xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision e3b9f1e8)
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 xfrm_dst **bundle, int nr);
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 (policy->walk.dead ||
613 		    xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
614 			/* skip socket policies */
615 			continue;
616 		}
617 		newpos = NULL;
618 		chain = policy_hash_bysel(net, &policy->selector,
619 					  policy->family,
620 					  xfrm_policy_id2dir(policy->index));
621 		hlist_for_each_entry(pol, chain, bydst) {
622 			if (policy->priority >= pol->priority)
623 				newpos = &pol->bydst;
624 			else
625 				break;
626 		}
627 		if (newpos)
628 			hlist_add_behind(&policy->bydst, newpos);
629 		else
630 			hlist_add_head(&policy->bydst, chain);
631 	}
632 
633 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
634 
635 	mutex_unlock(&hash_resize_mutex);
636 }
637 
638 void xfrm_policy_hash_rebuild(struct net *net)
639 {
640 	schedule_work(&net->xfrm.policy_hthresh.work);
641 }
642 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
643 
644 /* Generate new index... KAME seems to generate them ordered by cost
645  * of an absolute inpredictability of ordering of rules. This will not pass. */
646 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
647 {
648 	static u32 idx_generator;
649 
650 	for (;;) {
651 		struct hlist_head *list;
652 		struct xfrm_policy *p;
653 		u32 idx;
654 		int found;
655 
656 		if (!index) {
657 			idx = (idx_generator | dir);
658 			idx_generator += 8;
659 		} else {
660 			idx = index;
661 			index = 0;
662 		}
663 
664 		if (idx == 0)
665 			idx = 8;
666 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
667 		found = 0;
668 		hlist_for_each_entry(p, list, byidx) {
669 			if (p->index == idx) {
670 				found = 1;
671 				break;
672 			}
673 		}
674 		if (!found)
675 			return idx;
676 	}
677 }
678 
679 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
680 {
681 	u32 *p1 = (u32 *) s1;
682 	u32 *p2 = (u32 *) s2;
683 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
684 	int i;
685 
686 	for (i = 0; i < len; i++) {
687 		if (p1[i] != p2[i])
688 			return 1;
689 	}
690 
691 	return 0;
692 }
693 
694 static void xfrm_policy_requeue(struct xfrm_policy *old,
695 				struct xfrm_policy *new)
696 {
697 	struct xfrm_policy_queue *pq = &old->polq;
698 	struct sk_buff_head list;
699 
700 	if (skb_queue_empty(&pq->hold_queue))
701 		return;
702 
703 	__skb_queue_head_init(&list);
704 
705 	spin_lock_bh(&pq->hold_queue.lock);
706 	skb_queue_splice_init(&pq->hold_queue, &list);
707 	if (del_timer(&pq->hold_timer))
708 		xfrm_pol_put(old);
709 	spin_unlock_bh(&pq->hold_queue.lock);
710 
711 	pq = &new->polq;
712 
713 	spin_lock_bh(&pq->hold_queue.lock);
714 	skb_queue_splice(&list, &pq->hold_queue);
715 	pq->timeout = XFRM_QUEUE_TMO_MIN;
716 	if (!mod_timer(&pq->hold_timer, jiffies))
717 		xfrm_pol_hold(new);
718 	spin_unlock_bh(&pq->hold_queue.lock);
719 }
720 
721 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
722 				   struct xfrm_policy *pol)
723 {
724 	u32 mark = policy->mark.v & policy->mark.m;
725 
726 	if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
727 		return true;
728 
729 	if ((mark & pol->mark.m) == pol->mark.v &&
730 	    policy->priority == pol->priority)
731 		return true;
732 
733 	return false;
734 }
735 
736 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
737 {
738 	struct net *net = xp_net(policy);
739 	struct xfrm_policy *pol;
740 	struct xfrm_policy *delpol;
741 	struct hlist_head *chain;
742 	struct hlist_node *newpos;
743 
744 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
745 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
746 	delpol = NULL;
747 	newpos = NULL;
748 	hlist_for_each_entry(pol, chain, bydst) {
749 		if (pol->type == policy->type &&
750 		    !selector_cmp(&pol->selector, &policy->selector) &&
751 		    xfrm_policy_mark_match(policy, pol) &&
752 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
753 		    !WARN_ON(delpol)) {
754 			if (excl) {
755 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
756 				return -EEXIST;
757 			}
758 			delpol = pol;
759 			if (policy->priority > pol->priority)
760 				continue;
761 		} else if (policy->priority >= pol->priority) {
762 			newpos = &pol->bydst;
763 			continue;
764 		}
765 		if (delpol)
766 			break;
767 	}
768 	if (newpos)
769 		hlist_add_behind(&policy->bydst, newpos);
770 	else
771 		hlist_add_head(&policy->bydst, chain);
772 	__xfrm_policy_link(policy, dir);
773 
774 	/* After previous checking, family can either be AF_INET or AF_INET6 */
775 	if (policy->family == AF_INET)
776 		rt_genid_bump_ipv4(net);
777 	else
778 		rt_genid_bump_ipv6(net);
779 
780 	if (delpol) {
781 		xfrm_policy_requeue(delpol, policy);
782 		__xfrm_policy_unlink(delpol, dir);
783 	}
784 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
785 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
786 	policy->curlft.add_time = get_seconds();
787 	policy->curlft.use_time = 0;
788 	if (!mod_timer(&policy->timer, jiffies + HZ))
789 		xfrm_pol_hold(policy);
790 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
791 
792 	if (delpol)
793 		xfrm_policy_kill(delpol);
794 	else if (xfrm_bydst_should_resize(net, dir, NULL))
795 		schedule_work(&net->xfrm.policy_hash_work);
796 
797 	return 0;
798 }
799 EXPORT_SYMBOL(xfrm_policy_insert);
800 
801 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
802 					  int dir, struct xfrm_selector *sel,
803 					  struct xfrm_sec_ctx *ctx, int delete,
804 					  int *err)
805 {
806 	struct xfrm_policy *pol, *ret;
807 	struct hlist_head *chain;
808 
809 	*err = 0;
810 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
811 	chain = policy_hash_bysel(net, sel, sel->family, dir);
812 	ret = NULL;
813 	hlist_for_each_entry(pol, chain, bydst) {
814 		if (pol->type == type &&
815 		    (mark & pol->mark.m) == pol->mark.v &&
816 		    !selector_cmp(sel, &pol->selector) &&
817 		    xfrm_sec_ctx_match(ctx, pol->security)) {
818 			xfrm_pol_hold(pol);
819 			if (delete) {
820 				*err = security_xfrm_policy_delete(
821 								pol->security);
822 				if (*err) {
823 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
824 					return pol;
825 				}
826 				__xfrm_policy_unlink(pol, dir);
827 			}
828 			ret = pol;
829 			break;
830 		}
831 	}
832 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
833 
834 	if (ret && delete)
835 		xfrm_policy_kill(ret);
836 	return ret;
837 }
838 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
839 
840 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
841 				     int dir, u32 id, int delete, int *err)
842 {
843 	struct xfrm_policy *pol, *ret;
844 	struct hlist_head *chain;
845 
846 	*err = -ENOENT;
847 	if (xfrm_policy_id2dir(id) != dir)
848 		return NULL;
849 
850 	*err = 0;
851 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
852 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
853 	ret = NULL;
854 	hlist_for_each_entry(pol, chain, byidx) {
855 		if (pol->type == type && pol->index == id &&
856 		    (mark & pol->mark.m) == pol->mark.v) {
857 			xfrm_pol_hold(pol);
858 			if (delete) {
859 				*err = security_xfrm_policy_delete(
860 								pol->security);
861 				if (*err) {
862 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
863 					return pol;
864 				}
865 				__xfrm_policy_unlink(pol, dir);
866 			}
867 			ret = pol;
868 			break;
869 		}
870 	}
871 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
872 
873 	if (ret && delete)
874 		xfrm_policy_kill(ret);
875 	return ret;
876 }
877 EXPORT_SYMBOL(xfrm_policy_byid);
878 
879 #ifdef CONFIG_SECURITY_NETWORK_XFRM
880 static inline int
881 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
882 {
883 	int dir, err = 0;
884 
885 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
886 		struct xfrm_policy *pol;
887 		int i;
888 
889 		hlist_for_each_entry(pol,
890 				     &net->xfrm.policy_inexact[dir], bydst) {
891 			if (pol->type != type)
892 				continue;
893 			err = security_xfrm_policy_delete(pol->security);
894 			if (err) {
895 				xfrm_audit_policy_delete(pol, 0, task_valid);
896 				return err;
897 			}
898 		}
899 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
900 			hlist_for_each_entry(pol,
901 					     net->xfrm.policy_bydst[dir].table + i,
902 					     bydst) {
903 				if (pol->type != type)
904 					continue;
905 				err = security_xfrm_policy_delete(
906 								pol->security);
907 				if (err) {
908 					xfrm_audit_policy_delete(pol, 0,
909 								 task_valid);
910 					return err;
911 				}
912 			}
913 		}
914 	}
915 	return err;
916 }
917 #else
918 static inline int
919 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
920 {
921 	return 0;
922 }
923 #endif
924 
925 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
926 {
927 	int dir, err = 0, cnt = 0;
928 
929 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
930 
931 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
932 	if (err)
933 		goto out;
934 
935 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
936 		struct xfrm_policy *pol;
937 		int i;
938 
939 	again1:
940 		hlist_for_each_entry(pol,
941 				     &net->xfrm.policy_inexact[dir], bydst) {
942 			if (pol->type != type)
943 				continue;
944 			__xfrm_policy_unlink(pol, dir);
945 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
946 			cnt++;
947 
948 			xfrm_audit_policy_delete(pol, 1, task_valid);
949 
950 			xfrm_policy_kill(pol);
951 
952 			spin_lock_bh(&net->xfrm.xfrm_policy_lock);
953 			goto again1;
954 		}
955 
956 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
957 	again2:
958 			hlist_for_each_entry(pol,
959 					     net->xfrm.policy_bydst[dir].table + i,
960 					     bydst) {
961 				if (pol->type != type)
962 					continue;
963 				__xfrm_policy_unlink(pol, dir);
964 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
965 				cnt++;
966 
967 				xfrm_audit_policy_delete(pol, 1, task_valid);
968 				xfrm_policy_kill(pol);
969 
970 				spin_lock_bh(&net->xfrm.xfrm_policy_lock);
971 				goto again2;
972 			}
973 		}
974 
975 	}
976 	if (!cnt)
977 		err = -ESRCH;
978 out:
979 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
980 	return err;
981 }
982 EXPORT_SYMBOL(xfrm_policy_flush);
983 
984 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
985 		     int (*func)(struct xfrm_policy *, int, int, void*),
986 		     void *data)
987 {
988 	struct xfrm_policy *pol;
989 	struct xfrm_policy_walk_entry *x;
990 	int error = 0;
991 
992 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
993 	    walk->type != XFRM_POLICY_TYPE_ANY)
994 		return -EINVAL;
995 
996 	if (list_empty(&walk->walk.all) && walk->seq != 0)
997 		return 0;
998 
999 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1000 	if (list_empty(&walk->walk.all))
1001 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1002 	else
1003 		x = list_first_entry(&walk->walk.all,
1004 				     struct xfrm_policy_walk_entry, all);
1005 
1006 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1007 		if (x->dead)
1008 			continue;
1009 		pol = container_of(x, struct xfrm_policy, walk);
1010 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1011 		    walk->type != pol->type)
1012 			continue;
1013 		error = func(pol, xfrm_policy_id2dir(pol->index),
1014 			     walk->seq, data);
1015 		if (error) {
1016 			list_move_tail(&walk->walk.all, &x->all);
1017 			goto out;
1018 		}
1019 		walk->seq++;
1020 	}
1021 	if (walk->seq == 0) {
1022 		error = -ENOENT;
1023 		goto out;
1024 	}
1025 	list_del_init(&walk->walk.all);
1026 out:
1027 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1028 	return error;
1029 }
1030 EXPORT_SYMBOL(xfrm_policy_walk);
1031 
1032 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1033 {
1034 	INIT_LIST_HEAD(&walk->walk.all);
1035 	walk->walk.dead = 1;
1036 	walk->type = type;
1037 	walk->seq = 0;
1038 }
1039 EXPORT_SYMBOL(xfrm_policy_walk_init);
1040 
1041 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1042 {
1043 	if (list_empty(&walk->walk.all))
1044 		return;
1045 
1046 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1047 	list_del(&walk->walk.all);
1048 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1049 }
1050 EXPORT_SYMBOL(xfrm_policy_walk_done);
1051 
1052 /*
1053  * Find policy to apply to this flow.
