xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision f79e4d5f92a129a1159c973735007d4ddc8541f3)
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/cpu.h>
28 #include <linux/audit.h>
29 #include <net/dst.h>
30 #include <net/flow.h>
31 #include <net/xfrm.h>
32 #include <net/ip.h>
33 #ifdef CONFIG_XFRM_STATISTICS
34 #include <net/snmp.h>
35 #endif
36 
37 #include "xfrm_hash.h"
38 
39 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
40 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
41 #define XFRM_MAX_QUEUE_LEN	100
42 
43 struct xfrm_flo {
44 	struct dst_entry *dst_orig;
45 	u8 flags;
46 };
47 
48 static DEFINE_PER_CPU(struct xfrm_dst *, xfrm_last_dst);
49 static struct work_struct *xfrm_pcpu_work __read_mostly;
50 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
51 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
52 						__read_mostly;
53 
54 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
55 static __read_mostly seqcount_t xfrm_policy_hash_generation;
56 
57 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
58 static int stale_bundle(struct dst_entry *dst);
59 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
60 static void xfrm_policy_queue_process(struct timer_list *t);
61 
62 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
63 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
64 						int dir);
65 
66 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
67 {
68 	return refcount_inc_not_zero(&policy->refcnt);
69 }
70 
71 static inline bool
72 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
73 {
74 	const struct flowi4 *fl4 = &fl->u.ip4;
75 
76 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
77 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
78 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
79 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
80 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
81 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
82 }
83 
84 static inline bool
85 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
86 {
87 	const struct flowi6 *fl6 = &fl->u.ip6;
88 
89 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
90 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
91 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
92 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
93 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
94 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
95 }
96 
97 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
98 			 unsigned short family)
99 {
100 	switch (family) {
101 	case AF_INET:
102 		return __xfrm4_selector_match(sel, fl);
103 	case AF_INET6:
104 		return __xfrm6_selector_match(sel, fl);
105 	}
106 	return false;
107 }
108 
109 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
110 {
111 	const struct xfrm_policy_afinfo *afinfo;
112 
113 	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
114 		return NULL;
115 	rcu_read_lock();
116 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
117 	if (unlikely(!afinfo))
118 		rcu_read_unlock();
119 	return afinfo;
120 }
121 
122 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
123 				    const xfrm_address_t *saddr,
124 				    const xfrm_address_t *daddr,
125 				    int family, u32 mark)
126 {
127 	const struct xfrm_policy_afinfo *afinfo;
128 	struct dst_entry *dst;
129 
130 	afinfo = xfrm_policy_get_afinfo(family);
131 	if (unlikely(afinfo == NULL))
132 		return ERR_PTR(-EAFNOSUPPORT);
133 
134 	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
135 
136 	rcu_read_unlock();
137 
138 	return dst;
139 }
140 EXPORT_SYMBOL(__xfrm_dst_lookup);
141 
142 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
143 						int tos, int oif,
144 						xfrm_address_t *prev_saddr,
145 						xfrm_address_t *prev_daddr,
146 						int family, u32 mark)
147 {
148 	struct net *net = xs_net(x);
149 	xfrm_address_t *saddr = &x->props.saddr;
150 	xfrm_address_t *daddr = &x->id.daddr;
151 	struct dst_entry *dst;
152 
153 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
154 		saddr = x->coaddr;
155 		daddr = prev_daddr;
156 	}
157 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
158 		saddr = prev_saddr;
159 		daddr = x->coaddr;
160 	}
161 
162 	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
163 
164 	if (!IS_ERR(dst)) {
165 		if (prev_saddr != saddr)
166 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
167 		if (prev_daddr != daddr)
168 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
169 	}
170 
171 	return dst;
172 }
173 
174 static inline unsigned long make_jiffies(long secs)
175 {
176 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
177 		return MAX_SCHEDULE_TIMEOUT-1;
178 	else
179 		return secs*HZ;
180 }
181 
182 static void xfrm_policy_timer(struct timer_list *t)
183 {
184 	struct xfrm_policy *xp = from_timer(xp, t, timer);
185 	unsigned long now = get_seconds();
186 	long next = LONG_MAX;
187 	int warn = 0;
188 	int dir;
189 
190 	read_lock(&xp->lock);
191 
192 	if (unlikely(xp->walk.dead))
193 		goto out;
194 
195 	dir = xfrm_policy_id2dir(xp->index);
196 
197 	if (xp->lft.hard_add_expires_seconds) {
198 		long tmo = xp->lft.hard_add_expires_seconds +
199 			xp->curlft.add_time - now;
200 		if (tmo <= 0)
201 			goto expired;
202 		if (tmo < next)
203 			next = tmo;
204 	}
205 	if (xp->lft.hard_use_expires_seconds) {
206 		long tmo = xp->lft.hard_use_expires_seconds +
207 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
208 		if (tmo <= 0)
209 			goto expired;
210 		if (tmo < next)
211 			next = tmo;
212 	}
213 	if (xp->lft.soft_add_expires_seconds) {
214 		long tmo = xp->lft.soft_add_expires_seconds +
215 			xp->curlft.add_time - now;
216 		if (tmo <= 0) {
217 			warn = 1;
218 			tmo = XFRM_KM_TIMEOUT;
219 		}
220 		if (tmo < next)
221 			next = tmo;
222 	}
223 	if (xp->lft.soft_use_expires_seconds) {
224 		long tmo = xp->lft.soft_use_expires_seconds +
225 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
226 		if (tmo <= 0) {
227 			warn = 1;
228 			tmo = XFRM_KM_TIMEOUT;
229 		}
230 		if (tmo < next)
231 			next = tmo;
232 	}
233 
234 	if (warn)
235 		km_policy_expired(xp, dir, 0, 0);
236 	if (next != LONG_MAX &&
237 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
238 		xfrm_pol_hold(xp);
239 
240 out:
241 	read_unlock(&xp->lock);
242 	xfrm_pol_put(xp);
243 	return;
244 
245 expired:
246 	read_unlock(&xp->lock);
247 	if (!xfrm_policy_delete(xp, dir))
248 		km_policy_expired(xp, dir, 1, 0);
249 	xfrm_pol_put(xp);
250 }
251 
252 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
253  * SPD calls.
254  */
255 
256 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
257 {
258 	struct xfrm_policy *policy;
259 
260 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
261 
262 	if (policy) {
263 		write_pnet(&policy->xp_net, net);
264 		INIT_LIST_HEAD(&policy->walk.all);
265 		INIT_HLIST_NODE(&policy->bydst);
266 		INIT_HLIST_NODE(&policy->byidx);
267 		rwlock_init(&policy->lock);
268 		refcount_set(&policy->refcnt, 1);
269 		skb_queue_head_init(&policy->polq.hold_queue);
270 		timer_setup(&policy->timer, xfrm_policy_timer, 0);
271 		timer_setup(&policy->polq.hold_timer,
272 			    xfrm_policy_queue_process, 0);
273 	}
274 	return policy;
275 }
276 EXPORT_SYMBOL(xfrm_policy_alloc);
277 
278 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
279 {
280 	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
281 
282 	security_xfrm_policy_free(policy->security);
283 	kfree(policy);
284 }
285 
286 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
287 
288 void xfrm_policy_destroy(struct xfrm_policy *policy)
289 {
290 	BUG_ON(!policy->walk.dead);
291 
292 	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
293 		BUG();
294 
295 	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
296 }
297 EXPORT_SYMBOL(xfrm_policy_destroy);
298 
299 /* Rule must be locked. Release descendant resources, announce
300  * entry dead. The rule must be unlinked from lists to the moment.
301  */
302 
303 static void xfrm_policy_kill(struct xfrm_policy *policy)
304 {
305 	policy->walk.dead = 1;
306 
307 	atomic_inc(&policy->genid);
308 
309 	if (del_timer(&policy->polq.hold_timer))
310 		xfrm_pol_put(policy);
311 	skb_queue_purge(&policy->polq.hold_queue);
312 
313 	if (del_timer(&policy->timer))
314 		xfrm_pol_put(policy);
315 
316 	xfrm_pol_put(policy);
317 }
318 
319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
320 
321 static inline unsigned int idx_hash(struct net *net, u32 index)
322 {
323 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
324 }
325 
326 /* calculate policy hash thresholds */
327 static void __get_hash_thresh(struct net *net,
328 			      unsigned short family, int dir,
329 			      u8 *dbits, u8 *sbits)
330 {
331 	switch (family) {
332 	case AF_INET:
333 		*dbits = net->xfrm.policy_bydst[dir].dbits4;
334 		*sbits = net->xfrm.policy_bydst[dir].sbits4;
335 		break;
336 
337 	case AF_INET6:
338 		*dbits = net->xfrm.policy_bydst[dir].dbits6;
339 		*sbits = net->xfrm.policy_bydst[dir].sbits6;
340 		break;
341 
342 	default:
343 		*dbits = 0;
344 		*sbits = 0;
345 	}
346 }
347 
348 static struct hlist_head *policy_hash_bysel(struct net *net,
349 					    const struct xfrm_selector *sel,
350 					    unsigned short family, int dir)
351 {
352 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
353 	unsigned int hash;
354 	u8 dbits;
355 	u8 sbits;
356 
357 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
358 	hash = __sel_hash(sel, family, hmask, dbits, sbits);
359 
360 	if (hash == hmask + 1)
361 		return &net->xfrm.policy_inexact[dir];
362 
363 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
364 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
365 }
366 
367 static struct hlist_head *policy_hash_direct(struct net *net,
368 					     const xfrm_address_t *daddr,
369 					     const xfrm_address_t *saddr,
370 					     unsigned short family, int dir)
371 {
372 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
373 	unsigned int hash;
374 	u8 dbits;
375 	u8 sbits;
376 
377 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
378 	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
379 
380 	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
381 		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
382 }
383 
384 static void xfrm_dst_hash_transfer(struct net *net,
385 				   struct hlist_head *list,
386 				   struct hlist_head *ndsttable,
387 				   unsigned int nhashmask,
388 				   int dir)
389 {
390 	struct hlist_node *tmp, *entry0 = NULL;
391 	struct xfrm_policy *pol;
392 	unsigned int h0 = 0;
393 	u8 dbits;
394 	u8 sbits;
395 
396 redo:
397 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
398 		unsigned int h;
399 
400 		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
401 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
402 				pol->family, nhashmask, dbits, sbits);
403 		if (!entry0) {
404 			hlist_del_rcu(&pol->bydst);
405 			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
406 			h0 = h;
407 		} else {
408 			if (h != h0)
409 				continue;
410 			hlist_del_rcu(&pol->bydst);
411 			hlist_add_behind_rcu(&pol->bydst, entry0);
412 		}
413 		entry0 = &pol->bydst;
414 	}
415 	if (!hlist_empty(list)) {
416 		entry0 = NULL;
417 		goto redo;
418 	}
419 }
420 
421 static void xfrm_idx_hash_transfer(struct hlist_head *list,
422 				   struct hlist_head *nidxtable,
423 				   unsigned int nhashmask)
424 {
425 	struct hlist_node *tmp;
426 	struct xfrm_policy *pol;
427 
428 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
429 		unsigned int h;
430 
431 		h = __idx_hash(pol->index, nhashmask);
432 		hlist_add_head(&pol->byidx, nidxtable+h);
433 	}
434 }
435 
436 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
437 {
438 	return ((old_hmask + 1) << 1) - 1;
439 }
440 
441 static void xfrm_bydst_resize(struct net *net, int dir)
442 {
