uwb: document UWB and WUSB sysfs files
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / xfrm / xfrm_policy.c
blobb7754b1b73a459fee91885c36d74d1d38b3fc9da
1 /*
2 * xfrm_policy.c
4 * Changes:
5 * Mitsuru KANDA @USAGI
6 * Kazunori MIYAZAWA @USAGI
7 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8 * IPv6 support
9 * Kazunori MIYAZAWA @USAGI
10 * YOSHIFUJI Hideaki
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>
28 #include <net/dst.h>
29 #include <net/xfrm.h>
30 #include <net/ip.h>
31 #ifdef CONFIG_XFRM_STATISTICS
32 #include <net/snmp.h>
33 #endif
35 #include "xfrm_hash.h"
37 int sysctl_xfrm_larval_drop __read_mostly;
39 #ifdef CONFIG_XFRM_STATISTICS
40 DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
41 EXPORT_SYMBOL(xfrm_statistics);
42 #endif
44 DEFINE_MUTEX(xfrm_cfg_mutex);
45 EXPORT_SYMBOL(xfrm_cfg_mutex);
47 static DEFINE_RWLOCK(xfrm_policy_lock);
49 static struct list_head xfrm_policy_bytype[XFRM_POLICY_TYPE_MAX];
50 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
51 EXPORT_SYMBOL(xfrm_policy_count);
53 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
54 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
56 static struct kmem_cache *xfrm_dst_cache __read_mostly;
58 static struct work_struct xfrm_policy_gc_work;
59 static HLIST_HEAD(xfrm_policy_gc_list);
60 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
62 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
63 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
64 static void xfrm_init_pmtu(struct dst_entry *dst);
66 static inline int
67 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
69 return addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
70 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
71 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
72 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
73 (fl->proto == sel->proto || !sel->proto) &&
74 (fl->oif == sel->ifindex || !sel->ifindex);
77 static inline int
78 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
80 return addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
81 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
82 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
83 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
84 (fl->proto == sel->proto || !sel->proto) &&
85 (fl->oif == sel->ifindex || !sel->ifindex);
88 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
89 unsigned short family)
91 switch (family) {
92 case AF_INET:
93 return __xfrm4_selector_match(sel, fl);
94 case AF_INET6:
95 return __xfrm6_selector_match(sel, fl);
97 return 0;
100 static inline struct dst_entry *__xfrm_dst_lookup(int tos,
101 xfrm_address_t *saddr,
102 xfrm_address_t *daddr,
103 int family)
105 struct xfrm_policy_afinfo *afinfo;
106 struct dst_entry *dst;
108 afinfo = xfrm_policy_get_afinfo(family);
109 if (unlikely(afinfo == NULL))
110 return ERR_PTR(-EAFNOSUPPORT);
112 dst = afinfo->dst_lookup(tos, saddr, daddr);
114 xfrm_policy_put_afinfo(afinfo);
116 return dst;
119 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
120 xfrm_address_t *prev_saddr,
121 xfrm_address_t *prev_daddr,
122 int family)
124 xfrm_address_t *saddr = &x->props.saddr;
125 xfrm_address_t *daddr = &x->id.daddr;
126 struct dst_entry *dst;
128 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
129 saddr = x->coaddr;
130 daddr = prev_daddr;
132 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
133 saddr = prev_saddr;
134 daddr = x->coaddr;
137 dst = __xfrm_dst_lookup(tos, saddr, daddr, family);
139 if (!IS_ERR(dst)) {
140 if (prev_saddr != saddr)
141 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
142 if (prev_daddr != daddr)
143 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
146 return dst;
149 static inline unsigned long make_jiffies(long secs)
151 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
152 return MAX_SCHEDULE_TIMEOUT-1;
153 else
154 return secs*HZ;
157 static void xfrm_policy_timer(unsigned long data)
159 struct xfrm_policy *xp = (struct xfrm_policy*)data;
160 unsigned long now = get_seconds();
161 long next = LONG_MAX;
162 int warn = 0;
163 int dir;
165 read_lock(&xp->lock);
167 if (xp->dead)
168 goto out;
170 dir = xfrm_policy_id2dir(xp->index);
172 if (xp->lft.hard_add_expires_seconds) {
173 long tmo = xp->lft.hard_add_expires_seconds +
174 xp->curlft.add_time - now;
175 if (tmo <= 0)
176 goto expired;
177 if (tmo < next)
178 next = tmo;
180 if (xp->lft.hard_use_expires_seconds) {
181 long tmo = xp->lft.hard_use_expires_seconds +
182 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
183 if (tmo <= 0)
184 goto expired;
185 if (tmo < next)
186 next = tmo;
188 if (xp->lft.soft_add_expires_seconds) {
189 long tmo = xp->lft.soft_add_expires_seconds +
190 xp->curlft.add_time - now;
191 if (tmo <= 0) {
192 warn = 1;
193 tmo = XFRM_KM_TIMEOUT;
195 if (tmo < next)
196 next = tmo;
198 if (xp->lft.soft_use_expires_seconds) {
199 long tmo = xp->lft.soft_use_expires_seconds +
200 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
201 if (tmo <= 0) {
202 warn = 1;
203 tmo = XFRM_KM_TIMEOUT;
205 if (tmo < next)
206 next = tmo;
209 if (warn)
210 km_policy_expired(xp, dir, 0, 0);
211 if (next != LONG_MAX &&
212 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
213 xfrm_pol_hold(xp);
215 out:
216 read_unlock(&xp->lock);
217 xfrm_pol_put(xp);
218 return;
220 expired:
221 read_unlock(&xp->lock);
222 if (!xfrm_policy_delete(xp, dir))
223 km_policy_expired(xp, dir, 1, 0);
224 xfrm_pol_put(xp);
228 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
229 * SPD calls.
232 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
234 struct xfrm_policy *policy;
236 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
238 if (policy) {
239 INIT_LIST_HEAD(&policy->bytype);
240 INIT_HLIST_NODE(&policy->bydst);
241 INIT_HLIST_NODE(&policy->byidx);
242 rwlock_init(&policy->lock);
243 atomic_set(&policy->refcnt, 1);
244 setup_timer(&policy->timer, xfrm_policy_timer,
245 (unsigned long)policy);
247 return policy;
249 EXPORT_SYMBOL(xfrm_policy_alloc);
251 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
253 void xfrm_policy_destroy(struct xfrm_policy *policy)
255 BUG_ON(!policy->dead);
257 BUG_ON(policy->bundles);
259 if (del_timer(&policy->timer))
260 BUG();
262 write_lock_bh(&xfrm_policy_lock);
263 list_del(&policy->bytype);
264 write_unlock_bh(&xfrm_policy_lock);
266 security_xfrm_policy_free(policy->security);
267 kfree(policy);
269 EXPORT_SYMBOL(xfrm_policy_destroy);
271 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
273 struct dst_entry *dst;
275 while ((dst = policy->bundles) != NULL) {
276 policy->bundles = dst->next;
277 dst_free(dst);
280 if (del_timer(&policy->timer))
281 atomic_dec(&policy->refcnt);
283 if (atomic_read(&policy->refcnt) > 1)
284 flow_cache_flush();
286 xfrm_pol_put(policy);
289 static void xfrm_policy_gc_task(struct work_struct *work)
291 struct xfrm_policy *policy;
292 struct hlist_node *entry, *tmp;
293 struct hlist_head gc_list;
295 spin_lock_bh(&xfrm_policy_gc_lock);
296 gc_list.first = xfrm_policy_gc_list.first;
297 INIT_HLIST_HEAD(&xfrm_policy_gc_list);
298 spin_unlock_bh(&xfrm_policy_gc_lock);
300 hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
301 xfrm_policy_gc_kill(policy);
304 /* Rule must be locked. Release descentant resources, announce
305 * entry dead. The rule must be unlinked from lists to the moment.
308 static void xfrm_policy_kill(struct xfrm_policy *policy)
310 int dead;
312 write_lock_bh(&policy->lock);
313 dead = policy->dead;
314 policy->dead = 1;
315 write_unlock_bh(&policy->lock);
317 if (unlikely(dead)) {
318 WARN_ON(1);
319 return;
322 spin_lock(&xfrm_policy_gc_lock);
323 hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
324 spin_unlock(&xfrm_policy_gc_lock);
326 schedule_work(&xfrm_policy_gc_work);
329 struct xfrm_policy_hash {
330 struct hlist_head *table;
331 unsigned int hmask;
334 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
335 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
336 static struct hlist_head *xfrm_policy_byidx __read_mostly;
337 static unsigned int xfrm_idx_hmask __read_mostly;
338 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
340 static inline unsigned int idx_hash(u32 index)
342 return __idx_hash(index, xfrm_idx_hmask);
345 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
347 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
348 unsigned int hash = __sel_hash(sel, family, hmask);
350 return (hash == hmask + 1 ?