1054  *
1055  * Returns 0 if policy found, else an -errno.
1056  */
1057 static int xfrm_policy_match(const struct xfrm_policy *pol,
1058 			     const struct flowi *fl,
1059 			     u8 type, u16 family, int dir)
1060 {
1061 	const struct xfrm_selector *sel = &pol->selector;
1062 	int ret = -ESRCH;
1063 	bool match;
1064 
1065 	if (pol->family != family ||
1066 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1067 	    pol->type != type)
1068 		return ret;
1069 
1070 	match = xfrm_selector_match(sel, fl, family);
1071 	if (match)
1072 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1073 						  dir);
1074 
1075 	return ret;
1076 }
1077 
1078 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1079 						     const struct flowi *fl,
1080 						     u16 family, u8 dir)
1081 {
1082 	int err;
1083 	struct xfrm_policy *pol, *ret;
1084 	const xfrm_address_t *daddr, *saddr;
1085 	struct hlist_head *chain;
1086 	unsigned int sequence;
1087 	u32 priority;
1088 
1089 	daddr = xfrm_flowi_daddr(fl, family);
1090 	saddr = xfrm_flowi_saddr(fl, family);
1091 	if (unlikely(!daddr || !saddr))
1092 		return NULL;
1093 
1094 	rcu_read_lock();
1095  retry:
1096 	do {
1097 		sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1098 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
1099 	} while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1100 
1101 	priority = ~0U;
1102 	ret = NULL;
1103 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1104 		err = xfrm_policy_match(pol, fl, type, family, dir);
1105 		if (err) {
1106 			if (err == -ESRCH)
1107 				continue;
1108 			else {
1109 				ret = ERR_PTR(err);
1110 				goto fail;
1111 			}
1112 		} else {
1113 			ret = pol;
1114 			priority = ret->priority;
1115 			break;
1116 		}
1117 	}
1118 	chain = &net->xfrm.policy_inexact[dir];
1119 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1120 		if ((pol->priority >= priority) && ret)
1121 			break;
1122 
1123 		err = xfrm_policy_match(pol, fl, type, family, dir);
1124 		if (err) {
1125 			if (err == -ESRCH)
1126 				continue;
1127 			else {
1128 				ret = ERR_PTR(err);
1129 				goto fail;
1130 			}
1131 		} else {
1132 			ret = pol;
1133 			break;
1134 		}
1135 	}
1136 
1137 	if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1138 		goto retry;
1139 
1140 	if (ret && !xfrm_pol_hold_rcu(ret))
1141 		goto retry;
1142 fail:
1143 	rcu_read_unlock();
1144 
1145 	return ret;
1146 }
1147 
1148 static struct xfrm_policy *
1149 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1150 {
1151 #ifdef CONFIG_XFRM_SUB_POLICY
1152 	struct xfrm_policy *pol;
1153 
1154 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1155 	if (pol != NULL)
1156 		return pol;
1157 #endif
1158 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1159 }
1160 
1161 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1162 						 const struct flowi *fl, u16 family)
1163 {
1164 	struct xfrm_policy *pol;
1165 
1166 	rcu_read_lock();
1167  again:
1168 	pol = rcu_dereference(sk->sk_policy[dir]);
1169 	if (pol != NULL) {
1170 		bool match;
1171 		int err = 0;
1172 
1173 		if (pol->family != family) {
1174 			pol = NULL;
1175 			goto out;
1176 		}
1177 
1178 		match = xfrm_selector_match(&pol->selector, fl, family);
1179 		if (match) {
1180 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1181 				pol = NULL;
1182 				goto out;
1183 			}
1184 			err = security_xfrm_policy_lookup(pol->security,
1185 						      fl->flowi_secid,
1186 						      dir);
1187 			if (!err) {
1188 				if (!xfrm_pol_hold_rcu(pol))
1189 					goto again;
1190 			} else if (err == -ESRCH) {
1191 				pol = NULL;
1192 			} else {
1193 				pol = ERR_PTR(err);
1194 			}
1195 		} else
1196 			pol = NULL;
1197 	}
1198 out:
1199 	rcu_read_unlock();
1200 	return pol;
1201 }
1202 
1203 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1204 {
1205 	struct net *net = xp_net(pol);
1206 
1207 	list_add(&pol->walk.all, &net->xfrm.policy_all);
1208 	net->xfrm.policy_count[dir]++;
1209 	xfrm_pol_hold(pol);
1210 }
1211 
1212 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1213 						int dir)
1214 {
1215 	struct net *net = xp_net(pol);
1216 
1217 	if (list_empty(&pol->walk.all))
1218 		return NULL;
1219 
1220 	/* Socket policies are not hashed. */
1221 	if (!hlist_unhashed(&pol->bydst)) {
1222 		hlist_del_rcu(&pol->bydst);
1223 		hlist_del(&pol->byidx);
1224 	}
1225 
1226 	list_del_init(&pol->walk.all);
1227 	net->xfrm.policy_count[dir]--;
1228 
1229 	return pol;
1230 }
1231 
1232 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1233 {
1234 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1235 }
1236 
1237 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1238 {
1239 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1240 }
1241 
1242 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1243 {
1244 	struct net *net = xp_net(pol);
1245 
1246 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1247 	pol = __xfrm_policy_unlink(pol, dir);
1248 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1249 	if (pol) {
1250 		xfrm_policy_kill(pol);
1251 		return 0;
1252 	}
1253 	return -ENOENT;
1254 }
1255 EXPORT_SYMBOL(xfrm_policy_delete);
1256 
1257 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1258 {
1259 	struct net *net = sock_net(sk);
1260 	struct xfrm_policy *old_pol;
1261 
1262 #ifdef CONFIG_XFRM_SUB_POLICY
1263 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1264 		return -EINVAL;
1265 #endif
1266 
1267 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1268 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1269 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1270 	if (pol) {
1271 		pol->curlft.add_time = get_seconds();
1272 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1273 		xfrm_sk_policy_link(pol, dir);
1274 	}
1275 	rcu_assign_pointer(sk->sk_policy[dir], pol);
1276 	if (old_pol) {
1277 		if (pol)
1278 			xfrm_policy_requeue(old_pol, pol);
1279 
1280 		/* Unlinking succeeds always. This is the only function
1281 		 * allowed to delete or replace socket policy.
