xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 232b0b08)
1 /*
2  * xfrm_policy.c
3  *
4  * Changes:
5  *	Mitsuru KANDA @USAGI
6  * 	Kazunori MIYAZAWA @USAGI
7  * 	Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8  * 		IPv6 support
9  * 	Kazunori MIYAZAWA @USAGI
10  * 	YOSHIFUJI Hideaki
11  * 		Split up af-specific portion
12  *	Derek Atkins <derek@ihtfp.com>		Add the post_input processor
13  *
14  */
15 
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
28 #include <net/dst.h>
29 #include <net/flow.h>
30 #include <net/xfrm.h>
31 #include <net/ip.h>
32 #ifdef CONFIG_XFRM_STATISTICS
33 #include <net/snmp.h>
34 #endif
35 
36 #include "xfrm_hash.h"
37 
38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
40 #define XFRM_MAX_QUEUE_LEN	100
41 
42 struct xfrm_flo {
43 	struct dst_entry *dst_orig;
44 	u8 flags;
45 };
46 
47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
48 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
49 						__read_mostly;
50 
51 static struct kmem_cache *xfrm_dst_cache __read_mostly;
52 static __read_mostly seqcount_t xfrm_policy_hash_generation;
53 
54 static void xfrm_init_pmtu(struct dst_entry *dst);
55 static int stale_bundle(struct dst_entry *dst);
56 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
57 static void xfrm_policy_queue_process(unsigned long arg);
58 
59 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
61 						int dir);
62 
63 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
64 {
65 	return atomic_inc_not_zero(&policy->refcnt);
66 }
67 
68 static inline bool
69 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
70 {
71 	const struct flowi4 *fl4 = &fl->u.ip4;
72 
73 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
74 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
75 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
76 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
77 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
78 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
79 }
80 
81 static inline bool
82 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
83 {
84 	const struct flowi6 *fl6 = &fl->u.ip6;
85 
86 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
87 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
88 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
89 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
90 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
91 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
92 }
93 
94 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
95 			 unsigned short family)
96 {
97 	switch (family) {
98 	case AF_INET:
99 		return __xfrm4_selector_match(sel, fl);
100 	case AF_INET6:
101 		return __xfrm6_selector_match(sel, fl);
102 	}
103 	return false;
104 }
105 
106 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
107 {
108 	const struct xfrm_policy_afinfo *afinfo;
109 
110 	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
111 		return NULL;
112 	rcu_read_lock();
113 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
114 	if (unlikely(!afinfo))
115 		rcu_read_unlock();
116 	return afinfo;
117 }
118 
119 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net,
120 						  int tos, int oif,
121 						  const xfrm_address_t *saddr,
122 						  const xfrm_address_t *daddr,
123 						  int family)
124 {
125 	const struct xfrm_policy_afinfo *afinfo;
126 	struct dst_entry *dst;
127 
128 	afinfo = xfrm_policy_get_afinfo(family);
129 	if (unlikely(afinfo == NULL))
130 		return ERR_PTR(-EAFNOSUPPORT);
131 
132 	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr);
133 
134 	rcu_read_unlock();
135 
136 	return dst;
137 }
138 
139 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
140 						int tos, int oif,
141 						xfrm_address_t *prev_saddr,
142 						xfrm_address_t *prev_daddr,
143 						int family)
144 {
145 	struct net *net = xs_net(x);
146 	xfrm_address_t *saddr = &x->props.saddr;
147 	xfrm_address_t *daddr = &x->id.daddr;
148 	struct dst_entry *dst;
149 
150 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
151 		saddr = x->coaddr;
152 		daddr = prev_daddr;
153 	}
154 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
155 		saddr = prev_saddr;
156 		daddr = x->coaddr;
157 	}
158 
159 	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family);
160 
161 	if (!IS_ERR(dst)) {
162 		if (prev_saddr != saddr)
163 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
164 		if (prev_daddr != daddr)
165 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
166 	}
167 
168 	return dst;
169 }
170 
171 static inline unsigned long make_jiffies(long secs)
172 {
173 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
174 		return MAX_SCHEDULE_TIMEOUT-1;
175 	else
176 		return secs*HZ;
177 }
178 
179 static void xfrm_policy_timer(unsigned long data)
180 {
181 	struct xfrm_policy *xp = (struct xfrm_policy *)data;
182 	unsigned long now = get_seconds();
183 	long next = LONG_MAX;
184 	int warn = 0;
185 	int dir;
186 
187 	read_lock(&xp->lock);
188 
189 	if (unlikely(xp->walk.dead))
190 		goto out;
191 
192 	dir = xfrm_policy_id2dir(xp->index);
193 
194 	if (xp->lft.hard_add_expires_seconds) {
195 		long tmo = xp->lft.hard_add_expires_seconds +
196 			xp->curlft.add_time - now;
197 		if (tmo <= 0)
198 			goto expired;
199 		if (tmo < next)
200 			next = tmo;
201 	}
202 	if (xp->lft.hard_use_expires_seconds) {
203 		long tmo = xp->lft.hard_use_expires_seconds +
204 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
205 		if (tmo <= 0)
206 			goto expired;
207 		if (tmo < next)
208 			next = tmo;
209 	}
210 	if (xp->lft.soft_add_expires_seconds) {
211 		long tmo = xp->lft.soft_add_expires_seconds +
212 			xp->curlft.add_time - now;
213 		if (tmo <= 0) {
214 			warn = 1;
215 			tmo = XFRM_KM_TIMEOUT;
216 		}
217 		if (tmo < next)
218 			next = tmo;
219 	}
220 	if (xp->lft.soft_use_expires_seconds) {
221 		long tmo = xp->lft.soft_use_expires_seconds +
222 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
223 		if (tmo <= 0) {
224 			warn = 1;
225 			tmo = XFRM_KM_TIMEOUT;
226 		}
227 		if (tmo < next)
228 			next = tmo;
229 	}
230 
231 	if (warn)
232 		km_policy_expired(xp, dir, 0, 0);
233 	if (next != LONG_MAX &&
234 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
235 		xfrm_pol_hold(xp);
236 
237 out:
238 	read_unlock(&xp->lock);
239 	xfrm_pol_put(xp);
240 	return;
241 
242 expired:
243 	read_unlock(&xp->lock);
244 	if (!xfrm_policy_delete(xp, dir))
245 		km_policy_expired(xp, dir, 1, 0);
246 	xfrm_pol_put(xp);
247 }
248 
249 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
250 {
251 	struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
252 
253 	if (unlikely(pol->walk.dead))
254 		flo = NULL;
255 	else
256 		xfrm_pol_hold(pol);
257 
258 	return flo;
259 }
260 
261 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
262 {
263 	struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
264 
265 	return !pol->walk.dead;
266 }
267 
268 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
269 {
270 	xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
271 }
272 
273 static const struct flow_cache_ops xfrm_policy_fc_ops = {
274 	.get = xfrm_policy_flo_get,
275 	.check = xfrm_policy_flo_check,
276 	.delete = xfrm_policy_flo_delete,
277 };
278 
279 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
280  * SPD calls.
281  */
282 
283 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
284 {
285 	struct xfrm_policy *policy;
286 
287 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
288 
289 	if (policy) {
290 		write_pnet(&policy->xp_net, net);
291 		INIT_LIST_HEAD(&policy->walk.all);
292 		INIT_HLIST_NODE(&policy->bydst);
293 		INIT_HLIST_NODE(&policy->byidx);
294 		rwlock_init(&policy->lock);
295 		atomic_set(&policy->refcnt, 1);
296 		skb_queue_head_init(&policy->polq.hold_queue);
297 		setup_timer(&policy->timer, xfrm_policy_timer,
298 				(unsigned long)policy);
299 		setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
300 			    (unsigned long)policy);
301 		policy->flo.ops = &xfrm_policy_fc_ops;
302 	}
303 	return policy;
304 }
305 EXPORT_SYMBOL(xfrm_policy_alloc);
306 
307 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
308 {
309 	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
310 
311 	security_xfrm_policy_free(policy->security);
312 	kfree(policy);
313 }
314 
315 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
316 
317 void xfrm_policy_destroy(struct xfrm_policy *policy)
318 {
319 	BUG_ON(!policy->walk.dead);
320 
321 	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
322 		BUG();
323 
324 	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
325 }
326 EXPORT_SYMBOL(xfrm_policy_destroy);
327 
328 /* Rule must be locked. Release descendant resources, announce
329  * entry dead. The rule must be unlinked from lists to the moment.
330  */
331 
332 static void xfrm_policy_kill(struct xfrm_policy *policy)
333 {
334 	policy->walk.dead = 1;
335 
336 	atomic_inc(&policy->genid);
337 
338 	if (del_timer(&policy->polq.hold_timer))
339 		xfrm_pol_put(policy);
340 	skb_queue_purge(&policy->polq.hold_queue);
341 
342 	if (del_timer(&policy->timer))
343 		xfrm_pol_put(policy);
344 
345 	xfrm_pol_put(policy);
346 }
347 
348 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
349 
350 static inline unsigned int idx_hash(struct net *net, u32 index)
351 {
352 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
353 }
354 
355 /* calculate policy hash thresholds */
356 static void __get_hash_thresh(struct net *net,
357 			      unsigned short family, int dir,
358 			      u8 *dbits, u8 *sbits)
359 {
360 	switch (family) {
361 	case AF_INET:
362 		*dbits = net->xfrm.policy_bydst[dir].dbits4;
363 		*sbits = net->xfrm.policy_bydst[dir].sbits4;
364 		break;
365 
366 	case AF_INET6:
367 		*dbits = net->xfrm.policy_bydst[dir].dbits6;
368 		*sbits = net->xfrm.policy_bydst[dir].sbits6;
369 		break;
370 
371 	default:
372 		*dbits = 0;
373 		*sbits = 0;
374 	}
375 }
376 
377 static struct hlist_head *policy_hash_bysel(struct net *net,
378 					    const struct xfrm_selector *sel,
379 					    unsigned short family, int dir)
380 {
381 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
382 	unsigned int hash;
383 	u8 dbits;
384 	u8 sbits;
385 
386 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
387 	hash = __sel_hash(sel, family, hmask, dbits, sbits);
388 
389 	if (hash == hmask + 1)
390 		return &net->xfrm.policy_inexact[dir];
391 
392 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
393 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
394 }
395 
396 static struct hlist_head *policy_hash_direct(struct net *net,
397 					     const xfrm_address_t *daddr,
398 					     const xfrm_address_t *saddr,
399 					     unsigned short family, int dir)
400 {
401 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
402 	unsigned int hash;
403 	u8 dbits;
404 	u8 sbits;
405 
406 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
407 	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
408 
409 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
410 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
411 }
412 
413 static void xfrm_dst_hash_transfer(struct net *net,
414 				   struct hlist_head *list,
415 				   struct hlist_head *ndsttable,
416 				   unsigned int nhashmask,
417 				   int dir)
418 {
419 	struct hlist_node *tmp, *entry0 = NULL;
420 	struct xfrm_policy *pol;
421 	unsigned int h0 = 0;
422 	u8 dbits;
423 	u8 sbits;
424 
425 redo:
426 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
427 		unsigned int h;
428 
429 		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
430 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
431 				pol->family, nhashmask, dbits, sbits);
432 		if (!entry0) {
433 			hlist_del_rcu(&pol->bydst);
434 			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
435 			h0 = h;
436 		} else {
437 			if (h != h0)
438 				continue;
439 			hlist_del_rcu(&pol->bydst);
440 			hlist_add_behind_rcu(&pol->bydst, entry0);
441 		}
442 		entry0 = &pol->bydst;
443 	}
444 	if (!hlist_empty(list)) {
445 		entry0 = NULL;
446 		goto redo;
447 	}
448 }
449 
450 static void xfrm_idx_hash_transfer(struct hlist_head *list,
451 				   struct hlist_head *nidxtable,
452 				   unsigned int nhashmask)
453 {
454 	struct hlist_node *tmp;
455 	struct xfrm_policy *pol;
456 
457 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
458 		unsigned int h;
459 
460 		h = __idx_hash(pol->index, nhashmask);
461 		hlist_add_head(&pol->byidx, nidxtable+h);
462 	}
463 }
464 
465 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
466 {
467 	return ((old_hmask + 1) << 1) - 1;
468 }
469 