351 &xfrm_policy_inexact[dir] :
352 xfrm_policy_bydst[dir].table + hash);
355 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
357 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
358 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
360 return xfrm_policy_bydst[dir].table + hash;
363 static void xfrm_dst_hash_transfer(struct hlist_head *list,
364 struct hlist_head *ndsttable,
365 unsigned int nhashmask)
367 struct hlist_node *entry, *tmp, *entry0 = NULL;
368 struct xfrm_policy *pol;
369 unsigned int h0 = 0;
371 redo:
372 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
373 unsigned int h;
375 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
376 pol->family, nhashmask);
377 if (!entry0) {
378 hlist_del(entry);
379 hlist_add_head(&pol->bydst, ndsttable+h);
380 h0 = h;
381 } else {
382 if (h != h0)
383 continue;
384 hlist_del(entry);
385 hlist_add_after(entry0, &pol->bydst);
387 entry0 = entry;
389 if (!hlist_empty(list)) {
390 entry0 = NULL;
391 goto redo;
395 static void xfrm_idx_hash_transfer(struct hlist_head *list,
396 struct hlist_head *nidxtable,
397 unsigned int nhashmask)
399 struct hlist_node *entry, *tmp;
400 struct xfrm_policy *pol;
402 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
403 unsigned int h;
405 h = __idx_hash(pol->index, nhashmask);
406 hlist_add_head(&pol->byidx, nidxtable+h);
410 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
412 return ((old_hmask + 1) << 1) - 1;
415 static void xfrm_bydst_resize(int dir)
417 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
418 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
419 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
420 struct hlist_head *odst = xfrm_policy_bydst[dir].table;
421 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
422 int i;
424 if (!ndst)
425 return;
427 write_lock_bh(&xfrm_policy_lock);
429 for (i = hmask; i >= 0; i--)
430 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
432 xfrm_policy_bydst[dir].table = ndst;
433 xfrm_policy_bydst[dir].hmask = nhashmask;
435 write_unlock_bh(&xfrm_policy_lock);
437 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
440 static void xfrm_byidx_resize(int total)
442 unsigned int hmask = xfrm_idx_hmask;
443 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
444 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
445 struct hlist_head *oidx = xfrm_policy_byidx;
446 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
447 int i;
449 if (!nidx)
450 return;
452 write_lock_bh(&xfrm_policy_lock);
454 for (i = hmask; i >= 0; i--)
455 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
457 xfrm_policy_byidx = nidx;
458 xfrm_idx_hmask = nhashmask;
460 write_unlock_bh(&xfrm_policy_lock);
462 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
465 static inline int xfrm_bydst_should_resize(int dir, int *total)
467 unsigned int cnt = xfrm_policy_count[dir];
468 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
470 if (total)
471 *total += cnt;
473 if ((hmask + 1) < xfrm_policy_hashmax &&
474 cnt > hmask)
475 return 1;
477 return 0;
480 static inline int xfrm_byidx_should_resize(int total)
482 unsigned int hmask = xfrm_idx_hmask;
484 if ((hmask + 1) < xfrm_policy_hashmax &&
485 total > hmask)
486 return 1;
488 return 0;
491 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
493 read_lock_bh(&xfrm_policy_lock);
494 si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
495 si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
496 si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
497 si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
498 si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
499 si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
500 si->spdhcnt = xfrm_idx_hmask;
501 si->spdhmcnt = xfrm_policy_hashmax;
502 read_unlock_bh(&xfrm_policy_lock);
504 EXPORT_SYMBOL(xfrm_spd_getinfo);
506 static DEFINE_MUTEX(hash_resize_mutex);
507 static void xfrm_hash_resize(struct work_struct *__unused)
509 int dir, total;
511 mutex_lock(&hash_resize_mutex);
513 total = 0;
514 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
515 if (xfrm_bydst_should_resize(dir, &total))
516 xfrm_bydst_resize(dir);
518 if (xfrm_byidx_should_resize(total))
519 xfrm_byidx_resize(total);
521 mutex_unlock(&hash_resize_mutex);
524 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
526 /* Generate new index... KAME seems to generate them ordered by cost
527 * of an absolute inpredictability of ordering of rules. This will not pass. */
528 static u32 xfrm_gen_index(u8 type, int dir)
530 static u32 idx_generator;
532 for (;;) {
533 struct hlist_node *entry;
534 struct hlist_head *list;
535 struct xfrm_policy *p;
536 u32 idx;
537 int found;
539 idx = (idx_generator | dir);
540 idx_generator += 8;
541 if (idx == 0)
542 idx = 8;
543 list = xfrm_policy_byidx + idx_hash(idx);
544 found = 0;
545 hlist_for_each_entry(p, entry, list, byidx) {
546 if (p->index == idx) {
547 found = 1;
548 break;
551 if (!found)
552 return idx;
556 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
558 u32 *p1 = (u32 *) s1;
559 u32 *p2 = (u32 *) s2;
560 int len = sizeof(struct xfrm_selector) / sizeof(u32);
561 int i;
563 for (i = 0; i < len; i++) {
564 if (p1[i] != p2[i])
565 return 1;
568 return 0;
571 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
573 struct xfrm_policy *pol;
574 struct xfrm_policy *delpol;
575 struct hlist_head *chain;
576 struct hlist_node *entry, *newpos;
577 struct dst_entry *gc_list;
579 write_lock_bh(&xfrm_policy_lock);
580 chain = policy_hash_bysel(&policy->selector, policy->family, dir);
581 delpol = NULL;
582 newpos = NULL;
583 hlist_for_each_entry(pol, entry, chain, bydst) {
584 if (pol->type == policy->type &&
585 !selector_cmp(&pol->selector, &policy->selector) &&
586 xfrm_sec_ctx_match(pol->security, policy->security) &&
587 !WARN_ON(delpol)) {
588 if (excl) {
589 write_unlock_bh(&xfrm_policy_lock);
590 return -EEXIST;
592 delpol = pol;
593 if (policy->priority > pol->priority)
594 continue;
595 } else if (policy->priority >= pol->priority) {
596 newpos = &pol->bydst;
597 continue;
599 if (delpol)
600 break;
602 if (newpos)
603 hlist_add_after(newpos, &policy->bydst);
604 else
605 hlist_add_head(&policy->bydst, chain);
606 xfrm_pol_hold(policy);
607 xfrm_policy_count[dir]++;
608 atomic_inc(&flow_cache_genid);
609 if (delpol) {
610 hlist_del(&delpol->bydst);
611 hlist_del(&delpol->byidx);
612 xfrm_policy_count[dir]--;
614 policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
615 hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
616 policy->curlft.add_time = get_seconds();
617 policy->curlft.use_time = 0;
618 if (!mod_timer(&policy->timer, jiffies + HZ))
619 xfrm_pol_hold(policy);
620 list_add_tail(&policy->bytype, &xfrm_policy_bytype[policy->type]);
