xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 5104d265)
1 /*
2  * xfrm_policy.c
3  *
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
13  *
14  */
15 
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/flow.h>
30 #include <net/xfrm.h>
31 #include <net/ip.h>
32 #ifdef CONFIG_XFRM_STATISTICS
33 #include <net/snmp.h>
34 #endif
35 
36 #include "xfrm_hash.h"
37 
38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
40 #define XFRM_MAX_QUEUE_LEN	100
41 
42 DEFINE_MUTEX(xfrm_cfg_mutex);
43 EXPORT_SYMBOL(xfrm_cfg_mutex);
44 
45 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
46 static struct dst_entry *xfrm_policy_sk_bundles;
47 static DEFINE_RWLOCK(xfrm_policy_lock);
48 
49 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
50 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
51 						__read_mostly;
52 
53 static struct kmem_cache *xfrm_dst_cache __read_mostly;
54 
55 static void xfrm_init_pmtu(struct dst_entry *dst);
56 static int stale_bundle(struct dst_entry *dst);
57 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
58 static void xfrm_policy_queue_process(unsigned long arg);
59 
60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
61 						int dir);
62 
63 static inline bool
64 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
65 {
66 	const struct flowi4 *fl4 = &fl->u.ip4;
67 
68 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
69 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
70 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
71 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
72 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
73 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
74 }
75 
76 static inline bool
77 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
78 {
79 	const struct flowi6 *fl6 = &fl->u.ip6;
80 
81 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
82 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
83 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
84 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
85 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
86 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
87 }
88 
89 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
90 			 unsigned short family)
91 {
92 	switch (family) {
93 	case AF_INET:
94 		return __xfrm4_selector_match(sel, fl);
95 	case AF_INET6:
96 		return __xfrm6_selector_match(sel, fl);
97 	}
98 	return false;
99 }
100 
101 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
102 {
103 	struct xfrm_policy_afinfo *afinfo;
104 
105 	if (unlikely(family >= NPROTO))
106 		return NULL;
107 	rcu_read_lock();
108 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
109 	if (unlikely(!afinfo))
110 		rcu_read_unlock();
111 	return afinfo;
112 }
113 
114 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
115 {
116 	rcu_read_unlock();
117 }
118 
119 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
120 						  const xfrm_address_t *saddr,
121 						  const xfrm_address_t *daddr,
122 						  int family)
123 {
124 	struct xfrm_policy_afinfo *afinfo;
125 	struct dst_entry *dst;
126 
127 	afinfo = xfrm_policy_get_afinfo(family);
128 	if (unlikely(afinfo == NULL))
129 		return ERR_PTR(-EAFNOSUPPORT);
130 
131 	dst = afinfo->dst_lookup(net, tos, saddr, daddr);
132 
133 	xfrm_policy_put_afinfo(afinfo);
134 
135 	return dst;
136 }
137 
138 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
139 						xfrm_address_t *prev_saddr,
140 						xfrm_address_t *prev_daddr,
141 						int family)
142 {
143 	struct net *net = xs_net(x);
144 	xfrm_address_t *saddr = &x->props.saddr;
145 	xfrm_address_t *daddr = &x->id.daddr;
146 	struct dst_entry *dst;
147 
148 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
149 		saddr = x->coaddr;
150 		daddr = prev_daddr;
151 	}
152 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
153 		saddr = prev_saddr;
154 		daddr = x->coaddr;
155 	}
156 
157 	dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
158 
159 	if (!IS_ERR(dst)) {
160 		if (prev_saddr != saddr)
161 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
162 		if (prev_daddr != daddr)
163 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
164 	}
165 
166 	return dst;
167 }
168 
169 static inline unsigned long make_jiffies(long secs)
170 {
171 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
172 		return MAX_SCHEDULE_TIMEOUT-1;
173 	else
174 		return secs*HZ;
175 }
176 
177 static void xfrm_policy_timer(unsigned long data)
178 {
179 	struct xfrm_policy *xp = (struct xfrm_policy*)data;
180 	unsigned long now = get_seconds();
181 	long next = LONG_MAX;
182 	int warn = 0;
183 	int dir;
184 
185 	read_lock(&xp->lock);
186 
187 	if (unlikely(xp->walk.dead))
188 		goto out;
189 
190 	dir = xfrm_policy_id2dir(xp->index);
191 
192 	if (xp->lft.hard_add_expires_seconds) {
193 		long tmo = xp->lft.hard_add_expires_seconds +
194 			xp->curlft.add_time - now;
195 		if (tmo <= 0)
196 			goto expired;
197 		if (tmo < next)
198 			next = tmo;
199 	}
200 	if (xp->lft.hard_use_expires_seconds) {
201 		long tmo = xp->lft.hard_use_expires_seconds +
202 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
203 		if (tmo <= 0)
204 			goto expired;
205 		if (tmo < next)
206 			next = tmo;
207 	}
208 	if (xp->lft.soft_add_expires_seconds) {
209 		long tmo = xp->lft.soft_add_expires_seconds +
210 			xp->curlft.add_time - now;
211 		if (tmo <= 0) {
212 			warn = 1;
213 			tmo = XFRM_KM_TIMEOUT;
214 		}
215 		if (tmo < next)
216 			next = tmo;
217 	}
218 	if (xp->lft.soft_use_expires_seconds) {
219 		long tmo = xp->lft.soft_use_expires_seconds +
220 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
221 		if (tmo <= 0) {
222 			warn = 1;
223 			tmo = XFRM_KM_TIMEOUT;
224 		}
225 		if (tmo < next)
226 			next = tmo;
227 	}
228 
229 	if (warn)
230 		km_policy_expired(xp, dir, 0, 0);
231 	if (next != LONG_MAX &&
232 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
233 		xfrm_pol_hold(xp);
234 
235 out:
236 	read_unlock(&xp->lock);
237 	xfrm_pol_put(xp);
238 	return;
239 
240 expired:
241 	read_unlock(&xp->lock);
242 	if (!xfrm_policy_delete(xp, dir))
243 		km_policy_expired(xp, dir, 1, 0);
244 	xfrm_pol_put(xp);
245 }
246 
247 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
248 {
249 	struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
250 
251 	if (unlikely(pol->walk.dead))
252 		flo = NULL;
253 	else
254 		xfrm_pol_hold(pol);
255 
256 	return flo;
257 }
258 
259 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
260 {
261 	struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
262 
263 	return !pol->walk.dead;
264 }
265 
266 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
267 {
268 	xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
269 }
270 
271 static const struct flow_cache_ops xfrm_policy_fc_ops = {
272 	.get = xfrm_policy_flo_get,
273 	.check = xfrm_policy_flo_check,
274 	.delete = xfrm_policy_flo_delete,
275 };
276 
277 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
278  * SPD calls.
279  */
280 
281 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
282 {
283 	struct xfrm_policy *policy;
284 
285 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
286 
287 	if (policy) {
288 		write_pnet(&policy->xp_net, net);
289 		INIT_LIST_HEAD(&policy->walk.all);
290 		INIT_HLIST_NODE(&policy->bydst);
291 		INIT_HLIST_NODE(&policy->byidx);
292 		rwlock_init(&policy->lock);
293 		atomic_set(&policy->refcnt, 1);
294 		skb_queue_head_init(&policy->polq.hold_queue);
295 		setup_timer(&policy->timer, xfrm_policy_timer,
296 				(unsigned long)policy);
297 		setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
298 			    (unsigned long)policy);
299 		policy->flo.ops = &xfrm_policy_fc_ops;
300 	}
301 	return policy;
302 }
303 EXPORT_SYMBOL(xfrm_policy_alloc);
304 
305 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
306 
307 void xfrm_policy_destroy(struct xfrm_policy *policy)
308 {
309 	BUG_ON(!policy->walk.dead);
310 
311 	if (del_timer(&policy->timer))
312 		BUG();
313 
314 	security_xfrm_policy_free(policy->security);
315 	kfree(policy);
316 }
317 EXPORT_SYMBOL(xfrm_policy_destroy);
318 
319 static void xfrm_queue_purge(struct sk_buff_head *list)
320 {
321 	struct sk_buff *skb;
322 
323 	while ((skb = skb_dequeue(list)) != NULL) {
324 		dev_put(skb->dev);
325 		kfree_skb(skb);
326 	}
327 }
328 
329 /* Rule must be locked. Release descentant resources, announce
330  * entry dead. The rule must be unlinked from lists to the moment.
331  */
332 
333 static void xfrm_policy_kill(struct xfrm_policy *policy)
334 {
335 	policy->walk.dead = 1;
336 
337 	atomic_inc(&policy->genid);
338 
339 	del_timer(&policy->polq.hold_timer);
340 	xfrm_queue_purge(&policy->polq.hold_queue);
341 
342 	if (del_timer(&policy->timer))
343 		xfrm_pol_put(policy);
344 
345 	xfrm_pol_put(policy);
346 }
347 
348 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
349 
350 static inline unsigned int idx_hash(struct net *net, u32 index)
351 {
352 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
353 }
354 
355 static struct hlist_head *policy_hash_bysel(struct net *net,
356 					    const struct xfrm_selector *sel,
357 					    unsigned short family, int dir)
358 {
359 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
360 	unsigned int hash = __sel_hash(sel, family, hmask);
361 
362 	return (hash == hmask + 1 ?