470 static void xfrm_bydst_resize(struct net *net, int dir)
471 {
472 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
473 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
474 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
475 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
476 	struct hlist_head *odst;
477 	int i;
478 
479 	if (!ndst)
480 		return;
481 
482 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
483 	write_seqcount_begin(&xfrm_policy_hash_generation);
484 
485 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
486 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
487 
488 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
489 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
490 
491 	for (i = hmask; i >= 0; i--)
492 		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
493 
494 	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
495 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
496 
497 	write_seqcount_end(&xfrm_policy_hash_generation);
498 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
499 
500 	synchronize_rcu();
501 
502 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
503 }
504 
505 static void xfrm_byidx_resize(struct net *net, int total)
506 {
507 	unsigned int hmask = net->xfrm.policy_idx_hmask;
508 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
509 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
510 	struct hlist_head *oidx = net->xfrm.policy_byidx;
511 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
512 	int i;
513 
514 	if (!nidx)
515 		return;
516 
517 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
518 
519 	for (i = hmask; i >= 0; i--)
520 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
521 
522 	net->xfrm.policy_byidx = nidx;
523 	net->xfrm.policy_idx_hmask = nhashmask;
524 
525 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
526 
527 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
528 }
529 
530 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
531 {
532 	unsigned int cnt = net->xfrm.policy_count[dir];
533 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
534 
535 	if (total)
536 		*total += cnt;
537 
538 	if ((hmask + 1) < xfrm_policy_hashmax &&
539 	    cnt > hmask)
540 		return 1;
541 
542 	return 0;
543 }
544 
545 static inline int xfrm_byidx_should_resize(struct net *net, int total)
546 {
547 	unsigned int hmask = net->xfrm.policy_idx_hmask;
548 
549 	if ((hmask + 1) < xfrm_policy_hashmax &&
550 	    total > hmask)
551 		return 1;
552 
553 	return 0;
554 }
555 
556 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
557 {
558 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
559 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
560 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
561 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
562 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
563 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
564 	si->spdhcnt = net->xfrm.policy_idx_hmask;
565 	si->spdhmcnt = xfrm_policy_hashmax;
566 }
567 EXPORT_SYMBOL(xfrm_spd_getinfo);
568 
569 static DEFINE_MUTEX(hash_resize_mutex);
570 static void xfrm_hash_resize(struct work_struct *work)
571 {
572 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
573 	int dir, total;
574 
575 	mutex_lock(&hash_resize_mutex);
576 
577 	total = 0;
578 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
579 		if (xfrm_bydst_should_resize(net, dir, &total))
580 			xfrm_bydst_resize(net, dir);
581 	}
582 	if (xfrm_byidx_should_resize(net, total))
583 		xfrm_byidx_resize(net, total);
584 
585 	mutex_unlock(&hash_resize_mutex);
586 }
587 
588 static void xfrm_hash_rebuild(struct work_struct *work)
589 {
590 	struct net *net = container_of(work, struct net,
591 				       xfrm.policy_hthresh.work);
592 	unsigned int hmask;
593 	struct xfrm_policy *pol;
594 	struct xfrm_policy *policy;
595 	struct hlist_head *chain;
596 	struct hlist_head *odst;
597 	struct hlist_node *newpos;
598 	int i;
599 	int dir;
600 	unsigned seq;
601 	u8 lbits4, rbits4, lbits6, rbits6;
602 
603 	mutex_lock(&hash_resize_mutex);
604 
605 	/* read selector prefixlen thresholds */
606 	do {
607 		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
608 
609 		lbits4 = net->xfrm.policy_hthresh.lbits4;
610 		rbits4 = net->xfrm.policy_hthresh.rbits4;
611 		lbits6 = net->xfrm.policy_hthresh.lbits6;
612 		rbits6 = net->xfrm.policy_hthresh.rbits6;
613 	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
614 
615 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
616 
617 	/* reset the bydst and inexact table in all directions */
618 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
619 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
620 		hmask = net->xfrm.policy_bydst[dir].hmask;
621 		odst = net->xfrm.policy_bydst[dir].table;
622 		for (i = hmask; i >= 0; i--)
623 			INIT_HLIST_HEAD(odst + i);
624 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
625 			/* dir out => dst = remote, src = local */
626 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
627 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
628 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
629 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
630 		} else {
631 			/* dir in/fwd => dst = local, src = remote */
632 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
633 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
634 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
635 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
636 		}
637 	}
638 
639 	/* re-insert all policies by order of creation */
640 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
641 		if (xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
642 			/* skip socket policies */
643 			continue;
644 		}
645 		newpos = NULL;
646 		chain = policy_hash_bysel(net, &policy->selector,
647 					  policy->family,
648 					  xfrm_policy_id2dir(policy->index));
649 		hlist_for_each_entry(pol, chain, bydst) {
650 			if (policy->priority >= pol->priority)
651 				newpos = &pol->bydst;
652 			else
653 				break;
654 		}
655 		if (newpos)
656 			hlist_add_behind(&policy->bydst, newpos);
657 		else
658 			hlist_add_head(&policy->bydst, chain);
659 	}
660 
661 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
662 
663 	mutex_unlock(&hash_resize_mutex);
664 }
665 
666 void xfrm_policy_hash_rebuild(struct net *net)
667 {
668 	schedule_work(&net->xfrm.policy_hthresh.work);
669 }
670 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
671 
672 /* Generate new index... KAME seems to generate them ordered by cost
673  * of an absolute inpredictability of ordering of rules. This will not pass. */
674 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
675 {
676 	static u32 idx_generator;
677 
678 	for (;;) {
679 		struct hlist_head *list;
680 		struct xfrm_policy *p;
681 		u32 idx;
682 		int found;
683 
684 		if (!index) {
685 			idx = (idx_generator | dir);
686 			idx_generator += 8;
687 		} else {
688 			idx = index;
689 			index = 0;
690 		}
691 
692 		if (idx == 0)
693 			idx = 8;
694 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
695 		found = 0;
696 		hlist_for_each_entry(p, list, byidx) {
697 			if (p->index == idx) {
698 				found = 1;
699 				break;
700 			}
701 		}
702 		if (!found)
703 			return idx;
704 	}
705 }
706 
707 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
708 {
709 	u32 *p1 = (u32 *) s1;
710 	u32 *p2 = (u32 *) s2;
711 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
712 	int i;
713 
714 	for (i = 0; i < len; i++) {
715 		if (p1[i] != p2[i])
716 			return 1;
717 	}
718 
719 	return 0;
720 }
721 
722 static void xfrm_policy_requeue(struct xfrm_policy *old,
723 				struct xfrm_policy *new)
724 {
725 	struct xfrm_policy_queue *pq = &old->polq;
726 	struct sk_buff_head list;
727 
728 	if (skb_queue_empty(&pq->hold_queue))
729 		return;
730 
731 	__skb_queue_head_init(&list);
732 
733 	spin_lock_bh(&pq->hold_queue.lock);
734 	skb_queue_splice_init(&pq->hold_queue, &list);
735 	if (del_timer(&pq->hold_timer))
736 		xfrm_pol_put(old);
737 	spin_unlock_bh(&pq->hold_queue.lock);
738 
739 	pq = &new->polq;
740 
741 	spin_lock_bh(&pq->hold_queue.lock);
742 	skb_queue_splice(&list, &pq->hold_queue);
743 	pq->timeout = XFRM_QUEUE_TMO_MIN;
744 	if (!mod_timer(&pq->hold_timer, jiffies))
745 		xfrm_pol_hold(new);
746 	spin_unlock_bh(&pq->hold_queue.lock);
747 }
748 
749 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
750 				   struct xfrm_policy *pol)
751 {
752 	u32 mark = policy->mark.v & policy->mark.m;
753 
754 	if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
755 		return true;
756 
757 	if ((mark & pol->mark.m) == pol->mark.v &&
758 	    policy->priority == pol->priority)
759 		return true;
760 
761 	return false;
762 }
763 
764 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
765 {
766 	struct net *net = xp_net(policy);
767 	struct xfrm_policy *pol;
768 	struct xfrm_policy *delpol;
769 	struct hlist_head *chain;
770 	struct hlist_node *newpos;
771 
772 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
773 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
774 	delpol = NULL;
775 	newpos = NULL;
776 	hlist_for_each_entry(pol, chain, bydst) {
777 		if (pol->type == policy->type &&
778 		    !selector_cmp(&pol->selector, &policy->selector) &&
779 		    xfrm_policy_mark_match(policy, pol) &&
780 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
781 		    !WARN_ON(delpol)) {
782 			if (excl) {
783 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
784 				return -EEXIST;
785 			}
786 			delpol = pol;
787 			if (policy->priority > pol->priority)
788 				continue;
789 		} else if (policy->priority >= pol->priority) {
790 			newpos = &pol->bydst;
791 			continue;
792 		}
793 		if (delpol)
794 			break;
795 	}
796 	if (newpos)
797 		hlist_add_behind(&policy->bydst, newpos);
798 	else
799 		hlist_add_head(&policy->bydst, chain);
800 	__xfrm_policy_link(policy, dir);
801 	atomic_inc(&net->xfrm.flow_cache_genid);
802 
803 	/* After previous checking, family can either be AF_INET or AF_INET6 */
804 	if (policy->family == AF_INET)
805 		rt_genid_bump_ipv4(net);
806 	else
807 		rt_genid_bump_ipv6(net);
808 
809 	if (delpol) {
810 		xfrm_policy_requeue(delpol, policy);
811 		__xfrm_policy_unlink(delpol, dir);
812 	}
813 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
814 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
815 	policy->curlft.add_time = get_seconds();
816 	policy->curlft.use_time = 0;
817 	if (!mod_timer(&policy->timer, jiffies + HZ))
818 		xfrm_pol_hold(policy);
819 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
820 
821 	if (delpol)
822 		xfrm_policy_kill(delpol);
823 	else if (xfrm_bydst_should_resize(net, dir, NULL))
824 		schedule_work(&net->xfrm.policy_hash_work);
825 
826 	return 0;
827 }
828 EXPORT_SYMBOL(xfrm_policy_insert);
829 
830 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
831 					  int dir, struct xfrm_selector *sel,
832 					  struct xfrm_sec_ctx *ctx, int delete,
833 					  int *err)
834 {
835 	struct xfrm_policy *pol, *ret;
836 	struct hlist_head *chain;
837 
838 	*err = 0;
839 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
840 	chain = policy_hash_bysel(net, sel, sel->family, dir);
841 	ret = NULL;
842 	hlist_for_each_entry(pol, chain, bydst) {
843 		if (pol->type == type &&
844 		    (mark & pol->mark.m) == pol->mark.v &&
845 		    !selector_cmp(sel, &pol->selector) &&
846 		    xfrm_sec_ctx_match(ctx, pol->security)) {
847 			xfrm_pol_hold(pol);
848 			if (delete) {
849 				*err = security_xfrm_policy_delete(
850 								pol->security);
851 				if (*err) {
852 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
853 					return pol;
854 				}
855 				__xfrm_policy_unlink(pol, dir);
856 			}
857 			ret = pol;
858 			break;
859 		}
860 	}
861 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
862 
863 	if (ret && delete)
864 		xfrm_policy_kill(ret);
865 	return ret;
866 }
867 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
868 
869 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
870 				     int dir, u32 id, int delete, int *err)
871 {
872 	struct xfrm_policy *pol, *ret;
873 	struct hlist_head *chain;
874 
875 	*err = -ENOENT;
876 	if (xfrm_policy_id2dir(id) != dir)
877 		return NULL;
878 
879 	*err = 0;
880 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
881 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
882 	ret = NULL;
883 	hlist_for_each_entry(pol, chain, byidx) {
884 		if (pol->type == type && pol->index == id &&
885 		    (mark & pol->mark.m) == pol->mark.v) {
886 			xfrm_pol_hold(pol);
887 			if (delete) {
888 				*err = security_xfrm_policy_delete(
