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/slab.h>
17 #include <linux/kmod.h>
18 #include <linux/list.h>
19 #include <linux/spinlock.h>
20 #include <linux/workqueue.h>
21 #include <linux/notifier.h>
22 #include <linux/netdevice.h>
23 #include <linux/netfilter.h>
24 #include <linux/module.h>
25 #include <linux/cache.h>
28 #include <linux/audit.h>
30 #include "xfrm_hash.h"
32 int sysctl_xfrm_larval_drop;
34 DEFINE_MUTEX(xfrm_cfg_mutex);
35 EXPORT_SYMBOL(xfrm_cfg_mutex);
37 static DEFINE_RWLOCK(xfrm_policy_lock);
39 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
40 EXPORT_SYMBOL(xfrm_policy_count);
42 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
43 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
45 static struct kmem_cache *xfrm_dst_cache __read_mostly;
47 static struct work_struct xfrm_policy_gc_work;
48 static HLIST_HEAD(xfrm_policy_gc_list);
49 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
51 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
52 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
53 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
54 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
57 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
59 return addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
60 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
61 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
62 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
63 (fl->proto == sel->proto || !sel->proto) &&
64 (fl->oif == sel->ifindex || !sel->ifindex);
68 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
70 return addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
71 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
72 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
73 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
74 (fl->proto == sel->proto || !sel->proto) &&
75 (fl->oif == sel->ifindex || !sel->ifindex);
78 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
79 unsigned short family)
83 return __xfrm4_selector_match(sel, fl);
85 return __xfrm6_selector_match(sel, fl);
90 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
92 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
93 struct xfrm_type **typemap;
96 if (unlikely(afinfo == NULL))
98 typemap = afinfo->type_map;
100 if (likely(typemap[type->proto] == NULL))
101 typemap[type->proto] = type;
104 xfrm_policy_unlock_afinfo(afinfo);
107 EXPORT_SYMBOL(xfrm_register_type);
109 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
111 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
112 struct xfrm_type **typemap;
115 if (unlikely(afinfo == NULL))
116 return -EAFNOSUPPORT;
117 typemap = afinfo->type_map;
119 if (unlikely(typemap[type->proto] != type))
122 typemap[type->proto] = NULL;
123 xfrm_policy_unlock_afinfo(afinfo);
126 EXPORT_SYMBOL(xfrm_unregister_type);
128 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
130 struct xfrm_policy_afinfo *afinfo;
131 struct xfrm_type **typemap;
132 struct xfrm_type *type;
133 int modload_attempted = 0;
136 afinfo = xfrm_policy_get_afinfo(family);
137 if (unlikely(afinfo == NULL))
139 typemap = afinfo->type_map;
141 type = typemap[proto];
142 if (unlikely(type && !try_module_get(type->owner)))
144 if (!type && !modload_attempted) {
145 xfrm_policy_put_afinfo(afinfo);
146 request_module("xfrm-type-%d-%d",
147 (int) family, (int) proto);
148 modload_attempted = 1;
152 xfrm_policy_put_afinfo(afinfo);
156 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
157 unsigned short family)
159 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
162 if (unlikely(afinfo == NULL))
163 return -EAFNOSUPPORT;
165 if (likely(afinfo->dst_lookup != NULL))
166 err = afinfo->dst_lookup(dst, fl);
169 xfrm_policy_put_afinfo(afinfo);
172 EXPORT_SYMBOL(xfrm_dst_lookup);
174 void xfrm_put_type(struct xfrm_type *type)
176 module_put(type->owner);
179 int xfrm_register_mode(struct xfrm_mode *mode, int family)
181 struct xfrm_policy_afinfo *afinfo;
182 struct xfrm_mode **modemap;
185 if (unlikely(mode->encap >= XFRM_MODE_MAX))
188 afinfo = xfrm_policy_lock_afinfo(family);
189 if (unlikely(afinfo == NULL))
190 return -EAFNOSUPPORT;
193 modemap = afinfo->mode_map;
194 if (likely(modemap[mode->encap] == NULL)) {
195 modemap[mode->encap] = mode;
199 xfrm_policy_unlock_afinfo(afinfo);
202 EXPORT_SYMBOL(xfrm_register_mode);
204 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
206 struct xfrm_policy_afinfo *afinfo;
207 struct xfrm_mode **modemap;
210 if (unlikely(mode->encap >= XFRM_MODE_MAX))
213 afinfo = xfrm_policy_lock_afinfo(family);
214 if (unlikely(afinfo == NULL))
215 return -EAFNOSUPPORT;
218 modemap = afinfo->mode_map;
219 if (likely(modemap[mode->encap] == mode)) {
220 modemap[mode->encap] = NULL;
224 xfrm_policy_unlock_afinfo(afinfo);
227 EXPORT_SYMBOL(xfrm_unregister_mode);
229 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
231 struct xfrm_policy_afinfo *afinfo;
232 struct xfrm_mode *mode;
233 int modload_attempted = 0;
235 if (unlikely(encap >= XFRM_MODE_MAX))
239 afinfo = xfrm_policy_get_afinfo(family);
240 if (unlikely(afinfo == NULL))
243 mode = afinfo->mode_map[encap];
244 if (unlikely(mode && !try_module_get(mode->owner)))
246 if (!mode && !modload_attempted) {
247 xfrm_policy_put_afinfo(afinfo);
248 request_module("xfrm-mode-%d-%d", family, encap);
249 modload_attempted = 1;
253 xfrm_policy_put_afinfo(afinfo);
257 void xfrm_put_mode(struct xfrm_mode *mode)
259 module_put(mode->owner);
262 static inline unsigned long make_jiffies(long secs)
264 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
265 return MAX_SCHEDULE_TIMEOUT-1;
270 static void xfrm_policy_timer(unsigned long data)
272 struct xfrm_policy *xp = (struct xfrm_policy*)data;
273 unsigned long now = get_seconds();
274 long next = LONG_MAX;
278 read_lock(&xp->lock);
283 dir = xfrm_policy_id2dir(xp->index);
285 if (xp->lft.hard_add_expires_seconds) {
286 long tmo = xp->lft.hard_add_expires_seconds +
287 xp->curlft.add_time - now;
293 if (xp->lft.hard_use_expires_seconds) {
294 long tmo = xp->lft.hard_use_expires_seconds +
295 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
301 if (xp->lft.soft_add_expires_seconds) {
302 long tmo = xp->lft.soft_add_expires_seconds +
303 xp->curlft.add_time - now;
306 tmo = XFRM_KM_TIMEOUT;
311 if (xp->lft.soft_use_expires_seconds) {
312 long tmo = xp->lft.soft_use_expires_seconds +
