xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 9dae47aba0a055f761176d9297371d5bb24289ec)
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 __read_mostly;
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 (xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
613 			/* skip socket policies */
614 			continue;
615 		}
616 		newpos = NULL;
617 		chain = policy_hash_bysel(net, &policy->selector,
618 					  policy->family,
619 					  xfrm_policy_id2dir(policy->index));
620 		hlist_for_each_entry(pol, chain, bydst) {
621 			if (policy->priority >= pol->priority)
622 				newpos = &pol->bydst;
623 			else
624 				break;
625 		}
626 		if (newpos)
627 			hlist_add_behind(&policy->bydst, newpos);
628 		else
629 			hlist_add_head(&policy->bydst, chain);
630 	}
631 
632 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
633 
634 	mutex_unlock(&hash_resize_mutex);
635 }
636 
637 void xfrm_policy_hash_rebuild(struct net *net)
638 {
639 	schedule_work(&net->xfrm.policy_hthresh.work);
640 }
641 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
642 
643 /* Generate new index... KAME seems to generate them ordered by cost
644  * of an absolute inpredictability of ordering of rules. This will not pass. */
645 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
646 {
647 	static u32 idx_generator;
648 
649 	for (;;) {
650 		struct hlist_head *list;
651 		struct xfrm_policy *p;
652 		u32 idx;
653 		int found;
654 
655 		if (!index) {
656 			idx = (idx_generator | dir);
657 			idx_generator += 8;
658 		} else {
659 			idx = index;
660 			index = 0;
661 		}
662 
663 		if (idx == 0)
664 			idx = 8;
665 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
666 		found = 0;
667 		hlist_for_each_entry(p, list, byidx) {
668 			if (p->index == idx) {
669 				found = 1;
670 				break;
671 			}
672 		}
673 		if (!found)
674 			return idx;
675 	}
676 }
677 
678 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
679 {
680 	u32 *p1 = (u32 *) s1;
681 	u32 *p2 = (u32 *) s2;
682 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
683 	int i;
684 
685 	for (i = 0; i < len; i++) {
686 		if (p1[i] != p2[i])
687 			return 1;
688 	}
689 
690 	return 0;
691 }
692 
693 static void xfrm_policy_requeue(struct xfrm_policy *old,
694 				struct xfrm_policy *new)
695 {
696 	struct xfrm_policy_queue *pq = &old->polq;
697 	struct sk_buff_head list;
698 
699 	if (skb_queue_empty(&pq->hold_queue))
700 		return;
701 
702 	__skb_queue_head_init(&list);
703 
704 	spin_lock_bh(&pq->hold_queue.lock);
705 	skb_queue_splice_init(&pq->hold_queue, &list);
706 	if (del_timer(&pq->hold_timer))
707 		xfrm_pol_put(old);
708 	spin_unlock_bh(&pq->hold_queue.lock);
709 
710 	pq = &new->polq;
711 
712 	spin_lock_bh(&pq->hold_queue.lock);
713 	skb_queue_splice(&list, &pq->hold_queue);
714 	pq->timeout = XFRM_QUEUE_TMO_MIN;
715 	if (!mod_timer(&pq->hold_timer, jiffies))
716 		xfrm_pol_hold(new);
717 	spin_unlock_bh(&pq->hold_queue.lock);
718 }
719 
720 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
721 				   struct xfrm_policy *pol)
722 {
723 	u32 mark = policy->mark.v & policy->mark.m;
724 
725 	if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
726 		return true;
727 
728 	if ((mark & pol->mark.m) == pol->mark.v &&
729 	    policy->priority == pol->priority)
730 		return true;
731 
732 	return false;
733 }
734 
735 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
736 {
737 	struct net *net = xp_net(policy);
738 	struct xfrm_policy *pol;
739 	struct xfrm_policy *delpol;
740 	struct hlist_head *chain;
741 	struct hlist_node *newpos;
742 
743 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
744 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
745 	delpol = NULL;
746 	newpos = NULL;
747 	hlist_for_each_entry(pol, chain, bydst) {
748 		if (pol->type == policy->type &&
749 		    !selector_cmp(&pol->selector, &policy->selector) &&
750 		    xfrm_policy_mark_match(policy, pol) &&
751 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
752 		    !WARN_ON(delpol)) {
753 			if (excl) {
754 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
755 				return -EEXIST;
756 			}
757 			delpol = pol;
758 			if (policy->priority > pol->priority)
759 				continue;
760 		} else if (policy->priority >= pol->priority) {
761 			newpos = &pol->bydst;
762 			continue;
763 		}
764 		if (delpol)
765 			break;
766 	}
767 	if (newpos)
768 		hlist_add_behind(&policy->bydst, newpos);
769 	else
770 		hlist_add_head(&policy->bydst, chain);
771 	__xfrm_policy_link(policy, dir);
772 
773 	/* After previous checking, family can either be AF_INET or AF_INET6 */
774 	if (policy->family == AF_INET)
775 		rt_genid_bump_ipv4(net);
776 	else
777 		rt_genid_bump_ipv6(net);
778 
779 	if (delpol) {
780 		xfrm_policy_requeue(delpol, policy);
781 		__xfrm_policy_unlink(delpol, dir);
782 	}
783 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
784 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
785 	policy->curlft.add_time = get_seconds();
786 	policy->curlft.use_time = 0;
787 	if (!mod_timer(&policy->timer, jiffies + HZ))
788 		xfrm_pol_hold(policy);
789 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
790 
791 	if (delpol)
792 		xfrm_policy_kill(delpol);
793 	else if (xfrm_bydst_should_resize(net, dir, NULL))
794 		schedule_work(&net->xfrm.policy_hash_work);
795 
796 	return 0;
797 }
798 EXPORT_SYMBOL(xfrm_policy_insert);
799 
800 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
801 					  int dir, struct xfrm_selector *sel,
802 					  struct xfrm_sec_ctx *ctx, int delete,
803 					  int *err)
804 {
805 	struct xfrm_policy *pol, *ret;
806 	struct hlist_head *chain;
807 
808 	*err = 0;
809 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
810 	chain = policy_hash_bysel(net, sel, sel->family, dir);
811 	ret = NULL;
812 	hlist_for_each_entry(pol, chain, bydst) {
813 		if (pol->type == type &&
814 		    (mark & pol->mark.m) == pol->mark.v &&
815 		    !selector_cmp(sel, &pol->selector) &&
816 		    xfrm_sec_ctx_match(ctx, pol->security)) {
817 			xfrm_pol_hold(pol);
818 			if (delete) {
819 				*err = security_xfrm_policy_delete(
820 								pol->security);
821 				if (*err) {
822 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
823 					return pol;
824 				}
825 				__xfrm_policy_unlink(pol, dir);
826 			}
827 			ret = pol;
828 			break;
829 		}
830 	}
831 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
832 
833 	if (ret && delete)
834 		xfrm_policy_kill(ret);
835 	return ret;
836 }
837 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
838 
839 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
840 				     int dir, u32 id, int delete, int *err)
841 {
842 	struct xfrm_policy *pol, *ret;
843 	struct hlist_head *chain;
844 
845 	*err = -ENOENT;
846 	if (xfrm_policy_id2dir(id) != dir)
847 		return NULL;
848 
849 	*err = 0;
850 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
851 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
852 	ret = NULL;
853 	hlist_for_each_entry(pol, chain, byidx) {
854 		if (pol->type == type && pol->index == id &&
855 		    (mark & pol->mark.m) == pol->mark.v) {
856 			xfrm_pol_hold(pol);
857 			if (delete) {
858 				*err = security_xfrm_policy_delete(
859 								pol->security);
860 				if (*err) {
861 					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
862 					return pol;
863 				}
864 				__xfrm_policy_unlink(pol, dir);
865 			}
866 			ret = pol;
867 			break;
868 		}
869 	}
870 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
871 
872 	if (ret && delete)
873 		xfrm_policy_kill(ret);
874 	return ret;
875 }
876 EXPORT_SYMBOL(xfrm_policy_byid);
877 
878 #ifdef CONFIG_SECURITY_NETWORK_XFRM
879 static inline int
880 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
881 {
882 	int dir, err = 0;
883 
884 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
885 		struct xfrm_policy *pol;
886 		int i;
887 
888 		hlist_for_each_entry(pol,
889 				     &net->xfrm.policy_inexact[dir], bydst) {
890 			if (pol->type != type)
891 				continue;
892 			err = security_xfrm_policy_delete(pol->security);
893 			if (err) {
894 				xfrm_audit_policy_delete(pol, 0, task_valid);
895 				return err;
896 			}
897 		}
898 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
899 			hlist_for_each_entry(pol,
900 					     net->xfrm.policy_bydst[dir].table + i,
901 					     bydst) {
902 				if (pol->type != type)
903 					continue;
904 				err = security_xfrm_policy_delete(
905 								pol->security);
906 				if (err) {
907 					xfrm_audit_policy_delete(pol, 0,
908 								 task_valid);
909 					return err;
910 				}
911 			}
912 		}
913 	}
914 	return err;
915 }
916 #else
917 static inline int
918 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
919 {
920 	return 0;
921 }
922 #endif
923 
924 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
925 {
926 	int dir, err = 0, cnt = 0;
927 
928 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
929 
930 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
931 	if (err)
932 		goto out;
933 
934 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
935 		struct xfrm_policy *pol;
936 		int i;
937 
938 	again1:
939 		hlist_for_each_entry(pol,
940 				     &net->xfrm.policy_inexact[dir], bydst) {
941 			if (pol->type != type)
942 				continue;
943 			__xfrm_policy_unlink(pol, dir);
944 			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
945 			cnt++;
946 
947 			xfrm_audit_policy_delete(pol, 1, task_valid);
948 
949 			xfrm_policy_kill(pol);
950 
951 			spin_lock_bh(&net->xfrm.xfrm_policy_lock);
952 			goto again1;
953 		}
954 
955 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
956 	again2:
957 			hlist_for_each_entry(pol,
958 					     net->xfrm.policy_bydst[dir].table + i,
959 					     bydst) {
960 				if (pol->type != type)
961 					continue;
962 				__xfrm_policy_unlink(pol, dir);
963 				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
964 				cnt++;
965 
966 				xfrm_audit_policy_delete(pol, 1, task_valid);
967 				xfrm_policy_kill(pol);
968 
969 				spin_lock_bh(&net->xfrm.xfrm_policy_lock);
970 				goto again2;
971 			}
972 		}
973 
974 	}
975 	if (!cnt)
976 		err = -ESRCH;
977 	else
978 		xfrm_policy_cache_flush();
979 out:
980 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
981 	return err;
982 }
983 EXPORT_SYMBOL(xfrm_policy_flush);
984 
985 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
986 		     int (*func)(struct xfrm_policy *, int, int, void*),
987 		     void *data)
988 {
989 	struct xfrm_policy *pol;
990 	struct xfrm_policy_walk_entry *x;
991 	int error = 0;
992 
993 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
994 	    walk->type != XFRM_POLICY_TYPE_ANY)
995 		return -EINVAL;
996 
997 	if (list_empty(&walk->walk.all) && walk->seq != 0)
998 		return 0;
999 
1000 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1001 	if (list_empty(&walk->walk.all))
1002 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1003 	else
1004 		x = list_first_entry(&walk->walk.all,
1005 				     struct xfrm_policy_walk_entry, all);
1006 
1007 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1008 		if (x->dead)
1009 			continue;
1010 		pol = container_of(x, struct xfrm_policy, walk);
1011 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1012 		    walk->type != pol->type)
1013 			continue;
1014 		error = func(pol, xfrm_policy_id2dir(pol->index),
1015 			     walk->seq, data);
1016 		if (error) {
1017 			list_move_tail(&walk->walk.all, &x->all);
1018 			goto out;
1019 		}
1020 		walk->seq++;
1021 	}
1022 	if (walk->seq == 0) {
1023 		error = -ENOENT;
1024 		goto out;
1025 	}
1026 	list_del_init(&walk->walk.all);
1027 out:
1028 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1029 	return error;
1030 }
1031 EXPORT_SYMBOL(xfrm_policy_walk);
1032 
1033 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1034 {
1035 	INIT_LIST_HEAD(&walk->walk.all);
1036 	walk->walk.dead = 1;
1037 	walk->type = type;
1038 	walk->seq = 0;
1039 }
1040 EXPORT_SYMBOL(xfrm_policy_walk_init);
1041 
1042 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1043 {
1044 	if (list_empty(&walk->walk.all))
1045 		return;
1046 
1047 	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1048 	list_del(&walk->walk.all);
1049 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1050 }
1051 EXPORT_SYMBOL(xfrm_policy_walk_done);
1052 
1053 /*
1054  * Find policy to apply to this flow.
