xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 2eb3ed33e55d003d721d4d1a5e72fe323c12b4c0)
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/cpu.h>
28 #include <linux/audit.h>
29 #include <net/dst.h>
30 #include <net/flow.h>
31 #include <net/xfrm.h>
32 #include <net/ip.h>
33 #ifdef CONFIG_XFRM_STATISTICS
34 #include <net/snmp.h>
35 #endif
36 
37 #include "xfrm_hash.h"
38 
39 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
40 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
41 #define XFRM_MAX_QUEUE_LEN	100
42 
43 struct xfrm_flo {
44 	struct dst_entry *dst_orig;
45 	u8 flags;
46 };
47 
48 static DEFINE_PER_CPU(struct xfrm_dst *, xfrm_last_dst);
49 static struct work_struct *xfrm_pcpu_work __read_mostly;
50 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
51 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
52 						__read_mostly;
53 
54 static struct kmem_cache *xfrm_dst_cache __read_mostly;
55 static __read_mostly seqcount_t xfrm_policy_hash_generation;
56 
57 static void xfrm_init_pmtu(struct dst_entry *dst);
58 static int stale_bundle(struct dst_entry *dst);
59 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
60 static void xfrm_policy_queue_process(struct timer_list *t);
61 
62 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
63 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
64 						int dir);
65 
66 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
67 {
68 	return refcount_inc_not_zero(&policy->refcnt);
69 }
70 
71 static inline bool
72 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
73 {
74 	const struct flowi4 *fl4 = &fl->u.ip4;
75 
76 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
77 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
78 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
79 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
80 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
81 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
82 }
83 
84 static inline bool
85 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
86 {
87 	const struct flowi6 *fl6 = &fl->u.ip6;
88 
89 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
90 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
91 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
92 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
93 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
94 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
95 }
96 
97 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
98 			 unsigned short family)
99 {
100 	switch (family) {
101 	case AF_INET:
102 		return __xfrm4_selector_match(sel, fl);
103 	case AF_INET6:
104 		return __xfrm6_selector_match(sel, fl);
105 	}
106 	return false;
107 }
108 
109 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
110 {
111 	const struct xfrm_policy_afinfo *afinfo;
112 
113 	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
114 		return NULL;
115 	rcu_read_lock();
116 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
117 	if (unlikely(!afinfo))
118 		rcu_read_unlock();
119 	return afinfo;
120 }
121 
122 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
123 				    const xfrm_address_t *saddr,
124 				    const xfrm_address_t *daddr,
125 				    int family, u32 mark)
126 {
127 	const struct xfrm_policy_afinfo *afinfo;
128 	struct dst_entry *dst;
129 
130 	afinfo = xfrm_policy_get_afinfo(family);
131 	if (unlikely(afinfo == NULL))
132 		return ERR_PTR(-EAFNOSUPPORT);
133 
134 	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
135 
136 	rcu_read_unlock();
137 
138 	return dst;
139 }
140 EXPORT_SYMBOL(__xfrm_dst_lookup);
141 
142 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
143 						int tos, int oif,
144 						xfrm_address_t *prev_saddr,
145 						xfrm_address_t *prev_daddr,
146 						int family, u32 mark)
147 {
148 	struct net *net = xs_net(x);
149 	xfrm_address_t *saddr = &x->props.saddr;
150 	xfrm_address_t *daddr = &x->id.daddr;
151 	struct dst_entry *dst;
152 
153 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
154 		saddr = x->coaddr;
155 		daddr = prev_daddr;
156 	}
157 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
158 		saddr = prev_saddr;
159 		daddr = x->coaddr;
160 	}
161 
162 	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
163 
164 	if (!IS_ERR(dst)) {
165 		if (prev_saddr != saddr)
166 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
167 		if (prev_daddr != daddr)
168 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
169 	}
170 
171 	return dst;
172 }
173 
174 static inline unsigned long make_jiffies(long secs)
175 {
176 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
177 		return MAX_SCHEDULE_TIMEOUT-1;
178 	else
179 		return secs*HZ;
180 }
181 
182 static void xfrm_policy_timer(struct timer_list *t)
183 {
184 	struct xfrm_policy *xp = from_timer(xp, t, timer);
185 	unsigned long now = get_seconds();
186 	long next = LONG_MAX;
187 	int warn = 0;
188 	int dir;
189 
190 	read_lock(&xp->lock);
191 
192 	if (unlikely(xp->walk.dead))
193 		goto out;
194 
195 	dir = xfrm_policy_id2dir(xp->index);
196 
197 	if (xp->lft.hard_add_expires_seconds) {
198 		long tmo = xp->lft.hard_add_expires_seconds +
199 			xp->curlft.add_time - now;
200 		if (tmo <= 0)
201 			goto expired;
202 		if (tmo < next)
203 			next = tmo;
204 	}
205 	if (xp->lft.hard_use_expires_seconds) {
206 		long tmo = xp->lft.hard_use_expires_seconds +
207 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
208 		if (tmo <= 0)
209 			goto expired;
210 		if (tmo < next)
211 			next = tmo;
212 	}
213 	if (xp->lft.soft_add_expires_seconds) {
214 		long tmo = xp->lft.soft_add_expires_seconds +
215 			xp->curlft.add_time - now;
216 		if (tmo <= 0) {
217 			warn = 1;
218 			tmo = XFRM_KM_TIMEOUT;
219 		}
220 		if (tmo < next)
221 			next = tmo;
222 	}
223 	if (xp->lft.soft_use_expires_seconds) {
224 		long tmo = xp->lft.soft_use_expires_seconds +
225 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
226 		if (tmo <= 0) {
227 			warn = 1;
228 			tmo = XFRM_KM_TIMEOUT;
229 		}
230 		if (tmo < next)
231 			next = tmo;
232 	}
233 
234 	if (warn)
235 		km_policy_expired(xp, dir, 0, 0);
236 	if (next != LONG_MAX &&
237 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
238 		xfrm_pol_hold(xp);
239 
240 out:
241 	read_unlock(&xp->lock);
242 	xfrm_pol_put(xp);
243 	return;
244 
245 expired:
246 	read_unlock(&xp->lock);
247 	if (!xfrm_policy_delete(xp, dir))
248 		km_policy_expired(xp, dir, 1, 0);
249 	xfrm_pol_put(xp);
250 }
251 
252 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
253  * SPD calls.
254  */
255 
256 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
257 {
258 	struct xfrm_policy *policy;
259 
260 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
261 
262 	if (policy) {
263 		write_pnet(&policy->xp_net, net);
264 		INIT_LIST_HEAD(&policy->walk.all);
265 		INIT_HLIST_NODE(&policy->bydst);
266 		INIT_HLIST_NODE(&policy->byidx);
267 		rwlock_init(&policy->lock);
268 		refcount_set(&policy->refcnt, 1);
269 		skb_queue_head_init(&policy->polq.hold_queue);
270 		timer_setup(&policy->timer, xfrm_policy_timer, 0);
271 		timer_setup(&policy->polq.hold_timer,
272 			    xfrm_policy_queue_process, 0);
273 	}
274 	return policy;
275 }
276 EXPORT_SYMBOL(xfrm_policy_alloc);
277 
278 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
279 {
280 	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
281 
282 	security_xfrm_policy_free(policy->security);
283 	kfree(policy);
284 }
285 
286 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
287 
288 void xfrm_policy_destroy(struct xfrm_policy *policy)
289 {
290 	BUG_ON(!policy->walk.dead);
291 
292 	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
293 		BUG();
294 
295 	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
296 }
297 EXPORT_SYMBOL(xfrm_policy_destroy);
298 
299 /* Rule must be locked. Release descendant resources, announce
300  * entry dead. The rule must be unlinked from lists to the moment.
301  */
302 
303 static void xfrm_policy_kill(struct xfrm_policy *policy)
304 {
305 	policy->walk.dead = 1;
306 
307 	atomic_inc(&policy->genid);
308 
309 	if (del_timer(&policy->polq.hold_timer))
310 		xfrm_pol_put(policy);
311 	skb_queue_purge(&policy->polq.hold_queue);
312 
313 	if (del_timer(&policy->timer))
314 		xfrm_pol_put(policy);
315 
316 	xfrm_pol_put(policy);
317 }
318 
319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
320 
321 static inline unsigned int idx_hash(struct net *net, u32 index)
322 {
323 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
324 }
325 
326 /* calculate policy hash thresholds */
327 static void __get_hash_thresh(struct net *net,
328 			      unsigned short family, int dir,
329 			      u8 *dbits, u8 *sbits)
330 {
331 	switch (family) {
332 	case AF_INET:
333 		*dbits = net->xfrm.policy_bydst[dir].dbits4;
334 		*sbits = net->xfrm.policy_bydst[dir].sbits4;
335 		break;
336 
337 	case AF_INET6:
338 		*dbits = net->xfrm.policy_bydst[dir].dbits6;
339 		*sbits = net->xfrm.policy_bydst[dir].sbits6;
340 		break;
341 
342 	default:
343 		*dbits = 0;
344 		*sbits = 0;
345 	}
346 }
347 
348 static struct hlist_head *policy_hash_bysel(struct net *net,
349 					    const struct xfrm_selector *sel,
350 					    unsigned short family, int dir)
351 {
352 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
353 	unsigned int hash;
354 	u8 dbits;
355 	u8 sbits;
356 
357 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
358 	hash = __sel_hash(sel, family, hmask, dbits, sbits);
359 
360 	if (hash == hmask + 1)
361 		return &net->xfrm.policy_inexact[dir];
362 
363 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
364 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + 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;
374 	u8 dbits;
375 	u8 sbits;
376 
377 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
378 	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
379 
380 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
381 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
382 }
383 
384 static void xfrm_dst_hash_transfer(struct net *net,
385 				   struct hlist_head *list,
386 				   struct hlist_head *ndsttable,
387 				   unsigned int nhashmask,
388 				   int dir)
389 {
390 	struct hlist_node *tmp, *entry0 = NULL;
391 	struct xfrm_policy *pol;
392 	unsigned int h0 = 0;
393 	u8 dbits;
394 	u8 sbits;
395 
396 redo:
397 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
398 		unsigned int h;
399 
400 		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
401 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
402 				pol->family, nhashmask, dbits, sbits);
403 		if (!entry0) {
404 			hlist_del_rcu(&pol->bydst);
405 			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
406 			h0 = h;
407 		} else {
408 			if (h != h0)
409 				continue;
410 			hlist_del_rcu(&pol->bydst);
411 			hlist_add_behind_rcu(&pol->bydst, entry0);
412 		}
413 		entry0 = &pol->bydst;
414 	}
415 	if (!hlist_empty(list)) {
416 		entry0 = NULL;
417 		goto redo;
418 	}
419 }
420 
421 static void xfrm_idx_hash_transfer(struct hlist_head *list,
422 				   struct hlist_head *nidxtable,
423 				   unsigned int nhashmask)
424 {
425 	struct hlist_node *tmp;
426 	struct xfrm_policy *pol;
427 
428 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
429 		unsigned int h;
430 
431 		h = __idx_hash(pol->index, nhashmask);
432 		hlist_add_head(&pol->byidx, nidxtable+h);
433 	}
434 }
435 
436 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
437 {
438 	return ((old_hmask + 1) << 1) - 1;
439 }
440 
441 static void xfrm_bydst_resize(struct net *net, int dir)
442 {
443 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
444 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
445 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
