1184 *p1 = sk->sk_policy[1];
1186 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1187 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1189 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1195 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1196 unsigned short family)
1199 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1201 if (unlikely(afinfo == NULL))
1203 err = afinfo->get_saddr(local, remote);
1204 xfrm_policy_put_afinfo(afinfo);
1208 /* Resolve list of templates for the flow, given policy. */
1211 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1212 struct xfrm_state **xfrm,
1213 unsigned short family)
1217 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1218 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1221 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1222 struct xfrm_state *x;
1223 xfrm_address_t *remote = daddr;
1224 xfrm_address_t *local = saddr;
1225 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1227 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1228 tmpl->mode == XFRM_MODE_BEET) {
1229 remote = &tmpl->id.daddr;
1230 local = &tmpl->saddr;
1231 family = tmpl->encap_family;
1232 if (xfrm_addr_any(local, family)) {
1233 error = xfrm_get_saddr(&tmp, remote, family);
1240 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1242 if (x && x->km.state == XFRM_STATE_VALID) {
1249 error = (x->km.state == XFRM_STATE_ERROR ?
1253 else if (error == -ESRCH)
1256 if (!tmpl->optional)
1262 for (nx--; nx>=0; nx--)
1263 xfrm_state_put(xfrm[nx]);
1268 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1269 struct xfrm_state **xfrm,
1270 unsigned short family)
1272 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1273 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1279 for (i = 0; i < npols; i++) {
1280 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1285 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1293 /* found states are sorted for outbound processing */
1295 xfrm_state_sort(xfrm, tpp, cnx, family);
1300 for (cnx--; cnx>=0; cnx--)
1301 xfrm_state_put(tpp[cnx]);
1306 /* Check that the bundle accepts the flow and its components are
1310 static struct dst_entry *
1311 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1313 struct dst_entry *x;
1314 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1315 if (unlikely(afinfo == NULL))
1316 return ERR_PTR(-EINVAL);
1317 x = afinfo->find_bundle(fl, policy);
1318 xfrm_policy_put_afinfo(afinfo);
1322 static inline int xfrm_get_tos(struct flowi *fl, int family)
1324 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1330 tos = afinfo->get_tos(fl);
1332 xfrm_policy_put_afinfo(afinfo);
1337 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1339 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1340 struct xfrm_dst *xdst;
1343 return ERR_PTR(-EINVAL);
1345 xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1347 xfrm_policy_put_afinfo(afinfo);
1352 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1355 struct xfrm_policy_afinfo *afinfo =
1356 xfrm_policy_get_afinfo(dst->ops->family);
1362 err = afinfo->init_path(path, dst, nfheader_len);
1364 xfrm_policy_put_afinfo(afinfo);
1369 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1371 struct xfrm_policy_afinfo *afinfo =
1372 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1378 err = afinfo->fill_dst(xdst, dev);
1380 xfrm_policy_put_afinfo(afinfo);
1385 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1386 * all the metrics... Shortly, bundle a bundle.
1389 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1390 struct xfrm_state **xfrm, int nx,
1392 struct dst_entry *dst)
1394 unsigned long now = jiffies;
1395 struct net_device *dev;
1396 struct dst_entry *dst_prev = NULL;
1397 struct dst_entry *dst0 = NULL;
1401 int nfheader_len = 0;
1402 int trailer_len = 0;
1404 int family = policy->selector.family;
1405 xfrm_address_t saddr, daddr;
1407 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1409 tos = xfrm_get_tos(fl, family);
1416 for (; i < nx; i++) {
1417 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1418 struct dst_entry *dst1 = &xdst->u.dst;
1420 err = PTR_ERR(xdst);
1429 dst_prev->child = dst_clone(dst1);
1430 dst1->flags |= DST_NOHASH;
1434 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1436 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1437 family = xfrm[i]->props.family;
1438 dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1446 dst1->xfrm = xfrm[i];
1447 xdst->genid = xfrm[i]->genid;
1449 dst1->obsolete = -1;
1450 dst1->flags |= DST_HOST;
1451 dst1->lastuse = now;
1453 dst1->input = dst_discard;
1454 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1456 dst1->next = dst_prev;
1459 header_len += xfrm[i]->props.header_len;
1460 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1461 nfheader_len += xfrm[i]->props.header_len;
1462 trailer_len += xfrm[i]->props.trailer_len;
1465 dst_prev->child = dst;
1473 /* Copy neighbout for reachability confirmation */
1474 dst0->neighbour = neigh_clone(dst->neighbour);
1476 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1477 xfrm_init_pmtu(dst_prev);
1479 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1480 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1482 err = xfrm_fill_dst(xdst, dev);
1486 dst_prev->header_len = header_len;
1487 dst_prev->trailer_len = trailer_len;
1488 header_len -= xdst->u.dst.xfrm->props.header_len;
1489 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1497 xfrm_state_put(xfrm[i]);
1501 dst0 = ERR_PTR(err);
1506 xfrm_dst_alloc_copy(void **target, void *src, int size)
1509 *target = kmalloc(size, GFP_ATOMIC);
1513 memcpy(*target, src, size);
1518 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1520 #ifdef CONFIG_XFRM_SUB_POLICY
1521 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1522 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1530 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1532 #ifdef CONFIG_XFRM_SUB_POLICY
1533 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1534 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1540 static int stale_bundle(struct dst_entry *dst);
1542 /* Main function: finds/creates a bundle for given flow.
