xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 6db6b729)
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->walk.dead || !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 		if (policy->walk.dead)
1260 			continue;
1261 
1262 		dir = xfrm_policy_id2dir(policy->index);
1263 		if (dir >= XFRM_POLICY_MAX)
1264 			continue;
1265 
1266 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1267 			if (policy->family == AF_INET) {
1268 				dbits = rbits4;
1269 				sbits = lbits4;
1270 			} else {
1271 				dbits = rbits6;
1272 				sbits = lbits6;
1273 			}
1274 		} else {
1275 			if (policy->family == AF_INET) {
1276 				dbits = lbits4;
1277 				sbits = rbits4;
1278 			} else {
1279 				dbits = lbits6;
1280 				sbits = rbits6;
1281 			}
1282 		}
1283 
1284 		if (policy->selector.prefixlen_d < dbits ||
1285 		    policy->selector.prefixlen_s < sbits)
1286 			continue;
1287 
1288 		bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1289 		if (!bin)
1290 			goto out_unlock;
1291 
1292 		if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1293 			goto out_unlock;
1294 	}
1295 
1296 	/* reset the bydst and inexact table in all directions */
1297 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1298 		struct hlist_node *n;
1299 
1300 		hlist_for_each_entry_safe(policy, n,
1301 					  &net->xfrm.policy_inexact[dir],
1302 					  bydst_inexact_list) {
1303 			hlist_del_rcu(&policy->bydst);
1304 			hlist_del_init(&policy->bydst_inexact_list);
1305 		}
1306 
1307 		hmask = net->xfrm.policy_bydst[dir].hmask;
1308 		odst = net->xfrm.policy_bydst[dir].table;
1309 		for (i = hmask; i >= 0; i--) {
1310 			hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1311 				hlist_del_rcu(&policy->bydst);
1312 		}
1313 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1314 			/* dir out => dst = remote, src = local */
1315 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1316 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1317 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1318 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1319 		} else {
1320 			/* dir in/fwd => dst = local, src = remote */
1321 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1322 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1323 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1324 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1325 		}
1326 	}
1327 
1328 	/* re-insert all policies by order of creation */
1329 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1330 		if (policy->walk.dead)
1331 			continue;
1332 		dir = xfrm_policy_id2dir(policy->index);
1333 		if (dir >= XFRM_POLICY_MAX) {
1334 			/* skip socket policies */
1335 			continue;
1336 		}
1337 		newpos = NULL;
1338 		chain = policy_hash_bysel(net, &policy->selector,
1339 					  policy->family, dir);
1340 
1341 		if (!chain) {
1342 			void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1343 
1344 			WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1345 			continue;
1346 		}
1347 
1348 		hlist_for_each_entry(pol, chain, bydst) {
1349 			if (policy->priority >= pol->priority)
1350 				newpos = &pol->bydst;
1351 			else
1352 				break;
1353 		}
1354 		if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1355 			hlist_add_behind_rcu(&policy->bydst, newpos);
1356 		else
1357 			hlist_add_head_rcu(&policy->bydst, chain);
1358 	}
1359 
1360 out_unlock:
1361 	__xfrm_policy_inexact_flush(net);
1362 	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1363 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1364 
1365 	mutex_unlock(&hash_resize_mutex);
1366 }
1367 
1368 void xfrm_policy_hash_rebuild(struct net *net)
1369 {
1370 	schedule_work(&net->xfrm.policy_hthresh.work);
1371 }
1372 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1373 
1374 /* Generate new index... KAME seems to generate them ordered by cost
1375  * of an absolute inpredictability of ordering of rules. This will not pass. */
1376 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1377 {
1378 	for (;;) {
1379 		struct hlist_head *list;
1380 		struct xfrm_policy *p;
1381 		u32 idx;
1382 		int found;
1383 
1384 		if (!index) {
1385 			idx = (net->xfrm.idx_generator | dir);
1386 			net->xfrm.idx_generator += 8;
1387 		} else {
1388 			idx = index;
1389 			index = 0;
1390 		}
1391 
1392 		if (idx == 0)
1393 			idx = 8;
1394 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
1395 		found = 0;
1396 		hlist_for_each_entry(p, list, byidx) {
1397 			if (p->index == idx) {
1398 				found = 1;
1399 				break;
1400 			}
1401 		}
1402 		if (!found)
1403 			return idx;
1404 	}
1405 }
1406 
1407 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1408 {
1409 	u32 *p1 = (u32 *) s1;
1410 	u32 *p2 = (u32 *) s2;
1411 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
1412 	int i;
1413 
1414 	for (i = 0; i < len; i++) {
1415 		if (p1[i] != p2[i])
1416 			return 1;
1417 	}
1418 
1419 	return 0;
1420 }
1421 
1422 static void xfrm_policy_requeue(struct xfrm_policy *old,
1423 				struct xfrm_policy *new)
1424 {
1425 	struct xfrm_policy_queue *pq = &old->polq;
1426 	struct sk_buff_head list;
1427 
1428 	if (skb_queue_empty(&pq->hold_queue))
1429 		return;
1430 
1431 	__skb_queue_head_init(&list);
1432 
1433 	spin_lock_bh(&pq->hold_queue.lock);
1434 	skb_queue_splice_init(&pq->hold_queue, &list);
1435 	if (del_timer(&pq->hold_timer))
1436 		xfrm_pol_put(old);
1437 	spin_unlock_bh(&pq->hold_queue.lock);
1438 
1439 	pq = &new->polq;
1440 
1441 	spin_lock_bh(&pq->hold_queue.lock);
1442 	skb_queue_splice(&list, &pq->hold_queue);
1443 	pq->timeout = XFRM_QUEUE_TMO_MIN;
1444 	if (!mod_timer(&pq->hold_timer, jiffies))
1445 		xfrm_pol_hold(new);
1446 	spin_unlock_bh(&pq->hold_queue.lock);
1447 }
1448 
1449 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1450 					  struct xfrm_policy *pol)
1451 {
1452 	return mark->v == pol->mark.v && mark->m == pol->mark.m;
1453 }
1454 
1455 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1456 {
1457 	const struct xfrm_pol_inexact_key *k = data;
1458 	u32 a = k->type << 24 | k->dir << 16 | k->family;
1459 
1460 	return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1461 			    seed);
1462 }
1463 
1464 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1465 {
1466 	const struct xfrm_pol_inexact_bin *b = data;
1467 
1468 	return xfrm_pol_bin_key(&b->k, 0, seed);
1469 }
1470 
1471 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1472 			    const void *ptr)
1473 {
1474 	const struct xfrm_pol_inexact_key *key = arg->key;
1475 	const struct xfrm_pol_inexact_bin *b = ptr;
1476 	int ret;
1477 
1478 	if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1479 		return -1;
1480 
1481 	ret = b->k.dir ^ key->dir;
1482 	if (ret)
1483 		return ret;
1484 
1485 	ret = b->k.type ^ key->type;
1486 	if (ret)
1487 		return ret;
1488 
1489 	ret = b->k.family ^ key->family;
1490 	if (ret)
1491 		return ret;
1492 
1493 	return b->k.if_id ^ key->if_id;
1494 }
1495 
1496 static const struct rhashtable_params xfrm_pol_inexact_params = {
1497 	.head_offset		= offsetof(struct xfrm_pol_inexact_bin, head),
1498 	.hashfn			= xfrm_pol_bin_key,
1499 	.obj_hashfn		= xfrm_pol_bin_obj,
1500 	.obj_cmpfn		= xfrm_pol_bin_cmp,
1501 	.automatic_shrinking	= true,
1502 };
1503 
1504 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1505 					    struct xfrm_policy *policy)
1506 {
1507 	struct xfrm_policy *pol, *delpol = NULL;
1508 	struct hlist_node *newpos = NULL;
1509 	int i = 0;
1510 
1511 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1512 		if (pol->type == policy->type &&
1513 		    pol->if_id == policy->if_id &&
1514 		    !selector_cmp(&pol->selector, &policy->selector) &&
1515 		    xfrm_policy_mark_match(&policy->mark, pol) &&
1516 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1517 		    !WARN_ON(delpol)) {
1518 			delpol = pol;
1519 			if (policy->priority > pol->priority)
1520 				continue;
1521 		} else if (policy->priority >= pol->priority) {
1522 			newpos = &pol->bydst_inexact_list;
1523 			continue;
1524 		}
1525 		if (delpol)
1526 			break;
1527 	}
1528 
1529 	if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1530 		hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1531 	else
1532 		hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1533 
1534 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1535 		pol->pos = i;
1536 		i++;
1537 	}
1538 }
1539 
1540 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1541 						   struct xfrm_policy *policy,
1542 						   bool excl)
1543 {
1544 	struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1545 
1546 	hlist_for_each_entry(pol, chain, bydst) {
1547 		if (pol->type == policy->type &&
1548 		    pol->if_id == policy->if_id &&
1549 		    !selector_cmp(&pol->selector, &policy->selector) &&
1550 		    xfrm_policy_mark_match(&policy->mark, pol) &&
1551 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1552 		    !WARN_ON(delpol)) {
1553 			if (excl)
1554 				return ERR_PTR(-EEXIST);
1555 			delpol = pol;
1556 			if (policy->priority > pol->priority)
1557 				continue;
1558 		} else if (policy->priority >= pol->priority) {
1559 			newpos = pol;
1560 			continue;
1561 		}
1562 		if (delpol)
1563 			break;
1564 	}
1565 
1566 	if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1567 		hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1568 	else
1569 		/* Packet offload policies enter to the head
1570 		 * to speed-up lookups.