446 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
447 	struct hlist_head *odst;
448 	int i;
449 
450 	if (!ndst)
451 		return;
452 
453 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
454 	write_seqcount_begin(&xfrm_policy_hash_generation);
455 
456 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
457 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
458 
459 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
460 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
461 
462 	for (i = hmask; i >= 0; i--)
463 		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
464 
465 	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
466 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
467 
468 	write_seqcount_end(&xfrm_policy_hash_generation);
469 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
470 
471 	synchronize_rcu();
472 
473 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
474 }
475 
476 static void xfrm_byidx_resize(struct net *net, int total)
477 {
478 	unsigned int hmask = net->xfrm.policy_idx_hmask;
479 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
480 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
481 	struct hlist_head *oidx = net->xfrm.policy_byidx;
482 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
483 	int i;
484 
485 	if (!nidx)
486 		return;
487 
488 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
489 
490 	for (i = hmask; i >= 0; i--)
491 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
492 
493 	net->xfrm.policy_byidx = nidx;
494 	net->xfrm.policy_idx_hmask = nhashmask;
495 
496 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
497 
498 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
499 }
500 
501 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
502 {
503 	unsigned int cnt = net->xfrm.policy_count[dir];
504 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
505 
506 	if (total)
507 		*total += cnt;
508 
509 	if ((hmask + 1) < xfrm_policy_hashmax &&
510 	    cnt > hmask)
511 		return 1;
512 
513 	return 0;
514 }
515 
516 static inline int xfrm_byidx_should_resize(struct net *net, int total)
517 {
518 	unsigned int hmask = net->xfrm.policy_idx_hmask;
519 
520 	if ((hmask + 1) < xfrm_policy_hashmax &&
521 	    total > hmask)
522 		return 1;
523 
524 	return 0;
525 }
526 
527 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
528 {
529 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
530 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
531 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
532 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
533 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
534 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
535 	si->spdhcnt = net->xfrm.policy_idx_hmask;
536 	si->spdhmcnt = xfrm_policy_hashmax;
537 }
538 EXPORT_SYMBOL(xfrm_spd_getinfo);
539 
540 static DEFINE_MUTEX(hash_resize_mutex);
541 static void xfrm_hash_resize(struct work_struct *work)
542 {
543 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
544 	int dir, total;
545 
546 	mutex_lock(&hash_resize_mutex);
547 
548 	total = 0;
549 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
550 		if (xfrm_bydst_should_resize(net, dir, &total))
551 			xfrm_bydst_resize(net, dir);
552 	}
553 	if (xfrm_byidx_should_resize(net, total))
554 		xfrm_byidx_resize(net, total);
555 
556 	mutex_unlock(&hash_resize_mutex);
557 }
558 
559 static void xfrm_hash_rebuild(struct work_struct *work)
560 {
561 	struct net *net = container_of(work, struct net,
562 				       xfrm.policy_hthresh.work);
563 	unsigned int hmask;
564 	struct xfrm_policy *pol;
565 	struct xfrm_policy *policy;
566 	struct hlist_head *chain;
567 	struct hlist_head *odst;
568 	struct hlist_node *newpos;
569 	int i;
570 	int dir;
571 	unsigned seq;
572 	u8 lbits4, rbits4, lbits6, rbits6;
573 
574 	mutex_lock(&hash_resize_mutex);
575 
576 	/* read selector prefixlen thresholds */
577 	do {
578 		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
579 
580 		lbits4 = net->xfrm.policy_hthresh.lbits4;
581 		rbits4 = net->xfrm.policy_hthresh.rbits4;
582 		lbits6 = net->xfrm.policy_hthresh.lbits6;
583 		rbits6 = net->xfrm.policy_hthresh.rbits6;
584 	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
585 
586 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
587 
588 	/* reset the bydst and inexact table in all directions */
589 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
590 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
591 		hmask = net->xfrm.policy_bydst[dir].hmask;
592 		odst = net->xfrm.policy_bydst[dir].table;
593 		for (i = hmask; i >= 0; i--)
594 			INIT_HLIST_HEAD(odst + i);
595 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
596 			/* dir out => dst = remote, src = local */
597 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
598 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
599 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
600 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
601 		} else {
602 			/* dir in/fwd => dst = local, src = remote */
603 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
604 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
605 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
606 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
607 		}
608 	}
609 
610 	/* re-insert all policies by order of creation */
611 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
612 		if (xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
613 			/* skip socket policies */
614 			continue;
615 		}
616 		newpos = NULL;
617 		chain = policy_hash_bysel(net, &policy->selector,
618 					  policy->family,
619 					  xfrm_policy_id2dir(policy->index));
620 		hlist_for_each_entry(pol, chain, bydst) {
621 			if (policy->priority >= pol->priority)
622 				newpos = &pol->bydst;
623 			else
624 				break;
625 		}
626 		if (newpos)
627 			hlist_add_behind(&policy->bydst, newpos);
628 		else
629 			hlist_add_head(&policy->bydst, chain);
630 	}
631 
632 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
633 
634 	mutex_unlock(&hash_resize_mutex);
635 }
636 
637 void xfrm_policy_hash_rebuild(struct net *net)
638 {
639 	schedule_work(&net->xfrm.policy_hthresh.work);
640 }
641 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
642 
643 /* Generate new index... KAME seems to generate them ordered by cost
644  * of an absolute inpredictability of ordering of rules. This will not pass. */
645 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
646 {
647 	static u32 idx_generator;
648 
649 	for (;;) {
650 		struct hlist_head *list;
651 		struct xfrm_policy *p;
652 		u32 idx;
653 		int found;
654 
655 		if (!index) {
656 			idx = (idx_generator | dir);
657 			idx_generator += 8;
658 		} else {
659 			idx = index;
660 			index = 0;
661 		}
662 
663 		if (idx == 0)
664 			idx = 8;
665 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
666 		found = 0;
667 		hlist_for_each_entry(p, list, byidx) {
668 			if (p->index == idx) {
669 				found = 1;
670 				break;
671 			}
672 		}
673 		if (!found)
674 			return idx;
675 	}
676 }
677 
678 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
679 {
680 	u32 *p1 = (u32 *) s1;
681 	u32 *p2 = (u32 *) s2;
682 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
683 	int i;
684 
685 	for (i = 0; i < len; i++) {
686 		if (p1[i] != p2[i])
687 			return 1;
688 	}
689 
690 	return 0;
691 }
692 
693 static void xfrm_policy_requeue(struct xfrm_policy *old,
694 				struct xfrm_policy *new)
695 {
696 	struct xfrm_policy_queue *pq = &old->polq;
697 	struct sk_buff_head list;
698 
699 	if (skb_queue_empty(&pq->hold_queue))
700 		return;
701 
702 	__skb_queue_head_init(&list);
703 
704 	spin_lock_bh(&pq->hold_queue.lock);
705 	skb_queue_splice_init(&pq->hold_queue, &list);
706 	if (del_timer(&pq->hold_timer))
707 		xfrm_pol_put(old);
708 	spin_unlock_bh(&pq->hold_queue.lock);
709 
710 	pq = &new->polq;
711 
712 	spin_lock_bh(&pq->hold_queue.lock);
713 	skb_queue_splice(&list, &pq->hold_queue);
714 	pq->timeout = XFRM_QUEUE_TMO_MIN;
715 	if (!mod_timer(&pq->hold_timer, jiffies))
716 		xfrm_pol_hold(new);
717 	spin_unlock_bh(&pq->hold_queue.lock);
718 }
719 
720 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
721 				   struct xfrm_policy *pol)
722 {
723 	u32 mark = policy->mark.v & policy->mark.m;
724 
725 	if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
726 		return true;
727 
728 	if ((mark & pol->mark.m) == pol->mark.v &&
729 	    policy->priority == pol->priority)
730 		return true;
731 
732 	return false;
733 }
734 
735 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
736 {
737 	struct net *net = xp_net(policy);
738 	struct xfrm_policy *pol;
739 	struct xfrm_policy *delpol;
740 	struct hlist_head *chain;
741 	struct hlist_node *newpos;
742 
743 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
744 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
745 	delpol = NULL;
746 	newpos = NULL;
747 	hlist_for_each_entry(pol, chain, bydst) {
748 		if (pol->type == policy->type &&
749 		    !selector_cmp(&pol->selector, &policy->selector) &&
750 		    xfrm_policy_mark_match(policy, pol) &&
751 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
752 		    !WARN_ON(delpol)) {
753 			if (excl) {
754 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
755 				return -EEXIST;
756 			}
757 			delpol = pol;
758 			if (policy->priority > pol->priority)
759 				continue;
760 		} else if (policy->priority >= pol->priority) {
761 			newpos = &pol->bydst;
762 			continue;
763 		}
764 		if (delpol)
765 			break;
766 	}
767 	if (newpos)
768 		hlist_add_behind(&policy->bydst, newpos);
769 	else
770 		hlist_add_head(&policy->bydst, chain);
771 	__xfrm_policy_link(policy, dir);
772 
773 	/* After previous checking, family can either be AF_INET or AF_INET6 */
774 	if (policy->family == AF_INET)
775 		rt_genid_bump_ipv4(net);
776 	else
777 		rt_genid_bump_ipv6(net);
778 
779 	if (delpol) {
780 		xfrm_policy_requeue(delpol, policy);
781 		__xfrm_policy_unlink(delpol, dir);
782 	}
783 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
784 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
785 	policy->curlft.add_time = get_seconds();
786 	policy->curlft.use_time = 0;
787 	if (!mod_timer(&policy->timer, jiffies + HZ))
788 		xfrm_pol_hold(policy);
789 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
790 
791 	if (delpol)
792 		xfrm_policy_kill(delpol);
793 	else if (xfrm_bydst_should_resize(net, dir, NULL))
794 		schedule_work(&net->xfrm.policy_hash_work);
795 
796 	return 0;
797 }
798 EXPORT_SYMBOL(xfrm_policy_insert);
799 
800 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
801 					  int dir, struct xfrm_selector *sel,
802 					  struct xfrm_sec_ctx *ctx, int delete,
803 					  int *err)
804 {
805 	struct xfrm_policy *pol, *ret;
806 	struct hlist_head *chain;
807 
808 	*err = 0;
809 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
810 	chain = policy_hash_bysel(net, sel, sel->family, dir);
811 	ret = NULL;
812 	hlist_for_each_entry(pol, chain, bydst) {
813 		if (pol->type == type &&
814 		    (mark & pol->mark.m) == pol->mark.v &&
815 		    !selector_cmp(sel, &pol->selector) &&
816 		    xfrm_sec_ctx_match(ctx, pol->security)) {
817 			xfrm_pol_hold(pol);
818 			if (delete) {
819 				*err = security_xfrm_policy_delete(
820 								pol->security);
821 				if (*err) {
822 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
823 					return pol;
824 				}
825 				__xfrm_policy_unlink(pol, dir);
826 			}
827 			ret = pol;
828 			break;
829 		}
830 	}
831 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
832 
833 	if (ret && delete)
834 		xfrm_policy_kill(ret);
835 	return ret;
836 }
837 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