313 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
316 tmo = XFRM_KM_TIMEOUT;
323 km_policy_expired(xp, dir, 0, 0);
324 if (next != LONG_MAX &&
325 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
329 read_unlock(&xp->lock);
334 read_unlock(&xp->lock);
335 if (!xfrm_policy_delete(xp, dir))
336 km_policy_expired(xp, dir, 1, 0);
341 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
345 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
347 struct xfrm_policy *policy;
349 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
352 INIT_HLIST_NODE(&policy->bydst);
353 INIT_HLIST_NODE(&policy->byidx);
354 rwlock_init(&policy->lock);
355 atomic_set(&policy->refcnt, 1);
356 init_timer(&policy->timer);
357 policy->timer.data = (unsigned long)policy;
358 policy->timer.function = xfrm_policy_timer;
362 EXPORT_SYMBOL(xfrm_policy_alloc);
364 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
366 void __xfrm_policy_destroy(struct xfrm_policy *policy)
368 BUG_ON(!policy->dead);
370 BUG_ON(policy->bundles);
372 if (del_timer(&policy->timer))
375 security_xfrm_policy_free(policy);
378 EXPORT_SYMBOL(__xfrm_policy_destroy);
380 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
382 struct dst_entry *dst;
384 while ((dst = policy->bundles) != NULL) {
385 policy->bundles = dst->next;
389 if (del_timer(&policy->timer))
390 atomic_dec(&policy->refcnt);
392 if (atomic_read(&policy->refcnt) > 1)
395 xfrm_pol_put(policy);
398 static void xfrm_policy_gc_task(struct work_struct *work)
400 struct xfrm_policy *policy;
401 struct hlist_node *entry, *tmp;
402 struct hlist_head gc_list;
404 spin_lock_bh(&xfrm_policy_gc_lock);
405 gc_list.first = xfrm_policy_gc_list.first;
406 INIT_HLIST_HEAD(&xfrm_policy_gc_list);
407 spin_unlock_bh(&xfrm_policy_gc_lock);
409 hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
410 xfrm_policy_gc_kill(policy);
413 /* Rule must be locked. Release descentant resources, announce
414 * entry dead. The rule must be unlinked from lists to the moment.
417 static void xfrm_policy_kill(struct xfrm_policy *policy)
421 write_lock_bh(&policy->lock);
424 write_unlock_bh(&policy->lock);
426 if (unlikely(dead)) {
431 spin_lock(&xfrm_policy_gc_lock);
432 hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
433 spin_unlock(&xfrm_policy_gc_lock);
435 schedule_work(&xfrm_policy_gc_work);
438 struct xfrm_policy_hash {
439 struct hlist_head *table;
443 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
444 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
445 static struct hlist_head *xfrm_policy_byidx __read_mostly;
446 static unsigned int xfrm_idx_hmask __read_mostly;
447 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
449 static inline unsigned int idx_hash(u32 index)
451 return __idx_hash(index, xfrm_idx_hmask);
454 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
456 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
457 unsigned int hash = __sel_hash(sel, family, hmask);
459 return (hash == hmask + 1 ?
460 &xfrm_policy_inexact[dir] :
461 xfrm_policy_bydst[dir].table + hash);
464 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
466 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
467 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
469 return xfrm_policy_bydst[dir].table + hash;
472 static void xfrm_dst_hash_transfer(struct hlist_head *list,
473 struct hlist_head *ndsttable,
474 unsigned int nhashmask)
476 struct hlist_node *entry, *tmp;
477 struct xfrm_policy *pol;
479 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
482 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
483 pol->family, nhashmask);
484 hlist_add_head(&pol->bydst, ndsttable+h);
488 static void xfrm_idx_hash_transfer(struct hlist_head *list,
489 struct hlist_head *nidxtable,
490 unsigned int nhashmask)
492 struct hlist_node *entry, *tmp;
493 struct xfrm_policy *pol;
495 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
498 h = __idx_hash(pol->index, nhashmask);
499 hlist_add_head(&pol->byidx, nidxtable+h);
503 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
505 return ((old_hmask + 1) << 1) - 1;
508 static void xfrm_bydst_resize(int dir)
510 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
511 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
512 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
513 struct hlist_head *odst = xfrm_policy_bydst[dir].table;
514 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
520 write_lock_bh(&xfrm_policy_lock);
522 for (i = hmask; i >= 0; i--)
523 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
525 xfrm_policy_bydst[dir].table = ndst;
526 xfrm_policy_bydst[dir].hmask = nhashmask;
528 write_unlock_bh(&xfrm_policy_lock);
530 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
533 static void xfrm_byidx_resize(int total)
535 unsigned int hmask = xfrm_idx_hmask;
536 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
537 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
538 struct hlist_head *oidx = xfrm_policy_byidx;
539 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
545 write_lock_bh(&xfrm_policy_lock);
547 for (i = hmask; i >= 0; i--)
548 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
550 xfrm_policy_byidx = nidx;
551 xfrm_idx_hmask = nhashmask;
553 write_unlock_bh(&xfrm_policy_lock);
555 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
558 static inline int xfrm_bydst_should_resize(int dir, int *total)
560 unsigned int cnt = xfrm_policy_count[dir];
561 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
566 if ((hmask + 1) < xfrm_policy_hashmax &&
573 static inline int xfrm_byidx_should_resize(int total)
575 unsigned int hmask = xfrm_idx_hmask;
577 if ((hmask + 1) < xfrm_policy_hashmax &&
584 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
586 read_lock_bh(&xfrm_policy_lock);
587 si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
588 si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
589 si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
590 si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
591 si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
592 si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
593 si->spdhcnt = xfrm_idx_hmask;
594 si->spdhmcnt = xfrm_policy_hashmax;
595 read_unlock_bh(&xfrm_policy_lock);
597 EXPORT_SYMBOL(xfrm_spd_getinfo);
599 static DEFINE_MUTEX(hash_resize_mutex);
