xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 206e8c00752fbe9cc463184236ac64b2a532cda5)
1 /*
2  * xfrm_policy.c
3  *
4  * Changes:
5  *	Mitsuru KANDA @USAGI
6  * 	Kazunori MIYAZAWA @USAGI
7  * 	Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8  * 		IPv6 support
9  * 	Kazunori MIYAZAWA @USAGI
10  * 	YOSHIFUJI Hideaki
11  * 		Split up af-specific portion
12  *	Derek Atkins <derek@ihtfp.com>		Add the post_input processor
13  *
14  */
15 
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
28 #include <net/dst.h>
29 #include <net/flow.h>
30 #include <net/xfrm.h>
31 #include <net/ip.h>
32 #ifdef CONFIG_XFRM_STATISTICS
33 #include <net/snmp.h>
34 #endif
35 
36 #include "xfrm_hash.h"
37 
38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
40 #define XFRM_MAX_QUEUE_LEN	100
41 
42 struct xfrm_flo {
43 	struct dst_entry *dst_orig;
44 	u8 flags;
45 };
46 
47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
48 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
49 						__read_mostly;
50 
51 static struct kmem_cache *xfrm_dst_cache __read_mostly;
52 
53 static void xfrm_init_pmtu(struct dst_entry *dst);
54 static int stale_bundle(struct dst_entry *dst);
55 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
56 static void xfrm_policy_queue_process(unsigned long arg);
57 
58 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
59 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
60 						int dir);
61 
62 static inline bool
63 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
64 {
65 	const struct flowi4 *fl4 = &fl->u.ip4;
66 
67 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
68 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
69 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
70 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
71 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
72 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
73 }
74 
75 static inline bool
76 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
77 {
78 	const struct flowi6 *fl6 = &fl->u.ip6;
79 
80 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
81 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
82 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
83 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
84 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
85 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
86 }
87 
88 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
89 			 unsigned short family)
90 {
91 	switch (family) {
92 	case AF_INET:
93 		return __xfrm4_selector_match(sel, fl);
94 	case AF_INET6:
95 		return __xfrm6_selector_match(sel, fl);
96 	}
97 	return false;
98 }
99 
100 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
101 {
102 	struct xfrm_policy_afinfo *afinfo;
103 
104 	if (unlikely(family >= NPROTO))
105 		return NULL;
106 	rcu_read_lock();
107 	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
108 	if (unlikely(!afinfo))
109 		rcu_read_unlock();
110 	return afinfo;
111 }
112 
113 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
114 {
115 	rcu_read_unlock();
116 }
117 
118 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net,
119 						  int tos, int oif,
120 						  const xfrm_address_t *saddr,
121 						  const xfrm_address_t *daddr,
122 						  int family)
123 {
124 	struct xfrm_policy_afinfo *afinfo;
125 	struct dst_entry *dst;
126 
127 	afinfo = xfrm_policy_get_afinfo(family);
128 	if (unlikely(afinfo == NULL))
129 		return ERR_PTR(-EAFNOSUPPORT);
130 
131 	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr);
132 
133 	xfrm_policy_put_afinfo(afinfo);
134 
135 	return dst;
136 }
137 
138 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
139 						int tos, int oif,
140 						xfrm_address_t *prev_saddr,
141 						xfrm_address_t *prev_daddr,
142 						int family)
143 {
144 	struct net *net = xs_net(x);
145 	xfrm_address_t *saddr = &x->props.saddr;
146 	xfrm_address_t *daddr = &x->id.daddr;
147 	struct dst_entry *dst;
148 
149 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
150 		saddr = x->coaddr;
151 		daddr = prev_daddr;
152 	}
153 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
154 		saddr = prev_saddr;
155 		daddr = x->coaddr;
156 	}
157 
158 	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family);
159 
160 	if (!IS_ERR(dst)) {
161 		if (prev_saddr != saddr)
162 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
163 		if (prev_daddr != daddr)
164 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
165 	}
166 
167 	return dst;
168 }
169 
170 static inline unsigned long make_jiffies(long secs)
171 {
172 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
173 		return MAX_SCHEDULE_TIMEOUT-1;
174 	else
175 		return secs*HZ;
176 }
177 
178 static void xfrm_policy_timer(unsigned long data)
179 {
180 	struct xfrm_policy *xp = (struct xfrm_policy *)data;
181 	unsigned long now = get_seconds();
182 	long next = LONG_MAX;
183 	int warn = 0;
184 	int dir;
185 
186 	read_lock(&xp->lock);
187 
188 	if (unlikely(xp->walk.dead))
189 		goto out;
190 
191 	dir = xfrm_policy_id2dir(xp->index);
