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 xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
331 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
333 static inline unsigned int idx_hash(u32 index)
335 return __idx_hash(index, init_net.xfrm.policy_idx_hmask);
338 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
340 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
341 unsigned int hash = __sel_hash(sel, family, hmask);
343 return (hash == hmask + 1 ?
344 &init_net.xfrm.policy_inexact[dir] :
345 xfrm_policy_bydst[dir].table + hash);
348 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
350 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
351 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
353 return xfrm_policy_bydst[dir].table + hash;
356 static void xfrm_dst_hash_transfer(struct hlist_head *list,
357 struct hlist_head *ndsttable,
358 unsigned int nhashmask)
360 struct hlist_node *entry, *tmp, *entry0 = NULL;
361 struct xfrm_policy *pol;
365 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
368 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
369 pol->family, nhashmask);
372 hlist_add_head(&pol->bydst, ndsttable+h);
378 hlist_add_after(entry0, &pol->bydst);
382 if (!hlist_empty(list)) {
388 static void xfrm_idx_hash_transfer(struct hlist_head *list,
389 struct hlist_head *nidxtable,
390 unsigned int nhashmask)
392 struct hlist_node *entry, *tmp;
393 struct xfrm_policy *pol;
395 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
398 h = __idx_hash(pol->index, nhashmask);
399 hlist_add_head(&pol->byidx, nidxtable+h);
403 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
405 return ((old_hmask + 1) << 1) - 1;
408 static void xfrm_bydst_resize(int dir)
410 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
411 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
412 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
413 struct hlist_head *odst = xfrm_policy_bydst[dir].table;
414 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
420 write_lock_bh(&xfrm_policy_lock);
422 for (i = hmask; i >= 0; i--)
423 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
425 xfrm_policy_bydst[dir].table = ndst;
426 xfrm_policy_bydst[dir].hmask = nhashmask;
428 write_unlock_bh(&xfrm_policy_lock);
430 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
433 static void xfrm_byidx_resize(int total)
435 unsigned int hmask = init_net.xfrm.policy_idx_hmask;
436 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
437 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
438 struct hlist_head *oidx = init_net.xfrm.policy_byidx;
439 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
445 write_lock_bh(&xfrm_policy_lock);
447 for (i = hmask; i >= 0; i--)
448 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
450 init_net.xfrm.policy_byidx = nidx;
451 init_net.xfrm.policy_idx_hmask = nhashmask;
453 write_unlock_bh(&xfrm_policy_lock);
455 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
458 static inline int xfrm_bydst_should_resize(int dir, int *total)
460 unsigned int cnt = xfrm_policy_count[dir];
461 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
466 if ((hmask + 1) < xfrm_policy_hashmax &&
473 static inline int xfrm_byidx_should_resize(int total)
475 unsigned int hmask = init_net.xfrm.policy_idx_hmask;
477 if ((hmask + 1) < xfrm_policy_hashmax &&
484 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
486 read_lock_bh(&xfrm_policy_lock);
487 si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
488 si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
489 si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
490 si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
491 si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
492 si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
493 si->spdhcnt = init_net.xfrm.policy_idx_hmask;
494 si->spdhmcnt = xfrm_policy_hashmax;
495 read_unlock_bh(&xfrm_policy_lock);
497 EXPORT_SYMBOL(xfrm_spd_getinfo);
499 static DEFINE_MUTEX(hash_resize_mutex);
500 static void xfrm_hash_resize(struct work_struct *__unused)
504 mutex_lock(&hash_resize_mutex);
507 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
508 if (xfrm_bydst_should_resize(dir, &total))
509 xfrm_bydst_resize(dir);
511 if (xfrm_byidx_should_resize(total))
512 xfrm_byidx_resize(total);
514 mutex_unlock(&hash_resize_mutex);
517 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
519 /* Generate new index... KAME seems to generate them ordered by cost
520 * of an absolute inpredictability of ordering of rules. This will not pass. */
521 static u32 xfrm_gen_index(int dir)
523 static u32 idx_generator;
526 struct hlist_node *entry;
527 struct hlist_head *list;
528 struct xfrm_policy *p;
532 idx = (idx_generator | dir);
536 list = init_net.xfrm.policy_byidx + idx_hash(idx);
538 hlist_for_each_entry(p, entry, list, byidx) {
539 if (p->index == idx) {
549 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
551 u32 *p1 = (u32 *) s1;
552 u32 *p2 = (u32 *) s2;
553 int len = sizeof(struct xfrm_selector) / sizeof(u32);
556 for (i = 0; i < len; i++) {
564 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
566 struct xfrm_policy *pol;
567 struct xfrm_policy *delpol;
568 struct hlist_head *chain;
569 struct hlist_node *entry, *newpos;
570 struct dst_entry *gc_list;
572 write_lock_bh(&xfrm_policy_lock);
573 chain = policy_hash_bysel(&policy->selector, policy->family, dir);
576 hlist_for_each_entry(pol, entry, chain, bydst) {
577 if (pol->type == policy->type &&
578 !selector_cmp(&pol->selector, &policy->selector) &&
579 xfrm_sec_ctx_match(pol->security, policy->security) &&
582 write_unlock_bh(&xfrm_policy_lock);
586 if (policy->priority > pol->priority)
588 } else if (policy->priority >= pol->priority) {
589 newpos = &pol->bydst;
596 hlist_add_after(newpos, &policy->bydst);
598 hlist_add_head(&policy->bydst, chain);
599 xfrm_pol_hold(policy);
600 xfrm_policy_count[dir]++;
601 atomic_inc(&flow_cache_genid);
603 hlist_del(&delpol->bydst);
604 hlist_del(&delpol->byidx);
