xref: /linux/net/xfrm/xfrm_policy.c (revision 2363088eba2ecccfb643725e4864af73c4226a04)
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/inet_ecn.h>
35 #include <net/xfrm.h>
36 #include <net/ip.h>
37 #include <net/gre.h>
38 #if IS_ENABLED(CONFIG_IPV6_MIP6)
39 #include <net/mip6.h>
40 #endif
41 #ifdef CONFIG_XFRM_STATISTICS
42 #include <net/snmp.h>
43 #endif
44 #ifdef CONFIG_XFRM_ESPINTCP
45 #include <net/espintcp.h>
46 #endif
47 
48 #include "xfrm_hash.h"
49 
50 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
51 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
52 #define XFRM_MAX_QUEUE_LEN	100
53 
54 struct xfrm_flo {
55 	struct dst_entry *dst_orig;
56 	u8 flags;
57 };
58 
59 /* prefixes smaller than this are stored in lists, not trees. */
60 #define INEXACT_PREFIXLEN_IPV4	16
61 #define INEXACT_PREFIXLEN_IPV6	48
62 
63 struct xfrm_pol_inexact_node {
64 	struct rb_node node;
65 	union {
66 		xfrm_address_t addr;
67 		struct rcu_head rcu;
68 	};
69 	u8 prefixlen;
70 
71 	struct rb_root root;
72 
73 	/* the policies matching this node, can be empty list */
74 	struct hlist_head hhead;
75 };
76 
77 /* xfrm inexact policy search tree:
78  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
79  *  |
80  * +---- root_d: sorted by daddr:prefix
81  * |                 |
82  * |        xfrm_pol_inexact_node
83  * |                 |
84  * |                 +- root: sorted by saddr/prefix
85  * |                 |              |
86  * |                 |         xfrm_pol_inexact_node
87  * |                 |              |
88  * |                 |              + root: unused
89  * |                 |              |
90  * |                 |              + hhead: saddr:daddr policies
91  * |                 |
92  * |                 +- coarse policies and all any:daddr policies
93  * |
94  * +---- root_s: sorted by saddr:prefix
95  * |                 |
96  * |        xfrm_pol_inexact_node
97  * |                 |
98  * |                 + root: unused
99  * |                 |
100  * |                 + hhead: saddr:any policies
101  * |
102  * +---- coarse policies and all any:any policies
103  *
104  * Lookups return four candidate lists:
105  * 1. any:any list from top-level xfrm_pol_inexact_bin
106  * 2. any:daddr list from daddr tree
107  * 3. saddr:daddr list from 2nd level daddr tree
108  * 4. saddr:any list from saddr tree
109  *
110  * This result set then needs to be searched for the policy with
111  * the lowest priority.  If two results have same prio, youngest one wins.
112  */
113 
114 struct xfrm_pol_inexact_key {
115 	possible_net_t net;
116 	u32 if_id;
117 	u16 family;
118 	u8 dir, type;
119 };
120 
121 struct xfrm_pol_inexact_bin {
122 	struct xfrm_pol_inexact_key k;
123 	struct rhash_head head;
124 	/* list containing '*:*' policies */
125 	struct hlist_head hhead;
126 
127 	seqcount_spinlock_t count;
128 	/* tree sorted by daddr/prefix */
129 	struct rb_root root_d;
130 
131 	/* tree sorted by saddr/prefix */
132 	struct rb_root root_s;
133 
134 	/* slow path below */
135 	struct list_head inexact_bins;
136 	struct rcu_head rcu;
137 };
138 
139 enum xfrm_pol_inexact_candidate_type {
140 	XFRM_POL_CAND_BOTH,
141 	XFRM_POL_CAND_SADDR,
142 	XFRM_POL_CAND_DADDR,
143 	XFRM_POL_CAND_ANY,
144 
145 	XFRM_POL_CAND_MAX,
146 };
147 
148 struct xfrm_pol_inexact_candidates {
149 	struct hlist_head *res[XFRM_POL_CAND_MAX];
150 };
151 
152 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
153 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
154 
155 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
156 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
157 						__read_mostly;
158 
159 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
160 
161 static struct rhashtable xfrm_policy_inexact_table;
162 static const struct rhashtable_params xfrm_pol_inexact_params;
163 
164 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
165 static int stale_bundle(struct dst_entry *dst);
166 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
167 static void xfrm_policy_queue_process(struct timer_list *t);
168 
169 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
170 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
171 						int dir);
172 
173 static struct xfrm_pol_inexact_bin *
174 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
175 			   u32 if_id);
176 
177 static struct xfrm_pol_inexact_bin *
178 xfrm_policy_inexact_lookup_rcu(struct net *net,
179 			       u8 type, u16 family, u8 dir, u32 if_id);
180 static struct xfrm_policy *
181 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
182 			bool excl);
183 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
184 					    struct xfrm_policy *policy);
185 
186 static bool
187 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
188 				    struct xfrm_pol_inexact_bin *b,
189 				    const xfrm_address_t *saddr,
190 				    const xfrm_address_t *daddr);
191 
192 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
193 {
194 	return refcount_inc_not_zero(&policy->refcnt);
195 }
196 
197 static inline bool
198 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
199 {
200 	const struct flowi4 *fl4 = &fl->u.ip4;
201 
202 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
203 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
204 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
205 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
206 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
207 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
208 }
209 
210 static inline bool
211 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
212 {
213 	const struct flowi6 *fl6 = &fl->u.ip6;
214 
215 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
216 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
217 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
218 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
219 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
220 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
221 }
222 
223 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
224 			 unsigned short family)
225 {
226 	switch (family) {
227 	case AF_INET:
228 		return __xfrm4_selector_match(sel, fl);
229 	case AF_INET6:
230 		return __xfrm6_selector_match(sel, fl);
231 	}
232 	return false;
233 }
234 
235 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
236 {
237 	const struct xfrm_policy_afinfo *afinfo;
238 
239 	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
240 		return NULL;
241 	rcu_read_lock();
242 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
243 	if (unlikely(!afinfo))
244 		rcu_read_unlock();
245 	return afinfo;
246 }
247 
248 /* Called with rcu_read_lock(). */
249 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
250 {
251 	return rcu_dereference(xfrm_if_cb);
252 }
253 
254 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
255 				    const xfrm_address_t *saddr,
256 				    const xfrm_address_t *daddr,
257 				    int family, u32 mark)
258 {
259 	const struct xfrm_policy_afinfo *afinfo;
260 	struct dst_entry *dst;
261 
262 	afinfo = xfrm_policy_get_afinfo(family);
263 	if (unlikely(afinfo == NULL))
264 		return ERR_PTR(-EAFNOSUPPORT);
265 
266 	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
267 
268 	rcu_read_unlock();
269 
270 	return dst;
271 }
272 EXPORT_SYMBOL(__xfrm_dst_lookup);
273 
274 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
275 						int tos, int oif,
276 						xfrm_address_t *prev_saddr,
277 						xfrm_address_t *prev_daddr,
278 						int family, u32 mark)
279 {
280 	struct net *net = xs_net(x);
281 	xfrm_address_t *saddr = &x->props.saddr;
282 	xfrm_address_t *daddr = &x->id.daddr;
283 	struct dst_entry *dst;
284 
285 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
286 		saddr = x->coaddr;
287 		daddr = prev_daddr;
288 	}
289 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
290 		saddr = prev_saddr;
291 		daddr = x->coaddr;
292 	}
293 
294 	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
295 
296 	if (!IS_ERR(dst)) {
297 		if (prev_saddr != saddr)
298 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
299 		if (prev_daddr != daddr)
300 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
301 	}
302 
303 	return dst;
304 }
305 
306 static inline unsigned long make_jiffies(long secs)
307 {
308 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
309 		return MAX_SCHEDULE_TIMEOUT-1;
310 	else
311 		return secs*HZ;
312 }
313 
314 static void xfrm_policy_timer(struct timer_list *t)
315 {
316 	struct xfrm_policy *xp = from_timer(xp, t, timer);
317 	time64_t now = ktime_get_real_seconds();
318 	time64_t next = TIME64_MAX;
319 	int warn = 0;
320 	int dir;
321 
322 	read_lock(&xp->lock);
323 
324 	if (unlikely(xp->walk.dead))
325 		goto out;
326 
327 	dir = xfrm_policy_id2dir(xp->index);
328 
329 	if (xp->lft.hard_add_expires_seconds) {
330 		time64_t tmo = xp->lft.hard_add_expires_seconds +
331 			xp->curlft.add_time - now;
332 		if (tmo <= 0)
333 			goto expired;
334 		if (tmo < next)
335 			next = tmo;
336 	}
337 	if (xp->lft.hard_use_expires_seconds) {
338 		time64_t tmo = xp->lft.hard_use_expires_seconds +
339 			(READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
340 		if (tmo <= 0)
341 			goto expired;
342 		if (tmo < next)
343 			next = tmo;
344 	}
345 	if (xp->lft.soft_add_expires_seconds) {
346 		time64_t tmo = xp->lft.soft_add_expires_seconds +
347 			xp->curlft.add_time - now;
348 		if (tmo <= 0) {
349 			warn = 1;
350 			tmo = XFRM_KM_TIMEOUT;
351 		}
352 		if (tmo < next)
353 			next = tmo;
354 	}
355 	if (xp->lft.soft_use_expires_seconds) {
356 		time64_t tmo = xp->lft.soft_use_expires_seconds +
357 			(READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
358 		if (tmo <= 0) {
359 			warn = 1;
360 			tmo = XFRM_KM_TIMEOUT;
361 		}
362 		if (tmo < next)
363 			next = tmo;
364 	}
365 
366 	if (warn)
367 		km_policy_expired(xp, dir, 0, 0);
368 	if (next != TIME64_MAX &&
369 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
370 		xfrm_pol_hold(xp);
371 
372 out:
373 	read_unlock(&xp->lock);
374 	xfrm_pol_put(xp);
375 	return;
376 
377 expired:
378 	read_unlock(&xp->lock);
379 	if (!xfrm_policy_delete(xp, dir))
380 		km_policy_expired(xp, dir, 1, 0);
381 	xfrm_pol_put(xp);
382 }
383 
384 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
385  * SPD calls.
386  */
387 
388 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
389 {
390 	struct xfrm_policy *policy;
391 
392 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
393 
394 	if (policy) {
395 		write_pnet(&policy->xp_net, net);
396 		INIT_LIST_HEAD(&policy->walk.all);
397 		INIT_HLIST_NODE(&policy->bydst_inexact_list);
398 		INIT_HLIST_NODE(&policy->bydst);
399 		INIT_HLIST_NODE(&policy->byidx);
400 		rwlock_init(&policy->lock);
401 		refcount_set(&policy->refcnt, 1);
402 		skb_queue_head_init(&policy->polq.hold_queue);
403 		timer_setup(&policy->timer, xfrm_policy_timer, 0);
404 		timer_setup(&policy->polq.hold_timer,
405 			    xfrm_policy_queue_process, 0);
406 	}
407 	return policy;
408 }
409 EXPORT_SYMBOL(xfrm_policy_alloc);
410 
411 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
412 {
413 	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
414 
415 	security_xfrm_policy_free(policy->security);
416 	kfree(policy);
417 }
418 
419 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
420 
421 void xfrm_policy_destroy(struct xfrm_policy *policy)
422 {
423 	BUG_ON(!policy->walk.dead);
424 
425 	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
426 		BUG();
427 
428 	xfrm_dev_policy_free(policy);
429 	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
430 }
431 EXPORT_SYMBOL(xfrm_policy_destroy);
432 
433 /* Rule must be locked. Release descendant resources, announce
434  * entry dead. The rule must be unlinked from lists to the moment.
435  */
436 
437 static void xfrm_policy_kill(struct xfrm_policy *policy)
438 {
439 	write_lock_bh(&policy->lock);
440 	policy->walk.dead = 1;
441 	write_unlock_bh(&policy->lock);
442 
443 	atomic_inc(&policy->genid);
444 
445 	if (del_timer(&policy->polq.hold_timer))
446 		xfrm_pol_put(policy);
447 	skb_queue_purge(&policy->polq.hold_queue);
448 
449 	if (del_timer(&policy->timer))
450 		xfrm_pol_put(policy);
451 
452 	xfrm_pol_put(policy);
453 }
454 
455 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
456 
457 static inline unsigned int idx_hash(struct net *net, u32 index)
458 {
459 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
460 }
461 
462 /* calculate policy hash thresholds */
463 static void __get_hash_thresh(struct net *net,
464 			      unsigned short family, int dir,
465 			      u8 *dbits, u8 *sbits)
466 {
467 	switch (family) {
468 	case AF_INET:
469 		*dbits = net->xfrm.policy_bydst[dir].dbits4;
470 		*sbits = net->xfrm.policy_bydst[dir].sbits4;
471 		break;
472 
473 	case AF_INET6:
474 		*dbits = net->xfrm.policy_bydst[dir].dbits6;
475 		*sbits = net->xfrm.policy_bydst[dir].sbits6;
476 		break;
477 
478 	default:
479 		*dbits = 0;
480 		*sbits = 0;
481 	}
482 }
483 
484 static struct hlist_head *policy_hash_bysel(struct net *net,
485 					    const struct xfrm_selector *sel,
486 					    unsigned short family, int dir)
487 {
488 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
489 	unsigned int hash;
490 	u8 dbits;
491 	u8 sbits;
492 
493 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
494 	hash = __sel_hash(sel, family, hmask, dbits, sbits);
495 
496 	if (hash == hmask + 1)
497 		return NULL;
498 
499 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
500 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
501 }
502 
503 static struct hlist_head *policy_hash_direct(struct net *net,
504 					     const xfrm_address_t *daddr,
505 					     const xfrm_address_t *saddr,
506 					     unsigned short family, int dir)
507 {
508 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
509 	unsigned int hash;
510 	u8 dbits;
511 	u8 sbits;
512 
513 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
514 	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
515 
516 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
517 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
518 }
519 
520 static void xfrm_dst_hash_transfer(struct net *net,
521 				   struct hlist_head *list,
522 				   struct hlist_head *ndsttable,
523 				   unsigned int nhashmask,
524 				   int dir)
525 {
526 	struct hlist_node *tmp, *entry0 = NULL;
527 	struct xfrm_policy *pol;
528 	unsigned int h0 = 0;
529 	u8 dbits;
530 	u8 sbits;
531 
532 redo:
533 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
534 		unsigned int h;
535 
536 		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
537 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
538 				pol->family, nhashmask, dbits, sbits);
539 		if (!entry0 || pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET) {
540 			hlist_del_rcu(&pol->bydst);
541 			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
542 			h0 = h;
543 		} else {
544 			if (h != h0)
545 				continue;
546 			hlist_del_rcu(&pol->bydst);
547 			hlist_add_behind_rcu(&pol->bydst, entry0);
548 		}
549 		entry0 = &pol->bydst;
550 	}
551 	if (!hlist_empty(list)) {
552 		entry0 = NULL;
553 		goto redo;
554 	}
555 }
556 
557 static void xfrm_idx_hash_transfer(struct hlist_head *list,
558 				   struct hlist_head *nidxtable,
559 				   unsigned int nhashmask)
560 {
561 	struct hlist_node *tmp;
562 	struct xfrm_policy *pol;
563 
564 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
565 		unsigned int h;
566 
567 		h = __idx_hash(pol->index, nhashmask);
568 		hlist_add_head(&pol->byidx, nidxtable+h);
569 	}
570 }
571 
572 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
573 {
574 	return ((old_hmask + 1) << 1) - 1;
575 }
576 
577 static void xfrm_bydst_resize(struct net *net, int dir)
578 {
579 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
580 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
581 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
582 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
583 	struct hlist_head *odst;
584 	int i;
585 
586 	if (!ndst)
587 		return;
588 
589 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
590 	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
591 
592 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
593 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
594 
595 	for (i = hmask; i >= 0; i--)
596 		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
597 
598 	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
599 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
600 
601 	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
602 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
603 
604 	synchronize_rcu();
605 
606 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
607 }
608 
609 static void xfrm_byidx_resize(struct net *net)
610 {
611 	unsigned int hmask = net->xfrm.policy_idx_hmask;
612 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
613 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
614 	struct hlist_head *oidx = net->xfrm.policy_byidx;
615 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
616 	int i;
617 
618 	if (!nidx)
619 		return;
620 
621 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
622 
623 	for (i = hmask; i >= 0; i--)
624 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
625 
626 	net->xfrm.policy_byidx = nidx;
627 	net->xfrm.policy_idx_hmask = nhashmask;
628 
629 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
630 
631 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
632 }
633 
634 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
635 {
636 	unsigned int cnt = net->xfrm.policy_count[dir];
637 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
638 
639 	if (total)
640 		*total += cnt;
641 
642 	if ((hmask + 1) < xfrm_policy_hashmax &&
643 	    cnt > hmask)
644 		return 1;
645 
646 	return 0;
647 }
648 
649 static inline int xfrm_byidx_should_resize(struct net *net, int total)
650 {
651 	unsigned int hmask = net->xfrm.policy_idx_hmask;
652 
653 	if ((hmask + 1) < xfrm_policy_hashmax &&
654 	    total > hmask)
655 		return 1;
656 
657 	return 0;
658 }
659 
660 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
661 {
662 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
663 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
664 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
665 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
666 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
667 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
668 	si->spdhcnt = net->xfrm.policy_idx_hmask;
669 	si->spdhmcnt = xfrm_policy_hashmax;
670 }
671 EXPORT_SYMBOL(xfrm_spd_getinfo);
672 
673 static DEFINE_MUTEX(hash_resize_mutex);
674 static void xfrm_hash_resize(struct work_struct *work)
675 {
676 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
677 	int dir, total;
678 
679 	mutex_lock(&hash_resize_mutex);
680 
681 	total = 0;
682 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
683 		if (xfrm_bydst_should_resize(net, dir, &total))
684 			xfrm_bydst_resize(net, dir);
685 	}
686 	if (xfrm_byidx_should_resize(net, total))
687 		xfrm_byidx_resize(net);
688 
689 	mutex_unlock(&hash_resize_mutex);
690 }
691 
692 /* Make sure *pol can be inserted into fastbin.