1544 * At the moment we eat a raw IP route. Mostly to speed up lookups
1545 * on interfaces with disabled IPsec.
1547 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1548 struct sock *sk, int flags)
1550 struct xfrm_policy *policy;
1551 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1556 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1557 struct dst_entry *dst, *dst_orig = *dst_p;
1562 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1565 genid = atomic_read(&flow_cache_genid);
1567 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1573 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1574 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1575 err = PTR_ERR(policy);
1576 if (IS_ERR(policy)) {
1577 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1583 /* To accelerate a bit... */
1584 if ((dst_orig->flags & DST_NOXFRM) ||
1585 !xfrm_policy_count[XFRM_POLICY_OUT])
1588 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1589 dir, xfrm_policy_lookup);
1590 err = PTR_ERR(policy);
1591 if (IS_ERR(policy)) {
1592 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1600 family = dst_orig->ops->family;
1603 xfrm_nr += pols[0]->xfrm_nr;
1606 if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1609 policy->curlft.use_time = get_seconds();
1611 switch (policy->action) {
1613 case XFRM_POLICY_BLOCK:
1614 /* Prohibit the flow */
1615 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1619 case XFRM_POLICY_ALLOW:
1620 #ifndef CONFIG_XFRM_SUB_POLICY
1621 if (policy->xfrm_nr == 0) {
1622 /* Flow passes not transformed. */
1623 xfrm_pol_put(policy);
1628 /* Try to find matching bundle.
1630 * LATER: help from flow cache. It is optional, this
1631 * is required only for output policy.
1633 dst = xfrm_find_bundle(fl, policy, family);
1635 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1643 #ifdef CONFIG_XFRM_SUB_POLICY
1644 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1645 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1649 if (IS_ERR(pols[1])) {
1650 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1651 err = PTR_ERR(pols[1]);
1654 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1655 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1660 xfrm_nr += pols[1]->xfrm_nr;
1665 * Because neither flowi nor bundle information knows about
1666 * transformation template size. On more than one policy usage
1667 * we can realize whether all of them is bypass or not after
1668 * they are searched. See above not-transformed bypass
1669 * is surrounded by non-sub policy configuration, too.
1672 /* Flow passes not transformed. */
1673 xfrm_pols_put(pols, npols);
1678 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1680 if (unlikely(nx<0)) {
1682 if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1683 /* EREMOTE tells the caller to generate
1684 * a one-shot blackhole route.
1686 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1687 xfrm_pol_put(policy);
1690 if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1691 DECLARE_WAITQUEUE(wait, current);
1693 add_wait_queue(&init_net.xfrm.km_waitq, &wait);
1694 set_current_state(TASK_INTERRUPTIBLE);
1696 set_current_state(TASK_RUNNING);
1697 remove_wait_queue(&init_net.xfrm.km_waitq, &wait);
1699 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1701 if (nx == -EAGAIN && signal_pending(current)) {
1702 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1706 if (nx == -EAGAIN ||
1707 genid != atomic_read(&flow_cache_genid)) {
1708 xfrm_pols_put(pols, npols);
1714 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1719 /* Flow passes not transformed. */
1720 xfrm_pols_put(pols, npols);
1724 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1727 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1731 for (pi = 0; pi < npols; pi++) {
1732 read_lock_bh(&pols[pi]->lock);
1733 pol_dead |= pols[pi]->walk.dead;
1734 read_unlock_bh(&pols[pi]->lock);
1737 write_lock_bh(&policy->lock);
1738 if (unlikely(pol_dead || stale_bundle(dst))) {
1739 /* Wow! While we worked on resolving, this
1740 * policy has gone. Retry. It is not paranoia,
1741 * we just cannot enlist new bundle to dead object.