1282 		 */
1283 		xfrm_sk_policy_unlink(old_pol, dir);
1284 	}
1285 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1286 
1287 	if (old_pol) {
1288 		xfrm_policy_kill(old_pol);
1289 	}
1290 	return 0;
1291 }
1292 
1293 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1294 {
1295 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1296 	struct net *net = xp_net(old);
1297 
1298 	if (newp) {
1299 		newp->selector = old->selector;
1300 		if (security_xfrm_policy_clone(old->security,
1301 					       &newp->security)) {
1302 			kfree(newp);
1303 			return NULL;  /* ENOMEM */
1304 		}
1305 		newp->lft = old->lft;
1306 		newp->curlft = old->curlft;
1307 		newp->mark = old->mark;
1308 		newp->action = old->action;
1309 		newp->flags = old->flags;
1310 		newp->xfrm_nr = old->xfrm_nr;
1311 		newp->index = old->index;
1312 		newp->type = old->type;
1313 		newp->family = old->family;
1314 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1315 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1316 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1317 		xfrm_sk_policy_link(newp, dir);
1318 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1319 		xfrm_pol_put(newp);
1320 	}
1321 	return newp;
1322 }
1323 
1324 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1325 {
1326 	const struct xfrm_policy *p;
1327 	struct xfrm_policy *np;
1328 	int i, ret = 0;
1329 
1330 	rcu_read_lock();
1331 	for (i = 0; i < 2; i++) {
1332 		p = rcu_dereference(osk->sk_policy[i]);
1333 		if (p) {
1334 			np = clone_policy(p, i);
1335 			if (unlikely(!np)) {
1336 				ret = -ENOMEM;
1337 				break;
1338 			}
1339 			rcu_assign_pointer(sk->sk_policy[i], np);
1340 		}
1341 	}
1342 	rcu_read_unlock();
1343 	return ret;
1344 }
1345 
1346 static int
1347 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1348 	       xfrm_address_t *remote, unsigned short family, u32 mark)
1349 {
1350 	int err;
1351 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1352 
1353 	if (unlikely(afinfo == NULL))
1354 		return -EINVAL;
1355 	err = afinfo->get_saddr(net, oif, local, remote, mark);
1356 	rcu_read_unlock();
1357 	return err;
1358 }
1359 
1360 /* Resolve list of templates for the flow, given policy. */
1361 
1362 static int
1363 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1364 		      struct xfrm_state **xfrm, unsigned short family)
1365 {
1366 	struct net *net = xp_net(policy);
1367 	int nx;
1368 	int i, error;
1369 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1370 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1371 	xfrm_address_t tmp;
1372 
1373 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1374 		struct xfrm_state *x;
1375 		xfrm_address_t *remote = daddr;
1376 		xfrm_address_t *local  = saddr;
1377 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1378 
1379 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1380 		    tmpl->mode == XFRM_MODE_BEET) {
1381 			remote = &tmpl->id.daddr;
1382 			local = &tmpl->saddr;
1383 			if (xfrm_addr_any(local, tmpl->encap_family)) {
1384 				error = xfrm_get_saddr(net, fl->flowi_oif,
1385 						       &tmp, remote,
1386 						       tmpl->encap_family, 0);
1387 				if (error)
1388 					goto fail;
1389 				local = &tmp;
1390 			}
1391 		}
1392 
1393 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1394 
1395 		if (x && x->km.state == XFRM_STATE_VALID) {
1396 			xfrm[nx++] = x;
1397 			daddr = remote;
1398 			saddr = local;
1399 			continue;
1400 		}
1401 		if (x) {
1402 			error = (x->km.state == XFRM_STATE_ERROR ?
1403 				 -EINVAL : -EAGAIN);
1404 			xfrm_state_put(x);
1405 		} else if (error == -ESRCH) {
1406 			error = -EAGAIN;
1407 		}
1408 
1409 		if (!tmpl->optional)
1410 			goto fail;
1411 	}
1412 	return nx;
1413 
1414 fail:
1415 	for (nx--; nx >= 0; nx--)
1416 		xfrm_state_put(xfrm[nx]);
1417 	return error;
1418 }
1419 
1420 static int
1421 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1422 		  struct xfrm_state **xfrm, unsigned short family)
1423 {
1424 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1425 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1426 	int cnx = 0;
1427 	int error;
1428 	int ret;
1429 	int i;
1430 
1431 	for (i = 0; i < npols; i++) {
1432 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1433 			error = -ENOBUFS;
1434 			goto fail;
1435 		}
1436 
1437 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1438 		if (ret < 0) {
1439 			error = ret;
1440 			goto fail;
1441 		} else
1442 			cnx += ret;
1443 	}
1444 
1445 	/* found states are sorted for outbound processing */
1446 	if (npols > 1)
1447 		xfrm_state_sort(xfrm, tpp, cnx, family);
1448 
1449 	return cnx;
1450 
1451  fail:
1452 	for (cnx--; cnx >= 0; cnx--)
1453 		xfrm_state_put(tpp[cnx]);
1454 	return error;
1455 
1456 }
1457 
1458 static int xfrm_get_tos(const struct flowi *fl, int family)
1459 {
1460 	const struct xfrm_policy_afinfo *afinfo;
1461 	int tos = 0;
1462 
1463 	afinfo = xfrm_policy_get_afinfo(family);
1464 	tos = afinfo ? afinfo->get_tos(fl) : 0;
1465 
1466 	rcu_read_unlock();
1467 
1468 	return tos;
1469 }
1470 
1471 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1472 {
1473 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1474 	struct dst_ops *dst_ops;
1475 	struct xfrm_dst *xdst;
1476 
1477 	if (!afinfo)
1478 		return ERR_PTR(-EINVAL);
1479 
1480 	switch (family) {
1481 	case AF_INET:
1482 		dst_ops = &net->xfrm.xfrm4_dst_ops;
1483 		break;
1484 #if IS_ENABLED(CONFIG_IPV6)
1485 	case AF_INET6:
1486 		dst_ops = &net->xfrm.xfrm6_dst_ops;
1487 		break;
1488 #endif
1489 	default:
1490 		BUG();
1491 	}
1492 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1493 
1494 	if (likely(xdst)) {
1495 		struct dst_entry *dst = &xdst->u.dst;
1496 
1497 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1498 	} else
1499 		xdst = ERR_PTR(-ENOBUFS);
1500 
1501 	rcu_read_unlock();
1502 
1503 	return xdst;
1504 }
1505 
1506 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1507 				 int nfheader_len)
1508 {
1509 	const struct xfrm_policy_afinfo *afinfo =
1510 		xfrm_policy_get_afinfo(dst->ops->family);
1511 	int err;
1512 
1513 	if (!afinfo)
1514 		return -EINVAL;
1515 
1516 	err = afinfo->init_path(path, dst, nfheader_len);
1517 
1518 	rcu_read_unlock();
1519 
1520 	return err;
1521 }
1522 
1523 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1524 				const struct flowi *fl)
1525 {
1526 	const struct xfrm_policy_afinfo *afinfo =
1527 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1528 	int err;
1529 
1530 	if (!afinfo)
1531 		return -EINVAL;
1532 
1533 	err = afinfo->fill_dst(xdst, dev, fl);
1534 
1535 	rcu_read_unlock();
1536 
1537 	return err;
1538 }
1539 
1540 
1541 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1542  * all the metrics... Shortly, bundle a bundle.
1543  */
1544 
1545 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1546 					    struct xfrm_state **xfrm,
1547 					    struct xfrm_dst **bundle,
1548 					    int nx,
1549 					    const struct flowi *fl,
1550 					    struct dst_entry *dst)
1551 {
1552 	struct net *net = xp_net(policy);
1553 	unsigned long now = jiffies;
1554 	struct net_device *dev;
1555 	struct xfrm_mode *inner_mode;
1556 	struct xfrm_dst *xdst_prev = NULL;
1557 	struct xfrm_dst *xdst0 = NULL;
1558 	int i = 0;
1559 	int err;
1560 	int header_len = 0;
1561 	int nfheader_len = 0;
1562 	int trailer_len = 0;
1563 	int tos;
1564 	int family = policy->selector.family;
1565 	xfrm_address_t saddr, daddr;
1566 
1567 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1568 
1569 	tos = xfrm_get_tos(fl, family);
1570 
1571 	dst_hold(dst);
1572 
1573 	for (; i < nx; i++) {
1574 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1575 		struct dst_entry *dst1 = &xdst->u.dst;
1576 
1577 		err = PTR_ERR(xdst);
1578 		if (IS_ERR(xdst)) {
1579 			dst_release(dst);
1580 			goto put_states;
1581 		}
1582 
1583 		bundle[i] = xdst;
1584 		if (!xdst_prev)
1585 			xdst0 = xdst;
1586 		else
1587 			/* Ref count is taken during xfrm_alloc_dst()
1588 			 * No need to do dst_clone() on dst1
1589 			 */
1590 			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
1591 
1592 		if (xfrm[i]->sel.family == AF_UNSPEC) {
1593 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
1594 							xfrm_af2proto(family));
1595 			if (!inner_mode) {
1596 				err = -EAFNOSUPPORT;
1597 				dst_release(dst);
1598 				goto put_states;
1599 			}
1600 		} else
1601 			inner_mode = xfrm[i]->inner_mode;
1602 
1603 		xdst->route = dst;
1604 		dst_copy_metrics(dst1, dst);
1605 
1606 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1607 			family = xfrm[i]->props.family;
1608 			dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1609 					      &saddr, &daddr, family,
1610 					      xfrm[i]->props.output_mark);
1611 			err = PTR_ERR(dst);
1612 			if (IS_ERR(dst))
1613 				goto put_states;
