xref: /linux/net/xfrm/xfrm_policy.c (revision ff40b5769a50fab654a70575ff0f49853b799b0e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * xfrm_policy.c
4  *
5  * Changes:
6  *	Mitsuru KANDA @USAGI
7  * 	Kazunori MIYAZAWA @USAGI
8  * 	Kunihiro Ishiguro <kunihiro@ipinfusion.com>
9  * 		IPv6 support
10  * 	Kazunori MIYAZAWA @USAGI
11  * 	YOSHIFUJI Hideaki
12  * 		Split up af-specific portion
13  *	Derek Atkins <derek@ihtfp.com>		Add the post_input processor
14  *
15  */
16 
17 #include <linux/err.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/workqueue.h>
23 #include <linux/notifier.h>
24 #include <linux/netdevice.h>
25 #include <linux/netfilter.h>
26 #include <linux/module.h>
27 #include <linux/cache.h>
28 #include <linux/cpu.h>
29 #include <linux/audit.h>
30 #include <linux/rhashtable.h>
31 #include <linux/if_tunnel.h>
32 #include <net/dst.h>
33 #include <net/flow.h>
34 #include <net/inet_ecn.h>
35 #include <net/xfrm.h>
36 #include <net/ip.h>
37 #include <net/gre.h>
38 #if IS_ENABLED(CONFIG_IPV6_MIP6)
39 #include <net/mip6.h>
40 #endif
41 #ifdef CONFIG_XFRM_STATISTICS
42 #include <net/snmp.h>
43 #endif
44 #ifdef CONFIG_XFRM_ESPINTCP
45 #include <net/espintcp.h>
46 #endif
47 
48 #include "xfrm_hash.h"
49 
50 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
51 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
52 #define XFRM_MAX_QUEUE_LEN	100
53 
54 struct xfrm_flo {
55 	struct dst_entry *dst_orig;
56 	u8 flags;
57 };
58 
59 /* prefixes smaller than this are stored in lists, not trees. */
60 #define INEXACT_PREFIXLEN_IPV4	16
61 #define INEXACT_PREFIXLEN_IPV6	48
62 
63 struct xfrm_pol_inexact_node {
64 	struct rb_node node;
65 	union {
66 		xfrm_address_t addr;
67 		struct rcu_head rcu;
68 	};
69 	u8 prefixlen;
70 
71 	struct rb_root root;
72 
73 	/* the policies matching this node, can be empty list */
74 	struct hlist_head hhead;
75 };
76 
77 /* xfrm inexact policy search tree:
78  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
79  *  |
80  * +---- root_d: sorted by daddr:prefix
81  * |                 |
82  * |        xfrm_pol_inexact_node
83  * |                 |
84  * |                 +- root: sorted by saddr/prefix
85  * |                 |              |
86  * |                 |         xfrm_pol_inexact_node
87  * |                 |              |
88  * |                 |              + root: unused
89  * |                 |              |
90  * |                 |              + hhead: saddr:daddr policies
91  * |                 |
92  * |                 +- coarse policies and all any:daddr policies
93  * |
94  * +---- root_s: sorted by saddr:prefix
95  * |                 |
96  * |        xfrm_pol_inexact_node
97  * |                 |
98  * |                 + root: unused
99  * |                 |
100  * |                 + hhead: saddr:any policies
101  * |
102  * +---- coarse policies and all any:any policies
103  *
104  * Lookups return four candidate lists:
105  * 1. any:any list from top-level xfrm_pol_inexact_bin
106  * 2. any:daddr list from daddr tree
107  * 3. saddr:daddr list from 2nd level daddr tree
108  * 4. saddr:any list from saddr tree
109  *
110  * This result set then needs to be searched for the policy with
111  * the lowest priority.  If two results have same prio, youngest one wins.
112  */
113 
114 struct xfrm_pol_inexact_key {
115 	possible_net_t net;
116 	u32 if_id;
117 	u16 family;
118 	u8 dir, type;
119 };
120 
121 struct xfrm_pol_inexact_bin {
122 	struct xfrm_pol_inexact_key k;
123 	struct rhash_head head;
124 	/* list containing '*:*' policies */
125 	struct hlist_head hhead;
126 
127 	seqcount_spinlock_t count;
128 	/* tree sorted by daddr/prefix */
129 	struct rb_root root_d;
130 
131 	/* tree sorted by saddr/prefix */
132 	struct rb_root root_s;
133 
134 	/* slow path below */
135 	struct list_head inexact_bins;
136 	struct rcu_head rcu;
137 };
138 
139 enum xfrm_pol_inexact_candidate_type {
140 	XFRM_POL_CAND_BOTH,
141 	XFRM_POL_CAND_SADDR,
142 	XFRM_POL_CAND_DADDR,
143 	XFRM_POL_CAND_ANY,
144 
145 	XFRM_POL_CAND_MAX,
146 };
147 
148 struct xfrm_pol_inexact_candidates {
149 	struct hlist_head *res[XFRM_POL_CAND_MAX];
150 };
151 
152 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
153 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
154 
155 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
156 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
157 						__read_mostly;
158 
159 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
160 
161 static struct rhashtable xfrm_policy_inexact_table;
162 static const struct rhashtable_params xfrm_pol_inexact_params;
163 
164 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
165 static int stale_bundle(struct dst_entry *dst);
166 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
167 static void xfrm_policy_queue_process(struct timer_list *t);
168 
169 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
170 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
171 						int dir);
172 
173 static struct xfrm_pol_inexact_bin *
174 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
175 			   u32 if_id);
176 
177 static struct xfrm_pol_inexact_bin *
178 xfrm_policy_inexact_lookup_rcu(struct net *net,
179 			       u8 type, u16 family, u8 dir, u32 if_id);
180 static struct xfrm_policy *
181 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
182 			bool excl);
183 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
184 					    struct xfrm_policy *policy);
185 
186 static bool
187 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
188 				    struct xfrm_pol_inexact_bin *b,
189 				    const xfrm_address_t *saddr,
190 				    const xfrm_address_t *daddr);
191 
192 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
193 {
194 	return refcount_inc_not_zero(&policy->refcnt);
195 }
196 
197 static inline bool
198 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
199 {
200 	const struct flowi4 *fl4 = &fl->u.ip4;
201 
202 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
203 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
204 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
205 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
206 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
207 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
208 }
209 
210 static inline bool
211 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
212 {
213 	const struct flowi6 *fl6 = &fl->u.ip6;
214 
215 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
216 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
217 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
218 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
219 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
220 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
221 }
222 
223 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
224 			 unsigned short family)
225 {
226 	switch (family) {
227 	case AF_INET:
228 		return __xfrm4_selector_match(sel, fl);
229 	case AF_INET6:
230 		return __xfrm6_selector_match(sel, fl);
231 	}
232 	return false;
233 }
234 
235 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
236 {
237 	const struct xfrm_policy_afinfo *afinfo;
238 
239 	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
240 		return NULL;
241 	rcu_read_lock();
242 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
243 	if (unlikely(!afinfo))
244 		rcu_read_unlock();
245 	return afinfo;
246 }
247 
248 /* Called with rcu_read_lock(). */
249 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
250 {
251 	return rcu_dereference(xfrm_if_cb);
252 }
253 
254 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
255 				    const xfrm_address_t *saddr,
256 				    const xfrm_address_t *daddr,
257 				    int family, u32 mark)
258 {
259 	const struct xfrm_policy_afinfo *afinfo;
260 	struct dst_entry *dst;
261 
262 	afinfo = xfrm_policy_get_afinfo(family);
263 	if (unlikely(afinfo == NULL))
264 		return ERR_PTR(-EAFNOSUPPORT);
265 
266 	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
267 
268 	rcu_read_unlock();
269 
270 	return dst;
271 }
272 EXPORT_SYMBOL(__xfrm_dst_lookup);
273 
274 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
275 						int tos, int oif,
276 						xfrm_address_t *prev_saddr,
277 						xfrm_address_t *prev_daddr,
278 						int family, u32 mark)
279 {
280 	struct net *net = xs_net(x);
281 	xfrm_address_t *saddr = &x->props.saddr;
282 	xfrm_address_t *daddr = &x->id.daddr;
283 	struct dst_entry *dst;
284 
285 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
286 		saddr = x->coaddr;
287 		daddr = prev_daddr;
288 	}
289 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
290 		saddr = prev_saddr;
291 		daddr = x->coaddr;
292 	}
293 
294 	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
295 
296 	if (!IS_ERR(dst)) {
297 		if (prev_saddr != saddr)
298 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
299 		if (prev_daddr != daddr)
300 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
301 	}
302 
303 	return dst;
304 }
305 
306 static inline unsigned long make_jiffies(long secs)
307 {
308 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
309 		return MAX_SCHEDULE_TIMEOUT-1;
310 	else
311 		return secs*HZ;
312 }
313 
314 static void xfrm_policy_timer(struct timer_list *t)
315 {
316 	struct xfrm_policy *xp = from_timer(xp, t, timer);
317 	time64_t now = ktime_get_real_seconds();
318 	time64_t next = TIME64_MAX;
319 	int warn = 0;
320 	int dir;
321 
322 	read_lock(&xp->lock);
323 
324 	if (unlikely(xp->walk.dead))
325 		goto out;
326 
327 	dir = xfrm_policy_id2dir(xp->index);
328 
329 	if (xp->lft.hard_add_expires_seconds) {
330 		time64_t tmo = xp->lft.hard_add_expires_seconds +
331 			xp->curlft.add_time - now;
332 		if (tmo <= 0)
333 			goto expired;
334 		if (tmo < next)
335 			next = tmo;
336 	}
337 	if (xp->lft.hard_use_expires_seconds) {
338 		time64_t tmo = xp->lft.hard_use_expires_seconds +
339 			(READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
340 		if (tmo <= 0)
341 			goto expired;
342 		if (tmo < next)
343 			next = tmo;
344 	}
345 	if (xp->lft.soft_add_expires_seconds) {
346 		time64_t tmo = xp->lft.soft_add_expires_seconds +
347 			xp->curlft.add_time - now;
348 		if (tmo <= 0) {
349 			warn = 1;
350 			tmo = XFRM_KM_TIMEOUT;
351 		}
352 		if (tmo < next)
353 			next = tmo;
354 	}
355 	if (xp->lft.soft_use_expires_seconds) {
356 		time64_t tmo = xp->lft.soft_use_expires_seconds +
357 			(READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
358 		if (tmo <= 0) {
359 			warn = 1;
360 			tmo = XFRM_KM_TIMEOUT;
361 		}
362 		if (tmo < next)
363 			next = tmo;
364 	}
365 
366 	if (warn)
367 		km_policy_expired(xp, dir, 0, 0);
368 	if (next != TIME64_MAX &&
369 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
370 		xfrm_pol_hold(xp);
371 
372 out:
373 	read_unlock(&xp->lock);
374 	xfrm_pol_put(xp);
375 	return;
376 
377 expired:
378 	read_unlock(&xp->lock);
379 	if (!xfrm_policy_delete(xp, dir))
380 		km_policy_expired(xp, dir, 1, 0);
381 	xfrm_pol_put(xp);
382 }
383 
384 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
385  * SPD calls.
386  */
387 
388 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
389 {
390 	struct xfrm_policy *policy;
391 
392 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
393 
394 	if (policy) {
395 		write_pnet(&policy->xp_net, net);
396 		INIT_LIST_HEAD(&policy->walk.all);
397 		INIT_HLIST_NODE(&policy->bydst_inexact_list);
398 		INIT_HLIST_NODE(&policy->bydst);
399 		INIT_HLIST_NODE(&policy->byidx);
400 		rwlock_init(&policy->lock);
401 		refcount_set(&policy->refcnt, 1);
402 		skb_queue_head_init(&policy->polq.hold_queue);
403 		timer_setup(&policy->timer, xfrm_policy_timer, 0);
404 		timer_setup(&policy->polq.hold_timer,
405 			    xfrm_policy_queue_process, 0);
406 	}
407 	return policy;
408 }
409 EXPORT_SYMBOL(xfrm_policy_alloc);
410 
411 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
412 {
413 	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
414 
415 	security_xfrm_policy_free(policy->security);
416 	kfree(policy);
417 }
418 
419 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
420 
421 void xfrm_policy_destroy(struct xfrm_policy *policy)
422 {
423 	BUG_ON(!policy->walk.dead);
424 
425 	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
426 		BUG();
427 
428 	xfrm_dev_policy_free(policy);
429 	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
430 }
431 EXPORT_SYMBOL(xfrm_policy_destroy);
432 
433 /* Rule must be locked. Release descendant resources, announce
434  * entry dead. The rule must be unlinked from lists to the moment.
435  */
436 
437 static void xfrm_policy_kill(struct xfrm_policy *policy)
438 {
439 	write_lock_bh(&policy->lock);
440 	policy->walk.dead = 1;
441 	write_unlock_bh(&policy->lock);
442 
443 	atomic_inc(&policy->genid);
444 
445 	if (del_timer(&policy->polq.hold_timer))
446 		xfrm_pol_put(policy);
447 	skb_queue_purge(&policy->polq.hold_queue);
448 
449 	if (del_timer(&policy->timer))
450 		xfrm_pol_put(policy);
451 
452 	xfrm_pol_put(policy);
453 }
454 
455 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
456 
457 static inline unsigned int idx_hash(struct net *net, u32 index)
458 {
459 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
460 }
461 
462 /* calculate policy hash thresholds */
463 static void __get_hash_thresh(struct net *net,
464 			      unsigned short family, int dir,
465 			      u8 *dbits, u8 *sbits)
466 {
467 	switch (family) {
468 	case AF_INET:
469 		*dbits = net->xfrm.policy_bydst[dir].dbits4;
470 		*sbits = net->xfrm.policy_bydst[dir].sbits4;
471 		break;
472 
473 	case AF_INET6:
474 		*dbits = net->xfrm.policy_bydst[dir].dbits6;
475 		*sbits = net->xfrm.policy_bydst[dir].sbits6;
476 		break;
477 
478 	default:
479 		*dbits = 0;
480 		*sbits = 0;
481 	}
482 }
483 
484 static struct hlist_head *policy_hash_bysel(struct net *net,
485 					    const struct xfrm_selector *sel,
486 					    unsigned short family, int dir)
487 {
488 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
489 	unsigned int hash;
490 	u8 dbits;
491 	u8 sbits;
492 
493 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
494 	hash = __sel_hash(sel, family, hmask, dbits, sbits);
495 
496 	if (hash == hmask + 1)
497 		return NULL;
498 
499 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
500 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
501 }
502 
503 static struct hlist_head *policy_hash_direct(struct net *net,
504 					     const xfrm_address_t *daddr,
505 					     const xfrm_address_t *saddr,
506 					     unsigned short family, int dir)
507 {
508 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
509 	unsigned int hash;
510 	u8 dbits;
511 	u8 sbits;
512 
513 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
514 	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
515 
516 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
517 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
518 }
519 
520 static void xfrm_dst_hash_transfer(struct net *net,
521 				   struct hlist_head *list,
522 				   struct hlist_head *ndsttable,
523 				   unsigned int nhashmask,
524 				   int dir)
525 {
526 	struct hlist_node *tmp, *entry0 = NULL;
527 	struct xfrm_policy *pol;
528 	unsigned int h0 = 0;
529 	u8 dbits;
530 	u8 sbits;
531 
532 redo:
533 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
534 		unsigned int h;
535 
536 		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
537 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
538 				pol->family, nhashmask, dbits, sbits);
539 		if (!entry0 || pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET) {
540 			hlist_del_rcu(&pol->bydst);
541 			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
542 			h0 = h;
543 		} else {
544 			if (h != h0)
545 				continue;
546 			hlist_del_rcu(&pol->bydst);
547 			hlist_add_behind_rcu(&pol->bydst, entry0);
548 		}
549 		entry0 = &pol->bydst;
550 	}
551 	if (!hlist_empty(list)) {
552 		entry0 = NULL;
553 		goto redo;
554 	}
555 }
556 
557 static void xfrm_idx_hash_transfer(struct hlist_head *list,
558 				   struct hlist_head *nidxtable,
559 				   unsigned int nhashmask)
560 {
561 	struct hlist_node *tmp;
562 	struct xfrm_policy *pol;
563 
564 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
565 		unsigned int h;
566 
567 		h = __idx_hash(pol->index, nhashmask);
568 		hlist_add_head(&pol->byidx, nidxtable+h);
569 	}
570 }
571 
572 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
573 {
574 	return ((old_hmask + 1) << 1) - 1;
575 }
576 
577 static void xfrm_bydst_resize(struct net *net, int dir)
578 {
579 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
580 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
581 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
582 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
583 	struct hlist_head *odst;
584 	int i;
585 
586 	if (!ndst)
587 		return;
588 
589 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
590 	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
591 
592 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
593 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
594 
595 	for (i = hmask; i >= 0; i--)
596 		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
597 
598 	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
599 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
600 
601 	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
602 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
603 
604 	synchronize_rcu();
605 
606 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
607 }
608 
609 static void xfrm_byidx_resize(struct net *net)
610 {
611 	unsigned int hmask = net->xfrm.policy_idx_hmask;
612 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
613 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
614 	struct hlist_head *oidx = net->xfrm.policy_byidx;
615 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
616 	int i;
617 
618 	if (!nidx)
619 		return;
620 
621 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
622 
623 	for (i = hmask; i >= 0; i--)
624 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
625 
626 	net->xfrm.policy_byidx = nidx;
627 	net->xfrm.policy_idx_hmask = nhashmask;
628 
629 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
630 
631 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
632 }
633 
634 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
635 {
636 	unsigned int cnt = net->xfrm.policy_count[dir];
637 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
638 
639 	if (total)
640 		*total += cnt;
641 
642 	if ((hmask + 1) < xfrm_policy_hashmax &&
643 	    cnt > hmask)
644 		return 1;
645 
646 	return 0;
647 }
648 
649 static inline int xfrm_byidx_should_resize(struct net *net, int total)
650 {
651 	unsigned int hmask = net->xfrm.policy_idx_hmask;
652 
653 	if ((hmask + 1) < xfrm_policy_hashmax &&
654 	    total > hmask)
655 		return 1;
656 
657 	return 0;
658 }
659 
660 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
661 {
662 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
663 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
664 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
665 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
666 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
667 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
668 	si->spdhcnt = net->xfrm.policy_idx_hmask;
669 	si->spdhmcnt = xfrm_policy_hashmax;
670 }
671 EXPORT_SYMBOL(xfrm_spd_getinfo);
672 
673 static DEFINE_MUTEX(hash_resize_mutex);
674 static void xfrm_hash_resize(struct work_struct *work)
675 {
676 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
677 	int dir, total;
678 
679 	mutex_lock(&hash_resize_mutex);
680 
681 	total = 0;
682 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
683 		if (xfrm_bydst_should_resize(net, dir, &total))
684 			xfrm_bydst_resize(net, dir);
685 	}
686 	if (xfrm_byidx_should_resize(net, total))
687 		xfrm_byidx_resize(net);
688 
689 	mutex_unlock(&hash_resize_mutex);
690 }
691 
692 /* Make sure *pol can be inserted into fastbin.