693  * Useful to check that later insert requests will be successful
694  * (provided xfrm_policy_lock is held throughout).
695  */
696 static struct xfrm_pol_inexact_bin *
697 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
698 {
699 	struct xfrm_pol_inexact_bin *bin, *prev;
700 	struct xfrm_pol_inexact_key k = {
701 		.family = pol->family,
702 		.type = pol->type,
703 		.dir = dir,
704 		.if_id = pol->if_id,
705 	};
706 	struct net *net = xp_net(pol);
707 
708 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
709 
710 	write_pnet(&k.net, net);
711 	bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
712 				     xfrm_pol_inexact_params);
713 	if (bin)
714 		return bin;
715 
716 	bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
717 	if (!bin)
718 		return NULL;
719 
720 	bin->k = k;
721 	INIT_HLIST_HEAD(&bin->hhead);
722 	bin->root_d = RB_ROOT;
723 	bin->root_s = RB_ROOT;
724 	seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
725 
726 	prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
727 						&bin->k, &bin->head,
728 						xfrm_pol_inexact_params);
729 	if (!prev) {
730 		list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
731 		return bin;
732 	}
733 
734 	kfree(bin);
735 
736 	return IS_ERR(prev) ? NULL : prev;
737 }
738 
739 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
740 					       int family, u8 prefixlen)
741 {
742 	if (xfrm_addr_any(addr, family))
743 		return true;
744 
745 	if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
746 		return true;
747 
748 	if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
749 		return true;
750 
751 	return false;
752 }
753 
754 static bool
755 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
756 {
757 	const xfrm_address_t *addr;
758 	bool saddr_any, daddr_any;
759 	u8 prefixlen;
760 
761 	addr = &policy->selector.saddr;
762 	prefixlen = policy->selector.prefixlen_s;
763 
764 	saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
765 						       policy->family,
766 						       prefixlen);
767 	addr = &policy->selector.daddr;
768 	prefixlen = policy->selector.prefixlen_d;
769 	daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
770 						       policy->family,
771 						       prefixlen);
772 	return saddr_any && daddr_any;
773 }
774 
775 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
776 				       const xfrm_address_t *addr, u8 prefixlen)
777 {
778 	node->addr = *addr;
779 	node->prefixlen = prefixlen;
780 }
781 
782 static struct xfrm_pol_inexact_node *
783 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
784 {
785 	struct xfrm_pol_inexact_node *node;
786 
787 	node = kzalloc(sizeof(*node), GFP_ATOMIC);
788 	if (node)
789 		xfrm_pol_inexact_node_init(node, addr, prefixlen);
790 
791 	return node;
792 }
793 
794 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
795 				  const xfrm_address_t *b,
796 				  u8 prefixlen, u16 family)
797 {
798 	u32 ma, mb, mask;
799 	unsigned int pdw, pbi;
800 	int delta = 0;
801 
802 	switch (family) {
803 	case AF_INET:
804 		if (prefixlen == 0)
805 			return 0;
806 		mask = ~0U << (32 - prefixlen);
807 		ma = ntohl(a->a4) & mask;
808 		mb = ntohl(b->a4) & mask;
809 		if (ma < mb)
810 			delta = -1;
811 		else if (ma > mb)
812 			delta = 1;
813 		break;
814 	case AF_INET6:
815 		pdw = prefixlen >> 5;
816 		pbi = prefixlen & 0x1f;
817 
818 		if (pdw) {
819 			delta = memcmp(a->a6, b->a6, pdw << 2);
820 			if (delta)
821 				return delta;
822 		}
823 		if (pbi) {
824 			mask = ~0U << (32 - pbi);
825 			ma = ntohl(a->a6[pdw]) & mask;
826 			mb = ntohl(b->a6[pdw]) & mask;
827 			if (ma < mb)
828 				delta = -1;
829 			else if (ma > mb)
830 				delta = 1;
831 		}
832 		break;
833 	default:
834 		break;
835 	}
836 
837 	return delta;
838 }
839 
840 static void xfrm_policy_inexact_list_reinsert(struct net *net,
841 					      struct xfrm_pol_inexact_node *n,
842 					      u16 family)
843 {
844 	unsigned int matched_s, matched_d;
845 	struct xfrm_policy *policy, *p;
846 
847 	matched_s = 0;
848 	matched_d = 0;
849 
850 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
851 		struct hlist_node *newpos = NULL;
852 		bool matches_s, matches_d;
853 
854 		if (!policy->bydst_reinsert)
855 			continue;
856 
857 		WARN_ON_ONCE(policy->family != family);
858 
859 		policy->bydst_reinsert = false;
860 		hlist_for_each_entry(p, &n->hhead, bydst) {
861 			if (policy->priority > p->priority)
862 				newpos = &p->bydst;
863 			else if (policy->priority == p->priority &&
864 				 policy->pos > p->pos)
865 				newpos = &p->bydst;
866 			else
867 				break;
868 		}
869 
870 		if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
871 			hlist_add_behind_rcu(&policy->bydst, newpos);
872 		else
873 			hlist_add_head_rcu(&policy->bydst, &n->hhead);
874 
875 		/* paranoia checks follow.
876 		 * Check that the reinserted policy matches at least
877 		 * saddr or daddr for current node prefix.
878 		 *
879 		 * Matching both is fine, matching saddr in one policy
880 		 * (but not daddr) and then matching only daddr in another
881 		 * is a bug.
882 		 */
883 		matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
884 						   &n->addr,
885 						   n->prefixlen,
886 						   family) == 0;
887 		matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
888 						   &n->addr,
889 						   n->prefixlen,
890 						   family) == 0;
891 		if (matches_s && matches_d)
892 			continue;
893 
894 		WARN_ON_ONCE(!matches_s && !matches_d);
895 		if (matches_s)
896 			matched_s++;
897 		if (matches_d)
898 			matched_d++;
899 		WARN_ON_ONCE(matched_s && matched_d);
900 	}
901 }
902 
903 static void xfrm_policy_inexact_node_reinsert(struct net *net,
904 					      struct xfrm_pol_inexact_node *n,
905 					      struct rb_root *new,
906 					      u16 family)
907 {
908 	struct xfrm_pol_inexact_node *node;
909 	struct rb_node **p, *parent;
910 
911 	/* we should not have another subtree here */
912 	WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
913 restart:
914 	parent = NULL;
915 	p = &new->rb_node;
916 	while (*p) {
917 		u8 prefixlen;
918 		int delta;
919 
920 		parent = *p;
921 		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
922 
923 		prefixlen = min(node->prefixlen, n->prefixlen);
924 
925 		delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
926 					       prefixlen, family);
927 		if (delta < 0) {
928 			p = &parent->rb_left;
929 		} else if (delta > 0) {
930 			p = &parent->rb_right;
931 		} else {
932 			bool same_prefixlen = node->prefixlen == n->prefixlen;
933 			struct xfrm_policy *tmp;
934 
935 			hlist_for_each_entry(tmp, &n->hhead, bydst) {
936 				tmp->bydst_reinsert = true;
937 				hlist_del_rcu(&tmp->bydst);
938 			}
939 
940 			node->prefixlen = prefixlen;
941 
942 			xfrm_policy_inexact_list_reinsert(net, node, family);
943 
944 			if (same_prefixlen) {
945 				kfree_rcu(n, rcu);
946 				return;
947 			}
948 
949 			rb_erase(*p, new);
950 			kfree_rcu(n, rcu);
951 			n = node;
952 			goto restart;
953 		}
954 	}
955 
956 	rb_link_node_rcu(&n->node, parent, p);
957 	rb_insert_color(&n->node, new);
958 }
959 
960 /* merge nodes v and n */
961 static void xfrm_policy_inexact_node_merge(struct net *net,
962 					   struct xfrm_pol_inexact_node *v,
963 					   struct xfrm_pol_inexact_node *n,
964 					   u16 family)
965 {
966 	struct xfrm_pol_inexact_node *node;
967 	struct xfrm_policy *tmp;
968 	struct rb_node *rnode;
969 
970 	/* To-be-merged node v has a subtree.
971 	 *
972 	 * Dismantle it and insert its nodes to n->root.
973 	 */
974 	while ((rnode = rb_first(&v->root)) != NULL) {
975 		node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
976 		rb_erase(&node->node, &v->root);
977 		xfrm_policy_inexact_node_reinsert(net, node, &n->root,
978 						  family);
979 	}
980 
981 	hlist_for_each_entry(tmp, &v->hhead, bydst) {
982 		tmp->bydst_reinsert = true;
983 		hlist_del_rcu(&tmp->bydst);
984 	}
985 
986 	xfrm_policy_inexact_list_reinsert(net, n, family);
987 }
988 
989 static struct xfrm_pol_inexact_node *
990 xfrm_policy_inexact_insert_node(struct net *net,
991 				struct rb_root *root,
992 				xfrm_address_t *addr,
993 				u16 family, u8 prefixlen, u8 dir)
994 {
995 	struct xfrm_pol_inexact_node *cached = NULL;
996 	struct rb_node **p, *parent = NULL;
997 	struct xfrm_pol_inexact_node *node;
998 
999 	p = &root->rb_node;
1000 	while (*p) {
1001 		int delta;
1002 
1003 		parent = *p;
1004 		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1005 
1006 		delta = xfrm_policy_addr_delta(addr, &node->addr,
1007 					       node->prefixlen,
1008 					       family);
1009 		if (delta == 0 && prefixlen >= node->prefixlen) {
1010 			WARN_ON_ONCE(cached); /* ipsec policies got lost */
1011 			return node;
1012 		}
1013 
1014 		if (delta < 0)
1015 			p = &parent->rb_left;
1016 		else
1017 			p = &parent->rb_right;
1018 
1019 		if (prefixlen < node->prefixlen) {
1020 			delta = xfrm_policy_addr_delta(addr, &node->addr,
1021 						       prefixlen,
1022 						       family);
1023 			if (delta)
1024 				continue;
1025 
1026 			/* This node is a subnet of the new prefix. It needs
1027 			 * to be removed and re-inserted with the smaller
1028 			 * prefix and all nodes that are now also covered
1029 			 * by the reduced prefixlen.
1030 			 */
1031 			rb_erase(&node->node, root);
1032 
1033 			if (!cached) {
1034 				xfrm_pol_inexact_node_init(node, addr,
1035 							   prefixlen);
1036 				cached = node;
1037 			} else {
1038 				/* This node also falls within the new
1039 				 * prefixlen. Merge the to-be-reinserted
1040 				 * node and this one.
1041 				 */
1042 				xfrm_policy_inexact_node_merge(net, node,
1043 							       cached, family);
1044 				kfree_rcu(node, rcu);
1045 			}
1046 
1047 			/* restart */
1048 			p = &root->rb_node;
1049 			parent = NULL;
1050 		}
1051 	}
1052 
1053 	node = cached;
1054 	if (!node) {
1055 		node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1056 		if (!node)
1057 			return NULL;
1058 	}
1059 
1060 	rb_link_node_rcu(&node->node, parent, p);
1061 	rb_insert_color(&node->node, root);
1062 
1063 	return node;
1064 }
1065 
1066 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1067 {
1068 	struct xfrm_pol_inexact_node *node;
1069 	struct rb_node *rn = rb_first(r);
1070 
1071 	while (rn) {
1072 		node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1073 
1074 		xfrm_policy_inexact_gc_tree(&node->root, rm);
1075 		rn = rb_next(rn);
1076 
1077 		if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1078 			WARN_ON_ONCE(rm);
1079 			continue;
1080 		}
1081 
1082 		rb_erase(&node->node, r);
1083 		kfree_rcu(node, rcu);
1084 	}
1085 }
1086 
1087 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1088 {
1089 	write_seqcount_begin(&b->count);
1090 	xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1091 	xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1092 	write_seqcount_end(&b->count);
1093 
1094 	if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1095 	    !hlist_empty(&b->hhead)) {
1096 		WARN_ON_ONCE(net_exit);
1097 		return;
1098 	}
1099 
1100 	if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1101 				   xfrm_pol_inexact_params) == 0) {
1102 		list_del(&b->inexact_bins);
1103 		kfree_rcu(b, rcu);
1104 	}
1105 }
1106 
1107 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1108 {
1109 	struct net *net = read_pnet(&b->k.net);
1110 
1111 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1112 	__xfrm_policy_inexact_prune_bin(b, false);
1113 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1114 }
1115 
1116 static void __xfrm_policy_inexact_flush(struct net *net)
1117 {
1118 	struct xfrm_pol_inexact_bin *bin, *t;
1119 
1120 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1121 
1122 	list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1123 		__xfrm_policy_inexact_prune_bin(bin, false);
1124 }
1125 
1126 static struct hlist_head *
1127 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1128 				struct xfrm_policy *policy, u8 dir)
1129 {
1130 	struct xfrm_pol_inexact_node *n;
1131 	struct net *net;
1132 
1133 	net = xp_net(policy);
1134 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1135 
1136 	if (xfrm_policy_inexact_insert_use_any_list(policy))
1137 		return &bin->hhead;
1138 
1139 	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1140 					       policy->family,
1141 					       policy->selector.prefixlen_d)) {
1142 		write_seqcount_begin(&bin->count);
1143 		n = xfrm_policy_inexact_insert_node(net,
1144 						    &bin->root_s,
1145 						    &policy->selector.saddr,
1146 						    policy->family,
1147 						    policy->selector.prefixlen_s,
1148 						    dir);
1149 		write_seqcount_end(&bin->count);
1150 		if (!n)
1151 			return NULL;
1152 
1153 		return &n->hhead;
1154 	}
1155 
1156 	/* daddr is fixed */
1157 	write_seqcount_begin(&bin->count);
1158 	n = xfrm_policy_inexact_insert_node(net,
1159 					    &bin->root_d,
1160 					    &policy->selector.daddr,
1161 					    policy->family,
1162 					    policy->selector.prefixlen_d, dir);
1163 	write_seqcount_end(&bin->count);
1164 	if (!n)
1165 		return NULL;
1166 
1167 	/* saddr is wildcard */
1168 	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1169 					       policy->family,
1170 					       policy->selector.prefixlen_s))
1171 		return &n->hhead;
1172 
1173 	write_seqcount_begin(&bin->count);
1174 	n = xfrm_policy_inexact_insert_node(net,
1175 					    &n->root,
1176 					    &policy->selector.saddr,
1177 					    policy->family,
1178 					    policy->selector.prefixlen_s, dir);
1179 	write_seqcount_end(&bin->count);
1180 	if (!n)
1181 		return NULL;
1182 
1183 	return &n->hhead;
1184 }
1185 
1186 static struct xfrm_policy *
1187 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1188 {
1189 	struct xfrm_pol_inexact_bin *bin;
1190 	struct xfrm_policy *delpol;
1191 	struct hlist_head *chain;
1192 	struct net *net;
1193 
1194 	bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1195 	if (!bin)
1196 		return ERR_PTR(-ENOMEM);
1197 
1198 	net = xp_net(policy);
1199 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1200 
1201 	chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1202 	if (!chain) {
1203 		__xfrm_policy_inexact_prune_bin(bin, false);
1204 		return ERR_PTR(-ENOMEM);
1205 	}
1206 
1207 	delpol = xfrm_policy_insert_list(chain, policy, excl);
1208 	if (delpol && excl) {
1209 		__xfrm_policy_inexact_prune_bin(bin, false);
1210 		return ERR_PTR(-EEXIST);
1211 	}
1212 
1213 	chain = &net->xfrm.policy_inexact[dir];
1214 	xfrm_policy_insert_inexact_list(chain, policy);
1215 
1216 	if (delpol)
1217 		__xfrm_policy_inexact_prune_bin(bin, false);
1218 
1219 	return delpol;
1220 }
1221 
1222 static void xfrm_hash_rebuild(struct work_struct *work)
1223 {
1224 	struct net *net = container_of(work, struct net,
1225 				       xfrm.policy_hthresh.work);
1226 	unsigned int hmask;
1227 	struct xfrm_policy *pol;
1228 	struct xfrm_policy *policy;
1229 	struct hlist_head *chain;
1230 	struct hlist_head *odst;
1231 	struct hlist_node *newpos;
1232 	int i;
1233 	int dir;
1234 	unsigned seq;
1235 	u8 lbits4, rbits4, lbits6, rbits6;
1236 
1237 	mutex_lock(&hash_resize_mutex);
1238 
1239 	/* read selector prefixlen thresholds */
1240 	do {
1241 		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1242 
1243 		lbits4 = net->xfrm.policy_hthresh.lbits4;
1244 		rbits4 = net->xfrm.policy_hthresh.rbits4;
1245 		lbits6 = net->xfrm.policy_hthresh.lbits6;
1246 		rbits6 = net->xfrm.policy_hthresh.rbits6;
1247 	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1248 
1249 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1250 	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1251 
1252 	/* make sure that we can insert the indirect policies again before
1253 	 * we start with destructive action.