838 
839 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
840 				     int dir, u32 id, int delete, int *err)
841 {
842 	struct xfrm_policy *pol, *ret;
843 	struct hlist_head *chain;
844 
845 	*err = -ENOENT;
846 	if (xfrm_policy_id2dir(id) != dir)
847 		return NULL;
848 
849 	*err = 0;
850 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
851 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
852 	ret = NULL;
853 	hlist_for_each_entry(pol, chain, byidx) {
854 		if (pol->type == type && pol->index == id &&
855 		    (mark & pol->mark.m) == pol->mark.v) {
856 			xfrm_pol_hold(pol);
857 			if (delete) {
858 				*err = security_xfrm_policy_delete(
859 								pol->security);
860 				if (*err) {
861 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
862 					return pol;
863 				}
864 				__xfrm_policy_unlink(pol, dir);
865 			}
866 			ret = pol;
867 			break;
868 		}
869 	}
870 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
871 
872 	if (ret && delete)
873 		xfrm_policy_kill(ret);
874 	return ret;
875 }
876 EXPORT_SYMBOL(xfrm_policy_byid);
877 
878 #ifdef CONFIG_SECURITY_NETWORK_XFRM
879 static inline int
880 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
881 {
882 	int dir, err = 0;
883 
884 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
885 		struct xfrm_policy *pol;
886 		int i;
887 
888 		hlist_for_each_entry(pol,
889 				     &net->xfrm.policy_inexact[dir], bydst) {
890 			if (pol->type != type)
891 				continue;
892 			err = security_xfrm_policy_delete(pol->security);
893 			if (err) {
894 				xfrm_audit_policy_delete(pol, 0, task_valid);
895 				return err;
896 			}
897 		}
898 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
899 			hlist_for_each_entry(pol,
900 					     net->xfrm.policy_bydst[dir].table + i,
901 					     bydst) {
902 				if (pol->type != type)
903 					continue;
904 				err = security_xfrm_policy_delete(
905 								pol->security);
906 				if (err) {
907 					xfrm_audit_policy_delete(pol, 0,
908 								 task_valid);
909 					return err;
910 				}
911 			}
912 		}
913 	}
914 	return err;
915 }
916 #else
917 static inline int
918 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
919 {
920 	return 0;
921 }
922 #endif
923 
924 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
925 {
926 	int dir, err = 0, cnt = 0;
927 
928 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
929 
930 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
931 	if (err)
932 		goto out;
933 
934 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
935 		struct xfrm_policy *pol;
936 		int i;
937 
938 	again1:
939 		hlist_for_each_entry(pol,
940 				     &net->xfrm.policy_inexact[dir], bydst) {
941 			if (pol->type != type)
942 				continue;
943 			__xfrm_policy_unlink(pol, dir);
944 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
945 			cnt++;
946 
947 			xfrm_audit_policy_delete(pol, 1, task_valid);
948 
949 			xfrm_policy_kill(pol);
950 
951 			spin_lock_bh(&net->xfrm.xfrm_policy_lock);
952 			goto again1;
953 		}
954 
955 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
956 	again2:
957 			hlist_for_each_entry(pol,
958 					     net->xfrm.policy_bydst[dir].table + i,
959 					     bydst) {
960 				if (pol->type != type)
961 					continue;
962 				__xfrm_policy_unlink(pol, dir);
963 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
964 				cnt++;
965 
966 				xfrm_audit_policy_delete(pol, 1, task_valid);
967 				xfrm_policy_kill(pol);
968 
969 				spin_lock_bh(&net->xfrm.xfrm_policy_lock);
970 				goto again2;
971 			}
972 		}
973 
974 	}
975 	if (!cnt)
976 		err = -ESRCH;
977 	else
978 		xfrm_policy_cache_flush();
979 out:
980 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
981 	return err;
982 }
983 EXPORT_SYMBOL(xfrm_policy_flush);
984 
985 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
986 		     int (*func)(struct xfrm_policy *, int, int, void*),
987 		     void *data)
988 {
989 	struct xfrm_policy *pol;
990 	struct xfrm_policy_walk_entry *x;
991 	int error = 0;
992 
993 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
994 	    walk->type != XFRM_POLICY_TYPE_ANY)
995 		return -EINVAL;
996 
997 	if (list_empty(&walk->walk.all) && walk->seq != 0)
998 		return 0;
999 
1000 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1001 	if (list_empty(&walk->walk.all))
1002 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1003 	else
1004 		x = list_first_entry(&walk->walk.all,
1005 				     struct xfrm_policy_walk_entry, all);
1006 
1007 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1008 		if (x->dead)
1009 			continue;
1010 		pol = container_of(x, struct xfrm_policy, walk);
1011 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1012 		    walk->type != pol->type)
1013 			continue;
1014 		error = func(pol, xfrm_policy_id2dir(pol->index),
1015 			     walk->seq, data);
1016 		if (error) {
1017 			list_move_tail(&walk->walk.all, &x->all);
1018 			goto out;
1019 		}
1020 		walk->seq++;
1021 	}
1022 	if (walk->seq == 0) {
1023 		error = -ENOENT;
1024 		goto out;
1025 	}
1026 	list_del_init(&walk->walk.all);
1027 out:
1028 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1029 	return error;
1030 }
1031 EXPORT_SYMBOL(xfrm_policy_walk);
1032 
1033 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1034 {
1035 	INIT_LIST_HEAD(&walk->walk.all);
1036 	walk->walk.dead = 1;
1037 	walk->type = type;
1038 	walk->seq = 0;
1039 }
1040 EXPORT_SYMBOL(xfrm_policy_walk_init);
1041 
1042 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1043 {
1044 	if (list_empty(&walk->walk.all))
1045 		return;
1046 
1047 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1048 	list_del(&walk->walk.all);
1049 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1050 }
1051 EXPORT_SYMBOL(xfrm_policy_walk_done);
1052 
1053 /*
1054  * Find policy to apply to this flow.
1055  *
1056  * Returns 0 if policy found, else an -errno.
1057  */
1058 static int xfrm_policy_match(const struct xfrm_policy *pol,
1059 			     const struct flowi *fl,
1060 			     u8 type, u16 family, int dir)
1061 {
1062 	const struct xfrm_selector *sel = &pol->selector;
1063 	int ret = -ESRCH;
1064 	bool match;
1065 
1066 	if (pol->family != family ||
1067 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1068 	    pol->type != type)
1069 		return ret;
1070 
1071 	match = xfrm_selector_match(sel, fl, family);
1072 	if (match)
1073 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1074 						  dir);
1075 
1076 	return ret;
1077 }
1078 
1079 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1080 						     const struct flowi *fl,
1081 						     u16 family, u8 dir)
1082 {
1083 	int err;
1084 	struct xfrm_policy *pol, *ret;
1085 	const xfrm_address_t *daddr, *saddr;
1086 	struct hlist_head *chain;
1087 	unsigned int sequence;
1088 	u32 priority;
1089 
1090 	daddr = xfrm_flowi_daddr(fl, family);
1091 	saddr = xfrm_flowi_saddr(fl, family);
1092 	if (unlikely(!daddr || !saddr))
1093 		return NULL;
1094 
1095 	rcu_read_lock();
1096  retry:
1097 	do {
1098 		sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1099 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
1100 	} while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1101 
1102 	priority = ~0U;
1103 	ret = NULL;
1104 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1105 		err = xfrm_policy_match(pol, fl, type, family, dir);
1106 		if (err) {
1107 			if (err == -ESRCH)
1108 				continue;
1109 			else {
1110 				ret = ERR_PTR(err);
1111 				goto fail;
1112 			}
1113 		} else {
1114 			ret = pol;
1115 			priority = ret->priority;
1116 			break;
1117 		}
1118 	}
1119 	chain = &net->xfrm.policy_inexact[dir];
1120 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1121 		if ((pol->priority >= priority) && ret)
1122 			break;
1123 
1124 		err = xfrm_policy_match(pol, fl, type, family, dir);
1125 		if (err) {
1126 			if (err == -ESRCH)
1127 				continue;
1128 			else {
1129 				ret = ERR_PTR(err);
1130 				goto fail;
1131 			}
1132 		} else {
1133 			ret = pol;
1134 			break;
1135 		}
1136 	}
1137 
1138 	if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1139 		goto retry;
1140 
1141 	if (ret && !xfrm_pol_hold_rcu(ret))
1142 		goto retry;
1143 fail:
1144 	rcu_read_unlock();
1145 
1146 	return ret;
1147 }
1148 
1149 static struct xfrm_policy *
1150 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1151 {
1152 #ifdef CONFIG_XFRM_SUB_POLICY
1153 	struct xfrm_policy *pol;
1154 
1155 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1156 	if (pol != NULL)
1157 		return pol;
1158 #endif
1159 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1160 }
1161 
1162 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1163 						 const struct flowi *fl, u16 family)
1164 {
1165 	struct xfrm_policy *pol;
1166 
1167 	rcu_read_lock();
1168  again:
1169 	pol = rcu_dereference(sk->sk_policy[dir]);
1170 	if (pol != NULL) {
1171 		bool match = xfrm_selector_match(&pol->selector, fl, family);
1172 		int err = 0;
1173 
1174 		if (match) {
1175 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1176 				pol = NULL;
1177 				goto out;
1178 			}
1179 			err = security_xfrm_policy_lookup(pol->security,
1180 						      fl->flowi_secid,
1181 						      dir);
1182 			if (!err) {
1183 				if (!xfrm_pol_hold_rcu(pol))
1184 					goto again;
1185 			} else if (err == -ESRCH) {
1186 				pol = NULL;
1187 			} else {
1188 				pol = ERR_PTR(err);
1189 			}
1190 		} else
1191 			pol = NULL;
1192 	}
1193 out:
1194 	rcu_read_unlock();
1195 	return pol;
1196 }
1197 
1198 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1199 {
1200 	struct net *net = xp_net(pol);
1201 
1202 	list_add(&pol->walk.all, &net->xfrm.policy_all);
1203 	net->xfrm.policy_count[dir]++;
1204 	xfrm_pol_hold(pol);
1205 }
1206 
1207 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1208 						int dir)
1209 {
1210 	struct net *net = xp_net(pol);
1211 
1212 	if (list_empty(&pol->walk.all))
1213 		return NULL;
1214 
1215 	/* Socket policies are not hashed. */
1216 	if (!hlist_unhashed(&pol->bydst)) {
1217 		hlist_del_rcu(&pol->bydst);
1218 		hlist_del(&pol->byidx);
1219 	}
1220 
1221 	list_del_init(&pol->walk.all);
1222 	net->xfrm.policy_count[dir]--;
1223 
1224 	return pol;
1225 }
1226 
1227 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1228 {
1229 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1230 }
1231 
1232 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1233 {
1234 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1235 }
1236 
1237 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1238 {
1239 	struct net *net = xp_net(pol);
1240 
1241 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1242 	pol = __xfrm_policy_unlink(pol, dir);
1243 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1244 	if (pol) {
1245 		xfrm_policy_kill(pol);
1246 		return 0;
1247 	}
1248 	return -ENOENT;
1249 }
1250 EXPORT_SYMBOL(xfrm_policy_delete);
1251 
1252 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1253 {
1254 	struct net *net = xp_net(pol);
1255 	struct xfrm_policy *old_pol;
1256 
1257 #ifdef CONFIG_XFRM_SUB_POLICY
1258 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1259 		return -EINVAL;
1260 #endif
1261 
1262 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1263 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1264 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1265 	if (pol) {
1266 		pol->curlft.add_time = get_seconds();
1267 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1268 		xfrm_sk_policy_link(pol, dir);
1269 	}
1270 	rcu_assign_pointer(sk->sk_policy[dir], pol);
1271 	if (old_pol) {
1272 		if (pol)
1273 			xfrm_policy_requeue(old_pol, pol);
1274 
1275 		/* Unlinking succeeds always. This is the only function
1276 		 * allowed to delete or replace socket policy.