1614 		} else
1615 			dst_hold(dst);
1616 
1617 		dst1->xfrm = xfrm[i];
1618 		xdst->xfrm_genid = xfrm[i]->genid;
1619 
1620 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1621 		dst1->flags |= DST_HOST;
1622 		dst1->lastuse = now;
1623 
1624 		dst1->input = dst_discard;
1625 		dst1->output = inner_mode->afinfo->output;
1626 
1627 		xdst_prev = xdst;
1628 
1629 		header_len += xfrm[i]->props.header_len;
1630 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1631 			nfheader_len += xfrm[i]->props.header_len;
1632 		trailer_len += xfrm[i]->props.trailer_len;
1633 	}
1634 
1635 	xfrm_dst_set_child(xdst_prev, dst);
1636 	xdst0->path = dst;
1637 
1638 	err = -ENODEV;
1639 	dev = dst->dev;
1640 	if (!dev)
1641 		goto free_dst;
1642 
1643 	xfrm_init_path(xdst0, dst, nfheader_len);
1644 	xfrm_init_pmtu(bundle, nx);
1645 
1646 	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
1647 	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
1648 		err = xfrm_fill_dst(xdst_prev, dev, fl);
1649 		if (err)
1650 			goto free_dst;
1651 
1652 		xdst_prev->u.dst.header_len = header_len;
1653 		xdst_prev->u.dst.trailer_len = trailer_len;
1654 		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
1655 		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
1656 	}
1657 
1658 out:
1659 	return &xdst0->u.dst;
1660 
1661 put_states:
1662 	for (; i < nx; i++)
1663 		xfrm_state_put(xfrm[i]);
1664 free_dst:
1665 	if (xdst0)
1666 		dst_release_immediate(&xdst0->u.dst);
1667 	xdst0 = ERR_PTR(err);
1668 	goto out;
1669 }
1670 
1671 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1672 				struct xfrm_policy **pols,
1673 				int *num_pols, int *num_xfrms)
1674 {
1675 	int i;
1676 
1677 	if (*num_pols == 0 || !pols[0]) {
1678 		*num_pols = 0;
1679 		*num_xfrms = 0;
1680 		return 0;
1681 	}
1682 	if (IS_ERR(pols[0]))
1683 		return PTR_ERR(pols[0]);
1684 
1685 	*num_xfrms = pols[0]->xfrm_nr;
1686 
1687 #ifdef CONFIG_XFRM_SUB_POLICY
1688 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1689 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1690 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1691 						    XFRM_POLICY_TYPE_MAIN,
1692 						    fl, family,
1693 						    XFRM_POLICY_OUT);
1694 		if (pols[1]) {
1695 			if (IS_ERR(pols[1])) {
1696 				xfrm_pols_put(pols, *num_pols);
1697 				return PTR_ERR(pols[1]);
1698 			}
1699 			(*num_pols)++;
1700 			(*num_xfrms) += pols[1]->xfrm_nr;
1701 		}
1702 	}
1703 #endif
1704 	for (i = 0; i < *num_pols; i++) {
1705 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
1706 			*num_xfrms = -1;
1707 			break;
1708 		}
1709 	}
1710 
1711 	return 0;
1712 
1713 }
1714 
1715 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1716 {
1717 	this_cpu_write(xfrm_last_dst, xdst);
1718 	if (old)
1719 		dst_release(&old->u.dst);
1720 }
1721 
1722 static void __xfrm_pcpu_work_fn(void)
1723 {
1724 	struct xfrm_dst *old;
1725 
1726 	old = this_cpu_read(xfrm_last_dst);
1727 	if (old && !xfrm_bundle_ok(old))
1728 		xfrm_last_dst_update(NULL, old);
1729 }
1730 
1731 static void xfrm_pcpu_work_fn(struct work_struct *work)
1732 {
1733 	local_bh_disable();
1734 	rcu_read_lock();
1735 	__xfrm_pcpu_work_fn();
1736 	rcu_read_unlock();
1737 	local_bh_enable();
1738 }
1739 
1740 void xfrm_policy_cache_flush(void)
1741 {
1742 	struct xfrm_dst *old;
1743 	bool found = 0;
1744 	int cpu;
1745 
1746 	might_sleep();
1747 
1748 	local_bh_disable();
1749 	rcu_read_lock();
1750 	for_each_possible_cpu(cpu) {
1751 		old = per_cpu(xfrm_last_dst, cpu);
1752 		if (old && !xfrm_bundle_ok(old)) {
1753 			if (smp_processor_id() == cpu) {
1754 				__xfrm_pcpu_work_fn();
1755 				continue;
1756 			}
1757 			found = true;
1758 			break;
1759 		}
1760 	}
1761 
1762 	rcu_read_unlock();
1763 	local_bh_enable();
1764 
1765 	if (!found)
1766 		return;
1767 
1768 	get_online_cpus();
1769 
1770 	for_each_possible_cpu(cpu) {
1771 		bool bundle_release;
1772 
1773 		rcu_read_lock();
1774 		old = per_cpu(xfrm_last_dst, cpu);
1775 		bundle_release = old && !xfrm_bundle_ok(old);
1776 		rcu_read_unlock();
1777 
1778 		if (!bundle_release)
1779 			continue;
1780 
1781 		if (cpu_online(cpu)) {
1782 			schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1783 			continue;
1784 		}
1785 
1786 		rcu_read_lock();
1787 		old = per_cpu(xfrm_last_dst, cpu);
1788 		if (old && !xfrm_bundle_ok(old)) {
1789 			per_cpu(xfrm_last_dst, cpu) = NULL;
1790 			dst_release(&old->u.dst);
1791 		}
1792 		rcu_read_unlock();
1793 	}
1794 
1795 	put_online_cpus();
1796 }
1797 
1798 static bool xfrm_xdst_can_reuse(struct xfrm_dst *xdst,
1799 				struct xfrm_state * const xfrm[],
1800 				int num)
1801 {
1802 	const struct dst_entry *dst = &xdst->u.dst;
1803 	int i;
1804 
1805 	if (xdst->num_xfrms != num)
1806 		return false;
1807 
1808 	for (i = 0; i < num; i++) {
1809 		if (!dst || dst->xfrm != xfrm[i])
1810 			return false;
1811 		dst = xfrm_dst_child(dst);
1812 	}
1813 
1814 	return xfrm_bundle_ok(xdst);
1815 }
1816 
1817 static struct xfrm_dst *
1818 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1819 			       const struct flowi *fl, u16 family,
1820 			       struct dst_entry *dst_orig)
1821 {
1822 	struct net *net = xp_net(pols[0]);
1823 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1824 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
1825 	struct xfrm_dst *xdst, *old;
1826 	struct dst_entry *dst;
1827 	int err;
1828 
1829 	/* Try to instantiate a bundle */
1830 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1831 	if (err <= 0) {
1832 		if (err != 0 && err != -EAGAIN)
1833 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1834 		return ERR_PTR(err);
1835 	}
1836 
1837 	xdst = this_cpu_read(xfrm_last_dst);
1838 	if (xdst &&
1839 	    xdst->u.dst.dev == dst_orig->dev &&
1840 	    xdst->num_pols == num_pols &&
1841 	    memcmp(xdst->pols, pols,
1842 		   sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1843 	    xfrm_xdst_can_reuse(xdst, xfrm, err)) {
1844 		dst_hold(&xdst->u.dst);
1845 		xfrm_pols_put(pols, num_pols);
1846 		while (err > 0)
1847 			xfrm_state_put(xfrm[--err]);
1848 		return xdst;
1849 	}
1850 
1851 	old = xdst;
1852 
1853 	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
1854 	if (IS_ERR(dst)) {
1855 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1856 		return ERR_CAST(dst);
1857 	}
1858 
1859 	xdst = (struct xfrm_dst *)dst;
1860 	xdst->num_xfrms = err;
1861 	xdst->num_pols = num_pols;
1862 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1863 	xdst->policy_genid = atomic_read(&pols[0]->genid);
1864 
1865 	atomic_set(&xdst->u.dst.__refcnt, 2);
1866 	xfrm_last_dst_update(xdst, old);
1867 
1868 	return xdst;
1869 }
1870 
1871 static void xfrm_policy_queue_process(struct timer_list *t)
1872 {
1873 	struct sk_buff *skb;
1874 	struct sock *sk;
1875 	struct dst_entry *dst;
1876 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1877 	struct net *net = xp_net(pol);
1878 	struct xfrm_policy_queue *pq = &pol->polq;
1879 	struct flowi fl;
1880 	struct sk_buff_head list;
1881 
1882 	spin_lock(&pq->hold_queue.lock);
1883 	skb = skb_peek(&pq->hold_queue);
1884 	if (!skb) {
1885 		spin_unlock(&pq->hold_queue.lock);
1886 		goto out;
1887 	}
1888 	dst = skb_dst(skb);
1889 	sk = skb->sk;
1890 	xfrm_decode_session(skb, &fl, dst->ops->family);
1891 	spin_unlock(&pq->hold_queue.lock);
1892 
1893 	dst_hold(xfrm_dst_path(dst));
1894 	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, 0);
1895 	if (IS_ERR(dst))
1896 		goto purge_queue;
1897 
1898 	if (dst->flags & DST_XFRM_QUEUE) {
1899 		dst_release(dst);
1900 
1901 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1902 			goto purge_queue;
1903 
1904 		pq->timeout = pq->timeout << 1;
1905 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1906 			xfrm_pol_hold(pol);
1907 	goto out;
1908 	}
1909 
1910 	dst_release(dst);
1911 
1912 	__skb_queue_head_init(&list);
1913 
1914 	spin_lock(&pq->hold_queue.lock);
1915 	pq->timeout = 0;
1916 	skb_queue_splice_init(&pq->hold_queue, &list);
1917 	spin_unlock(&pq->hold_queue.lock);
1918 
1919 	while (!skb_queue_empty(&list)) {
1920 		skb = __skb_dequeue(&list);
1921 
1922 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1923 		dst_hold(xfrm_dst_path(skb_dst(skb)));
1924 		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
1925 		if (IS_ERR(dst)) {
1926 			kfree_skb(skb);
1927 			continue;
1928 		}
1929 
1930 		nf_reset(skb);
1931 		skb_dst_drop(skb);
1932 		skb_dst_set(skb, dst);
1933 
1934 		dst_output(net, skb->sk, skb);
1935 	}
1936 
1937 out:
1938 	xfrm_pol_put(pol);
1939 	return;
1940 
1941 purge_queue:
1942 	pq->timeout = 0;
1943 	skb_queue_purge(&pq->hold_queue);
1944 	xfrm_pol_put(pol);
1945 }
1946 
1947 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1948 {
1949 	unsigned long sched_next;
1950 	struct dst_entry *dst = skb_dst(skb);