1254 	 */
1255 	list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1256 		struct xfrm_pol_inexact_bin *bin;
1257 		u8 dbits, sbits;
1258 
1259 		dir = xfrm_policy_id2dir(policy->index);
1260 		if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1261 			continue;
1262 
1263 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1264 			if (policy->family == AF_INET) {
1265 				dbits = rbits4;
1266 				sbits = lbits4;
1267 			} else {
1268 				dbits = rbits6;
1269 				sbits = lbits6;
1270 			}
1271 		} else {
1272 			if (policy->family == AF_INET) {
1273 				dbits = lbits4;
1274 				sbits = rbits4;
1275 			} else {
1276 				dbits = lbits6;
1277 				sbits = rbits6;
1278 			}
1279 		}
1280 
1281 		if (policy->selector.prefixlen_d < dbits ||
1282 		    policy->selector.prefixlen_s < sbits)
1283 			continue;
1284 
1285 		bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1286 		if (!bin)
1287 			goto out_unlock;
1288 
1289 		if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1290 			goto out_unlock;
1291 	}
1292 
1293 	/* reset the bydst and inexact table in all directions */
1294 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1295 		struct hlist_node *n;
1296 
1297 		hlist_for_each_entry_safe(policy, n,
1298 					  &net->xfrm.policy_inexact[dir],
1299 					  bydst_inexact_list) {
1300 			hlist_del_rcu(&policy->bydst);
1301 			hlist_del_init(&policy->bydst_inexact_list);
1302 		}
1303 
1304 		hmask = net->xfrm.policy_bydst[dir].hmask;
1305 		odst = net->xfrm.policy_bydst[dir].table;
1306 		for (i = hmask; i >= 0; i--) {
1307 			hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1308 				hlist_del_rcu(&policy->bydst);
1309 		}
1310 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1311 			/* dir out => dst = remote, src = local */
1312 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1313 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1314 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1315 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1316 		} else {
1317 			/* dir in/fwd => dst = local, src = remote */
1318 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1319 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1320 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1321 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1322 		}
1323 	}
1324 
1325 	/* re-insert all policies by order of creation */
1326 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1327 		if (policy->walk.dead)
1328 			continue;
1329 		dir = xfrm_policy_id2dir(policy->index);
1330 		if (dir >= XFRM_POLICY_MAX) {
1331 			/* skip socket policies */
1332 			continue;
1333 		}
1334 		newpos = NULL;
1335 		chain = policy_hash_bysel(net, &policy->selector,
1336 					  policy->family, dir);
1337 
1338 		if (!chain) {
1339 			void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1340 
1341 			WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1342 			continue;
1343 		}
1344 
1345 		hlist_for_each_entry(pol, chain, bydst) {
1346 			if (policy->priority >= pol->priority)
1347 				newpos = &pol->bydst;
1348 			else
1349 				break;
1350 		}
1351 		if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1352 			hlist_add_behind_rcu(&policy->bydst, newpos);
1353 		else
1354 			hlist_add_head_rcu(&policy->bydst, chain);
1355 	}
1356 
1357 out_unlock:
1358 	__xfrm_policy_inexact_flush(net);
1359 	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1360 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1361 
1362 	mutex_unlock(&hash_resize_mutex);
1363 }
1364 
1365 void xfrm_policy_hash_rebuild(struct net *net)
1366 {
1367 	schedule_work(&net->xfrm.policy_hthresh.work);
1368 }
1369 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1370 
1371 /* Generate new index... KAME seems to generate them ordered by cost
1372  * of an absolute inpredictability of ordering of rules. This will not pass. */
1373 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1374 {
1375 	static u32 idx_generator;
1376 
1377 	for (;;) {
1378 		struct hlist_head *list;
1379 		struct xfrm_policy *p;
1380 		u32 idx;
1381 		int found;
1382 
1383 		if (!index) {
1384 			idx = (idx_generator | dir);
1385 			idx_generator += 8;
1386 		} else {
1387 			idx = index;
1388 			index = 0;
1389 		}
1390 
1391 		if (idx == 0)
1392 			idx = 8;
1393 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
1394 		found = 0;
1395 		hlist_for_each_entry(p, list, byidx) {
1396 			if (p->index == idx) {
1397 				found = 1;
1398 				break;
1399 			}
1400 		}
1401 		if (!found)
1402 			return idx;
1403 	}
1404 }
1405 
1406 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1407 {
1408 	u32 *p1 = (u32 *) s1;
1409 	u32 *p2 = (u32 *) s2;
1410 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
1411 	int i;
1412 
1413 	for (i = 0; i < len; i++) {
1414 		if (p1[i] != p2[i])
1415 			return 1;
1416 	}
1417 
1418 	return 0;
1419 }
1420 
1421 static void xfrm_policy_requeue(struct xfrm_policy *old,
1422 				struct xfrm_policy *new)
1423 {
1424 	struct xfrm_policy_queue *pq = &old->polq;
1425 	struct sk_buff_head list;
1426 
1427 	if (skb_queue_empty(&pq->hold_queue))
1428 		return;
1429 
1430 	__skb_queue_head_init(&list);
1431 
1432 	spin_lock_bh(&pq->hold_queue.lock);
1433 	skb_queue_splice_init(&pq->hold_queue, &list);
1434 	if (del_timer(&pq->hold_timer))
1435 		xfrm_pol_put(old);
1436 	spin_unlock_bh(&pq->hold_queue.lock);
1437 
1438 	pq = &new->polq;
1439 
1440 	spin_lock_bh(&pq->hold_queue.lock);
1441 	skb_queue_splice(&list, &pq->hold_queue);
1442 	pq->timeout = XFRM_QUEUE_TMO_MIN;
1443 	if (!mod_timer(&pq->hold_timer, jiffies))
1444 		xfrm_pol_hold(new);
1445 	spin_unlock_bh(&pq->hold_queue.lock);
1446 }
1447 
1448 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1449 					  struct xfrm_policy *pol)
1450 {
1451 	return mark->v == pol->mark.v && mark->m == pol->mark.m;
1452 }
1453 
1454 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1455 {
1456 	const struct xfrm_pol_inexact_key *k = data;
1457 	u32 a = k->type << 24 | k->dir << 16 | k->family;
1458 
1459 	return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1460 			    seed);
1461 }
1462 
1463 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1464 {
1465 	const struct xfrm_pol_inexact_bin *b = data;
1466 
1467 	return xfrm_pol_bin_key(&b->k, 0, seed);
1468 }
1469 
1470 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1471 			    const void *ptr)
1472 {
1473 	const struct xfrm_pol_inexact_key *key = arg->key;
1474 	const struct xfrm_pol_inexact_bin *b = ptr;
1475 	int ret;
1476 
1477 	if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1478 		return -1;
1479 
1480 	ret = b->k.dir ^ key->dir;
1481 	if (ret)
1482 		return ret;
1483 
1484 	ret = b->k.type ^ key->type;
1485 	if (ret)
1486 		return ret;
1487 
1488 	ret = b->k.family ^ key->family;
1489 	if (ret)
1490 		return ret;
1491 
1492 	return b->k.if_id ^ key->if_id;
1493 }
1494 
1495 static const struct rhashtable_params xfrm_pol_inexact_params = {
1496 	.head_offset		= offsetof(struct xfrm_pol_inexact_bin, head),
1497 	.hashfn			= xfrm_pol_bin_key,
1498 	.obj_hashfn		= xfrm_pol_bin_obj,
1499 	.obj_cmpfn		= xfrm_pol_bin_cmp,
1500 	.automatic_shrinking	= true,
1501 };
1502 
1503 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1504 					    struct xfrm_policy *policy)
1505 {
1506 	struct xfrm_policy *pol, *delpol = NULL;
1507 	struct hlist_node *newpos = NULL;
1508 	int i = 0;
1509 
1510 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1511 		if (pol->type == policy->type &&
1512 		    pol->if_id == policy->if_id &&
1513 		    !selector_cmp(&pol->selector, &policy->selector) &&
1514 		    xfrm_policy_mark_match(&policy->mark, pol) &&
1515 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1516 		    !WARN_ON(delpol)) {
1517 			delpol = pol;
1518 			if (policy->priority > pol->priority)
1519 				continue;
1520 		} else if (policy->priority >= pol->priority) {
1521 			newpos = &pol->bydst_inexact_list;
1522 			continue;
1523 		}
1524 		if (delpol)
1525 			break;
1526 	}
1527 
1528 	if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1529 		hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1530 	else
1531 		hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1532 
1533 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1534 		pol->pos = i;
1535 		i++;
1536 	}
1537 }
1538 
1539 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1540 						   struct xfrm_policy *policy,
1541 						   bool excl)
1542 {
1543 	struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1544 
1545 	hlist_for_each_entry(pol, chain, bydst) {
1546 		if (pol->type == policy->type &&
1547 		    pol->if_id == policy->if_id &&
1548 		    !selector_cmp(&pol->selector, &policy->selector) &&
1549 		    xfrm_policy_mark_match(&policy->mark, pol) &&
1550 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1551 		    !WARN_ON(delpol)) {
1552 			if (excl)
1553 				return ERR_PTR(-EEXIST);
1554 			delpol = pol;
1555 			if (policy->priority > pol->priority)
1556 				continue;
1557 		} else if (policy->priority >= pol->priority) {
1558 			newpos = pol;
1559 			continue;
1560 		}
1561 		if (delpol)
1562 			break;
1563 	}
1564 
1565 	if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1566 		hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1567 	else
1568 		/* Packet offload policies enter to the head
1569 		 * to speed-up lookups.
1570 		 */
1571 		hlist_add_head_rcu(&policy->bydst, chain);
1572 
1573 	return delpol;
1574 }
1575 
1576 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1577 {
1578 	struct net *net = xp_net(policy);
1579 	struct xfrm_policy *delpol;
1580 	struct hlist_head *chain;
1581 
1582 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1583 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1584 	if (chain)
1585 		delpol = xfrm_policy_insert_list(chain, policy, excl);
1586 	else
1587 		delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1588 
1589 	if (IS_ERR(delpol)) {
1590 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1591 		return PTR_ERR(delpol);
1592 	}
1593 
1594 	__xfrm_policy_link(policy, dir);
1595 
1596 	/* After previous checking, family can either be AF_INET or AF_INET6 */
1597 	if (policy->family == AF_INET)
1598 		rt_genid_bump_ipv4(net);
1599 	else
1600 		rt_genid_bump_ipv6(net);
1601 
1602 	if (delpol) {
1603 		xfrm_policy_requeue(delpol, policy);
1604 		__xfrm_policy_unlink(delpol, dir);
1605 	}
1606 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1607 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1608 	policy->curlft.add_time = ktime_get_real_seconds();
1609 	policy->curlft.use_time = 0;
1610 	if (!mod_timer(&policy->timer, jiffies + HZ))
1611 		xfrm_pol_hold(policy);
1612 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1613 
1614 	if (delpol)
1615 		xfrm_policy_kill(delpol);
1616 	else if (xfrm_bydst_should_resize(net, dir, NULL))
1617 		schedule_work(&net->xfrm.policy_hash_work);
1618 
1619 	return 0;
1620 }
1621 EXPORT_SYMBOL(xfrm_policy_insert);
1622 
1623 static struct xfrm_policy *
1624 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1625 			u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1626 			struct xfrm_sec_ctx *ctx)
1627 {
1628 	struct xfrm_policy *pol;
1629 
1630 	if (!chain)
1631 		return NULL;
1632 
1633 	hlist_for_each_entry(pol, chain, bydst) {
1634 		if (pol->type == type &&
1635 		    pol->if_id == if_id &&
1636 		    xfrm_policy_mark_match(mark, pol) &&
1637 		    !selector_cmp(sel, &pol->selector) &&
1638 		    xfrm_sec_ctx_match(ctx, pol->security))
1639 			return pol;
1640 	}
1641 
1642 	return NULL;
1643 }
1644 
1645 struct xfrm_policy *
1646 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1647 		      u8 type, int dir, struct xfrm_selector *sel,
1648 		      struct xfrm_sec_ctx *ctx, int delete, int *err)
1649 {
1650 	struct xfrm_pol_inexact_bin *bin = NULL;
1651 	struct xfrm_policy *pol, *ret = NULL;
1652 	struct hlist_head *chain;
1653 
1654 	*err = 0;
1655 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1656 	chain = policy_hash_bysel(net, sel, sel->family, dir);
1657 	if (!chain) {
1658 		struct xfrm_pol_inexact_candidates cand;
1659 		int i;
1660 
1661 		bin = xfrm_policy_inexact_lookup(net, type,
1662 						 sel->family, dir, if_id);
1663 		if (!bin) {
1664 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1665 			return NULL;
1666 		}
1667 
1668 		if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1669 							 &sel->saddr,
1670 							 &sel->daddr)) {
1671 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1672 			return NULL;
1673 		}
1674 
1675 		pol = NULL;
1676 		for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1677 			struct xfrm_policy *tmp;
1678 
1679 			tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1680 						      if_id, type, dir,
1681 						      sel, ctx);
1682 			if (!tmp)
1683 				continue;
1684 
1685 			if (!pol || tmp->pos < pol->pos)
1686 				pol = tmp;
1687 		}
1688 	} else {
1689 		pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1690 					      sel, ctx);
1691 	}
1692 
1693 	if (pol) {
1694 		xfrm_pol_hold(pol);
1695 		if (delete) {
1696 			*err = security_xfrm_policy_delete(pol->security);
1697 			if (*err) {
1698 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1699 				return pol;
1700 			}
1701 			__xfrm_policy_unlink(pol, dir);
1702 		}
1703 		ret = pol;
1704 	}
1705 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1706 
1707 	if (ret && delete)
1708 		xfrm_policy_kill(ret);
1709 	if (bin && delete)
1710 		xfrm_policy_inexact_prune_bin(bin);
1711 	return ret;
1712 }
1713 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1714 
1715 struct xfrm_policy *
1716 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1717 		 u8 type, int dir, u32 id, int delete, int *err)
1718 {
1719 	struct xfrm_policy *pol, *ret;
1720 	struct hlist_head *chain;
1721 
1722 	*err = -ENOENT;
1723 	if (xfrm_policy_id2dir(id) != dir)
1724 		return NULL;
1725 
1726 	*err = 0;
1727 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1728 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
1729 	ret = NULL;
1730 	hlist_for_each_entry(pol, chain, byidx) {
1731 		if (pol->type == type && pol->index == id &&
1732 		    pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1733 			xfrm_pol_hold(pol);
1734 			if (delete) {
1735 				*err = security_xfrm_policy_delete(
1736 								pol->security);
1737 				if (*err) {
1738 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1739 					return pol;
1740 				}
1741 				__xfrm_policy_unlink(pol, dir);
1742 			}
1743 			ret = pol;
1744 			break;
1745 		}
1746 	}
1747 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1748 
1749 	if (ret && delete)
1750 		xfrm_policy_kill(ret);
1751 	return ret;
1752 }
1753 EXPORT_SYMBOL(xfrm_policy_byid);
1754 
1755 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1756 static inline int
1757 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1758 {
1759 	struct xfrm_policy *pol;
1760 	int err = 0;
1761 
1762 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1763 		if (pol->walk.dead ||
1764 		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1765 		    pol->type != type)
1766 			continue;
1767 
1768 		err = security_xfrm_policy_delete(pol->security);
1769 		if (err) {
1770 			xfrm_audit_policy_delete(pol, 0, task_valid);
1771 			return err;
1772 		}
1773 	}
1774 	return err;
1775 }
1776 
1777 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1778 						     struct net_device *dev,
1779 						     bool task_valid)
1780 {
1781 	struct xfrm_policy *pol;
1782 	int err = 0;
1783 