1055  *
1056  * Returns 0 if policy found, else an -errno.
1057  */
1058 static int xfrm_policy_match(const struct xfrm_policy *pol,
1059 			     const struct flowi *fl,
1060 			     u8 type, u16 family, int dir)
1061 {
1062 	const struct xfrm_selector *sel = &pol->selector;
1063 	int ret = -ESRCH;
1064 	bool match;
1065 
1066 	if (pol->family != family ||
1067 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1068 	    pol->type != type)
1069 		return ret;
1070 
1071 	match = xfrm_selector_match(sel, fl, family);
1072 	if (match)
1073 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1074 						  dir);
1075 
1076 	return ret;
1077 }
1078 
1079 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1080 						     const struct flowi *fl,
1081 						     u16 family, u8 dir)
1082 {
1083 	int err;
1084 	struct xfrm_policy *pol, *ret;
1085 	const xfrm_address_t *daddr, *saddr;
1086 	struct hlist_head *chain;
1087 	unsigned int sequence;
1088 	u32 priority;
1089 
1090 	daddr = xfrm_flowi_daddr(fl, family);
1091 	saddr = xfrm_flowi_saddr(fl, family);
1092 	if (unlikely(!daddr || !saddr))
1093 		return NULL;
1094 
1095 	rcu_read_lock();
1096  retry:
1097 	do {
1098 		sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1099 		chain = policy_hash_direct(net, daddr, saddr, family, dir);
1100 	} while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1101 
1102 	priority = ~0U;
1103 	ret = NULL;
1104 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1105 		err = xfrm_policy_match(pol, fl, type, family, dir);
1106 		if (err) {
1107 			if (err == -ESRCH)
1108 				continue;
1109 			else {
1110 				ret = ERR_PTR(err);
1111 				goto fail;
1112 			}
1113 		} else {
1114 			ret = pol;
1115 			priority = ret->priority;
1116 			break;
1117 		}
1118 	}
1119 	chain = &net->xfrm.policy_inexact[dir];
1120 	hlist_for_each_entry_rcu(pol, chain, bydst) {
1121 		if ((pol->priority >= priority) && ret)
1122 			break;
1123 
1124 		err = xfrm_policy_match(pol, fl, type, family, dir);
1125 		if (err) {
1126 			if (err == -ESRCH)
1127 				continue;
1128 			else {
1129 				ret = ERR_PTR(err);
1130 				goto fail;
1131 			}
1132 		} else {
1133 			ret = pol;
1134 			break;
1135 		}
1136 	}
1137 
1138 	if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1139 		goto retry;
1140 
1141 	if (ret && !xfrm_pol_hold_rcu(ret))
1142 		goto retry;
1143 fail:
1144 	rcu_read_unlock();
1145 
1146 	return ret;
1147 }
1148 
1149 static struct xfrm_policy *
1150 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1151 {
1152 #ifdef CONFIG_XFRM_SUB_POLICY
1153 	struct xfrm_policy *pol;
1154 
1155 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1156 	if (pol != NULL)
1157 		return pol;
1158 #endif
1159 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1160 }
1161 
1162 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1163 						 const struct flowi *fl, u16 family)
1164 {
1165 	struct xfrm_policy *pol;
1166 
1167 	rcu_read_lock();
1168  again:
1169 	pol = rcu_dereference(sk->sk_policy[dir]);
1170 	if (pol != NULL) {
1171 		bool match;
1172 		int err = 0;
1173 
1174 		if (pol->family != family) {
1175 			pol = NULL;
1176 			goto out;
1177 		}
1178 
1179 		match = xfrm_selector_match(&pol->selector, fl, family);
1180 		if (match) {
1181 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1182 				pol = NULL;
1183 				goto out;
1184 			}
1185 			err = security_xfrm_policy_lookup(pol->security,
1186 						      fl->flowi_secid,
1187 						      dir);
1188 			if (!err) {
1189 				if (!xfrm_pol_hold_rcu(pol))
1190 					goto again;
1191 			} else if (err == -ESRCH) {
1192 				pol = NULL;
1193 			} else {
1194 				pol = ERR_PTR(err);
1195 			}
1196 		} else
1197 			pol = NULL;
1198 	}
1199 out:
1200 	rcu_read_unlock();
1201 	return pol;
1202 }
1203 
1204 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1205 {
1206 	struct net *net = xp_net(pol);
1207 
1208 	list_add(&pol->walk.all, &net->xfrm.policy_all);
1209 	net->xfrm.policy_count[dir]++;
1210 	xfrm_pol_hold(pol);
1211 }
1212 
1213 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1214 						int dir)
1215 {
1216 	struct net *net = xp_net(pol);
1217 
1218 	if (list_empty(&pol->walk.all))
1219 		return NULL;
1220 
1221 	/* Socket policies are not hashed. */
1222 	if (!hlist_unhashed(&pol->bydst)) {
1223 		hlist_del_rcu(&pol->bydst);
1224 		hlist_del(&pol->byidx);
1225 	}
1226 
1227 	list_del_init(&pol->walk.all);
1228 	net->xfrm.policy_count[dir]--;
1229 
1230 	return pol;
1231 }
1232 
1233 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1234 {
1235 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1236 }
1237 
1238 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1239 {
1240 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1241 }
1242 
1243 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1244 {
1245 	struct net *net = xp_net(pol);
1246 
1247 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1248 	pol = __xfrm_policy_unlink(pol, dir);
1249 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1250 	if (pol) {
1251 		xfrm_policy_kill(pol);
1252 		return 0;
1253 	}
1254 	return -ENOENT;
1255 }
1256 EXPORT_SYMBOL(xfrm_policy_delete);
1257 
1258 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1259 {
1260 	struct net *net = sock_net(sk);
1261 	struct xfrm_policy *old_pol;
1262 
1263 #ifdef CONFIG_XFRM_SUB_POLICY
1264 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1265 		return -EINVAL;
1266 #endif
1267 
1268 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1269 	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1270 				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1271 	if (pol) {
1272 		pol->curlft.add_time = get_seconds();
1273 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1274 		xfrm_sk_policy_link(pol, dir);
1275 	}
1276 	rcu_assign_pointer(sk->sk_policy[dir], pol);
1277 	if (old_pol) {
1278 		if (pol)
1279 			xfrm_policy_requeue(old_pol, pol);
1280 
1281 		/* Unlinking succeeds always. This is the only function
1282 		 * allowed to delete or replace socket policy.
