xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 752beb5e)
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 <linux/rhashtable.h>
30 #include <linux/if_tunnel.h>
31 #include <net/dst.h>
32 #include <net/flow.h>
33 #include <net/xfrm.h>
34 #include <net/ip.h>
35 #if IS_ENABLED(CONFIG_IPV6_MIP6)
36 #include <net/mip6.h>
37 #endif
38 #ifdef CONFIG_XFRM_STATISTICS
39 #include <net/snmp.h>
40 #endif
41 
42 #include "xfrm_hash.h"
43 
44 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
45 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
46 #define XFRM_MAX_QUEUE_LEN	100
47 
48 struct xfrm_flo {
49 	struct dst_entry *dst_orig;
50 	u8 flags;
51 };
52 
53 /* prefixes smaller than this are stored in lists, not trees. */
54 #define INEXACT_PREFIXLEN_IPV4	16
55 #define INEXACT_PREFIXLEN_IPV6	48
56 
57 struct xfrm_pol_inexact_node {
58 	struct rb_node node;
59 	union {
60 		xfrm_address_t addr;
61 		struct rcu_head rcu;
62 	};
63 	u8 prefixlen;
64 
65 	struct rb_root root;
66 
67 	/* the policies matching this node, can be empty list */
68 	struct hlist_head hhead;
69 };
70 
71 /* xfrm inexact policy search tree:
72  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
73  *  |
74  * +---- root_d: sorted by daddr:prefix
75  * |                 |
76  * |        xfrm_pol_inexact_node
77  * |                 |
78  * |                 +- root: sorted by saddr/prefix
79  * |                 |              |
80  * |                 |         xfrm_pol_inexact_node
81  * |                 |              |
82  * |                 |              + root: unused
83  * |                 |              |
84  * |                 |              + hhead: saddr:daddr policies
85  * |                 |
86  * |                 +- coarse policies and all any:daddr policies
87  * |
88  * +---- root_s: sorted by saddr:prefix
89  * |                 |
90  * |        xfrm_pol_inexact_node
91  * |                 |
92  * |                 + root: unused
93  * |                 |
94  * |                 + hhead: saddr:any policies
95  * |
96  * +---- coarse policies and all any:any policies
97  *
98  * Lookups return four candidate lists:
99  * 1. any:any list from top-level xfrm_pol_inexact_bin
100  * 2. any:daddr list from daddr tree
101  * 3. saddr:daddr list from 2nd level daddr tree
102  * 4. saddr:any list from saddr tree
103  *
104  * This result set then needs to be searched for the policy with
105  * the lowest priority.  If two results have same prio, youngest one wins.
106  */
107 
108 struct xfrm_pol_inexact_key {
109 	possible_net_t net;
110 	u32 if_id;
111 	u16 family;
112 	u8 dir, type;
113 };
114 
115 struct xfrm_pol_inexact_bin {
116 	struct xfrm_pol_inexact_key k;
117 	struct rhash_head head;
118 	/* list containing '*:*' policies */
119 	struct hlist_head hhead;
120 
121 	seqcount_t count;
122 	/* tree sorted by daddr/prefix */
123 	struct rb_root root_d;
124 
125 	/* tree sorted by saddr/prefix */
126 	struct rb_root root_s;
127 
128 	/* slow path below */
129 	struct list_head inexact_bins;
130 	struct rcu_head rcu;
131 };
132 
133 enum xfrm_pol_inexact_candidate_type {
134 	XFRM_POL_CAND_BOTH,
135 	XFRM_POL_CAND_SADDR,
136 	XFRM_POL_CAND_DADDR,
137 	XFRM_POL_CAND_ANY,
138 
139 	XFRM_POL_CAND_MAX,
140 };
141 
142 struct xfrm_pol_inexact_candidates {
143 	struct hlist_head *res[XFRM_POL_CAND_MAX];
144 };
145 
146 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
147 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
148 
149 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
150 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
151 						__read_mostly;
152 
153 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
154 static __read_mostly seqcount_t xfrm_policy_hash_generation;
155 
156 static struct rhashtable xfrm_policy_inexact_table;
157 static const struct rhashtable_params xfrm_pol_inexact_params;
158 
159 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
160 static int stale_bundle(struct dst_entry *dst);
161 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
162 static void xfrm_policy_queue_process(struct timer_list *t);
163 
164 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
165 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
166 						int dir);
167 
168 static struct xfrm_pol_inexact_bin *
169 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
170 			   u32 if_id);
171 
172 static struct xfrm_pol_inexact_bin *
173 xfrm_policy_inexact_lookup_rcu(struct net *net,
174 			       u8 type, u16 family, u8 dir, u32 if_id);
175 static struct xfrm_policy *
176 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
177 			bool excl);
178 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
179 					    struct xfrm_policy *policy);
180 
181 static bool
182 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
183 				    struct xfrm_pol_inexact_bin *b,
184 				    const xfrm_address_t *saddr,
185 				    const xfrm_address_t *daddr);
186 
187 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
188 {
189 	return refcount_inc_not_zero(&policy->refcnt);
190 }
191 
192 static inline bool
193 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
194 {
195 	const struct flowi4 *fl4 = &fl->u.ip4;
196 
197 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
198 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
199 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
200 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
201 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
202 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
203 }
204 
205 static inline bool
206 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
207 {
208 	const struct flowi6 *fl6 = &fl->u.ip6;
209 
210 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
211 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
212 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
213 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
214 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
215 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
216 }
217 
218 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
219 			 unsigned short family)
220 {
221 	switch (family) {
222 	case AF_INET:
223 		return __xfrm4_selector_match(sel, fl);
224 	case AF_INET6:
225 		return __xfrm6_selector_match(sel, fl);
226 	}
227 	return false;
228 }
229 
230 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
231 {
232 	const struct xfrm_policy_afinfo *afinfo;
233 
234 	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
235 		return NULL;
236 	rcu_read_lock();
237 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
238 	if (unlikely(!afinfo))
239 		rcu_read_unlock();
240 	return afinfo;
241 }
242 
243 /* Called with rcu_read_lock(). */
244 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
245 {
246 	return rcu_dereference(xfrm_if_cb);
247 }
248 
249 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
250 				    const xfrm_address_t *saddr,
251 				    const xfrm_address_t *daddr,
252 				    int family, u32 mark)
253 {
254 	const struct xfrm_policy_afinfo *afinfo;
255 	struct dst_entry *dst;
256 
257 	afinfo = xfrm_policy_get_afinfo(family);
258 	if (unlikely(afinfo == NULL))
259 		return ERR_PTR(-EAFNOSUPPORT);
260 
261 	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
262 
263 	rcu_read_unlock();
264 
265 	return dst;
266 }
267 EXPORT_SYMBOL(__xfrm_dst_lookup);
268 
269 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
270 						int tos, int oif,
271 						xfrm_address_t *prev_saddr,
272 						xfrm_address_t *prev_daddr,
273 						int family, u32 mark)
274 {
275 	struct net *net = xs_net(x);
276 	xfrm_address_t *saddr = &x->props.saddr;
277 	xfrm_address_t *daddr = &x->id.daddr;
278 	struct dst_entry *dst;
279 
280 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
281 		saddr = x->coaddr;
282 		daddr = prev_daddr;
283 	}
284 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
285 		saddr = prev_saddr;
286 		daddr = x->coaddr;
287 	}
288 
289 	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
290 
291 	if (!IS_ERR(dst)) {
292 		if (prev_saddr != saddr)
293 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
294 		if (prev_daddr != daddr)
295 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
296 	}
297 
298 	return dst;
299 }
300 
301 static inline unsigned long make_jiffies(long secs)
302 {
303 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
304 		return MAX_SCHEDULE_TIMEOUT-1;
305 	else
306 		return secs*HZ;
307 }
308 
309 static void xfrm_policy_timer(struct timer_list *t)
310 {
311 	struct xfrm_policy *xp = from_timer(xp, t, timer);
312 	time64_t now = ktime_get_real_seconds();
313 	time64_t next = TIME64_MAX;
314 	int warn = 0;
315 	int dir;
316 
317 	read_lock(&xp->lock);
318 
319 	if (unlikely(xp->walk.dead))
320 		goto out;
321 
322 	dir = xfrm_policy_id2dir(xp->index);
323 
324 	if (xp->lft.hard_add_expires_seconds) {
325 		time64_t tmo = xp->lft.hard_add_expires_seconds +
326 			xp->curlft.add_time - now;
327 		if (tmo <= 0)
328 			goto expired;
329 		if (tmo < next)
330 			next = tmo;
331 	}
332 	if (xp->lft.hard_use_expires_seconds) {
333 		time64_t tmo = xp->lft.hard_use_expires_seconds +
334 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
335 		if (tmo <= 0)
336 			goto expired;
337 		if (tmo < next)
338 			next = tmo;
339 	}
340 	if (xp->lft.soft_add_expires_seconds) {
341 		time64_t tmo = xp->lft.soft_add_expires_seconds +
342 			xp->curlft.add_time - now;
343 		if (tmo <= 0) {
344 			warn = 1;
345 			tmo = XFRM_KM_TIMEOUT;
346 		}
347 		if (tmo < next)
348 			next = tmo;
349 	}
350 	if (xp->lft.soft_use_expires_seconds) {
351 		time64_t tmo = xp->lft.soft_use_expires_seconds +
352 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
353 		if (tmo <= 0) {
354 			warn = 1;
355 			tmo = XFRM_KM_TIMEOUT;
356 		}
357 		if (tmo < next)
358 			next = tmo;
359 	}
360 
361 	if (warn)
362 		km_policy_expired(xp, dir, 0, 0);
363 	if (next != TIME64_MAX &&
364 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
365 		xfrm_pol_hold(xp);
366 
367 out:
368 	read_unlock(&xp->lock);
369 	xfrm_pol_put(xp);
370 	return;
371 
372 expired:
373 	read_unlock(&xp->lock);
374 	if (!xfrm_policy_delete(xp, dir))
375 		km_policy_expired(xp, dir, 1, 0);
376 	xfrm_pol_put(xp);
377 }
378 
379 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
380  * SPD calls.
381  */
382 
383 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
384 {
385 	struct xfrm_policy *policy;
386 
387 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
388 
389 	if (policy) {
390 		write_pnet(&policy->xp_net, net);
391 		INIT_LIST_HEAD(&policy->walk.all);
392 		INIT_HLIST_NODE(&policy->bydst_inexact_list);
393 		INIT_HLIST_NODE(&policy->bydst);
394 		INIT_HLIST_NODE(&policy->byidx);
395 		rwlock_init(&policy->lock);
396 		refcount_set(&policy->refcnt, 1);
397 		skb_queue_head_init(&policy->polq.hold_queue);
398 		timer_setup(&policy->timer, xfrm_policy_timer, 0);
399 		timer_setup(&policy->polq.hold_timer,
400 			    xfrm_policy_queue_process, 0);
401 	}
402 	return policy;
403 }
404 EXPORT_SYMBOL(xfrm_policy_alloc);
405 
406 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
407 {
408 	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
409 
410 	security_xfrm_policy_free(policy->security);
411 	kfree(policy);
412 }
413 
414 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
415 
416 void xfrm_policy_destroy(struct xfrm_policy *policy)
417 {
418 	BUG_ON(!policy->walk.dead);
419 
420 	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
421 		BUG();
422 
423 	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
424 }
425 EXPORT_SYMBOL(xfrm_policy_destroy);
426 
427 /* Rule must be locked. Release descendant resources, announce
428  * entry dead. The rule must be unlinked from lists to the moment.
429  */
430 
431 static void xfrm_policy_kill(struct xfrm_policy *policy)
432 {
433 	policy->walk.dead = 1;
434 
435 	atomic_inc(&policy->genid);
436 
437 	if (del_timer(&policy->polq.hold_timer))
438 		xfrm_pol_put(policy);
439 	skb_queue_purge(&policy->polq.hold_queue);
440 
441 	if (del_timer(&policy->timer))
442 		xfrm_pol_put(policy);
443 
444 	xfrm_pol_put(policy);
445 }
446 
447 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
448 
449 static inline unsigned int idx_hash(struct net *net, u32 index)
450 {
451 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
452 }
453 
454 /* calculate policy hash thresholds */
455 static void __get_hash_thresh(struct net *net,
456 			      unsigned short family, int dir,
457 			      u8 *dbits, u8 *sbits)
458 {
459 	switch (family) {
460 	case AF_INET:
461 		*dbits = net->xfrm.policy_bydst[dir].dbits4;
462 		*sbits = net->xfrm.policy_bydst[dir].sbits4;
463 		break;
464 
465 	case AF_INET6:
466 		*dbits = net->xfrm.policy_bydst[dir].dbits6;
467 		*sbits = net->xfrm.policy_bydst[dir].sbits6;
468 		break;
469 
470 	default:
471 		*dbits = 0;
472 		*sbits = 0;
473 	}
474 }
475 
476 static struct hlist_head *policy_hash_bysel(struct net *net,
477 					    const struct xfrm_selector *sel,
478 					    unsigned short family, int dir)
479 {
480 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
481 	unsigned int hash;
482 	u8 dbits;
483 	u8 sbits;
484 
485 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
486 	hash = __sel_hash(sel, family, hmask, dbits, sbits);
487 
488 	if (hash == hmask + 1)
489 		return NULL;
490 
491 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
492 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
493 }
494 
495 static struct hlist_head *policy_hash_direct(struct net *net,
496 					     const xfrm_address_t *daddr,
497 					     const xfrm_address_t *saddr,
498 					     unsigned short family, int dir)
499 {
500 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
501 	unsigned int hash;
502 	u8 dbits;
503 	u8 sbits;
504 
505 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
506 	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
507 
508 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
509 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
510 }
511 
512 static void xfrm_dst_hash_transfer(struct net *net,
513 				   struct hlist_head *list,
514 				   struct hlist_head *ndsttable,
515 				   unsigned int nhashmask,
516 				   int dir)
517 {
518 	struct hlist_node *tmp, *entry0 = NULL;
519 	struct xfrm_policy *pol;
520 	unsigned int h0 = 0;
521 	u8 dbits;
522 	u8 sbits;
523 
524 redo:
525 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
526 		unsigned int h;
527 
528 		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
529 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
530 				pol->family, nhashmask, dbits, sbits);
531 		if (!entry0) {
532 			hlist_del_rcu(&pol->bydst);
533 			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
534 			h0 = h;
535 		} else {
536 			if (h != h0)
537 				continue;
538 			hlist_del_rcu(&pol->bydst);
539 			hlist_add_behind_rcu(&pol->bydst, entry0);
540 		}
541 		entry0 = &pol->bydst;
542 	}
543 	if (!hlist_empty(list)) {
544 		entry0 = NULL;
545 		goto redo;
546 	}
547 }
548 
549 static void xfrm_idx_hash_transfer(struct hlist_head *list,
550 				   struct hlist_head *nidxtable,
551 				   unsigned int nhashmask)
552 {
553 	struct hlist_node *tmp;
554 	struct xfrm_policy *pol;
555 
556 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
557 		unsigned int h;
558 
559 		h = __idx_hash(pol->index, nhashmask);
560 		hlist_add_head(&pol->byidx, nidxtable+h);
561 	}
562 }
563 
564 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
565 {
566 	return ((old_hmask + 1) << 1) - 1;
567 }
568 
569 static void xfrm_bydst_resize(struct net *net, int dir)
570 {
571 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
572 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
573 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
574 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
575 	struct hlist_head *odst;
576 	int i;
577 
578 	if (!ndst)
579 		return;
580 
581 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
582 	write_seqcount_begin(&xfrm_policy_hash_generation);
583 
584 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
585 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
586 
587 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
588 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
589 
590 	for (i = hmask; i >= 0; i--)
591 		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
592 
593 	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
594 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
595 
596 	write_seqcount_end(&xfrm_policy_hash_generation);
597 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
598 
599 	synchronize_rcu();
600 
601 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
602 }
603 
604 static void xfrm_byidx_resize(struct net *net, int total)
605 {
606 	unsigned int hmask = net->xfrm.policy_idx_hmask;
607 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
608 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
609 	struct hlist_head *oidx = net->xfrm.policy_byidx;
610 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
611 	int i;
612 
613 	if (!nidx)
614 		return;
615 
616 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
617 
618 	for (i = hmask; i >= 0; i--)
619 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
620 
621 	net->xfrm.policy_byidx = nidx;
622 	net->xfrm.policy_idx_hmask = nhashmask;
623 
624 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
625 
626 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
627 }
628 
629 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
630 {
631 	unsigned int cnt = net->xfrm.policy_count[dir];
632 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
633 
634 	if (total)
635 		*total += cnt;
636 
637 	if ((hmask + 1) < xfrm_policy_hashmax &&
638 	    cnt > hmask)
639 		return 1;
640 
641 	return 0;
642 }
643 
644 static inline int xfrm_byidx_should_resize(struct net *net, int total)
645 {
646 	unsigned int hmask = net->xfrm.policy_idx_hmask;
647 
648 	if ((hmask + 1) < xfrm_policy_hashmax &&
649 	    total > hmask)
650 		return 1;
651 
652 	return 0;
653 }
654 
655 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
656 {
657 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
658 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
659 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
660 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
661 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
662 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
663 	si->spdhcnt = net->xfrm.policy_idx_hmask;
664 	si->spdhmcnt = xfrm_policy_hashmax;
665 }
666 EXPORT_SYMBOL(xfrm_spd_getinfo);
667 
668 static DEFINE_MUTEX(hash_resize_mutex);
669 static void xfrm_hash_resize(struct work_struct *work)
670 {
671 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
672 	int dir, total;
673 
674 	mutex_lock(&hash_resize_mutex);
675 
676 	total = 0;
677 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
678 		if (xfrm_bydst_should_resize(net, dir, &total))
679 			xfrm_bydst_resize(net, dir);
680 	}
681 	if (xfrm_byidx_should_resize(net, total))
682 		xfrm_byidx_resize(net, total);
683 
684 	mutex_unlock(&hash_resize_mutex);
685 }
686 
687 /* Make sure *pol can be inserted into fastbin.