443 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
444 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
445 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
446 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
447 	struct hlist_head *odst;
448 	int i;
449 
450 	if (!ndst)
451 		return;
452 
453 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
454 	write_seqcount_begin(&xfrm_policy_hash_generation);
455 
456 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
457 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
458 
459 	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
460 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
461 
462 	for (i = hmask; i >= 0; i--)
463 		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
464 
465 	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
466 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
467 
468 	write_seqcount_end(&xfrm_policy_hash_generation);
469 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
470 
471 	synchronize_rcu();
472 
473 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
474 }
475 
476 static void xfrm_byidx_resize(struct net *net, int total)
477 {
478 	unsigned int hmask = net->xfrm.policy_idx_hmask;
479 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
480 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
481 	struct hlist_head *oidx = net->xfrm.policy_byidx;
482 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
483 	int i;
484 
485 	if (!nidx)
486 		return;
487 
488 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
489 
490 	for (i = hmask; i >= 0; i--)
491 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
492 
493 	net->xfrm.policy_byidx = nidx;
494 	net->xfrm.policy_idx_hmask = nhashmask;
495 
496 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
497 
498 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
499 }
500 
501 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
502 {
503 	unsigned int cnt = net->xfrm.policy_count[dir];
504 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
505 
506 	if (total)
507 		*total += cnt;
508 
509 	if ((hmask + 1) < xfrm_policy_hashmax &&
510 	    cnt > hmask)
511 		return 1;
512 
513 	return 0;
514 }
515 
516 static inline int xfrm_byidx_should_resize(struct net *net, int total)
517 {
518 	unsigned int hmask = net->xfrm.policy_idx_hmask;
519 
520 	if ((hmask + 1) < xfrm_policy_hashmax &&
521 	    total > hmask)
522 		return 1;
523 
524 	return 0;
525 }
526 
527 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
528 {
529 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
530 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
531 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
532 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
533 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
534 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
535 	si->spdhcnt = net->xfrm.policy_idx_hmask;
536 	si->spdhmcnt = xfrm_policy_hashmax;
537 }
538 EXPORT_SYMBOL(xfrm_spd_getinfo);
539 
540 static DEFINE_MUTEX(hash_resize_mutex);
541 static void xfrm_hash_resize(struct work_struct *work)
542 {
543 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
544 	int dir, total;
545 
546 	mutex_lock(&hash_resize_mutex);
547 
548 	total = 0;
549 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
550 		if (xfrm_bydst_should_resize(net, dir, &total))
551 			xfrm_bydst_resize(net, dir);
552 	}
553 	if (xfrm_byidx_should_resize(net, total))
554 		xfrm_byidx_resize(net, total);
555 
556 	mutex_unlock(&hash_resize_mutex);
557 }
558 
559 static void xfrm_hash_rebuild(struct work_struct *work)
560 {
561 	struct net *net = container_of(work, struct net,
562 				       xfrm.policy_hthresh.work);
563 	unsigned int hmask;
564 	struct xfrm_policy *pol;
565 	struct xfrm_policy *policy;
566 	struct hlist_head *chain;
567 	struct hlist_head *odst;
568 	struct hlist_node *newpos;
569 	int i;
570 	int dir;
571 	unsigned seq;
572 	u8 lbits4, rbits4, lbits6, rbits6;
573 
574 	mutex_lock(&hash_resize_mutex);
575 
576 	/* read selector prefixlen thresholds */
577 	do {
578 		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
579 
580 		lbits4 = net->xfrm.policy_hthresh.lbits4;
581 		rbits4 = net->xfrm.policy_hthresh.rbits4;
582 		lbits6 = net->xfrm.policy_hthresh.lbits6;
583 		rbits6 = net->xfrm.policy_hthresh.rbits6;
584 	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
585 
586 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
587 
588 	/* reset the bydst and inexact table in all directions */
589 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
590 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
591 		hmask = net->xfrm.policy_bydst[dir].hmask;
592 		odst = net->xfrm.policy_bydst[dir].table;
593 		for (i = hmask; i >= 0; i--)
594 			INIT_HLIST_HEAD(odst + i);
595 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
596 			/* dir out => dst = remote, src = local */
597 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
598 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
599 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
600 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
601 		} else {
602 			/* dir in/fwd => dst = local, src = remote */
603 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
604 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
605 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
606 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
607 		}
608 	}
609 
610 	/* re-insert all policies by order of creation */
611 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
612 		if (policy->walk.dead ||
613 		    xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
614 			/* skip socket policies */
615 			continue;
616 		}
617 		newpos = NULL;
618 		chain = policy_hash_bysel(net, &policy->selector,
619 					  policy->family,
620 					  xfrm_policy_id2dir(policy->index));
621 		hlist_for_each_entry(pol, chain, bydst) {
622 			if (policy->priority >= pol->priority)
623 				newpos = &pol->bydst;
624 			else
625 				break;
626 		}
627 		if (newpos)
628 			hlist_add_behind(&policy->bydst, newpos);
629 		else
630 			hlist_add_head(&policy->bydst, chain);
631 	}
632 
633 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
634 
635 	mutex_unlock(&hash_resize_mutex);
636 }
637 
638 void xfrm_policy_hash_rebuild(struct net *net)
639 {
640 	schedule_work(&net->xfrm.policy_hthresh.work);
641 }
642 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
643 
644 /* Generate new index... KAME seems to generate them ordered by cost
645  * of an absolute inpredictability of ordering of rules. This will not pass. */
646 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
647 {
648 	static u32 idx_generator;
649 
650 	for (;;) {
651 		struct hlist_head *list;
652 		struct xfrm_policy *p;
653 		u32 idx;
654 		int found;
655 
656 		if (!index) {
657 			idx = (idx_generator | dir);
658 			idx_generator += 8;
659 		} else {
660 			idx = index;
661 			index = 0;
662 		}
663 
664 		if (idx == 0)
665 			idx = 8;
666 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
667 		found = 0;
668 		hlist_for_each_entry(p, list, byidx) {
669 			if (p->index == idx) {
670 				found = 1;
671 				break;
672 			}
673 		}
674 		if (!found)
675 			return idx;
676 	}
677 }
678 
679 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
680 {
681 	u32 *p1 = (u32 *) s1;
682 	u32 *p2 = (u32 *) s2;
683 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
684 	int i;
685 
686 	for (i = 0; i < len; i++) {
687 		if (p1[i] != p2[i])
688 			return 1;
689 	}
690 
691 	return 0;
692 }
693 
694 static void xfrm_policy_requeue(struct xfrm_policy *old,
695 				struct xfrm_policy *new)
696 {
697 	struct xfrm_policy_queue *pq = &old->polq;
698 	struct sk_buff_head list;
699 
700 	if (skb_queue_empty(&pq->hold_queue))
701 		return;
702 
703 	__skb_queue_head_init(&list);
704 
705 	spin_lock_bh(&pq->hold_queue.lock);
706 	skb_queue_splice_init(&pq->hold_queue, &list);
707 	if (del_timer(&pq->hold_timer))
708 		xfrm_pol_put(old);
709 	spin_unlock_bh(&pq->hold_queue.lock);
710 
711 	pq = &new->polq;
712 
713 	spin_lock_bh(&pq->hold_queue.lock);
714 	skb_queue_splice(&list, &pq->hold_queue);
715 	pq->timeout = XFRM_QUEUE_TMO_MIN;
716 	if (!mod_timer(&pq->hold_timer, jiffies))
717 		xfrm_pol_hold(new);
718 	spin_unlock_bh(&pq->hold_queue.lock);
719 }
720 
721 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
722 				   struct xfrm_policy *pol)
723 {
724 	u32 mark = policy->mark.v & policy->mark.m;
725 
726 	if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
727 		return true;
728 
729 	if ((mark & pol->mark.m) == pol->mark.v &&
730 	    policy->priority == pol->priority)
731 		return true;
732 
733 	return false;
734 }
735 
736 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
737 {
738 	struct net *net = xp_net(policy);
739 	struct xfrm_policy *pol;
740 	struct xfrm_policy *delpol;
741 	struct hlist_head *chain;
742 	struct hlist_node *newpos;
743 
744 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
745 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
746 	delpol = NULL;
747 	newpos = NULL;
748 	hlist_for_each_entry(pol, chain, bydst) {
749 		if (pol->type == policy->type &&
750 		    !selector_cmp(&pol->selector, &policy->selector) &&
751 		    xfrm_policy_mark_match(policy, pol) &&
752 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
753 		    !WARN_ON(delpol)) {
754 			if (excl) {
755 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
756 				return -EEXIST;
757 			}
758 			delpol = pol;
759 			if (policy->priority > pol->priority)
760 				continue;
761 		} else if (policy->priority >= pol->priority) {
762 			newpos = &pol->bydst;
763 			continue;
764 		}
765 		if (delpol)
766 			break;
767 	}
768 	if (newpos)
769 		hlist_add_behind(&policy->bydst, newpos);
770 	else
771 		hlist_add_head(&policy->bydst, chain);
772 	__xfrm_policy_link(policy, dir);
773 
774 	/* After previous checking, family can either be AF_INET or AF_INET6 */
775 	if (policy->family == AF_INET)
776 		rt_genid_bump_ipv4(net);
777 	else
778 		rt_genid_bump_ipv6(net);
779 
780 	if (delpol) {
781 		xfrm_policy_requeue(delpol, policy);
782 		__xfrm_policy_unlink(delpol, dir);
783 	}
784 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
785 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
786 	policy->curlft.add_time = get_seconds();
787 	policy->curlft.use_time = 0;
788 	if (!mod_timer(&policy->timer, jiffies + HZ))
789 		xfrm_pol_hold(policy);
790 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
791 
792 	if (delpol)
793 		xfrm_policy_kill(delpol);
794 	else if (xfrm_bydst_should_resize(net, dir, NULL))
795 		schedule_work(&net->xfrm.policy_hash_work);
796 
797 	return 0;
798 }
799 EXPORT_SYMBOL(xfrm_policy_insert);
800 
801 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
802 					  int dir, struct xfrm_selector *sel,
803 					  struct xfrm_sec_ctx *ctx, int delete,
804 					  int *err)
805 {
806 	struct xfrm_policy *pol, *ret;
807 	struct hlist_head *chain;
808 
809 	*err = 0;
810 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
811 	chain = policy_hash_bysel(net, sel, sel->family, dir);
812 	ret = NULL;
813 	hlist_for_each_entry(pol, chain, bydst) {
814 		if (pol->type == type &&
815 		    (mark & pol->mark.m) == pol->mark.v &&
816 		    !selector_cmp(sel, &pol->selector) &&
817 		    xfrm_sec_ctx_match(ctx, pol->security)) {
818 			xfrm_pol_hold(pol);
819 			if (delete) {
820 				*err = security_xfrm_policy_delete(
821 								pol->security);
822 				if (*err) {
823 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
824 					return pol;
825 				}
826 				__xfrm_policy_unlink(pol, dir);
827 			}
828 			ret = pol;
829 			break;
830 		}
831 	}
832 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
833 
834 	if (ret && delete)
835 		xfrm_policy_kill(ret);
836 	return ret;
837 }