192 
193 	if (xp->lft.hard_add_expires_seconds) {
194 		long tmo = xp->lft.hard_add_expires_seconds +
195 			xp->curlft.add_time - now;
196 		if (tmo <= 0)
197 			goto expired;
198 		if (tmo < next)
199 			next = tmo;
200 	}
201 	if (xp->lft.hard_use_expires_seconds) {
202 		long tmo = xp->lft.hard_use_expires_seconds +
203 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
204 		if (tmo <= 0)
205 			goto expired;
206 		if (tmo < next)
207 			next = tmo;
208 	}
209 	if (xp->lft.soft_add_expires_seconds) {
210 		long tmo = xp->lft.soft_add_expires_seconds +
211 			xp->curlft.add_time - now;
212 		if (tmo <= 0) {
213 			warn = 1;
214 			tmo = XFRM_KM_TIMEOUT;
215 		}
216 		if (tmo < next)
217 			next = tmo;
218 	}
219 	if (xp->lft.soft_use_expires_seconds) {
220 		long tmo = xp->lft.soft_use_expires_seconds +
221 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
222 		if (tmo <= 0) {
223 			warn = 1;
224 			tmo = XFRM_KM_TIMEOUT;
225 		}
226 		if (tmo < next)
227 			next = tmo;
228 	}
229 
230 	if (warn)
231 		km_policy_expired(xp, dir, 0, 0);
232 	if (next != LONG_MAX &&
233 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
234 		xfrm_pol_hold(xp);
235 
236 out:
237 	read_unlock(&xp->lock);
238 	xfrm_pol_put(xp);
239 	return;
240 
241 expired:
242 	read_unlock(&xp->lock);
243 	if (!xfrm_policy_delete(xp, dir))
244 		km_policy_expired(xp, dir, 1, 0);
245 	xfrm_pol_put(xp);
246 }
247 
248 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
249 {
250 	struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
251 
252 	if (unlikely(pol->walk.dead))
253 		flo = NULL;
254 	else
255 		xfrm_pol_hold(pol);
256 
257 	return flo;
258 }
259 
260 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
261 {
262 	struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
263 
264 	return !pol->walk.dead;
265 }
266 
267 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
268 {
269 	xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
270 }
271 
272 static const struct flow_cache_ops xfrm_policy_fc_ops = {
273 	.get = xfrm_policy_flo_get,
274 	.check = xfrm_policy_flo_check,
275 	.delete = xfrm_policy_flo_delete,
276 };
277 
278 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
279  * SPD calls.
280  */
281 
282 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
283 {
284 	struct xfrm_policy *policy;
285 
286 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
287 
288 	if (policy) {
289 		write_pnet(&policy->xp_net, net);
290 		INIT_LIST_HEAD(&policy->walk.all);
291 		INIT_HLIST_NODE(&policy->bydst);
292 		INIT_HLIST_NODE(&policy->byidx);
293 		rwlock_init(&policy->lock);
294 		atomic_set(&policy->refcnt, 1);
295 		skb_queue_head_init(&policy->polq.hold_queue);
296 		setup_timer(&policy->timer, xfrm_policy_timer,
297 				(unsigned long)policy);
298 		setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
299 			    (unsigned long)policy);
300 		policy->flo.ops = &xfrm_policy_fc_ops;
301 	}
302 	return policy;
303 }
304 EXPORT_SYMBOL(xfrm_policy_alloc);
305 
306 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
307 
308 void xfrm_policy_destroy(struct xfrm_policy *policy)
309 {
310 	BUG_ON(!policy->walk.dead);
311 
312 	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
313 		BUG();
314 
315 	security_xfrm_policy_free(policy->security);
316 	kfree(policy);
317 }
318 EXPORT_SYMBOL(xfrm_policy_destroy);
319 
320 /* Rule must be locked. Release descentant resources, announce
321  * entry dead. The rule must be unlinked from lists to the moment.
322  */
323 
324 static void xfrm_policy_kill(struct xfrm_policy *policy)
325 {
326 	policy->walk.dead = 1;
327 
328 	atomic_inc(&policy->genid);
329 
330 	if (del_timer(&policy->polq.hold_timer))
331 		xfrm_pol_put(policy);
332 	skb_queue_purge(&policy->polq.hold_queue);
333 
334 	if (del_timer(&policy->timer))
335 		xfrm_pol_put(policy);
336 
337 	xfrm_pol_put(policy);
338 }
339 
340 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
341 
342 static inline unsigned int idx_hash(struct net *net, u32 index)
343 {
344 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
345 }
346 
347 /* calculate policy hash thresholds */
348 static void __get_hash_thresh(struct net *net,
349 			      unsigned short family, int dir,
350 			      u8 *dbits, u8 *sbits)
351 {
352 	switch (family) {
353 	case AF_INET:
354 		*dbits = net->xfrm.policy_bydst[dir].dbits4;
355 		*sbits = net->xfrm.policy_bydst[dir].sbits4;
356 		break;
357 
358 	case AF_INET6:
359 		*dbits = net->xfrm.policy_bydst[dir].dbits6;
360 		*sbits = net->xfrm.policy_bydst[dir].sbits6;
361 		break;
362 
363 	default:
364 		*dbits = 0;
365 		*sbits = 0;
366 	}
367 }
368 
369 static struct hlist_head *policy_hash_bysel(struct net *net,
370 					    const struct xfrm_selector *sel,
371 					    unsigned short family, int dir)
372 {
373 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
374 	unsigned int hash;
375 	u8 dbits;
376 	u8 sbits;
377 
378 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
379 	hash = __sel_hash(sel, family, hmask, dbits, sbits);
380 
381 	return (hash == hmask + 1 ?