693  * Useful to check that later insert requests will be successful
694  * (provided xfrm_policy_lock is held throughout).
695  */
696 static struct xfrm_pol_inexact_bin *
697 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
698 {
699 	struct xfrm_pol_inexact_bin *bin, *prev;
700 	struct xfrm_pol_inexact_key k = {
701 		.family = pol->family,
702 		.type = pol->type,
703 		.dir = dir,
704 		.if_id = pol->if_id,
705 	};
706 	struct net *net = xp_net(pol);
707 
708 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
709 
710 	write_pnet(&k.net, net);
711 	bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
712 				     xfrm_pol_inexact_params);
713 	if (bin)
714 		return bin;
715 
716 	bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
717 	if (!bin)
718 		return NULL;
719 
720 	bin->k = k;
721 	INIT_HLIST_HEAD(&bin->hhead);
722 	bin->root_d = RB_ROOT;
723 	bin->root_s = RB_ROOT;
724 	seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
725 
726 	prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
727 						&bin->k, &bin->head,
728 						xfrm_pol_inexact_params);
729 	if (!prev) {
730 		list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
731 		return bin;
732 	}
733 
734 	kfree(bin);
735 
736 	return IS_ERR(prev) ? NULL : prev;
737 }
738 
739 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
740 					       int family, u8 prefixlen)
741 {
742 	if (xfrm_addr_any(addr, family))
743 		return true;
744 
745 	if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
746 		return true;
747 
748 	if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
749 		return true;
750 
751 	return false;
752 }
753 
754 static bool
755 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
756 {
757 	const xfrm_address_t *addr;
758 	bool saddr_any, daddr_any;
759 	u8 prefixlen;
760 
761 	addr = &policy->selector.saddr;
762 	prefixlen = policy->selector.prefixlen_s;
763 
764 	saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
765 						       policy->family,
766 						       prefixlen);
767 	addr = &policy->selector.daddr;
768 	prefixlen = policy->selector.prefixlen_d;
769 	daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
770 						       policy->family,
771 						       prefixlen);
772 	return saddr_any && daddr_any;
773 }
774 
775 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
776 				       const xfrm_address_t *addr, u8 prefixlen)
777 {
778 	node->addr = *addr;
779 	node->prefixlen = prefixlen;
780 }
781 
782 static struct xfrm_pol_inexact_node *
783 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
784 {
785 	struct xfrm_pol_inexact_node *node;
786 
787 	node = kzalloc(sizeof(*node), GFP_ATOMIC);
788 	if (node)
789 		xfrm_pol_inexact_node_init(node, addr, prefixlen);
790 
791 	return node;
792 }
793 
794 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
795 				  const xfrm_address_t *b,
796 				  u8 prefixlen, u16 family)
797 {
798 	u32 ma, mb, mask;
799 	unsigned int pdw, pbi;
800 	int delta = 0;
801 
802 	switch (family) {
803 	case AF_INET:
804 		if (prefixlen == 0)
805 			return 0;
806 		mask = ~0U << (32 - prefixlen);
807 		ma = ntohl(a->a4) & mask;
808 		mb = ntohl(b->a4) & mask;
809 		if (ma < mb)
810 			delta = -1;
811 		else if (ma > mb)
812 			delta = 1;
813 		break;
814 	case AF_INET6:
815 		pdw = prefixlen >> 5;
816 		pbi = prefixlen & 0x1f;
817 
818 		if (pdw) {
819 			delta = memcmp(a->a6, b->a6, pdw << 2);
820 			if (delta)
821 				return delta;
822 		}
823 		if (pbi) {
824 			mask = ~0U << (32 - pbi);
825 			ma = ntohl(a->a6[pdw]) & mask;
826 			mb = ntohl(b->a6[pdw]) & mask;
827 			if (ma < mb)
828 				delta = -1;
829 			else if (ma > mb)
830 				delta = 1;
831 		}
832 		break;
833 	default:
834 		break;
835 	}
836 
837 	return delta;
838 }
839 
840 static void xfrm_policy_inexact_list_reinsert(struct net *net,
841 					      struct xfrm_pol_inexact_node *n,
842 					      u16 family)
843 {
844 	unsigned int matched_s, matched_d;
845 	struct xfrm_policy *policy, *p;
846 
847 	matched_s = 0;
848 	matched_d = 0;
849 
850 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
851 		struct hlist_node *newpos = NULL;
852 		bool matches_s, matches_d;
853 
854 		if (!policy->bydst_reinsert)
855 			continue;
856 
857 		WARN_ON_ONCE(policy->family != family);
858 
859 		policy->bydst_reinsert = false;
860 		hlist_for_each_entry(p, &n->hhead, bydst) {
861 			if (policy->priority > p->priority)
862 				newpos = &p->bydst;
863 			else if (policy->priority == p->priority &&
864 				 policy->pos > p->pos)
865 				newpos = &p->bydst;
866 			else
867 				break;
868 		}
869 
870 		if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
871 			hlist_add_behind_rcu(&policy->bydst, newpos);
872 		else
873 			hlist_add_head_rcu(&policy->bydst, &n->hhead);
874 
875 		/* paranoia checks follow.
876 		 * Check that the reinserted policy matches at least
877 		 * saddr or daddr for current node prefix.
878 		 *
879 		 * Matching both is fine, matching saddr in one policy
880 		 * (but not daddr) and then matching only daddr in another
881 		 * is a bug.
882 		 */
883 		matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
884 						   &n->addr,
885 						   n->prefixlen,
886 						   family) == 0;
887 		matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
888 						   &n->addr,
889 						   n->prefixlen,
890 						   family) == 0;
891 		if (matches_s && matches_d)
892 			continue;
893 
894 		WARN_ON_ONCE(!matches_s && !matches_d);
895 		if (matches_s)
896 			matched_s++;
897 		if (matches_d)
898 			matched_d++;
899 		WARN_ON_ONCE(matched_s && matched_d);
900 	}
901 }
902 
903 static void xfrm_policy_inexact_node_reinsert(struct net *net,
904 					      struct xfrm_pol_inexact_node *n,
905 					      struct rb_root *new,
906 					      u16 family)
907 {
908 	struct xfrm_pol_inexact_node *node;
909 	struct rb_node **p, *parent;
910 
911 	/* we should not have another subtree here */
912 	WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
913 restart:
914 	parent = NULL;
915 	p = &new->rb_node;
916 	while (*p) {
917 		u8 prefixlen;
918 		int delta;
919 
920 		parent = *p;
921 		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
922 
923 		prefixlen = min(node->prefixlen, n->prefixlen);
924 
925 		delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
926 					       prefixlen, family);
927 		if (delta < 0) {
928 			p = &parent->rb_left;
929 		} else if (delta > 0) {
930 			p = &parent->rb_right;
931 		} else {
932 			bool same_prefixlen = node->prefixlen == n->prefixlen;
933 			struct xfrm_policy *tmp;
934 
935 			hlist_for_each_entry(tmp, &n->hhead, bydst) {
936 				tmp->bydst_reinsert = true;
937 				hlist_del_rcu(&tmp->bydst);
938 			}
939 
940 			node->prefixlen = prefixlen;
941 
942 			xfrm_policy_inexact_list_reinsert(net, node, family);
943 
944 			if (same_prefixlen) {
945 				kfree_rcu(n, rcu);
946 				return;
947 			}
948 
949 			rb_erase(*p, new);
950 			kfree_rcu(n, rcu);
951 			n = node;
952 			goto restart;
953 		}
954 	}
955 
956 	rb_link_node_rcu(&n->node, parent, p);
957 	rb_insert_color(&n->node, new);
958 }
959 
960 /* merge nodes v and n */
961 static void xfrm_policy_inexact_node_merge(struct net *net,
962 					   struct xfrm_pol_inexact_node *v,
963 					   struct xfrm_pol_inexact_node *n,
964 					   u16 family)
965 {
966 	struct xfrm_pol_inexact_node *node;
967 	struct xfrm_policy *tmp;
968 	struct rb_node *rnode;
969 
970 	/* To-be-merged node v has a subtree.
971 	 *
972 	 * Dismantle it and insert its nodes to n->root.
973 	 */
974 	while ((rnode = rb_first(&v->root)) != NULL) {
975 		node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
976 		rb_erase(&node->node, &v->root);
977 		xfrm_policy_inexact_node_reinsert(net, node, &n->root,
978 						  family);
979 	}
980 
981 	hlist_for_each_entry(tmp, &v->hhead, bydst) {
982 		tmp->bydst_reinsert = true;
983 		hlist_del_rcu(&tmp->bydst);
984 	}
985 
986 	xfrm_policy_inexact_list_reinsert(net, n, family);
987 }
988 
989 static struct xfrm_pol_inexact_node *
990 xfrm_policy_inexact_insert_node(struct net *net,
991 				struct rb_root *root,
992 				xfrm_address_t *addr,
993 				u16 family, u8 prefixlen, u8 dir)
994 {
995 	struct xfrm_pol_inexact_node *cached = NULL;
996 	struct rb_node **p, *parent = NULL;
997 	struct xfrm_pol_inexact_node *node;
998 
999 	p = &root->rb_node;
1000 	while (*p) {
1001 		int delta;
1002 
1003 		parent = *p;
1004 		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1005 
1006 		delta = xfrm_policy_addr_delta(addr, &node->addr,
1007 					       node->prefixlen,
1008 					       family);
1009 		if (delta == 0 && prefixlen >= node->prefixlen) {
1010 			WARN_ON_ONCE(cached); /* ipsec policies got lost */
1011 			return node;
1012 		}
1013 
1014 		if (delta < 0)
1015 			p = &parent->rb_left;
1016 		else
1017 			p = &parent->rb_right;
1018 
1019 		if (prefixlen < node->prefixlen) {
1020 			delta = xfrm_policy_addr_delta(addr, &node->addr,
1021 						       prefixlen,
1022 						       family);
1023 			if (delta)
1024 				continue;
1025 
1026 			/* This node is a subnet of the new prefix. It needs
1027 			 * to be removed and re-inserted with the smaller
1028 			 * prefix and all nodes that are now also covered
1029 			 * by the reduced prefixlen.
1030 			 */
1031 			rb_erase(&node->node, root);
1032 
1033 			if (!cached) {
1034 				xfrm_pol_inexact_node_init(node, addr,
1035 							   prefixlen);
1036 				cached = node;
1037 			} else {
1038 				/* This node also falls within the new
1039 				 * prefixlen. Merge the to-be-reinserted
1040 				 * node and this one.
1041 				 */
1042 				xfrm_policy_inexact_node_merge(net, node,
1043 							       cached, family);
1044 				kfree_rcu(node, rcu);
1045 			}
1046 
1047 			/* restart */
1048 			p = &root->rb_node;
1049 			parent = NULL;
1050 		}
1051 	}
1052 
1053 	node = cached;
1054 	if (!node) {
1055 		node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1056 		if (!node)
1057 			return NULL;
1058 	}
1059 
1060 	rb_link_node_rcu(&node->node, parent, p);
1061 	rb_insert_color(&node->node, root);
1062 
1063 	return node;
1064 }
1065 
1066 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1067 {
1068 	struct xfrm_pol_inexact_node *node;
1069 	struct rb_node *rn = rb_first(r);
1070 
1071 	while (rn) {
1072 		node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1073 
1074 		xfrm_policy_inexact_gc_tree(&node->root, rm);
1075 		rn = rb_next(rn);
1076 
1077 		if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1078 			WARN_ON_ONCE(rm);
1079 			continue;
1080 		}
1081 
1082 		rb_erase(&node->node, r);
1083 		kfree_rcu(node, rcu);
1084 	}
1085 }
1086 
1087 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1088 {
1089 	write_seqcount_begin(&b->count);
1090 	xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1091 	xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1092 	write_seqcount_end(&b->count);
1093 
1094 	if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1095 	    !hlist_empty(&b->hhead)) {
1096 		WARN_ON_ONCE(net_exit);
1097 		return;
1098 	}
1099 
1100 	if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1101 				   xfrm_pol_inexact_params) == 0) {
1102 		list_del(&b->inexact_bins);
1103 		kfree_rcu(b, rcu);
1104 	}
1105 }
1106 
1107 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1108 {
1109 	struct net *net = read_pnet(&b->k.net);
1110 
1111 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1112 	__xfrm_policy_inexact_prune_bin(b, false);
1113 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1114 }
1115 
1116 static void __xfrm_policy_inexact_flush(struct net *net)
1117 {
1118 	struct xfrm_pol_inexact_bin *bin, *t;
1119 
1120 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1121 
1122 	list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1123 		__xfrm_policy_inexact_prune_bin(bin, false);
1124 }
1125 
1126 static struct hlist_head *
1127 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1128 				struct xfrm_policy *policy, u8 dir)
1129 {
1130 	struct xfrm_pol_inexact_node *n;
1131 	struct net *net;
1132 
1133 	net = xp_net(policy);
1134 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1135 
1136 	if (xfrm_policy_inexact_insert_use_any_list(policy))
1137 		return &bin->hhead;
1138 
1139 	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1140 					       policy->family,
1141 					       policy->selector.prefixlen_d)) {
1142 		write_seqcount_begin(&bin->count);
1143 		n = xfrm_policy_inexact_insert_node(net,
1144 						    &bin->root_s,
1145 						    &policy->selector.saddr,
1146 						    policy->family,
1147 						    policy->selector.prefixlen_s,
1148 						    dir);
1149 		write_seqcount_end(&bin->count);
1150 		if (!n)
1151 			return NULL;
1152 
1153 		return &n->hhead;
1154 	}
1155 
1156 	/* daddr is fixed */
1157 	write_seqcount_begin(&bin->count);
1158 	n = xfrm_policy_inexact_insert_node(net,
1159 					    &bin->root_d,
1160 					    &policy->selector.daddr,
1161 					    policy->family,
1162 					    policy->selector.prefixlen_d, dir);
1163 	write_seqcount_end(&bin->count);
1164 	if (!n)
1165 		return NULL;
1166 
1167 	/* saddr is wildcard */
1168 	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1169 					       policy->family,
1170 					       policy->selector.prefixlen_s))
1171 		return &n->hhead;
1172 
1173 	write_seqcount_begin(&bin->count);
1174 	n = xfrm_policy_inexact_insert_node(net,
1175 					    &n->root,
1176 					    &policy->selector.saddr,
1177 					    policy->family,
1178 					    policy->selector.prefixlen_s, dir);
1179 	write_seqcount_end(&bin->count);
1180 	if (!n)
1181 		return NULL;
1182 
1183 	return &n->hhead;
1184 }
1185 
1186 static struct xfrm_policy *
1187 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1188 {
1189 	struct xfrm_pol_inexact_bin *bin;
1190 	struct xfrm_policy *delpol;
1191 	struct hlist_head *chain;
1192 	struct net *net;
1193 
1194 	bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1195 	if (!bin)
1196 		return ERR_PTR(-ENOMEM);
1197 
1198 	net = xp_net(policy);
1199 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1200 
1201 	chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1202 	if (!chain) {
1203 		__xfrm_policy_inexact_prune_bin(bin, false);
1204 		return ERR_PTR(-ENOMEM);
1205 	}
1206 
1207 	delpol = xfrm_policy_insert_list(chain, policy, excl);
1208 	if (delpol && excl) {
1209 		__xfrm_policy_inexact_prune_bin(bin, false);
1210 		return ERR_PTR(-EEXIST);
1211 	}
1212 
1213 	chain = &net->xfrm.policy_inexact[dir];
1214 	xfrm_policy_insert_inexact_list(chain, policy);
1215 
1216 	if (delpol)
1217 		__xfrm_policy_inexact_prune_bin(bin, false);
1218 
1219 	return delpol;
1220 }
1221 
1222 static void xfrm_hash_rebuild(struct work_struct *work)
1223 {
1224 	struct net *net = container_of(work, struct net,
1225 				       xfrm.policy_hthresh.work);
1226 	unsigned int hmask;
1227 	struct xfrm_policy *pol;
1228 	struct xfrm_policy *policy;
1229 	struct hlist_head *chain;
1230 	struct hlist_head *odst;
1231 	struct hlist_node *newpos;
1232 	int i;
1233 	int dir;
1234 	unsigned seq;
1235 	u8 lbits4, rbits4, lbits6, rbits6;
1236 
1237 	mutex_lock(&hash_resize_mutex);
1238 
1239 	/* read selector prefixlen thresholds */
1240 	do {
1241 		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1242 
1243 		lbits4 = net->xfrm.policy_hthresh.lbits4;
1244 		rbits4 = net->xfrm.policy_hthresh.rbits4;
1245 		lbits6 = net->xfrm.policy_hthresh.lbits6;
1246 		rbits6 = net->xfrm.policy_hthresh.rbits6;
1247 	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1248 
1249 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1250 	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1251 
1252 	/* make sure that we can insert the indirect policies again before
1253 	 * we start with destructive action.