363 		&net->xfrm.policy_inexact[dir] :
364 		net->xfrm.policy_bydst[dir].table + hash);
365 }
366 
367 static struct hlist_head *policy_hash_direct(struct net *net,
368 					     const xfrm_address_t *daddr,
369 					     const xfrm_address_t *saddr,
370 					     unsigned short family, int dir)
371 {
372 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
373 	unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
374 
375 	return net->xfrm.policy_bydst[dir].table + hash;
376 }
377 
378 static void xfrm_dst_hash_transfer(struct hlist_head *list,
379 				   struct hlist_head *ndsttable,
380 				   unsigned int nhashmask)
381 {
382 	struct hlist_node *tmp, *entry0 = NULL;
383 	struct xfrm_policy *pol;
384 	unsigned int h0 = 0;
385 
386 redo:
387 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
388 		unsigned int h;
389 
390 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
391 				pol->family, nhashmask);
392 		if (!entry0) {
393 			hlist_del(&pol->bydst);
394 			hlist_add_head(&pol->bydst, ndsttable+h);
395 			h0 = h;
396 		} else {
397 			if (h != h0)
398 				continue;
399 			hlist_del(&pol->bydst);
400 			hlist_add_after(entry0, &pol->bydst);
401 		}
402 		entry0 = &pol->bydst;
403 	}
404 	if (!hlist_empty(list)) {
405 		entry0 = NULL;
406 		goto redo;
407 	}
408 }
409 
410 static void xfrm_idx_hash_transfer(struct hlist_head *list,
411 				   struct hlist_head *nidxtable,
412 				   unsigned int nhashmask)
413 {
414 	struct hlist_node *tmp;
415 	struct xfrm_policy *pol;
416 
417 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
418 		unsigned int h;
419 
420 		h = __idx_hash(pol->index, nhashmask);
421 		hlist_add_head(&pol->byidx, nidxtable+h);
422 	}
423 }
424 
425 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
426 {
427 	return ((old_hmask + 1) << 1) - 1;
428 }
429 
430 static void xfrm_bydst_resize(struct net *net, int dir)
431 {
432 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
433 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
434 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
435 	struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
436 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
437 	int i;
438 
439 	if (!ndst)
440 		return;
441 
442 	write_lock_bh(&xfrm_policy_lock);
443 
444 	for (i = hmask; i >= 0; i--)
445 		xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
446 
447 	net->xfrm.policy_bydst[dir].table = ndst;
448 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
449 
450 	write_unlock_bh(&xfrm_policy_lock);
451 
452 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
453 }
454 
455 static void xfrm_byidx_resize(struct net *net, int total)
456 {
457 	unsigned int hmask = net->xfrm.policy_idx_hmask;
458 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
459 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
460 	struct hlist_head *oidx = net->xfrm.policy_byidx;
461 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
462 	int i;
463 
464 	if (!nidx)
465 		return;
466 
467 	write_lock_bh(&xfrm_policy_lock);
468 
469 	for (i = hmask; i >= 0; i--)
470 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
471 
472 	net->xfrm.policy_byidx = nidx;
473 	net->xfrm.policy_idx_hmask = nhashmask;
474 
475 	write_unlock_bh(&xfrm_policy_lock);
476 
477 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
478 }
479 
480 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
481 {
482 	unsigned int cnt = net->xfrm.policy_count[dir];
483 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
484 
485 	if (total)
486 		*total += cnt;
487 
488 	if ((hmask + 1) < xfrm_policy_hashmax &&
489 	    cnt > hmask)
490 		return 1;
491 
492 	return 0;
493 }
494 
495 static inline int xfrm_byidx_should_resize(struct net *net, int total)
496 {
497 	unsigned int hmask = net->xfrm.policy_idx_hmask;
498 
499 	if ((hmask + 1) < xfrm_policy_hashmax &&
500 	    total > hmask)
501 		return 1;
502 
503 	return 0;
504 }
505 
506 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
507 {
508 	read_lock_bh(&xfrm_policy_lock);
509 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
510 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
511 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
512 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
513 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
514 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
515 	si->spdhcnt = net->xfrm.policy_idx_hmask;
516 	si->spdhmcnt = xfrm_policy_hashmax;
517 	read_unlock_bh(&xfrm_policy_lock);
518 }
519 EXPORT_SYMBOL(xfrm_spd_getinfo);
520 
521 static DEFINE_MUTEX(hash_resize_mutex);
522 static void xfrm_hash_resize(struct work_struct *work)
523 {
524 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
525 	int dir, total;
526 
527 	mutex_lock(&hash_resize_mutex);
528 
529 	total = 0;
530 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
531 		if (xfrm_bydst_should_resize(net, dir, &total))
532 			xfrm_bydst_resize(net, dir);
533 	}
534 	if (xfrm_byidx_should_resize(net, total))
535 		xfrm_byidx_resize(net, total);
536 
537 	mutex_unlock(&hash_resize_mutex);
538 }
539 
540 /* Generate new index... KAME seems to generate them ordered by cost
541  * of an absolute inpredictability of ordering of rules. This will not pass. */
542 static u32 xfrm_gen_index(struct net *net, int dir)
543 {
544 	static u32 idx_generator;
545 
546 	for (;;) {
547 		struct hlist_head *list;
548 		struct xfrm_policy *p;
549 		u32 idx;
550 		int found;
551 
552 		idx = (idx_generator | dir);
553 		idx_generator += 8;
554 		if (idx == 0)
555 			idx = 8;
556 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
557 		found = 0;
558 		hlist_for_each_entry(p, list, byidx) {
559 			if (p->index == idx) {
560 				found = 1;
561 				break;
562 			}
563 		}
564 		if (!found)
565 			return idx;
566 	}
567 }
568 
569 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
570 {
571 	u32 *p1 = (u32 *) s1;
572 	u32 *p2 = (u32 *) s2;
573 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
574 	int i;
575 
576 	for (i = 0; i < len; i++) {
577 		if (p1[i] != p2[i])
578 			return 1;
579 	}
580 
581 	return 0;
582 }
583 
584 static void xfrm_policy_requeue(struct xfrm_policy *old,
585 				struct xfrm_policy *new)
586 {
587 	struct xfrm_policy_queue *pq = &old->polq;
588 	struct sk_buff_head list;
589 
590 	__skb_queue_head_init(&list);
591 
592 	spin_lock_bh(&pq->hold_queue.lock);
593 	skb_queue_splice_init(&pq->hold_queue, &list);
594 	del_timer(&pq->hold_timer);
595 	spin_unlock_bh(&pq->hold_queue.lock);
596 
597 	if (skb_queue_empty(&list))
598 		return;
599 
600 	pq = &new->polq;
601 
602 	spin_lock_bh(&pq->hold_queue.lock);
603 	skb_queue_splice(&list, &pq->hold_queue);
604 	pq->timeout = XFRM_QUEUE_TMO_MIN;
605 	mod_timer(&pq->hold_timer, jiffies);
606 	spin_unlock_bh(&pq->hold_queue.lock);
607 }
608 
609 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
610 				   struct xfrm_policy *pol)
611 {
612 	u32 mark = policy->mark.v & policy->mark.m;
613 
614 	if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
615 		return true;
616 
617 	if ((mark & pol->mark.m) == pol->mark.v &&
618 	    policy->priority == pol->priority)
619 		return true;
620 
621 	return false;
622 }
623 
624 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
625 {
626 	struct net *net = xp_net(policy);
627 	struct xfrm_policy *pol;
628 	struct xfrm_policy *delpol;
629 	struct hlist_head *chain;
630 	struct hlist_node *newpos;
631 
632 	write_lock_bh(&xfrm_policy_lock);
633 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
634 	delpol = NULL;
635 	newpos = NULL;
636 	hlist_for_each_entry(pol, chain, bydst) {
637 		if (pol->type == policy->type &&
638 		    !selector_cmp(&pol->selector, &policy->selector) &&
639 		    xfrm_policy_mark_match(policy, pol) &&
640 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
641 		    !WARN_ON(delpol)) {
642 			if (excl) {
643 				write_unlock_bh(&xfrm_policy_lock);
644 				return -EEXIST;
645 			}
646 			delpol = pol;
647 			if (policy->priority > pol->priority)
648 				continue;
649 		} else if (policy->priority >= pol->priority) {
650 			newpos = &pol->bydst;
651 			continue;
652 		}
653 		if (delpol)
654 			break;
655 	}
656 	if (newpos)
657 		hlist_add_after(newpos, &policy->bydst);
658 	else
659 		hlist_add_head(&policy->bydst, chain);
660 	xfrm_pol_hold(policy);
661 	net->xfrm.policy_count[dir]++;
662 	atomic_inc(&flow_cache_genid);
663 	rt_genid_bump(net);
664 	if (delpol) {
665 		xfrm_policy_requeue(delpol, policy);
666 		__xfrm_policy_unlink(delpol, dir);
667 	}
668 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
669 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
670 	policy->curlft.add_time = get_seconds();
671 	policy->curlft.use_time = 0;
672 	if (!mod_timer(&policy->timer, jiffies + HZ))
673 		xfrm_pol_hold(policy);
674 	list_add(&policy->walk.all, &net->xfrm.policy_all);
675 	write_unlock_bh(&xfrm_policy_lock);
676 
677 	if (delpol)
678 		xfrm_policy_kill(delpol);
679 	else if (xfrm_bydst_should_resize(net, dir, NULL))
680 		schedule_work(&net->xfrm.policy_hash_work);
681 
682 	return 0;
683 }
684 EXPORT_SYMBOL(xfrm_policy_insert);
685 
686 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
687 					  int dir, struct xfrm_selector *sel,
688 					  struct xfrm_sec_ctx *ctx, int delete,
689 					  int *err)
690 {
691 	struct xfrm_policy *pol, *ret;
692 	struct hlist_head *chain;
693 
694 	*err = 0;
695 	write_lock_bh(&xfrm_policy_lock);
696 	chain = policy_hash_bysel(net, sel, sel->family, dir);
697 	ret = NULL;
698 	hlist_for_each_entry(pol, chain, bydst) {
699 		if (pol->type == type &&
700 		    (mark & pol->mark.m) == pol->mark.v &&
701 		    !selector_cmp(sel, &pol->selector) &&
702 		    xfrm_sec_ctx_match(ctx, pol->security)) {
703 			xfrm_pol_hold(pol);
704 			if (delete) {
705 				*err = security_xfrm_policy_delete(
706 								pol->security);
707 				if (*err) {
708 					write_unlock_bh(&xfrm_policy_lock);
709 					return pol;
710 				}
711 				__xfrm_policy_unlink(pol, dir);
712 			}
713 			ret = pol;
714 			break;
715 		}
716 	}
717 	write_unlock_bh(&xfrm_policy_lock);
718 
719 	if (ret && delete)
720 		xfrm_policy_kill(ret);
721 	return ret;
722 }
723 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
724 
725 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
726 				     int dir, u32 id, int delete, int *err)
727 {
728 	struct xfrm_policy *pol, *ret;
729 	struct hlist_head *chain;
730 
731 	*err = -ENOENT;
732 	if (xfrm_policy_id2dir(id) != dir)
733 		return NULL;
734 
735 	*err = 0;
736 	write_lock_bh(&xfrm_policy_lock);
737 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
738 	ret = NULL;
739 	hlist_for_each_entry(pol, chain, byidx) {
740 		if (pol->type == type && pol->index == id &&
741 		    (mark & pol->mark.m) == pol->mark.v) {
742 			xfrm_pol_hold(pol);
743 			if (delete) {
744 				*err = security_xfrm_policy_delete(
745 								pol->security);
746 				if (*err) {
747 					write_unlock_bh(&xfrm_policy_lock);
748 					return pol;
749 				}
750 				__xfrm_policy_unlink(pol, dir);
751 			}
752 			ret = pol;
753 			break;
754 		}
755 	}
756 	write_unlock_bh(&xfrm_policy_lock);
757 
758 	if (ret && delete)
759 		xfrm_policy_kill(ret);
760 	return ret;
761 }
762 EXPORT_SYMBOL(xfrm_policy_byid);
763 
764 #ifdef CONFIG_SECURITY_NETWORK_XFRM
765 static inline int
766 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
767 {
768 	int dir, err = 0;
769 
770 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
771 		struct xfrm_policy *pol;
772 		int i;
773 
774 		hlist_for_each_entry(pol,
775 				     &net->xfrm.policy_inexact[dir], bydst) {
776 			if (pol->type != type)
777 				continue;
778 			err = security_xfrm_policy_delete(pol->security);
779 			if (err) {
780 				xfrm_audit_policy_delete(pol, 0,
781 							 audit_info->loginuid,
782 							 audit_info->sessionid,
783 							 audit_info->secid);
784 				return err;
785 			}
786 		}
787 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
788 			hlist_for_each_entry(pol,
789 					     net->xfrm.policy_bydst[dir].table + i,
790 					     bydst) {
791 				if (pol->type != type)
792 					continue;
793 				err = security_xfrm_policy_delete(
794 								pol->security);
795 				if (err) {
796 					xfrm_audit_policy_delete(pol, 0,
797 							audit_info->loginuid,
798 							audit_info->sessionid,
799 							audit_info->secid);
800 					return err;
801 				}
802 			}
803 		}
804 	}
805 	return err;
806 }
807 #else
808 static inline int
809 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
810 {
811 	return 0;
812 }
813 #endif
814 
815 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
816 {
817 	int dir, err = 0, cnt = 0;
818 
819 	write_lock_bh(&xfrm_policy_lock);
820 
821 	err = xfrm_policy_flush_secctx_check(net, type, audit_info);
822 	if (err)
823 		goto out;
824 
825 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
826 		struct xfrm_policy *pol;
827 		int i;
828 
829 	again1:
830 		hlist_for_each_entry(pol,
831 				     &net->xfrm.policy_inexact[dir], bydst) {
832 			if (pol->type != type)
833 				continue;
834 			__xfrm_policy_unlink(pol, dir);
835 			write_unlock_bh(&xfrm_policy_lock);
836 			cnt++;
837 
838 			xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
839 						 audit_info->sessionid,
840 						 audit_info->secid);
841 
842 			xfrm_policy_kill(pol);
843 
844 			write_lock_bh(&xfrm_policy_lock);
845 			goto again1;
846 		}
847 
848 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
849 	again2:
850 			hlist_for_each_entry(pol,
851 					     net->xfrm.policy_bydst[dir].table + i,
852 					     bydst) {
853 				if (pol->type != type)
854 					continue;
855 				__xfrm_policy_unlink(pol, dir);
856 				write_unlock_bh(&xfrm_policy_lock);
857 				cnt++;
858 
859 				xfrm_audit_policy_delete(pol, 1,
860 							 audit_info->loginuid,
861 							 audit_info->sessionid,
862 							 audit_info->secid);
863 				xfrm_policy_kill(pol);
864 
865 				write_lock_bh(&xfrm_policy_lock);
866 				goto again2;
867 			}
868 		}
869 
870 	}
871 	if (!cnt)
872 		err = -ESRCH;
873 out:
874 	write_unlock_bh(&xfrm_policy_lock);
875 	return err;
876 }
877 EXPORT_SYMBOL(xfrm_policy_flush);
878 
879 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
880 		     int (*func)(struct xfrm_policy *, int, int, void*),
881 		     void *data)
882 {
883 	struct xfrm_policy *pol;
884 	struct xfrm_policy_walk_entry *x;
885 	int error = 0;
886 
887 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
888 	    walk->type != XFRM_POLICY_TYPE_ANY)
889 		return -EINVAL;
890 
891 	if (list_empty(&walk->walk.all) && walk->seq != 0)
892 		return 0;
893 
894 	write_lock_bh(&xfrm_policy_lock);
895 	if (list_empty(&walk->walk.all))
896 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
897 	else
898 		x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
899 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
900 		if (x->dead)
901 			continue;
902 		pol = container_of(x, struct xfrm_policy, walk);
903 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
904 		    walk->type != pol->type)
905 			continue;
906 		error = func(pol, xfrm_policy_id2dir(pol->index),
907 			     walk->seq, data);
908 		if (error) {
909 			list_move_tail(&walk->walk.all, &x->all);
910 			goto out;
911 		}
912 		walk->seq++;
913 	}
914 	if (walk->seq == 0) {
915 		error = -ENOENT;
916 		goto out;
917 	}
918 	list_del_init(&walk->walk.all);
919 out:
920 	write_unlock_bh(&xfrm_policy_lock);
921 	return error;
922 }
923 EXPORT_SYMBOL(xfrm_policy_walk);
924 
925 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
926 {
927 	INIT_LIST_HEAD(&walk->walk.all);
928 	walk->walk.dead = 1;
929 	walk->type = type;
930 	walk->seq = 0;
931 }
932 EXPORT_SYMBOL(xfrm_policy_walk_init);
933 
934 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
935 {
936 	if (list_empty(&walk->walk.all))
937 		return;
938 
939 	write_lock_bh(&xfrm_policy_lock);
940 	list_del(&walk->walk.all);
941 	write_unlock_bh(&xfrm_policy_lock);
942 }
943 EXPORT_SYMBOL(xfrm_policy_walk_done);
944 
945 /*
946  * Find policy to apply to this flow.