688  * Useful to check that later insert requests will be sucessful
689  * (provided xfrm_policy_lock is held throughout).
690  */
691 static struct xfrm_pol_inexact_bin *
692 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
693 {
694 	struct xfrm_pol_inexact_bin *bin, *prev;
695 	struct xfrm_pol_inexact_key k = {
696 		.family = pol->family,
697 		.type = pol->type,
698 		.dir = dir,
699 		.if_id = pol->if_id,
700 	};
701 	struct net *net = xp_net(pol);
702 
703 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
704 
705 	write_pnet(&k.net, net);
706 	bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
707 				     xfrm_pol_inexact_params);
708 	if (bin)
709 		return bin;
710 
711 	bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
712 	if (!bin)
713 		return NULL;
714 
715 	bin->k = k;
716 	INIT_HLIST_HEAD(&bin->hhead);
717 	bin->root_d = RB_ROOT;
718 	bin->root_s = RB_ROOT;
719 	seqcount_init(&bin->count);
720 
721 	prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
722 						&bin->k, &bin->head,
723 						xfrm_pol_inexact_params);
724 	if (!prev) {
725 		list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
726 		return bin;
727 	}
728 
729 	kfree(bin);
730 
731 	return IS_ERR(prev) ? NULL : prev;
732 }
733 
734 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
735 					       int family, u8 prefixlen)
736 {
737 	if (xfrm_addr_any(addr, family))
738 		return true;
739 
740 	if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
741 		return true;
742 
743 	if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
744 		return true;
745 
746 	return false;
747 }
748 
749 static bool
750 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
751 {
752 	const xfrm_address_t *addr;
753 	bool saddr_any, daddr_any;
754 	u8 prefixlen;
755 
756 	addr = &policy->selector.saddr;
757 	prefixlen = policy->selector.prefixlen_s;
758 
759 	saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
760 						       policy->family,
761 						       prefixlen);
762 	addr = &policy->selector.daddr;
763 	prefixlen = policy->selector.prefixlen_d;
764 	daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
765 						       policy->family,
766 						       prefixlen);
767 	return saddr_any && daddr_any;
768 }
769 
770 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
771 				       const xfrm_address_t *addr, u8 prefixlen)
772 {
773 	node->addr = *addr;
774 	node->prefixlen = prefixlen;
775 }
776 
777 static struct xfrm_pol_inexact_node *
778 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
779 {
780 	struct xfrm_pol_inexact_node *node;
781 
782 	node = kzalloc(sizeof(*node), GFP_ATOMIC);
783 	if (node)
784 		xfrm_pol_inexact_node_init(node, addr, prefixlen);
785 
786 	return node;
787 }
788 
789 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
790 				  const xfrm_address_t *b,
791 				  u8 prefixlen, u16 family)
792 {
793 	unsigned int pdw, pbi;
794 	int delta = 0;
795 
796 	switch (family) {
797 	case AF_INET:
798 		if (sizeof(long) == 4 && prefixlen == 0)
799 			return ntohl(a->a4) - ntohl(b->a4);
800 		return (ntohl(a->a4) & ((~0UL << (32 - prefixlen)))) -
801 		       (ntohl(b->a4) & ((~0UL << (32 - prefixlen))));
802 	case AF_INET6:
803 		pdw = prefixlen >> 5;
804 		pbi = prefixlen & 0x1f;
805 
806 		if (pdw) {
807 			delta = memcmp(a->a6, b->a6, pdw << 2);
808 			if (delta)
809 				return delta;
810 		}
811 		if (pbi) {
812 			u32 mask = ~0u << (32 - pbi);
813 
814 			delta = (ntohl(a->a6[pdw]) & mask) -
815 				(ntohl(b->a6[pdw]) & mask);
816 		}
817 		break;
818 	default:
819 		break;
820 	}
821 
822 	return delta;
823 }
824 
825 static void xfrm_policy_inexact_list_reinsert(struct net *net,
826 					      struct xfrm_pol_inexact_node *n,
827 					      u16 family)
828 {
829 	unsigned int matched_s, matched_d;
830 	struct xfrm_policy *policy, *p;
831 
832 	matched_s = 0;
833 	matched_d = 0;
834 
835 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
836 		struct hlist_node *newpos = NULL;
837 		bool matches_s, matches_d;
838 
839 		if (!policy->bydst_reinsert)
840 			continue;
841 
842 		WARN_ON_ONCE(policy->family != family);
843 
844 		policy->bydst_reinsert = false;
845 		hlist_for_each_entry(p, &n->hhead, bydst) {
846 			if (policy->priority > p->priority)
847 				newpos = &p->bydst;
848 			else if (policy->priority == p->priority &&
849 				 policy->pos > p->pos)
850 				newpos = &p->bydst;
851 			else
852 				break;
853 		}
854 
855 		if (newpos)
856 			hlist_add_behind_rcu(&policy->bydst, newpos);
857 		else
858 			hlist_add_head_rcu(&policy->bydst, &n->hhead);
859 
860 		/* paranoia checks follow.
861 		 * Check that the reinserted policy matches at least
862 		 * saddr or daddr for current node prefix.
863 		 *
864 		 * Matching both is fine, matching saddr in one policy
865 		 * (but not daddr) and then matching only daddr in another
866 		 * is a bug.
867 		 */
868 		matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
869 						   &n->addr,
870 						   n->prefixlen,
871 						   family) == 0;
872 		matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
873 						   &n->addr,
874 						   n->prefixlen,
875 						   family) == 0;
876 		if (matches_s && matches_d)
877 			continue;
878 
879 		WARN_ON_ONCE(!matches_s && !matches_d);
880 		if (matches_s)
881 			matched_s++;
882 		if (matches_d)
883 			matched_d++;
884 		WARN_ON_ONCE(matched_s && matched_d);
885 	}
886 }
887 
888 static void xfrm_policy_inexact_node_reinsert(struct net *net,
889 					      struct xfrm_pol_inexact_node *n,
890 					      struct rb_root *new,
891 					      u16 family)
892 {
893 	struct xfrm_pol_inexact_node *node;
894 	struct rb_node **p, *parent;
895 
896 	/* we should not have another subtree here */
897 	WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
898 restart:
899 	parent = NULL;
900 	p = &new->rb_node;
901 	while (*p) {
902 		u8 prefixlen;
903 		int delta;
904 
905 		parent = *p;
906 		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
907 
908 		prefixlen = min(node->prefixlen, n->prefixlen);
909 
910 		delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
911 					       prefixlen, family);
912 		if (delta < 0) {
913 			p = &parent->rb_left;
914 		} else if (delta > 0) {
915 			p = &parent->rb_right;
916 		} else {
917 			struct xfrm_policy *tmp;
918 
919 			hlist_for_each_entry(tmp, &n->hhead, bydst) {
920 				tmp->bydst_reinsert = true;
921 				hlist_del_rcu(&tmp->bydst);
922 			}
923 
924 			xfrm_policy_inexact_list_reinsert(net, node, family);
925 
926 			if (node->prefixlen == n->prefixlen) {
927 				kfree_rcu(n, rcu);
928 				return;
929 			}
930 
931 			rb_erase(*p, new);
932 			kfree_rcu(n, rcu);
933 			n = node;
934 			n->prefixlen = prefixlen;
935 			goto restart;
936 		}
937 	}
938 
939 	rb_link_node_rcu(&n->node, parent, p);
940 	rb_insert_color(&n->node, new);
941 }
942 
943 /* merge nodes v and n */
944 static void xfrm_policy_inexact_node_merge(struct net *net,
945 					   struct xfrm_pol_inexact_node *v,
946 					   struct xfrm_pol_inexact_node *n,
947 					   u16 family)
948 {
949 	struct xfrm_pol_inexact_node *node;
950 	struct xfrm_policy *tmp;
951 	struct rb_node *rnode;
952 
953 	/* To-be-merged node v has a subtree.
954 	 *
955 	 * Dismantle it and insert its nodes to n->root.
956 	 */
957 	while ((rnode = rb_first(&v->root)) != NULL) {
958 		node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
959 		rb_erase(&node->node, &v->root);
960 		xfrm_policy_inexact_node_reinsert(net, node, &n->root,
961 						  family);
962 	}
963 
964 	hlist_for_each_entry(tmp, &v->hhead, bydst) {
965 		tmp->bydst_reinsert = true;
966 		hlist_del_rcu(&tmp->bydst);
967 	}
968 
969 	xfrm_policy_inexact_list_reinsert(net, n, family);
970 }
971 
972 static struct xfrm_pol_inexact_node *
973 xfrm_policy_inexact_insert_node(struct net *net,
974 				struct rb_root *root,
975 				xfrm_address_t *addr,
976 				u16 family, u8 prefixlen, u8 dir)
977 {
978 	struct xfrm_pol_inexact_node *cached = NULL;
979 	struct rb_node **p, *parent = NULL;
980 	struct xfrm_pol_inexact_node *node;
981 
982 	p = &root->rb_node;
983 	while (*p) {
984 		int delta;
985 
986 		parent = *p;
987 		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
988 
989 		delta = xfrm_policy_addr_delta(addr, &node->addr,
990 					       node->prefixlen,
991 					       family);
992 		if (delta == 0 && prefixlen >= node->prefixlen) {
993 			WARN_ON_ONCE(cached); /* ipsec policies got lost */
994 			return node;
995 		}
996 
997 		if (delta < 0)
998 			p = &parent->rb_left;
999 		else
1000 			p = &parent->rb_right;
1001 
1002 		if (prefixlen < node->prefixlen) {
1003 			delta = xfrm_policy_addr_delta(addr, &node->addr,
1004 						       prefixlen,
1005 						       family);
1006 			if (delta)
1007 				continue;
1008 
1009 			/* This node is a subnet of the new prefix. It needs
1010 			 * to be removed and re-inserted with the smaller
1011 			 * prefix and all nodes that are now also covered
1012 			 * by the reduced prefixlen.
1013 			 */
1014 			rb_erase(&node->node, root);
1015 
1016 			if (!cached) {
1017 				xfrm_pol_inexact_node_init(node, addr,
1018 							   prefixlen);
1019 				cached = node;
1020 			} else {
1021 				/* This node also falls within the new
1022 				 * prefixlen. Merge the to-be-reinserted
1023 				 * node and this one.
1024 				 */
1025 				xfrm_policy_inexact_node_merge(net, node,
1026 							       cached, family);
1027 				kfree_rcu(node, rcu);
1028 			}
1029 
1030 			/* restart */
1031 			p = &root->rb_node;
1032 			parent = NULL;
1033 		}
1034 	}
1035 
1036 	node = cached;
1037 	if (!node) {
1038 		node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1039 		if (!node)
1040 			return NULL;
1041 	}
1042 
1043 	rb_link_node_rcu(&node->node, parent, p);
1044 	rb_insert_color(&node->node, root);
1045 
1046 	return node;
1047 }
1048 
1049 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1050 {
1051 	struct xfrm_pol_inexact_node *node;
1052 	struct rb_node *rn = rb_first(r);
1053 
1054 	while (rn) {
1055 		node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1056 
1057 		xfrm_policy_inexact_gc_tree(&node->root, rm);
1058 		rn = rb_next(rn);
1059 
1060 		if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1061 			WARN_ON_ONCE(rm);
1062 			continue;
1063 		}
1064 
1065 		rb_erase(&node->node, r);
1066 		kfree_rcu(node, rcu);
1067 	}
1068 }
1069 
1070 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1071 {
1072 	write_seqcount_begin(&b->count);
1073 	xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1074 	xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1075 	write_seqcount_end(&b->count);
1076 
1077 	if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1078 	    !hlist_empty(&b->hhead)) {
1079 		WARN_ON_ONCE(net_exit);
1080 		return;
1081 	}
1082 
1083 	if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1084 				   xfrm_pol_inexact_params) == 0) {
1085 		list_del(&b->inexact_bins);
1086 		kfree_rcu(b, rcu);
1087 	}
1088 }
1089 
1090 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1091 {
1092 	struct net *net = read_pnet(&b->k.net);
1093 
1094 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1095 	__xfrm_policy_inexact_prune_bin(b, false);
1096 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1097 }
1098 
1099 static void __xfrm_policy_inexact_flush(struct net *net)
1100 {
1101 	struct xfrm_pol_inexact_bin *bin, *t;
1102 
1103 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1104 
1105 	list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1106 		__xfrm_policy_inexact_prune_bin(bin, false);
1107 }
1108 
1109 static struct hlist_head *
1110 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1111 				struct xfrm_policy *policy, u8 dir)
1112 {
1113 	struct xfrm_pol_inexact_node *n;
1114 	struct net *net;
1115 
1116 	net = xp_net(policy);
1117 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1118 
1119 	if (xfrm_policy_inexact_insert_use_any_list(policy))
1120 		return &bin->hhead;
1121 
1122 	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1123 					       policy->family,
1124 					       policy->selector.prefixlen_d)) {
1125 		write_seqcount_begin(&bin->count);
1126 		n = xfrm_policy_inexact_insert_node(net,
1127 						    &bin->root_s,
1128 						    &policy->selector.saddr,
1129 						    policy->family,
1130 						    policy->selector.prefixlen_s,
1131 						    dir);
1132 		write_seqcount_end(&bin->count);
1133 		if (!n)
1134 			return NULL;
1135 
1136 		return &n->hhead;
1137 	}
1138 
1139 	/* daddr is fixed */
1140 	write_seqcount_begin(&bin->count);
1141 	n = xfrm_policy_inexact_insert_node(net,
1142 					    &bin->root_d,
1143 					    &policy->selector.daddr,
1144 					    policy->family,
1145 					    policy->selector.prefixlen_d, dir);
1146 	write_seqcount_end(&bin->count);
1147 	if (!n)
1148 		return NULL;
1149 
1150 	/* saddr is wildcard */
1151 	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1152 					       policy->family,
1153 					       policy->selector.prefixlen_s))
1154 		return &n->hhead;
1155 
1156 	write_seqcount_begin(&bin->count);
1157 	n = xfrm_policy_inexact_insert_node(net,
1158 					    &n->root,
1159 					    &policy->selector.saddr,
1160 					    policy->family,
1161 					    policy->selector.prefixlen_s, dir);
1162 	write_seqcount_end(&bin->count);
1163 	if (!n)
1164 		return NULL;
1165 
1166 	return &n->hhead;
1167 }
1168 
1169 static struct xfrm_policy *
1170 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1171 {
1172 	struct xfrm_pol_inexact_bin *bin;
1173 	struct xfrm_policy *delpol;
1174 	struct hlist_head *chain;
1175 	struct net *net;
1176 
1177 	bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1178 	if (!bin)
1179 		return ERR_PTR(-ENOMEM);
1180 
1181 	net = xp_net(policy);
1182 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1183 
1184 	chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1185 	if (!chain) {
1186 		__xfrm_policy_inexact_prune_bin(bin, false);
1187 		return ERR_PTR(-ENOMEM);
1188 	}
1189 
1190 	delpol = xfrm_policy_insert_list(chain, policy, excl);
1191 	if (delpol && excl) {
1192 		__xfrm_policy_inexact_prune_bin(bin, false);
1193 		return ERR_PTR(-EEXIST);
1194 	}
1195 
1196 	chain = &net->xfrm.policy_inexact[dir];
1197 	xfrm_policy_insert_inexact_list(chain, policy);
1198 
1199 	if (delpol)
1200 		__xfrm_policy_inexact_prune_bin(bin, false);
1201 
1202 	return delpol;
1203 }
1204 
1205 static void xfrm_hash_rebuild(struct work_struct *work)
1206 {
1207 	struct net *net = container_of(work, struct net,
1208 				       xfrm.policy_hthresh.work);
1209 	unsigned int hmask;
1210 	struct xfrm_policy *pol;
1211 	struct xfrm_policy *policy;
1212 	struct hlist_head *chain;
1213 	struct hlist_head *odst;
1214 	struct hlist_node *newpos;
1215 	int i;
1216 	int dir;
1217 	unsigned seq;
1218 	u8 lbits4, rbits4, lbits6, rbits6;
1219 
1220 	mutex_lock(&hash_resize_mutex);
1221 
1222 	/* read selector prefixlen thresholds */
1223 	do {
1224 		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1225 
1226 		lbits4 = net->xfrm.policy_hthresh.lbits4;
1227 		rbits4 = net->xfrm.policy_hthresh.rbits4;
1228 		lbits6 = net->xfrm.policy_hthresh.lbits6;
1229 		rbits6 = net->xfrm.policy_hthresh.rbits6;
1230 	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1231 
1232 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1233 	write_seqcount_begin(&xfrm_policy_hash_generation);
1234 
1235 	/* make sure that we can insert the indirect policies again before
1236 	 * we start with destructive action.