1283 		 */
1284 		xfrm_sk_policy_unlink(old_pol, dir);
1285 	}
1286 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1287 
1288 	if (old_pol) {
1289 		xfrm_policy_kill(old_pol);
1290 	}
1291 	return 0;
1292 }
1293 
1294 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1295 {
1296 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1297 	struct net *net = xp_net(old);
1298 
1299 	if (newp) {
1300 		newp->selector = old->selector;
1301 		if (security_xfrm_policy_clone(old->security,
1302 					       &newp->security)) {
1303 			kfree(newp);
1304 			return NULL;  /* ENOMEM */
1305 		}
1306 		newp->lft = old->lft;
1307 		newp->curlft = old->curlft;
1308 		newp->mark = old->mark;
1309 		newp->action = old->action;
1310 		newp->flags = old->flags;
1311 		newp->xfrm_nr = old->xfrm_nr;
1312 		newp->index = old->index;
1313 		newp->type = old->type;
1314 		newp->family = old->family;
1315 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1316 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1317 		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1318 		xfrm_sk_policy_link(newp, dir);
1319 		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1320 		xfrm_pol_put(newp);
1321 	}
1322 	return newp;
1323 }
1324 
1325 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1326 {
1327 	const struct xfrm_policy *p;
1328 	struct xfrm_policy *np;
1329 	int i, ret = 0;
1330 
1331 	rcu_read_lock();
1332 	for (i = 0; i < 2; i++) {
1333 		p = rcu_dereference(osk->sk_policy[i]);
1334 		if (p) {
1335 			np = clone_policy(p, i);
1336 			if (unlikely(!np)) {
1337 				ret = -ENOMEM;
1338 				break;
1339 			}
1340 			rcu_assign_pointer(sk->sk_policy[i], np);
1341 		}
1342 	}
1343 	rcu_read_unlock();
1344 	return ret;
1345 }
1346 
1347 static int
1348 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1349 	       xfrm_address_t *remote, unsigned short family, u32 mark)
1350 {
1351 	int err;
1352 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1353 
1354 	if (unlikely(afinfo == NULL))
1355 		return -EINVAL;
1356 	err = afinfo->get_saddr(net, oif, local, remote, mark);
1357 	rcu_read_unlock();
1358 	return err;
1359 }
1360 
1361 /* Resolve list of templates for the flow, given policy. */
1362 
1363 static int
1364 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1365 		      struct xfrm_state **xfrm, unsigned short family)
1366 {
1367 	struct net *net = xp_net(policy);
1368 	int nx;
1369 	int i, error;
1370 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1371 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1372 	xfrm_address_t tmp;
1373 
1374 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1375 		struct xfrm_state *x;
1376 		xfrm_address_t *remote = daddr;
1377 		xfrm_address_t *local  = saddr;
1378 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1379 
1380 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1381 		    tmpl->mode == XFRM_MODE_BEET) {
1382 			remote = &tmpl->id.daddr;
1383 			local = &tmpl->saddr;
1384 			if (xfrm_addr_any(local, tmpl->encap_family)) {
1385 				error = xfrm_get_saddr(net, fl->flowi_oif,
1386 						       &tmp, remote,
1387 						       tmpl->encap_family, 0);
1388 				if (error)
1389 					goto fail;
1390 				local = &tmp;
1391 			}
1392 		}
1393 
1394 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1395 
1396 		if (x && x->km.state == XFRM_STATE_VALID) {
1397 			xfrm[nx++] = x;
1398 			daddr = remote;
1399 			saddr = local;
1400 			continue;
1401 		}
1402 		if (x) {
1403 			error = (x->km.state == XFRM_STATE_ERROR ?
1404 				 -EINVAL : -EAGAIN);
1405 			xfrm_state_put(x);
1406 		} else if (error == -ESRCH) {
1407 			error = -EAGAIN;
1408 		}
1409 
1410 		if (!tmpl->optional)
1411 			goto fail;
1412 	}
1413 	return nx;
1414 
1415 fail:
1416 	for (nx--; nx >= 0; nx--)
1417 		xfrm_state_put(xfrm[nx]);
1418 	return error;
1419 }
1420 
1421 static int
1422 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1423 		  struct xfrm_state **xfrm, unsigned short family)
1424 {
1425 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1426 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1427 	int cnx = 0;
1428 	int error;
1429 	int ret;
1430 	int i;
1431 
1432 	for (i = 0; i < npols; i++) {
1433 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1434 			error = -ENOBUFS;
1435 			goto fail;
1436 		}
1437 
1438 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1439 		if (ret < 0) {
1440 			error = ret;
1441 			goto fail;
1442 		} else
1443 			cnx += ret;
1444 	}
1445 
1446 	/* found states are sorted for outbound processing */
1447 	if (npols > 1)
1448 		xfrm_state_sort(xfrm, tpp, cnx, family);
1449 
1450 	return cnx;
1451 
1452  fail:
1453 	for (cnx--; cnx >= 0; cnx--)
1454 		xfrm_state_put(tpp[cnx]);
1455 	return error;
1456 
1457 }
1458 
1459 static int xfrm_get_tos(const struct flowi *fl, int family)
1460 {
1461 	const struct xfrm_policy_afinfo *afinfo;
1462 	int tos = 0;
1463 
1464 	afinfo = xfrm_policy_get_afinfo(family);
1465 	tos = afinfo ? afinfo->get_tos(fl) : 0;
1466 
1467 	rcu_read_unlock();
1468 
1469 	return tos;
1470 }
1471 
1472 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1473 {
1474 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1475 	struct dst_ops *dst_ops;
1476 	struct xfrm_dst *xdst;
1477 
1478 	if (!afinfo)
1479 		return ERR_PTR(-EINVAL);
1480 
1481 	switch (family) {
1482 	case AF_INET:
1483 		dst_ops = &net->xfrm.xfrm4_dst_ops;
1484 		break;
1485 #if IS_ENABLED(CONFIG_IPV6)
1486 	case AF_INET6:
1487 		dst_ops = &net->xfrm.xfrm6_dst_ops;
1488 		break;
1489 #endif
1490 	default:
1491 		BUG();
1492 	}
1493 	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1494 
1495 	if (likely(xdst)) {
1496 		struct dst_entry *dst = &xdst->u.dst;
1497 
1498 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1499 	} else
1500 		xdst = ERR_PTR(-ENOBUFS);
1501 
1502 	rcu_read_unlock();
1503 
1504 	return xdst;
1505 }
1506 
1507 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1508 				 int nfheader_len)
1509 {
1510 	const struct xfrm_policy_afinfo *afinfo =
1511 		xfrm_policy_get_afinfo(dst->ops->family);
1512 	int err;
1513 
1514 	if (!afinfo)
1515 		return -EINVAL;
1516 
1517 	err = afinfo->init_path(path, dst, nfheader_len);
1518 
1519 	rcu_read_unlock();
1520 
1521 	return err;
1522 }
1523 
1524 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1525 				const struct flowi *fl)
1526 {
1527 	const struct xfrm_policy_afinfo *afinfo =
1528 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1529 	int err;
1530 
1531 	if (!afinfo)
1532 		return -EINVAL;
1533 
1534 	err = afinfo->fill_dst(xdst, dev, fl);
1535 
1536 	rcu_read_unlock();
1537 
1538 	return err;
1539 }
1540 
1541 
1542 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1543  * all the metrics... Shortly, bundle a bundle.
1544  */
1545 
1546 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1547 					    struct xfrm_state **xfrm,
1548 					    struct xfrm_dst **bundle,
1549 					    int nx,
1550 					    const struct flowi *fl,
1551 					    struct dst_entry *dst)
1552 {
1553 	struct net *net = xp_net(policy);
1554 	unsigned long now = jiffies;
1555 	struct net_device *dev;
1556 	struct xfrm_mode *inner_mode;
1557 	struct xfrm_dst *xdst_prev = NULL;
1558 	struct xfrm_dst *xdst0 = NULL;
1559 	int i = 0;
1560 	int err;
1561 	int header_len = 0;
1562 	int nfheader_len = 0;
1563 	int trailer_len = 0;
1564 	int tos;
1565 	int family = policy->selector.family;
1566 	xfrm_address_t saddr, daddr;
1567 
1568 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1569 
1570 	tos = xfrm_get_tos(fl, family);
1571 
1572 	dst_hold(dst);
1573 
1574 	for (; i < nx; i++) {
1575 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1576 		struct dst_entry *dst1 = &xdst->u.dst;
1577 
1578 		err = PTR_ERR(xdst);
1579 		if (IS_ERR(xdst)) {
1580 			dst_release(dst);
1581 			goto put_states;
1582 		}
1583 
1584 		bundle[i] = xdst;
1585 		if (!xdst_prev)
1586 			xdst0 = xdst;
1587 		else
1588 			/* Ref count is taken during xfrm_alloc_dst()
1589 			 * No need to do dst_clone() on dst1
1590 			 */
1591 			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
1592 
1593 		if (xfrm[i]->sel.family == AF_UNSPEC) {
1594 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
1595 							xfrm_af2proto(family));
1596 			if (!inner_mode) {
1597 				err = -EAFNOSUPPORT;
1598 				dst_release(dst);
1599 				goto put_states;
1600 			}
1601 		} else
1602 			inner_mode = xfrm[i]->inner_mode;
1603 
1604 		xdst->route = dst;
1605 		dst_copy_metrics(dst1, dst);
1606 
1607 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1608 			family = xfrm[i]->props.family;
1609 			dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1610 					      &saddr, &daddr, family,
1611 					      xfrm[i]->props.output_mark);
1612 			err = PTR_ERR(dst);