1277 		 */
1278 		xfrm_sk_policy_unlink(old_pol, dir);
1279 	}
1280 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1281 
1282 	if (old_pol) {
1283 		xfrm_policy_kill(old_pol);
1284 	}
1285 	return 0;
1286 }
1287 
1288 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1289 {
1290 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1291 	struct net *net = xp_net(old);
1292 
1293 	if (newp) {
1294 		newp->selector = old->selector;
1295 		if (security_xfrm_policy_clone(old->security,
1296 					       &newp->security)) {
1297 			kfree(newp);
1298 			return NULL;  /* ENOMEM */
1299 		}
1300 		newp->lft = old->lft;
1301 		newp->curlft = old->curlft;
1302 		newp->mark = old->mark;
1303 		newp->action = old->action;
1304 		newp->flags = old->flags;
1305 		newp->xfrm_nr = old->xfrm_nr;
1306 		newp->index = old->index;
1307 		newp->type = old->type;
1308 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1309 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1310 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1311 		xfrm_sk_policy_link(newp, dir);
1312 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1313 		xfrm_pol_put(newp);
1314 	}
1315 	return newp;
1316 }
1317 
1318 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1319 {
1320 	const struct xfrm_policy *p;
1321 	struct xfrm_policy *np;
1322 	int i, ret = 0;
1323 
1324 	rcu_read_lock();
1325 	for (i = 0; i < 2; i++) {
1326 		p = rcu_dereference(osk->sk_policy[i]);
1327 		if (p) {
1328 			np = clone_policy(p, i);
1329 			if (unlikely(!np)) {
1330 				ret = -ENOMEM;
1331 				break;
1332 			}
1333 			rcu_assign_pointer(sk->sk_policy[i], np);
1334 		}
1335 	}
1336 	rcu_read_unlock();
1337 	return ret;
1338 }
1339 
1340 static int
1341 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1342 	       xfrm_address_t *remote, unsigned short family, u32 mark)
1343 {
1344 	int err;
1345 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1346 
1347 	if (unlikely(afinfo == NULL))
1348 		return -EINVAL;
1349 	err = afinfo->get_saddr(net, oif, local, remote, mark);
1350 	rcu_read_unlock();
1351 	return err;
1352 }
1353 
1354 /* Resolve list of templates for the flow, given policy. */
1355 
1356 static int
1357 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1358 		      struct xfrm_state **xfrm, unsigned short family)
1359 {
1360 	struct net *net = xp_net(policy);
1361 	int nx;
1362 	int i, error;
1363 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1364 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1365 	xfrm_address_t tmp;
1366 
1367 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1368 		struct xfrm_state *x;
1369 		xfrm_address_t *remote = daddr;
1370 		xfrm_address_t *local  = saddr;
1371 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1372 
1373 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1374 		    tmpl->mode == XFRM_MODE_BEET) {
1375 			remote = &tmpl->id.daddr;
1376 			local = &tmpl->saddr;
1377 			if (xfrm_addr_any(local, tmpl->encap_family)) {
1378 				error = xfrm_get_saddr(net, fl->flowi_oif,
1379 						       &tmp, remote,
1380 						       tmpl->encap_family, 0);
1381 				if (error)
1382 					goto fail;
1383 				local = &tmp;
1384 			}
1385 		}
1386 
1387 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1388 
1389 		if (x && x->km.state == XFRM_STATE_VALID) {
1390 			xfrm[nx++] = x;
1391 			daddr = remote;
1392 			saddr = local;
1393 			continue;
1394 		}
1395 		if (x) {
1396 			error = (x->km.state == XFRM_STATE_ERROR ?
1397 				 -EINVAL : -EAGAIN);
1398 			xfrm_state_put(x);
1399 		} else if (error == -ESRCH) {
1400 			error = -EAGAIN;
1401 		}
1402 
1403 		if (!tmpl->optional)
1404 			goto fail;
1405 	}
1406 	return nx;
1407 
1408 fail:
1409 	for (nx--; nx >= 0; nx--)
1410 		xfrm_state_put(xfrm[nx]);
1411 	return error;
1412 }
1413 
1414 static int
1415 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1416 		  struct xfrm_state **xfrm, unsigned short family)
1417 {
1418 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1419 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1420 	int cnx = 0;
1421 	int error;
1422 	int ret;
1423 	int i;
1424 
1425 	for (i = 0; i < npols; i++) {
1426 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1427 			error = -ENOBUFS;
1428 			goto fail;
1429 		}
1430 
1431 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1432 		if (ret < 0) {
1433 			error = ret;
1434 			goto fail;
1435 		} else
1436 			cnx += ret;
1437 	}
1438 
1439 	/* found states are sorted for outbound processing */
1440 	if (npols > 1)
1441 		xfrm_state_sort(xfrm, tpp, cnx, family);
1442 
1443 	return cnx;
1444 
1445  fail:
1446 	for (cnx--; cnx >= 0; cnx--)
1447 		xfrm_state_put(tpp[cnx]);
1448 	return error;
1449 
1450 }
1451 
1452 static int xfrm_get_tos(const struct flowi *fl, int family)
1453 {
1454 	const struct xfrm_policy_afinfo *afinfo;
1455 	int tos = 0;
1456 
1457 	afinfo = xfrm_policy_get_afinfo(family);
1458 	tos = afinfo ? afinfo->get_tos(fl) : 0;
1459 
1460 	rcu_read_unlock();
1461 
1462 	return tos;
1463 }
1464 
1465 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1466 {
1467 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1468 	struct dst_ops *dst_ops;
1469 	struct xfrm_dst *xdst;
1470 
1471 	if (!afinfo)
1472 		return ERR_PTR(-EINVAL);
1473 
1474 	switch (family) {
1475 	case AF_INET:
1476 		dst_ops = &net->xfrm.xfrm4_dst_ops;
1477 		break;
1478 #if IS_ENABLED(CONFIG_IPV6)
1479 	case AF_INET6:
1480 		dst_ops = &net->xfrm.xfrm6_dst_ops;
1481 		break;
1482 #endif
1483 	default:
1484 		BUG();
1485 	}
1486 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1487 
1488 	if (likely(xdst)) {
1489 		struct dst_entry *dst = &xdst->u.dst;
1490 
1491 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1492 	} else
1493 		xdst = ERR_PTR(-ENOBUFS);
1494 
1495 	rcu_read_unlock();
1496 
1497 	return xdst;
1498 }
1499 
1500 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1501 				 int nfheader_len)
1502 {
1503 	const struct xfrm_policy_afinfo *afinfo =
1504 		xfrm_policy_get_afinfo(dst->ops->family);
1505 	int err;
1506 
1507 	if (!afinfo)
1508 		return -EINVAL;
1509 
1510 	err = afinfo->init_path(path, dst, nfheader_len);
1511 
1512 	rcu_read_unlock();
1513 
1514 	return err;
1515 }
1516 
1517 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1518 				const struct flowi *fl)
1519 {
1520 	const struct xfrm_policy_afinfo *afinfo =
1521 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1522 	int err;
1523 
1524 	if (!afinfo)
1525 		return -EINVAL;
1526 
1527 	err = afinfo->fill_dst(xdst, dev, fl);
1528 
1529 	rcu_read_unlock();
1530 
1531 	return err;
1532 }
1533 
1534 
1535 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1536  * all the metrics... Shortly, bundle a bundle.
1537  */
1538 
1539 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1540 					    struct xfrm_state **xfrm, int nx,
1541 					    const struct flowi *fl,
1542 					    struct dst_entry *dst)
1543 {
1544 	struct net *net = xp_net(policy);
1545 	unsigned long now = jiffies;
1546 	struct net_device *dev;
1547 	struct xfrm_mode *inner_mode;
1548 	struct dst_entry *dst_prev = NULL;
1549 	struct dst_entry *dst0 = NULL;
1550 	int i = 0;
1551 	int err;
1552 	int header_len = 0;
1553 	int nfheader_len = 0;
1554 	int trailer_len = 0;
1555 	int tos;
1556 	int family = policy->selector.family;
1557 	xfrm_address_t saddr, daddr;
1558 
1559 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1560 
1561 	tos = xfrm_get_tos(fl, family);
1562 
1563 	dst_hold(dst);
1564 
1565 	for (; i < nx; i++) {
1566 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1567 		struct dst_entry *dst1 = &xdst->u.dst;
1568 
1569 		err = PTR_ERR(xdst);
1570 		if (IS_ERR(xdst)) {
1571 			dst_release(dst);
1572 			goto put_states;
1573 		}
1574 
1575 		if (!dst_prev)
1576 			dst0 = dst1;
1577 		else
1578 			/* Ref count is taken during xfrm_alloc_dst()
1579 			 * No need to do dst_clone() on dst1
1580 			 */
1581 			dst_prev->child = dst1;
1582 
1583 		if (xfrm[i]->sel.family == AF_UNSPEC) {
1584 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
1585 							xfrm_af2proto(family));
1586 			if (!inner_mode) {
1587 				err = -EAFNOSUPPORT;
1588 				dst_release(dst);
1589 				goto put_states;
1590 			}
1591 		} else
1592 			inner_mode = xfrm[i]->inner_mode;
1593 
1594 		xdst->route = dst;
1595 		dst_copy_metrics(dst1, dst);
1596 
1597 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1598 			family = xfrm[i]->props.family;
1599 			dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1600 					      &saddr, &daddr, family,
1601 					      xfrm[i]->props.output_mark);
1602 			err = PTR_ERR(dst);
1603 			if (IS_ERR(dst))
1604 				goto put_states;
1605 		} else
1606 			dst_hold(dst);
1607 
1608 		dst1->xfrm = xfrm[i];
1609 		xdst->xfrm_genid = xfrm[i]->genid;
1610 
1611 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1612 		dst1->flags |= DST_HOST;
1613 		dst1->lastuse = now;
1614 
1615 		dst1->input = dst_discard;
1616 		dst1->output = inner_mode->afinfo->output;
1617 
1618 		dst1->next = dst_prev;
1619 		dst_prev = dst1;
1620 
1621 		header_len += xfrm[i]->props.header_len;
1622 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1623 			nfheader_len += xfrm[i]->props.header_len;
1624 		trailer_len += xfrm[i]->props.trailer_len;
1625 	}
1626 
1627 	dst_prev->child = dst;
1628 	dst0->path = dst;
1629 
1630 	err = -ENODEV;
1631 	dev = dst->dev;
1632 	if (!dev)
1633 		goto free_dst;
1634 
1635 	xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1636 	xfrm_init_pmtu(dst_prev);
1637 
1638 	for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1639 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1640 
1641 		err = xfrm_fill_dst(xdst, dev, fl);
1642 		if (err)
1643 			goto free_dst;
1644 
1645 		dst_prev->header_len = header_len;
1646 		dst_prev->trailer_len = trailer_len;
1647 		header_len -= xdst->u.dst.xfrm->props.header_len;
1648 		trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1649 	}
1650 
1651 out:
1652 	return dst0;
1653 
1654 put_states:
1655 	for (; i < nx; i++)
1656 		xfrm_state_put(xfrm[i]);
1657 free_dst:
1658 	if (dst0)
1659 		dst_release_immediate(dst0);
1660 	dst0 = ERR_PTR(err);
1661 	goto out;
1662 }
1663 
1664 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1665 				struct xfrm_policy **pols,
1666 				int *num_pols, int *num_xfrms)
1667 {
1668 	int i;
1669 
1670 	if (*num_pols == 0 || !pols[0]) {
1671 		*num_pols = 0;
1672 		*num_xfrms = 0;
1673 		return 0;
1674 	}
1675 	if (IS_ERR(pols[0]))
1676 		return PTR_ERR(pols[0]);
1677 
1678 	*num_xfrms = pols[0]->xfrm_nr;
1679 
1680 #ifdef CONFIG_XFRM_SUB_POLICY
1681 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1682 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1683 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1684 						    XFRM_POLICY_TYPE_MAIN,
1685 						    fl, family,
1686 						    XFRM_POLICY_OUT);