1237 	 */
1238 	list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1239 		struct xfrm_pol_inexact_bin *bin;
1240 		u8 dbits, sbits;
1241 
1242 		dir = xfrm_policy_id2dir(policy->index);
1243 		if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1244 			continue;
1245 
1246 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1247 			if (policy->family == AF_INET) {
1248 				dbits = rbits4;
1249 				sbits = lbits4;
1250 			} else {
1251 				dbits = rbits6;
1252 				sbits = lbits6;
1253 			}
1254 		} else {
1255 			if (policy->family == AF_INET) {
1256 				dbits = lbits4;
1257 				sbits = rbits4;
1258 			} else {
1259 				dbits = lbits6;
1260 				sbits = rbits6;
1261 			}
1262 		}
1263 
1264 		if (policy->selector.prefixlen_d < dbits ||
1265 		    policy->selector.prefixlen_s < sbits)
1266 			continue;
1267 
1268 		bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1269 		if (!bin)
1270 			goto out_unlock;
1271 
1272 		if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1273 			goto out_unlock;
1274 	}
1275 
1276 	/* reset the bydst and inexact table in all directions */
1277 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1278 		struct hlist_node *n;
1279 
1280 		hlist_for_each_entry_safe(policy, n,
1281 					  &net->xfrm.policy_inexact[dir],
1282 					  bydst_inexact_list)
1283 			hlist_del_init(&policy->bydst_inexact_list);
1284 
1285 		hmask = net->xfrm.policy_bydst[dir].hmask;
1286 		odst = net->xfrm.policy_bydst[dir].table;
1287 		for (i = hmask; i >= 0; i--)
1288 			INIT_HLIST_HEAD(odst + i);
1289 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1290 			/* dir out => dst = remote, src = local */
1291 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1292 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1293 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1294 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1295 		} else {
1296 			/* dir in/fwd => dst = local, src = remote */
1297 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1298 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1299 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1300 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1301 		}
1302 	}
1303 
1304 	/* re-insert all policies by order of creation */
1305 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1306 		if (policy->walk.dead)
1307 			continue;
1308 		dir = xfrm_policy_id2dir(policy->index);
1309 		if (dir >= XFRM_POLICY_MAX) {
1310 			/* skip socket policies */
1311 			continue;
1312 		}
1313 		newpos = NULL;
1314 		chain = policy_hash_bysel(net, &policy->selector,
1315 					  policy->family, dir);
1316 
1317 		hlist_del_rcu(&policy->bydst);
1318 
1319 		if (!chain) {
1320 			void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1321 
1322 			WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1323 			continue;
1324 		}
1325 
1326 		hlist_for_each_entry(pol, chain, bydst) {
1327 			if (policy->priority >= pol->priority)
1328 				newpos = &pol->bydst;
1329 			else
1330 				break;
1331 		}
1332 		if (newpos)
1333 			hlist_add_behind_rcu(&policy->bydst, newpos);
1334 		else
1335 			hlist_add_head_rcu(&policy->bydst, chain);
1336 	}
1337 
1338 out_unlock:
1339 	__xfrm_policy_inexact_flush(net);
1340 	write_seqcount_end(&xfrm_policy_hash_generation);
1341 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1342 
1343 	mutex_unlock(&hash_resize_mutex);
1344 }
1345 
1346 void xfrm_policy_hash_rebuild(struct net *net)
1347 {
1348 	schedule_work(&net->xfrm.policy_hthresh.work);
1349 }
1350 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1351 
1352 /* Generate new index... KAME seems to generate them ordered by cost
1353  * of an absolute inpredictability of ordering of rules. This will not pass. */
1354 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1355 {
1356 	static u32 idx_generator;
1357 
1358 	for (;;) {
1359 		struct hlist_head *list;
1360 		struct xfrm_policy *p;
1361 		u32 idx;
1362 		int found;
1363 
1364 		if (!index) {
1365 			idx = (idx_generator | dir);
1366 			idx_generator += 8;
1367 		} else {
1368 			idx = index;
1369 			index = 0;
1370 		}
1371 
1372 		if (idx == 0)
1373 			idx = 8;
1374 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
1375 		found = 0;
1376 		hlist_for_each_entry(p, list, byidx) {
1377 			if (p->index == idx) {
1378 				found = 1;
1379 				break;
1380 			}
1381 		}
1382 		if (!found)
1383 			return idx;
1384 	}
1385 }
1386 
1387 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1388 {
1389 	u32 *p1 = (u32 *) s1;
1390 	u32 *p2 = (u32 *) s2;
1391 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
1392 	int i;
1393 
1394 	for (i = 0; i < len; i++) {
1395 		if (p1[i] != p2[i])
1396 			return 1;
1397 	}
1398 
1399 	return 0;
1400 }
1401 
1402 static void xfrm_policy_requeue(struct xfrm_policy *old,
1403 				struct xfrm_policy *new)
1404 {
1405 	struct xfrm_policy_queue *pq = &old->polq;
1406 	struct sk_buff_head list;
1407 
1408 	if (skb_queue_empty(&pq->hold_queue))
1409 		return;
1410 
1411 	__skb_queue_head_init(&list);
1412 
1413 	spin_lock_bh(&pq->hold_queue.lock);
1414 	skb_queue_splice_init(&pq->hold_queue, &list);
1415 	if (del_timer(&pq->hold_timer))
1416 		xfrm_pol_put(old);
1417 	spin_unlock_bh(&pq->hold_queue.lock);
1418 
1419 	pq = &new->polq;
1420 
1421 	spin_lock_bh(&pq->hold_queue.lock);
1422 	skb_queue_splice(&list, &pq->hold_queue);
1423 	pq->timeout = XFRM_QUEUE_TMO_MIN;
1424 	if (!mod_timer(&pq->hold_timer, jiffies))
1425 		xfrm_pol_hold(new);
1426 	spin_unlock_bh(&pq->hold_queue.lock);
1427 }
1428 
1429 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
1430 				   struct xfrm_policy *pol)
1431 {
1432 	u32 mark = policy->mark.v & policy->mark.m;
1433 
1434 	if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
1435 		return true;
1436 
1437 	if ((mark & pol->mark.m) == pol->mark.v &&
1438 	    policy->priority == pol->priority)
1439 		return true;
1440 
1441 	return false;
1442 }
1443 
1444 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1445 {
1446 	const struct xfrm_pol_inexact_key *k = data;
1447 	u32 a = k->type << 24 | k->dir << 16 | k->family;
1448 
1449 	return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1450 			    seed);
1451 }
1452 
1453 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1454 {
1455 	const struct xfrm_pol_inexact_bin *b = data;
1456 
1457 	return xfrm_pol_bin_key(&b->k, 0, seed);
1458 }
1459 
1460 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1461 			    const void *ptr)
1462 {
1463 	const struct xfrm_pol_inexact_key *key = arg->key;
1464 	const struct xfrm_pol_inexact_bin *b = ptr;
1465 	int ret;
1466 
1467 	if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1468 		return -1;
1469 
1470 	ret = b->k.dir ^ key->dir;
1471 	if (ret)
1472 		return ret;
1473 
1474 	ret = b->k.type ^ key->type;
1475 	if (ret)
1476 		return ret;
1477 
1478 	ret = b->k.family ^ key->family;
1479 	if (ret)
1480 		return ret;
1481 
1482 	return b->k.if_id ^ key->if_id;
1483 }
1484 
1485 static const struct rhashtable_params xfrm_pol_inexact_params = {
1486 	.head_offset		= offsetof(struct xfrm_pol_inexact_bin, head),
1487 	.hashfn			= xfrm_pol_bin_key,
1488 	.obj_hashfn		= xfrm_pol_bin_obj,
1489 	.obj_cmpfn		= xfrm_pol_bin_cmp,
1490 	.automatic_shrinking	= true,
1491 };
1492 
1493 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1494 					    struct xfrm_policy *policy)
1495 {
1496 	struct xfrm_policy *pol, *delpol = NULL;
1497 	struct hlist_node *newpos = NULL;
1498 	int i = 0;
1499 
1500 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1501 		if (pol->type == policy->type &&
1502 		    pol->if_id == policy->if_id &&
1503 		    !selector_cmp(&pol->selector, &policy->selector) &&
1504 		    xfrm_policy_mark_match(policy, pol) &&
1505 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1506 		    !WARN_ON(delpol)) {
1507 			delpol = pol;
1508 			if (policy->priority > pol->priority)
1509 				continue;
1510 		} else if (policy->priority >= pol->priority) {
1511 			newpos = &pol->bydst_inexact_list;
1512 			continue;
1513 		}
1514 		if (delpol)
1515 			break;
1516 	}
1517 
1518 	if (newpos)
1519 		hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1520 	else
1521 		hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1522 
1523 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1524 		pol->pos = i;
1525 		i++;
1526 	}
1527 }
1528 
1529 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1530 						   struct xfrm_policy *policy,
1531 						   bool excl)
1532 {
1533 	struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1534 
1535 	hlist_for_each_entry(pol, chain, bydst) {
1536 		if (pol->type == policy->type &&
1537 		    pol->if_id == policy->if_id &&
1538 		    !selector_cmp(&pol->selector, &policy->selector) &&
1539 		    xfrm_policy_mark_match(policy, pol) &&
1540 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1541 		    !WARN_ON(delpol)) {
1542 			if (excl)
1543 				return ERR_PTR(-EEXIST);
1544 			delpol = pol;
1545 			if (policy->priority > pol->priority)
1546 				continue;
1547 		} else if (policy->priority >= pol->priority) {
1548 			newpos = pol;
1549 			continue;
1550 		}
1551 		if (delpol)
1552 			break;
1553 	}
1554 
1555 	if (newpos)
1556 		hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1557 	else
1558 		hlist_add_head_rcu(&policy->bydst, chain);
1559 
1560 	return delpol;
1561 }
1562 
1563 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1564 {
1565 	struct net *net = xp_net(policy);
1566 	struct xfrm_policy *delpol;
1567 	struct hlist_head *chain;
1568 
1569 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1570 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1571 	if (chain)
1572 		delpol = xfrm_policy_insert_list(chain, policy, excl);
1573 	else
1574 		delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1575 
1576 	if (IS_ERR(delpol)) {
1577 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1578 		return PTR_ERR(delpol);
1579 	}
1580 
1581 	__xfrm_policy_link(policy, dir);
1582 
1583 	/* After previous checking, family can either be AF_INET or AF_INET6 */
1584 	if (policy->family == AF_INET)
1585 		rt_genid_bump_ipv4(net);
1586 	else
1587 		rt_genid_bump_ipv6(net);
1588 
1589 	if (delpol) {
1590 		xfrm_policy_requeue(delpol, policy);
1591 		__xfrm_policy_unlink(delpol, dir);
1592 	}
1593 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1594 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1595 	policy->curlft.add_time = ktime_get_real_seconds();
1596 	policy->curlft.use_time = 0;
1597 	if (!mod_timer(&policy->timer, jiffies + HZ))
1598 		xfrm_pol_hold(policy);
1599 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1600 
1601 	if (delpol)
1602 		xfrm_policy_kill(delpol);
1603 	else if (xfrm_bydst_should_resize(net, dir, NULL))
1604 		schedule_work(&net->xfrm.policy_hash_work);
1605 
1606 	return 0;
1607 }
1608 EXPORT_SYMBOL(xfrm_policy_insert);
1609 
1610 static struct xfrm_policy *
1611 __xfrm_policy_bysel_ctx(struct hlist_head *chain, u32 mark, u32 if_id,
1612 			u8 type, int dir,
1613 			struct xfrm_selector *sel,
1614 			struct xfrm_sec_ctx *ctx)
1615 {
1616 	struct xfrm_policy *pol;
1617 
1618 	if (!chain)
1619 		return NULL;
1620 
1621 	hlist_for_each_entry(pol, chain, bydst) {
1622 		if (pol->type == type &&
1623 		    pol->if_id == if_id &&
1624 		    (mark & pol->mark.m) == pol->mark.v &&
1625 		    !selector_cmp(sel, &pol->selector) &&
1626 		    xfrm_sec_ctx_match(ctx, pol->security))
1627 			return pol;
1628 	}
1629 
1630 	return NULL;
1631 }
1632 
1633 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u32 if_id,
1634 					  u8 type, int dir,
1635 					  struct xfrm_selector *sel,
1636 					  struct xfrm_sec_ctx *ctx, int delete,
1637 					  int *err)
1638 {
1639 	struct xfrm_pol_inexact_bin *bin = NULL;
1640 	struct xfrm_policy *pol, *ret = NULL;
1641 	struct hlist_head *chain;
1642 
1643 	*err = 0;
1644 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1645 	chain = policy_hash_bysel(net, sel, sel->family, dir);
1646 	if (!chain) {
1647 		struct xfrm_pol_inexact_candidates cand;
1648 		int i;
1649 
1650 		bin = xfrm_policy_inexact_lookup(net, type,
1651 						 sel->family, dir, if_id);
1652 		if (!bin) {
1653 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1654 			return NULL;
1655 		}
1656 
1657 		if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1658 							 &sel->saddr,
1659 							 &sel->daddr)) {
1660 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1661 			return NULL;
1662 		}
1663 
1664 		pol = NULL;
1665 		for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1666 			struct xfrm_policy *tmp;
1667 
1668 			tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1669 						      if_id, type, dir,
1670 						      sel, ctx);
1671 			if (!tmp)
1672 				continue;
1673 
1674 			if (!pol || tmp->pos < pol->pos)
1675 				pol = tmp;
1676 		}
1677 	} else {
1678 		pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1679 					      sel, ctx);
1680 	}
1681 
1682 	if (pol) {
1683 		xfrm_pol_hold(pol);
1684 		if (delete) {
1685 			*err = security_xfrm_policy_delete(pol->security);
1686 			if (*err) {
1687 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1688 				return pol;
1689 			}
1690 			__xfrm_policy_unlink(pol, dir);
1691 		}
1692 		ret = pol;
1693 	}
1694 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1695 
1696 	if (ret && delete)
1697 		xfrm_policy_kill(ret);
1698 	if (bin && delete)
1699 		xfrm_policy_inexact_prune_bin(bin);
1700 	return ret;
1701 }
1702 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1703 
1704 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u32 if_id,
1705 				     u8 type, int dir, u32 id, int delete,
1706 				     int *err)
1707 {
1708 	struct xfrm_policy *pol, *ret;
1709 	struct hlist_head *chain;
1710 
1711 	*err = -ENOENT;
1712 	if (xfrm_policy_id2dir(id) != dir)
1713 		return NULL;
1714 
1715 	*err = 0;
1716 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1717 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
1718 	ret = NULL;
1719 	hlist_for_each_entry(pol, chain, byidx) {
1720 		if (pol->type == type && pol->index == id &&
1721 		    pol->if_id == if_id &&
1722 		    (mark & pol->mark.m) == pol->mark.v) {
1723 			xfrm_pol_hold(pol);
1724 			if (delete) {
1725 				*err = security_xfrm_policy_delete(
1726 								pol->security);
1727 				if (*err) {
1728 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1729 					return pol;
1730 				}
1731 				__xfrm_policy_unlink(pol, dir);
1732 			}
1733 			ret = pol;
1734 			break;
1735 		}
1736 	}
1737 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1738 
1739 	if (ret && delete)
1740 		xfrm_policy_kill(ret);
1741 	return ret;
1742 }
1743 EXPORT_SYMBOL(xfrm_policy_byid);
1744 
1745 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1746 static inline int
1747 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1748 {
1749 	struct xfrm_policy *pol;
1750 	int err = 0;
1751 
1752 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1753 		if (pol->walk.dead ||
1754 		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1755 		    pol->type != type)
1756 			continue;
1757 
1758 		err = security_xfrm_policy_delete(pol->security);
1759 		if (err) {
1760 			xfrm_audit_policy_delete(pol, 0, task_valid);
1761 			return err;
1762 		}
1763 	}
1764 	return err;
1765 }
1766 #else
1767 static inline int
1768 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1769 {
1770 	return 0;
1771 }
1772 #endif
1773 
1774 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1775 {
1776 	int dir, err = 0, cnt = 0;
1777 	struct xfrm_policy *pol;
1778 
1779 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1780 
1781 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1782 	if (err)
1783 		goto out;
1784 
1785 again:
1786 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1787 		dir = xfrm_policy_id2dir(pol->index);
1788 		if (pol->walk.dead ||
1789 		    dir >= XFRM_POLICY_MAX ||
1790 		    pol->type != type)
1791 			continue;
1792 
1793 		__xfrm_policy_unlink(pol, dir);
1794 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1795 		cnt++;
1796 		xfrm_audit_policy_delete(pol, 1, task_valid);
1797 		xfrm_policy_kill(pol);
1798 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1799 		goto again;
1800 	}
1801 	if (cnt)
1802 		__xfrm_policy_inexact_flush(net);
1803 	else
1804 		err = -ESRCH;
1805 out:
1806 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1807 	return err;
1808 }
1809 EXPORT_SYMBOL(xfrm_policy_flush);
1810 
1811 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1812 		     int (*func)(struct xfrm_policy *, int, int, void*),
1813 		     void *data)
1814 {
1815 	struct xfrm_policy *pol;
1816 	struct xfrm_policy_walk_entry *x;
1817 	int error = 0;
1818 
1819 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1820 	    walk->type != XFRM_POLICY_TYPE_ANY)
1821 		return -EINVAL;
1822 
1823 	if (list_empty(&walk->walk.all) && walk->seq != 0)
1824 		return 0;
1825 
1826 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1827 	if (list_empty(&walk->walk.all))
1828 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1829 	else
1830 		x = list_first_entry(&walk->walk.all,
1831 				     struct xfrm_policy_walk_entry, all);
1832 
1833 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1834 		if (x->dead)
1835 			continue;
1836 		pol = container_of(x, struct xfrm_policy, walk);
1837 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1838 		    walk->type != pol->type)
1839 			continue;
1840 		error = func(pol, xfrm_policy_id2dir(pol->index),
1841 			     walk->seq, data);
1842 		if (error) {
1843 			list_move_tail(&walk->walk.all, &x->all);
1844 			goto out;
1845 		}
1846 		walk->seq++;
1847 	}
1848 	if (walk->seq == 0) {
1849 		error = -ENOENT;
1850 		goto out;
1851 	}
1852 	list_del_init(&walk->walk.all);
1853 out:
1854 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1855 	return error;
1856 }
1857 EXPORT_SYMBOL(xfrm_policy_walk);
1858 
1859 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1860 {
1861 	INIT_LIST_HEAD(&walk->walk.all);
1862 	walk->walk.dead = 1;
1863 	walk->type = type;
1864 	walk->seq = 0;
1865 }
1866 EXPORT_SYMBOL(xfrm_policy_walk_init);
1867 
1868 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1869 {
1870 	if (list_empty(&walk->walk.all))
1871 		return;
1872 
1873 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1874 	list_del(&walk->walk.all);
1875 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1876 }
1877 EXPORT_SYMBOL(xfrm_policy_walk_done);
1878 
1879 /*
1880  * Find policy to apply to this flow.