605 list_del(&delpol->walk.all);
606 xfrm_policy_count[dir]--;
608 policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
609 hlist_add_head(&policy->byidx, init_net.xfrm.policy_byidx+idx_hash(policy->index));
610 policy->curlft.add_time = get_seconds();
611 policy->curlft.use_time = 0;
612 if (!mod_timer(&policy->timer, jiffies + HZ))
613 xfrm_pol_hold(policy);
614 list_add(&policy->walk.all, &init_net.xfrm.policy_all);
615 write_unlock_bh(&xfrm_policy_lock);
618 xfrm_policy_kill(delpol);
619 else if (xfrm_bydst_should_resize(dir, NULL))
620 schedule_work(&xfrm_hash_work);
622 read_lock_bh(&xfrm_policy_lock);
624 entry = &policy->bydst;
625 hlist_for_each_entry_continue(policy, entry, bydst) {
626 struct dst_entry *dst;
628 write_lock(&policy->lock);
629 dst = policy->bundles;
631 struct dst_entry *tail = dst;
634 tail->next = gc_list;
637 policy->bundles = NULL;
639 write_unlock(&policy->lock);
641 read_unlock_bh(&xfrm_policy_lock);
644 struct dst_entry *dst = gc_list;
652 EXPORT_SYMBOL(xfrm_policy_insert);
654 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
655 struct xfrm_selector *sel,
656 struct xfrm_sec_ctx *ctx, int delete,
659 struct xfrm_policy *pol, *ret;
660 struct hlist_head *chain;
661 struct hlist_node *entry;
664 write_lock_bh(&xfrm_policy_lock);
665 chain = policy_hash_bysel(sel, sel->family, dir);
667 hlist_for_each_entry(pol, entry, chain, bydst) {
668 if (pol->type == type &&
669 !selector_cmp(sel, &pol->selector) &&
670 xfrm_sec_ctx_match(ctx, pol->security)) {
673 *err = security_xfrm_policy_delete(
676 write_unlock_bh(&xfrm_policy_lock);
679 hlist_del(&pol->bydst);
680 hlist_del(&pol->byidx);
681 list_del(&pol->walk.all);
682 xfrm_policy_count[dir]--;
688 write_unlock_bh(&xfrm_policy_lock);
691 atomic_inc(&flow_cache_genid);
692 xfrm_policy_kill(ret);
696 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
698 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
701 struct xfrm_policy *pol, *ret;
702 struct hlist_head *chain;
703 struct hlist_node *entry;
706 if (xfrm_policy_id2dir(id) != dir)
710 write_lock_bh(&xfrm_policy_lock);
711 chain = init_net.xfrm.policy_byidx + idx_hash(id);
713 hlist_for_each_entry(pol, entry, chain, byidx) {
714 if (pol->type == type && pol->index == id) {
717 *err = security_xfrm_policy_delete(
720 write_unlock_bh(&xfrm_policy_lock);
723 hlist_del(&pol->bydst);
724 hlist_del(&pol->byidx);
725 list_del(&pol->walk.all);
726 xfrm_policy_count[dir]--;
732 write_unlock_bh(&xfrm_policy_lock);
735 atomic_inc(&flow_cache_genid);
736 xfrm_policy_kill(ret);
740 EXPORT_SYMBOL(xfrm_policy_byid);
742 #ifdef CONFIG_SECURITY_NETWORK_XFRM
744 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
748 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
749 struct xfrm_policy *pol;
750 struct hlist_node *entry;
753 hlist_for_each_entry(pol, entry,
754 &init_net.xfrm.policy_inexact[dir], bydst) {
755 if (pol->type != type)
757 err = security_xfrm_policy_delete(pol->security);
759 xfrm_audit_policy_delete(pol, 0,
760 audit_info->loginuid,
761 audit_info->sessionid,
766 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
767 hlist_for_each_entry(pol, entry,
768 xfrm_policy_bydst[dir].table + i,
770 if (pol->type != type)
772 err = security_xfrm_policy_delete(
775 xfrm_audit_policy_delete(pol, 0,
776 audit_info->loginuid,
777 audit_info->sessionid,
788 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
794 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
798 write_lock_bh(&xfrm_policy_lock);
800 err = xfrm_policy_flush_secctx_check(type, audit_info);
804 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
805 struct xfrm_policy *pol;
806 struct hlist_node *entry;
811 hlist_for_each_entry(pol, entry,
812 &init_net.xfrm.policy_inexact[dir], bydst) {
813 if (pol->type != type)
815 hlist_del(&pol->bydst);
816 hlist_del(&pol->byidx);
817 write_unlock_bh(&xfrm_policy_lock);
819 xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
820 audit_info->sessionid,
823 xfrm_policy_kill(pol);
826 write_lock_bh(&xfrm_policy_lock);
830 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
832 hlist_for_each_entry(pol, entry,
833 xfrm_policy_bydst[dir].table + i,
835 if (pol->type != type)
837 hlist_del(&pol->bydst);
838 hlist_del(&pol->byidx);
839 list_del(&pol->walk.all);
840 write_unlock_bh(&xfrm_policy_lock);
842 xfrm_audit_policy_delete(pol, 1,
843 audit_info->loginuid,
844 audit_info->sessionid,
846 xfrm_policy_kill(pol);
849 write_lock_bh(&xfrm_policy_lock);
854 xfrm_policy_count[dir] -= killed;
856 atomic_inc(&flow_cache_genid);
858 write_unlock_bh(&xfrm_policy_lock);
861 EXPORT_SYMBOL(xfrm_policy_flush);
863 int xfrm_policy_walk(struct xfrm_policy_walk *walk,
864 int (*func)(struct xfrm_policy *, int, int, void*),
867 struct xfrm_policy *pol;
868 struct xfrm_policy_walk_entry *x;
871 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
872 walk->type != XFRM_POLICY_TYPE_ANY)
875 if (list_empty(&walk->walk.all) && walk->seq != 0)
878 write_lock_bh(&xfrm_policy_lock);
879 if (list_empty(&walk->walk.all))
880 x = list_first_entry(&init_net.xfrm.policy_all, struct xfrm_policy_walk_entry, all);
882 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
883 list_for_each_entry_from(x, &init_net.xfrm.policy_all, all) {
886 pol = container_of(x, struct xfrm_policy, walk);
887 if (walk->type != XFRM_POLICY_TYPE_ANY &&
888 walk->type != pol->type)
890 error = func(pol, xfrm_policy_id2dir(pol->index),
893 list_move_tail(&walk->walk.all, &x->all);
898 if (walk->seq == 0) {
902 list_del_init(&walk->walk.all);
904 write_unlock_bh(&xfrm_policy_lock);
907 EXPORT_SYMBOL(xfrm_policy_walk);
909 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
911 INIT_LIST_HEAD(&walk->walk.all);
916 EXPORT_SYMBOL(xfrm_policy_walk_init);
918 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
920 if (list_empty(&walk->walk.all))
923 write_lock_bh(&xfrm_policy_lock);
924 list_del(&walk->walk.all);
925 write_unlock_bh(&xfrm_policy_lock);
927 EXPORT_SYMBOL(xfrm_policy_walk_done);
930 * Find policy to apply to this flow.