621 write_unlock_bh(&xfrm_policy_lock);
623 if (delpol)
624 xfrm_policy_kill(delpol);
625 else if (xfrm_bydst_should_resize(dir, NULL))
626 schedule_work(&xfrm_hash_work);
628 read_lock_bh(&xfrm_policy_lock);
629 gc_list = NULL;
630 entry = &policy->bydst;
631 hlist_for_each_entry_continue(policy, entry, bydst) {
632 struct dst_entry *dst;
634 write_lock(&policy->lock);
635 dst = policy->bundles;
636 if (dst) {
637 struct dst_entry *tail = dst;
638 while (tail->next)
639 tail = tail->next;
640 tail->next = gc_list;
641 gc_list = dst;
643 policy->bundles = NULL;
645 write_unlock(&policy->lock);
647 read_unlock_bh(&xfrm_policy_lock);
649 while (gc_list) {
650 struct dst_entry *dst = gc_list;
652 gc_list = dst->next;
653 dst_free(dst);
656 return 0;
658 EXPORT_SYMBOL(xfrm_policy_insert);
660 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
661 struct xfrm_selector *sel,
662 struct xfrm_sec_ctx *ctx, int delete,
663 int *err)
665 struct xfrm_policy *pol, *ret;
666 struct hlist_head *chain;
667 struct hlist_node *entry;
669 *err = 0;
670 write_lock_bh(&xfrm_policy_lock);
671 chain = policy_hash_bysel(sel, sel->family, dir);
672 ret = NULL;
673 hlist_for_each_entry(pol, entry, chain, bydst) {
674 if (pol->type == type &&
675 !selector_cmp(sel, &pol->selector) &&
676 xfrm_sec_ctx_match(ctx, pol->security)) {
677 xfrm_pol_hold(pol);
678 if (delete) {
679 *err = security_xfrm_policy_delete(
680 pol->security);
681 if (*err) {
682 write_unlock_bh(&xfrm_policy_lock);
683 return pol;
685 hlist_del(&pol->bydst);
686 hlist_del(&pol->byidx);
687 xfrm_policy_count[dir]--;
689 ret = pol;
690 break;
693 write_unlock_bh(&xfrm_policy_lock);
695 if (ret && delete) {
696 atomic_inc(&flow_cache_genid);
697 xfrm_policy_kill(ret);
699 return ret;
701 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
703 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
704 int *err)
706 struct xfrm_policy *pol, *ret;
707 struct hlist_head *chain;
708 struct hlist_node *entry;
710 *err = -ENOENT;
711 if (xfrm_policy_id2dir(id) != dir)
712 return NULL;
714 *err = 0;
715 write_lock_bh(&xfrm_policy_lock);
716 chain = xfrm_policy_byidx + idx_hash(id);
717 ret = NULL;
718 hlist_for_each_entry(pol, entry, chain, byidx) {
719 if (pol->type == type && pol->index == id) {
720 xfrm_pol_hold(pol);
721 if (delete) {
722 *err = security_xfrm_policy_delete(
723 pol->security);
724 if (*err) {
725 write_unlock_bh(&xfrm_policy_lock);
726 return pol;
728 hlist_del(&pol->bydst);
729 hlist_del(&pol->byidx);
730 xfrm_policy_count[dir]--;
732 ret = pol;
733 break;
736 write_unlock_bh(&xfrm_policy_lock);
738 if (ret && delete) {
739 atomic_inc(&flow_cache_genid);
740 xfrm_policy_kill(ret);
742 return ret;
744 EXPORT_SYMBOL(xfrm_policy_byid);
746 #ifdef CONFIG_SECURITY_NETWORK_XFRM
747 static inline int
748 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
750 int dir, err = 0;
752 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
753 struct xfrm_policy *pol;
754 struct hlist_node *entry;
755 int i;
757 hlist_for_each_entry(pol, entry,
758 &xfrm_policy_inexact[dir], bydst) {
759 if (pol->type != type)
760 continue;
761 err = security_xfrm_policy_delete(pol->security);
762 if (err) {
763 xfrm_audit_policy_delete(pol, 0,
764 audit_info->loginuid,
765 audit_info->sessionid,
766 audit_info->secid);
767 return err;
770 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
771 hlist_for_each_entry(pol, entry,
772 xfrm_policy_bydst[dir].table + i,
773 bydst) {
774 if (pol->type != type)
775 continue;
776 err = security_xfrm_policy_delete(
777 pol->security);
778 if (err) {
779 xfrm_audit_policy_delete(pol, 0,
780 audit_info->loginuid,
781 audit_info->sessionid,
782 audit_info->secid);
783 return err;
788 return err;
790 #else
791 static inline int
792 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
794 return 0;
796 #endif
798 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
800 int dir, err = 0;
802 write_lock_bh(&xfrm_policy_lock);
804 err = xfrm_policy_flush_secctx_check(type, audit_info);
805 if (err)
806 goto out;
808 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
809 struct xfrm_policy *pol;
810 struct hlist_node *entry;
811 int i, killed;
813 killed = 0;
814 again1:
815 hlist_for_each_entry(pol, entry,
816 &xfrm_policy_inexact[dir], bydst) {
817 if (pol->type != type)
818 continue;
819 hlist_del(&pol->bydst);
820 hlist_del(&pol->byidx);
821 write_unlock_bh(&xfrm_policy_lock);
823 xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
824 audit_info->sessionid,
825 audit_info->secid);
827 xfrm_policy_kill(pol);
828 killed++;
830 write_lock_bh(&xfrm_policy_lock);
831 goto again1;
834 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
835 again2:
836 hlist_for_each_entry(pol, entry,
837 xfrm_policy_bydst[dir].table + i,
838 bydst) {
839 if (pol->type != type)
840 continue;
841 hlist_del(&pol->bydst);
842 hlist_del(&pol->byidx);
843 write_unlock_bh(&xfrm_policy_lock);
845 xfrm_audit_policy_delete(pol, 1,
846 audit_info->loginuid,
847 audit_info->sessionid,
848 audit_info->secid);
849 xfrm_policy_kill(pol);
850 killed++;
852 write_lock_bh(&xfrm_policy_lock);
853 goto again2;
857 xfrm_policy_count[dir] -= killed;
859 atomic_inc(&flow_cache_genid);
860 out:
861 write_unlock_bh(&xfrm_policy_lock);
862 return err;
864 EXPORT_SYMBOL(xfrm_policy_flush);
866 int xfrm_policy_walk(struct xfrm_policy_walk *walk,
867 int (*func)(struct xfrm_policy *, int, int, void*),
868 void *data)
870 struct xfrm_policy *old, *pol, *last = NULL;
871 int error = 0;
873 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
874 walk->type != XFRM_POLICY_TYPE_ANY)
875 return -EINVAL;
877 if (walk->policy == NULL && walk->count != 0)
878 return 0;
880 old = pol = walk->policy;
881 walk->policy = NULL;
882 read_lock_bh(&xfrm_policy_lock);
884 for (; walk->cur_type < XFRM_POLICY_TYPE_MAX; walk->cur_type++) {
885 if (walk->type != walk->cur_type &&
886 walk->type != XFRM_POLICY_TYPE_ANY)
887 continue;
889 if (pol == NULL) {
890 pol = list_first_entry(&xfrm_policy_bytype[walk->cur_type],
891 struct xfrm_policy, bytype);
893 list_for_each_entry_from(pol, &xfrm_policy_bytype[walk->cur_type], bytype) {
894 if (pol->dead)
895 continue;
896 if (last) {
897 error = func(last, xfrm_policy_id2dir(last->index),
898 walk->count, data);
899 if (error) {
900 xfrm_pol_hold(last);
901 walk->policy = last;
902 goto out;
905 last = pol;
906 walk->count++;
908 pol = NULL;
910 if (walk->count == 0) {
911 error = -ENOENT;
912 goto out;
914 if (last)
915 error = func(last, xfrm_policy_id2dir(last->index), 0, data);
916 out:
917 read_unlock_bh(&xfrm_policy_lock);
918 if (old != NULL)
919 xfrm_pol_put(old);
920 return error;
922 EXPORT_SYMBOL(xfrm_policy_walk);
925 * Find policy to apply to this flow.
927 * Returns 0 if policy found, else an -errno.