1881  *
1882  * Returns 0 if policy found, else an -errno.
1883  */
1884 static int xfrm_policy_match(const struct xfrm_policy *pol,
1885 			     const struct flowi *fl,
1886 			     u8 type, u16 family, int dir, u32 if_id)
1887 {
1888 	const struct xfrm_selector *sel = &pol->selector;
1889 	int ret = -ESRCH;
1890 	bool match;
1891 
1892 	if (pol->family != family ||
1893 	    pol->if_id != if_id ||
1894 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1895 	    pol->type != type)
1896 		return ret;
1897 
1898 	match = xfrm_selector_match(sel, fl, family);
1899 	if (match)
1900 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1901 						  dir);
1902 	return ret;
1903 }
1904 
1905 static struct xfrm_pol_inexact_node *
1906 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1907 				seqcount_t *count,
1908 				const xfrm_address_t *addr, u16 family)
1909 {
1910 	const struct rb_node *parent;
1911 	int seq;
1912 
1913 again:
1914 	seq = read_seqcount_begin(count);
1915 
1916 	parent = rcu_dereference_raw(r->rb_node);
1917 	while (parent) {
1918 		struct xfrm_pol_inexact_node *node;
1919 		int delta;
1920 
1921 		node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1922 
1923 		delta = xfrm_policy_addr_delta(addr, &node->addr,
1924 					       node->prefixlen, family);
1925 		if (delta < 0) {
1926 			parent = rcu_dereference_raw(parent->rb_left);
1927 			continue;
1928 		} else if (delta > 0) {
1929 			parent = rcu_dereference_raw(parent->rb_right);
1930 			continue;
1931 		}
1932 
1933 		return node;
1934 	}
1935 
1936 	if (read_seqcount_retry(count, seq))
1937 		goto again;
1938 
1939 	return NULL;
1940 }
1941 
1942 static bool
1943 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1944 				    struct xfrm_pol_inexact_bin *b,
1945 				    const xfrm_address_t *saddr,
1946 				    const xfrm_address_t *daddr)
1947 {
1948 	struct xfrm_pol_inexact_node *n;
1949 	u16 family;
1950 
1951 	if (!b)
1952 		return false;
1953 
1954 	family = b->k.family;
1955 	memset(cand, 0, sizeof(*cand));
1956 	cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1957 
1958 	n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1959 					    family);
1960 	if (n) {
1961 		cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1962 		n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1963 						    family);
1964 		if (n)
1965 			cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1966 	}
1967 
1968 	n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1969 					    family);
1970 	if (n)
1971 		cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1972 
1973 	return true;
1974 }
1975 
1976 static struct xfrm_pol_inexact_bin *
1977 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1978 			       u8 dir, u32 if_id)
1979 {
1980 	struct xfrm_pol_inexact_key k = {
1981 		.family = family,
1982 		.type = type,
1983 		.dir = dir,
1984 		.if_id = if_id,
1985 	};
1986 
1987 	write_pnet(&k.net, net);
1988 
1989 	return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1990 				 xfrm_pol_inexact_params);
1991 }
1992 
1993 static struct xfrm_pol_inexact_bin *
1994 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1995 			   u8 dir, u32 if_id)
1996 {
1997 	struct xfrm_pol_inexact_bin *bin;
1998 
1999 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2000 
2001 	rcu_read_lock();
2002 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2003 	rcu_read_unlock();
2004 
2005 	return bin;
2006 }
2007 
2008 static struct xfrm_policy *
2009 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2010 			      struct xfrm_policy *prefer,
2011 			      const struct flowi *fl,
2012 			      u8 type, u16 family, int dir, u32 if_id)
2013 {
2014 	u32 priority = prefer ? prefer->priority : ~0u;
2015 	struct xfrm_policy *pol;
2016 
2017 	if (!chain)
2018 		return NULL;
2019 
2020 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2021 		int err;
2022 
2023 		if (pol->priority > priority)
2024 			break;
2025 
2026 		err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2027 		if (err) {
2028 			if (err != -ESRCH)
2029 				return ERR_PTR(err);
2030 
2031 			continue;
2032 		}
2033 
2034 		if (prefer) {
2035 			/* matches.  Is it older than *prefer? */
2036 			if (pol->priority == priority &&
2037 			    prefer->pos < pol->pos)
2038 				return prefer;
2039 		}
2040 
2041 		return pol;
2042 	}
2043 
2044 	return NULL;
2045 }
2046 
2047 static struct xfrm_policy *
2048 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2049 			    struct xfrm_policy *prefer,
2050 			    const struct flowi *fl,
2051 			    u8 type, u16 family, int dir, u32 if_id)
2052 {
2053 	struct xfrm_policy *tmp;
2054 	int i;
2055 
2056 	for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2057 		tmp = __xfrm_policy_eval_candidates(cand->res[i],
2058 						    prefer,
2059 						    fl, type, family, dir,
2060 						    if_id);
2061 		if (!tmp)
2062 			continue;
2063 
2064 		if (IS_ERR(tmp))
2065 			return tmp;
2066 		prefer = tmp;
2067 	}
2068 
2069 	return prefer;
2070 }
2071 
2072 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2073 						     const struct flowi *fl,
2074 						     u16 family, u8 dir,
2075 						     u32 if_id)
2076 {
2077 	struct xfrm_pol_inexact_candidates cand;
2078 	const xfrm_address_t *daddr, *saddr;
2079 	struct xfrm_pol_inexact_bin *bin;
2080 	struct xfrm_policy *pol, *ret;
2081 	struct hlist_head *chain;
2082 	unsigned int sequence;
2083 	int err;
2084 
2085 	daddr = xfrm_flowi_daddr(fl, family);
2086 	saddr = xfrm_flowi_saddr(fl, family);
2087 	if (unlikely(!daddr || !saddr))
2088 		return NULL;
2089 
2090 	rcu_read_lock();
2091  retry:
2092 	do {
2093 		sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2094 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
2095 	} while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2096 
2097 	ret = NULL;
2098 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2099 		err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2100 		if (err) {
2101 			if (err == -ESRCH)
2102 				continue;
2103 			else {
2104 				ret = ERR_PTR(err);
2105 				goto fail;
2106 			}
2107 		} else {
2108 			ret = pol;
2109 			break;
2110 		}
2111 	}
2112 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2113 	if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2114 							 daddr))
2115 		goto skip_inexact;
2116 
2117 	pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2118 					  family, dir, if_id);
2119 	if (pol) {
2120 		ret = pol;
2121 		if (IS_ERR(pol))
2122 			goto fail;
2123 	}
2124 
2125 skip_inexact:
2126 	if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2127 		goto retry;
2128 
2129 	if (ret && !xfrm_pol_hold_rcu(ret))
2130 		goto retry;
2131 fail:
2132 	rcu_read_unlock();
2133 
2134 	return ret;
2135 }
2136 
2137 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2138 					      const struct flowi *fl,
2139 					      u16 family, u8 dir, u32 if_id)
2140 {
2141 #ifdef CONFIG_XFRM_SUB_POLICY
2142 	struct xfrm_policy *pol;
2143 
2144 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2145 					dir, if_id);
2146 	if (pol != NULL)
2147 		return pol;
2148 #endif
2149 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2150 					 dir, if_id);
2151 }
2152 
2153 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2154 						 const struct flowi *fl,
2155 						 u16 family, u32 if_id)
2156 {
2157 	struct xfrm_policy *pol;
2158 
2159 	rcu_read_lock();
2160  again:
2161 	pol = rcu_dereference(sk->sk_policy[dir]);
2162 	if (pol != NULL) {
2163 		bool match;
2164 		int err = 0;
2165 
2166 		if (pol->family != family) {
2167 			pol = NULL;
2168 			goto out;
2169 		}
2170 
2171 		match = xfrm_selector_match(&pol->selector, fl, family);
2172 		if (match) {
2173 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2174 			    pol->if_id != if_id) {
2175 				pol = NULL;
2176 				goto out;
2177 			}
2178 			err = security_xfrm_policy_lookup(pol->security,
2179 						      fl->flowi_secid,
2180 						      dir);
2181 			if (!err) {
2182 				if (!xfrm_pol_hold_rcu(pol))
2183 					goto again;
2184 			} else if (err == -ESRCH) {
2185 				pol = NULL;
2186 			} else {
2187 				pol = ERR_PTR(err);
2188 			}
2189 		} else
2190 			pol = NULL;
2191 	}
2192 out:
2193 	rcu_read_unlock();
2194 	return pol;
2195 }
2196 
2197 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2198 {
2199 	struct net *net = xp_net(pol);
2200 
2201 	list_add(&pol->walk.all, &net->xfrm.policy_all);
2202 	net->xfrm.policy_count[dir]++;
2203 	xfrm_pol_hold(pol);
2204 }
2205 
2206 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2207 						int dir)
2208 {
2209 	struct net *net = xp_net(pol);
2210 
2211 	if (list_empty(&pol->walk.all))
2212 		return NULL;
2213 
2214 	/* Socket policies are not hashed. */
2215 	if (!hlist_unhashed(&pol->bydst)) {
2216 		hlist_del_rcu(&pol->bydst);
2217 		hlist_del_init(&pol->bydst_inexact_list);
2218 		hlist_del(&pol->byidx);
2219 	}
2220 
2221 	list_del_init(&pol->walk.all);
2222 	net->xfrm.policy_count[dir]--;
2223 
2224 	return pol;
2225 }
2226 
2227 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2228 {
2229 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2230 }
2231 
2232 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2233 {
2234 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2235 }
2236 
2237 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2238 {
2239 	struct net *net = xp_net(pol);
2240 
2241 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2242 	pol = __xfrm_policy_unlink(pol, dir);
2243 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2244 	if (pol) {
2245 		xfrm_policy_kill(pol);
2246 		return 0;
2247 	}
2248 	return -ENOENT;
2249 }
2250 EXPORT_SYMBOL(xfrm_policy_delete);
2251 
2252 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2253 {
2254 	struct net *net = sock_net(sk);
2255 	struct xfrm_policy *old_pol;
2256 
2257 #ifdef CONFIG_XFRM_SUB_POLICY
2258 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2259 		return -EINVAL;
2260 #endif
2261 
2262 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2263 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2264 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2265 	if (pol) {
2266 		pol->curlft.add_time = ktime_get_real_seconds();
2267 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2268 		xfrm_sk_policy_link(pol, dir);
2269 	}
2270 	rcu_assign_pointer(sk->sk_policy[dir], pol);
2271 	if (old_pol) {
2272 		if (pol)
2273 			xfrm_policy_requeue(old_pol, pol);
2274 
2275 		/* Unlinking succeeds always. This is the only function
2276 		 * allowed to delete or replace socket policy.