1687 		if (pols[1]) {
1688 			if (IS_ERR(pols[1])) {
1689 				xfrm_pols_put(pols, *num_pols);
1690 				return PTR_ERR(pols[1]);
1691 			}
1692 			(*num_pols)++;
1693 			(*num_xfrms) += pols[1]->xfrm_nr;
1694 		}
1695 	}
1696 #endif
1697 	for (i = 0; i < *num_pols; i++) {
1698 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
1699 			*num_xfrms = -1;
1700 			break;
1701 		}
1702 	}
1703 
1704 	return 0;
1705 
1706 }
1707 
1708 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1709 {
1710 	this_cpu_write(xfrm_last_dst, xdst);
1711 	if (old)
1712 		dst_release(&old->u.dst);
1713 }
1714 
1715 static void __xfrm_pcpu_work_fn(void)
1716 {
1717 	struct xfrm_dst *old;
1718 
1719 	old = this_cpu_read(xfrm_last_dst);
1720 	if (old && !xfrm_bundle_ok(old))
1721 		xfrm_last_dst_update(NULL, old);
1722 }
1723 
1724 static void xfrm_pcpu_work_fn(struct work_struct *work)
1725 {
1726 	local_bh_disable();
1727 	rcu_read_lock();
1728 	__xfrm_pcpu_work_fn();
1729 	rcu_read_unlock();
1730 	local_bh_enable();
1731 }
1732 
1733 void xfrm_policy_cache_flush(void)
1734 {
1735 	struct xfrm_dst *old;
1736 	bool found = 0;
1737 	int cpu;
1738 
1739 	local_bh_disable();
1740 	rcu_read_lock();
1741 	for_each_possible_cpu(cpu) {
1742 		old = per_cpu(xfrm_last_dst, cpu);
1743 		if (old && !xfrm_bundle_ok(old)) {
1744 			if (smp_processor_id() == cpu) {
1745 				__xfrm_pcpu_work_fn();
1746 				continue;
1747 			}
1748 			found = true;
1749 			break;
1750 		}
1751 	}
1752 
1753 	rcu_read_unlock();
1754 	local_bh_enable();
1755 
1756 	if (!found)
1757 		return;
1758 
1759 	get_online_cpus();
1760 
1761 	for_each_possible_cpu(cpu) {
1762 		bool bundle_release;
1763 
1764 		rcu_read_lock();
1765 		old = per_cpu(xfrm_last_dst, cpu);
1766 		bundle_release = old && !xfrm_bundle_ok(old);
1767 		rcu_read_unlock();
1768 
1769 		if (!bundle_release)
1770 			continue;
1771 
1772 		if (cpu_online(cpu)) {
1773 			schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1774 			continue;
1775 		}
1776 
1777 		rcu_read_lock();
1778 		old = per_cpu(xfrm_last_dst, cpu);
1779 		if (old && !xfrm_bundle_ok(old)) {
1780 			per_cpu(xfrm_last_dst, cpu) = NULL;
1781 			dst_release(&old->u.dst);
1782 		}
1783 		rcu_read_unlock();
1784 	}
1785 
1786 	put_online_cpus();
1787 }
1788 
1789 static bool xfrm_pol_dead(struct xfrm_dst *xdst)
1790 {
1791 	unsigned int num_pols = xdst->num_pols;
1792 	unsigned int pol_dead = 0, i;
1793 
1794 	for (i = 0; i < num_pols; i++)
1795 		pol_dead |= xdst->pols[i]->walk.dead;
1796 
1797 	/* Mark DST_OBSOLETE_DEAD to fail the next xfrm_dst_check() */
1798 	if (pol_dead)
1799 		xdst->u.dst.obsolete = DST_OBSOLETE_DEAD;
1800 
1801 	return pol_dead;
1802 }
1803 
1804 static struct xfrm_dst *
1805 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1806 			       const struct flowi *fl, u16 family,
1807 			       struct dst_entry *dst_orig)
1808 {
1809 	struct net *net = xp_net(pols[0]);
1810 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1811 	struct xfrm_dst *xdst, *old;
1812 	struct dst_entry *dst;
1813 	int err;
1814 
1815 	xdst = this_cpu_read(xfrm_last_dst);
1816 	if (xdst &&
1817 	    xdst->u.dst.dev == dst_orig->dev &&
1818 	    xdst->num_pols == num_pols &&
1819 	    !xfrm_pol_dead(xdst) &&
1820 	    memcmp(xdst->pols, pols,
1821 		   sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1822 	    xfrm_bundle_ok(xdst)) {
1823 		dst_hold(&xdst->u.dst);
1824 		return xdst;
1825 	}
1826 
1827 	old = xdst;
1828 	/* Try to instantiate a bundle */
1829 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1830 	if (err <= 0) {
1831 		if (err != 0 && err != -EAGAIN)
1832 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1833 		return ERR_PTR(err);
1834 	}
1835 
1836 	dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1837 	if (IS_ERR(dst)) {
1838 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1839 		return ERR_CAST(dst);
1840 	}
1841 
1842 	xdst = (struct xfrm_dst *)dst;
1843 	xdst->num_xfrms = err;
1844 	xdst->num_pols = num_pols;
1845 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1846 	xdst->policy_genid = atomic_read(&pols[0]->genid);
1847 
1848 	atomic_set(&xdst->u.dst.__refcnt, 2);
1849 	xfrm_last_dst_update(xdst, old);
1850 
1851 	return xdst;
1852 }
1853 
1854 static void xfrm_policy_queue_process(struct timer_list *t)
1855 {
1856 	struct sk_buff *skb;
1857 	struct sock *sk;
1858 	struct dst_entry *dst;
1859 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1860 	struct net *net = xp_net(pol);
1861 	struct xfrm_policy_queue *pq = &pol->polq;
1862 	struct flowi fl;
1863 	struct sk_buff_head list;
1864 
1865 	spin_lock(&pq->hold_queue.lock);
1866 	skb = skb_peek(&pq->hold_queue);
1867 	if (!skb) {
1868 		spin_unlock(&pq->hold_queue.lock);
1869 		goto out;
1870 	}
1871 	dst = skb_dst(skb);
1872 	sk = skb->sk;
1873 	xfrm_decode_session(skb, &fl, dst->ops->family);
1874 	spin_unlock(&pq->hold_queue.lock);
1875 
1876 	dst_hold(dst->path);
1877 	dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1878 	if (IS_ERR(dst))
1879 		goto purge_queue;
1880 
1881 	if (dst->flags & DST_XFRM_QUEUE) {
1882 		dst_release(dst);
1883 
1884 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1885 			goto purge_queue;
1886 
1887 		pq->timeout = pq->timeout << 1;
1888 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1889 			xfrm_pol_hold(pol);
1890 	goto out;
1891 	}
1892 
1893 	dst_release(dst);
1894 
1895 	__skb_queue_head_init(&list);
1896 
1897 	spin_lock(&pq->hold_queue.lock);
1898 	pq->timeout = 0;
1899 	skb_queue_splice_init(&pq->hold_queue, &list);
1900 	spin_unlock(&pq->hold_queue.lock);
1901 
1902 	while (!skb_queue_empty(&list)) {
1903 		skb = __skb_dequeue(&list);
1904 
1905 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1906 		dst_hold(skb_dst(skb)->path);
1907 		dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1908 		if (IS_ERR(dst)) {
1909 			kfree_skb(skb);
1910 			continue;
1911 		}
1912 
1913 		nf_reset(skb);
1914 		skb_dst_drop(skb);
1915 		skb_dst_set(skb, dst);
1916 
1917 		dst_output(net, skb->sk, skb);
1918 	}
1919 
1920 out:
1921 	xfrm_pol_put(pol);
1922 	return;
1923 
1924 purge_queue:
1925 	pq->timeout = 0;
1926 	skb_queue_purge(&pq->hold_queue);
1927 	xfrm_pol_put(pol);
1928 }
1929 
1930 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1931 {
1932 	unsigned long sched_next;
1933 	struct dst_entry *dst = skb_dst(skb);
1934 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1935 	struct xfrm_policy *pol = xdst->pols[0];
1936 	struct xfrm_policy_queue *pq = &pol->polq;
1937 
1938 	if (unlikely(skb_fclone_busy(sk, skb))) {
1939 		kfree_skb(skb);
1940 		return 0;
1941 	}
1942 
1943 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1944 		kfree_skb(skb);
1945 		return -EAGAIN;
1946 	}
1947 
1948 	skb_dst_force(skb);
1949 
1950 	spin_lock_bh(&pq->hold_queue.lock);
1951 
1952 	if (!pq->timeout)
1953 		pq->timeout = XFRM_QUEUE_TMO_MIN;
1954 
1955 	sched_next = jiffies + pq->timeout;
1956 
1957 	if (del_timer(&pq->hold_timer)) {
1958 		if (time_before(pq->hold_timer.expires, sched_next))
1959 			sched_next = pq->hold_timer.expires;
1960 		xfrm_pol_put(pol);
1961 	}
1962 
1963 	__skb_queue_tail(&pq->hold_queue, skb);
1964 	if (!mod_timer(&pq->hold_timer, sched_next))
1965 		xfrm_pol_hold(pol);
1966 
1967 	spin_unlock_bh(&pq->hold_queue.lock);
1968 
1969 	return 0;
1970 }
1971 
1972 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1973 						 struct xfrm_flo *xflo,
1974 						 const struct flowi *fl,
1975 						 int num_xfrms,
1976 						 u16 family)
1977 {
1978 	int err;
1979 	struct net_device *dev;
1980 	struct dst_entry *dst;
1981 	struct dst_entry *dst1;
1982 	struct xfrm_dst *xdst;
1983 
1984 	xdst = xfrm_alloc_dst(net, family);
1985 	if (IS_ERR(xdst))
1986 		return xdst;
1987 
1988 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
1989 	    net->xfrm.sysctl_larval_drop ||
1990 	    num_xfrms <= 0)
1991 		return xdst;
1992 
1993 	dst = xflo->dst_orig;
1994 	dst1 = &xdst->u.dst;
1995 	dst_hold(dst);
1996 	xdst->route = dst;
1997 
1998 	dst_copy_metrics(dst1, dst);
1999 
2000 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2001 	dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2002 	dst1->lastuse = jiffies;
2003 
2004 	dst1->input = dst_discard;
2005 	dst1->output = xdst_queue_output;
2006 
2007 	dst_hold(dst);
2008 	dst1->child = dst;
2009 	dst1->path = dst;
2010 
2011 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2012 
2013 	err = -ENODEV;
2014 	dev = dst->dev;
2015 	if (!dev)
2016 		goto free_dst;
2017 
2018 	err = xfrm_fill_dst(xdst, dev, fl);
2019 	if (err)
2020 		goto free_dst;
2021 
2022 out:
2023 	return xdst;
2024 
2025 free_dst:
2026 	dst_release(dst1);
2027 	xdst = ERR_PTR(err);
2028 	goto out;
2029 }
2030 
2031 static struct xfrm_dst *
2032 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2033 {
2034 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2035 	int num_pols = 0, num_xfrms = 0, err;
2036 	struct xfrm_dst *xdst;
2037 
2038 	/* Resolve policies to use if we couldn't get them from
2039 	 * previous cache entry */
2040 	num_pols = 1;
2041 	pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2042 	err = xfrm_expand_policies(fl, family, pols,
2043 					   &num_pols, &num_xfrms);
2044 	if (err < 0)
2045 		goto inc_error;
2046 	if (num_pols == 0)
2047 		return NULL;
2048 	if (num_xfrms <= 0)
2049 		goto make_dummy_bundle;
2050 
2051 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2052 						  xflo->dst_orig);
2053 	if (IS_ERR(xdst)) {
2054 		err = PTR_ERR(xdst);
2055 		if (err != -EAGAIN)
2056 			goto error;
2057 		goto make_dummy_bundle;
2058 	} else if (xdst == NULL) {
2059 		num_xfrms = 0;
2060 		goto make_dummy_bundle;
2061 	}
2062 
2063 	return xdst;
2064 
2065 make_dummy_bundle:
2066 	/* We found policies, but there's no bundles to instantiate:
2067 	 * either because the policy blocks, has no transformations or
2068 	 * we could not build template (no xfrm_states).*/
2069 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2070 	if (IS_ERR(xdst)) {
2071 		xfrm_pols_put(pols, num_pols);
2072 		return ERR_CAST(xdst);
2073 	}
2074 	xdst->num_pols = num_pols;
2075 	xdst->num_xfrms = num_xfrms;
2076 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2077 
2078 	return xdst;
2079 
2080 inc_error:
2081 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2082 error:
2083 	xfrm_pols_put(pols, num_pols);
2084 	return ERR_PTR(err);
2085 }
2086 
2087 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2088 					struct dst_entry *dst_orig)
2089 {
2090 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2091 	struct dst_entry *ret;
2092 
2093 	if (!afinfo) {
2094 		dst_release(dst_orig);
2095 		return ERR_PTR(-EINVAL);
2096 	} else {
2097 		ret = afinfo->blackhole_route(net, dst_orig);
2098 	}
2099 	rcu_read_unlock();
2100 
2101 	return ret;
2102 }
2103 
2104 /* Main function: finds/creates a bundle for given flow.