1784 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1785 		if (pol->walk.dead ||
1786 		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1787 		    pol->xdo.dev != dev)
1788 			continue;
1789 
1790 		err = security_xfrm_policy_delete(pol->security);
1791 		if (err) {
1792 			xfrm_audit_policy_delete(pol, 0, task_valid);
1793 			return err;
1794 		}
1795 	}
1796 	return err;
1797 }
1798 #else
1799 static inline int
1800 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1801 {
1802 	return 0;
1803 }
1804 
1805 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1806 						     struct net_device *dev,
1807 						     bool task_valid)
1808 {
1809 	return 0;
1810 }
1811 #endif
1812 
1813 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1814 {
1815 	int dir, err = 0, cnt = 0;
1816 	struct xfrm_policy *pol;
1817 
1818 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1819 
1820 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1821 	if (err)
1822 		goto out;
1823 
1824 again:
1825 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1826 		dir = xfrm_policy_id2dir(pol->index);
1827 		if (pol->walk.dead ||
1828 		    dir >= XFRM_POLICY_MAX ||
1829 		    pol->type != type)
1830 			continue;
1831 
1832 		__xfrm_policy_unlink(pol, dir);
1833 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1834 		xfrm_dev_policy_delete(pol);
1835 		cnt++;
1836 		xfrm_audit_policy_delete(pol, 1, task_valid);
1837 		xfrm_policy_kill(pol);
1838 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1839 		goto again;
1840 	}
1841 	if (cnt)
1842 		__xfrm_policy_inexact_flush(net);
1843 	else
1844 		err = -ESRCH;
1845 out:
1846 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1847 	return err;
1848 }
1849 EXPORT_SYMBOL(xfrm_policy_flush);
1850 
1851 int xfrm_dev_policy_flush(struct net *net, struct net_device *dev,
1852 			  bool task_valid)
1853 {
1854 	int dir, err = 0, cnt = 0;
1855 	struct xfrm_policy *pol;
1856 
1857 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1858 
1859 	err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid);
1860 	if (err)
1861 		goto out;
1862 
1863 again:
1864 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1865 		dir = xfrm_policy_id2dir(pol->index);
1866 		if (pol->walk.dead ||
1867 		    dir >= XFRM_POLICY_MAX ||
1868 		    pol->xdo.dev != dev)
1869 			continue;
1870 
1871 		__xfrm_policy_unlink(pol, dir);
1872 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1873 		xfrm_dev_policy_delete(pol);
1874 		cnt++;
1875 		xfrm_audit_policy_delete(pol, 1, task_valid);
1876 		xfrm_policy_kill(pol);
1877 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1878 		goto again;
1879 	}
1880 	if (cnt)
1881 		__xfrm_policy_inexact_flush(net);
1882 	else
1883 		err = -ESRCH;
1884 out:
1885 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1886 	return err;
1887 }
1888 EXPORT_SYMBOL(xfrm_dev_policy_flush);
1889 
1890 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1891 		     int (*func)(struct xfrm_policy *, int, int, void*),
1892 		     void *data)
1893 {
1894 	struct xfrm_policy *pol;
1895 	struct xfrm_policy_walk_entry *x;
1896 	int error = 0;
1897 
1898 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1899 	    walk->type != XFRM_POLICY_TYPE_ANY)
1900 		return -EINVAL;
1901 
1902 	if (list_empty(&walk->walk.all) && walk->seq != 0)
1903 		return 0;
1904 
1905 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1906 	if (list_empty(&walk->walk.all))
1907 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1908 	else
1909 		x = list_first_entry(&walk->walk.all,
1910 				     struct xfrm_policy_walk_entry, all);
1911 
1912 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1913 		if (x->dead)
1914 			continue;
1915 		pol = container_of(x, struct xfrm_policy, walk);
1916 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1917 		    walk->type != pol->type)
1918 			continue;
1919 		error = func(pol, xfrm_policy_id2dir(pol->index),
1920 			     walk->seq, data);
1921 		if (error) {
1922 			list_move_tail(&walk->walk.all, &x->all);
1923 			goto out;
1924 		}
1925 		walk->seq++;
1926 	}
1927 	if (walk->seq == 0) {
1928 		error = -ENOENT;
1929 		goto out;
1930 	}
1931 	list_del_init(&walk->walk.all);
1932 out:
1933 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1934 	return error;
1935 }
1936 EXPORT_SYMBOL(xfrm_policy_walk);
1937 
1938 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1939 {
1940 	INIT_LIST_HEAD(&walk->walk.all);
1941 	walk->walk.dead = 1;
1942 	walk->type = type;
1943 	walk->seq = 0;
1944 }
1945 EXPORT_SYMBOL(xfrm_policy_walk_init);
1946 
1947 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1948 {
1949 	if (list_empty(&walk->walk.all))
1950 		return;
1951 
1952 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1953 	list_del(&walk->walk.all);
1954 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1955 }
1956 EXPORT_SYMBOL(xfrm_policy_walk_done);
1957 
1958 /*
1959  * Find policy to apply to this flow.
1960  *
1961  * Returns 0 if policy found, else an -errno.
1962  */
1963 static int xfrm_policy_match(const struct xfrm_policy *pol,
1964 			     const struct flowi *fl,
1965 			     u8 type, u16 family, u32 if_id)
1966 {
1967 	const struct xfrm_selector *sel = &pol->selector;
1968 	int ret = -ESRCH;
1969 	bool match;
1970 
1971 	if (pol->family != family ||
1972 	    pol->if_id != if_id ||
1973 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1974 	    pol->type != type)
1975 		return ret;
1976 
1977 	match = xfrm_selector_match(sel, fl, family);
1978 	if (match)
1979 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1980 	return ret;
1981 }
1982 
1983 static struct xfrm_pol_inexact_node *
1984 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1985 				seqcount_spinlock_t *count,
1986 				const xfrm_address_t *addr, u16 family)
1987 {
1988 	const struct rb_node *parent;
1989 	int seq;
1990 
1991 again:
1992 	seq = read_seqcount_begin(count);
1993 
1994 	parent = rcu_dereference_raw(r->rb_node);
1995 	while (parent) {
1996 		struct xfrm_pol_inexact_node *node;
1997 		int delta;
1998 
1999 		node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
2000 
2001 		delta = xfrm_policy_addr_delta(addr, &node->addr,
2002 					       node->prefixlen, family);
2003 		if (delta < 0) {
2004 			parent = rcu_dereference_raw(parent->rb_left);
2005 			continue;
2006 		} else if (delta > 0) {
2007 			parent = rcu_dereference_raw(parent->rb_right);
2008 			continue;
2009 		}
2010 
2011 		return node;
2012 	}
2013 
2014 	if (read_seqcount_retry(count, seq))
2015 		goto again;
2016 
2017 	return NULL;
2018 }
2019 
2020 static bool
2021 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
2022 				    struct xfrm_pol_inexact_bin *b,
2023 				    const xfrm_address_t *saddr,
2024 				    const xfrm_address_t *daddr)
2025 {
2026 	struct xfrm_pol_inexact_node *n;
2027 	u16 family;
2028 
2029 	if (!b)
2030 		return false;
2031 
2032 	family = b->k.family;
2033 	memset(cand, 0, sizeof(*cand));
2034 	cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
2035 
2036 	n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
2037 					    family);
2038 	if (n) {
2039 		cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
2040 		n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
2041 						    family);
2042 		if (n)
2043 			cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
2044 	}
2045 
2046 	n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
2047 					    family);
2048 	if (n)
2049 		cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
2050 
2051 	return true;
2052 }
2053 
2054 static struct xfrm_pol_inexact_bin *
2055 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
2056 			       u8 dir, u32 if_id)
2057 {
2058 	struct xfrm_pol_inexact_key k = {
2059 		.family = family,
2060 		.type = type,
2061 		.dir = dir,
2062 		.if_id = if_id,
2063 	};
2064 
2065 	write_pnet(&k.net, net);
2066 
2067 	return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
2068 				 xfrm_pol_inexact_params);
2069 }
2070 
2071 static struct xfrm_pol_inexact_bin *
2072 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2073 			   u8 dir, u32 if_id)
2074 {
2075 	struct xfrm_pol_inexact_bin *bin;
2076 
2077 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2078 
2079 	rcu_read_lock();
2080 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2081 	rcu_read_unlock();
2082 
2083 	return bin;
2084 }
2085 
2086 static struct xfrm_policy *
2087 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2088 			      struct xfrm_policy *prefer,
2089 			      const struct flowi *fl,
2090 			      u8 type, u16 family, u32 if_id)
2091 {
2092 	u32 priority = prefer ? prefer->priority : ~0u;
2093 	struct xfrm_policy *pol;
2094 
2095 	if (!chain)
2096 		return NULL;
2097 
2098 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2099 		int err;
2100 
2101 		if (pol->priority > priority)
2102 			break;
2103 
2104 		err = xfrm_policy_match(pol, fl, type, family, if_id);
2105 		if (err) {
2106 			if (err != -ESRCH)
2107 				return ERR_PTR(err);
2108 
2109 			continue;
2110 		}
2111 
2112 		if (prefer) {
2113 			/* matches.  Is it older than *prefer? */
2114 			if (pol->priority == priority &&
2115 			    prefer->pos < pol->pos)
2116 				return prefer;
2117 		}
2118 
2119 		return pol;
2120 	}
2121 
2122 	return NULL;
2123 }
2124 
2125 static struct xfrm_policy *
2126 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2127 			    struct xfrm_policy *prefer,
2128 			    const struct flowi *fl,
2129 			    u8 type, u16 family, u32 if_id)
2130 {
2131 	struct xfrm_policy *tmp;
2132 	int i;
2133 
2134 	for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2135 		tmp = __xfrm_policy_eval_candidates(cand->res[i],
2136 						    prefer,
2137 						    fl, type, family, if_id);
2138 		if (!tmp)
2139 			continue;
2140 
2141 		if (IS_ERR(tmp))
2142 			return tmp;
2143 		prefer = tmp;
2144 	}
2145 
2146 	return prefer;
2147 }
2148 
2149 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2150 						     const struct flowi *fl,
2151 						     u16 family, u8 dir,
2152 						     u32 if_id)
2153 {
2154 	struct xfrm_pol_inexact_candidates cand;
2155 	const xfrm_address_t *daddr, *saddr;
2156 	struct xfrm_pol_inexact_bin *bin;
2157 	struct xfrm_policy *pol, *ret;
2158 	struct hlist_head *chain;
2159 	unsigned int sequence;
2160 	int err;
2161 
2162 	daddr = xfrm_flowi_daddr(fl, family);
2163 	saddr = xfrm_flowi_saddr(fl, family);
2164 	if (unlikely(!daddr || !saddr))
2165 		return NULL;
2166 
2167 	rcu_read_lock();
2168  retry:
2169 	do {
2170 		sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2171 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
2172 	} while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2173 
2174 	ret = NULL;
2175 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2176 		err = xfrm_policy_match(pol, fl, type, family, if_id);
2177 		if (err) {
2178 			if (err == -ESRCH)
2179 				continue;
2180 			else {
2181 				ret = ERR_PTR(err);
2182 				goto fail;
2183 			}
2184 		} else {
2185 			ret = pol;
2186 			break;
2187 		}
2188 	}
2189 	if (ret && ret->xdo.type == XFRM_DEV_OFFLOAD_PACKET)
2190 		goto skip_inexact;
2191 
2192 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2193 	if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2194 							 daddr))
2195 		goto skip_inexact;
2196 
2197 	pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2198 					  family, if_id);
2199 	if (pol) {
2200 		ret = pol;
2201 		if (IS_ERR(pol))
2202 			goto fail;
2203 	}
2204 
2205 skip_inexact:
2206 	if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2207 		goto retry;
2208 
2209 	if (ret && !xfrm_pol_hold_rcu(ret))
2210 		goto retry;
2211 fail:
2212 	rcu_read_unlock();
2213 
2214 	return ret;
2215 }
2216 
2217 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2218 					      const struct flowi *fl,
2219 					      u16 family, u8 dir, u32 if_id)
2220 {
2221 #ifdef CONFIG_XFRM_SUB_POLICY
2222 	struct xfrm_policy *pol;
2223 
2224 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2225 					dir, if_id);
2226 	if (pol != NULL)
2227 		return pol;
2228 #endif
2229 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2230 					 dir, if_id);
2231 }
2232 
2233 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2234 						 const struct flowi *fl,
2235 						 u16 family, u32 if_id)
2236 {
2237 	struct xfrm_policy *pol;
2238 
2239 	rcu_read_lock();
2240  again:
2241 	pol = rcu_dereference(sk->sk_policy[dir]);
2242 	if (pol != NULL) {
2243 		bool match;
2244 		int err = 0;
2245 
2246 		if (pol->family != family) {
2247 			pol = NULL;
2248 			goto out;
2249 		}
2250 
2251 		match = xfrm_selector_match(&pol->selector, fl, family);
2252 		if (match) {
2253 			if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v ||
2254 			    pol->if_id != if_id) {
2255 				pol = NULL;
2256 				goto out;
2257 			}
2258 			err = security_xfrm_policy_lookup(pol->security,
2259 						      fl->flowi_secid);
2260 			if (!err) {
2261 				if (!xfrm_pol_hold_rcu(pol))
2262 					goto again;
2263 			} else if (err == -ESRCH) {
2264 				pol = NULL;
2265 			} else {
2266 				pol = ERR_PTR(err);
2267 			}
2268 		} else
2269 			pol = NULL;
2270 	}
2271 out:
2272 	rcu_read_unlock();
2273 	return pol;
2274 }
2275 
2276 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2277 {
2278 	struct net *net = xp_net(pol);
2279 
2280 	list_add(&pol->walk.all, &net->xfrm.policy_all);
2281 	net->xfrm.policy_count[dir]++;
2282 	xfrm_pol_hold(pol);
2283 }
2284 
2285 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2286 						int dir)
2287 {
2288 	struct net *net = xp_net(pol);
2289 
2290 	if (list_empty(&pol->walk.all))
2291 		return NULL;
2292 
2293 	/* Socket policies are not hashed. */
2294 	if (!hlist_unhashed(&pol->bydst)) {
2295 		hlist_del_rcu(&pol->bydst);
2296 		hlist_del_init(&pol->bydst_inexact_list);
2297 		hlist_del(&pol->byidx);
2298 	}
2299 
2300 	list_del_init(&pol->walk.all);
2301 	net->xfrm.policy_count[dir]--;
2302 
2303 	return pol;
2304 }
2305 
2306 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2307 {
2308 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2309 }
2310 
2311 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2312 {
2313 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2314 }
2315 
2316 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2317 {
2318 	struct net *net = xp_net(pol);
2319 
2320 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2321 	pol = __xfrm_policy_unlink(pol, dir);
2322 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2323 	if (pol) {
2324 		xfrm_dev_policy_delete(pol);
2325 		xfrm_policy_kill(pol);
2326 		return 0;
2327 	}
2328 	return -ENOENT;
2329 }
2330 EXPORT_SYMBOL(xfrm_policy_delete);
2331 
2332 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2333 {
2334 	struct net *net = sock_net(sk);
2335 	struct xfrm_policy *old_pol;
2336 
2337 #ifdef CONFIG_XFRM_SUB_POLICY
2338 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2339 		return -EINVAL;
2340 #endif
2341 
2342 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2343 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2344 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2345 	if (pol) {
2346 		pol->curlft.add_time = ktime_get_real_seconds();
2347 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2348 		xfrm_sk_policy_link(pol, dir);
2349 	}
2350 	rcu_assign_pointer(sk->sk_policy[dir], pol);
2351 	if (old_pol) {
2352 		if (pol)
2353 			xfrm_policy_requeue(old_pol, pol);
2354 
2355 		/* Unlinking succeeds always. This is the only function
2356 		 * allowed to delete or replace socket policy.