1571 		 */
1572 		hlist_add_head_rcu(&policy->bydst, chain);
1573 
1574 	return delpol;
1575 }
1576 
1577 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1578 {
1579 	struct net *net = xp_net(policy);
1580 	struct xfrm_policy *delpol;
1581 	struct hlist_head *chain;
1582 
1583 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1584 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1585 	if (chain)
1586 		delpol = xfrm_policy_insert_list(chain, policy, excl);
1587 	else
1588 		delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1589 
1590 	if (IS_ERR(delpol)) {
1591 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1592 		return PTR_ERR(delpol);
1593 	}
1594 
1595 	__xfrm_policy_link(policy, dir);
1596 
1597 	/* After previous checking, family can either be AF_INET or AF_INET6 */
1598 	if (policy->family == AF_INET)
1599 		rt_genid_bump_ipv4(net);
1600 	else
1601 		rt_genid_bump_ipv6(net);
1602 
1603 	if (delpol) {
1604 		xfrm_policy_requeue(delpol, policy);
1605 		__xfrm_policy_unlink(delpol, dir);
1606 	}
1607 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1608 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1609 	policy->curlft.add_time = ktime_get_real_seconds();
1610 	policy->curlft.use_time = 0;
1611 	if (!mod_timer(&policy->timer, jiffies + HZ))
1612 		xfrm_pol_hold(policy);
1613 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1614 
1615 	if (delpol)
1616 		xfrm_policy_kill(delpol);
1617 	else if (xfrm_bydst_should_resize(net, dir, NULL))
1618 		schedule_work(&net->xfrm.policy_hash_work);
1619 
1620 	return 0;
1621 }
1622 EXPORT_SYMBOL(xfrm_policy_insert);
1623 
1624 static struct xfrm_policy *
1625 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1626 			u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1627 			struct xfrm_sec_ctx *ctx)
1628 {
1629 	struct xfrm_policy *pol;
1630 
1631 	if (!chain)
1632 		return NULL;
1633 
1634 	hlist_for_each_entry(pol, chain, bydst) {
1635 		if (pol->type == type &&
1636 		    pol->if_id == if_id &&
1637 		    xfrm_policy_mark_match(mark, pol) &&
1638 		    !selector_cmp(sel, &pol->selector) &&
1639 		    xfrm_sec_ctx_match(ctx, pol->security))
1640 			return pol;
1641 	}
1642 
1643 	return NULL;
1644 }
1645 
1646 struct xfrm_policy *
1647 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1648 		      u8 type, int dir, struct xfrm_selector *sel,
1649 		      struct xfrm_sec_ctx *ctx, int delete, int *err)
1650 {
1651 	struct xfrm_pol_inexact_bin *bin = NULL;
1652 	struct xfrm_policy *pol, *ret = NULL;
1653 	struct hlist_head *chain;
1654 
1655 	*err = 0;
1656 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1657 	chain = policy_hash_bysel(net, sel, sel->family, dir);
1658 	if (!chain) {
1659 		struct xfrm_pol_inexact_candidates cand;
1660 		int i;
1661 
1662 		bin = xfrm_policy_inexact_lookup(net, type,
1663 						 sel->family, dir, if_id);
1664 		if (!bin) {
1665 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1666 			return NULL;
1667 		}
1668 
1669 		if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1670 							 &sel->saddr,
1671 							 &sel->daddr)) {
1672 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1673 			return NULL;
1674 		}
1675 
1676 		pol = NULL;
1677 		for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1678 			struct xfrm_policy *tmp;
1679 
1680 			tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1681 						      if_id, type, dir,
1682 						      sel, ctx);
1683 			if (!tmp)
1684 				continue;
1685 
1686 			if (!pol || tmp->pos < pol->pos)
1687 				pol = tmp;
1688 		}
1689 	} else {
1690 		pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1691 					      sel, ctx);
1692 	}
1693 
1694 	if (pol) {
1695 		xfrm_pol_hold(pol);
1696 		if (delete) {
1697 			*err = security_xfrm_policy_delete(pol->security);
1698 			if (*err) {
1699 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1700 				return pol;
1701 			}
1702 			__xfrm_policy_unlink(pol, dir);
1703 		}
1704 		ret = pol;
1705 	}
1706 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1707 
1708 	if (ret && delete)
1709 		xfrm_policy_kill(ret);
1710 	if (bin && delete)
1711 		xfrm_policy_inexact_prune_bin(bin);
1712 	return ret;
1713 }
1714 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1715 
1716 struct xfrm_policy *
1717 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1718 		 u8 type, int dir, u32 id, int delete, int *err)
1719 {
1720 	struct xfrm_policy *pol, *ret;
1721 	struct hlist_head *chain;
1722 
1723 	*err = -ENOENT;
1724 	if (xfrm_policy_id2dir(id) != dir)
1725 		return NULL;
1726 
1727 	*err = 0;
1728 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1729 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
1730 	ret = NULL;
1731 	hlist_for_each_entry(pol, chain, byidx) {
1732 		if (pol->type == type && pol->index == id &&
1733 		    pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1734 			xfrm_pol_hold(pol);
1735 			if (delete) {
1736 				*err = security_xfrm_policy_delete(
1737 								pol->security);
1738 				if (*err) {
1739 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1740 					return pol;
1741 				}
1742 				__xfrm_policy_unlink(pol, dir);
1743 			}
1744 			ret = pol;
1745 			break;
1746 		}
1747 	}
1748 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1749 
1750 	if (ret && delete)
1751 		xfrm_policy_kill(ret);
1752 	return ret;
1753 }
1754 EXPORT_SYMBOL(xfrm_policy_byid);
1755 
1756 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1757 static inline int
1758 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1759 {
1760 	struct xfrm_policy *pol;
1761 	int err = 0;
1762 
1763 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1764 		if (pol->walk.dead ||
1765 		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1766 		    pol->type != type)
1767 			continue;
1768 
1769 		err = security_xfrm_policy_delete(pol->security);
1770 		if (err) {
1771 			xfrm_audit_policy_delete(pol, 0, task_valid);
1772 			return err;
1773 		}
1774 	}
1775 	return err;
1776 }
1777 
1778 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1779 						     struct net_device *dev,
1780 						     bool task_valid)
1781 {
1782 	struct xfrm_policy *pol;
1783 	int err = 0;
1784 
1785 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1786 		if (pol->walk.dead ||
1787 		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1788 		    pol->xdo.dev != dev)
1789 			continue;
1790 
1791 		err = security_xfrm_policy_delete(pol->security);
1792 		if (err) {
1793 			xfrm_audit_policy_delete(pol, 0, task_valid);
1794 			return err;
1795 		}
1796 	}
1797 	return err;
1798 }
1799 #else
1800 static inline int
1801 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1802 {
1803 	return 0;
1804 }
1805 
1806 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1807 						     struct net_device *dev,
1808 						     bool task_valid)
1809 {
1810 	return 0;
1811 }
1812 #endif
1813 
1814 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1815 {
1816 	int dir, err = 0, cnt = 0;
1817 	struct xfrm_policy *pol;
1818 
1819 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1820 
1821 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1822 	if (err)
1823 		goto out;
1824 
1825 again:
1826 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1827 		if (pol->walk.dead)
1828 			continue;
1829 
1830 		dir = xfrm_policy_id2dir(pol->index);
1831 		if (dir >= XFRM_POLICY_MAX ||
1832 		    pol->type != type)
1833 			continue;
1834 
1835 		__xfrm_policy_unlink(pol, dir);
1836 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1837 		xfrm_dev_policy_delete(pol);
1838 		cnt++;
1839 		xfrm_audit_policy_delete(pol, 1, task_valid);
1840 		xfrm_policy_kill(pol);
1841 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1842 		goto again;
1843 	}
1844 	if (cnt)
1845 		__xfrm_policy_inexact_flush(net);
1846 	else
1847 		err = -ESRCH;
1848 out:
1849 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1850 	return err;
1851 }
1852 EXPORT_SYMBOL(xfrm_policy_flush);
1853 
1854 int xfrm_dev_policy_flush(struct net *net, struct net_device *dev,
1855 			  bool task_valid)
1856 {
1857 	int dir, err = 0, cnt = 0;
1858 	struct xfrm_policy *pol;
1859 
1860 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1861 
1862 	err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid);
1863 	if (err)
1864 		goto out;
1865 
1866 again:
1867 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1868 		if (pol->walk.dead)
1869 			continue;
1870 
1871 		dir = xfrm_policy_id2dir(pol->index);
1872 		if (dir >= XFRM_POLICY_MAX ||
1873 		    pol->xdo.dev != dev)
1874 			continue;
1875 
1876 		__xfrm_policy_unlink(pol, dir);
1877 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1878 		xfrm_dev_policy_delete(pol);
1879 		cnt++;
1880 		xfrm_audit_policy_delete(pol, 1, task_valid);
1881 		xfrm_policy_kill(pol);
1882 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1883 		goto again;
1884 	}
1885 	if (cnt)
1886 		__xfrm_policy_inexact_flush(net);
1887 	else
1888 		err = -ESRCH;
1889 out:
1890 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1891 	return err;
1892 }
1893 EXPORT_SYMBOL(xfrm_dev_policy_flush);
1894 
1895 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1896 		     int (*func)(struct xfrm_policy *, int, int, void*),
1897 		     void *data)
1898 {
1899 	struct xfrm_policy *pol;
1900 	struct xfrm_policy_walk_entry *x;
1901 	int error = 0;
1902 
1903 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1904 	    walk->type != XFRM_POLICY_TYPE_ANY)
1905 		return -EINVAL;
1906 
1907 	if (list_empty(&walk->walk.all) && walk->seq != 0)
1908 		return 0;
1909 
1910 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1911 	if (list_empty(&walk->walk.all))
1912 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1913 	else
1914 		x = list_first_entry(&walk->walk.all,
1915 				     struct xfrm_policy_walk_entry, all);
1916 
1917 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1918 		if (x->dead)
1919 			continue;
1920 		pol = container_of(x, struct xfrm_policy, walk);
1921 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1922 		    walk->type != pol->type)
1923 			continue;
1924 		error = func(pol, xfrm_policy_id2dir(pol->index),
1925 			     walk->seq, data);
1926 		if (error) {
1927 			list_move_tail(&walk->walk.all, &x->all);
1928 			goto out;
1929 		}
1930 		walk->seq++;
1931 	}
1932 	if (walk->seq == 0) {
1933 		error = -ENOENT;
1934 		goto out;
1935 	}
1936 	list_del_init(&walk->walk.all);
1937 out:
1938 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1939 	return error;
1940 }
1941 EXPORT_SYMBOL(xfrm_policy_walk);
1942 
1943 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1944 {
1945 	INIT_LIST_HEAD(&walk->walk.all);
1946 	walk->walk.dead = 1;
1947 	walk->type = type;
1948 	walk->seq = 0;
1949 }
1950 EXPORT_SYMBOL(xfrm_policy_walk_init);
1951 
1952 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1953 {
1954 	if (list_empty(&walk->walk.all))
1955 		return;
1956 
1957 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1958 	list_del(&walk->walk.all);
1959 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1960 }
1961 EXPORT_SYMBOL(xfrm_policy_walk_done);
1962 
1963 /*
1964  * Find policy to apply to this flow.
1965  *
1966  * Returns 0 if policy found, else an -errno.
1967  */
1968 static int xfrm_policy_match(const struct xfrm_policy *pol,
1969 			     const struct flowi *fl,
1970 			     u8 type, u16 family, u32 if_id)
1971 {
1972 	const struct xfrm_selector *sel = &pol->selector;
1973 	int ret = -ESRCH;
1974 	bool match;
1975 
1976 	if (pol->family != family ||
1977 	    pol->if_id != if_id ||
1978 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1979 	    pol->type != type)
1980 		return ret;
1981 
1982 	match = xfrm_selector_match(sel, fl, family);
1983 	if (match)
1984 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1985 	return ret;
1986 }
1987 
1988 static struct xfrm_pol_inexact_node *
1989 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1990 				seqcount_spinlock_t *count,
1991 				const xfrm_address_t *addr, u16 family)
1992 {
1993 	const struct rb_node *parent;
1994 	int seq;
1995 
1996 again:
1997 	seq = read_seqcount_begin(count);
1998 
1999 	parent = rcu_dereference_raw(r->rb_node);
2000 	while (parent) {
2001 		struct xfrm_pol_inexact_node *node;
2002 		int delta;
2003 
2004 		node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
2005 
2006 		delta = xfrm_policy_addr_delta(addr, &node->addr,
2007 					       node->prefixlen, family);
2008 		if (delta < 0) {
2009 			parent = rcu_dereference_raw(parent->rb_left);
2010 			continue;
2011 		} else if (delta > 0) {
2012 			parent = rcu_dereference_raw(parent->rb_right);
2013 			continue;
2014 		}
2015 
2016 		return node;
2017 	}
2018 
2019 	if (read_seqcount_retry(count, seq))
2020 		goto again;
2021 
2022 	return NULL;
2023 }
2024 
2025 static bool
2026 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
2027 				    struct xfrm_pol_inexact_bin *b,
2028 				    const xfrm_address_t *saddr,
2029 				    const xfrm_address_t *daddr)
2030 {
2031 	struct xfrm_pol_inexact_node *n;
2032 	u16 family;
2033 
2034 	if (!b)
2035 		return false;
2036 
2037 	family = b->k.family;
2038 	memset(cand, 0, sizeof(*cand));
2039 	cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
2040 
2041 	n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
2042 					    family);
2043 	if (n) {
2044 		cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
2045 		n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
2046 						    family);
2047 		if (n)
2048 			cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
2049 	}
2050 
2051 	n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
2052 					    family);
2053 	if (n)
2054 		cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
2055 
2056 	return true;
2057 }
2058 
2059 static struct xfrm_pol_inexact_bin *
2060 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
2061 			       u8 dir, u32 if_id)
2062 {
2063 	struct xfrm_pol_inexact_key k = {
2064 		.family = family,
2065 		.type = type,
2066 		.dir = dir,
2067 		.if_id = if_id,
2068 	};
2069 
2070 	write_pnet(&k.net, net);
2071 
2072 	return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