947  *
948  * Returns 0 if policy found, else an -errno.
949  */
950 static int xfrm_policy_match(const struct xfrm_policy *pol,
951 			     const struct flowi *fl,
952 			     u8 type, u16 family, int dir)
953 {
954 	const struct xfrm_selector *sel = &pol->selector;
955 	int ret = -ESRCH;
956 	bool match;
957 
958 	if (pol->family != family ||
959 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
960 	    pol->type != type)
961 		return ret;
962 
963 	match = xfrm_selector_match(sel, fl, family);
964 	if (match)
965 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
966 						  dir);
967 
968 	return ret;
969 }
970 
971 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
972 						     const struct flowi *fl,
973 						     u16 family, u8 dir)
974 {
975 	int err;
976 	struct xfrm_policy *pol, *ret;
977 	const xfrm_address_t *daddr, *saddr;
978 	struct hlist_head *chain;
979 	u32 priority = ~0U;
980 
981 	daddr = xfrm_flowi_daddr(fl, family);
982 	saddr = xfrm_flowi_saddr(fl, family);
983 	if (unlikely(!daddr || !saddr))
984 		return NULL;
985 
986 	read_lock_bh(&xfrm_policy_lock);
987 	chain = policy_hash_direct(net, daddr, saddr, family, dir);
988 	ret = NULL;
989 	hlist_for_each_entry(pol, chain, bydst) {
990 		err = xfrm_policy_match(pol, fl, type, family, dir);
991 		if (err) {
992 			if (err == -ESRCH)
993 				continue;
994 			else {
995 				ret = ERR_PTR(err);
996 				goto fail;
997 			}
998 		} else {
999 			ret = pol;
1000 			priority = ret->priority;
1001 			break;
1002 		}
1003 	}
1004 	chain = &net->xfrm.policy_inexact[dir];
1005 	hlist_for_each_entry(pol, chain, bydst) {
1006 		err = xfrm_policy_match(pol, fl, type, family, dir);
1007 		if (err) {
1008 			if (err == -ESRCH)
1009 				continue;
1010 			else {
1011 				ret = ERR_PTR(err);
1012 				goto fail;
1013 			}
1014 		} else if (pol->priority < priority) {
1015 			ret = pol;
1016 			break;
1017 		}
1018 	}
1019 	if (ret)
1020 		xfrm_pol_hold(ret);
1021 fail:
1022 	read_unlock_bh(&xfrm_policy_lock);
1023 
1024 	return ret;
1025 }
1026 
1027 static struct xfrm_policy *
1028 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1029 {
1030 #ifdef CONFIG_XFRM_SUB_POLICY
1031 	struct xfrm_policy *pol;
1032 
1033 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1034 	if (pol != NULL)
1035 		return pol;
1036 #endif
1037 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1038 }
1039 
1040 static int flow_to_policy_dir(int dir)
1041 {
1042 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1043 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1044 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
1045 		return dir;
1046 
1047 	switch (dir) {
1048 	default:
1049 	case FLOW_DIR_IN:
1050 		return XFRM_POLICY_IN;
1051 	case FLOW_DIR_OUT:
1052 		return XFRM_POLICY_OUT;
1053 	case FLOW_DIR_FWD:
1054 		return XFRM_POLICY_FWD;
1055 	}
1056 }
1057 
1058 static struct flow_cache_object *
1059 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1060 		   u8 dir, struct flow_cache_object *old_obj, void *ctx)
1061 {
1062 	struct xfrm_policy *pol;
1063 
1064 	if (old_obj)
1065 		xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1066 
1067 	pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1068 	if (IS_ERR_OR_NULL(pol))
1069 		return ERR_CAST(pol);
1070 
1071 	/* Resolver returns two references:
1072 	 * one for cache and one for caller of flow_cache_lookup() */
1073 	xfrm_pol_hold(pol);
1074 
1075 	return &pol->flo;
1076 }
1077 
1078 static inline int policy_to_flow_dir(int dir)
1079 {
1080 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1081 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1082 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
1083 		return dir;
1084 	switch (dir) {
1085 	default:
1086 	case XFRM_POLICY_IN:
1087 		return FLOW_DIR_IN;
1088 	case XFRM_POLICY_OUT:
1089 		return FLOW_DIR_OUT;
1090 	case XFRM_POLICY_FWD:
1091 		return FLOW_DIR_FWD;
1092 	}
1093 }
1094 
1095 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1096 						 const struct flowi *fl)
1097 {
1098 	struct xfrm_policy *pol;
1099 
1100 	read_lock_bh(&xfrm_policy_lock);
1101 	if ((pol = sk->sk_policy[dir]) != NULL) {
1102 		bool match = xfrm_selector_match(&pol->selector, fl,
1103 						 sk->sk_family);
1104 		int err = 0;
1105 
1106 		if (match) {
1107 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1108 				pol = NULL;
1109 				goto out;
1110 			}
1111 			err = security_xfrm_policy_lookup(pol->security,
1112 						      fl->flowi_secid,
1113 						      policy_to_flow_dir(dir));
1114 			if (!err)
1115 				xfrm_pol_hold(pol);
1116 			else if (err == -ESRCH)
1117 				pol = NULL;
1118 			else
1119 				pol = ERR_PTR(err);
1120 		} else
1121 			pol = NULL;
1122 	}
1123 out:
1124 	read_unlock_bh(&xfrm_policy_lock);
1125 	return pol;
1126 }
1127 
1128 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1129 {
1130 	struct net *net = xp_net(pol);
1131 	struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1132 						     pol->family, dir);
1133 
1134 	list_add(&pol->walk.all, &net->xfrm.policy_all);
1135 	hlist_add_head(&pol->bydst, chain);
1136 	hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1137 	net->xfrm.policy_count[dir]++;
1138 	xfrm_pol_hold(pol);
1139 
1140 	if (xfrm_bydst_should_resize(net, dir, NULL))
1141 		schedule_work(&net->xfrm.policy_hash_work);
1142 }
1143 
1144 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1145 						int dir)
1146 {
1147 	struct net *net = xp_net(pol);
1148 
1149 	if (hlist_unhashed(&pol->bydst))
1150 		return NULL;
1151 
1152 	hlist_del(&pol->bydst);
1153 	hlist_del(&pol->byidx);
1154 	list_del(&pol->walk.all);
1155 	net->xfrm.policy_count[dir]--;
1156 
1157 	return pol;
1158 }
1159 
1160 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1161 {
1162 	write_lock_bh(&xfrm_policy_lock);
1163 	pol = __xfrm_policy_unlink(pol, dir);
1164 	write_unlock_bh(&xfrm_policy_lock);
1165 	if (pol) {
1166 		xfrm_policy_kill(pol);
1167 		return 0;
1168 	}
1169 	return -ENOENT;
1170 }
1171 EXPORT_SYMBOL(xfrm_policy_delete);
1172 
1173 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1174 {
1175 	struct net *net = xp_net(pol);
1176 	struct xfrm_policy *old_pol;
1177 
1178 #ifdef CONFIG_XFRM_SUB_POLICY
1179 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1180 		return -EINVAL;
1181 #endif
1182 
1183 	write_lock_bh(&xfrm_policy_lock);
1184 	old_pol = sk->sk_policy[dir];
1185 	sk->sk_policy[dir] = pol;
1186 	if (pol) {
1187 		pol->curlft.add_time = get_seconds();
1188 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1189 		__xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1190 	}
1191 	if (old_pol) {
1192 		if (pol)
1193 			xfrm_policy_requeue(old_pol, pol);
1194 
1195 		/* Unlinking succeeds always. This is the only function
1196 		 * allowed to delete or replace socket policy.
1197 		 */
1198 		__xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1199 	}
1200 	write_unlock_bh(&xfrm_policy_lock);
1201 
1202 	if (old_pol) {
1203 		xfrm_policy_kill(old_pol);
1204 	}
1205 	return 0;
1206 }
1207 
1208 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1209 {
1210 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1211 
1212 	if (newp) {
1213 		newp->selector = old->selector;
1214 		if (security_xfrm_policy_clone(old->security,
1215 					       &newp->security)) {
1216 			kfree(newp);
1217 			return NULL;  /* ENOMEM */
1218 		}
1219 		newp->lft = old->lft;
1220 		newp->curlft = old->curlft;
1221 		newp->mark = old->mark;
1222 		newp->action = old->action;
1223 		newp->flags = old->flags;
1224 		newp->xfrm_nr = old->xfrm_nr;
1225 		newp->index = old->index;
1226 		newp->type = old->type;
1227 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1228 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1229 		write_lock_bh(&xfrm_policy_lock);
1230 		__xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1231 		write_unlock_bh(&xfrm_policy_lock);
1232 		xfrm_pol_put(newp);
1233 	}
1234 	return newp;
1235 }
1236 
1237 int __xfrm_sk_clone_policy(struct sock *sk)
1238 {
1239 	struct xfrm_policy *p0 = sk->sk_policy[0],
1240 			   *p1 = sk->sk_policy[1];
1241 
1242 	sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1243 	if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1244 		return -ENOMEM;
1245 	if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1246 		return -ENOMEM;
1247 	return 0;
1248 }
1249 
1250 static int
1251 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1252 	       unsigned short family)
1253 {
1254 	int err;
1255 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1256 
1257 	if (unlikely(afinfo == NULL))
1258 		return -EINVAL;
1259 	err = afinfo->get_saddr(net, local, remote);
1260 	xfrm_policy_put_afinfo(afinfo);
1261 	return err;
1262 }
1263 
1264 /* Resolve list of templates for the flow, given policy. */
1265 
1266 static int
1267 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1268 		      struct xfrm_state **xfrm, unsigned short family)
1269 {
1270 	struct net *net = xp_net(policy);
1271 	int nx;
1272 	int i, error;
1273 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1274 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1275 	xfrm_address_t tmp;
1276 
1277 	for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1278 		struct xfrm_state *x;
1279 		xfrm_address_t *remote = daddr;
1280 		xfrm_address_t *local  = saddr;
1281 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1282 
1283 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1284 		    tmpl->mode == XFRM_MODE_BEET) {
1285 			remote = &tmpl->id.daddr;
1286 			local = &tmpl->saddr;
1287 			if (xfrm_addr_any(local, tmpl->encap_family)) {
1288 				error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1289 				if (error)
1290 					goto fail;
1291 				local = &tmp;
1292 			}
1293 		}
1294 
1295 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1296 
1297 		if (x && x->km.state == XFRM_STATE_VALID) {
1298 			xfrm[nx++] = x;
1299 			daddr = remote;
1300 			saddr = local;
1301 			continue;
1302 		}
1303 		if (x) {
1304 			error = (x->km.state == XFRM_STATE_ERROR ?