889 								pol->security);
890 				if (*err) {
891 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
892 					return pol;
893 				}
894 				__xfrm_policy_unlink(pol, dir);
895 			}
896 			ret = pol;
897 			break;
898 		}
899 	}
900 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
901 
902 	if (ret && delete)
903 		xfrm_policy_kill(ret);
904 	return ret;
905 }
906 EXPORT_SYMBOL(xfrm_policy_byid);
907 
908 #ifdef CONFIG_SECURITY_NETWORK_XFRM
909 static inline int
910 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
911 {
912 	int dir, err = 0;
913 
914 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
915 		struct xfrm_policy *pol;
916 		int i;
917 
918 		hlist_for_each_entry(pol,
919 				     &net->xfrm.policy_inexact[dir], bydst) {
920 			if (pol->type != type)
921 				continue;
922 			err = security_xfrm_policy_delete(pol->security);
923 			if (err) {
924 				xfrm_audit_policy_delete(pol, 0, task_valid);
925 				return err;
926 			}
927 		}
928 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
929 			hlist_for_each_entry(pol,
930 					     net->xfrm.policy_bydst[dir].table + i,
931 					     bydst) {
932 				if (pol->type != type)
933 					continue;
934 				err = security_xfrm_policy_delete(
935 								pol->security);
936 				if (err) {
937 					xfrm_audit_policy_delete(pol, 0,
938 								 task_valid);
939 					return err;
940 				}
941 			}
942 		}
943 	}
944 	return err;
945 }
946 #else
947 static inline int
948 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
949 {
950 	return 0;
951 }
952 #endif
953 
954 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
955 {
956 	int dir, err = 0, cnt = 0;
957 
958 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
959 
960 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
961 	if (err)
962 		goto out;
963 
964 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
965 		struct xfrm_policy *pol;
966 		int i;
967 
968 	again1:
969 		hlist_for_each_entry(pol,
970 				     &net->xfrm.policy_inexact[dir], bydst) {
971 			if (pol->type != type)
972 				continue;
973 			__xfrm_policy_unlink(pol, dir);
974 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
975 			cnt++;
976 
977 			xfrm_audit_policy_delete(pol, 1, task_valid);
978 
979 			xfrm_policy_kill(pol);
980 
981 			spin_lock_bh(&net->xfrm.xfrm_policy_lock);
982 			goto again1;
983 		}
984 
985 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
986 	again2:
987 			hlist_for_each_entry(pol,
988 					     net->xfrm.policy_bydst[dir].table + i,
989 					     bydst) {
990 				if (pol->type != type)
991 					continue;
992 				__xfrm_policy_unlink(pol, dir);
993 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
994 				cnt++;
995 
996 				xfrm_audit_policy_delete(pol, 1, task_valid);
997 				xfrm_policy_kill(pol);
998 
999 				spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1000 				goto again2;
1001 			}
1002 		}
1003 
1004 	}
1005 	if (!cnt)
1006 		err = -ESRCH;
1007 out:
1008 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1009 	return err;
1010 }
1011 EXPORT_SYMBOL(xfrm_policy_flush);
1012 
1013 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1014 		     int (*func)(struct xfrm_policy *, int, int, void*),
1015 		     void *data)
1016 {
1017 	struct xfrm_policy *pol;
1018 	struct xfrm_policy_walk_entry *x;
1019 	int error = 0;
1020 
1021 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1022 	    walk->type != XFRM_POLICY_TYPE_ANY)
1023 		return -EINVAL;
1024 
1025 	if (list_empty(&walk->walk.all) && walk->seq != 0)
1026 		return 0;
1027 
1028 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1029 	if (list_empty(&walk->walk.all))
1030 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1031 	else
1032 		x = list_first_entry(&walk->walk.all,
1033 				     struct xfrm_policy_walk_entry, all);
1034 
1035 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1036 		if (x->dead)
1037 			continue;
1038 		pol = container_of(x, struct xfrm_policy, walk);
1039 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1040 		    walk->type != pol->type)
1041 			continue;
1042 		error = func(pol, xfrm_policy_id2dir(pol->index),
1043 			     walk->seq, data);
1044 		if (error) {
1045 			list_move_tail(&walk->walk.all, &x->all);
1046 			goto out;
1047 		}
1048 		walk->seq++;
1049 	}
1050 	if (walk->seq == 0) {
1051 		error = -ENOENT;
1052 		goto out;
1053 	}
1054 	list_del_init(&walk->walk.all);
1055 out:
1056 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1057 	return error;
1058 }
1059 EXPORT_SYMBOL(xfrm_policy_walk);
1060 
1061 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1062 {
1063 	INIT_LIST_HEAD(&walk->walk.all);
1064 	walk->walk.dead = 1;
1065 	walk->type = type;
1066 	walk->seq = 0;
1067 }
1068 EXPORT_SYMBOL(xfrm_policy_walk_init);
1069 
1070 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1071 {
1072 	if (list_empty(&walk->walk.all))
1073 		return;
1074 
1075 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1076 	list_del(&walk->walk.all);
1077 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1078 }
1079 EXPORT_SYMBOL(xfrm_policy_walk_done);
1080 
1081 /*
1082  * Find policy to apply to this flow.
1083  *
1084  * Returns 0 if policy found, else an -errno.
1085  */
1086 static int xfrm_policy_match(const struct xfrm_policy *pol,
1087 			     const struct flowi *fl,
1088 			     u8 type, u16 family, int dir)
1089 {
1090 	const struct xfrm_selector *sel = &pol->selector;
1091 	int ret = -ESRCH;
1092 	bool match;
1093 
1094 	if (pol->family != family ||
1095 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1096 	    pol->type != type)
1097 		return ret;
1098 
1099 	match = xfrm_selector_match(sel, fl, family);
1100 	if (match)
1101 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1102 						  dir);
1103 
1104 	return ret;
1105 }
1106 
1107 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1108 						     const struct flowi *fl,
1109 						     u16 family, u8 dir)
1110 {
1111 	int err;
1112 	struct xfrm_policy *pol, *ret;
1113 	const xfrm_address_t *daddr, *saddr;
1114 	struct hlist_head *chain;
1115 	unsigned int sequence;
1116 	u32 priority;
1117 
1118 	daddr = xfrm_flowi_daddr(fl, family);
1119 	saddr = xfrm_flowi_saddr(fl, family);
1120 	if (unlikely(!daddr || !saddr))
1121 		return NULL;
1122 
1123 	rcu_read_lock();
1124  retry:
1125 	do {
1126 		sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1127 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
1128 	} while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1129 
1130 	priority = ~0U;
1131 	ret = NULL;
1132 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1133 		err = xfrm_policy_match(pol, fl, type, family, dir);
1134 		if (err) {
1135 			if (err == -ESRCH)
1136 				continue;
1137 			else {
1138 				ret = ERR_PTR(err);
1139 				goto fail;
1140 			}
1141 		} else {
1142 			ret = pol;
1143 			priority = ret->priority;
1144 			break;
1145 		}
1146 	}
1147 	chain = &net->xfrm.policy_inexact[dir];
1148 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1149 		if ((pol->priority >= priority) && ret)
1150 			break;
1151 
1152 		err = xfrm_policy_match(pol, fl, type, family, dir);
1153 		if (err) {
1154 			if (err == -ESRCH)
1155 				continue;
1156 			else {
1157 				ret = ERR_PTR(err);
1158 				goto fail;
1159 			}
1160 		} else {
1161 			ret = pol;
1162 			break;
1163 		}
1164 	}
1165 
1166 	if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1167 		goto retry;
1168 
1169 	if (ret && !xfrm_pol_hold_rcu(ret))
1170 		goto retry;
1171 fail:
1172 	rcu_read_unlock();
1173 
1174 	return ret;
1175 }
1176 
1177 static struct xfrm_policy *
1178 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1179 {
1180 #ifdef CONFIG_XFRM_SUB_POLICY
1181 	struct xfrm_policy *pol;
1182 
1183 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1184 	if (pol != NULL)
1185 		return pol;
1186 #endif
1187 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1188 }
1189 
1190 static int flow_to_policy_dir(int dir)
1191 {
1192 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1193 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1194 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
1195 		return dir;
1196 
1197 	switch (dir) {
1198 	default:
1199 	case FLOW_DIR_IN:
1200 		return XFRM_POLICY_IN;
1201 	case FLOW_DIR_OUT:
1202 		return XFRM_POLICY_OUT;
1203 	case FLOW_DIR_FWD:
1204 		return XFRM_POLICY_FWD;
1205 	}
1206 }
1207 
1208 static struct flow_cache_object *
1209 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1210 		   u8 dir, struct flow_cache_object *old_obj, void *ctx)
1211 {
1212 	struct xfrm_policy *pol;
1213 
1214 	if (old_obj)
1215 		xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1216 
1217 	pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1218 	if (IS_ERR_OR_NULL(pol))
1219 		return ERR_CAST(pol);
1220 
1221 	/* Resolver returns two references:
1222 	 * one for cache and one for caller of flow_cache_lookup() */
1223 	xfrm_pol_hold(pol);
1224 
1225 	return &pol->flo;
1226 }
1227 
1228 static inline int policy_to_flow_dir(int dir)
1229 {
1230 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1231 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1232 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
1233 		return dir;
1234 	switch (dir) {
1235 	default:
1236 	case XFRM_POLICY_IN:
1237 		return FLOW_DIR_IN;
1238 	case XFRM_POLICY_OUT:
1239 		return FLOW_DIR_OUT;
1240 	case XFRM_POLICY_FWD:
1241 		return FLOW_DIR_FWD;
1242 	}
1243 }
1244 
1245 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1246 						 const struct flowi *fl, u16 family)
1247 {
1248 	struct xfrm_policy *pol;
1249 
1250 	rcu_read_lock();
1251  again:
1252 	pol = rcu_dereference(sk->sk_policy[dir]);
1253 	if (pol != NULL) {
1254 		bool match = xfrm_selector_match(&pol->selector, fl, family);
1255 		int err = 0;
1256 
1257 		if (match) {
1258 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1259 				pol = NULL;
1260 				goto out;
1261 			}
1262 			err = security_xfrm_policy_lookup(pol->security,
1263 						      fl->flowi_secid,
1264 						      policy_to_flow_dir(dir));
1265 			if (!err) {
1266 				if (!xfrm_pol_hold_rcu(pol))
1267 					goto again;
1268 			} else if (err == -ESRCH) {
1269 				pol = NULL;
1270 			} else {
1271 				pol = ERR_PTR(err);
1272 			}
1273 		} else
1274 			pol = NULL;
1275 	}
1276 out:
1277 	rcu_read_unlock();
1278 	return pol;
1279 }
1280 
1281 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1282 {
1283 	struct net *net = xp_net(pol);
1284 
1285 	list_add(&pol->walk.all, &net->xfrm.policy_all);
1286 	net->xfrm.policy_count[dir]++;
1287 	xfrm_pol_hold(pol);
1288 }
1289 
1290 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1291 						int dir)
1292 {
1293 	struct net *net = xp_net(pol);
1294 
1295 	if (list_empty(&pol->walk.all))
1296 		return NULL;
1297 
1298 	/* Socket policies are not hashed. */
1299 	if (!hlist_unhashed(&pol->bydst)) {
1300 		hlist_del_rcu(&pol->bydst);
1301 		hlist_del(&pol->byidx);
1302 	}
1303 
1304 	list_del_init(&pol->walk.all);
1305 	net->xfrm.policy_count[dir]--;
1306 
1307 	return pol;
1308 }
1309 
1310 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1311 {
1312 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1313 }
1314 
1315 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1316 {
1317 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1318 }
1319 
1320 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1321 {
1322 	struct net *net = xp_net(pol);
1323 
1324 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1325 	pol = __xfrm_policy_unlink(pol, dir);
1326 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1327 	if (pol) {
1328 		xfrm_policy_kill(pol);
1329 		return 0;
1330 	}
1331 	return -ENOENT;
1332 }
1333 EXPORT_SYMBOL(xfrm_policy_delete);
1334 
1335 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1336 {
1337 	struct net *net = xp_net(pol);
1338 	struct xfrm_policy *old_pol;
1339 
1340 #ifdef CONFIG_XFRM_SUB_POLICY
1341 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1342 		return -EINVAL;
1343 #endif
1344 
1345 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1346 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1347 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1348 	if (pol) {
1349 		pol->curlft.add_time = get_seconds();
1350 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1351 		xfrm_sk_policy_link(pol, dir);
1352 	}
1353 	rcu_assign_pointer(sk->sk_policy[dir], pol);
1354 	if (old_pol) {
1355 		if (pol)
1356 			xfrm_policy_requeue(old_pol, pol);
1357 
1358 		/* Unlinking succeeds always. This is the only function
1359 		 * allowed to delete or replace socket policy.