1742 * We can't enlist stable bundles either.
1744 write_unlock_bh(&policy->lock);
1748 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1750 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1751 err = -EHOSTUNREACH;
1756 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1758 err = xfrm_dst_update_origin(dst, fl);
1759 if (unlikely(err)) {
1760 write_unlock_bh(&policy->lock);
1762 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1766 dst->next = policy->bundles;
1767 policy->bundles = dst;
1769 write_unlock_bh(&policy->lock);
1772 dst_release(dst_orig);
1773 xfrm_pols_put(pols, npols);
1777 xfrm_pols_put(pols, npols);
1779 dst_release(dst_orig);
1785 if (flags & XFRM_LOOKUP_ICMP)
1789 EXPORT_SYMBOL(__xfrm_lookup);
1791 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1792 struct sock *sk, int flags)
1794 int err = __xfrm_lookup(dst_p, fl, sk, flags);
1796 if (err == -EREMOTE) {
1797 dst_release(*dst_p);
1804 EXPORT_SYMBOL(xfrm_lookup);
1807 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1809 struct xfrm_state *x;
1811 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1813 x = skb->sp->xvec[idx];
1814 if (!x->type->reject)
1816 return x->type->reject(x, skb, fl);
1819 /* When skb is transformed back to its "native" form, we have to
1820 * check policy restrictions. At the moment we make this in maximally
1821 * stupid way. Shame on me. :-) Of course, connected sockets must
1822 * have policy cached at them.
1826 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1827 unsigned short family)
1829 if (xfrm_state_kern(x))
1830 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1831 return x->id.proto == tmpl->id.proto &&
1832 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1833 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1834 x->props.mode == tmpl->mode &&
1835 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1836 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1837 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1838 xfrm_state_addr_cmp(tmpl, x, family));
1842 * 0 or more than 0 is returned when validation is succeeded (either bypass
1843 * because of optional transport mode, or next index of the mathced secpath
1844 * state with the template.
1845 * -1 is returned when no matching template is found.
1846 * Otherwise "-2 - errored_index" is returned.
1849 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1850 unsigned short family)
1854 if (tmpl->optional) {
1855 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1859 for (; idx < sp->len; idx++) {
1860 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1862 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1871 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1872 unsigned int family, int reverse)
1874 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1877 if (unlikely(afinfo == NULL))
1878 return -EAFNOSUPPORT;
1880 afinfo->decode_session(skb, fl, reverse);
1881 err = security_xfrm_decode_session(skb, &fl->secid);
1882 xfrm_policy_put_afinfo(afinfo);
1885 EXPORT_SYMBOL(__xfrm_decode_session);
1887 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1889 for (; k < sp->len; k++) {
1890 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1899 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1900 unsigned short family)
1902 struct xfrm_policy *pol;
1903 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1912 reverse = dir & ~XFRM_POLICY_MASK;
1913 dir &= XFRM_POLICY_MASK;
1914 fl_dir = policy_to_flow_dir(dir);
1916 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1917 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1921 nf_nat_decode_session(skb, &fl, family);
1923 /* First, check used SA against their selectors. */
1927 for (i=skb->sp->len-1; i>=0; i--) {
1928 struct xfrm_state *x = skb->sp->xvec[i];
1929 if (!xfrm_selector_match(&x->sel, &fl, family)) {
1930 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1937 if (sk && sk->sk_policy[dir]) {
1938 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1940 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1946 pol = flow_cache_lookup(&fl, family, fl_dir,
1947 xfrm_policy_lookup);
1950 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1955 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1956 xfrm_secpath_reject(xerr_idx, skb, &fl);
1957 XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1963 pol->curlft.use_time = get_seconds();
1967 #ifdef CONFIG_XFRM_SUB_POLICY
1968 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1969 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1973 if (IS_ERR(pols[1])) {
1974 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1977 pols[1]->curlft.use_time = get_seconds();
1983 if (pol->action == XFRM_POLICY_ALLOW) {
1984 struct sec_path *sp;
1985 static struct sec_path dummy;
1986 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1987 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1988 struct xfrm_tmpl **tpp = tp;
1992 if ((sp = skb->sp) == NULL)
1995 for (pi = 0; pi < npols; pi++) {
1996 if (pols[pi] != pol &&
1997 pols[pi]->action != XFRM_POLICY_ALLOW) {
1998 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2001 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2002 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
2005 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2006 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2010 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2014 /* For each tunnel xfrm, find the first matching tmpl.