929 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
930 u8 type, u16 family, int dir)
932 struct xfrm_selector *sel = &pol->selector;
933 int match, ret = -ESRCH;
935 if (pol->family != family ||
936 pol->type != type)
937 return ret;
939 match = xfrm_selector_match(sel, fl, family);
940 if (match)
941 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
942 dir);
944 return ret;
947 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
948 u16 family, u8 dir)
950 int err;
951 struct xfrm_policy *pol, *ret;
952 xfrm_address_t *daddr, *saddr;
953 struct hlist_node *entry;
954 struct hlist_head *chain;
955 u32 priority = ~0U;
957 daddr = xfrm_flowi_daddr(fl, family);
958 saddr = xfrm_flowi_saddr(fl, family);
959 if (unlikely(!daddr || !saddr))
960 return NULL;
962 read_lock_bh(&xfrm_policy_lock);
963 chain = policy_hash_direct(daddr, saddr, family, dir);
964 ret = NULL;
965 hlist_for_each_entry(pol, entry, chain, bydst) {
966 err = xfrm_policy_match(pol, fl, type, family, dir);
967 if (err) {
968 if (err == -ESRCH)
969 continue;
970 else {
971 ret = ERR_PTR(err);
972 goto fail;
974 } else {
975 ret = pol;
976 priority = ret->priority;
977 break;
980 chain = &xfrm_policy_inexact[dir];
981 hlist_for_each_entry(pol, entry, chain, bydst) {
982 err = xfrm_policy_match(pol, fl, type, family, dir);
983 if (err) {
984 if (err == -ESRCH)
985 continue;
986 else {
987 ret = ERR_PTR(err);
988 goto fail;
990 } else if (pol->priority < priority) {
991 ret = pol;
992 break;
995 if (ret)
996 xfrm_pol_hold(ret);
997 fail:
998 read_unlock_bh(&xfrm_policy_lock);
1000 return ret;
1003 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1004 void **objp, atomic_t **obj_refp)
1006 struct xfrm_policy *pol;
1007 int err = 0;
1009 #ifdef CONFIG_XFRM_SUB_POLICY
1010 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1011 if (IS_ERR(pol)) {
1012 err = PTR_ERR(pol);
1013 pol = NULL;
1015 if (pol || err)
1016 goto end;
1017 #endif
1018 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1019 if (IS_ERR(pol)) {
1020 err = PTR_ERR(pol);
1021 pol = NULL;
1023 #ifdef CONFIG_XFRM_SUB_POLICY
1024 end:
1025 #endif
1026 if ((*objp = (void *) pol) != NULL)
1027 *obj_refp = &pol->refcnt;
1028 return err;
1031 static inline int policy_to_flow_dir(int dir)
1033 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1034 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1035 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1036 return dir;
1037 switch (dir) {
1038 default:
1039 case XFRM_POLICY_IN:
1040 return FLOW_DIR_IN;
1041 case XFRM_POLICY_OUT:
1042 return FLOW_DIR_OUT;
1043 case XFRM_POLICY_FWD:
1044 return FLOW_DIR_FWD;
1048 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1050 struct xfrm_policy *pol;
1052 read_lock_bh(&xfrm_policy_lock);
1053 if ((pol = sk->sk_policy[dir]) != NULL) {
1054 int match = xfrm_selector_match(&pol->selector, fl,
1055 sk->sk_family);
1056 int err = 0;
1058 if (match) {
1059 err = security_xfrm_policy_lookup(pol->security,
1060 fl->secid,
1061 policy_to_flow_dir(dir));
1062 if (!err)
1063 xfrm_pol_hold(pol);
1064 else if (err == -ESRCH)
1065 pol = NULL;
1066 else
1067 pol = ERR_PTR(err);
1068 } else
1069 pol = NULL;
1071 read_unlock_bh(&xfrm_policy_lock);
1072 return pol;
1075 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1077 struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1078 pol->family, dir);
1080 list_add_tail(&pol->bytype, &xfrm_policy_bytype[pol->type]);
1081 hlist_add_head(&pol->bydst, chain);
1082 hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1083 xfrm_policy_count[dir]++;
1084 xfrm_pol_hold(pol);
1086 if (xfrm_bydst_should_resize(dir, NULL))
1087 schedule_work(&xfrm_hash_work);
1090 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1091 int dir)
1093 if (hlist_unhashed(&pol->bydst))
1094 return NULL;
1096 hlist_del(&pol->bydst);
1097 hlist_del(&pol->byidx);
1098 xfrm_policy_count[dir]--;
1100 return pol;
1103 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1105 write_lock_bh(&xfrm_policy_lock);
1106 pol = __xfrm_policy_unlink(pol, dir);
1107 write_unlock_bh(&xfrm_policy_lock);
1108 if (pol) {
1109 if (dir < XFRM_POLICY_MAX)
1110 atomic_inc(&flow_cache_genid);
1111 xfrm_policy_kill(pol);
1112 return 0;
1114 return -ENOENT;
1116 EXPORT_SYMBOL(xfrm_policy_delete);
1118 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1120 struct xfrm_policy *old_pol;
1122 #ifdef CONFIG_XFRM_SUB_POLICY
1123 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1124 return -EINVAL;
1125 #endif
1127 write_lock_bh(&xfrm_policy_lock);
1128 old_pol = sk->sk_policy[dir];
1129 sk->sk_policy[dir] = pol;
1130 if (pol) {
1131 pol->curlft.add_time = get_seconds();
1132 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1133 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1135 if (old_pol)
1136 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1137 write_unlock_bh(&xfrm_policy_lock);
1139 if (old_pol) {
1140 xfrm_policy_kill(old_pol);
1142 return 0;
1145 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1147 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1149 if (newp) {
1150 newp->selector = old->selector;
1151 if (security_xfrm_policy_clone(old->security,
1152 &newp->security)) {
1153 kfree(newp);
1154 return NULL; /* ENOMEM */
1156 newp->lft = old->lft;
1157 newp->curlft = old->curlft;
1158 newp->action = old->action;
1159 newp->flags = old->flags;
1160 newp->xfrm_nr = old->xfrm_nr;
1161 newp->index = old->index;
1162 newp->type = old->type;
1163 memcpy(newp->xfrm_vec, old->xfrm_vec,
1164 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1165 write_lock_bh(&xfrm_policy_lock);
1166 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1167 write_unlock_bh(&xfrm_policy_lock);
1168 xfrm_pol_put(newp);
1170 return newp;
1173 int __xfrm_sk_clone_policy(struct sock *sk)
1175 struct xfrm_policy *p0 = sk->sk_policy[0],
1176 *p1 = sk->sk_policy[1];
1178 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1179 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1180 return -ENOMEM;
1181 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1182 return -ENOMEM;
1183 return 0;
1186 static int
1187 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1188 unsigned short family)
1190 int err;
1191 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1193 if (unlikely(afinfo == NULL))
1194 return -EINVAL;
1195 err = afinfo->get_saddr(local, remote);
1196 xfrm_policy_put_afinfo(afinfo);
1197 return err;
1200 /* Resolve list of templates for the flow, given policy. */
1202 static int
1203 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1204 struct xfrm_state **xfrm,
1205 unsigned short family)
1207 int nx;
1208 int i, error;
1209 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1210 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1211 xfrm_address_t tmp;
1213 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1214 struct xfrm_state *x;
1215 xfrm_address_t *remote = daddr;
1216 xfrm_address_t *local = saddr;
1217 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1219 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1220 tmpl->mode == XFRM_MODE_BEET) {
1221 remote = &tmpl->id.daddr;
1222 local = &tmpl->saddr;
1223 family = tmpl->encap_family;
1224 if (xfrm_addr_any(local, family)) {
1225 error = xfrm_get_saddr(&tmp, remote, family);
1226 if (error)
1227 goto fail;
1228 local = &tmp;
1232 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1234 if (x && x->km.state == XFRM_STATE_VALID) {
1235 xfrm[nx++] = x;
1236 daddr = remote;
1237 saddr = local;
1238 continue;
1240 if (x) {
1241 error = (x->km.state == XFRM_STATE_ERROR ?
1242 -EINVAL : -EAGAIN);
1243 xfrm_state_put(x);
1246 if (!tmpl->optional)
1247 goto fail;
1249 return nx;
1251 fail:
1252 for (nx--; nx>=0; nx--)
1253 xfrm_state_put(xfrm[nx]);
1254 return error;
1257 static int
1258 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1259 struct xfrm_state **xfrm,
1260 unsigned short family)
1262 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1263 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1264 int cnx = 0;
1265 int error;
1266 int ret;
1267 int i;
1269 for (i = 0; i < npols; i++) {
1270 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1271 error = -ENOBUFS;
1272 goto fail;
1275 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1276 if (ret < 0) {
1277 error = ret;
1278 goto fail;
1279 } else
1280 cnx += ret;
1283 /* found states are sorted for outbound processing */
1284 if (npols > 1)
1285 xfrm_state_sort(xfrm, tpp, cnx, family);
1287 return cnx;
1289 fail:
1290 for (cnx--; cnx>=0; cnx--)
1291 xfrm_state_put(tpp[cnx]);
1292 return error;
1296 /* Check that the bundle accepts the flow and its components are
1297 * still valid.
1300 static struct dst_entry *
1301 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1303 struct dst_entry *x;
1304 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1305 if (unlikely(afinfo == NULL))
1306 return ERR_PTR(-EINVAL);
1307 x = afinfo->find_bundle(fl, policy);
1308 xfrm_policy_put_afinfo(afinfo);
1309 return x;
1312 static inline int xfrm_get_tos(struct flowi *fl, int family)
1314 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1315 int tos;
1317 if (!afinfo)
1318 return -EINVAL;
1320 tos = afinfo->get_tos(fl);
1322 xfrm_policy_put_afinfo(afinfo);
1324 return tos;
1327 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1329 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1330 struct xfrm_dst *xdst;
1332 if (!afinfo)
1333 return ERR_PTR(-EINVAL);
1335 xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1337 xfrm_policy_put_afinfo(afinfo);
1339 return xdst;
1342 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1343 int nfheader_len)
1345 struct xfrm_policy_afinfo *afinfo =
1346 xfrm_policy_get_afinfo(dst->ops->family);
1347 int err;
1349 if (!afinfo)
1350 return -EINVAL;
1352 err = afinfo->init_path(path, dst, nfheader_len);
1354 xfrm_policy_put_afinfo(afinfo);
1356 return err;
1359 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1361 struct xfrm_policy_afinfo *afinfo =
1362 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1363 int err;
1365 if (!afinfo)
1366 return -EINVAL;
1368 err = afinfo->fill_dst(xdst, dev);
1370 xfrm_policy_put_afinfo(afinfo);
1372 return err;
1375 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1376 * all the metrics... Shortly, bundle a bundle.