2277 		 */
2278 		xfrm_sk_policy_unlink(old_pol, dir);
2279 	}
2280 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2281 
2282 	if (old_pol) {
2283 		xfrm_policy_kill(old_pol);
2284 	}
2285 	return 0;
2286 }
2287 
2288 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2289 {
2290 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2291 	struct net *net = xp_net(old);
2292 
2293 	if (newp) {
2294 		newp->selector = old->selector;
2295 		if (security_xfrm_policy_clone(old->security,
2296 					       &newp->security)) {
2297 			kfree(newp);
2298 			return NULL;  /* ENOMEM */
2299 		}
2300 		newp->lft = old->lft;
2301 		newp->curlft = old->curlft;
2302 		newp->mark = old->mark;
2303 		newp->if_id = old->if_id;
2304 		newp->action = old->action;
2305 		newp->flags = old->flags;
2306 		newp->xfrm_nr = old->xfrm_nr;
2307 		newp->index = old->index;
2308 		newp->type = old->type;
2309 		newp->family = old->family;
2310 		memcpy(newp->xfrm_vec, old->xfrm_vec,
2311 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2312 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2313 		xfrm_sk_policy_link(newp, dir);
2314 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2315 		xfrm_pol_put(newp);
2316 	}
2317 	return newp;
2318 }
2319 
2320 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2321 {
2322 	const struct xfrm_policy *p;
2323 	struct xfrm_policy *np;
2324 	int i, ret = 0;
2325 
2326 	rcu_read_lock();
2327 	for (i = 0; i < 2; i++) {
2328 		p = rcu_dereference(osk->sk_policy[i]);
2329 		if (p) {
2330 			np = clone_policy(p, i);
2331 			if (unlikely(!np)) {
2332 				ret = -ENOMEM;
2333 				break;
2334 			}
2335 			rcu_assign_pointer(sk->sk_policy[i], np);
2336 		}
2337 	}
2338 	rcu_read_unlock();
2339 	return ret;
2340 }
2341 
2342 static int
2343 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2344 	       xfrm_address_t *remote, unsigned short family, u32 mark)
2345 {
2346 	int err;
2347 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2348 
2349 	if (unlikely(afinfo == NULL))
2350 		return -EINVAL;
2351 	err = afinfo->get_saddr(net, oif, local, remote, mark);
2352 	rcu_read_unlock();
2353 	return err;
2354 }
2355 
2356 /* Resolve list of templates for the flow, given policy. */
2357 
2358 static int
2359 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2360 		      struct xfrm_state **xfrm, unsigned short family)
2361 {
2362 	struct net *net = xp_net(policy);
2363 	int nx;
2364 	int i, error;
2365 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2366 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2367 	xfrm_address_t tmp;
2368 
2369 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2370 		struct xfrm_state *x;
2371 		xfrm_address_t *remote = daddr;
2372 		xfrm_address_t *local  = saddr;
2373 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2374 
2375 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
2376 		    tmpl->mode == XFRM_MODE_BEET) {
2377 			remote = &tmpl->id.daddr;
2378 			local = &tmpl->saddr;
2379 			if (xfrm_addr_any(local, tmpl->encap_family)) {
2380 				error = xfrm_get_saddr(net, fl->flowi_oif,
2381 						       &tmp, remote,
2382 						       tmpl->encap_family, 0);
2383 				if (error)
2384 					goto fail;
2385 				local = &tmp;
2386 			}
2387 		}
2388 
2389 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2390 				    family, policy->if_id);
2391 
2392 		if (x && x->km.state == XFRM_STATE_VALID) {
2393 			xfrm[nx++] = x;
2394 			daddr = remote;
2395 			saddr = local;
2396 			continue;
2397 		}
2398 		if (x) {
2399 			error = (x->km.state == XFRM_STATE_ERROR ?
2400 				 -EINVAL : -EAGAIN);
2401 			xfrm_state_put(x);
2402 		} else if (error == -ESRCH) {
2403 			error = -EAGAIN;
2404 		}
2405 
2406 		if (!tmpl->optional)
2407 			goto fail;
2408 	}
2409 	return nx;
2410 
2411 fail:
2412 	for (nx--; nx >= 0; nx--)
2413 		xfrm_state_put(xfrm[nx]);
2414 	return error;
2415 }
2416 
2417 static int
2418 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2419 		  struct xfrm_state **xfrm, unsigned short family)
2420 {
2421 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
2422 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2423 	int cnx = 0;
2424 	int error;
2425 	int ret;
2426 	int i;
2427 
2428 	for (i = 0; i < npols; i++) {
2429 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2430 			error = -ENOBUFS;
2431 			goto fail;
2432 		}
2433 
2434 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2435 		if (ret < 0) {
2436 			error = ret;
2437 			goto fail;
2438 		} else
2439 			cnx += ret;
2440 	}
2441 
2442 	/* found states are sorted for outbound processing */
2443 	if (npols > 1)
2444 		xfrm_state_sort(xfrm, tpp, cnx, family);
2445 
2446 	return cnx;
2447 
2448  fail:
2449 	for (cnx--; cnx >= 0; cnx--)
2450 		xfrm_state_put(tpp[cnx]);
2451 	return error;
2452 
2453 }
2454 
2455 static int xfrm_get_tos(const struct flowi *fl, int family)
2456 {
2457 	if (family == AF_INET)
2458 		return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2459 
2460 	return 0;
2461 }
2462 
2463 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2464 {
2465 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2466 	struct dst_ops *dst_ops;
2467 	struct xfrm_dst *xdst;
2468 
2469 	if (!afinfo)
2470 		return ERR_PTR(-EINVAL);
2471 
2472 	switch (family) {
2473 	case AF_INET:
2474 		dst_ops = &net->xfrm.xfrm4_dst_ops;
2475 		break;
2476 #if IS_ENABLED(CONFIG_IPV6)
2477 	case AF_INET6:
2478 		dst_ops = &net->xfrm.xfrm6_dst_ops;
2479 		break;
2480 #endif
2481 	default:
2482 		BUG();
2483 	}
2484 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2485 
2486 	if (likely(xdst)) {
2487 		struct dst_entry *dst = &xdst->u.dst;
2488 
2489 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2490 	} else
2491 		xdst = ERR_PTR(-ENOBUFS);
2492 
2493 	rcu_read_unlock();
2494 
2495 	return xdst;
2496 }
2497 
2498 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2499 			   int nfheader_len)
2500 {
2501 	if (dst->ops->family == AF_INET6) {
2502 		struct rt6_info *rt = (struct rt6_info *)dst;
2503 		path->path_cookie = rt6_get_cookie(rt);
2504 		path->u.rt6.rt6i_nfheader_len = nfheader_len;
2505 	}
2506 }
2507 
2508 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2509 				const struct flowi *fl)
2510 {
2511 	const struct xfrm_policy_afinfo *afinfo =
2512 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2513 	int err;
2514 
2515 	if (!afinfo)
2516 		return -EINVAL;
2517 
2518 	err = afinfo->fill_dst(xdst, dev, fl);
2519 
2520 	rcu_read_unlock();
2521 
2522 	return err;
2523 }
2524 
2525 
2526 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2527  * all the metrics... Shortly, bundle a bundle.
2528  */
2529 
2530 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2531 					    struct xfrm_state **xfrm,
2532 					    struct xfrm_dst **bundle,
2533 					    int nx,
2534 					    const struct flowi *fl,
2535 					    struct dst_entry *dst)
2536 {
2537 	const struct xfrm_state_afinfo *afinfo;
2538 	const struct xfrm_mode *inner_mode;
2539 	struct net *net = xp_net(policy);
2540 	unsigned long now = jiffies;
2541 	struct net_device *dev;
2542 	struct xfrm_dst *xdst_prev = NULL;
2543 	struct xfrm_dst *xdst0 = NULL;
2544 	int i = 0;
2545 	int err;
2546 	int header_len = 0;
2547 	int nfheader_len = 0;
2548 	int trailer_len = 0;
2549 	int tos;
2550 	int family = policy->selector.family;
2551 	xfrm_address_t saddr, daddr;
2552 
2553 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2554 
2555 	tos = xfrm_get_tos(fl, family);
2556 
2557 	dst_hold(dst);
2558 
2559 	for (; i < nx; i++) {
2560 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2561 		struct dst_entry *dst1 = &xdst->u.dst;
2562 
2563 		err = PTR_ERR(xdst);
2564 		if (IS_ERR(xdst)) {
2565 			dst_release(dst);
2566 			goto put_states;
2567 		}
2568 
2569 		bundle[i] = xdst;
2570 		if (!xdst_prev)
2571 			xdst0 = xdst;
2572 		else
2573 			/* Ref count is taken during xfrm_alloc_dst()
2574 			 * No need to do dst_clone() on dst1
2575 			 */
2576 			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2577 
2578 		if (xfrm[i]->sel.family == AF_UNSPEC) {
2579 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
2580 							xfrm_af2proto(family));
2581 			if (!inner_mode) {
2582 				err = -EAFNOSUPPORT;
2583 				dst_release(dst);
2584 				goto put_states;
2585 			}
2586 		} else
2587 			inner_mode = &xfrm[i]->inner_mode;
2588 
2589 		xdst->route = dst;
2590 		dst_copy_metrics(dst1, dst);
2591 
2592 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2593 			__u32 mark = 0;
2594 
2595 			if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2596 				mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2597 
2598 			family = xfrm[i]->props.family;
2599 			dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2600 					      &saddr, &daddr, family, mark);
2601 			err = PTR_ERR(dst);
2602 			if (IS_ERR(dst))
2603 				goto put_states;
2604 		} else
2605 			dst_hold(dst);
2606 
2607 		dst1->xfrm = xfrm[i];
2608 		xdst->xfrm_genid = xfrm[i]->genid;
2609 
2610 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2611 		dst1->flags |= DST_HOST;
2612 		dst1->lastuse = now;
2613 
2614 		dst1->input = dst_discard;
2615 
2616 		rcu_read_lock();
2617 		afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2618 		if (likely(afinfo))
2619 			dst1->output = afinfo->output;
2620 		else
2621 			dst1->output = dst_discard_out;
2622 		rcu_read_unlock();
2623 
2624 		xdst_prev = xdst;
2625 
2626 		header_len += xfrm[i]->props.header_len;
2627 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2628 			nfheader_len += xfrm[i]->props.header_len;
2629 		trailer_len += xfrm[i]->props.trailer_len;
2630 	}
2631 
2632 	xfrm_dst_set_child(xdst_prev, dst);
2633 	xdst0->path = dst;
2634 
2635 	err = -ENODEV;
2636 	dev = dst->dev;
2637 	if (!dev)
2638 		goto free_dst;
2639 
2640 	xfrm_init_path(xdst0, dst, nfheader_len);
2641 	xfrm_init_pmtu(bundle, nx);
2642 
2643 	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2644 	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2645 		err = xfrm_fill_dst(xdst_prev, dev, fl);
2646 		if (err)
2647 			goto free_dst;
2648 
2649 		xdst_prev->u.dst.header_len = header_len;
2650 		xdst_prev->u.dst.trailer_len = trailer_len;
2651 		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2652 		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2653 	}
2654 
2655 	return &xdst0->u.dst;
2656 
2657 put_states:
2658 	for (; i < nx; i++)
2659 		xfrm_state_put(xfrm[i]);
2660 free_dst:
2661 	if (xdst0)
2662 		dst_release_immediate(&xdst0->u.dst);
2663 
2664 	return ERR_PTR(err);
2665 }
2666 
2667 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2668 				struct xfrm_policy **pols,
2669 				int *num_pols, int *num_xfrms)
2670 {
2671 	int i;
2672 
2673 	if (*num_pols == 0 || !pols[0]) {
2674 		*num_pols = 0;
2675 		*num_xfrms = 0;
2676 		return 0;
2677 	}
2678 	if (IS_ERR(pols[0]))
2679 		return PTR_ERR(pols[0]);
2680 
2681 	*num_xfrms = pols[0]->xfrm_nr;
2682 
2683 #ifdef CONFIG_XFRM_SUB_POLICY
2684 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2685 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2686 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2687 						    XFRM_POLICY_TYPE_MAIN,
2688 						    fl, family,
2689 						    XFRM_POLICY_OUT,
2690 						    pols[0]->if_id);
2691 		if (pols[1]) {
2692 			if (IS_ERR(pols[1])) {
2693 				xfrm_pols_put(pols, *num_pols);
2694 				return PTR_ERR(pols[1]);
2695 			}
2696 			(*num_pols)++;
2697 			(*num_xfrms) += pols[1]->xfrm_nr;
2698 		}
2699 	}
2700 #endif
2701 	for (i = 0; i < *num_pols; i++) {
2702 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
2703 			*num_xfrms = -1;
2704 			break;
2705 		}
2706 	}
2707 
2708 	return 0;
2709 
2710 }
2711 
2712 static struct xfrm_dst *
2713 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2714 			       const struct flowi *fl, u16 family,
2715 			       struct dst_entry *dst_orig)
2716 {
2717 	struct net *net = xp_net(pols[0]);
2718 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2719 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2720 	struct xfrm_dst *xdst;
2721 	struct dst_entry *dst;
2722 	int err;
2723 
2724 	/* Try to instantiate a bundle */
2725 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2726 	if (err <= 0) {
2727 		if (err == 0)
2728 			return NULL;
2729 
2730 		if (err != -EAGAIN)
2731 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2732 		return ERR_PTR(err);
2733 	}
2734 
2735 	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2736 	if (IS_ERR(dst)) {
2737 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2738 		return ERR_CAST(dst);
2739 	}
2740 
2741 	xdst = (struct xfrm_dst *)dst;
2742 	xdst->num_xfrms = err;
2743 	xdst->num_pols = num_pols;
2744 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2745 	xdst->policy_genid = atomic_read(&pols[0]->genid);
2746 
2747 	return xdst;
2748 }
2749 
2750 static void xfrm_policy_queue_process(struct timer_list *t)
2751 {
2752 	struct sk_buff *skb;
2753 	struct sock *sk;
2754 	struct dst_entry *dst;
2755 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2756 	struct net *net = xp_net(pol);
2757 	struct xfrm_policy_queue *pq = &pol->polq;
2758 	struct flowi fl;
2759 	struct sk_buff_head list;
2760 
2761 	spin_lock(&pq->hold_queue.lock);
2762 	skb = skb_peek(&pq->hold_queue);
2763 	if (!skb) {
2764 		spin_unlock(&pq->hold_queue.lock);
2765 		goto out;
2766 	}
2767 	dst = skb_dst(skb);
2768 	sk = skb->sk;
2769 	xfrm_decode_session(skb, &fl, dst->ops->family);
2770 	spin_unlock(&pq->hold_queue.lock);
2771 
2772 	dst_hold(xfrm_dst_path(dst));
2773 	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2774 	if (IS_ERR(dst))
2775 		goto purge_queue;
2776 
2777 	if (dst->flags & DST_XFRM_QUEUE) {
2778 		dst_release(dst);
2779 
2780 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2781 			goto purge_queue;
2782 
2783 		pq->timeout = pq->timeout << 1;
2784 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2785 			xfrm_pol_hold(pol);
2786 		goto out;
2787 	}
2788 
2789 	dst_release(dst);
2790 
2791 	__skb_queue_head_init(&list);
2792 
2793 	spin_lock(&pq->hold_queue.lock);
2794 	pq->timeout = 0;
2795 	skb_queue_splice_init(&pq->hold_queue, &list);
2796 	spin_unlock(&pq->hold_queue.lock);
2797 
2798 	while (!skb_queue_empty(&list)) {
2799 		skb = __skb_dequeue(&list);
2800 
2801 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2802 		dst_hold(xfrm_dst_path(skb_dst(skb)));
2803 		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2804 		if (IS_ERR(dst)) {
2805 			kfree_skb(skb);
2806 			continue;
2807 		}
2808 
2809 		nf_reset(skb);
2810 		skb_dst_drop(skb);
2811 		skb_dst_set(skb, dst);
2812 
2813 		dst_output(net, skb->sk, skb);
2814 	}
2815 
2816 out:
2817 	xfrm_pol_put(pol);
2818 	return;
2819 
2820 purge_queue:
2821 	pq->timeout = 0;
2822 	skb_queue_purge(&pq->hold_queue);
2823 	xfrm_pol_put(pol);
2824 }
2825 
2826 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2827 {
2828 	unsigned long sched_next;
2829 	struct dst_entry *dst = skb_dst(skb);
2830 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2831 	struct xfrm_policy *pol = xdst->pols[0];
2832 	struct xfrm_policy_queue *pq = &pol->polq;
2833 
2834 	if (unlikely(skb_fclone_busy(sk, skb))) {
2835 		kfree_skb(skb);
2836 		return 0;
2837 	}
2838 
2839 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2840 		kfree_skb(skb);
2841 		return -EAGAIN;
2842 	}
2843 
2844 	skb_dst_force(skb);
2845 
2846 	spin_lock_bh(&pq->hold_queue.lock);
2847 
2848 	if (!pq->timeout)
2849 		pq->timeout = XFRM_QUEUE_TMO_MIN;
2850 
2851 	sched_next = jiffies + pq->timeout;
2852 
2853 	if (del_timer(&pq->hold_timer)) {
2854 		if (time_before(pq->hold_timer.expires, sched_next))
2855 			sched_next = pq->hold_timer.expires;
2856 		xfrm_pol_put(pol);
2857 	}
2858 
2859 	__skb_queue_tail(&pq->hold_queue, skb);
2860 	if (!mod_timer(&pq->hold_timer, sched_next))
2861 		xfrm_pol_hold(pol);
2862 
2863 	spin_unlock_bh(&pq->hold_queue.lock);
2864 
2865 	return 0;
2866 }
2867 
2868 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2869 						 struct xfrm_flo *xflo,
2870 						 const struct flowi *fl,
2871 						 int num_xfrms,
2872 						 u16 family)
2873 {
2874 	int err;
2875 	struct net_device *dev;
2876 	struct dst_entry *dst;
2877 	struct dst_entry *dst1;
2878 	struct xfrm_dst *xdst;
2879 
2880 	xdst = xfrm_alloc_dst(net, family);
2881 	if (IS_ERR(xdst))
2882 		return xdst;
2883 
2884 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2885 	    net->xfrm.sysctl_larval_drop ||
2886 	    num_xfrms <= 0)
2887 		return xdst;
2888 
2889 	dst = xflo->dst_orig;
2890 	dst1 = &xdst->u.dst;
2891 	dst_hold(dst);
2892 	xdst->route = dst;
2893 
2894 	dst_copy_metrics(dst1, dst);
2895 
2896 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2897 	dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2898 	dst1->lastuse = jiffies;
2899 
2900 	dst1->input = dst_discard;
2901 	dst1->output = xdst_queue_output;
2902 
2903 	dst_hold(dst);
2904 	xfrm_dst_set_child(xdst, dst);
2905 	xdst->path = dst;
2906 
2907 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2908 
2909 	err = -ENODEV;
2910 	dev = dst->dev;
2911 	if (!dev)
2912 		goto free_dst;
2913 
2914 	err = xfrm_fill_dst(xdst, dev, fl);
2915 	if (err)
2916 		goto free_dst;
2917 
2918 out:
2919 	return xdst;
2920 
2921 free_dst:
2922 	dst_release(dst1);
2923 	xdst = ERR_PTR(err);
2924 	goto out;
2925 }
2926 
2927 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2928 					   const struct flowi *fl,
2929 					   u16 family, u8 dir,
2930 					   struct xfrm_flo *xflo, u32 if_id)
2931 {
2932 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2933 	int num_pols = 0, num_xfrms = 0, err;
2934 	struct xfrm_dst *xdst;
2935 
2936 	/* Resolve policies to use if we couldn't get them from
2937 	 * previous cache entry */
2938 	num_pols = 1;
2939 	pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2940 	err = xfrm_expand_policies(fl, family, pols,
2941 					   &num_pols, &num_xfrms);
2942 	if (err < 0)
2943 		goto inc_error;
2944 	if (num_pols == 0)
2945 		return NULL;
2946 	if (num_xfrms <= 0)
2947 		goto make_dummy_bundle;
2948 
2949 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2950 					      xflo->dst_orig);
2951 	if (IS_ERR(xdst)) {
2952 		err = PTR_ERR(xdst);
2953 		if (err == -EREMOTE) {
2954 			xfrm_pols_put(pols, num_pols);
2955 			return NULL;
2956 		}
2957 
2958 		if (err != -EAGAIN)
2959 			goto error;
2960 		goto make_dummy_bundle;
2961 	} else if (xdst == NULL) {
2962 		num_xfrms = 0;
2963 		goto make_dummy_bundle;
2964 	}
2965 
2966 	return xdst;
2967 
2968 make_dummy_bundle:
2969 	/* We found policies, but there's no bundles to instantiate:
2970 	 * either because the policy blocks, has no transformations or
2971 	 * we could not build template (no xfrm_states).*/
2972 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2973 	if (IS_ERR(xdst)) {
2974 		xfrm_pols_put(pols, num_pols);
2975 		return ERR_CAST(xdst);
2976 	}
2977 	xdst->num_pols = num_pols;
2978 	xdst->num_xfrms = num_xfrms;
2979 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2980 
2981 	return xdst;
2982 
2983 inc_error:
2984 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2985 error:
2986 	xfrm_pols_put(pols, num_pols);
2987 	return ERR_PTR(err);
2988 }
2989 
2990 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2991 					struct dst_entry *dst_orig)
2992 {
2993 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2994 	struct dst_entry *ret;
2995 
2996 	if (!afinfo) {
2997 		dst_release(dst_orig);
2998 		return ERR_PTR(-EINVAL);
2999 	} else {
3000 		ret = afinfo->blackhole_route(net, dst_orig);
3001 	}
3002 	rcu_read_unlock();
3003 
3004 	return ret;
3005 }
3006 
3007 /* Finds/creates a bundle for given flow and if_id
3008  *
3009  * At the moment we eat a raw IP route. Mostly to speed up lookups
3010  * on interfaces with disabled IPsec.