382 		&net->xfrm.policy_inexact[dir] :
383 		net->xfrm.policy_bydst[dir].table + hash);
384 }
385 
386 static struct hlist_head *policy_hash_direct(struct net *net,
387 					     const xfrm_address_t *daddr,
388 					     const xfrm_address_t *saddr,
389 					     unsigned short family, int dir)
390 {
391 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
392 	unsigned int hash;
393 	u8 dbits;
394 	u8 sbits;
395 
396 	__get_hash_thresh(net, family, dir, &dbits, &sbits);
397 	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
398 
399 	return net->xfrm.policy_bydst[dir].table + hash;
400 }
401 
402 static void xfrm_dst_hash_transfer(struct net *net,
403 				   struct hlist_head *list,
404 				   struct hlist_head *ndsttable,
405 				   unsigned int nhashmask,
406 				   int dir)
407 {
408 	struct hlist_node *tmp, *entry0 = NULL;
409 	struct xfrm_policy *pol;
410 	unsigned int h0 = 0;
411 	u8 dbits;
412 	u8 sbits;
413 
414 redo:
415 	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
416 		unsigned int h;
417 
418 		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
419 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
420 				pol->family, nhashmask, dbits, sbits);
421 		if (!entry0) {
422 			hlist_del(&pol->bydst);
423 			hlist_add_head(&pol->bydst, ndsttable+h);
424 			h0 = h;
425 		} else {
426 			if (h != h0)
427 				continue;
428 			hlist_del(&pol->bydst);
429 			hlist_add_behind(&pol->bydst, entry0);
430 		}
431 		entry0 = &pol->bydst;
432 	}
433 	if (!hlist_empty(list)) {
434 		entry0 = NULL;
435 		goto redo;
436 	}
437 }
438 
439 static void xfrm_idx_hash_transfer(struct hlist_head *list,
440 				   struct hlist_head *nidxtable,
441 				   unsigned int nhashmask)
442 {
443 	struct hlist_node *tmp;
444 	struct xfrm_policy *pol;
445 
446 	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
447 		unsigned int h;
448 
449 		h = __idx_hash(pol->index, nhashmask);
450 		hlist_add_head(&pol->byidx, nidxtable+h);
451 	}
452 }
453 
454 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
455 {
456 	return ((old_hmask + 1) << 1) - 1;
457 }
458 
459 static void xfrm_bydst_resize(struct net *net, int dir)
460 {
461 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
462 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
463 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
464 	struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
465 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
466 	int i;
467 
468 	if (!ndst)
469 		return;
470 
471 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
472 
473 	for (i = hmask; i >= 0; i--)
474 		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
475 
476 	net->xfrm.policy_bydst[dir].table = ndst;
477 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
478 
479 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
480 
481 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
482 }
483 
484 static void xfrm_byidx_resize(struct net *net, int total)
485 {
486 	unsigned int hmask = net->xfrm.policy_idx_hmask;
487 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
488 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
489 	struct hlist_head *oidx = net->xfrm.policy_byidx;
490 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
491 	int i;
492 
493 	if (!nidx)
494 		return;
495 
496 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
497 
498 	for (i = hmask; i >= 0; i--)
499 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
500 
501 	net->xfrm.policy_byidx = nidx;
502 	net->xfrm.policy_idx_hmask = nhashmask;
503 
504 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
505 
506 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
507 }
508 
509 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
510 {
511 	unsigned int cnt = net->xfrm.policy_count[dir];
512 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
513 
514 	if (total)
515 		*total += cnt;
516 
517 	if ((hmask + 1) < xfrm_policy_hashmax &&
518 	    cnt > hmask)
519 		return 1;
520 
521 	return 0;
522 }
523 
524 static inline int xfrm_byidx_should_resize(struct net *net, int total)
525 {
526 	unsigned int hmask = net->xfrm.policy_idx_hmask;
527 
528 	if ((hmask + 1) < xfrm_policy_hashmax &&
529 	    total > hmask)
530 		return 1;
531 
532 	return 0;
533 }
534 
535 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
536 {
537 	read_lock_bh(&net->xfrm.xfrm_policy_lock);
538 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
539 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
540 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
541 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
542 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
543 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
544 	si->spdhcnt = net->xfrm.policy_idx_hmask;
545 	si->spdhmcnt = xfrm_policy_hashmax;
546 	read_unlock_bh(&net->xfrm.xfrm_policy_lock);
547 }