932 * Returns 0 if policy found, else an -errno.
934 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
935 u8 type, u16 family, int dir)
937 struct xfrm_selector *sel = &pol->selector;
938 int match, ret = -ESRCH;
940 if (pol->family != family ||
944 match = xfrm_selector_match(sel, fl, family);
946 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
952 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
956 struct xfrm_policy *pol, *ret;
957 xfrm_address_t *daddr, *saddr;
958 struct hlist_node *entry;
959 struct hlist_head *chain;
962 daddr = xfrm_flowi_daddr(fl, family);
963 saddr = xfrm_flowi_saddr(fl, family);
964 if (unlikely(!daddr || !saddr))
967 read_lock_bh(&xfrm_policy_lock);
968 chain = policy_hash_direct(daddr, saddr, family, dir);
970 hlist_for_each_entry(pol, entry, chain, bydst) {
971 err = xfrm_policy_match(pol, fl, type, family, dir);
981 priority = ret->priority;
985 chain = &init_net.xfrm.policy_inexact[dir];
986 hlist_for_each_entry(pol, entry, chain, bydst) {
987 err = xfrm_policy_match(pol, fl, type, family, dir);
995 } else if (pol->priority < priority) {
1003 read_unlock_bh(&xfrm_policy_lock);
1008 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1009 void **objp, atomic_t **obj_refp)
1011 struct xfrm_policy *pol;
1014 #ifdef CONFIG_XFRM_SUB_POLICY
1015 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1023 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1028 #ifdef CONFIG_XFRM_SUB_POLICY
1031 if ((*objp = (void *) pol) != NULL)
1032 *obj_refp = &pol->refcnt;
1036 static inline int policy_to_flow_dir(int dir)
1038 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1039 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1040 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1044 case XFRM_POLICY_IN:
1046 case XFRM_POLICY_OUT:
1047 return FLOW_DIR_OUT;
1048 case XFRM_POLICY_FWD:
1049 return FLOW_DIR_FWD;
1053 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1055 struct xfrm_policy *pol;
1057 read_lock_bh(&xfrm_policy_lock);
1058 if ((pol = sk->sk_policy[dir]) != NULL) {
1059 int match = xfrm_selector_match(&pol->selector, fl,
1064 err = security_xfrm_policy_lookup(pol->security,
1066 policy_to_flow_dir(dir));
1069 else if (err == -ESRCH)
1076 read_unlock_bh(&xfrm_policy_lock);
1080 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1082 struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1085 list_add(&pol->walk.all, &init_net.xfrm.policy_all);
1086 hlist_add_head(&pol->bydst, chain);
1087 hlist_add_head(&pol->byidx, init_net.xfrm.policy_byidx+idx_hash(pol->index));
1088 xfrm_policy_count[dir]++;
1091 if (xfrm_bydst_should_resize(dir, NULL))
1092 schedule_work(&xfrm_hash_work);
1095 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1098 if (hlist_unhashed(&pol->bydst))
1101 hlist_del(&pol->bydst);
1102 hlist_del(&pol->byidx);
1103 list_del(&pol->walk.all);
1104 xfrm_policy_count[dir]--;
1109 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1111 write_lock_bh(&xfrm_policy_lock);
1112 pol = __xfrm_policy_unlink(pol, dir);
1113 write_unlock_bh(&xfrm_policy_lock);
1115 if (dir < XFRM_POLICY_MAX)
1116 atomic_inc(&flow_cache_genid);
1117 xfrm_policy_kill(pol);
1122 EXPORT_SYMBOL(xfrm_policy_delete);
1124 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1126 struct xfrm_policy *old_pol;
1128 #ifdef CONFIG_XFRM_SUB_POLICY
1129 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1133 write_lock_bh(&xfrm_policy_lock);
1134 old_pol = sk->sk_policy[dir];
1135 sk->sk_policy[dir] = pol;
1137 pol->curlft.add_time = get_seconds();
1138 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
1139 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1142 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1143 write_unlock_bh(&xfrm_policy_lock);
1146 xfrm_policy_kill(old_pol);
1151 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1153 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1156 newp->selector = old->selector;
1157 if (security_xfrm_policy_clone(old->security,
1160 return NULL; /* ENOMEM */
1162 newp->lft = old->lft;
1163 newp->curlft = old->curlft;
1164 newp->action = old->action;
1165 newp->flags = old->flags;
1166 newp->xfrm_nr = old->xfrm_nr;
1167 newp->index = old->index;
1168 newp->type = old->type;
1169 memcpy(newp->xfrm_vec, old->xfrm_vec,
1170 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1171 write_lock_bh(&xfrm_policy_lock);
1172 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1173 write_unlock_bh(&xfrm_policy_lock);
1179 int __xfrm_sk_clone_policy(struct sock *sk)
1181 struct xfrm_policy *p0 = sk->sk_policy[0],
1182 *p1 = sk->sk_policy[1];
1184 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1185 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1187 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1193 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1194 unsigned short family)
1197 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1199 if (unlikely(afinfo == NULL))
1201 err = afinfo->get_saddr(local, remote);
1202 xfrm_policy_put_afinfo(afinfo);
1206 /* Resolve list of templates for the flow, given policy. */
1209 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1210 struct xfrm_state **xfrm,
1211 unsigned short family)
1215 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1216 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1219 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1220 struct xfrm_state *x;
1221 xfrm_address_t *remote = daddr;
1222 xfrm_address_t *local = saddr;
1223 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1225 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1226 tmpl->mode == XFRM_MODE_BEET) {
1227 remote = &tmpl->id.daddr;
1228 local = &tmpl->saddr;
1229 family = tmpl->encap_family;
1230 if (xfrm_addr_any(local, family)) {
1231 error = xfrm_get_saddr(&tmp, remote, family);
1238 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1240 if (x && x->km.state == XFRM_STATE_VALID) {
1247 error = (x->km.state == XFRM_STATE_ERROR ?