2073 				 xfrm_pol_inexact_params);
2074 }
2075 
2076 static struct xfrm_pol_inexact_bin *
2077 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2078 			   u8 dir, u32 if_id)
2079 {
2080 	struct xfrm_pol_inexact_bin *bin;
2081 
2082 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2083 
2084 	rcu_read_lock();
2085 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2086 	rcu_read_unlock();
2087 
2088 	return bin;
2089 }
2090 
2091 static struct xfrm_policy *
2092 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2093 			      struct xfrm_policy *prefer,
2094 			      const struct flowi *fl,
2095 			      u8 type, u16 family, u32 if_id)
2096 {
2097 	u32 priority = prefer ? prefer->priority : ~0u;
2098 	struct xfrm_policy *pol;
2099 
2100 	if (!chain)
2101 		return NULL;
2102 
2103 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2104 		int err;
2105 
2106 		if (pol->priority > priority)
2107 			break;
2108 
2109 		err = xfrm_policy_match(pol, fl, type, family, if_id);
2110 		if (err) {
2111 			if (err != -ESRCH)
2112 				return ERR_PTR(err);
2113 
2114 			continue;
2115 		}
2116 
2117 		if (prefer) {
2118 			/* matches.  Is it older than *prefer? */
2119 			if (pol->priority == priority &&
2120 			    prefer->pos < pol->pos)
2121 				return prefer;
2122 		}
2123 
2124 		return pol;
2125 	}
2126 
2127 	return NULL;
2128 }
2129 
2130 static struct xfrm_policy *
2131 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2132 			    struct xfrm_policy *prefer,
2133 			    const struct flowi *fl,
2134 			    u8 type, u16 family, u32 if_id)
2135 {
2136 	struct xfrm_policy *tmp;
2137 	int i;
2138 
2139 	for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2140 		tmp = __xfrm_policy_eval_candidates(cand->res[i],
2141 						    prefer,
2142 						    fl, type, family, if_id);
2143 		if (!tmp)
2144 			continue;
2145 
2146 		if (IS_ERR(tmp))
2147 			return tmp;
2148 		prefer = tmp;
2149 	}
2150 
2151 	return prefer;
2152 }
2153 
2154 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2155 						     const struct flowi *fl,
2156 						     u16 family, u8 dir,
2157 						     u32 if_id)
2158 {
2159 	struct xfrm_pol_inexact_candidates cand;
2160 	const xfrm_address_t *daddr, *saddr;
2161 	struct xfrm_pol_inexact_bin *bin;
2162 	struct xfrm_policy *pol, *ret;
2163 	struct hlist_head *chain;
2164 	unsigned int sequence;
2165 	int err;
2166 
2167 	daddr = xfrm_flowi_daddr(fl, family);
2168 	saddr = xfrm_flowi_saddr(fl, family);
2169 	if (unlikely(!daddr || !saddr))
2170 		return NULL;
2171 
2172 	rcu_read_lock();
2173  retry:
2174 	do {
2175 		sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2176 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
2177 	} while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2178 
2179 	ret = NULL;
2180 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2181 		err = xfrm_policy_match(pol, fl, type, family, if_id);
2182 		if (err) {
2183 			if (err == -ESRCH)
2184 				continue;
2185 			else {
2186 				ret = ERR_PTR(err);
2187 				goto fail;
2188 			}
2189 		} else {
2190 			ret = pol;
2191 			break;
2192 		}
2193 	}
2194 	if (ret && ret->xdo.type == XFRM_DEV_OFFLOAD_PACKET)
2195 		goto skip_inexact;
2196 
2197 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2198 	if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2199 							 daddr))
2200 		goto skip_inexact;
2201 
2202 	pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2203 					  family, if_id);
2204 	if (pol) {
2205 		ret = pol;
2206 		if (IS_ERR(pol))
2207 			goto fail;
2208 	}
2209 
2210 skip_inexact:
2211 	if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2212 		goto retry;
2213 
2214 	if (ret && !xfrm_pol_hold_rcu(ret))
2215 		goto retry;
2216 fail:
2217 	rcu_read_unlock();
2218 
2219 	return ret;
2220 }
2221 
2222 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2223 					      const struct flowi *fl,
2224 					      u16 family, u8 dir, u32 if_id)
2225 {
2226 #ifdef CONFIG_XFRM_SUB_POLICY
2227 	struct xfrm_policy *pol;
2228 
2229 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2230 					dir, if_id);
2231 	if (pol != NULL)
2232 		return pol;
2233 #endif
2234 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2235 					 dir, if_id);
2236 }
2237 
2238 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2239 						 const struct flowi *fl,
2240 						 u16 family, u32 if_id)
2241 {
2242 	struct xfrm_policy *pol;
2243 
2244 	rcu_read_lock();
2245  again:
2246 	pol = rcu_dereference(sk->sk_policy[dir]);
2247 	if (pol != NULL) {
2248 		bool match;
2249 		int err = 0;
2250 
2251 		if (pol->family != family) {
2252 			pol = NULL;
2253 			goto out;
2254 		}
2255 
2256 		match = xfrm_selector_match(&pol->selector, fl, family);
2257 		if (match) {
2258 			if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v ||
2259 			    pol->if_id != if_id) {
2260 				pol = NULL;
2261 				goto out;
2262 			}
2263 			err = security_xfrm_policy_lookup(pol->security,
2264 						      fl->flowi_secid);
2265 			if (!err) {
2266 				if (!xfrm_pol_hold_rcu(pol))
2267 					goto again;
2268 			} else if (err == -ESRCH) {
2269 				pol = NULL;
2270 			} else {
2271 				pol = ERR_PTR(err);
2272 			}
2273 		} else
2274 			pol = NULL;
2275 	}
2276 out:
2277 	rcu_read_unlock();
2278 	return pol;
2279 }
2280 
2281 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2282 {
2283 	struct net *net = xp_net(pol);
2284 
2285 	list_add(&pol->walk.all, &net->xfrm.policy_all);
2286 	net->xfrm.policy_count[dir]++;
2287 	xfrm_pol_hold(pol);
2288 }
2289 
2290 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2291 						int dir)
2292 {
2293 	struct net *net = xp_net(pol);
2294 
2295 	if (list_empty(&pol->walk.all))
2296 		return NULL;
2297 
2298 	/* Socket policies are not hashed. */
2299 	if (!hlist_unhashed(&pol->bydst)) {
2300 		hlist_del_rcu(&pol->bydst);
2301 		hlist_del_init(&pol->bydst_inexact_list);
2302 		hlist_del(&pol->byidx);
2303 	}
2304 
2305 	list_del_init(&pol->walk.all);
2306 	net->xfrm.policy_count[dir]--;
2307 
2308 	return pol;
2309 }
2310 
2311 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2312 {
2313 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2314 }
2315 
2316 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2317 {
2318 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2319 }
2320 
2321 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2322 {
2323 	struct net *net = xp_net(pol);
2324 
2325 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2326 	pol = __xfrm_policy_unlink(pol, dir);
2327 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2328 	if (pol) {
2329 		xfrm_dev_policy_delete(pol);
2330 		xfrm_policy_kill(pol);
2331 		return 0;
2332 	}
2333 	return -ENOENT;
2334 }
2335 EXPORT_SYMBOL(xfrm_policy_delete);
2336 
2337 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2338 {
2339 	struct net *net = sock_net(sk);
2340 	struct xfrm_policy *old_pol;
2341 
2342 #ifdef CONFIG_XFRM_SUB_POLICY
2343 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2344 		return -EINVAL;
2345 #endif
2346 
2347 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2348 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2349 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2350 	if (pol) {
2351 		pol->curlft.add_time = ktime_get_real_seconds();
2352 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2353 		xfrm_sk_policy_link(pol, dir);
2354 	}
2355 	rcu_assign_pointer(sk->sk_policy[dir], pol);
2356 	if (old_pol) {
2357 		if (pol)
2358 			xfrm_policy_requeue(old_pol, pol);
2359 
2360 		/* Unlinking succeeds always. This is the only function
2361 		 * allowed to delete or replace socket policy.
2362 		 */
2363 		xfrm_sk_policy_unlink(old_pol, dir);
2364 	}
2365 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2366 
2367 	if (old_pol) {
2368 		xfrm_policy_kill(old_pol);
2369 	}
2370 	return 0;
2371 }
2372 
2373 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2374 {
2375 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2376 	struct net *net = xp_net(old);
2377 
2378 	if (newp) {
2379 		newp->selector = old->selector;
2380 		if (security_xfrm_policy_clone(old->security,
2381 					       &newp->security)) {
2382 			kfree(newp);
2383 			return NULL;  /* ENOMEM */
2384 		}
2385 		newp->lft = old->lft;
2386 		newp->curlft = old->curlft;
2387 		newp->mark = old->mark;
2388 		newp->if_id = old->if_id;
2389 		newp->action = old->action;
2390 		newp->flags = old->flags;
2391 		newp->xfrm_nr = old->xfrm_nr;
2392 		newp->index = old->index;
2393 		newp->type = old->type;
2394 		newp->family = old->family;
2395 		memcpy(newp->xfrm_vec, old->xfrm_vec,
2396 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2397 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2398 		xfrm_sk_policy_link(newp, dir);
2399 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2400 		xfrm_pol_put(newp);
2401 	}
2402 	return newp;
2403 }
2404 
2405 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2406 {
2407 	const struct xfrm_policy *p;
2408 	struct xfrm_policy *np;
2409 	int i, ret = 0;
2410 
2411 	rcu_read_lock();
2412 	for (i = 0; i < 2; i++) {
2413 		p = rcu_dereference(osk->sk_policy[i]);
2414 		if (p) {
2415 			np = clone_policy(p, i);
2416 			if (unlikely(!np)) {
2417 				ret = -ENOMEM;
2418 				break;
2419 			}
2420 			rcu_assign_pointer(sk->sk_policy[i], np);
2421 		}
2422 	}
2423 	rcu_read_unlock();
2424 	return ret;
2425 }
2426 
2427 static int
2428 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2429 	       xfrm_address_t *remote, unsigned short family, u32 mark)
2430 {
2431 	int err;
2432 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2433 
2434 	if (unlikely(afinfo == NULL))
2435 		return -EINVAL;
2436 	err = afinfo->get_saddr(net, oif, local, remote, mark);
2437 	rcu_read_unlock();
2438 	return err;
2439 }
2440 
2441 /* Resolve list of templates for the flow, given policy. */
2442 
2443 static int
2444 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2445 		      struct xfrm_state **xfrm, unsigned short family)
2446 {
2447 	struct net *net = xp_net(policy);
2448 	int nx;
2449 	int i, error;
2450 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2451 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2452 	xfrm_address_t tmp;
2453 
2454 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2455 		struct xfrm_state *x;
2456 		xfrm_address_t *remote = daddr;
2457 		xfrm_address_t *local  = saddr;
2458 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2459 
2460 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
2461 		    tmpl->mode == XFRM_MODE_BEET) {
2462 			remote = &tmpl->id.daddr;
2463 			local = &tmpl->saddr;
2464 			if (xfrm_addr_any(local, tmpl->encap_family)) {
2465 				error = xfrm_get_saddr(net, fl->flowi_oif,
2466 						       &tmp, remote,
2467 						       tmpl->encap_family, 0);
2468 				if (error)
2469 					goto fail;
2470 				local = &tmp;
2471 			}
2472 		}
2473 
2474 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2475 				    family, policy->if_id);
2476 
2477 		if (x && x->km.state == XFRM_STATE_VALID) {
2478 			xfrm[nx++] = x;
2479 			daddr = remote;
2480 			saddr = local;
2481 			continue;
2482 		}
2483 		if (x) {
2484 			error = (x->km.state == XFRM_STATE_ERROR ?
2485 				 -EINVAL : -EAGAIN);
2486 			xfrm_state_put(x);
2487 		} else if (error == -ESRCH) {
2488 			error = -EAGAIN;
2489 		}
2490 
2491 		if (!tmpl->optional)
2492 			goto fail;
2493 	}
2494 	return nx;
2495 
2496 fail:
2497 	for (nx--; nx >= 0; nx--)
2498 		xfrm_state_put(xfrm[nx]);
2499 	return error;
2500 }
2501 
2502 static int
2503 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2504 		  struct xfrm_state **xfrm, unsigned short family)
2505 {
2506 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
2507 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2508 	int cnx = 0;
2509 	int error;
2510 	int ret;
2511 	int i;
2512 
2513 	for (i = 0; i < npols; i++) {
2514 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2515 			error = -ENOBUFS;
2516 			goto fail;
2517 		}
2518 
2519 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2520 		if (ret < 0) {
2521 			error = ret;
2522 			goto fail;
2523 		} else
2524 			cnx += ret;
2525 	}
2526 
2527 	/* found states are sorted for outbound processing */
2528 	if (npols > 1)
2529 		xfrm_state_sort(xfrm, tpp, cnx, family);
2530 
2531 	return cnx;
2532 
2533  fail:
2534 	for (cnx--; cnx >= 0; cnx--)
2535 		xfrm_state_put(tpp[cnx]);
2536 	return error;
2537 
2538 }
2539 
2540 static int xfrm_get_tos(const struct flowi *fl, int family)
2541 {
2542 	if (family == AF_INET)
2543 		return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2544 
2545 	return 0;
2546 }
2547 
2548 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2549 {
2550 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2551 	struct dst_ops *dst_ops;
2552 	struct xfrm_dst *xdst;
2553 
2554 	if (!afinfo)
2555 		return ERR_PTR(-EINVAL);
2556 
2557 	switch (family) {
2558 	case AF_INET:
2559 		dst_ops = &net->xfrm.xfrm4_dst_ops;
2560 		break;
2561 #if IS_ENABLED(CONFIG_IPV6)
2562 	case AF_INET6:
2563 		dst_ops = &net->xfrm.xfrm6_dst_ops;
2564 		break;
2565 #endif
2566 	default:
2567 		BUG();
2568 	}
2569 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2570 
2571 	if (likely(xdst)) {
2572 		memset_after(xdst, 0, u.dst);
2573 	} else
2574 		xdst = ERR_PTR(-ENOBUFS);
2575 
2576 	rcu_read_unlock();
2577 
2578 	return xdst;
2579 }
2580 
2581 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2582 			   int nfheader_len)
2583 {
2584 	if (dst->ops->family == AF_INET6) {
2585 		struct rt6_info *rt = (struct rt6_info *)dst;
2586 		path->path_cookie = rt6_get_cookie(rt);
2587 		path->u.rt6.rt6i_nfheader_len = nfheader_len;
2588 	}
2589 }
2590 
2591 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2592 				const struct flowi *fl)
2593 {
2594 	const struct xfrm_policy_afinfo *afinfo =
2595 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2596 	int err;
2597 
2598 	if (!afinfo)
2599 		return -EINVAL;
2600 
2601 	err = afinfo->fill_dst(xdst, dev, fl);
2602 
2603 	rcu_read_unlock();
2604 
2605 	return err;
2606 }
2607 
2608 
2609 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2610  * all the metrics... Shortly, bundle a bundle.