2015 * For each tmpl before that, find corresponding xfrm.
2016 * Order is _important_. Later we will implement
2017 * some barriers, but at the moment barriers
2018 * are implied between each two transformations.
2020 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2021 k = xfrm_policy_ok(tpp[i], sp, k, family);
2024 /* "-2 - errored_index" returned */
2026 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2031 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2032 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2036 xfrm_pols_put(pols, npols);
2039 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2042 xfrm_secpath_reject(xerr_idx, skb, &fl);
2044 xfrm_pols_put(pols, npols);
2047 EXPORT_SYMBOL(__xfrm_policy_check);
2049 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2053 if (xfrm_decode_session(skb, &fl, family) < 0) {
2054 /* XXX: we should have something like FWDHDRERROR here. */
2055 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2059 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2061 EXPORT_SYMBOL(__xfrm_route_forward);
2063 /* Optimize later using cookies and generation ids. */
2065 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2067 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2068 * to "-1" to force all XFRM destinations to get validated by
2069 * dst_ops->check on every use. We do this because when a
2070 * normal route referenced by an XFRM dst is obsoleted we do
2071 * not go looking around for all parent referencing XFRM dsts
2072 * so that we can invalidate them. It is just too much work.
2073 * Instead we make the checks here on every use. For example:
2075 * XFRM dst A --> IPv4 dst X
2077 * X is the "xdst->route" of A (X is also the "dst->path" of A
2078 * in this example). If X is marked obsolete, "A" will not
2079 * notice. That's what we are validating here via the
2080 * stale_bundle() check.
2082 * When a policy's bundle is pruned, we dst_free() the XFRM
2083 * dst which causes it's ->obsolete field to be set to a
2084 * positive non-zero integer. If an XFRM dst has been pruned
2085 * like this, we want to force a new route lookup.
2087 if (dst->obsolete < 0 && !stale_bundle(dst))
2093 static int stale_bundle(struct dst_entry *dst)
2095 return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2098 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2100 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2101 dst->dev = dev_net(dev)->loopback_dev;
2106 EXPORT_SYMBOL(xfrm_dst_ifdown);
2108 static void xfrm_link_failure(struct sk_buff *skb)
2110 /* Impossible. Such dst must be popped before reaches point of failure. */
2114 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2117 if (dst->obsolete) {
2125 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2127 struct dst_entry *dst, **dstp;
2129 write_lock(&pol->lock);
2130 dstp = &pol->bundles;
2131 while ((dst=*dstp) != NULL) {
2134 dst->next = *gc_list_p;
2140 write_unlock(&pol->lock);
2143 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2145 struct dst_entry *gc_list = NULL;
2148 read_lock_bh(&xfrm_policy_lock);
2149 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2150 struct xfrm_policy *pol;
2151 struct hlist_node *entry;
2152 struct hlist_head *table;
2155 hlist_for_each_entry(pol, entry,
2156 &xfrm_policy_inexact[dir], bydst)
2157 prune_one_bundle(pol, func, &gc_list);
2159 table = xfrm_policy_bydst[dir].table;
2160 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2161 hlist_for_each_entry(pol, entry, table + i, bydst)
2162 prune_one_bundle(pol, func, &gc_list);
2165 read_unlock_bh(&xfrm_policy_lock);
2168 struct dst_entry *dst = gc_list;
2169 gc_list = dst->next;
2174 static int unused_bundle(struct dst_entry *dst)
2176 return !atomic_read(&dst->__refcnt);
2179 static void __xfrm_garbage_collect(void)
2181 xfrm_prune_bundles(unused_bundle);
2184 static int xfrm_flush_bundles(void)
2186 xfrm_prune_bundles(stale_bundle);
2190 static void xfrm_init_pmtu(struct dst_entry *dst)
2193 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2194 u32 pmtu, route_mtu_cached;
2196 pmtu = dst_mtu(dst->child);
2197 xdst->child_mtu_cached = pmtu;
2199 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2201 route_mtu_cached = dst_mtu(xdst->route);