1251 else if (error == -ESRCH)
1254 if (!tmpl->optional)
1260 for (nx--; nx>=0; nx--)
1261 xfrm_state_put(xfrm[nx]);
1266 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1267 struct xfrm_state **xfrm,
1268 unsigned short family)
1270 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1271 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1277 for (i = 0; i < npols; i++) {
1278 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1283 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1291 /* found states are sorted for outbound processing */
1293 xfrm_state_sort(xfrm, tpp, cnx, family);
1298 for (cnx--; cnx>=0; cnx--)
1299 xfrm_state_put(tpp[cnx]);
1304 /* Check that the bundle accepts the flow and its components are
1308 static struct dst_entry *
1309 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1311 struct dst_entry *x;
1312 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1313 if (unlikely(afinfo == NULL))
1314 return ERR_PTR(-EINVAL);
1315 x = afinfo->find_bundle(fl, policy);
1316 xfrm_policy_put_afinfo(afinfo);
1320 static inline int xfrm_get_tos(struct flowi *fl, int family)
1322 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1328 tos = afinfo->get_tos(fl);
1330 xfrm_policy_put_afinfo(afinfo);
1335 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1337 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1338 struct xfrm_dst *xdst;
1341 return ERR_PTR(-EINVAL);
1343 xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1345 xfrm_policy_put_afinfo(afinfo);
1350 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1353 struct xfrm_policy_afinfo *afinfo =
1354 xfrm_policy_get_afinfo(dst->ops->family);
1360 err = afinfo->init_path(path, dst, nfheader_len);
1362 xfrm_policy_put_afinfo(afinfo);
1367 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1369 struct xfrm_policy_afinfo *afinfo =
1370 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1376 err = afinfo->fill_dst(xdst, dev);
1378 xfrm_policy_put_afinfo(afinfo);
1383 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1384 * all the metrics... Shortly, bundle a bundle.
1387 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1388 struct xfrm_state **xfrm, int nx,
1390 struct dst_entry *dst)
1392 unsigned long now = jiffies;
1393 struct net_device *dev;
1394 struct dst_entry *dst_prev = NULL;
1395 struct dst_entry *dst0 = NULL;
1399 int nfheader_len = 0;
1400 int trailer_len = 0;
1402 int family = policy->selector.family;
1403 xfrm_address_t saddr, daddr;
1405 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1407 tos = xfrm_get_tos(fl, family);
1414 for (; i < nx; i++) {
1415 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1416 struct dst_entry *dst1 = &xdst->u.dst;
1418 err = PTR_ERR(xdst);
1427 dst_prev->child = dst_clone(dst1);
1428 dst1->flags |= DST_NOHASH;
1432 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1434 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1435 family = xfrm[i]->props.family;
1436 dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1444 dst1->xfrm = xfrm[i];
1445 xdst->genid = xfrm[i]->genid;
1447 dst1->obsolete = -1;
1448 dst1->flags |= DST_HOST;
1449 dst1->lastuse = now;
1451 dst1->input = dst_discard;
1452 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1454 dst1->next = dst_prev;
1457 header_len += xfrm[i]->props.header_len;
1458 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1459 nfheader_len += xfrm[i]->props.header_len;
1460 trailer_len += xfrm[i]->props.trailer_len;
1463 dst_prev->child = dst;
1471 /* Copy neighbout for reachability confirmation */
1472 dst0->neighbour = neigh_clone(dst->neighbour);
1474 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1475 xfrm_init_pmtu(dst_prev);
1477 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1478 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1480 err = xfrm_fill_dst(xdst, dev);
1484 dst_prev->header_len = header_len;
1485 dst_prev->trailer_len = trailer_len;
1486 header_len -= xdst->u.dst.xfrm->props.header_len;
1487 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1495 xfrm_state_put(xfrm[i]);
1499 dst0 = ERR_PTR(err);
1504 xfrm_dst_alloc_copy(void **target, void *src, int size)
1507 *target = kmalloc(size, GFP_ATOMIC);
1511 memcpy(*target, src, size);
1516 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1518 #ifdef CONFIG_XFRM_SUB_POLICY
1519 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1520 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1528 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1530 #ifdef CONFIG_XFRM_SUB_POLICY
1531 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1532 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1538 static int stale_bundle(struct dst_entry *dst);
1540 /* Main function: finds/creates a bundle for given flow.
1542 * At the moment we eat a raw IP route. Mostly to speed up lookups
1543 * on interfaces with disabled IPsec.