1613 			if (IS_ERR(dst))
1614 				goto put_states;
1615 		} else
1616 			dst_hold(dst);
1617 
1618 		dst1->xfrm = xfrm[i];
1619 		xdst->xfrm_genid = xfrm[i]->genid;
1620 
1621 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1622 		dst1->flags |= DST_HOST;
1623 		dst1->lastuse = now;
1624 
1625 		dst1->input = dst_discard;
1626 		dst1->output = inner_mode->afinfo->output;
1627 
1628 		xdst_prev = xdst;
1629 
1630 		header_len += xfrm[i]->props.header_len;
1631 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1632 			nfheader_len += xfrm[i]->props.header_len;
1633 		trailer_len += xfrm[i]->props.trailer_len;
1634 	}
1635 
1636 	xfrm_dst_set_child(xdst_prev, dst);
1637 	xdst0->path = dst;
1638 
1639 	err = -ENODEV;
1640 	dev = dst->dev;
1641 	if (!dev)
1642 		goto free_dst;
1643 
1644 	xfrm_init_path(xdst0, dst, nfheader_len);
1645 	xfrm_init_pmtu(bundle, nx);
1646 
1647 	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
1648 	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
1649 		err = xfrm_fill_dst(xdst_prev, dev, fl);
1650 		if (err)
1651 			goto free_dst;
1652 
1653 		xdst_prev->u.dst.header_len = header_len;
1654 		xdst_prev->u.dst.trailer_len = trailer_len;
1655 		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
1656 		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
1657 	}
1658 
1659 out:
1660 	return &xdst0->u.dst;
1661 
1662 put_states:
1663 	for (; i < nx; i++)
1664 		xfrm_state_put(xfrm[i]);
1665 free_dst:
1666 	if (xdst0)
1667 		dst_release_immediate(&xdst0->u.dst);
1668 	xdst0 = ERR_PTR(err);
1669 	goto out;
1670 }
1671 
1672 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1673 				struct xfrm_policy **pols,
1674 				int *num_pols, int *num_xfrms)
1675 {
1676 	int i;
1677 
1678 	if (*num_pols == 0 || !pols[0]) {
1679 		*num_pols = 0;
1680 		*num_xfrms = 0;
1681 		return 0;
1682 	}
1683 	if (IS_ERR(pols[0]))
1684 		return PTR_ERR(pols[0]);
1685 
1686 	*num_xfrms = pols[0]->xfrm_nr;
1687 
1688 #ifdef CONFIG_XFRM_SUB_POLICY
1689 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1690 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1691 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1692 						    XFRM_POLICY_TYPE_MAIN,
1693 						    fl, family,
1694 						    XFRM_POLICY_OUT);
1695 		if (pols[1]) {
1696 			if (IS_ERR(pols[1])) {
1697 				xfrm_pols_put(pols, *num_pols);
1698 				return PTR_ERR(pols[1]);
1699 			}
1700 			(*num_pols)++;
1701 			(*num_xfrms) += pols[1]->xfrm_nr;
1702 		}
1703 	}
1704 #endif
1705 	for (i = 0; i < *num_pols; i++) {
1706 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
1707 			*num_xfrms = -1;
1708 			break;
1709 		}
1710 	}
1711 
1712 	return 0;
1713 
1714 }
1715 
1716 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1717 {
1718 	this_cpu_write(xfrm_last_dst, xdst);
1719 	if (old)
1720 		dst_release(&old->u.dst);
1721 }
1722 
1723 static void __xfrm_pcpu_work_fn(void)
1724 {
1725 	struct xfrm_dst *old;
1726 
1727 	old = this_cpu_read(xfrm_last_dst);
1728 	if (old && !xfrm_bundle_ok(old))
1729 		xfrm_last_dst_update(NULL, old);
1730 }
1731 
1732 static void xfrm_pcpu_work_fn(struct work_struct *work)
1733 {
1734 	local_bh_disable();
1735 	rcu_read_lock();
1736 	__xfrm_pcpu_work_fn();
1737 	rcu_read_unlock();
1738 	local_bh_enable();
1739 }
1740 
1741 void xfrm_policy_cache_flush(void)
1742 {
1743 	struct xfrm_dst *old;
1744 	bool found = 0;
1745 	int cpu;
1746 
1747 	local_bh_disable();
1748 	rcu_read_lock();
1749 	for_each_possible_cpu(cpu) {
1750 		old = per_cpu(xfrm_last_dst, cpu);
1751 		if (old && !xfrm_bundle_ok(old)) {
1752 			if (smp_processor_id() == cpu) {
1753 				__xfrm_pcpu_work_fn();
1754 				continue;
1755 			}
1756 			found = true;
1757 			break;
1758 		}
1759 	}
1760 
1761 	rcu_read_unlock();
1762 	local_bh_enable();
1763 
1764 	if (!found)
1765 		return;
1766 
1767 	get_online_cpus();
1768 
1769 	for_each_possible_cpu(cpu) {
1770 		bool bundle_release;
1771 
1772 		rcu_read_lock();
1773 		old = per_cpu(xfrm_last_dst, cpu);
1774 		bundle_release = old && !xfrm_bundle_ok(old);
1775 		rcu_read_unlock();
1776 
1777 		if (!bundle_release)
1778 			continue;
1779 
1780 		if (cpu_online(cpu)) {
1781 			schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1782 			continue;
1783 		}
1784 
1785 		rcu_read_lock();
1786 		old = per_cpu(xfrm_last_dst, cpu);
1787 		if (old && !xfrm_bundle_ok(old)) {
1788 			per_cpu(xfrm_last_dst, cpu) = NULL;
1789 			dst_release(&old->u.dst);
1790 		}
1791 		rcu_read_unlock();
1792 	}
1793 
1794 	put_online_cpus();
1795 }
1796 
1797 static bool xfrm_xdst_can_reuse(struct xfrm_dst *xdst,
1798 				struct xfrm_state * const xfrm[],
1799 				int num)
1800 {
1801 	const struct dst_entry *dst = &xdst->u.dst;
1802 	int i;
1803 
1804 	if (xdst->num_xfrms != num)
1805 		return false;
1806 
1807 	for (i = 0; i < num; i++) {
1808 		if (!dst || dst->xfrm != xfrm[i])
1809 			return false;
1810 		dst = xfrm_dst_child(dst);
1811 	}
1812 
1813 	return xfrm_bundle_ok(xdst);
1814 }
1815 
1816 static struct xfrm_dst *
1817 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1818 			       const struct flowi *fl, u16 family,
1819 			       struct dst_entry *dst_orig)
1820 {
1821 	struct net *net = xp_net(pols[0]);
1822 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1823 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
1824 	struct xfrm_dst *xdst, *old;
1825 	struct dst_entry *dst;
1826 	int err;
1827 
1828 	/* Try to instantiate a bundle */
1829 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1830 	if (err <= 0) {
1831 		if (err != 0 && err != -EAGAIN)
1832 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1833 		return ERR_PTR(err);
1834 	}
1835 
1836 	xdst = this_cpu_read(xfrm_last_dst);
1837 	if (xdst &&
1838 	    xdst->u.dst.dev == dst_orig->dev &&
1839 	    xdst->num_pols == num_pols &&
1840 	    memcmp(xdst->pols, pols,
1841 		   sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1842 	    xfrm_xdst_can_reuse(xdst, xfrm, err)) {
1843 		dst_hold(&xdst->u.dst);
1844 		xfrm_pols_put(pols, num_pols);
1845 		while (err > 0)
1846 			xfrm_state_put(xfrm[--err]);
1847 		return xdst;
1848 	}
1849 
1850 	old = xdst;
1851 
1852 	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
1853 	if (IS_ERR(dst)) {
1854 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1855 		return ERR_CAST(dst);
1856 	}
1857 
1858 	xdst = (struct xfrm_dst *)dst;
1859 	xdst->num_xfrms = err;
1860 	xdst->num_pols = num_pols;
1861 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1862 	xdst->policy_genid = atomic_read(&pols[0]->genid);
1863 
1864 	atomic_set(&xdst->u.dst.__refcnt, 2);
1865 	xfrm_last_dst_update(xdst, old);
1866 
1867 	return xdst;
1868 }
1869 
1870 static void xfrm_policy_queue_process(struct timer_list *t)
1871 {
1872 	struct sk_buff *skb;
1873 	struct sock *sk;
1874 	struct dst_entry *dst;
1875 	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1876 	struct net *net = xp_net(pol);
1877 	struct xfrm_policy_queue *pq = &pol->polq;
1878 	struct flowi fl;
1879 	struct sk_buff_head list;
1880 
1881 	spin_lock(&pq->hold_queue.lock);
1882 	skb = skb_peek(&pq->hold_queue);
1883 	if (!skb) {
1884 		spin_unlock(&pq->hold_queue.lock);
1885 		goto out;
1886 	}
1887 	dst = skb_dst(skb);
1888 	sk = skb->sk;
1889 	xfrm_decode_session(skb, &fl, dst->ops->family);
1890 	spin_unlock(&pq->hold_queue.lock);
1891 
1892 	dst_hold(xfrm_dst_path(dst));
1893 	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, 0);
1894 	if (IS_ERR(dst))
1895 		goto purge_queue;
1896 
1897 	if (dst->flags & DST_XFRM_QUEUE) {
1898 		dst_release(dst);
1899 
1900 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1901 			goto purge_queue;
1902 
1903 		pq->timeout = pq->timeout << 1;
1904 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1905 			xfrm_pol_hold(pol);
1906 	goto out;
1907 	}
1908 
1909 	dst_release(dst);
1910 
1911 	__skb_queue_head_init(&list);
1912 
1913 	spin_lock(&pq->hold_queue.lock);
1914 	pq->timeout = 0;
1915 	skb_queue_splice_init(&pq->hold_queue, &list);
1916 	spin_unlock(&pq->hold_queue.lock);
1917 
1918 	while (!skb_queue_empty(&list)) {
1919 		skb = __skb_dequeue(&list);
1920 
1921 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1922 		dst_hold(xfrm_dst_path(skb_dst(skb)));
1923 		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
1924 		if (IS_ERR(dst)) {
1925 			kfree_skb(skb);
1926 			continue;
1927 		}
1928 
1929 		nf_reset(skb);
1930 		skb_dst_drop(skb);
1931 		skb_dst_set(skb, dst);
1932 
1933 		dst_output(net, skb->sk, skb);
1934 	}
1935 
1936 out:
1937 	xfrm_pol_put(pol);
1938 	return;
1939 
1940 purge_queue:
1941 	pq->timeout = 0;
1942 	skb_queue_purge(&pq->hold_queue);
1943 	xfrm_pol_put(pol);
1944 }
1945 
1946 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1947 {