2202 xdst->route_mtu_cached = route_mtu_cached;
2204 if (pmtu > route_mtu_cached)
2205 pmtu = route_mtu_cached;
2207 dst->metrics[RTAX_MTU-1] = pmtu;
2208 } while ((dst = dst->next));
2211 /* Check that the bundle accepts the flow and its components are
2215 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2216 struct flowi *fl, int family, int strict)
2218 struct dst_entry *dst = &first->u.dst;
2219 struct xfrm_dst *last;
2222 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2223 (dst->dev && !netif_running(dst->dev)))
2225 #ifdef CONFIG_XFRM_SUB_POLICY
2227 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2229 if (first->partner &&
2230 !xfrm_selector_match(first->partner, fl, family))
2238 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2240 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2243 !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2245 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2247 if (xdst->genid != dst->xfrm->genid)
2251 !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2252 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2255 mtu = dst_mtu(dst->child);
2256 if (xdst->child_mtu_cached != mtu) {
2258 xdst->child_mtu_cached = mtu;
2261 if (!dst_check(xdst->route, xdst->route_cookie))
2263 mtu = dst_mtu(xdst->route);
2264 if (xdst->route_mtu_cached != mtu) {
2266 xdst->route_mtu_cached = mtu;
2270 } while (dst->xfrm);
2275 mtu = last->child_mtu_cached;
2279 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2280 if (mtu > last->route_mtu_cached)
2281 mtu = last->route_mtu_cached;
2282 dst->metrics[RTAX_MTU-1] = mtu;
2287 last = (struct xfrm_dst *)last->u.dst.next;
2288 last->child_mtu_cached = mtu;
2294 EXPORT_SYMBOL(xfrm_bundle_ok);
2296 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2299 if (unlikely(afinfo == NULL))
2301 if (unlikely(afinfo->family >= NPROTO))
2302 return -EAFNOSUPPORT;
2303 write_lock_bh(&xfrm_policy_afinfo_lock);
2304 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2307 struct dst_ops *dst_ops = afinfo->dst_ops;
2308 if (likely(dst_ops->kmem_cachep == NULL))
2309 dst_ops->kmem_cachep = xfrm_dst_cache;
2310 if (likely(dst_ops->check == NULL))
2311 dst_ops->check = xfrm_dst_check;
2312 if (likely(dst_ops->negative_advice == NULL))
2313 dst_ops->negative_advice = xfrm_negative_advice;
2314 if (likely(dst_ops->link_failure == NULL))
2315 dst_ops->link_failure = xfrm_link_failure;
2316 if (likely(afinfo->garbage_collect == NULL))
2317 afinfo->garbage_collect = __xfrm_garbage_collect;
2318 xfrm_policy_afinfo[afinfo->family] = afinfo;
2320 write_unlock_bh(&xfrm_policy_afinfo_lock);
2323 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2325 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2328 if (unlikely(afinfo == NULL))
2330 if (unlikely(afinfo->family >= NPROTO))
2331 return -EAFNOSUPPORT;
2332 write_lock_bh(&xfrm_policy_afinfo_lock);
2333 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2334 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2337 struct dst_ops *dst_ops = afinfo->dst_ops;
2338 xfrm_policy_afinfo[afinfo->family] = NULL;
2339 dst_ops->kmem_cachep = NULL;
2340 dst_ops->check = NULL;
2341 dst_ops->negative_advice = NULL;
2342 dst_ops->link_failure = NULL;
2343 afinfo->garbage_collect = NULL;
2346 write_unlock_bh(&xfrm_policy_afinfo_lock);
2349 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2351 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2353 struct xfrm_policy_afinfo *afinfo;
2354 if (unlikely(family >= NPROTO))
2356 read_lock(&xfrm_policy_afinfo_lock);
2357 afinfo = xfrm_policy_afinfo[family];
2358 if (unlikely(!afinfo))
2359 read_unlock(&xfrm_policy_afinfo_lock);
2363 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2365 read_unlock(&xfrm_policy_afinfo_lock);
2368 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2370 struct net_device *dev = ptr;
2372 if (!net_eq(dev_net(dev), &init_net))
2377 xfrm_flush_bundles();
2382 static struct notifier_block xfrm_dev_notifier = {
2383 .notifier_call = xfrm_dev_event,
2386 #ifdef CONFIG_XFRM_STATISTICS
2387 static int __init xfrm_statistics_init(void)
2389 if (snmp_mib_init((void **)xfrm_statistics,