1545 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1546 struct sock *sk, int flags)
1548 struct xfrm_policy *policy;
1549 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1554 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1555 struct dst_entry *dst, *dst_orig = *dst_p;
1560 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1563 genid = atomic_read(&flow_cache_genid);
1565 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1571 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1572 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1573 err = PTR_ERR(policy);
1574 if (IS_ERR(policy)) {
1575 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1581 /* To accelerate a bit... */
1582 if ((dst_orig->flags & DST_NOXFRM) ||
1583 !xfrm_policy_count[XFRM_POLICY_OUT])
1586 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1587 dir, xfrm_policy_lookup);
1588 err = PTR_ERR(policy);
1589 if (IS_ERR(policy)) {
1590 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1598 family = dst_orig->ops->family;
1601 xfrm_nr += pols[0]->xfrm_nr;
1604 if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1607 policy->curlft.use_time = get_seconds();
1609 switch (policy->action) {
1611 case XFRM_POLICY_BLOCK:
1612 /* Prohibit the flow */
1613 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1617 case XFRM_POLICY_ALLOW:
1618 #ifndef CONFIG_XFRM_SUB_POLICY
1619 if (policy->xfrm_nr == 0) {
1620 /* Flow passes not transformed. */
1621 xfrm_pol_put(policy);
1626 /* Try to find matching bundle.
1628 * LATER: help from flow cache. It is optional, this
1629 * is required only for output policy.
1631 dst = xfrm_find_bundle(fl, policy, family);
1633 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1641 #ifdef CONFIG_XFRM_SUB_POLICY
1642 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1643 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1647 if (IS_ERR(pols[1])) {
1648 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1649 err = PTR_ERR(pols[1]);
1652 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1653 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1658 xfrm_nr += pols[1]->xfrm_nr;
1663 * Because neither flowi nor bundle information knows about
1664 * transformation template size. On more than one policy usage
1665 * we can realize whether all of them is bypass or not after
1666 * they are searched. See above not-transformed bypass
1667 * is surrounded by non-sub policy configuration, too.
1670 /* Flow passes not transformed. */
1671 xfrm_pols_put(pols, npols);
1676 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1678 if (unlikely(nx<0)) {
1680 if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1681 /* EREMOTE tells the caller to generate
1682 * a one-shot blackhole route.
1684 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1685 xfrm_pol_put(policy);
1688 if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1689 DECLARE_WAITQUEUE(wait, current);
1691 add_wait_queue(&init_net.xfrm.km_waitq, &wait);
1692 set_current_state(TASK_INTERRUPTIBLE);
1694 set_current_state(TASK_RUNNING);
1695 remove_wait_queue(&init_net.xfrm.km_waitq, &wait);
1697 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1699 if (nx == -EAGAIN && signal_pending(current)) {
1700 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1704 if (nx == -EAGAIN ||
1705 genid != atomic_read(&flow_cache_genid)) {
1706 xfrm_pols_put(pols, npols);
1712 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1717 /* Flow passes not transformed. */
1718 xfrm_pols_put(pols, npols);
1722 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1725 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1729 for (pi = 0; pi < npols; pi++) {
1730 read_lock_bh(&pols[pi]->lock);
1731 pol_dead |= pols[pi]->walk.dead;
1732 read_unlock_bh(&pols[pi]->lock);
1735 write_lock_bh(&policy->lock);
1736 if (unlikely(pol_dead || stale_bundle(dst))) {
1737 /* Wow! While we worked on resolving, this
1738 * policy has gone. Retry. It is not paranoia,
1739 * we just cannot enlist new bundle to dead object.
1740 * We can't enlist stable bundles either.
1742 write_unlock_bh(&policy->lock);
1746 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1748 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1749 err = -EHOSTUNREACH;
1754 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1756 err = xfrm_dst_update_origin(dst, fl);
1757 if (unlikely(err)) {
1758 write_unlock_bh(&policy->lock);
1760 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1764 dst->next = policy->bundles;
1765 policy->bundles = dst;
1767 write_unlock_bh(&policy->lock);
1770 dst_release(dst_orig);
1771 xfrm_pols_put(pols, npols);
1775 xfrm_pols_put(pols, npols);
1777 dst_release(dst_orig);
1783 if (flags & XFRM_LOOKUP_ICMP)
1787 EXPORT_SYMBOL(__xfrm_lookup);
1789 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1790 struct sock *sk, int flags)
1792 int err = __xfrm_lookup(dst_p, fl, sk, flags);
1794 if (err == -EREMOTE) {
1795 dst_release(*dst_p);
1802 EXPORT_SYMBOL(xfrm_lookup);
1805 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1807 struct xfrm_state *x;
1809 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1811 x = skb->sp->xvec[idx];
1812 if (!x->type->reject)
1814 return x->type->reject(x, skb, fl);
1817 /* When skb is transformed back to its "native" form, we have to
1818 * check policy restrictions. At the moment we make this in maximally
1819 * stupid way. Shame on me. :-) Of course, connected sockets must
1820 * have policy cached at them.
1824 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1825 unsigned short family)
1827 if (xfrm_state_kern(x))
1828 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1829 return x->id.proto == tmpl->id.proto &&
1830 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1831 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1832 x->props.mode == tmpl->mode &&
1833 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1834 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1835 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1836 xfrm_state_addr_cmp(tmpl, x, family));
1840 * 0 or more than 0 is returned when validation is succeeded (either bypass
1841 * because of optional transport mode, or next index of the mathced secpath
1842 * state with the template.
1843 * -1 is returned when no matching template is found.
1844 * Otherwise "-2 - errored_index" is returned.
