xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 901181b7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * xfrm_policy.c
4  *
5  * Changes:
6  *	Mitsuru KANDA @USAGI
7  * 	Kazunori MIYAZAWA @USAGI
8  * 	Kunihiro Ishiguro <kunihiro@ipinfusion.com>
9  * 		IPv6 support
10  * 	Kazunori MIYAZAWA @USAGI
11  * 	YOSHIFUJI Hideaki
12  * 		Split up af-specific portion
13  *	Derek Atkins <derek@ihtfp.com>		Add the post_input processor
14  *
15  */
16 
17 #include <linux/err.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/workqueue.h>
23 #include <linux/notifier.h>
24 #include <linux/netdevice.h>
25 #include <linux/netfilter.h>
26 #include <linux/module.h>
27 #include <linux/cache.h>
28 #include <linux/cpu.h>
29 #include <linux/audit.h>
30 #include <linux/rhashtable.h>
31 #include <linux/if_tunnel.h>
32 #include <net/dst.h>
33 #include <net/flow.h>
34 #include <net/xfrm.h>
35 #include <net/ip.h>
36 #if IS_ENABLED(CONFIG_IPV6_MIP6)
37 #include <net/mip6.h>
38 #endif
39 #ifdef CONFIG_XFRM_STATISTICS
40 #include <net/snmp.h>
41 #endif
42 #ifdef CONFIG_XFRM_ESPINTCP
43 #include <net/espintcp.h>
44 #endif
45 
46 #include "xfrm_hash.h"
47 
48 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
49 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
50 #define XFRM_MAX_QUEUE_LEN	100
51 
52 struct xfrm_flo {
53 	struct dst_entry *dst_orig;
54 	u8 flags;
55 };
56 
57 /* prefixes smaller than this are stored in lists, not trees. */
58 #define INEXACT_PREFIXLEN_IPV4	16
59 #define INEXACT_PREFIXLEN_IPV6	48
60 
61 struct xfrm_pol_inexact_node {
62 	struct rb_node node;
63 	union {
64 		xfrm_address_t addr;
65 		struct rcu_head rcu;
66 	};
67 	u8 prefixlen;
68 
69 	struct rb_root root;
70 
71 	/* the policies matching this node, can be empty list */
72 	struct hlist_head hhead;
73 };
74 
75 /* xfrm inexact policy search tree:
76  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
77  *  |
78  * +---- root_d: sorted by daddr:prefix
79  * |                 |
80  * |        xfrm_pol_inexact_node
81  * |                 |
82  * |                 +- root: sorted by saddr/prefix
83  * |                 |              |
84  * |                 |         xfrm_pol_inexact_node
85  * |                 |              |
86  * |                 |              + root: unused
87  * |                 |              |
88  * |                 |              + hhead: saddr:daddr policies
89  * |                 |
90  * |                 +- coarse policies and all any:daddr policies
91  * |
92  * +---- root_s: sorted by saddr:prefix
93  * |                 |
94  * |        xfrm_pol_inexact_node
95  * |                 |
96  * |                 + root: unused
97  * |                 |
98  * |                 + hhead: saddr:any policies
99  * |
100  * +---- coarse policies and all any:any policies
101  *
102  * Lookups return four candidate lists:
103  * 1. any:any list from top-level xfrm_pol_inexact_bin
104  * 2. any:daddr list from daddr tree
105  * 3. saddr:daddr list from 2nd level daddr tree
106  * 4. saddr:any list from saddr tree
107  *
108  * This result set then needs to be searched for the policy with
109  * the lowest priority.  If two results have same prio, youngest one wins.
110  */
111 
112 struct xfrm_pol_inexact_key {
113 	possible_net_t net;
114 	u32 if_id;
115 	u16 family;
116 	u8 dir, type;
117 };
118 
119 struct xfrm_pol_inexact_bin {
120 	struct xfrm_pol_inexact_key k;
121 	struct rhash_head head;
122 	/* list containing '*:*' policies */
123 	struct hlist_head hhead;
124 
125 	seqcount_spinlock_t count;
126 	/* tree sorted by daddr/prefix */
127 	struct rb_root root_d;
128 
129 	/* tree sorted by saddr/prefix */
130 	struct rb_root root_s;
131 
132 	/* slow path below */
133 	struct list_head inexact_bins;
134 	struct rcu_head rcu;
135 };
136 
137 enum xfrm_pol_inexact_candidate_type {
138 	XFRM_POL_CAND_BOTH,
139 	XFRM_POL_CAND_SADDR,
140 	XFRM_POL_CAND_DADDR,
141 	XFRM_POL_CAND_ANY,
142 
143 	XFRM_POL_CAND_MAX,
144 };
145 
146 struct xfrm_pol_inexact_candidates {
147 	struct hlist_head *res[XFRM_POL_CAND_MAX];
148 };
149 
150 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
151 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
152 
153 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
154 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
155 						__read_mostly;
156 
157 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
158 
159 static struct rhashtable xfrm_policy_inexact_table;
160 static const struct rhashtable_params xfrm_pol_inexact_params;
161 
162 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
163 static int stale_bundle(struct dst_entry *dst);
164 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
165 static void xfrm_policy_queue_process(struct timer_list *t);
166 
167 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
168 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
169 						int dir);
170 
171 static struct xfrm_pol_inexact_bin *
172 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
173 			   u32 if_id);
174 
175 static struct xfrm_pol_inexact_bin *
176 xfrm_policy_inexact_lookup_rcu(struct net *net,
177 			       u8 type, u16 family, u8 dir, u32 if_id);
178 static struct xfrm_policy *
179 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
180 			bool excl);
181 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
182 					    struct xfrm_policy *policy);
183 
184 static bool
185 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
186 				    struct xfrm_pol_inexact_bin *b,
187 				    const xfrm_address_t *saddr,
188 				    const xfrm_address_t *daddr);
189 
190 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
191 {
192 	return refcount_inc_not_zero(&policy->refcnt);
193 }
194 
195 static inline bool
196 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
197 {
198 	const struct flowi4 *fl4 = &fl->u.ip4;
199 
200 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
201 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
202 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
203 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
204 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
205 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
206 }
207 
208 static inline bool
209 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
210 {
211 	const struct flowi6 *fl6 = &fl->u.ip6;
212 
213 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
214 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
215 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
216 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
217 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
218 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
219 }
220 
221 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
222 			 unsigned short family)
223 {
224 	switch (family) {
225 	case AF_INET:
226 		return __xfrm4_selector_match(sel, fl);
227 	case AF_INET6:
228 		return __xfrm6_selector_match(sel, fl);
229 	}
230 	return false;
231 }
232 
233 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
234 {
235 	const struct xfrm_policy_afinfo *afinfo;
236 
237 	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
238 		return NULL;
239 	rcu_read_lock();
240 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
241 	if (unlikely(!afinfo))
242 		rcu_read_unlock();
243 	return afinfo;
244 }
245 
246 /* Called with rcu_read_lock(). */
247 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
248 {
249 	return rcu_dereference(xfrm_if_cb);
250 }
251 
252 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
253 				    const xfrm_address_t *saddr,
254 				    const xfrm_address_t *daddr,
255 				    int family, u32 mark)
256 {
257 	const struct xfrm_policy_afinfo *afinfo;
258 	struct dst_entry *dst;
259 
260 	afinfo = xfrm_policy_get_afinfo(family);
261 	if (unlikely(afinfo == NULL))
262 		return ERR_PTR(-EAFNOSUPPORT);
263 
264 	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
265 
266 	rcu_read_unlock();
267 
268 	return dst;
269 }
270 EXPORT_SYMBOL(__xfrm_dst_lookup);
271 
272 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
273 						int tos, int oif,
274 						xfrm_address_t *prev_saddr,
275 						xfrm_address_t *prev_daddr,
276 						int family, u32 mark)
277 {
278 	struct net *net = xs_net(x);
279 	xfrm_address_t *saddr = &x->props.saddr;
280 	xfrm_address_t *daddr = &x->id.daddr;
281 	struct dst_entry *dst;
282 
283 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
284 		saddr = x->coaddr;
285 		daddr = prev_daddr;
286 	}
287 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
288 		saddr = prev_saddr;
289 		daddr = x->coaddr;
290 	}
291 
292 	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
293 
294 	if (!IS_ERR(dst)) {
295 		if (prev_saddr != saddr)
296 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
297 		if (prev_daddr != daddr)
298 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
299 	}
300 
301 	return dst;
302 }
303 
304 static inline unsigned long make_jiffies(long secs)
305 {
306 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
307 		return MAX_SCHEDULE_TIMEOUT-1;
308 	else
309 		return secs*HZ;
310 }
311 
312 static void xfrm_policy_timer(struct timer_list *t)
313 {
314 	struct xfrm_policy *xp = from_timer(xp, t, timer);
315 	time64_t now = ktime_get_real_seconds();
316 	time64_t next = TIME64_MAX;
317 	int warn = 0;
318 	int dir;
319 
320 	read_lock(&xp->lock);
321 
322 	if (unlikely(xp->walk.dead))
323 		goto out;
324 
325 	dir = xfrm_policy_id2dir(xp->index);
326 
327 	if (xp->lft.hard_add_expires_seconds) {
328 		time64_t tmo = xp->lft.hard_add_expires_seconds +
329 			xp->curlft.add_time - now;
330 		if (tmo <= 0)
331 			goto expired;
332 		if (tmo < next)
333 			next = tmo;
334 	}
335 	if (xp->lft.hard_use_expires_seconds) {
336 		time64_t tmo = xp->lft.hard_use_expires_seconds +
337 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
338 		if (tmo <= 0)
339 			goto expired;
340 		if (tmo < next)
341 			next = tmo;
342 	}
343 	if (xp->lft.soft_add_expires_seconds) {
344 		time64_t tmo = xp->lft.soft_add_expires_seconds +
345 			xp->curlft.add_time - now;
346 		if (tmo <= 0) {
347 			warn = 1;
348 			tmo = XFRM_KM_TIMEOUT;
349 		}
350 		if (tmo < next)
351 			next = tmo;
352 	}
353 	if (xp->lft.soft_use_expires_seconds) {
354 		time64_t tmo = xp->lft.soft_use_expires_seconds +
355 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
356 		if (tmo <= 0) {
357 			warn = 1;
358 			tmo = XFRM_KM_TIMEOUT;
359 		}
360 		if (tmo < next)
361 			next = tmo;
362 	}
363 
364 	if (warn)
365 		km_policy_expired(xp, dir, 0, 0);
366 	if (next != TIME64_MAX &&
367 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
368 		xfrm_pol_hold(xp);
369 
370 out:
371 	read_unlock(&xp->lock);
372 	xfrm_pol_put(xp);
373 	return;
374 
375 expired:
376 	read_unlock(&xp->lock);
377 	if (!xfrm_policy_delete(xp, dir))
378 		km_policy_expired(xp, dir, 1, 0);
379 	xfrm_pol_put(xp);
380 }
381 
382 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
383  * SPD calls.
384  */
385 
386 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
387 {
388 	struct xfrm_policy *policy;
389 
390 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
391 
392 	if (policy) {
393 		write_pnet(&policy->xp_net, net);
394 		INIT_LIST_HEAD(&policy->walk.all);
395 		INIT_HLIST_NODE(&policy->bydst_inexact_list);
396 		INIT_HLIST_NODE(&policy->bydst);
397 		INIT_HLIST_NODE(&policy->byidx);
398 		rwlock_init(&policy->lock);
399 		refcount_set(&policy->refcnt, 1);
400 		skb_queue_head_init(&policy->polq.hold_queue);
401 		timer_setup(&policy->timer, xfrm_policy_timer, 0);
402 		timer_setup(&policy->polq.hold_timer,
403 			    xfrm_policy_queue_process, 0);
404 	}
405 	return policy;
406 }
407 EXPORT_SYMBOL(xfrm_policy_alloc);
408 
409 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
410 {
411 	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
412 
413 	security_xfrm_policy_free(policy->security);
414 	kfree(policy);
415 }
416 
417 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
418 
419 void xfrm_policy_destroy(struct xfrm_policy *policy)
420 {
421 	BUG_ON(!policy->walk.dead);
422 
423 	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
424 		BUG();
425 
426 	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
427 }
428 EXPORT_SYMBOL(xfrm_policy_destroy);
429 
430 /* Rule must be locked. Release descendant resources, announce
431  * entry dead. The rule must be unlinked from lists to the moment.
432  */
433 
434 static void xfrm_policy_kill(struct xfrm_policy *policy)
435 {
436 	write_lock_bh(&policy->lock);
437 	policy->walk.dead = 1;
438 	write_unlock_bh(&policy->lock);
439 
440 	atomic_inc(&policy->genid);
441 
442 	if (del_timer(&policy->polq.hold_timer))
443 		xfrm_pol_put(policy);
444 	skb_queue_purge(&policy->polq.hold_queue);
445 
446 	if (del_timer(&policy->timer))
447 		xfrm_pol_put(policy);
448 
449 	xfrm_pol_put(policy);
450 }
451 
452 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
453 
454 static inline unsigned int idx_hash(struct net *net, u32 index)
455 {
456 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
457 }
458 
459 /* calculate policy hash thresholds */
460 static void __get_hash_thresh(struct net *net,
461 			      unsigned short family, int dir,
462 			      u8 *dbits, u8 *sbits)
463 {
464 	switch (family) {
465 	case AF_INET:
466 		*dbits = net->xfrm.policy_bydst[dir].dbits4;
467 		*sbits = net->xfrm.policy_bydst[dir].sbits4;
468 		break;
469 
470 	case AF_INET6:
471 		*dbits = net->xfrm.policy_bydst[dir].dbits6;
472 		*sbits = net->xfrm.policy_bydst[dir].sbits6;
473 		break;
474 
475 	default:
476 		*dbits = 0;
477 		*sbits = 0;
478 	}
479 }
480 
481 static struct hlist_head *policy_hash_bysel(struct net *net,
482 					    const struct xfrm_selector *sel,
483 					    unsigned short family, int dir)
484 {
485 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
486 	unsigned int hash;
487 	u8 dbits;
488 	u8 sbits;
489 
490 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
491 	hash = __sel_hash(sel, family, hmask, dbits, sbits);
492 
493 	if (hash == hmask + 1)
494 		return NULL;
495 
496 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
497 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
498 }
499 
500 static struct hlist_head *policy_hash_direct(struct net *net,
501 					     const xfrm_address_t *daddr,
502 					     const xfrm_address_t *saddr,
503 					     unsigned short family, int dir)
504 {
505 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
506 	unsigned int hash;
507 	u8 dbits;
508 	u8 sbits;
509 
510 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
511 	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
512 
513 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
514 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
515 }
516 
517 static void xfrm_dst_hash_transfer(struct net *net,
518 				   struct hlist_head *list,
519 				   struct hlist_head *ndsttable,
520 				   unsigned int nhashmask,
521 				   int dir)
522 {
523 	struct hlist_node *tmp, *entry0 = NULL;
524 	struct xfrm_policy *pol;
525 	unsigned int h0 = 0;
526 	u8 dbits;
527 	u8 sbits;
528 
529 redo:
530 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
531 		unsigned int h;
532 
533 		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
534 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
535 				pol->family, nhashmask, dbits, sbits);
536 		if (!entry0) {
537 			hlist_del_rcu(&pol->bydst);
538 			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
539 			h0 = h;
540 		} else {
541 			if (h != h0)
542 				continue;
543 			hlist_del_rcu(&pol->bydst);
544 			hlist_add_behind_rcu(&pol->bydst, entry0);
545 		}
546 		entry0 = &pol->bydst;
547 	}
548 	if (!hlist_empty(list)) {
549 		entry0 = NULL;
550 		goto redo;
551 	}
552 }
553 
554 static void xfrm_idx_hash_transfer(struct hlist_head *list,
555 				   struct hlist_head *nidxtable,
556 				   unsigned int nhashmask)
557 {
558 	struct hlist_node *tmp;
559 	struct xfrm_policy *pol;
560 
561 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
562 		unsigned int h;
563 
564 		h = __idx_hash(pol->index, nhashmask);
565 		hlist_add_head(&pol->byidx, nidxtable+h);
566 	}
567 }
568 
569 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
570 {
571 	return ((old_hmask + 1) << 1) - 1;
572 }
573 
574 static void xfrm_bydst_resize(struct net *net, int dir)
575 {
576 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
577 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
578 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
579 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
580 	struct hlist_head *odst;
581 	int i;
582 
583 	if (!ndst)
584 		return;
585 
586 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
587 	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
588 
589 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
590 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
591 
592 	for (i = hmask; i >= 0; i--)
593 		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
594 
595 	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
596 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
597 
598 	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
599 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
600 
601 	synchronize_rcu();
602 
603 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
604 }
605 
606 static void xfrm_byidx_resize(struct net *net, int total)
607 {
608 	unsigned int hmask = net->xfrm.policy_idx_hmask;
609 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
610 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
611 	struct hlist_head *oidx = net->xfrm.policy_byidx;
612 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
613 	int i;
614 
615 	if (!nidx)
616 		return;
617 
618 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
619 
620 	for (i = hmask; i >= 0; i--)
621 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
622 
623 	net->xfrm.policy_byidx = nidx;
624 	net->xfrm.policy_idx_hmask = nhashmask;
625 
626 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
627 
628 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
629 }
630 
631 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
632 {
633 	unsigned int cnt = net->xfrm.policy_count[dir];
634 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
635 
636 	if (total)
637 		*total += cnt;
638 
639 	if ((hmask + 1) < xfrm_policy_hashmax &&
640 	    cnt > hmask)
641 		return 1;
642 
643 	return 0;
644 }
645 
646 static inline int xfrm_byidx_should_resize(struct net *net, int total)
647 {
648 	unsigned int hmask = net->xfrm.policy_idx_hmask;
649 
650 	if ((hmask + 1) < xfrm_policy_hashmax &&
651 	    total > hmask)
652 		return 1;
653 
654 	return 0;
655 }
656 
657 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
658 {
659 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
660 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
661 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
662 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
663 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
664 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
665 	si->spdhcnt = net->xfrm.policy_idx_hmask;
666 	si->spdhmcnt = xfrm_policy_hashmax;
667 }
668 EXPORT_SYMBOL(xfrm_spd_getinfo);
669 
670 static DEFINE_MUTEX(hash_resize_mutex);
671 static void xfrm_hash_resize(struct work_struct *work)
672 {
673 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
674 	int dir, total;
675 
676 	mutex_lock(&hash_resize_mutex);
677 
678 	total = 0;
679 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
680 		if (xfrm_bydst_should_resize(net, dir, &total))
681 			xfrm_bydst_resize(net, dir);
682 	}
683 	if (xfrm_byidx_should_resize(net, total))
684 		xfrm_byidx_resize(net, total);
685 
686 	mutex_unlock(&hash_resize_mutex);
687 }
688 
689 /* Make sure *pol can be inserted into fastbin.