1254 	 */
1255 	list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1256 		struct xfrm_pol_inexact_bin *bin;
1257 		u8 dbits, sbits;
1258 
1259 		dir = xfrm_policy_id2dir(policy->index);
1260 		if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1261 			continue;
1262 
1263 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1264 			if (policy->family == AF_INET) {
1265 				dbits = rbits4;
1266 				sbits = lbits4;
1267 			} else {
1268 				dbits = rbits6;
1269 				sbits = lbits6;
1270 			}
1271 		} else {
1272 			if (policy->family == AF_INET) {
1273 				dbits = lbits4;
1274 				sbits = rbits4;
1275 			} else {
1276 				dbits = lbits6;
1277 				sbits = rbits6;
1278 			}
1279 		}
1280 
1281 		if (policy->selector.prefixlen_d < dbits ||
1282 		    policy->selector.prefixlen_s < sbits)
1283 			continue;
1284 
1285 		bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1286 		if (!bin)
1287 			goto out_unlock;
1288 
1289 		if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1290 			goto out_unlock;
1291 	}
1292 
1293 	/* reset the bydst and inexact table in all directions */
1294 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1295 		struct hlist_node *n;
1296 
1297 		hlist_for_each_entry_safe(policy, n,
1298 					  &net->xfrm.policy_inexact[dir],
1299 					  bydst_inexact_list) {
1300 			hlist_del_rcu(&policy->bydst);
1301 			hlist_del_init(&policy->bydst_inexact_list);
1302 		}
1303 
1304 		hmask = net->xfrm.policy_bydst[dir].hmask;
1305 		odst = net->xfrm.policy_bydst[dir].table;
1306 		for (i = hmask; i >= 0; i--) {
1307 			hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1308 				hlist_del_rcu(&policy->bydst);
1309 		}
1310 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1311 			/* dir out => dst = remote, src = local */
1312 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1313 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1314 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1315 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1316 		} else {
1317 			/* dir in/fwd => dst = local, src = remote */
1318 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1319 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1320 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1321 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1322 		}
1323 	}
1324 
1325 	/* re-insert all policies by order of creation */
1326 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1327 		if (policy->walk.dead)
1328 			continue;
1329 		dir = xfrm_policy_id2dir(policy->index);
1330 		if (dir >= XFRM_POLICY_MAX) {
1331 			/* skip socket policies */
1332 			continue;
1333 		}
1334 		newpos = NULL;
1335 		chain = policy_hash_bysel(net, &policy->selector,
1336 					  policy->family, dir);
1337 
1338 		if (!chain) {
1339 			void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1340 
1341 			WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1342 			continue;
1343 		}
1344 
1345 		hlist_for_each_entry(pol, chain, bydst) {
1346 			if (policy->priority >= pol->priority)
1347 				newpos = &pol->bydst;
1348 			else
1349 				break;
1350 		}
1351 		if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1352 			hlist_add_behind_rcu(&policy->bydst, newpos);
1353 		else
1354 			hlist_add_head_rcu(&policy->bydst, chain);
1355 	}
1356 
1357 out_unlock:
1358 	__xfrm_policy_inexact_flush(net);
1359 	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1360 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1361 
1362 	mutex_unlock(&hash_resize_mutex);
1363 }
1364 
1365 void xfrm_policy_hash_rebuild(struct net *net)
1366 {
1367 	schedule_work(&net->xfrm.policy_hthresh.work);
1368 }
1369 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1370 
1371 /* Generate new index... KAME seems to generate them ordered by cost
1372  * of an absolute inpredictability of ordering of rules. This will not pass. */
1373 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1374 {
1375 	static u32 idx_generator;
1376 
1377 	for (;;) {
1378 		struct hlist_head *list;
1379 		struct xfrm_policy *p;
1380 		u32 idx;
1381 		int found;
1382 
1383 		if (!index) {
1384 			idx = (idx_generator | dir);
1385 			idx_generator += 8;
1386 		} else {
1387 			idx = index;
1388 			index = 0;
1389 		}
1390 
1391 		if (idx == 0)
1392 			idx = 8;
1393 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
1394 		found = 0;
1395 		hlist_for_each_entry(p, list, byidx) {
1396 			if (p->index == idx) {
1397 				found = 1;
1398 				break;
1399 			}
1400 		}
1401 		if (!found)
1402 			return idx;
1403 	}
1404 }
1405 
1406 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1407 {
1408 	u32 *p1 = (u32 *) s1;
1409 	u32 *p2 = (u32 *) s2;
1410 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
1411 	int i;
1412 
1413 	for (i = 0; i < len; i++) {
1414 		if (p1[i] != p2[i])
1415 			return 1;
1416 	}
1417 
1418 	return 0;
1419 }
1420 
1421 static void xfrm_policy_requeue(struct xfrm_policy *old,
1422 				struct xfrm_policy *new)
1423 {
1424 	struct xfrm_policy_queue *pq = &old->polq;
1425 	struct sk_buff_head list;
1426 
1427 	if (skb_queue_empty(&pq->hold_queue))
1428 		return;
1429 
1430 	__skb_queue_head_init(&list);
1431 
1432 	spin_lock_bh(&pq->hold_queue.lock);
1433 	skb_queue_splice_init(&pq->hold_queue, &list);
1434 	if (del_timer(&pq->hold_timer))
1435 		xfrm_pol_put(old);
1436 	spin_unlock_bh(&pq->hold_queue.lock);
1437 
1438 	pq = &new->polq;
1439 
1440 	spin_lock_bh(&pq->hold_queue.lock);
1441 	skb_queue_splice(&list, &pq->hold_queue);
1442 	pq->timeout = XFRM_QUEUE_TMO_MIN;
1443 	if (!mod_timer(&pq->hold_timer, jiffies))
1444 		xfrm_pol_hold(new);
1445 	spin_unlock_bh(&pq->hold_queue.lock);
1446 }
1447 
1448 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1449 					  struct xfrm_policy *pol)
1450 {
1451 	return mark->v == pol->mark.v && mark->m == pol->mark.m;
1452 }
1453 
1454 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1455 {
1456 	const struct xfrm_pol_inexact_key *k = data;
1457 	u32 a = k->type << 24 | k->dir << 16 | k->family;
1458 
1459 	return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1460 			    seed);
1461 }
1462 
1463 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1464 {
1465 	const struct xfrm_pol_inexact_bin *b = data;
1466 
1467 	return xfrm_pol_bin_key(&b->k, 0, seed);
1468 }
1469 
1470 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1471 			    const void *ptr)
1472 {
1473 	const struct xfrm_pol_inexact_key *key = arg->key;
1474 	const struct xfrm_pol_inexact_bin *b = ptr;
1475 	int ret;
1476 
1477 	if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1478 		return -1;
1479 
1480 	ret = b->k.dir ^ key->dir;
1481 	if (ret)
1482 		return ret;
1483 
1484 	ret = b->k.type ^ key->type;
1485 	if (ret)
1486 		return ret;
1487 
1488 	ret = b->k.family ^ key->family;
1489 	if (ret)
1490 		return ret;
1491 
1492 	return b->k.if_id ^ key->if_id;
1493 }
1494 
1495 static const struct rhashtable_params xfrm_pol_inexact_params = {
1496 	.head_offset		= offsetof(struct xfrm_pol_inexact_bin, head),
1497 	.hashfn			= xfrm_pol_bin_key,
1498 	.obj_hashfn		= xfrm_pol_bin_obj,
1499 	.obj_cmpfn		= xfrm_pol_bin_cmp,
1500 	.automatic_shrinking	= true,
1501 };
1502 
1503 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1504 					    struct xfrm_policy *policy)
1505 {
1506 	struct xfrm_policy *pol, *delpol = NULL;
1507 	struct hlist_node *newpos = NULL;
1508 	int i = 0;
1509 
1510 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1511 		if (pol->type == policy->type &&
1512 		    pol->if_id == policy->if_id &&
1513 		    !selector_cmp(&pol->selector, &policy->selector) &&
1514 		    xfrm_policy_mark_match(&policy->mark, pol) &&
1515 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1516 		    !WARN_ON(delpol)) {
1517 			delpol = pol;
1518 			if (policy->priority > pol->priority)
1519 				continue;
1520 		} else if (policy->priority >= pol->priority) {
1521 			newpos = &pol->bydst_inexact_list;
1522 			continue;
1523 		}
1524 		if (delpol)
1525 			break;
1526 	}
1527 
1528 	if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1529 		hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1530 	else
1531 		hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1532 
1533 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1534 		pol->pos = i;
1535 		i++;
1536 	}
1537 }
1538 
1539 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1540 						   struct xfrm_policy *policy,
1541 						   bool excl)
1542 {
1543 	struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1544 
1545 	hlist_for_each_entry(pol, chain, bydst) {
1546 		if (pol->type == policy->type &&
1547 		    pol->if_id == policy->if_id &&
1548 		    !selector_cmp(&pol->selector, &policy->selector) &&
1549 		    xfrm_policy_mark_match(&policy->mark, pol) &&
1550 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1551 		    !WARN_ON(delpol)) {
1552 			if (excl)
1553 				return ERR_PTR(-EEXIST);
1554 			delpol = pol;
1555 			if (policy->priority > pol->priority)
1556 				continue;
1557 		} else if (policy->priority >= pol->priority) {
1558 			newpos = pol;
1559 			continue;
1560 		}
1561 		if (delpol)
1562 			break;
1563 	}
1564 
1565 	if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1566 		hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1567 	else
1568 		/* Packet offload policies enter to the head
1569 		 * to speed-up lookups.
1570 		 */
1571 		hlist_add_head_rcu(&policy->bydst, chain);
1572 
1573 	return delpol;
1574 }
1575 
1576 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1577 {
1578 	struct net *net = xp_net(policy);
1579 	struct xfrm_policy *delpol;
1580 	struct hlist_head *chain;
1581 
1582 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1583 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1584 	if (chain)
1585 		delpol = xfrm_policy_insert_list(chain, policy, excl);
1586 	else
1587 		delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1588 
1589 	if (IS_ERR(delpol)) {
1590 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1591 		return PTR_ERR(delpol);
1592 	}
1593 
1594 	__xfrm_policy_link(policy, dir);
1595 
1596 	/* After previous checking, family can either be AF_INET or AF_INET6 */
1597 	if (policy->family == AF_INET)
1598 		rt_genid_bump_ipv4(net);
1599 	else
1600 		rt_genid_bump_ipv6(net);
1601 
1602 	if (delpol) {
1603 		xfrm_policy_requeue(delpol, policy);
1604 		__xfrm_policy_unlink(delpol, dir);
1605 	}
1606 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1607 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1608 	policy->curlft.add_time = ktime_get_real_seconds();
1609 	policy->curlft.use_time = 0;
1610 	if (!mod_timer(&policy->timer, jiffies + HZ))
1611 		xfrm_pol_hold(policy);
1612 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1613 
1614 	if (delpol)
1615 		xfrm_policy_kill(delpol);
1616 	else if (xfrm_bydst_should_resize(net, dir, NULL))
1617 		schedule_work(&net->xfrm.policy_hash_work);
1618 
1619 	return 0;
1620 }
1621 EXPORT_SYMBOL(xfrm_policy_insert);
1622 
1623 static struct xfrm_policy *
1624 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1625 			u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1626 			struct xfrm_sec_ctx *ctx)
1627 {
1628 	struct xfrm_policy *pol;
1629 
1630 	if (!chain)
1631 		return NULL;
1632 
1633 	hlist_for_each_entry(pol, chain, bydst) {
1634 		if (pol->type == type &&
1635 		    pol->if_id == if_id &&
1636 		    xfrm_policy_mark_match(mark, pol) &&
1637 		    !selector_cmp(sel, &pol->selector) &&
1638 		    xfrm_sec_ctx_match(ctx, pol->security))
1639 			return pol;
1640 	}
1641 
1642 	return NULL;
1643 }
1644 
1645 struct xfrm_policy *
1646 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1647 		      u8 type, int dir, struct xfrm_selector *sel,
1648 		      struct xfrm_sec_ctx *ctx, int delete, int *err)
1649 {
1650 	struct xfrm_pol_inexact_bin *bin = NULL;
1651 	struct xfrm_policy *pol, *ret = NULL;
1652 	struct hlist_head *chain;
1653 
1654 	*err = 0;
1655 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1656 	chain = policy_hash_bysel(net, sel, sel->family, dir);
1657 	if (!chain) {
1658 		struct xfrm_pol_inexact_candidates cand;
1659 		int i;
1660 
1661 		bin = xfrm_policy_inexact_lookup(net, type,
1662 						 sel->family, dir, if_id);
1663 		if (!bin) {
1664 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1665 			return NULL;
1666 		}
1667 
1668 		if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1669 							 &sel->saddr,
1670 							 &sel->daddr)) {
1671 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1672 			return NULL;
1673 		}
1674 
1675 		pol = NULL;
1676 		for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1677 			struct xfrm_policy *tmp;
1678 
1679 			tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1680 						      if_id, type, dir,
1681 						      sel, ctx);
1682 			if (!tmp)
1683 				continue;
1684 
1685 			if (!pol || tmp->pos < pol->pos)
1686 				pol = tmp;
1687 		}
1688 	} else {
1689 		pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1690 					      sel, ctx);
1691 	}
1692 
1693 	if (pol) {
1694 		xfrm_pol_hold(pol);
1695 		if (delete) {
1696 			*err = security_xfrm_policy_delete(pol->security);
1697 			if (*err) {
1698 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1699 				return pol;
1700 			}
1701 			__xfrm_policy_unlink(pol, dir);
1702 		}
1703 		ret = pol;
1704 	}
1705 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1706 
1707 	if (ret && delete)
1708 		xfrm_policy_kill(ret);
1709 	if (bin && delete)
1710 		xfrm_policy_inexact_prune_bin(bin);
1711 	return ret;
1712 }
1713 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1714 
1715 struct xfrm_policy *
1716 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1717 		 u8 type, int dir, u32 id, int delete, int *err)
1718 {
1719 	struct xfrm_policy *pol, *ret;
1720 	struct hlist_head *chain;
1721 
1722 	*err = -ENOENT;
1723 	if (xfrm_policy_id2dir(id) != dir)
1724 		return NULL;
1725 
1726 	*err = 0;
1727 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1728 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
1729 	ret = NULL;
1730 	hlist_for_each_entry(pol, chain, byidx) {
1731 		if (pol->type == type && pol->index == id &&
1732 		    pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1733 			xfrm_pol_hold(pol);
1734 			if (delete) {
1735 				*err = security_xfrm_policy_delete(
1736 								pol->security);
1737 				if (*err) {
1738 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1739 					return pol;
1740 				}
1741 				__xfrm_policy_unlink(pol, dir);
1742 			}
1743 			ret = pol;
1744 			break;
1745 		}
1746 	}
1747 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1748 
1749 	if (ret && delete)
1750 		xfrm_policy_kill(ret);
1751 	return ret;
1752 }
1753 EXPORT_SYMBOL(xfrm_policy_byid);
1754 
1755 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1756 static inline int
1757 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1758 {
1759 	struct xfrm_policy *pol;
1760 	int err = 0;
1761 
1762 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1763 		if (pol->walk.dead ||
1764 		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1765 		    pol->type != type)
1766 			continue;
1767 
1768 		err = security_xfrm_policy_delete(pol->security);
1769 		if (err) {
1770 			xfrm_audit_policy_delete(pol, 0, task_valid);
1771 			return err;
1772 		}
1773 	}
1774 	return err;
1775 }
1776 
1777 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1778 						     struct net_device *dev,
1779 						     bool task_valid)
1780 {
1781 	struct xfrm_policy *pol;