3011  *
3012  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3013  * compatibility
3014  */
3015 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3016 					struct dst_entry *dst_orig,
3017 					const struct flowi *fl,
3018 					const struct sock *sk,
3019 					int flags, u32 if_id)
3020 {
3021 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3022 	struct xfrm_dst *xdst;
3023 	struct dst_entry *dst, *route;
3024 	u16 family = dst_orig->ops->family;
3025 	u8 dir = XFRM_POLICY_OUT;
3026 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3027 
3028 	dst = NULL;
3029 	xdst = NULL;
3030 	route = NULL;
3031 
3032 	sk = sk_const_to_full_sk(sk);
3033 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3034 		num_pols = 1;
3035 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3036 						if_id);
3037 		err = xfrm_expand_policies(fl, family, pols,
3038 					   &num_pols, &num_xfrms);
3039 		if (err < 0)
3040 			goto dropdst;
3041 
3042 		if (num_pols) {
3043 			if (num_xfrms <= 0) {
3044 				drop_pols = num_pols;
3045 				goto no_transform;
3046 			}
3047 
3048 			xdst = xfrm_resolve_and_create_bundle(
3049 					pols, num_pols, fl,
3050 					family, dst_orig);
3051 
3052 			if (IS_ERR(xdst)) {
3053 				xfrm_pols_put(pols, num_pols);
3054 				err = PTR_ERR(xdst);
3055 				if (err == -EREMOTE)
3056 					goto nopol;
3057 
3058 				goto dropdst;
3059 			} else if (xdst == NULL) {
3060 				num_xfrms = 0;
3061 				drop_pols = num_pols;
3062 				goto no_transform;
3063 			}
3064 
3065 			route = xdst->route;
3066 		}
3067 	}
3068 
3069 	if (xdst == NULL) {
3070 		struct xfrm_flo xflo;
3071 
3072 		xflo.dst_orig = dst_orig;
3073 		xflo.flags = flags;
3074 
3075 		/* To accelerate a bit...  */
3076 		if ((dst_orig->flags & DST_NOXFRM) ||
3077 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
3078 			goto nopol;
3079 
3080 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3081 		if (xdst == NULL)
3082 			goto nopol;
3083 		if (IS_ERR(xdst)) {
3084 			err = PTR_ERR(xdst);
3085 			goto dropdst;
3086 		}
3087 
3088 		num_pols = xdst->num_pols;
3089 		num_xfrms = xdst->num_xfrms;
3090 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3091 		route = xdst->route;
3092 	}
3093 
3094 	dst = &xdst->u.dst;
3095 	if (route == NULL && num_xfrms > 0) {
3096 		/* The only case when xfrm_bundle_lookup() returns a
3097 		 * bundle with null route, is when the template could
3098 		 * not be resolved. It means policies are there, but
3099 		 * bundle could not be created, since we don't yet
3100 		 * have the xfrm_state's. We need to wait for KM to
3101 		 * negotiate new SA's or bail out with error.*/
3102 		if (net->xfrm.sysctl_larval_drop) {
3103 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3104 			err = -EREMOTE;
3105 			goto error;
3106 		}
3107 
3108 		err = -EAGAIN;
3109 
3110 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3111 		goto error;
3112 	}
3113 
3114 no_transform:
3115 	if (num_pols == 0)
3116 		goto nopol;
3117 
3118 	if ((flags & XFRM_LOOKUP_ICMP) &&
3119 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3120 		err = -ENOENT;
3121 		goto error;
3122 	}
3123 
3124 	for (i = 0; i < num_pols; i++)
3125 		pols[i]->curlft.use_time = ktime_get_real_seconds();
3126 
3127 	if (num_xfrms < 0) {
3128 		/* Prohibit the flow */
3129 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3130 		err = -EPERM;
3131 		goto error;
3132 	} else if (num_xfrms > 0) {
3133 		/* Flow transformed */
3134 		dst_release(dst_orig);
3135 	} else {
3136 		/* Flow passes untransformed */
3137 		dst_release(dst);
3138 		dst = dst_orig;
3139 	}
3140 ok:
3141 	xfrm_pols_put(pols, drop_pols);
3142 	if (dst && dst->xfrm &&
3143 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3144 		dst->flags |= DST_XFRM_TUNNEL;
3145 	return dst;
3146 
3147 nopol:
3148 	if (!(flags & XFRM_LOOKUP_ICMP)) {
3149 		dst = dst_orig;
3150 		goto ok;
3151 	}
3152 	err = -ENOENT;
3153 error:
3154 	dst_release(dst);
3155 dropdst:
3156 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3157 		dst_release(dst_orig);
3158 	xfrm_pols_put(pols, drop_pols);
3159 	return ERR_PTR(err);
3160 }
3161 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3162 
3163 /* Main function: finds/creates a bundle for given flow.
3164  *
3165  * At the moment we eat a raw IP route. Mostly to speed up lookups
3166  * on interfaces with disabled IPsec.
3167  */
3168 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3169 			      const struct flowi *fl, const struct sock *sk,
3170 			      int flags)
3171 {
3172 	return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3173 }
3174 EXPORT_SYMBOL(xfrm_lookup);
3175 
3176 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3177  * Otherwise we may send out blackholed packets.
3178  */
3179 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3180 				    const struct flowi *fl,
3181 				    const struct sock *sk, int flags)
3182 {
3183 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3184 					    flags | XFRM_LOOKUP_QUEUE |
3185 					    XFRM_LOOKUP_KEEP_DST_REF);
3186 
3187 	if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
3188 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
3189 
3190 	if (IS_ERR(dst))
3191 		dst_release(dst_orig);
3192 
3193 	return dst;
3194 }
3195 EXPORT_SYMBOL(xfrm_lookup_route);
3196 
3197 static inline int
3198 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3199 {
3200 	struct sec_path *sp = skb_sec_path(skb);
3201 	struct xfrm_state *x;
3202 
3203 	if (!sp || idx < 0 || idx >= sp->len)
3204 		return 0;
3205 	x = sp->xvec[idx];
3206 	if (!x->type->reject)
3207 		return 0;
3208 	return x->type->reject(x, skb, fl);
3209 }
3210 
3211 /* When skb is transformed back to its "native" form, we have to
3212  * check policy restrictions. At the moment we make this in maximally
3213  * stupid way. Shame on me. :-) Of course, connected sockets must
3214  * have policy cached at them.
3215  */
3216 
3217 static inline int
3218 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3219 	      unsigned short family)
3220 {
3221 	if (xfrm_state_kern(x))
3222 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3223 	return	x->id.proto == tmpl->id.proto &&
3224 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3225 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3226 		x->props.mode == tmpl->mode &&
3227 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3228 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3229 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
3230 		  xfrm_state_addr_cmp(tmpl, x, family));
3231 }
3232 
3233 /*
3234  * 0 or more than 0 is returned when validation is succeeded (either bypass
3235  * because of optional transport mode, or next index of the mathced secpath
3236  * state with the template.
3237  * -1 is returned when no matching template is found.
3238  * Otherwise "-2 - errored_index" is returned.