600 static void xfrm_hash_resize(struct work_struct *__unused)
604 mutex_lock(&hash_resize_mutex);
607 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
608 if (xfrm_bydst_should_resize(dir, &total))
609 xfrm_bydst_resize(dir);
611 if (xfrm_byidx_should_resize(total))
612 xfrm_byidx_resize(total);
614 mutex_unlock(&hash_resize_mutex);
617 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
619 /* Generate new index... KAME seems to generate them ordered by cost
620 * of an absolute inpredictability of ordering of rules. This will not pass. */
621 static u32 xfrm_gen_index(u8 type, int dir)
623 static u32 idx_generator;
626 struct hlist_node *entry;
627 struct hlist_head *list;
628 struct xfrm_policy *p;
632 idx = (idx_generator | dir);
636 list = xfrm_policy_byidx + idx_hash(idx);
638 hlist_for_each_entry(p, entry, list, byidx) {
639 if (p->index == idx) {
649 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
651 u32 *p1 = (u32 *) s1;
652 u32 *p2 = (u32 *) s2;
653 int len = sizeof(struct xfrm_selector) / sizeof(u32);
656 for (i = 0; i < len; i++) {
664 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
666 struct xfrm_policy *pol;
667 struct xfrm_policy *delpol;
668 struct hlist_head *chain;
669 struct hlist_node *entry, *newpos;
670 struct dst_entry *gc_list;
672 write_lock_bh(&xfrm_policy_lock);
673 chain = policy_hash_bysel(&policy->selector, policy->family, dir);
676 hlist_for_each_entry(pol, entry, chain, bydst) {
677 if (pol->type == policy->type &&
678 !selector_cmp(&pol->selector, &policy->selector) &&
679 xfrm_sec_ctx_match(pol->security, policy->security) &&
682 write_unlock_bh(&xfrm_policy_lock);
686 if (policy->priority > pol->priority)
688 } else if (policy->priority >= pol->priority) {
689 newpos = &pol->bydst;
696 hlist_add_after(newpos, &policy->bydst);
698 hlist_add_head(&policy->bydst, chain);
699 xfrm_pol_hold(policy);
700 xfrm_policy_count[dir]++;
701 atomic_inc(&flow_cache_genid);
703 hlist_del(&delpol->bydst);
704 hlist_del(&delpol->byidx);
705 xfrm_policy_count[dir]--;
707 policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
708 hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
709 policy->curlft.add_time = get_seconds();
710 policy->curlft.use_time = 0;
711 if (!mod_timer(&policy->timer, jiffies + HZ))
712 xfrm_pol_hold(policy);
713 write_unlock_bh(&xfrm_policy_lock);
716 xfrm_policy_kill(delpol);
717 else if (xfrm_bydst_should_resize(dir, NULL))
718 schedule_work(&xfrm_hash_work);
720 read_lock_bh(&xfrm_policy_lock);
722 entry = &policy->bydst;
723 hlist_for_each_entry_continue(policy, entry, bydst) {
724 struct dst_entry *dst;
726 write_lock(&policy->lock);
727 dst = policy->bundles;
729 struct dst_entry *tail = dst;
732 tail->next = gc_list;
735 policy->bundles = NULL;
737 write_unlock(&policy->lock);
739 read_unlock_bh(&xfrm_policy_lock);
742 struct dst_entry *dst = gc_list;
750 EXPORT_SYMBOL(xfrm_policy_insert);
752 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
753 struct xfrm_selector *sel,
754 struct xfrm_sec_ctx *ctx, int delete,
757 struct xfrm_policy *pol, *ret;
758 struct hlist_head *chain;
759 struct hlist_node *entry;
762 write_lock_bh(&xfrm_policy_lock);
763 chain = policy_hash_bysel(sel, sel->family, dir);
765 hlist_for_each_entry(pol, entry, chain, bydst) {
766 if (pol->type == type &&
767 !selector_cmp(sel, &pol->selector) &&
768 xfrm_sec_ctx_match(ctx, pol->security)) {
771 *err = security_xfrm_policy_delete(pol);
773 write_unlock_bh(&xfrm_policy_lock);
776 hlist_del(&pol->bydst);
777 hlist_del(&pol->byidx);
778 xfrm_policy_count[dir]--;
784 write_unlock_bh(&xfrm_policy_lock);
787 atomic_inc(&flow_cache_genid);
788 xfrm_policy_kill(ret);
792 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
794 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
797 struct xfrm_policy *pol, *ret;
798 struct hlist_head *chain;
799 struct hlist_node *entry;
802 if (xfrm_policy_id2dir(id) != dir)
806 write_lock_bh(&xfrm_policy_lock);
807 chain = xfrm_policy_byidx + idx_hash(id);
809 hlist_for_each_entry(pol, entry, chain, byidx) {
810 if (pol->type == type && pol->index == id) {
813 *err = security_xfrm_policy_delete(pol);
815 write_unlock_bh(&xfrm_policy_lock);
818 hlist_del(&pol->bydst);
819 hlist_del(&pol->byidx);
820 xfrm_policy_count[dir]--;
826 write_unlock_bh(&xfrm_policy_lock);
829 atomic_inc(&flow_cache_genid);
830 xfrm_policy_kill(ret);
834 EXPORT_SYMBOL(xfrm_policy_byid);
836 void xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
840 write_lock_bh(&xfrm_policy_lock);
841 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
842 struct xfrm_policy *pol;
843 struct hlist_node *entry;
848 hlist_for_each_entry(pol, entry,
849 &xfrm_policy_inexact[dir], bydst) {
850 if (pol->type != type)
852 hlist_del(&pol->bydst);
853 hlist_del(&pol->byidx);
854 write_unlock_bh(&xfrm_policy_lock);
856 xfrm_audit_log(audit_info->loginuid, audit_info->secid,
857 AUDIT_MAC_IPSEC_DELSPD, 1, pol, NULL);
859 xfrm_policy_kill(pol);
862 write_lock_bh(&xfrm_policy_lock);
866 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
868 hlist_for_each_entry(pol, entry,
869 xfrm_policy_bydst[dir].table + i,
871 if (pol->type != type)
873 hlist_del(&pol->bydst);
874 hlist_del(&pol->byidx);
875 write_unlock_bh(&xfrm_policy_lock);
877 xfrm_audit_log(audit_info->loginuid,
879 AUDIT_MAC_IPSEC_DELSPD, 1,
882 xfrm_policy_kill(pol);
885 write_lock_bh(&xfrm_policy_lock);
890 xfrm_policy_count[dir] -= killed;
892 atomic_inc(&flow_cache_genid);
893 write_unlock_bh(&xfrm_policy_lock);
895 EXPORT_SYMBOL(xfrm_policy_flush);
897 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
900 struct xfrm_policy *pol, *last = NULL;
901 struct hlist_node *entry;
902 int dir, last_dir = 0, count, error;
904 read_lock_bh(&xfrm_policy_lock);
907 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
908 struct hlist_head *table = xfrm_policy_bydst[dir].table;
911 hlist_for_each_entry(pol, entry,
912 &xfrm_policy_inexact[dir], bydst) {
913 if (pol->type != type)
916 error = func(last, last_dir % XFRM_POLICY_MAX,
925 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
926 hlist_for_each_entry(pol, entry, table + i, bydst) {
927 if (pol->type != type)
930 error = func(last, last_dir % XFRM_POLICY_MAX,
945 error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
947 read_unlock_bh(&xfrm_policy_lock);
950 EXPORT_SYMBOL(xfrm_policy_walk);
953 * Find policy to apply to this flow.