2357 		 */
2358 		xfrm_sk_policy_unlink(old_pol, dir);
2359 	}
2360 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2361 
2362 	if (old_pol) {
2363 		xfrm_policy_kill(old_pol);
2364 	}
2365 	return 0;
2366 }
2367 
2368 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2369 {
2370 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2371 	struct net *net = xp_net(old);
2372 
2373 	if (newp) {
2374 		newp->selector = old->selector;
2375 		if (security_xfrm_policy_clone(old->security,
2376 					       &newp->security)) {
2377 			kfree(newp);
2378 			return NULL;  /* ENOMEM */
2379 		}
2380 		newp->lft = old->lft;
2381 		newp->curlft = old->curlft;
2382 		newp->mark = old->mark;
2383 		newp->if_id = old->if_id;
2384 		newp->action = old->action;
2385 		newp->flags = old->flags;
2386 		newp->xfrm_nr = old->xfrm_nr;
2387 		newp->index = old->index;
2388 		newp->type = old->type;
2389 		newp->family = old->family;
2390 		memcpy(newp->xfrm_vec, old->xfrm_vec,
2391 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2392 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2393 		xfrm_sk_policy_link(newp, dir);
2394 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2395 		xfrm_pol_put(newp);
2396 	}
2397 	return newp;
2398 }
2399 
2400 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2401 {
2402 	const struct xfrm_policy *p;
2403 	struct xfrm_policy *np;
2404 	int i, ret = 0;
2405 
2406 	rcu_read_lock();
2407 	for (i = 0; i < 2; i++) {
2408 		p = rcu_dereference(osk->sk_policy[i]);
2409 		if (p) {
2410 			np = clone_policy(p, i);
2411 			if (unlikely(!np)) {
2412 				ret = -ENOMEM;
2413 				break;
2414 			}
2415 			rcu_assign_pointer(sk->sk_policy[i], np);
2416 		}
2417 	}
2418 	rcu_read_unlock();
2419 	return ret;
2420 }
2421 
2422 static int
2423 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2424 	       xfrm_address_t *remote, unsigned short family, u32 mark)
2425 {
2426 	int err;
2427 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2428 
2429 	if (unlikely(afinfo == NULL))
2430 		return -EINVAL;
2431 	err = afinfo->get_saddr(net, oif, local, remote, mark);
2432 	rcu_read_unlock();
2433 	return err;
2434 }
2435 
2436 /* Resolve list of templates for the flow, given policy. */
2437 
2438 static int
2439 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2440 		      struct xfrm_state **xfrm, unsigned short family)
2441 {
2442 	struct net *net = xp_net(policy);
2443 	int nx;
2444 	int i, error;
2445 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2446 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2447 	xfrm_address_t tmp;
2448 
2449 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2450 		struct xfrm_state *x;
2451 		xfrm_address_t *remote = daddr;
2452 		xfrm_address_t *local  = saddr;
2453 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2454 
2455 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
2456 		    tmpl->mode == XFRM_MODE_BEET) {
2457 			remote = &tmpl->id.daddr;
2458 			local = &tmpl->saddr;
2459 			if (xfrm_addr_any(local, tmpl->encap_family)) {
2460 				error = xfrm_get_saddr(net, fl->flowi_oif,
2461 						       &tmp, remote,
2462 						       tmpl->encap_family, 0);
2463 				if (error)
2464 					goto fail;
2465 				local = &tmp;
2466 			}
2467 		}
2468 
2469 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2470 				    family, policy->if_id);
2471 
2472 		if (x && x->km.state == XFRM_STATE_VALID) {
2473 			xfrm[nx++] = x;
2474 			daddr = remote;
2475 			saddr = local;
2476 			continue;
2477 		}
2478 		if (x) {
2479 			error = (x->km.state == XFRM_STATE_ERROR ?
2480 				 -EINVAL : -EAGAIN);
2481 			xfrm_state_put(x);
2482 		} else if (error == -ESRCH) {
2483 			error = -EAGAIN;
2484 		}
2485 
2486 		if (!tmpl->optional)
2487 			goto fail;
2488 	}
2489 	return nx;
2490 
2491 fail:
2492 	for (nx--; nx >= 0; nx--)
2493 		xfrm_state_put(xfrm[nx]);
2494 	return error;
2495 }
2496 
2497 static int
2498 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2499 		  struct xfrm_state **xfrm, unsigned short family)
2500 {
2501 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
2502 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2503 	int cnx = 0;
2504 	int error;
2505 	int ret;
2506 	int i;
2507 
2508 	for (i = 0; i < npols; i++) {
2509 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2510 			error = -ENOBUFS;
2511 			goto fail;
2512 		}
2513 
2514 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2515 		if (ret < 0) {
2516 			error = ret;
2517 			goto fail;
2518 		} else
2519 			cnx += ret;
2520 	}
2521 
2522 	/* found states are sorted for outbound processing */
2523 	if (npols > 1)
2524 		xfrm_state_sort(xfrm, tpp, cnx, family);
2525 
2526 	return cnx;
2527 
2528  fail:
2529 	for (cnx--; cnx >= 0; cnx--)
2530 		xfrm_state_put(tpp[cnx]);
2531 	return error;
2532 
2533 }
2534 
2535 static int xfrm_get_tos(const struct flowi *fl, int family)
2536 {
2537 	if (family == AF_INET)
2538 		return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2539 
2540 	return 0;
2541 }
2542 
2543 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2544 {
2545 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2546 	struct dst_ops *dst_ops;
2547 	struct xfrm_dst *xdst;
2548 
2549 	if (!afinfo)
2550 		return ERR_PTR(-EINVAL);
2551 
2552 	switch (family) {
2553 	case AF_INET:
2554 		dst_ops = &net->xfrm.xfrm4_dst_ops;
2555 		break;
2556 #if IS_ENABLED(CONFIG_IPV6)
2557 	case AF_INET6:
2558 		dst_ops = &net->xfrm.xfrm6_dst_ops;
2559 		break;
2560 #endif
2561 	default:
2562 		BUG();
2563 	}
2564 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2565 
2566 	if (likely(xdst)) {
2567 		memset_after(xdst, 0, u.dst);
2568 	} else
2569 		xdst = ERR_PTR(-ENOBUFS);
2570 
2571 	rcu_read_unlock();
2572 
2573 	return xdst;
2574 }
2575 
2576 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2577 			   int nfheader_len)
2578 {
2579 	if (dst->ops->family == AF_INET6) {
2580 		struct rt6_info *rt = (struct rt6_info *)dst;
2581 		path->path_cookie = rt6_get_cookie(rt);
2582 		path->u.rt6.rt6i_nfheader_len = nfheader_len;
2583 	}
2584 }
2585 
2586 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2587 				const struct flowi *fl)
2588 {
2589 	const struct xfrm_policy_afinfo *afinfo =
2590 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2591 	int err;
2592 
2593 	if (!afinfo)
2594 		return -EINVAL;
2595 
2596 	err = afinfo->fill_dst(xdst, dev, fl);
2597 
2598 	rcu_read_unlock();
2599 
2600 	return err;
2601 }
2602 
2603 
2604 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2605  * all the metrics... Shortly, bundle a bundle.
2606  */
2607 
2608 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2609 					    struct xfrm_state **xfrm,
2610 					    struct xfrm_dst **bundle,
2611 					    int nx,
2612 					    const struct flowi *fl,
2613 					    struct dst_entry *dst)
2614 {
2615 	const struct xfrm_state_afinfo *afinfo;
2616 	const struct xfrm_mode *inner_mode;
2617 	struct net *net = xp_net(policy);
2618 	unsigned long now = jiffies;
2619 	struct net_device *dev;
2620 	struct xfrm_dst *xdst_prev = NULL;
2621 	struct xfrm_dst *xdst0 = NULL;
2622 	int i = 0;
2623 	int err;
2624 	int header_len = 0;
2625 	int nfheader_len = 0;
2626 	int trailer_len = 0;
2627 	int tos;
2628 	int family = policy->selector.family;
2629 	xfrm_address_t saddr, daddr;
2630 
2631 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2632 
2633 	tos = xfrm_get_tos(fl, family);
2634 
2635 	dst_hold(dst);
2636 
2637 	for (; i < nx; i++) {
2638 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2639 		struct dst_entry *dst1 = &xdst->u.dst;
2640 
2641 		err = PTR_ERR(xdst);
2642 		if (IS_ERR(xdst)) {
2643 			dst_release(dst);
2644 			goto put_states;
2645 		}
2646 
2647 		bundle[i] = xdst;
2648 		if (!xdst_prev)
2649 			xdst0 = xdst;
2650 		else
2651 			/* Ref count is taken during xfrm_alloc_dst()
2652 			 * No need to do dst_clone() on dst1
2653 			 */
2654 			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2655 
2656 		if (xfrm[i]->sel.family == AF_UNSPEC) {
2657 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
2658 							xfrm_af2proto(family));
2659 			if (!inner_mode) {
2660 				err = -EAFNOSUPPORT;
2661 				dst_release(dst);
2662 				goto put_states;
2663 			}
2664 		} else
2665 			inner_mode = &xfrm[i]->inner_mode;
2666 
2667 		xdst->route = dst;
2668 		dst_copy_metrics(dst1, dst);
2669 
2670 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2671 			__u32 mark = 0;
2672 			int oif;
2673 
2674 			if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2675 				mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2676 
2677 			family = xfrm[i]->props.family;
2678 			oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2679 			dst = xfrm_dst_lookup(xfrm[i], tos, oif,
2680 					      &saddr, &daddr, family, mark);
2681 			err = PTR_ERR(dst);
2682 			if (IS_ERR(dst))
2683 				goto put_states;
2684 		} else
2685 			dst_hold(dst);
2686 
2687 		dst1->xfrm = xfrm[i];
2688 		xdst->xfrm_genid = xfrm[i]->genid;
2689 
2690 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2691 		dst1->lastuse = now;
2692 
2693 		dst1->input = dst_discard;
2694 
2695 		rcu_read_lock();
2696 		afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2697 		if (likely(afinfo))
2698 			dst1->output = afinfo->output;
2699 		else
2700 			dst1->output = dst_discard_out;
2701 		rcu_read_unlock();
2702 
2703 		xdst_prev = xdst;
2704 
2705 		header_len += xfrm[i]->props.header_len;
2706 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2707 			nfheader_len += xfrm[i]->props.header_len;
2708 		trailer_len += xfrm[i]->props.trailer_len;
2709 	}
2710 
2711 	xfrm_dst_set_child(xdst_prev, dst);
2712 	xdst0->path = dst;
2713 
2714 	err = -ENODEV;
2715 	dev = dst->dev;
2716 	if (!dev)
2717 		goto free_dst;
2718 
2719 	xfrm_init_path(xdst0, dst, nfheader_len);
2720 	xfrm_init_pmtu(bundle, nx);
2721 
2722 	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2723 	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2724 		err = xfrm_fill_dst(xdst_prev, dev, fl);
2725 		if (err)
2726 			goto free_dst;
2727 
2728 		xdst_prev->u.dst.header_len = header_len;
2729 		xdst_prev->u.dst.trailer_len = trailer_len;
2730 		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2731 		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2732 	}
2733 
2734 	return &xdst0->u.dst;
2735 
2736 put_states:
2737 	for (; i < nx; i++)
2738 		xfrm_state_put(xfrm[i]);
2739 free_dst:
2740 	if (xdst0)
2741 		dst_release_immediate(&xdst0->u.dst);
2742 
2743 	return ERR_PTR(err);
2744 }
2745 
2746 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2747 				struct xfrm_policy **pols,
2748 				int *num_pols, int *num_xfrms)
2749 {
2750 	int i;
2751 
2752 	if (*num_pols == 0 || !pols[0]) {
2753 		*num_pols = 0;
2754 		*num_xfrms = 0;
2755 		return 0;
2756 	}
2757 	if (IS_ERR(pols[0])) {
2758 		*num_pols = 0;
2759 		return PTR_ERR(pols[0]);
2760 	}
2761 
2762 	*num_xfrms = pols[0]->xfrm_nr;
2763 
2764 #ifdef CONFIG_XFRM_SUB_POLICY
2765 	if (pols[0]->action == XFRM_POLICY_ALLOW &&
2766 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2767 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2768 						    XFRM_POLICY_TYPE_MAIN,
2769 						    fl, family,
2770 						    XFRM_POLICY_OUT,
2771 						    pols[0]->if_id);
2772 		if (pols[1]) {
2773 			if (IS_ERR(pols[1])) {
2774 				xfrm_pols_put(pols, *num_pols);
2775 				*num_pols = 0;
2776 				return PTR_ERR(pols[1]);
2777 			}
2778 			(*num_pols)++;
2779 			(*num_xfrms) += pols[1]->xfrm_nr;
2780 		}
2781 	}
2782 #endif
2783 	for (i = 0; i < *num_pols; i++) {
2784 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
2785 			*num_xfrms = -1;
2786 			break;
2787 		}
2788 	}
2789 
2790 	return 0;
2791 
2792 }
2793 
2794 static struct xfrm_dst *
2795 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2796 			       const struct flowi *fl, u16 family,
2797 			       struct dst_entry *dst_orig)
2798 {
2799 	struct net *net = xp_net(pols[0]);
2800 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2801 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2802 	struct xfrm_dst *xdst;
2803 	struct dst_entry *dst;
2804 	int err;
2805 
2806 	/* Try to instantiate a bundle */
2807 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2808 	if (err <= 0) {
2809 		if (err == 0)
2810 			return NULL;
2811 
2812 		if (err != -EAGAIN)
2813 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2814 		return ERR_PTR(err);
2815 	}
2816 
2817 	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2818 	if (IS_ERR(dst)) {
2819 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2820 		return ERR_CAST(dst);
2821 	}
2822 
2823 	xdst = (struct xfrm_dst *)dst;
2824 	xdst->num_xfrms = err;
2825 	xdst->num_pols = num_pols;
2826 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2827 	xdst->policy_genid = atomic_read(&pols[0]->genid);
2828 
2829 	return xdst;
2830 }
2831 
2832 static void xfrm_policy_queue_process(struct timer_list *t)
2833 {
2834 	struct sk_buff *skb;
2835 	struct sock *sk;
2836 	struct dst_entry *dst;
2837 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2838 	struct net *net = xp_net(pol);
2839 	struct xfrm_policy_queue *pq = &pol->polq;
2840 	struct flowi fl;
2841 	struct sk_buff_head list;
2842 	__u32 skb_mark;
2843 
2844 	spin_lock(&pq->hold_queue.lock);
2845 	skb = skb_peek(&pq->hold_queue);
2846 	if (!skb) {
2847 		spin_unlock(&pq->hold_queue.lock);
2848 		goto out;
2849 	}
2850 	dst = skb_dst(skb);
2851 	sk = skb->sk;
2852 
2853 	/* Fixup the mark to support VTI. */
2854 	skb_mark = skb->mark;
2855 	skb->mark = pol->mark.v;
2856 	xfrm_decode_session(skb, &fl, dst->ops->family);