1782 	int err = 0;
1783 
1784 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1785 		if (pol->walk.dead ||
1786 		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1787 		    pol->xdo.dev != dev)
1788 			continue;
1789 
1790 		err = security_xfrm_policy_delete(pol->security);
1791 		if (err) {
1792 			xfrm_audit_policy_delete(pol, 0, task_valid);
1793 			return err;
1794 		}
1795 	}
1796 	return err;
1797 }
1798 #else
1799 static inline int
1800 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1801 {
1802 	return 0;
1803 }
1804 
1805 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1806 						     struct net_device *dev,
1807 						     bool task_valid)
1808 {
1809 	return 0;
1810 }
1811 #endif
1812 
1813 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1814 {
1815 	int dir, err = 0, cnt = 0;
1816 	struct xfrm_policy *pol;
1817 
1818 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1819 
1820 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1821 	if (err)
1822 		goto out;
1823 
1824 again:
1825 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1826 		dir = xfrm_policy_id2dir(pol->index);
1827 		if (pol->walk.dead ||
1828 		    dir >= XFRM_POLICY_MAX ||
1829 		    pol->type != type)
1830 			continue;
1831 
1832 		__xfrm_policy_unlink(pol, dir);
1833 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1834 		cnt++;
1835 		xfrm_audit_policy_delete(pol, 1, task_valid);
1836 		xfrm_policy_kill(pol);
1837 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1838 		goto again;
1839 	}
1840 	if (cnt)
1841 		__xfrm_policy_inexact_flush(net);
1842 	else
1843 		err = -ESRCH;
1844 out:
1845 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1846 	return err;
1847 }
1848 EXPORT_SYMBOL(xfrm_policy_flush);
1849 
1850 int xfrm_dev_policy_flush(struct net *net, struct net_device *dev,
1851 			  bool task_valid)
1852 {
1853 	int dir, err = 0, cnt = 0;
1854 	struct xfrm_policy *pol;
1855 
1856 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1857 
1858 	err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid);
1859 	if (err)
1860 		goto out;
1861 
1862 again:
1863 	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1864 		dir = xfrm_policy_id2dir(pol->index);
1865 		if (pol->walk.dead ||
1866 		    dir >= XFRM_POLICY_MAX ||
1867 		    pol->xdo.dev != dev)
1868 			continue;
1869 
1870 		__xfrm_policy_unlink(pol, dir);
1871 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1872 		cnt++;
1873 		xfrm_audit_policy_delete(pol, 1, task_valid);
1874 		xfrm_policy_kill(pol);
1875 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1876 		goto again;
1877 	}
1878 	if (cnt)
1879 		__xfrm_policy_inexact_flush(net);
1880 	else
1881 		err = -ESRCH;
1882 out:
1883 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1884 	return err;
1885 }
1886 EXPORT_SYMBOL(xfrm_dev_policy_flush);
1887 
1888 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1889 		     int (*func)(struct xfrm_policy *, int, int, void*),
1890 		     void *data)
1891 {
1892 	struct xfrm_policy *pol;
1893 	struct xfrm_policy_walk_entry *x;
1894 	int error = 0;
1895 
1896 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1897 	    walk->type != XFRM_POLICY_TYPE_ANY)
1898 		return -EINVAL;
1899 
1900 	if (list_empty(&walk->walk.all) && walk->seq != 0)
1901 		return 0;
1902 
1903 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1904 	if (list_empty(&walk->walk.all))
1905 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1906 	else
1907 		x = list_first_entry(&walk->walk.all,
1908 				     struct xfrm_policy_walk_entry, all);
1909 
1910 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1911 		if (x->dead)
1912 			continue;
1913 		pol = container_of(x, struct xfrm_policy, walk);
1914 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1915 		    walk->type != pol->type)
1916 			continue;
1917 		error = func(pol, xfrm_policy_id2dir(pol->index),
1918 			     walk->seq, data);
1919 		if (error) {
1920 			list_move_tail(&walk->walk.all, &x->all);
1921 			goto out;
1922 		}
1923 		walk->seq++;
1924 	}
1925 	if (walk->seq == 0) {
1926 		error = -ENOENT;
1927 		goto out;
1928 	}
1929 	list_del_init(&walk->walk.all);
1930 out:
1931 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1932 	return error;
1933 }
1934 EXPORT_SYMBOL(xfrm_policy_walk);
1935 
1936 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1937 {
1938 	INIT_LIST_HEAD(&walk->walk.all);
1939 	walk->walk.dead = 1;
1940 	walk->type = type;
1941 	walk->seq = 0;
1942 }
1943 EXPORT_SYMBOL(xfrm_policy_walk_init);
1944 
1945 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1946 {
1947 	if (list_empty(&walk->walk.all))
1948 		return;
1949 
1950 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1951 	list_del(&walk->walk.all);
1952 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1953 }
1954 EXPORT_SYMBOL(xfrm_policy_walk_done);
1955 
1956 /*
1957  * Find policy to apply to this flow.
1958  *
1959  * Returns 0 if policy found, else an -errno.
1960  */
1961 static int xfrm_policy_match(const struct xfrm_policy *pol,
1962 			     const struct flowi *fl,
1963 			     u8 type, u16 family, u32 if_id)
1964 {
1965 	const struct xfrm_selector *sel = &pol->selector;
1966 	int ret = -ESRCH;
1967 	bool match;
1968 
1969 	if (pol->family != family ||
1970 	    pol->if_id != if_id ||
1971 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1972 	    pol->type != type)
1973 		return ret;
1974 
1975 	match = xfrm_selector_match(sel, fl, family);
1976 	if (match)
1977 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1978 	return ret;
1979 }
1980 
1981 static struct xfrm_pol_inexact_node *
1982 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1983 				seqcount_spinlock_t *count,
1984 				const xfrm_address_t *addr, u16 family)
1985 {
1986 	const struct rb_node *parent;
1987 	int seq;
1988 
1989 again:
1990 	seq = read_seqcount_begin(count);
1991 
1992 	parent = rcu_dereference_raw(r->rb_node);
1993 	while (parent) {
1994 		struct xfrm_pol_inexact_node *node;
1995 		int delta;
1996 
1997 		node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1998 
1999 		delta = xfrm_policy_addr_delta(addr, &node->addr,
2000 					       node->prefixlen, family);
2001 		if (delta < 0) {
2002 			parent = rcu_dereference_raw(parent->rb_left);
2003 			continue;
2004 		} else if (delta > 0) {
2005 			parent = rcu_dereference_raw(parent->rb_right);
2006 			continue;
2007 		}
2008 
2009 		return node;
2010 	}
2011 
2012 	if (read_seqcount_retry(count, seq))
2013 		goto again;
2014 
2015 	return NULL;
2016 }
2017 
2018 static bool
2019 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
2020 				    struct xfrm_pol_inexact_bin *b,
2021 				    const xfrm_address_t *saddr,
2022 				    const xfrm_address_t *daddr)
2023 {
2024 	struct xfrm_pol_inexact_node *n;
2025 	u16 family;
2026 
2027 	if (!b)
2028 		return false;
2029 
2030 	family = b->k.family;
2031 	memset(cand, 0, sizeof(*cand));
2032 	cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
2033 
2034 	n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
2035 					    family);
2036 	if (n) {
2037 		cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
2038 		n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
2039 						    family);
2040 		if (n)
2041 			cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
2042 	}
2043 
2044 	n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
2045 					    family);
2046 	if (n)
2047 		cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
2048 
2049 	return true;
2050 }
2051 
2052 static struct xfrm_pol_inexact_bin *
2053 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
2054 			       u8 dir, u32 if_id)
2055 {
2056 	struct xfrm_pol_inexact_key k = {
2057 		.family = family,
2058 		.type = type,
2059 		.dir = dir,
2060 		.if_id = if_id,
2061 	};
2062 
2063 	write_pnet(&k.net, net);
2064 
2065 	return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
2066 				 xfrm_pol_inexact_params);
2067 }
2068 
2069 static struct xfrm_pol_inexact_bin *
2070 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2071 			   u8 dir, u32 if_id)
2072 {
2073 	struct xfrm_pol_inexact_bin *bin;
2074 
2075 	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2076 
2077 	rcu_read_lock();
2078 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2079 	rcu_read_unlock();
2080 
2081 	return bin;
2082 }
2083 
2084 static struct xfrm_policy *
2085 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2086 			      struct xfrm_policy *prefer,
2087 			      const struct flowi *fl,
2088 			      u8 type, u16 family, u32 if_id)
2089 {
2090 	u32 priority = prefer ? prefer->priority : ~0u;
2091 	struct xfrm_policy *pol;
2092 
2093 	if (!chain)
2094 		return NULL;
2095 
2096 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2097 		int err;
2098 
2099 		if (pol->priority > priority)
2100 			break;
2101 
2102 		err = xfrm_policy_match(pol, fl, type, family, if_id);
2103 		if (err) {
2104 			if (err != -ESRCH)
2105 				return ERR_PTR(err);
2106 
2107 			continue;
2108 		}
2109 
2110 		if (prefer) {
2111 			/* matches.  Is it older than *prefer? */
2112 			if (pol->priority == priority &&
2113 			    prefer->pos < pol->pos)
2114 				return prefer;
2115 		}
2116 
2117 		return pol;
2118 	}
2119 
2120 	return NULL;
2121 }
2122 
2123 static struct xfrm_policy *
2124 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2125 			    struct xfrm_policy *prefer,
2126 			    const struct flowi *fl,
2127 			    u8 type, u16 family, u32 if_id)
2128 {
2129 	struct xfrm_policy *tmp;
2130 	int i;
2131 
2132 	for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2133 		tmp = __xfrm_policy_eval_candidates(cand->res[i],
2134 						    prefer,
2135 						    fl, type, family, if_id);
2136 		if (!tmp)
2137 			continue;
2138 
2139 		if (IS_ERR(tmp))
2140 			return tmp;
2141 		prefer = tmp;
2142 	}
2143 
2144 	return prefer;
2145 }
2146 
2147 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2148 						     const struct flowi *fl,
2149 						     u16 family, u8 dir,
2150 						     u32 if_id)
2151 {
2152 	struct xfrm_pol_inexact_candidates cand;
2153 	const xfrm_address_t *daddr, *saddr;
2154 	struct xfrm_pol_inexact_bin *bin;
2155 	struct xfrm_policy *pol, *ret;
2156 	struct hlist_head *chain;
2157 	unsigned int sequence;
2158 	int err;
2159 
2160 	daddr = xfrm_flowi_daddr(fl, family);
2161 	saddr = xfrm_flowi_saddr(fl, family);
2162 	if (unlikely(!daddr || !saddr))
2163 		return NULL;
2164 
2165 	rcu_read_lock();
2166  retry:
2167 	do {
2168 		sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2169 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
2170 	} while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2171 
2172 	ret = NULL;
2173 	hlist_for_each_entry_rcu(pol, chain, bydst) {
2174 		err = xfrm_policy_match(pol, fl, type, family, if_id);
2175 		if (err) {
2176 			if (err == -ESRCH)
2177 				continue;
2178 			else {
2179 				ret = ERR_PTR(err);
2180 				goto fail;
2181 			}
2182 		} else {
2183 			ret = pol;
2184 			break;
2185 		}
2186 	}
2187 	if (ret && ret->xdo.type == XFRM_DEV_OFFLOAD_PACKET)
2188 		goto skip_inexact;
2189 
2190 	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2191 	if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2192 							 daddr))
2193 		goto skip_inexact;
2194 
2195 	pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2196 					  family, if_id);
2197 	if (pol) {
2198 		ret = pol;
2199 		if (IS_ERR(pol))
2200 			goto fail;
2201 	}
2202 
2203 skip_inexact:
2204 	if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2205 		goto retry;
2206 
2207 	if (ret && !xfrm_pol_hold_rcu(ret))
2208 		goto retry;
2209 fail:
2210 	rcu_read_unlock();
2211 
2212 	return ret;
2213 }
2214 
2215 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2216 					      const struct flowi *fl,
2217 					      u16 family, u8 dir, u32 if_id)
2218 {
2219 #ifdef CONFIG_XFRM_SUB_POLICY
2220 	struct xfrm_policy *pol;
2221 
2222 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2223 					dir, if_id);
2224 	if (pol != NULL)
2225 		return pol;
2226 #endif
2227 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2228 					 dir, if_id);
2229 }
2230 
2231 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2232 						 const struct flowi *fl,
2233 						 u16 family, u32 if_id)
2234 {
2235 	struct xfrm_policy *pol;
2236 
2237 	rcu_read_lock();
2238  again:
2239 	pol = rcu_dereference(sk->sk_policy[dir]);
2240 	if (pol != NULL) {
2241 		bool match;
2242 		int err = 0;
2243 
2244 		if (pol->family != family) {
2245 			pol = NULL;
2246 			goto out;
2247 		}
2248 
2249 		match = xfrm_selector_match(&pol->selector, fl, family);
2250 		if (match) {
2251 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2252 			    pol->if_id != if_id) {
2253 				pol = NULL;
2254 				goto out;
2255 			}
2256 			err = security_xfrm_policy_lookup(pol->security,
2257 						      fl->flowi_secid);
2258 			if (!err) {
2259 				if (!xfrm_pol_hold_rcu(pol))
2260 					goto again;
2261 			} else if (err == -ESRCH) {
2262 				pol = NULL;
2263 			} else {
2264 				pol = ERR_PTR(err);
2265 			}
2266 		} else
2267 			pol = NULL;
2268 	}
2269 out:
2270 	rcu_read_unlock();
2271 	return pol;
2272 }
2273 
2274 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2275 {
2276 	struct net *net = xp_net(pol);
2277 
2278 	list_add(&pol->walk.all, &net->xfrm.policy_all);
2279 	net->xfrm.policy_count[dir]++;
2280 	xfrm_pol_hold(pol);
2281 }
2282 
2283 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2284 						int dir)
2285 {
2286 	struct net *net = xp_net(pol);
2287 
2288 	if (list_empty(&pol->walk.all))
2289 		return NULL;
2290 
2291 	/* Socket policies are not hashed. */
2292 	if (!hlist_unhashed(&pol->bydst)) {
2293 		hlist_del_rcu(&pol->bydst);
2294 		hlist_del_init(&pol->bydst_inexact_list);
2295 		hlist_del(&pol->byidx);
2296 	}
2297 
2298 	list_del_init(&pol->walk.all);
2299 	net->xfrm.policy_count[dir]--;
2300 
2301 	return pol;
2302 }
2303 
2304 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2305 {
2306 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2307 }
2308 
2309 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2310 {
2311 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2312 }
2313 
2314 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2315 {
2316 	struct net *net = xp_net(pol);
2317 
2318 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2319 	pol = __xfrm_policy_unlink(pol, dir);
2320 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2321 	if (pol) {
2322 		xfrm_dev_policy_delete(pol);
2323 		xfrm_policy_kill(pol);
2324 		return 0;
2325 	}
2326 	return -ENOENT;
2327 }
2328 EXPORT_SYMBOL(xfrm_policy_delete);
2329 
2330 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2331 {
2332 	struct net *net = sock_net(sk);
2333 	struct xfrm_policy *old_pol;
2334 
2335 #ifdef CONFIG_XFRM_SUB_POLICY
2336 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2337 		return -EINVAL;
2338 #endif
2339 
2340 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2341 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2342 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2343 	if (pol) {
2344 		pol->curlft.add_time = ktime_get_real_seconds();
2345 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2346 		xfrm_sk_policy_link(pol, dir);
2347 	}
2348 	rcu_assign_pointer(sk->sk_policy[dir], pol);
2349 	if (old_pol) {
2350 		if (pol)
2351 			xfrm_policy_requeue(old_pol, pol);
2352 
2353 		/* Unlinking succeeds always. This is the only function
2354 		 * allowed to delete or replace socket policy.