838 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
839 
840 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
841 				     int dir, u32 id, int delete, int *err)
842 {
843 	struct xfrm_policy *pol, *ret;
844 	struct hlist_head *chain;
845 
846 	*err = -ENOENT;
847 	if (xfrm_policy_id2dir(id) != dir)
848 		return NULL;
849 
850 	*err = 0;
851 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
852 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
853 	ret = NULL;
854 	hlist_for_each_entry(pol, chain, byidx) {
855 		if (pol->type == type && pol->index == id &&
856 		    (mark & pol->mark.m) == pol->mark.v) {
857 			xfrm_pol_hold(pol);
858 			if (delete) {
859 				*err = security_xfrm_policy_delete(
860 								pol->security);
861 				if (*err) {
862 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
863 					return pol;
864 				}
865 				__xfrm_policy_unlink(pol, dir);
866 			}
867 			ret = pol;
868 			break;
869 		}
870 	}
871 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
872 
873 	if (ret && delete)
874 		xfrm_policy_kill(ret);
875 	return ret;
876 }
877 EXPORT_SYMBOL(xfrm_policy_byid);
878 
879 #ifdef CONFIG_SECURITY_NETWORK_XFRM
880 static inline int
881 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
882 {
883 	int dir, err = 0;
884 
885 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
886 		struct xfrm_policy *pol;
887 		int i;
888 
889 		hlist_for_each_entry(pol,
890 				     &net->xfrm.policy_inexact[dir], bydst) {
891 			if (pol->type != type)
892 				continue;
893 			err = security_xfrm_policy_delete(pol->security);
894 			if (err) {
895 				xfrm_audit_policy_delete(pol, 0, task_valid);
896 				return err;
897 			}
898 		}
899 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
900 			hlist_for_each_entry(pol,
901 					     net->xfrm.policy_bydst[dir].table + i,
902 					     bydst) {
903 				if (pol->type != type)
904 					continue;
905 				err = security_xfrm_policy_delete(
906 								pol->security);
907 				if (err) {
908 					xfrm_audit_policy_delete(pol, 0,
909 								 task_valid);
910 					return err;
911 				}
912 			}
913 		}
914 	}
915 	return err;
916 }
917 #else
918 static inline int
919 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
920 {
921 	return 0;
922 }
923 #endif
924 
925 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
926 {
927 	int dir, err = 0, cnt = 0;
928 
929 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
930 
931 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
932 	if (err)
933 		goto out;
934 
935 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
936 		struct xfrm_policy *pol;
937 		int i;
938 
939 	again1:
940 		hlist_for_each_entry(pol,
941 				     &net->xfrm.policy_inexact[dir], bydst) {
942 			if (pol->type != type)
943 				continue;
944 			__xfrm_policy_unlink(pol, dir);
945 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
946 			cnt++;
947 
948 			xfrm_audit_policy_delete(pol, 1, task_valid);
949 
950 			xfrm_policy_kill(pol);
951 
952 			spin_lock_bh(&net->xfrm.xfrm_policy_lock);
953 			goto again1;
954 		}
955 
956 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
957 	again2:
958 			hlist_for_each_entry(pol,
959 					     net->xfrm.policy_bydst[dir].table + i,
960 					     bydst) {
961 				if (pol->type != type)
962 					continue;
963 				__xfrm_policy_unlink(pol, dir);
964 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
965 				cnt++;
966 
967 				xfrm_audit_policy_delete(pol, 1, task_valid);
968 				xfrm_policy_kill(pol);
969 
970 				spin_lock_bh(&net->xfrm.xfrm_policy_lock);
971 				goto again2;
972 			}
973 		}
974 
975 	}
976 	if (!cnt)
977 		err = -ESRCH;
978 out:
979 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
980 	return err;
981 }
982 EXPORT_SYMBOL(xfrm_policy_flush);
983 
984 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
985 		     int (*func)(struct xfrm_policy *, int, int, void*),
986 		     void *data)
987 {
988 	struct xfrm_policy *pol;
989 	struct xfrm_policy_walk_entry *x;
990 	int error = 0;
991 
992 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
993 	    walk->type != XFRM_POLICY_TYPE_ANY)
994 		return -EINVAL;
995 
996 	if (list_empty(&walk->walk.all) && walk->seq != 0)
997 		return 0;
998 
999 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1000 	if (list_empty(&walk->walk.all))
1001 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1002 	else
1003 		x = list_first_entry(&walk->walk.all,
1004 				     struct xfrm_policy_walk_entry, all);
1005 
1006 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1007 		if (x->dead)
1008 			continue;
1009 		pol = container_of(x, struct xfrm_policy, walk);
1010 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1011 		    walk->type != pol->type)
1012 			continue;
1013 		error = func(pol, xfrm_policy_id2dir(pol->index),
1014 			     walk->seq, data);
1015 		if (error) {
1016 			list_move_tail(&walk->walk.all, &x->all);
1017 			goto out;
1018 		}
1019 		walk->seq++;
1020 	}
1021 	if (walk->seq == 0) {
1022 		error = -ENOENT;
1023 		goto out;
1024 	}
1025 	list_del_init(&walk->walk.all);
1026 out:
1027 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1028 	return error;
1029 }
1030 EXPORT_SYMBOL(xfrm_policy_walk);
1031 
1032 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1033 {
1034 	INIT_LIST_HEAD(&walk->walk.all);
1035 	walk->walk.dead = 1;
1036 	walk->type = type;
1037 	walk->seq = 0;
1038 }
1039 EXPORT_SYMBOL(xfrm_policy_walk_init);
1040 
1041 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1042 {
1043 	if (list_empty(&walk->walk.all))
1044 		return;
1045 
1046 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1047 	list_del(&walk->walk.all);
1048 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1049 }
1050 EXPORT_SYMBOL(xfrm_policy_walk_done);
1051 
1052 /*
1053  * Find policy to apply to this flow.
1054  *
1055  * Returns 0 if policy found, else an -errno.
1056  */
1057 static int xfrm_policy_match(const struct xfrm_policy *pol,
1058 			     const struct flowi *fl,
1059 			     u8 type, u16 family, int dir)
1060 {
1061 	const struct xfrm_selector *sel = &pol->selector;
1062 	int ret = -ESRCH;
1063 	bool match;
1064 
1065 	if (pol->family != family ||
1066 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1067 	    pol->type != type)
1068 		return ret;
1069 
1070 	match = xfrm_selector_match(sel, fl, family);
1071 	if (match)
1072 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1073 						  dir);
1074 
1075 	return ret;
1076 }
1077 
1078 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1079 						     const struct flowi *fl,
1080 						     u16 family, u8 dir)
1081 {
1082 	int err;
1083 	struct xfrm_policy *pol, *ret;
1084 	const xfrm_address_t *daddr, *saddr;
1085 	struct hlist_head *chain;
1086 	unsigned int sequence;
1087 	u32 priority;
1088 
1089 	daddr = xfrm_flowi_daddr(fl, family);
1090 	saddr = xfrm_flowi_saddr(fl, family);
1091 	if (unlikely(!daddr || !saddr))
1092 		return NULL;
1093 
1094 	rcu_read_lock();
1095  retry:
1096 	do {
1097 		sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1098 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
1099 	} while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1100 
1101 	priority = ~0U;
1102 	ret = NULL;
1103 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1104 		err = xfrm_policy_match(pol, fl, type, family, dir);
1105 		if (err) {
1106 			if (err == -ESRCH)
1107 				continue;
1108 			else {
1109 				ret = ERR_PTR(err);
1110 				goto fail;
1111 			}
1112 		} else {
1113 			ret = pol;
1114 			priority = ret->priority;
1115 			break;
1116 		}
1117 	}
1118 	chain = &net->xfrm.policy_inexact[dir];
1119 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1120 		if ((pol->priority >= priority) && ret)
1121 			break;
1122 
1123 		err = xfrm_policy_match(pol, fl, type, family, dir);
1124 		if (err) {
1125 			if (err == -ESRCH)
1126 				continue;
1127 			else {
1128 				ret = ERR_PTR(err);
1129 				goto fail;
1130 			}
1131 		} else {
1132 			ret = pol;
1133 			break;
1134 		}
1135 	}
1136 
1137 	if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1138 		goto retry;
1139 
1140 	if (ret && !xfrm_pol_hold_rcu(ret))
1141 		goto retry;
1142 fail:
1143 	rcu_read_unlock();
1144 
1145 	return ret;
1146 }
1147 
1148 static struct xfrm_policy *
1149 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1150 {
1151 #ifdef CONFIG_XFRM_SUB_POLICY
1152 	struct xfrm_policy *pol;
1153 
1154 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1155 	if (pol != NULL)
1156 		return pol;
1157 #endif
1158 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1159 }
1160 
1161 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1162 						 const struct flowi *fl, u16 family)
1163 {
1164 	struct xfrm_policy *pol;
1165 
1166 	rcu_read_lock();
1167  again:
1168 	pol = rcu_dereference(sk->sk_policy[dir]);
1169 	if (pol != NULL) {
1170 		bool match;
1171 		int err = 0;
1172 
1173 		if (pol->family != family) {
1174 			pol = NULL;
1175 			goto out;
1176 		}
1177 
1178 		match = xfrm_selector_match(&pol->selector, fl, family);
1179 		if (match) {
1180 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1181 				pol = NULL;
1182 				goto out;
1183 			}
1184 			err = security_xfrm_policy_lookup(pol->security,
1185 						      fl->flowi_secid,
1186 						      dir);
1187 			if (!err) {
1188 				if (!xfrm_pol_hold_rcu(pol))
1189 					goto again;
1190 			} else if (err == -ESRCH) {
1191 				pol = NULL;
1192 			} else {
1193 				pol = ERR_PTR(err);
1194 			}
1195 		} else
1196 			pol = NULL;
1197 	}
1198 out:
1199 	rcu_read_unlock();
1200 	return pol;
1201 }
1202 
1203 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1204 {
1205 	struct net *net = xp_net(pol);
1206 
1207 	list_add(&pol->walk.all, &net->xfrm.policy_all);
1208 	net->xfrm.policy_count[dir]++;
1209 	xfrm_pol_hold(pol);
1210 }
1211 
1212 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1213 						int dir)
1214 {
1215 	struct net *net = xp_net(pol);
1216 
1217 	if (list_empty(&pol->walk.all))
1218 		return NULL;
1219 
1220 	/* Socket policies are not hashed. */
1221 	if (!hlist_unhashed(&pol->bydst)) {
1222 		hlist_del_rcu(&pol->bydst);
1223 		hlist_del(&pol->byidx);
1224 	}
1225 
1226 	list_del_init(&pol->walk.all);
1227 	net->xfrm.policy_count[dir]--;
1228 
1229 	return pol;
1230 }
1231 
1232 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1233 {
1234 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1235 }
1236 
1237 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1238 {
1239 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1240 }
1241 
1242 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1243 {
1244 	struct net *net = xp_net(pol);
1245 
1246 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1247 	pol = __xfrm_policy_unlink(pol, dir);
1248 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1249 	if (pol) {
1250 		xfrm_policy_kill(pol);
1251 		return 0;
1252 	}
1253 	return -ENOENT;
1254 }
1255 EXPORT_SYMBOL(xfrm_policy_delete);
1256 
1257 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1258 {
1259 	struct net *net = sock_net(sk);
1260 	struct xfrm_policy *old_pol;
1261 
1262 #ifdef CONFIG_XFRM_SUB_POLICY
1263 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1264 		return -EINVAL;
1265 #endif
1266 
1267 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1268 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1269 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1270 	if (pol) {
1271 		pol->curlft.add_time = get_seconds();
1272 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1273 		xfrm_sk_policy_link(pol, dir);
1274 	}
1275 	rcu_assign_pointer(sk->sk_policy[dir], pol);
1276 	if (old_pol) {
1277 		if (pol)
1278 			xfrm_policy_requeue(old_pol, pol);
1279 
1280 		/* Unlinking succeeds always. This is the only function
1281 		 * allowed to delete or replace socket policy.