2611  */
2612 
2613 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2614 					    struct xfrm_state **xfrm,
2615 					    struct xfrm_dst **bundle,
2616 					    int nx,
2617 					    const struct flowi *fl,
2618 					    struct dst_entry *dst)
2619 {
2620 	const struct xfrm_state_afinfo *afinfo;
2621 	const struct xfrm_mode *inner_mode;
2622 	struct net *net = xp_net(policy);
2623 	unsigned long now = jiffies;
2624 	struct net_device *dev;
2625 	struct xfrm_dst *xdst_prev = NULL;
2626 	struct xfrm_dst *xdst0 = NULL;
2627 	int i = 0;
2628 	int err;
2629 	int header_len = 0;
2630 	int nfheader_len = 0;
2631 	int trailer_len = 0;
2632 	int tos;
2633 	int family = policy->selector.family;
2634 	xfrm_address_t saddr, daddr;
2635 
2636 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2637 
2638 	tos = xfrm_get_tos(fl, family);
2639 
2640 	dst_hold(dst);
2641 
2642 	for (; i < nx; i++) {
2643 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2644 		struct dst_entry *dst1 = &xdst->u.dst;
2645 
2646 		err = PTR_ERR(xdst);
2647 		if (IS_ERR(xdst)) {
2648 			dst_release(dst);
2649 			goto put_states;
2650 		}
2651 
2652 		bundle[i] = xdst;
2653 		if (!xdst_prev)
2654 			xdst0 = xdst;
2655 		else
2656 			/* Ref count is taken during xfrm_alloc_dst()
2657 			 * No need to do dst_clone() on dst1
2658 			 */
2659 			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2660 
2661 		if (xfrm[i]->sel.family == AF_UNSPEC) {
2662 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
2663 							xfrm_af2proto(family));
2664 			if (!inner_mode) {
2665 				err = -EAFNOSUPPORT;
2666 				dst_release(dst);
2667 				goto put_states;
2668 			}
2669 		} else
2670 			inner_mode = &xfrm[i]->inner_mode;
2671 
2672 		xdst->route = dst;
2673 		dst_copy_metrics(dst1, dst);
2674 
2675 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2676 			__u32 mark = 0;
2677 			int oif;
2678 
2679 			if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2680 				mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2681 
2682 			family = xfrm[i]->props.family;
2683 			oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2684 			dst = xfrm_dst_lookup(xfrm[i], tos, oif,
2685 					      &saddr, &daddr, family, mark);
2686 			err = PTR_ERR(dst);
2687 			if (IS_ERR(dst))
2688 				goto put_states;
2689 		} else
2690 			dst_hold(dst);
2691 
2692 		dst1->xfrm = xfrm[i];
2693 		xdst->xfrm_genid = xfrm[i]->genid;
2694 
2695 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2696 		dst1->lastuse = now;
2697 
2698 		dst1->input = dst_discard;
2699 
2700 		rcu_read_lock();
2701 		afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2702 		if (likely(afinfo))
2703 			dst1->output = afinfo->output;
2704 		else
2705 			dst1->output = dst_discard_out;
2706 		rcu_read_unlock();
2707 
2708 		xdst_prev = xdst;
2709 
2710 		header_len += xfrm[i]->props.header_len;
2711 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2712 			nfheader_len += xfrm[i]->props.header_len;
2713 		trailer_len += xfrm[i]->props.trailer_len;
2714 	}
2715 
2716 	xfrm_dst_set_child(xdst_prev, dst);
2717 	xdst0->path = dst;
2718 
2719 	err = -ENODEV;
2720 	dev = dst->dev;
2721 	if (!dev)
2722 		goto free_dst;
2723 
2724 	xfrm_init_path(xdst0, dst, nfheader_len);
2725 	xfrm_init_pmtu(bundle, nx);
2726 
2727 	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2728 	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2729 		err = xfrm_fill_dst(xdst_prev, dev, fl);
2730 		if (err)
2731 			goto free_dst;
2732 
2733 		xdst_prev->u.dst.header_len = header_len;
2734 		xdst_prev->u.dst.trailer_len = trailer_len;
2735 		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2736 		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2737 	}
2738 
2739 	return &xdst0->u.dst;
2740 
2741 put_states:
2742 	for (; i < nx; i++)
2743 		xfrm_state_put(xfrm[i]);
2744 free_dst:
2745 	if (xdst0)
2746 		dst_release_immediate(&xdst0->u.dst);
2747 
2748 	return ERR_PTR(err);
2749 }
2750 
2751 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2752 				struct xfrm_policy **pols,
2753 				int *num_pols, int *num_xfrms)
2754 {
2755 	int i;
2756 
2757 	if (*num_pols == 0 || !pols[0]) {
2758 		*num_pols = 0;
2759 		*num_xfrms = 0;
2760 		return 0;
2761 	}
2762 	if (IS_ERR(pols[0])) {
2763 		*num_pols = 0;
2764 		return PTR_ERR(pols[0]);
2765 	}
2766 
2767 	*num_xfrms = pols[0]->xfrm_nr;
2768 
2769 #ifdef CONFIG_XFRM_SUB_POLICY
2770 	if (pols[0]->action == XFRM_POLICY_ALLOW &&
2771 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2772 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2773 						    XFRM_POLICY_TYPE_MAIN,
2774 						    fl, family,
2775 						    XFRM_POLICY_OUT,
2776 						    pols[0]->if_id);
2777 		if (pols[1]) {
2778 			if (IS_ERR(pols[1])) {
2779 				xfrm_pols_put(pols, *num_pols);
2780 				*num_pols = 0;
2781 				return PTR_ERR(pols[1]);
2782 			}
2783 			(*num_pols)++;
2784 			(*num_xfrms) += pols[1]->xfrm_nr;
2785 		}
2786 	}
2787 #endif
2788 	for (i = 0; i < *num_pols; i++) {
2789 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
2790 			*num_xfrms = -1;
2791 			break;
2792 		}
2793 	}
2794 
2795 	return 0;
2796 
2797 }
2798 
2799 static struct xfrm_dst *
2800 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2801 			       const struct flowi *fl, u16 family,
2802 			       struct dst_entry *dst_orig)
2803 {
2804 	struct net *net = xp_net(pols[0]);
2805 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2806 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2807 	struct xfrm_dst *xdst;
2808 	struct dst_entry *dst;
2809 	int err;
2810 
2811 	/* Try to instantiate a bundle */
2812 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2813 	if (err <= 0) {
2814 		if (err == 0)
2815 			return NULL;
2816 
2817 		if (err != -EAGAIN)
2818 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2819 		return ERR_PTR(err);
2820 	}
2821 
2822 	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2823 	if (IS_ERR(dst)) {
2824 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2825 		return ERR_CAST(dst);
2826 	}
2827 
2828 	xdst = (struct xfrm_dst *)dst;
2829 	xdst->num_xfrms = err;
2830 	xdst->num_pols = num_pols;
2831 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2832 	xdst->policy_genid = atomic_read(&pols[0]->genid);
2833 
2834 	return xdst;
2835 }
2836 
2837 static void xfrm_policy_queue_process(struct timer_list *t)
2838 {
2839 	struct sk_buff *skb;
2840 	struct sock *sk;
2841 	struct dst_entry *dst;
2842 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2843 	struct net *net = xp_net(pol);
2844 	struct xfrm_policy_queue *pq = &pol->polq;
2845 	struct flowi fl;
2846 	struct sk_buff_head list;
2847 	__u32 skb_mark;
2848 
2849 	spin_lock(&pq->hold_queue.lock);
2850 	skb = skb_peek(&pq->hold_queue);
2851 	if (!skb) {
2852 		spin_unlock(&pq->hold_queue.lock);
2853 		goto out;
2854 	}
2855 	dst = skb_dst(skb);
2856 	sk = skb->sk;
2857 
2858 	/* Fixup the mark to support VTI. */
2859 	skb_mark = skb->mark;
2860 	skb->mark = pol->mark.v;
2861 	xfrm_decode_session(skb, &fl, dst->ops->family);
2862 	skb->mark = skb_mark;
2863 	spin_unlock(&pq->hold_queue.lock);
2864 
2865 	dst_hold(xfrm_dst_path(dst));
2866 	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2867 	if (IS_ERR(dst))
2868 		goto purge_queue;
2869 
2870 	if (dst->flags & DST_XFRM_QUEUE) {
2871 		dst_release(dst);
2872 
2873 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2874 			goto purge_queue;
2875 
2876 		pq->timeout = pq->timeout << 1;
2877 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2878 			xfrm_pol_hold(pol);
2879 		goto out;
2880 	}
2881 
2882 	dst_release(dst);
2883 
2884 	__skb_queue_head_init(&list);
2885 
2886 	spin_lock(&pq->hold_queue.lock);
2887 	pq->timeout = 0;
2888 	skb_queue_splice_init(&pq->hold_queue, &list);
2889 	spin_unlock(&pq->hold_queue.lock);
2890 
2891 	while (!skb_queue_empty(&list)) {
2892 		skb = __skb_dequeue(&list);
2893 
2894 		/* Fixup the mark to support VTI. */
2895 		skb_mark = skb->mark;
2896 		skb->mark = pol->mark.v;
2897 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2898 		skb->mark = skb_mark;
2899 
2900 		dst_hold(xfrm_dst_path(skb_dst(skb)));
2901 		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2902 		if (IS_ERR(dst)) {
2903 			kfree_skb(skb);
2904 			continue;
2905 		}
2906 
2907 		nf_reset_ct(skb);
2908 		skb_dst_drop(skb);
2909 		skb_dst_set(skb, dst);
2910 
2911 		dst_output(net, skb->sk, skb);
2912 	}
2913 
2914 out:
2915 	xfrm_pol_put(pol);
2916 	return;
2917 
2918 purge_queue:
2919 	pq->timeout = 0;
2920 	skb_queue_purge(&pq->hold_queue);
2921 	xfrm_pol_put(pol);
2922 }
2923 
2924 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2925 {
2926 	unsigned long sched_next;
2927 	struct dst_entry *dst = skb_dst(skb);
2928 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2929 	struct xfrm_policy *pol = xdst->pols[0];
2930 	struct xfrm_policy_queue *pq = &pol->polq;
2931 
2932 	if (unlikely(skb_fclone_busy(sk, skb))) {
2933 		kfree_skb(skb);
2934 		return 0;
2935 	}
2936 
2937 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2938 		kfree_skb(skb);
2939 		return -EAGAIN;
2940 	}
2941 
2942 	skb_dst_force(skb);
2943 
2944 	spin_lock_bh(&pq->hold_queue.lock);
2945 
2946 	if (!pq->timeout)
2947 		pq->timeout = XFRM_QUEUE_TMO_MIN;
2948 
2949 	sched_next = jiffies + pq->timeout;
2950 
2951 	if (del_timer(&pq->hold_timer)) {
2952 		if (time_before(pq->hold_timer.expires, sched_next))
2953 			sched_next = pq->hold_timer.expires;
2954 		xfrm_pol_put(pol);
2955 	}
2956 
2957 	__skb_queue_tail(&pq->hold_queue, skb);
2958 	if (!mod_timer(&pq->hold_timer, sched_next))
2959 		xfrm_pol_hold(pol);
2960 
2961 	spin_unlock_bh(&pq->hold_queue.lock);
2962 
2963 	return 0;
2964 }
2965 
2966 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2967 						 struct xfrm_flo *xflo,
2968 						 const struct flowi *fl,
2969 						 int num_xfrms,
2970 						 u16 family)
2971 {
2972 	int err;
2973 	struct net_device *dev;
2974 	struct dst_entry *dst;
2975 	struct dst_entry *dst1;
2976 	struct xfrm_dst *xdst;
2977 
2978 	xdst = xfrm_alloc_dst(net, family);
2979 	if (IS_ERR(xdst))
2980 		return xdst;
2981 
2982 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2983 	    net->xfrm.sysctl_larval_drop ||
2984 	    num_xfrms <= 0)
2985 		return xdst;
2986 
2987 	dst = xflo->dst_orig;
2988 	dst1 = &xdst->u.dst;
2989 	dst_hold(dst);