1951 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1952 	struct xfrm_policy *pol = xdst->pols[0];
1953 	struct xfrm_policy_queue *pq = &pol->polq;
1954 
1955 	if (unlikely(skb_fclone_busy(sk, skb))) {
1956 		kfree_skb(skb);
1957 		return 0;
1958 	}
1959 
1960 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1961 		kfree_skb(skb);
1962 		return -EAGAIN;
1963 	}
1964 
1965 	skb_dst_force(skb);
1966 
1967 	spin_lock_bh(&pq->hold_queue.lock);
1968 
1969 	if (!pq->timeout)
1970 		pq->timeout = XFRM_QUEUE_TMO_MIN;
1971 
1972 	sched_next = jiffies + pq->timeout;
1973 
1974 	if (del_timer(&pq->hold_timer)) {
1975 		if (time_before(pq->hold_timer.expires, sched_next))
1976 			sched_next = pq->hold_timer.expires;
1977 		xfrm_pol_put(pol);
1978 	}
1979 
1980 	__skb_queue_tail(&pq->hold_queue, skb);
1981 	if (!mod_timer(&pq->hold_timer, sched_next))
1982 		xfrm_pol_hold(pol);
1983 
1984 	spin_unlock_bh(&pq->hold_queue.lock);
1985 
1986 	return 0;
1987 }
1988 
1989 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1990 						 struct xfrm_flo *xflo,
1991 						 const struct flowi *fl,
1992 						 int num_xfrms,
1993 						 u16 family)
1994 {
1995 	int err;
1996 	struct net_device *dev;
1997 	struct dst_entry *dst;
1998 	struct dst_entry *dst1;
1999 	struct xfrm_dst *xdst;
2000 
2001 	xdst = xfrm_alloc_dst(net, family);
2002 	if (IS_ERR(xdst))
2003 		return xdst;
2004 
2005 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2006 	    net->xfrm.sysctl_larval_drop ||
2007 	    num_xfrms <= 0)
2008 		return xdst;
2009 
2010 	dst = xflo->dst_orig;
2011 	dst1 = &xdst->u.dst;
2012 	dst_hold(dst);
2013 	xdst->route = dst;
2014 
2015 	dst_copy_metrics(dst1, dst);
2016 
2017 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2018 	dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2019 	dst1->lastuse = jiffies;
2020 
2021 	dst1->input = dst_discard;
2022 	dst1->output = xdst_queue_output;
2023 
2024 	dst_hold(dst);
2025 	xfrm_dst_set_child(xdst, dst);
2026 	xdst->path = dst;
2027 
2028 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2029 
2030 	err = -ENODEV;
2031 	dev = dst->dev;
2032 	if (!dev)
2033 		goto free_dst;
2034 
2035 	err = xfrm_fill_dst(xdst, dev, fl);
2036 	if (err)
2037 		goto free_dst;
2038 
2039 out:
2040 	return xdst;
2041 
2042 free_dst:
2043 	dst_release(dst1);
2044 	xdst = ERR_PTR(err);
2045 	goto out;
2046 }
2047 
2048 static struct xfrm_dst *
2049 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2050 {
2051 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2052 	int num_pols = 0, num_xfrms = 0, err;
2053 	struct xfrm_dst *xdst;
2054 
2055 	/* Resolve policies to use if we couldn't get them from
2056 	 * previous cache entry */
2057 	num_pols = 1;
2058 	pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2059 	err = xfrm_expand_policies(fl, family, pols,
2060 					   &num_pols, &num_xfrms);
2061 	if (err < 0)
2062 		goto inc_error;
2063 	if (num_pols == 0)
2064 		return NULL;
2065 	if (num_xfrms <= 0)
2066 		goto make_dummy_bundle;
2067 
2068 	local_bh_disable();
2069 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2070 					      xflo->dst_orig);
2071 	local_bh_enable();
2072 
2073 	if (IS_ERR(xdst)) {
2074 		err = PTR_ERR(xdst);
2075 		if (err != -EAGAIN)
2076 			goto error;
2077 		goto make_dummy_bundle;
2078 	} else if (xdst == NULL) {
2079 		num_xfrms = 0;
2080 		goto make_dummy_bundle;
2081 	}
2082 
2083 	return xdst;
2084 
2085 make_dummy_bundle:
2086 	/* We found policies, but there's no bundles to instantiate:
2087 	 * either because the policy blocks, has no transformations or
2088 	 * we could not build template (no xfrm_states).*/
2089 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2090 	if (IS_ERR(xdst)) {
2091 		xfrm_pols_put(pols, num_pols);
2092 		return ERR_CAST(xdst);
2093 	}
2094 	xdst->num_pols = num_pols;
2095 	xdst->num_xfrms = num_xfrms;
2096 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2097 
2098 	return xdst;
2099 
2100 inc_error:
2101 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2102 error:
2103 	xfrm_pols_put(pols, num_pols);
2104 	return ERR_PTR(err);
2105 }
2106 
2107 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2108 					struct dst_entry *dst_orig)
2109 {
2110 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2111 	struct dst_entry *ret;
2112 
2113 	if (!afinfo) {
2114 		dst_release(dst_orig);
2115 		return ERR_PTR(-EINVAL);
2116 	} else {
2117 		ret = afinfo->blackhole_route(net, dst_orig);
2118 	}
2119 	rcu_read_unlock();
2120 
2121 	return ret;
2122 }
2123 
2124 /* Main function: finds/creates a bundle for given flow.
2125  *
2126  * At the moment we eat a raw IP route. Mostly to speed up lookups
2127  * on interfaces with disabled IPsec.
2128  */
2129 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2130 			      const struct flowi *fl,
2131 			      const struct sock *sk, int flags)
2132 {
2133 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2134 	struct xfrm_dst *xdst;
2135 	struct dst_entry *dst, *route;
2136 	u16 family = dst_orig->ops->family;
2137 	u8 dir = XFRM_POLICY_OUT;
2138 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2139 
2140 	dst = NULL;
2141 	xdst = NULL;
2142 	route = NULL;
2143 
2144 	sk = sk_const_to_full_sk(sk);
2145 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2146 		num_pols = 1;
2147 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2148 		err = xfrm_expand_policies(fl, family, pols,
2149 					   &num_pols, &num_xfrms);
2150 		if (err < 0)
2151 			goto dropdst;
2152 
2153 		if (num_pols) {
2154 			if (num_xfrms <= 0) {
2155 				drop_pols = num_pols;
2156 				goto no_transform;
2157 			}
2158 
2159 			local_bh_disable();
2160 			xdst = xfrm_resolve_and_create_bundle(
2161 					pols, num_pols, fl,
2162 					family, dst_orig);
2163 			local_bh_enable();
2164 
2165 			if (IS_ERR(xdst)) {
2166 				xfrm_pols_put(pols, num_pols);
2167 				err = PTR_ERR(xdst);
2168 				goto dropdst;
2169 			} else if (xdst == NULL) {
2170 				num_xfrms = 0;
2171 				drop_pols = num_pols;
2172 				goto no_transform;
2173 			}
2174 
2175 			route = xdst->route;
2176 		}
2177 	}
2178 
2179 	if (xdst == NULL) {
2180 		struct xfrm_flo xflo;
2181 
2182 		xflo.dst_orig = dst_orig;
2183 		xflo.flags = flags;
2184 
2185 		/* To accelerate a bit...  */
2186 		if ((dst_orig->flags & DST_NOXFRM) ||
2187 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
2188 			goto nopol;
2189 
2190 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2191 		if (xdst == NULL)
2192 			goto nopol;
2193 		if (IS_ERR(xdst)) {
2194 			err = PTR_ERR(xdst);
2195 			goto dropdst;
2196 		}
2197 
2198 		num_pols = xdst->num_pols;
2199 		num_xfrms = xdst->num_xfrms;
2200 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2201 		route = xdst->route;
2202 	}
2203 
2204 	dst = &xdst->u.dst;
2205 	if (route == NULL && num_xfrms > 0) {
2206 		/* The only case when xfrm_bundle_lookup() returns a
2207 		 * bundle with null route, is when the template could
2208 		 * not be resolved. It means policies are there, but
2209 		 * bundle could not be created, since we don't yet
2210 		 * have the xfrm_state's. We need to wait for KM to
2211 		 * negotiate new SA's or bail out with error.*/
2212 		if (net->xfrm.sysctl_larval_drop) {
2213 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2214 			err = -EREMOTE;
2215 			goto error;
2216 		}
2217 
2218 		err = -EAGAIN;
2219 
2220 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2221 		goto error;
2222 	}
2223 
2224 no_transform:
2225 	if (num_pols == 0)
2226 		goto nopol;
2227 
2228 	if ((flags & XFRM_LOOKUP_ICMP) &&
2229 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2230 		err = -ENOENT;
2231 		goto error;
2232 	}
2233 
2234 	for (i = 0; i < num_pols; i++)
2235 		pols[i]->curlft.use_time = get_seconds();
2236 
2237 	if (num_xfrms < 0) {
2238 		/* Prohibit the flow */
2239 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2240 		err = -EPERM;
2241 		goto error;
2242 	} else if (num_xfrms > 0) {
2243 		/* Flow transformed */
2244 		dst_release(dst_orig);
2245 	} else {
2246 		/* Flow passes untransformed */
2247 		dst_release(dst);
2248 		dst = dst_orig;
2249 	}
2250 ok:
2251 	xfrm_pols_put(pols, drop_pols);
2252 	if (dst && dst->xfrm &&
2253 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2254 		dst->flags |= DST_XFRM_TUNNEL;
2255 	return dst;
2256 
2257 nopol:
2258 	if (!(flags & XFRM_LOOKUP_ICMP)) {
2259 		dst = dst_orig;
2260 		goto ok;
2261 	}
2262 	err = -ENOENT;
2263 error:
2264 	dst_release(dst);
2265 dropdst:
2266 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2267 		dst_release(dst_orig);
2268 	xfrm_pols_put(pols, drop_pols);
2269 	return ERR_PTR(err);
2270 }
2271 EXPORT_SYMBOL(xfrm_lookup);
2272 
2273 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2274  * Otherwise we may send out blackholed packets.