1360 		 */
1361 		xfrm_sk_policy_unlink(old_pol, dir);
1362 	}
1363 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1364 
1365 	if (old_pol) {
1366 		xfrm_policy_kill(old_pol);
1367 	}
1368 	return 0;
1369 }
1370 
1371 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1372 {
1373 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1374 	struct net *net = xp_net(old);
1375 
1376 	if (newp) {
1377 		newp->selector = old->selector;
1378 		if (security_xfrm_policy_clone(old->security,
1379 					       &newp->security)) {
1380 			kfree(newp);
1381 			return NULL;  /* ENOMEM */
1382 		}
1383 		newp->lft = old->lft;
1384 		newp->curlft = old->curlft;
1385 		newp->mark = old->mark;
1386 		newp->action = old->action;
1387 		newp->flags = old->flags;
1388 		newp->xfrm_nr = old->xfrm_nr;
1389 		newp->index = old->index;
1390 		newp->type = old->type;
1391 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1392 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1393 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1394 		xfrm_sk_policy_link(newp, dir);
1395 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1396 		xfrm_pol_put(newp);
1397 	}
1398 	return newp;
1399 }
1400 
1401 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1402 {
1403 	const struct xfrm_policy *p;
1404 	struct xfrm_policy *np;
1405 	int i, ret = 0;
1406 
1407 	rcu_read_lock();
1408 	for (i = 0; i < 2; i++) {
1409 		p = rcu_dereference(osk->sk_policy[i]);
1410 		if (p) {
1411 			np = clone_policy(p, i);
1412 			if (unlikely(!np)) {
1413 				ret = -ENOMEM;
1414 				break;
1415 			}
1416 			rcu_assign_pointer(sk->sk_policy[i], np);
1417 		}
1418 	}
1419 	rcu_read_unlock();
1420 	return ret;
1421 }
1422 
1423 static int
1424 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1425 	       xfrm_address_t *remote, unsigned short family)
1426 {
1427 	int err;
1428 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1429 
1430 	if (unlikely(afinfo == NULL))
1431 		return -EINVAL;
1432 	err = afinfo->get_saddr(net, oif, local, remote);
1433 	rcu_read_unlock();
1434 	return err;
1435 }
1436 
1437 /* Resolve list of templates for the flow, given policy. */
1438 
1439 static int
1440 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1441 		      struct xfrm_state **xfrm, unsigned short family)
1442 {
1443 	struct net *net = xp_net(policy);
1444 	int nx;
1445 	int i, error;
1446 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1447 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1448 	xfrm_address_t tmp;
1449 
1450 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1451 		struct xfrm_state *x;
1452 		xfrm_address_t *remote = daddr;
1453 		xfrm_address_t *local  = saddr;
1454 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1455 
1456 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1457 		    tmpl->mode == XFRM_MODE_BEET) {
1458 			remote = &tmpl->id.daddr;
1459 			local = &tmpl->saddr;
1460 			if (xfrm_addr_any(local, tmpl->encap_family)) {
1461 				error = xfrm_get_saddr(net, fl->flowi_oif,
1462 						       &tmp, remote,
1463 						       tmpl->encap_family);
1464 				if (error)
1465 					goto fail;
1466 				local = &tmp;
1467 			}
1468 		}
1469 
1470 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1471 
1472 		if (x && x->km.state == XFRM_STATE_VALID) {
1473 			xfrm[nx++] = x;
1474 			daddr = remote;
1475 			saddr = local;
1476 			continue;
1477 		}
1478 		if (x) {
1479 			error = (x->km.state == XFRM_STATE_ERROR ?
1480 				 -EINVAL : -EAGAIN);
1481 			xfrm_state_put(x);
1482 		} else if (error == -ESRCH) {
1483 			error = -EAGAIN;
1484 		}
1485 
1486 		if (!tmpl->optional)
1487 			goto fail;
1488 	}
1489 	return nx;
1490 
1491 fail:
1492 	for (nx--; nx >= 0; nx--)
1493 		xfrm_state_put(xfrm[nx]);
1494 	return error;
1495 }
1496 
1497 static int
1498 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1499 		  struct xfrm_state **xfrm, unsigned short family)
1500 {
1501 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1502 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1503 	int cnx = 0;
1504 	int error;
1505 	int ret;
1506 	int i;
1507 
1508 	for (i = 0; i < npols; i++) {
1509 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1510 			error = -ENOBUFS;
1511 			goto fail;
1512 		}
1513 
1514 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1515 		if (ret < 0) {
1516 			error = ret;
1517 			goto fail;
1518 		} else
1519 			cnx += ret;
1520 	}
1521 
1522 	/* found states are sorted for outbound processing */
1523 	if (npols > 1)
1524 		xfrm_state_sort(xfrm, tpp, cnx, family);
1525 
1526 	return cnx;
1527 
1528  fail:
1529 	for (cnx--; cnx >= 0; cnx--)
1530 		xfrm_state_put(tpp[cnx]);
1531 	return error;
1532 
1533 }
1534 
1535 static int xfrm_get_tos(const struct flowi *fl, int family)
1536 {
1537 	const struct xfrm_policy_afinfo *afinfo;
1538 	int tos = 0;
1539 
1540 	afinfo = xfrm_policy_get_afinfo(family);
1541 	tos = afinfo ? afinfo->get_tos(fl) : 0;
1542 
1543 	rcu_read_unlock();
1544 
1545 	return tos;
1546 }
1547 
1548 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1549 {
1550 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1551 	struct dst_entry *dst = &xdst->u.dst;
1552 
1553 	if (xdst->route == NULL) {
1554 		/* Dummy bundle - if it has xfrms we were not
1555 		 * able to build bundle as template resolution failed.
1556 		 * It means we need to try again resolving. */
1557 		if (xdst->num_xfrms > 0)
1558 			return NULL;
1559 	} else if (dst->flags & DST_XFRM_QUEUE) {
1560 		return NULL;
1561 	} else {
1562 		/* Real bundle */
1563 		if (stale_bundle(dst))
1564 			return NULL;
1565 	}
1566 
1567 	dst_hold(dst);
1568 	return flo;
1569 }
1570 
1571 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1572 {
1573 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1574 	struct dst_entry *dst = &xdst->u.dst;
1575 
1576 	if (!xdst->route)
1577 		return 0;
1578 	if (stale_bundle(dst))
1579 		return 0;
1580 
1581 	return 1;
1582 }
1583 
1584 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1585 {
1586 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1587 	struct dst_entry *dst = &xdst->u.dst;
1588 
1589 	dst_free(dst);
1590 }
1591 
1592 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1593 	.get = xfrm_bundle_flo_get,
1594 	.check = xfrm_bundle_flo_check,
1595 	.delete = xfrm_bundle_flo_delete,
1596 };
1597 
1598 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1599 {
1600 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1601 	struct dst_ops *dst_ops;
1602 	struct xfrm_dst *xdst;
1603 
1604 	if (!afinfo)
1605 		return ERR_PTR(-EINVAL);
1606 
1607 	switch (family) {
1608 	case AF_INET:
1609 		dst_ops = &net->xfrm.xfrm4_dst_ops;
1610 		break;
1611 #if IS_ENABLED(CONFIG_IPV6)
1612 	case AF_INET6:
1613 		dst_ops = &net->xfrm.xfrm6_dst_ops;
1614 		break;
1615 #endif
1616 	default:
1617 		BUG();
1618 	}
1619 	xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1620 
1621 	if (likely(xdst)) {
1622 		struct dst_entry *dst = &xdst->u.dst;
1623 
1624 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1625 		xdst->flo.ops = &xfrm_bundle_fc_ops;
1626 	} else
1627 		xdst = ERR_PTR(-ENOBUFS);
1628 
1629 	rcu_read_unlock();
1630 
1631 	return xdst;
1632 }
1633 
1634 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1635 				 int nfheader_len)
1636 {
1637 	const struct xfrm_policy_afinfo *afinfo =
1638 		xfrm_policy_get_afinfo(dst->ops->family);
1639 	int err;
1640 
1641 	if (!afinfo)
1642 		return -EINVAL;
1643 
1644 	err = afinfo->init_path(path, dst, nfheader_len);
1645 
1646 	rcu_read_unlock();
1647 
1648 	return err;
1649 }
1650 
1651 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1652 				const struct flowi *fl)
1653 {
1654 	const struct xfrm_policy_afinfo *afinfo =
1655 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1656 	int err;
1657 
1658 	if (!afinfo)
1659 		return -EINVAL;
1660 
1661 	err = afinfo->fill_dst(xdst, dev, fl);
1662 
1663 	rcu_read_unlock();
1664 
1665 	return err;
1666 }
1667 
1668 
1669 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1670  * all the metrics... Shortly, bundle a bundle.