1948 	unsigned long sched_next;
1949 	struct dst_entry *dst = skb_dst(skb);
1950 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1951 	struct xfrm_policy *pol = xdst->pols[0];
1952 	struct xfrm_policy_queue *pq = &pol->polq;
1953 
1954 	if (unlikely(skb_fclone_busy(sk, skb))) {
1955 		kfree_skb(skb);
1956 		return 0;
1957 	}
1958 
1959 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1960 		kfree_skb(skb);
1961 		return -EAGAIN;
1962 	}
1963 
1964 	skb_dst_force(skb);
1965 
1966 	spin_lock_bh(&pq->hold_queue.lock);
1967 
1968 	if (!pq->timeout)
1969 		pq->timeout = XFRM_QUEUE_TMO_MIN;
1970 
1971 	sched_next = jiffies + pq->timeout;
1972 
1973 	if (del_timer(&pq->hold_timer)) {
1974 		if (time_before(pq->hold_timer.expires, sched_next))
1975 			sched_next = pq->hold_timer.expires;
1976 		xfrm_pol_put(pol);
1977 	}
1978 
1979 	__skb_queue_tail(&pq->hold_queue, skb);
1980 	if (!mod_timer(&pq->hold_timer, sched_next))
1981 		xfrm_pol_hold(pol);
1982 
1983 	spin_unlock_bh(&pq->hold_queue.lock);
1984 
1985 	return 0;
1986 }
1987 
1988 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1989 						 struct xfrm_flo *xflo,
1990 						 const struct flowi *fl,
1991 						 int num_xfrms,
1992 						 u16 family)
1993 {
1994 	int err;
1995 	struct net_device *dev;
1996 	struct dst_entry *dst;
1997 	struct dst_entry *dst1;
1998 	struct xfrm_dst *xdst;
1999 
2000 	xdst = xfrm_alloc_dst(net, family);
2001 	if (IS_ERR(xdst))
2002 		return xdst;
2003 
2004 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2005 	    net->xfrm.sysctl_larval_drop ||
2006 	    num_xfrms <= 0)
2007 		return xdst;
2008 
2009 	dst = xflo->dst_orig;
2010 	dst1 = &xdst->u.dst;
2011 	dst_hold(dst);
2012 	xdst->route = dst;
2013 
2014 	dst_copy_metrics(dst1, dst);
2015 
2016 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2017 	dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2018 	dst1->lastuse = jiffies;
2019 
2020 	dst1->input = dst_discard;
2021 	dst1->output = xdst_queue_output;
2022 
2023 	dst_hold(dst);
2024 	xfrm_dst_set_child(xdst, dst);
2025 	xdst->path = dst;
2026 
2027 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2028 
2029 	err = -ENODEV;
2030 	dev = dst->dev;
2031 	if (!dev)
2032 		goto free_dst;
2033 
2034 	err = xfrm_fill_dst(xdst, dev, fl);
2035 	if (err)
2036 		goto free_dst;
2037 
2038 out:
2039 	return xdst;
2040 
2041 free_dst:
2042 	dst_release(dst1);
2043 	xdst = ERR_PTR(err);
2044 	goto out;
2045 }
2046 
2047 static struct xfrm_dst *
2048 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2049 {
2050 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2051 	int num_pols = 0, num_xfrms = 0, err;
2052 	struct xfrm_dst *xdst;
2053 
2054 	/* Resolve policies to use if we couldn't get them from
2055 	 * previous cache entry */
2056 	num_pols = 1;
2057 	pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2058 	err = xfrm_expand_policies(fl, family, pols,
2059 					   &num_pols, &num_xfrms);
2060 	if (err < 0)
2061 		goto inc_error;
2062 	if (num_pols == 0)
2063 		return NULL;
2064 	if (num_xfrms <= 0)
2065 		goto make_dummy_bundle;
2066 
2067 	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2068 						  xflo->dst_orig);
2069 	if (IS_ERR(xdst)) {
2070 		err = PTR_ERR(xdst);
2071 		if (err != -EAGAIN)
2072 			goto error;
2073 		goto make_dummy_bundle;
2074 	} else if (xdst == NULL) {
2075 		num_xfrms = 0;
2076 		goto make_dummy_bundle;
2077 	}
2078 
2079 	return xdst;
2080 
2081 make_dummy_bundle:
2082 	/* We found policies, but there's no bundles to instantiate:
2083 	 * either because the policy blocks, has no transformations or
2084 	 * we could not build template (no xfrm_states).*/
2085 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2086 	if (IS_ERR(xdst)) {
2087 		xfrm_pols_put(pols, num_pols);
2088 		return ERR_CAST(xdst);
2089 	}
2090 	xdst->num_pols = num_pols;
2091 	xdst->num_xfrms = num_xfrms;
2092 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2093 
2094 	return xdst;
2095 
2096 inc_error:
2097 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2098 error:
2099 	xfrm_pols_put(pols, num_pols);
2100 	return ERR_PTR(err);
2101 }
2102 
2103 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2104 					struct dst_entry *dst_orig)
2105 {
2106 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2107 	struct dst_entry *ret;
2108 
2109 	if (!afinfo) {
2110 		dst_release(dst_orig);
2111 		return ERR_PTR(-EINVAL);
2112 	} else {
2113 		ret = afinfo->blackhole_route(net, dst_orig);
2114 	}
2115 	rcu_read_unlock();
2116 
2117 	return ret;
2118 }
2119 
2120 /* Main function: finds/creates a bundle for given flow.
2121  *
2122  * At the moment we eat a raw IP route. Mostly to speed up lookups
2123  * on interfaces with disabled IPsec.
2124  */
2125 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2126 			      const struct flowi *fl,
2127 			      const struct sock *sk, int flags)
2128 {
2129 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2130 	struct xfrm_dst *xdst;
2131 	struct dst_entry *dst, *route;
2132 	u16 family = dst_orig->ops->family;
2133 	u8 dir = XFRM_POLICY_OUT;
2134 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2135 
2136 	dst = NULL;
2137 	xdst = NULL;
2138 	route = NULL;
2139 
2140 	sk = sk_const_to_full_sk(sk);
2141 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2142 		num_pols = 1;
2143 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2144 		err = xfrm_expand_policies(fl, family, pols,
2145 					   &num_pols, &num_xfrms);
2146 		if (err < 0)
2147 			goto dropdst;
2148 
2149 		if (num_pols) {
2150 			if (num_xfrms <= 0) {
2151 				drop_pols = num_pols;
2152 				goto no_transform;
2153 			}
2154 
2155 			xdst = xfrm_resolve_and_create_bundle(
2156 					pols, num_pols, fl,
2157 					family, dst_orig);
2158 			if (IS_ERR(xdst)) {
2159 				xfrm_pols_put(pols, num_pols);
2160 				err = PTR_ERR(xdst);
2161 				goto dropdst;
2162 			} else if (xdst == NULL) {
2163 				num_xfrms = 0;
2164 				drop_pols = num_pols;
2165 				goto no_transform;
2166 			}
2167 
2168 			route = xdst->route;
2169 		}
2170 	}
2171 
2172 	if (xdst == NULL) {
2173 		struct xfrm_flo xflo;
2174 
2175 		xflo.dst_orig = dst_orig;
2176 		xflo.flags = flags;
2177 
2178 		/* To accelerate a bit...  */
2179 		if ((dst_orig->flags & DST_NOXFRM) ||
2180 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
2181 			goto nopol;
2182 
2183 		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2184 		if (xdst == NULL)
2185 			goto nopol;
2186 		if (IS_ERR(xdst)) {
2187 			err = PTR_ERR(xdst);
2188 			goto dropdst;
2189 		}
2190 
2191 		num_pols = xdst->num_pols;
2192 		num_xfrms = xdst->num_xfrms;
2193 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2194 		route = xdst->route;
2195 	}
2196 
2197 	dst = &xdst->u.dst;
2198 	if (route == NULL && num_xfrms > 0) {
2199 		/* The only case when xfrm_bundle_lookup() returns a
2200 		 * bundle with null route, is when the template could
2201 		 * not be resolved. It means policies are there, but
2202 		 * bundle could not be created, since we don't yet
2203 		 * have the xfrm_state's. We need to wait for KM to
2204 		 * negotiate new SA's or bail out with error.*/
2205 		if (net->xfrm.sysctl_larval_drop) {
2206 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2207 			err = -EREMOTE;
2208 			goto error;
2209 		}
2210 
2211 		err = -EAGAIN;
2212 
2213 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2214 		goto error;
2215 	}
2216 
2217 no_transform:
2218 	if (num_pols == 0)
2219 		goto nopol;
2220 
2221 	if ((flags & XFRM_LOOKUP_ICMP) &&
2222 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2223 		err = -ENOENT;
2224 		goto error;
2225 	}
2226 
2227 	for (i = 0; i < num_pols; i++)
2228 		pols[i]->curlft.use_time = get_seconds();
2229 
2230 	if (num_xfrms < 0) {
2231 		/* Prohibit the flow */
2232 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2233 		err = -EPERM;
2234 		goto error;
2235 	} else if (num_xfrms > 0) {
2236 		/* Flow transformed */
2237 		dst_release(dst_orig);
2238 	} else {
2239 		/* Flow passes untransformed */
2240 		dst_release(dst);
2241 		dst = dst_orig;
2242 	}
2243 ok:
2244 	xfrm_pols_put(pols, drop_pols);
2245 	if (dst && dst->xfrm &&
2246 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2247 		dst->flags |= DST_XFRM_TUNNEL;
2248 	return dst;
2249 
2250 nopol:
2251 	if (!(flags & XFRM_LOOKUP_ICMP)) {
2252 		dst = dst_orig;
2253 		goto ok;
2254 	}
2255 	err = -ENOENT;
2256 error:
2257 	dst_release(dst);
2258 dropdst:
2259 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2260 		dst_release(dst_orig);
2261 	xfrm_pols_put(pols, drop_pols);
2262 	return ERR_PTR(err);
2263 }
2264 EXPORT_SYMBOL(xfrm_lookup);
2265 
2266 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2267  * Otherwise we may send out blackholed packets.