2275  */
2276 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2277 				    const struct flowi *fl,
2278 				    const struct sock *sk, int flags)
2279 {
2280 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2281 					    flags | XFRM_LOOKUP_QUEUE |
2282 					    XFRM_LOOKUP_KEEP_DST_REF);
2283 
2284 	if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2285 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
2286 
2287 	return dst;
2288 }
2289 EXPORT_SYMBOL(xfrm_lookup_route);
2290 
2291 static inline int
2292 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2293 {
2294 	struct xfrm_state *x;
2295 
2296 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2297 		return 0;
2298 	x = skb->sp->xvec[idx];
2299 	if (!x->type->reject)
2300 		return 0;
2301 	return x->type->reject(x, skb, fl);
2302 }
2303 
2304 /* When skb is transformed back to its "native" form, we have to
2305  * check policy restrictions. At the moment we make this in maximally
2306  * stupid way. Shame on me. :-) Of course, connected sockets must
2307  * have policy cached at them.
2308  */
2309 
2310 static inline int
2311 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2312 	      unsigned short family)
2313 {
2314 	if (xfrm_state_kern(x))
2315 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2316 	return	x->id.proto == tmpl->id.proto &&
2317 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2318 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2319 		x->props.mode == tmpl->mode &&
2320 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2321 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2322 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
2323 		  xfrm_state_addr_cmp(tmpl, x, family));
2324 }
2325 
2326 /*
2327  * 0 or more than 0 is returned when validation is succeeded (either bypass
2328  * because of optional transport mode, or next index of the mathced secpath
2329  * state with the template.
2330  * -1 is returned when no matching template is found.
2331  * Otherwise "-2 - errored_index" is returned.
2332  */
2333 static inline int
2334 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2335 	       unsigned short family)
2336 {
2337 	int idx = start;
2338 
2339 	if (tmpl->optional) {
2340 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
2341 			return start;
2342 	} else
2343 		start = -1;
2344 	for (; idx < sp->len; idx++) {
2345 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2346 			return ++idx;
2347 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2348 			if (start == -1)
2349 				start = -2-idx;
2350 			break;
2351 		}
2352 	}
2353 	return start;
2354 }
2355 
2356 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2357 			  unsigned int family, int reverse)
2358 {
2359 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2360 	int err;
2361 
2362 	if (unlikely(afinfo == NULL))
2363 		return -EAFNOSUPPORT;
2364 
2365 	afinfo->decode_session(skb, fl, reverse);
2366 	err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2367 	rcu_read_unlock();
2368 	return err;
2369 }
2370 EXPORT_SYMBOL(__xfrm_decode_session);
2371 
2372 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2373 {
2374 	for (; k < sp->len; k++) {
2375 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2376 			*idxp = k;
2377 			return 1;
2378 		}
2379 	}
2380 
2381 	return 0;
2382 }
2383 
2384 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2385 			unsigned short family)
2386 {
2387 	struct net *net = dev_net(skb->dev);
2388 	struct xfrm_policy *pol;
2389 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2390 	int npols = 0;
2391 	int xfrm_nr;
2392 	int pi;
2393 	int reverse;
2394 	struct flowi fl;
2395 	int xerr_idx = -1;
2396 
2397 	reverse = dir & ~XFRM_POLICY_MASK;
2398 	dir &= XFRM_POLICY_MASK;
2399 
2400 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2401 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2402 		return 0;
2403 	}
2404 
2405 	nf_nat_decode_session(skb, &fl, family);
2406 
2407 	/* First, check used SA against their selectors. */
2408 	if (skb->sp) {
2409 		int i;
2410 
2411 		for (i = skb->sp->len-1; i >= 0; i--) {
2412 			struct xfrm_state *x = skb->sp->xvec[i];
2413 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
2414 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2415 				return 0;
2416 			}
2417 		}
2418 	}
2419 
2420 	pol = NULL;
2421 	sk = sk_to_full_sk(sk);
2422 	if (sk && sk->sk_policy[dir]) {
2423 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2424 		if (IS_ERR(pol)) {
2425 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2426 			return 0;
2427 		}
2428 	}
2429 
2430 	if (!pol)
2431 		pol = xfrm_policy_lookup(net, &fl, family, dir);
2432 
2433 	if (IS_ERR(pol)) {
2434 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2435 		return 0;
2436 	}
2437 
2438 	if (!pol) {
2439 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2440 			xfrm_secpath_reject(xerr_idx, skb, &fl);
2441 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2442 			return 0;
2443 		}
2444 		return 1;
2445 	}
2446 
2447 	pol->curlft.use_time = get_seconds();
2448 
2449 	pols[0] = pol;
2450 	npols++;
2451 #ifdef CONFIG_XFRM_SUB_POLICY
2452 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2453 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2454 						    &fl, family,
2455 						    XFRM_POLICY_IN);
2456 		if (pols[1]) {
2457 			if (IS_ERR(pols[1])) {
2458 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2459 				return 0;
2460 			}
2461 			pols[1]->curlft.use_time = get_seconds();
2462 			npols++;
2463 		}
2464 	}
2465 #endif
2466 
2467 	if (pol->action == XFRM_POLICY_ALLOW) {
2468 		struct sec_path *sp;
2469 		static struct sec_path dummy;
2470 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2471 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2472 		struct xfrm_tmpl **tpp = tp;
2473 		int ti = 0;
2474 		int i, k;
2475 
2476 		if ((sp = skb->sp) == NULL)
2477 			sp = &dummy;
2478 
2479 		for (pi = 0; pi < npols; pi++) {
2480 			if (pols[pi] != pol &&
2481 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
2482 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2483 				goto reject;
2484 			}
2485 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2486 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2487 				goto reject_error;
2488 			}
2489 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
2490 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
2491 		}
2492 		xfrm_nr = ti;
2493 		if (npols > 1) {
2494 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2495 			tpp = stp;
2496 		}
2497 
2498 		/* For each tunnel xfrm, find the first matching tmpl.
2499 		 * For each tmpl before that, find corresponding xfrm.
2500 		 * Order is _important_. Later we will implement
2501 		 * some barriers, but at the moment barriers
2502 		 * are implied between each two transformations.
2503 		 */
2504 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2505 			k = xfrm_policy_ok(tpp[i], sp, k, family);
2506 			if (k < 0) {
2507 				if (k < -1)
2508 					/* "-2 - errored_index" returned */
2509 					xerr_idx = -(2+k);
2510 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2511 				goto reject;
2512 			}
2513 		}
2514 
2515 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2516 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2517 			goto reject;
2518 		}
2519 
2520 		xfrm_pols_put(pols, npols);
2521 		return 1;
2522 	}
2523 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2524 
2525 reject:
2526 	xfrm_secpath_reject(xerr_idx, skb, &fl);
2527 reject_error:
2528 	xfrm_pols_put(pols, npols);
2529 	return 0;
2530 }
2531 EXPORT_SYMBOL(__xfrm_policy_check);
2532 
2533 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2534 {
2535 	struct net *net = dev_net(skb->dev);
2536 	struct flowi fl;
2537 	struct dst_entry *dst;
2538 	int res = 1;
2539 
2540 	if (xfrm_decode_session(skb, &fl, family) < 0) {
2541 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2542 		return 0;
2543 	}
2544 
2545 	skb_dst_force(skb);
2546 
2547 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2548 	if (IS_ERR(dst)) {
2549 		res = 0;
2550 		dst = NULL;
2551 	}
2552 	skb_dst_set(skb, dst);
2553 	return res;
2554 }
2555 EXPORT_SYMBOL(__xfrm_route_forward);
2556 
2557 /* Optimize later using cookies and generation ids. */
2558 
2559 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2560 {
2561 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2562 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2563 	 * get validated by dst_ops->check on every use.  We do this
2564 	 * because when a normal route referenced by an XFRM dst is
2565 	 * obsoleted we do not go looking around for all parent
2566 	 * referencing XFRM dsts so that we can invalidate them.  It
2567 	 * is just too much work.  Instead we make the checks here on
2568 	 * every use.  For example:
2569 	 *
2570 	 *	XFRM dst A --> IPv4 dst X
2571 	 *
2572 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
2573 	 * in this example).  If X is marked obsolete, "A" will not
2574 	 * notice.  That's what we are validating here via the
2575 	 * stale_bundle() check.
2576 	 *
2577 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2578 	 * be marked on it.
2579 	 * This will force stale_bundle() to fail on any xdst bundle with
2580 	 * this dst linked in it.
2581 	 */
2582 	if (dst->obsolete < 0 && !stale_bundle(dst))
2583 		return dst;
2584 
2585 	return NULL;
2586 }
2587 
2588 static int stale_bundle(struct dst_entry *dst)
2589 {
2590 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2591 }
2592 
2593 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2594 {
2595 	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
2596 		dst->dev = dev_net(dev)->loopback_dev;
2597 		dev_hold(dst->dev);
2598 		dev_put(dev);
2599 	}
2600 }
2601 EXPORT_SYMBOL(xfrm_dst_ifdown);
2602 
2603 static void xfrm_link_failure(struct sk_buff *skb)
2604 {
2605 	/* Impossible. Such dst must be popped before reaches point of failure. */
2606 }
2607 
2608 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2609 {
2610 	if (dst) {
2611 		if (dst->obsolete) {
2612 			dst_release(dst);
2613 			dst = NULL;
2614 		}
2615 	}
2616 	return dst;
2617 }
2618 
2619 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
2620 {
2621 	while (nr--) {
2622 		struct xfrm_dst *xdst = bundle[nr];
2623 		u32 pmtu, route_mtu_cached;
2624 		struct dst_entry *dst;
2625 
2626 		dst = &xdst->u.dst;
2627 		pmtu = dst_mtu(xfrm_dst_child(dst));
2628 		xdst->child_mtu_cached = pmtu;
2629 
2630 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2631 
2632 		route_mtu_cached = dst_mtu(xdst->route);
2633 		xdst->route_mtu_cached = route_mtu_cached;
2634 
2635 		if (pmtu > route_mtu_cached)
2636 			pmtu = route_mtu_cached;
2637 
2638 		dst_metric_set(dst, RTAX_MTU, pmtu);
2639 	}
2640 }
2641 
2642 /* Check that the bundle accepts the flow and its components are
2643  * still valid.