1671  */
1672 
1673 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1674 					    struct xfrm_state **xfrm, int nx,
1675 					    const struct flowi *fl,
1676 					    struct dst_entry *dst)
1677 {
1678 	struct net *net = xp_net(policy);
1679 	unsigned long now = jiffies;
1680 	struct net_device *dev;
1681 	struct xfrm_mode *inner_mode;
1682 	struct dst_entry *dst_prev = NULL;
1683 	struct dst_entry *dst0 = NULL;
1684 	int i = 0;
1685 	int err;
1686 	int header_len = 0;
1687 	int nfheader_len = 0;
1688 	int trailer_len = 0;
1689 	int tos;
1690 	int family = policy->selector.family;
1691 	xfrm_address_t saddr, daddr;
1692 
1693 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1694 
1695 	tos = xfrm_get_tos(fl, family);
1696 
1697 	dst_hold(dst);
1698 
1699 	for (; i < nx; i++) {
1700 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1701 		struct dst_entry *dst1 = &xdst->u.dst;
1702 
1703 		err = PTR_ERR(xdst);
1704 		if (IS_ERR(xdst)) {
1705 			dst_release(dst);
1706 			goto put_states;
1707 		}
1708 
1709 		if (xfrm[i]->sel.family == AF_UNSPEC) {
1710 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
1711 							xfrm_af2proto(family));
1712 			if (!inner_mode) {
1713 				err = -EAFNOSUPPORT;
1714 				dst_release(dst);
1715 				goto put_states;
1716 			}
1717 		} else
1718 			inner_mode = xfrm[i]->inner_mode;
1719 
1720 		if (!dst_prev)
1721 			dst0 = dst1;
1722 		else {
1723 			dst_prev->child = dst_clone(dst1);
1724 			dst1->flags |= DST_NOHASH;
1725 		}
1726 
1727 		xdst->route = dst;
1728 		dst_copy_metrics(dst1, dst);
1729 
1730 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1731 			family = xfrm[i]->props.family;
1732 			dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1733 					      &saddr, &daddr, family);
1734 			err = PTR_ERR(dst);
1735 			if (IS_ERR(dst))
1736 				goto put_states;
1737 		} else
1738 			dst_hold(dst);
1739 
1740 		dst1->xfrm = xfrm[i];
1741 		xdst->xfrm_genid = xfrm[i]->genid;
1742 
1743 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1744 		dst1->flags |= DST_HOST;
1745 		dst1->lastuse = now;
1746 
1747 		dst1->input = dst_discard;
1748 		dst1->output = inner_mode->afinfo->output;
1749 
1750 		dst1->next = dst_prev;
1751 		dst_prev = dst1;
1752 
1753 		header_len += xfrm[i]->props.header_len;
1754 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1755 			nfheader_len += xfrm[i]->props.header_len;
1756 		trailer_len += xfrm[i]->props.trailer_len;
1757 	}
1758 
1759 	dst_prev->child = dst;
1760 	dst0->path = dst;
1761 
1762 	err = -ENODEV;
1763 	dev = dst->dev;
1764 	if (!dev)
1765 		goto free_dst;
1766 
1767 	xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1768 	xfrm_init_pmtu(dst_prev);
1769 
1770 	for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1771 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1772 
1773 		err = xfrm_fill_dst(xdst, dev, fl);
1774 		if (err)
1775 			goto free_dst;
1776 
1777 		dst_prev->header_len = header_len;
1778 		dst_prev->trailer_len = trailer_len;
1779 		header_len -= xdst->u.dst.xfrm->props.header_len;
1780 		trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1781 	}
1782 
1783 out:
1784 	return dst0;
1785 
1786 put_states:
1787 	for (; i < nx; i++)
1788 		xfrm_state_put(xfrm[i]);
1789 free_dst:
1790 	if (dst0)
1791 		dst_free(dst0);
1792 	dst0 = ERR_PTR(err);
1793 	goto out;
1794 }
1795 
1796 #ifdef CONFIG_XFRM_SUB_POLICY
1797 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1798 {
1799 	if (!*target) {
1800 		*target = kmalloc(size, GFP_ATOMIC);
1801 		if (!*target)
1802 			return -ENOMEM;
1803 	}
1804 
1805 	memcpy(*target, src, size);
1806 	return 0;
1807 }
1808 #endif
1809 
1810 static int xfrm_dst_update_parent(struct dst_entry *dst,
1811 				  const struct xfrm_selector *sel)
1812 {
1813 #ifdef CONFIG_XFRM_SUB_POLICY
1814 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1815 	return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1816 				   sel, sizeof(*sel));
1817 #else
1818 	return 0;
1819 #endif
1820 }
1821 
1822 static int xfrm_dst_update_origin(struct dst_entry *dst,
1823 				  const struct flowi *fl)
1824 {
1825 #ifdef CONFIG_XFRM_SUB_POLICY
1826 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1827 	return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1828 #else
1829 	return 0;
1830 #endif
1831 }
1832 
1833 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1834 				struct xfrm_policy **pols,
1835 				int *num_pols, int *num_xfrms)
1836 {
1837 	int i;
1838 
1839 	if (*num_pols == 0 || !pols[0]) {
1840 		*num_pols = 0;
1841 		*num_xfrms = 0;
1842 		return 0;
1843 	}
1844 	if (IS_ERR(pols[0]))
1845 		return PTR_ERR(pols[0]);
1846 
1847 	*num_xfrms = pols[0]->xfrm_nr;
1848 
1849 #ifdef CONFIG_XFRM_SUB_POLICY
1850 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1851 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1852 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1853 						    XFRM_POLICY_TYPE_MAIN,
1854 						    fl, family,
1855 						    XFRM_POLICY_OUT);
1856 		if (pols[1]) {
1857 			if (IS_ERR(pols[1])) {
1858 				xfrm_pols_put(pols, *num_pols);
1859 				return PTR_ERR(pols[1]);
1860 			}
1861 			(*num_pols)++;
1862 			(*num_xfrms) += pols[1]->xfrm_nr;
1863 		}
1864 	}
1865 #endif
1866 	for (i = 0; i < *num_pols; i++) {
1867 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
1868 			*num_xfrms = -1;
1869 			break;
1870 		}
1871 	}
1872 
1873 	return 0;
1874 
1875 }
1876 
1877 static struct xfrm_dst *
1878 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1879 			       const struct flowi *fl, u16 family,
1880 			       struct dst_entry *dst_orig)
1881 {
1882 	struct net *net = xp_net(pols[0]);
1883 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1884 	struct dst_entry *dst;
1885 	struct xfrm_dst *xdst;
1886 	int err;
1887 
1888 	/* Try to instantiate a bundle */
1889 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1890 	if (err <= 0) {
1891 		if (err != 0 && err != -EAGAIN)
1892 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1893 		return ERR_PTR(err);
1894 	}
1895 
1896 	dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1897 	if (IS_ERR(dst)) {
1898 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1899 		return ERR_CAST(dst);
1900 	}
1901 
1902 	xdst = (struct xfrm_dst *)dst;
1903 	xdst->num_xfrms = err;
1904 	if (num_pols > 1)
1905 		err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1906 	else
1907 		err = xfrm_dst_update_origin(dst, fl);
1908 	if (unlikely(err)) {
1909 		dst_free(dst);
1910 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1911 		return ERR_PTR(err);
1912 	}
1913 
1914 	xdst->num_pols = num_pols;
1915 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1916 	xdst->policy_genid = atomic_read(&pols[0]->genid);
1917 
1918 	return xdst;
1919 }
1920 
1921 static void xfrm_policy_queue_process(unsigned long arg)
1922 {
1923 	struct sk_buff *skb;
1924 	struct sock *sk;
1925 	struct dst_entry *dst;
1926 	struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1927 	struct net *net = xp_net(pol);
1928 	struct xfrm_policy_queue *pq = &pol->polq;
1929 	struct flowi fl;
1930 	struct sk_buff_head list;
1931 
1932 	spin_lock(&pq->hold_queue.lock);
1933 	skb = skb_peek(&pq->hold_queue);
1934 	if (!skb) {
1935 		spin_unlock(&pq->hold_queue.lock);
1936 		goto out;
1937 	}
1938 	dst = skb_dst(skb);
1939 	sk = skb->sk;
1940 	xfrm_decode_session(skb, &fl, dst->ops->family);
1941 	spin_unlock(&pq->hold_queue.lock);
1942 
1943 	dst_hold(dst->path);
1944 	dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1945 	if (IS_ERR(dst))
1946 		goto purge_queue;
1947 
1948 	if (dst->flags & DST_XFRM_QUEUE) {
1949 		dst_release(dst);
1950 
1951 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1952 			goto purge_queue;
1953 
1954 		pq->timeout = pq->timeout << 1;
1955 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1956 			xfrm_pol_hold(pol);
1957 	goto out;
1958 	}
1959 
1960 	dst_release(dst);
1961 
1962 	__skb_queue_head_init(&list);
1963 
1964 	spin_lock(&pq->hold_queue.lock);
1965 	pq->timeout = 0;
1966 	skb_queue_splice_init(&pq->hold_queue, &list);
1967 	spin_unlock(&pq->hold_queue.lock);
1968 
1969 	while (!skb_queue_empty(&list)) {
1970 		skb = __skb_dequeue(&list);
1971 
1972 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1973 		dst_hold(skb_dst(skb)->path);
1974 		dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1975 		if (IS_ERR(dst)) {
1976 			kfree_skb(skb);
1977 			continue;
1978 		}
1979 
1980 		nf_reset(skb);
1981 		skb_dst_drop(skb);
1982 		skb_dst_set(skb, dst);
1983 
1984 		dst_output(net, skb->sk, skb);
1985 	}
1986 
1987 out:
1988 	xfrm_pol_put(pol);
1989 	return;
1990 
1991 purge_queue:
1992 	pq->timeout = 0;
1993 	skb_queue_purge(&pq->hold_queue);
1994 	xfrm_pol_put(pol);
1995 }
1996 
1997 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1998 {
1999 	unsigned long sched_next;
2000 	struct dst_entry *dst = skb_dst(skb);
2001 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2002 	struct xfrm_policy *pol = xdst->pols[0];
2003 	struct xfrm_policy_queue *pq = &pol->polq;
2004 
2005 	if (unlikely(skb_fclone_busy(sk, skb))) {
2006 		kfree_skb(skb);
2007 		return 0;
2008 	}
2009 
2010 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2011 		kfree_skb(skb);
2012 		return -EAGAIN;
2013 	}
2014 
2015 	skb_dst_force(skb);
2016 
2017 	spin_lock_bh(&pq->hold_queue.lock);
2018 
2019 	if (!pq->timeout)
2020 		pq->timeout = XFRM_QUEUE_TMO_MIN;
2021 
2022 	sched_next = jiffies + pq->timeout;
2023 
2024 	if (del_timer(&pq->hold_timer)) {
2025 		if (time_before(pq->hold_timer.expires, sched_next))
2026 			sched_next = pq->hold_timer.expires;
2027 		xfrm_pol_put(pol);
2028 	}
2029 
2030 	__skb_queue_tail(&pq->hold_queue, skb);
2031 	if (!mod_timer(&pq->hold_timer, sched_next))
2032 		xfrm_pol_hold(pol);
2033 
2034 	spin_unlock_bh(&pq->hold_queue.lock);
2035 
2036 	return 0;
2037 }
2038 
2039 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2040 						 struct xfrm_flo *xflo,
2041 						 const struct flowi *fl,
2042 						 int num_xfrms,
2043 						 u16 family)
2044 {
2045 	int err;
2046 	struct net_device *dev;
2047 	struct dst_entry *dst;
2048 	struct dst_entry *dst1;
2049 	struct xfrm_dst *xdst;
2050 
2051 	xdst = xfrm_alloc_dst(net, family);
2052 	if (IS_ERR(xdst))
2053 		return xdst;
2054 
2055 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2056 	    net->xfrm.sysctl_larval_drop ||
2057 	    num_xfrms <= 0)
2058 		return xdst;
2059 
2060 	dst = xflo->dst_orig;
2061 	dst1 = &xdst->u.dst;
2062 	dst_hold(dst);
2063 	xdst->route = dst;
2064 
2065 	dst_copy_metrics(dst1, dst);
2066 
2067 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2068 	dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2069 	dst1->lastuse = jiffies;
2070 
2071 	dst1->input = dst_discard;
2072 	dst1->output = xdst_queue_output;
2073 
2074 	dst_hold(dst);
2075 	dst1->child = dst;
2076 	dst1->path = dst;
2077 
2078 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2079 
2080 	err = -ENODEV;
2081 	dev = dst->dev;
2082 	if (!dev)
2083 		goto free_dst;
2084 
2085 	err = xfrm_fill_dst(xdst, dev, fl);
2086 	if (err)
2087 		goto free_dst;
2088 
2089 out:
2090 	return xdst;
2091 
2092 free_dst:
2093 	dst_release(dst1);
2094 	xdst = ERR_PTR(err);
2095 	goto out;
2096 }
2097 
2098 static struct flow_cache_object *
2099 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2100 		   struct flow_cache_object *oldflo, void *ctx)
2101 {
2102 	struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2103 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2104 	struct xfrm_dst *xdst, *new_xdst;
2105 	int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2106 
2107 	/* Check if the policies from old bundle are usable */
2108 	xdst = NULL;
2109 	if (oldflo) {
2110 		xdst = container_of(oldflo, struct xfrm_dst, flo);
2111 		num_pols = xdst->num_pols;
2112 		num_xfrms = xdst->num_xfrms;
2113 		pol_dead = 0;
2114 		for (i = 0; i < num_pols; i++) {
2115 			pols[i] = xdst->pols[i];
2116 			pol_dead |= pols[i]->walk.dead;
2117 		}
2118 		if (pol_dead) {
2119 			dst_free(&xdst->u.dst);
2120 			xdst = NULL;
2121 			num_pols = 0;
2122 			num_xfrms = 0;
2123 			oldflo = NULL;
2124 		}
2125 	}
2126 
2127 	/* Resolve policies to use if we couldn't get them from
2128 	 * previous cache entry */
2129 	if (xdst == NULL) {
2130 		num_pols = 1;
2131 		pols[0] = __xfrm_policy_lookup(net, fl, family,
2132 					       flow_to_policy_dir(dir));
2133 		err = xfrm_expand_policies(fl, family, pols,
2134 					   &num_pols, &num_xfrms);
2135 		if (err < 0)
2136 			goto inc_error;
2137 		if (num_pols == 0)
2138 			return NULL;
2139 		if (num_xfrms <= 0)
2140 			goto make_dummy_bundle;
2141 	}
2142 
2143 	new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2144 						  xflo->dst_orig);
2145 	if (IS_ERR(new_xdst)) {
2146 		err = PTR_ERR(new_xdst);
2147 		if (err != -EAGAIN)
2148 			goto error;
2149 		if (oldflo == NULL)
2150 			goto make_dummy_bundle;
2151 		dst_hold(&xdst->u.dst);
2152 		return oldflo;
2153 	} else if (new_xdst == NULL) {
2154 		num_xfrms = 0;
2155 		if (oldflo == NULL)
2156 			goto make_dummy_bundle;
2157 		xdst->num_xfrms = 0;
2158 		dst_hold(&xdst->u.dst);
2159 		return oldflo;
2160 	}
2161 
2162 	/* Kill the previous bundle */
2163 	if (xdst) {
2164 		/* The policies were stolen for newly generated bundle */
2165 		xdst->num_pols = 0;
2166 		dst_free(&xdst->u.dst);
2167 	}
2168 
2169 	/* Flow cache does not have reference, it dst_free()'s,
2170 	 * but we do need to return one reference for original caller */
2171 	dst_hold(&new_xdst->u.dst);
2172 	return &new_xdst->flo;
2173 
2174 make_dummy_bundle:
2175 	/* We found policies, but there's no bundles to instantiate:
2176 	 * either because the policy blocks, has no transformations or
2177 	 * we could not build template (no xfrm_states).*/
2178 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2179 	if (IS_ERR(xdst)) {
2180 		xfrm_pols_put(pols, num_pols);
2181 		return ERR_CAST(xdst);
2182 	}
2183 	xdst->num_pols = num_pols;
2184 	xdst->num_xfrms = num_xfrms;
2185 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2186 
2187 	dst_hold(&xdst->u.dst);
2188 	return &xdst->flo;
2189 
2190 inc_error:
2191 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2192 error:
2193 	if (xdst != NULL)
2194 		dst_free(&xdst->u.dst);
2195 	else
2196 		xfrm_pols_put(pols, num_pols);
2197 	return ERR_PTR(err);
2198 }
2199 
2200 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2201 					struct dst_entry *dst_orig)
2202 {
2203 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2204 	struct dst_entry *ret;
2205 
2206 	if (!afinfo) {
2207 		dst_release(dst_orig);
2208 		return ERR_PTR(-EINVAL);
2209 	} else {
2210 		ret = afinfo->blackhole_route(net, dst_orig);
2211 	}
2212 	rcu_read_unlock();
2213 
2214 	return ret;
2215 }
2216 
2217 /* Main function: finds/creates a bundle for given flow.