955 * Returns 0 if policy found, else an -errno.
957 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
958 u8 type, u16 family, int dir)
960 struct xfrm_selector *sel = &pol->selector;
961 int match, ret = -ESRCH;
963 if (pol->family != family ||
967 match = xfrm_selector_match(sel, fl, family);
969 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
974 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
978 struct xfrm_policy *pol, *ret;
979 xfrm_address_t *daddr, *saddr;
980 struct hlist_node *entry;
981 struct hlist_head *chain;
984 daddr = xfrm_flowi_daddr(fl, family);
985 saddr = xfrm_flowi_saddr(fl, family);
986 if (unlikely(!daddr || !saddr))
989 read_lock_bh(&xfrm_policy_lock);
990 chain = policy_hash_direct(daddr, saddr, family, dir);
992 hlist_for_each_entry(pol, entry, chain, bydst) {
993 err = xfrm_policy_match(pol, fl, type, family, dir);
1003 priority = ret->priority;
1007 chain = &xfrm_policy_inexact[dir];
1008 hlist_for_each_entry(pol, entry, chain, bydst) {
1009 err = xfrm_policy_match(pol, fl, type, family, dir);
1017 } else if (pol->priority < priority) {
1025 read_unlock_bh(&xfrm_policy_lock);
1030 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1031 void **objp, atomic_t **obj_refp)
1033 struct xfrm_policy *pol;
1036 #ifdef CONFIG_XFRM_SUB_POLICY
1037 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1045 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1050 #ifdef CONFIG_XFRM_SUB_POLICY
1053 if ((*objp = (void *) pol) != NULL)
1054 *obj_refp = &pol->refcnt;
1058 static inline int policy_to_flow_dir(int dir)
1060 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1061 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1062 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1066 case XFRM_POLICY_IN:
1068 case XFRM_POLICY_OUT:
1069 return FLOW_DIR_OUT;
1070 case XFRM_POLICY_FWD:
1071 return FLOW_DIR_FWD;
1075 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1077 struct xfrm_policy *pol;
1079 read_lock_bh(&xfrm_policy_lock);
1080 if ((pol = sk->sk_policy[dir]) != NULL) {
1081 int match = xfrm_selector_match(&pol->selector, fl,
1086 err = security_xfrm_policy_lookup(pol, fl->secid,
1087 policy_to_flow_dir(dir));
1090 else if (err == -ESRCH)
1097 read_unlock_bh(&xfrm_policy_lock);
1101 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1103 struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1106 hlist_add_head(&pol->bydst, chain);
1107 hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1108 xfrm_policy_count[dir]++;
1111 if (xfrm_bydst_should_resize(dir, NULL))
1112 schedule_work(&xfrm_hash_work);
1115 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1118 if (hlist_unhashed(&pol->bydst))
1121 hlist_del(&pol->bydst);
1122 hlist_del(&pol->byidx);
1123 xfrm_policy_count[dir]--;
1128 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1130 write_lock_bh(&xfrm_policy_lock);
1131 pol = __xfrm_policy_unlink(pol, dir);
1132 write_unlock_bh(&xfrm_policy_lock);
1134 if (dir < XFRM_POLICY_MAX)
1135 atomic_inc(&flow_cache_genid);
1136 xfrm_policy_kill(pol);
1141 EXPORT_SYMBOL(xfrm_policy_delete);
1143 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1145 struct xfrm_policy *old_pol;
1147 #ifdef CONFIG_XFRM_SUB_POLICY
1148 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1152 write_lock_bh(&xfrm_policy_lock);
1153 old_pol = sk->sk_policy[dir];
1154 sk->sk_policy[dir] = pol;
1156 pol->curlft.add_time = get_seconds();
1157 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1158 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1161 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1162 write_unlock_bh(&xfrm_policy_lock);
1165 xfrm_policy_kill(old_pol);
1170 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1172 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1175 newp->selector = old->selector;
1176 if (security_xfrm_policy_clone(old, newp)) {
1178 return NULL; /* ENOMEM */
1180 newp->lft = old->lft;
1181 newp->curlft = old->curlft;
1182 newp->action = old->action;
1183 newp->flags = old->flags;
1184 newp->xfrm_nr = old->xfrm_nr;
1185 newp->index = old->index;
1186 newp->type = old->type;
1187 memcpy(newp->xfrm_vec, old->xfrm_vec,
1188 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1189 write_lock_bh(&xfrm_policy_lock);
1190 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1191 write_unlock_bh(&xfrm_policy_lock);
1197 int __xfrm_sk_clone_policy(struct sock *sk)
1199 struct xfrm_policy *p0 = sk->sk_policy[0],
1200 *p1 = sk->sk_policy[1];
1202 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1203 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1205 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1211 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1212 unsigned short family)
1215 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1217 if (unlikely(afinfo == NULL))
1219 err = afinfo->get_saddr(local, remote);
1220 xfrm_policy_put_afinfo(afinfo);
1224 /* Resolve list of templates for the flow, given policy. */
1227 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1228 struct xfrm_state **xfrm,
1229 unsigned short family)
1233 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1234 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1237 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1238 struct xfrm_state *x;
1239 xfrm_address_t *remote = daddr;
1240 xfrm_address_t *local = saddr;
1241 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1243 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1244 remote = &tmpl->id.daddr;
1245 local = &tmpl->saddr;
1246 family = tmpl->encap_family;
1247 if (xfrm_addr_any(local, family)) {
1248 error = xfrm_get_saddr(&tmp, remote, family);
1255 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1257 if (x && x->km.state == XFRM_STATE_VALID) {
1264 error = (x->km.state == XFRM_STATE_ERROR ?