2105  *
2106  * At the moment we eat a raw IP route. Mostly to speed up lookups
2107  * on interfaces with disabled IPsec.
2108  */
2109 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2110 			      const struct flowi *fl,
2111 			      const struct sock *sk, int flags)
2112 {
2113 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2114 	struct xfrm_dst *xdst;
2115 	struct dst_entry *dst, *route;
2116 	u16 family = dst_orig->ops->family;
2117 	u8 dir = XFRM_POLICY_OUT;
2118 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2119 
2120 	dst = NULL;
2121 	xdst = NULL;
2122 	route = NULL;
2123 
2124 	sk = sk_const_to_full_sk(sk);
2125 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2126 		num_pols = 1;
2127 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2128 		err = xfrm_expand_policies(fl, family, pols,
2129 					   &num_pols, &num_xfrms);
2130 		if (err < 0)
2131 			goto dropdst;
2132 
2133 		if (num_pols) {
2134 			if (num_xfrms <= 0) {
2135 				drop_pols = num_pols;
2136 				goto no_transform;
2137 			}
2138 
2139 			xdst = xfrm_resolve_and_create_bundle(
2140 					pols, num_pols, fl,
2141 					family, dst_orig);
2142 			if (IS_ERR(xdst)) {
2143 				xfrm_pols_put(pols, num_pols);
2144 				err = PTR_ERR(xdst);
2145 				goto dropdst;
2146 			} else if (xdst == NULL) {
2147 				num_xfrms = 0;
2148 				drop_pols = num_pols;
2149 				goto no_transform;
2150 			}
2151 
2152 			route = xdst->route;
2153 		}
2154 	}
2155 
2156 	if (xdst == NULL) {
2157 		struct xfrm_flo xflo;
2158 
2159 		xflo.dst_orig = dst_orig;
2160 		xflo.flags = flags;
2161 
2162 		/* To accelerate a bit...  */
2163 		if ((dst_orig->flags & DST_NOXFRM) ||
2164 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
2165 			goto nopol;
2166 
2167 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2168 		if (xdst == NULL)
2169 			goto nopol;
2170 		if (IS_ERR(xdst)) {
2171 			err = PTR_ERR(xdst);
2172 			goto dropdst;
2173 		}
2174 
2175 		num_pols = xdst->num_pols;
2176 		num_xfrms = xdst->num_xfrms;
2177 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2178 		route = xdst->route;
2179 	}
2180 
2181 	dst = &xdst->u.dst;
2182 	if (route == NULL && num_xfrms > 0) {
2183 		/* The only case when xfrm_bundle_lookup() returns a
2184 		 * bundle with null route, is when the template could
2185 		 * not be resolved. It means policies are there, but
2186 		 * bundle could not be created, since we don't yet
2187 		 * have the xfrm_state's. We need to wait for KM to
2188 		 * negotiate new SA's or bail out with error.*/
2189 		if (net->xfrm.sysctl_larval_drop) {
2190 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2191 			err = -EREMOTE;
2192 			goto error;
2193 		}
2194 
2195 		err = -EAGAIN;
2196 
2197 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2198 		goto error;
2199 	}
2200 
2201 no_transform:
2202 	if (num_pols == 0)
2203 		goto nopol;
2204 
2205 	if ((flags & XFRM_LOOKUP_ICMP) &&
2206 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2207 		err = -ENOENT;
2208 		goto error;
2209 	}
2210 
2211 	for (i = 0; i < num_pols; i++)
2212 		pols[i]->curlft.use_time = get_seconds();
2213 
2214 	if (num_xfrms < 0) {
2215 		/* Prohibit the flow */
2216 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2217 		err = -EPERM;
2218 		goto error;
2219 	} else if (num_xfrms > 0) {
2220 		/* Flow transformed */
2221 		dst_release(dst_orig);
2222 	} else {
2223 		/* Flow passes untransformed */
2224 		dst_release(dst);
2225 		dst = dst_orig;
2226 	}
2227 ok:
2228 	xfrm_pols_put(pols, drop_pols);
2229 	if (dst && dst->xfrm &&
2230 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2231 		dst->flags |= DST_XFRM_TUNNEL;
2232 	return dst;
2233 
2234 nopol:
2235 	if (!(flags & XFRM_LOOKUP_ICMP)) {
2236 		dst = dst_orig;
2237 		goto ok;
2238 	}
2239 	err = -ENOENT;
2240 error:
2241 	dst_release(dst);
2242 dropdst:
2243 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2244 		dst_release(dst_orig);
2245 	xfrm_pols_put(pols, drop_pols);
2246 	return ERR_PTR(err);
2247 }
2248 EXPORT_SYMBOL(xfrm_lookup);
2249 
2250 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2251  * Otherwise we may send out blackholed packets.
2252  */
2253 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2254 				    const struct flowi *fl,
2255 				    const struct sock *sk, int flags)
2256 {
2257 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2258 					    flags | XFRM_LOOKUP_QUEUE |
2259 					    XFRM_LOOKUP_KEEP_DST_REF);
2260 
2261 	if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2262 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
2263 
2264 	return dst;
2265 }
2266 EXPORT_SYMBOL(xfrm_lookup_route);
2267 
2268 static inline int
2269 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2270 {
2271 	struct xfrm_state *x;
2272 
2273 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2274 		return 0;
2275 	x = skb->sp->xvec[idx];
2276 	if (!x->type->reject)
2277 		return 0;
2278 	return x->type->reject(x, skb, fl);
2279 }
2280 
2281 /* When skb is transformed back to its "native" form, we have to
2282  * check policy restrictions. At the moment we make this in maximally
2283  * stupid way. Shame on me. :-) Of course, connected sockets must
2284  * have policy cached at them.
2285  */
2286 
2287 static inline int
2288 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2289 	      unsigned short family)
2290 {
2291 	if (xfrm_state_kern(x))
2292 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2293 	return	x->id.proto == tmpl->id.proto &&
2294 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2295 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2296 		x->props.mode == tmpl->mode &&
2297 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2298 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2299 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
2300 		  xfrm_state_addr_cmp(tmpl, x, family));
2301 }
2302 
2303 /*
2304  * 0 or more than 0 is returned when validation is succeeded (either bypass
2305  * because of optional transport mode, or next index of the mathced secpath
2306  * state with the template.
2307  * -1 is returned when no matching template is found.
2308  * Otherwise "-2 - errored_index" is returned.
2309  */
2310 static inline int
2311 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2312 	       unsigned short family)
2313 {
2314 	int idx = start;
2315 
2316 	if (tmpl->optional) {
2317 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
2318 			return start;
2319 	} else
2320 		start = -1;
2321 	for (; idx < sp->len; idx++) {
2322 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2323 			return ++idx;
2324 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2325 			if (start == -1)
2326 				start = -2-idx;
2327 			break;
2328 		}
2329 	}
2330 	return start;
2331 }
2332 
2333 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2334 			  unsigned int family, int reverse)
2335 {
2336 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2337 	int err;
2338 
2339 	if (unlikely(afinfo == NULL))
2340 		return -EAFNOSUPPORT;
2341 
2342 	afinfo->decode_session(skb, fl, reverse);
2343 	err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2344 	rcu_read_unlock();
2345 	return err;
2346 }
2347 EXPORT_SYMBOL(__xfrm_decode_session);
2348 
2349 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2350 {
2351 	for (; k < sp->len; k++) {
2352 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2353 			*idxp = k;
2354 			return 1;
2355 		}
2356 	}
2357 
2358 	return 0;
2359 }
2360 
2361 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2362 			unsigned short family)
2363 {
2364 	struct net *net = dev_net(skb->dev);
2365 	struct xfrm_policy *pol;
2366 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2367 	int npols = 0;
2368 	int xfrm_nr;
2369 	int pi;
2370 	int reverse;
2371 	struct flowi fl;
2372 	int xerr_idx = -1;
2373 
2374 	reverse = dir & ~XFRM_POLICY_MASK;
2375 	dir &= XFRM_POLICY_MASK;
2376 
2377 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2378 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2379 		return 0;
2380 	}
2381 
2382 	nf_nat_decode_session(skb, &fl, family);
2383 
2384 	/* First, check used SA against their selectors. */
2385 	if (skb->sp) {
2386 		int i;
2387 
2388 		for (i = skb->sp->len-1; i >= 0; i--) {
2389 			struct xfrm_state *x = skb->sp->xvec[i];
2390 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
2391 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2392 				return 0;
2393 			}
2394 		}
2395 	}
2396 
2397 	pol = NULL;
2398 	sk = sk_to_full_sk(sk);
2399 	if (sk && sk->sk_policy[dir]) {
2400 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2401 		if (IS_ERR(pol)) {
2402 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2403 			return 0;
2404 		}
2405 	}
2406 
2407 	if (!pol)
2408 		pol = xfrm_policy_lookup(net, &fl, family, dir);
2409 
2410 	if (IS_ERR(pol)) {
2411 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2412 		return 0;
2413 	}
2414 
2415 	if (!pol) {
2416 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2417 			xfrm_secpath_reject(xerr_idx, skb, &fl);
2418 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2419 			return 0;
2420 		}
2421 		return 1;
2422 	}
2423 
2424 	pol->curlft.use_time = get_seconds();
2425 
2426 	pols[0] = pol;
2427 	npols++;
2428 #ifdef CONFIG_XFRM_SUB_POLICY
2429 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2430 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2431 						    &fl, family,
2432 						    XFRM_POLICY_IN);
2433 		if (pols[1]) {
2434 			if (IS_ERR(pols[1])) {
2435 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2436 				return 0;
2437 			}
2438 			pols[1]->curlft.use_time = get_seconds();
2439 			npols++;
2440 		}
2441 	}
2442 #endif
2443 
2444 	if (pol->action == XFRM_POLICY_ALLOW) {
2445 		struct sec_path *sp;
2446 		static struct sec_path dummy;
2447 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2448 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2449 		struct xfrm_tmpl **tpp = tp;
2450 		int ti = 0;
2451 		int i, k;
2452 
2453 		if ((sp = skb->sp) == NULL)
2454 			sp = &dummy;
2455 
2456 		for (pi = 0; pi < npols; pi++) {
2457 			if (pols[pi] != pol &&
2458 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
2459 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2460 				goto reject;
2461 			}
2462 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2463 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2464 				goto reject_error;
2465 			}
2466 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
2467 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
2468 		}
2469 		xfrm_nr = ti;
2470 		if (npols > 1) {
2471 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2472 			tpp = stp;
2473 		}
2474 
2475 		/* For each tunnel xfrm, find the first matching tmpl.
2476 		 * For each tmpl before that, find corresponding xfrm.
2477 		 * Order is _important_. Later we will implement
2478 		 * some barriers, but at the moment barriers
2479 		 * are implied between each two transformations.