1305 				 -EINVAL : -EAGAIN);
1306 			xfrm_state_put(x);
1307 		}
1308 		else if (error == -ESRCH)
1309 			error = -EAGAIN;
1310 
1311 		if (!tmpl->optional)
1312 			goto fail;
1313 	}
1314 	return nx;
1315 
1316 fail:
1317 	for (nx--; nx>=0; nx--)
1318 		xfrm_state_put(xfrm[nx]);
1319 	return error;
1320 }
1321 
1322 static int
1323 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1324 		  struct xfrm_state **xfrm, unsigned short family)
1325 {
1326 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1327 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1328 	int cnx = 0;
1329 	int error;
1330 	int ret;
1331 	int i;
1332 
1333 	for (i = 0; i < npols; i++) {
1334 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1335 			error = -ENOBUFS;
1336 			goto fail;
1337 		}
1338 
1339 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1340 		if (ret < 0) {
1341 			error = ret;
1342 			goto fail;
1343 		} else
1344 			cnx += ret;
1345 	}
1346 
1347 	/* found states are sorted for outbound processing */
1348 	if (npols > 1)
1349 		xfrm_state_sort(xfrm, tpp, cnx, family);
1350 
1351 	return cnx;
1352 
1353  fail:
1354 	for (cnx--; cnx>=0; cnx--)
1355 		xfrm_state_put(tpp[cnx]);
1356 	return error;
1357 
1358 }
1359 
1360 /* Check that the bundle accepts the flow and its components are
1361  * still valid.
1362  */
1363 
1364 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1365 {
1366 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1367 	int tos;
1368 
1369 	if (!afinfo)
1370 		return -EINVAL;
1371 
1372 	tos = afinfo->get_tos(fl);
1373 
1374 	xfrm_policy_put_afinfo(afinfo);
1375 
1376 	return tos;
1377 }
1378 
1379 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1380 {
1381 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1382 	struct dst_entry *dst = &xdst->u.dst;
1383 
1384 	if (xdst->route == NULL) {
1385 		/* Dummy bundle - if it has xfrms we were not
1386 		 * able to build bundle as template resolution failed.
1387 		 * It means we need to try again resolving. */
1388 		if (xdst->num_xfrms > 0)
1389 			return NULL;
1390 	} else if (dst->flags & DST_XFRM_QUEUE) {
1391 		return NULL;
1392 	} else {
1393 		/* Real bundle */
1394 		if (stale_bundle(dst))
1395 			return NULL;
1396 	}
1397 
1398 	dst_hold(dst);
1399 	return flo;
1400 }
1401 
1402 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1403 {
1404 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1405 	struct dst_entry *dst = &xdst->u.dst;
1406 
1407 	if (!xdst->route)
1408 		return 0;
1409 	if (stale_bundle(dst))
1410 		return 0;
1411 
1412 	return 1;
1413 }
1414 
1415 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1416 {
1417 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1418 	struct dst_entry *dst = &xdst->u.dst;
1419 
1420 	dst_free(dst);
1421 }
1422 
1423 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1424 	.get = xfrm_bundle_flo_get,
1425 	.check = xfrm_bundle_flo_check,
1426 	.delete = xfrm_bundle_flo_delete,
1427 };
1428 
1429 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1430 {
1431 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1432 	struct dst_ops *dst_ops;
1433 	struct xfrm_dst *xdst;
1434 
1435 	if (!afinfo)
1436 		return ERR_PTR(-EINVAL);
1437 
1438 	switch (family) {
1439 	case AF_INET:
1440 		dst_ops = &net->xfrm.xfrm4_dst_ops;
1441 		break;
1442 #if IS_ENABLED(CONFIG_IPV6)
1443 	case AF_INET6:
1444 		dst_ops = &net->xfrm.xfrm6_dst_ops;
1445 		break;
1446 #endif
1447 	default:
1448 		BUG();
1449 	}
1450 	xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1451 
1452 	if (likely(xdst)) {
1453 		struct dst_entry *dst = &xdst->u.dst;
1454 
1455 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1456 		xdst->flo.ops = &xfrm_bundle_fc_ops;
1457 		if (afinfo->init_dst)
1458 			afinfo->init_dst(net, xdst);
1459 	} else
1460 		xdst = ERR_PTR(-ENOBUFS);
1461 
1462 	xfrm_policy_put_afinfo(afinfo);
1463 
1464 	return xdst;
1465 }
1466 
1467 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1468 				 int nfheader_len)
1469 {
1470 	struct xfrm_policy_afinfo *afinfo =
1471 		xfrm_policy_get_afinfo(dst->ops->family);
1472 	int err;
1473 
1474 	if (!afinfo)
1475 		return -EINVAL;
1476 
1477 	err = afinfo->init_path(path, dst, nfheader_len);
1478 
1479 	xfrm_policy_put_afinfo(afinfo);
1480 
1481 	return err;
1482 }
1483 
1484 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1485 				const struct flowi *fl)
1486 {
1487 	struct xfrm_policy_afinfo *afinfo =
1488 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1489 	int err;
1490 
1491 	if (!afinfo)
1492 		return -EINVAL;
1493 
1494 	err = afinfo->fill_dst(xdst, dev, fl);
1495 
1496 	xfrm_policy_put_afinfo(afinfo);
1497 
1498 	return err;
1499 }
1500 
1501 
1502 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1503  * all the metrics... Shortly, bundle a bundle.
1504  */
1505 
1506 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1507 					    struct xfrm_state **xfrm, int nx,
1508 					    const struct flowi *fl,
1509 					    struct dst_entry *dst)
1510 {
1511 	struct net *net = xp_net(policy);
1512 	unsigned long now = jiffies;
1513 	struct net_device *dev;
1514 	struct xfrm_mode *inner_mode;
1515 	struct dst_entry *dst_prev = NULL;
1516 	struct dst_entry *dst0 = NULL;
1517 	int i = 0;
1518 	int err;
1519 	int header_len = 0;
1520 	int nfheader_len = 0;
1521 	int trailer_len = 0;
1522 	int tos;
1523 	int family = policy->selector.family;
1524 	xfrm_address_t saddr, daddr;
1525 
1526 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1527 
1528 	tos = xfrm_get_tos(fl, family);
1529 	err = tos;
1530 	if (tos < 0)
1531 		goto put_states;
1532 
1533 	dst_hold(dst);
1534 
1535 	for (; i < nx; i++) {
1536 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1537 		struct dst_entry *dst1 = &xdst->u.dst;
1538 
1539 		err = PTR_ERR(xdst);
1540 		if (IS_ERR(xdst)) {
1541 			dst_release(dst);
1542 			goto put_states;
1543 		}
1544 
1545 		if (xfrm[i]->sel.family == AF_UNSPEC) {
1546 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
1547 							xfrm_af2proto(family));
1548 			if (!inner_mode) {
1549 				err = -EAFNOSUPPORT;
1550 				dst_release(dst);
1551 				goto put_states;
1552 			}
1553 		} else
1554 			inner_mode = xfrm[i]->inner_mode;
1555 
1556 		if (!dst_prev)
1557 			dst0 = dst1;
1558 		else {
1559 			dst_prev->child = dst_clone(dst1);
1560 			dst1->flags |= DST_NOHASH;
1561 		}
1562 
1563 		xdst->route = dst;
1564 		dst_copy_metrics(dst1, dst);
1565 
1566 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1567 			family = xfrm[i]->props.family;
1568 			dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1569 					      family);
1570 			err = PTR_ERR(dst);
1571 			if (IS_ERR(dst))
1572 				goto put_states;
1573 		} else
1574 			dst_hold(dst);
1575 
1576 		dst1->xfrm = xfrm[i];
1577 		xdst->xfrm_genid = xfrm[i]->genid;
1578 
1579 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1580 		dst1->flags |= DST_HOST;
1581 		dst1->lastuse = now;
1582 
1583 		dst1->input = dst_discard;
1584 		dst1->output = inner_mode->afinfo->output;
1585 
1586 		dst1->next = dst_prev;
1587 		dst_prev = dst1;
1588 
1589 		header_len += xfrm[i]->props.header_len;
1590 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1591 			nfheader_len += xfrm[i]->props.header_len;
1592 		trailer_len += xfrm[i]->props.trailer_len;
1593 	}
1594 
1595 	dst_prev->child = dst;
1596 	dst0->path = dst;
1597 
1598 	err = -ENODEV;
1599 	dev = dst->dev;
1600 	if (!dev)
1601 		goto free_dst;
1602 
1603 	xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1604 	xfrm_init_pmtu(dst_prev);
1605 
1606 	for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1607 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1608 
1609 		err = xfrm_fill_dst(xdst, dev, fl);
1610 		if (err)
1611 			goto free_dst;
1612 
1613 		dst_prev->header_len = header_len;
1614 		dst_prev->trailer_len = trailer_len;
1615 		header_len -= xdst->u.dst.xfrm->props.header_len;
1616 		trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1617 	}
1618 
1619 out:
1620 	return dst0;
1621 
1622 put_states:
1623 	for (; i < nx; i++)
1624 		xfrm_state_put(xfrm[i]);
1625 free_dst:
1626 	if (dst0)
1627 		dst_free(dst0);
1628 	dst0 = ERR_PTR(err);
1629 	goto out;
1630 }
1631 
1632 static int inline
1633 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1634 {
1635 	if (!*target) {
1636 		*target = kmalloc(size, GFP_ATOMIC);
1637 		if (!*target)
1638 			return -ENOMEM;
1639 	}
1640 	memcpy(*target, src, size);
1641 	return 0;
1642 }
1643 
1644 static int inline
1645 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1646 {
1647 #ifdef CONFIG_XFRM_SUB_POLICY
1648 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1649 	return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1650 				   sel, sizeof(*sel));
1651 #else
1652 	return 0;
1653 #endif
1654 }
1655 
1656 static int inline
1657 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1658 {
1659 #ifdef CONFIG_XFRM_SUB_POLICY
1660 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1661 	return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1662 #else
1663 	return 0;
1664 #endif
1665 }
1666 
1667 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1668 				struct xfrm_policy **pols,
1669 				int *num_pols, int *num_xfrms)
1670 {
1671 	int i;
1672 
1673 	if (*num_pols == 0 || !pols[0]) {
1674 		*num_pols = 0;
1675 		*num_xfrms = 0;
1676 		return 0;
1677 	}
1678 	if (IS_ERR(pols[0]))
1679 		return PTR_ERR(pols[0]);
1680 
1681 	*num_xfrms = pols[0]->xfrm_nr;
1682 
1683 #ifdef CONFIG_XFRM_SUB_POLICY
1684 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1685 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1686 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1687 						    XFRM_POLICY_TYPE_MAIN,
1688 						    fl, family,