548 EXPORT_SYMBOL(xfrm_spd_getinfo);
549 
550 static DEFINE_MUTEX(hash_resize_mutex);
551 static void xfrm_hash_resize(struct work_struct *work)
552 {
553 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
554 	int dir, total;
555 
556 	mutex_lock(&hash_resize_mutex);
557 
558 	total = 0;
559 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
560 		if (xfrm_bydst_should_resize(net, dir, &total))
561 			xfrm_bydst_resize(net, dir);
562 	}
563 	if (xfrm_byidx_should_resize(net, total))
564 		xfrm_byidx_resize(net, total);
565 
566 	mutex_unlock(&hash_resize_mutex);
567 }
568 
569 static void xfrm_hash_rebuild(struct work_struct *work)
570 {
571 	struct net *net = container_of(work, struct net,
572 				       xfrm.policy_hthresh.work);
573 	unsigned int hmask;
574 	struct xfrm_policy *pol;
575 	struct xfrm_policy *policy;
576 	struct hlist_head *chain;
577 	struct hlist_head *odst;
578 	struct hlist_node *newpos;
579 	int i;
580 	int dir;
581 	unsigned seq;
582 	u8 lbits4, rbits4, lbits6, rbits6;
583 
584 	mutex_lock(&hash_resize_mutex);
585 
586 	/* read selector prefixlen thresholds */
587 	do {
588 		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
589 
590 		lbits4 = net->xfrm.policy_hthresh.lbits4;
591 		rbits4 = net->xfrm.policy_hthresh.rbits4;
592 		lbits6 = net->xfrm.policy_hthresh.lbits6;
593 		rbits6 = net->xfrm.policy_hthresh.rbits6;
594 	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
595 
596 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
597 
598 	/* reset the bydst and inexact table in all directions */
599 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
600 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
601 		hmask = net->xfrm.policy_bydst[dir].hmask;
602 		odst = net->xfrm.policy_bydst[dir].table;
603 		for (i = hmask; i >= 0; i--)
604 			INIT_HLIST_HEAD(odst + i);
605 		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
606 			/* dir out => dst = remote, src = local */
607 			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
608 			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
609 			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
610 			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
611 		} else {
612 			/* dir in/fwd => dst = local, src = remote */
613 			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
614 			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
615 			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
616 			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
617 		}
618 	}
619 
620 	/* re-insert all policies by order of creation */
621 	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
622 		newpos = NULL;
623 		chain = policy_hash_bysel(net, &policy->selector,
624 					  policy->family,
625 					  xfrm_policy_id2dir(policy->index));
626 		hlist_for_each_entry(pol, chain, bydst) {
627 			if (policy->priority >= pol->priority)
628 				newpos = &pol->bydst;
629 			else
630 				break;
631 		}
632 		if (newpos)
633 			hlist_add_behind(&policy->bydst, newpos);
634 		else
635 			hlist_add_head(&policy->bydst, chain);
636 	}
637 
638 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
639 
640 	mutex_unlock(&hash_resize_mutex);
641 }
642 
643 void xfrm_policy_hash_rebuild(struct net *net)
644 {
645 	schedule_work(&net->xfrm.policy_hthresh.work);
646 }
647 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
648 
649 /* Generate new index... KAME seems to generate them ordered by cost
650  * of an absolute inpredictability of ordering of rules. This will not pass. */
651 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
652 {
653 	static u32 idx_generator;
654 
655 	for (;;) {
656 		struct hlist_head *list;
657 		struct xfrm_policy *p;
658 		u32 idx;
659 		int found;
660 
661 		if (!index) {
662 			idx = (idx_generator | dir);
663 			idx_generator += 8;
664 		} else {
665 			idx = index;
666 			index = 0;
667 		}
668 
669 		if (idx == 0)
670 			idx = 8;
671 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
672 		found = 0;
673 		hlist_for_each_entry(p, list, byidx) {
674 			if (p->index == idx) {
675 				found = 1;
676 				break;
677 			}
678 		}
679 		if (!found)
680 			return idx;
681 	}
682 }
683 
684 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
685 {
686 	u32 *p1 = (u32 *) s1;
687 	u32 *p2 = (u32 *) s2;
688 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
689 	int i;
690 
691 	for (i = 0; i < len; i++) {
692 		if (p1[i] != p2[i])
693 			return 1;
694 	}
695 
696 	return 0;
697 }
698 
699 static void xfrm_policy_requeue(struct xfrm_policy *old,
700 				struct xfrm_policy *new)
701 {
702 	struct xfrm_policy_queue *pq = &old->polq;
703 	struct sk_buff_head list;
704 
705 	if (skb_queue_empty(&pq->hold_queue))
706 		return;
707 
708 	__skb_queue_head_init(&list);
709 
710 	spin_lock_bh(&pq->hold_queue.lock);
711 	skb_queue_splice_init(&pq->hold_queue, &list);
712 	if (del_timer(&pq->hold_timer))