2480 		 */
2481 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2482 			k = xfrm_policy_ok(tpp[i], sp, k, family);
2483 			if (k < 0) {
2484 				if (k < -1)
2485 					/* "-2 - errored_index" returned */
2486 					xerr_idx = -(2+k);
2487 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2488 				goto reject;
2489 			}
2490 		}
2491 
2492 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2493 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2494 			goto reject;
2495 		}
2496 
2497 		xfrm_pols_put(pols, npols);
2498 		return 1;
2499 	}
2500 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2501 
2502 reject:
2503 	xfrm_secpath_reject(xerr_idx, skb, &fl);
2504 reject_error:
2505 	xfrm_pols_put(pols, npols);
2506 	return 0;
2507 }
2508 EXPORT_SYMBOL(__xfrm_policy_check);
2509 
2510 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2511 {
2512 	struct net *net = dev_net(skb->dev);
2513 	struct flowi fl;
2514 	struct dst_entry *dst;
2515 	int res = 1;
2516 
2517 	if (xfrm_decode_session(skb, &fl, family) < 0) {
2518 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2519 		return 0;
2520 	}
2521 
2522 	skb_dst_force(skb);
2523 
2524 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2525 	if (IS_ERR(dst)) {
2526 		res = 0;
2527 		dst = NULL;
2528 	}
2529 	skb_dst_set(skb, dst);
2530 	return res;
2531 }
2532 EXPORT_SYMBOL(__xfrm_route_forward);
2533 
2534 /* Optimize later using cookies and generation ids. */
2535 
2536 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2537 {
2538 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2539 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2540 	 * get validated by dst_ops->check on every use.  We do this
2541 	 * because when a normal route referenced by an XFRM dst is
2542 	 * obsoleted we do not go looking around for all parent
2543 	 * referencing XFRM dsts so that we can invalidate them.  It
2544 	 * is just too much work.  Instead we make the checks here on
2545 	 * every use.  For example:
2546 	 *
2547 	 *	XFRM dst A --> IPv4 dst X
2548 	 *
2549 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
2550 	 * in this example).  If X is marked obsolete, "A" will not
2551 	 * notice.  That's what we are validating here via the
2552 	 * stale_bundle() check.
2553 	 *
2554 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2555 	 * be marked on it.
2556 	 * This will force stale_bundle() to fail on any xdst bundle with
2557 	 * this dst linked in it.
2558 	 */
2559 	if (dst->obsolete < 0 && !stale_bundle(dst))
2560 		return dst;
2561 
2562 	return NULL;
2563 }
2564 
2565 static int stale_bundle(struct dst_entry *dst)
2566 {
2567 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2568 }
2569 
2570 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2571 {
2572 	while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2573 		dst->dev = dev_net(dev)->loopback_dev;
2574 		dev_hold(dst->dev);
2575 		dev_put(dev);
2576 	}
2577 }
2578 EXPORT_SYMBOL(xfrm_dst_ifdown);
2579 
2580 static void xfrm_link_failure(struct sk_buff *skb)
2581 {
2582 	/* Impossible. Such dst must be popped before reaches point of failure. */
2583 }
2584 
2585 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2586 {
2587 	if (dst) {
2588 		if (dst->obsolete) {
2589 			dst_release(dst);
2590 			dst = NULL;
2591 		}
2592 	}
2593 	return dst;
2594 }
2595 
2596 static void xfrm_init_pmtu(struct dst_entry *dst)
2597 {
2598 	do {
2599 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2600 		u32 pmtu, route_mtu_cached;
2601 
2602 		pmtu = dst_mtu(dst->child);
2603 		xdst->child_mtu_cached = pmtu;
2604 
2605 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2606 
2607 		route_mtu_cached = dst_mtu(xdst->route);
2608 		xdst->route_mtu_cached = route_mtu_cached;
2609 
2610 		if (pmtu > route_mtu_cached)
2611 			pmtu = route_mtu_cached;
2612 
2613 		dst_metric_set(dst, RTAX_MTU, pmtu);
2614 	} while ((dst = dst->next));
2615 }
2616 
2617 /* Check that the bundle accepts the flow and its components are
2618  * still valid.
2619  */
2620 
2621 static int xfrm_bundle_ok(struct xfrm_dst *first)
2622 {
2623 	struct dst_entry *dst = &first->u.dst;
2624 	struct xfrm_dst *last;
2625 	u32 mtu;
2626 
2627 	if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2628 	    (dst->dev && !netif_running(dst->dev)))
2629 		return 0;
2630 
2631 	if (dst->flags & DST_XFRM_QUEUE)
2632 		return 1;
2633 
2634 	last = NULL;
2635 
2636 	do {
2637 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2638 
2639 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2640 			return 0;
2641 		if (xdst->xfrm_genid != dst->xfrm->genid)
2642 			return 0;
2643 		if (xdst->num_pols > 0 &&
2644 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2645 			return 0;
2646 
2647 		mtu = dst_mtu(dst->child);
2648 		if (xdst->child_mtu_cached != mtu) {
2649 			last = xdst;
2650 			xdst->child_mtu_cached = mtu;
2651 		}
2652 
2653 		if (!dst_check(xdst->route, xdst->route_cookie))
2654 			return 0;
2655 		mtu = dst_mtu(xdst->route);
2656 		if (xdst->route_mtu_cached != mtu) {
2657 			last = xdst;
2658 			xdst->route_mtu_cached = mtu;
2659 		}
2660 
2661 		dst = dst->child;
2662 	} while (dst->xfrm);
2663 
2664 	if (likely(!last))
2665 		return 1;
2666 
2667 	mtu = last->child_mtu_cached;
2668 	for (;;) {
2669 		dst = &last->u.dst;
2670 
2671 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2672 		if (mtu > last->route_mtu_cached)
2673 			mtu = last->route_mtu_cached;
2674 		dst_metric_set(dst, RTAX_MTU, mtu);
2675 
2676 		if (last == first)
2677 			break;
2678 
2679 		last = (struct xfrm_dst *)last->u.dst.next;
2680 		last->child_mtu_cached = mtu;
2681 	}
2682 
2683 	return 1;
2684 }
2685 
2686 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2687 {
2688 	return dst_metric_advmss(dst->path);
2689 }
2690 
2691 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2692 {
2693 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2694 
2695 	return mtu ? : dst_mtu(dst->path);
2696 }
2697 
2698 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2699 					const void *daddr)
2700 {
2701 	const struct dst_entry *path = dst->path;
2702 
2703 	for (; dst != path; dst = dst->child) {
2704 		const struct xfrm_state *xfrm = dst->xfrm;
2705 
2706 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2707 			continue;
2708 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2709 			daddr = xfrm->coaddr;
2710 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2711 			daddr = &xfrm->id.daddr;
2712 	}
2713 	return daddr;
2714 }
2715 
2716 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2717 					   struct sk_buff *skb,
2718 					   const void *daddr)
2719 {
2720 	const struct dst_entry *path = dst->path;
2721 
2722 	if (!skb)
2723 		daddr = xfrm_get_dst_nexthop(dst, daddr);
2724 	return path->ops->neigh_lookup(path, skb, daddr);
2725 }
2726 
2727 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2728 {
2729 	const struct dst_entry *path = dst->path;
2730 
2731 	daddr = xfrm_get_dst_nexthop(dst, daddr);
2732 	path->ops->confirm_neigh(path, daddr);
2733 }
2734 
2735 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2736 {
2737 	int err = 0;
2738 
2739 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2740 		return -EAFNOSUPPORT;
2741 
2742 	spin_lock(&xfrm_policy_afinfo_lock);
2743 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
2744 		err = -EEXIST;
2745 	else {
2746 		struct dst_ops *dst_ops = afinfo->dst_ops;
2747 		if (likely(dst_ops->kmem_cachep == NULL))
2748 			dst_ops->kmem_cachep = xfrm_dst_cache;
2749 		if (likely(dst_ops->check == NULL))
2750 			dst_ops->check = xfrm_dst_check;
2751 		if (likely(dst_ops->default_advmss == NULL))
2752 			dst_ops->default_advmss = xfrm_default_advmss;
2753 		if (likely(dst_ops->mtu == NULL))
2754 			dst_ops->mtu = xfrm_mtu;
2755 		if (likely(dst_ops->negative_advice == NULL))
2756 			dst_ops->negative_advice = xfrm_negative_advice;
2757 		if (likely(dst_ops->link_failure == NULL))
2758 			dst_ops->link_failure = xfrm_link_failure;
2759 		if (likely(dst_ops->neigh_lookup == NULL))
2760 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
2761 		if (likely(!dst_ops->confirm_neigh))
2762 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
2763 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2764 	}
2765 	spin_unlock(&xfrm_policy_afinfo_lock);
2766 
2767 	return err;
2768 }
2769 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2770 
2771 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2772 {
2773 	struct dst_ops *dst_ops = afinfo->dst_ops;
2774 	int i;
2775 
2776 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2777 		if (xfrm_policy_afinfo[i] != afinfo)
2778 			continue;
2779 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2780 		break;
2781 	}
2782 
2783 	synchronize_rcu();
2784 
2785 	dst_ops->kmem_cachep = NULL;
2786 	dst_ops->check = NULL;
2787 	dst_ops->negative_advice = NULL;
2788 	dst_ops->link_failure = NULL;
2789 }
2790 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2791 
2792 #ifdef CONFIG_XFRM_STATISTICS
2793 static int __net_init xfrm_statistics_init(struct net *net)
2794 {
2795 	int rv;
2796 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2797 	if (!net->mib.xfrm_statistics)
2798 		return -ENOMEM;
2799 	rv = xfrm_proc_init(net);
2800 	if (rv < 0)
2801 		free_percpu(net->mib.xfrm_statistics);
2802 	return rv;
2803 }
2804 
2805 static void xfrm_statistics_fini(struct net *net)
2806 {
2807 	xfrm_proc_fini(net);
2808 	free_percpu(net->mib.xfrm_statistics);
2809 }
2810 #else
2811 static int __net_init xfrm_statistics_init(struct net *net)
2812 {
2813 	return 0;
2814 }
2815 
2816 static void xfrm_statistics_fini(struct net *net)
2817 {
2818 }
2819 #endif
2820 
2821 static int __net_init xfrm_policy_init(struct net *net)
2822 {
2823 	unsigned int hmask, sz;
2824 	int dir;
2825 
2826 	if (net_eq(net, &init_net))
2827 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2828 					   sizeof(struct xfrm_dst),
2829 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2830 					   NULL);
2831 
2832 	hmask = 8 - 1;
2833 	sz = (hmask+1) * sizeof(struct hlist_head);
2834 
2835 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2836 	if (!net->xfrm.policy_byidx)
2837 		goto out_byidx;
2838 	net->xfrm.policy_idx_hmask = hmask;
2839 
2840 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2841 		struct xfrm_policy_hash *htab;
2842 
2843 		net->xfrm.policy_count[dir] = 0;
2844 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2845 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2846 
2847 		htab = &net->xfrm.policy_bydst[dir];
2848 		htab->table = xfrm_hash_alloc(sz);
2849 		if (!htab->table)
2850 			goto out_bydst;
2851 		htab->hmask = hmask;
2852 		htab->dbits4 = 32;
2853 		htab->sbits4 = 32;
2854 		htab->dbits6 = 128;
2855 		htab->sbits6 = 128;
2856 	}
2857 	net->xfrm.policy_hthresh.lbits4 = 32;
2858 	net->xfrm.policy_hthresh.rbits4 = 32;
2859 	net->xfrm.policy_hthresh.lbits6 = 128;
2860 	net->xfrm.policy_hthresh.rbits6 = 128;
2861 
2862 	seqlock_init(&net->xfrm.policy_hthresh.lock);
2863 
2864 	INIT_LIST_HEAD(&net->xfrm.policy_all);
2865 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2866 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2867 	if (net_eq(net, &init_net))
2868 		xfrm_dev_init();
2869 	return 0;
2870 
2871 out_bydst:
2872 	for (dir--; dir >= 0; dir--) {
2873 		struct xfrm_policy_hash *htab;
2874 
2875 		htab = &net->xfrm.policy_bydst[dir];
2876 		xfrm_hash_free(htab->table, sz);