1379 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1380 struct xfrm_state **xfrm, int nx,
1381 struct flowi *fl,
1382 struct dst_entry *dst)
1384 unsigned long now = jiffies;
1385 struct net_device *dev;
1386 struct dst_entry *dst_prev = NULL;
1387 struct dst_entry *dst0 = NULL;
1388 int i = 0;
1389 int err;
1390 int header_len = 0;
1391 int nfheader_len = 0;
1392 int trailer_len = 0;
1393 int tos;
1394 int family = policy->selector.family;
1395 xfrm_address_t saddr, daddr;
1397 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1399 tos = xfrm_get_tos(fl, family);
1400 err = tos;
1401 if (tos < 0)
1402 goto put_states;
1404 dst_hold(dst);
1406 for (; i < nx; i++) {
1407 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1408 struct dst_entry *dst1 = &xdst->u.dst;
1410 err = PTR_ERR(xdst);
1411 if (IS_ERR(xdst)) {
1412 dst_release(dst);
1413 goto put_states;
1416 if (!dst_prev)
1417 dst0 = dst1;
1418 else {
1419 dst_prev->child = dst_clone(dst1);
1420 dst1->flags |= DST_NOHASH;
1423 xdst->route = dst;
1424 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1426 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1427 family = xfrm[i]->props.family;
1428 dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1429 family);
1430 err = PTR_ERR(dst);
1431 if (IS_ERR(dst))
1432 goto put_states;
1433 } else
1434 dst_hold(dst);
1436 dst1->xfrm = xfrm[i];
1437 xdst->genid = xfrm[i]->genid;
1439 dst1->obsolete = -1;
1440 dst1->flags |= DST_HOST;
1441 dst1->lastuse = now;
1443 dst1->input = dst_discard;
1444 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1446 dst1->next = dst_prev;
1447 dst_prev = dst1;
1449 header_len += xfrm[i]->props.header_len;
1450 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1451 nfheader_len += xfrm[i]->props.header_len;
1452 trailer_len += xfrm[i]->props.trailer_len;
1455 dst_prev->child = dst;
1456 dst0->path = dst;
1458 err = -ENODEV;
1459 dev = dst->dev;
1460 if (!dev)
1461 goto free_dst;
1463 /* Copy neighbout for reachability confirmation */
1464 dst0->neighbour = neigh_clone(dst->neighbour);
1466 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1467 xfrm_init_pmtu(dst_prev);
1469 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1470 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1472 err = xfrm_fill_dst(xdst, dev);
1473 if (err)
1474 goto free_dst;
1476 dst_prev->header_len = header_len;
1477 dst_prev->trailer_len = trailer_len;
1478 header_len -= xdst->u.dst.xfrm->props.header_len;
1479 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1482 out:
1483 return dst0;
1485 put_states:
1486 for (; i < nx; i++)
1487 xfrm_state_put(xfrm[i]);
1488 free_dst:
1489 if (dst0)
1490 dst_free(dst0);
1491 dst0 = ERR_PTR(err);
1492 goto out;
1495 static int inline
1496 xfrm_dst_alloc_copy(void **target, void *src, int size)
1498 if (!*target) {
1499 *target = kmalloc(size, GFP_ATOMIC);
1500 if (!*target)
1501 return -ENOMEM;
1503 memcpy(*target, src, size);
1504 return 0;
1507 static int inline
1508 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1510 #ifdef CONFIG_XFRM_SUB_POLICY
1511 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1512 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1513 sel, sizeof(*sel));
1514 #else
1515 return 0;
1516 #endif
1519 static int inline
1520 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1522 #ifdef CONFIG_XFRM_SUB_POLICY
1523 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1524 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1525 #else
1526 return 0;
1527 #endif
1530 static int stale_bundle(struct dst_entry *dst);
1532 /* Main function: finds/creates a bundle for given flow.
1534 * At the moment we eat a raw IP route. Mostly to speed up lookups
1535 * on interfaces with disabled IPsec.
1537 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1538 struct sock *sk, int flags)
1540 struct xfrm_policy *policy;
1541 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1542 int npols;
1543 int pol_dead;
1544 int xfrm_nr;
1545 int pi;
1546 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1547 struct dst_entry *dst, *dst_orig = *dst_p;
1548 int nx = 0;
1549 int err;
1550 u32 genid;
1551 u16 family;
1552 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1554 restart:
1555 genid = atomic_read(&flow_cache_genid);
1556 policy = NULL;
1557 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1558 pols[pi] = NULL;
1559 npols = 0;
1560 pol_dead = 0;
1561 xfrm_nr = 0;
1563 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1564 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1565 err = PTR_ERR(policy);
1566 if (IS_ERR(policy)) {
1567 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1568 goto dropdst;
1572 if (!policy) {
1573 /* To accelerate a bit... */
1574 if ((dst_orig->flags & DST_NOXFRM) ||
1575 !xfrm_policy_count[XFRM_POLICY_OUT])
1576 goto nopol;
1578 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1579 dir, xfrm_policy_lookup);
1580 err = PTR_ERR(policy);
1581 if (IS_ERR(policy)) {
1582 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1583 goto dropdst;
1587 if (!policy)
1588 goto nopol;
1590 family = dst_orig->ops->family;
1591 pols[0] = policy;
1592 npols ++;
1593 xfrm_nr += pols[0]->xfrm_nr;
1595 err = -ENOENT;
1596 if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1597 goto error;
1599 policy->curlft.use_time = get_seconds();
1601 switch (policy->action) {
1602 default:
1603 case XFRM_POLICY_BLOCK:
1604 /* Prohibit the flow */
1605 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1606 err = -EPERM;
1607 goto error;
1609 case XFRM_POLICY_ALLOW:
1610 #ifndef CONFIG_XFRM_SUB_POLICY
1611 if (policy->xfrm_nr == 0) {
1612 /* Flow passes not transformed. */
1613 xfrm_pol_put(policy);
1614 return 0;
1616 #endif
1618 /* Try to find matching bundle.
1620 * LATER: help from flow cache. It is optional, this
1621 * is required only for output policy.
1623 dst = xfrm_find_bundle(fl, policy, family);
1624 if (IS_ERR(dst)) {
1625 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1626 err = PTR_ERR(dst);
1627 goto error;
1630 if (dst)
1631 break;
1633 #ifdef CONFIG_XFRM_SUB_POLICY
1634 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1635 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1636 fl, family,
1637 XFRM_POLICY_OUT);
1638 if (pols[1]) {
1639 if (IS_ERR(pols[1])) {
1640 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1641 err = PTR_ERR(pols[1]);
1642 goto error;
1644 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1645 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1646 err = -EPERM;
1647 goto error;
1649 npols ++;
1650 xfrm_nr += pols[1]->xfrm_nr;
1655 * Because neither flowi nor bundle information knows about
1656 * transformation template size. On more than one policy usage
1657 * we can realize whether all of them is bypass or not after
1658 * they are searched. See above not-transformed bypass
1659 * is surrounded by non-sub policy configuration, too.
1661 if (xfrm_nr == 0) {
1662 /* Flow passes not transformed. */
1663 xfrm_pols_put(pols, npols);
1664 return 0;
1667 #endif
1668 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1670 if (unlikely(nx<0)) {
1671 err = nx;
1672 if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1673 /* EREMOTE tells the caller to generate
1674 * a one-shot blackhole route.
1676 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1677 xfrm_pol_put(policy);
1678 return -EREMOTE;
1680 if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1681 DECLARE_WAITQUEUE(wait, current);
1683 add_wait_queue(&km_waitq, &wait);
1684 set_current_state(TASK_INTERRUPTIBLE);
1685 schedule();
1686 set_current_state(TASK_RUNNING);
1687 remove_wait_queue(&km_waitq, &wait);
1689 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1691 if (nx == -EAGAIN && signal_pending(current)) {
1692 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1693 err = -ERESTART;
1694 goto error;
1696 if (nx == -EAGAIN ||
1697 genid != atomic_read(&flow_cache_genid)) {
1698 xfrm_pols_put(pols, npols);
1699 goto restart;
1701 err = nx;
1703 if (err < 0) {
1704 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1705 goto error;
1708 if (nx == 0) {
1709 /* Flow passes not transformed. */
1710 xfrm_pols_put(pols, npols);
1711 return 0;
1714 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1715 err = PTR_ERR(dst);
1716 if (IS_ERR(dst)) {
1717 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1718 goto error;
1721 for (pi = 0; pi < npols; pi++) {
1722 read_lock_bh(&pols[pi]->lock);
1723 pol_dead |= pols[pi]->dead;
1724 read_unlock_bh(&pols[pi]->lock);
1727 write_lock_bh(&policy->lock);
1728 if (unlikely(pol_dead || stale_bundle(dst))) {
1729 /* Wow! While we worked on resolving, this
1730 * policy has gone. Retry. It is not paranoia,
1731 * we just cannot enlist new bundle to dead object.