713 		xfrm_pol_put(old);
714 	spin_unlock_bh(&pq->hold_queue.lock);
715 
716 	pq = &new->polq;
717 
718 	spin_lock_bh(&pq->hold_queue.lock);
719 	skb_queue_splice(&list, &pq->hold_queue);
720 	pq->timeout = XFRM_QUEUE_TMO_MIN;
721 	if (!mod_timer(&pq->hold_timer, jiffies))
722 		xfrm_pol_hold(new);
723 	spin_unlock_bh(&pq->hold_queue.lock);
724 }
725 
726 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
727 				   struct xfrm_policy *pol)
728 {
729 	u32 mark = policy->mark.v & policy->mark.m;
730 
731 	if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
732 		return true;
733 
734 	if ((mark & pol->mark.m) == pol->mark.v &&
735 	    policy->priority == pol->priority)
736 		return true;
737 
738 	return false;
739 }
740 
741 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
742 {
743 	struct net *net = xp_net(policy);
744 	struct xfrm_policy *pol;
745 	struct xfrm_policy *delpol;
746 	struct hlist_head *chain;
747 	struct hlist_node *newpos;
748 
749 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
750 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
751 	delpol = NULL;
752 	newpos = NULL;
753 	hlist_for_each_entry(pol, chain, bydst) {
754 		if (pol->type == policy->type &&
755 		    !selector_cmp(&pol->selector, &policy->selector) &&
756 		    xfrm_policy_mark_match(policy, pol) &&
757 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
758 		    !WARN_ON(delpol)) {
759 			if (excl) {
760 				write_unlock_bh(&net->xfrm.xfrm_policy_lock);
761 				return -EEXIST;
762 			}
763 			delpol = pol;
764 			if (policy->priority > pol->priority)
765 				continue;
766 		} else if (policy->priority >= pol->priority) {
767 			newpos = &pol->bydst;
768 			continue;
769 		}
770 		if (delpol)
771 			break;
772 	}
773 	if (newpos)
774 		hlist_add_behind(&policy->bydst, newpos);
775 	else
776 		hlist_add_head(&policy->bydst, chain);
777 	__xfrm_policy_link(policy, dir);
778 	atomic_inc(&net->xfrm.flow_cache_genid);
779 
780 	/* After previous checking, family can either be AF_INET or AF_INET6 */
781 	if (policy->family == AF_INET)
782 		rt_genid_bump_ipv4(net);
783 	else
784 		rt_genid_bump_ipv6(net);
785 
786 	if (delpol) {
787 		xfrm_policy_requeue(delpol, policy);
788 		__xfrm_policy_unlink(delpol, dir);
789 	}
790 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
791 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
792 	policy->curlft.add_time = get_seconds();
793 	policy->curlft.use_time = 0;
794 	if (!mod_timer(&policy->timer, jiffies + HZ))
795 		xfrm_pol_hold(policy);
796 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
797 
798 	if (delpol)
799 		xfrm_policy_kill(delpol);
800 	else if (xfrm_bydst_should_resize(net, dir, NULL))
801 		schedule_work(&net->xfrm.policy_hash_work);
802 
803 	return 0;
804 }
805 EXPORT_SYMBOL(xfrm_policy_insert);
806 
807 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
808 					  int dir, struct xfrm_selector *sel,
809 					  struct xfrm_sec_ctx *ctx, int delete,
810 					  int *err)
811 {
812 	struct xfrm_policy *pol, *ret;
813 	struct hlist_head *chain;
814 
815 	*err = 0;
816 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
817 	chain = policy_hash_bysel(net, sel, sel->family, dir);
818 	ret = NULL;
819 	hlist_for_each_entry(pol, chain, bydst) {
820 		if (pol->type == type &&
821 		    (mark & pol->mark.m) == pol->mark.v &&
822 		    !selector_cmp(sel, &pol->selector) &&
823 		    xfrm_sec_ctx_match(ctx, pol->security)) {
824 			xfrm_pol_hold(pol);
825 			if (delete) {
826 				*err = security_xfrm_policy_delete(
827 								pol->security);
828 				if (*err) {
829 					write_unlock_bh(&net->xfrm.xfrm_policy_lock);
830 					return pol;
831 				}
832 				__xfrm_policy_unlink(pol, dir);
833 			}
834 			ret = pol;
835 			break;
836 		}
837 	}
838 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
839 
840 	if (ret && delete)
841 		xfrm_policy_kill(ret);
842 	return ret;
843 }
844 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
845 
846 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
847 				     int dir, u32 id, int delete, int *err)
848 {
849 	struct xfrm_policy *pol, *ret;
850 	struct hlist_head *chain;
851 
852 	*err = -ENOENT;
853 	if (xfrm_policy_id2dir(id) != dir)
854 		return NULL;
855 
856 	*err = 0;
857 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
858 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
859 	ret = NULL;
860 	hlist_for_each_entry(pol, chain, byidx) {
861 		if (pol->type == type && pol->index == id &&
862 		    (mark & pol->mark.m) == pol->mark.v) {
863 			xfrm_pol_hold(pol);
864 			if (delete) {
865 				*err = security_xfrm_policy_delete(
866 								pol->security);
867 				if (*err) {
868 					write_unlock_bh(&net->xfrm.xfrm_policy_lock);
869 					return pol;
870 				}
871 				__xfrm_policy_unlink(pol, dir);
872 			}
873 			ret = pol;
874 			break;
875 		}
876 	}