1847 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1848 unsigned short family)
1852 if (tmpl->optional) {
1853 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1857 for (; idx < sp->len; idx++) {
1858 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1860 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1869 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1870 unsigned int family, int reverse)
1872 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1875 if (unlikely(afinfo == NULL))
1876 return -EAFNOSUPPORT;
1878 afinfo->decode_session(skb, fl, reverse);
1879 err = security_xfrm_decode_session(skb, &fl->secid);
1880 xfrm_policy_put_afinfo(afinfo);
1883 EXPORT_SYMBOL(__xfrm_decode_session);
1885 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1887 for (; k < sp->len; k++) {
1888 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1897 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1898 unsigned short family)
1900 struct xfrm_policy *pol;
1901 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1910 reverse = dir & ~XFRM_POLICY_MASK;
1911 dir &= XFRM_POLICY_MASK;
1912 fl_dir = policy_to_flow_dir(dir);
1914 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1915 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1919 nf_nat_decode_session(skb, &fl, family);
1921 /* First, check used SA against their selectors. */
1925 for (i=skb->sp->len-1; i>=0; i--) {
1926 struct xfrm_state *x = skb->sp->xvec[i];
1927 if (!xfrm_selector_match(&x->sel, &fl, family)) {
1928 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1935 if (sk && sk->sk_policy[dir]) {
1936 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1938 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1944 pol = flow_cache_lookup(&fl, family, fl_dir,
1945 xfrm_policy_lookup);
1948 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1953 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1954 xfrm_secpath_reject(xerr_idx, skb, &fl);
1955 XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1961 pol->curlft.use_time = get_seconds();
1965 #ifdef CONFIG_XFRM_SUB_POLICY
1966 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1967 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1971 if (IS_ERR(pols[1])) {
1972 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1975 pols[1]->curlft.use_time = get_seconds();
1981 if (pol->action == XFRM_POLICY_ALLOW) {
1982 struct sec_path *sp;
1983 static struct sec_path dummy;
1984 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1985 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1986 struct xfrm_tmpl **tpp = tp;
1990 if ((sp = skb->sp) == NULL)
1993 for (pi = 0; pi < npols; pi++) {
1994 if (pols[pi] != pol &&
1995 pols[pi]->action != XFRM_POLICY_ALLOW) {
1996 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1999 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2000 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
2003 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2004 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2008 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2012 /* For each tunnel xfrm, find the first matching tmpl.
2013 * For each tmpl before that, find corresponding xfrm.
2014 * Order is _important_. Later we will implement
2015 * some barriers, but at the moment barriers
2016 * are implied between each two transformations.
2018 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2019 k = xfrm_policy_ok(tpp[i], sp, k, family);
2022 /* "-2 - errored_index" returned */
2024 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2029 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2030 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2034 xfrm_pols_put(pols, npols);
2037 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2040 xfrm_secpath_reject(xerr_idx, skb, &fl);
2042 xfrm_pols_put(pols, npols);
2045 EXPORT_SYMBOL(__xfrm_policy_check);
2047 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2051 if (xfrm_decode_session(skb, &fl, family) < 0) {
2052 /* XXX: we should have something like FWDHDRERROR here. */
2053 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2057 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2059 EXPORT_SYMBOL(__xfrm_route_forward);
2061 /* Optimize later using cookies and generation ids. */
2063 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2065 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2066 * to "-1" to force all XFRM destinations to get validated by
2067 * dst_ops->check on every use. We do this because when a
2068 * normal route referenced by an XFRM dst is obsoleted we do
2069 * not go looking around for all parent referencing XFRM dsts
2070 * so that we can invalidate them. It is just too much work.
2071 * Instead we make the checks here on every use. For example:
2073 * XFRM dst A --> IPv4 dst X
2075 * X is the "xdst->route" of A (X is also the "dst->path" of A
2076 * in this example). If X is marked obsolete, "A" will not
2077 * notice. That's what we are validating here via the
2078 * stale_bundle() check.
2080 * When a policy's bundle is pruned, we dst_free() the XFRM
2081 * dst which causes it's ->obsolete field to be set to a
2082 * positive non-zero integer. If an XFRM dst has been pruned
2083 * like this, we want to force a new route lookup.
2085 if (dst->obsolete < 0 && !stale_bundle(dst))
2091 static int stale_bundle(struct dst_entry *dst)
2093 return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2096 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2098 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2099 dst->dev = dev_net(dev)->loopback_dev;
2104 EXPORT_SYMBOL(xfrm_dst_ifdown);
2106 static void xfrm_link_failure(struct sk_buff *skb)
2108 /* Impossible. Such dst must be popped before reaches point of failure. */
2112 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2115 if (dst->obsolete) {
2123 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2125 struct dst_entry *dst, **dstp;
2127 write_lock(&pol->lock);
2128 dstp = &pol->bundles;
2129 while ((dst=*dstp) != NULL) {
2132 dst->next = *gc_list_p;
2138 write_unlock(&pol->lock);
2141 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2143 struct dst_entry *gc_list = NULL;
2146 read_lock_bh(&xfrm_policy_lock);
2147 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2148 struct xfrm_policy *pol;
2149 struct hlist_node *entry;
2150 struct hlist_head *table;
2153 hlist_for_each_entry(pol, entry,
2154 &init_net.xfrm.policy_inexact[dir], bydst)
2155 prune_one_bundle(pol, func, &gc_list);
2157 table = xfrm_policy_bydst[dir].table;
2158 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2159 hlist_for_each_entry(pol, entry, table + i, bydst)
2160 prune_one_bundle(pol, func, &gc_list);