690  * Useful to check that later insert requests will be successful
691  * (provided xfrm_policy_lock is held throughout).
692  */
693 static struct xfrm_pol_inexact_bin *
694 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
695 {
696 	struct xfrm_pol_inexact_bin *bin, *prev;
697 	struct xfrm_pol_inexact_key k = {
698 		.family = pol->family,
699 		.type = pol->type,
700 		.dir = dir,
701 		.if_id = pol->if_id,
702 	};
703 	struct net *net = xp_net(pol);
704 
705 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
706 
707 	write_pnet(&k.net, net);
708 	bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
709 				     xfrm_pol_inexact_params);
710 	if (bin)
711 		return bin;
712 
713 	bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
714 	if (!bin)
715 		return NULL;
716 
717 	bin->k = k;
718 	INIT_HLIST_HEAD(&bin->hhead);
719 	bin->root_d = RB_ROOT;
720 	bin->root_s = RB_ROOT;
721 	seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
722 
723 	prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
724 						&bin->k, &bin->head,
725 						xfrm_pol_inexact_params);
726 	if (!prev) {
727 		list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
728 		return bin;
729 	}
730 
731 	kfree(bin);
732 
733 	return IS_ERR(prev) ? NULL : prev;
734 }
735 
736 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
737 					       int family, u8 prefixlen)
738 {
739 	if (xfrm_addr_any(addr, family))
740 		return true;
741 
742 	if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
743 		return true;
744 
745 	if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
746 		return true;
747 
748 	return false;
749 }
750 
751 static bool
752 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
753 {
754 	const xfrm_address_t *addr;
755 	bool saddr_any, daddr_any;
756 	u8 prefixlen;
757 
758 	addr = &policy->selector.saddr;
759 	prefixlen = policy->selector.prefixlen_s;
760 
761 	saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
762 						       policy->family,
763 						       prefixlen);
764 	addr = &policy->selector.daddr;
765 	prefixlen = policy->selector.prefixlen_d;
766 	daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
767 						       policy->family,
768 						       prefixlen);
769 	return saddr_any && daddr_any;
770 }
771 
772 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
773 				       const xfrm_address_t *addr, u8 prefixlen)
774 {
775 	node->addr = *addr;
776 	node->prefixlen = prefixlen;
777 }
778 
779 static struct xfrm_pol_inexact_node *
780 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
781 {
782 	struct xfrm_pol_inexact_node *node;
783 
784 	node = kzalloc(sizeof(*node), GFP_ATOMIC);
785 	if (node)
786 		xfrm_pol_inexact_node_init(node, addr, prefixlen);
787 
788 	return node;
789 }
790 
791 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
792 				  const xfrm_address_t *b,
793 				  u8 prefixlen, u16 family)
794 {
795 	u32 ma, mb, mask;
796 	unsigned int pdw, pbi;
797 	int delta = 0;
798 
799 	switch (family) {
800 	case AF_INET:
801 		if (prefixlen == 0)
802 			return 0;
803 		mask = ~0U << (32 - prefixlen);
804 		ma = ntohl(a->a4) & mask;
805 		mb = ntohl(b->a4) & mask;
806 		if (ma < mb)
807 			delta = -1;
808 		else if (ma > mb)
809 			delta = 1;
810 		break;
811 	case AF_INET6:
812 		pdw = prefixlen >> 5;
813 		pbi = prefixlen & 0x1f;
814 
815 		if (pdw) {
816 			delta = memcmp(a->a6, b->a6, pdw << 2);
817 			if (delta)
818 				return delta;
819 		}
820 		if (pbi) {
821 			mask = ~0U << (32 - pbi);
822 			ma = ntohl(a->a6[pdw]) & mask;
823 			mb = ntohl(b->a6[pdw]) & mask;
824 			if (ma < mb)
825 				delta = -1;
826 			else if (ma > mb)
827 				delta = 1;
828 		}
829 		break;
830 	default:
831 		break;
832 	}
833 
834 	return delta;
835 }
836 
837 static void xfrm_policy_inexact_list_reinsert(struct net *net,
838 					      struct xfrm_pol_inexact_node *n,
839 					      u16 family)
840 {
841 	unsigned int matched_s, matched_d;
842 	struct xfrm_policy *policy, *p;
843 
844 	matched_s = 0;
845 	matched_d = 0;
846 
847 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
848 		struct hlist_node *newpos = NULL;
849 		bool matches_s, matches_d;
850 
851 		if (!policy->bydst_reinsert)
852 			continue;
853 
854 		WARN_ON_ONCE(policy->family != family);
855 
856 		policy->bydst_reinsert = false;
857 		hlist_for_each_entry(p, &n->hhead, bydst) {
858 			if (policy->priority > p->priority)
859 				newpos = &p->bydst;
860 			else if (policy->priority == p->priority &&
861 				 policy->pos > p->pos)
862 				newpos = &p->bydst;
863 			else
864 				break;
865 		}
866 
867 		if (newpos)
868 			hlist_add_behind_rcu(&policy->bydst, newpos);
869 		else
870 			hlist_add_head_rcu(&policy->bydst, &n->hhead);
871 
872 		/* paranoia checks follow.
873 		 * Check that the reinserted policy matches at least
874 		 * saddr or daddr for current node prefix.
875 		 *
876 		 * Matching both is fine, matching saddr in one policy
877 		 * (but not daddr) and then matching only daddr in another
878 		 * is a bug.
879 		 */
880 		matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
881 						   &n->addr,
882 						   n->prefixlen,
883 						   family) == 0;
884 		matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
885 						   &n->addr,
886 						   n->prefixlen,
887 						   family) == 0;
888 		if (matches_s && matches_d)
889 			continue;
890 
891 		WARN_ON_ONCE(!matches_s && !matches_d);
892 		if (matches_s)
893 			matched_s++;
894 		if (matches_d)
895 			matched_d++;
896 		WARN_ON_ONCE(matched_s && matched_d);
897 	}
898 }
899 
900 static void xfrm_policy_inexact_node_reinsert(struct net *net,
901 					      struct xfrm_pol_inexact_node *n,
902 					      struct rb_root *new,
903 					      u16 family)
904 {
905 	struct xfrm_pol_inexact_node *node;
906 	struct rb_node **p, *parent;
907 
908 	/* we should not have another subtree here */
909 	WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
910 restart:
911 	parent = NULL;
912 	p = &new->rb_node;
913 	while (*p) {
914 		u8 prefixlen;
915 		int delta;
916 
917 		parent = *p;
918 		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
919 
920 		prefixlen = min(node->prefixlen, n->prefixlen);
921 
922 		delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
923 					       prefixlen, family);
924 		if (delta < 0) {
925 			p = &parent->rb_left;
926 		} else if (delta > 0) {
927 			p = &parent->rb_right;
928 		} else {
929 			bool same_prefixlen = node->prefixlen == n->prefixlen;
930 			struct xfrm_policy *tmp;
931 
932 			hlist_for_each_entry(tmp, &n->hhead, bydst) {
933 				tmp->bydst_reinsert = true;
934 				hlist_del_rcu(&tmp->bydst);
935 			}
936 
937 			node->prefixlen = prefixlen;
938 
939 			xfrm_policy_inexact_list_reinsert(net, node, family);
940 
941 			if (same_prefixlen) {
942 				kfree_rcu(n, rcu);
943 				return;
944 			}
945 
946 			rb_erase(*p, new);
947 			kfree_rcu(n, rcu);
948 			n = node;
949 			goto restart;
950 		}
951 	}
952 
953 	rb_link_node_rcu(&n->node, parent, p);
954 	rb_insert_color(&n->node, new);
955 }
956 
957 /* merge nodes v and n */
958 static void xfrm_policy_inexact_node_merge(struct net *net,
959 					   struct xfrm_pol_inexact_node *v,
960 					   struct xfrm_pol_inexact_node *n,
961 					   u16 family)
962 {
963 	struct xfrm_pol_inexact_node *node;
964 	struct xfrm_policy *tmp;
965 	struct rb_node *rnode;
966 
967 	/* To-be-merged node v has a subtree.
968 	 *
969 	 * Dismantle it and insert its nodes to n->root.
970 	 */
971 	while ((rnode = rb_first(&v->root)) != NULL) {
972 		node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
973 		rb_erase(&node->node, &v->root);
974 		xfrm_policy_inexact_node_reinsert(net, node, &n->root,
975 						  family);
976 	}
977 
978 	hlist_for_each_entry(tmp, &v->hhead, bydst) {
979 		tmp->bydst_reinsert = true;
980 		hlist_del_rcu(&tmp->bydst);
981 	}
982 
983 	xfrm_policy_inexact_list_reinsert(net, n, family);
984 }
985 
986 static struct xfrm_pol_inexact_node *
987 xfrm_policy_inexact_insert_node(struct net *net,
988 				struct rb_root *root,
989 				xfrm_address_t *addr,
990 				u16 family, u8 prefixlen, u8 dir)
991 {
992 	struct xfrm_pol_inexact_node *cached = NULL;
993 	struct rb_node **p, *parent = NULL;
994 	struct xfrm_pol_inexact_node *node;
995 
996 	p = &root->rb_node;
997 	while (*p) {
998 		int delta;
999 
1000 		parent = *p;
1001 		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1002 
1003 		delta = xfrm_policy_addr_delta(addr, &node->addr,
1004 					       node->prefixlen,
1005 					       family);
1006 		if (delta == 0 && prefixlen >= node->prefixlen) {
1007 			WARN_ON_ONCE(cached); /* ipsec policies got lost */
1008 			return node;
1009 		}
1010 
1011 		if (delta < 0)
1012 			p = &parent->rb_left;
1013 		else
1014 			p = &parent->rb_right;
1015 
1016 		if (prefixlen < node->prefixlen) {
1017 			delta = xfrm_policy_addr_delta(addr, &node->addr,
1018 						       prefixlen,
1019 						       family);
1020 			if (delta)
1021 				continue;
1022 
1023 			/* This node is a subnet of the new prefix. It needs
1024 			 * to be removed and re-inserted with the smaller
1025 			 * prefix and all nodes that are now also covered
1026 			 * by the reduced prefixlen.
1027 			 */
1028 			rb_erase(&node->node, root);
1029 
1030 			if (!cached) {
1031 				xfrm_pol_inexact_node_init(node, addr,
1032 							   prefixlen);
1033 				cached = node;
1034 			} else {
1035 				/* This node also falls within the new
1036 				 * prefixlen. Merge the to-be-reinserted
1037 				 * node and this one.