1269 if (!tmpl->optional)
1275 for (nx--; nx>=0; nx--)
1276 xfrm_state_put(xfrm[nx]);
1281 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1282 struct xfrm_state **xfrm,
1283 unsigned short family)
1285 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1286 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1292 for (i = 0; i < npols; i++) {
1293 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1298 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1306 /* found states are sorted for outbound processing */
1308 xfrm_state_sort(xfrm, tpp, cnx, family);
1313 for (cnx--; cnx>=0; cnx--)
1314 xfrm_state_put(tpp[cnx]);
1319 /* Check that the bundle accepts the flow and its components are
1323 static struct dst_entry *
1324 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1326 struct dst_entry *x;
1327 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1328 if (unlikely(afinfo == NULL))
1329 return ERR_PTR(-EINVAL);
1330 x = afinfo->find_bundle(fl, policy);
1331 xfrm_policy_put_afinfo(afinfo);
1335 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1336 * all the metrics... Shortly, bundle a bundle.
1340 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1341 struct flowi *fl, struct dst_entry **dst_p,
1342 unsigned short family)
1345 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1346 if (unlikely(afinfo == NULL))
1348 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1349 xfrm_policy_put_afinfo(afinfo);
1354 xfrm_dst_alloc_copy(void **target, void *src, int size)
1357 *target = kmalloc(size, GFP_ATOMIC);
1361 memcpy(*target, src, size);
1366 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1368 #ifdef CONFIG_XFRM_SUB_POLICY
1369 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1370 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1378 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1380 #ifdef CONFIG_XFRM_SUB_POLICY
1381 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1382 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1388 static int stale_bundle(struct dst_entry *dst);
1390 /* Main function: finds/creates a bundle for given flow.
1392 * At the moment we eat a raw IP route. Mostly to speed up lookups
1393 * on interfaces with disabled IPsec.
1395 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1396 struct sock *sk, int flags)
1398 struct xfrm_policy *policy;
1399 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1404 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1405 struct dst_entry *dst, *dst_orig = *dst_p;
1410 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1413 genid = atomic_read(&flow_cache_genid);
1415 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1421 if (sk && sk->sk_policy[1]) {
1422 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1424 return PTR_ERR(policy);
1428 /* To accelerate a bit... */
1429 if ((dst_orig->flags & DST_NOXFRM) ||
1430 !xfrm_policy_count[XFRM_POLICY_OUT])
1433 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1434 dir, xfrm_policy_lookup);
1436 return PTR_ERR(policy);
1442 family = dst_orig->ops->family;
1443 policy->curlft.use_time = get_seconds();
1446 xfrm_nr += pols[0]->xfrm_nr;
1448 switch (policy->action) {
1449 case XFRM_POLICY_BLOCK:
1450 /* Prohibit the flow */
1454 case XFRM_POLICY_ALLOW:
1455 #ifndef CONFIG_XFRM_SUB_POLICY
1456 if (policy->xfrm_nr == 0) {
1457 /* Flow passes not transformed. */
1458 xfrm_pol_put(policy);
1463 /* Try to find matching bundle.
1465 * LATER: help from flow cache. It is optional, this
1466 * is required only for output policy.
1468 dst = xfrm_find_bundle(fl, policy, family);
1477 #ifdef CONFIG_XFRM_SUB_POLICY
1478 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1479 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1483 if (IS_ERR(pols[1])) {
1484 err = PTR_ERR(pols[1]);
1487 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1492 xfrm_nr += pols[1]->xfrm_nr;
1497 * Because neither flowi nor bundle information knows about
1498 * transformation template size. On more than one policy usage
1499 * we can realize whether all of them is bypass or not after
1500 * they are searched. See above not-transformed bypass
1501 * is surrounded by non-sub policy configuration, too.
1504 /* Flow passes not transformed. */
1505 xfrm_pols_put(pols, npols);
1510 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1512 if (unlikely(nx<0)) {
1514 if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1515 /* EREMOTE tells the caller to generate
1516 * a one-shot blackhole route.
1518 xfrm_pol_put(policy);
1521 if (err == -EAGAIN && flags) {
1522 DECLARE_WAITQUEUE(wait, current);
1524 add_wait_queue(&km_waitq, &wait);
1525 set_current_state(TASK_INTERRUPTIBLE);
1527 set_current_state(TASK_RUNNING);
1528 remove_wait_queue(&km_waitq, &wait);
1530 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1532 if (nx == -EAGAIN && signal_pending(current)) {
1536 if (nx == -EAGAIN ||
1537 genid != atomic_read(&flow_cache_genid)) {
1538 xfrm_pols_put(pols, npols);
1547 /* Flow passes not transformed. */
1548 xfrm_pols_put(pols, npols);
1553 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1555 if (unlikely(err)) {
1557 for (i=0; i<nx; i++)
1558 xfrm_state_put(xfrm[i]);
1562 for (pi = 0; pi < npols; pi++) {
1563 read_lock_bh(&pols[pi]->lock);
1564 pol_dead |= pols[pi]->dead;
1565 read_unlock_bh(&pols[pi]->lock);
1568 write_lock_bh(&policy->lock);
1569 if (unlikely(pol_dead || stale_bundle(dst))) {
1570 /* Wow! While we worked on resolving, this
1571 * policy has gone. Retry. It is not paranoia,
1572 * we just cannot enlist new bundle to dead object.
1573 * We can't enlist stable bundles either.
1575 write_unlock_bh(&policy->lock);
1579 err = -EHOSTUNREACH;
1584 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1586 err = xfrm_dst_update_origin(dst, fl);
1587 if (unlikely(err)) {
1588 write_unlock_bh(&policy->lock);
1594 dst->next = policy->bundles;
1595 policy->bundles = dst;
1597 write_unlock_bh(&policy->lock);
1600 dst_release(dst_orig);
1601 xfrm_pols_put(pols, npols);
1605 dst_release(dst_orig);
1606 xfrm_pols_put(pols, npols);
1610 EXPORT_SYMBOL(__xfrm_lookup);
1612 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1613 struct sock *sk, int flags)
1615 int err = __xfrm_lookup(dst_p, fl, sk, flags);
1617 if (err == -EREMOTE) {
1618 dst_release(*dst_p);
1625 EXPORT_SYMBOL(xfrm_lookup);
1628 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1630 struct xfrm_state *x;
1633 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1635 x = skb->sp->xvec[idx];
1636 if (!x->type->reject)
1639 err = x->type->reject(x, skb, fl);
1644 /* When skb is transformed back to its "native" form, we have to
1645 * check policy restrictions. At the moment we make this in maximally
1646 * stupid way. Shame on me. :-) Of course, connected sockets must
1647 * have policy cached at them.
1651 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1652 unsigned short family)
1654 if (xfrm_state_kern(x))
1655 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1656 return x->id.proto == tmpl->id.proto &&
1657 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1658 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1659 x->props.mode == tmpl->mode &&
1660 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1661 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1662 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1663 xfrm_state_addr_cmp(tmpl, x, family));
1667 * 0 or more than 0 is returned when validation is succeeded (either bypass
1668 * because of optional transport mode, or next index of the mathced secpath
1669 * state with the template.