1732 * We can't enlist stable bundles either.
1734 write_unlock_bh(&policy->lock);
1735 dst_free(dst);
1737 if (pol_dead)
1738 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1739 else
1740 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1741 err = -EHOSTUNREACH;
1742 goto error;
1745 if (npols > 1)
1746 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1747 else
1748 err = xfrm_dst_update_origin(dst, fl);
1749 if (unlikely(err)) {
1750 write_unlock_bh(&policy->lock);
1751 dst_free(dst);
1752 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1753 goto error;
1756 dst->next = policy->bundles;
1757 policy->bundles = dst;
1758 dst_hold(dst);
1759 write_unlock_bh(&policy->lock);
1761 *dst_p = dst;
1762 dst_release(dst_orig);
1763 xfrm_pols_put(pols, npols);
1764 return 0;
1766 error:
1767 xfrm_pols_put(pols, npols);
1768 dropdst:
1769 dst_release(dst_orig);
1770 *dst_p = NULL;
1771 return err;
1773 nopol:
1774 err = -ENOENT;
1775 if (flags & XFRM_LOOKUP_ICMP)
1776 goto dropdst;
1777 return 0;
1779 EXPORT_SYMBOL(__xfrm_lookup);
1781 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1782 struct sock *sk, int flags)
1784 int err = __xfrm_lookup(dst_p, fl, sk, flags);
1786 if (err == -EREMOTE) {
1787 dst_release(*dst_p);
1788 *dst_p = NULL;
1789 err = -EAGAIN;
1792 return err;
1794 EXPORT_SYMBOL(xfrm_lookup);
1796 static inline int
1797 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1799 struct xfrm_state *x;
1801 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1802 return 0;
1803 x = skb->sp->xvec[idx];
1804 if (!x->type->reject)
1805 return 0;
1806 return x->type->reject(x, skb, fl);
1809 /* When skb is transformed back to its "native" form, we have to
1810 * check policy restrictions. At the moment we make this in maximally
1811 * stupid way. Shame on me. :-) Of course, connected sockets must
1812 * have policy cached at them.
1815 static inline int
1816 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1817 unsigned short family)
1819 if (xfrm_state_kern(x))
1820 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1821 return x->id.proto == tmpl->id.proto &&
1822 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1823 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1824 x->props.mode == tmpl->mode &&
1825 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1826 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1827 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1828 xfrm_state_addr_cmp(tmpl, x, family));
1832 * 0 or more than 0 is returned when validation is succeeded (either bypass
1833 * because of optional transport mode, or next index of the mathced secpath
1834 * state with the template.
1835 * -1 is returned when no matching template is found.
1836 * Otherwise "-2 - errored_index" is returned.
1838 static inline int
1839 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1840 unsigned short family)
1842 int idx = start;
1844 if (tmpl->optional) {
1845 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1846 return start;
1847 } else
1848 start = -1;
1849 for (; idx < sp->len; idx++) {
1850 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1851 return ++idx;
1852 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1853 if (start == -1)
1854 start = -2-idx;
1855 break;
1858 return start;
1861 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1862 unsigned int family, int reverse)
1864 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1865 int err;
1867 if (unlikely(afinfo == NULL))
1868 return -EAFNOSUPPORT;
1870 afinfo->decode_session(skb, fl, reverse);
1871 err = security_xfrm_decode_session(skb, &fl->secid);
1872 xfrm_policy_put_afinfo(afinfo);
1873 return err;
1875 EXPORT_SYMBOL(__xfrm_decode_session);
1877 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1879 for (; k < sp->len; k++) {
1880 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1881 *idxp = k;
1882 return 1;
1886 return 0;
1889 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1890 unsigned short family)
1892 struct xfrm_policy *pol;
1893 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1894 int npols = 0;
1895 int xfrm_nr;
1896 int pi;
1897 int reverse;
1898 struct flowi fl;
1899 u8 fl_dir;
1900 int xerr_idx = -1;
1902 reverse = dir & ~XFRM_POLICY_MASK;
1903 dir &= XFRM_POLICY_MASK;
1904 fl_dir = policy_to_flow_dir(dir);
1906 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1907 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1908 return 0;
1911 nf_nat_decode_session(skb, &fl, family);
1913 /* First, check used SA against their selectors. */
1914 if (skb->sp) {
1915 int i;
1917 for (i=skb->sp->len-1; i>=0; i--) {
1918 struct xfrm_state *x = skb->sp->xvec[i];
1919 if (!xfrm_selector_match(&x->sel, &fl, family)) {
1920 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1921 return 0;
1926 pol = NULL;
1927 if (sk && sk->sk_policy[dir]) {
1928 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1929 if (IS_ERR(pol)) {
1930 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1931 return 0;
1935 if (!pol)
1936 pol = flow_cache_lookup(&fl, family, fl_dir,
1937 xfrm_policy_lookup);
1939 if (IS_ERR(pol)) {
1940 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1941 return 0;
1944 if (!pol) {
1945 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1946 xfrm_secpath_reject(xerr_idx, skb, &fl);
1947 XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1948 return 0;
1950 return 1;
1953 pol->curlft.use_time = get_seconds();
1955 pols[0] = pol;
1956 npols ++;
1957 #ifdef CONFIG_XFRM_SUB_POLICY
1958 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1959 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1960 &fl, family,
1961 XFRM_POLICY_IN);
1962 if (pols[1]) {
1963 if (IS_ERR(pols[1])) {
1964 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1965 return 0;
1967 pols[1]->curlft.use_time = get_seconds();
1968 npols ++;
1971 #endif
1973 if (pol->action == XFRM_POLICY_ALLOW) {
1974 struct sec_path *sp;
1975 static struct sec_path dummy;
1976 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1977 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1978 struct xfrm_tmpl **tpp = tp;
1979 int ti = 0;
1980 int i, k;
1982 if ((sp = skb->sp) == NULL)
1983 sp = &dummy;
1985 for (pi = 0; pi < npols; pi++) {
1986 if (pols[pi] != pol &&
1987 pols[pi]->action != XFRM_POLICY_ALLOW) {
1988 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1989 goto reject;
1991 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1992 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1993 goto reject_error;
1995 for (i = 0; i < pols[pi]->xfrm_nr; i++)
1996 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1998 xfrm_nr = ti;
1999 if (npols > 1) {
2000 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2001 tpp = stp;
2004 /* For each tunnel xfrm, find the first matching tmpl.
2005 * For each tmpl before that, find corresponding xfrm.
2006 * Order is _important_. Later we will implement
2007 * some barriers, but at the moment barriers
2008 * are implied between each two transformations.
2010 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2011 k = xfrm_policy_ok(tpp[i], sp, k, family);
2012 if (k < 0) {
2013 if (k < -1)
2014 /* "-2 - errored_index" returned */
2015 xerr_idx = -(2+k);
2016 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2017 goto reject;
2021 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2022 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2023 goto reject;
2026 xfrm_pols_put(pols, npols);
2027 return 1;
2029 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2031 reject:
2032 xfrm_secpath_reject(xerr_idx, skb, &fl);
2033 reject_error:
2034 xfrm_pols_put(pols, npols);
2035 return 0;
2037 EXPORT_SYMBOL(__xfrm_policy_check);
2039 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2041 struct flowi fl;
2043 if (xfrm_decode_session(skb, &fl, family) < 0) {
2044 /* XXX: we should have something like FWDHDRERROR here. */
2045 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2046 return 0;
2049 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2051 EXPORT_SYMBOL(__xfrm_route_forward);
2053 /* Optimize later using cookies and generation ids. */
2055 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2057 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2058 * to "-1" to force all XFRM destinations to get validated by
2059 * dst_ops->check on every use. We do this because when a
2060 * normal route referenced by an XFRM dst is obsoleted we do
2061 * not go looking around for all parent referencing XFRM dsts
2062 * so that we can invalidate them. It is just too much work.
2063 * Instead we make the checks here on every use. For example:
2065 * XFRM dst A --> IPv4 dst X
2067 * X is the "xdst->route" of A (X is also the "dst->path" of A
2068 * in this example). If X is marked obsolete, "A" will not
2069 * notice. That's what we are validating here via the
2070 * stale_bundle() check.
2072 * When a policy's bundle is pruned, we dst_free() the XFRM
2073 * dst which causes it's ->obsolete field to be set to a
2074 * positive non-zero integer. If an XFRM dst has been pruned
2075 * like this, we want to force a new route lookup.