2218  *
2219  * At the moment we eat a raw IP route. Mostly to speed up lookups
2220  * on interfaces with disabled IPsec.
2221  */
2222 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2223 			      const struct flowi *fl,
2224 			      const struct sock *sk, int flags)
2225 {
2226 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2227 	struct flow_cache_object *flo;
2228 	struct xfrm_dst *xdst;
2229 	struct dst_entry *dst, *route;
2230 	u16 family = dst_orig->ops->family;
2231 	u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2232 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2233 
2234 	dst = NULL;
2235 	xdst = NULL;
2236 	route = NULL;
2237 
2238 	sk = sk_const_to_full_sk(sk);
2239 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2240 		num_pols = 1;
2241 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2242 		err = xfrm_expand_policies(fl, family, pols,
2243 					   &num_pols, &num_xfrms);
2244 		if (err < 0)
2245 			goto dropdst;
2246 
2247 		if (num_pols) {
2248 			if (num_xfrms <= 0) {
2249 				drop_pols = num_pols;
2250 				goto no_transform;
2251 			}
2252 
2253 			xdst = xfrm_resolve_and_create_bundle(
2254 					pols, num_pols, fl,
2255 					family, dst_orig);
2256 			if (IS_ERR(xdst)) {
2257 				xfrm_pols_put(pols, num_pols);
2258 				err = PTR_ERR(xdst);
2259 				goto dropdst;
2260 			} else if (xdst == NULL) {
2261 				num_xfrms = 0;
2262 				drop_pols = num_pols;
2263 				goto no_transform;
2264 			}
2265 
2266 			dst_hold(&xdst->u.dst);
2267 			xdst->u.dst.flags |= DST_NOCACHE;
2268 			route = xdst->route;
2269 		}
2270 	}
2271 
2272 	if (xdst == NULL) {
2273 		struct xfrm_flo xflo;
2274 
2275 		xflo.dst_orig = dst_orig;
2276 		xflo.flags = flags;
2277 
2278 		/* To accelerate a bit...  */
2279 		if ((dst_orig->flags & DST_NOXFRM) ||
2280 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
2281 			goto nopol;
2282 
2283 		flo = flow_cache_lookup(net, fl, family, dir,
2284 					xfrm_bundle_lookup, &xflo);
2285 		if (flo == NULL)
2286 			goto nopol;
2287 		if (IS_ERR(flo)) {
2288 			err = PTR_ERR(flo);
2289 			goto dropdst;
2290 		}
2291 		xdst = container_of(flo, struct xfrm_dst, flo);
2292 
2293 		num_pols = xdst->num_pols;
2294 		num_xfrms = xdst->num_xfrms;
2295 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2296 		route = xdst->route;
2297 	}
2298 
2299 	dst = &xdst->u.dst;
2300 	if (route == NULL && num_xfrms > 0) {
2301 		/* The only case when xfrm_bundle_lookup() returns a
2302 		 * bundle with null route, is when the template could
2303 		 * not be resolved. It means policies are there, but
2304 		 * bundle could not be created, since we don't yet
2305 		 * have the xfrm_state's. We need to wait for KM to
2306 		 * negotiate new SA's or bail out with error.*/
2307 		if (net->xfrm.sysctl_larval_drop) {
2308 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2309 			err = -EREMOTE;
2310 			goto error;
2311 		}
2312 
2313 		err = -EAGAIN;
2314 
2315 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2316 		goto error;
2317 	}
2318 
2319 no_transform:
2320 	if (num_pols == 0)
2321 		goto nopol;
2322 
2323 	if ((flags & XFRM_LOOKUP_ICMP) &&
2324 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2325 		err = -ENOENT;
2326 		goto error;
2327 	}
2328 
2329 	for (i = 0; i < num_pols; i++)
2330 		pols[i]->curlft.use_time = get_seconds();
2331 
2332 	if (num_xfrms < 0) {
2333 		/* Prohibit the flow */
2334 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2335 		err = -EPERM;
2336 		goto error;
2337 	} else if (num_xfrms > 0) {
2338 		/* Flow transformed */
2339 		dst_release(dst_orig);
2340 	} else {
2341 		/* Flow passes untransformed */
2342 		dst_release(dst);
2343 		dst = dst_orig;
2344 	}
2345 ok:
2346 	xfrm_pols_put(pols, drop_pols);
2347 	if (dst && dst->xfrm &&
2348 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2349 		dst->flags |= DST_XFRM_TUNNEL;
2350 	return dst;
2351 
2352 nopol:
2353 	if (!(flags & XFRM_LOOKUP_ICMP)) {
2354 		dst = dst_orig;
2355 		goto ok;
2356 	}
2357 	err = -ENOENT;
2358 error:
2359 	dst_release(dst);
2360 dropdst:
2361 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2362 		dst_release(dst_orig);
2363 	xfrm_pols_put(pols, drop_pols);
2364 	return ERR_PTR(err);
2365 }
2366 EXPORT_SYMBOL(xfrm_lookup);
2367 
2368 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2369  * Otherwise we may send out blackholed packets.
2370  */
2371 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2372 				    const struct flowi *fl,
2373 				    const struct sock *sk, int flags)
2374 {
2375 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2376 					    flags | XFRM_LOOKUP_QUEUE |
2377 					    XFRM_LOOKUP_KEEP_DST_REF);
2378 
2379 	if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2380 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
2381 
2382 	return dst;
2383 }
2384 EXPORT_SYMBOL(xfrm_lookup_route);
2385 
2386 static inline int
2387 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2388 {
2389 	struct xfrm_state *x;
2390 
2391 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2392 		return 0;
2393 	x = skb->sp->xvec[idx];
2394 	if (!x->type->reject)
2395 		return 0;
2396 	return x->type->reject(x, skb, fl);
2397 }
2398 
2399 /* When skb is transformed back to its "native" form, we have to
2400  * check policy restrictions. At the moment we make this in maximally
2401  * stupid way. Shame on me. :-) Of course, connected sockets must
2402  * have policy cached at them.
2403  */
2404 
2405 static inline int
2406 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2407 	      unsigned short family)
2408 {
2409 	if (xfrm_state_kern(x))
2410 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2411 	return	x->id.proto == tmpl->id.proto &&
2412 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2413 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2414 		x->props.mode == tmpl->mode &&
2415 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2416 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2417 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
2418 		  xfrm_state_addr_cmp(tmpl, x, family));
2419 }
2420 
2421 /*
2422  * 0 or more than 0 is returned when validation is succeeded (either bypass
2423  * because of optional transport mode, or next index of the mathced secpath
2424  * state with the template.
2425  * -1 is returned when no matching template is found.
2426  * Otherwise "-2 - errored_index" is returned.
2427  */
2428 static inline int
2429 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2430 	       unsigned short family)
2431 {
2432 	int idx = start;
2433 
2434 	if (tmpl->optional) {
2435 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
2436 			return start;
2437 	} else
2438 		start = -1;
2439 	for (; idx < sp->len; idx++) {
2440 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2441 			return ++idx;
2442 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2443 			if (start == -1)
2444 				start = -2-idx;
2445 			break;
2446 		}
2447 	}
2448 	return start;
2449 }
2450 
2451 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2452 			  unsigned int family, int reverse)
2453 {
2454 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2455 	int err;
2456 
2457 	if (unlikely(afinfo == NULL))
2458 		return -EAFNOSUPPORT;
2459 
2460 	afinfo->decode_session(skb, fl, reverse);
2461 	err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2462 	rcu_read_unlock();
2463 	return err;
2464 }
2465 EXPORT_SYMBOL(__xfrm_decode_session);
2466 
2467 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2468 {
2469 	for (; k < sp->len; k++) {
2470 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2471 			*idxp = k;
2472 			return 1;
2473 		}
2474 	}
2475 
2476 	return 0;
2477 }
2478 
2479 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2480 			unsigned short family)
2481 {
2482 	struct net *net = dev_net(skb->dev);
2483 	struct xfrm_policy *pol;
2484 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2485 	int npols = 0;
2486 	int xfrm_nr;
2487 	int pi;
2488 	int reverse;
2489 	struct flowi fl;
2490 	u8 fl_dir;
2491 	int xerr_idx = -1;
2492 
2493 	reverse = dir & ~XFRM_POLICY_MASK;
2494 	dir &= XFRM_POLICY_MASK;
2495 	fl_dir = policy_to_flow_dir(dir);
2496 
2497 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2498 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2499 		return 0;
2500 	}
2501 
2502 	nf_nat_decode_session(skb, &fl, family);
2503 
2504 	/* First, check used SA against their selectors. */
2505 	if (skb->sp) {
2506 		int i;
2507 
2508 		for (i = skb->sp->len-1; i >= 0; i--) {
2509 			struct xfrm_state *x = skb->sp->xvec[i];
2510 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
2511 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2512 				return 0;
2513 			}
2514 		}
2515 	}
2516 
2517 	pol = NULL;
2518 	sk = sk_to_full_sk(sk);
2519 	if (sk && sk->sk_policy[dir]) {
2520 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2521 		if (IS_ERR(pol)) {
2522 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2523 			return 0;
2524 		}
2525 	}
2526 
2527 	if (!pol) {
2528 		struct flow_cache_object *flo;
2529 
2530 		flo = flow_cache_lookup(net, &fl, family, fl_dir,
2531 					xfrm_policy_lookup, NULL);
2532 		if (IS_ERR_OR_NULL(flo))
2533 			pol = ERR_CAST(flo);
2534 		else
2535 			pol = container_of(flo, struct xfrm_policy, flo);
2536 	}
2537 
2538 	if (IS_ERR(pol)) {
2539 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2540 		return 0;
2541 	}
2542 
2543 	if (!pol) {
2544 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2545 			xfrm_secpath_reject(xerr_idx, skb, &fl);
2546 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2547 			return 0;
2548 		}
2549 		return 1;
2550 	}
2551 
2552 	pol->curlft.use_time = get_seconds();
2553 
2554 	pols[0] = pol;
2555 	npols++;
2556 #ifdef CONFIG_XFRM_SUB_POLICY
2557 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2558 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2559 						    &fl, family,
2560 						    XFRM_POLICY_IN);
2561 		if (pols[1]) {
2562 			if (IS_ERR(pols[1])) {
2563 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2564 				return 0;
2565 			}
2566 			pols[1]->curlft.use_time = get_seconds();
2567 			npols++;
2568 		}
2569 	}
2570 #endif
2571 
2572 	if (pol->action == XFRM_POLICY_ALLOW) {
2573 		struct sec_path *sp;
2574 		static struct sec_path dummy;
2575 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2576 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2577 		struct xfrm_tmpl **tpp = tp;
2578 		int ti = 0;
2579 		int i, k;
2580 
2581 		if ((sp = skb->sp) == NULL)
2582 			sp = &dummy;
2583 
2584 		for (pi = 0; pi < npols; pi++) {
2585 			if (pols[pi] != pol &&
2586 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
2587 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2588 				goto reject;
2589 			}
2590 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2591 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2592 				goto reject_error;
2593 			}
2594 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
2595 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
2596 		}
2597 		xfrm_nr = ti;
2598 		if (npols > 1) {
2599 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2600 			tpp = stp;
2601 		}
2602 
2603 		/* For each tunnel xfrm, find the first matching tmpl.