2355 		 */
2356 		xfrm_sk_policy_unlink(old_pol, dir);
2357 	}
2358 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2359 
2360 	if (old_pol) {
2361 		xfrm_policy_kill(old_pol);
2362 	}
2363 	return 0;
2364 }
2365 
2366 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2367 {
2368 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2369 	struct net *net = xp_net(old);
2370 
2371 	if (newp) {
2372 		newp->selector = old->selector;
2373 		if (security_xfrm_policy_clone(old->security,
2374 					       &newp->security)) {
2375 			kfree(newp);
2376 			return NULL;  /* ENOMEM */
2377 		}
2378 		newp->lft = old->lft;
2379 		newp->curlft = old->curlft;
2380 		newp->mark = old->mark;
2381 		newp->if_id = old->if_id;
2382 		newp->action = old->action;
2383 		newp->flags = old->flags;
2384 		newp->xfrm_nr = old->xfrm_nr;
2385 		newp->index = old->index;
2386 		newp->type = old->type;
2387 		newp->family = old->family;
2388 		memcpy(newp->xfrm_vec, old->xfrm_vec,
2389 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2390 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2391 		xfrm_sk_policy_link(newp, dir);
2392 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2393 		xfrm_pol_put(newp);
2394 	}
2395 	return newp;
2396 }
2397 
2398 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2399 {
2400 	const struct xfrm_policy *p;
2401 	struct xfrm_policy *np;
2402 	int i, ret = 0;
2403 
2404 	rcu_read_lock();
2405 	for (i = 0; i < 2; i++) {
2406 		p = rcu_dereference(osk->sk_policy[i]);
2407 		if (p) {
2408 			np = clone_policy(p, i);
2409 			if (unlikely(!np)) {
2410 				ret = -ENOMEM;
2411 				break;
2412 			}
2413 			rcu_assign_pointer(sk->sk_policy[i], np);
2414 		}
2415 	}
2416 	rcu_read_unlock();
2417 	return ret;
2418 }
2419 
2420 static int
2421 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2422 	       xfrm_address_t *remote, unsigned short family, u32 mark)
2423 {
2424 	int err;
2425 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2426 
2427 	if (unlikely(afinfo == NULL))
2428 		return -EINVAL;
2429 	err = afinfo->get_saddr(net, oif, local, remote, mark);
2430 	rcu_read_unlock();
2431 	return err;
2432 }
2433 
2434 /* Resolve list of templates for the flow, given policy. */
2435 
2436 static int
2437 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2438 		      struct xfrm_state **xfrm, unsigned short family)
2439 {
2440 	struct net *net = xp_net(policy);
2441 	int nx;
2442 	int i, error;
2443 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2444 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2445 	xfrm_address_t tmp;
2446 
2447 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2448 		struct xfrm_state *x;
2449 		xfrm_address_t *remote = daddr;
2450 		xfrm_address_t *local  = saddr;
2451 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2452 
2453 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
2454 		    tmpl->mode == XFRM_MODE_BEET) {
2455 			remote = &tmpl->id.daddr;
2456 			local = &tmpl->saddr;
2457 			if (xfrm_addr_any(local, tmpl->encap_family)) {
2458 				error = xfrm_get_saddr(net, fl->flowi_oif,
2459 						       &tmp, remote,
2460 						       tmpl->encap_family, 0);
2461 				if (error)
2462 					goto fail;
2463 				local = &tmp;
2464 			}
2465 		}
2466 
2467 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2468 				    family, policy->if_id);
2469 
2470 		if (x && x->km.state == XFRM_STATE_VALID) {
2471 			xfrm[nx++] = x;
2472 			daddr = remote;
2473 			saddr = local;
2474 			continue;
2475 		}
2476 		if (x) {
2477 			error = (x->km.state == XFRM_STATE_ERROR ?
2478 				 -EINVAL : -EAGAIN);
2479 			xfrm_state_put(x);
2480 		} else if (error == -ESRCH) {
2481 			error = -EAGAIN;
2482 		}
2483 
2484 		if (!tmpl->optional)
2485 			goto fail;
2486 	}
2487 	return nx;
2488 
2489 fail:
2490 	for (nx--; nx >= 0; nx--)
2491 		xfrm_state_put(xfrm[nx]);
2492 	return error;
2493 }
2494 
2495 static int
2496 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2497 		  struct xfrm_state **xfrm, unsigned short family)
2498 {
2499 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
2500 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2501 	int cnx = 0;
2502 	int error;
2503 	int ret;
2504 	int i;
2505 
2506 	for (i = 0; i < npols; i++) {
2507 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2508 			error = -ENOBUFS;
2509 			goto fail;
2510 		}
2511 
2512 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2513 		if (ret < 0) {
2514 			error = ret;
2515 			goto fail;
2516 		} else
2517 			cnx += ret;
2518 	}
2519 
2520 	/* found states are sorted for outbound processing */
2521 	if (npols > 1)
2522 		xfrm_state_sort(xfrm, tpp, cnx, family);
2523 
2524 	return cnx;
2525 
2526  fail:
2527 	for (cnx--; cnx >= 0; cnx--)
2528 		xfrm_state_put(tpp[cnx]);
2529 	return error;
2530 
2531 }
2532 
2533 static int xfrm_get_tos(const struct flowi *fl, int family)
2534 {
2535 	if (family == AF_INET)
2536 		return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2537 
2538 	return 0;
2539 }
2540 
2541 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2542 {
2543 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2544 	struct dst_ops *dst_ops;
2545 	struct xfrm_dst *xdst;
2546 
2547 	if (!afinfo)
2548 		return ERR_PTR(-EINVAL);
2549 
2550 	switch (family) {
2551 	case AF_INET:
2552 		dst_ops = &net->xfrm.xfrm4_dst_ops;
2553 		break;
2554 #if IS_ENABLED(CONFIG_IPV6)
2555 	case AF_INET6:
2556 		dst_ops = &net->xfrm.xfrm6_dst_ops;
2557 		break;
2558 #endif
2559 	default:
2560 		BUG();
2561 	}
2562 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2563 
2564 	if (likely(xdst)) {
2565 		memset_after(xdst, 0, u.dst);
2566 	} else
2567 		xdst = ERR_PTR(-ENOBUFS);
2568 
2569 	rcu_read_unlock();
2570 
2571 	return xdst;
2572 }
2573 
2574 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2575 			   int nfheader_len)
2576 {
2577 	if (dst->ops->family == AF_INET6) {
2578 		struct rt6_info *rt = (struct rt6_info *)dst;
2579 		path->path_cookie = rt6_get_cookie(rt);
2580 		path->u.rt6.rt6i_nfheader_len = nfheader_len;
2581 	}
2582 }
2583 
2584 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2585 				const struct flowi *fl)
2586 {
2587 	const struct xfrm_policy_afinfo *afinfo =
2588 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2589 	int err;
2590 
2591 	if (!afinfo)
2592 		return -EINVAL;
2593 
2594 	err = afinfo->fill_dst(xdst, dev, fl);
2595 
2596 	rcu_read_unlock();
2597 
2598 	return err;
2599 }
2600 
2601 
2602 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2603  * all the metrics... Shortly, bundle a bundle.
2604  */
2605 
2606 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2607 					    struct xfrm_state **xfrm,
2608 					    struct xfrm_dst **bundle,
2609 					    int nx,
2610 					    const struct flowi *fl,
2611 					    struct dst_entry *dst)
2612 {
2613 	const struct xfrm_state_afinfo *afinfo;
2614 	const struct xfrm_mode *inner_mode;
2615 	struct net *net = xp_net(policy);
2616 	unsigned long now = jiffies;
2617 	struct net_device *dev;
2618 	struct xfrm_dst *xdst_prev = NULL;
2619 	struct xfrm_dst *xdst0 = NULL;
2620 	int i = 0;
2621 	int err;
2622 	int header_len = 0;
2623 	int nfheader_len = 0;
2624 	int trailer_len = 0;
2625 	int tos;
2626 	int family = policy->selector.family;
2627 	xfrm_address_t saddr, daddr;
2628 
2629 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2630 
2631 	tos = xfrm_get_tos(fl, family);
2632 
2633 	dst_hold(dst);
2634 
2635 	for (; i < nx; i++) {
2636 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2637 		struct dst_entry *dst1 = &xdst->u.dst;
2638 
2639 		err = PTR_ERR(xdst);
2640 		if (IS_ERR(xdst)) {
2641 			dst_release(dst);
2642 			goto put_states;
2643 		}
2644 
2645 		bundle[i] = xdst;
2646 		if (!xdst_prev)
2647 			xdst0 = xdst;
2648 		else
2649 			/* Ref count is taken during xfrm_alloc_dst()
2650 			 * No need to do dst_clone() on dst1
2651 			 */
2652 			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2653 
2654 		if (xfrm[i]->sel.family == AF_UNSPEC) {
2655 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
2656 							xfrm_af2proto(family));
2657 			if (!inner_mode) {
2658 				err = -EAFNOSUPPORT;
2659 				dst_release(dst);
2660 				goto put_states;
2661 			}
2662 		} else
2663 			inner_mode = &xfrm[i]->inner_mode;
2664 
2665 		xdst->route = dst;
2666 		dst_copy_metrics(dst1, dst);
2667 
2668 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2669 			__u32 mark = 0;
2670 			int oif;
2671 
2672 			if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2673 				mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2674 
2675 			family = xfrm[i]->props.family;
2676 			oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2677 			dst = xfrm_dst_lookup(xfrm[i], tos, oif,
2678 					      &saddr, &daddr, family, mark);
2679 			err = PTR_ERR(dst);
2680 			if (IS_ERR(dst))
2681 				goto put_states;
2682 		} else
2683 			dst_hold(dst);
2684 
2685 		dst1->xfrm = xfrm[i];
2686 		xdst->xfrm_genid = xfrm[i]->genid;
2687 
2688 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2689 		dst1->lastuse = now;
2690 
2691 		dst1->input = dst_discard;
2692 
2693 		rcu_read_lock();
2694 		afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2695 		if (likely(afinfo))
2696 			dst1->output = afinfo->output;
2697 		else
2698 			dst1->output = dst_discard_out;
2699 		rcu_read_unlock();
2700 
2701 		xdst_prev = xdst;
2702 
2703 		header_len += xfrm[i]->props.header_len;
2704 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2705 			nfheader_len += xfrm[i]->props.header_len;
2706 		trailer_len += xfrm[i]->props.trailer_len;
2707 	}
2708 
2709 	xfrm_dst_set_child(xdst_prev, dst);
2710 	xdst0->path = dst;
2711 
2712 	err = -ENODEV;
2713 	dev = dst->dev;
2714 	if (!dev)
2715 		goto free_dst;
2716 
2717 	xfrm_init_path(xdst0, dst, nfheader_len);
2718 	xfrm_init_pmtu(bundle, nx);
2719 
2720 	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2721 	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2722 		err = xfrm_fill_dst(xdst_prev, dev, fl);
2723 		if (err)
2724 			goto free_dst;
2725 
2726 		xdst_prev->u.dst.header_len = header_len;
2727 		xdst_prev->u.dst.trailer_len = trailer_len;
2728 		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2729 		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2730 	}
2731 
2732 	return &xdst0->u.dst;
2733 
2734 put_states:
2735 	for (; i < nx; i++)
2736 		xfrm_state_put(xfrm[i]);
2737 free_dst:
2738 	if (xdst0)
2739 		dst_release_immediate(&xdst0->u.dst);
2740 
2741 	return ERR_PTR(err);
2742 }
2743 
2744 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2745 				struct xfrm_policy **pols,
2746 				int *num_pols, int *num_xfrms)
2747 {
2748 	int i;
2749 
2750 	if (*num_pols == 0 || !pols[0]) {
2751 		*num_pols = 0;
2752 		*num_xfrms = 0;
2753 		return 0;
2754 	}
2755 	if (IS_ERR(pols[0])) {
2756 		*num_pols = 0;
2757 		return PTR_ERR(pols[0]);
2758 	}
2759 
2760 	*num_xfrms = pols[0]->xfrm_nr;
2761 
2762 #ifdef CONFIG_XFRM_SUB_POLICY
2763 	if (pols[0]->action == XFRM_POLICY_ALLOW &&
2764 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2765 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2766 						    XFRM_POLICY_TYPE_MAIN,
2767 						    fl, family,
2768 						    XFRM_POLICY_OUT,
2769 						    pols[0]->if_id);
2770 		if (pols[1]) {
2771 			if (IS_ERR(pols[1])) {
2772 				xfrm_pols_put(pols, *num_pols);
2773 				*num_pols = 0;
2774 				return PTR_ERR(pols[1]);
2775 			}
2776 			(*num_pols)++;
2777 			(*num_xfrms) += pols[1]->xfrm_nr;
2778 		}
2779 	}
2780 #endif
2781 	for (i = 0; i < *num_pols; i++) {
2782 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
2783 			*num_xfrms = -1;
2784 			break;
2785 		}
2786 	}
2787 
2788 	return 0;
2789 
2790 }
2791 
2792 static struct xfrm_dst *
2793 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2794 			       const struct flowi *fl, u16 family,
2795 			       struct dst_entry *dst_orig)
2796 {
2797 	struct net *net = xp_net(pols[0]);
2798 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2799 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2800 	struct xfrm_dst *xdst;
2801 	struct dst_entry *dst;
2802 	int err;
2803 
2804 	/* Try to instantiate a bundle */
2805 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2806 	if (err <= 0) {
2807 		if (err == 0)
2808 			return NULL;
2809 
2810 		if (err != -EAGAIN)
2811 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2812 		return ERR_PTR(err);
2813 	}
2814 
2815 	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2816 	if (IS_ERR(dst)) {
2817 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2818 		return ERR_CAST(dst);
2819 	}
2820 
2821 	xdst = (struct xfrm_dst *)dst;
2822 	xdst->num_xfrms = err;
2823 	xdst->num_pols = num_pols;
2824 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2825 	xdst->policy_genid = atomic_read(&pols[0]->genid);
2826 
2827 	return xdst;
2828 }
2829 
2830 static void xfrm_policy_queue_process(struct timer_list *t)
2831 {
2832 	struct sk_buff *skb;
2833 	struct sock *sk;
2834 	struct dst_entry *dst;
2835 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2836 	struct net *net = xp_net(pol);
2837 	struct xfrm_policy_queue *pq = &pol->polq;
2838 	struct flowi fl;
2839 	struct sk_buff_head list;
2840 	__u32 skb_mark;
2841 
2842 	spin_lock(&pq->hold_queue.lock);
2843 	skb = skb_peek(&pq->hold_queue);
2844 	if (!skb) {
2845 		spin_unlock(&pq->hold_queue.lock);
2846 		goto out;
2847 	}
2848 	dst = skb_dst(skb);
2849 	sk = skb->sk;
2850 
2851 	/* Fixup the mark to support VTI. */
2852 	skb_mark = skb->mark;
2853 	skb->mark = pol->mark.v;
2854 	xfrm_decode_session(skb, &fl, dst->ops->family);