3239  */
3240 static inline int
3241 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3242 	       unsigned short family)
3243 {
3244 	int idx = start;
3245 
3246 	if (tmpl->optional) {
3247 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
3248 			return start;
3249 	} else
3250 		start = -1;
3251 	for (; idx < sp->len; idx++) {
3252 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3253 			return ++idx;
3254 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3255 			if (start == -1)
3256 				start = -2-idx;
3257 			break;
3258 		}
3259 	}
3260 	return start;
3261 }
3262 
3263 static void
3264 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3265 {
3266 	const struct iphdr *iph = ip_hdr(skb);
3267 	u8 *xprth = skb_network_header(skb) + iph->ihl * 4;
3268 	struct flowi4 *fl4 = &fl->u.ip4;
3269 	int oif = 0;
3270 
3271 	if (skb_dst(skb))
3272 		oif = skb_dst(skb)->dev->ifindex;
3273 
3274 	memset(fl4, 0, sizeof(struct flowi4));
3275 	fl4->flowi4_mark = skb->mark;
3276 	fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3277 
3278 	if (!ip_is_fragment(iph)) {
3279 		switch (iph->protocol) {
3280 		case IPPROTO_UDP:
3281 		case IPPROTO_UDPLITE:
3282 		case IPPROTO_TCP:
3283 		case IPPROTO_SCTP:
3284 		case IPPROTO_DCCP:
3285 			if (xprth + 4 < skb->data ||
3286 			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3287 				__be16 *ports;
3288 
3289 				xprth = skb_network_header(skb) + iph->ihl * 4;
3290 				ports = (__be16 *)xprth;
3291 
3292 				fl4->fl4_sport = ports[!!reverse];
3293 				fl4->fl4_dport = ports[!reverse];
3294 			}
3295 			break;
3296 		case IPPROTO_ICMP:
3297 			if (xprth + 2 < skb->data ||
3298 			    pskb_may_pull(skb, xprth + 2 - skb->data)) {
3299 				u8 *icmp;
3300 
3301 				xprth = skb_network_header(skb) + iph->ihl * 4;
3302 				icmp = xprth;
3303 
3304 				fl4->fl4_icmp_type = icmp[0];
3305 				fl4->fl4_icmp_code = icmp[1];
3306 			}
3307 			break;
3308 		case IPPROTO_ESP:
3309 			if (xprth + 4 < skb->data ||
3310 			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3311 				__be32 *ehdr;
3312 
3313 				xprth = skb_network_header(skb) + iph->ihl * 4;
3314 				ehdr = (__be32 *)xprth;
3315 
3316 				fl4->fl4_ipsec_spi = ehdr[0];
3317 			}
3318 			break;
3319 		case IPPROTO_AH:
3320 			if (xprth + 8 < skb->data ||
3321 			    pskb_may_pull(skb, xprth + 8 - skb->data)) {
3322 				__be32 *ah_hdr;
3323 
3324 				xprth = skb_network_header(skb) + iph->ihl * 4;
3325 				ah_hdr = (__be32 *)xprth;
3326 
3327 				fl4->fl4_ipsec_spi = ah_hdr[1];
3328 			}
3329 			break;
3330 		case IPPROTO_COMP:
3331 			if (xprth + 4 < skb->data ||
3332 			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3333 				__be16 *ipcomp_hdr;
3334 
3335 				xprth = skb_network_header(skb) + iph->ihl * 4;
3336 				ipcomp_hdr = (__be16 *)xprth;
3337 
3338 				fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
3339 			}
3340 			break;
3341 		case IPPROTO_GRE:
3342 			if (xprth + 12 < skb->data ||
3343 			    pskb_may_pull(skb, xprth + 12 - skb->data)) {
3344 				__be16 *greflags;
3345 				__be32 *gre_hdr;
3346 
3347 				xprth = skb_network_header(skb) + iph->ihl * 4;
3348 				greflags = (__be16 *)xprth;
3349 				gre_hdr = (__be32 *)xprth;
3350 
3351 				if (greflags[0] & GRE_KEY) {
3352 					if (greflags[0] & GRE_CSUM)
3353 						gre_hdr++;
3354 					fl4->fl4_gre_key = gre_hdr[1];
3355 				}
3356 			}
3357 			break;
3358 		default:
3359 			fl4->fl4_ipsec_spi = 0;
3360 			break;
3361 		}
3362 	}
3363 	fl4->flowi4_proto = iph->protocol;
3364 	fl4->daddr = reverse ? iph->saddr : iph->daddr;
3365 	fl4->saddr = reverse ? iph->daddr : iph->saddr;
3366 	fl4->flowi4_tos = iph->tos;
3367 }
3368 
3369 #if IS_ENABLED(CONFIG_IPV6)
3370 static void
3371 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3372 {
3373 	struct flowi6 *fl6 = &fl->u.ip6;
3374 	int onlyproto = 0;
3375 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
3376 	u32 offset = sizeof(*hdr);
3377 	struct ipv6_opt_hdr *exthdr;
3378 	const unsigned char *nh = skb_network_header(skb);
3379 	u16 nhoff = IP6CB(skb)->nhoff;
3380 	int oif = 0;
3381 	u8 nexthdr;
3382 
3383 	if (!nhoff)
3384 		nhoff = offsetof(struct ipv6hdr, nexthdr);
3385 
3386 	nexthdr = nh[nhoff];
3387 
3388 	if (skb_dst(skb))
3389 		oif = skb_dst(skb)->dev->ifindex;
3390 
3391 	memset(fl6, 0, sizeof(struct flowi6));
3392 	fl6->flowi6_mark = skb->mark;
3393 	fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3394 
3395 	fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3396 	fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3397 
3398 	while (nh + offset + sizeof(*exthdr) < skb->data ||
3399 	       pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3400 		nh = skb_network_header(skb);
3401 		exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3402 
3403 		switch (nexthdr) {
3404 		case NEXTHDR_FRAGMENT:
3405 			onlyproto = 1;
3406 			/* fall through */
3407 		case NEXTHDR_ROUTING:
3408 		case NEXTHDR_HOP:
3409 		case NEXTHDR_DEST:
3410 			offset += ipv6_optlen(exthdr);
3411 			nexthdr = exthdr->nexthdr;
3412 			exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3413 			break;
3414 		case IPPROTO_UDP:
3415 		case IPPROTO_UDPLITE:
3416 		case IPPROTO_TCP:
3417 		case IPPROTO_SCTP:
3418 		case IPPROTO_DCCP:
3419 			if (!onlyproto && (nh + offset + 4 < skb->data ||
3420 			     pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3421 				__be16 *ports;
3422 
3423 				nh = skb_network_header(skb);
3424 				ports = (__be16 *)(nh + offset);
3425 				fl6->fl6_sport = ports[!!reverse];
3426 				fl6->fl6_dport = ports[!reverse];
3427 			}
3428 			fl6->flowi6_proto = nexthdr;
3429 			return;
3430 		case IPPROTO_ICMPV6:
3431 			if (!onlyproto && (nh + offset + 2 < skb->data ||
3432 			    pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3433 				u8 *icmp;
3434 
3435 				nh = skb_network_header(skb);
3436 				icmp = (u8 *)(nh + offset);
3437 				fl6->fl6_icmp_type = icmp[0];
3438 				fl6->fl6_icmp_code = icmp[1];
3439 			}
3440 			fl6->flowi6_proto = nexthdr;
3441 			return;
3442 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3443 		case IPPROTO_MH:
3444 			offset += ipv6_optlen(exthdr);
3445 			if (!onlyproto && (nh + offset + 3 < skb->data ||
3446 			    pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3447 				struct ip6_mh *mh;
3448 
3449 				nh = skb_network_header(skb);
3450 				mh = (struct ip6_mh *)(nh + offset);
3451 				fl6->fl6_mh_type = mh->ip6mh_type;
3452 			}
3453 			fl6->flowi6_proto = nexthdr;
3454 			return;
3455 #endif
3456 		/* XXX Why are there these headers? */
3457 		case IPPROTO_AH:
3458 		case IPPROTO_ESP:
3459 		case IPPROTO_COMP:
3460 		default:
3461 			fl6->fl6_ipsec_spi = 0;
3462 			fl6->flowi6_proto = nexthdr;
3463 			return;
3464 		}
3465 	}
3466 }
3467 #endif
3468 
3469 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3470 			  unsigned int family, int reverse)
3471 {
3472 	switch (family) {
3473 	case AF_INET:
3474 		decode_session4(skb, fl, reverse);
3475 		break;
3476 #if IS_ENABLED(CONFIG_IPV6)
3477 	case AF_INET6:
3478 		decode_session6(skb, fl, reverse);
3479 		break;
3480 #endif
3481 	default:
3482 		return -EAFNOSUPPORT;
3483 	}
3484 
3485 	return security_xfrm_decode_session(skb, &fl->flowi_secid);
3486 }
3487 EXPORT_SYMBOL(__xfrm_decode_session);
3488 
3489 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3490 {
3491 	for (; k < sp->len; k++) {
3492 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3493 			*idxp = k;
3494 			return 1;
3495 		}
3496 	}
3497 
3498 	return 0;
3499 }
3500 
3501 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3502 			unsigned short family)
3503 {
3504 	struct net *net = dev_net(skb->dev);
3505 	struct xfrm_policy *pol;
3506 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3507 	int npols = 0;
3508 	int xfrm_nr;
3509 	int pi;
3510 	int reverse;
3511 	struct flowi fl;
3512 	int xerr_idx = -1;
3513 	const struct xfrm_if_cb *ifcb;
3514 	struct sec_path *sp;
3515 	struct xfrm_if *xi;
3516 	u32 if_id = 0;
3517 
3518 	rcu_read_lock();
3519 	ifcb = xfrm_if_get_cb();
3520 
3521 	if (ifcb) {
3522 		xi = ifcb->decode_session(skb, family);
3523 		if (xi) {
3524 			if_id = xi->p.if_id;
3525 			net = xi->net;
3526 		}
3527 	}
3528 	rcu_read_unlock();
3529 
3530 	reverse = dir & ~XFRM_POLICY_MASK;
3531 	dir &= XFRM_POLICY_MASK;
3532 
3533 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3534 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3535 		return 0;
3536 	}
3537 
3538 	nf_nat_decode_session(skb, &fl, family);
3539 
3540 	/* First, check used SA against their selectors. */
3541 	sp = skb_sec_path(skb);
3542 	if (sp) {
3543 		int i;
3544 
3545 		for (i = sp->len - 1; i >= 0; i--) {
3546 			struct xfrm_state *x = sp->xvec[i];
3547 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
3548 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3549 				return 0;
3550 			}
3551 		}
3552 	}
3553 
3554 	pol = NULL;
3555 	sk = sk_to_full_sk(sk);
3556 	if (sk && sk->sk_policy[dir]) {
3557 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3558 		if (IS_ERR(pol)) {
3559 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3560 			return 0;
3561 		}
3562 	}
3563 
3564 	if (!pol)
3565 		pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3566 
3567 	if (IS_ERR(pol)) {
3568 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3569 		return 0;
3570 	}
3571 
3572 	if (!pol) {
3573 		if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3574 			xfrm_secpath_reject(xerr_idx, skb, &fl);
3575 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3576 			return 0;
3577 		}
3578 		return 1;
3579 	}
3580 
3581 	pol->curlft.use_time = ktime_get_real_seconds();
3582 
3583 	pols[0] = pol;
3584 	npols++;
3585 #ifdef CONFIG_XFRM_SUB_POLICY
3586 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3587 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3588 						    &fl, family,
3589 						    XFRM_POLICY_IN, if_id);
3590 		if (pols[1]) {
3591 			if (IS_ERR(pols[1])) {
3592 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3593 				return 0;
3594 			}
3595 			pols[1]->curlft.use_time = ktime_get_real_seconds();
3596 			npols++;
3597 		}
3598 	}
3599 #endif
3600 
3601 	if (pol->action == XFRM_POLICY_ALLOW) {
3602 		static struct sec_path dummy;
3603 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3604 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3605 		struct xfrm_tmpl **tpp = tp;
3606 		int ti = 0;
3607 		int i, k;
3608 
3609 		sp = skb_sec_path(skb);
3610 		if (!sp)
3611 			sp = &dummy;
3612 
3613 		for (pi = 0; pi < npols; pi++) {
3614 			if (pols[pi] != pol &&
3615 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
3616 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3617 				goto reject;
3618 			}
3619 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3620 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3621 				goto reject_error;
3622 			}
3623 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
3624 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
3625 		}
3626 		xfrm_nr = ti;
3627 		if (npols > 1) {
3628 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
3629 			tpp = stp;
3630 		}
3631 
3632 		/* For each tunnel xfrm, find the first matching tmpl.
3633 		 * For each tmpl before that, find corresponding xfrm.
3634 		 * Order is _important_. Later we will implement
3635 		 * some barriers, but at the moment barriers
3636 		 * are implied between each two transformations.
3637 		 */
3638 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3639 			k = xfrm_policy_ok(tpp[i], sp, k, family);
3640 			if (k < 0) {
3641 				if (k < -1)
3642 					/* "-2 - errored_index" returned */
3643 					xerr_idx = -(2+k);
3644 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3645 				goto reject;
3646 			}
3647 		}
3648 
3649 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3650 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3651 			goto reject;
3652 		}
3653 
3654 		xfrm_pols_put(pols, npols);
3655 		return 1;
3656 	}
3657 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3658 
3659 reject:
3660 	xfrm_secpath_reject(xerr_idx, skb, &fl);
3661 reject_error:
3662 	xfrm_pols_put(pols, npols);
3663 	return 0;
3664 }
3665 EXPORT_SYMBOL(__xfrm_policy_check);
3666 
3667 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3668 {
3669 	struct net *net = dev_net(skb->dev);
3670 	struct flowi fl;
3671 	struct dst_entry *dst;
3672 	int res = 1;
3673 
3674 	if (xfrm_decode_session(skb, &fl, family) < 0) {
3675 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3676 		return 0;
3677 	}
3678 
3679 	skb_dst_force(skb);
3680 	if (!skb_dst(skb)) {
3681 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3682 		return 0;
3683 	}
3684 
3685 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3686 	if (IS_ERR(dst)) {
3687 		res = 0;
3688 		dst = NULL;
3689 	}
3690 	skb_dst_set(skb, dst);
3691 	return res;
3692 }
3693 EXPORT_SYMBOL(__xfrm_route_forward);
3694 
3695 /* Optimize later using cookies and generation ids. */
3696 
3697 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3698 {
3699 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3700 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3701 	 * get validated by dst_ops->check on every use.  We do this
3702 	 * because when a normal route referenced by an XFRM dst is
3703 	 * obsoleted we do not go looking around for all parent
3704 	 * referencing XFRM dsts so that we can invalidate them.  It
3705 	 * is just too much work.  Instead we make the checks here on
3706 	 * every use.  For example:
3707 	 *
3708 	 *	XFRM dst A --> IPv4 dst X
3709 	 *
3710 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
3711 	 * in this example).  If X is marked obsolete, "A" will not
3712 	 * notice.  That's what we are validating here via the
3713 	 * stale_bundle() check.
3714 	 *
3715 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3716 	 * be marked on it.
3717 	 * This will force stale_bundle() to fail on any xdst bundle with
3718 	 * this dst linked in it.
3719 	 */
3720 	if (dst->obsolete < 0 && !stale_bundle(dst))
3721 		return dst;
3722 
3723 	return NULL;
3724 }
3725 
3726 static int stale_bundle(struct dst_entry *dst)
3727 {
3728 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3729 }
3730 
3731 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3732 {
3733 	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3734 		dst->dev = dev_net(dev)->loopback_dev;
3735 		dev_hold(dst->dev);
3736 		dev_put(dev);
3737 	}
3738 }
3739 EXPORT_SYMBOL(xfrm_dst_ifdown);
3740 
3741 static void xfrm_link_failure(struct sk_buff *skb)
3742 {
3743 	/* Impossible. Such dst must be popped before reaches point of failure. */
3744 }
3745 
3746 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3747 {
3748 	if (dst) {
3749 		if (dst->obsolete) {
3750 			dst_release(dst);
3751 			dst = NULL;
3752 		}
3753 	}
3754 	return dst;
3755 }
3756 
3757 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3758 {
3759 	while (nr--) {
3760 		struct xfrm_dst *xdst = bundle[nr];
3761 		u32 pmtu, route_mtu_cached;
3762 		struct dst_entry *dst;
3763 
3764 		dst = &xdst->u.dst;
3765 		pmtu = dst_mtu(xfrm_dst_child(dst));
3766 		xdst->child_mtu_cached = pmtu;
3767 
3768 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3769 
3770 		route_mtu_cached = dst_mtu(xdst->route);
3771 		xdst->route_mtu_cached = route_mtu_cached;
3772 
3773 		if (pmtu > route_mtu_cached)
3774 			pmtu = route_mtu_cached;
3775 
3776 		dst_metric_set(dst, RTAX_MTU, pmtu);
3777 	}
3778 }
3779 
3780 /* Check that the bundle accepts the flow and its components are
3781  * still valid.