2857 	skb->mark = skb_mark;
2858 	spin_unlock(&pq->hold_queue.lock);
2859 
2860 	dst_hold(xfrm_dst_path(dst));
2861 	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2862 	if (IS_ERR(dst))
2863 		goto purge_queue;
2864 
2865 	if (dst->flags & DST_XFRM_QUEUE) {
2866 		dst_release(dst);
2867 
2868 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2869 			goto purge_queue;
2870 
2871 		pq->timeout = pq->timeout << 1;
2872 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2873 			xfrm_pol_hold(pol);
2874 		goto out;
2875 	}
2876 
2877 	dst_release(dst);
2878 
2879 	__skb_queue_head_init(&list);
2880 
2881 	spin_lock(&pq->hold_queue.lock);
2882 	pq->timeout = 0;
2883 	skb_queue_splice_init(&pq->hold_queue, &list);
2884 	spin_unlock(&pq->hold_queue.lock);
2885 
2886 	while (!skb_queue_empty(&list)) {
2887 		skb = __skb_dequeue(&list);
2888 
2889 		/* Fixup the mark to support VTI. */
2890 		skb_mark = skb->mark;
2891 		skb->mark = pol->mark.v;
2892 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2893 		skb->mark = skb_mark;
2894 
2895 		dst_hold(xfrm_dst_path(skb_dst(skb)));
2896 		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2897 		if (IS_ERR(dst)) {
2898 			kfree_skb(skb);
2899 			continue;
2900 		}
2901 
2902 		nf_reset_ct(skb);
2903 		skb_dst_drop(skb);
2904 		skb_dst_set(skb, dst);
2905 
2906 		dst_output(net, skb->sk, skb);
2907 	}
2908 
2909 out:
2910 	xfrm_pol_put(pol);
2911 	return;
2912 
2913 purge_queue:
2914 	pq->timeout = 0;
2915 	skb_queue_purge(&pq->hold_queue);
2916 	xfrm_pol_put(pol);
2917 }
2918 
2919 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2920 {
2921 	unsigned long sched_next;
2922 	struct dst_entry *dst = skb_dst(skb);
2923 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2924 	struct xfrm_policy *pol = xdst->pols[0];
2925 	struct xfrm_policy_queue *pq = &pol->polq;
2926 
2927 	if (unlikely(skb_fclone_busy(sk, skb))) {
2928 		kfree_skb(skb);
2929 		return 0;
2930 	}
2931 
2932 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2933 		kfree_skb(skb);
2934 		return -EAGAIN;
2935 	}
2936 
2937 	skb_dst_force(skb);
2938 
2939 	spin_lock_bh(&pq->hold_queue.lock);
2940 
2941 	if (!pq->timeout)
2942 		pq->timeout = XFRM_QUEUE_TMO_MIN;
2943 
2944 	sched_next = jiffies + pq->timeout;
2945 
2946 	if (del_timer(&pq->hold_timer)) {
2947 		if (time_before(pq->hold_timer.expires, sched_next))
2948 			sched_next = pq->hold_timer.expires;
2949 		xfrm_pol_put(pol);
2950 	}
2951 
2952 	__skb_queue_tail(&pq->hold_queue, skb);
2953 	if (!mod_timer(&pq->hold_timer, sched_next))
2954 		xfrm_pol_hold(pol);
2955 
2956 	spin_unlock_bh(&pq->hold_queue.lock);
2957 
2958 	return 0;
2959 }
2960 
2961 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2962 						 struct xfrm_flo *xflo,
2963 						 const struct flowi *fl,
2964 						 int num_xfrms,
2965 						 u16 family)
2966 {
2967 	int err;
2968 	struct net_device *dev;
2969 	struct dst_entry *dst;
2970 	struct dst_entry *dst1;
2971 	struct xfrm_dst *xdst;
2972 
2973 	xdst = xfrm_alloc_dst(net, family);
2974 	if (IS_ERR(xdst))
2975 		return xdst;
2976 
2977 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2978 	    net->xfrm.sysctl_larval_drop ||
2979 	    num_xfrms <= 0)
2980 		return xdst;
2981 
2982 	dst = xflo->dst_orig;
2983 	dst1 = &xdst->u.dst;
2984 	dst_hold(dst);
2985 	xdst->route = dst;
2986 
2987 	dst_copy_metrics(dst1, dst);
2988 
2989 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2990 	dst1->flags |= DST_XFRM_QUEUE;
2991 	dst1->lastuse = jiffies;
2992 
2993 	dst1->input = dst_discard;
2994 	dst1->output = xdst_queue_output;
2995 
2996 	dst_hold(dst);
2997 	xfrm_dst_set_child(xdst, dst);
2998 	xdst->path = dst;
2999 
3000 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
3001 
3002 	err = -ENODEV;
3003 	dev = dst->dev;
3004 	if (!dev)
3005 		goto free_dst;
3006 
3007 	err = xfrm_fill_dst(xdst, dev, fl);
3008 	if (err)
3009 		goto free_dst;
3010 
3011 out:
3012 	return xdst;
3013 
3014 free_dst:
3015 	dst_release(dst1);
3016 	xdst = ERR_PTR(err);
3017 	goto out;
3018 }
3019 
3020 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
3021 					   const struct flowi *fl,
3022 					   u16 family, u8 dir,
3023 					   struct xfrm_flo *xflo, u32 if_id)
3024 {
3025 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3026 	int num_pols = 0, num_xfrms = 0, err;
3027 	struct xfrm_dst *xdst;
3028 
3029 	/* Resolve policies to use if we couldn't get them from
3030 	 * previous cache entry */
3031 	num_pols = 1;
3032 	pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
3033 	err = xfrm_expand_policies(fl, family, pols,
3034 					   &num_pols, &num_xfrms);
3035 	if (err < 0)
3036 		goto inc_error;
3037 	if (num_pols == 0)
3038 		return NULL;
3039 	if (num_xfrms <= 0)
3040 		goto make_dummy_bundle;
3041 
3042 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
3043 					      xflo->dst_orig);
3044 	if (IS_ERR(xdst)) {
3045 		err = PTR_ERR(xdst);
3046 		if (err == -EREMOTE) {
3047 			xfrm_pols_put(pols, num_pols);
3048 			return NULL;
3049 		}
3050 
3051 		if (err != -EAGAIN)
3052 			goto error;
3053 		goto make_dummy_bundle;
3054 	} else if (xdst == NULL) {
3055 		num_xfrms = 0;
3056 		goto make_dummy_bundle;
3057 	}
3058 
3059 	return xdst;
3060 
3061 make_dummy_bundle:
3062 	/* We found policies, but there's no bundles to instantiate:
3063 	 * either because the policy blocks, has no transformations or
3064 	 * we could not build template (no xfrm_states).*/
3065 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
3066 	if (IS_ERR(xdst)) {
3067 		xfrm_pols_put(pols, num_pols);
3068 		return ERR_CAST(xdst);
3069 	}
3070 	xdst->num_pols = num_pols;
3071 	xdst->num_xfrms = num_xfrms;
3072 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
3073 
3074 	return xdst;
3075 
3076 inc_error:
3077 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3078 error:
3079 	xfrm_pols_put(pols, num_pols);
3080 	return ERR_PTR(err);
3081 }
3082 
3083 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3084 					struct dst_entry *dst_orig)
3085 {
3086 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3087 	struct dst_entry *ret;
3088 
3089 	if (!afinfo) {
3090 		dst_release(dst_orig);
3091 		return ERR_PTR(-EINVAL);
3092 	} else {
3093 		ret = afinfo->blackhole_route(net, dst_orig);
3094 	}
3095 	rcu_read_unlock();
3096 
3097 	return ret;
3098 }
3099 
3100 /* Finds/creates a bundle for given flow and if_id
3101  *
3102  * At the moment we eat a raw IP route. Mostly to speed up lookups
3103  * on interfaces with disabled IPsec.
3104  *
3105  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3106  * compatibility
3107  */
3108 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3109 					struct dst_entry *dst_orig,
3110 					const struct flowi *fl,
3111 					const struct sock *sk,
3112 					int flags, u32 if_id)
3113 {
3114 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3115 	struct xfrm_dst *xdst;
3116 	struct dst_entry *dst, *route;
3117 	u16 family = dst_orig->ops->family;
3118 	u8 dir = XFRM_POLICY_OUT;
3119 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3120 
3121 	dst = NULL;
3122 	xdst = NULL;
3123 	route = NULL;
3124 
3125 	sk = sk_const_to_full_sk(sk);
3126 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3127 		num_pols = 1;
3128 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3129 						if_id);
3130 		err = xfrm_expand_policies(fl, family, pols,
3131 					   &num_pols, &num_xfrms);
3132 		if (err < 0)
3133 			goto dropdst;
3134 
3135 		if (num_pols) {
3136 			if (num_xfrms <= 0) {
3137 				drop_pols = num_pols;
3138 				goto no_transform;
3139 			}
3140 
3141 			xdst = xfrm_resolve_and_create_bundle(
3142 					pols, num_pols, fl,
3143 					family, dst_orig);
3144 
3145 			if (IS_ERR(xdst)) {
3146 				xfrm_pols_put(pols, num_pols);
3147 				err = PTR_ERR(xdst);
3148 				if (err == -EREMOTE)
3149 					goto nopol;
3150 
3151 				goto dropdst;
3152 			} else if (xdst == NULL) {
3153 				num_xfrms = 0;
3154 				drop_pols = num_pols;
3155 				goto no_transform;
3156 			}
3157 
3158 			route = xdst->route;
3159 		}
3160 	}
3161 
3162 	if (xdst == NULL) {
3163 		struct xfrm_flo xflo;
3164 
3165 		xflo.dst_orig = dst_orig;
3166 		xflo.flags = flags;
3167 
3168 		/* To accelerate a bit...  */
3169 		if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3170 			       !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3171 			goto nopol;
3172 
3173 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3174 		if (xdst == NULL)
3175 			goto nopol;
3176 		if (IS_ERR(xdst)) {
3177 			err = PTR_ERR(xdst);
3178 			goto dropdst;
3179 		}
3180 
3181 		num_pols = xdst->num_pols;
3182 		num_xfrms = xdst->num_xfrms;
3183 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3184 		route = xdst->route;
3185 	}
3186 
3187 	dst = &xdst->u.dst;
3188 	if (route == NULL && num_xfrms > 0) {
3189 		/* The only case when xfrm_bundle_lookup() returns a
3190 		 * bundle with null route, is when the template could
3191 		 * not be resolved. It means policies are there, but
3192 		 * bundle could not be created, since we don't yet
3193 		 * have the xfrm_state's. We need to wait for KM to
3194 		 * negotiate new SA's or bail out with error.*/
3195 		if (net->xfrm.sysctl_larval_drop) {
3196 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3197 			err = -EREMOTE;
3198 			goto error;
3199 		}
3200 
3201 		err = -EAGAIN;
3202 
3203 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3204 		goto error;
3205 	}
3206 
3207 no_transform:
3208 	if (num_pols == 0)
3209 		goto nopol;
3210 
3211 	if ((flags & XFRM_LOOKUP_ICMP) &&
3212 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3213 		err = -ENOENT;
3214 		goto error;
3215 	}
3216 
3217 	for (i = 0; i < num_pols; i++)
3218 		pols[i]->curlft.use_time = ktime_get_real_seconds();
3219 
3220 	if (num_xfrms < 0) {
3221 		/* Prohibit the flow */
3222 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3223 		err = -EPERM;
3224 		goto error;
3225 	} else if (num_xfrms > 0) {
3226 		/* Flow transformed */
3227 		dst_release(dst_orig);
3228 	} else {
3229 		/* Flow passes untransformed */
3230 		dst_release(dst);
3231 		dst = dst_orig;
3232 	}
3233 ok:
3234 	xfrm_pols_put(pols, drop_pols);
3235 	if (dst && dst->xfrm &&
3236 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3237 		dst->flags |= DST_XFRM_TUNNEL;
3238 	return dst;
3239 
3240 nopol:
3241 	if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3242 	    net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3243 		err = -EPERM;
3244 		goto error;
3245 	}
3246 	if (!(flags & XFRM_LOOKUP_ICMP)) {
3247 		dst = dst_orig;
3248 		goto ok;
3249 	}
3250 	err = -ENOENT;
3251 error:
3252 	dst_release(dst);
3253 dropdst:
3254 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3255 		dst_release(dst_orig);
3256 	xfrm_pols_put(pols, drop_pols);
3257 	return ERR_PTR(err);
3258 }
3259 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3260 
3261 /* Main function: finds/creates a bundle for given flow.
3262  *
3263  * At the moment we eat a raw IP route. Mostly to speed up lookups
3264  * on interfaces with disabled IPsec.
3265  */
3266 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3267 			      const struct flowi *fl, const struct sock *sk,
3268 			      int flags)
3269 {
3270 	return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3271 }
3272 EXPORT_SYMBOL(xfrm_lookup);
3273 
3274 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3275  * Otherwise we may send out blackholed packets.
3276  */
3277 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3278 				    const struct flowi *fl,
3279 				    const struct sock *sk, int flags)
3280 {
3281 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3282 					    flags | XFRM_LOOKUP_QUEUE |
3283 					    XFRM_LOOKUP_KEEP_DST_REF);
3284 
3285 	if (PTR_ERR(dst) == -EREMOTE)
3286 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
3287 
3288 	if (IS_ERR(dst))
3289 		dst_release(dst_orig);
3290 
3291 	return dst;
3292 }
3293 EXPORT_SYMBOL(xfrm_lookup_route);
3294 
3295 static inline int
3296 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3297 {
3298 	struct sec_path *sp = skb_sec_path(skb);
3299 	struct xfrm_state *x;
3300 
3301 	if (!sp || idx < 0 || idx >= sp->len)
3302 		return 0;
3303 	x = sp->xvec[idx];
3304 	if (!x->type->reject)
3305 		return 0;
3306 	return x->type->reject(x, skb, fl);
3307 }
3308 
3309 /* When skb is transformed back to its "native" form, we have to
3310  * check policy restrictions. At the moment we make this in maximally
3311  * stupid way. Shame on me. :-) Of course, connected sockets must
3312  * have policy cached at them.
3313  */
3314 
3315 static inline int
3316 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3317 	      unsigned short family, u32 if_id)
3318 {
3319 	if (xfrm_state_kern(x))
3320 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3321 	return	x->id.proto == tmpl->id.proto &&
3322 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3323 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3324 		x->props.mode == tmpl->mode &&
3325 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3326 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3327 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
3328 		  xfrm_state_addr_cmp(tmpl, x, family)) &&
3329 		(if_id == 0 || if_id == x->if_id);
3330 }
3331 
3332 /*
3333  * 0 or more than 0 is returned when validation is succeeded (either bypass
3334  * because of optional transport mode, or next index of the matched secpath
3335  * state with the template.
3336  * -1 is returned when no matching template is found.
3337  * Otherwise "-2 - errored_index" is returned.