1282 		 */
1283 		xfrm_sk_policy_unlink(old_pol, dir);
1284 	}
1285 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1286 
1287 	if (old_pol) {
1288 		xfrm_policy_kill(old_pol);
1289 	}
1290 	return 0;
1291 }
1292 
1293 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1294 {
1295 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1296 	struct net *net = xp_net(old);
1297 
1298 	if (newp) {
1299 		newp->selector = old->selector;
1300 		if (security_xfrm_policy_clone(old->security,
1301 					       &newp->security)) {
1302 			kfree(newp);
1303 			return NULL;  /* ENOMEM */
1304 		}
1305 		newp->lft = old->lft;
1306 		newp->curlft = old->curlft;
1307 		newp->mark = old->mark;
1308 		newp->action = old->action;
1309 		newp->flags = old->flags;
1310 		newp->xfrm_nr = old->xfrm_nr;
1311 		newp->index = old->index;
1312 		newp->type = old->type;
1313 		newp->family = old->family;
1314 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1315 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1316 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1317 		xfrm_sk_policy_link(newp, dir);
1318 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1319 		xfrm_pol_put(newp);
1320 	}
1321 	return newp;
1322 }
1323 
1324 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1325 {
1326 	const struct xfrm_policy *p;
1327 	struct xfrm_policy *np;
1328 	int i, ret = 0;
1329 
1330 	rcu_read_lock();
1331 	for (i = 0; i < 2; i++) {
1332 		p = rcu_dereference(osk->sk_policy[i]);
1333 		if (p) {
1334 			np = clone_policy(p, i);
1335 			if (unlikely(!np)) {
1336 				ret = -ENOMEM;
1337 				break;
1338 			}
1339 			rcu_assign_pointer(sk->sk_policy[i], np);
1340 		}
1341 	}
1342 	rcu_read_unlock();
1343 	return ret;
1344 }
1345 
1346 static int
1347 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1348 	       xfrm_address_t *remote, unsigned short family, u32 mark)
1349 {
1350 	int err;
1351 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1352 
1353 	if (unlikely(afinfo == NULL))
1354 		return -EINVAL;
1355 	err = afinfo->get_saddr(net, oif, local, remote, mark);
1356 	rcu_read_unlock();
1357 	return err;
1358 }
1359 
1360 /* Resolve list of templates for the flow, given policy. */
1361 
1362 static int
1363 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1364 		      struct xfrm_state **xfrm, unsigned short family)
1365 {
1366 	struct net *net = xp_net(policy);
1367 	int nx;
1368 	int i, error;
1369 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1370 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1371 	xfrm_address_t tmp;
1372 
1373 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1374 		struct xfrm_state *x;
1375 		xfrm_address_t *remote = daddr;
1376 		xfrm_address_t *local  = saddr;
1377 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1378 
1379 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1380 		    tmpl->mode == XFRM_MODE_BEET) {
1381 			remote = &tmpl->id.daddr;
1382 			local = &tmpl->saddr;
1383 			if (xfrm_addr_any(local, tmpl->encap_family)) {
1384 				error = xfrm_get_saddr(net, fl->flowi_oif,
1385 						       &tmp, remote,
1386 						       tmpl->encap_family, 0);
1387 				if (error)
1388 					goto fail;
1389 				local = &tmp;
1390 			}
1391 		}
1392 
1393 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1394 
1395 		if (x && x->km.state == XFRM_STATE_VALID) {
1396 			xfrm[nx++] = x;
1397 			daddr = remote;
1398 			saddr = local;
1399 			continue;
1400 		}
1401 		if (x) {
1402 			error = (x->km.state == XFRM_STATE_ERROR ?
1403 				 -EINVAL : -EAGAIN);
1404 			xfrm_state_put(x);
1405 		} else if (error == -ESRCH) {
1406 			error = -EAGAIN;
1407 		}
1408 
1409 		if (!tmpl->optional)
1410 			goto fail;
1411 	}
1412 	return nx;
1413 
1414 fail:
1415 	for (nx--; nx >= 0; nx--)
1416 		xfrm_state_put(xfrm[nx]);
1417 	return error;
1418 }
1419 
1420 static int
1421 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1422 		  struct xfrm_state **xfrm, unsigned short family)
1423 {
1424 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1425 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1426 	int cnx = 0;
1427 	int error;
1428 	int ret;
1429 	int i;
1430 
1431 	for (i = 0; i < npols; i++) {
1432 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1433 			error = -ENOBUFS;
1434 			goto fail;
1435 		}
1436 
1437 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1438 		if (ret < 0) {
1439 			error = ret;
1440 			goto fail;
1441 		} else
1442 			cnx += ret;
1443 	}
1444 
1445 	/* found states are sorted for outbound processing */
1446 	if (npols > 1)
1447 		xfrm_state_sort(xfrm, tpp, cnx, family);
1448 
1449 	return cnx;
1450 
1451  fail:
1452 	for (cnx--; cnx >= 0; cnx--)
1453 		xfrm_state_put(tpp[cnx]);
1454 	return error;
1455 
1456 }
1457 
1458 static int xfrm_get_tos(const struct flowi *fl, int family)
1459 {
1460 	const struct xfrm_policy_afinfo *afinfo;
1461 	int tos;
1462 
1463 	afinfo = xfrm_policy_get_afinfo(family);
1464 	if (!afinfo)
1465 		return 0;
1466 
1467 	tos = afinfo->get_tos(fl);
1468 
1469 	rcu_read_unlock();
1470 
1471 	return tos;
1472 }
1473 
1474 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1475 {
1476 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1477 	struct dst_ops *dst_ops;
1478 	struct xfrm_dst *xdst;
1479 
1480 	if (!afinfo)
1481 		return ERR_PTR(-EINVAL);
1482 
1483 	switch (family) {
1484 	case AF_INET:
1485 		dst_ops = &net->xfrm.xfrm4_dst_ops;
1486 		break;
1487 #if IS_ENABLED(CONFIG_IPV6)
1488 	case AF_INET6:
1489 		dst_ops = &net->xfrm.xfrm6_dst_ops;
1490 		break;
1491 #endif
1492 	default:
1493 		BUG();
1494 	}
1495 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1496 
1497 	if (likely(xdst)) {
1498 		struct dst_entry *dst = &xdst->u.dst;
1499 
1500 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1501 	} else
1502 		xdst = ERR_PTR(-ENOBUFS);
1503 
1504 	rcu_read_unlock();
1505 
1506 	return xdst;
1507 }
1508 
1509 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1510 				 int nfheader_len)
1511 {
1512 	const struct xfrm_policy_afinfo *afinfo =
1513 		xfrm_policy_get_afinfo(dst->ops->family);
1514 	int err;
1515 
1516 	if (!afinfo)
1517 		return -EINVAL;
1518 
1519 	err = afinfo->init_path(path, dst, nfheader_len);
1520 
1521 	rcu_read_unlock();
1522 
1523 	return err;
1524 }
1525 
1526 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1527 				const struct flowi *fl)
1528 {
1529 	const struct xfrm_policy_afinfo *afinfo =
1530 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1531 	int err;
1532 
1533 	if (!afinfo)
1534 		return -EINVAL;
1535 
1536 	err = afinfo->fill_dst(xdst, dev, fl);
1537 
1538 	rcu_read_unlock();
1539 
1540 	return err;
1541 }
1542 
1543 
1544 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1545  * all the metrics... Shortly, bundle a bundle.
1546  */
1547 
1548 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1549 					    struct xfrm_state **xfrm,
1550 					    struct xfrm_dst **bundle,
1551 					    int nx,
1552 					    const struct flowi *fl,
1553 					    struct dst_entry *dst)
1554 {
1555 	struct net *net = xp_net(policy);
1556 	unsigned long now = jiffies;
1557 	struct net_device *dev;
1558 	struct xfrm_mode *inner_mode;
1559 	struct xfrm_dst *xdst_prev = NULL;
1560 	struct xfrm_dst *xdst0 = NULL;
1561 	int i = 0;
1562 	int err;
1563 	int header_len = 0;
1564 	int nfheader_len = 0;
1565 	int trailer_len = 0;
1566 	int tos;
1567 	int family = policy->selector.family;
1568 	xfrm_address_t saddr, daddr;
1569 
1570 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1571 
1572 	tos = xfrm_get_tos(fl, family);
1573 
1574 	dst_hold(dst);
1575 
1576 	for (; i < nx; i++) {
1577 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1578 		struct dst_entry *dst1 = &xdst->u.dst;
1579 
1580 		err = PTR_ERR(xdst);
1581 		if (IS_ERR(xdst)) {
1582 			dst_release(dst);
1583 			goto put_states;
1584 		}
1585 
1586 		bundle[i] = xdst;
1587 		if (!xdst_prev)
1588 			xdst0 = xdst;
1589 		else
1590 			/* Ref count is taken during xfrm_alloc_dst()
1591 			 * No need to do dst_clone() on dst1
1592 			 */
1593 			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
1594 
1595 		if (xfrm[i]->sel.family == AF_UNSPEC) {
1596 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
1597 							xfrm_af2proto(family));
1598 			if (!inner_mode) {
1599 				err = -EAFNOSUPPORT;
1600 				dst_release(dst);
1601 				goto put_states;
1602 			}
1603 		} else
1604 			inner_mode = xfrm[i]->inner_mode;
1605 
1606 		xdst->route = dst;
1607 		dst_copy_metrics(dst1, dst);
1608 
1609 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1610 			family = xfrm[i]->props.family;
1611 			dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1612 					      &saddr, &daddr, family,
1613 					      xfrm[i]->props.output_mark);
1614 			err = PTR_ERR(dst);
1615 			if (IS_ERR(dst))
1616 				goto put_states;
1617 		} else
1618 			dst_hold(dst);
1619 
1620 		dst1->xfrm = xfrm[i];
1621 		xdst->xfrm_genid = xfrm[i]->genid;
1622 
1623 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1624 		dst1->flags |= DST_HOST;
1625 		dst1->lastuse = now;
1626 
1627 		dst1->input = dst_discard;
1628 		dst1->output = inner_mode->afinfo->output;
1629 
1630 		xdst_prev = xdst;
1631 
1632 		header_len += xfrm[i]->props.header_len;
1633 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1634 			nfheader_len += xfrm[i]->props.header_len;
1635 		trailer_len += xfrm[i]->props.trailer_len;
1636 	}
1637 
1638 	xfrm_dst_set_child(xdst_prev, dst);
1639 	xdst0->path = dst;
1640 
1641 	err = -ENODEV;
1642 	dev = dst->dev;
1643 	if (!dev)
1644 		goto free_dst;
1645 
1646 	xfrm_init_path(xdst0, dst, nfheader_len);
1647 	xfrm_init_pmtu(bundle, nx);
1648 
1649 	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
1650 	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
1651 		err = xfrm_fill_dst(xdst_prev, dev, fl);
1652 		if (err)
1653 			goto free_dst;
1654 
1655 		xdst_prev->u.dst.header_len = header_len;
1656 		xdst_prev->u.dst.trailer_len = trailer_len;
1657 		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
1658 		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
1659 	}
1660 
1661 	return &xdst0->u.dst;
1662 
1663 put_states:
1664 	for (; i < nx; i++)
1665 		xfrm_state_put(xfrm[i]);
1666 free_dst:
1667 	if (xdst0)
1668 		dst_release_immediate(&xdst0->u.dst);
1669 
1670 	return ERR_PTR(err);
1671 }
1672 
1673 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1674 				struct xfrm_policy **pols,
1675 				int *num_pols, int *num_xfrms)
1676 {
1677 	int i;
1678 
1679 	if (*num_pols == 0 || !pols[0]) {
1680 		*num_pols = 0;
1681 		*num_xfrms = 0;
1682 		return 0;
1683 	}
1684 	if (IS_ERR(pols[0]))
1685 		return PTR_ERR(pols[0]);
1686 
1687 	*num_xfrms = pols[0]->xfrm_nr;
1688 
1689 #ifdef CONFIG_XFRM_SUB_POLICY
1690 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1691 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1692 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1693 						    XFRM_POLICY_TYPE_MAIN,
1694 						    fl, family,
1695 						    XFRM_POLICY_OUT);
1696 		if (pols[1]) {
1697 			if (IS_ERR(pols[1])) {