877 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
878 
879 	if (ret && delete)
880 		xfrm_policy_kill(ret);
881 	return ret;
882 }
883 EXPORT_SYMBOL(xfrm_policy_byid);
884 
885 #ifdef CONFIG_SECURITY_NETWORK_XFRM
886 static inline int
887 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
888 {
889 	int dir, err = 0;
890 
891 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
892 		struct xfrm_policy *pol;
893 		int i;
894 
895 		hlist_for_each_entry(pol,
896 				     &net->xfrm.policy_inexact[dir], bydst) {
897 			if (pol->type != type)
898 				continue;
899 			err = security_xfrm_policy_delete(pol->security);
900 			if (err) {
901 				xfrm_audit_policy_delete(pol, 0, task_valid);
902 				return err;
903 			}
904 		}
905 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
906 			hlist_for_each_entry(pol,
907 					     net->xfrm.policy_bydst[dir].table + i,
908 					     bydst) {
909 				if (pol->type != type)
910 					continue;
911 				err = security_xfrm_policy_delete(
912 								pol->security);
913 				if (err) {
914 					xfrm_audit_policy_delete(pol, 0,
915 								 task_valid);
916 					return err;
917 				}
918 			}
919 		}
920 	}
921 	return err;
922 }
923 #else
924 static inline int
925 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
926 {
927 	return 0;
928 }
929 #endif
930 
931 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
932 {
933 	int dir, err = 0, cnt = 0;
934 
935 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
936 
937 	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
938 	if (err)
939 		goto out;
940 
941 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
942 		struct xfrm_policy *pol;
943 		int i;
944 
945 	again1:
946 		hlist_for_each_entry(pol,
947 				     &net->xfrm.policy_inexact[dir], bydst) {
948 			if (pol->type != type)
949 				continue;
950 			__xfrm_policy_unlink(pol, dir);
951 			write_unlock_bh(&net->xfrm.xfrm_policy_lock);
952 			cnt++;
953 
954 			xfrm_audit_policy_delete(pol, 1, task_valid);
955 
956 			xfrm_policy_kill(pol);
957 
958 			write_lock_bh(&net->xfrm.xfrm_policy_lock);
959 			goto again1;
960 		}
961 
962 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
963 	again2:
964 			hlist_for_each_entry(pol,
965 					     net->xfrm.policy_bydst[dir].table + i,
966 					     bydst) {
967 				if (pol->type != type)
968 					continue;
969 				__xfrm_policy_unlink(pol, dir);
970 				write_unlock_bh(&net->xfrm.xfrm_policy_lock);
971 				cnt++;
972 
973 				xfrm_audit_policy_delete(pol, 1, task_valid);
974 				xfrm_policy_kill(pol);
975 
976 				write_lock_bh(&net->xfrm.xfrm_policy_lock);
977 				goto again2;
978 			}
979 		}
980 
981 	}
982 	if (!cnt)
983 		err = -ESRCH;
984 out:
985 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
986 	return err;
987 }
988 EXPORT_SYMBOL(xfrm_policy_flush);
989 
990 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
991 		     int (*func)(struct xfrm_policy *, int, int, void*),
992 		     void *data)
993 {
994 	struct xfrm_policy *pol;
995 	struct xfrm_policy_walk_entry *x;
996 	int error = 0;
997 
998 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
999 	    walk->type != XFRM_POLICY_TYPE_ANY)
1000 		return -EINVAL;
1001 
1002 	if (list_empty(&walk->walk.all) && walk->seq != 0)
1003 		return 0;
1004 
1005 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
1006 	if (list_empty(&walk->walk.all))
1007 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1008 	else
1009 		x = list_first_entry(&walk->walk.all,
1010 				     struct xfrm_policy_walk_entry, all);
1011 
1012 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1013 		if (x->dead)
1014 			continue;
1015 		pol = container_of(x, struct xfrm_policy, walk);
1016 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1017 		    walk->type != pol->type)
1018 			continue;
1019 		error = func(pol, xfrm_policy_id2dir(pol->index),
1020 			     walk->seq, data);
1021 		if (error) {
1022 			list_move_tail(&walk->walk.all, &x->all);
1023 			goto out;
1024 		}
1025 		walk->seq++;
1026 	}
1027 	if (walk->seq == 0) {
1028 		error = -ENOENT;
1029 		goto out;
1030 	}
1031 	list_del_init(&walk->walk.all);
1032 out:
1033 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1034 	return error;
1035 }
1036 EXPORT_SYMBOL(xfrm_policy_walk);
1037 
1038 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1039 {
1040 	INIT_LIST_HEAD(&walk->walk.all);
1041 	walk->walk.dead = 1;
1042 	walk->type = type;
1043 	walk->seq = 0;
1044 }
1045 EXPORT_SYMBOL(xfrm_policy_walk_init);
1046 
1047 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1048 {
1049 	if (list_empty(&walk->walk.all))
1050 		return;
1051 
1052 	write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1053 	list_del(&walk->walk.all);
1054 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1055 }
1056 EXPORT_SYMBOL(xfrm_policy_walk_done);
1057 
1058 /*
1059  * Find policy to apply to this flow.