1670 * -1 is returned when no matching template is found.
1671 * Otherwise "-2 - errored_index" is returned.
1674 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1675 unsigned short family)
1679 if (tmpl->optional) {
1680 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1684 for (; idx < sp->len; idx++) {
1685 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1687 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1697 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1699 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1702 if (unlikely(afinfo == NULL))
1703 return -EAFNOSUPPORT;
1705 afinfo->decode_session(skb, fl);
1706 err = security_xfrm_decode_session(skb, &fl->secid);
1707 xfrm_policy_put_afinfo(afinfo);
1710 EXPORT_SYMBOL(xfrm_decode_session);
1712 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1714 for (; k < sp->len; k++) {
1715 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1724 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1725 unsigned short family)
1727 struct xfrm_policy *pol;
1728 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1733 u8 fl_dir = policy_to_flow_dir(dir);
1736 if (xfrm_decode_session(skb, &fl, family) < 0)
1738 nf_nat_decode_session(skb, &fl, family);
1740 /* First, check used SA against their selectors. */
1744 for (i=skb->sp->len-1; i>=0; i--) {
1745 struct xfrm_state *x = skb->sp->xvec[i];
1746 if (!xfrm_selector_match(&x->sel, &fl, family))
1752 if (sk && sk->sk_policy[dir]) {
1753 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1759 pol = flow_cache_lookup(&fl, family, fl_dir,
1760 xfrm_policy_lookup);
1766 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1767 xfrm_secpath_reject(xerr_idx, skb, &fl);
1773 pol->curlft.use_time = get_seconds();
1777 #ifdef CONFIG_XFRM_SUB_POLICY
1778 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1779 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1783 if (IS_ERR(pols[1]))
1785 pols[1]->curlft.use_time = get_seconds();
1791 if (pol->action == XFRM_POLICY_ALLOW) {
1792 struct sec_path *sp;
1793 static struct sec_path dummy;
1794 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1795 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1796 struct xfrm_tmpl **tpp = tp;
1800 if ((sp = skb->sp) == NULL)
1803 for (pi = 0; pi < npols; pi++) {
1804 if (pols[pi] != pol &&
1805 pols[pi]->action != XFRM_POLICY_ALLOW)
1807 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1809 for (i = 0; i < pols[pi]->xfrm_nr; i++)
1810 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1814 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1818 /* For each tunnel xfrm, find the first matching tmpl.
1819 * For each tmpl before that, find corresponding xfrm.
1820 * Order is _important_. Later we will implement
1821 * some barriers, but at the moment barriers
1822 * are implied between each two transformations.
1824 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1825 k = xfrm_policy_ok(tpp[i], sp, k, family);
1828 /* "-2 - errored_index" returned */
1834 if (secpath_has_nontransport(sp, k, &xerr_idx))
1837 xfrm_pols_put(pols, npols);
1842 xfrm_secpath_reject(xerr_idx, skb, &fl);
1844 xfrm_pols_put(pols, npols);
1847 EXPORT_SYMBOL(__xfrm_policy_check);
1849 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1853 if (xfrm_decode_session(skb, &fl, family) < 0)
1856 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1858 EXPORT_SYMBOL(__xfrm_route_forward);
1860 /* Optimize later using cookies and generation ids. */
1862 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1864 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1865 * to "-1" to force all XFRM destinations to get validated by
1866 * dst_ops->check on every use. We do this because when a
1867 * normal route referenced by an XFRM dst is obsoleted we do
1868 * not go looking around for all parent referencing XFRM dsts
1869 * so that we can invalidate them. It is just too much work.
1870 * Instead we make the checks here on every use. For example:
1872 * XFRM dst A --> IPv4 dst X
1874 * X is the "xdst->route" of A (X is also the "dst->path" of A
1875 * in this example). If X is marked obsolete, "A" will not
1876 * notice. That's what we are validating here via the
1877 * stale_bundle() check.
1879 * When a policy's bundle is pruned, we dst_free() the XFRM
1880 * dst which causes it's ->obsolete field to be set to a
1881 * positive non-zero integer. If an XFRM dst has been pruned
1882 * like this, we want to force a new route lookup.
1884 if (dst->obsolete < 0 && !stale_bundle(dst))
1890 static int stale_bundle(struct dst_entry *dst)
1892 return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1895 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1897 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1898 dst->dev = &loopback_dev;
1899 dev_hold(&loopback_dev);
1903 EXPORT_SYMBOL(xfrm_dst_ifdown);
1905 static void xfrm_link_failure(struct sk_buff *skb)
1907 /* Impossible. Such dst must be popped before reaches point of failure. */
1911 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1914 if (dst->obsolete) {
1922 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1924 struct dst_entry *dst, **dstp;
1926 write_lock(&pol->lock);
1927 dstp = &pol->bundles;
1928 while ((dst=*dstp) != NULL) {
1931 dst->next = *gc_list_p;
1937 write_unlock(&pol->lock);