1698 				xfrm_pols_put(pols, *num_pols);
1699 				return PTR_ERR(pols[1]);
1700 			}
1701 			(*num_pols)++;
1702 			(*num_xfrms) += pols[1]->xfrm_nr;
1703 		}
1704 	}
1705 #endif
1706 	for (i = 0; i < *num_pols; i++) {
1707 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
1708 			*num_xfrms = -1;
1709 			break;
1710 		}
1711 	}
1712 
1713 	return 0;
1714 
1715 }
1716 
1717 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1718 {
1719 	this_cpu_write(xfrm_last_dst, xdst);
1720 	if (old)
1721 		dst_release(&old->u.dst);
1722 }
1723 
1724 static void __xfrm_pcpu_work_fn(void)
1725 {
1726 	struct xfrm_dst *old;
1727 
1728 	old = this_cpu_read(xfrm_last_dst);
1729 	if (old && !xfrm_bundle_ok(old))
1730 		xfrm_last_dst_update(NULL, old);
1731 }
1732 
1733 static void xfrm_pcpu_work_fn(struct work_struct *work)
1734 {
1735 	local_bh_disable();
1736 	rcu_read_lock();
1737 	__xfrm_pcpu_work_fn();
1738 	rcu_read_unlock();
1739 	local_bh_enable();
1740 }
1741 
1742 void xfrm_policy_cache_flush(void)
1743 {
1744 	struct xfrm_dst *old;
1745 	bool found = false;
1746 	int cpu;
1747 
1748 	might_sleep();
1749 
1750 	local_bh_disable();
1751 	rcu_read_lock();
1752 	for_each_possible_cpu(cpu) {
1753 		old = per_cpu(xfrm_last_dst, cpu);
1754 		if (old && !xfrm_bundle_ok(old)) {
1755 			if (smp_processor_id() == cpu) {
1756 				__xfrm_pcpu_work_fn();
1757 				continue;
1758 			}
1759 			found = true;
1760 			break;
1761 		}
1762 	}
1763 
1764 	rcu_read_unlock();
1765 	local_bh_enable();
1766 
1767 	if (!found)
1768 		return;
1769 
1770 	get_online_cpus();
1771 
1772 	for_each_possible_cpu(cpu) {
1773 		bool bundle_release;
1774 
1775 		rcu_read_lock();
1776 		old = per_cpu(xfrm_last_dst, cpu);
1777 		bundle_release = old && !xfrm_bundle_ok(old);
1778 		rcu_read_unlock();
1779 
1780 		if (!bundle_release)
1781 			continue;
1782 
1783 		if (cpu_online(cpu)) {
1784 			schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1785 			continue;
1786 		}
1787 
1788 		rcu_read_lock();
1789 		old = per_cpu(xfrm_last_dst, cpu);
1790 		if (old && !xfrm_bundle_ok(old)) {
1791 			per_cpu(xfrm_last_dst, cpu) = NULL;
1792 			dst_release(&old->u.dst);
1793 		}
1794 		rcu_read_unlock();
1795 	}
1796 
1797 	put_online_cpus();
1798 }
1799 
1800 static bool xfrm_xdst_can_reuse(struct xfrm_dst *xdst,
1801 				struct xfrm_state * const xfrm[],
1802 				int num)
1803 {
1804 	const struct dst_entry *dst = &xdst->u.dst;
1805 	int i;
1806 
1807 	if (xdst->num_xfrms != num)
1808 		return false;
1809 
1810 	for (i = 0; i < num; i++) {
1811 		if (!dst || dst->xfrm != xfrm[i])
1812 			return false;
1813 		dst = xfrm_dst_child(dst);
1814 	}
1815 
1816 	return xfrm_bundle_ok(xdst);
1817 }
1818 
1819 static struct xfrm_dst *
1820 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1821 			       const struct flowi *fl, u16 family,
1822 			       struct dst_entry *dst_orig)
1823 {
1824 	struct net *net = xp_net(pols[0]);
1825 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1826 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
1827 	struct xfrm_dst *xdst, *old;
1828 	struct dst_entry *dst;
1829 	int err;
1830 
1831 	/* Try to instantiate a bundle */
1832 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1833 	if (err <= 0) {
1834 		if (err != 0 && err != -EAGAIN)
1835 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1836 		return ERR_PTR(err);
1837 	}
1838 
1839 	xdst = this_cpu_read(xfrm_last_dst);
1840 	if (xdst &&
1841 	    xdst->u.dst.dev == dst_orig->dev &&
1842 	    xdst->num_pols == num_pols &&
1843 	    memcmp(xdst->pols, pols,
1844 		   sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1845 	    xfrm_xdst_can_reuse(xdst, xfrm, err)) {
1846 		dst_hold(&xdst->u.dst);
1847 		xfrm_pols_put(pols, num_pols);
1848 		while (err > 0)
1849 			xfrm_state_put(xfrm[--err]);
1850 		return xdst;
1851 	}
1852 
1853 	old = xdst;
1854 
1855 	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
1856 	if (IS_ERR(dst)) {
1857 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1858 		return ERR_CAST(dst);
1859 	}
1860 
1861 	xdst = (struct xfrm_dst *)dst;
1862 	xdst->num_xfrms = err;
1863 	xdst->num_pols = num_pols;
1864 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1865 	xdst->policy_genid = atomic_read(&pols[0]->genid);
1866 
1867 	atomic_set(&xdst->u.dst.__refcnt, 2);
1868 	xfrm_last_dst_update(xdst, old);
1869 
1870 	return xdst;
1871 }
1872 
1873 static void xfrm_policy_queue_process(struct timer_list *t)
1874 {
1875 	struct sk_buff *skb;
1876 	struct sock *sk;
1877 	struct dst_entry *dst;
1878 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1879 	struct net *net = xp_net(pol);
1880 	struct xfrm_policy_queue *pq = &pol->polq;
1881 	struct flowi fl;
1882 	struct sk_buff_head list;
1883 
1884 	spin_lock(&pq->hold_queue.lock);
1885 	skb = skb_peek(&pq->hold_queue);
1886 	if (!skb) {
1887 		spin_unlock(&pq->hold_queue.lock);
1888 		goto out;
1889 	}
1890 	dst = skb_dst(skb);
1891 	sk = skb->sk;
1892 	xfrm_decode_session(skb, &fl, dst->ops->family);
1893 	spin_unlock(&pq->hold_queue.lock);
1894 
1895 	dst_hold(xfrm_dst_path(dst));
1896 	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
1897 	if (IS_ERR(dst))
1898 		goto purge_queue;
1899 
1900 	if (dst->flags & DST_XFRM_QUEUE) {
1901 		dst_release(dst);
1902 
1903 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1904 			goto purge_queue;
1905 
1906 		pq->timeout = pq->timeout << 1;
1907 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1908 			xfrm_pol_hold(pol);
1909 	goto out;
1910 	}
1911 
1912 	dst_release(dst);
1913 
1914 	__skb_queue_head_init(&list);
1915 
1916 	spin_lock(&pq->hold_queue.lock);
1917 	pq->timeout = 0;
1918 	skb_queue_splice_init(&pq->hold_queue, &list);
1919 	spin_unlock(&pq->hold_queue.lock);
1920 
1921 	while (!skb_queue_empty(&list)) {
1922 		skb = __skb_dequeue(&list);
1923 
1924 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1925 		dst_hold(xfrm_dst_path(skb_dst(skb)));
1926 		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
1927 		if (IS_ERR(dst)) {
1928 			kfree_skb(skb);
1929 			continue;
1930 		}
1931 
1932 		nf_reset(skb);
1933 		skb_dst_drop(skb);
1934 		skb_dst_set(skb, dst);
1935 
1936 		dst_output(net, skb->sk, skb);
1937 	}
1938 
1939 out:
1940 	xfrm_pol_put(pol);
1941 	return;
1942 
1943 purge_queue:
1944 	pq->timeout = 0;
1945 	skb_queue_purge(&pq->hold_queue);
1946 	xfrm_pol_put(pol);
1947 }
1948 
1949 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1950 {
1951 	unsigned long sched_next;
1952 	struct dst_entry *dst = skb_dst(skb);
1953 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1954 	struct xfrm_policy *pol = xdst->pols[0];
1955 	struct xfrm_policy_queue *pq = &pol->polq;
1956 
1957 	if (unlikely(skb_fclone_busy(sk, skb))) {
1958 		kfree_skb(skb);
1959 		return 0;
1960 	}
1961 
1962 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1963 		kfree_skb(skb);
1964 		return -EAGAIN;
1965 	}
1966 
1967 	skb_dst_force(skb);
1968 
1969 	spin_lock_bh(&pq->hold_queue.lock);
1970 
1971 	if (!pq->timeout)
1972 		pq->timeout = XFRM_QUEUE_TMO_MIN;
1973 
1974 	sched_next = jiffies + pq->timeout;
1975 
1976 	if (del_timer(&pq->hold_timer)) {
1977 		if (time_before(pq->hold_timer.expires, sched_next))
1978 			sched_next = pq->hold_timer.expires;
1979 		xfrm_pol_put(pol);
1980 	}
1981 
1982 	__skb_queue_tail(&pq->hold_queue, skb);
1983 	if (!mod_timer(&pq->hold_timer, sched_next))
1984 		xfrm_pol_hold(pol);
1985 
1986 	spin_unlock_bh(&pq->hold_queue.lock);
1987 
1988 	return 0;
1989 }
1990 
1991 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1992 						 struct xfrm_flo *xflo,
1993 						 const struct flowi *fl,
1994 						 int num_xfrms,
1995 						 u16 family)
1996 {
1997 	int err;
1998 	struct net_device *dev;
1999 	struct dst_entry *dst;
2000 	struct dst_entry *dst1;
2001 	struct xfrm_dst *xdst;
2002 
2003 	xdst = xfrm_alloc_dst(net, family);
2004 	if (IS_ERR(xdst))
2005 		return xdst;
2006 
2007 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2008 	    net->xfrm.sysctl_larval_drop ||
2009 	    num_xfrms <= 0)
2010 		return xdst;
2011 
2012 	dst = xflo->dst_orig;
2013 	dst1 = &xdst->u.dst;
2014 	dst_hold(dst);
2015 	xdst->route = dst;
2016 
2017 	dst_copy_metrics(dst1, dst);
2018 
2019 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2020 	dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2021 	dst1->lastuse = jiffies;
2022 
2023 	dst1->input = dst_discard;
2024 	dst1->output = xdst_queue_output;
2025 
2026 	dst_hold(dst);
2027 	xfrm_dst_set_child(xdst, dst);
2028 	xdst->path = dst;
2029 
2030 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2031 
2032 	err = -ENODEV;
2033 	dev = dst->dev;
2034 	if (!dev)
2035 		goto free_dst;
2036 
2037 	err = xfrm_fill_dst(xdst, dev, fl);
2038 	if (err)
2039 		goto free_dst;
2040 
2041 out:
2042 	return xdst;
2043 
2044 free_dst:
2045 	dst_release(dst1);
2046 	xdst = ERR_PTR(err);
2047 	goto out;
2048 }
2049 
2050 static struct xfrm_dst *
2051 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2052 {
2053 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2054 	int num_pols = 0, num_xfrms = 0, err;
2055 	struct xfrm_dst *xdst;
2056 
2057 	/* Resolve policies to use if we couldn't get them from
2058 	 * previous cache entry */
2059 	num_pols = 1;
2060 	pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2061 	err = xfrm_expand_policies(fl, family, pols,
2062 					   &num_pols, &num_xfrms);
2063 	if (err < 0)
2064 		goto inc_error;
2065 	if (num_pols == 0)
2066 		return NULL;
2067 	if (num_xfrms <= 0)
2068 		goto make_dummy_bundle;
2069 
2070 	local_bh_disable();
2071 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2072 					      xflo->dst_orig);
2073 	local_bh_enable();
2074 
2075 	if (IS_ERR(xdst)) {
2076 		err = PTR_ERR(xdst);
2077 		if (err != -EAGAIN)
2078 			goto error;
2079 		goto make_dummy_bundle;
2080 	} else if (xdst == NULL) {
2081 		num_xfrms = 0;
2082 		goto make_dummy_bundle;
2083 	}
2084 
2085 	return xdst;
2086 
2087 make_dummy_bundle:
2088 	/* We found policies, but there's no bundles to instantiate:
2089 	 * either because the policy blocks, has no transformations or
2090 	 * we could not build template (no xfrm_states).*/
2091 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2092 	if (IS_ERR(xdst)) {
2093 		xfrm_pols_put(pols, num_pols);
2094 		return ERR_CAST(xdst);
2095 	}
2096 	xdst->num_pols = num_pols;
2097 	xdst->num_xfrms = num_xfrms;
2098 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2099 
2100 	return xdst;
2101 
2102 inc_error:
2103 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2104 error:
2105 	xfrm_pols_put(pols, num_pols);
2106 	return ERR_PTR(err);
2107 }
2108 
2109 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2110 					struct dst_entry *dst_orig)
2111 {
2112 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2113 	struct dst_entry *ret;
2114 
2115 	if (!afinfo) {
2116 		dst_release(dst_orig);
2117 		return ERR_PTR(-EINVAL);
2118 	} else {
2119 		ret = afinfo->blackhole_route(net, dst_orig);
2120 	}
2121 	rcu_read_unlock();
2122 
2123 	return ret;
2124 }
2125 
2126 /* Main function: finds/creates a bundle for given flow.