1038 				 */
1039 				xfrm_policy_inexact_node_merge(net, node,
1040 							       cached, family);
1041 				kfree_rcu(node, rcu);
1042 			}
1043 
1044 			/* restart */
1045 			p = &root->rb_node;
1046 			parent = NULL;
1047 		}
1048 	}
1049 
1050 	node = cached;
1051 	if (!node) {
1052 		node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1053 		if (!node)
1054 			return NULL;
1055 	}
1056 
1057 	rb_link_node_rcu(&node->node, parent, p);
1058 	rb_insert_color(&node->node, root);
1059 
1060 	return node;
1061 }
1062 
1063 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1064 {
1065 	struct xfrm_pol_inexact_node *node;
1066 	struct rb_node *rn = rb_first(r);
1067 
1068 	while (rn) {
1069 		node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1070 
1071 		xfrm_policy_inexact_gc_tree(&node->root, rm);
1072 		rn = rb_next(rn);
1073 
1074 		if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1075 			WARN_ON_ONCE(rm);
1076 			continue;
1077 		}
1078 
1079 		rb_erase(&node->node, r);
1080 		kfree_rcu(node, rcu);
1081 	}
1082 }
1083 
1084 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1085 {
1086 	write_seqcount_begin(&b->count);
1087 	xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1088 	xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1089 	write_seqcount_end(&b->count);
1090 
1091 	if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1092 	    !hlist_empty(&b->hhead)) {
1093 		WARN_ON_ONCE(net_exit);
1094 		return;
1095 	}
1096 
1097 	if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1098 				   xfrm_pol_inexact_params) == 0) {
1099 		list_del(&b->inexact_bins);
1100 		kfree_rcu(b, rcu);
1101 	}
1102 }
1103 
1104 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1105 {
1106 	struct net *net = read_pnet(&b->k.net);
1107 
1108 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1109 	__xfrm_policy_inexact_prune_bin(b, false);
1110 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1111 }
1112 
1113 static void __xfrm_policy_inexact_flush(struct net *net)
1114 {
1115 	struct xfrm_pol_inexact_bin *bin, *t;
1116 
1117 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1118 
1119 	list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1120 		__xfrm_policy_inexact_prune_bin(bin, false);
1121 }
1122 
1123 static struct hlist_head *
1124 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1125 				struct xfrm_policy *policy, u8 dir)
1126 {
1127 	struct xfrm_pol_inexact_node *n;
1128 	struct net *net;
1129 
1130 	net = xp_net(policy);
1131 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1132 
1133 	if (xfrm_policy_inexact_insert_use_any_list(policy))
1134 		return &bin->hhead;
1135 
1136 	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1137 					       policy->family,
1138 					       policy->selector.prefixlen_d)) {
1139 		write_seqcount_begin(&bin->count);
1140 		n = xfrm_policy_inexact_insert_node(net,
1141 						    &bin->root_s,
1142 						    &policy->selector.saddr,
1143 						    policy->family,
1144 						    policy->selector.prefixlen_s,
1145 						    dir);
1146 		write_seqcount_end(&bin->count);
1147 		if (!n)
1148 			return NULL;
1149 
1150 		return &n->hhead;
1151 	}
1152 
1153 	/* daddr is fixed */
1154 	write_seqcount_begin(&bin->count);
1155 	n = xfrm_policy_inexact_insert_node(net,
1156 					    &bin->root_d,
1157 					    &policy->selector.daddr,
1158 					    policy->family,
1159 					    policy->selector.prefixlen_d, dir);
1160 	write_seqcount_end(&bin->count);
1161 	if (!n)
1162 		return NULL;
1163 
1164 	/* saddr is wildcard */
1165 	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1166 					       policy->family,
1167 					       policy->selector.prefixlen_s))
1168 		return &n->hhead;
1169 
1170 	write_seqcount_begin(&bin->count);
1171 	n = xfrm_policy_inexact_insert_node(net,
1172 					    &n->root,
1173 					    &policy->selector.saddr,
1174 					    policy->family,
1175 					    policy->selector.prefixlen_s, dir);
1176 	write_seqcount_end(&bin->count);
1177 	if (!n)
1178 		return NULL;
1179 
1180 	return &n->hhead;
1181 }
1182 
1183 static struct xfrm_policy *
1184 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1185 {
1186 	struct xfrm_pol_inexact_bin *bin;
1187 	struct xfrm_policy *delpol;
1188 	struct hlist_head *chain;
1189 	struct net *net;
1190 
1191 	bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1192 	if (!bin)
1193 		return ERR_PTR(-ENOMEM);
1194 
1195 	net = xp_net(policy);
1196 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1197 
1198 	chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1199 	if (!chain) {
1200 		__xfrm_policy_inexact_prune_bin(bin, false);
1201 		return ERR_PTR(-ENOMEM);
1202 	}
1203 
1204 	delpol = xfrm_policy_insert_list(chain, policy, excl);
1205 	if (delpol && excl) {
1206 		__xfrm_policy_inexact_prune_bin(bin, false);
1207 		return ERR_PTR(-EEXIST);
1208 	}
1209 
1210 	chain = &net->xfrm.policy_inexact[dir];
1211 	xfrm_policy_insert_inexact_list(chain, policy);
1212 
1213 	if (delpol)
1214 		__xfrm_policy_inexact_prune_bin(bin, false);
1215 
1216 	return delpol;
1217 }
1218 
1219 static void xfrm_hash_rebuild(struct work_struct *work)
1220 {
1221 	struct net *net = container_of(work, struct net,
1222 				       xfrm.policy_hthresh.work);
1223 	unsigned int hmask;
1224 	struct xfrm_policy *pol;
1225 	struct xfrm_policy *policy;
1226 	struct hlist_head *chain;
1227 	struct hlist_head *odst;
1228 	struct hlist_node *newpos;
1229 	int i;
1230 	int dir;
1231 	unsigned seq;
1232 	u8 lbits4, rbits4, lbits6, rbits6;
1233 
1234 	mutex_lock(&hash_resize_mutex);
1235 
1236 	/* read selector prefixlen thresholds */
1237 	do {
1238 		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1239 
1240 		lbits4 = net->xfrm.policy_hthresh.lbits4;
1241 		rbits4 = net->xfrm.policy_hthresh.rbits4;
1242 		lbits6 = net->xfrm.policy_hthresh.lbits6;
1243 		rbits6 = net->xfrm.policy_hthresh.rbits6;
1244 	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1245 
1246 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1247 	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1248 
1249 	/* make sure that we can insert the indirect policies again before
1250 	 * we start with destructive action.
1251 	 */
1252 	list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1253 		struct xfrm_pol_inexact_bin *bin;
1254 		u8 dbits, sbits;
1255 
1256 		dir = xfrm_policy_id2dir(policy->index);
1257 		if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1258 			continue;
1259 
1260 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1261 			if (policy->family == AF_INET) {
1262 				dbits = rbits4;
1263 				sbits = lbits4;
1264 			} else {
1265 				dbits = rbits6;
1266 				sbits = lbits6;
1267 			}
1268 		} else {
1269 			if (policy->family == AF_INET) {
1270 				dbits = lbits4;
1271 				sbits = rbits4;
1272 			} else {
1273 				dbits = lbits6;
1274 				sbits = rbits6;
1275 			}
1276 		}
1277 
1278 		if (policy->selector.prefixlen_d < dbits ||
1279 		    policy->selector.prefixlen_s < sbits)
1280 			continue;
1281 
1282 		bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1283 		if (!bin)
1284 			goto out_unlock;
1285 
1286 		if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1287 			goto out_unlock;
1288 	}
1289 
1290 	/* reset the bydst and inexact table in all directions */
1291 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1292 		struct hlist_node *n;
1293 
1294 		hlist_for_each_entry_safe(policy, n,
1295 					  &net->xfrm.policy_inexact[dir],
1296 					  bydst_inexact_list) {
1297 			hlist_del_rcu(&policy->bydst);
1298 			hlist_del_init(&policy->bydst_inexact_list);
1299 		}
1300 
1301 		hmask = net->xfrm.policy_bydst[dir].hmask;
1302 		odst = net->xfrm.policy_bydst[dir].table;
1303 		for (i = hmask; i >= 0; i--) {
1304 			hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1305 				hlist_del_rcu(&policy->bydst);
1306 		}
1307 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1308 			/* dir out => dst = remote, src = local */
1309 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1310 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1311 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1312 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1313 		} else {
1314 			/* dir in/fwd => dst = local, src = remote */
1315 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1316 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1317 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1318 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1319 		}
1320 	}
1321 
1322 	/* re-insert all policies by order of creation */
1323 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1324 		if (policy->walk.dead)
1325 			continue;
1326 		dir = xfrm_policy_id2dir(policy->index);
1327 		if (dir >= XFRM_POLICY_MAX) {
1328 			/* skip socket policies */
1329 			continue;
1330 		}
1331 		newpos = NULL;
1332 		chain = policy_hash_bysel(net, &policy->selector,
1333 					  policy->family, dir);
1334 
1335 		if (!chain) {
1336 			void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1337 
1338 			WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1339 			continue;
1340 		}
1341 
1342 		hlist_for_each_entry(pol, chain, bydst) {
1343 			if (policy->priority >= pol->priority)
1344 				newpos = &pol->bydst;
1345 			else
1346 				break;
1347 		}
1348 		if (newpos)
1349 			hlist_add_behind_rcu(&policy->bydst, newpos);
1350 		else
1351 			hlist_add_head_rcu(&policy->bydst, chain);
1352 	}
1353 
1354 out_unlock:
1355 	__xfrm_policy_inexact_flush(net);
1356 	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1357 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1358 
1359 	mutex_unlock(&hash_resize_mutex);
1360 }
1361 
1362 void xfrm_policy_hash_rebuild(struct net *net)
1363 {
1364 	schedule_work(&net->xfrm.policy_hthresh.work);
1365 }
1366 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1367 
1368 /* Generate new index... KAME seems to generate them ordered by cost
1369  * of an absolute inpredictability of ordering of rules. This will not pass. */
1370 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1371 {
1372 	static u32 idx_generator;
1373 
1374 	for (;;) {
1375 		struct hlist_head *list;
1376 		struct xfrm_policy *p;
1377 		u32 idx;
1378 		int found;
1379 
1380 		if (!index) {
1381 			idx = (idx_generator | dir);
1382 			idx_generator += 8;
1383 		} else {
1384 			idx = index;
1385 			index = 0;
1386 		}
1387 
1388 		if (idx == 0)
1389 			idx = 8;
1390 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
1391 		found = 0;
1392 		hlist_for_each_entry(p, list, byidx) {
1393 			if (p->index == idx) {
1394 				found = 1;
1395 				break;
1396 			}
1397 		}
1398 		if (!found)
1399 			return idx;
1400 	}
1401 }
1402 
1403 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1404 {
1405 	u32 *p1 = (u32 *) s1;
1406 	u32 *p2 = (u32 *) s2;
1407 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
1408 	int i;
1409 
1410 	for (i = 0; i < len; i++) {
1411 		if (p1[i] != p2[i])
1412 			return 1;
1413 	}
1414 
1415 	return 0;
1416 }
1417 
1418 static void xfrm_policy_requeue(struct xfrm_policy *old,
1419 				struct xfrm_policy *new)
1420 {
1421 	struct xfrm_policy_queue *pq = &old->polq;
1422 	struct sk_buff_head list;
1423 
1424 	if (skb_queue_empty(&pq->hold_queue))
1425 		return;
1426 
1427 	__skb_queue_head_init(&list);
1428 
1429 	spin_lock_bh(&pq->hold_queue.lock);
1430 	skb_queue_splice_init(&pq->hold_queue, &list);
1431 	if (del_timer(&pq->hold_timer))
1432 		xfrm_pol_put(old);
1433 	spin_unlock_bh(&pq->hold_queue.lock);
1434 
1435 	pq = &new->polq;
1436 
1437 	spin_lock_bh(&pq->hold_queue.lock);
1438 	skb_queue_splice(&list, &pq->hold_queue);
1439 	pq->timeout = XFRM_QUEUE_TMO_MIN;
1440 	if (!mod_timer(&pq->hold_timer, jiffies))
1441 		xfrm_pol_hold(new);
1442 	spin_unlock_bh(&pq->hold_queue.lock);
1443 }
1444 
1445 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1446 					  struct xfrm_policy *pol)
1447 {
1448 	return mark->v == pol->mark.v && mark->m == pol->mark.m;
1449 }
1450 
1451 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1452 {
1453 	const struct xfrm_pol_inexact_key *k = data;
1454 	u32 a = k->type << 24 | k->dir << 16 | k->family;
1455 
1456 	return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1457 			    seed);
1458 }
1459 
1460 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1461 {
1462 	const struct xfrm_pol_inexact_bin *b = data;
1463 
1464 	return xfrm_pol_bin_key(&b->k, 0, seed);
1465 }
1466 
1467 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1468 			    const void *ptr)
1469 {
1470 	const struct xfrm_pol_inexact_key *key = arg->key;
1471 	const struct xfrm_pol_inexact_bin *b = ptr;
1472 	int ret;
1473 
1474 	if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1475 		return -1;
1476 
1477 	ret = b->k.dir ^ key->dir;
1478 	if (ret)
1479 		return ret;
1480 
1481 	ret = b->k.type ^ key->type;
1482 	if (ret)
1483 		return ret;
1484 
1485 	ret = b->k.family ^ key->family;
1486 	if (ret)
1487 		return ret;
1488 
1489 	return b->k.if_id ^ key->if_id;
1490 }
1491 
1492 static const struct rhashtable_params xfrm_pol_inexact_params = {
1493 	.head_offset		= offsetof(struct xfrm_pol_inexact_bin, head),
1494 	.hashfn			= xfrm_pol_bin_key,
1495 	.obj_hashfn		= xfrm_pol_bin_obj,
1496 	.obj_cmpfn		= xfrm_pol_bin_cmp,
1497 	.automatic_shrinking	= true,
1498 };
1499 
1500 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1501 					    struct xfrm_policy *policy)
1502 {
1503 	struct xfrm_policy *pol, *delpol = NULL;
1504 	struct hlist_node *newpos = NULL;
1505 	int i = 0;
1506 
1507 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1508 		if (pol->type == policy->type &&
1509 		    pol->if_id == policy->if_id &&
1510 		    !selector_cmp(&pol->selector, &policy->selector) &&
1511 		    xfrm_policy_mark_match(&policy->mark, pol) &&
1512 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1513 		    !WARN_ON(delpol)) {
1514 			delpol = pol;
1515 			if (policy->priority > pol->priority)
1516 				continue;
1517 		} else if (policy->priority >= pol->priority) {
1518 			newpos = &pol->bydst_inexact_list;
1519 			continue;
1520 		}
1521 		if (delpol)
1522 			break;
1523 	}
1524 
1525 	if (newpos)
1526 		hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1527 	else
1528 		hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1529 
1530 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1531 		pol->pos = i;
1532 		i++;
1533 	}
1534 }
1535 
1536 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1537 						   struct xfrm_policy *policy,
1538 						   bool excl)
1539 {
1540 	struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1541 
1542 	hlist_for_each_entry(pol, chain, bydst) {
1543 		if (pol->type == policy->type &&
1544 		    pol->if_id == policy->if_id &&
1545 		    !selector_cmp(&pol->selector, &policy->selector) &&
1546 		    xfrm_policy_mark_match(&policy->mark, pol) &&
1547 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1548 		    !WARN_ON(delpol)) {
1549 			if (excl)
1550 				return ERR_PTR(-EEXIST);
1551 			delpol = pol;
1552 			if (policy->priority > pol->priority)
1553 				continue;
1554 		} else if (policy->priority >= pol->priority) {
1555 			newpos = pol;
1556 			continue;
1557 		}
1558 		if (delpol)
1559 			break;
1560 	}
1561 
1562 	if (newpos)
1563 		hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1564 	else
1565 		hlist_add_head_rcu(&policy->bydst, chain);
1566 
1567 	return delpol;
1568 }
1569 
1570 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1571 {
1572 	struct net *net = xp_net(policy);
1573 	struct xfrm_policy *delpol;
1574 	struct hlist_head *chain;
1575 
1576 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1577 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1578 	if (chain)
1579 		delpol = xfrm_policy_insert_list(chain, policy, excl);
1580 	else
1581 		delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1582 
1583 	if (IS_ERR(delpol)) {
1584 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1585 		return PTR_ERR(delpol);
1586 	}
1587 
1588 	__xfrm_policy_link(policy, dir);
1589 
1590 	/* After previous checking, family can either be AF_INET or AF_INET6 */
1591 	if (policy->family == AF_INET)
1592 		rt_genid_bump_ipv4(net);
1593 	else
1594 		rt_genid_bump_ipv6(net);
1595 
1596 	if (delpol) {
1597 		xfrm_policy_requeue(delpol, policy);
1598 		__xfrm_policy_unlink(delpol, dir);
1599 	}
1600 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1601 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1602 	policy->curlft.add_time = ktime_get_real_seconds();
1603 	policy->curlft.use_time = 0;
1604 	if (!mod_timer(&policy->timer, jiffies + HZ))
1605 		xfrm_pol_hold(policy);
1606 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1607 
1608 	if (delpol)
1609 		xfrm_policy_kill(delpol);
1610 	else if (xfrm_bydst_should_resize(net, dir, NULL))
1611 		schedule_work(&net->xfrm.policy_hash_work);
1612 
1613 	return 0;
1614 }
1615 EXPORT_SYMBOL(xfrm_policy_insert);
1616 
1617 static struct xfrm_policy *
1618 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1619 			u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1620 			struct xfrm_sec_ctx *ctx)
1621 {
1622 	struct xfrm_policy *pol;
1623 
1624 	if (!chain)
1625 		return NULL;
1626 
1627 	hlist_for_each_entry(pol, chain, bydst) {
1628 		if (pol->type == type &&
1629 		    pol->if_id == if_id &&
1630 		    xfrm_policy_mark_match(mark, pol) &&
1631 		    !selector_cmp(sel, &pol->selector) &&
1632 		    xfrm_sec_ctx_match(ctx, pol->security))
1633 			return pol;
1634 	}
1635 
1636 	return NULL;
1637 }