1060  *
1061  * Returns 0 if policy found, else an -errno.
1062  */
1063 static int xfrm_policy_match(const struct xfrm_policy *pol,
1064 			     const struct flowi *fl,
1065 			     u8 type, u16 family, int dir)
1066 {
1067 	const struct xfrm_selector *sel = &pol->selector;
1068 	int ret = -ESRCH;
1069 	bool match;
1070 
1071 	if (pol->family != family ||
1072 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1073 	    pol->type != type)
1074 		return ret;
1075 
1076 	match = xfrm_selector_match(sel, fl, family);
1077 	if (match)
1078 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1079 						  dir);
1080 
1081 	return ret;
1082 }
1083 
1084 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1085 						     const struct flowi *fl,
1086 						     u16 family, u8 dir)
1087 {
1088 	int err;
1089 	struct xfrm_policy *pol, *ret;
1090 	const xfrm_address_t *daddr, *saddr;
1091 	struct hlist_head *chain;
1092 	u32 priority = ~0U;
1093 
1094 	daddr = xfrm_flowi_daddr(fl, family);
1095 	saddr = xfrm_flowi_saddr(fl, family);
1096 	if (unlikely(!daddr || !saddr))
1097 		return NULL;
1098 
1099 	read_lock_bh(&net->xfrm.xfrm_policy_lock);
1100 	chain = policy_hash_direct(net, daddr, saddr, family, dir);
1101 	ret = NULL;
1102 	hlist_for_each_entry(pol, chain, bydst) {
1103 		err = xfrm_policy_match(pol, fl, type, family, dir);
1104 		if (err) {
1105 			if (err == -ESRCH)
1106 				continue;
1107 			else {
1108 				ret = ERR_PTR(err);
1109 				goto fail;
1110 			}
1111 		} else {
1112 			ret = pol;
1113 			priority = ret->priority;
1114 			break;
1115 		}
1116 	}
1117 	chain = &net->xfrm.policy_inexact[dir];
1118 	hlist_for_each_entry(pol, chain, bydst) {
1119 		if ((pol->priority >= priority) && ret)
1120 			break;
1121 
1122 		err = xfrm_policy_match(pol, fl, type, family, dir);
1123 		if (err) {
1124 			if (err == -ESRCH)
1125 				continue;
1126 			else {
1127 				ret = ERR_PTR(err);
1128 				goto fail;
1129 			}
1130 		} else {
1131 			ret = pol;
1132 			break;
1133 		}
1134 	}
1135 
1136 	xfrm_pol_hold(ret);
1137 fail:
1138 	read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1139 
1140 	return ret;
1141 }
1142 
1143 static struct xfrm_policy *
1144 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1145 {
1146 #ifdef CONFIG_XFRM_SUB_POLICY
1147 	struct xfrm_policy *pol;
1148 
1149 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1150 	if (pol != NULL)
1151 		return pol;
1152 #endif
1153 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1154 }
1155 
1156 static int flow_to_policy_dir(int dir)
1157 {
1158 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1159 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1160 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
1161 		return dir;
1162 
1163 	switch (dir) {
1164 	default:
1165 	case FLOW_DIR_IN:
1166 		return XFRM_POLICY_IN;
1167 	case FLOW_DIR_OUT:
1168 		return XFRM_POLICY_OUT;
1169 	case FLOW_DIR_FWD:
1170 		return XFRM_POLICY_FWD;
1171 	}
1172 }
1173 
1174 static struct flow_cache_object *
1175 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1176 		   u8 dir, struct flow_cache_object *old_obj, void *ctx)
1177 {
1178 	struct xfrm_policy *pol;
1179 
1180 	if (old_obj)
1181 		xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1182 
1183 	pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1184 	if (IS_ERR_OR_NULL(pol))
1185 		return ERR_CAST(pol);
1186 
1187 	/* Resolver returns two references:
1188 	 * one for cache and one for caller of flow_cache_lookup() */
1189 	xfrm_pol_hold(pol);
1190 
1191 	return &pol->flo;
1192 }
1193 
1194 static inline int policy_to_flow_dir(int dir)
1195 {
1196 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1197 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1198 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
1199 		return dir;
1200 	switch (dir) {
1201 	default:
1202 	case XFRM_POLICY_IN:
1203 		return FLOW_DIR_IN;
1204 	case XFRM_POLICY_OUT:
1205 		return FLOW_DIR_OUT;
1206 	case XFRM_POLICY_FWD:
1207 		return FLOW_DIR_FWD;
1208 	}
1209 }
1210 
1211 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1212 						 const struct flowi *fl)
1213 {
1214 	struct xfrm_policy *pol;
1215 	struct net *net = sock_net(sk);
1216 
1217 	read_lock_bh(&net->xfrm.xfrm_policy_lock);
1218 	if ((pol = sk->sk_policy[dir]) != NULL) {
1219 		bool match = xfrm_selector_match(&pol->selector, fl,
1220 						 sk->sk_family);
1221 		int err = 0;
1222 
1223 		if (match) {
1224 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1225 				pol = NULL;
1226 				goto out;