1689 						    XFRM_POLICY_OUT);
1690 		if (pols[1]) {
1691 			if (IS_ERR(pols[1])) {
1692 				xfrm_pols_put(pols, *num_pols);
1693 				return PTR_ERR(pols[1]);
1694 			}
1695 			(*num_pols) ++;
1696 			(*num_xfrms) += pols[1]->xfrm_nr;
1697 		}
1698 	}
1699 #endif
1700 	for (i = 0; i < *num_pols; i++) {
1701 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
1702 			*num_xfrms = -1;
1703 			break;
1704 		}
1705 	}
1706 
1707 	return 0;
1708 
1709 }
1710 
1711 static struct xfrm_dst *
1712 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1713 			       const struct flowi *fl, u16 family,
1714 			       struct dst_entry *dst_orig)
1715 {
1716 	struct net *net = xp_net(pols[0]);
1717 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1718 	struct dst_entry *dst;
1719 	struct xfrm_dst *xdst;
1720 	int err;
1721 
1722 	/* Try to instantiate a bundle */
1723 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1724 	if (err <= 0) {
1725 		if (err != 0 && err != -EAGAIN)
1726 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1727 		return ERR_PTR(err);
1728 	}
1729 
1730 	dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1731 	if (IS_ERR(dst)) {
1732 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1733 		return ERR_CAST(dst);
1734 	}
1735 
1736 	xdst = (struct xfrm_dst *)dst;
1737 	xdst->num_xfrms = err;
1738 	if (num_pols > 1)
1739 		err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1740 	else
1741 		err = xfrm_dst_update_origin(dst, fl);
1742 	if (unlikely(err)) {
1743 		dst_free(dst);
1744 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1745 		return ERR_PTR(err);
1746 	}
1747 
1748 	xdst->num_pols = num_pols;
1749 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1750 	xdst->policy_genid = atomic_read(&pols[0]->genid);
1751 
1752 	return xdst;
1753 }
1754 
1755 static void xfrm_policy_queue_process(unsigned long arg)
1756 {
1757 	int err = 0;
1758 	struct sk_buff *skb;
1759 	struct sock *sk;
1760 	struct dst_entry *dst;
1761 	struct net_device *dev;
1762 	struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1763 	struct xfrm_policy_queue *pq = &pol->polq;
1764 	struct flowi fl;
1765 	struct sk_buff_head list;
1766 
1767 	spin_lock(&pq->hold_queue.lock);
1768 	skb = skb_peek(&pq->hold_queue);
1769 	dst = skb_dst(skb);
1770 	sk = skb->sk;
1771 	xfrm_decode_session(skb, &fl, dst->ops->family);
1772 	spin_unlock(&pq->hold_queue.lock);
1773 
1774 	dst_hold(dst->path);
1775 	dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1776 			  sk, 0);
1777 	if (IS_ERR(dst))
1778 		goto purge_queue;
1779 
1780 	if (dst->flags & DST_XFRM_QUEUE) {
1781 		dst_release(dst);
1782 
1783 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1784 			goto purge_queue;
1785 
1786 		pq->timeout = pq->timeout << 1;
1787 		mod_timer(&pq->hold_timer, jiffies + pq->timeout);
1788 		return;
1789 	}
1790 
1791 	dst_release(dst);
1792 
1793 	__skb_queue_head_init(&list);
1794 
1795 	spin_lock(&pq->hold_queue.lock);
1796 	pq->timeout = 0;
1797 	skb_queue_splice_init(&pq->hold_queue, &list);
1798 	spin_unlock(&pq->hold_queue.lock);
1799 
1800 	while (!skb_queue_empty(&list)) {
1801 		skb = __skb_dequeue(&list);
1802 
1803 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1804 		dst_hold(skb_dst(skb)->path);
1805 		dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1806 				  &fl, skb->sk, 0);
1807 		if (IS_ERR(dst)) {
1808 			dev_put(skb->dev);
1809 			kfree_skb(skb);
1810 			continue;
1811 		}
1812 
1813 		nf_reset(skb);
1814 		skb_dst_drop(skb);
1815 		skb_dst_set(skb, dst);
1816 
1817 		dev = skb->dev;
1818 		err = dst_output(skb);
1819 		dev_put(dev);
1820 	}
1821 
1822 	return;
1823 
1824 purge_queue:
1825 	pq->timeout = 0;
1826 	xfrm_queue_purge(&pq->hold_queue);
1827 }
1828 
1829 static int xdst_queue_output(struct sk_buff *skb)
1830 {
1831 	unsigned long sched_next;
1832 	struct dst_entry *dst = skb_dst(skb);
1833 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1834 	struct xfrm_policy_queue *pq = &xdst->pols[0]->polq;
1835 
1836 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1837 		kfree_skb(skb);
1838 		return -EAGAIN;
1839 	}
1840 
1841 	skb_dst_force(skb);
1842 	dev_hold(skb->dev);
1843 
1844 	spin_lock_bh(&pq->hold_queue.lock);
1845 
1846 	if (!pq->timeout)
1847 		pq->timeout = XFRM_QUEUE_TMO_MIN;
1848 
1849 	sched_next = jiffies + pq->timeout;
1850 
1851 	if (del_timer(&pq->hold_timer)) {
1852 		if (time_before(pq->hold_timer.expires, sched_next))
1853 			sched_next = pq->hold_timer.expires;
1854 	}
1855 
1856 	__skb_queue_tail(&pq->hold_queue, skb);
1857 	mod_timer(&pq->hold_timer, sched_next);
1858 
1859 	spin_unlock_bh(&pq->hold_queue.lock);
1860 
1861 	return 0;
1862 }
1863 
1864 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1865 						 struct dst_entry *dst,
1866 						 const struct flowi *fl,
1867 						 int num_xfrms,
1868 						 u16 family)
1869 {
1870 	int err;
1871 	struct net_device *dev;
1872 	struct dst_entry *dst1;
1873 	struct xfrm_dst *xdst;
1874 
1875 	xdst = xfrm_alloc_dst(net, family);
1876 	if (IS_ERR(xdst))
1877 		return xdst;
1878 
1879 	if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0 ||
1880 	    (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP))
1881 		return xdst;
1882 
1883 	dst1 = &xdst->u.dst;
1884 	dst_hold(dst);
1885 	xdst->route = dst;
1886 
1887 	dst_copy_metrics(dst1, dst);
1888 
1889 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1890 	dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1891 	dst1->lastuse = jiffies;
1892 
1893 	dst1->input = dst_discard;
1894 	dst1->output = xdst_queue_output;
1895 
1896 	dst_hold(dst);
1897 	dst1->child = dst;
1898 	dst1->path = dst;
1899 
1900 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1901 
1902 	err = -ENODEV;
1903 	dev = dst->dev;
1904 	if (!dev)
1905 		goto free_dst;
1906 
1907 	err = xfrm_fill_dst(xdst, dev, fl);
1908 	if (err)
1909 		goto free_dst;
1910 
1911 out:
1912 	return xdst;
1913 
1914 free_dst:
1915 	dst_release(dst1);
1916 	xdst = ERR_PTR(err);
1917 	goto out;
1918 }
1919 
1920 static struct flow_cache_object *
1921 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1922 		   struct flow_cache_object *oldflo, void *ctx)
1923 {
1924 	struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1925 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1926 	struct xfrm_dst *xdst, *new_xdst;
1927 	int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1928 
1929 	/* Check if the policies from old bundle are usable */
1930 	xdst = NULL;
1931 	if (oldflo) {
1932 		xdst = container_of(oldflo, struct xfrm_dst, flo);
1933 		num_pols = xdst->num_pols;
1934 		num_xfrms = xdst->num_xfrms;
1935 		pol_dead = 0;
1936 		for (i = 0; i < num_pols; i++) {
1937 			pols[i] = xdst->pols[i];
1938 			pol_dead |= pols[i]->walk.dead;
1939 		}
1940 		if (pol_dead) {
1941 			dst_free(&xdst->u.dst);
1942 			xdst = NULL;
1943 			num_pols = 0;
1944 			num_xfrms = 0;
1945 			oldflo = NULL;
1946 		}
1947 	}
1948 
1949 	/* Resolve policies to use if we couldn't get them from
1950 	 * previous cache entry */
1951 	if (xdst == NULL) {
1952 		num_pols = 1;
1953 		pols[0] = __xfrm_policy_lookup(net, fl, family,
1954 					       flow_to_policy_dir(dir));
1955 		err = xfrm_expand_policies(fl, family, pols,
1956 					   &num_pols, &num_xfrms);
1957 		if (err < 0)
1958 			goto inc_error;
1959 		if (num_pols == 0)
1960 			return NULL;
1961 		if (num_xfrms <= 0)
1962 			goto make_dummy_bundle;
1963 	}
1964 
1965 	new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1966 	if (IS_ERR(new_xdst)) {
1967 		err = PTR_ERR(new_xdst);
1968 		if (err != -EAGAIN)
1969 			goto error;
1970 		if (oldflo == NULL)
1971 			goto make_dummy_bundle;
1972 		dst_hold(&xdst->u.dst);
1973 		return oldflo;
1974 	} else if (new_xdst == NULL) {
1975 		num_xfrms = 0;
1976 		if (oldflo == NULL)
1977 			goto make_dummy_bundle;
1978 		xdst->num_xfrms = 0;
1979 		dst_hold(&xdst->u.dst);
1980 		return oldflo;
1981 	}
1982 
1983 	/* Kill the previous bundle */
1984 	if (xdst) {
1985 		/* The policies were stolen for newly generated bundle */
1986 		xdst->num_pols = 0;
1987 		dst_free(&xdst->u.dst);
1988 	}
1989 
1990 	/* Flow cache does not have reference, it dst_free()'s,
1991 	 * but we do need to return one reference for original caller */
1992 	dst_hold(&new_xdst->u.dst);
1993 	return &new_xdst->flo;
1994 
1995 make_dummy_bundle:
1996 	/* We found policies, but there's no bundles to instantiate:
1997 	 * either because the policy blocks, has no transformations or
1998 	 * we could not build template (no xfrm_states).*/
1999 	xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
2000 	if (IS_ERR(xdst)) {
2001 		xfrm_pols_put(pols, num_pols);
2002 		return ERR_CAST(xdst);
2003 	}
2004 	xdst->num_pols = num_pols;
2005 	xdst->num_xfrms = num_xfrms;
2006 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
2007 
2008 	dst_hold(&xdst->u.dst);
2009 	return &xdst->flo;
2010 
2011 inc_error:
2012 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2013 error:
2014 	if (xdst != NULL)
2015 		dst_free(&xdst->u.dst);
2016 	else
2017 		xfrm_pols_put(pols, num_pols);
2018 	return ERR_PTR(err);
2019 }
2020 
2021 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2022 					struct dst_entry *dst_orig)
2023 {
2024 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2025 	struct dst_entry *ret;
2026 
2027 	if (!afinfo) {
2028 		dst_release(dst_orig);
2029 		return ERR_PTR(-EINVAL);
2030 	} else {
2031 		ret = afinfo->blackhole_route(net, dst_orig);
2032 	}
2033 	xfrm_policy_put_afinfo(afinfo);
2034 
2035 	return ret;
2036 }
2037 
2038 /* Main function: finds/creates a bundle for given flow.
2039  *
2040  * At the moment we eat a raw IP route. Mostly to speed up lookups
2041  * on interfaces with disabled IPsec.