2877 	}
2878 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2879 out_byidx:
2880 	return -ENOMEM;
2881 }
2882 
2883 static void xfrm_policy_fini(struct net *net)
2884 {
2885 	unsigned int sz;
2886 	int dir;
2887 
2888 	flush_work(&net->xfrm.policy_hash_work);
2889 #ifdef CONFIG_XFRM_SUB_POLICY
2890 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2891 #endif
2892 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2893 
2894 	WARN_ON(!list_empty(&net->xfrm.policy_all));
2895 
2896 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2897 		struct xfrm_policy_hash *htab;
2898 
2899 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2900 
2901 		htab = &net->xfrm.policy_bydst[dir];
2902 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2903 		WARN_ON(!hlist_empty(htab->table));
2904 		xfrm_hash_free(htab->table, sz);
2905 	}
2906 
2907 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2908 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2909 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2910 }
2911 
2912 static int __net_init xfrm_net_init(struct net *net)
2913 {
2914 	int rv;
2915 
2916 	/* Initialize the per-net locks here */
2917 	spin_lock_init(&net->xfrm.xfrm_state_lock);
2918 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
2919 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
2920 
2921 	rv = xfrm_statistics_init(net);
2922 	if (rv < 0)
2923 		goto out_statistics;
2924 	rv = xfrm_state_init(net);
2925 	if (rv < 0)
2926 		goto out_state;
2927 	rv = xfrm_policy_init(net);
2928 	if (rv < 0)
2929 		goto out_policy;
2930 	rv = xfrm_sysctl_init(net);
2931 	if (rv < 0)
2932 		goto out_sysctl;
2933 
2934 	return 0;
2935 
2936 out_sysctl:
2937 	xfrm_policy_fini(net);
2938 out_policy:
2939 	xfrm_state_fini(net);
2940 out_state:
2941 	xfrm_statistics_fini(net);
2942 out_statistics:
2943 	return rv;
2944 }
2945 
2946 static void __net_exit xfrm_net_exit(struct net *net)
2947 {
2948 	xfrm_sysctl_fini(net);
2949 	xfrm_policy_fini(net);
2950 	xfrm_state_fini(net);
2951 	xfrm_statistics_fini(net);
2952 }
2953 
2954 static struct pernet_operations __net_initdata xfrm_net_ops = {
2955 	.init = xfrm_net_init,
2956 	.exit = xfrm_net_exit,
2957 };
2958 
2959 void __init xfrm_init(void)
2960 {
2961 	int i;
2962 
2963 	xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2964 				       GFP_KERNEL);
2965 	BUG_ON(!xfrm_pcpu_work);
2966 
2967 	for (i = 0; i < NR_CPUS; i++)
2968 		INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
2969 
2970 	register_pernet_subsys(&xfrm_net_ops);
2971 	seqcount_init(&xfrm_policy_hash_generation);
2972 	xfrm_input_init();
2973 }
2974 
2975 #ifdef CONFIG_AUDITSYSCALL
2976 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2977 					 struct audit_buffer *audit_buf)
2978 {
2979 	struct xfrm_sec_ctx *ctx = xp->security;
2980 	struct xfrm_selector *sel = &xp->selector;
2981 
2982 	if (ctx)
2983 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2984 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2985 
2986 	switch (sel->family) {
2987 	case AF_INET:
2988 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2989 		if (sel->prefixlen_s != 32)
2990 			audit_log_format(audit_buf, " src_prefixlen=%d",
2991 					 sel->prefixlen_s);
2992 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2993 		if (sel->prefixlen_d != 32)
2994 			audit_log_format(audit_buf, " dst_prefixlen=%d",
2995 					 sel->prefixlen_d);
2996 		break;
2997 	case AF_INET6:
2998 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2999 		if (sel->prefixlen_s != 128)
3000 			audit_log_format(audit_buf, " src_prefixlen=%d",
3001 					 sel->prefixlen_s);
3002 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3003 		if (sel->prefixlen_d != 128)
3004 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3005 					 sel->prefixlen_d);
3006 		break;
3007 	}
3008 }
3009 
3010 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3011 {
3012 	struct audit_buffer *audit_buf;
3013 
3014 	audit_buf = xfrm_audit_start("SPD-add");
3015 	if (audit_buf == NULL)
3016 		return;
3017 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3018 	audit_log_format(audit_buf, " res=%u", result);
3019 	xfrm_audit_common_policyinfo(xp, audit_buf);
3020 	audit_log_end(audit_buf);
3021 }
3022 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3023 
3024 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3025 			      bool task_valid)
3026 {
3027 	struct audit_buffer *audit_buf;
3028 
3029 	audit_buf = xfrm_audit_start("SPD-delete");
3030 	if (audit_buf == NULL)
3031 		return;
3032 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3033 	audit_log_format(audit_buf, " res=%u", result);
3034 	xfrm_audit_common_policyinfo(xp, audit_buf);
3035 	audit_log_end(audit_buf);
3036 }
3037 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3038 #endif
3039 
3040 #ifdef CONFIG_XFRM_MIGRATE
3041 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3042 					const struct xfrm_selector *sel_tgt)
3043 {
3044 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3045 		if (sel_tgt->family == sel_cmp->family &&
3046 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3047 				    sel_cmp->family) &&
3048 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3049 				    sel_cmp->family) &&
3050 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3051 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3052 			return true;
3053 		}
3054 	} else {
3055 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3056 			return true;
3057 		}
3058 	}
3059 	return false;
3060 }
3061 
3062 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3063 						    u8 dir, u8 type, struct net *net)
3064 {
3065 	struct xfrm_policy *pol, *ret = NULL;
3066 	struct hlist_head *chain;
3067 	u32 priority = ~0U;
3068 
3069 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3070 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3071 	hlist_for_each_entry(pol, chain, bydst) {
3072 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3073 		    pol->type == type) {
3074 			ret = pol;
3075 			priority = ret->priority;
3076 			break;
3077 		}
3078 	}
3079 	chain = &net->xfrm.policy_inexact[dir];
3080 	hlist_for_each_entry(pol, chain, bydst) {
3081 		if ((pol->priority >= priority) && ret)
3082 			break;
3083 
3084 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3085 		    pol->type == type) {
3086 			ret = pol;
3087 			break;
3088 		}
3089 	}
3090 
3091 	xfrm_pol_hold(ret);
3092 
3093 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3094 
3095 	return ret;
3096 }
3097 
3098 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3099 {
3100 	int match = 0;
3101 
3102 	if (t->mode == m->mode && t->id.proto == m->proto &&
3103 	    (m->reqid == 0 || t->reqid == m->reqid)) {
3104 		switch (t->mode) {
3105 		case XFRM_MODE_TUNNEL:
3106 		case XFRM_MODE_BEET:
3107 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3108 					    m->old_family) &&
3109 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
3110 					    m->old_family)) {
3111 				match = 1;
3112 			}
3113 			break;
3114 		case XFRM_MODE_TRANSPORT:
3115 			/* in case of transport mode, template does not store
3116 			   any IP addresses, hence we just compare mode and
3117 			   protocol */
3118 			match = 1;
3119 			break;
3120 		default:
3121 			break;
3122 		}
3123 	}
3124 	return match;
3125 }
3126 
3127 /* update endpoint address(es) of template(s) */
3128 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3129 			       struct xfrm_migrate *m, int num_migrate)
3130 {
3131 	struct xfrm_migrate *mp;
3132 	int i, j, n = 0;
3133 
3134 	write_lock_bh(&pol->lock);
3135 	if (unlikely(pol->walk.dead)) {
3136 		/* target policy has been deleted */
3137 		write_unlock_bh(&pol->lock);
3138 		return -ENOENT;
3139 	}
3140 
3141 	for (i = 0; i < pol->xfrm_nr; i++) {
3142 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3143 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3144 				continue;
3145 			n++;
3146 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3147 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3148 				continue;
3149 			/* update endpoints */
3150 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3151 			       sizeof(pol->xfrm_vec[i].id.daddr));
3152 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3153 			       sizeof(pol->xfrm_vec[i].saddr));
3154 			pol->xfrm_vec[i].encap_family = mp->new_family;
3155 			/* flush bundles */
3156 			atomic_inc(&pol->genid);
3157 		}
3158 	}
3159 
3160 	write_unlock_bh(&pol->lock);
3161 
3162 	if (!n)
3163 		return -ENODATA;
3164 
3165 	return 0;
3166 }
3167 
3168 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3169 {
3170 	int i, j;
3171 
3172 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3173 		return -EINVAL;
3174 
3175 	for (i = 0; i < num_migrate; i++) {
3176 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3177 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3178 			return -EINVAL;
3179 
3180 		/* check if there is any duplicated entry */
3181 		for (j = i + 1; j < num_migrate; j++) {
3182 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3183 				    sizeof(m[i].old_daddr)) &&
3184 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3185 				    sizeof(m[i].old_saddr)) &&
3186 			    m[i].proto == m[j].proto &&
3187 			    m[i].mode == m[j].mode &&
3188 			    m[i].reqid == m[j].reqid &&
3189 			    m[i].old_family == m[j].old_family)
3190 				return -EINVAL;
3191 		}
3192 	}
3193 
3194 	return 0;
3195 }
3196 
3197 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3198 		 struct xfrm_migrate *m, int num_migrate,
3199 		 struct xfrm_kmaddress *k, struct net *net,
3200 		 struct xfrm_encap_tmpl *encap)
3201 {
3202 	int i, err, nx_cur = 0, nx_new = 0;
3203 	struct xfrm_policy *pol = NULL;
3204 	struct xfrm_state *x, *xc;
3205 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3206 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3207 	struct xfrm_migrate *mp;
3208 
3209 	/* Stage 0 - sanity checks */
3210 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3211 		goto out;
3212 
3213 	if (dir >= XFRM_POLICY_MAX) {
3214 		err = -EINVAL;
3215 		goto out;
3216 	}
3217 
3218 	/* Stage 1 - find policy */
3219 	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3220 		err = -ENOENT;
3221 		goto out;
3222 	}
3223 
3224 	/* Stage 2 - find and update state(s) */
3225 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3226 		if ((x = xfrm_migrate_state_find(mp, net))) {
3227 			x_cur[nx_cur] = x;
3228 			nx_cur++;
3229 			xc = xfrm_state_migrate(x, mp, encap);
3230 			if (xc) {
3231 				x_new[nx_new] = xc;
3232 				nx_new++;
3233 			} else {
3234 				err = -ENODATA;
3235 				goto restore_state;
3236 			}
3237 		}
3238 	}
3239 
3240 	/* Stage 3 - update policy */
3241 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3242 		goto restore_state;
3243 
3244 	/* Stage 4 - delete old state(s) */
3245 	if (nx_cur) {
3246 		xfrm_states_put(x_cur, nx_cur);
3247 		xfrm_states_delete(x_cur, nx_cur);
3248 	}
3249 
3250 	/* Stage 5 - announce */
3251 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
3252 
3253 	xfrm_pol_put(pol);
3254 
3255 	return 0;
3256 out:
3257 	return err;
3258 
3259 restore_state:
3260 	if (pol)
3261 		xfrm_pol_put(pol);
3262 	if (nx_cur)
3263 		xfrm_states_put(x_cur, nx_cur);
3264 	if (nx_new)
3265 		xfrm_states_delete(x_new, nx_new);
3266 
3267 	return err;
3268 }
3269 EXPORT_SYMBOL(xfrm_migrate);
3270 #endif
3271