2163 read_unlock_bh(&xfrm_policy_lock);
2166 struct dst_entry *dst = gc_list;
2167 gc_list = dst->next;
2172 static int unused_bundle(struct dst_entry *dst)
2174 return !atomic_read(&dst->__refcnt);
2177 static void __xfrm_garbage_collect(void)
2179 xfrm_prune_bundles(unused_bundle);
2182 static int xfrm_flush_bundles(void)
2184 xfrm_prune_bundles(stale_bundle);
2188 static void xfrm_init_pmtu(struct dst_entry *dst)
2191 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2192 u32 pmtu, route_mtu_cached;
2194 pmtu = dst_mtu(dst->child);
2195 xdst->child_mtu_cached = pmtu;
2197 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2199 route_mtu_cached = dst_mtu(xdst->route);
2200 xdst->route_mtu_cached = route_mtu_cached;
2202 if (pmtu > route_mtu_cached)
2203 pmtu = route_mtu_cached;
2205 dst->metrics[RTAX_MTU-1] = pmtu;
2206 } while ((dst = dst->next));
2209 /* Check that the bundle accepts the flow and its components are
2213 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2214 struct flowi *fl, int family, int strict)
2216 struct dst_entry *dst = &first->u.dst;
2217 struct xfrm_dst *last;
2220 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2221 (dst->dev && !netif_running(dst->dev)))
2223 #ifdef CONFIG_XFRM_SUB_POLICY
2225 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2227 if (first->partner &&
2228 !xfrm_selector_match(first->partner, fl, family))
2236 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2238 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2241 !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2243 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2245 if (xdst->genid != dst->xfrm->genid)
2249 !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2250 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2253 mtu = dst_mtu(dst->child);
2254 if (xdst->child_mtu_cached != mtu) {
2256 xdst->child_mtu_cached = mtu;
2259 if (!dst_check(xdst->route, xdst->route_cookie))
2261 mtu = dst_mtu(xdst->route);
2262 if (xdst->route_mtu_cached != mtu) {
2264 xdst->route_mtu_cached = mtu;
2268 } while (dst->xfrm);
2273 mtu = last->child_mtu_cached;
2277 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2278 if (mtu > last->route_mtu_cached)
2279 mtu = last->route_mtu_cached;
2280 dst->metrics[RTAX_MTU-1] = mtu;
2285 last = (struct xfrm_dst *)last->u.dst.next;
2286 last->child_mtu_cached = mtu;
2292 EXPORT_SYMBOL(xfrm_bundle_ok);
2294 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2297 if (unlikely(afinfo == NULL))
2299 if (unlikely(afinfo->family >= NPROTO))
2300 return -EAFNOSUPPORT;
2301 write_lock_bh(&xfrm_policy_afinfo_lock);
2302 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2305 struct dst_ops *dst_ops = afinfo->dst_ops;
2306 if (likely(dst_ops->kmem_cachep == NULL))
2307 dst_ops->kmem_cachep = xfrm_dst_cache;
2308 if (likely(dst_ops->check == NULL))
2309 dst_ops->check = xfrm_dst_check;
2310 if (likely(dst_ops->negative_advice == NULL))
2311 dst_ops->negative_advice = xfrm_negative_advice;
2312 if (likely(dst_ops->link_failure == NULL))
2313 dst_ops->link_failure = xfrm_link_failure;
2314 if (likely(afinfo->garbage_collect == NULL))
2315 afinfo->garbage_collect = __xfrm_garbage_collect;
2316 xfrm_policy_afinfo[afinfo->family] = afinfo;
2318 write_unlock_bh(&xfrm_policy_afinfo_lock);
2321 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2323 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2326 if (unlikely(afinfo == NULL))
2328 if (unlikely(afinfo->family >= NPROTO))
2329 return -EAFNOSUPPORT;
2330 write_lock_bh(&xfrm_policy_afinfo_lock);
2331 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2332 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2335 struct dst_ops *dst_ops = afinfo->dst_ops;
2336 xfrm_policy_afinfo[afinfo->family] = NULL;
2337 dst_ops->kmem_cachep = NULL;
2338 dst_ops->check = NULL;
2339 dst_ops->negative_advice = NULL;
2340 dst_ops->link_failure = NULL;
2341 afinfo->garbage_collect = NULL;
2344 write_unlock_bh(&xfrm_policy_afinfo_lock);
2347 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2349 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2351 struct xfrm_policy_afinfo *afinfo;
2352 if (unlikely(family >= NPROTO))
2354 read_lock(&xfrm_policy_afinfo_lock);
2355 afinfo = xfrm_policy_afinfo[family];
2356 if (unlikely(!afinfo))
2357 read_unlock(&xfrm_policy_afinfo_lock);
2361 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2363 read_unlock(&xfrm_policy_afinfo_lock);
2366 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2368 struct net_device *dev = ptr;
2370 if (!net_eq(dev_net(dev), &init_net))
2375 xfrm_flush_bundles();
2380 static struct notifier_block xfrm_dev_notifier = {
2381 .notifier_call = xfrm_dev_event,
2384 #ifdef CONFIG_XFRM_STATISTICS
2385 static int __init xfrm_statistics_init(void)
2387 if (snmp_mib_init((void **)xfrm_statistics,
2388 sizeof(struct linux_xfrm_mib)) < 0)
2394 static int __net_init xfrm_policy_init(struct net *net)
2396 unsigned int hmask, sz;
2399 if (net_eq(net, &init_net))
2400 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2401 sizeof(struct xfrm_dst),
2402 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2406 sz = (hmask+1) * sizeof(struct hlist_head);
2408 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2409 if (!net->xfrm.policy_byidx)
2411 net->xfrm.policy_idx_hmask = hmask;
2413 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2414 struct xfrm_policy_hash *htab;
2416 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2418 htab = &xfrm_policy_bydst[dir];
2419 htab->table = xfrm_hash_alloc(sz);
2420 htab->hmask = hmask;
2422 panic("XFRM: failed to allocate bydst hash\n");
2425 INIT_LIST_HEAD(&net->xfrm.policy_all);
2426 if (net_eq(net, &init_net))
2427 register_netdevice_notifier(&xfrm_dev_notifier);
2434 static void xfrm_policy_fini(struct net *net)
2439 WARN_ON(!list_empty(&net->xfrm.policy_all));
2441 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2442 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2445 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2446 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2447 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2450 static int __net_init xfrm_net_init(struct net *net)
2454 rv = xfrm_state_init(net);
2457 rv = xfrm_policy_init(net);
2463 xfrm_state_fini(net);
2468 static void __net_exit xfrm_net_exit(struct net *net)
2470 xfrm_policy_fini(net);
2471 xfrm_state_fini(net);
2474 static struct pernet_operations __net_initdata xfrm_net_ops = {
2475 .init = xfrm_net_init,
2476 .exit = xfrm_net_exit,
2479 void __init xfrm_init(void)
2481 register_pernet_subsys(&xfrm_net_ops);
2482 #ifdef CONFIG_XFRM_STATISTICS
2483 xfrm_statistics_init();