2855 	skb->mark = skb_mark;
2856 	spin_unlock(&pq->hold_queue.lock);
2857 
2858 	dst_hold(xfrm_dst_path(dst));
2859 	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2860 	if (IS_ERR(dst))
2861 		goto purge_queue;
2862 
2863 	if (dst->flags & DST_XFRM_QUEUE) {
2864 		dst_release(dst);
2865 
2866 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2867 			goto purge_queue;
2868 
2869 		pq->timeout = pq->timeout << 1;
2870 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2871 			xfrm_pol_hold(pol);
2872 		goto out;
2873 	}
2874 
2875 	dst_release(dst);
2876 
2877 	__skb_queue_head_init(&list);
2878 
2879 	spin_lock(&pq->hold_queue.lock);
2880 	pq->timeout = 0;
2881 	skb_queue_splice_init(&pq->hold_queue, &list);
2882 	spin_unlock(&pq->hold_queue.lock);
2883 
2884 	while (!skb_queue_empty(&list)) {
2885 		skb = __skb_dequeue(&list);
2886 
2887 		/* Fixup the mark to support VTI. */
2888 		skb_mark = skb->mark;
2889 		skb->mark = pol->mark.v;
2890 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2891 		skb->mark = skb_mark;
2892 
2893 		dst_hold(xfrm_dst_path(skb_dst(skb)));
2894 		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2895 		if (IS_ERR(dst)) {
2896 			kfree_skb(skb);
2897 			continue;
2898 		}
2899 
2900 		nf_reset_ct(skb);
2901 		skb_dst_drop(skb);
2902 		skb_dst_set(skb, dst);
2903 
2904 		dst_output(net, skb->sk, skb);
2905 	}
2906 
2907 out:
2908 	xfrm_pol_put(pol);
2909 	return;
2910 
2911 purge_queue:
2912 	pq->timeout = 0;
2913 	skb_queue_purge(&pq->hold_queue);
2914 	xfrm_pol_put(pol);
2915 }
2916 
2917 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2918 {
2919 	unsigned long sched_next;
2920 	struct dst_entry *dst = skb_dst(skb);
2921 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2922 	struct xfrm_policy *pol = xdst->pols[0];
2923 	struct xfrm_policy_queue *pq = &pol->polq;
2924 
2925 	if (unlikely(skb_fclone_busy(sk, skb))) {
2926 		kfree_skb(skb);
2927 		return 0;
2928 	}
2929 
2930 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2931 		kfree_skb(skb);
2932 		return -EAGAIN;
2933 	}
2934 
2935 	skb_dst_force(skb);
2936 
2937 	spin_lock_bh(&pq->hold_queue.lock);
2938 
2939 	if (!pq->timeout)
2940 		pq->timeout = XFRM_QUEUE_TMO_MIN;
2941 
2942 	sched_next = jiffies + pq->timeout;
2943 
2944 	if (del_timer(&pq->hold_timer)) {
2945 		if (time_before(pq->hold_timer.expires, sched_next))
2946 			sched_next = pq->hold_timer.expires;
2947 		xfrm_pol_put(pol);
2948 	}
2949 
2950 	__skb_queue_tail(&pq->hold_queue, skb);
2951 	if (!mod_timer(&pq->hold_timer, sched_next))
2952 		xfrm_pol_hold(pol);
2953 
2954 	spin_unlock_bh(&pq->hold_queue.lock);
2955 
2956 	return 0;
2957 }
2958 
2959 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2960 						 struct xfrm_flo *xflo,
2961 						 const struct flowi *fl,
2962 						 int num_xfrms,
2963 						 u16 family)
2964 {
2965 	int err;
2966 	struct net_device *dev;
2967 	struct dst_entry *dst;
2968 	struct dst_entry *dst1;
2969 	struct xfrm_dst *xdst;
2970 
2971 	xdst = xfrm_alloc_dst(net, family);
2972 	if (IS_ERR(xdst))
2973 		return xdst;
2974 
2975 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2976 	    net->xfrm.sysctl_larval_drop ||
2977 	    num_xfrms <= 0)
2978 		return xdst;
2979 
2980 	dst = xflo->dst_orig;
2981 	dst1 = &xdst->u.dst;
2982 	dst_hold(dst);
2983 	xdst->route = dst;
2984 
2985 	dst_copy_metrics(dst1, dst);
2986 
2987 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2988 	dst1->flags |= DST_XFRM_QUEUE;
2989 	dst1->lastuse = jiffies;
2990 
2991 	dst1->input = dst_discard;
2992 	dst1->output = xdst_queue_output;
2993 
2994 	dst_hold(dst);
2995 	xfrm_dst_set_child(xdst, dst);
2996 	xdst->path = dst;
2997 
2998 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2999 
3000 	err = -ENODEV;
3001 	dev = dst->dev;
3002 	if (!dev)
3003 		goto free_dst;
3004 
3005 	err = xfrm_fill_dst(xdst, dev, fl);
3006 	if (err)
3007 		goto free_dst;
3008 
3009 out:
3010 	return xdst;
3011 
3012 free_dst:
3013 	dst_release(dst1);
3014 	xdst = ERR_PTR(err);
3015 	goto out;
3016 }
3017 
3018 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
3019 					   const struct flowi *fl,
3020 					   u16 family, u8 dir,
3021 					   struct xfrm_flo *xflo, u32 if_id)
3022 {
3023 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3024 	int num_pols = 0, num_xfrms = 0, err;
3025 	struct xfrm_dst *xdst;
3026 
3027 	/* Resolve policies to use if we couldn't get them from
3028 	 * previous cache entry */
3029 	num_pols = 1;
3030 	pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
3031 	err = xfrm_expand_policies(fl, family, pols,
3032 					   &num_pols, &num_xfrms);
3033 	if (err < 0)
3034 		goto inc_error;
3035 	if (num_pols == 0)
3036 		return NULL;
3037 	if (num_xfrms <= 0)
3038 		goto make_dummy_bundle;
3039 
3040 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
3041 					      xflo->dst_orig);
3042 	if (IS_ERR(xdst)) {
3043 		err = PTR_ERR(xdst);
3044 		if (err == -EREMOTE) {
3045 			xfrm_pols_put(pols, num_pols);
3046 			return NULL;
3047 		}
3048 
3049 		if (err != -EAGAIN)
3050 			goto error;
3051 		goto make_dummy_bundle;
3052 	} else if (xdst == NULL) {
3053 		num_xfrms = 0;
3054 		goto make_dummy_bundle;
3055 	}
3056 
3057 	return xdst;
3058 
3059 make_dummy_bundle:
3060 	/* We found policies, but there's no bundles to instantiate:
3061 	 * either because the policy blocks, has no transformations or
3062 	 * we could not build template (no xfrm_states).*/
3063 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
3064 	if (IS_ERR(xdst)) {
3065 		xfrm_pols_put(pols, num_pols);
3066 		return ERR_CAST(xdst);
3067 	}
3068 	xdst->num_pols = num_pols;
3069 	xdst->num_xfrms = num_xfrms;
3070 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
3071 
3072 	return xdst;
3073 
3074 inc_error:
3075 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3076 error:
3077 	xfrm_pols_put(pols, num_pols);
3078 	return ERR_PTR(err);
3079 }
3080 
3081 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3082 					struct dst_entry *dst_orig)
3083 {
3084 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3085 	struct dst_entry *ret;
3086 
3087 	if (!afinfo) {
3088 		dst_release(dst_orig);
3089 		return ERR_PTR(-EINVAL);
3090 	} else {
3091 		ret = afinfo->blackhole_route(net, dst_orig);
3092 	}
3093 	rcu_read_unlock();
3094 
3095 	return ret;
3096 }
3097 
3098 /* Finds/creates a bundle for given flow and if_id
3099  *
3100  * At the moment we eat a raw IP route. Mostly to speed up lookups
3101  * on interfaces with disabled IPsec.
3102  *
3103  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3104  * compatibility
3105  */
3106 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3107 					struct dst_entry *dst_orig,
3108 					const struct flowi *fl,
3109 					const struct sock *sk,
3110 					int flags, u32 if_id)
3111 {
3112 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3113 	struct xfrm_dst *xdst;
3114 	struct dst_entry *dst, *route;
3115 	u16 family = dst_orig->ops->family;
3116 	u8 dir = XFRM_POLICY_OUT;
3117 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3118 
3119 	dst = NULL;
3120 	xdst = NULL;
3121 	route = NULL;
3122 
3123 	sk = sk_const_to_full_sk(sk);
3124 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3125 		num_pols = 1;
3126 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3127 						if_id);
3128 		err = xfrm_expand_policies(fl, family, pols,
3129 					   &num_pols, &num_xfrms);
3130 		if (err < 0)
3131 			goto dropdst;
3132 
3133 		if (num_pols) {
3134 			if (num_xfrms <= 0) {
3135 				drop_pols = num_pols;
3136 				goto no_transform;
3137 			}
3138 
3139 			xdst = xfrm_resolve_and_create_bundle(
3140 					pols, num_pols, fl,
3141 					family, dst_orig);
3142 
3143 			if (IS_ERR(xdst)) {
3144 				xfrm_pols_put(pols, num_pols);
3145 				err = PTR_ERR(xdst);
3146 				if (err == -EREMOTE)
3147 					goto nopol;
3148 
3149 				goto dropdst;
3150 			} else if (xdst == NULL) {
3151 				num_xfrms = 0;
3152 				drop_pols = num_pols;
3153 				goto no_transform;
3154 			}
3155 
3156 			route = xdst->route;
3157 		}
3158 	}
3159 
3160 	if (xdst == NULL) {
3161 		struct xfrm_flo xflo;
3162 
3163 		xflo.dst_orig = dst_orig;
3164 		xflo.flags = flags;
3165 
3166 		/* To accelerate a bit...  */
3167 		if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3168 			       !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3169 			goto nopol;
3170 
3171 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3172 		if (xdst == NULL)
3173 			goto nopol;
3174 		if (IS_ERR(xdst)) {
3175 			err = PTR_ERR(xdst);
3176 			goto dropdst;
3177 		}
3178 
3179 		num_pols = xdst->num_pols;
3180 		num_xfrms = xdst->num_xfrms;
3181 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3182 		route = xdst->route;
3183 	}
3184 
3185 	dst = &xdst->u.dst;
3186 	if (route == NULL && num_xfrms > 0) {
3187 		/* The only case when xfrm_bundle_lookup() returns a
3188 		 * bundle with null route, is when the template could
3189 		 * not be resolved. It means policies are there, but
3190 		 * bundle could not be created, since we don't yet
3191 		 * have the xfrm_state's. We need to wait for KM to
3192 		 * negotiate new SA's or bail out with error.*/
3193 		if (net->xfrm.sysctl_larval_drop) {
3194 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3195 			err = -EREMOTE;
3196 			goto error;
3197 		}
3198 
3199 		err = -EAGAIN;
3200 
3201 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3202 		goto error;
3203 	}
3204 
3205 no_transform:
3206 	if (num_pols == 0)
3207 		goto nopol;
3208 
3209 	if ((flags & XFRM_LOOKUP_ICMP) &&
3210 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3211 		err = -ENOENT;
3212 		goto error;
3213 	}
3214 
3215 	for (i = 0; i < num_pols; i++)
3216 		pols[i]->curlft.use_time = ktime_get_real_seconds();
3217 
3218 	if (num_xfrms < 0) {
3219 		/* Prohibit the flow */
3220 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3221 		err = -EPERM;
3222 		goto error;
3223 	} else if (num_xfrms > 0) {
3224 		/* Flow transformed */
3225 		dst_release(dst_orig);
3226 	} else {
3227 		/* Flow passes untransformed */
3228 		dst_release(dst);
3229 		dst = dst_orig;
3230 	}
3231 ok:
3232 	xfrm_pols_put(pols, drop_pols);
3233 	if (dst && dst->xfrm &&
3234 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3235 		dst->flags |= DST_XFRM_TUNNEL;
3236 	return dst;
3237 
3238 nopol:
3239 	if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3240 	    net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3241 		err = -EPERM;
3242 		goto error;
3243 	}
3244 	if (!(flags & XFRM_LOOKUP_ICMP)) {
3245 		dst = dst_orig;
3246 		goto ok;
3247 	}
3248 	err = -ENOENT;
3249 error:
3250 	dst_release(dst);
3251 dropdst:
3252 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3253 		dst_release(dst_orig);
3254 	xfrm_pols_put(pols, drop_pols);
3255 	return ERR_PTR(err);
3256 }
3257 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3258 
3259 /* Main function: finds/creates a bundle for given flow.
3260  *
3261  * At the moment we eat a raw IP route. Mostly to speed up lookups
3262  * on interfaces with disabled IPsec.
3263  */
3264 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3265 			      const struct flowi *fl, const struct sock *sk,
3266 			      int flags)
3267 {
3268 	return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3269 }
3270 EXPORT_SYMBOL(xfrm_lookup);
3271 
3272 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3273  * Otherwise we may send out blackholed packets.
3274  */
3275 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3276 				    const struct flowi *fl,
3277 				    const struct sock *sk, int flags)
3278 {
3279 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3280 					    flags | XFRM_LOOKUP_QUEUE |
3281 					    XFRM_LOOKUP_KEEP_DST_REF);
3282 
3283 	if (PTR_ERR(dst) == -EREMOTE)
3284 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
3285 
3286 	if (IS_ERR(dst))
3287 		dst_release(dst_orig);
3288 
3289 	return dst;
3290 }
3291 EXPORT_SYMBOL(xfrm_lookup_route);
3292 
3293 static inline int
3294 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3295 {
3296 	struct sec_path *sp = skb_sec_path(skb);
3297 	struct xfrm_state *x;
3298 
3299 	if (!sp || idx < 0 || idx >= sp->len)
3300 		return 0;
3301 	x = sp->xvec[idx];
3302 	if (!x->type->reject)
3303 		return 0;
3304 	return x->type->reject(x, skb, fl);
3305 }
3306 
3307 /* When skb is transformed back to its "native" form, we have to
3308  * check policy restrictions. At the moment we make this in maximally
3309  * stupid way. Shame on me. :-) Of course, connected sockets must
3310  * have policy cached at them.
3311  */
3312 
3313 static inline int
3314 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3315 	      unsigned short family, u32 if_id)
3316 {
3317 	if (xfrm_state_kern(x))
3318 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3319 	return	x->id.proto == tmpl->id.proto &&
3320 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3321 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3322 		x->props.mode == tmpl->mode &&
3323 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3324 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3325 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
3326 		  xfrm_state_addr_cmp(tmpl, x, family)) &&
3327 		(if_id == 0 || if_id == x->if_id);
3328 }
3329 
3330 /*
3331  * 0 or more than 0 is returned when validation is succeeded (either bypass
3332  * because of optional transport mode, or next index of the matched secpath
3333  * state with the template.
3334  * -1 is returned when no matching template is found.
3335  * Otherwise "-2 - errored_index" is returned.