2077 if (dst->obsolete < 0 && !stale_bundle(dst))
2078 return dst;
2080 return NULL;
2083 static int stale_bundle(struct dst_entry *dst)
2085 return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2088 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2090 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2091 dst->dev = dev_net(dev)->loopback_dev;
2092 dev_hold(dst->dev);
2093 dev_put(dev);
2096 EXPORT_SYMBOL(xfrm_dst_ifdown);
2098 static void xfrm_link_failure(struct sk_buff *skb)
2100 /* Impossible. Such dst must be popped before reaches point of failure. */
2101 return;
2104 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2106 if (dst) {
2107 if (dst->obsolete) {
2108 dst_release(dst);
2109 dst = NULL;
2112 return dst;
2115 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2117 struct dst_entry *dst, **dstp;
2119 write_lock(&pol->lock);
2120 dstp = &pol->bundles;
2121 while ((dst=*dstp) != NULL) {
2122 if (func(dst)) {
2123 *dstp = dst->next;
2124 dst->next = *gc_list_p;
2125 *gc_list_p = dst;
2126 } else {
2127 dstp = &dst->next;
2130 write_unlock(&pol->lock);
2133 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2135 struct dst_entry *gc_list = NULL;
2136 int dir;
2138 read_lock_bh(&xfrm_policy_lock);
2139 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2140 struct xfrm_policy *pol;
2141 struct hlist_node *entry;
2142 struct hlist_head *table;
2143 int i;
2145 hlist_for_each_entry(pol, entry,
2146 &xfrm_policy_inexact[dir], bydst)
2147 prune_one_bundle(pol, func, &gc_list);
2149 table = xfrm_policy_bydst[dir].table;
2150 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2151 hlist_for_each_entry(pol, entry, table + i, bydst)
2152 prune_one_bundle(pol, func, &gc_list);
2155 read_unlock_bh(&xfrm_policy_lock);
2157 while (gc_list) {
2158 struct dst_entry *dst = gc_list;
2159 gc_list = dst->next;
2160 dst_free(dst);
2164 static int unused_bundle(struct dst_entry *dst)
2166 return !atomic_read(&dst->__refcnt);
2169 static void __xfrm_garbage_collect(void)
2171 xfrm_prune_bundles(unused_bundle);
2174 static int xfrm_flush_bundles(void)
2176 xfrm_prune_bundles(stale_bundle);
2177 return 0;
2180 static void xfrm_init_pmtu(struct dst_entry *dst)
2182 do {
2183 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2184 u32 pmtu, route_mtu_cached;
2186 pmtu = dst_mtu(dst->child);
2187 xdst->child_mtu_cached = pmtu;
2189 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2191 route_mtu_cached = dst_mtu(xdst->route);
2192 xdst->route_mtu_cached = route_mtu_cached;
2194 if (pmtu > route_mtu_cached)
2195 pmtu = route_mtu_cached;
2197 dst->metrics[RTAX_MTU-1] = pmtu;
2198 } while ((dst = dst->next));
2201 /* Check that the bundle accepts the flow and its components are
2202 * still valid.
2205 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2206 struct flowi *fl, int family, int strict)
2208 struct dst_entry *dst = &first->u.dst;
2209 struct xfrm_dst *last;
2210 u32 mtu;
2212 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2213 (dst->dev && !netif_running(dst->dev)))
2214 return 0;
2215 #ifdef CONFIG_XFRM_SUB_POLICY
2216 if (fl) {
2217 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2218 return 0;
2219 if (first->partner &&
2220 !xfrm_selector_match(first->partner, fl, family))
2221 return 0;
2223 #endif
2225 last = NULL;
2227 do {
2228 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2230 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2231 return 0;
2232 if (fl && pol &&
2233 !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2234 return 0;
2235 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2236 return 0;
2237 if (xdst->genid != dst->xfrm->genid)
2238 return 0;
2240 if (strict && fl &&
2241 !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2242 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2243 return 0;
2245 mtu = dst_mtu(dst->child);
2246 if (xdst->child_mtu_cached != mtu) {
2247 last = xdst;
2248 xdst->child_mtu_cached = mtu;
2251 if (!dst_check(xdst->route, xdst->route_cookie))
2252 return 0;
2253 mtu = dst_mtu(xdst->route);
2254 if (xdst->route_mtu_cached != mtu) {
2255 last = xdst;
2256 xdst->route_mtu_cached = mtu;
2259 dst = dst->child;
2260 } while (dst->xfrm);
2262 if (likely(!last))
2263 return 1;
2265 mtu = last->child_mtu_cached;
2266 for (;;) {
2267 dst = &last->u.dst;
2269 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2270 if (mtu > last->route_mtu_cached)
2271 mtu = last->route_mtu_cached;
2272 dst->metrics[RTAX_MTU-1] = mtu;
2274 if (last == first)
2275 break;
2277 last = (struct xfrm_dst *)last->u.dst.next;
2278 last->child_mtu_cached = mtu;
2281 return 1;
2284 EXPORT_SYMBOL(xfrm_bundle_ok);
2286 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2288 int err = 0;
2289 if (unlikely(afinfo == NULL))
2290 return -EINVAL;
2291 if (unlikely(afinfo->family >= NPROTO))
2292 return -EAFNOSUPPORT;
2293 write_lock_bh(&xfrm_policy_afinfo_lock);
2294 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2295 err = -ENOBUFS;
2296 else {
2297 struct dst_ops *dst_ops = afinfo->dst_ops;
2298 if (likely(dst_ops->kmem_cachep == NULL))
2299 dst_ops->kmem_cachep = xfrm_dst_cache;
2300 if (likely(dst_ops->check == NULL))
2301 dst_ops->check = xfrm_dst_check;
2302 if (likely(dst_ops->negative_advice == NULL))
2303 dst_ops->negative_advice = xfrm_negative_advice;
2304 if (likely(dst_ops->link_failure == NULL))
2305 dst_ops->link_failure = xfrm_link_failure;
2306 if (likely(afinfo->garbage_collect == NULL))
2307 afinfo->garbage_collect = __xfrm_garbage_collect;
2308 xfrm_policy_afinfo[afinfo->family] = afinfo;
2310 write_unlock_bh(&xfrm_policy_afinfo_lock);
2311 return err;
2313 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2315 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2317 int err = 0;
2318 if (unlikely(afinfo == NULL))
2319 return -EINVAL;
2320 if (unlikely(afinfo->family >= NPROTO))
2321 return -EAFNOSUPPORT;
2322 write_lock_bh(&xfrm_policy_afinfo_lock);
2323 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2324 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2325 err = -EINVAL;
2326 else {
2327 struct dst_ops *dst_ops = afinfo->dst_ops;
2328 xfrm_policy_afinfo[afinfo->family] = NULL;
2329 dst_ops->kmem_cachep = NULL;
2330 dst_ops->check = NULL;
2331 dst_ops->negative_advice = NULL;
2332 dst_ops->link_failure = NULL;
2333 afinfo->garbage_collect = NULL;
2336 write_unlock_bh(&xfrm_policy_afinfo_lock);
2337 return err;
2339 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2341 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2343 struct xfrm_policy_afinfo *afinfo;
2344 if (unlikely(family >= NPROTO))
2345 return NULL;
2346 read_lock(&xfrm_policy_afinfo_lock);
2347 afinfo = xfrm_policy_afinfo[family];
2348 if (unlikely(!afinfo))
2349 read_unlock(&xfrm_policy_afinfo_lock);
2350 return afinfo;
2353 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2355 read_unlock(&xfrm_policy_afinfo_lock);
2358 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2360 struct net_device *dev = ptr;
2362 if (!net_eq(dev_net(dev), &init_net))
2363 return NOTIFY_DONE;
2365 switch (event) {
2366 case NETDEV_DOWN:
2367 xfrm_flush_bundles();
2369 return NOTIFY_DONE;
2372 static struct notifier_block xfrm_dev_notifier = {
2373 xfrm_dev_event,
2374 NULL,
2378 #ifdef CONFIG_XFRM_STATISTICS
2379 static int __init xfrm_statistics_init(void)
2381 if (snmp_mib_init((void **)xfrm_statistics,
2382 sizeof(struct linux_xfrm_mib)) < 0)
2383 return -ENOMEM;
2384 return 0;
2386 #endif
2388 static void __init xfrm_policy_init(void)
2390 unsigned int hmask, sz;
2391 int dir;
2393 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2394 sizeof(struct xfrm_dst),
2395 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2396 NULL);
2398 hmask = 8 - 1;
2399 sz = (hmask+1) * sizeof(struct hlist_head);
2401 xfrm_policy_byidx = xfrm_hash_alloc(sz);
2402 xfrm_idx_hmask = hmask;
2403 if (!xfrm_policy_byidx)
2404 panic("XFRM: failed to allocate byidx hash\n");
2406 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2407 struct xfrm_policy_hash *htab;
2409 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2411 htab = &xfrm_policy_bydst[dir];
2412 htab->table = xfrm_hash_alloc(sz);
2413 htab->hmask = hmask;
2414 if (!htab->table)
2415 panic("XFRM: failed to allocate bydst hash\n");
2418 for (dir = 0; dir < XFRM_POLICY_TYPE_MAX; dir++)
2419 INIT_LIST_HEAD(&xfrm_policy_bytype[dir]);