2604 		 * For each tmpl before that, find corresponding xfrm.
2605 		 * Order is _important_. Later we will implement
2606 		 * some barriers, but at the moment barriers
2607 		 * are implied between each two transformations.
2608 		 */
2609 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2610 			k = xfrm_policy_ok(tpp[i], sp, k, family);
2611 			if (k < 0) {
2612 				if (k < -1)
2613 					/* "-2 - errored_index" returned */
2614 					xerr_idx = -(2+k);
2615 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2616 				goto reject;
2617 			}
2618 		}
2619 
2620 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2621 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2622 			goto reject;
2623 		}
2624 
2625 		xfrm_pols_put(pols, npols);
2626 		return 1;
2627 	}
2628 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2629 
2630 reject:
2631 	xfrm_secpath_reject(xerr_idx, skb, &fl);
2632 reject_error:
2633 	xfrm_pols_put(pols, npols);
2634 	return 0;
2635 }
2636 EXPORT_SYMBOL(__xfrm_policy_check);
2637 
2638 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2639 {
2640 	struct net *net = dev_net(skb->dev);
2641 	struct flowi fl;
2642 	struct dst_entry *dst;
2643 	int res = 1;
2644 
2645 	if (xfrm_decode_session(skb, &fl, family) < 0) {
2646 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2647 		return 0;
2648 	}
2649 
2650 	skb_dst_force(skb);
2651 
2652 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2653 	if (IS_ERR(dst)) {
2654 		res = 0;
2655 		dst = NULL;
2656 	}
2657 	skb_dst_set(skb, dst);
2658 	return res;
2659 }
2660 EXPORT_SYMBOL(__xfrm_route_forward);
2661 
2662 /* Optimize later using cookies and generation ids. */
2663 
2664 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2665 {
2666 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2667 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2668 	 * get validated by dst_ops->check on every use.  We do this
2669 	 * because when a normal route referenced by an XFRM dst is
2670 	 * obsoleted we do not go looking around for all parent
2671 	 * referencing XFRM dsts so that we can invalidate them.  It
2672 	 * is just too much work.  Instead we make the checks here on
2673 	 * every use.  For example:
2674 	 *
2675 	 *	XFRM dst A --> IPv4 dst X
2676 	 *
2677 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
2678 	 * in this example).  If X is marked obsolete, "A" will not
2679 	 * notice.  That's what we are validating here via the
2680 	 * stale_bundle() check.
2681 	 *
2682 	 * When a policy's bundle is pruned, we dst_free() the XFRM
2683 	 * dst which causes it's ->obsolete field to be set to
2684 	 * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2685 	 * this, we want to force a new route lookup.
2686 	 */
2687 	if (dst->obsolete < 0 && !stale_bundle(dst))
2688 		return dst;
2689 
2690 	return NULL;
2691 }
2692 
2693 static int stale_bundle(struct dst_entry *dst)
2694 {
2695 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2696 }
2697 
2698 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2699 {
2700 	while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2701 		dst->dev = dev_net(dev)->loopback_dev;
2702 		dev_hold(dst->dev);
2703 		dev_put(dev);
2704 	}
2705 }
2706 EXPORT_SYMBOL(xfrm_dst_ifdown);
2707 
2708 static void xfrm_link_failure(struct sk_buff *skb)
2709 {
2710 	/* Impossible. Such dst must be popped before reaches point of failure. */
2711 }
2712 
2713 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2714 {
2715 	if (dst) {
2716 		if (dst->obsolete) {
2717 			dst_release(dst);
2718 			dst = NULL;
2719 		}
2720 	}
2721 	return dst;
2722 }
2723 
2724 void xfrm_garbage_collect(struct net *net)
2725 {
2726 	flow_cache_flush(net);
2727 }
2728 EXPORT_SYMBOL(xfrm_garbage_collect);
2729 
2730 void xfrm_garbage_collect_deferred(struct net *net)
2731 {
2732 	flow_cache_flush_deferred(net);
2733 }
2734 EXPORT_SYMBOL(xfrm_garbage_collect_deferred);
2735 
2736 static void xfrm_init_pmtu(struct dst_entry *dst)
2737 {
2738 	do {
2739 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2740 		u32 pmtu, route_mtu_cached;
2741 
2742 		pmtu = dst_mtu(dst->child);
2743 		xdst->child_mtu_cached = pmtu;
2744 
2745 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2746 
2747 		route_mtu_cached = dst_mtu(xdst->route);
2748 		xdst->route_mtu_cached = route_mtu_cached;
2749 
2750 		if (pmtu > route_mtu_cached)
2751 			pmtu = route_mtu_cached;
2752 
2753 		dst_metric_set(dst, RTAX_MTU, pmtu);
2754 	} while ((dst = dst->next));
2755 }
2756 
2757 /* Check that the bundle accepts the flow and its components are
2758  * still valid.
2759  */
2760 
2761 static int xfrm_bundle_ok(struct xfrm_dst *first)
2762 {
2763 	struct dst_entry *dst = &first->u.dst;
2764 	struct xfrm_dst *last;
2765 	u32 mtu;
2766 
2767 	if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2768 	    (dst->dev && !netif_running(dst->dev)))
2769 		return 0;
2770 
2771 	if (dst->flags & DST_XFRM_QUEUE)
2772 		return 1;
2773 
2774 	last = NULL;
2775 
2776 	do {
2777 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2778 
2779 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2780 			return 0;
2781 		if (xdst->xfrm_genid != dst->xfrm->genid)
2782 			return 0;
2783 		if (xdst->num_pols > 0 &&
2784 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2785 			return 0;
2786 
2787 		mtu = dst_mtu(dst->child);
2788 		if (xdst->child_mtu_cached != mtu) {
2789 			last = xdst;
2790 			xdst->child_mtu_cached = mtu;
2791 		}
2792 
2793 		if (!dst_check(xdst->route, xdst->route_cookie))
2794 			return 0;
2795 		mtu = dst_mtu(xdst->route);
2796 		if (xdst->route_mtu_cached != mtu) {
2797 			last = xdst;
2798 			xdst->route_mtu_cached = mtu;
2799 		}
2800 
2801 		dst = dst->child;
2802 	} while (dst->xfrm);
2803 
2804 	if (likely(!last))
2805 		return 1;
2806 
2807 	mtu = last->child_mtu_cached;
2808 	for (;;) {
2809 		dst = &last->u.dst;
2810 
2811 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2812 		if (mtu > last->route_mtu_cached)
2813 			mtu = last->route_mtu_cached;
2814 		dst_metric_set(dst, RTAX_MTU, mtu);
2815 
2816 		if (last == first)
2817 			break;
2818 
2819 		last = (struct xfrm_dst *)last->u.dst.next;
2820 		last->child_mtu_cached = mtu;
2821 	}
2822 
2823 	return 1;
2824 }
2825 
2826 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2827 {
2828 	return dst_metric_advmss(dst->path);
2829 }
2830 
2831 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2832 {
2833 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2834 
2835 	return mtu ? : dst_mtu(dst->path);
2836 }
2837 
2838 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2839 					const void *daddr)
2840 {
2841 	const struct dst_entry *path = dst->path;
2842 
2843 	for (; dst != path; dst = dst->child) {
2844 		const struct xfrm_state *xfrm = dst->xfrm;
2845 
2846 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2847 			continue;
2848 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2849 			daddr = xfrm->coaddr;
2850 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2851 			daddr = &xfrm->id.daddr;
2852 	}
2853 	return daddr;
2854 }
2855 
2856 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2857 					   struct sk_buff *skb,
2858 					   const void *daddr)
2859 {
2860 	const struct dst_entry *path = dst->path;
2861 
2862 	if (!skb)
2863 		daddr = xfrm_get_dst_nexthop(dst, daddr);
2864 	return path->ops->neigh_lookup(path, skb, daddr);
2865 }
2866 
2867 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2868 {
2869 	const struct dst_entry *path = dst->path;
2870 
2871 	daddr = xfrm_get_dst_nexthop(dst, daddr);
2872 	path->ops->confirm_neigh(path, daddr);
2873 }
2874 
2875 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2876 {
2877 	int err = 0;
2878 
2879 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2880 		return -EAFNOSUPPORT;
2881 
2882 	spin_lock(&xfrm_policy_afinfo_lock);
2883 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
2884 		err = -EEXIST;
2885 	else {
2886 		struct dst_ops *dst_ops = afinfo->dst_ops;
2887 		if (likely(dst_ops->kmem_cachep == NULL))
2888 			dst_ops->kmem_cachep = xfrm_dst_cache;
2889 		if (likely(dst_ops->check == NULL))
2890 			dst_ops->check = xfrm_dst_check;
2891 		if (likely(dst_ops->default_advmss == NULL))
2892 			dst_ops->default_advmss = xfrm_default_advmss;
2893 		if (likely(dst_ops->mtu == NULL))
2894 			dst_ops->mtu = xfrm_mtu;
2895 		if (likely(dst_ops->negative_advice == NULL))
2896 			dst_ops->negative_advice = xfrm_negative_advice;
2897 		if (likely(dst_ops->link_failure == NULL))
2898 			dst_ops->link_failure = xfrm_link_failure;
2899 		if (likely(dst_ops->neigh_lookup == NULL))
2900 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
2901 		if (likely(!dst_ops->confirm_neigh))
2902 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
2903 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2904 	}
2905 	spin_unlock(&xfrm_policy_afinfo_lock);
2906 
2907 	return err;
2908 }
2909 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2910 
2911 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2912 {
2913 	struct dst_ops *dst_ops = afinfo->dst_ops;
2914 	int i;
2915 
2916 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2917 		if (xfrm_policy_afinfo[i] != afinfo)
2918 			continue;
2919 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2920 		break;
2921 	}
2922 
2923 	synchronize_rcu();
2924 
2925 	dst_ops->kmem_cachep = NULL;
2926 	dst_ops->check = NULL;
2927 	dst_ops->negative_advice = NULL;
2928 	dst_ops->link_failure = NULL;
2929 }
2930 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2931 
2932 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2933 {
2934 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2935 
2936 	switch (event) {
2937 	case NETDEV_DOWN:
2938 		xfrm_garbage_collect(dev_net(dev));
2939 	}
2940 	return NOTIFY_DONE;
2941 }
2942 
2943 static struct notifier_block xfrm_dev_notifier = {
2944 	.notifier_call	= xfrm_dev_event,
2945 };
2946 
2947 #ifdef CONFIG_XFRM_STATISTICS
2948 static int __net_init xfrm_statistics_init(struct net *net)
2949 {
2950 	int rv;
2951 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2952 	if (!net->mib.xfrm_statistics)
2953 		return -ENOMEM;
2954 	rv = xfrm_proc_init(net);
2955 	if (rv < 0)
2956 		free_percpu(net->mib.xfrm_statistics);
2957 	return rv;
2958 }
2959 
2960 static void xfrm_statistics_fini(struct net *net)
2961 {
2962 	xfrm_proc_fini(net);
2963 	free_percpu(net->mib.xfrm_statistics);
2964 }
2965 #else
2966 static int __net_init xfrm_statistics_init(struct net *net)
2967 {
2968 	return 0;
2969 }
2970 
2971 static void xfrm_statistics_fini(struct net *net)
2972 {
2973 }
2974 #endif
2975 
2976 static int __net_init xfrm_policy_init(struct net *net)
2977 {
2978 	unsigned int hmask, sz;
2979 	int dir;
2980 
2981 	if (net_eq(net, &init_net))
2982 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2983 					   sizeof(struct xfrm_dst),
2984 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2985 					   NULL);
2986 
2987 	hmask = 8 - 1;
2988 	sz = (hmask+1) * sizeof(struct hlist_head);
2989 
2990 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2991 	if (!net->xfrm.policy_byidx)
2992 		goto out_byidx;
2993 	net->xfrm.policy_idx_hmask = hmask;
2994 
2995 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2996 		struct xfrm_policy_hash *htab;
2997 
2998 		net->xfrm.policy_count[dir] = 0;
2999 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
3000 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
3001 
3002 		htab = &net->xfrm.policy_bydst[dir];
3003 		htab->table = xfrm_hash_alloc(sz);
3004 		if (!htab->table)
3005 			goto out_bydst;
3006 		htab->hmask = hmask;
3007 		htab->dbits4 = 32;
3008 		htab->sbits4 = 32;
3009 		htab->dbits6 = 128;
3010 		htab->sbits6 = 128;
3011 	}
3012 	net->xfrm.policy_hthresh.lbits4 = 32;
3013 	net->xfrm.policy_hthresh.rbits4 = 32;
3014 	net->xfrm.policy_hthresh.lbits6 = 128;