3338  */
3339 static inline int
3340 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3341 	       unsigned short family, u32 if_id)
3342 {
3343 	int idx = start;
3344 
3345 	if (tmpl->optional) {
3346 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
3347 			return start;
3348 	} else
3349 		start = -1;
3350 	for (; idx < sp->len; idx++) {
3351 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3352 			return ++idx;
3353 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3354 			if (idx < sp->verified_cnt) {
3355 				/* Secpath entry previously verified, consider optional and
3356 				 * continue searching
3357 				 */
3358 				continue;
3359 			}
3360 
3361 			if (start == -1)
3362 				start = -2-idx;
3363 			break;
3364 		}
3365 	}
3366 	return start;
3367 }
3368 
3369 static void
3370 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3371 {
3372 	const struct iphdr *iph = ip_hdr(skb);
3373 	int ihl = iph->ihl;
3374 	u8 *xprth = skb_network_header(skb) + ihl * 4;
3375 	struct flowi4 *fl4 = &fl->u.ip4;
3376 	int oif = 0;
3377 
3378 	if (skb_dst(skb) && skb_dst(skb)->dev)
3379 		oif = skb_dst(skb)->dev->ifindex;
3380 
3381 	memset(fl4, 0, sizeof(struct flowi4));
3382 	fl4->flowi4_mark = skb->mark;
3383 	fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3384 
3385 	fl4->flowi4_proto = iph->protocol;
3386 	fl4->daddr = reverse ? iph->saddr : iph->daddr;
3387 	fl4->saddr = reverse ? iph->daddr : iph->saddr;
3388 	fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3389 
3390 	if (!ip_is_fragment(iph)) {
3391 		switch (iph->protocol) {
3392 		case IPPROTO_UDP:
3393 		case IPPROTO_UDPLITE:
3394 		case IPPROTO_TCP:
3395 		case IPPROTO_SCTP:
3396 		case IPPROTO_DCCP:
3397 			if (xprth + 4 < skb->data ||
3398 			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3399 				__be16 *ports;
3400 
3401 				xprth = skb_network_header(skb) + ihl * 4;
3402 				ports = (__be16 *)xprth;
3403 
3404 				fl4->fl4_sport = ports[!!reverse];
3405 				fl4->fl4_dport = ports[!reverse];
3406 			}
3407 			break;
3408 		case IPPROTO_ICMP:
3409 			if (xprth + 2 < skb->data ||
3410 			    pskb_may_pull(skb, xprth + 2 - skb->data)) {
3411 				u8 *icmp;
3412 
3413 				xprth = skb_network_header(skb) + ihl * 4;
3414 				icmp = xprth;
3415 
3416 				fl4->fl4_icmp_type = icmp[0];
3417 				fl4->fl4_icmp_code = icmp[1];
3418 			}
3419 			break;
3420 		case IPPROTO_GRE:
3421 			if (xprth + 12 < skb->data ||
3422 			    pskb_may_pull(skb, xprth + 12 - skb->data)) {
3423 				__be16 *greflags;
3424 				__be32 *gre_hdr;
3425 
3426 				xprth = skb_network_header(skb) + ihl * 4;
3427 				greflags = (__be16 *)xprth;
3428 				gre_hdr = (__be32 *)xprth;
3429 
3430 				if (greflags[0] & GRE_KEY) {
3431 					if (greflags[0] & GRE_CSUM)
3432 						gre_hdr++;
3433 					fl4->fl4_gre_key = gre_hdr[1];
3434 				}
3435 			}
3436 			break;
3437 		default:
3438 			break;
3439 		}
3440 	}
3441 }
3442 
3443 #if IS_ENABLED(CONFIG_IPV6)
3444 static void
3445 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3446 {
3447 	struct flowi6 *fl6 = &fl->u.ip6;
3448 	int onlyproto = 0;
3449 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
3450 	u32 offset = sizeof(*hdr);
3451 	struct ipv6_opt_hdr *exthdr;
3452 	const unsigned char *nh = skb_network_header(skb);
3453 	u16 nhoff = IP6CB(skb)->nhoff;
3454 	int oif = 0;
3455 	u8 nexthdr;
3456 
3457 	if (!nhoff)
3458 		nhoff = offsetof(struct ipv6hdr, nexthdr);
3459 
3460 	nexthdr = nh[nhoff];
3461 
3462 	if (skb_dst(skb) && skb_dst(skb)->dev)
3463 		oif = skb_dst(skb)->dev->ifindex;
3464 
3465 	memset(fl6, 0, sizeof(struct flowi6));
3466 	fl6->flowi6_mark = skb->mark;
3467 	fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3468 
3469 	fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3470 	fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3471 
3472 	while (nh + offset + sizeof(*exthdr) < skb->data ||
3473 	       pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3474 		nh = skb_network_header(skb);
3475 		exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3476 
3477 		switch (nexthdr) {
3478 		case NEXTHDR_FRAGMENT:
3479 			onlyproto = 1;
3480 			fallthrough;
3481 		case NEXTHDR_ROUTING:
3482 		case NEXTHDR_HOP:
3483 		case NEXTHDR_DEST:
3484 			offset += ipv6_optlen(exthdr);
3485 			nexthdr = exthdr->nexthdr;
3486 			break;
3487 		case IPPROTO_UDP:
3488 		case IPPROTO_UDPLITE:
3489 		case IPPROTO_TCP:
3490 		case IPPROTO_SCTP:
3491 		case IPPROTO_DCCP:
3492 			if (!onlyproto && (nh + offset + 4 < skb->data ||
3493 			     pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3494 				__be16 *ports;
3495 
3496 				nh = skb_network_header(skb);
3497 				ports = (__be16 *)(nh + offset);
3498 				fl6->fl6_sport = ports[!!reverse];
3499 				fl6->fl6_dport = ports[!reverse];
3500 			}
3501 			fl6->flowi6_proto = nexthdr;
3502 			return;
3503 		case IPPROTO_ICMPV6:
3504 			if (!onlyproto && (nh + offset + 2 < skb->data ||
3505 			    pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3506 				u8 *icmp;
3507 
3508 				nh = skb_network_header(skb);
3509 				icmp = (u8 *)(nh + offset);
3510 				fl6->fl6_icmp_type = icmp[0];
3511 				fl6->fl6_icmp_code = icmp[1];
3512 			}
3513 			fl6->flowi6_proto = nexthdr;
3514 			return;
3515 		case IPPROTO_GRE:
3516 			if (!onlyproto &&
3517 			    (nh + offset + 12 < skb->data ||
3518 			     pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3519 				struct gre_base_hdr *gre_hdr;
3520 				__be32 *gre_key;
3521 
3522 				nh = skb_network_header(skb);
3523 				gre_hdr = (struct gre_base_hdr *)(nh + offset);
3524 				gre_key = (__be32 *)(gre_hdr + 1);
3525 
3526 				if (gre_hdr->flags & GRE_KEY) {
3527 					if (gre_hdr->flags & GRE_CSUM)
3528 						gre_key++;
3529 					fl6->fl6_gre_key = *gre_key;
3530 				}
3531 			}
3532 			fl6->flowi6_proto = nexthdr;
3533 			return;
3534 
3535 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3536 		case IPPROTO_MH:
3537 			offset += ipv6_optlen(exthdr);
3538 			if (!onlyproto && (nh + offset + 3 < skb->data ||
3539 			    pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3540 				struct ip6_mh *mh;
3541 
3542 				nh = skb_network_header(skb);
3543 				mh = (struct ip6_mh *)(nh + offset);
3544 				fl6->fl6_mh_type = mh->ip6mh_type;
3545 			}
3546 			fl6->flowi6_proto = nexthdr;
3547 			return;
3548 #endif
3549 		default:
3550 			fl6->flowi6_proto = nexthdr;
3551 			return;
3552 		}
3553 	}
3554 }
3555 #endif
3556 
3557 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3558 			  unsigned int family, int reverse)
3559 {
3560 	switch (family) {
3561 	case AF_INET:
3562 		decode_session4(skb, fl, reverse);
3563 		break;
3564 #if IS_ENABLED(CONFIG_IPV6)
3565 	case AF_INET6:
3566 		decode_session6(skb, fl, reverse);
3567 		break;
3568 #endif
3569 	default:
3570 		return -EAFNOSUPPORT;
3571 	}
3572 
3573 	return security_xfrm_decode_session(skb, &fl->flowi_secid);
3574 }
3575 EXPORT_SYMBOL(__xfrm_decode_session);
3576 
3577 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3578 {
3579 	for (; k < sp->len; k++) {
3580 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3581 			*idxp = k;
3582 			return 1;
3583 		}
3584 	}
3585 
3586 	return 0;
3587 }
3588 
3589 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3590 			unsigned short family)
3591 {
3592 	struct net *net = dev_net(skb->dev);
3593 	struct xfrm_policy *pol;
3594 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3595 	int npols = 0;
3596 	int xfrm_nr;
3597 	int pi;
3598 	int reverse;
3599 	struct flowi fl;
3600 	int xerr_idx = -1;
3601 	const struct xfrm_if_cb *ifcb;
3602 	struct sec_path *sp;
3603 	u32 if_id = 0;
3604 
3605 	rcu_read_lock();
3606 	ifcb = xfrm_if_get_cb();
3607 
3608 	if (ifcb) {
3609 		struct xfrm_if_decode_session_result r;
3610 
3611 		if (ifcb->decode_session(skb, family, &r)) {
3612 			if_id = r.if_id;
3613 			net = r.net;
3614 		}
3615 	}
3616 	rcu_read_unlock();
3617 
3618 	reverse = dir & ~XFRM_POLICY_MASK;
3619 	dir &= XFRM_POLICY_MASK;
3620 
3621 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3622 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3623 		return 0;
3624 	}
3625 
3626 	nf_nat_decode_session(skb, &fl, family);
3627 
3628 	/* First, check used SA against their selectors. */
3629 	sp = skb_sec_path(skb);
3630 	if (sp) {
3631 		int i;
3632 
3633 		for (i = sp->len - 1; i >= 0; i--) {
3634 			struct xfrm_state *x = sp->xvec[i];
3635 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
3636 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3637 				return 0;
3638 			}
3639 		}
3640 	}
3641 
3642 	pol = NULL;
3643 	sk = sk_to_full_sk(sk);
3644 	if (sk && sk->sk_policy[dir]) {
3645 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3646 		if (IS_ERR(pol)) {
3647 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3648 			return 0;
3649 		}
3650 	}
3651 
3652 	if (!pol)
3653 		pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3654 
3655 	if (IS_ERR(pol)) {
3656 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3657 		return 0;
3658 	}
3659 
3660 	if (!pol) {
3661 		if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3662 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3663 			return 0;
3664 		}
3665 
3666 		if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3667 			xfrm_secpath_reject(xerr_idx, skb, &fl);
3668 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3669 			return 0;
3670 		}
3671 		return 1;
3672 	}
3673 
3674 	/* This lockless write can happen from different cpus. */
3675 	WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3676 
3677 	pols[0] = pol;
3678 	npols++;
3679 #ifdef CONFIG_XFRM_SUB_POLICY
3680 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3681 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3682 						    &fl, family,
3683 						    XFRM_POLICY_IN, if_id);
3684 		if (pols[1]) {
3685 			if (IS_ERR(pols[1])) {
3686 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3687 				xfrm_pol_put(pols[0]);
3688 				return 0;
3689 			}
3690 			/* This write can happen from different cpus. */
3691 			WRITE_ONCE(pols[1]->curlft.use_time,
3692 				   ktime_get_real_seconds());
3693 			npols++;
3694 		}
3695 	}
3696 #endif
3697 
3698 	if (pol->action == XFRM_POLICY_ALLOW) {
3699 		static struct sec_path dummy;
3700 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3701 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3702 		struct xfrm_tmpl **tpp = tp;
3703 		int ti = 0;
3704 		int i, k;
3705 
3706 		sp = skb_sec_path(skb);
3707 		if (!sp)
3708 			sp = &dummy;
3709 
3710 		for (pi = 0; pi < npols; pi++) {
3711 			if (pols[pi] != pol &&
3712 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
3713 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3714 				goto reject;
3715 			}
3716 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3717 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3718 				goto reject_error;
3719 			}
3720 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
3721 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
3722 		}
3723 		xfrm_nr = ti;
3724 
3725 		if (npols > 1) {
3726 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3727 			tpp = stp;
3728 		}
3729 
3730 		/* For each tunnel xfrm, find the first matching tmpl.
3731 		 * For each tmpl before that, find corresponding xfrm.
3732 		 * Order is _important_. Later we will implement
3733 		 * some barriers, but at the moment barriers
3734 		 * are implied between each two transformations.
3735 		 * Upon success, marks secpath entries as having been
3736 		 * verified to allow them to be skipped in future policy
3737 		 * checks (e.g. nested tunnels).
3738 		 */
3739 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3740 			k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3741 			if (k < 0) {
3742 				if (k < -1)
3743 					/* "-2 - errored_index" returned */
3744 					xerr_idx = -(2+k);
3745 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3746 				goto reject;
3747 			}
3748 		}
3749 
3750 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3751 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3752 			goto reject;
3753 		}
3754 
3755 		xfrm_pols_put(pols, npols);
3756 		sp->verified_cnt = k;
3757 
3758 		return 1;
3759 	}
3760 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3761 
3762 reject:
3763 	xfrm_secpath_reject(xerr_idx, skb, &fl);
3764 reject_error:
3765 	xfrm_pols_put(pols, npols);
3766 	return 0;
3767 }
3768 EXPORT_SYMBOL(__xfrm_policy_check);
3769 
3770 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3771 {
3772 	struct net *net = dev_net(skb->dev);
3773 	struct flowi fl;
3774 	struct dst_entry *dst;
3775 	int res = 1;
3776 
3777 	if (xfrm_decode_session(skb, &fl, family) < 0) {
3778 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3779 		return 0;
3780 	}
3781 
3782 	skb_dst_force(skb);
3783 	if (!skb_dst(skb)) {
3784 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3785 		return 0;
3786 	}
3787 
3788 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3789 	if (IS_ERR(dst)) {
3790 		res = 0;
3791 		dst = NULL;
3792 	}
3793 	skb_dst_set(skb, dst);
3794 	return res;
3795 }
3796 EXPORT_SYMBOL(__xfrm_route_forward);
3797 
3798 /* Optimize later using cookies and generation ids. */
3799 
3800 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3801 {
3802 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3803 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3804 	 * get validated by dst_ops->check on every use.  We do this
3805 	 * because when a normal route referenced by an XFRM dst is
3806 	 * obsoleted we do not go looking around for all parent
3807 	 * referencing XFRM dsts so that we can invalidate them.  It
3808 	 * is just too much work.  Instead we make the checks here on
3809 	 * every use.  For example:
3810 	 *
3811 	 *	XFRM dst A --> IPv4 dst X
3812 	 *
3813 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
3814 	 * in this example).  If X is marked obsolete, "A" will not
3815 	 * notice.  That's what we are validating here via the
3816 	 * stale_bundle() check.
3817 	 *
3818 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3819 	 * be marked on it.
3820 	 * This will force stale_bundle() to fail on any xdst bundle with
3821 	 * this dst linked in it.
3822 	 */
3823 	if (dst->obsolete < 0 && !stale_bundle(dst))
3824 		return dst;
3825 
3826 	return NULL;
3827 }
3828 
3829 static int stale_bundle(struct dst_entry *dst)
3830 {
3831 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3832 }
3833 
3834 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3835 {
3836 	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3837 		dst->dev = blackhole_netdev;
3838 		dev_hold(dst->dev);
3839 		dev_put(dev);
3840 	}
3841 }
3842 EXPORT_SYMBOL(xfrm_dst_ifdown);
3843 
3844 static void xfrm_link_failure(struct sk_buff *skb)
3845 {
3846 	/* Impossible. Such dst must be popped before reaches point of failure. */
3847 }
3848 
3849 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3850 {
3851 	if (dst) {
3852 		if (dst->obsolete) {
3853 			dst_release(dst);
3854 			dst = NULL;
3855 		}
3856 	}
3857 	return dst;
3858 }
3859 
3860 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3861 {
3862 	while (nr--) {
3863 		struct xfrm_dst *xdst = bundle[nr];
3864 		u32 pmtu, route_mtu_cached;
3865 		struct dst_entry *dst;
3866 
3867 		dst = &xdst->u.dst;
3868 		pmtu = dst_mtu(xfrm_dst_child(dst));
3869 		xdst->child_mtu_cached = pmtu;
3870 
3871 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3872 
3873 		route_mtu_cached = dst_mtu(xdst->route);
3874 		xdst->route_mtu_cached = route_mtu_cached;
3875 
3876 		if (pmtu > route_mtu_cached)
3877 			pmtu = route_mtu_cached;
3878 
3879 		dst_metric_set(dst, RTAX_MTU, pmtu);
3880 	}
3881 }
3882 
3883 /* Check that the bundle accepts the flow and its components are
3884  * still valid.