3336  */
3337 static inline int
3338 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3339 	       unsigned short family, u32 if_id)
3340 {
3341 	int idx = start;
3342 
3343 	if (tmpl->optional) {
3344 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
3345 			return start;
3346 	} else
3347 		start = -1;
3348 	for (; idx < sp->len; idx++) {
3349 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3350 			return ++idx;
3351 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3352 			if (start == -1)
3353 				start = -2-idx;
3354 			break;
3355 		}
3356 	}
3357 	return start;
3358 }
3359 
3360 static void
3361 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3362 {
3363 	const struct iphdr *iph = ip_hdr(skb);
3364 	int ihl = iph->ihl;
3365 	u8 *xprth = skb_network_header(skb) + ihl * 4;
3366 	struct flowi4 *fl4 = &fl->u.ip4;
3367 	int oif = 0;
3368 
3369 	if (skb_dst(skb) && skb_dst(skb)->dev)
3370 		oif = skb_dst(skb)->dev->ifindex;
3371 
3372 	memset(fl4, 0, sizeof(struct flowi4));
3373 	fl4->flowi4_mark = skb->mark;
3374 	fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3375 
3376 	fl4->flowi4_proto = iph->protocol;
3377 	fl4->daddr = reverse ? iph->saddr : iph->daddr;
3378 	fl4->saddr = reverse ? iph->daddr : iph->saddr;
3379 	fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3380 
3381 	if (!ip_is_fragment(iph)) {
3382 		switch (iph->protocol) {
3383 		case IPPROTO_UDP:
3384 		case IPPROTO_UDPLITE:
3385 		case IPPROTO_TCP:
3386 		case IPPROTO_SCTP:
3387 		case IPPROTO_DCCP:
3388 			if (xprth + 4 < skb->data ||
3389 			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3390 				__be16 *ports;
3391 
3392 				xprth = skb_network_header(skb) + ihl * 4;
3393 				ports = (__be16 *)xprth;
3394 
3395 				fl4->fl4_sport = ports[!!reverse];
3396 				fl4->fl4_dport = ports[!reverse];
3397 			}
3398 			break;
3399 		case IPPROTO_ICMP:
3400 			if (xprth + 2 < skb->data ||
3401 			    pskb_may_pull(skb, xprth + 2 - skb->data)) {
3402 				u8 *icmp;
3403 
3404 				xprth = skb_network_header(skb) + ihl * 4;
3405 				icmp = xprth;
3406 
3407 				fl4->fl4_icmp_type = icmp[0];
3408 				fl4->fl4_icmp_code = icmp[1];
3409 			}
3410 			break;
3411 		case IPPROTO_GRE:
3412 			if (xprth + 12 < skb->data ||
3413 			    pskb_may_pull(skb, xprth + 12 - skb->data)) {
3414 				__be16 *greflags;
3415 				__be32 *gre_hdr;
3416 
3417 				xprth = skb_network_header(skb) + ihl * 4;
3418 				greflags = (__be16 *)xprth;
3419 				gre_hdr = (__be32 *)xprth;
3420 
3421 				if (greflags[0] & GRE_KEY) {
3422 					if (greflags[0] & GRE_CSUM)
3423 						gre_hdr++;
3424 					fl4->fl4_gre_key = gre_hdr[1];
3425 				}
3426 			}
3427 			break;
3428 		default:
3429 			break;
3430 		}
3431 	}
3432 }
3433 
3434 #if IS_ENABLED(CONFIG_IPV6)
3435 static void
3436 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3437 {
3438 	struct flowi6 *fl6 = &fl->u.ip6;
3439 	int onlyproto = 0;
3440 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
3441 	u32 offset = sizeof(*hdr);
3442 	struct ipv6_opt_hdr *exthdr;
3443 	const unsigned char *nh = skb_network_header(skb);
3444 	u16 nhoff = IP6CB(skb)->nhoff;
3445 	int oif = 0;
3446 	u8 nexthdr;
3447 
3448 	if (!nhoff)
3449 		nhoff = offsetof(struct ipv6hdr, nexthdr);
3450 
3451 	nexthdr = nh[nhoff];
3452 
3453 	if (skb_dst(skb) && skb_dst(skb)->dev)
3454 		oif = skb_dst(skb)->dev->ifindex;
3455 
3456 	memset(fl6, 0, sizeof(struct flowi6));
3457 	fl6->flowi6_mark = skb->mark;
3458 	fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3459 
3460 	fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3461 	fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3462 
3463 	while (nh + offset + sizeof(*exthdr) < skb->data ||
3464 	       pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3465 		nh = skb_network_header(skb);
3466 		exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3467 
3468 		switch (nexthdr) {
3469 		case NEXTHDR_FRAGMENT:
3470 			onlyproto = 1;
3471 			fallthrough;
3472 		case NEXTHDR_ROUTING:
3473 		case NEXTHDR_HOP:
3474 		case NEXTHDR_DEST:
3475 			offset += ipv6_optlen(exthdr);
3476 			nexthdr = exthdr->nexthdr;
3477 			break;
3478 		case IPPROTO_UDP:
3479 		case IPPROTO_UDPLITE:
3480 		case IPPROTO_TCP:
3481 		case IPPROTO_SCTP:
3482 		case IPPROTO_DCCP:
3483 			if (!onlyproto && (nh + offset + 4 < skb->data ||
3484 			     pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3485 				__be16 *ports;
3486 
3487 				nh = skb_network_header(skb);
3488 				ports = (__be16 *)(nh + offset);
3489 				fl6->fl6_sport = ports[!!reverse];
3490 				fl6->fl6_dport = ports[!reverse];
3491 			}
3492 			fl6->flowi6_proto = nexthdr;
3493 			return;
3494 		case IPPROTO_ICMPV6:
3495 			if (!onlyproto && (nh + offset + 2 < skb->data ||
3496 			    pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3497 				u8 *icmp;
3498 
3499 				nh = skb_network_header(skb);
3500 				icmp = (u8 *)(nh + offset);
3501 				fl6->fl6_icmp_type = icmp[0];
3502 				fl6->fl6_icmp_code = icmp[1];
3503 			}
3504 			fl6->flowi6_proto = nexthdr;
3505 			return;
3506 		case IPPROTO_GRE:
3507 			if (!onlyproto &&
3508 			    (nh + offset + 12 < skb->data ||
3509 			     pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3510 				struct gre_base_hdr *gre_hdr;
3511 				__be32 *gre_key;
3512 
3513 				nh = skb_network_header(skb);
3514 				gre_hdr = (struct gre_base_hdr *)(nh + offset);
3515 				gre_key = (__be32 *)(gre_hdr + 1);
3516 
3517 				if (gre_hdr->flags & GRE_KEY) {
3518 					if (gre_hdr->flags & GRE_CSUM)
3519 						gre_key++;
3520 					fl6->fl6_gre_key = *gre_key;
3521 				}
3522 			}
3523 			fl6->flowi6_proto = nexthdr;
3524 			return;
3525 
3526 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3527 		case IPPROTO_MH:
3528 			offset += ipv6_optlen(exthdr);
3529 			if (!onlyproto && (nh + offset + 3 < skb->data ||
3530 			    pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3531 				struct ip6_mh *mh;
3532 
3533 				nh = skb_network_header(skb);
3534 				mh = (struct ip6_mh *)(nh + offset);
3535 				fl6->fl6_mh_type = mh->ip6mh_type;
3536 			}
3537 			fl6->flowi6_proto = nexthdr;
3538 			return;
3539 #endif
3540 		default:
3541 			fl6->flowi6_proto = nexthdr;
3542 			return;
3543 		}
3544 	}
3545 }
3546 #endif
3547 
3548 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3549 			  unsigned int family, int reverse)
3550 {
3551 	switch (family) {
3552 	case AF_INET:
3553 		decode_session4(skb, fl, reverse);
3554 		break;
3555 #if IS_ENABLED(CONFIG_IPV6)
3556 	case AF_INET6:
3557 		decode_session6(skb, fl, reverse);
3558 		break;
3559 #endif
3560 	default:
3561 		return -EAFNOSUPPORT;
3562 	}
3563 
3564 	return security_xfrm_decode_session(skb, &fl->flowi_secid);
3565 }
3566 EXPORT_SYMBOL(__xfrm_decode_session);
3567 
3568 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3569 {
3570 	for (; k < sp->len; k++) {
3571 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3572 			*idxp = k;
3573 			return 1;
3574 		}
3575 	}
3576 
3577 	return 0;
3578 }
3579 
3580 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3581 			unsigned short family)
3582 {
3583 	struct net *net = dev_net(skb->dev);
3584 	struct xfrm_policy *pol;
3585 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3586 	int npols = 0;
3587 	int xfrm_nr;
3588 	int pi;
3589 	int reverse;
3590 	struct flowi fl;
3591 	int xerr_idx = -1;
3592 	const struct xfrm_if_cb *ifcb;
3593 	struct sec_path *sp;
3594 	u32 if_id = 0;
3595 
3596 	rcu_read_lock();
3597 	ifcb = xfrm_if_get_cb();
3598 
3599 	if (ifcb) {
3600 		struct xfrm_if_decode_session_result r;
3601 
3602 		if (ifcb->decode_session(skb, family, &r)) {
3603 			if_id = r.if_id;
3604 			net = r.net;
3605 		}
3606 	}
3607 	rcu_read_unlock();
3608 
3609 	reverse = dir & ~XFRM_POLICY_MASK;
3610 	dir &= XFRM_POLICY_MASK;
3611 
3612 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3613 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3614 		return 0;
3615 	}
3616 
3617 	nf_nat_decode_session(skb, &fl, family);
3618 
3619 	/* First, check used SA against their selectors. */
3620 	sp = skb_sec_path(skb);
3621 	if (sp) {
3622 		int i;
3623 
3624 		for (i = sp->len - 1; i >= 0; i--) {
3625 			struct xfrm_state *x = sp->xvec[i];
3626 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
3627 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3628 				return 0;
3629 			}
3630 		}
3631 	}
3632 
3633 	pol = NULL;
3634 	sk = sk_to_full_sk(sk);
3635 	if (sk && sk->sk_policy[dir]) {
3636 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3637 		if (IS_ERR(pol)) {
3638 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3639 			return 0;
3640 		}
3641 	}
3642 
3643 	if (!pol)
3644 		pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3645 
3646 	if (IS_ERR(pol)) {
3647 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3648 		return 0;
3649 	}
3650 
3651 	if (!pol) {
3652 		if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3653 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3654 			return 0;
3655 		}
3656 
3657 		if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3658 			xfrm_secpath_reject(xerr_idx, skb, &fl);
3659 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3660 			return 0;
3661 		}
3662 		return 1;
3663 	}
3664 
3665 	/* This lockless write can happen from different cpus. */
3666 	WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3667 
3668 	pols[0] = pol;
3669 	npols++;
3670 #ifdef CONFIG_XFRM_SUB_POLICY
3671 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3672 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3673 						    &fl, family,
3674 						    XFRM_POLICY_IN, if_id);
3675 		if (pols[1]) {
3676 			if (IS_ERR(pols[1])) {
3677 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3678 				xfrm_pol_put(pols[0]);
3679 				return 0;
3680 			}
3681 			/* This write can happen from different cpus. */
3682 			WRITE_ONCE(pols[1]->curlft.use_time,
3683 				   ktime_get_real_seconds());
3684 			npols++;
3685 		}
3686 	}
3687 #endif
3688 
3689 	if (pol->action == XFRM_POLICY_ALLOW) {
3690 		static struct sec_path dummy;
3691 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3692 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3693 		struct xfrm_tmpl **tpp = tp;
3694 		int ti = 0;
3695 		int i, k;
3696 
3697 		sp = skb_sec_path(skb);
3698 		if (!sp)
3699 			sp = &dummy;
3700 
3701 		for (pi = 0; pi < npols; pi++) {
3702 			if (pols[pi] != pol &&
3703 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
3704 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3705 				goto reject;
3706 			}
3707 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3708 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3709 				goto reject_error;
3710 			}
3711 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
3712 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
3713 		}
3714 		xfrm_nr = ti;
3715 
3716 		if (npols > 1) {
3717 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3718 			tpp = stp;
3719 		}
3720 
3721 		/* For each tunnel xfrm, find the first matching tmpl.
3722 		 * For each tmpl before that, find corresponding xfrm.
3723 		 * Order is _important_. Later we will implement
3724 		 * some barriers, but at the moment barriers
3725 		 * are implied between each two transformations.
3726 		 */
3727 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3728 			k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3729 			if (k < 0) {
3730 				if (k < -1)
3731 					/* "-2 - errored_index" returned */
3732 					xerr_idx = -(2+k);
3733 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3734 				goto reject;
3735 			}
3736 		}
3737 
3738 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3739 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3740 			goto reject;
3741 		}
3742 
3743 		xfrm_pols_put(pols, npols);
3744 		return 1;
3745 	}
3746 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3747 
3748 reject:
3749 	xfrm_secpath_reject(xerr_idx, skb, &fl);
3750 reject_error:
3751 	xfrm_pols_put(pols, npols);
3752 	return 0;
3753 }
3754 EXPORT_SYMBOL(__xfrm_policy_check);
3755 
3756 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3757 {
3758 	struct net *net = dev_net(skb->dev);
3759 	struct flowi fl;
3760 	struct dst_entry *dst;
3761 	int res = 1;
3762 
3763 	if (xfrm_decode_session(skb, &fl, family) < 0) {
3764 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3765 		return 0;
3766 	}
3767 
3768 	skb_dst_force(skb);
3769 	if (!skb_dst(skb)) {
3770 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3771 		return 0;
3772 	}
3773 
3774 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3775 	if (IS_ERR(dst)) {
3776 		res = 0;
3777 		dst = NULL;
3778 	}
3779 	skb_dst_set(skb, dst);
3780 	return res;
3781 }
3782 EXPORT_SYMBOL(__xfrm_route_forward);
3783 
3784 /* Optimize later using cookies and generation ids. */
3785 
3786 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3787 {
3788 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3789 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3790 	 * get validated by dst_ops->check on every use.  We do this
3791 	 * because when a normal route referenced by an XFRM dst is
3792 	 * obsoleted we do not go looking around for all parent
3793 	 * referencing XFRM dsts so that we can invalidate them.  It
3794 	 * is just too much work.  Instead we make the checks here on
3795 	 * every use.  For example:
3796 	 *
3797 	 *	XFRM dst A --> IPv4 dst X
3798 	 *
3799 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
3800 	 * in this example).  If X is marked obsolete, "A" will not
3801 	 * notice.  That's what we are validating here via the
3802 	 * stale_bundle() check.
3803 	 *
3804 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3805 	 * be marked on it.
3806 	 * This will force stale_bundle() to fail on any xdst bundle with
3807 	 * this dst linked in it.
3808 	 */
3809 	if (dst->obsolete < 0 && !stale_bundle(dst))
3810 		return dst;
3811 
3812 	return NULL;
3813 }
3814 
3815 static int stale_bundle(struct dst_entry *dst)
3816 {
3817 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3818 }
3819 
3820 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3821 {
3822 	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3823 		dst->dev = blackhole_netdev;
3824 		dev_hold(dst->dev);
3825 		dev_put(dev);
3826 	}
3827 }
3828 EXPORT_SYMBOL(xfrm_dst_ifdown);
3829 
3830 static void xfrm_link_failure(struct sk_buff *skb)
3831 {
3832 	/* Impossible. Such dst must be popped before reaches point of failure. */
3833 }
3834 
3835 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3836 {
3837 	if (dst) {
3838 		if (dst->obsolete) {
3839 			dst_release(dst);
3840 			dst = NULL;
3841 		}
3842 	}
3843 	return dst;
3844 }
3845 
3846 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3847 {
3848 	while (nr--) {
3849 		struct xfrm_dst *xdst = bundle[nr];
3850 		u32 pmtu, route_mtu_cached;
3851 		struct dst_entry *dst;
3852 
3853 		dst = &xdst->u.dst;
3854 		pmtu = dst_mtu(xfrm_dst_child(dst));
3855 		xdst->child_mtu_cached = pmtu;
3856 
3857 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3858 
3859 		route_mtu_cached = dst_mtu(xdst->route);
3860 		xdst->route_mtu_cached = route_mtu_cached;
3861 
3862 		if (pmtu > route_mtu_cached)
3863 			pmtu = route_mtu_cached;
3864 
3865 		dst_metric_set(dst, RTAX_MTU, pmtu);
3866 	}
3867 }
3868 
3869 /* Check that the bundle accepts the flow and its components are
3870  * still valid.