2268  */
2269 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2270 				    const struct flowi *fl,
2271 				    const struct sock *sk, int flags)
2272 {
2273 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2274 					    flags | XFRM_LOOKUP_QUEUE |
2275 					    XFRM_LOOKUP_KEEP_DST_REF);
2276 
2277 	if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2278 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
2279 
2280 	return dst;
2281 }
2282 EXPORT_SYMBOL(xfrm_lookup_route);
2283 
2284 static inline int
2285 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2286 {
2287 	struct xfrm_state *x;
2288 
2289 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2290 		return 0;
2291 	x = skb->sp->xvec[idx];
2292 	if (!x->type->reject)
2293 		return 0;
2294 	return x->type->reject(x, skb, fl);
2295 }
2296 
2297 /* When skb is transformed back to its "native" form, we have to
2298  * check policy restrictions. At the moment we make this in maximally
2299  * stupid way. Shame on me. :-) Of course, connected sockets must
2300  * have policy cached at them.
2301  */
2302 
2303 static inline int
2304 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2305 	      unsigned short family)
2306 {
2307 	if (xfrm_state_kern(x))
2308 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2309 	return	x->id.proto == tmpl->id.proto &&
2310 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2311 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2312 		x->props.mode == tmpl->mode &&
2313 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2314 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2315 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
2316 		  xfrm_state_addr_cmp(tmpl, x, family));
2317 }
2318 
2319 /*
2320  * 0 or more than 0 is returned when validation is succeeded (either bypass
2321  * because of optional transport mode, or next index of the mathced secpath
2322  * state with the template.
2323  * -1 is returned when no matching template is found.
2324  * Otherwise "-2 - errored_index" is returned.
2325  */
2326 static inline int
2327 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2328 	       unsigned short family)
2329 {
2330 	int idx = start;
2331 
2332 	if (tmpl->optional) {
2333 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
2334 			return start;
2335 	} else
2336 		start = -1;
2337 	for (; idx < sp->len; idx++) {
2338 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2339 			return ++idx;
2340 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2341 			if (start == -1)
2342 				start = -2-idx;
2343 			break;
2344 		}
2345 	}
2346 	return start;
2347 }
2348 
2349 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2350 			  unsigned int family, int reverse)
2351 {
2352 	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2353 	int err;
2354 
2355 	if (unlikely(afinfo == NULL))
2356 		return -EAFNOSUPPORT;
2357 
2358 	afinfo->decode_session(skb, fl, reverse);
2359 	err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2360 	rcu_read_unlock();
2361 	return err;
2362 }
2363 EXPORT_SYMBOL(__xfrm_decode_session);
2364 
2365 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2366 {
2367 	for (; k < sp->len; k++) {
2368 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2369 			*idxp = k;
2370 			return 1;
2371 		}
2372 	}
2373 
2374 	return 0;
2375 }
2376 
2377 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2378 			unsigned short family)
2379 {
2380 	struct net *net = dev_net(skb->dev);
2381 	struct xfrm_policy *pol;
2382 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2383 	int npols = 0;
2384 	int xfrm_nr;
2385 	int pi;
2386 	int reverse;
2387 	struct flowi fl;
2388 	int xerr_idx = -1;
2389 
2390 	reverse = dir & ~XFRM_POLICY_MASK;
2391 	dir &= XFRM_POLICY_MASK;
2392 
2393 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2394 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2395 		return 0;
2396 	}
2397 
2398 	nf_nat_decode_session(skb, &fl, family);
2399 
2400 	/* First, check used SA against their selectors. */
2401 	if (skb->sp) {
2402 		int i;
2403 
2404 		for (i = skb->sp->len-1; i >= 0; i--) {
2405 			struct xfrm_state *x = skb->sp->xvec[i];
2406 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
2407 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2408 				return 0;
2409 			}
2410 		}
2411 	}
2412 
2413 	pol = NULL;
2414 	sk = sk_to_full_sk(sk);
2415 	if (sk && sk->sk_policy[dir]) {
2416 		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2417 		if (IS_ERR(pol)) {
2418 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2419 			return 0;
2420 		}
2421 	}
2422 
2423 	if (!pol)
2424 		pol = xfrm_policy_lookup(net, &fl, family, dir);
2425 
2426 	if (IS_ERR(pol)) {
2427 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2428 		return 0;
2429 	}
2430 
2431 	if (!pol) {
2432 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2433 			xfrm_secpath_reject(xerr_idx, skb, &fl);
2434 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2435 			return 0;
2436 		}
2437 		return 1;
2438 	}
2439 
2440 	pol->curlft.use_time = get_seconds();
2441 
2442 	pols[0] = pol;
2443 	npols++;
2444 #ifdef CONFIG_XFRM_SUB_POLICY
2445 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2446 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2447 						    &fl, family,
2448 						    XFRM_POLICY_IN);
2449 		if (pols[1]) {
2450 			if (IS_ERR(pols[1])) {
2451 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2452 				return 0;
2453 			}
2454 			pols[1]->curlft.use_time = get_seconds();
2455 			npols++;
2456 		}
2457 	}
2458 #endif
2459 
2460 	if (pol->action == XFRM_POLICY_ALLOW) {
2461 		struct sec_path *sp;
2462 		static struct sec_path dummy;
2463 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2464 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2465 		struct xfrm_tmpl **tpp = tp;
2466 		int ti = 0;
2467 		int i, k;
2468 
2469 		if ((sp = skb->sp) == NULL)
2470 			sp = &dummy;
2471 
2472 		for (pi = 0; pi < npols; pi++) {
2473 			if (pols[pi] != pol &&
2474 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
2475 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2476 				goto reject;
2477 			}
2478 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2479 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2480 				goto reject_error;
2481 			}
2482 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
2483 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
2484 		}
2485 		xfrm_nr = ti;
2486 		if (npols > 1) {
2487 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2488 			tpp = stp;
2489 		}
2490 
2491 		/* For each tunnel xfrm, find the first matching tmpl.
2492 		 * For each tmpl before that, find corresponding xfrm.
2493 		 * Order is _important_. Later we will implement
2494 		 * some barriers, but at the moment barriers
2495 		 * are implied between each two transformations.
2496 		 */
2497 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2498 			k = xfrm_policy_ok(tpp[i], sp, k, family);
2499 			if (k < 0) {
2500 				if (k < -1)
2501 					/* "-2 - errored_index" returned */
2502 					xerr_idx = -(2+k);
2503 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2504 				goto reject;
2505 			}
2506 		}
2507 
2508 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2509 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2510 			goto reject;
2511 		}
2512 
2513 		xfrm_pols_put(pols, npols);
2514 		return 1;
2515 	}
2516 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2517 
2518 reject:
2519 	xfrm_secpath_reject(xerr_idx, skb, &fl);
2520 reject_error:
2521 	xfrm_pols_put(pols, npols);
2522 	return 0;
2523 }
2524 EXPORT_SYMBOL(__xfrm_policy_check);
2525 
2526 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2527 {
2528 	struct net *net = dev_net(skb->dev);
2529 	struct flowi fl;
2530 	struct dst_entry *dst;
2531 	int res = 1;
2532 
2533 	if (xfrm_decode_session(skb, &fl, family) < 0) {
2534 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2535 		return 0;
2536 	}
2537 
2538 	skb_dst_force(skb);
2539 
2540 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2541 	if (IS_ERR(dst)) {
2542 		res = 0;
2543 		dst = NULL;
2544 	}
2545 	skb_dst_set(skb, dst);
2546 	return res;
2547 }
2548 EXPORT_SYMBOL(__xfrm_route_forward);
2549 
2550 /* Optimize later using cookies and generation ids. */
2551 
2552 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2553 {
2554 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2555 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2556 	 * get validated by dst_ops->check on every use.  We do this
2557 	 * because when a normal route referenced by an XFRM dst is
2558 	 * obsoleted we do not go looking around for all parent
2559 	 * referencing XFRM dsts so that we can invalidate them.  It
2560 	 * is just too much work.  Instead we make the checks here on
2561 	 * every use.  For example:
2562 	 *
2563 	 *	XFRM dst A --> IPv4 dst X
2564 	 *
2565 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
2566 	 * in this example).  If X is marked obsolete, "A" will not
2567 	 * notice.  That's what we are validating here via the
2568 	 * stale_bundle() check.
2569 	 *
2570 	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2571 	 * be marked on it.
2572 	 * This will force stale_bundle() to fail on any xdst bundle with
2573 	 * this dst linked in it.
2574 	 */
2575 	if (dst->obsolete < 0 && !stale_bundle(dst))
2576 		return dst;
2577 
2578 	return NULL;
2579 }
2580 
2581 static int stale_bundle(struct dst_entry *dst)
2582 {
2583 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2584 }
2585 
2586 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2587 {
2588 	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
2589 		dst->dev = dev_net(dev)->loopback_dev;
2590 		dev_hold(dst->dev);
2591 		dev_put(dev);
2592 	}
2593 }
2594 EXPORT_SYMBOL(xfrm_dst_ifdown);
2595 
2596 static void xfrm_link_failure(struct sk_buff *skb)
2597 {
2598 	/* Impossible. Such dst must be popped before reaches point of failure. */
2599 }
2600 
2601 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2602 {
2603 	if (dst) {
2604 		if (dst->obsolete) {
2605 			dst_release(dst);
2606 			dst = NULL;
2607 		}
2608 	}
2609 	return dst;
2610 }
2611 
2612 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
2613 {
2614 	while (nr--) {
2615 		struct xfrm_dst *xdst = bundle[nr];
2616 		u32 pmtu, route_mtu_cached;
2617 		struct dst_entry *dst;
2618 
2619 		dst = &xdst->u.dst;
2620 		pmtu = dst_mtu(xfrm_dst_child(dst));
2621 		xdst->child_mtu_cached = pmtu;
2622 
2623 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2624 
2625 		route_mtu_cached = dst_mtu(xdst->route);
2626 		xdst->route_mtu_cached = route_mtu_cached;
2627 
2628 		if (pmtu > route_mtu_cached)
2629 			pmtu = route_mtu_cached;
2630 
2631 		dst_metric_set(dst, RTAX_MTU, pmtu);
2632 	}
2633 }
2634 
2635 /* Check that the bundle accepts the flow and its components are
2636  * still valid.