2644  */
2645 
2646 static int xfrm_bundle_ok(struct xfrm_dst *first)
2647 {
2648 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2649 	struct dst_entry *dst = &first->u.dst;
2650 	struct xfrm_dst *xdst;
2651 	int start_from, nr;
2652 	u32 mtu;
2653 
2654 	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
2655 	    (dst->dev && !netif_running(dst->dev)))
2656 		return 0;
2657 
2658 	if (dst->flags & DST_XFRM_QUEUE)
2659 		return 1;
2660 
2661 	start_from = nr = 0;
2662 	do {
2663 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2664 
2665 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2666 			return 0;
2667 		if (xdst->xfrm_genid != dst->xfrm->genid)
2668 			return 0;
2669 		if (xdst->num_pols > 0 &&
2670 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2671 			return 0;
2672 
2673 		bundle[nr++] = xdst;
2674 
2675 		mtu = dst_mtu(xfrm_dst_child(dst));
2676 		if (xdst->child_mtu_cached != mtu) {
2677 			start_from = nr;
2678 			xdst->child_mtu_cached = mtu;
2679 		}
2680 
2681 		if (!dst_check(xdst->route, xdst->route_cookie))
2682 			return 0;
2683 		mtu = dst_mtu(xdst->route);
2684 		if (xdst->route_mtu_cached != mtu) {
2685 			start_from = nr;
2686 			xdst->route_mtu_cached = mtu;
2687 		}
2688 
2689 		dst = xfrm_dst_child(dst);
2690 	} while (dst->xfrm);
2691 
2692 	if (likely(!start_from))
2693 		return 1;
2694 
2695 	xdst = bundle[start_from - 1];
2696 	mtu = xdst->child_mtu_cached;
2697 	while (start_from--) {
2698 		dst = &xdst->u.dst;
2699 
2700 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2701 		if (mtu > xdst->route_mtu_cached)
2702 			mtu = xdst->route_mtu_cached;
2703 		dst_metric_set(dst, RTAX_MTU, mtu);
2704 		if (!start_from)
2705 			break;
2706 
2707 		xdst = bundle[start_from - 1];
2708 		xdst->child_mtu_cached = mtu;
2709 	}
2710 
2711 	return 1;
2712 }
2713 
2714 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2715 {
2716 	return dst_metric_advmss(xfrm_dst_path(dst));
2717 }
2718 
2719 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2720 {
2721 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2722 
2723 	return mtu ? : dst_mtu(xfrm_dst_path(dst));
2724 }
2725 
2726 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2727 					const void *daddr)
2728 {
2729 	while (dst->xfrm) {
2730 		const struct xfrm_state *xfrm = dst->xfrm;
2731 
2732 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2733 			continue;
2734 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2735 			daddr = xfrm->coaddr;
2736 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2737 			daddr = &xfrm->id.daddr;
2738 
2739 		dst = xfrm_dst_child(dst);
2740 	}
2741 	return daddr;
2742 }
2743 
2744 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2745 					   struct sk_buff *skb,
2746 					   const void *daddr)
2747 {
2748 	const struct dst_entry *path = xfrm_dst_path(dst);
2749 
2750 	if (!skb)
2751 		daddr = xfrm_get_dst_nexthop(dst, daddr);
2752 	return path->ops->neigh_lookup(path, skb, daddr);
2753 }
2754 
2755 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2756 {
2757 	const struct dst_entry *path = xfrm_dst_path(dst);
2758 
2759 	daddr = xfrm_get_dst_nexthop(dst, daddr);
2760 	path->ops->confirm_neigh(path, daddr);
2761 }
2762 
2763 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2764 {
2765 	int err = 0;
2766 
2767 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2768 		return -EAFNOSUPPORT;
2769 
2770 	spin_lock(&xfrm_policy_afinfo_lock);
2771 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
2772 		err = -EEXIST;
2773 	else {
2774 		struct dst_ops *dst_ops = afinfo->dst_ops;
2775 		if (likely(dst_ops->kmem_cachep == NULL))
2776 			dst_ops->kmem_cachep = xfrm_dst_cache;
2777 		if (likely(dst_ops->check == NULL))
2778 			dst_ops->check = xfrm_dst_check;
2779 		if (likely(dst_ops->default_advmss == NULL))
2780 			dst_ops->default_advmss = xfrm_default_advmss;
2781 		if (likely(dst_ops->mtu == NULL))
2782 			dst_ops->mtu = xfrm_mtu;
2783 		if (likely(dst_ops->negative_advice == NULL))
2784 			dst_ops->negative_advice = xfrm_negative_advice;
2785 		if (likely(dst_ops->link_failure == NULL))
2786 			dst_ops->link_failure = xfrm_link_failure;
2787 		if (likely(dst_ops->neigh_lookup == NULL))
2788 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
2789 		if (likely(!dst_ops->confirm_neigh))
2790 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
2791 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2792 	}
2793 	spin_unlock(&xfrm_policy_afinfo_lock);
2794 
2795 	return err;
2796 }
2797 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2798 
2799 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2800 {
2801 	struct dst_ops *dst_ops = afinfo->dst_ops;
2802 	int i;
2803 
2804 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2805 		if (xfrm_policy_afinfo[i] != afinfo)
2806 			continue;
2807 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2808 		break;
2809 	}
2810 
2811 	synchronize_rcu();
2812 
2813 	dst_ops->kmem_cachep = NULL;
2814 	dst_ops->check = NULL;
2815 	dst_ops->negative_advice = NULL;
2816 	dst_ops->link_failure = NULL;
2817 }
2818 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2819 
2820 #ifdef CONFIG_XFRM_STATISTICS
2821 static int __net_init xfrm_statistics_init(struct net *net)
2822 {
2823 	int rv;
2824 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2825 	if (!net->mib.xfrm_statistics)
2826 		return -ENOMEM;
2827 	rv = xfrm_proc_init(net);
2828 	if (rv < 0)
2829 		free_percpu(net->mib.xfrm_statistics);
2830 	return rv;
2831 }
2832 
2833 static void xfrm_statistics_fini(struct net *net)
2834 {
2835 	xfrm_proc_fini(net);
2836 	free_percpu(net->mib.xfrm_statistics);
2837 }
2838 #else
2839 static int __net_init xfrm_statistics_init(struct net *net)
2840 {
2841 	return 0;
2842 }
2843 
2844 static void xfrm_statistics_fini(struct net *net)
2845 {
2846 }
2847 #endif
2848 
2849 static int __net_init xfrm_policy_init(struct net *net)
2850 {
2851 	unsigned int hmask, sz;
2852 	int dir;
2853 
2854 	if (net_eq(net, &init_net))
2855 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2856 					   sizeof(struct xfrm_dst),
2857 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2858 					   NULL);
2859 
2860 	hmask = 8 - 1;
2861 	sz = (hmask+1) * sizeof(struct hlist_head);
2862 
2863 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2864 	if (!net->xfrm.policy_byidx)
2865 		goto out_byidx;
2866 	net->xfrm.policy_idx_hmask = hmask;
2867 
2868 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2869 		struct xfrm_policy_hash *htab;
2870 
2871 		net->xfrm.policy_count[dir] = 0;
2872 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2873 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2874 
2875 		htab = &net->xfrm.policy_bydst[dir];
2876 		htab->table = xfrm_hash_alloc(sz);
2877 		if (!htab->table)
2878 			goto out_bydst;
2879 		htab->hmask = hmask;
2880 		htab->dbits4 = 32;
2881 		htab->sbits4 = 32;
2882 		htab->dbits6 = 128;
2883 		htab->sbits6 = 128;
2884 	}
2885 	net->xfrm.policy_hthresh.lbits4 = 32;
2886 	net->xfrm.policy_hthresh.rbits4 = 32;
2887 	net->xfrm.policy_hthresh.lbits6 = 128;
2888 	net->xfrm.policy_hthresh.rbits6 = 128;
2889 
2890 	seqlock_init(&net->xfrm.policy_hthresh.lock);
2891 
2892 	INIT_LIST_HEAD(&net->xfrm.policy_all);
2893 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2894 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2895 	if (net_eq(net, &init_net))
2896 		xfrm_dev_init();
2897 	return 0;
2898 
2899 out_bydst:
2900 	for (dir--; dir >= 0; dir--) {
2901 		struct xfrm_policy_hash *htab;
2902 
2903 		htab = &net->xfrm.policy_bydst[dir];
2904 		xfrm_hash_free(htab->table, sz);
2905 	}
2906 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2907 out_byidx:
2908 	return -ENOMEM;
2909 }
2910 
2911 static void xfrm_policy_fini(struct net *net)
2912 {
2913 	unsigned int sz;
2914 	int dir;
2915 
2916 	flush_work(&net->xfrm.policy_hash_work);
2917 #ifdef CONFIG_XFRM_SUB_POLICY
2918 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2919 #endif
2920 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2921 
2922 	WARN_ON(!list_empty(&net->xfrm.policy_all));
2923 
2924 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2925 		struct xfrm_policy_hash *htab;
2926 
2927 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2928 
2929 		htab = &net->xfrm.policy_bydst[dir];
2930 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2931 		WARN_ON(!hlist_empty(htab->table));
2932 		xfrm_hash_free(htab->table, sz);
2933 	}
2934 
2935 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2936 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2937 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2938 }
2939 
2940 static int __net_init xfrm_net_init(struct net *net)
2941 {
2942 	int rv;
2943 
2944 	/* Initialize the per-net locks here */
2945 	spin_lock_init(&net->xfrm.xfrm_state_lock);
2946 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
2947 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
2948 
2949 	rv = xfrm_statistics_init(net);
2950 	if (rv < 0)
2951 		goto out_statistics;
2952 	rv = xfrm_state_init(net);
2953 	if (rv < 0)
2954 		goto out_state;
2955 	rv = xfrm_policy_init(net);
2956 	if (rv < 0)
2957 		goto out_policy;
2958 	rv = xfrm_sysctl_init(net);
2959 	if (rv < 0)
2960 		goto out_sysctl;
2961 
2962 	return 0;
2963 
2964 out_sysctl:
2965 	xfrm_policy_fini(net);
2966 out_policy:
2967 	xfrm_state_fini(net);
2968 out_state:
2969 	xfrm_statistics_fini(net);
2970 out_statistics:
2971 	return rv;
2972 }
2973 
2974 static void __net_exit xfrm_net_exit(struct net *net)