2990 	xdst->route = dst;
2991 
2992 	dst_copy_metrics(dst1, dst);
2993 
2994 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2995 	dst1->flags |= DST_XFRM_QUEUE;
2996 	dst1->lastuse = jiffies;
2997 
2998 	dst1->input = dst_discard;
2999 	dst1->output = xdst_queue_output;
3000 
3001 	dst_hold(dst);
3002 	xfrm_dst_set_child(xdst, dst);
3003 	xdst->path = dst;
3004 
3005 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
3006 
3007 	err = -ENODEV;
3008 	dev = dst->dev;
3009 	if (!dev)
3010 		goto free_dst;
3011 
3012 	err = xfrm_fill_dst(xdst, dev, fl);
3013 	if (err)
3014 		goto free_dst;
3015 
3016 out:
3017 	return xdst;
3018 
3019 free_dst:
3020 	dst_release(dst1);
3021 	xdst = ERR_PTR(err);
3022 	goto out;
3023 }
3024 
3025 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
3026 					   const struct flowi *fl,
3027 					   u16 family, u8 dir,
3028 					   struct xfrm_flo *xflo, u32 if_id)
3029 {
3030 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3031 	int num_pols = 0, num_xfrms = 0, err;
3032 	struct xfrm_dst *xdst;
3033 
3034 	/* Resolve policies to use if we couldn't get them from
3035 	 * previous cache entry */
3036 	num_pols = 1;
3037 	pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
3038 	err = xfrm_expand_policies(fl, family, pols,
3039 					   &num_pols, &num_xfrms);
3040 	if (err < 0)
3041 		goto inc_error;
3042 	if (num_pols == 0)
3043 		return NULL;
3044 	if (num_xfrms <= 0)
3045 		goto make_dummy_bundle;
3046 
3047 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
3048 					      xflo->dst_orig);
3049 	if (IS_ERR(xdst)) {
3050 		err = PTR_ERR(xdst);
3051 		if (err == -EREMOTE) {
3052 			xfrm_pols_put(pols, num_pols);
3053 			return NULL;
3054 		}
3055 
3056 		if (err != -EAGAIN)
3057 			goto error;
3058 		goto make_dummy_bundle;
3059 	} else if (xdst == NULL) {
3060 		num_xfrms = 0;
3061 		goto make_dummy_bundle;
3062 	}
3063 
3064 	return xdst;
3065 
3066 make_dummy_bundle:
3067 	/* We found policies, but there's no bundles to instantiate:
3068 	 * either because the policy blocks, has no transformations or
3069 	 * we could not build template (no xfrm_states).*/
3070 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
3071 	if (IS_ERR(xdst)) {
3072 		xfrm_pols_put(pols, num_pols);
3073 		return ERR_CAST(xdst);
3074 	}
3075 	xdst->num_pols = num_pols;
3076 	xdst->num_xfrms = num_xfrms;
3077 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
3078 
3079 	return xdst;
3080 
3081 inc_error:
3082 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3083 error:
3084 	xfrm_pols_put(pols, num_pols);
3085 	return ERR_PTR(err);
3086 }
3087 
3088 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3089 					struct dst_entry *dst_orig)
3090 {
3091 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3092 	struct dst_entry *ret;
3093 
3094 	if (!afinfo) {
3095 		dst_release(dst_orig);
3096 		return ERR_PTR(-EINVAL);
3097 	} else {
3098 		ret = afinfo->blackhole_route(net, dst_orig);
3099 	}
3100 	rcu_read_unlock();
3101 
3102 	return ret;
3103 }
3104 
3105 /* Finds/creates a bundle for given flow and if_id
3106  *
3107  * At the moment we eat a raw IP route. Mostly to speed up lookups
3108  * on interfaces with disabled IPsec.
3109  *
3110  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3111  * compatibility
3112  */
3113 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3114 					struct dst_entry *dst_orig,
3115 					const struct flowi *fl,
3116 					const struct sock *sk,
3117 					int flags, u32 if_id)
3118 {
3119 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3120 	struct xfrm_dst *xdst;
3121 	struct dst_entry *dst, *route;
3122 	u16 family = dst_orig->ops->family;
3123 	u8 dir = XFRM_POLICY_OUT;
3124 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3125 
3126 	dst = NULL;
3127 	xdst = NULL;
3128 	route = NULL;
3129 
3130 	sk = sk_const_to_full_sk(sk);
3131 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3132 		num_pols = 1;
3133 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3134 						if_id);
3135 		err = xfrm_expand_policies(fl, family, pols,
3136 					   &num_pols, &num_xfrms);
3137 		if (err < 0)
3138 			goto dropdst;
3139 
3140 		if (num_pols) {
3141 			if (num_xfrms <= 0) {
3142 				drop_pols = num_pols;
3143 				goto no_transform;
3144 			}
3145 
3146 			xdst = xfrm_resolve_and_create_bundle(
3147 					pols, num_pols, fl,
3148 					family, dst_orig);
3149 
3150 			if (IS_ERR(xdst)) {
3151 				xfrm_pols_put(pols, num_pols);
3152 				err = PTR_ERR(xdst);
3153 				if (err == -EREMOTE)
3154 					goto nopol;
3155 
3156 				goto dropdst;
3157 			} else if (xdst == NULL) {
3158 				num_xfrms = 0;
3159 				drop_pols = num_pols;
3160 				goto no_transform;
3161 			}
3162 
3163 			route = xdst->route;
3164 		}
3165 	}
3166 
3167 	if (xdst == NULL) {
3168 		struct xfrm_flo xflo;
3169 
3170 		xflo.dst_orig = dst_orig;
3171 		xflo.flags = flags;
3172 
3173 		/* To accelerate a bit...  */
3174 		if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3175 			       !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3176 			goto nopol;
3177 
3178 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3179 		if (xdst == NULL)
3180 			goto nopol;
3181 		if (IS_ERR(xdst)) {
3182 			err = PTR_ERR(xdst);
3183 			goto dropdst;
3184 		}
3185 
3186 		num_pols = xdst->num_pols;
3187 		num_xfrms = xdst->num_xfrms;
3188 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3189 		route = xdst->route;
3190 	}
3191 
3192 	dst = &xdst->u.dst;
3193 	if (route == NULL && num_xfrms > 0) {
3194 		/* The only case when xfrm_bundle_lookup() returns a
3195 		 * bundle with null route, is when the template could
3196 		 * not be resolved. It means policies are there, but
3197 		 * bundle could not be created, since we don't yet
3198 		 * have the xfrm_state's. We need to wait for KM to
3199 		 * negotiate new SA's or bail out with error.*/
3200 		if (net->xfrm.sysctl_larval_drop) {
3201 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3202 			err = -EREMOTE;
3203 			goto error;
3204 		}
3205 
3206 		err = -EAGAIN;
3207 
3208 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3209 		goto error;
3210 	}
3211 
3212 no_transform:
3213 	if (num_pols == 0)
3214 		goto nopol;
3215 
3216 	if ((flags & XFRM_LOOKUP_ICMP) &&
3217 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3218 		err = -ENOENT;
3219 		goto error;
3220 	}
3221 
3222 	for (i = 0; i < num_pols; i++)
3223 		WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds());
3224 
3225 	if (num_xfrms < 0) {
3226 		/* Prohibit the flow */
3227 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3228 		err = -EPERM;
3229 		goto error;
3230 	} else if (num_xfrms > 0) {
3231 		/* Flow transformed */
3232 		dst_release(dst_orig);
3233 	} else {
3234 		/* Flow passes untransformed */
3235 		dst_release(dst);
3236 		dst = dst_orig;
3237 	}
3238 ok:
3239 	xfrm_pols_put(pols, drop_pols);
3240 	if (dst && dst->xfrm &&
3241 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3242 		dst->flags |= DST_XFRM_TUNNEL;
3243 	return dst;
3244 
3245 nopol:
3246 	if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3247 	    net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3248 		err = -EPERM;
3249 		goto error;
3250 	}
3251 	if (!(flags & XFRM_LOOKUP_ICMP)) {
3252 		dst = dst_orig;
3253 		goto ok;
3254 	}
3255 	err = -ENOENT;
3256 error:
3257 	dst_release(dst);
3258 dropdst:
3259 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3260 		dst_release(dst_orig);
3261 	xfrm_pols_put(pols, drop_pols);
3262 	return ERR_PTR(err);
3263 }
3264 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3265 
3266 /* Main function: finds/creates a bundle for given flow.
3267  *
3268  * At the moment we eat a raw IP route. Mostly to speed up lookups
3269  * on interfaces with disabled IPsec.
3270  */
3271 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3272 			      const struct flowi *fl, const struct sock *sk,
3273 			      int flags)
3274 {
3275 	return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3276 }
3277 EXPORT_SYMBOL(xfrm_lookup);
3278 
3279 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3280  * Otherwise we may send out blackholed packets.
3281  */
3282 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3283 				    const struct flowi *fl,
3284 				    const struct sock *sk, int flags)
3285 {
3286 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3287 					    flags | XFRM_LOOKUP_QUEUE |
3288 					    XFRM_LOOKUP_KEEP_DST_REF);
3289 
3290 	if (PTR_ERR(dst) == -EREMOTE)
3291 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
3292 
3293 	if (IS_ERR(dst))
3294 		dst_release(dst_orig);
3295 
3296 	return dst;
3297 }
3298 EXPORT_SYMBOL(xfrm_lookup_route);
3299 
3300 static inline int
3301 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3302 {
3303 	struct sec_path *sp = skb_sec_path(skb);
3304 	struct xfrm_state *x;
3305 
3306 	if (!sp || idx < 0 || idx >= sp->len)
3307 		return 0;
3308 	x = sp->xvec[idx];
3309 	if (!x->type->reject)
3310 		return 0;
3311 	return x->type->reject(x, skb, fl);
3312 }
3313 
3314 /* When skb is transformed back to its "native" form, we have to
3315  * check policy restrictions. At the moment we make this in maximally
3316  * stupid way. Shame on me. :-) Of course, connected sockets must
3317  * have policy cached at them.
3318  */
3319 
3320 static inline int
3321 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3322 	      unsigned short family, u32 if_id)
3323 {
3324 	if (xfrm_state_kern(x))
3325 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3326 	return	x->id.proto == tmpl->id.proto &&
3327 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3328 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3329 		x->props.mode == tmpl->mode &&
3330 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3331 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3332 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
3333 		  xfrm_state_addr_cmp(tmpl, x, family)) &&
3334 		(if_id == 0 || if_id == x->if_id);
3335 }
3336 
3337 /*
3338  * 0 or more than 0 is returned when validation is succeeded (either bypass
3339  * because of optional transport mode, or next index of the matched secpath
3340  * state with the template.
3341  * -1 is returned when no matching template is found.
3342  * Otherwise "-2 - errored_index" is returned.