2390 sizeof(struct linux_xfrm_mib)) < 0)
2396 static int __net_init xfrm_policy_init(struct net *net)
2398 unsigned int hmask, sz;
2401 if (net_eq(net, &init_net))
2402 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2403 sizeof(struct xfrm_dst),
2404 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2408 sz = (hmask+1) * sizeof(struct hlist_head);
2410 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2411 if (!net->xfrm.policy_byidx)
2413 xfrm_idx_hmask = hmask;
2415 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2416 struct xfrm_policy_hash *htab;
2418 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2420 htab = &xfrm_policy_bydst[dir];
2421 htab->table = xfrm_hash_alloc(sz);
2422 htab->hmask = hmask;
2424 panic("XFRM: failed to allocate bydst hash\n");
2427 INIT_LIST_HEAD(&net->xfrm.policy_all);
2428 if (net_eq(net, &init_net))
2429 register_netdevice_notifier(&xfrm_dev_notifier);
2436 static void xfrm_policy_fini(struct net *net)
2440 WARN_ON(!list_empty(&net->xfrm.policy_all));
2442 sz = (xfrm_idx_hmask + 1) * sizeof(struct hlist_head);
2443 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2444 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2447 static int __net_init xfrm_net_init(struct net *net)
2451 rv = xfrm_state_init(net);
2454 rv = xfrm_policy_init(net);
2460 xfrm_state_fini(net);
2465 static void __net_exit xfrm_net_exit(struct net *net)
2467 xfrm_policy_fini(net);
2468 xfrm_state_fini(net);
2471 static struct pernet_operations __net_initdata xfrm_net_ops = {
2472 .init = xfrm_net_init,
2473 .exit = xfrm_net_exit,
2476 void __init xfrm_init(void)
2478 register_pernet_subsys(&xfrm_net_ops);
2479 #ifdef CONFIG_XFRM_STATISTICS
2480 xfrm_statistics_init();
2483 #ifdef CONFIG_XFRM_STATISTICS
2488 #ifdef CONFIG_AUDITSYSCALL
2489 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2490 struct audit_buffer *audit_buf)
2492 struct xfrm_sec_ctx *ctx = xp->security;
2493 struct xfrm_selector *sel = &xp->selector;
2496 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2497 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2499 switch(sel->family) {
2501 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2502 if (sel->prefixlen_s != 32)
2503 audit_log_format(audit_buf, " src_prefixlen=%d",
2505 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2506 if (sel->prefixlen_d != 32)
2507 audit_log_format(audit_buf, " dst_prefixlen=%d",
2511 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2512 if (sel->prefixlen_s != 128)
2513 audit_log_format(audit_buf, " src_prefixlen=%d",
2515 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2516 if (sel->prefixlen_d != 128)
2517 audit_log_format(audit_buf, " dst_prefixlen=%d",
2523 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2524 uid_t auid, u32 sessionid, u32 secid)
2526 struct audit_buffer *audit_buf;
2528 audit_buf = xfrm_audit_start("SPD-add");
2529 if (audit_buf == NULL)
2531 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2532 audit_log_format(audit_buf, " res=%u", result);
2533 xfrm_audit_common_policyinfo(xp, audit_buf);
2534 audit_log_end(audit_buf);
2536 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2538 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2539 uid_t auid, u32 sessionid, u32 secid)
2541 struct audit_buffer *audit_buf;
2543 audit_buf = xfrm_audit_start("SPD-delete");
2544 if (audit_buf == NULL)
2546 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2547 audit_log_format(audit_buf, " res=%u", result);
2548 xfrm_audit_common_policyinfo(xp, audit_buf);
2549 audit_log_end(audit_buf);
2551 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2554 #ifdef CONFIG_XFRM_MIGRATE
2555 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2556 struct xfrm_selector *sel_tgt)
2558 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2559 if (sel_tgt->family == sel_cmp->family &&
2560 xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2561 sel_cmp->family) == 0 &&
2562 xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2563 sel_cmp->family) == 0 &&
2564 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2565 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2569 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2576 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2579 struct xfrm_policy *pol, *ret = NULL;