2637  */
2638 
2639 static int xfrm_bundle_ok(struct xfrm_dst *first)
2640 {
2641 	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2642 	struct dst_entry *dst = &first->u.dst;
2643 	struct xfrm_dst *xdst;
2644 	int start_from, nr;
2645 	u32 mtu;
2646 
2647 	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
2648 	    (dst->dev && !netif_running(dst->dev)))
2649 		return 0;
2650 
2651 	if (dst->flags & DST_XFRM_QUEUE)
2652 		return 1;
2653 
2654 	start_from = nr = 0;
2655 	do {
2656 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2657 
2658 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2659 			return 0;
2660 		if (xdst->xfrm_genid != dst->xfrm->genid)
2661 			return 0;
2662 		if (xdst->num_pols > 0 &&
2663 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2664 			return 0;
2665 
2666 		bundle[nr++] = xdst;
2667 
2668 		mtu = dst_mtu(xfrm_dst_child(dst));
2669 		if (xdst->child_mtu_cached != mtu) {
2670 			start_from = nr;
2671 			xdst->child_mtu_cached = mtu;
2672 		}
2673 
2674 		if (!dst_check(xdst->route, xdst->route_cookie))
2675 			return 0;
2676 		mtu = dst_mtu(xdst->route);
2677 		if (xdst->route_mtu_cached != mtu) {
2678 			start_from = nr;
2679 			xdst->route_mtu_cached = mtu;
2680 		}
2681 
2682 		dst = xfrm_dst_child(dst);
2683 	} while (dst->xfrm);
2684 
2685 	if (likely(!start_from))
2686 		return 1;
2687 
2688 	xdst = bundle[start_from - 1];
2689 	mtu = xdst->child_mtu_cached;
2690 	while (start_from--) {
2691 		dst = &xdst->u.dst;
2692 
2693 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2694 		if (mtu > xdst->route_mtu_cached)
2695 			mtu = xdst->route_mtu_cached;
2696 		dst_metric_set(dst, RTAX_MTU, mtu);
2697 		if (!start_from)
2698 			break;
2699 
2700 		xdst = bundle[start_from - 1];
2701 		xdst->child_mtu_cached = mtu;
2702 	}
2703 
2704 	return 1;
2705 }
2706 
2707 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2708 {
2709 	return dst_metric_advmss(xfrm_dst_path(dst));
2710 }
2711 
2712 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2713 {
2714 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2715 
2716 	return mtu ? : dst_mtu(xfrm_dst_path(dst));
2717 }
2718 
2719 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2720 					const void *daddr)
2721 {
2722 	while (dst->xfrm) {
2723 		const struct xfrm_state *xfrm = dst->xfrm;
2724 
2725 		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2726 			continue;
2727 		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2728 			daddr = xfrm->coaddr;
2729 		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2730 			daddr = &xfrm->id.daddr;
2731 
2732 		dst = xfrm_dst_child(dst);
2733 	}
2734 	return daddr;
2735 }
2736 
2737 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2738 					   struct sk_buff *skb,
2739 					   const void *daddr)
2740 {
2741 	const struct dst_entry *path = xfrm_dst_path(dst);
2742 
2743 	if (!skb)
2744 		daddr = xfrm_get_dst_nexthop(dst, daddr);
2745 	return path->ops->neigh_lookup(path, skb, daddr);
2746 }
2747 
2748 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2749 {
2750 	const struct dst_entry *path = xfrm_dst_path(dst);
2751 
2752 	daddr = xfrm_get_dst_nexthop(dst, daddr);
2753 	path->ops->confirm_neigh(path, daddr);
2754 }
2755 
2756 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2757 {
2758 	int err = 0;
2759 
2760 	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2761 		return -EAFNOSUPPORT;
2762 
2763 	spin_lock(&xfrm_policy_afinfo_lock);
2764 	if (unlikely(xfrm_policy_afinfo[family] != NULL))
2765 		err = -EEXIST;
2766 	else {
2767 		struct dst_ops *dst_ops = afinfo->dst_ops;
2768 		if (likely(dst_ops->kmem_cachep == NULL))
2769 			dst_ops->kmem_cachep = xfrm_dst_cache;
2770 		if (likely(dst_ops->check == NULL))
2771 			dst_ops->check = xfrm_dst_check;
2772 		if (likely(dst_ops->default_advmss == NULL))
2773 			dst_ops->default_advmss = xfrm_default_advmss;
2774 		if (likely(dst_ops->mtu == NULL))
2775 			dst_ops->mtu = xfrm_mtu;
2776 		if (likely(dst_ops->negative_advice == NULL))
2777 			dst_ops->negative_advice = xfrm_negative_advice;
2778 		if (likely(dst_ops->link_failure == NULL))
2779 			dst_ops->link_failure = xfrm_link_failure;
2780 		if (likely(dst_ops->neigh_lookup == NULL))
2781 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
2782 		if (likely(!dst_ops->confirm_neigh))
2783 			dst_ops->confirm_neigh = xfrm_confirm_neigh;
2784 		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2785 	}
2786 	spin_unlock(&xfrm_policy_afinfo_lock);
2787 
2788 	return err;
2789 }
2790 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2791 
2792 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2793 {
2794 	struct dst_ops *dst_ops = afinfo->dst_ops;
2795 	int i;
2796 
2797 	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2798 		if (xfrm_policy_afinfo[i] != afinfo)
2799 			continue;
2800 		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2801 		break;
2802 	}
2803 
2804 	synchronize_rcu();
2805 
2806 	dst_ops->kmem_cachep = NULL;
2807 	dst_ops->check = NULL;
2808 	dst_ops->negative_advice = NULL;
2809 	dst_ops->link_failure = NULL;
2810 }
2811 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2812 
2813 #ifdef CONFIG_XFRM_STATISTICS
2814 static int __net_init xfrm_statistics_init(struct net *net)
2815 {
2816 	int rv;
2817 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2818 	if (!net->mib.xfrm_statistics)
2819 		return -ENOMEM;
2820 	rv = xfrm_proc_init(net);
2821 	if (rv < 0)
2822 		free_percpu(net->mib.xfrm_statistics);
2823 	return rv;
2824 }
2825 
2826 static void xfrm_statistics_fini(struct net *net)
2827 {
2828 	xfrm_proc_fini(net);
2829 	free_percpu(net->mib.xfrm_statistics);
2830 }
2831 #else
2832 static int __net_init xfrm_statistics_init(struct net *net)
2833 {
2834 	return 0;
2835 }
2836 
2837 static void xfrm_statistics_fini(struct net *net)
2838 {
2839 }
2840 #endif
2841 
2842 static int __net_init xfrm_policy_init(struct net *net)
2843 {
2844 	unsigned int hmask, sz;
2845 	int dir;
2846 
2847 	if (net_eq(net, &init_net))
2848 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2849 					   sizeof(struct xfrm_dst),
2850 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2851 					   NULL);
2852 
2853 	hmask = 8 - 1;
2854 	sz = (hmask+1) * sizeof(struct hlist_head);
2855 
2856 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2857 	if (!net->xfrm.policy_byidx)
2858 		goto out_byidx;
2859 	net->xfrm.policy_idx_hmask = hmask;
2860 
2861 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2862 		struct xfrm_policy_hash *htab;
2863 
2864 		net->xfrm.policy_count[dir] = 0;
2865 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2866 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2867 
2868 		htab = &net->xfrm.policy_bydst[dir];
2869 		htab->table = xfrm_hash_alloc(sz);
2870 		if (!htab->table)
2871 			goto out_bydst;
2872 		htab->hmask = hmask;
2873 		htab->dbits4 = 32;
2874 		htab->sbits4 = 32;
2875 		htab->dbits6 = 128;
2876 		htab->sbits6 = 128;
2877 	}
2878 	net->xfrm.policy_hthresh.lbits4 = 32;
2879 	net->xfrm.policy_hthresh.rbits4 = 32;
2880 	net->xfrm.policy_hthresh.lbits6 = 128;
2881 	net->xfrm.policy_hthresh.rbits6 = 128;
2882 
2883 	seqlock_init(&net->xfrm.policy_hthresh.lock);
2884 
2885 	INIT_LIST_HEAD(&net->xfrm.policy_all);
2886 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2887 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2888 	if (net_eq(net, &init_net))
2889 		xfrm_dev_init();
2890 	return 0;
2891 
2892 out_bydst:
2893 	for (dir--; dir >= 0; dir--) {
2894 		struct xfrm_policy_hash *htab;
2895 
2896 		htab = &net->xfrm.policy_bydst[dir];
2897 		xfrm_hash_free(htab->table, sz);
2898 	}
2899 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2900 out_byidx:
2901 	return -ENOMEM;
2902 }
2903 
2904 static void xfrm_policy_fini(struct net *net)
2905 {
2906 	unsigned int sz;
2907 	int dir;
2908 
2909 	flush_work(&net->xfrm.policy_hash_work);
2910 #ifdef CONFIG_XFRM_SUB_POLICY
2911 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2912 #endif
2913 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2914 
2915 	WARN_ON(!list_empty(&net->xfrm.policy_all));
2916 
2917 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2918 		struct xfrm_policy_hash *htab;
2919 
2920 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2921 
2922 		htab = &net->xfrm.policy_bydst[dir];
2923 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2924 		WARN_ON(!hlist_empty(htab->table));
2925 		xfrm_hash_free(htab->table, sz);
2926 	}
2927 
2928 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2929 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2930 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2931 }
2932 
2933 static int __net_init xfrm_net_init(struct net *net)
2934 {
2935 	int rv;
2936 
2937 	/* Initialize the per-net locks here */
2938 	spin_lock_init(&net->xfrm.xfrm_state_lock);
2939 	spin_lock_init(&net->xfrm.xfrm_policy_lock);
2940 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
2941 
2942 	rv = xfrm_statistics_init(net);
2943 	if (rv < 0)
2944 		goto out_statistics;
2945 	rv = xfrm_state_init(net);
2946 	if (rv < 0)
2947 		goto out_state;
2948 	rv = xfrm_policy_init(net);
2949 	if (rv < 0)
2950 		goto out_policy;
2951 	rv = xfrm_sysctl_init(net);
2952 	if (rv < 0)
2953 		goto out_sysctl;
2954 
2955 	return 0;
2956 
2957 out_sysctl:
2958 	xfrm_policy_fini(net);
2959 out_policy:
2960 	xfrm_state_fini(net);
2961 out_state:
2962 	xfrm_statistics_fini(net);
2963 out_statistics:
2964 	return rv;
2965 }
2966 
2967 static void __net_exit xfrm_net_exit(struct net *net)