2042  */
2043 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2044 			      const struct flowi *fl,
2045 			      struct sock *sk, int flags)
2046 {
2047 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2048 	struct flow_cache_object *flo;
2049 	struct xfrm_dst *xdst;
2050 	struct dst_entry *dst, *route;
2051 	u16 family = dst_orig->ops->family;
2052 	u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2053 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2054 
2055 restart:
2056 	dst = NULL;
2057 	xdst = NULL;
2058 	route = NULL;
2059 
2060 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2061 		num_pols = 1;
2062 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2063 		err = xfrm_expand_policies(fl, family, pols,
2064 					   &num_pols, &num_xfrms);
2065 		if (err < 0)
2066 			goto dropdst;
2067 
2068 		if (num_pols) {
2069 			if (num_xfrms <= 0) {
2070 				drop_pols = num_pols;
2071 				goto no_transform;
2072 			}
2073 
2074 			xdst = xfrm_resolve_and_create_bundle(
2075 					pols, num_pols, fl,
2076 					family, dst_orig);
2077 			if (IS_ERR(xdst)) {
2078 				xfrm_pols_put(pols, num_pols);
2079 				err = PTR_ERR(xdst);
2080 				goto dropdst;
2081 			} else if (xdst == NULL) {
2082 				num_xfrms = 0;
2083 				drop_pols = num_pols;
2084 				goto no_transform;
2085 			}
2086 
2087 			dst_hold(&xdst->u.dst);
2088 
2089 			spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2090 			xdst->u.dst.next = xfrm_policy_sk_bundles;
2091 			xfrm_policy_sk_bundles = &xdst->u.dst;
2092 			spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2093 
2094 			route = xdst->route;
2095 		}
2096 	}
2097 
2098 	if (xdst == NULL) {
2099 		/* To accelerate a bit...  */
2100 		if ((dst_orig->flags & DST_NOXFRM) ||
2101 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
2102 			goto nopol;
2103 
2104 		flo = flow_cache_lookup(net, fl, family, dir,
2105 					xfrm_bundle_lookup, dst_orig);
2106 		if (flo == NULL)
2107 			goto nopol;
2108 		if (IS_ERR(flo)) {
2109 			err = PTR_ERR(flo);
2110 			goto dropdst;
2111 		}
2112 		xdst = container_of(flo, struct xfrm_dst, flo);
2113 
2114 		num_pols = xdst->num_pols;
2115 		num_xfrms = xdst->num_xfrms;
2116 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
2117 		route = xdst->route;
2118 	}
2119 
2120 	dst = &xdst->u.dst;
2121 	if (route == NULL && num_xfrms > 0) {
2122 		/* The only case when xfrm_bundle_lookup() returns a
2123 		 * bundle with null route, is when the template could
2124 		 * not be resolved. It means policies are there, but
2125 		 * bundle could not be created, since we don't yet
2126 		 * have the xfrm_state's. We need to wait for KM to
2127 		 * negotiate new SA's or bail out with error.*/
2128 		if (net->xfrm.sysctl_larval_drop) {
2129 			/* EREMOTE tells the caller to generate
2130 			 * a one-shot blackhole route. */
2131 			dst_release(dst);
2132 			xfrm_pols_put(pols, drop_pols);
2133 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2134 
2135 			return make_blackhole(net, family, dst_orig);
2136 		}
2137 		if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
2138 			DECLARE_WAITQUEUE(wait, current);
2139 
2140 			add_wait_queue(&net->xfrm.km_waitq, &wait);
2141 			set_current_state(TASK_INTERRUPTIBLE);
2142 			schedule();
2143 			set_current_state(TASK_RUNNING);
2144 			remove_wait_queue(&net->xfrm.km_waitq, &wait);
2145 
2146 			if (!signal_pending(current)) {
2147 				dst_release(dst);
2148 				goto restart;
2149 			}
2150 
2151 			err = -ERESTART;
2152 		} else
2153 			err = -EAGAIN;
2154 
2155 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2156 		goto error;
2157 	}
2158 
2159 no_transform:
2160 	if (num_pols == 0)
2161 		goto nopol;
2162 
2163 	if ((flags & XFRM_LOOKUP_ICMP) &&
2164 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2165 		err = -ENOENT;
2166 		goto error;
2167 	}
2168 
2169 	for (i = 0; i < num_pols; i++)
2170 		pols[i]->curlft.use_time = get_seconds();
2171 
2172 	if (num_xfrms < 0) {
2173 		/* Prohibit the flow */
2174 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2175 		err = -EPERM;
2176 		goto error;
2177 	} else if (num_xfrms > 0) {
2178 		/* Flow transformed */
2179 		dst_release(dst_orig);
2180 	} else {
2181 		/* Flow passes untransformed */
2182 		dst_release(dst);
2183 		dst = dst_orig;
2184 	}
2185 ok:
2186 	xfrm_pols_put(pols, drop_pols);
2187 	if (dst && dst->xfrm &&
2188 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2189 		dst->flags |= DST_XFRM_TUNNEL;
2190 	return dst;
2191 
2192 nopol:
2193 	if (!(flags & XFRM_LOOKUP_ICMP)) {
2194 		dst = dst_orig;
2195 		goto ok;
2196 	}
2197 	err = -ENOENT;
2198 error:
2199 	dst_release(dst);
2200 dropdst:
2201 	dst_release(dst_orig);
2202 	xfrm_pols_put(pols, drop_pols);
2203 	return ERR_PTR(err);
2204 }
2205 EXPORT_SYMBOL(xfrm_lookup);
2206 
2207 static inline int
2208 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2209 {
2210 	struct xfrm_state *x;
2211 
2212 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2213 		return 0;
2214 	x = skb->sp->xvec[idx];
2215 	if (!x->type->reject)
2216 		return 0;
2217 	return x->type->reject(x, skb, fl);
2218 }
2219 
2220 /* When skb is transformed back to its "native" form, we have to
2221  * check policy restrictions. At the moment we make this in maximally
2222  * stupid way. Shame on me. :-) Of course, connected sockets must
2223  * have policy cached at them.
2224  */
2225 
2226 static inline int
2227 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2228 	      unsigned short family)
2229 {
2230 	if (xfrm_state_kern(x))
2231 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2232 	return	x->id.proto == tmpl->id.proto &&
2233 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2234 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2235 		x->props.mode == tmpl->mode &&
2236 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2237 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2238 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
2239 		  xfrm_state_addr_cmp(tmpl, x, family));
2240 }
2241 
2242 /*
2243  * 0 or more than 0 is returned when validation is succeeded (either bypass
2244  * because of optional transport mode, or next index of the mathced secpath
2245  * state with the template.
2246  * -1 is returned when no matching template is found.
2247  * Otherwise "-2 - errored_index" is returned.
2248  */
2249 static inline int
2250 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2251 	       unsigned short family)
2252 {
2253 	int idx = start;
2254 
2255 	if (tmpl->optional) {
2256 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
2257 			return start;
2258 	} else
2259 		start = -1;
2260 	for (; idx < sp->len; idx++) {
2261 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2262 			return ++idx;
2263 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2264 			if (start == -1)
2265 				start = -2-idx;
2266 			break;
2267 		}
2268 	}
2269 	return start;
2270 }
2271 
2272 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2273 			  unsigned int family, int reverse)
2274 {
2275 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2276 	int err;
2277 
2278 	if (unlikely(afinfo == NULL))
2279 		return -EAFNOSUPPORT;
2280 
2281 	afinfo->decode_session(skb, fl, reverse);
2282 	err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2283 	xfrm_policy_put_afinfo(afinfo);
2284 	return err;
2285 }
2286 EXPORT_SYMBOL(__xfrm_decode_session);
2287 
2288 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2289 {
2290 	for (; k < sp->len; k++) {
2291 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2292 			*idxp = k;
2293 			return 1;
2294 		}
2295 	}
2296 
2297 	return 0;
2298 }
2299 
2300 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2301 			unsigned short family)
2302 {
2303 	struct net *net = dev_net(skb->dev);
2304 	struct xfrm_policy *pol;
2305 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2306 	int npols = 0;
2307 	int xfrm_nr;
2308 	int pi;
2309 	int reverse;
2310 	struct flowi fl;
2311 	u8 fl_dir;
2312 	int xerr_idx = -1;
2313 
2314 	reverse = dir & ~XFRM_POLICY_MASK;
2315 	dir &= XFRM_POLICY_MASK;
2316 	fl_dir = policy_to_flow_dir(dir);
2317 
2318 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2319 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2320 		return 0;
2321 	}
2322 
2323 	nf_nat_decode_session(skb, &fl, family);
2324 
2325 	/* First, check used SA against their selectors. */
2326 	if (skb->sp) {
2327 		int i;
2328 
2329 		for (i=skb->sp->len-1; i>=0; i--) {
2330 			struct xfrm_state *x = skb->sp->xvec[i];
2331 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
2332 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2333 				return 0;
2334 			}
2335 		}
2336 	}
2337 
2338 	pol = NULL;
2339 	if (sk && sk->sk_policy[dir]) {
2340 		pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2341 		if (IS_ERR(pol)) {
2342 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2343 			return 0;
2344 		}
2345 	}
2346 
2347 	if (!pol) {
2348 		struct flow_cache_object *flo;
2349 
2350 		flo = flow_cache_lookup(net, &fl, family, fl_dir,
2351 					xfrm_policy_lookup, NULL);
2352 		if (IS_ERR_OR_NULL(flo))
2353 			pol = ERR_CAST(flo);
2354 		else
2355 			pol = container_of(flo, struct xfrm_policy, flo);
2356 	}
2357 
2358 	if (IS_ERR(pol)) {
2359 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2360 		return 0;
2361 	}
2362 
2363 	if (!pol) {
2364 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2365 			xfrm_secpath_reject(xerr_idx, skb, &fl);
2366 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2367 			return 0;
2368 		}
2369 		return 1;
2370 	}
2371 
2372 	pol->curlft.use_time = get_seconds();
2373 
2374 	pols[0] = pol;
2375 	npols ++;
2376 #ifdef CONFIG_XFRM_SUB_POLICY
2377 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2378 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2379 						    &fl, family,
2380 						    XFRM_POLICY_IN);
2381 		if (pols[1]) {
2382 			if (IS_ERR(pols[1])) {
2383 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2384 				return 0;
2385 			}
2386 			pols[1]->curlft.use_time = get_seconds();
2387 			npols ++;
2388 		}
2389 	}
2390 #endif
2391 
2392 	if (pol->action == XFRM_POLICY_ALLOW) {
2393 		struct sec_path *sp;
2394 		static struct sec_path dummy;
2395 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2396 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2397 		struct xfrm_tmpl **tpp = tp;
2398 		int ti = 0;
2399 		int i, k;
2400 
2401 		if ((sp = skb->sp) == NULL)
2402 			sp = &dummy;
2403 
2404 		for (pi = 0; pi < npols; pi++) {
2405 			if (pols[pi] != pol &&
2406 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
2407 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2408 				goto reject;
2409 			}
2410 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2411 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2412 				goto reject_error;
2413 			}
2414 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
2415 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
2416 		}
2417 		xfrm_nr = ti;
2418 		if (npols > 1) {
2419 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2420 			tpp = stp;
2421 		}
2422 
2423 		/* For each tunnel xfrm, find the first matching tmpl.
2424 		 * For each tmpl before that, find corresponding xfrm.
2425 		 * Order is _important_. Later we will implement
2426 		 * some barriers, but at the moment barriers
2427 		 * are implied between each two transformations.
2428 		 */
2429 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2430 			k = xfrm_policy_ok(tpp[i], sp, k, family);
2431 			if (k < 0) {
2432 				if (k < -1)
2433 					/* "-2 - errored_index" returned */
2434 					xerr_idx = -(2+k);
2435 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2436 				goto reject;
2437 			}
2438 		}
2439 
2440 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2441 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2442 			goto reject;
2443 		}
2444 
2445 		xfrm_pols_put(pols, npols);
2446 		return 1;
2447 	}
2448 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2449 
2450 reject:
2451 	xfrm_secpath_reject(xerr_idx, skb, &fl);
2452 reject_error:
2453 	xfrm_pols_put(pols, npols);
2454 	return 0;
2455 }
2456 EXPORT_SYMBOL(__xfrm_policy_check);
2457 
2458 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2459 {
2460 	struct net *net = dev_net(skb->dev);
2461 	struct flowi fl;
2462 	struct dst_entry *dst;
2463 	int res = 1;
2464 
2465 	if (xfrm_decode_session(skb, &fl, family) < 0) {
2466 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2467 		return 0;
2468 	}
2469 
2470 	skb_dst_force(skb);
2471 
2472 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2473 	if (IS_ERR(dst)) {
2474 		res = 0;
2475 		dst = NULL;
2476 	}
2477 	skb_dst_set(skb, dst);
2478 	return res;
2479 }
2480 EXPORT_SYMBOL(__xfrm_route_forward);
2481 
2482 /* Optimize later using cookies and generation ids. */
2483 
2484 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2485 {
2486 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2487 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2488 	 * get validated by dst_ops->check on every use.  We do this
2489 	 * because when a normal route referenced by an XFRM dst is
2490 	 * obsoleted we do not go looking around for all parent
2491 	 * referencing XFRM dsts so that we can invalidate them.  It
2492 	 * is just too much work.  Instead we make the checks here on
2493 	 * every use.  For example:
2494 	 *
2495 	 *	XFRM dst A --> IPv4 dst X
2496 	 *
2497 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
2498 	 * in this example).  If X is marked obsolete, "A" will not
2499 	 * notice.  That's what we are validating here via the
2500 	 * stale_bundle() check.