3015 	net->xfrm.policy_hthresh.rbits6 = 128;
3016 
3017 	seqlock_init(&net->xfrm.policy_hthresh.lock);
3018 
3019 	INIT_LIST_HEAD(&net->xfrm.policy_all);
3020 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
3021 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
3022 	if (net_eq(net, &init_net))
3023 		register_netdevice_notifier(&xfrm_dev_notifier);
3024 	return 0;
3025 
3026 out_bydst:
3027 	for (dir--; dir >= 0; dir--) {
3028 		struct xfrm_policy_hash *htab;
3029 
3030 		htab = &net->xfrm.policy_bydst[dir];
3031 		xfrm_hash_free(htab->table, sz);
3032 	}
3033 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
3034 out_byidx:
3035 	return -ENOMEM;
3036 }
3037 
3038 static void xfrm_policy_fini(struct net *net)
3039 {
3040 	unsigned int sz;
3041 	int dir;
3042 
3043 	flush_work(&net->xfrm.policy_hash_work);
3044 #ifdef CONFIG_XFRM_SUB_POLICY
3045 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3046 #endif
3047 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3048 
3049 	WARN_ON(!list_empty(&net->xfrm.policy_all));
3050 
3051 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3052 		struct xfrm_policy_hash *htab;
3053 
3054 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3055 
3056 		htab = &net->xfrm.policy_bydst[dir];
3057 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3058 		WARN_ON(!hlist_empty(htab->table));
3059 		xfrm_hash_free(htab->table, sz);
3060 	}
3061 
3062 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3063 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3064 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
3065 }
3066 
3067 static int __net_init xfrm_net_init(struct net *net)
3068 {
3069 	int rv;
3070 
3071 	/* Initialize the per-net locks here */
3072 	spin_lock_init(&net->xfrm.xfrm_state_lock);
3073 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
3074 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
3075 
3076 	rv = xfrm_statistics_init(net);
3077 	if (rv < 0)
3078 		goto out_statistics;
3079 	rv = xfrm_state_init(net);
3080 	if (rv < 0)
3081 		goto out_state;
3082 	rv = xfrm_policy_init(net);
3083 	if (rv < 0)
3084 		goto out_policy;
3085 	rv = xfrm_sysctl_init(net);
3086 	if (rv < 0)
3087 		goto out_sysctl;
3088 	rv = flow_cache_init(net);
3089 	if (rv < 0)
3090 		goto out;
3091 
3092 	return 0;
3093 
3094 out:
3095 	xfrm_sysctl_fini(net);
3096 out_sysctl:
3097 	xfrm_policy_fini(net);
3098 out_policy:
3099 	xfrm_state_fini(net);
3100 out_state:
3101 	xfrm_statistics_fini(net);
3102 out_statistics:
3103 	return rv;
3104 }
3105 
3106 static void __net_exit xfrm_net_exit(struct net *net)
3107 {
3108 	flow_cache_fini(net);
3109 	xfrm_sysctl_fini(net);
3110 	xfrm_policy_fini(net);
3111 	xfrm_state_fini(net);
3112 	xfrm_statistics_fini(net);
3113 }
3114 
3115 static struct pernet_operations __net_initdata xfrm_net_ops = {
3116 	.init = xfrm_net_init,
3117 	.exit = xfrm_net_exit,
3118 };
3119 
3120 void __init xfrm_init(void)
3121 {
3122 	flow_cache_hp_init();
3123 	register_pernet_subsys(&xfrm_net_ops);
3124 	seqcount_init(&xfrm_policy_hash_generation);
3125 	xfrm_input_init();
3126 }
3127 
3128 #ifdef CONFIG_AUDITSYSCALL
3129 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3130 					 struct audit_buffer *audit_buf)
3131 {
3132 	struct xfrm_sec_ctx *ctx = xp->security;
3133 	struct xfrm_selector *sel = &xp->selector;
3134 
3135 	if (ctx)
3136 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3137 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3138 
3139 	switch (sel->family) {
3140 	case AF_INET:
3141 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3142 		if (sel->prefixlen_s != 32)
3143 			audit_log_format(audit_buf, " src_prefixlen=%d",
3144 					 sel->prefixlen_s);
3145 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3146 		if (sel->prefixlen_d != 32)
3147 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3148 					 sel->prefixlen_d);
3149 		break;
3150 	case AF_INET6:
3151 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3152 		if (sel->prefixlen_s != 128)
3153 			audit_log_format(audit_buf, " src_prefixlen=%d",
3154 					 sel->prefixlen_s);
3155 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3156 		if (sel->prefixlen_d != 128)
3157 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3158 					 sel->prefixlen_d);
3159 		break;
3160 	}
3161 }
3162 
3163 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3164 {
3165 	struct audit_buffer *audit_buf;
3166 
3167 	audit_buf = xfrm_audit_start("SPD-add");
3168 	if (audit_buf == NULL)
3169 		return;
3170 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3171 	audit_log_format(audit_buf, " res=%u", result);
3172 	xfrm_audit_common_policyinfo(xp, audit_buf);
3173 	audit_log_end(audit_buf);
3174 }
3175 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3176 
3177 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3178 			      bool task_valid)
3179 {
3180 	struct audit_buffer *audit_buf;
3181 
3182 	audit_buf = xfrm_audit_start("SPD-delete");
3183 	if (audit_buf == NULL)
3184 		return;
3185 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3186 	audit_log_format(audit_buf, " res=%u", result);
3187 	xfrm_audit_common_policyinfo(xp, audit_buf);
3188 	audit_log_end(audit_buf);
3189 }
3190 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3191 #endif
3192 
3193 #ifdef CONFIG_XFRM_MIGRATE
3194 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3195 					const struct xfrm_selector *sel_tgt)
3196 {
3197 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3198 		if (sel_tgt->family == sel_cmp->family &&
3199 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3200 				    sel_cmp->family) &&
3201 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3202 				    sel_cmp->family) &&
3203 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3204 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3205 			return true;
3206 		}
3207 	} else {
3208 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3209 			return true;
3210 		}
3211 	}
3212 	return false;
3213 }
3214 
3215 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3216 						    u8 dir, u8 type, struct net *net)
3217 {
3218 	struct xfrm_policy *pol, *ret = NULL;
3219 	struct hlist_head *chain;
3220 	u32 priority = ~0U;
3221 
3222 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3223 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3224 	hlist_for_each_entry(pol, chain, bydst) {
3225 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3226 		    pol->type == type) {
3227 			ret = pol;
3228 			priority = ret->priority;
3229 			break;
3230 		}
3231 	}
3232 	chain = &net->xfrm.policy_inexact[dir];
3233 	hlist_for_each_entry(pol, chain, bydst) {
3234 		if ((pol->priority >= priority) && ret)
3235 			break;
3236 
3237 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3238 		    pol->type == type) {
3239 			ret = pol;
3240 			break;
3241 		}
3242 	}
3243 
3244 	xfrm_pol_hold(ret);
3245 
3246 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3247 
3248 	return ret;
3249 }
3250 
3251 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3252 {
3253 	int match = 0;
3254 
3255 	if (t->mode == m->mode && t->id.proto == m->proto &&
3256 	    (m->reqid == 0 || t->reqid == m->reqid)) {
3257 		switch (t->mode) {
3258 		case XFRM_MODE_TUNNEL:
3259 		case XFRM_MODE_BEET:
3260 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3261 					    m->old_family) &&
3262 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
3263 					    m->old_family)) {
3264 				match = 1;
3265 			}
3266 			break;
3267 		case XFRM_MODE_TRANSPORT:
3268 			/* in case of transport mode, template does not store
3269 			   any IP addresses, hence we just compare mode and
3270 			   protocol */
3271 			match = 1;
3272 			break;
3273 		default:
3274 			break;
3275 		}
3276 	}
3277 	return match;
3278 }
3279 
3280 /* update endpoint address(es) of template(s) */
3281 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3282 			       struct xfrm_migrate *m, int num_migrate)
3283 {
3284 	struct xfrm_migrate *mp;
3285 	int i, j, n = 0;
3286 
3287 	write_lock_bh(&pol->lock);
3288 	if (unlikely(pol->walk.dead)) {
3289 		/* target policy has been deleted */
3290 		write_unlock_bh(&pol->lock);
3291 		return -ENOENT;
3292 	}
3293 
3294 	for (i = 0; i < pol->xfrm_nr; i++) {
3295 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3296 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3297 				continue;
3298 			n++;
3299 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3300 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3301 				continue;
3302 			/* update endpoints */
3303 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3304 			       sizeof(pol->xfrm_vec[i].id.daddr));
3305 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3306 			       sizeof(pol->xfrm_vec[i].saddr));
3307 			pol->xfrm_vec[i].encap_family = mp->new_family;
3308 			/* flush bundles */
3309 			atomic_inc(&pol->genid);
3310 		}
3311 	}
3312 
3313 	write_unlock_bh(&pol->lock);
3314 
3315 	if (!n)
3316 		return -ENODATA;
3317 
3318 	return 0;
3319 }
3320 
3321 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3322 {
3323 	int i, j;
3324 
3325 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3326 		return -EINVAL;
3327 
3328 	for (i = 0; i < num_migrate; i++) {
3329 		if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3330 				    m[i].old_family) &&
3331 		    xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3332 				    m[i].old_family))
3333 			return -EINVAL;
3334 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3335 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3336 			return -EINVAL;
3337 
3338 		/* check if there is any duplicated entry */
3339 		for (j = i + 1; j < num_migrate; j++) {
3340 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3341 				    sizeof(m[i].old_daddr)) &&
3342 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3343 				    sizeof(m[i].old_saddr)) &&
3344 			    m[i].proto == m[j].proto &&
3345 			    m[i].mode == m[j].mode &&
3346 			    m[i].reqid == m[j].reqid &&
3347 			    m[i].old_family == m[j].old_family)
3348 				return -EINVAL;
3349 		}
3350 	}
3351 
3352 	return 0;
3353 }
3354 
3355 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3356 		 struct xfrm_migrate *m, int num_migrate,
3357 		 struct xfrm_kmaddress *k, struct net *net)
3358 {
3359 	int i, err, nx_cur = 0, nx_new = 0;
3360 	struct xfrm_policy *pol = NULL;
3361 	struct xfrm_state *x, *xc;
3362 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3363 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3364 	struct xfrm_migrate *mp;
3365 
3366 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3367 		goto out;
3368 
3369 	/* Stage 1 - find policy */
3370 	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3371 		err = -ENOENT;
3372 		goto out;
3373 	}
3374 
3375 	/* Stage 2 - find and update state(s) */
3376 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3377 		if ((x = xfrm_migrate_state_find(mp, net))) {
3378 			x_cur[nx_cur] = x;
3379 			nx_cur++;
3380 			if ((xc = xfrm_state_migrate(x, mp))) {
3381 				x_new[nx_new] = xc;
3382 				nx_new++;
3383 			} else {
3384 				err = -ENODATA;
3385 				goto restore_state;
3386 			}
3387 		}
3388 	}
3389 
3390 	/* Stage 3 - update policy */
3391 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3392 		goto restore_state;
3393 
3394 	/* Stage 4 - delete old state(s) */
3395 	if (nx_cur) {
3396 		xfrm_states_put(x_cur, nx_cur);
3397 		xfrm_states_delete(x_cur, nx_cur);
3398 	}
3399 
3400 	/* Stage 5 - announce */
3401 	km_migrate(sel, dir, type, m, num_migrate, k);
3402 
3403 	xfrm_pol_put(pol);
3404 
3405 	return 0;
3406 out:
3407 	return err;
3408 
3409 restore_state:
3410 	if (pol)
3411 		xfrm_pol_put(pol);
3412 	if (nx_cur)
3413 		xfrm_states_put(x_cur, nx_cur);
3414 	if (nx_new)
3415 		xfrm_states_delete(x_new, nx_new);
3416 
3417 	return err;
3418 }
3419 EXPORT_SYMBOL(xfrm_migrate);
3420 #endif
3421