3343  */
3344 static inline int
3345 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3346 	       unsigned short family, u32 if_id)
3347 {
3348 	int idx = start;
3349 
3350 	if (tmpl->optional) {
3351 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
3352 			return start;
3353 	} else
3354 		start = -1;
3355 	for (; idx < sp->len; idx++) {
3356 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3357 			return ++idx;
3358 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3359 			if (idx < sp->verified_cnt) {
3360 				/* Secpath entry previously verified, consider optional and
3361 				 * continue searching
3362 				 */
3363 				continue;
3364 			}
3365 
3366 			if (start == -1)
3367 				start = -2-idx;
3368 			break;
3369 		}
3370 	}
3371 	return start;
3372 }
3373 
3374 static void
3375 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3376 {
3377 	const struct iphdr *iph = ip_hdr(skb);
3378 	int ihl = iph->ihl;
3379 	u8 *xprth = skb_network_header(skb) + ihl * 4;
3380 	struct flowi4 *fl4 = &fl->u.ip4;
3381 	int oif = 0;
3382 
3383 	if (skb_dst(skb) && skb_dst(skb)->dev)
3384 		oif = skb_dst(skb)->dev->ifindex;
3385 
3386 	memset(fl4, 0, sizeof(struct flowi4));
3387 	fl4->flowi4_mark = skb->mark;
3388 	fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3389 
3390 	fl4->flowi4_proto = iph->protocol;
3391 	fl4->daddr = reverse ? iph->saddr : iph->daddr;
3392 	fl4->saddr = reverse ? iph->daddr : iph->saddr;
3393 	fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3394 
3395 	if (!ip_is_fragment(iph)) {
3396 		switch (iph->protocol) {
3397 		case IPPROTO_UDP:
3398 		case IPPROTO_UDPLITE:
3399 		case IPPROTO_TCP:
3400 		case IPPROTO_SCTP:
3401 		case IPPROTO_DCCP:
3402 			if (xprth + 4 < skb->data ||
3403 			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3404 				__be16 *ports;
3405 
3406 				xprth = skb_network_header(skb) + ihl * 4;
3407 				ports = (__be16 *)xprth;
3408 
3409 				fl4->fl4_sport = ports[!!reverse];
3410 				fl4->fl4_dport = ports[!reverse];
3411 			}
3412 			break;
3413 		case IPPROTO_ICMP:
3414 			if (xprth + 2 < skb->data ||
3415 			    pskb_may_pull(skb, xprth + 2 - skb->data)) {
3416 				u8 *icmp;
3417 
3418 				xprth = skb_network_header(skb) + ihl * 4;
3419 				icmp = xprth;
3420 
3421 				fl4->fl4_icmp_type = icmp[0];
3422 				fl4->fl4_icmp_code = icmp[1];
3423 			}
3424 			break;
3425 		case IPPROTO_GRE:
3426 			if (xprth + 12 < skb->data ||
3427 			    pskb_may_pull(skb, xprth + 12 - skb->data)) {
3428 				__be16 *greflags;
3429 				__be32 *gre_hdr;
3430 
3431 				xprth = skb_network_header(skb) + ihl * 4;
3432 				greflags = (__be16 *)xprth;
3433 				gre_hdr = (__be32 *)xprth;
3434 
3435 				if (greflags[0] & GRE_KEY) {
3436 					if (greflags[0] & GRE_CSUM)
3437 						gre_hdr++;
3438 					fl4->fl4_gre_key = gre_hdr[1];
3439 				}
3440 			}
3441 			break;
3442 		default:
3443 			break;
3444 		}
3445 	}
3446 }
3447 
3448 #if IS_ENABLED(CONFIG_IPV6)
3449 static void
3450 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3451 {
3452 	struct flowi6 *fl6 = &fl->u.ip6;
3453 	int onlyproto = 0;
3454 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
3455 	u32 offset = sizeof(*hdr);
3456 	struct ipv6_opt_hdr *exthdr;
3457 	const unsigned char *nh = skb_network_header(skb);
3458 	u16 nhoff = IP6CB(skb)->nhoff;
3459 	int oif = 0;
3460 	u8 nexthdr;
3461 
3462 	if (!nhoff)
3463 		nhoff = offsetof(struct ipv6hdr, nexthdr);
3464 
3465 	nexthdr = nh[nhoff];
3466 
3467 	if (skb_dst(skb) && skb_dst(skb)->dev)
3468 		oif = skb_dst(skb)->dev->ifindex;
3469 
3470 	memset(fl6, 0, sizeof(struct flowi6));
3471 	fl6->flowi6_mark = skb->mark;
3472 	fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3473 
3474 	fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3475 	fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3476 
3477 	while (nh + offset + sizeof(*exthdr) < skb->data ||
3478 	       pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3479 		nh = skb_network_header(skb);
3480 		exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3481 
3482 		switch (nexthdr) {
3483 		case NEXTHDR_FRAGMENT:
3484 			onlyproto = 1;
3485 			fallthrough;
3486 		case NEXTHDR_ROUTING:
3487 		case NEXTHDR_HOP:
3488 		case NEXTHDR_DEST:
3489 			offset += ipv6_optlen(exthdr);
3490 			nexthdr = exthdr->nexthdr;
3491 			break;
3492 		case IPPROTO_UDP:
3493 		case IPPROTO_UDPLITE:
3494 		case IPPROTO_TCP:
3495 		case IPPROTO_SCTP:
3496 		case IPPROTO_DCCP:
3497 			if (!onlyproto && (nh + offset + 4 < skb->data ||
3498 			     pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3499 				__be16 *ports;
3500 
3501 				nh = skb_network_header(skb);
3502 				ports = (__be16 *)(nh + offset);
3503 				fl6->fl6_sport = ports[!!reverse];
3504 				fl6->fl6_dport = ports[!reverse];
3505 			}
3506 			fl6->flowi6_proto = nexthdr;
3507 			return;
3508 		case IPPROTO_ICMPV6:
3509 			if (!onlyproto && (nh + offset + 2 < skb->data ||
3510 			    pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3511 				u8 *icmp;
3512 
3513 				nh = skb_network_header(skb);
3514 				icmp = (u8 *)(nh + offset);
3515 				fl6->fl6_icmp_type = icmp[0];
3516 				fl6->fl6_icmp_code = icmp[1];
3517 			}
3518 			fl6->flowi6_proto = nexthdr;
3519 			return;
3520 		case IPPROTO_GRE:
3521 			if (!onlyproto &&
3522 			    (nh + offset + 12 < skb->data ||
3523 			     pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3524 				struct gre_base_hdr *gre_hdr;
3525 				__be32 *gre_key;
3526 
3527 				nh = skb_network_header(skb);
3528 				gre_hdr = (struct gre_base_hdr *)(nh + offset);
3529 				gre_key = (__be32 *)(gre_hdr + 1);
3530 
3531 				if (gre_hdr->flags & GRE_KEY) {
3532 					if (gre_hdr->flags & GRE_CSUM)
3533 						gre_key++;
3534 					fl6->fl6_gre_key = *gre_key;
3535 				}
3536 			}
3537 			fl6->flowi6_proto = nexthdr;
3538 			return;
3539 
3540 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3541 		case IPPROTO_MH:
3542 			offset += ipv6_optlen(exthdr);
3543 			if (!onlyproto && (nh + offset + 3 < skb->data ||
3544 			    pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3545 				struct ip6_mh *mh;
3546 
3547 				nh = skb_network_header(skb);
3548 				mh = (struct ip6_mh *)(nh + offset);
3549 				fl6->fl6_mh_type = mh->ip6mh_type;
3550 			}
3551 			fl6->flowi6_proto = nexthdr;
3552 			return;
3553 #endif
3554 		default:
3555 			fl6->flowi6_proto = nexthdr;
3556 			return;
3557 		}
3558 	}
3559 }
3560 #endif
3561 
3562 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3563 			  unsigned int family, int reverse)
3564 {
3565 	switch (family) {
3566 	case AF_INET:
3567 		decode_session4(skb, fl, reverse);
3568 		break;
3569 #if IS_ENABLED(CONFIG_IPV6)
3570 	case AF_INET6:
3571 		decode_session6(skb, fl, reverse);
3572 		break;
3573 #endif
3574 	default:
3575 		return -EAFNOSUPPORT;
3576 	}
3577 
3578 	return security_xfrm_decode_session(skb, &fl->flowi_secid);
3579 }
3580 EXPORT_SYMBOL(__xfrm_decode_session);
3581 
3582 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3583 {
3584 	for (; k < sp->len; k++) {
3585 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3586 			*idxp = k;
3587 			return 1;
3588 		}
3589 	}
3590 
3591 	return 0;
3592 }
3593 
3594 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3595 			unsigned short family)
3596 {
3597 	struct net *net = dev_net(skb->dev);
3598 	struct xfrm_policy *pol;
3599 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3600 	int npols = 0;
3601 	int xfrm_nr;
3602 	int pi;
3603 	int reverse;
3604 	struct flowi fl;
3605 	int xerr_idx = -1;
3606 	const struct xfrm_if_cb *ifcb;
3607 	struct sec_path *sp;
3608 	u32 if_id = 0;
3609 
3610 	rcu_read_lock();
3611 	ifcb = xfrm_if_get_cb();
3612 
3613 	if (ifcb) {
3614 		struct xfrm_if_decode_session_result r;
3615 
3616 		if (ifcb->decode_session(skb, family, &r)) {
3617 			if_id = r.if_id;
3618 			net = r.net;
3619 		}
3620 	}
3621 	rcu_read_unlock();
3622 
3623 	reverse = dir & ~XFRM_POLICY_MASK;
3624 	dir &= XFRM_POLICY_MASK;
3625 
3626 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3627 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3628 		return 0;
3629 	}
3630 
3631 	nf_nat_decode_session(skb, &fl, family);
3632 
3633 	/* First, check used SA against their selectors. */
3634 	sp = skb_sec_path(skb);
3635 	if (sp) {
3636 		int i;
3637 
3638 		for (i = sp->len - 1; i >= 0; i--) {
3639 			struct xfrm_state *x = sp->xvec[i];
3640 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
3641 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3642 				return 0;
3643 			}
3644 		}
3645 	}
3646 
3647 	pol = NULL;
3648 	sk = sk_to_full_sk(sk);
3649 	if (sk && sk->sk_policy[dir]) {
3650 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3651 		if (IS_ERR(pol)) {
3652 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3653 			return 0;
3654 		}
3655 	}
3656 
3657 	if (!pol)
3658 		pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3659 
3660 	if (IS_ERR(pol)) {
3661 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3662 		return 0;
3663 	}
3664 
3665 	if (!pol) {
3666 		if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3667 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3668 			return 0;
3669 		}
3670 
3671 		if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3672 			xfrm_secpath_reject(xerr_idx, skb, &fl);
3673 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3674 			return 0;
3675 		}
3676 		return 1;
3677 	}
3678 
3679 	/* This lockless write can happen from different cpus. */
3680 	WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3681 
3682 	pols[0] = pol;
3683 	npols++;
3684 #ifdef CONFIG_XFRM_SUB_POLICY
3685 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3686 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3687 						    &fl, family,
3688 						    XFRM_POLICY_IN, if_id);
3689 		if (pols[1]) {
3690 			if (IS_ERR(pols[1])) {
3691 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3692 				xfrm_pol_put(pols[0]);
3693 				return 0;
3694 			}
3695 			/* This write can happen from different cpus. */
3696 			WRITE_ONCE(pols[1]->curlft.use_time,
3697 				   ktime_get_real_seconds());
3698 			npols++;
3699 		}
3700 	}
3701 #endif
3702 
3703 	if (pol->action == XFRM_POLICY_ALLOW) {
3704 		static struct sec_path dummy;
3705 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3706 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3707 		struct xfrm_tmpl **tpp = tp;
3708 		int ti = 0;
3709 		int i, k;
3710 
3711 		sp = skb_sec_path(skb);
3712 		if (!sp)
3713 			sp = &dummy;
3714 
3715 		for (pi = 0; pi < npols; pi++) {
3716 			if (pols[pi] != pol &&
3717 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
3718 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3719 				goto reject;
3720 			}
3721 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3722 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3723 				goto reject_error;
3724 			}
3725 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
3726 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
3727 		}
3728 		xfrm_nr = ti;
3729 
3730 		if (npols > 1) {
3731 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3732 			tpp = stp;
3733 		}
3734 
3735 		/* For each tunnel xfrm, find the first matching tmpl.
3736 		 * For each tmpl before that, find corresponding xfrm.
3737 		 * Order is _important_. Later we will implement
3738 		 * some barriers, but at the moment barriers
3739 		 * are implied between each two transformations.
3740 		 * Upon success, marks secpath entries as having been
3741 		 * verified to allow them to be skipped in future policy
3742 		 * checks (e.g. nested tunnels).