2501 	 *
2502 	 * When a policy's bundle is pruned, we dst_free() the XFRM
2503 	 * dst which causes it's ->obsolete field to be set to
2504 	 * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2505 	 * this, we want to force a new route lookup.
2506 	 */
2507 	if (dst->obsolete < 0 && !stale_bundle(dst))
2508 		return dst;
2509 
2510 	return NULL;
2511 }
2512 
2513 static int stale_bundle(struct dst_entry *dst)
2514 {
2515 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2516 }
2517 
2518 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2519 {
2520 	while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2521 		dst->dev = dev_net(dev)->loopback_dev;
2522 		dev_hold(dst->dev);
2523 		dev_put(dev);
2524 	}
2525 }
2526 EXPORT_SYMBOL(xfrm_dst_ifdown);
2527 
2528 static void xfrm_link_failure(struct sk_buff *skb)
2529 {
2530 	/* Impossible. Such dst must be popped before reaches point of failure. */
2531 }
2532 
2533 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2534 {
2535 	if (dst) {
2536 		if (dst->obsolete) {
2537 			dst_release(dst);
2538 			dst = NULL;
2539 		}
2540 	}
2541 	return dst;
2542 }
2543 
2544 static void __xfrm_garbage_collect(struct net *net)
2545 {
2546 	struct dst_entry *head, *next;
2547 
2548 	spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2549 	head = xfrm_policy_sk_bundles;
2550 	xfrm_policy_sk_bundles = NULL;
2551 	spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2552 
2553 	while (head) {
2554 		next = head->next;
2555 		dst_free(head);
2556 		head = next;
2557 	}
2558 }
2559 
2560 void xfrm_garbage_collect(struct net *net)
2561 {
2562 	flow_cache_flush();
2563 	__xfrm_garbage_collect(net);
2564 }
2565 EXPORT_SYMBOL(xfrm_garbage_collect);
2566 
2567 static void xfrm_garbage_collect_deferred(struct net *net)
2568 {
2569 	flow_cache_flush_deferred();
2570 	__xfrm_garbage_collect(net);
2571 }
2572 
2573 static void xfrm_init_pmtu(struct dst_entry *dst)
2574 {
2575 	do {
2576 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2577 		u32 pmtu, route_mtu_cached;
2578 
2579 		pmtu = dst_mtu(dst->child);
2580 		xdst->child_mtu_cached = pmtu;
2581 
2582 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2583 
2584 		route_mtu_cached = dst_mtu(xdst->route);
2585 		xdst->route_mtu_cached = route_mtu_cached;
2586 
2587 		if (pmtu > route_mtu_cached)
2588 			pmtu = route_mtu_cached;
2589 
2590 		dst_metric_set(dst, RTAX_MTU, pmtu);
2591 	} while ((dst = dst->next));
2592 }
2593 
2594 /* Check that the bundle accepts the flow and its components are
2595  * still valid.
2596  */
2597 
2598 static int xfrm_bundle_ok(struct xfrm_dst *first)
2599 {
2600 	struct dst_entry *dst = &first->u.dst;
2601 	struct xfrm_dst *last;
2602 	u32 mtu;
2603 
2604 	if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2605 	    (dst->dev && !netif_running(dst->dev)))
2606 		return 0;
2607 
2608 	if (dst->flags & DST_XFRM_QUEUE)
2609 		return 1;
2610 
2611 	last = NULL;
2612 
2613 	do {
2614 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2615 
2616 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2617 			return 0;
2618 		if (xdst->xfrm_genid != dst->xfrm->genid)
2619 			return 0;
2620 		if (xdst->num_pols > 0 &&
2621 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2622 			return 0;
2623 
2624 		mtu = dst_mtu(dst->child);
2625 		if (xdst->child_mtu_cached != mtu) {
2626 			last = xdst;
2627 			xdst->child_mtu_cached = mtu;
2628 		}
2629 
2630 		if (!dst_check(xdst->route, xdst->route_cookie))
2631 			return 0;
2632 		mtu = dst_mtu(xdst->route);
2633 		if (xdst->route_mtu_cached != mtu) {
2634 			last = xdst;
2635 			xdst->route_mtu_cached = mtu;
2636 		}
2637 
2638 		dst = dst->child;
2639 	} while (dst->xfrm);
2640 
2641 	if (likely(!last))
2642 		return 1;
2643 
2644 	mtu = last->child_mtu_cached;
2645 	for (;;) {
2646 		dst = &last->u.dst;
2647 
2648 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2649 		if (mtu > last->route_mtu_cached)
2650 			mtu = last->route_mtu_cached;
2651 		dst_metric_set(dst, RTAX_MTU, mtu);
2652 
2653 		if (last == first)
2654 			break;
2655 
2656 		last = (struct xfrm_dst *)last->u.dst.next;
2657 		last->child_mtu_cached = mtu;
2658 	}
2659 
2660 	return 1;
2661 }
2662 
2663 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2664 {
2665 	return dst_metric_advmss(dst->path);
2666 }
2667 
2668 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2669 {
2670 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2671 
2672 	return mtu ? : dst_mtu(dst->path);
2673 }
2674 
2675 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2676 					   struct sk_buff *skb,
2677 					   const void *daddr)
2678 {
2679 	return dst->path->ops->neigh_lookup(dst, skb, daddr);
2680 }
2681 
2682 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2683 {
2684 	struct net *net;
2685 	int err = 0;
2686 	if (unlikely(afinfo == NULL))
2687 		return -EINVAL;
2688 	if (unlikely(afinfo->family >= NPROTO))
2689 		return -EAFNOSUPPORT;
2690 	spin_lock(&xfrm_policy_afinfo_lock);
2691 	if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2692 		err = -ENOBUFS;
2693 	else {
2694 		struct dst_ops *dst_ops = afinfo->dst_ops;
2695 		if (likely(dst_ops->kmem_cachep == NULL))
2696 			dst_ops->kmem_cachep = xfrm_dst_cache;
2697 		if (likely(dst_ops->check == NULL))
2698 			dst_ops->check = xfrm_dst_check;
2699 		if (likely(dst_ops->default_advmss == NULL))
2700 			dst_ops->default_advmss = xfrm_default_advmss;
2701 		if (likely(dst_ops->mtu == NULL))
2702 			dst_ops->mtu = xfrm_mtu;
2703 		if (likely(dst_ops->negative_advice == NULL))
2704 			dst_ops->negative_advice = xfrm_negative_advice;
2705 		if (likely(dst_ops->link_failure == NULL))
2706 			dst_ops->link_failure = xfrm_link_failure;
2707 		if (likely(dst_ops->neigh_lookup == NULL))
2708 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
2709 		if (likely(afinfo->garbage_collect == NULL))
2710 			afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2711 		rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2712 	}
2713 	spin_unlock(&xfrm_policy_afinfo_lock);
2714 
2715 	rtnl_lock();
2716 	for_each_net(net) {
2717 		struct dst_ops *xfrm_dst_ops;
2718 
2719 		switch (afinfo->family) {
2720 		case AF_INET:
2721 			xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2722 			break;
2723 #if IS_ENABLED(CONFIG_IPV6)
2724 		case AF_INET6:
2725 			xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2726 			break;
2727 #endif
2728 		default:
2729 			BUG();
2730 		}
2731 		*xfrm_dst_ops = *afinfo->dst_ops;
2732 	}
2733 	rtnl_unlock();
2734 
2735 	return err;
2736 }
2737 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2738 
2739 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2740 {
2741 	int err = 0;
2742 	if (unlikely(afinfo == NULL))
2743 		return -EINVAL;
2744 	if (unlikely(afinfo->family >= NPROTO))
2745 		return -EAFNOSUPPORT;
2746 	spin_lock(&xfrm_policy_afinfo_lock);
2747 	if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2748 		if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2749 			err = -EINVAL;
2750 		else
2751 			RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2752 					 NULL);
2753 	}
2754 	spin_unlock(&xfrm_policy_afinfo_lock);
2755 	if (!err) {
2756 		struct dst_ops *dst_ops = afinfo->dst_ops;
2757 
2758 		synchronize_rcu();
2759 
2760 		dst_ops->kmem_cachep = NULL;
2761 		dst_ops->check = NULL;
2762 		dst_ops->negative_advice = NULL;
2763 		dst_ops->link_failure = NULL;
2764 		afinfo->garbage_collect = NULL;
2765 	}
2766 	return err;
2767 }
2768 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2769 
2770 static void __net_init xfrm_dst_ops_init(struct net *net)
2771 {
2772 	struct xfrm_policy_afinfo *afinfo;
2773 
2774 	rcu_read_lock();
2775 	afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2776 	if (afinfo)
2777 		net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2778 #if IS_ENABLED(CONFIG_IPV6)
2779 	afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2780 	if (afinfo)
2781 		net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2782 #endif
2783 	rcu_read_unlock();
2784 }
2785 
2786 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2787 {
2788 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2789 
2790 	switch (event) {
2791 	case NETDEV_DOWN:
2792 		xfrm_garbage_collect(dev_net(dev));
2793 	}
2794 	return NOTIFY_DONE;
2795 }
2796 
2797 static struct notifier_block xfrm_dev_notifier = {
2798 	.notifier_call	= xfrm_dev_event,
2799 };
2800 
2801 #ifdef CONFIG_XFRM_STATISTICS
2802 static int __net_init xfrm_statistics_init(struct net *net)
2803 {
2804 	int rv;
2805 
2806 	if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2807 			  sizeof(struct linux_xfrm_mib),
2808 			  __alignof__(struct linux_xfrm_mib)) < 0)
2809 		return -ENOMEM;
2810 	rv = xfrm_proc_init(net);
2811 	if (rv < 0)
2812 		snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2813 	return rv;
2814 }
2815 
2816 static void xfrm_statistics_fini(struct net *net)
2817 {
2818 	xfrm_proc_fini(net);
2819 	snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2820 }
2821 #else
2822 static int __net_init xfrm_statistics_init(struct net *net)
2823 {
2824 	return 0;
2825 }
2826 
2827 static void xfrm_statistics_fini(struct net *net)
2828 {
2829 }
2830 #endif
2831 
2832 static int __net_init xfrm_policy_init(struct net *net)
2833 {
2834 	unsigned int hmask, sz;
2835 	int dir;
2836 
2837 	if (net_eq(net, &init_net))
2838 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2839 					   sizeof(struct xfrm_dst),
2840 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2841 					   NULL);
2842 
2843 	hmask = 8 - 1;
2844 	sz = (hmask+1) * sizeof(struct hlist_head);
2845 
2846 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2847 	if (!net->xfrm.policy_byidx)
2848 		goto out_byidx;
2849 	net->xfrm.policy_idx_hmask = hmask;
2850 
2851 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2852 		struct xfrm_policy_hash *htab;
2853 
2854 		net->xfrm.policy_count[dir] = 0;
2855 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2856 
2857 		htab = &net->xfrm.policy_bydst[dir];
2858 		htab->table = xfrm_hash_alloc(sz);
2859 		if (!htab->table)
2860 			goto out_bydst;
2861 		htab->hmask = hmask;
2862 	}
2863 
2864 	INIT_LIST_HEAD(&net->xfrm.policy_all);
2865 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2866 	if (net_eq(net, &init_net))
2867 		register_netdevice_notifier(&xfrm_dev_notifier);
2868 	return 0;
2869 
2870 out_bydst:
2871 	for (dir--; dir >= 0; dir--) {
2872 		struct xfrm_policy_hash *htab;
2873 
2874 		htab = &net->xfrm.policy_bydst[dir];
2875 		xfrm_hash_free(htab->table, sz);
2876 	}
2877 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2878 out_byidx:
2879 	return -ENOMEM;
2880 }
2881 
2882 static void xfrm_policy_fini(struct net *net)
2883 {
2884 	struct xfrm_audit audit_info;
2885 	unsigned int sz;
2886 	int dir;
2887 
2888 	flush_work(&net->xfrm.policy_hash_work);