3885  */
3886 
3887 static int xfrm_bundle_ok(struct xfrm_dst *first)
3888 {
3889 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3890 	struct dst_entry *dst = &first->u.dst;
3891 	struct xfrm_dst *xdst;
3892 	int start_from, nr;
3893 	u32 mtu;
3894 
3895 	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3896 	    (dst->dev && !netif_running(dst->dev)))
3897 		return 0;
3898 
3899 	if (dst->flags & DST_XFRM_QUEUE)
3900 		return 1;
3901 
3902 	start_from = nr = 0;
3903 	do {
3904 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3905 
3906 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
3907 			return 0;
3908 		if (xdst->xfrm_genid != dst->xfrm->genid)
3909 			return 0;
3910 		if (xdst->num_pols > 0 &&
3911 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3912 			return 0;
3913 
3914 		bundle[nr++] = xdst;
3915 
3916 		mtu = dst_mtu(xfrm_dst_child(dst));
3917 		if (xdst->child_mtu_cached != mtu) {
3918 			start_from = nr;
3919 			xdst->child_mtu_cached = mtu;
3920 		}
3921 
3922 		if (!dst_check(xdst->route, xdst->route_cookie))
3923 			return 0;
3924 		mtu = dst_mtu(xdst->route);
3925 		if (xdst->route_mtu_cached != mtu) {
3926 			start_from = nr;
3927 			xdst->route_mtu_cached = mtu;
3928 		}
3929 
3930 		dst = xfrm_dst_child(dst);
3931 	} while (dst->xfrm);
3932 
3933 	if (likely(!start_from))
3934 		return 1;
3935 
3936 	xdst = bundle[start_from - 1];
3937 	mtu = xdst->child_mtu_cached;
3938 	while (start_from--) {
3939 		dst = &xdst->u.dst;
3940 
3941 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
3942 		if (mtu > xdst->route_mtu_cached)
3943 			mtu = xdst->route_mtu_cached;
3944 		dst_metric_set(dst, RTAX_MTU, mtu);
3945 		if (!start_from)
3946 			break;
3947 
3948 		xdst = bundle[start_from - 1];
3949 		xdst->child_mtu_cached = mtu;
3950 	}
3951 
3952 	return 1;
3953 }
3954 
3955 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3956 {
3957 	return dst_metric_advmss(xfrm_dst_path(dst));
3958 }
3959 
3960 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3961 {
3962 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3963 
3964 	return mtu ? : dst_mtu(xfrm_dst_path(dst));
3965 }
3966 
3967 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3968 					const void *daddr)
3969 {
3970 	while (dst->xfrm) {
3971 		const struct xfrm_state *xfrm = dst->xfrm;
3972 
3973 		dst = xfrm_dst_child(dst);
3974 
3975 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3976 			continue;
3977 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3978 			daddr = xfrm->coaddr;
3979 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3980 			daddr = &xfrm->id.daddr;
3981 	}
3982 	return daddr;
3983 }
3984 
3985 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3986 					   struct sk_buff *skb,
3987 					   const void *daddr)
3988 {
3989 	const struct dst_entry *path = xfrm_dst_path(dst);
3990 
3991 	if (!skb)
3992 		daddr = xfrm_get_dst_nexthop(dst, daddr);
3993 	return path->ops->neigh_lookup(path, skb, daddr);
3994 }
3995 
3996 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3997 {
3998 	const struct dst_entry *path = xfrm_dst_path(dst);
3999 
4000 	daddr = xfrm_get_dst_nexthop(dst, daddr);
4001 	path->ops->confirm_neigh(path, daddr);
4002 }
4003 
4004 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
4005 {
4006 	int err = 0;
4007 
4008 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
4009 		return -EAFNOSUPPORT;
4010 
4011 	spin_lock(&xfrm_policy_afinfo_lock);
4012 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
4013 		err = -EEXIST;
4014 	else {
4015 		struct dst_ops *dst_ops = afinfo->dst_ops;
4016 		if (likely(dst_ops->kmem_cachep == NULL))
4017 			dst_ops->kmem_cachep = xfrm_dst_cache;
4018 		if (likely(dst_ops->check == NULL))
4019 			dst_ops->check = xfrm_dst_check;
4020 		if (likely(dst_ops->default_advmss == NULL))
4021 			dst_ops->default_advmss = xfrm_default_advmss;
4022 		if (likely(dst_ops->mtu == NULL))
4023 			dst_ops->mtu = xfrm_mtu;
4024 		if (likely(dst_ops->negative_advice == NULL))
4025 			dst_ops->negative_advice = xfrm_negative_advice;
4026 		if (likely(dst_ops->link_failure == NULL))
4027 			dst_ops->link_failure = xfrm_link_failure;
4028 		if (likely(dst_ops->neigh_lookup == NULL))
4029 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
4030 		if (likely(!dst_ops->confirm_neigh))
4031 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
4032 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
4033 	}
4034 	spin_unlock(&xfrm_policy_afinfo_lock);
4035 
4036 	return err;
4037 }
4038 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
4039 
4040 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
4041 {
4042 	struct dst_ops *dst_ops = afinfo->dst_ops;
4043 	int i;
4044 
4045 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
4046 		if (xfrm_policy_afinfo[i] != afinfo)
4047 			continue;
4048 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
4049 		break;
4050 	}
4051 
4052 	synchronize_rcu();
4053 
4054 	dst_ops->kmem_cachep = NULL;
4055 	dst_ops->check = NULL;
4056 	dst_ops->negative_advice = NULL;
4057 	dst_ops->link_failure = NULL;
4058 }
4059 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
4060 
4061 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
4062 {
4063 	spin_lock(&xfrm_if_cb_lock);
4064 	rcu_assign_pointer(xfrm_if_cb, ifcb);
4065 	spin_unlock(&xfrm_if_cb_lock);
4066 }
4067 EXPORT_SYMBOL(xfrm_if_register_cb);
4068 
4069 void xfrm_if_unregister_cb(void)
4070 {
4071 	RCU_INIT_POINTER(xfrm_if_cb, NULL);
4072 	synchronize_rcu();
4073 }
4074 EXPORT_SYMBOL(xfrm_if_unregister_cb);
4075 
4076 #ifdef CONFIG_XFRM_STATISTICS
4077 static int __net_init xfrm_statistics_init(struct net *net)
4078 {
4079 	int rv;
4080 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4081 	if (!net->mib.xfrm_statistics)
4082 		return -ENOMEM;
4083 	rv = xfrm_proc_init(net);
4084 	if (rv < 0)
4085 		free_percpu(net->mib.xfrm_statistics);
4086 	return rv;
4087 }
4088 
4089 static void xfrm_statistics_fini(struct net *net)
4090 {
4091 	xfrm_proc_fini(net);
4092 	free_percpu(net->mib.xfrm_statistics);
4093 }
4094 #else
4095 static int __net_init xfrm_statistics_init(struct net *net)
4096 {
4097 	return 0;
4098 }
4099 
4100 static void xfrm_statistics_fini(struct net *net)
4101 {
4102 }
4103 #endif
4104 
4105 static int __net_init xfrm_policy_init(struct net *net)
4106 {
4107 	unsigned int hmask, sz;
4108 	int dir, err;
4109 
4110 	if (net_eq(net, &init_net)) {
4111 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4112 					   sizeof(struct xfrm_dst),
4113 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4114 					   NULL);
4115 		err = rhashtable_init(&xfrm_policy_inexact_table,
4116 				      &xfrm_pol_inexact_params);
4117 		BUG_ON(err);
4118 	}
4119 
4120 	hmask = 8 - 1;
4121 	sz = (hmask+1) * sizeof(struct hlist_head);
4122 
4123 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4124 	if (!net->xfrm.policy_byidx)
4125 		goto out_byidx;
4126 	net->xfrm.policy_idx_hmask = hmask;
4127 
4128 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4129 		struct xfrm_policy_hash *htab;
4130 
4131 		net->xfrm.policy_count[dir] = 0;
4132 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4133 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4134 
4135 		htab = &net->xfrm.policy_bydst[dir];
4136 		htab->table = xfrm_hash_alloc(sz);
4137 		if (!htab->table)
4138 			goto out_bydst;
4139 		htab->hmask = hmask;
4140 		htab->dbits4 = 32;
4141 		htab->sbits4 = 32;
4142 		htab->dbits6 = 128;
4143 		htab->sbits6 = 128;
4144 	}
4145 	net->xfrm.policy_hthresh.lbits4 = 32;
4146 	net->xfrm.policy_hthresh.rbits4 = 32;
4147 	net->xfrm.policy_hthresh.lbits6 = 128;
4148 	net->xfrm.policy_hthresh.rbits6 = 128;
4149 
4150 	seqlock_init(&net->xfrm.policy_hthresh.lock);
4151 
4152 	INIT_LIST_HEAD(&net->xfrm.policy_all);
4153 	INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4154 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4155 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4156 	return 0;
4157 
4158 out_bydst:
4159 	for (dir--; dir >= 0; dir--) {
4160 		struct xfrm_policy_hash *htab;
4161 
4162 		htab = &net->xfrm.policy_bydst[dir];
4163 		xfrm_hash_free(htab->table, sz);
4164 	}
4165 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4166 out_byidx:
4167 	return -ENOMEM;
4168 }
4169 
4170 static void xfrm_policy_fini(struct net *net)
4171 {
4172 	struct xfrm_pol_inexact_bin *b, *t;
4173 	unsigned int sz;
4174 	int dir;
4175 
4176 	flush_work(&net->xfrm.policy_hash_work);
4177 #ifdef CONFIG_XFRM_SUB_POLICY
4178 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4179 #endif
4180 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4181 
4182 	WARN_ON(!list_empty(&net->xfrm.policy_all));
4183 
4184 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4185 		struct xfrm_policy_hash *htab;
4186 
4187 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4188 
4189 		htab = &net->xfrm.policy_bydst[dir];
4190 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4191 		WARN_ON(!hlist_empty(htab->table));
4192 		xfrm_hash_free(htab->table, sz);
4193 	}
4194 
4195 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4196 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4197 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4198 
4199 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4200 	list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4201 		__xfrm_policy_inexact_prune_bin(b, true);
4202 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4203 }
4204 
4205 static int __net_init xfrm_net_init(struct net *net)
4206 {
4207 	int rv;
4208 
4209 	/* Initialize the per-net locks here */
4210 	spin_lock_init(&net->xfrm.xfrm_state_lock);
4211 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
4212 	seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4213 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
4214 	net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4215 	net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4216 	net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4217 
4218 	rv = xfrm_statistics_init(net);
4219 	if (rv < 0)
4220 		goto out_statistics;
4221 	rv = xfrm_state_init(net);
4222 	if (rv < 0)
4223 		goto out_state;
4224 	rv = xfrm_policy_init(net);
4225 	if (rv < 0)
4226 		goto out_policy;
4227 	rv = xfrm_sysctl_init(net);
4228 	if (rv < 0)
4229 		goto out_sysctl;
4230 
4231 	return 0;
4232 
4233 out_sysctl:
4234 	xfrm_policy_fini(net);
4235 out_policy:
4236 	xfrm_state_fini(net);
4237 out_state:
4238 	xfrm_statistics_fini(net);
4239 out_statistics:
4240 	return rv;
4241 }
4242 
4243 static void __net_exit xfrm_net_exit(struct net *net)
4244 {
4245 	xfrm_sysctl_fini(net);
4246 	xfrm_policy_fini(net);
4247 	xfrm_state_fini(net);
4248 	xfrm_statistics_fini(net);
4249 }
4250 
4251 static struct pernet_operations __net_initdata xfrm_net_ops = {
4252 	.init = xfrm_net_init,
4253 	.exit = xfrm_net_exit,
4254 };
4255 
4256 void __init xfrm_init(void)
4257 {
4258 	register_pernet_subsys(&xfrm_net_ops);
4259 	xfrm_dev_init();
4260 	xfrm_input_init();
4261 
4262 #ifdef CONFIG_XFRM_ESPINTCP
4263 	espintcp_init();
4264 #endif
4265 }
4266 
4267 #ifdef CONFIG_AUDITSYSCALL
4268 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4269 					 struct audit_buffer *audit_buf)
4270 {
4271 	struct xfrm_sec_ctx *ctx = xp->security;
4272 	struct xfrm_selector *sel = &xp->selector;
4273 
4274 	if (ctx)
4275 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4276 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4277 
4278 	switch (sel->family) {
4279 	case AF_INET:
4280 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4281 		if (sel->prefixlen_s != 32)
4282 			audit_log_format(audit_buf, " src_prefixlen=%d",
4283 					 sel->prefixlen_s);
4284 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4285 		if (sel->prefixlen_d != 32)
4286 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4287 					 sel->prefixlen_d);
4288 		break;
4289 	case AF_INET6:
4290 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4291 		if (sel->prefixlen_s != 128)
4292 			audit_log_format(audit_buf, " src_prefixlen=%d",
4293 					 sel->prefixlen_s);
4294 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4295 		if (sel->prefixlen_d != 128)
4296 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4297 					 sel->prefixlen_d);
4298 		break;
4299 	}
4300 }
4301 
4302 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4303 {
4304 	struct audit_buffer *audit_buf;
4305 
4306 	audit_buf = xfrm_audit_start("SPD-add");
4307 	if (audit_buf == NULL)
4308 		return;
4309 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4310 	audit_log_format(audit_buf, " res=%u", result);
4311 	xfrm_audit_common_policyinfo(xp, audit_buf);
4312 	audit_log_end(audit_buf);
4313 }
4314 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4315 
4316 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4317 			      bool task_valid)
4318 {
4319 	struct audit_buffer *audit_buf;
4320 
4321 	audit_buf = xfrm_audit_start("SPD-delete");
4322 	if (audit_buf == NULL)
4323 		return;
4324 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4325 	audit_log_format(audit_buf, " res=%u", result);
4326 	xfrm_audit_common_policyinfo(xp, audit_buf);
4327 	audit_log_end(audit_buf);
4328 }
4329 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4330 #endif
4331 
4332 #ifdef CONFIG_XFRM_MIGRATE
4333 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4334 					const struct xfrm_selector *sel_tgt)
4335 {
4336 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4337 		if (sel_tgt->family == sel_cmp->family &&
4338 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4339 				    sel_cmp->family) &&
4340 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4341 				    sel_cmp->family) &&
4342 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4343 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4344 			return true;
4345 		}
4346 	} else {
4347 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4348 			return true;
4349 		}
4350 	}
4351 	return false;
4352 }
4353 
4354 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4355 						    u8 dir, u8 type, struct net *net, u32 if_id)
4356 {
4357 	struct xfrm_policy *pol, *ret = NULL;
4358 	struct hlist_head *chain;
4359 	u32 priority = ~0U;
4360 
4361 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4362 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4363 	hlist_for_each_entry(pol, chain, bydst) {
4364 		if ((if_id == 0 || pol->if_id == if_id) &&
4365 		    xfrm_migrate_selector_match(sel, &pol->selector) &&
4366 		    pol->type == type) {
4367 			ret = pol;
4368 			priority = ret->priority;
4369 			break;
4370 		}
4371 	}
4372 	chain = &net->xfrm.policy_inexact[dir];
4373 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4374 		if ((pol->priority >= priority) && ret)
4375 			break;
4376 
4377 		if ((if_id == 0 || pol->if_id == if_id) &&
4378 		    xfrm_migrate_selector_match(sel, &pol->selector) &&
4379 		    pol->type == type) {
4380 			ret = pol;
4381 			break;
4382 		}
4383 	}
4384 
4385 	xfrm_pol_hold(ret);
4386 
4387 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4388 
4389 	return ret;
4390 }
4391 
4392 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4393 {
4394 	int match = 0;
4395 
4396 	if (t->mode == m->mode && t->id.proto == m->proto &&
4397 	    (m->reqid == 0 || t->reqid == m->reqid)) {
4398 		switch (t->mode) {
4399 		case XFRM_MODE_TUNNEL:
4400 		case XFRM_MODE_BEET:
4401 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4402 					    m->old_family) &&
4403 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
4404 					    m->old_family)) {
4405 				match = 1;
4406 			}
4407 			break;
4408 		case XFRM_MODE_TRANSPORT:
4409 			/* in case of transport mode, template does not store
4410 			   any IP addresses, hence we just compare mode and
4411 			   protocol */
4412 			match = 1;
4413 			break;
4414 		default:
4415 			break;
4416 		}
4417 	}
4418 	return match;
4419 }
4420 
4421 /* update endpoint address(es) of template(s) */
4422 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4423 			       struct xfrm_migrate *m, int num_migrate,
4424 			       struct netlink_ext_ack *extack)
4425 {
4426 	struct xfrm_migrate *mp;
4427 	int i, j, n = 0;
4428 
4429 	write_lock_bh(&pol->lock);
4430 	if (unlikely(pol->walk.dead)) {
4431 		/* target policy has been deleted */
4432 		NL_SET_ERR_MSG(extack, "Target policy not found");
4433 		write_unlock_bh(&pol->lock);
4434 		return -ENOENT;
4435 	}
4436 
4437 	for (i = 0; i < pol->xfrm_nr; i++) {
4438 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4439 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4440 				continue;
4441 			n++;
4442 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4443 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4444 				continue;
4445 			/* update endpoints */
4446 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4447 			       sizeof(pol->xfrm_vec[i].id.daddr));
4448 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4449 			       sizeof(pol->xfrm_vec[i].saddr));
4450 			pol->xfrm_vec[i].encap_family = mp->new_family;
4451 			/* flush bundles */
4452 			atomic_inc(&pol->genid);
4453 		}
4454 	}
4455 
4456 	write_unlock_bh(&pol->lock);
4457 
4458 	if (!n)
4459 		return -ENODATA;
4460 
4461 	return 0;
4462 }
4463 
4464 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate,
4465 			      struct netlink_ext_ack *extack)
4466 {
4467 	int i, j;
4468 
4469 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) {
4470 		NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)");
4471 		return -EINVAL;
4472 	}
4473 
4474 	for (i = 0; i < num_migrate; i++) {
4475 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4476 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) {
4477 			NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null");
4478 			return -EINVAL;
4479 		}
4480 
4481 		/* check if there is any duplicated entry */
4482 		for (j = i + 1; j < num_migrate; j++) {
4483 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4484 				    sizeof(m[i].old_daddr)) &&
4485 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4486 				    sizeof(m[i].old_saddr)) &&
4487 			    m[i].proto == m[j].proto &&
4488 			    m[i].mode == m[j].mode &&
4489 			    m[i].reqid == m[j].reqid &&
4490 			    m[i].old_family == m[j].old_family) {
4491 				NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique");
4492 				return -EINVAL;
4493 			}
4494 		}
4495 	}
4496 
4497 	return 0;
4498 }
4499 
4500 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4501 		 struct xfrm_migrate *m, int num_migrate,
4502 		 struct xfrm_kmaddress *k, struct net *net,
4503 		 struct xfrm_encap_tmpl *encap, u32 if_id,
4504 		 struct netlink_ext_ack *extack)
4505 {
4506 	int i, err, nx_cur = 0, nx_new = 0;
4507 	struct xfrm_policy *pol = NULL;
4508 	struct xfrm_state *x, *xc;
4509 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4510 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4511 	struct xfrm_migrate *mp;
4512 
4513 	/* Stage 0 - sanity checks */
4514 	err = xfrm_migrate_check(m, num_migrate, extack);
4515 	if (err < 0)
4516 		goto out;
4517 
4518 	if (dir >= XFRM_POLICY_MAX) {
4519 		NL_SET_ERR_MSG(extack, "Invalid policy direction");
4520 		err = -EINVAL;
4521 		goto out;
4522 	}
4523 
4524 	/* Stage 1 - find policy */
4525 	pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id);
4526 	if (!pol) {
4527 		NL_SET_ERR_MSG(extack, "Target policy not found");
4528 		err = -ENOENT;
4529 		goto out;
4530 	}
4531 
4532 	/* Stage 2 - find and update state(s) */
4533 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4534 		if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4535 			x_cur[nx_cur] = x;
4536 			nx_cur++;
4537 			xc = xfrm_state_migrate(x, mp, encap);
4538 			if (xc) {
4539 				x_new[nx_new] = xc;
4540 				nx_new++;
4541 			} else {
4542 				err = -ENODATA;
4543 				goto restore_state;
4544 			}
4545 		}
4546 	}
4547 
4548 	/* Stage 3 - update policy */
4549 	err = xfrm_policy_migrate(pol, m, num_migrate, extack);
4550 	if (err < 0)
4551 		goto restore_state;
4552 
4553 	/* Stage 4 - delete old state(s) */
4554 	if (nx_cur) {
4555 		xfrm_states_put(x_cur, nx_cur);
4556 		xfrm_states_delete(x_cur, nx_cur);
4557 	}
4558 
4559 	/* Stage 5 - announce */
4560 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
4561 
4562 	xfrm_pol_put(pol);
4563 
4564 	return 0;
4565 out:
4566 	return err;
4567 
4568 restore_state:
4569 	if (pol)
4570 		xfrm_pol_put(pol);
4571 	if (nx_cur)
4572 		xfrm_states_put(x_cur, nx_cur);
4573 	if (nx_new)
4574 		xfrm_states_delete(x_new, nx_new);
4575 
4576 	return err;
4577 }
4578 EXPORT_SYMBOL(xfrm_migrate);
4579 #endif
4580