1638 
1639 struct xfrm_policy *
1640 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1641 		      u8 type, int dir, struct xfrm_selector *sel,
1642 		      struct xfrm_sec_ctx *ctx, int delete, int *err)
1643 {
1644 	struct xfrm_pol_inexact_bin *bin = NULL;
1645 	struct xfrm_policy *pol, *ret = NULL;
1646 	struct hlist_head *chain;
1647 
1648 	*err = 0;
1649 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1650 	chain = policy_hash_bysel(net, sel, sel->family, dir);
1651 	if (!chain) {
1652 		struct xfrm_pol_inexact_candidates cand;
1653 		int i;
1654 
1655 		bin = xfrm_policy_inexact_lookup(net, type,
1656 						 sel->family, dir, if_id);
1657 		if (!bin) {
1658 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1659 			return NULL;
1660 		}
1661 
1662 		if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1663 							 &sel->saddr,
1664 							 &sel->daddr)) {
1665 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1666 			return NULL;
1667 		}
1668 
1669 		pol = NULL;
1670 		for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1671 			struct xfrm_policy *tmp;
1672 
1673 			tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1674 						      if_id, type, dir,
1675 						      sel, ctx);
1676 			if (!tmp)
1677 				continue;
1678 
1679 			if (!pol || tmp->pos < pol->pos)
1680 				pol = tmp;
1681 		}
1682 	} else {
1683 		pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1684 					      sel, ctx);
1685 	}
1686 
1687 	if (pol) {
1688 		xfrm_pol_hold(pol);
1689 		if (delete) {
1690 			*err = security_xfrm_policy_delete(pol->security);
1691 			if (*err) {
1692 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1693 				return pol;
1694 			}
1695 			__xfrm_policy_unlink(pol, dir);
1696 		}
1697 		ret = pol;
1698 	}
1699 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1700 
1701 	if (ret && delete)
1702 		xfrm_policy_kill(ret);
1703 	if (bin && delete)
1704 		xfrm_policy_inexact_prune_bin(bin);
1705 	return ret;
1706 }
1707 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1708 
1709 struct xfrm_policy *
1710 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1711 		 u8 type, int dir, u32 id, int delete, int *err)
1712 {
1713 	struct xfrm_policy *pol, *ret;
1714 	struct hlist_head *chain;
1715 
1716 	*err = -ENOENT;
1717 	if (xfrm_policy_id2dir(id) != dir)
1718 		return NULL;
1719 
1720 	*err = 0;
1721 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1722 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
1723 	ret = NULL;
1724 	hlist_for_each_entry(pol, chain, byidx) {
1725 		if (pol->type == type && pol->index == id &&
1726 		    pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1727 			xfrm_pol_hold(pol);
1728 			if (delete) {
1729 				*err = security_xfrm_policy_delete(
1730 								pol->security);
1731 				if (*err) {
1732 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1733 					return pol;
1734 				}
1735 				__xfrm_policy_unlink(pol, dir);
1736 			}
1737 			ret = pol;
1738 			break;
1739 		}
1740 	}
1741 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1742 
1743 	if (ret && delete)
1744 		xfrm_policy_kill(ret);
1745 	return ret;
1746 }
1747 EXPORT_SYMBOL(xfrm_policy_byid);
1748 
1749 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1750 static inline int
1751 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1752 {
1753 	struct xfrm_policy *pol;
1754 	int err = 0;
1755 
1756 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1757 		if (pol->walk.dead ||
1758 		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1759 		    pol->type != type)
1760 			continue;
1761 
1762 		err = security_xfrm_policy_delete(pol->security);
1763 		if (err) {
1764 			xfrm_audit_policy_delete(pol, 0, task_valid);
1765 			return err;
1766 		}
1767 	}
1768 	return err;
1769 }
1770 #else
1771 static inline int
1772 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1773 {
1774 	return 0;
1775 }
1776 #endif
1777 
1778 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1779 {
1780 	int dir, err = 0, cnt = 0;
1781 	struct xfrm_policy *pol;
1782 
1783 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1784 
1785 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1786 	if (err)
1787 		goto out;
1788 
1789 again:
1790 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1791 		dir = xfrm_policy_id2dir(pol->index);
1792 		if (pol->walk.dead ||
1793 		    dir >= XFRM_POLICY_MAX ||
1794 		    pol->type != type)
1795 			continue;
1796 
1797 		__xfrm_policy_unlink(pol, dir);
1798 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1799 		cnt++;
1800 		xfrm_audit_policy_delete(pol, 1, task_valid);
1801 		xfrm_policy_kill(pol);
1802 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1803 		goto again;
1804 	}
1805 	if (cnt)
1806 		__xfrm_policy_inexact_flush(net);
1807 	else
1808 		err = -ESRCH;
1809 out:
1810 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1811 	return err;
1812 }
1813 EXPORT_SYMBOL(xfrm_policy_flush);
1814 
1815 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1816 		     int (*func)(struct xfrm_policy *, int, int, void*),
1817 		     void *data)
1818 {
1819 	struct xfrm_policy *pol;
1820 	struct xfrm_policy_walk_entry *x;
1821 	int error = 0;
1822 
1823 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1824 	    walk->type != XFRM_POLICY_TYPE_ANY)
1825 		return -EINVAL;
1826 
1827 	if (list_empty(&walk->walk.all) && walk->seq != 0)
1828 		return 0;
1829 
1830 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1831 	if (list_empty(&walk->walk.all))
1832 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1833 	else
1834 		x = list_first_entry(&walk->walk.all,
1835 				     struct xfrm_policy_walk_entry, all);
1836 
1837 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1838 		if (x->dead)
1839 			continue;
1840 		pol = container_of(x, struct xfrm_policy, walk);
1841 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1842 		    walk->type != pol->type)
1843 			continue;
1844 		error = func(pol, xfrm_policy_id2dir(pol->index),
1845 			     walk->seq, data);
1846 		if (error) {
1847 			list_move_tail(&walk->walk.all, &x->all);
1848 			goto out;
1849 		}
1850 		walk->seq++;
1851 	}
1852 	if (walk->seq == 0) {
1853 		error = -ENOENT;
1854 		goto out;
1855 	}
1856 	list_del_init(&walk->walk.all);
1857 out:
1858 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1859 	return error;
1860 }
1861 EXPORT_SYMBOL(xfrm_policy_walk);
1862 
1863 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1864 {
1865 	INIT_LIST_HEAD(&walk->walk.all);
1866 	walk->walk.dead = 1;
1867 	walk->type = type;
1868 	walk->seq = 0;
1869 }
1870 EXPORT_SYMBOL(xfrm_policy_walk_init);
1871 
1872 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1873 {
1874 	if (list_empty(&walk->walk.all))
1875 		return;
1876 
1877 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1878 	list_del(&walk->walk.all);
1879 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1880 }
1881 EXPORT_SYMBOL(xfrm_policy_walk_done);
1882 
1883 /*
1884  * Find policy to apply to this flow.
1885  *
1886  * Returns 0 if policy found, else an -errno.
1887  */
1888 static int xfrm_policy_match(const struct xfrm_policy *pol,
1889 			     const struct flowi *fl,
1890 			     u8 type, u16 family, int dir, u32 if_id)
1891 {
1892 	const struct xfrm_selector *sel = &pol->selector;
1893 	int ret = -ESRCH;
1894 	bool match;
1895 
1896 	if (pol->family != family ||
1897 	    pol->if_id != if_id ||
1898 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1899 	    pol->type != type)
1900 		return ret;
1901 
1902 	match = xfrm_selector_match(sel, fl, family);
1903 	if (match)
1904 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1905 	return ret;
1906 }
1907 
1908 static struct xfrm_pol_inexact_node *
1909 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1910 				seqcount_spinlock_t *count,
1911 				const xfrm_address_t *addr, u16 family)
1912 {
1913 	const struct rb_node *parent;
1914 	int seq;
1915 
1916 again:
1917 	seq = read_seqcount_begin(count);
1918 
1919 	parent = rcu_dereference_raw(r->rb_node);
1920 	while (parent) {
1921 		struct xfrm_pol_inexact_node *node;
1922 		int delta;
1923 
1924 		node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1925 
1926 		delta = xfrm_policy_addr_delta(addr, &node->addr,
1927 					       node->prefixlen, family);
1928 		if (delta < 0) {
1929 			parent = rcu_dereference_raw(parent->rb_left);
1930 			continue;
1931 		} else if (delta > 0) {
1932 			parent = rcu_dereference_raw(parent->rb_right);
1933 			continue;
1934 		}
1935 
1936 		return node;
1937 	}
1938 
1939 	if (read_seqcount_retry(count, seq))
1940 		goto again;
1941 
1942 	return NULL;
1943 }
1944 
1945 static bool
1946 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1947 				    struct xfrm_pol_inexact_bin *b,
1948 				    const xfrm_address_t *saddr,
1949 				    const xfrm_address_t *daddr)
1950 {
1951 	struct xfrm_pol_inexact_node *n;
1952 	u16 family;
1953 
1954 	if (!b)
1955 		return false;
1956 
1957 	family = b->k.family;
1958 	memset(cand, 0, sizeof(*cand));
1959 	cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1960 
1961 	n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1962 					    family);
1963 	if (n) {
1964 		cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1965 		n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1966 						    family);
1967 		if (n)
1968 			cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1969 	}
1970 
1971 	n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1972 					    family);
1973 	if (n)
1974 		cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1975 
1976 	return true;
1977 }
1978 
1979 static struct xfrm_pol_inexact_bin *
1980 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1981 			       u8 dir, u32 if_id)
1982 {
1983 	struct xfrm_pol_inexact_key k = {
1984 		.family = family,
1985 		.type = type,
1986 		.dir = dir,
1987 		.if_id = if_id,
1988 	};
1989 
1990 	write_pnet(&k.net, net);
1991 
1992 	return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1993 				 xfrm_pol_inexact_params);
1994 }
1995 
1996 static struct xfrm_pol_inexact_bin *
1997 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1998 			   u8 dir, u32 if_id)
1999 {
2000 	struct xfrm_pol_inexact_bin *bin;
2001 
2002 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2003 
2004 	rcu_read_lock();
2005 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2006 	rcu_read_unlock();
2007 
2008 	return bin;
2009 }
2010 
2011 static struct xfrm_policy *
2012 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2013 			      struct xfrm_policy *prefer,
2014 			      const struct flowi *fl,
2015 			      u8 type, u16 family, int dir, u32 if_id)
2016 {
2017 	u32 priority = prefer ? prefer->priority : ~0u;
2018 	struct xfrm_policy *pol;
2019 
2020 	if (!chain)
2021 		return NULL;
2022 
2023 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2024 		int err;
2025 
2026 		if (pol->priority > priority)
2027 			break;
2028 
2029 		err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2030 		if (err) {
2031 			if (err != -ESRCH)
2032 				return ERR_PTR(err);
2033 
2034 			continue;
2035 		}
2036 
2037 		if (prefer) {
2038 			/* matches.  Is it older than *prefer? */
2039 			if (pol->priority == priority &&
2040 			    prefer->pos < pol->pos)
2041 				return prefer;
2042 		}
2043 
2044 		return pol;
2045 	}
2046 
2047 	return NULL;
2048 }
2049 
2050 static struct xfrm_policy *
2051 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2052 			    struct xfrm_policy *prefer,
2053 			    const struct flowi *fl,
2054 			    u8 type, u16 family, int dir, u32 if_id)
2055 {
2056 	struct xfrm_policy *tmp;
2057 	int i;
2058 
2059 	for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2060 		tmp = __xfrm_policy_eval_candidates(cand->res[i],
2061 						    prefer,
2062 						    fl, type, family, dir,
2063 						    if_id);
2064 		if (!tmp)
2065 			continue;
2066 
2067 		if (IS_ERR(tmp))
2068 			return tmp;
2069 		prefer = tmp;
2070 	}
2071 
2072 	return prefer;
2073 }
2074 
2075 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2076 						     const struct flowi *fl,
2077 						     u16 family, u8 dir,
2078 						     u32 if_id)
2079 {
2080 	struct xfrm_pol_inexact_candidates cand;
2081 	const xfrm_address_t *daddr, *saddr;
2082 	struct xfrm_pol_inexact_bin *bin;
2083 	struct xfrm_policy *pol, *ret;
2084 	struct hlist_head *chain;
2085 	unsigned int sequence;
2086 	int err;
2087 
2088 	daddr = xfrm_flowi_daddr(fl, family);
2089 	saddr = xfrm_flowi_saddr(fl, family);
2090 	if (unlikely(!daddr || !saddr))
2091 		return NULL;
2092 
2093 	rcu_read_lock();
2094  retry:
2095 	do {
2096 		sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2097 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
2098 	} while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2099 
2100 	ret = NULL;
2101 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2102 		err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2103 		if (err) {
2104 			if (err == -ESRCH)
2105 				continue;
2106 			else {
2107 				ret = ERR_PTR(err);
2108 				goto fail;
2109 			}
2110 		} else {
2111 			ret = pol;
2112 			break;
2113 		}
2114 	}
2115 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2116 	if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2117 							 daddr))
2118 		goto skip_inexact;
2119 
2120 	pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2121 					  family, dir, if_id);
2122 	if (pol) {
2123 		ret = pol;
2124 		if (IS_ERR(pol))
2125 			goto fail;
2126 	}
2127 
2128 skip_inexact:
2129 	if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2130 		goto retry;
2131 
2132 	if (ret && !xfrm_pol_hold_rcu(ret))
2133 		goto retry;
2134 fail:
2135 	rcu_read_unlock();
2136 
2137 	return ret;
2138 }
2139 
2140 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2141 					      const struct flowi *fl,
2142 					      u16 family, u8 dir, u32 if_id)
2143 {
2144 #ifdef CONFIG_XFRM_SUB_POLICY
2145 	struct xfrm_policy *pol;
2146 
2147 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2148 					dir, if_id);
2149 	if (pol != NULL)
2150 		return pol;
2151 #endif
2152 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2153 					 dir, if_id);
2154 }
2155 
2156 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2157 						 const struct flowi *fl,
2158 						 u16 family, u32 if_id)
2159 {
2160 	struct xfrm_policy *pol;
2161 
2162 	rcu_read_lock();
2163  again:
2164 	pol = rcu_dereference(sk->sk_policy[dir]);
2165 	if (pol != NULL) {
2166 		bool match;
2167 		int err = 0;
2168 
2169 		if (pol->family != family) {
2170 			pol = NULL;
2171 			goto out;
2172 		}
2173 
2174 		match = xfrm_selector_match(&pol->selector, fl, family);
2175 		if (match) {
2176 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2177 			    pol->if_id != if_id) {
2178 				pol = NULL;
2179 				goto out;
2180 			}
2181 			err = security_xfrm_policy_lookup(pol->security,
2182 						      fl->flowi_secid);
2183 			if (!err) {
2184 				if (!xfrm_pol_hold_rcu(pol))
2185 					goto again;
2186 			} else if (err == -ESRCH) {
2187 				pol = NULL;
2188 			} else {
2189 				pol = ERR_PTR(err);
2190 			}
2191 		} else
2192 			pol = NULL;
2193 	}
2194 out:
2195 	rcu_read_unlock();
2196 	return pol;
2197 }
2198 
2199 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2200 {
2201 	struct net *net = xp_net(pol);
2202 
2203 	list_add(&pol->walk.all, &net->xfrm.policy_all);
2204 	net->xfrm.policy_count[dir]++;
2205 	xfrm_pol_hold(pol);
2206 }
2207 
2208 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2209 						int dir)
2210 {
2211 	struct net *net = xp_net(pol);
2212 
2213 	if (list_empty(&pol->walk.all))
2214 		return NULL;
2215 
2216 	/* Socket policies are not hashed. */
2217 	if (!hlist_unhashed(&pol->bydst)) {
2218 		hlist_del_rcu(&pol->bydst);
2219 		hlist_del_init(&pol->bydst_inexact_list);
2220 		hlist_del(&pol->byidx);
2221 	}
2222 
2223 	list_del_init(&pol->walk.all);
2224 	net->xfrm.policy_count[dir]--;
2225 
2226 	return pol;
2227 }
2228 
2229 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2230 {
2231 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2232 }
2233 
2234 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2235 {
2236 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2237 }
2238 
2239 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2240 {
2241 	struct net *net = xp_net(pol);
2242 
2243 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2244 	pol = __xfrm_policy_unlink(pol, dir);
2245 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2246 	if (pol) {
2247 		xfrm_policy_kill(pol);
2248 		return 0;
2249 	}
2250 	return -ENOENT;
2251 }
2252 EXPORT_SYMBOL(xfrm_policy_delete);
2253 
2254 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2255 {
2256 	struct net *net = sock_net(sk);
2257 	struct xfrm_policy *old_pol;
2258 
2259 #ifdef CONFIG_XFRM_SUB_POLICY
2260 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2261 		return -EINVAL;
2262 #endif
2263 
2264 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2265 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2266 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2267 	if (pol) {
2268 		pol->curlft.add_time = ktime_get_real_seconds();
2269 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2270 		xfrm_sk_policy_link(pol, dir);
2271 	}
2272 	rcu_assign_pointer(sk->sk_policy[dir], pol);
2273 	if (old_pol) {
2274 		if (pol)
2275 			xfrm_policy_requeue(old_pol, pol);
2276 
2277 		/* Unlinking succeeds always. This is the only function
2278 		 * allowed to delete or replace socket policy.