2127  *
2128  * At the moment we eat a raw IP route. Mostly to speed up lookups
2129  * on interfaces with disabled IPsec.
2130  */
2131 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2132 			      const struct flowi *fl,
2133 			      const struct sock *sk, int flags)
2134 {
2135 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2136 	struct xfrm_dst *xdst;
2137 	struct dst_entry *dst, *route;
2138 	u16 family = dst_orig->ops->family;
2139 	u8 dir = XFRM_POLICY_OUT;
2140 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2141 
2142 	dst = NULL;
2143 	xdst = NULL;
2144 	route = NULL;
2145 
2146 	sk = sk_const_to_full_sk(sk);
2147 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2148 		num_pols = 1;
2149 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2150 		err = xfrm_expand_policies(fl, family, pols,
2151 					   &num_pols, &num_xfrms);
2152 		if (err < 0)
2153 			goto dropdst;
2154 
2155 		if (num_pols) {
2156 			if (num_xfrms <= 0) {
2157 				drop_pols = num_pols;
2158 				goto no_transform;
2159 			}
2160 
2161 			local_bh_disable();
2162 			xdst = xfrm_resolve_and_create_bundle(
2163 					pols, num_pols, fl,
2164 					family, dst_orig);
2165 			local_bh_enable();
2166 
2167 			if (IS_ERR(xdst)) {
2168 				xfrm_pols_put(pols, num_pols);
2169 				err = PTR_ERR(xdst);
2170 				goto dropdst;
2171 			} else if (xdst == NULL) {
2172 				num_xfrms = 0;
2173 				drop_pols = num_pols;
2174 				goto no_transform;
2175 			}
2176 
2177 			route = xdst->route;
2178 		}
2179 	}
2180 
2181 	if (xdst == NULL) {
2182 		struct xfrm_flo xflo;
2183 
2184 		xflo.dst_orig = dst_orig;
2185 		xflo.flags = flags;
2186 
2187 		/* To accelerate a bit...  */
2188 		if ((dst_orig->flags & DST_NOXFRM) ||
2189 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
2190 			goto nopol;
2191 
2192 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2193 		if (xdst == NULL)
2194 			goto nopol;
2195 		if (IS_ERR(xdst)) {
2196 			err = PTR_ERR(xdst);
2197 			goto dropdst;
2198 		}
2199 
2200 		num_pols = xdst->num_pols;
2201 		num_xfrms = xdst->num_xfrms;
2202 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2203 		route = xdst->route;
2204 	}
2205 
2206 	dst = &xdst->u.dst;
2207 	if (route == NULL && num_xfrms > 0) {
2208 		/* The only case when xfrm_bundle_lookup() returns a
2209 		 * bundle with null route, is when the template could
2210 		 * not be resolved. It means policies are there, but
2211 		 * bundle could not be created, since we don't yet
2212 		 * have the xfrm_state's. We need to wait for KM to
2213 		 * negotiate new SA's or bail out with error.*/
2214 		if (net->xfrm.sysctl_larval_drop) {
2215 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2216 			err = -EREMOTE;
2217 			goto error;
2218 		}
2219 
2220 		err = -EAGAIN;
2221 
2222 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2223 		goto error;
2224 	}
2225 
2226 no_transform:
2227 	if (num_pols == 0)
2228 		goto nopol;
2229 
2230 	if ((flags & XFRM_LOOKUP_ICMP) &&
2231 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2232 		err = -ENOENT;
2233 		goto error;
2234 	}
2235 
2236 	for (i = 0; i < num_pols; i++)
2237 		pols[i]->curlft.use_time = get_seconds();
2238 
2239 	if (num_xfrms < 0) {
2240 		/* Prohibit the flow */
2241 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2242 		err = -EPERM;
2243 		goto error;
2244 	} else if (num_xfrms > 0) {
2245 		/* Flow transformed */
2246 		dst_release(dst_orig);
2247 	} else {
2248 		/* Flow passes untransformed */
2249 		dst_release(dst);
2250 		dst = dst_orig;
2251 	}
2252 ok:
2253 	xfrm_pols_put(pols, drop_pols);
2254 	if (dst && dst->xfrm &&
2255 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2256 		dst->flags |= DST_XFRM_TUNNEL;
2257 	return dst;
2258 
2259 nopol:
2260 	if (!(flags & XFRM_LOOKUP_ICMP)) {
2261 		dst = dst_orig;
2262 		goto ok;
2263 	}
2264 	err = -ENOENT;
2265 error:
2266 	dst_release(dst);
2267 dropdst:
2268 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2269 		dst_release(dst_orig);
2270 	xfrm_pols_put(pols, drop_pols);
2271 	return ERR_PTR(err);
2272 }
2273 EXPORT_SYMBOL(xfrm_lookup);
2274 
2275 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2276  * Otherwise we may send out blackholed packets.
2277  */
2278 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2279 				    const struct flowi *fl,
2280 				    const struct sock *sk, int flags)
2281 {
2282 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2283 					    flags | XFRM_LOOKUP_QUEUE |
2284 					    XFRM_LOOKUP_KEEP_DST_REF);
2285 
2286 	if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2287 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
2288 
2289 	if (IS_ERR(dst))
2290 		dst_release(dst_orig);
2291 
2292 	return dst;
2293 }
2294 EXPORT_SYMBOL(xfrm_lookup_route);
2295 
2296 static inline int
2297 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2298 {
2299 	struct xfrm_state *x;
2300 
2301 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2302 		return 0;
2303 	x = skb->sp->xvec[idx];
2304 	if (!x->type->reject)
2305 		return 0;
2306 	return x->type->reject(x, skb, fl);
2307 }
2308 
2309 /* When skb is transformed back to its "native" form, we have to
2310  * check policy restrictions. At the moment we make this in maximally
2311  * stupid way. Shame on me. :-) Of course, connected sockets must
2312  * have policy cached at them.
2313  */
2314 
2315 static inline int
2316 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2317 	      unsigned short family)
2318 {
2319 	if (xfrm_state_kern(x))
2320 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2321 	return	x->id.proto == tmpl->id.proto &&
2322 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2323 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2324 		x->props.mode == tmpl->mode &&
2325 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2326 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2327 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
2328 		  xfrm_state_addr_cmp(tmpl, x, family));
2329 }
2330 
2331 /*
2332  * 0 or more than 0 is returned when validation is succeeded (either bypass
2333  * because of optional transport mode, or next index of the mathced secpath
2334  * state with the template.
2335  * -1 is returned when no matching template is found.
2336  * Otherwise "-2 - errored_index" is returned.
2337  */
2338 static inline int
2339 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2340 	       unsigned short family)
2341 {
2342 	int idx = start;
2343 
2344 	if (tmpl->optional) {
2345 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
2346 			return start;
2347 	} else
2348 		start = -1;
2349 	for (; idx < sp->len; idx++) {
2350 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2351 			return ++idx;
2352 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2353 			if (start == -1)
2354 				start = -2-idx;
2355 			break;
2356 		}
2357 	}
2358 	return start;
2359 }
2360 
2361 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2362 			  unsigned int family, int reverse)
2363 {
2364 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2365 	int err;
2366 
2367 	if (unlikely(afinfo == NULL))
2368 		return -EAFNOSUPPORT;
2369 
2370 	afinfo->decode_session(skb, fl, reverse);
2371 	err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2372 	rcu_read_unlock();
2373 	return err;
2374 }
2375 EXPORT_SYMBOL(__xfrm_decode_session);
2376 
2377 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2378 {
2379 	for (; k < sp->len; k++) {
2380 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2381 			*idxp = k;
2382 			return 1;
2383 		}
2384 	}
2385 
2386 	return 0;
2387 }
2388 
2389 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2390 			unsigned short family)
2391 {
2392 	struct net *net = dev_net(skb->dev);
2393 	struct xfrm_policy *pol;
2394 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2395 	int npols = 0;
2396 	int xfrm_nr;
2397 	int pi;
2398 	int reverse;
2399 	struct flowi fl;
2400 	int xerr_idx = -1;
2401 
2402 	reverse = dir & ~XFRM_POLICY_MASK;
2403 	dir &= XFRM_POLICY_MASK;
2404 
2405 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2406 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2407 		return 0;
2408 	}
2409 
2410 	nf_nat_decode_session(skb, &fl, family);
2411 
2412 	/* First, check used SA against their selectors. */
2413 	if (skb->sp) {
2414 		int i;
2415 
2416 		for (i = skb->sp->len-1; i >= 0; i--) {
2417 			struct xfrm_state *x = skb->sp->xvec[i];
2418 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
2419 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2420 				return 0;
2421 			}
2422 		}
2423 	}
2424 
2425 	pol = NULL;
2426 	sk = sk_to_full_sk(sk);
2427 	if (sk && sk->sk_policy[dir]) {
2428 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2429 		if (IS_ERR(pol)) {
2430 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2431 			return 0;
2432 		}
2433 	}
2434 
2435 	if (!pol)
2436 		pol = xfrm_policy_lookup(net, &fl, family, dir);
2437 
2438 	if (IS_ERR(pol)) {
2439 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2440 		return 0;
2441 	}
2442 
2443 	if (!pol) {
2444 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2445 			xfrm_secpath_reject(xerr_idx, skb, &fl);
2446 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2447 			return 0;
2448 		}
2449 		return 1;
2450 	}
2451 
2452 	pol->curlft.use_time = get_seconds();
2453 
2454 	pols[0] = pol;
2455 	npols++;
2456 #ifdef CONFIG_XFRM_SUB_POLICY
2457 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2458 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2459 						    &fl, family,
2460 						    XFRM_POLICY_IN);
2461 		if (pols[1]) {
2462 			if (IS_ERR(pols[1])) {
2463 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2464 				return 0;
2465 			}
2466 			pols[1]->curlft.use_time = get_seconds();
2467 			npols++;
2468 		}
2469 	}
2470 #endif
2471 
2472 	if (pol->action == XFRM_POLICY_ALLOW) {
2473 		struct sec_path *sp;
2474 		static struct sec_path dummy;
2475 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2476 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2477 		struct xfrm_tmpl **tpp = tp;
2478 		int ti = 0;
2479 		int i, k;
2480 
2481 		if ((sp = skb->sp) == NULL)
2482 			sp = &dummy;
2483 
2484 		for (pi = 0; pi < npols; pi++) {
2485 			if (pols[pi] != pol &&
2486 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
2487 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2488 				goto reject;
2489 			}
2490 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2491 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2492 				goto reject_error;
2493 			}
2494 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
2495 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
2496 		}
2497 		xfrm_nr = ti;
2498 		if (npols > 1) {
2499 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2500 			tpp = stp;
2501 		}
2502 
2503 		/* For each tunnel xfrm, find the first matching tmpl.
2504 		 * For each tmpl before that, find corresponding xfrm.
2505 		 * Order is _important_. Later we will implement
2506 		 * some barriers, but at the moment barriers
2507 		 * are implied between each two transformations.