2975 {
2976 	xfrm_sysctl_fini(net);
2977 	xfrm_policy_fini(net);
2978 	xfrm_state_fini(net);
2979 	xfrm_statistics_fini(net);
2980 }
2981 
2982 static struct pernet_operations __net_initdata xfrm_net_ops = {
2983 	.init = xfrm_net_init,
2984 	.exit = xfrm_net_exit,
2985 };
2986 
2987 void __init xfrm_init(void)
2988 {
2989 	int i;
2990 
2991 	xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2992 				       GFP_KERNEL);
2993 	BUG_ON(!xfrm_pcpu_work);
2994 
2995 	for (i = 0; i < NR_CPUS; i++)
2996 		INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
2997 
2998 	register_pernet_subsys(&xfrm_net_ops);
2999 	seqcount_init(&xfrm_policy_hash_generation);
3000 	xfrm_input_init();
3001 }
3002 
3003 #ifdef CONFIG_AUDITSYSCALL
3004 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3005 					 struct audit_buffer *audit_buf)
3006 {
3007 	struct xfrm_sec_ctx *ctx = xp->security;
3008 	struct xfrm_selector *sel = &xp->selector;
3009 
3010 	if (ctx)
3011 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3012 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3013 
3014 	switch (sel->family) {
3015 	case AF_INET:
3016 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3017 		if (sel->prefixlen_s != 32)
3018 			audit_log_format(audit_buf, " src_prefixlen=%d",
3019 					 sel->prefixlen_s);
3020 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3021 		if (sel->prefixlen_d != 32)
3022 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3023 					 sel->prefixlen_d);
3024 		break;
3025 	case AF_INET6:
3026 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3027 		if (sel->prefixlen_s != 128)
3028 			audit_log_format(audit_buf, " src_prefixlen=%d",
3029 					 sel->prefixlen_s);
3030 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3031 		if (sel->prefixlen_d != 128)
3032 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3033 					 sel->prefixlen_d);
3034 		break;
3035 	}
3036 }
3037 
3038 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3039 {
3040 	struct audit_buffer *audit_buf;
3041 
3042 	audit_buf = xfrm_audit_start("SPD-add");
3043 	if (audit_buf == NULL)
3044 		return;
3045 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3046 	audit_log_format(audit_buf, " res=%u", result);
3047 	xfrm_audit_common_policyinfo(xp, audit_buf);
3048 	audit_log_end(audit_buf);
3049 }
3050 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3051 
3052 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3053 			      bool task_valid)
3054 {
3055 	struct audit_buffer *audit_buf;
3056 
3057 	audit_buf = xfrm_audit_start("SPD-delete");
3058 	if (audit_buf == NULL)
3059 		return;
3060 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3061 	audit_log_format(audit_buf, " res=%u", result);
3062 	xfrm_audit_common_policyinfo(xp, audit_buf);
3063 	audit_log_end(audit_buf);
3064 }
3065 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3066 #endif
3067 
3068 #ifdef CONFIG_XFRM_MIGRATE
3069 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3070 					const struct xfrm_selector *sel_tgt)
3071 {
3072 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3073 		if (sel_tgt->family == sel_cmp->family &&
3074 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3075 				    sel_cmp->family) &&
3076 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3077 				    sel_cmp->family) &&
3078 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3079 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3080 			return true;
3081 		}
3082 	} else {
3083 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3084 			return true;
3085 		}
3086 	}
3087 	return false;
3088 }
3089 
3090 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3091 						    u8 dir, u8 type, struct net *net)
3092 {
3093 	struct xfrm_policy *pol, *ret = NULL;
3094 	struct hlist_head *chain;
3095 	u32 priority = ~0U;
3096 
3097 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3098 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3099 	hlist_for_each_entry(pol, chain, bydst) {
3100 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3101 		    pol->type == type) {
3102 			ret = pol;
3103 			priority = ret->priority;
3104 			break;
3105 		}
3106 	}
3107 	chain = &net->xfrm.policy_inexact[dir];
3108 	hlist_for_each_entry(pol, chain, bydst) {
3109 		if ((pol->priority >= priority) && ret)
3110 			break;
3111 
3112 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3113 		    pol->type == type) {
3114 			ret = pol;
3115 			break;
3116 		}
3117 	}
3118 
3119 	xfrm_pol_hold(ret);
3120 
3121 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3122 
3123 	return ret;
3124 }
3125 
3126 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3127 {
3128 	int match = 0;
3129 
3130 	if (t->mode == m->mode && t->id.proto == m->proto &&
3131 	    (m->reqid == 0 || t->reqid == m->reqid)) {
3132 		switch (t->mode) {
3133 		case XFRM_MODE_TUNNEL:
3134 		case XFRM_MODE_BEET:
3135 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3136 					    m->old_family) &&
3137 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
3138 					    m->old_family)) {
3139 				match = 1;
3140 			}
3141 			break;
3142 		case XFRM_MODE_TRANSPORT:
3143 			/* in case of transport mode, template does not store
3144 			   any IP addresses, hence we just compare mode and
3145 			   protocol */
3146 			match = 1;
3147 			break;
3148 		default:
3149 			break;
3150 		}
3151 	}
3152 	return match;
3153 }
3154 
3155 /* update endpoint address(es) of template(s) */
3156 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3157 			       struct xfrm_migrate *m, int num_migrate)
3158 {
3159 	struct xfrm_migrate *mp;
3160 	int i, j, n = 0;
3161 
3162 	write_lock_bh(&pol->lock);
3163 	if (unlikely(pol->walk.dead)) {
3164 		/* target policy has been deleted */
3165 		write_unlock_bh(&pol->lock);
3166 		return -ENOENT;
3167 	}
3168 
3169 	for (i = 0; i < pol->xfrm_nr; i++) {
3170 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3171 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3172 				continue;
3173 			n++;
3174 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3175 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3176 				continue;
3177 			/* update endpoints */
3178 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3179 			       sizeof(pol->xfrm_vec[i].id.daddr));
3180 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3181 			       sizeof(pol->xfrm_vec[i].saddr));
3182 			pol->xfrm_vec[i].encap_family = mp->new_family;
3183 			/* flush bundles */
3184 			atomic_inc(&pol->genid);
3185 		}
3186 	}
3187 
3188 	write_unlock_bh(&pol->lock);
3189 
3190 	if (!n)
3191 		return -ENODATA;
3192 
3193 	return 0;
3194 }
3195 
3196 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3197 {
3198 	int i, j;
3199 
3200 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3201 		return -EINVAL;
3202 
3203 	for (i = 0; i < num_migrate; i++) {
3204 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3205 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3206 			return -EINVAL;
3207 
3208 		/* check if there is any duplicated entry */
3209 		for (j = i + 1; j < num_migrate; j++) {
3210 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3211 				    sizeof(m[i].old_daddr)) &&
3212 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3213 				    sizeof(m[i].old_saddr)) &&
3214 			    m[i].proto == m[j].proto &&
3215 			    m[i].mode == m[j].mode &&
3216 			    m[i].reqid == m[j].reqid &&
3217 			    m[i].old_family == m[j].old_family)
3218 				return -EINVAL;
3219 		}
3220 	}
3221 
3222 	return 0;
3223 }
3224 
3225 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3226 		 struct xfrm_migrate *m, int num_migrate,
3227 		 struct xfrm_kmaddress *k, struct net *net,
3228 		 struct xfrm_encap_tmpl *encap)
3229 {
3230 	int i, err, nx_cur = 0, nx_new = 0;
3231 	struct xfrm_policy *pol = NULL;
3232 	struct xfrm_state *x, *xc;
3233 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3234 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3235 	struct xfrm_migrate *mp;
3236 
3237 	/* Stage 0 - sanity checks */
3238 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3239 		goto out;
3240 
3241 	if (dir >= XFRM_POLICY_MAX) {
3242 		err = -EINVAL;
3243 		goto out;
3244 	}
3245 
3246 	/* Stage 1 - find policy */
3247 	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3248 		err = -ENOENT;
3249 		goto out;
3250 	}
3251 
3252 	/* Stage 2 - find and update state(s) */
3253 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3254 		if ((x = xfrm_migrate_state_find(mp, net))) {
3255 			x_cur[nx_cur] = x;
3256 			nx_cur++;
3257 			xc = xfrm_state_migrate(x, mp, encap);
3258 			if (xc) {
3259 				x_new[nx_new] = xc;
3260 				nx_new++;
3261 			} else {
3262 				err = -ENODATA;
3263 				goto restore_state;
3264 			}
3265 		}
3266 	}
3267 
3268 	/* Stage 3 - update policy */
3269 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3270 		goto restore_state;
3271 
3272 	/* Stage 4 - delete old state(s) */
3273 	if (nx_cur) {
3274 		xfrm_states_put(x_cur, nx_cur);
3275 		xfrm_states_delete(x_cur, nx_cur);
3276 	}
3277 
3278 	/* Stage 5 - announce */
3279 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
3280 
3281 	xfrm_pol_put(pol);
3282 
3283 	return 0;
3284 out:
3285 	return err;
3286 
3287 restore_state:
3288 	if (pol)
3289 		xfrm_pol_put(pol);
3290 	if (nx_cur)
3291 		xfrm_states_put(x_cur, nx_cur);
3292 	if (nx_new)
3293 		xfrm_states_delete(x_new, nx_new);
3294 
3295 	return err;
3296 }
3297 EXPORT_SYMBOL(xfrm_migrate);
3298 #endif
3299