3743 		 */
3744 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3745 			k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3746 			if (k < 0) {
3747 				if (k < -1)
3748 					/* "-2 - errored_index" returned */
3749 					xerr_idx = -(2+k);
3750 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3751 				goto reject;
3752 			}
3753 		}
3754 
3755 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3756 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3757 			goto reject;
3758 		}
3759 
3760 		xfrm_pols_put(pols, npols);
3761 		sp->verified_cnt = k;
3762 
3763 		return 1;
3764 	}
3765 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3766 
3767 reject:
3768 	xfrm_secpath_reject(xerr_idx, skb, &fl);
3769 reject_error:
3770 	xfrm_pols_put(pols, npols);
3771 	return 0;
3772 }
3773 EXPORT_SYMBOL(__xfrm_policy_check);
3774 
3775 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3776 {
3777 	struct net *net = dev_net(skb->dev);
3778 	struct flowi fl;
3779 	struct dst_entry *dst;
3780 	int res = 1;
3781 
3782 	if (xfrm_decode_session(skb, &fl, family) < 0) {
3783 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3784 		return 0;
3785 	}
3786 
3787 	skb_dst_force(skb);
3788 	if (!skb_dst(skb)) {
3789 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3790 		return 0;
3791 	}
3792 
3793 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3794 	if (IS_ERR(dst)) {
3795 		res = 0;
3796 		dst = NULL;
3797 	}
3798 	skb_dst_set(skb, dst);
3799 	return res;
3800 }
3801 EXPORT_SYMBOL(__xfrm_route_forward);
3802 
3803 /* Optimize later using cookies and generation ids. */
3804 
3805 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3806 {
3807 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3808 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3809 	 * get validated by dst_ops->check on every use.  We do this
3810 	 * because when a normal route referenced by an XFRM dst is
3811 	 * obsoleted we do not go looking around for all parent
3812 	 * referencing XFRM dsts so that we can invalidate them.  It
3813 	 * is just too much work.  Instead we make the checks here on
3814 	 * every use.  For example:
3815 	 *
3816 	 *	XFRM dst A --> IPv4 dst X
3817 	 *
3818 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
3819 	 * in this example).  If X is marked obsolete, "A" will not
3820 	 * notice.  That's what we are validating here via the
3821 	 * stale_bundle() check.
3822 	 *
3823 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3824 	 * be marked on it.
3825 	 * This will force stale_bundle() to fail on any xdst bundle with
3826 	 * this dst linked in it.
3827 	 */
3828 	if (dst->obsolete < 0 && !stale_bundle(dst))
3829 		return dst;
3830 
3831 	return NULL;
3832 }
3833 
3834 static int stale_bundle(struct dst_entry *dst)
3835 {
3836 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3837 }
3838 
3839 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3840 {
3841 	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3842 		dst->dev = blackhole_netdev;
3843 		dev_hold(dst->dev);
3844 		dev_put(dev);
3845 	}
3846 }
3847 EXPORT_SYMBOL(xfrm_dst_ifdown);
3848 
3849 static void xfrm_link_failure(struct sk_buff *skb)
3850 {
3851 	/* Impossible. Such dst must be popped before reaches point of failure. */
3852 }
3853 
3854 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3855 {
3856 	if (dst) {
3857 		if (dst->obsolete) {
3858 			dst_release(dst);
3859 			dst = NULL;
3860 		}
3861 	}
3862 	return dst;
3863 }
3864 
3865 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3866 {
3867 	while (nr--) {
3868 		struct xfrm_dst *xdst = bundle[nr];
3869 		u32 pmtu, route_mtu_cached;
3870 		struct dst_entry *dst;
3871 
3872 		dst = &xdst->u.dst;
3873 		pmtu = dst_mtu(xfrm_dst_child(dst));
3874 		xdst->child_mtu_cached = pmtu;
3875 
3876 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3877 
3878 		route_mtu_cached = dst_mtu(xdst->route);
3879 		xdst->route_mtu_cached = route_mtu_cached;
3880 
3881 		if (pmtu > route_mtu_cached)
3882 			pmtu = route_mtu_cached;
3883 
3884 		dst_metric_set(dst, RTAX_MTU, pmtu);
3885 	}
3886 }
3887 
3888 /* Check that the bundle accepts the flow and its components are
3889  * still valid.
3890  */
3891 
3892 static int xfrm_bundle_ok(struct xfrm_dst *first)
3893 {
3894 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3895 	struct dst_entry *dst = &first->u.dst;
3896 	struct xfrm_dst *xdst;
3897 	int start_from, nr;
3898 	u32 mtu;
3899 
3900 	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3901 	    (dst->dev && !netif_running(dst->dev)))
3902 		return 0;
3903 
3904 	if (dst->flags & DST_XFRM_QUEUE)
3905 		return 1;
3906 
3907 	start_from = nr = 0;
3908 	do {
3909 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3910 
3911 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
3912 			return 0;
3913 		if (xdst->xfrm_genid != dst->xfrm->genid)
3914 			return 0;
3915 		if (xdst->num_pols > 0 &&
3916 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3917 			return 0;
3918 
3919 		bundle[nr++] = xdst;
3920 
3921 		mtu = dst_mtu(xfrm_dst_child(dst));
3922 		if (xdst->child_mtu_cached != mtu) {
3923 			start_from = nr;
3924 			xdst->child_mtu_cached = mtu;
3925 		}
3926 
3927 		if (!dst_check(xdst->route, xdst->route_cookie))
3928 			return 0;
3929 		mtu = dst_mtu(xdst->route);
3930 		if (xdst->route_mtu_cached != mtu) {
3931 			start_from = nr;
3932 			xdst->route_mtu_cached = mtu;
3933 		}
3934 
3935 		dst = xfrm_dst_child(dst);
3936 	} while (dst->xfrm);
3937 
3938 	if (likely(!start_from))
3939 		return 1;
3940 
3941 	xdst = bundle[start_from - 1];
3942 	mtu = xdst->child_mtu_cached;
3943 	while (start_from--) {
3944 		dst = &xdst->u.dst;
3945 
3946 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
3947 		if (mtu > xdst->route_mtu_cached)
3948 			mtu = xdst->route_mtu_cached;
3949 		dst_metric_set(dst, RTAX_MTU, mtu);
3950 		if (!start_from)
3951 			break;
3952 
3953 		xdst = bundle[start_from - 1];
3954 		xdst->child_mtu_cached = mtu;
3955 	}
3956 
3957 	return 1;
3958 }
3959 
3960 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3961 {
3962 	return dst_metric_advmss(xfrm_dst_path(dst));
3963 }
3964 
3965 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3966 {
3967 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3968 
3969 	return mtu ? : dst_mtu(xfrm_dst_path(dst));
3970 }
3971 
3972 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3973 					const void *daddr)
3974 {
3975 	while (dst->xfrm) {
3976 		const struct xfrm_state *xfrm = dst->xfrm;
3977 
3978 		dst = xfrm_dst_child(dst);
3979 
3980 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3981 			continue;
3982 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3983 			daddr = xfrm->coaddr;
3984 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3985 			daddr = &xfrm->id.daddr;
3986 	}
3987 	return daddr;
3988 }
3989 
3990 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3991 					   struct sk_buff *skb,
3992 					   const void *daddr)
3993 {
3994 	const struct dst_entry *path = xfrm_dst_path(dst);
3995 
3996 	if (!skb)
3997 		daddr = xfrm_get_dst_nexthop(dst, daddr);
3998 	return path->ops->neigh_lookup(path, skb, daddr);
3999 }
4000 
4001 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
4002 {
4003 	const struct dst_entry *path = xfrm_dst_path(dst);
4004 
4005 	daddr = xfrm_get_dst_nexthop(dst, daddr);
4006 	path->ops->confirm_neigh(path, daddr);
4007 }
4008 
4009 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
4010 {
4011 	int err = 0;
4012 
4013 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
4014 		return -EAFNOSUPPORT;
4015 
4016 	spin_lock(&xfrm_policy_afinfo_lock);
4017 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
4018 		err = -EEXIST;
4019 	else {
4020 		struct dst_ops *dst_ops = afinfo->dst_ops;
4021 		if (likely(dst_ops->kmem_cachep == NULL))
4022 			dst_ops->kmem_cachep = xfrm_dst_cache;
4023 		if (likely(dst_ops->check == NULL))
4024 			dst_ops->check = xfrm_dst_check;
4025 		if (likely(dst_ops->default_advmss == NULL))
4026 			dst_ops->default_advmss = xfrm_default_advmss;
4027 		if (likely(dst_ops->mtu == NULL))
4028 			dst_ops->mtu = xfrm_mtu;
4029 		if (likely(dst_ops->negative_advice == NULL))
4030 			dst_ops->negative_advice = xfrm_negative_advice;
4031 		if (likely(dst_ops->link_failure == NULL))
4032 			dst_ops->link_failure = xfrm_link_failure;
4033 		if (likely(dst_ops->neigh_lookup == NULL))
4034 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
4035 		if (likely(!dst_ops->confirm_neigh))
4036 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
4037 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
4038 	}
4039 	spin_unlock(&xfrm_policy_afinfo_lock);
4040 
4041 	return err;
4042 }
4043 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
4044 
4045 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
4046 {
4047 	struct dst_ops *dst_ops = afinfo->dst_ops;
4048 	int i;
4049 
4050 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
4051 		if (xfrm_policy_afinfo[i] != afinfo)
4052 			continue;
4053 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
4054 		break;
4055 	}
4056 
4057 	synchronize_rcu();
4058 
4059 	dst_ops->kmem_cachep = NULL;
4060 	dst_ops->check = NULL;
4061 	dst_ops->negative_advice = NULL;
4062 	dst_ops->link_failure = NULL;
4063 }
4064 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
4065 
4066 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
4067 {
4068 	spin_lock(&xfrm_if_cb_lock);
4069 	rcu_assign_pointer(xfrm_if_cb, ifcb);
4070 	spin_unlock(&xfrm_if_cb_lock);
4071 }
4072 EXPORT_SYMBOL(xfrm_if_register_cb);
4073 
4074 void xfrm_if_unregister_cb(void)
4075 {
4076 	RCU_INIT_POINTER(xfrm_if_cb, NULL);
4077 	synchronize_rcu();
4078 }
4079 EXPORT_SYMBOL(xfrm_if_unregister_cb);
4080 
4081 #ifdef CONFIG_XFRM_STATISTICS
4082 static int __net_init xfrm_statistics_init(struct net *net)
4083 {
4084 	int rv;
4085 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4086 	if (!net->mib.xfrm_statistics)
4087 		return -ENOMEM;
4088 	rv = xfrm_proc_init(net);
4089 	if (rv < 0)
4090 		free_percpu(net->mib.xfrm_statistics);
4091 	return rv;
4092 }
4093 
4094 static void xfrm_statistics_fini(struct net *net)
4095 {
4096 	xfrm_proc_fini(net);
4097 	free_percpu(net->mib.xfrm_statistics);
4098 }
4099 #else
4100 static int __net_init xfrm_statistics_init(struct net *net)
4101 {
4102 	return 0;
4103 }
4104 
4105 static void xfrm_statistics_fini(struct net *net)
4106 {
4107 }
4108 #endif
4109 
4110 static int __net_init xfrm_policy_init(struct net *net)
4111 {
4112 	unsigned int hmask, sz;
4113 	int dir, err;
4114 
4115 	if (net_eq(net, &init_net)) {
4116 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4117 					   sizeof(struct xfrm_dst),
4118 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4119 					   NULL);
4120 		err = rhashtable_init(&xfrm_policy_inexact_table,
4121 				      &xfrm_pol_inexact_params);
4122 		BUG_ON(err);
4123 	}
4124 
4125 	hmask = 8 - 1;
4126 	sz = (hmask+1) * sizeof(struct hlist_head);
4127 
4128 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4129 	if (!net->xfrm.policy_byidx)
4130 		goto out_byidx;
4131 	net->xfrm.policy_idx_hmask = hmask;
4132 
4133 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4134 		struct xfrm_policy_hash *htab;
4135 
4136 		net->xfrm.policy_count[dir] = 0;
4137 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4138 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4139 
4140 		htab = &net->xfrm.policy_bydst[dir];
4141 		htab->table = xfrm_hash_alloc(sz);
4142 		if (!htab->table)
4143 			goto out_bydst;
4144 		htab->hmask = hmask;
4145 		htab->dbits4 = 32;
4146 		htab->sbits4 = 32;
4147 		htab->dbits6 = 128;
4148 		htab->sbits6 = 128;
4149 	}
4150 	net->xfrm.policy_hthresh.lbits4 = 32;
4151 	net->xfrm.policy_hthresh.rbits4 = 32;
4152 	net->xfrm.policy_hthresh.lbits6 = 128;
4153 	net->xfrm.policy_hthresh.rbits6 = 128;
4154 
4155 	seqlock_init(&net->xfrm.policy_hthresh.lock);
4156 
4157 	INIT_LIST_HEAD(&net->xfrm.policy_all);
4158 	INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4159 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4160 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4161 	return 0;
4162 
4163 out_bydst:
4164 	for (dir--; dir >= 0; dir--) {
4165 		struct xfrm_policy_hash *htab;
4166 
4167 		htab = &net->xfrm.policy_bydst[dir];
4168 		xfrm_hash_free(htab->table, sz);
4169 	}
4170 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4171 out_byidx:
4172 	return -ENOMEM;
4173 }
4174 
4175 static void xfrm_policy_fini(struct net *net)
4176 {
4177 	struct xfrm_pol_inexact_bin *b, *t;
4178 	unsigned int sz;
4179 	int dir;
4180 
4181 	flush_work(&net->xfrm.policy_hash_work);
4182 #ifdef CONFIG_XFRM_SUB_POLICY
4183 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4184 #endif
4185 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4186 
4187 	WARN_ON(!list_empty(&net->xfrm.policy_all));
4188 
4189 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4190 		struct xfrm_policy_hash *htab;
4191 
4192 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4193 
4194 		htab = &net->xfrm.policy_bydst[dir];
4195 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4196 		WARN_ON(!hlist_empty(htab->table));
4197 		xfrm_hash_free(htab->table, sz);
4198 	}
4199 
4200 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4201 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4202 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4203 
4204 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4205 	list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4206 		__xfrm_policy_inexact_prune_bin(b, true);
4207 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4208 }
4209 
4210 static int __net_init xfrm_net_init(struct net *net)
4211 {
4212 	int rv;
4213 
4214 	/* Initialize the per-net locks here */
4215 	spin_lock_init(&net->xfrm.xfrm_state_lock);
4216 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
4217 	seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4218 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
4219 	net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4220 	net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4221 	net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4222 
4223 	rv = xfrm_statistics_init(net);
4224 	if (rv < 0)
4225 		goto out_statistics;
4226 	rv = xfrm_state_init(net);
4227 	if (rv < 0)
4228 		goto out_state;
4229 	rv = xfrm_policy_init(net);
4230 	if (rv < 0)
4231 		goto out_policy;
4232 	rv = xfrm_sysctl_init(net);
4233 	if (rv < 0)
4234 		goto out_sysctl;
4235 
4236 	return 0;
4237 
4238 out_sysctl:
4239 	xfrm_policy_fini(net);
4240 out_policy:
4241 	xfrm_state_fini(net);
4242 out_state:
4243 	xfrm_statistics_fini(net);
4244 out_statistics:
4245 	return rv;
4246 }
4247 
4248 static void __net_exit xfrm_net_exit(struct net *net)
4249 {
4250 	xfrm_sysctl_fini(net);
4251 	xfrm_policy_fini(net);
4252 	xfrm_state_fini(net);
4253 	xfrm_statistics_fini(net);
4254 }
4255 
4256 static struct pernet_operations __net_initdata xfrm_net_ops = {
4257 	.init = xfrm_net_init,
4258 	.exit = xfrm_net_exit,
4259 };
4260 
4261 void __init xfrm_init(void)
4262 {
4263 	register_pernet_subsys(&xfrm_net_ops);
4264 	xfrm_dev_init();
4265 	xfrm_input_init();
4266 
4267 #ifdef CONFIG_XFRM_ESPINTCP
4268 	espintcp_init();
4269 #endif
4270 }
4271 
4272 #ifdef CONFIG_AUDITSYSCALL
4273 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4274 					 struct audit_buffer *audit_buf)
4275 {
4276 	struct xfrm_sec_ctx *ctx = xp->security;
4277 	struct xfrm_selector *sel = &xp->selector;
4278 
4279 	if (ctx)
4280 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4281 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4282 
4283 	switch (sel->family) {
4284 	case AF_INET:
4285 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4286 		if (sel->prefixlen_s != 32)
4287 			audit_log_format(audit_buf, " src_prefixlen=%d",
4288 					 sel->prefixlen_s);
4289 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4290 		if (sel->prefixlen_d != 32)
4291 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4292 					 sel->prefixlen_d);
4293 		break;
4294 	case AF_INET6:
4295 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4296 		if (sel->prefixlen_s != 128)
4297 			audit_log_format(audit_buf, " src_prefixlen=%d",
4298 					 sel->prefixlen_s);
4299 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4300 		if (sel->prefixlen_d != 128)
4301 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4302 					 sel->prefixlen_d);
4303 		break;
4304 	}
4305 }
4306 
4307 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4308 {
4309 	struct audit_buffer *audit_buf;
4310 
4311 	audit_buf = xfrm_audit_start("SPD-add");
4312 	if (audit_buf == NULL)
4313 		return;
4314 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4315 	audit_log_format(audit_buf, " res=%u", result);
4316 	xfrm_audit_common_policyinfo(xp, audit_buf);
4317 	audit_log_end(audit_buf);
4318 }
4319 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4320 
4321 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4322 			      bool task_valid)
4323 {
4324 	struct audit_buffer *audit_buf;
4325 
4326 	audit_buf = xfrm_audit_start("SPD-delete");
4327 	if (audit_buf == NULL)
4328 		return;
4329 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4330 	audit_log_format(audit_buf, " res=%u", result);
4331 	xfrm_audit_common_policyinfo(xp, audit_buf);
4332 	audit_log_end(audit_buf);
4333 }
4334 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4335 #endif
4336 
4337 #ifdef CONFIG_XFRM_MIGRATE
4338 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4339 					const struct xfrm_selector *sel_tgt)
4340 {
4341 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4342 		if (sel_tgt->family == sel_cmp->family &&
4343 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4344 				    sel_cmp->family) &&
4345 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4346 				    sel_cmp->family) &&
4347 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4348 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4349 			return true;
4350 		}
4351 	} else {
4352 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4353 			return true;
4354 		}
4355 	}
4356 	return false;
4357 }
4358 
4359 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4360 						    u8 dir, u8 type, struct net *net, u32 if_id)
4361 {
4362 	struct xfrm_policy *pol, *ret = NULL;
4363 	struct hlist_head *chain;
4364 	u32 priority = ~0U;
4365 
4366 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4367 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4368 	hlist_for_each_entry(pol, chain, bydst) {
4369 		if ((if_id == 0 || pol->if_id == if_id) &&
4370 		    xfrm_migrate_selector_match(sel, &pol->selector) &&
4371 		    pol->type == type) {
4372 			ret = pol;
4373 			priority = ret->priority;
4374 			break;
4375 		}
4376 	}
4377 	chain = &net->xfrm.policy_inexact[dir];
4378 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4379 		if ((pol->priority >= priority) && ret)
4380 			break;
4381 
4382 		if ((if_id == 0 || pol->if_id == if_id) &&
4383 		    xfrm_migrate_selector_match(sel, &pol->selector) &&
4384 		    pol->type == type) {
4385 			ret = pol;
4386 			break;
4387 		}
4388 	}
4389 
4390 	xfrm_pol_hold(ret);
4391 
4392 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4393 
4394 	return ret;
4395 }
4396 
4397 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4398 {
4399 	int match = 0;
4400 
4401 	if (t->mode == m->mode && t->id.proto == m->proto &&
4402 	    (m->reqid == 0 || t->reqid == m->reqid)) {
4403 		switch (t->mode) {
4404 		case XFRM_MODE_TUNNEL:
4405 		case XFRM_MODE_BEET:
4406 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4407 					    m->old_family) &&
4408 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
4409 					    m->old_family)) {
4410 				match = 1;
4411 			}
4412 			break;
4413 		case XFRM_MODE_TRANSPORT:
4414 			/* in case of transport mode, template does not store
4415 			   any IP addresses, hence we just compare mode and
4416 			   protocol */
4417 			match = 1;
4418 			break;
4419 		default:
4420 			break;
4421 		}
4422 	}
4423 	return match;
4424 }
4425 
4426 /* update endpoint address(es) of template(s) */
4427 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4428 			       struct xfrm_migrate *m, int num_migrate,
4429 			       struct netlink_ext_ack *extack)
4430 {
4431 	struct xfrm_migrate *mp;
4432 	int i, j, n = 0;
4433 
4434 	write_lock_bh(&pol->lock);
4435 	if (unlikely(pol->walk.dead)) {
4436 		/* target policy has been deleted */
4437 		NL_SET_ERR_MSG(extack, "Target policy not found");
4438 		write_unlock_bh(&pol->lock);
4439 		return -ENOENT;
4440 	}
4441 
4442 	for (i = 0; i < pol->xfrm_nr; i++) {
4443 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4444 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4445 				continue;
4446 			n++;
4447 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4448 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4449 				continue;
4450 			/* update endpoints */
4451 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4452 			       sizeof(pol->xfrm_vec[i].id.daddr));
4453 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4454 			       sizeof(pol->xfrm_vec[i].saddr));
4455 			pol->xfrm_vec[i].encap_family = mp->new_family;
4456 			/* flush bundles */
4457 			atomic_inc(&pol->genid);
4458 		}
4459 	}
4460 
4461 	write_unlock_bh(&pol->lock);
4462 
4463 	if (!n)
4464 		return -ENODATA;
4465 
4466 	return 0;
4467 }
4468 
4469 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate,
4470 			      struct netlink_ext_ack *extack)
4471 {
4472 	int i, j;
4473 
4474 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) {
4475 		NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)");
4476 		return -EINVAL;
4477 	}
4478 
4479 	for (i = 0; i < num_migrate; i++) {
4480 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4481 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) {
4482 			NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null");
4483 			return -EINVAL;
4484 		}
4485 
4486 		/* check if there is any duplicated entry */
4487 		for (j = i + 1; j < num_migrate; j++) {
4488 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4489 				    sizeof(m[i].old_daddr)) &&
4490 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4491 				    sizeof(m[i].old_saddr)) &&
4492 			    m[i].proto == m[j].proto &&
4493 			    m[i].mode == m[j].mode &&
4494 			    m[i].reqid == m[j].reqid &&
4495 			    m[i].old_family == m[j].old_family) {
4496 				NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique");
4497 				return -EINVAL;
4498 			}
4499 		}
4500 	}
4501 
4502 	return 0;
4503 }
4504 
4505 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4506 		 struct xfrm_migrate *m, int num_migrate,
4507 		 struct xfrm_kmaddress *k, struct net *net,
4508 		 struct xfrm_encap_tmpl *encap, u32 if_id,
4509 		 struct netlink_ext_ack *extack)
4510 {
4511 	int i, err, nx_cur = 0, nx_new = 0;
4512 	struct xfrm_policy *pol = NULL;
4513 	struct xfrm_state *x, *xc;
4514 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4515 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4516 	struct xfrm_migrate *mp;
4517 
4518 	/* Stage 0 - sanity checks */
4519 	err = xfrm_migrate_check(m, num_migrate, extack);
4520 	if (err < 0)
4521 		goto out;
4522 
4523 	if (dir >= XFRM_POLICY_MAX) {
4524 		NL_SET_ERR_MSG(extack, "Invalid policy direction");
4525 		err = -EINVAL;
4526 		goto out;
4527 	}
4528 
4529 	/* Stage 1 - find policy */
4530 	pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id);
4531 	if (!pol) {
4532 		NL_SET_ERR_MSG(extack, "Target policy not found");
4533 		err = -ENOENT;
4534 		goto out;
4535 	}
4536 
4537 	/* Stage 2 - find and update state(s) */
4538 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4539 		if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4540 			x_cur[nx_cur] = x;
4541 			nx_cur++;
4542 			xc = xfrm_state_migrate(x, mp, encap);
4543 			if (xc) {
4544 				x_new[nx_new] = xc;
4545 				nx_new++;
4546 			} else {
4547 				err = -ENODATA;
4548 				goto restore_state;
4549 			}
4550 		}
4551 	}
4552 
4553 	/* Stage 3 - update policy */
4554 	err = xfrm_policy_migrate(pol, m, num_migrate, extack);
4555 	if (err < 0)
4556 		goto restore_state;
4557 
4558 	/* Stage 4 - delete old state(s) */
4559 	if (nx_cur) {
4560 		xfrm_states_put(x_cur, nx_cur);
4561 		xfrm_states_delete(x_cur, nx_cur);
4562 	}
4563 
4564 	/* Stage 5 - announce */
4565 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
4566 
4567 	xfrm_pol_put(pol);
4568 
4569 	return 0;
4570 out:
4571 	return err;
4572 
4573 restore_state:
4574 	if (pol)
4575 		xfrm_pol_put(pol);
4576 	if (nx_cur)
4577 		xfrm_states_put(x_cur, nx_cur);
4578 	if (nx_new)
4579 		xfrm_states_delete(x_new, nx_new);
4580 
4581 	return err;
4582 }
4583 EXPORT_SYMBOL(xfrm_migrate);
4584 #endif
4585