1940 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1942 struct dst_entry *gc_list = NULL;
1945 read_lock_bh(&xfrm_policy_lock);
1946 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1947 struct xfrm_policy *pol;
1948 struct hlist_node *entry;
1949 struct hlist_head *table;
1952 hlist_for_each_entry(pol, entry,
1953 &xfrm_policy_inexact[dir], bydst)
1954 prune_one_bundle(pol, func, &gc_list);
1956 table = xfrm_policy_bydst[dir].table;
1957 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1958 hlist_for_each_entry(pol, entry, table + i, bydst)
1959 prune_one_bundle(pol, func, &gc_list);
1962 read_unlock_bh(&xfrm_policy_lock);
1965 struct dst_entry *dst = gc_list;
1966 gc_list = dst->next;
1971 static int unused_bundle(struct dst_entry *dst)
1973 return !atomic_read(&dst->__refcnt);
1976 static void __xfrm_garbage_collect(void)
1978 xfrm_prune_bundles(unused_bundle);
1981 static int xfrm_flush_bundles(void)
1983 xfrm_prune_bundles(stale_bundle);
1987 void xfrm_init_pmtu(struct dst_entry *dst)
1990 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1991 u32 pmtu, route_mtu_cached;
1993 pmtu = dst_mtu(dst->child);
1994 xdst->child_mtu_cached = pmtu;
1996 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1998 route_mtu_cached = dst_mtu(xdst->route);
1999 xdst->route_mtu_cached = route_mtu_cached;
2001 if (pmtu > route_mtu_cached)
2002 pmtu = route_mtu_cached;
2004 dst->metrics[RTAX_MTU-1] = pmtu;
2005 } while ((dst = dst->next));
2008 EXPORT_SYMBOL(xfrm_init_pmtu);
2010 /* Check that the bundle accepts the flow and its components are
2014 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2015 struct flowi *fl, int family, int strict)
2017 struct dst_entry *dst = &first->u.dst;
2018 struct xfrm_dst *last;
2021 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2022 (dst->dev && !netif_running(dst->dev)))
2024 #ifdef CONFIG_XFRM_SUB_POLICY
2026 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2028 if (first->partner &&
2029 !xfrm_selector_match(first->partner, fl, family))
2037 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2039 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2042 !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2044 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2046 if (xdst->genid != dst->xfrm->genid)
2049 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
2050 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2053 mtu = dst_mtu(dst->child);
2054 if (xdst->child_mtu_cached != mtu) {
2056 xdst->child_mtu_cached = mtu;
2059 if (!dst_check(xdst->route, xdst->route_cookie))
2061 mtu = dst_mtu(xdst->route);
2062 if (xdst->route_mtu_cached != mtu) {
2064 xdst->route_mtu_cached = mtu;
2068 } while (dst->xfrm);
2073 mtu = last->child_mtu_cached;
2077 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2078 if (mtu > last->route_mtu_cached)
2079 mtu = last->route_mtu_cached;
2080 dst->metrics[RTAX_MTU-1] = mtu;
2085 last = last->u.next;
2086 last->child_mtu_cached = mtu;
2092 EXPORT_SYMBOL(xfrm_bundle_ok);
2094 #ifdef CONFIG_AUDITSYSCALL
2095 /* Audit addition and deletion of SAs and ipsec policy */
2097 void xfrm_audit_log(uid_t auid, u32 sid, int type, int result,
2098 struct xfrm_policy *xp, struct xfrm_state *x)
2103 struct xfrm_sec_ctx *sctx = NULL;
2104 struct audit_buffer *audit_buf;
2106 extern int audit_enabled;
2108 if (audit_enabled == 0)
2111 BUG_ON((type == AUDIT_MAC_IPSEC_ADDSA ||
2112 type == AUDIT_MAC_IPSEC_DELSA) && !x);
2113 BUG_ON((type == AUDIT_MAC_IPSEC_ADDSPD ||
2114 type == AUDIT_MAC_IPSEC_DELSPD) && !xp);
2116 audit_buf = audit_log_start(current->audit_context, GFP_ATOMIC, type);
2117 if (audit_buf == NULL)
2121 case AUDIT_MAC_IPSEC_ADDSA:
2122 audit_log_format(audit_buf, "SAD add: auid=%u", auid);
2124 case AUDIT_MAC_IPSEC_DELSA:
2125 audit_log_format(audit_buf, "SAD delete: auid=%u", auid);
2127 case AUDIT_MAC_IPSEC_ADDSPD:
2128 audit_log_format(audit_buf, "SPD add: auid=%u", auid);
2130 case AUDIT_MAC_IPSEC_DELSPD:
2131 audit_log_format(audit_buf, "SPD delete: auid=%u", auid);
2138 security_secid_to_secctx(sid, &secctx, &secctx_len) == 0)
2139 audit_log_format(audit_buf, " subj=%s", secctx);
2141 audit_log_task_context(audit_buf);
2144 family = xp->selector.family;
2146 sctx = xp->security;
2148 family = x->props.family;
2154 audit_log_format(audit_buf,
2155 " sec_alg=%u sec_doi=%u sec_obj=%s",
2156 sctx->ctx_alg, sctx->ctx_doi, sctx->ctx_str);
2161 struct in_addr saddr, daddr;
2163 saddr.s_addr = xp->selector.saddr.a4;
2164 daddr.s_addr = xp->selector.daddr.a4;
2166 saddr.s_addr = x->props.saddr.a4;
2167 daddr.s_addr = x->id.daddr.a4;
2169 audit_log_format(audit_buf,
2170 " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2171 NIPQUAD(saddr), NIPQUAD(daddr));
2176 struct in6_addr saddr6, daddr6;
2178 memcpy(&saddr6, xp->selector.saddr.a6,
2179 sizeof(struct in6_addr));
2180 memcpy(&daddr6, xp->selector.daddr.a6,
2181 sizeof(struct in6_addr));
2183 memcpy(&saddr6, x->props.saddr.a6,
2184 sizeof(struct in6_addr));
2185 memcpy(&daddr6, x->id.daddr.a6,
2186 sizeof(struct in6_addr));
2188 audit_log_format(audit_buf,
2189 " src=" NIP6_FMT " dst=" NIP6_FMT,
2190 NIP6(saddr6), NIP6(daddr6));
2196 audit_log_format(audit_buf, " spi=%lu(0x%lx) protocol=%s",
2197 (unsigned long)ntohl(x->id.spi),
2198 (unsigned long)ntohl(x->id.spi),
2199 x->id.proto == IPPROTO_AH ? "AH" :
2200 (x->id.proto == IPPROTO_ESP ?