2279 		 */
2280 		xfrm_sk_policy_unlink(old_pol, dir);
2281 	}
2282 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2283 
2284 	if (old_pol) {
2285 		xfrm_policy_kill(old_pol);
2286 	}
2287 	return 0;
2288 }
2289 
2290 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2291 {
2292 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2293 	struct net *net = xp_net(old);
2294 
2295 	if (newp) {
2296 		newp->selector = old->selector;
2297 		if (security_xfrm_policy_clone(old->security,
2298 					       &newp->security)) {
2299 			kfree(newp);
2300 			return NULL;  /* ENOMEM */
2301 		}
2302 		newp->lft = old->lft;
2303 		newp->curlft = old->curlft;
2304 		newp->mark = old->mark;
2305 		newp->if_id = old->if_id;
2306 		newp->action = old->action;
2307 		newp->flags = old->flags;
2308 		newp->xfrm_nr = old->xfrm_nr;
2309 		newp->index = old->index;
2310 		newp->type = old->type;
2311 		newp->family = old->family;
2312 		memcpy(newp->xfrm_vec, old->xfrm_vec,
2313 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2314 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2315 		xfrm_sk_policy_link(newp, dir);
2316 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2317 		xfrm_pol_put(newp);
2318 	}
2319 	return newp;
2320 }
2321 
2322 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2323 {
2324 	const struct xfrm_policy *p;
2325 	struct xfrm_policy *np;
2326 	int i, ret = 0;
2327 
2328 	rcu_read_lock();
2329 	for (i = 0; i < 2; i++) {
2330 		p = rcu_dereference(osk->sk_policy[i]);
2331 		if (p) {
2332 			np = clone_policy(p, i);
2333 			if (unlikely(!np)) {
2334 				ret = -ENOMEM;
2335 				break;
2336 			}
2337 			rcu_assign_pointer(sk->sk_policy[i], np);
2338 		}
2339 	}
2340 	rcu_read_unlock();
2341 	return ret;
2342 }
2343 
2344 static int
2345 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2346 	       xfrm_address_t *remote, unsigned short family, u32 mark)
2347 {
2348 	int err;
2349 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2350 
2351 	if (unlikely(afinfo == NULL))
2352 		return -EINVAL;
2353 	err = afinfo->get_saddr(net, oif, local, remote, mark);
2354 	rcu_read_unlock();
2355 	return err;
2356 }
2357 
2358 /* Resolve list of templates for the flow, given policy. */
2359 
2360 static int
2361 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2362 		      struct xfrm_state **xfrm, unsigned short family)
2363 {
2364 	struct net *net = xp_net(policy);
2365 	int nx;
2366 	int i, error;
2367 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2368 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2369 	xfrm_address_t tmp;
2370 
2371 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2372 		struct xfrm_state *x;
2373 		xfrm_address_t *remote = daddr;
2374 		xfrm_address_t *local  = saddr;
2375 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2376 
2377 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
2378 		    tmpl->mode == XFRM_MODE_BEET) {
2379 			remote = &tmpl->id.daddr;
2380 			local = &tmpl->saddr;
2381 			if (xfrm_addr_any(local, tmpl->encap_family)) {
2382 				error = xfrm_get_saddr(net, fl->flowi_oif,
2383 						       &tmp, remote,
2384 						       tmpl->encap_family, 0);
2385 				if (error)
2386 					goto fail;
2387 				local = &tmp;
2388 			}
2389 		}
2390 
2391 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2392 				    family, policy->if_id);
2393 
2394 		if (x && x->km.state == XFRM_STATE_VALID) {
2395 			xfrm[nx++] = x;
2396 			daddr = remote;
2397 			saddr = local;
2398 			continue;
2399 		}
2400 		if (x) {
2401 			error = (x->km.state == XFRM_STATE_ERROR ?
2402 				 -EINVAL : -EAGAIN);
2403 			xfrm_state_put(x);
2404 		} else if (error == -ESRCH) {
2405 			error = -EAGAIN;
2406 		}
2407 
2408 		if (!tmpl->optional)
2409 			goto fail;
2410 	}
2411 	return nx;
2412 
2413 fail:
2414 	for (nx--; nx >= 0; nx--)
2415 		xfrm_state_put(xfrm[nx]);
2416 	return error;
2417 }
2418 
2419 static int
2420 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2421 		  struct xfrm_state **xfrm, unsigned short family)
2422 {
2423 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
2424 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2425 	int cnx = 0;
2426 	int error;
2427 	int ret;
2428 	int i;
2429 
2430 	for (i = 0; i < npols; i++) {
2431 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2432 			error = -ENOBUFS;
2433 			goto fail;
2434 		}
2435 
2436 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2437 		if (ret < 0) {
2438 			error = ret;
2439 			goto fail;
2440 		} else
2441 			cnx += ret;
2442 	}
2443 
2444 	/* found states are sorted for outbound processing */
2445 	if (npols > 1)
2446 		xfrm_state_sort(xfrm, tpp, cnx, family);
2447 
2448 	return cnx;
2449 
2450  fail:
2451 	for (cnx--; cnx >= 0; cnx--)
2452 		xfrm_state_put(tpp[cnx]);
2453 	return error;
2454 
2455 }
2456 
2457 static int xfrm_get_tos(const struct flowi *fl, int family)
2458 {
2459 	if (family == AF_INET)
2460 		return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2461 
2462 	return 0;
2463 }
2464 
2465 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2466 {
2467 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2468 	struct dst_ops *dst_ops;
2469 	struct xfrm_dst *xdst;
2470 
2471 	if (!afinfo)
2472 		return ERR_PTR(-EINVAL);
2473 
2474 	switch (family) {
2475 	case AF_INET:
2476 		dst_ops = &net->xfrm.xfrm4_dst_ops;
2477 		break;
2478 #if IS_ENABLED(CONFIG_IPV6)
2479 	case AF_INET6:
2480 		dst_ops = &net->xfrm.xfrm6_dst_ops;
2481 		break;
2482 #endif
2483 	default:
2484 		BUG();
2485 	}
2486 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2487 
2488 	if (likely(xdst)) {
2489 		memset_after(xdst, 0, u.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->lastuse = now;
2612 
2613 		dst1->input = dst_discard;
2614 
2615 		rcu_read_lock();
2616 		afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2617 		if (likely(afinfo))
2618 			dst1->output = afinfo->output;
2619 		else
2620 			dst1->output = dst_discard_out;
2621 		rcu_read_unlock();
2622 
2623 		xdst_prev = xdst;
2624 
2625 		header_len += xfrm[i]->props.header_len;
2626 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2627 			nfheader_len += xfrm[i]->props.header_len;
2628 		trailer_len += xfrm[i]->props.trailer_len;
2629 	}
2630 
2631 	xfrm_dst_set_child(xdst_prev, dst);
2632 	xdst0->path = dst;
2633 
2634 	err = -ENODEV;
2635 	dev = dst->dev;
2636 	if (!dev)
2637 		goto free_dst;
2638 
2639 	xfrm_init_path(xdst0, dst, nfheader_len);
2640 	xfrm_init_pmtu(bundle, nx);
2641 
2642 	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2643 	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2644 		err = xfrm_fill_dst(xdst_prev, dev, fl);
2645 		if (err)
2646 			goto free_dst;
2647 
2648 		xdst_prev->u.dst.header_len = header_len;
2649 		xdst_prev->u.dst.trailer_len = trailer_len;
2650 		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2651 		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2652 	}
2653 
2654 	return &xdst0->u.dst;
2655 
2656 put_states:
2657 	for (; i < nx; i++)
2658 		xfrm_state_put(xfrm[i]);
2659 free_dst:
2660 	if (xdst0)
2661 		dst_release_immediate(&xdst0->u.dst);
2662 
2663 	return ERR_PTR(err);
2664 }
2665 
2666 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2667 				struct xfrm_policy **pols,
2668 				int *num_pols, int *num_xfrms)
2669 {
2670 	int i;
2671 
2672 	if (*num_pols == 0 || !pols[0]) {
2673 		*num_pols = 0;
2674 		*num_xfrms = 0;
2675 		return 0;
2676 	}
2677 	if (IS_ERR(pols[0]))
2678 		return PTR_ERR(pols[0]);
2679 
2680 	*num_xfrms = pols[0]->xfrm_nr;
2681 
2682 #ifdef CONFIG_XFRM_SUB_POLICY
2683 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2684 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2685 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2686 						    XFRM_POLICY_TYPE_MAIN,
2687 						    fl, family,
2688 						    XFRM_POLICY_OUT,
2689 						    pols[0]->if_id);
2690 		if (pols[1]) {
2691 			if (IS_ERR(pols[1])) {
2692 				xfrm_pols_put(pols, *num_pols);
2693 				return PTR_ERR(pols[1]);
2694 			}
2695 			(*num_pols)++;
2696 			(*num_xfrms) += pols[1]->xfrm_nr;
2697 		}
2698 	}
2699 #endif
2700 	for (i = 0; i < *num_pols; i++) {
2701 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
2702 			*num_xfrms = -1;
2703 			break;
2704 		}
2705 	}
2706 
2707 	return 0;
2708 
2709 }
2710 
2711 static struct xfrm_dst *
2712 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2713 			       const struct flowi *fl, u16 family,
2714 			       struct dst_entry *dst_orig)
2715 {
2716 	struct net *net = xp_net(pols[0]);
2717 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2718 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2719 	struct xfrm_dst *xdst;
2720 	struct dst_entry *dst;
2721 	int err;
2722 
2723 	/* Try to instantiate a bundle */
2724 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2725 	if (err <= 0) {
2726 		if (err == 0)
2727 			return NULL;
2728 
2729 		if (err != -EAGAIN)
2730 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2731 		return ERR_PTR(err);
2732 	}
2733 
2734 	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2735 	if (IS_ERR(dst)) {
2736 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2737 		return ERR_CAST(dst);
2738 	}
2739 
2740 	xdst = (struct xfrm_dst *)dst;
2741 	xdst->num_xfrms = err;
2742 	xdst->num_pols = num_pols;
2743 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2744 	xdst->policy_genid = atomic_read(&pols[0]->genid);
2745 
2746 	return xdst;
2747 }
2748 
2749 static void xfrm_policy_queue_process(struct timer_list *t)
2750 {
2751 	struct sk_buff *skb;
2752 	struct sock *sk;
2753 	struct dst_entry *dst;
2754 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2755 	struct net *net = xp_net(pol);
2756 	struct xfrm_policy_queue *pq = &pol->polq;
2757 	struct flowi fl;
2758 	struct sk_buff_head list;
2759 	__u32 skb_mark;
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 
2770 	/* Fixup the mark to support VTI. */
2771 	skb_mark = skb->mark;
2772 	skb->mark = pol->mark.v;
2773 	xfrm_decode_session(skb, &fl, dst->ops->family);
2774 	skb->mark = skb_mark;
2775 	spin_unlock(&pq->hold_queue.lock);
2776 
2777 	dst_hold(xfrm_dst_path(dst));
2778 	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2779 	if (IS_ERR(dst))
2780 		goto purge_queue;
2781 
2782 	if (dst->flags & DST_XFRM_QUEUE) {
2783 		dst_release(dst);
2784 
2785 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2786 			goto purge_queue;
2787 
2788 		pq->timeout = pq->timeout << 1;
2789 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2790 			xfrm_pol_hold(pol);
2791 		goto out;
2792 	}
2793 
2794 	dst_release(dst);
2795 
2796 	__skb_queue_head_init(&list);
2797 
2798 	spin_lock(&pq->hold_queue.lock);
2799 	pq->timeout = 0;
2800 	skb_queue_splice_init(&pq->hold_queue, &list);
2801 	spin_unlock(&pq->hold_queue.lock);
2802 
2803 	while (!skb_queue_empty(&list)) {
2804 		skb = __skb_dequeue(&list);
2805 
2806 		/* Fixup the mark to support VTI. */
2807 		skb_mark = skb->mark;
2808 		skb->mark = pol->mark.v;
2809 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2810 		skb->mark = skb_mark;
2811 
2812 		dst_hold(xfrm_dst_path(skb_dst(skb)));
2813 		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2814 		if (IS_ERR(dst)) {
2815 			kfree_skb(skb);
2816 			continue;
2817 		}
2818 
2819 		nf_reset_ct(skb);
2820 		skb_dst_drop(skb);
2821 		skb_dst_set(skb, dst);
2822 
2823 		dst_output(net, skb->sk, skb);
2824 	}
2825 
2826 out:
2827 	xfrm_pol_put(pol);
2828 	return;
2829 
2830 purge_queue:
2831 	pq->timeout = 0;
2832 	skb_queue_purge(&pq->hold_queue);
2833 	xfrm_pol_put(pol);
2834 }
2835 
2836 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2837 {
2838 	unsigned long sched_next;
2839 	struct dst_entry *dst = skb_dst(skb);
2840 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2841 	struct xfrm_policy *pol = xdst->pols[0];
2842 	struct xfrm_policy_queue *pq = &pol->polq;
2843 
2844 	if (unlikely(skb_fclone_busy(sk, skb))) {
2845 		kfree_skb(skb);
2846 		return 0;
2847 	}
2848 
2849 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2850 		kfree_skb(skb);
2851 		return -EAGAIN;
2852 	}
2853 
2854 	skb_dst_force(skb);
2855 
2856 	spin_lock_bh(&pq->hold_queue.lock);
2857 
2858 	if (!pq->timeout)
2859 		pq->timeout = XFRM_QUEUE_TMO_MIN;
2860 
2861 	sched_next = jiffies + pq->timeout;
2862 
2863 	if (del_timer(&pq->hold_timer)) {
2864 		if (time_before(pq->hold_timer.expires, sched_next))
2865 			sched_next = pq->hold_timer.expires;
2866 		xfrm_pol_put(pol);
2867 	}
2868 
2869 	__skb_queue_tail(&pq->hold_queue, skb);
2870 	if (!mod_timer(&pq->hold_timer, sched_next))
2871 		xfrm_pol_hold(pol);
2872 
2873 	spin_unlock_bh(&pq->hold_queue.lock);
2874 
2875 	return 0;
2876 }
2877 
2878 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2879 						 struct xfrm_flo *xflo,
2880 						 const struct flowi *fl,
2881 						 int num_xfrms,
2882 						 u16 family)
2883 {
2884 	int err;
2885 	struct net_device *dev;
2886 	struct dst_entry *dst;
2887 	struct dst_entry *dst1;
2888 	struct xfrm_dst *xdst;
2889 
2890 	xdst = xfrm_alloc_dst(net, family);
2891 	if (IS_ERR(xdst))
2892 		return xdst;
2893 
2894 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2895 	    net->xfrm.sysctl_larval_drop ||
2896 	    num_xfrms <= 0)
2897 		return xdst;
2898 
2899 	dst = xflo->dst_orig;
2900 	dst1 = &xdst->u.dst;
2901 	dst_hold(dst);
2902 	xdst->route = dst;
2903 
2904 	dst_copy_metrics(dst1, dst);
2905 
2906 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2907 	dst1->flags |= DST_XFRM_QUEUE;
2908 	dst1->lastuse = jiffies;
2909 
2910 	dst1->input = dst_discard;
2911 	dst1->output = xdst_queue_output;
2912 
2913 	dst_hold(dst);
2914 	xfrm_dst_set_child(xdst, dst);
2915 	xdst->path = dst;
2916 
2917 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2918 
2919 	err = -ENODEV;
2920 	dev = dst->dev;
2921 	if (!dev)
2922 		goto free_dst;
2923 
2924 	err = xfrm_fill_dst(xdst, dev, fl);
2925 	if (err)
2926 		goto free_dst;
2927 
2928 out:
2929 	return xdst;
2930 
2931 free_dst:
2932 	dst_release(dst1);
2933 	xdst = ERR_PTR(err);
2934 	goto out;
2935 }
2936 
2937 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2938 					   const struct flowi *fl,
2939 					   u16 family, u8 dir,
2940 					   struct xfrm_flo *xflo, u32 if_id)
2941 {
2942 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2943 	int num_pols = 0, num_xfrms = 0, err;
2944 	struct xfrm_dst *xdst;
2945 
2946 	/* Resolve policies to use if we couldn't get them from
2947 	 * previous cache entry */
2948 	num_pols = 1;
2949 	pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2950 	err = xfrm_expand_policies(fl, family, pols,
2951 					   &num_pols, &num_xfrms);
2952 	if (err < 0)
2953 		goto inc_error;
2954 	if (num_pols == 0)
2955 		return NULL;
2956 	if (num_xfrms <= 0)
2957 		goto make_dummy_bundle;
2958 
2959 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2960 					      xflo->dst_orig);
2961 	if (IS_ERR(xdst)) {
2962 		err = PTR_ERR(xdst);
2963 		if (err == -EREMOTE) {
2964 			xfrm_pols_put(pols, num_pols);
2965 			return NULL;
2966 		}
2967 
2968 		if (err != -EAGAIN)
2969 			goto error;
2970 		goto make_dummy_bundle;
2971 	} else if (xdst == NULL) {
2972 		num_xfrms = 0;
2973 		goto make_dummy_bundle;
2974 	}
2975 
2976 	return xdst;
2977 
2978 make_dummy_bundle:
2979 	/* We found policies, but there's no bundles to instantiate:
2980 	 * either because the policy blocks, has no transformations or
2981 	 * we could not build template (no xfrm_states).*/
2982 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2983 	if (IS_ERR(xdst)) {
2984 		xfrm_pols_put(pols, num_pols);
2985 		return ERR_CAST(xdst);
2986 	}
2987 	xdst->num_pols = num_pols;
2988 	xdst->num_xfrms = num_xfrms;
2989 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2990 
2991 	return xdst;
2992 
2993 inc_error:
2994 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2995 error:
2996 	xfrm_pols_put(pols, num_pols);
2997 	return ERR_PTR(err);
2998 }
2999 
3000 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3001 					struct dst_entry *dst_orig)
3002 {
3003 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3004 	struct dst_entry *ret;
3005 
3006 	if (!afinfo) {
3007 		dst_release(dst_orig);
3008 		return ERR_PTR(-EINVAL);
3009 	} else {
3010 		ret = afinfo->blackhole_route(net, dst_orig);
3011 	}
3012 	rcu_read_unlock();
3013 
3014 	return ret;
3015 }
3016 
3017 /* Finds/creates a bundle for given flow and if_id
3018  *
3019  * At the moment we eat a raw IP route. Mostly to speed up lookups
3020  * on interfaces with disabled IPsec.