2580 struct hlist_node *entry;
2581 struct hlist_head *chain;
2584 read_lock_bh(&xfrm_policy_lock);
2585 chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2586 hlist_for_each_entry(pol, entry, chain, bydst) {
2587 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2588 pol->type == type) {
2590 priority = ret->priority;
2594 chain = &xfrm_policy_inexact[dir];
2595 hlist_for_each_entry(pol, entry, chain, bydst) {
2596 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2597 pol->type == type &&
2598 pol->priority < priority) {
2607 read_unlock_bh(&xfrm_policy_lock);
2612 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2616 if (t->mode == m->mode && t->id.proto == m->proto &&
2617 (m->reqid == 0 || t->reqid == m->reqid)) {
2619 case XFRM_MODE_TUNNEL:
2620 case XFRM_MODE_BEET:
2621 if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2622 m->old_family) == 0 &&
2623 xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2624 m->old_family) == 0) {
2628 case XFRM_MODE_TRANSPORT:
2629 /* in case of transport mode, template does not store
2630 any IP addresses, hence we just compare mode and
2641 /* update endpoint address(es) of template(s) */
2642 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2643 struct xfrm_migrate *m, int num_migrate)
2645 struct xfrm_migrate *mp;
2646 struct dst_entry *dst;
2649 write_lock_bh(&pol->lock);
2650 if (unlikely(pol->walk.dead)) {
2651 /* target policy has been deleted */
2652 write_unlock_bh(&pol->lock);
2656 for (i = 0; i < pol->xfrm_nr; i++) {
2657 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2658 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2661 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2662 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2664 /* update endpoints */
2665 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2666 sizeof(pol->xfrm_vec[i].id.daddr));
2667 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2668 sizeof(pol->xfrm_vec[i].saddr));
2669 pol->xfrm_vec[i].encap_family = mp->new_family;
2671 while ((dst = pol->bundles) != NULL) {
2672 pol->bundles = dst->next;
2678 write_unlock_bh(&pol->lock);
2686 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2690 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2693 for (i = 0; i < num_migrate; i++) {
2694 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2695 m[i].old_family) == 0) &&
2696 (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2697 m[i].old_family) == 0))
2699 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2700 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2703 /* check if there is any duplicated entry */
2704 for (j = i + 1; j < num_migrate; j++) {
2705 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2706 sizeof(m[i].old_daddr)) &&
2707 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2708 sizeof(m[i].old_saddr)) &&
2709 m[i].proto == m[j].proto &&
2710 m[i].mode == m[j].mode &&
2711 m[i].reqid == m[j].reqid &&
2712 m[i].old_family == m[j].old_family)
2720 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2721 struct xfrm_migrate *m, int num_migrate,
2722 struct xfrm_kmaddress *k)
2724 int i, err, nx_cur = 0, nx_new = 0;
2725 struct xfrm_policy *pol = NULL;
2726 struct xfrm_state *x, *xc;
2727 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2728 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2729 struct xfrm_migrate *mp;
2731 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2734 /* Stage 1 - find policy */
2735 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2740 /* Stage 2 - find and update state(s) */
2741 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2742 if ((x = xfrm_migrate_state_find(mp))) {
2745 if ((xc = xfrm_state_migrate(x, mp))) {
2755 /* Stage 3 - update policy */
2756 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2759 /* Stage 4 - delete old state(s) */
2761 xfrm_states_put(x_cur, nx_cur);
2762 xfrm_states_delete(x_cur, nx_cur);
2765 /* Stage 5 - announce */
2766 km_migrate(sel, dir, type, m, num_migrate, k);
2778 xfrm_states_put(x_cur, nx_cur);
2780 xfrm_states_delete(x_new, nx_new);
2784 EXPORT_SYMBOL(xfrm_migrate);