2486 #ifdef CONFIG_XFRM_STATISTICS
2491 #ifdef CONFIG_AUDITSYSCALL
2492 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2493 struct audit_buffer *audit_buf)
2495 struct xfrm_sec_ctx *ctx = xp->security;
2496 struct xfrm_selector *sel = &xp->selector;
2499 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2500 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2502 switch(sel->family) {
2504 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2505 if (sel->prefixlen_s != 32)
2506 audit_log_format(audit_buf, " src_prefixlen=%d",
2508 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2509 if (sel->prefixlen_d != 32)
2510 audit_log_format(audit_buf, " dst_prefixlen=%d",
2514 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2515 if (sel->prefixlen_s != 128)
2516 audit_log_format(audit_buf, " src_prefixlen=%d",
2518 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2519 if (sel->prefixlen_d != 128)
2520 audit_log_format(audit_buf, " dst_prefixlen=%d",
2526 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2527 uid_t auid, u32 sessionid, u32 secid)
2529 struct audit_buffer *audit_buf;
2531 audit_buf = xfrm_audit_start("SPD-add");
2532 if (audit_buf == NULL)
2534 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2535 audit_log_format(audit_buf, " res=%u", result);
2536 xfrm_audit_common_policyinfo(xp, audit_buf);
2537 audit_log_end(audit_buf);
2539 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2541 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2542 uid_t auid, u32 sessionid, u32 secid)
2544 struct audit_buffer *audit_buf;
2546 audit_buf = xfrm_audit_start("SPD-delete");
2547 if (audit_buf == NULL)
2549 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2550 audit_log_format(audit_buf, " res=%u", result);
2551 xfrm_audit_common_policyinfo(xp, audit_buf);
2552 audit_log_end(audit_buf);
2554 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2557 #ifdef CONFIG_XFRM_MIGRATE
2558 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2559 struct xfrm_selector *sel_tgt)
2561 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2562 if (sel_tgt->family == sel_cmp->family &&
2563 xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2564 sel_cmp->family) == 0 &&
2565 xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2566 sel_cmp->family) == 0 &&
2567 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2568 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2572 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2579 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2582 struct xfrm_policy *pol, *ret = NULL;
2583 struct hlist_node *entry;
2584 struct hlist_head *chain;
2587 read_lock_bh(&xfrm_policy_lock);
2588 chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2589 hlist_for_each_entry(pol, entry, chain, bydst) {
2590 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2591 pol->type == type) {
2593 priority = ret->priority;
2597 chain = &init_net.xfrm.policy_inexact[dir];
2598 hlist_for_each_entry(pol, entry, chain, bydst) {
2599 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2600 pol->type == type &&
2601 pol->priority < priority) {
2610 read_unlock_bh(&xfrm_policy_lock);
2615 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2619 if (t->mode == m->mode && t->id.proto == m->proto &&
2620 (m->reqid == 0 || t->reqid == m->reqid)) {
2622 case XFRM_MODE_TUNNEL:
2623 case XFRM_MODE_BEET:
2624 if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2625 m->old_family) == 0 &&
2626 xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2627 m->old_family) == 0) {
2631 case XFRM_MODE_TRANSPORT:
2632 /* in case of transport mode, template does not store
2633 any IP addresses, hence we just compare mode and
2644 /* update endpoint address(es) of template(s) */
2645 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2646 struct xfrm_migrate *m, int num_migrate)
2648 struct xfrm_migrate *mp;
2649 struct dst_entry *dst;
2652 write_lock_bh(&pol->lock);
2653 if (unlikely(pol->walk.dead)) {
2654 /* target policy has been deleted */
2655 write_unlock_bh(&pol->lock);
2659 for (i = 0; i < pol->xfrm_nr; i++) {
2660 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2661 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2664 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2665 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2667 /* update endpoints */
2668 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2669 sizeof(pol->xfrm_vec[i].id.daddr));
2670 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2671 sizeof(pol->xfrm_vec[i].saddr));
2672 pol->xfrm_vec[i].encap_family = mp->new_family;
2674 while ((dst = pol->bundles) != NULL) {
2675 pol->bundles = dst->next;
2681 write_unlock_bh(&pol->lock);
2689 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2693 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2696 for (i = 0; i < num_migrate; i++) {
2697 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2698 m[i].old_family) == 0) &&
2699 (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2700 m[i].old_family) == 0))
2702 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2703 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2706 /* check if there is any duplicated entry */
2707 for (j = i + 1; j < num_migrate; j++) {
2708 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2709 sizeof(m[i].old_daddr)) &&
2710 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2711 sizeof(m[i].old_saddr)) &&
2712 m[i].proto == m[j].proto &&
2713 m[i].mode == m[j].mode &&
2714 m[i].reqid == m[j].reqid &&
2715 m[i].old_family == m[j].old_family)
2723 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2724 struct xfrm_migrate *m, int num_migrate,
2725 struct xfrm_kmaddress *k)
2727 int i, err, nx_cur = 0, nx_new = 0;
2728 struct xfrm_policy *pol = NULL;
2729 struct xfrm_state *x, *xc;
2730 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2731 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2732 struct xfrm_migrate *mp;
2734 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2737 /* Stage 1 - find policy */
2738 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2743 /* Stage 2 - find and update state(s) */
2744 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2745 if ((x = xfrm_migrate_state_find(mp))) {
2748 if ((xc = xfrm_state_migrate(x, mp))) {
2758 /* Stage 3 - update policy */
2759 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2762 /* Stage 4 - delete old state(s) */
2764 xfrm_states_put(x_cur, nx_cur);
2765 xfrm_states_delete(x_cur, nx_cur);
2768 /* Stage 5 - announce */
2769 km_migrate(sel, dir, type, m, num_migrate, k);
2781 xfrm_states_put(x_cur, nx_cur);
2783 xfrm_states_delete(x_new, nx_new);
2787 EXPORT_SYMBOL(xfrm_migrate);