3021  *
3022  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3023  * compatibility
3024  */
3025 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3026 					struct dst_entry *dst_orig,
3027 					const struct flowi *fl,
3028 					const struct sock *sk,
3029 					int flags, u32 if_id)
3030 {
3031 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3032 	struct xfrm_dst *xdst;
3033 	struct dst_entry *dst, *route;
3034 	u16 family = dst_orig->ops->family;
3035 	u8 dir = XFRM_POLICY_OUT;
3036 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3037 
3038 	dst = NULL;
3039 	xdst = NULL;
3040 	route = NULL;
3041 
3042 	sk = sk_const_to_full_sk(sk);
3043 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3044 		num_pols = 1;
3045 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3046 						if_id);
3047 		err = xfrm_expand_policies(fl, family, pols,
3048 					   &num_pols, &num_xfrms);
3049 		if (err < 0)
3050 			goto dropdst;
3051 
3052 		if (num_pols) {
3053 			if (num_xfrms <= 0) {
3054 				drop_pols = num_pols;
3055 				goto no_transform;
3056 			}
3057 
3058 			xdst = xfrm_resolve_and_create_bundle(
3059 					pols, num_pols, fl,
3060 					family, dst_orig);
3061 
3062 			if (IS_ERR(xdst)) {
3063 				xfrm_pols_put(pols, num_pols);
3064 				err = PTR_ERR(xdst);
3065 				if (err == -EREMOTE)
3066 					goto nopol;
3067 
3068 				goto dropdst;
3069 			} else if (xdst == NULL) {
3070 				num_xfrms = 0;
3071 				drop_pols = num_pols;
3072 				goto no_transform;
3073 			}
3074 
3075 			route = xdst->route;
3076 		}
3077 	}
3078 
3079 	if (xdst == NULL) {
3080 		struct xfrm_flo xflo;
3081 
3082 		xflo.dst_orig = dst_orig;
3083 		xflo.flags = flags;
3084 
3085 		/* To accelerate a bit...  */
3086 		if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3087 			       !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3088 			goto nopol;
3089 
3090 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3091 		if (xdst == NULL)
3092 			goto nopol;
3093 		if (IS_ERR(xdst)) {
3094 			err = PTR_ERR(xdst);
3095 			goto dropdst;
3096 		}
3097 
3098 		num_pols = xdst->num_pols;
3099 		num_xfrms = xdst->num_xfrms;
3100 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3101 		route = xdst->route;
3102 	}
3103 
3104 	dst = &xdst->u.dst;
3105 	if (route == NULL && num_xfrms > 0) {
3106 		/* The only case when xfrm_bundle_lookup() returns a
3107 		 * bundle with null route, is when the template could
3108 		 * not be resolved. It means policies are there, but
3109 		 * bundle could not be created, since we don't yet
3110 		 * have the xfrm_state's. We need to wait for KM to
3111 		 * negotiate new SA's or bail out with error.*/
3112 		if (net->xfrm.sysctl_larval_drop) {
3113 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3114 			err = -EREMOTE;
3115 			goto error;
3116 		}
3117 
3118 		err = -EAGAIN;
3119 
3120 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3121 		goto error;
3122 	}
3123 
3124 no_transform:
3125 	if (num_pols == 0)
3126 		goto nopol;
3127 
3128 	if ((flags & XFRM_LOOKUP_ICMP) &&
3129 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3130 		err = -ENOENT;
3131 		goto error;
3132 	}
3133 
3134 	for (i = 0; i < num_pols; i++)
3135 		pols[i]->curlft.use_time = ktime_get_real_seconds();
3136 
3137 	if (num_xfrms < 0) {
3138 		/* Prohibit the flow */
3139 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3140 		err = -EPERM;
3141 		goto error;
3142 	} else if (num_xfrms > 0) {
3143 		/* Flow transformed */
3144 		dst_release(dst_orig);
3145 	} else {
3146 		/* Flow passes untransformed */
3147 		dst_release(dst);
3148 		dst = dst_orig;
3149 	}
3150 ok:
3151 	xfrm_pols_put(pols, drop_pols);
3152 	if (dst && dst->xfrm &&
3153 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3154 		dst->flags |= DST_XFRM_TUNNEL;
3155 	return dst;
3156 
3157 nopol:
3158 	if (!(dst_orig->dev->flags & IFF_LOOPBACK) &&
3159 	    !xfrm_default_allow(net, dir)) {
3160 		err = -EPERM;
3161 		goto error;
3162 	}
3163 	if (!(flags & XFRM_LOOKUP_ICMP)) {
3164 		dst = dst_orig;
3165 		goto ok;
3166 	}
3167 	err = -ENOENT;
3168 error:
3169 	dst_release(dst);
3170 dropdst:
3171 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3172 		dst_release(dst_orig);
3173 	xfrm_pols_put(pols, drop_pols);
3174 	return ERR_PTR(err);
3175 }
3176 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3177 
3178 /* Main function: finds/creates a bundle for given flow.
3179  *
3180  * At the moment we eat a raw IP route. Mostly to speed up lookups
3181  * on interfaces with disabled IPsec.
3182  */
3183 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3184 			      const struct flowi *fl, const struct sock *sk,
3185 			      int flags)
3186 {
3187 	return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3188 }
3189 EXPORT_SYMBOL(xfrm_lookup);
3190 
3191 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3192  * Otherwise we may send out blackholed packets.
3193  */
3194 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3195 				    const struct flowi *fl,
3196 				    const struct sock *sk, int flags)
3197 {
3198 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3199 					    flags | XFRM_LOOKUP_QUEUE |
3200 					    XFRM_LOOKUP_KEEP_DST_REF);
3201 
3202 	if (PTR_ERR(dst) == -EREMOTE)
3203 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
3204 
3205 	if (IS_ERR(dst))
3206 		dst_release(dst_orig);
3207 
3208 	return dst;
3209 }
3210 EXPORT_SYMBOL(xfrm_lookup_route);
3211 
3212 static inline int
3213 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3214 {
3215 	struct sec_path *sp = skb_sec_path(skb);
3216 	struct xfrm_state *x;
3217 
3218 	if (!sp || idx < 0 || idx >= sp->len)
3219 		return 0;
3220 	x = sp->xvec[idx];
3221 	if (!x->type->reject)
3222 		return 0;
3223 	return x->type->reject(x, skb, fl);
3224 }
3225 
3226 /* When skb is transformed back to its "native" form, we have to
3227  * check policy restrictions. At the moment we make this in maximally
3228  * stupid way. Shame on me. :-) Of course, connected sockets must
3229  * have policy cached at them.
3230  */
3231 
3232 static inline int
3233 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3234 	      unsigned short family)
3235 {
3236 	if (xfrm_state_kern(x))
3237 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3238 	return	x->id.proto == tmpl->id.proto &&
3239 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3240 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3241 		x->props.mode == tmpl->mode &&
3242 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3243 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3244 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
3245 		  xfrm_state_addr_cmp(tmpl, x, family));
3246 }
3247 
3248 /*
3249  * 0 or more than 0 is returned when validation is succeeded (either bypass
3250  * because of optional transport mode, or next index of the matched secpath
3251  * state with the template.
3252  * -1 is returned when no matching template is found.
3253  * Otherwise "-2 - errored_index" is returned.