2968 {
2969 	xfrm_sysctl_fini(net);
2970 	xfrm_policy_fini(net);
2971 	xfrm_state_fini(net);
2972 	xfrm_statistics_fini(net);
2973 }
2974 
2975 static struct pernet_operations __net_initdata xfrm_net_ops = {
2976 	.init = xfrm_net_init,
2977 	.exit = xfrm_net_exit,
2978 };
2979 
2980 void __init xfrm_init(void)
2981 {
2982 	int i;
2983 
2984 	xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2985 				       GFP_KERNEL);
2986 	BUG_ON(!xfrm_pcpu_work);
2987 
2988 	for (i = 0; i < NR_CPUS; i++)
2989 		INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
2990 
2991 	register_pernet_subsys(&xfrm_net_ops);
2992 	seqcount_init(&xfrm_policy_hash_generation);
2993 	xfrm_input_init();
2994 }
2995 
2996 #ifdef CONFIG_AUDITSYSCALL
2997 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2998 					 struct audit_buffer *audit_buf)
2999 {
3000 	struct xfrm_sec_ctx *ctx = xp->security;
3001 	struct xfrm_selector *sel = &xp->selector;
3002 
3003 	if (ctx)
3004 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3005 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3006 
3007 	switch (sel->family) {
3008 	case AF_INET:
3009 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3010 		if (sel->prefixlen_s != 32)
3011 			audit_log_format(audit_buf, " src_prefixlen=%d",
3012 					 sel->prefixlen_s);
3013 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3014 		if (sel->prefixlen_d != 32)
3015 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3016 					 sel->prefixlen_d);
3017 		break;
3018 	case AF_INET6:
3019 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3020 		if (sel->prefixlen_s != 128)
3021 			audit_log_format(audit_buf, " src_prefixlen=%d",
3022 					 sel->prefixlen_s);
3023 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3024 		if (sel->prefixlen_d != 128)
3025 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3026 					 sel->prefixlen_d);
3027 		break;
3028 	}
3029 }
3030 
3031 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3032 {
3033 	struct audit_buffer *audit_buf;
3034 
3035 	audit_buf = xfrm_audit_start("SPD-add");
3036 	if (audit_buf == NULL)
3037 		return;
3038 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3039 	audit_log_format(audit_buf, " res=%u", result);
3040 	xfrm_audit_common_policyinfo(xp, audit_buf);
3041 	audit_log_end(audit_buf);
3042 }
3043 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3044 
3045 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3046 			      bool task_valid)
3047 {
3048 	struct audit_buffer *audit_buf;
3049 
3050 	audit_buf = xfrm_audit_start("SPD-delete");
3051 	if (audit_buf == NULL)
3052 		return;
3053 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3054 	audit_log_format(audit_buf, " res=%u", result);
3055 	xfrm_audit_common_policyinfo(xp, audit_buf);
3056 	audit_log_end(audit_buf);
3057 }
3058 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3059 #endif
3060 
3061 #ifdef CONFIG_XFRM_MIGRATE
3062 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3063 					const struct xfrm_selector *sel_tgt)
3064 {
3065 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3066 		if (sel_tgt->family == sel_cmp->family &&
3067 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3068 				    sel_cmp->family) &&
3069 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3070 				    sel_cmp->family) &&
3071 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3072 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3073 			return true;
3074 		}
3075 	} else {
3076 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3077 			return true;
3078 		}
3079 	}
3080 	return false;
3081 }
3082 
3083 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3084 						    u8 dir, u8 type, struct net *net)
3085 {
3086 	struct xfrm_policy *pol, *ret = NULL;
3087 	struct hlist_head *chain;
3088 	u32 priority = ~0U;
3089 
3090 	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3091 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3092 	hlist_for_each_entry(pol, chain, bydst) {
3093 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3094 		    pol->type == type) {
3095 			ret = pol;
3096 			priority = ret->priority;
3097 			break;
3098 		}
3099 	}
3100 	chain = &net->xfrm.policy_inexact[dir];
3101 	hlist_for_each_entry(pol, chain, bydst) {
3102 		if ((pol->priority >= priority) && ret)
3103 			break;
3104 
3105 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3106 		    pol->type == type) {
3107 			ret = pol;
3108 			break;
3109 		}
3110 	}
3111 
3112 	xfrm_pol_hold(ret);
3113 
3114 	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3115 
3116 	return ret;
3117 }
3118 
3119 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3120 {
3121 	int match = 0;
3122 
3123 	if (t->mode == m->mode && t->id.proto == m->proto &&
3124 	    (m->reqid == 0 || t->reqid == m->reqid)) {
3125 		switch (t->mode) {
3126 		case XFRM_MODE_TUNNEL:
3127 		case XFRM_MODE_BEET:
3128 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3129 					    m->old_family) &&
3130 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
3131 					    m->old_family)) {
3132 				match = 1;
3133 			}
3134 			break;
3135 		case XFRM_MODE_TRANSPORT:
3136 			/* in case of transport mode, template does not store
3137 			   any IP addresses, hence we just compare mode and
3138 			   protocol */
3139 			match = 1;
3140 			break;
3141 		default:
3142 			break;
3143 		}
3144 	}
3145 	return match;
3146 }
3147 
3148 /* update endpoint address(es) of template(s) */
3149 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3150 			       struct xfrm_migrate *m, int num_migrate)
3151 {
3152 	struct xfrm_migrate *mp;
3153 	int i, j, n = 0;
3154 
3155 	write_lock_bh(&pol->lock);
3156 	if (unlikely(pol->walk.dead)) {
3157 		/* target policy has been deleted */
3158 		write_unlock_bh(&pol->lock);
3159 		return -ENOENT;
3160 	}
3161 
3162 	for (i = 0; i < pol->xfrm_nr; i++) {
3163 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3164 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3165 				continue;
3166 			n++;
3167 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3168 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3169 				continue;
3170 			/* update endpoints */
3171 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3172 			       sizeof(pol->xfrm_vec[i].id.daddr));
3173 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3174 			       sizeof(pol->xfrm_vec[i].saddr));
3175 			pol->xfrm_vec[i].encap_family = mp->new_family;
3176 			/* flush bundles */
3177 			atomic_inc(&pol->genid);
3178 		}
3179 	}
3180 
3181 	write_unlock_bh(&pol->lock);
3182 
3183 	if (!n)
3184 		return -ENODATA;
3185 
3186 	return 0;
3187 }
3188 
3189 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3190 {
3191 	int i, j;
3192 
3193 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3194 		return -EINVAL;
3195 
3196 	for (i = 0; i < num_migrate; i++) {
3197 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3198 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3199 			return -EINVAL;
3200 
3201 		/* check if there is any duplicated entry */
3202 		for (j = i + 1; j < num_migrate; j++) {
3203 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3204 				    sizeof(m[i].old_daddr)) &&
3205 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3206 				    sizeof(m[i].old_saddr)) &&
3207 			    m[i].proto == m[j].proto &&
3208 			    m[i].mode == m[j].mode &&
3209 			    m[i].reqid == m[j].reqid &&
3210 			    m[i].old_family == m[j].old_family)
3211 				return -EINVAL;
3212 		}
3213 	}
3214 
3215 	return 0;
3216 }
3217 
3218 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3219 		 struct xfrm_migrate *m, int num_migrate,
3220 		 struct xfrm_kmaddress *k, struct net *net,
3221 		 struct xfrm_encap_tmpl *encap)
3222 {
3223 	int i, err, nx_cur = 0, nx_new = 0;
3224 	struct xfrm_policy *pol = NULL;
3225 	struct xfrm_state *x, *xc;
3226 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3227 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3228 	struct xfrm_migrate *mp;
3229 
3230 	/* Stage 0 - sanity checks */
3231 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3232 		goto out;
3233 
3234 	if (dir >= XFRM_POLICY_MAX) {
3235 		err = -EINVAL;
3236 		goto out;
3237 	}
3238 
3239 	/* Stage 1 - find policy */
3240 	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3241 		err = -ENOENT;
3242 		goto out;
3243 	}
3244 
3245 	/* Stage 2 - find and update state(s) */
3246 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3247 		if ((x = xfrm_migrate_state_find(mp, net))) {
3248 			x_cur[nx_cur] = x;
3249 			nx_cur++;
3250 			xc = xfrm_state_migrate(x, mp, encap);
3251 			if (xc) {
3252 				x_new[nx_new] = xc;
3253 				nx_new++;
3254 			} else {
3255 				err = -ENODATA;
3256 				goto restore_state;
3257 			}
3258 		}
3259 	}
3260 
3261 	/* Stage 3 - update policy */
3262 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3263 		goto restore_state;
3264 
3265 	/* Stage 4 - delete old state(s) */
3266 	if (nx_cur) {
3267 		xfrm_states_put(x_cur, nx_cur);
3268 		xfrm_states_delete(x_cur, nx_cur);
3269 	}
3270 
3271 	/* Stage 5 - announce */
3272 	km_migrate(sel, dir, type, m, num_migrate, k, encap);
3273 
3274 	xfrm_pol_put(pol);
3275 
3276 	return 0;
3277 out:
3278 	return err;
3279 
3280 restore_state:
3281 	if (pol)
3282 		xfrm_pol_put(pol);
3283 	if (nx_cur)
3284 		xfrm_states_put(x_cur, nx_cur);
3285 	if (nx_new)
3286 		xfrm_states_delete(x_new, nx_new);
3287 
3288 	return err;
3289 }
3290 EXPORT_SYMBOL(xfrm_migrate);
3291 #endif
3292