2508 		 */
2509 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2510 			k = xfrm_policy_ok(tpp[i], sp, k, family);
2511 			if (k < 0) {
2512 				if (k < -1)
2513 					/* "-2 - errored_index" returned */
2514 					xerr_idx = -(2+k);
2515 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2516 				goto reject;
2517 			}
2518 		}
2519 
2520 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2521 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2522 			goto reject;
2523 		}
2524 
2525 		xfrm_pols_put(pols, npols);
2526 		return 1;
2527 	}
2528 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2529 
2530 reject:
2531 	xfrm_secpath_reject(xerr_idx, skb, &fl);
2532 reject_error:
2533 	xfrm_pols_put(pols, npols);
2534 	return 0;
2535 }
2536 EXPORT_SYMBOL(__xfrm_policy_check);
2537 
2538 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2539 {
2540 	struct net *net = dev_net(skb->dev);
2541 	struct flowi fl;
2542 	struct dst_entry *dst;
2543 	int res = 1;
2544 
2545 	if (xfrm_decode_session(skb, &fl, family) < 0) {
2546 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2547 		return 0;
2548 	}
2549 
2550 	skb_dst_force(skb);
2551 
2552 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2553 	if (IS_ERR(dst)) {
2554 		res = 0;
2555 		dst = NULL;
2556 	}
2557 	skb_dst_set(skb, dst);
2558 	return res;
2559 }
2560 EXPORT_SYMBOL(__xfrm_route_forward);
2561 
2562 /* Optimize later using cookies and generation ids. */
2563 
2564 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2565 {
2566 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2567 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2568 	 * get validated by dst_ops->check on every use.  We do this
2569 	 * because when a normal route referenced by an XFRM dst is
2570 	 * obsoleted we do not go looking around for all parent
2571 	 * referencing XFRM dsts so that we can invalidate them.  It
2572 	 * is just too much work.  Instead we make the checks here on
2573 	 * every use.  For example:
2574 	 *
2575 	 *	XFRM dst A --> IPv4 dst X
2576 	 *
2577 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
2578 	 * in this example).  If X is marked obsolete, "A" will not
2579 	 * notice.  That's what we are validating here via the
2580 	 * stale_bundle() check.
2581 	 *
2582 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2583 	 * be marked on it.
2584 	 * This will force stale_bundle() to fail on any xdst bundle with
2585 	 * this dst linked in it.
2586 	 */
2587 	if (dst->obsolete < 0 && !stale_bundle(dst))
2588 		return dst;
2589 
2590 	return NULL;
2591 }
2592 
2593 static int stale_bundle(struct dst_entry *dst)
2594 {
2595 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2596 }
2597 
2598 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2599 {
2600 	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
2601 		dst->dev = dev_net(dev)->loopback_dev;
2602 		dev_hold(dst->dev);
2603 		dev_put(dev);
2604 	}
2605 }
2606 EXPORT_SYMBOL(xfrm_dst_ifdown);
2607 
2608 static void xfrm_link_failure(struct sk_buff *skb)
2609 {
2610 	/* Impossible. Such dst must be popped before reaches point of failure. */
2611 }
2612 
2613 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2614 {
2615 	if (dst) {
2616 		if (dst->obsolete) {
2617 			dst_release(dst);
2618 			dst = NULL;
2619 		}
2620 	}
2621 	return dst;
2622 }
2623 
2624 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
2625 {
2626 	while (nr--) {
2627 		struct xfrm_dst *xdst = bundle[nr];
2628 		u32 pmtu, route_mtu_cached;
2629 		struct dst_entry *dst;
2630 
2631 		dst = &xdst->u.dst;
2632 		pmtu = dst_mtu(xfrm_dst_child(dst));
2633 		xdst->child_mtu_cached = pmtu;
2634 
2635 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2636 
2637 		route_mtu_cached = dst_mtu(xdst->route);
2638 		xdst->route_mtu_cached = route_mtu_cached;
2639 
2640 		if (pmtu > route_mtu_cached)
2641 			pmtu = route_mtu_cached;
2642 
2643 		dst_metric_set(dst, RTAX_MTU, pmtu);
2644 	}
2645 }
2646 
2647 /* Check that the bundle accepts the flow and its components are
2648  * still valid.
2649  */
2650 
2651 static int xfrm_bundle_ok(struct xfrm_dst *first)
2652 {
2653 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2654 	struct dst_entry *dst = &first->u.dst;
2655 	struct xfrm_dst *xdst;
2656 	int start_from, nr;
2657 	u32 mtu;
2658 
2659 	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
2660 	    (dst->dev && !netif_running(dst->dev)))
2661 		return 0;
2662 
2663 	if (dst->flags & DST_XFRM_QUEUE)
2664 		return 1;
2665 
2666 	start_from = nr = 0;
2667 	do {
2668 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2669 
2670 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2671 			return 0;
2672 		if (xdst->xfrm_genid != dst->xfrm->genid)
2673 			return 0;
2674 		if (xdst->num_pols > 0 &&
2675 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2676 			return 0;
2677 
2678 		bundle[nr++] = xdst;
2679 
2680 		mtu = dst_mtu(xfrm_dst_child(dst));
2681 		if (xdst->child_mtu_cached != mtu) {
2682 			start_from = nr;
2683 			xdst->child_mtu_cached = mtu;
2684 		}
2685 
2686 		if (!dst_check(xdst->route, xdst->route_cookie))
2687 			return 0;
2688 		mtu = dst_mtu(xdst->route);
2689 		if (xdst->route_mtu_cached != mtu) {
2690 			start_from = nr;
2691 			xdst->route_mtu_cached = mtu;
2692 		}
2693 
2694 		dst = xfrm_dst_child(dst);
2695 	} while (dst->xfrm);
2696 
2697 	if (likely(!start_from))
2698 		return 1;
2699 
2700 	xdst = bundle[start_from - 1];
2701 	mtu = xdst->child_mtu_cached;
2702 	while (start_from--) {
2703 		dst = &xdst->u.dst;
2704 
2705 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2706 		if (mtu > xdst->route_mtu_cached)
2707 			mtu = xdst->route_mtu_cached;
2708 		dst_metric_set(dst, RTAX_MTU, mtu);
2709 		if (!start_from)
2710 			break;
2711 
2712 		xdst = bundle[start_from - 1];
2713 		xdst->child_mtu_cached = mtu;
2714 	}
2715 
2716 	return 1;
2717 }
2718 
2719 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2720 {
2721 	return dst_metric_advmss(xfrm_dst_path(dst));
2722 }
2723 
2724 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2725 {
2726 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2727 
2728 	return mtu ? : dst_mtu(xfrm_dst_path(dst));
2729 }
2730 
2731 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2732 					const void *daddr)
2733 {
2734 	while (dst->xfrm) {
2735 		const struct xfrm_state *xfrm = dst->xfrm;
2736 
2737 		dst = xfrm_dst_child(dst);
2738 
2739 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2740 			continue;
2741 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2742 			daddr = xfrm->coaddr;
2743 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2744 			daddr = &xfrm->id.daddr;
2745 	}
2746 	return daddr;
2747 }
2748 
2749 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2750 					   struct sk_buff *skb,
2751 					   const void *daddr)
2752 {
2753 	const struct dst_entry *path = xfrm_dst_path(dst);
2754 
2755 	if (!skb)
2756 		daddr = xfrm_get_dst_nexthop(dst, daddr);
2757 	return path->ops->neigh_lookup(path, skb, daddr);
2758 }
2759 
2760 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2761 {
2762 	const struct dst_entry *path = xfrm_dst_path(dst);
2763 
2764 	daddr = xfrm_get_dst_nexthop(dst, daddr);
2765 	path->ops->confirm_neigh(path, daddr);
2766 }
2767 
2768 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2769 {
2770 	int err = 0;
2771 
2772 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2773 		return -EAFNOSUPPORT;
2774 
2775 	spin_lock(&xfrm_policy_afinfo_lock);
2776 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
2777 		err = -EEXIST;
2778 	else {
2779 		struct dst_ops *dst_ops = afinfo->dst_ops;
2780 		if (likely(dst_ops->kmem_cachep == NULL))
2781 			dst_ops->kmem_cachep = xfrm_dst_cache;
2782 		if (likely(dst_ops->check == NULL))
2783 			dst_ops->check = xfrm_dst_check;
2784 		if (likely(dst_ops->default_advmss == NULL))
2785 			dst_ops->default_advmss = xfrm_default_advmss;
2786 		if (likely(dst_ops->mtu == NULL))
2787 			dst_ops->mtu = xfrm_mtu;
2788 		if (likely(dst_ops->negative_advice == NULL))
2789 			dst_ops->negative_advice = xfrm_negative_advice;
2790 		if (likely(dst_ops->link_failure == NULL))
2791 			dst_ops->link_failure = xfrm_link_failure;
2792 		if (likely(dst_ops->neigh_lookup == NULL))
2793 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
2794 		if (likely(!dst_ops->confirm_neigh))
2795 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
2796 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2797 	}
2798 	spin_unlock(&xfrm_policy_afinfo_lock);
2799 
2800 	return err;
2801 }
2802 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2803 
2804 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2805 {
2806 	struct dst_ops *dst_ops = afinfo->dst_ops;
2807 	int i;
2808 
2809 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2810 		if (xfrm_policy_afinfo[i] != afinfo)
2811 			continue;
2812 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2813 		break;
2814 	}
2815 
2816 	synchronize_rcu();
2817 
2818 	dst_ops->kmem_cachep = NULL;
2819 	dst_ops->check = NULL;
2820 	dst_ops->negative_advice = NULL;
2821 	dst_ops->link_failure = NULL;
2822 }
2823 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2824 
2825 #ifdef CONFIG_XFRM_STATISTICS
2826 static int __net_init xfrm_statistics_init(struct net *net)
2827 {
2828 	int rv;
2829 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2830 	if (!net->mib.xfrm_statistics)
2831 		return -ENOMEM;
2832 	rv = xfrm_proc_init(net);
2833 	if (rv < 0)
2834 		free_percpu(net->mib.xfrm_statistics);
2835 	return rv;
2836 }
2837 
2838 static void xfrm_statistics_fini(struct net *net)
2839 {
2840 	xfrm_proc_fini(net);
2841 	free_percpu(net->mib.xfrm_statistics);
2842 }
2843 #else
2844 static int __net_init xfrm_statistics_init(struct net *net)
2845 {
2846 	return 0;
2847 }
2848 
2849 static void xfrm_statistics_fini(struct net *net)
2850 {
2851 }
2852 #endif
2853 
2854 static int __net_init xfrm_policy_init(struct net *net)
2855 {
2856 	unsigned int hmask, sz;
2857 	int dir;
2858 
2859 	if (net_eq(net, &init_net))
2860 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2861 					   sizeof(struct xfrm_dst),
2862 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2863 					   NULL);
2864 
2865 	hmask = 8 - 1;
2866 	sz = (hmask+1) * sizeof(struct hlist_head);
2867 
2868 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2869 	if (!net->xfrm.policy_byidx)
2870 		goto out_byidx;
2871 	net->xfrm.policy_idx_hmask = hmask;
2872 
2873 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2874 		struct xfrm_policy_hash *htab;
2875 
2876 		net->xfrm.policy_count[dir] = 0;
2877 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2878 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2879 
2880 		htab = &net->xfrm.policy_bydst[dir];
2881 		htab->table = xfrm_hash_alloc(sz);
2882 		if (!htab->table)
2883 			goto out_bydst;
2884 		htab->hmask = hmask;
2885 		htab->dbits4 = 32;
2886 		htab->sbits4 = 32;
2887 		htab->dbits6 = 128;
2888 		htab->sbits6 = 128;
2889 	}
2890 	net->xfrm.policy_hthresh.lbits4 = 32;
2891 	net->xfrm.policy_hthresh.rbits4 = 32;
2892 	net->xfrm.policy_hthresh.lbits6 = 128;
2893 	net->xfrm.policy_hthresh.rbits6 = 128;
2894 
2895 	seqlock_init(&net->xfrm.policy_hthresh.lock);
2896 
2897 	INIT_LIST_HEAD(&net->xfrm.policy_all);
2898 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2899 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2900 	return 0;
2901 
2902 out_bydst:
2903 	for (dir--; dir >= 0; dir--) {
2904 		struct xfrm_policy_hash *htab;
2905 