2203 audit_log_format(audit_buf, " res=%u", result);
2204 audit_log_end(audit_buf);
2207 EXPORT_SYMBOL(xfrm_audit_log);
2208 #endif /* CONFIG_AUDITSYSCALL */
2210 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2213 if (unlikely(afinfo == NULL))
2215 if (unlikely(afinfo->family >= NPROTO))
2216 return -EAFNOSUPPORT;
2217 write_lock_bh(&xfrm_policy_afinfo_lock);
2218 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2221 struct dst_ops *dst_ops = afinfo->dst_ops;
2222 if (likely(dst_ops->kmem_cachep == NULL))
2223 dst_ops->kmem_cachep = xfrm_dst_cache;
2224 if (likely(dst_ops->check == NULL))
2225 dst_ops->check = xfrm_dst_check;
2226 if (likely(dst_ops->negative_advice == NULL))
2227 dst_ops->negative_advice = xfrm_negative_advice;
2228 if (likely(dst_ops->link_failure == NULL))
2229 dst_ops->link_failure = xfrm_link_failure;
2230 if (likely(afinfo->garbage_collect == NULL))
2231 afinfo->garbage_collect = __xfrm_garbage_collect;
2232 xfrm_policy_afinfo[afinfo->family] = afinfo;
2234 write_unlock_bh(&xfrm_policy_afinfo_lock);
2237 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2239 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2242 if (unlikely(afinfo == NULL))
2244 if (unlikely(afinfo->family >= NPROTO))
2245 return -EAFNOSUPPORT;
2246 write_lock_bh(&xfrm_policy_afinfo_lock);
2247 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2248 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2251 struct dst_ops *dst_ops = afinfo->dst_ops;
2252 xfrm_policy_afinfo[afinfo->family] = NULL;
2253 dst_ops->kmem_cachep = NULL;
2254 dst_ops->check = NULL;
2255 dst_ops->negative_advice = NULL;
2256 dst_ops->link_failure = NULL;
2257 afinfo->garbage_collect = NULL;
2260 write_unlock_bh(&xfrm_policy_afinfo_lock);
2263 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2265 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2267 struct xfrm_policy_afinfo *afinfo;
2268 if (unlikely(family >= NPROTO))
2270 read_lock(&xfrm_policy_afinfo_lock);
2271 afinfo = xfrm_policy_afinfo[family];
2272 if (unlikely(!afinfo))
2273 read_unlock(&xfrm_policy_afinfo_lock);
2277 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2279 read_unlock(&xfrm_policy_afinfo_lock);
2282 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2284 struct xfrm_policy_afinfo *afinfo;
2285 if (unlikely(family >= NPROTO))
2287 write_lock_bh(&xfrm_policy_afinfo_lock);
2288 afinfo = xfrm_policy_afinfo[family];
2289 if (unlikely(!afinfo))
2290 write_unlock_bh(&xfrm_policy_afinfo_lock);
2294 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2296 write_unlock_bh(&xfrm_policy_afinfo_lock);
2299 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2303 xfrm_flush_bundles();
2308 static struct notifier_block xfrm_dev_notifier = {
2314 static void __init xfrm_policy_init(void)
2316 unsigned int hmask, sz;
2319 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2320 sizeof(struct xfrm_dst),
2321 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2325 sz = (hmask+1) * sizeof(struct hlist_head);
2327 xfrm_policy_byidx = xfrm_hash_alloc(sz);
2328 xfrm_idx_hmask = hmask;
2329 if (!xfrm_policy_byidx)
2330 panic("XFRM: failed to allocate byidx hash\n");
2332 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2333 struct xfrm_policy_hash *htab;
2335 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2337 htab = &xfrm_policy_bydst[dir];
2338 htab->table = xfrm_hash_alloc(sz);
2339 htab->hmask = hmask;
2341 panic("XFRM: failed to allocate bydst hash\n");
2344 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2345 register_netdevice_notifier(&xfrm_dev_notifier);
2348 void __init xfrm_init(void)
2355 #ifdef CONFIG_XFRM_MIGRATE
2356 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2357 struct xfrm_selector *sel_tgt)
2359 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2360 if (sel_tgt->family == sel_cmp->family &&
2361 xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2362 sel_cmp->family) == 0 &&
2363 xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2364 sel_cmp->family) == 0 &&
2365 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2366 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2370 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2377 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2380 struct xfrm_policy *pol, *ret = NULL;
2381 struct hlist_node *entry;
2382 struct hlist_head *chain;
2385 read_lock_bh(&xfrm_policy_lock);
2386 chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2387 hlist_for_each_entry(pol, entry, chain, bydst) {
2388 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2389 pol->type == type) {
2391 priority = ret->priority;
2395 chain = &xfrm_policy_inexact[dir];
2396 hlist_for_each_entry(pol, entry, chain, bydst) {
2397 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2398 pol->type == type &&
2399 pol->priority < priority) {
2408 read_unlock_bh(&xfrm_policy_lock);
2413 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2417 if (t->mode == m->mode && t->id.proto == m->proto &&
2418 (m->reqid == 0 || t->reqid == m->reqid)) {
2420 case XFRM_MODE_TUNNEL:
2421 case XFRM_MODE_BEET:
2422 if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2423 m->old_family) == 0 &&
2424 xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2425 m->old_family) == 0) {
2429 case XFRM_MODE_TRANSPORT:
2430 /* in case of transport mode, template does not store
2431 any IP addresses, hence we just compare mode and
2442 /* update endpoint address(es) of template(s) */
2443 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2444 struct xfrm_migrate *m, int num_migrate)
2446 struct xfrm_migrate *mp;
2447 struct dst_entry *dst;
2450 write_lock_bh(&pol->lock);
2451 if (unlikely(pol->dead)) {
2452 /* target policy has been deleted */
2453 write_unlock_bh(&pol->lock);
2457 for (i = 0; i < pol->xfrm_nr; i++) {
2458 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2459 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2462 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL)
2464 /* update endpoints */
2465 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2466 sizeof(pol->xfrm_vec[i].id.daddr));
2467 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2468 sizeof(pol->xfrm_vec[i].saddr));
2469 pol->xfrm_vec[i].encap_family = mp->new_family;
2471 while ((dst = pol->bundles) != NULL) {
2472 pol->bundles = dst->next;
2478 write_unlock_bh(&pol->lock);
2486 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2490 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2493 for (i = 0; i < num_migrate; i++) {
2494 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2495 m[i].old_family) == 0) &&
2496 (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2497 m[i].old_family) == 0))
2499 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2500 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2503 /* check if there is any duplicated entry */
2504 for (j = i + 1; j < num_migrate; j++) {
2505 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2506 sizeof(m[i].old_daddr)) &&
2507 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2508 sizeof(m[i].old_saddr)) &&
2509 m[i].proto == m[j].proto &&
2510 m[i].mode == m[j].mode &&
2511 m[i].reqid == m[j].reqid &&
2512 m[i].old_family == m[j].old_family)
2520 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2521 struct xfrm_migrate *m, int num_migrate)
2523 int i, err, nx_cur = 0, nx_new = 0;
2524 struct xfrm_policy *pol = NULL;
2525 struct xfrm_state *x, *xc;
2526 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2527 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2528 struct xfrm_migrate *mp;
2530 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2533 /* Stage 1 - find policy */
2534 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2539 /* Stage 2 - find and update state(s) */
2540 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2541 if ((x = xfrm_migrate_state_find(mp))) {
2544 if ((xc = xfrm_state_migrate(x, mp))) {
2554 /* Stage 3 - update policy */
2555 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2558 /* Stage 4 - delete old state(s) */
2560 xfrm_states_put(x_cur, nx_cur);
2561 xfrm_states_delete(x_cur, nx_cur);
2564 /* Stage 5 - announce */
2565 km_migrate(sel, dir, type, m, num_migrate);
2577 xfrm_states_put(x_cur, nx_cur);
2579 xfrm_states_delete(x_new, nx_new);
2583 EXPORT_SYMBOL(xfrm_migrate);