3254  */
3255 static inline int
3256 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3257 	       unsigned short family)
3258 {
3259 	int idx = start;
3260 
3261 	if (tmpl->optional) {
3262 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
3263 			return start;
3264 	} else
3265 		start = -1;
3266 	for (; idx < sp->len; idx++) {
3267 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3268 			return ++idx;
3269 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3270 			if (start == -1)
3271 				start = -2-idx;
3272 			break;
3273 		}
3274 	}
3275 	return start;
3276 }
3277 
3278 static void
3279 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3280 {
3281 	const struct iphdr *iph = ip_hdr(skb);
3282 	int ihl = iph->ihl;
3283 	u8 *xprth = skb_network_header(skb) + ihl * 4;
3284 	struct flowi4 *fl4 = &fl->u.ip4;
3285 	int oif = 0;
3286 
3287 	if (skb_dst(skb) && skb_dst(skb)->dev)
3288 		oif = skb_dst(skb)->dev->ifindex;
3289 
3290 	memset(fl4, 0, sizeof(struct flowi4));
3291 	fl4->flowi4_mark = skb->mark;
3292 	fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3293 
3294 	fl4->flowi4_proto = iph->protocol;
3295 	fl4->daddr = reverse ? iph->saddr : iph->daddr;
3296 	fl4->saddr = reverse ? iph->daddr : iph->saddr;
3297 	fl4->flowi4_tos = iph->tos;
3298 
3299 	if (!ip_is_fragment(iph)) {
3300 		switch (iph->protocol) {
3301 		case IPPROTO_UDP:
3302 		case IPPROTO_UDPLITE:
3303 		case IPPROTO_TCP:
3304 		case IPPROTO_SCTP:
3305 		case IPPROTO_DCCP:
3306 			if (xprth + 4 < skb->data ||
3307 			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3308 				__be16 *ports;
3309 
3310 				xprth = skb_network_header(skb) + ihl * 4;
3311 				ports = (__be16 *)xprth;
3312 
3313 				fl4->fl4_sport = ports[!!reverse];
3314 				fl4->fl4_dport = ports[!reverse];
3315 			}
3316 			break;
3317 		case IPPROTO_ICMP:
3318 			if (xprth + 2 < skb->data ||
3319 			    pskb_may_pull(skb, xprth + 2 - skb->data)) {
3320 				u8 *icmp;
3321 
3322 				xprth = skb_network_header(skb) + ihl * 4;
3323 				icmp = xprth;
3324 
3325 				fl4->fl4_icmp_type = icmp[0];
3326 				fl4->fl4_icmp_code = icmp[1];
3327 			}
3328 			break;
3329 		case IPPROTO_GRE:
3330 			if (xprth + 12 < skb->data ||
3331 			    pskb_may_pull(skb, xprth + 12 - skb->data)) {
3332 				__be16 *greflags;
3333 				__be32 *gre_hdr;
3334 
3335 				xprth = skb_network_header(skb) + ihl * 4;
3336 				greflags = (__be16 *)xprth;
3337 				gre_hdr = (__be32 *)xprth;
3338 
3339 				if (greflags[0] & GRE_KEY) {
3340 					if (greflags[0] & GRE_CSUM)
3341 						gre_hdr++;
3342 					fl4->fl4_gre_key = gre_hdr[1];
3343 				}
3344 			}
3345 			break;
3346 		default:
3347 			break;
3348 		}
3349 	}
3350 }
3351 
3352 #if IS_ENABLED(CONFIG_IPV6)
3353 static void
3354 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3355 {
3356 	struct flowi6 *fl6 = &fl->u.ip6;
3357 	int onlyproto = 0;
3358 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
3359 	u32 offset = sizeof(*hdr);
3360 	struct ipv6_opt_hdr *exthdr;
3361 	const unsigned char *nh = skb_network_header(skb);
3362 	u16 nhoff = IP6CB(skb)->nhoff;
3363 	int oif = 0;
3364 	u8 nexthdr;
3365 
3366 	if (!nhoff)
3367 		nhoff = offsetof(struct ipv6hdr, nexthdr);
3368 
3369 	nexthdr = nh[nhoff];
3370 
3371 	if (skb_dst(skb) && skb_dst(skb)->dev)
3372 		oif = skb_dst(skb)->dev->ifindex;
3373 
3374 	memset(fl6, 0, sizeof(struct flowi6));
3375 	fl6->flowi6_mark = skb->mark;
3376 	fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3377 
3378 	fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3379 	fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3380 
3381 	while (nh + offset + sizeof(*exthdr) < skb->data ||
3382 	       pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3383 		nh = skb_network_header(skb);
3384 		exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3385 
3386 		switch (nexthdr) {
3387 		case NEXTHDR_FRAGMENT:
3388 			onlyproto = 1;
3389 			fallthrough;
3390 		case NEXTHDR_ROUTING:
3391 		case NEXTHDR_HOP:
3392 		case NEXTHDR_DEST:
3393 			offset += ipv6_optlen(exthdr);
3394 			nexthdr = exthdr->nexthdr;
3395 			exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3396 			break;
3397 		case IPPROTO_UDP:
3398 		case IPPROTO_UDPLITE:
3399 		case IPPROTO_TCP:
3400 		case IPPROTO_SCTP:
3401 		case IPPROTO_DCCP:
3402 			if (!onlyproto && (nh + offset + 4 < skb->data ||
3403 			     pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3404 				__be16 *ports;
3405 
3406 				nh = skb_network_header(skb);
3407 				ports = (__be16 *)(nh + offset);
3408 				fl6->fl6_sport = ports[!!reverse];
3409 				fl6->fl6_dport = ports[!reverse];
3410 			}
3411 			fl6->flowi6_proto = nexthdr;
3412 			return;
3413 		case IPPROTO_ICMPV6:
3414 			if (!onlyproto && (nh + offset + 2 < skb->data ||
3415 			    pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3416 				u8 *icmp;
3417 
3418 				nh = skb_network_header(skb);
3419 				icmp = (u8 *)(nh + offset);
3420 				fl6->fl6_icmp_type = icmp[0];
3421 				fl6->fl6_icmp_code = icmp[1];
3422 			}
3423 			fl6->flowi6_proto = nexthdr;
3424 			return;
3425 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3426 		case IPPROTO_MH:
3427 			offset += ipv6_optlen(exthdr);
3428 			if (!onlyproto && (nh + offset + 3 < skb->data ||
3429 			    pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3430 				struct ip6_mh *mh;
3431 
3432 				nh = skb_network_header(skb);
3433 				mh = (struct ip6_mh *)(nh + offset);
3434 				fl6->fl6_mh_type = mh->ip6mh_type;
3435 			}
3436 			fl6->flowi6_proto = nexthdr;
3437 			return;
3438 #endif
3439 		default:
3440 			fl6->flowi6_proto = nexthdr;
3441 			return;
3442 		}
3443 	}
3444 }
3445 #endif
3446 
3447 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3448 			  unsigned int family, int reverse)
3449 {
3450 	switch (family) {
3451 	case AF_INET:
3452 		decode_session4(skb, fl, reverse);
3453 		break;
3454 #if IS_ENABLED(CONFIG_IPV6)
3455 	case AF_INET6:
3456 		decode_session6(skb, fl, reverse);
3457 		break;
3458 #endif
3459 	default:
3460 		return -EAFNOSUPPORT;
3461 	}
3462 
3463 	return security_xfrm_decode_session(skb, &fl->flowi_secid);
3464 }
3465 EXPORT_SYMBOL(__xfrm_decode_session);
3466 
3467 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3468 {
3469 	for (; k < sp->len; k++) {
3470 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3471 			*idxp = k;
3472 			return 1;
3473 		}
3474 	}
3475 
3476 	return 0;
3477 }
3478 
3479 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3480 			unsigned short family)
3481 {
3482 	struct net *net = dev_net(skb->dev);
3483 	struct xfrm_policy *pol;
3484 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3485 	int npols = 0;
3486 	int xfrm_nr;
3487 	int pi;
3488 	int reverse;
3489 	struct flowi fl;
3490 	int xerr_idx = -1;
3491 	const struct xfrm_if_cb *ifcb;
3492 	struct sec_path *sp;
3493 	struct xfrm_if *xi;
3494 	u32 if_id = 0;
3495 
3496 	rcu_read_lock();
3497 	ifcb = xfrm_if_get_cb();
3498 
3499 	if (ifcb) {
3500 		xi = ifcb->decode_session(skb, family);
3501 		if (xi) {
3502 			if_id = xi->p.if_id;
3503 			net = xi->net;
3504 		}
3505 	}
3506 	rcu_read_unlock();
3507 
3508 	reverse = dir & ~XFRM_POLICY_MASK;
3509 	dir &= XFRM_POLICY_MASK;
3510 
3511 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3512 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3513 		return 0;
3514 	}
3515 
3516 	nf_nat_decode_session(skb, &fl, family);
3517 
3518 	/* First, check used SA against their selectors. */
3519 	sp = skb_sec_path(skb);
3520 	if (sp) {
3521 		int i;
3522 
3523 		for (i = sp->len - 1; i >= 0; i--) {
3524 			struct xfrm_state *x = sp->xvec[i];
3525 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
3526 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3527 				return 0;
3528 			}
3529 		}
3530 	}
3531 
3532 	pol = NULL;
3533 	sk = sk_to_full_sk(sk);
3534 	if (sk && sk->sk_policy[dir]) {
3535 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3536 		if (IS_ERR(pol)) {
3537 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3538 			return 0;
3539 		}
3540 	}
3541 
3542 	if (!pol)
3543 		pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3544 
3545 	if (IS_ERR(pol)) {
3546 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3547 		return 0;
3548 	}
3549 
3550 	if (!pol) {
3551 		if (!xfrm_default_allow(net, dir)) {
3552 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3553 			return 0;
3554 		}
3555 
3556 		if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3557 			xfrm_secpath_reject(xerr_idx, skb, &fl);
3558 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3559 			return 0;
3560 		}
3561 		return 1;
3562 	}
3563 
3564 	pol->curlft.use_time = ktime_get_real_seconds();
3565 
3566 	pols[0] = pol;
3567 	npols++;
3568 #ifdef CONFIG_XFRM_SUB_POLICY
3569 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3570 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3571 						    &fl, family,
3572 						    XFRM_POLICY_IN, if_id);
3573 		if (pols[1]) {
3574 			if (IS_ERR(pols[1])) {
3575 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3576 				return 0;
3577 			}
3578 			pols[1]->curlft.use_time = ktime_get_real_seconds();
3579 			npols++;
3580 		}
3581 	}
3582 #endif
3583 
3584 	if (pol->action == XFRM_POLICY_ALLOW) {
3585 		static struct sec_path dummy;
3586 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3587 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3588 		struct xfrm_tmpl **tpp = tp;
3589 		int ti = 0;
3590 		int i, k;
3591 
3592 		sp = skb_sec_path(skb);
3593 		if (!sp)
3594 			sp = &dummy;
3595 
3596 		for (pi = 0; pi < npols; pi++) {
3597 			if (pols[pi] != pol &&
3598 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
3599 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3600 				goto reject;
3601 			}
3602 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3603 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3604 				goto reject_error;
3605 			}
3606 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
3607 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
3608 		}
3609 		xfrm_nr = ti;
3610 
3611 		if (!xfrm_default_allow(net, dir) && !xfrm_nr) {
3612 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
3613 			goto reject;
3614 		}
3615 
3616 		if (npols > 1) {
3617 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3618 			tpp = stp;
3619 		}
3620 
3621 		/* For each tunnel xfrm, find the first matching tmpl.
3622 		 * For each tmpl before that, find corresponding xfrm.
3623 		 * Order is _important_. Later we will implement
3624 		 * some barriers, but at the moment barriers
3625 		 * are implied between each two transformations.
3626 		 */
3627 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3628 			k = xfrm_policy_ok(tpp[i], sp, k, family);
3629 			if (k < 0) {
3630 				if (k < -1)
3631 					/* "-2 - errored_index" returned */
3632 					xerr_idx = -(2+k);
3633 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3634 				goto reject;
3635 			}
3636 		}
3637 
3638 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3639 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3640 			goto reject;
3641 		}
3642 
3643 		xfrm_pols_put(pols, npols);
3644 		return 1;
3645 	}
3646 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3647 
3648 reject:
3649 	xfrm_secpath_reject(xerr_idx, skb, &fl);
3650 reject_error:
3651 	xfrm_pols_put(pols, npols);
3652 	return 0;
3653 }
3654 EXPORT_SYMBOL(__xfrm_policy_check);
3655 
3656 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3657 {
3658 	struct net *net = dev_net(skb->dev);
3659 	struct flowi fl;
3660 	struct dst_entry *dst;
3661 	int res = 1;
3662 
3663 	if (xfrm_decode_session(skb, &fl, family) < 0) {
3664 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3665 		return 0;
3666 	}
3667 
3668 	skb_dst_force(skb);
3669 	if (!skb_dst(skb)) {
3670 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3671 		return 0;
3672 	}
3673 
3674 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3675 	if (IS_ERR(dst)) {
3676 		res = 0;
3677 		dst = NULL;
3678 	}
3679 	skb_dst_set(skb, dst);
3680 	return res;
3681 }
3682 EXPORT_SYMBOL(__xfrm_route_forward);
3683 
3684 /* Optimize later using cookies and generation ids. */
3685 
3686 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3687 {
3688 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3689 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3690 	 * get validated by dst_ops->check on every use.  We do this
3691 	 * because when a normal route referenced by an XFRM dst is
3692 	 * obsoleted we do not go looking around for all parent
3693 	 * referencing XFRM dsts so that we can invalidate them.  It
3694 	 * is just too much work.  Instead we make the checks here on
3695 	 * every use.  For example:
3696 	 *
3697 	 *	XFRM dst A --> IPv4 dst X
3698 	 *
3699 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
3700 	 * in this example).  If X is marked obsolete, "A" will not
3701 	 * notice.  That's what we are validating here via the
3702 	 * stale_bundle() check.
3703 	 *
3704 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3705 	 * be marked on it.
3706 	 * This will force stale_bundle() to fail on any xdst bundle with
3707 	 * this dst linked in it.
3708 	 */
3709 	if (dst->obsolete < 0 && !stale_bundle(dst))
3710 		return dst;
3711 
3712 	return NULL;
3713 }
3714 
3715 static int stale_bundle(struct dst_entry *dst)
3716 {
3717 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3718 }
3719 
3720 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3721 {
3722 	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3723 		dst->dev = dev_net(dev)->loopback_dev;
3724 		dev_hold(dst->dev);
3725 		dev_put(dev);
3726 	}
3727 }
3728 EXPORT_SYMBOL(xfrm_dst_ifdown);
3729 
3730 static void xfrm_link_failure(struct sk_buff *skb)
3731 {
3732 	/* Impossible. Such dst must be popped before reaches point of failure. */
3733 }
3734 
3735 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3736 {
3737 	if (dst) {
3738 		if (dst->obsolete) {
3739 			dst_release(dst);
3740 			dst = NULL;
3741 		}
3742 	}
3743 	return dst;
3744 }
3745 
3746 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3747 {
3748 	while (nr--) {
3749 		struct xfrm_dst *xdst = bundle[nr];
3750 		u32 pmtu, route_mtu_cached;
3751 		struct dst_entry *dst;
3752 
3753 		dst = &xdst->u.dst;
3754 		pmtu = dst_mtu(xfrm_dst_child(dst));
3755 		xdst->child_mtu_cached = pmtu;
3756 
3757 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3758 
3759 		route_mtu_cached = dst_mtu(xdst->route);
3760 		xdst->route_mtu_cached = route_mtu_cached;
3761 
3762 		if (pmtu > route_mtu_cached)
3763 			pmtu = route_mtu_cached;
3764 
3765 		dst_metric_set(dst, RTAX_MTU, pmtu);
3766 	}
3767 }
3768 
3769 /* Check that the bundle accepts the flow and its components are
3770  * still valid.