2906 		htab = &net->xfrm.policy_bydst[dir];
2907 		xfrm_hash_free(htab->table, sz);
2908 	}
2909 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2910 out_byidx:
2911 	return -ENOMEM;
2912 }
2913 
2914 static void xfrm_policy_fini(struct net *net)
2915 {
2916 	unsigned int sz;
2917 	int dir;
2918 
2919 	flush_work(&net->xfrm.policy_hash_work);
2920 #ifdef CONFIG_XFRM_SUB_POLICY
2921 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2922 #endif
2923 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2924 
2925 	WARN_ON(!list_empty(&net->xfrm.policy_all));
2926 
2927 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2928 		struct xfrm_policy_hash *htab;
2929 
2930 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2931 
2932 		htab = &net->xfrm.policy_bydst[dir];
2933 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2934 		WARN_ON(!hlist_empty(htab->table));
2935 		xfrm_hash_free(htab->table, sz);
2936 	}
2937 
2938 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2939 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2940 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2941 }
2942 
2943 static int __net_init xfrm_net_init(struct net *net)
2944 {
2945 	int rv;
2946 
2947 	/* Initialize the per-net locks here */
2948 	spin_lock_init(&net->xfrm.xfrm_state_lock);
2949 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
2950 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
2951 
2952 	rv = xfrm_statistics_init(net);
2953 	if (rv < 0)
2954 		goto out_statistics;
2955 	rv = xfrm_state_init(net);
2956 	if (rv < 0)
2957 		goto out_state;
2958 	rv = xfrm_policy_init(net);
2959 	if (rv < 0)
2960 		goto out_policy;
2961 	rv = xfrm_sysctl_init(net);
2962 	if (rv < 0)
2963 		goto out_sysctl;
2964 
2965 	return 0;
2966 
2967 out_sysctl:
2968 	xfrm_policy_fini(net);
2969 out_policy:
2970 	xfrm_state_fini(net);
2971 out_state:
2972 	xfrm_statistics_fini(net);
2973 out_statistics:
2974 	return rv;
2975 }
2976 
2977 static void __net_exit xfrm_net_exit(struct net *net)
2978 {
2979 	xfrm_sysctl_fini(net);
2980 	xfrm_policy_fini(net);
2981 	xfrm_state_fini(net);
2982 	xfrm_statistics_fini(net);
2983 }
2984 
2985 static struct pernet_operations __net_initdata xfrm_net_ops = {
2986 	.init = xfrm_net_init,
2987 	.exit = xfrm_net_exit,
2988 };
2989 
2990 void __init xfrm_init(void)
2991 {
2992 	int i;
2993 
2994 	xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2995 				       GFP_KERNEL);
2996 	BUG_ON(!xfrm_pcpu_work);
2997 
2998 	for (i = 0; i < NR_CPUS; i++)
2999 		INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
3000 
3001 	register_pernet_subsys(&xfrm_net_ops);
3002 	xfrm_dev_init();
3003 	seqcount_init(&xfrm_policy_hash_generation);
3004 	xfrm_input_init();
3005 }
3006 
3007 #ifdef CONFIG_AUDITSYSCALL
3008 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3009 					 struct audit_buffer *audit_buf)
3010 {
3011 	struct xfrm_sec_ctx *ctx = xp->security;
3012 	struct xfrm_selector *sel = &xp->selector;
3013 
3014 	if (ctx)
3015 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3016 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3017 
3018 	switch (sel->family) {
3019 	case AF_INET:
3020 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3021 		if (sel->prefixlen_s != 32)
3022 			audit_log_format(audit_buf, " src_prefixlen=%d",
3023 					 sel->prefixlen_s);
3024 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3025 		if (sel->prefixlen_d != 32)
3026 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3027 					 sel->prefixlen_d);
3028 		break;
3029 	case AF_INET6:
3030 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3031 		if (sel->prefixlen_s != 128)
3032 			audit_log_format(audit_buf, " src_prefixlen=%d",
3033 					 sel->prefixlen_s);
3034 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3035 		if (sel->prefixlen_d != 128)
3036 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3037 					 sel->prefixlen_d);
3038 		break;
3039 	}
3040 }
3041 
3042 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3043 {
3044 	struct audit_buffer *audit_buf;
3045 
3046 	audit_buf = xfrm_audit_start("SPD-add");
3047 	if (audit_buf == NULL)
3048 		return;
3049 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3050 	audit_log_format(audit_buf, " res=%u", result);
3051 	xfrm_audit_common_policyinfo(xp, audit_buf);
3052 	audit_log_end(audit_buf);
3053 }
3054 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3055 
3056 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3057 			      bool task_valid)
3058 {
3059 	struct audit_buffer *audit_buf;
3060 
3061 	audit_buf = xfrm_audit_start("SPD-delete");
3062 	if (audit_buf == NULL)
3063 		return;
3064 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3065 	audit_log_format(audit_buf, " res=%u", result);
3066 	xfrm_audit_common_policyinfo(xp, audit_buf);
3067 	audit_log_end(audit_buf);
3068 }
3069 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3070 #endif
3071 
3072 #ifdef CONFIG_XFRM_MIGRATE
3073 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3074 					const struct xfrm_selector *sel_tgt)
3075 {
3076 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3077 		if (sel_tgt->family == sel_cmp->family &&
3078 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3079 				    sel_cmp->family) &&
3080 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3081 				    sel_cmp->family) &&
3082 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3083 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3084 			return true;
3085 		}
3086 	} else {
3087 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3088 			return true;
3089 		}
3090 	}
3091 	return false;
3092 }
3093 
3094 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3095 						    u8 dir, u8 type, struct net *net)
3096 {
3097 	struct xfrm_policy *pol, *ret = NULL;
3098 	struct hlist_head *chain;
3099 	u32 priority = ~0U;
3100 
3101 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3102 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3103 	hlist_for_each_entry(pol, chain, bydst) {
3104 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3105 		    pol->type == type) {
3106 			ret = pol;
3107 			priority = ret->priority;
3108 			break;
3109 		}
3110 	}
3111 	chain = &net->xfrm.policy_inexact[dir];
3112 	hlist_for_each_entry(pol, chain, bydst) {
3113 		if ((pol->priority >= priority) && ret)
3114 			break;
3115 
3116 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3117 		    pol->type == type) {
3118 			ret = pol;
3119 			break;
3120 		}
3121 	}
3122 
3123 	xfrm_pol_hold(ret);
3124 
3125 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3126 
3127 	return ret;
3128 }
3129 
3130 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3131 {
3132 	int match = 0;
3133 
3134 	if (t->mode == m->mode && t->id.proto == m->proto &&
3135 	    (m->reqid == 0 || t->reqid == m->reqid)) {
3136 		switch (t->mode) {
3137 		case XFRM_MODE_TUNNEL:
3138 		case XFRM_MODE_BEET:
3139 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3140 					    m->old_family) &&
3141 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
3142 					    m->old_family)) {
3143 				match = 1;
3144 			}
3145 			break;
3146 		case XFRM_MODE_TRANSPORT:
3147 			/* in case of transport mode, template does not store
3148 			   any IP addresses, hence we just compare mode and
3149 			   protocol */
3150 			match = 1;
3151 			break;
3152 		default:
3153 			break;
3154 		}
3155 	}
3156 	return match;
3157 }
3158 
3159 /* update endpoint address(es) of template(s) */
3160 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3161 			       struct xfrm_migrate *m, int num_migrate)
3162 {
3163 	struct xfrm_migrate *mp;
3164 	int i, j, n = 0;
3165 
3166 	write_lock_bh(&pol->lock);
3167 	if (unlikely(pol->walk.dead)) {
3168 		/* target policy has been deleted */
3169 		write_unlock_bh(&pol->lock);
3170 		return -ENOENT;
3171 	}
3172 
3173 	for (i = 0; i < pol->xfrm_nr; i++) {
3174 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3175 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3176 				continue;
3177 			n++;
3178 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3179 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3180 				continue;
3181 			/* update endpoints */
3182 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3183 			       sizeof(pol->xfrm_vec[i].id.daddr));
3184 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3185 			       sizeof(pol->xfrm_vec[i].saddr));
3186 			pol->xfrm_vec[i].encap_family = mp->new_family;
3187 			/* flush bundles */
3188 			atomic_inc(&pol->genid);
3189 		}
3190 	}
3191 
3192 	write_unlock_bh(&pol->lock);
3193 
3194 	if (!n)
3195 		return -ENODATA;
3196 
3197 	return 0;
3198 }
3199 
3200 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3201 {
3202 	int i, j;
3203 
3204 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3205 		return -EINVAL;
3206 
3207 	for (i = 0; i < num_migrate; i++) {
3208 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3209 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3210 			return -EINVAL;
3211 
3212 		/* check if there is any duplicated entry */
3213 		for (j = i + 1; j < num_migrate; j++) {
3214 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3215 				    sizeof(m[i].old_daddr)) &&
3216 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3217 				    sizeof(m[i].old_saddr)) &&
3218 			    m[i].proto == m[j].proto &&
3219 			    m[i].mode == m[j].mode &&
3220 			    m[i].reqid == m[j].reqid &&
3221 			    m[i].old_family == m[j].old_family)
3222 				return -EINVAL;
3223 		}
3224 	}
3225 
3226 	return 0;
3227 }
3228 
3229 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3230 		 struct xfrm_migrate *m, int num_migrate,
3231 		 struct xfrm_kmaddress *k, struct net *net,
3232 		 struct xfrm_encap_tmpl *encap)
3233 {
3234 	int i, err, nx_cur = 0, nx_new = 0;
3235 	struct xfrm_policy *pol = NULL;
3236 	struct xfrm_state *x, *xc;
3237 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3238 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3239 	struct xfrm_migrate *mp;
3240 
3241 	/* Stage 0 - sanity checks */
3242 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3243 		goto out;
3244 
3245 	if (dir >= XFRM_POLICY_MAX) {
3246 		err = -EINVAL;
3247 		goto out;
3248 	}
3249 
3250 	/* Stage 1 - find policy */
3251 	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3252 		err = -ENOENT;
3253 		goto out;
3254 	}
3255 
3256 	/* Stage 2 - find and update state(s) */
3257 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3258 		if ((x = xfrm_migrate_state_find(mp, net))) {
3259 			x_cur[nx_cur] = x;
3260 			nx_cur++;
3261 			xc = xfrm_state_migrate(x, mp, encap);
3262 			if (xc) {
3263 				x_new[nx_new] = xc;
3264 				nx_new++;
3265 			} else {
3266 				err = -ENODATA;
3267 				goto restore_state;
3268 			}
3269 		}
3270 	}
3271 
3272 	/* Stage 3 - update policy */
3273 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3274 		goto restore_state;
3275 
3276 	/* Stage 4 - delete old state(s) */
3277 	if (nx_cur) {
3278 		xfrm_states_put(x_cur, nx_cur);
3279 		xfrm_states_delete(x_cur, nx_cur);
3280 	}
3281 
3282 	/* Stage 5 - announce */
3283 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
3284 
3285 	xfrm_pol_put(pol);
3286 
3287 	return 0;
3288 out:
3289 	return err;
3290 
3291 restore_state:
3292 	if (pol)
3293 		xfrm_pol_put(pol);
3294 	if (nx_cur)
3295 		xfrm_states_put(x_cur, nx_cur);
3296 	if (nx_new)
3297 		xfrm_states_delete(x_new, nx_new);
3298 
3299 	return err;
3300 }
3301 EXPORT_SYMBOL(xfrm_migrate);
3302 #endif
3303