2421 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2422 register_netdevice_notifier(&xfrm_dev_notifier);
2425 void __init xfrm_init(void)
2427 #ifdef CONFIG_XFRM_STATISTICS
2428 xfrm_statistics_init();
2429 #endif
2430 xfrm_state_init();
2431 xfrm_policy_init();
2432 xfrm_input_init();
2433 #ifdef CONFIG_XFRM_STATISTICS
2434 xfrm_proc_init();
2435 #endif
2438 #ifdef CONFIG_AUDITSYSCALL
2439 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2440 struct audit_buffer *audit_buf)
2442 struct xfrm_sec_ctx *ctx = xp->security;
2443 struct xfrm_selector *sel = &xp->selector;
2445 if (ctx)
2446 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2447 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2449 switch(sel->family) {
2450 case AF_INET:
2451 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2452 NIPQUAD(sel->saddr.a4));
2453 if (sel->prefixlen_s != 32)
2454 audit_log_format(audit_buf, " src_prefixlen=%d",
2455 sel->prefixlen_s);
2456 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2457 NIPQUAD(sel->daddr.a4));
2458 if (sel->prefixlen_d != 32)
2459 audit_log_format(audit_buf, " dst_prefixlen=%d",
2460 sel->prefixlen_d);
2461 break;
2462 case AF_INET6:
2463 audit_log_format(audit_buf, " src=" NIP6_FMT,
2464 NIP6(*(struct in6_addr *)sel->saddr.a6));
2465 if (sel->prefixlen_s != 128)
2466 audit_log_format(audit_buf, " src_prefixlen=%d",
2467 sel->prefixlen_s);
2468 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2469 NIP6(*(struct in6_addr *)sel->daddr.a6));
2470 if (sel->prefixlen_d != 128)
2471 audit_log_format(audit_buf, " dst_prefixlen=%d",
2472 sel->prefixlen_d);
2473 break;
2477 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2478 uid_t auid, u32 sessionid, u32 secid)
2480 struct audit_buffer *audit_buf;
2482 audit_buf = xfrm_audit_start("SPD-add");
2483 if (audit_buf == NULL)
2484 return;
2485 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2486 audit_log_format(audit_buf, " res=%u", result);
2487 xfrm_audit_common_policyinfo(xp, audit_buf);
2488 audit_log_end(audit_buf);
2490 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2492 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2493 uid_t auid, u32 sessionid, u32 secid)
2495 struct audit_buffer *audit_buf;
2497 audit_buf = xfrm_audit_start("SPD-delete");
2498 if (audit_buf == NULL)
2499 return;
2500 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2501 audit_log_format(audit_buf, " res=%u", result);
2502 xfrm_audit_common_policyinfo(xp, audit_buf);
2503 audit_log_end(audit_buf);
2505 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2506 #endif
2508 #ifdef CONFIG_XFRM_MIGRATE
2509 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2510 struct xfrm_selector *sel_tgt)
2512 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2513 if (sel_tgt->family == sel_cmp->family &&
2514 xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2515 sel_cmp->family) == 0 &&
2516 xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2517 sel_cmp->family) == 0 &&
2518 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2519 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2520 return 1;
2522 } else {
2523 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2524 return 1;
2527 return 0;
2530 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2531 u8 dir, u8 type)
2533 struct xfrm_policy *pol, *ret = NULL;
2534 struct hlist_node *entry;
2535 struct hlist_head *chain;
2536 u32 priority = ~0U;
2538 read_lock_bh(&xfrm_policy_lock);
2539 chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2540 hlist_for_each_entry(pol, entry, chain, bydst) {
2541 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2542 pol->type == type) {
2543 ret = pol;
2544 priority = ret->priority;
2545 break;
2548 chain = &xfrm_policy_inexact[dir];
2549 hlist_for_each_entry(pol, entry, chain, bydst) {
2550 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2551 pol->type == type &&
2552 pol->priority < priority) {
2553 ret = pol;
2554 break;
2558 if (ret)
2559 xfrm_pol_hold(ret);
2561 read_unlock_bh(&xfrm_policy_lock);
2563 return ret;
2566 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2568 int match = 0;
2570 if (t->mode == m->mode && t->id.proto == m->proto &&
2571 (m->reqid == 0 || t->reqid == m->reqid)) {
2572 switch (t->mode) {
2573 case XFRM_MODE_TUNNEL:
2574 case XFRM_MODE_BEET:
2575 if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2576 m->old_family) == 0 &&
2577 xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2578 m->old_family) == 0) {
2579 match = 1;
2581 break;
2582 case XFRM_MODE_TRANSPORT:
2583 /* in case of transport mode, template does not store
2584 any IP addresses, hence we just compare mode and
2585 protocol */
2586 match = 1;
2587 break;
2588 default:
2589 break;
2592 return match;
2595 /* update endpoint address(es) of template(s) */
2596 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2597 struct xfrm_migrate *m, int num_migrate)
2599 struct xfrm_migrate *mp;
2600 struct dst_entry *dst;
2601 int i, j, n = 0;
2603 write_lock_bh(&pol->lock);
2604 if (unlikely(pol->dead)) {
2605 /* target policy has been deleted */
2606 write_unlock_bh(&pol->lock);
2607 return -ENOENT;
2610 for (i = 0; i < pol->xfrm_nr; i++) {
2611 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2612 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2613 continue;
2614 n++;
2615 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2616 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2617 continue;
2618 /* update endpoints */
2619 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2620 sizeof(pol->xfrm_vec[i].id.daddr));
2621 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2622 sizeof(pol->xfrm_vec[i].saddr));
2623 pol->xfrm_vec[i].encap_family = mp->new_family;
2624 /* flush bundles */
2625 while ((dst = pol->bundles) != NULL) {
2626 pol->bundles = dst->next;
2627 dst_free(dst);
2632 write_unlock_bh(&pol->lock);
2634 if (!n)
2635 return -ENODATA;
2637 return 0;
2640 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2642 int i, j;
2644 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2645 return -EINVAL;
2647 for (i = 0; i < num_migrate; i++) {
2648 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2649 m[i].old_family) == 0) &&
2650 (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2651 m[i].old_family) == 0))
2652 return -EINVAL;
2653 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2654 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2655 return -EINVAL;
2657 /* check if there is any duplicated entry */
2658 for (j = i + 1; j < num_migrate; j++) {
2659 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2660 sizeof(m[i].old_daddr)) &&
2661 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2662 sizeof(m[i].old_saddr)) &&
2663 m[i].proto == m[j].proto &&
2664 m[i].mode == m[j].mode &&
2665 m[i].reqid == m[j].reqid &&
2666 m[i].old_family == m[j].old_family)
2667 return -EINVAL;
2671 return 0;
2674 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2675 struct xfrm_migrate *m, int num_migrate)
2677 int i, err, nx_cur = 0, nx_new = 0;
2678 struct xfrm_policy *pol = NULL;
2679 struct xfrm_state *x, *xc;
2680 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2681 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2682 struct xfrm_migrate *mp;
2684 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2685 goto out;
2687 /* Stage 1 - find policy */
2688 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2689 err = -ENOENT;
2690 goto out;
2693 /* Stage 2 - find and update state(s) */
2694 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2695 if ((x = xfrm_migrate_state_find(mp))) {
2696 x_cur[nx_cur] = x;
2697 nx_cur++;
2698 if ((xc = xfrm_state_migrate(x, mp))) {
2699 x_new[nx_new] = xc;
2700 nx_new++;
2701 } else {
2702 err = -ENODATA;
2703 goto restore_state;
2708 /* Stage 3 - update policy */
2709 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2710 goto restore_state;
2712 /* Stage 4 - delete old state(s) */
2713 if (nx_cur) {
2714 xfrm_states_put(x_cur, nx_cur);
2715 xfrm_states_delete(x_cur, nx_cur);
2718 /* Stage 5 - announce */
2719 km_migrate(sel, dir, type, m, num_migrate);
2721 xfrm_pol_put(pol);
2723 return 0;
2724 out:
2725 return err;
2727 restore_state:
2728 if (pol)
2729 xfrm_pol_put(pol);
2730 if (nx_cur)
2731 xfrm_states_put(x_cur, nx_cur);
2732 if (nx_new)
2733 xfrm_states_delete(x_new, nx_new);
2735 return err;
2737 EXPORT_SYMBOL(xfrm_migrate);
2738 #endif