3771  */
3772 
3773 static int xfrm_bundle_ok(struct xfrm_dst *first)
3774 {
3775 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3776 	struct dst_entry *dst = &first->u.dst;
3777 	struct xfrm_dst *xdst;
3778 	int start_from, nr;
3779 	u32 mtu;
3780 
3781 	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3782 	    (dst->dev && !netif_running(dst->dev)))
3783 		return 0;
3784 
3785 	if (dst->flags & DST_XFRM_QUEUE)
3786 		return 1;
3787 
3788 	start_from = nr = 0;
3789 	do {
3790 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3791 
3792 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
3793 			return 0;
3794 		if (xdst->xfrm_genid != dst->xfrm->genid)
3795 			return 0;
3796 		if (xdst->num_pols > 0 &&
3797 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3798 			return 0;
3799 
3800 		bundle[nr++] = xdst;
3801 
3802 		mtu = dst_mtu(xfrm_dst_child(dst));
3803 		if (xdst->child_mtu_cached != mtu) {
3804 			start_from = nr;
3805 			xdst->child_mtu_cached = mtu;
3806 		}
3807 
3808 		if (!dst_check(xdst->route, xdst->route_cookie))
3809 			return 0;
3810 		mtu = dst_mtu(xdst->route);
3811 		if (xdst->route_mtu_cached != mtu) {
3812 			start_from = nr;
3813 			xdst->route_mtu_cached = mtu;
3814 		}
3815 
3816 		dst = xfrm_dst_child(dst);
3817 	} while (dst->xfrm);
3818 
3819 	if (likely(!start_from))
3820 		return 1;
3821 
3822 	xdst = bundle[start_from - 1];
3823 	mtu = xdst->child_mtu_cached;
3824 	while (start_from--) {
3825 		dst = &xdst->u.dst;
3826 
3827 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
3828 		if (mtu > xdst->route_mtu_cached)
3829 			mtu = xdst->route_mtu_cached;
3830 		dst_metric_set(dst, RTAX_MTU, mtu);
3831 		if (!start_from)
3832 			break;
3833 
3834 		xdst = bundle[start_from - 1];
3835 		xdst->child_mtu_cached = mtu;
3836 	}
3837 
3838 	return 1;
3839 }
3840 
3841 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3842 {
3843 	return dst_metric_advmss(xfrm_dst_path(dst));
3844 }
3845 
3846 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3847 {
3848 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3849 
3850 	return mtu ? : dst_mtu(xfrm_dst_path(dst));
3851 }
3852 
3853 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3854 					const void *daddr)
3855 {
3856 	while (dst->xfrm) {
3857 		const struct xfrm_state *xfrm = dst->xfrm;
3858 
3859 		dst = xfrm_dst_child(dst);
3860 
3861 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3862 			continue;
3863 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3864 			daddr = xfrm->coaddr;
3865 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3866 			daddr = &xfrm->id.daddr;
3867 	}
3868 	return daddr;
3869 }
3870 
3871 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3872 					   struct sk_buff *skb,
3873 					   const void *daddr)
3874 {
3875 	const struct dst_entry *path = xfrm_dst_path(dst);
3876 
3877 	if (!skb)
3878 		daddr = xfrm_get_dst_nexthop(dst, daddr);
3879 	return path->ops->neigh_lookup(path, skb, daddr);
3880 }
3881 
3882 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3883 {
3884 	const struct dst_entry *path = xfrm_dst_path(dst);
3885 
3886 	daddr = xfrm_get_dst_nexthop(dst, daddr);
3887 	path->ops->confirm_neigh(path, daddr);
3888 }
3889 
3890 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3891 {
3892 	int err = 0;
3893 
3894 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3895 		return -EAFNOSUPPORT;
3896 
3897 	spin_lock(&xfrm_policy_afinfo_lock);
3898 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
3899 		err = -EEXIST;
3900 	else {
3901 		struct dst_ops *dst_ops = afinfo->dst_ops;
3902 		if (likely(dst_ops->kmem_cachep == NULL))
3903 			dst_ops->kmem_cachep = xfrm_dst_cache;
3904 		if (likely(dst_ops->check == NULL))
3905 			dst_ops->check = xfrm_dst_check;
3906 		if (likely(dst_ops->default_advmss == NULL))
3907 			dst_ops->default_advmss = xfrm_default_advmss;
3908 		if (likely(dst_ops->mtu == NULL))
3909 			dst_ops->mtu = xfrm_mtu;
3910 		if (likely(dst_ops->negative_advice == NULL))
3911 			dst_ops->negative_advice = xfrm_negative_advice;
3912 		if (likely(dst_ops->link_failure == NULL))
3913 			dst_ops->link_failure = xfrm_link_failure;
3914 		if (likely(dst_ops->neigh_lookup == NULL))
3915 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
3916 		if (likely(!dst_ops->confirm_neigh))
3917 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
3918 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3919 	}
3920 	spin_unlock(&xfrm_policy_afinfo_lock);
3921 
3922 	return err;
3923 }
3924 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3925 
3926 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3927 {
3928 	struct dst_ops *dst_ops = afinfo->dst_ops;
3929 	int i;
3930 
3931 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3932 		if (xfrm_policy_afinfo[i] != afinfo)
3933 			continue;
3934 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3935 		break;
3936 	}
3937 
3938 	synchronize_rcu();
3939 
3940 	dst_ops->kmem_cachep = NULL;
3941 	dst_ops->check = NULL;
3942 	dst_ops->negative_advice = NULL;
3943 	dst_ops->link_failure = NULL;
3944 }
3945 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3946 
3947 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3948 {
3949 	spin_lock(&xfrm_if_cb_lock);
3950 	rcu_assign_pointer(xfrm_if_cb, ifcb);
3951 	spin_unlock(&xfrm_if_cb_lock);
3952 }
3953 EXPORT_SYMBOL(xfrm_if_register_cb);
3954 
3955 void xfrm_if_unregister_cb(void)
3956 {
3957 	RCU_INIT_POINTER(xfrm_if_cb, NULL);
3958 	synchronize_rcu();
3959 }
3960 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3961 
3962 #ifdef CONFIG_XFRM_STATISTICS
3963 static int __net_init xfrm_statistics_init(struct net *net)
3964 {
3965 	int rv;
3966 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3967 	if (!net->mib.xfrm_statistics)
3968 		return -ENOMEM;
3969 	rv = xfrm_proc_init(net);
3970 	if (rv < 0)
3971 		free_percpu(net->mib.xfrm_statistics);
3972 	return rv;
3973 }
3974 
3975 static void xfrm_statistics_fini(struct net *net)
3976 {
3977 	xfrm_proc_fini(net);
3978 	free_percpu(net->mib.xfrm_statistics);
3979 }
3980 #else
3981 static int __net_init xfrm_statistics_init(struct net *net)
3982 {
3983 	return 0;
3984 }
3985 
3986 static void xfrm_statistics_fini(struct net *net)
3987 {
3988 }
3989 #endif
3990 
3991 static int __net_init xfrm_policy_init(struct net *net)
3992 {
3993 	unsigned int hmask, sz;
3994 	int dir, err;
3995 
3996 	if (net_eq(net, &init_net)) {
3997 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
3998 					   sizeof(struct xfrm_dst),
3999 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4000 					   NULL);
4001 		err = rhashtable_init(&xfrm_policy_inexact_table,
4002 				      &xfrm_pol_inexact_params);
4003 		BUG_ON(err);
4004 	}
4005 
4006 	hmask = 8 - 1;
4007 	sz = (hmask+1) * sizeof(struct hlist_head);
4008 
4009 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4010 	if (!net->xfrm.policy_byidx)
4011 		goto out_byidx;
4012 	net->xfrm.policy_idx_hmask = hmask;
4013 
4014 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4015 		struct xfrm_policy_hash *htab;
4016 
4017 		net->xfrm.policy_count[dir] = 0;
4018 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4019 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4020 
4021 		htab = &net->xfrm.policy_bydst[dir];
4022 		htab->table = xfrm_hash_alloc(sz);
4023 		if (!htab->table)
4024 			goto out_bydst;
4025 		htab->hmask = hmask;
4026 		htab->dbits4 = 32;
4027 		htab->sbits4 = 32;
4028 		htab->dbits6 = 128;
4029 		htab->sbits6 = 128;
4030 	}
4031 	net->xfrm.policy_hthresh.lbits4 = 32;
4032 	net->xfrm.policy_hthresh.rbits4 = 32;
4033 	net->xfrm.policy_hthresh.lbits6 = 128;
4034 	net->xfrm.policy_hthresh.rbits6 = 128;
4035 
4036 	seqlock_init(&net->xfrm.policy_hthresh.lock);
4037 
4038 	INIT_LIST_HEAD(&net->xfrm.policy_all);
4039 	INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4040 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4041 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4042 	return 0;
4043 
4044 out_bydst:
4045 	for (dir--; dir >= 0; dir--) {
4046 		struct xfrm_policy_hash *htab;
4047 
4048 		htab = &net->xfrm.policy_bydst[dir];
4049 		xfrm_hash_free(htab->table, sz);
4050 	}
4051 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4052 out_byidx:
4053 	return -ENOMEM;
4054 }
4055 
4056 static void xfrm_policy_fini(struct net *net)
4057 {
4058 	struct xfrm_pol_inexact_bin *b, *t;
4059 	unsigned int sz;
4060 	int dir;
4061 
4062 	flush_work(&net->xfrm.policy_hash_work);
4063 #ifdef CONFIG_XFRM_SUB_POLICY
4064 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4065 #endif
4066 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4067 
4068 	WARN_ON(!list_empty(&net->xfrm.policy_all));
4069 
4070 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4071 		struct xfrm_policy_hash *htab;
4072 
4073 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4074 
4075 		htab = &net->xfrm.policy_bydst[dir];
4076 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4077 		WARN_ON(!hlist_empty(htab->table));
4078 		xfrm_hash_free(htab->table, sz);
4079 	}
4080 
4081 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4082 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4083 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4084 
4085 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4086 	list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4087 		__xfrm_policy_inexact_prune_bin(b, true);
4088 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4089 }
4090 
4091 static int __net_init xfrm_net_init(struct net *net)
4092 {
4093 	int rv;
4094 
4095 	/* Initialize the per-net locks here */
4096 	spin_lock_init(&net->xfrm.xfrm_state_lock);
4097 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
4098 	seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4099 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
4100 
4101 	rv = xfrm_statistics_init(net);
4102 	if (rv < 0)
4103 		goto out_statistics;
4104 	rv = xfrm_state_init(net);
4105 	if (rv < 0)
4106 		goto out_state;
4107 	rv = xfrm_policy_init(net);
4108 	if (rv < 0)
4109 		goto out_policy;
4110 	rv = xfrm_sysctl_init(net);
4111 	if (rv < 0)
4112 		goto out_sysctl;
4113 
4114 	return 0;
4115 
4116 out_sysctl:
4117 	xfrm_policy_fini(net);
4118 out_policy:
4119 	xfrm_state_fini(net);
4120 out_state:
4121 	xfrm_statistics_fini(net);
4122 out_statistics:
4123 	return rv;
4124 }
4125 
4126 static void __net_exit xfrm_net_exit(struct net *net)
4127 {
4128 	xfrm_sysctl_fini(net);
4129 	xfrm_policy_fini(net);
4130 	xfrm_state_fini(net);
4131 	xfrm_statistics_fini(net);
4132 }
4133 
4134 static struct pernet_operations __net_initdata xfrm_net_ops = {
4135 	.init = xfrm_net_init,
4136 	.exit = xfrm_net_exit,
4137 };
4138 
4139 void __init xfrm_init(void)
4140 {
4141 	register_pernet_subsys(&xfrm_net_ops);
4142 	xfrm_dev_init();
4143 	xfrm_input_init();
4144 
4145 #ifdef CONFIG_XFRM_ESPINTCP
4146 	espintcp_init();
4147 #endif
4148 }
4149 
4150 #ifdef CONFIG_AUDITSYSCALL
4151 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4152 					 struct audit_buffer *audit_buf)
4153 {
4154 	struct xfrm_sec_ctx *ctx = xp->security;
4155 	struct xfrm_selector *sel = &xp->selector;
4156 
4157 	if (ctx)
4158 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4159 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4160 
4161 	switch (sel->family) {
4162 	case AF_INET:
4163 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4164 		if (sel->prefixlen_s != 32)
4165 			audit_log_format(audit_buf, " src_prefixlen=%d",
4166 					 sel->prefixlen_s);
4167 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4168 		if (sel->prefixlen_d != 32)
4169 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4170 					 sel->prefixlen_d);
4171 		break;
4172 	case AF_INET6:
4173 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4174 		if (sel->prefixlen_s != 128)
4175 			audit_log_format(audit_buf, " src_prefixlen=%d",
4176 					 sel->prefixlen_s);
4177 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4178 		if (sel->prefixlen_d != 128)
4179 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4180 					 sel->prefixlen_d);
4181 		break;
4182 	}
4183 }
4184 
4185 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4186 {
4187 	struct audit_buffer *audit_buf;
4188 
4189 	audit_buf = xfrm_audit_start("SPD-add");
4190 	if (audit_buf == NULL)
4191 		return;
4192 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4193 	audit_log_format(audit_buf, " res=%u", result);
4194 	xfrm_audit_common_policyinfo(xp, audit_buf);
4195 	audit_log_end(audit_buf);
4196 }
4197 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4198 
4199 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4200 			      bool task_valid)
4201 {
4202 	struct audit_buffer *audit_buf;
4203 
4204 	audit_buf = xfrm_audit_start("SPD-delete");
4205 	if (audit_buf == NULL)
4206 		return;
4207 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4208 	audit_log_format(audit_buf, " res=%u", result);
4209 	xfrm_audit_common_policyinfo(xp, audit_buf);
4210 	audit_log_end(audit_buf);
4211 }
4212 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4213 #endif
4214 
4215 #ifdef CONFIG_XFRM_MIGRATE
4216 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4217 					const struct xfrm_selector *sel_tgt)
4218 {
4219 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4220 		if (sel_tgt->family == sel_cmp->family &&
4221 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4222 				    sel_cmp->family) &&
4223 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4224 				    sel_cmp->family) &&
4225 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4226 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4227 			return true;
4228 		}
4229 	} else {
4230 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4231 			return true;
4232 		}
4233 	}
4234 	return false;
4235 }
4236 
4237 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4238 						    u8 dir, u8 type, struct net *net)
4239 {
4240 	struct xfrm_policy *pol, *ret = NULL;
4241 	struct hlist_head *chain;
4242 	u32 priority = ~0U;
4243 
4244 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4245 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4246 	hlist_for_each_entry(pol, chain, bydst) {
4247 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4248 		    pol->type == type) {
4249 			ret = pol;
4250 			priority = ret->priority;
4251 			break;
4252 		}
4253 	}
4254 	chain = &net->xfrm.policy_inexact[dir];
4255 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4256 		if ((pol->priority >= priority) && ret)
4257 			break;
4258 
4259 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4260 		    pol->type == type) {
4261 			ret = pol;
4262 			break;
4263 		}
4264 	}
4265 
4266 	xfrm_pol_hold(ret);
4267 
4268 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4269 
4270 	return ret;
4271 }
4272 
4273 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4274 {
4275 	int match = 0;
4276 
4277 	if (t->mode == m->mode && t->id.proto == m->proto &&
4278 	    (m->reqid == 0 || t->reqid == m->reqid)) {
4279 		switch (t->mode) {
4280 		case XFRM_MODE_TUNNEL:
4281 		case XFRM_MODE_BEET:
4282 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4283 					    m->old_family) &&
4284 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
4285 					    m->old_family)) {
4286 				match = 1;
4287 			}
4288 			break;
4289 		case XFRM_MODE_TRANSPORT:
4290 			/* in case of transport mode, template does not store
4291 			   any IP addresses, hence we just compare mode and
4292 			   protocol */
4293 			match = 1;
4294 			break;
4295 		default:
4296 			break;
4297 		}
4298 	}
4299 	return match;
4300 }
4301 
4302 /* update endpoint address(es) of template(s) */
4303 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4304 			       struct xfrm_migrate *m, int num_migrate)
4305 {
4306 	struct xfrm_migrate *mp;
4307 	int i, j, n = 0;
4308 
4309 	write_lock_bh(&pol->lock);
4310 	if (unlikely(pol->walk.dead)) {
4311 		/* target policy has been deleted */
4312 		write_unlock_bh(&pol->lock);
4313 		return -ENOENT;
4314 	}
4315 
4316 	for (i = 0; i < pol->xfrm_nr; i++) {
4317 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4318 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4319 				continue;
4320 			n++;
4321 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4322 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4323 				continue;
4324 			/* update endpoints */
4325 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4326 			       sizeof(pol->xfrm_vec[i].id.daddr));
4327 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4328 			       sizeof(pol->xfrm_vec[i].saddr));
4329 			pol->xfrm_vec[i].encap_family = mp->new_family;
4330 			/* flush bundles */
4331 			atomic_inc(&pol->genid);
4332 		}
4333 	}
4334 
4335 	write_unlock_bh(&pol->lock);
4336 
4337 	if (!n)
4338 		return -ENODATA;
4339 
4340 	return 0;
4341 }
4342 
4343 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4344 {
4345 	int i, j;
4346 
4347 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4348 		return -EINVAL;
4349 
4350 	for (i = 0; i < num_migrate; i++) {
4351 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4352 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4353 			return -EINVAL;
4354 
4355 		/* check if there is any duplicated entry */
4356 		for (j = i + 1; j < num_migrate; j++) {
4357 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4358 				    sizeof(m[i].old_daddr)) &&
4359 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4360 				    sizeof(m[i].old_saddr)) &&
4361 			    m[i].proto == m[j].proto &&
4362 			    m[i].mode == m[j].mode &&
4363 			    m[i].reqid == m[j].reqid &&
4364 			    m[i].old_family == m[j].old_family)
4365 				return -EINVAL;
4366 		}
4367 	}
4368 
4369 	return 0;
4370 }
4371 
4372 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4373 		 struct xfrm_migrate *m, int num_migrate,
4374 		 struct xfrm_kmaddress *k, struct net *net,
4375 		 struct xfrm_encap_tmpl *encap)
4376 {
4377 	int i, err, nx_cur = 0, nx_new = 0;
4378 	struct xfrm_policy *pol = NULL;
4379 	struct xfrm_state *x, *xc;
4380 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4381 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4382 	struct xfrm_migrate *mp;
4383 
4384 	/* Stage 0 - sanity checks */
4385 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4386 		goto out;
4387 
4388 	if (dir >= XFRM_POLICY_MAX) {
4389 		err = -EINVAL;
4390 		goto out;
4391 	}
4392 
4393 	/* Stage 1 - find policy */
4394 	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4395 		err = -ENOENT;
4396 		goto out;
4397 	}
4398 
4399 	/* Stage 2 - find and update state(s) */
4400 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4401 		if ((x = xfrm_migrate_state_find(mp, net))) {
4402 			x_cur[nx_cur] = x;
4403 			nx_cur++;
4404 			xc = xfrm_state_migrate(x, mp, encap);
4405 			if (xc) {
4406 				x_new[nx_new] = xc;
4407 				nx_new++;
4408 			} else {
4409 				err = -ENODATA;
4410 				goto restore_state;
4411 			}
4412 		}
4413 	}
4414 
4415 	/* Stage 3 - update policy */
4416 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4417 		goto restore_state;
4418 
4419 	/* Stage 4 - delete old state(s) */
4420 	if (nx_cur) {
4421 		xfrm_states_put(x_cur, nx_cur);
4422 		xfrm_states_delete(x_cur, nx_cur);
4423 	}
4424 
4425 	/* Stage 5 - announce */
4426 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
4427 
4428 	xfrm_pol_put(pol);
4429 
4430 	return 0;
4431 out:
4432 	return err;
4433 
4434 restore_state:
4435 	if (pol)
4436 		xfrm_pol_put(pol);
4437 	if (nx_cur)
4438 		xfrm_states_put(x_cur, nx_cur);
4439 	if (nx_new)
4440 		xfrm_states_delete(x_new, nx_new);
4441 
4442 	return err;
4443 }
4444 EXPORT_SYMBOL(xfrm_migrate);
4445 #endif
4446