3782  */
3783 
3784 static int xfrm_bundle_ok(struct xfrm_dst *first)
3785 {
3786 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3787 	struct dst_entry *dst = &first->u.dst;
3788 	struct xfrm_dst *xdst;
3789 	int start_from, nr;
3790 	u32 mtu;
3791 
3792 	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3793 	    (dst->dev && !netif_running(dst->dev)))
3794 		return 0;
3795 
3796 	if (dst->flags & DST_XFRM_QUEUE)
3797 		return 1;
3798 
3799 	start_from = nr = 0;
3800 	do {
3801 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3802 
3803 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
3804 			return 0;
3805 		if (xdst->xfrm_genid != dst->xfrm->genid)
3806 			return 0;
3807 		if (xdst->num_pols > 0 &&
3808 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3809 			return 0;
3810 
3811 		bundle[nr++] = xdst;
3812 
3813 		mtu = dst_mtu(xfrm_dst_child(dst));
3814 		if (xdst->child_mtu_cached != mtu) {
3815 			start_from = nr;
3816 			xdst->child_mtu_cached = mtu;
3817 		}
3818 
3819 		if (!dst_check(xdst->route, xdst->route_cookie))
3820 			return 0;
3821 		mtu = dst_mtu(xdst->route);
3822 		if (xdst->route_mtu_cached != mtu) {
3823 			start_from = nr;
3824 			xdst->route_mtu_cached = mtu;
3825 		}
3826 
3827 		dst = xfrm_dst_child(dst);
3828 	} while (dst->xfrm);
3829 
3830 	if (likely(!start_from))
3831 		return 1;
3832 
3833 	xdst = bundle[start_from - 1];
3834 	mtu = xdst->child_mtu_cached;
3835 	while (start_from--) {
3836 		dst = &xdst->u.dst;
3837 
3838 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
3839 		if (mtu > xdst->route_mtu_cached)
3840 			mtu = xdst->route_mtu_cached;
3841 		dst_metric_set(dst, RTAX_MTU, mtu);
3842 		if (!start_from)
3843 			break;
3844 
3845 		xdst = bundle[start_from - 1];
3846 		xdst->child_mtu_cached = mtu;
3847 	}
3848 
3849 	return 1;
3850 }
3851 
3852 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3853 {
3854 	return dst_metric_advmss(xfrm_dst_path(dst));
3855 }
3856 
3857 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3858 {
3859 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3860 
3861 	return mtu ? : dst_mtu(xfrm_dst_path(dst));
3862 }
3863 
3864 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3865 					const void *daddr)
3866 {
3867 	while (dst->xfrm) {
3868 		const struct xfrm_state *xfrm = dst->xfrm;
3869 
3870 		dst = xfrm_dst_child(dst);
3871 
3872 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3873 			continue;
3874 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3875 			daddr = xfrm->coaddr;
3876 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3877 			daddr = &xfrm->id.daddr;
3878 	}
3879 	return daddr;
3880 }
3881 
3882 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3883 					   struct sk_buff *skb,
3884 					   const void *daddr)
3885 {
3886 	const struct dst_entry *path = xfrm_dst_path(dst);
3887 
3888 	if (!skb)
3889 		daddr = xfrm_get_dst_nexthop(dst, daddr);
3890 	return path->ops->neigh_lookup(path, skb, daddr);
3891 }
3892 
3893 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3894 {
3895 	const struct dst_entry *path = xfrm_dst_path(dst);
3896 
3897 	daddr = xfrm_get_dst_nexthop(dst, daddr);
3898 	path->ops->confirm_neigh(path, daddr);
3899 }
3900 
3901 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3902 {
3903 	int err = 0;
3904 
3905 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3906 		return -EAFNOSUPPORT;
3907 
3908 	spin_lock(&xfrm_policy_afinfo_lock);
3909 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
3910 		err = -EEXIST;
3911 	else {
3912 		struct dst_ops *dst_ops = afinfo->dst_ops;
3913 		if (likely(dst_ops->kmem_cachep == NULL))
3914 			dst_ops->kmem_cachep = xfrm_dst_cache;
3915 		if (likely(dst_ops->check == NULL))
3916 			dst_ops->check = xfrm_dst_check;
3917 		if (likely(dst_ops->default_advmss == NULL))
3918 			dst_ops->default_advmss = xfrm_default_advmss;
3919 		if (likely(dst_ops->mtu == NULL))
3920 			dst_ops->mtu = xfrm_mtu;
3921 		if (likely(dst_ops->negative_advice == NULL))
3922 			dst_ops->negative_advice = xfrm_negative_advice;
3923 		if (likely(dst_ops->link_failure == NULL))
3924 			dst_ops->link_failure = xfrm_link_failure;
3925 		if (likely(dst_ops->neigh_lookup == NULL))
3926 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
3927 		if (likely(!dst_ops->confirm_neigh))
3928 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
3929 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3930 	}
3931 	spin_unlock(&xfrm_policy_afinfo_lock);
3932 
3933 	return err;
3934 }
3935 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3936 
3937 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3938 {
3939 	struct dst_ops *dst_ops = afinfo->dst_ops;
3940 	int i;
3941 
3942 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3943 		if (xfrm_policy_afinfo[i] != afinfo)
3944 			continue;
3945 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3946 		break;
3947 	}
3948 
3949 	synchronize_rcu();
3950 
3951 	dst_ops->kmem_cachep = NULL;
3952 	dst_ops->check = NULL;
3953 	dst_ops->negative_advice = NULL;
3954 	dst_ops->link_failure = NULL;
3955 }
3956 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3957 
3958 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3959 {
3960 	spin_lock(&xfrm_if_cb_lock);
3961 	rcu_assign_pointer(xfrm_if_cb, ifcb);
3962 	spin_unlock(&xfrm_if_cb_lock);
3963 }
3964 EXPORT_SYMBOL(xfrm_if_register_cb);
3965 
3966 void xfrm_if_unregister_cb(void)
3967 {
3968 	RCU_INIT_POINTER(xfrm_if_cb, NULL);
3969 	synchronize_rcu();
3970 }
3971 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3972 
3973 #ifdef CONFIG_XFRM_STATISTICS
3974 static int __net_init xfrm_statistics_init(struct net *net)
3975 {
3976 	int rv;
3977 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3978 	if (!net->mib.xfrm_statistics)
3979 		return -ENOMEM;
3980 	rv = xfrm_proc_init(net);
3981 	if (rv < 0)
3982 		free_percpu(net->mib.xfrm_statistics);
3983 	return rv;
3984 }
3985 
3986 static void xfrm_statistics_fini(struct net *net)
3987 {
3988 	xfrm_proc_fini(net);
3989 	free_percpu(net->mib.xfrm_statistics);
3990 }
3991 #else
3992 static int __net_init xfrm_statistics_init(struct net *net)
3993 {
3994 	return 0;
3995 }
3996 
3997 static void xfrm_statistics_fini(struct net *net)
3998 {
3999 }
4000 #endif
4001 
4002 static int __net_init xfrm_policy_init(struct net *net)
4003 {
4004 	unsigned int hmask, sz;
4005 	int dir, err;
4006 
4007 	if (net_eq(net, &init_net)) {
4008 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4009 					   sizeof(struct xfrm_dst),
4010 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4011 					   NULL);
4012 		err = rhashtable_init(&xfrm_policy_inexact_table,
4013 				      &xfrm_pol_inexact_params);
4014 		BUG_ON(err);
4015 	}
4016 
4017 	hmask = 8 - 1;
4018 	sz = (hmask+1) * sizeof(struct hlist_head);
4019 
4020 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4021 	if (!net->xfrm.policy_byidx)
4022 		goto out_byidx;
4023 	net->xfrm.policy_idx_hmask = hmask;
4024 
4025 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4026 		struct xfrm_policy_hash *htab;
4027 
4028 		net->xfrm.policy_count[dir] = 0;
4029 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4030 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4031 
4032 		htab = &net->xfrm.policy_bydst[dir];
4033 		htab->table = xfrm_hash_alloc(sz);
4034 		if (!htab->table)
4035 			goto out_bydst;
4036 		htab->hmask = hmask;
4037 		htab->dbits4 = 32;
4038 		htab->sbits4 = 32;
4039 		htab->dbits6 = 128;
4040 		htab->sbits6 = 128;
4041 	}
4042 	net->xfrm.policy_hthresh.lbits4 = 32;
4043 	net->xfrm.policy_hthresh.rbits4 = 32;
4044 	net->xfrm.policy_hthresh.lbits6 = 128;
4045 	net->xfrm.policy_hthresh.rbits6 = 128;
4046 
4047 	seqlock_init(&net->xfrm.policy_hthresh.lock);
4048 
4049 	INIT_LIST_HEAD(&net->xfrm.policy_all);
4050 	INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4051 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4052 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4053 	return 0;
4054 
4055 out_bydst:
4056 	for (dir--; dir >= 0; dir--) {
4057 		struct xfrm_policy_hash *htab;
4058 
4059 		htab = &net->xfrm.policy_bydst[dir];
4060 		xfrm_hash_free(htab->table, sz);
4061 	}
4062 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4063 out_byidx:
4064 	return -ENOMEM;
4065 }
4066 
4067 static void xfrm_policy_fini(struct net *net)
4068 {
4069 	struct xfrm_pol_inexact_bin *b, *t;
4070 	unsigned int sz;
4071 	int dir;
4072 
4073 	flush_work(&net->xfrm.policy_hash_work);
4074 #ifdef CONFIG_XFRM_SUB_POLICY
4075 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4076 #endif
4077 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4078 
4079 	WARN_ON(!list_empty(&net->xfrm.policy_all));
4080 
4081 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4082 		struct xfrm_policy_hash *htab;
4083 
4084 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4085 
4086 		htab = &net->xfrm.policy_bydst[dir];
4087 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4088 		WARN_ON(!hlist_empty(htab->table));
4089 		xfrm_hash_free(htab->table, sz);
4090 	}
4091 
4092 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4093 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4094 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4095 
4096 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4097 	list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4098 		__xfrm_policy_inexact_prune_bin(b, true);
4099 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4100 }
4101 
4102 static int __net_init xfrm_net_init(struct net *net)
4103 {
4104 	int rv;
4105 
4106 	/* Initialize the per-net locks here */
4107 	spin_lock_init(&net->xfrm.xfrm_state_lock);
4108 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
4109 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
4110 
4111 	rv = xfrm_statistics_init(net);
4112 	if (rv < 0)
4113 		goto out_statistics;
4114 	rv = xfrm_state_init(net);
4115 	if (rv < 0)
4116 		goto out_state;
4117 	rv = xfrm_policy_init(net);
4118 	if (rv < 0)
4119 		goto out_policy;
4120 	rv = xfrm_sysctl_init(net);
4121 	if (rv < 0)
4122 		goto out_sysctl;
4123 
4124 	return 0;
4125 
4126 out_sysctl:
4127 	xfrm_policy_fini(net);
4128 out_policy:
4129 	xfrm_state_fini(net);
4130 out_state:
4131 	xfrm_statistics_fini(net);
4132 out_statistics:
4133 	return rv;
4134 }
4135 
4136 static void __net_exit xfrm_net_exit(struct net *net)
4137 {
4138 	xfrm_sysctl_fini(net);
4139 	xfrm_policy_fini(net);
4140 	xfrm_state_fini(net);
4141 	xfrm_statistics_fini(net);
4142 }
4143 
4144 static struct pernet_operations __net_initdata xfrm_net_ops = {
4145 	.init = xfrm_net_init,
4146 	.exit = xfrm_net_exit,
4147 };
4148 
4149 void __init xfrm_init(void)
4150 {
4151 	register_pernet_subsys(&xfrm_net_ops);
4152 	xfrm_dev_init();
4153 	seqcount_init(&xfrm_policy_hash_generation);
4154 	xfrm_input_init();
4155 
4156 	RCU_INIT_POINTER(xfrm_if_cb, NULL);
4157 	synchronize_rcu();
4158 }
4159 
4160 #ifdef CONFIG_AUDITSYSCALL
4161 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4162 					 struct audit_buffer *audit_buf)
4163 {
4164 	struct xfrm_sec_ctx *ctx = xp->security;
4165 	struct xfrm_selector *sel = &xp->selector;
4166 
4167 	if (ctx)
4168 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4169 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4170 
4171 	switch (sel->family) {
4172 	case AF_INET:
4173 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4174 		if (sel->prefixlen_s != 32)
4175 			audit_log_format(audit_buf, " src_prefixlen=%d",
4176 					 sel->prefixlen_s);
4177 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4178 		if (sel->prefixlen_d != 32)
4179 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4180 					 sel->prefixlen_d);
4181 		break;
4182 	case AF_INET6:
4183 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4184 		if (sel->prefixlen_s != 128)
4185 			audit_log_format(audit_buf, " src_prefixlen=%d",
4186 					 sel->prefixlen_s);
4187 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4188 		if (sel->prefixlen_d != 128)
4189 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4190 					 sel->prefixlen_d);
4191 		break;
4192 	}
4193 }
4194 
4195 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4196 {
4197 	struct audit_buffer *audit_buf;
4198 
4199 	audit_buf = xfrm_audit_start("SPD-add");
4200 	if (audit_buf == NULL)
4201 		return;
4202 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4203 	audit_log_format(audit_buf, " res=%u", result);
4204 	xfrm_audit_common_policyinfo(xp, audit_buf);
4205 	audit_log_end(audit_buf);
4206 }
4207 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4208 
4209 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4210 			      bool task_valid)
4211 {
4212 	struct audit_buffer *audit_buf;
4213 
4214 	audit_buf = xfrm_audit_start("SPD-delete");
4215 	if (audit_buf == NULL)
4216 		return;
4217 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4218 	audit_log_format(audit_buf, " res=%u", result);
4219 	xfrm_audit_common_policyinfo(xp, audit_buf);
4220 	audit_log_end(audit_buf);
4221 }
4222 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4223 #endif
4224 
4225 #ifdef CONFIG_XFRM_MIGRATE
4226 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4227 					const struct xfrm_selector *sel_tgt)
4228 {
4229 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4230 		if (sel_tgt->family == sel_cmp->family &&
4231 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4232 				    sel_cmp->family) &&
4233 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4234 				    sel_cmp->family) &&
4235 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4236 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4237 			return true;
4238 		}
4239 	} else {
4240 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4241 			return true;
4242 		}
4243 	}
4244 	return false;
4245 }
4246 
4247 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4248 						    u8 dir, u8 type, struct net *net)
4249 {
4250 	struct xfrm_policy *pol, *ret = NULL;
4251 	struct hlist_head *chain;
4252 	u32 priority = ~0U;
4253 
4254 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4255 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4256 	hlist_for_each_entry(pol, chain, bydst) {
4257 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4258 		    pol->type == type) {
4259 			ret = pol;
4260 			priority = ret->priority;
4261 			break;
4262 		}
4263 	}
4264 	chain = &net->xfrm.policy_inexact[dir];
4265 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4266 		if ((pol->priority >= priority) && ret)
4267 			break;
4268 
4269 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4270 		    pol->type == type) {
4271 			ret = pol;
4272 			break;
4273 		}
4274 	}
4275 
4276 	xfrm_pol_hold(ret);
4277 
4278 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4279 
4280 	return ret;
4281 }
4282 
4283 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4284 {
4285 	int match = 0;
4286 
4287 	if (t->mode == m->mode && t->id.proto == m->proto &&
4288 	    (m->reqid == 0 || t->reqid == m->reqid)) {
4289 		switch (t->mode) {
4290 		case XFRM_MODE_TUNNEL:
4291 		case XFRM_MODE_BEET:
4292 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4293 					    m->old_family) &&
4294 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
4295 					    m->old_family)) {
4296 				match = 1;
4297 			}
4298 			break;
4299 		case XFRM_MODE_TRANSPORT:
4300 			/* in case of transport mode, template does not store
4301 			   any IP addresses, hence we just compare mode and
4302 			   protocol */
4303 			match = 1;
4304 			break;
4305 		default:
4306 			break;
4307 		}
4308 	}
4309 	return match;
4310 }
4311 
4312 /* update endpoint address(es) of template(s) */
4313 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4314 			       struct xfrm_migrate *m, int num_migrate)
4315 {
4316 	struct xfrm_migrate *mp;
4317 	int i, j, n = 0;
4318 
4319 	write_lock_bh(&pol->lock);
4320 	if (unlikely(pol->walk.dead)) {
4321 		/* target policy has been deleted */
4322 		write_unlock_bh(&pol->lock);
4323 		return -ENOENT;
4324 	}
4325 
4326 	for (i = 0; i < pol->xfrm_nr; i++) {
4327 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4328 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4329 				continue;
4330 			n++;
4331 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4332 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4333 				continue;
4334 			/* update endpoints */
4335 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4336 			       sizeof(pol->xfrm_vec[i].id.daddr));
4337 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4338 			       sizeof(pol->xfrm_vec[i].saddr));
4339 			pol->xfrm_vec[i].encap_family = mp->new_family;
4340 			/* flush bundles */
4341 			atomic_inc(&pol->genid);
4342 		}
4343 	}
4344 
4345 	write_unlock_bh(&pol->lock);
4346 
4347 	if (!n)
4348 		return -ENODATA;
4349 
4350 	return 0;
4351 }
4352 
4353 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4354 {
4355 	int i, j;
4356 
4357 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4358 		return -EINVAL;
4359 
4360 	for (i = 0; i < num_migrate; i++) {
4361 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4362 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4363 			return -EINVAL;
4364 
4365 		/* check if there is any duplicated entry */
4366 		for (j = i + 1; j < num_migrate; j++) {
4367 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4368 				    sizeof(m[i].old_daddr)) &&
4369 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4370 				    sizeof(m[i].old_saddr)) &&
4371 			    m[i].proto == m[j].proto &&
4372 			    m[i].mode == m[j].mode &&
4373 			    m[i].reqid == m[j].reqid &&
4374 			    m[i].old_family == m[j].old_family)
4375 				return -EINVAL;
4376 		}
4377 	}
4378 
4379 	return 0;
4380 }
4381 
4382 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4383 		 struct xfrm_migrate *m, int num_migrate,
4384 		 struct xfrm_kmaddress *k, struct net *net,
4385 		 struct xfrm_encap_tmpl *encap)
4386 {
4387 	int i, err, nx_cur = 0, nx_new = 0;
4388 	struct xfrm_policy *pol = NULL;
4389 	struct xfrm_state *x, *xc;
4390 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4391 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4392 	struct xfrm_migrate *mp;
4393 
4394 	/* Stage 0 - sanity checks */
4395 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4396 		goto out;
4397 
4398 	if (dir >= XFRM_POLICY_MAX) {
4399 		err = -EINVAL;
4400 		goto out;
4401 	}
4402 
4403 	/* Stage 1 - find policy */
4404 	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4405 		err = -ENOENT;
4406 		goto out;
4407 	}
4408 
4409 	/* Stage 2 - find and update state(s) */
4410 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4411 		if ((x = xfrm_migrate_state_find(mp, net))) {
4412 			x_cur[nx_cur] = x;
4413 			nx_cur++;
4414 			xc = xfrm_state_migrate(x, mp, encap);
4415 			if (xc) {
4416 				x_new[nx_new] = xc;
4417 				nx_new++;
4418 			} else {
4419 				err = -ENODATA;
4420 				goto restore_state;
4421 			}
4422 		}
4423 	}
4424 
4425 	/* Stage 3 - update policy */
4426 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4427 		goto restore_state;
4428 
4429 	/* Stage 4 - delete old state(s) */
4430 	if (nx_cur) {
4431 		xfrm_states_put(x_cur, nx_cur);
4432 		xfrm_states_delete(x_cur, nx_cur);
4433 	}
4434 
4435 	/* Stage 5 - announce */
4436 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
4437 
4438 	xfrm_pol_put(pol);
4439 
4440 	return 0;
4441 out:
4442 	return err;
4443 
4444 restore_state:
4445 	if (pol)
4446 		xfrm_pol_put(pol);
4447 	if (nx_cur)
4448 		xfrm_states_put(x_cur, nx_cur);
4449 	if (nx_new)
4450 		xfrm_states_delete(x_new, nx_new);
4451 
4452 	return err;
4453 }
4454 EXPORT_SYMBOL(xfrm_migrate);
4455 #endif
4456