3871  */
3872 
3873 static int xfrm_bundle_ok(struct xfrm_dst *first)
3874 {
3875 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3876 	struct dst_entry *dst = &first->u.dst;
3877 	struct xfrm_dst *xdst;
3878 	int start_from, nr;
3879 	u32 mtu;
3880 
3881 	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3882 	    (dst->dev && !netif_running(dst->dev)))
3883 		return 0;
3884 
3885 	if (dst->flags & DST_XFRM_QUEUE)
3886 		return 1;
3887 
3888 	start_from = nr = 0;
3889 	do {
3890 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3891 
3892 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
3893 			return 0;
3894 		if (xdst->xfrm_genid != dst->xfrm->genid)
3895 			return 0;
3896 		if (xdst->num_pols > 0 &&
3897 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3898 			return 0;
3899 
3900 		bundle[nr++] = xdst;
3901 
3902 		mtu = dst_mtu(xfrm_dst_child(dst));
3903 		if (xdst->child_mtu_cached != mtu) {
3904 			start_from = nr;
3905 			xdst->child_mtu_cached = mtu;
3906 		}
3907 
3908 		if (!dst_check(xdst->route, xdst->route_cookie))
3909 			return 0;
3910 		mtu = dst_mtu(xdst->route);
3911 		if (xdst->route_mtu_cached != mtu) {
3912 			start_from = nr;
3913 			xdst->route_mtu_cached = mtu;
3914 		}
3915 
3916 		dst = xfrm_dst_child(dst);
3917 	} while (dst->xfrm);
3918 
3919 	if (likely(!start_from))
3920 		return 1;
3921 
3922 	xdst = bundle[start_from - 1];
3923 	mtu = xdst->child_mtu_cached;
3924 	while (start_from--) {
3925 		dst = &xdst->u.dst;
3926 
3927 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
3928 		if (mtu > xdst->route_mtu_cached)
3929 			mtu = xdst->route_mtu_cached;
3930 		dst_metric_set(dst, RTAX_MTU, mtu);
3931 		if (!start_from)
3932 			break;
3933 
3934 		xdst = bundle[start_from - 1];
3935 		xdst->child_mtu_cached = mtu;
3936 	}
3937 
3938 	return 1;
3939 }
3940 
3941 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3942 {
3943 	return dst_metric_advmss(xfrm_dst_path(dst));
3944 }
3945 
3946 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3947 {
3948 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3949 
3950 	return mtu ? : dst_mtu(xfrm_dst_path(dst));
3951 }
3952 
3953 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3954 					const void *daddr)
3955 {
3956 	while (dst->xfrm) {
3957 		const struct xfrm_state *xfrm = dst->xfrm;
3958 
3959 		dst = xfrm_dst_child(dst);
3960 
3961 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3962 			continue;
3963 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3964 			daddr = xfrm->coaddr;
3965 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3966 			daddr = &xfrm->id.daddr;
3967 	}
3968 	return daddr;
3969 }
3970 
3971 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3972 					   struct sk_buff *skb,
3973 					   const void *daddr)
3974 {
3975 	const struct dst_entry *path = xfrm_dst_path(dst);
3976 
3977 	if (!skb)
3978 		daddr = xfrm_get_dst_nexthop(dst, daddr);
3979 	return path->ops->neigh_lookup(path, skb, daddr);
3980 }
3981 
3982 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3983 {
3984 	const struct dst_entry *path = xfrm_dst_path(dst);
3985 
3986 	daddr = xfrm_get_dst_nexthop(dst, daddr);
3987 	path->ops->confirm_neigh(path, daddr);
3988 }
3989 
3990 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3991 {
3992 	int err = 0;
3993 
3994 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3995 		return -EAFNOSUPPORT;
3996 
3997 	spin_lock(&xfrm_policy_afinfo_lock);
3998 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
3999 		err = -EEXIST;
4000 	else {
4001 		struct dst_ops *dst_ops = afinfo->dst_ops;
4002 		if (likely(dst_ops->kmem_cachep == NULL))
4003 			dst_ops->kmem_cachep = xfrm_dst_cache;
4004 		if (likely(dst_ops->check == NULL))
4005 			dst_ops->check = xfrm_dst_check;
4006 		if (likely(dst_ops->default_advmss == NULL))
4007 			dst_ops->default_advmss = xfrm_default_advmss;
4008 		if (likely(dst_ops->mtu == NULL))
4009 			dst_ops->mtu = xfrm_mtu;
4010 		if (likely(dst_ops->negative_advice == NULL))
4011 			dst_ops->negative_advice = xfrm_negative_advice;
4012 		if (likely(dst_ops->link_failure == NULL))
4013 			dst_ops->link_failure = xfrm_link_failure;
4014 		if (likely(dst_ops->neigh_lookup == NULL))
4015 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
4016 		if (likely(!dst_ops->confirm_neigh))
4017 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
4018 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
4019 	}
4020 	spin_unlock(&xfrm_policy_afinfo_lock);
4021 
4022 	return err;
4023 }
4024 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
4025 
4026 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
4027 {
4028 	struct dst_ops *dst_ops = afinfo->dst_ops;
4029 	int i;
4030 
4031 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
4032 		if (xfrm_policy_afinfo[i] != afinfo)
4033 			continue;
4034 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
4035 		break;
4036 	}
4037 
4038 	synchronize_rcu();
4039 
4040 	dst_ops->kmem_cachep = NULL;
4041 	dst_ops->check = NULL;
4042 	dst_ops->negative_advice = NULL;
4043 	dst_ops->link_failure = NULL;
4044 }
4045 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
4046 
4047 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
4048 {
4049 	spin_lock(&xfrm_if_cb_lock);
4050 	rcu_assign_pointer(xfrm_if_cb, ifcb);
4051 	spin_unlock(&xfrm_if_cb_lock);
4052 }
4053 EXPORT_SYMBOL(xfrm_if_register_cb);
4054 
4055 void xfrm_if_unregister_cb(void)
4056 {
4057 	RCU_INIT_POINTER(xfrm_if_cb, NULL);
4058 	synchronize_rcu();
4059 }
4060 EXPORT_SYMBOL(xfrm_if_unregister_cb);
4061 
4062 #ifdef CONFIG_XFRM_STATISTICS
4063 static int __net_init xfrm_statistics_init(struct net *net)
4064 {
4065 	int rv;
4066 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4067 	if (!net->mib.xfrm_statistics)
4068 		return -ENOMEM;
4069 	rv = xfrm_proc_init(net);
4070 	if (rv < 0)
4071 		free_percpu(net->mib.xfrm_statistics);
4072 	return rv;
4073 }
4074 
4075 static void xfrm_statistics_fini(struct net *net)
4076 {
4077 	xfrm_proc_fini(net);
4078 	free_percpu(net->mib.xfrm_statistics);
4079 }
4080 #else
4081 static int __net_init xfrm_statistics_init(struct net *net)
4082 {
4083 	return 0;
4084 }
4085 
4086 static void xfrm_statistics_fini(struct net *net)
4087 {
4088 }
4089 #endif
4090 
4091 static int __net_init xfrm_policy_init(struct net *net)
4092 {
4093 	unsigned int hmask, sz;
4094 	int dir, err;
4095 
4096 	if (net_eq(net, &init_net)) {
4097 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4098 					   sizeof(struct xfrm_dst),
4099 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4100 					   NULL);
4101 		err = rhashtable_init(&xfrm_policy_inexact_table,
4102 				      &xfrm_pol_inexact_params);
4103 		BUG_ON(err);
4104 	}
4105 
4106 	hmask = 8 - 1;
4107 	sz = (hmask+1) * sizeof(struct hlist_head);
4108 
4109 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4110 	if (!net->xfrm.policy_byidx)
4111 		goto out_byidx;
4112 	net->xfrm.policy_idx_hmask = hmask;
4113 
4114 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4115 		struct xfrm_policy_hash *htab;
4116 
4117 		net->xfrm.policy_count[dir] = 0;
4118 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4119 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4120 
4121 		htab = &net->xfrm.policy_bydst[dir];
4122 		htab->table = xfrm_hash_alloc(sz);
4123 		if (!htab->table)
4124 			goto out_bydst;
4125 		htab->hmask = hmask;
4126 		htab->dbits4 = 32;
4127 		htab->sbits4 = 32;
4128 		htab->dbits6 = 128;
4129 		htab->sbits6 = 128;
4130 	}
4131 	net->xfrm.policy_hthresh.lbits4 = 32;
4132 	net->xfrm.policy_hthresh.rbits4 = 32;
4133 	net->xfrm.policy_hthresh.lbits6 = 128;
4134 	net->xfrm.policy_hthresh.rbits6 = 128;
4135 
4136 	seqlock_init(&net->xfrm.policy_hthresh.lock);
4137 
4138 	INIT_LIST_HEAD(&net->xfrm.policy_all);
4139 	INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4140 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4141 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4142 	return 0;
4143 
4144 out_bydst:
4145 	for (dir--; dir >= 0; dir--) {
4146 		struct xfrm_policy_hash *htab;
4147 
4148 		htab = &net->xfrm.policy_bydst[dir];
4149 		xfrm_hash_free(htab->table, sz);
4150 	}
4151 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4152 out_byidx:
4153 	return -ENOMEM;
4154 }
4155 
4156 static void xfrm_policy_fini(struct net *net)
4157 {
4158 	struct xfrm_pol_inexact_bin *b, *t;
4159 	unsigned int sz;
4160 	int dir;
4161 
4162 	flush_work(&net->xfrm.policy_hash_work);
4163 #ifdef CONFIG_XFRM_SUB_POLICY
4164 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4165 #endif
4166 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4167 
4168 	WARN_ON(!list_empty(&net->xfrm.policy_all));
4169 
4170 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4171 		struct xfrm_policy_hash *htab;
4172 
4173 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4174 
4175 		htab = &net->xfrm.policy_bydst[dir];
4176 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4177 		WARN_ON(!hlist_empty(htab->table));
4178 		xfrm_hash_free(htab->table, sz);
4179 	}
4180 
4181 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4182 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4183 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4184 
4185 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4186 	list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4187 		__xfrm_policy_inexact_prune_bin(b, true);
4188 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4189 }
4190 
4191 static int __net_init xfrm_net_init(struct net *net)
4192 {
4193 	int rv;
4194 
4195 	/* Initialize the per-net locks here */
4196 	spin_lock_init(&net->xfrm.xfrm_state_lock);
4197 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
4198 	seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4199 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
4200 	net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4201 	net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4202 	net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4203 
4204 	rv = xfrm_statistics_init(net);
4205 	if (rv < 0)
4206 		goto out_statistics;
4207 	rv = xfrm_state_init(net);
4208 	if (rv < 0)
4209 		goto out_state;
4210 	rv = xfrm_policy_init(net);
4211 	if (rv < 0)
4212 		goto out_policy;
4213 	rv = xfrm_sysctl_init(net);
4214 	if (rv < 0)
4215 		goto out_sysctl;
4216 
4217 	return 0;
4218 
4219 out_sysctl:
4220 	xfrm_policy_fini(net);
4221 out_policy:
4222 	xfrm_state_fini(net);
4223 out_state:
4224 	xfrm_statistics_fini(net);
4225 out_statistics:
4226 	return rv;
4227 }
4228 
4229 static void __net_exit xfrm_net_exit(struct net *net)
4230 {
4231 	xfrm_sysctl_fini(net);
4232 	xfrm_policy_fini(net);
4233 	xfrm_state_fini(net);
4234 	xfrm_statistics_fini(net);
4235 }
4236 
4237 static struct pernet_operations __net_initdata xfrm_net_ops = {
4238 	.init = xfrm_net_init,
4239 	.exit = xfrm_net_exit,
4240 };
4241 
4242 void __init xfrm_init(void)
4243 {
4244 	register_pernet_subsys(&xfrm_net_ops);
4245 	xfrm_dev_init();
4246 	xfrm_input_init();
4247 
4248 #ifdef CONFIG_XFRM_ESPINTCP
4249 	espintcp_init();
4250 #endif
4251 }
4252 
4253 #ifdef CONFIG_AUDITSYSCALL
4254 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4255 					 struct audit_buffer *audit_buf)
4256 {
4257 	struct xfrm_sec_ctx *ctx = xp->security;
4258 	struct xfrm_selector *sel = &xp->selector;
4259 
4260 	if (ctx)
4261 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4262 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4263 
4264 	switch (sel->family) {
4265 	case AF_INET:
4266 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4267 		if (sel->prefixlen_s != 32)
4268 			audit_log_format(audit_buf, " src_prefixlen=%d",
4269 					 sel->prefixlen_s);
4270 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4271 		if (sel->prefixlen_d != 32)
4272 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4273 					 sel->prefixlen_d);
4274 		break;
4275 	case AF_INET6:
4276 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4277 		if (sel->prefixlen_s != 128)
4278 			audit_log_format(audit_buf, " src_prefixlen=%d",
4279 					 sel->prefixlen_s);
4280 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4281 		if (sel->prefixlen_d != 128)
4282 			audit_log_format(audit_buf, " dst_prefixlen=%d",
4283 					 sel->prefixlen_d);
4284 		break;
4285 	}
4286 }
4287 
4288 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4289 {
4290 	struct audit_buffer *audit_buf;
4291 
4292 	audit_buf = xfrm_audit_start("SPD-add");
4293 	if (audit_buf == NULL)
4294 		return;
4295 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4296 	audit_log_format(audit_buf, " res=%u", result);
4297 	xfrm_audit_common_policyinfo(xp, audit_buf);
4298 	audit_log_end(audit_buf);
4299 }
4300 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4301 
4302 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4303 			      bool task_valid)
4304 {
4305 	struct audit_buffer *audit_buf;
4306 
4307 	audit_buf = xfrm_audit_start("SPD-delete");
4308 	if (audit_buf == NULL)
4309 		return;
4310 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4311 	audit_log_format(audit_buf, " res=%u", result);
4312 	xfrm_audit_common_policyinfo(xp, audit_buf);
4313 	audit_log_end(audit_buf);
4314 }
4315 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4316 #endif
4317 
4318 #ifdef CONFIG_XFRM_MIGRATE
4319 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4320 					const struct xfrm_selector *sel_tgt)
4321 {
4322 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4323 		if (sel_tgt->family == sel_cmp->family &&
4324 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4325 				    sel_cmp->family) &&
4326 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4327 				    sel_cmp->family) &&
4328 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4329 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4330 			return true;
4331 		}
4332 	} else {
4333 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4334 			return true;
4335 		}
4336 	}
4337 	return false;
4338 }
4339 
4340 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4341 						    u8 dir, u8 type, struct net *net, u32 if_id)
4342 {
4343 	struct xfrm_policy *pol, *ret = NULL;
4344 	struct hlist_head *chain;
4345 	u32 priority = ~0U;
4346 
4347 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4348 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4349 	hlist_for_each_entry(pol, chain, bydst) {
4350 		if ((if_id == 0 || pol->if_id == if_id) &&
4351 		    xfrm_migrate_selector_match(sel, &pol->selector) &&
4352 		    pol->type == type) {
4353 			ret = pol;
4354 			priority = ret->priority;
4355 			break;
4356 		}
4357 	}
4358 	chain = &net->xfrm.policy_inexact[dir];
4359 	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4360 		if ((pol->priority >= priority) && ret)
4361 			break;
4362 
4363 		if ((if_id == 0 || pol->if_id == if_id) &&
4364 		    xfrm_migrate_selector_match(sel, &pol->selector) &&
4365 		    pol->type == type) {
4366 			ret = pol;
4367 			break;
4368 		}
4369 	}
4370 
4371 	xfrm_pol_hold(ret);
4372 
4373 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4374 
4375 	return ret;
4376 }
4377 
4378 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4379 {
4380 	int match = 0;
4381 
4382 	if (t->mode == m->mode && t->id.proto == m->proto &&
4383 	    (m->reqid == 0 || t->reqid == m->reqid)) {
4384 		switch (t->mode) {
4385 		case XFRM_MODE_TUNNEL:
4386 		case XFRM_MODE_BEET:
4387 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4388 					    m->old_family) &&
4389 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
4390 					    m->old_family)) {
4391 				match = 1;
4392 			}
4393 			break;
4394 		case XFRM_MODE_TRANSPORT:
4395 			/* in case of transport mode, template does not store
4396 			   any IP addresses, hence we just compare mode and
4397 			   protocol */
4398 			match = 1;
4399 			break;
4400 		default:
4401 			break;
4402 		}
4403 	}
4404 	return match;
4405 }
4406 
4407 /* update endpoint address(es) of template(s) */
4408 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4409 			       struct xfrm_migrate *m, int num_migrate,
4410 			       struct netlink_ext_ack *extack)
4411 {
4412 	struct xfrm_migrate *mp;
4413 	int i, j, n = 0;
4414 
4415 	write_lock_bh(&pol->lock);
4416 	if (unlikely(pol->walk.dead)) {
4417 		/* target policy has been deleted */
4418 		NL_SET_ERR_MSG(extack, "Target policy not found");
4419 		write_unlock_bh(&pol->lock);
4420 		return -ENOENT;
4421 	}
4422 
4423 	for (i = 0; i < pol->xfrm_nr; i++) {
4424 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4425 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4426 				continue;
4427 			n++;
4428 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4429 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4430 				continue;
4431 			/* update endpoints */
4432 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4433 			       sizeof(pol->xfrm_vec[i].id.daddr));
4434 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4435 			       sizeof(pol->xfrm_vec[i].saddr));
4436 			pol->xfrm_vec[i].encap_family = mp->new_family;
4437 			/* flush bundles */
4438 			atomic_inc(&pol->genid);
4439 		}
4440 	}
4441 
4442 	write_unlock_bh(&pol->lock);
4443 
4444 	if (!n)
4445 		return -ENODATA;
4446 
4447 	return 0;
4448 }
4449 
4450 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate,
4451 			      struct netlink_ext_ack *extack)
4452 {
4453 	int i, j;
4454 
4455 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) {
4456 		NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)");
4457 		return -EINVAL;
4458 	}
4459 
4460 	for (i = 0; i < num_migrate; i++) {
4461 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4462 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) {
4463 			NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null");
4464 			return -EINVAL;
4465 		}
4466 
4467 		/* check if there is any duplicated entry */
4468 		for (j = i + 1; j < num_migrate; j++) {
4469 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4470 				    sizeof(m[i].old_daddr)) &&
4471 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4472 				    sizeof(m[i].old_saddr)) &&
4473 			    m[i].proto == m[j].proto &&
4474 			    m[i].mode == m[j].mode &&
4475 			    m[i].reqid == m[j].reqid &&
4476 			    m[i].old_family == m[j].old_family) {
4477 				NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique");
4478 				return -EINVAL;
4479 			}
4480 		}
4481 	}
4482 
4483 	return 0;
4484 }
4485 
4486 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4487 		 struct xfrm_migrate *m, int num_migrate,
4488 		 struct xfrm_kmaddress *k, struct net *net,
4489 		 struct xfrm_encap_tmpl *encap, u32 if_id,
4490 		 struct netlink_ext_ack *extack)
4491 {
4492 	int i, err, nx_cur = 0, nx_new = 0;
4493 	struct xfrm_policy *pol = NULL;
4494 	struct xfrm_state *x, *xc;
4495 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4496 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4497 	struct xfrm_migrate *mp;
4498 
4499 	/* Stage 0 - sanity checks */
4500 	err = xfrm_migrate_check(m, num_migrate, extack);
4501 	if (err < 0)
4502 		goto out;
4503 
4504 	if (dir >= XFRM_POLICY_MAX) {
4505 		NL_SET_ERR_MSG(extack, "Invalid policy direction");
4506 		err = -EINVAL;
4507 		goto out;
4508 	}
4509 
4510 	/* Stage 1 - find policy */
4511 	pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id);
4512 	if (!pol) {
4513 		NL_SET_ERR_MSG(extack, "Target policy not found");
4514 		err = -ENOENT;
4515 		goto out;
4516 	}
4517 
4518 	/* Stage 2 - find and update state(s) */
4519 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4520 		if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4521 			x_cur[nx_cur] = x;
4522 			nx_cur++;
4523 			xc = xfrm_state_migrate(x, mp, encap);
4524 			if (xc) {
4525 				x_new[nx_new] = xc;
4526 				nx_new++;
4527 			} else {
4528 				err = -ENODATA;
4529 				goto restore_state;
4530 			}
4531 		}
4532 	}
4533 
4534 	/* Stage 3 - update policy */
4535 	err = xfrm_policy_migrate(pol, m, num_migrate, extack);
4536 	if (err < 0)
4537 		goto restore_state;
4538 
4539 	/* Stage 4 - delete old state(s) */
4540 	if (nx_cur) {
4541 		xfrm_states_put(x_cur, nx_cur);
4542 		xfrm_states_delete(x_cur, nx_cur);
4543 	}
4544 
4545 	/* Stage 5 - announce */
4546 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
4547 
4548 	xfrm_pol_put(pol);
4549 
4550 	return 0;
4551 out:
4552 	return err;
4553 
4554 restore_state:
4555 	if (pol)
4556 		xfrm_pol_put(pol);
4557 	if (nx_cur)
4558 		xfrm_states_put(x_cur, nx_cur);
4559 	if (nx_new)
4560 		xfrm_states_delete(x_new, nx_new);
4561 
4562 	return err;
4563 }
4564 EXPORT_SYMBOL(xfrm_migrate);
4565 #endif
4566