2889 #ifdef CONFIG_XFRM_SUB_POLICY
2890 	audit_info.loginuid = INVALID_UID;
2891 	audit_info.sessionid = -1;
2892 	audit_info.secid = 0;
2893 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2894 #endif
2895 	audit_info.loginuid = INVALID_UID;
2896 	audit_info.sessionid = -1;
2897 	audit_info.secid = 0;
2898 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2899 
2900 	WARN_ON(!list_empty(&net->xfrm.policy_all));
2901 
2902 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2903 		struct xfrm_policy_hash *htab;
2904 
2905 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2906 
2907 		htab = &net->xfrm.policy_bydst[dir];
2908 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2909 		WARN_ON(!hlist_empty(htab->table));
2910 		xfrm_hash_free(htab->table, sz);
2911 	}
2912 
2913 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2914 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2915 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2916 }
2917 
2918 static int __net_init xfrm_net_init(struct net *net)
2919 {
2920 	int rv;
2921 
2922 	rv = xfrm_statistics_init(net);
2923 	if (rv < 0)
2924 		goto out_statistics;
2925 	rv = xfrm_state_init(net);
2926 	if (rv < 0)
2927 		goto out_state;
2928 	rv = xfrm_policy_init(net);
2929 	if (rv < 0)
2930 		goto out_policy;
2931 	xfrm_dst_ops_init(net);
2932 	rv = xfrm_sysctl_init(net);
2933 	if (rv < 0)
2934 		goto out_sysctl;
2935 	return 0;
2936 
2937 out_sysctl:
2938 	xfrm_policy_fini(net);
2939 out_policy:
2940 	xfrm_state_fini(net);
2941 out_state:
2942 	xfrm_statistics_fini(net);
2943 out_statistics:
2944 	return rv;
2945 }
2946 
2947 static void __net_exit xfrm_net_exit(struct net *net)
2948 {
2949 	xfrm_sysctl_fini(net);
2950 	xfrm_policy_fini(net);
2951 	xfrm_state_fini(net);
2952 	xfrm_statistics_fini(net);
2953 }
2954 
2955 static struct pernet_operations __net_initdata xfrm_net_ops = {
2956 	.init = xfrm_net_init,
2957 	.exit = xfrm_net_exit,
2958 };
2959 
2960 void __init xfrm_init(void)
2961 {
2962 	register_pernet_subsys(&xfrm_net_ops);
2963 	xfrm_input_init();
2964 }
2965 
2966 #ifdef CONFIG_AUDITSYSCALL
2967 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2968 					 struct audit_buffer *audit_buf)
2969 {
2970 	struct xfrm_sec_ctx *ctx = xp->security;
2971 	struct xfrm_selector *sel = &xp->selector;
2972 
2973 	if (ctx)
2974 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2975 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2976 
2977 	switch(sel->family) {
2978 	case AF_INET:
2979 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2980 		if (sel->prefixlen_s != 32)
2981 			audit_log_format(audit_buf, " src_prefixlen=%d",
2982 					 sel->prefixlen_s);
2983 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2984 		if (sel->prefixlen_d != 32)
2985 			audit_log_format(audit_buf, " dst_prefixlen=%d",
2986 					 sel->prefixlen_d);
2987 		break;
2988 	case AF_INET6:
2989 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2990 		if (sel->prefixlen_s != 128)
2991 			audit_log_format(audit_buf, " src_prefixlen=%d",
2992 					 sel->prefixlen_s);
2993 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2994 		if (sel->prefixlen_d != 128)
2995 			audit_log_format(audit_buf, " dst_prefixlen=%d",
2996 					 sel->prefixlen_d);
2997 		break;
2998 	}
2999 }
3000 
3001 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
3002 			   kuid_t auid, u32 sessionid, u32 secid)
3003 {
3004 	struct audit_buffer *audit_buf;
3005 
3006 	audit_buf = xfrm_audit_start("SPD-add");
3007 	if (audit_buf == NULL)
3008 		return;
3009 	xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3010 	audit_log_format(audit_buf, " res=%u", result);
3011 	xfrm_audit_common_policyinfo(xp, audit_buf);
3012 	audit_log_end(audit_buf);
3013 }
3014 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3015 
3016 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3017 			      kuid_t auid, u32 sessionid, u32 secid)
3018 {
3019 	struct audit_buffer *audit_buf;
3020 
3021 	audit_buf = xfrm_audit_start("SPD-delete");
3022 	if (audit_buf == NULL)
3023 		return;
3024 	xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3025 	audit_log_format(audit_buf, " res=%u", result);
3026 	xfrm_audit_common_policyinfo(xp, audit_buf);
3027 	audit_log_end(audit_buf);
3028 }
3029 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3030 #endif
3031 
3032 #ifdef CONFIG_XFRM_MIGRATE
3033 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3034 					const struct xfrm_selector *sel_tgt)
3035 {
3036 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3037 		if (sel_tgt->family == sel_cmp->family &&
3038 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3039 				    sel_cmp->family) &&
3040 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3041 				    sel_cmp->family) &&
3042 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3043 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3044 			return true;
3045 		}
3046 	} else {
3047 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3048 			return true;
3049 		}
3050 	}
3051 	return false;
3052 }
3053 
3054 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3055 						     u8 dir, u8 type)
3056 {
3057 	struct xfrm_policy *pol, *ret = NULL;
3058 	struct hlist_head *chain;
3059 	u32 priority = ~0U;
3060 
3061 	read_lock_bh(&xfrm_policy_lock);
3062 	chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
3063 	hlist_for_each_entry(pol, chain, bydst) {
3064 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3065 		    pol->type == type) {
3066 			ret = pol;
3067 			priority = ret->priority;
3068 			break;
3069 		}
3070 	}
3071 	chain = &init_net.xfrm.policy_inexact[dir];
3072 	hlist_for_each_entry(pol, chain, bydst) {
3073 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3074 		    pol->type == type &&
3075 		    pol->priority < priority) {
3076 			ret = pol;
3077 			break;
3078 		}
3079 	}
3080 
3081 	if (ret)
3082 		xfrm_pol_hold(ret);
3083 
3084 	read_unlock_bh(&xfrm_policy_lock);
3085 
3086 	return ret;
3087 }
3088 
3089 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3090 {
3091 	int match = 0;
3092 
3093 	if (t->mode == m->mode && t->id.proto == m->proto &&
3094 	    (m->reqid == 0 || t->reqid == m->reqid)) {
3095 		switch (t->mode) {
3096 		case XFRM_MODE_TUNNEL:
3097 		case XFRM_MODE_BEET:
3098 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3099 					    m->old_family) &&
3100 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
3101 					    m->old_family)) {
3102 				match = 1;
3103 			}
3104 			break;
3105 		case XFRM_MODE_TRANSPORT:
3106 			/* in case of transport mode, template does not store
3107 			   any IP addresses, hence we just compare mode and
3108 			   protocol */
3109 			match = 1;
3110 			break;
3111 		default:
3112 			break;
3113 		}
3114 	}
3115 	return match;
3116 }
3117 
3118 /* update endpoint address(es) of template(s) */
3119 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3120 			       struct xfrm_migrate *m, int num_migrate)
3121 {
3122 	struct xfrm_migrate *mp;
3123 	int i, j, n = 0;
3124 
3125 	write_lock_bh(&pol->lock);
3126 	if (unlikely(pol->walk.dead)) {
3127 		/* target policy has been deleted */
3128 		write_unlock_bh(&pol->lock);
3129 		return -ENOENT;
3130 	}
3131 
3132 	for (i = 0; i < pol->xfrm_nr; i++) {
3133 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3134 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3135 				continue;
3136 			n++;
3137 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3138 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3139 				continue;
3140 			/* update endpoints */
3141 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3142 			       sizeof(pol->xfrm_vec[i].id.daddr));
3143 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3144 			       sizeof(pol->xfrm_vec[i].saddr));
3145 			pol->xfrm_vec[i].encap_family = mp->new_family;
3146 			/* flush bundles */
3147 			atomic_inc(&pol->genid);
3148 		}
3149 	}
3150 
3151 	write_unlock_bh(&pol->lock);
3152 
3153 	if (!n)
3154 		return -ENODATA;
3155 
3156 	return 0;
3157 }
3158 
3159 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3160 {
3161 	int i, j;
3162 
3163 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3164 		return -EINVAL;
3165 
3166 	for (i = 0; i < num_migrate; i++) {
3167 		if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3168 				    m[i].old_family) &&
3169 		    xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3170 				    m[i].old_family))
3171 			return -EINVAL;
3172 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3173 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3174 			return -EINVAL;
3175 
3176 		/* check if there is any duplicated entry */
3177 		for (j = i + 1; j < num_migrate; j++) {
3178 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3179 				    sizeof(m[i].old_daddr)) &&
3180 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3181 				    sizeof(m[i].old_saddr)) &&
3182 			    m[i].proto == m[j].proto &&
3183 			    m[i].mode == m[j].mode &&
3184 			    m[i].reqid == m[j].reqid &&
3185 			    m[i].old_family == m[j].old_family)
3186 				return -EINVAL;
3187 		}
3188 	}
3189 
3190 	return 0;
3191 }
3192 
3193 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3194 		 struct xfrm_migrate *m, int num_migrate,
3195 		 struct xfrm_kmaddress *k)
3196 {
3197 	int i, err, nx_cur = 0, nx_new = 0;
3198 	struct xfrm_policy *pol = NULL;
3199 	struct xfrm_state *x, *xc;
3200 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3201 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3202 	struct xfrm_migrate *mp;
3203 
3204 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3205 		goto out;
3206 
3207 	/* Stage 1 - find policy */
3208 	if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
3209 		err = -ENOENT;
3210 		goto out;
3211 	}
3212 
3213 	/* Stage 2 - find and update state(s) */
3214 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3215 		if ((x = xfrm_migrate_state_find(mp))) {
3216 			x_cur[nx_cur] = x;
3217 			nx_cur++;
3218 			if ((xc = xfrm_state_migrate(x, mp))) {
3219 				x_new[nx_new] = xc;
3220 				nx_new++;
3221 			} else {
3222 				err = -ENODATA;
3223 				goto restore_state;
3224 			}
3225 		}
3226 	}
3227 
3228 	/* Stage 3 - update policy */
3229 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3230 		goto restore_state;
3231 
3232 	/* Stage 4 - delete old state(s) */
3233 	if (nx_cur) {
3234 		xfrm_states_put(x_cur, nx_cur);
3235 		xfrm_states_delete(x_cur, nx_cur);
3236 	}
3237 
3238 	/* Stage 5 - announce */
3239 	km_migrate(sel, dir, type, m, num_migrate, k);
3240 
3241 	xfrm_pol_put(pol);
3242 
3243 	return 0;
3244 out:
3245 	return err;
3246 
3247 restore_state:
3248 	if (pol)
3249 		xfrm_pol_put(pol);
3250 	if (nx_cur)
3251 		xfrm_states_put(x_cur, nx_cur);
3252 	if (nx_new)
3253 		xfrm_states_delete(x_new, nx_new);
3254 
3255 	return err;
3256 }
3257 EXPORT_SYMBOL(xfrm_migrate);
3258 #endif
3259