1227 			}
1228 			err = security_xfrm_policy_lookup(pol->security,
1229 						      fl->flowi_secid,
1230 						      policy_to_flow_dir(dir));
1231 			if (!err)
1232 				xfrm_pol_hold(pol);
1233 			else if (err == -ESRCH)
1234 				pol = NULL;
1235 			else
1236 				pol = ERR_PTR(err);
1237 		} else
1238 			pol = NULL;
1239 	}
1240 out:
1241 	read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1242 	return pol;
1243 }
1244 
1245 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1246 {
1247 	struct net *net = xp_net(pol);
1248 
1249 	list_add(&pol->walk.all, &net->xfrm.policy_all);
1250 	net->xfrm.policy_count[dir]++;
1251 	xfrm_pol_hold(pol);
1252 }
1253 
1254 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1255 						int dir)
1256 {
1257 	struct net *net = xp_net(pol);
1258 
1259 	if (list_empty(&pol->walk.all))
1260 		return NULL;
1261 
1262 	/* Socket policies are not hashed. */
1263 	if (!hlist_unhashed(&pol->bydst)) {
1264 		hlist_del(&pol->bydst);
1265 		hlist_del(&pol->byidx);
1266 	}
1267 
1268 	list_del_init(&pol->walk.all);
1269 	net->xfrm.policy_count[dir]--;
1270 
1271 	return pol;
1272 }
1273 
1274 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1275 {
1276 	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1277 }
1278 
1279 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1280 {
1281 	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1282 }
1283 
1284 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1285 {
1286 	struct net *net = xp_net(pol);
1287 
1288 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
1289 	pol = __xfrm_policy_unlink(pol, dir);
1290 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1291 	if (pol) {
1292 		xfrm_policy_kill(pol);
1293 		return 0;
1294 	}
1295 	return -ENOENT;
1296 }
1297 EXPORT_SYMBOL(xfrm_policy_delete);
1298 
1299 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1300 {
1301 	struct net *net = xp_net(pol);
1302 	struct xfrm_policy *old_pol;
1303 
1304 #ifdef CONFIG_XFRM_SUB_POLICY
1305 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1306 		return -EINVAL;
1307 #endif
1308 
1309 	write_lock_bh(&net->xfrm.xfrm_policy_lock);
1310 	old_pol = sk->sk_policy[dir];
1311 	sk->sk_policy[dir] = pol;
1312 	if (pol) {
1313 		pol->curlft.add_time = get_seconds();
1314 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1315 		xfrm_sk_policy_link(pol, dir);
1316 	}
1317 	if (old_pol) {
1318 		if (pol)
1319 			xfrm_policy_requeue(old_pol, pol);
1320 
1321 		/* Unlinking succeeds always. This is the only function
1322 		 * allowed to delete or replace socket policy.
1323 		 */
1324 		xfrm_sk_policy_unlink(old_pol, dir);
1325 	}
1326 	write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1327 
1328 	if (old_pol) {
1329 		xfrm_policy_kill(old_pol);
1330 	}
1331 	return 0;
1332 }
1333 
1334 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1335 {
1336 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1337 	struct net *net = xp_net(old);
1338 
1339 	if (newp) {
1340 		newp->selector = old->selector;
1341 		if (security_xfrm_policy_clone(old->security,
1342 					       &newp->security)) {
1343 			kfree(newp);
1344 			return NULL;  /* ENOMEM */
1345 		}
1346 		newp->lft = old->lft;
1347 		newp->curlft = old->curlft;
1348 		newp->mark = old->mark;
1349 		newp->action = old->action;
1350 		newp->flags = old->flags;
1351 		newp->xfrm_nr = old->xfrm_nr;
1352 		newp->index = old->index;
1353 		newp->type = old->type;
1354 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1355 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1356 		write_lock_bh(&net->xfrm.xfrm_policy_lock);
1357 		xfrm_sk_policy_link(newp, dir);
1358 		write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1359 		xfrm_pol_put(newp);
1360 	}
1361 	return newp;
1362 }
1363 
1364 int __xfrm_sk_clone_policy(struct sock *sk)
1365 {
1366 	struct xfrm_policy *p0 = sk->sk_policy[0],
1367 			   *p1 = sk->sk_policy[1];
1368 
1369 	sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1370 	if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1371 		return -ENOMEM;
1372 	if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1373 		return -ENOMEM;
1374 	return 0;
1375 }
1376 
1377 static int
1378 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1379 	       xfrm_address_t *remote, unsigned short family)
1380 {
1381 	int err;
1382 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1383 
1384 	if (unlikely(afinfo == NULL))
1385 		return -EINVAL;
1386 	err = afinfo->get_saddr(net, oif, local, remote);
1387 	xfrm_policy_put_afinfo(afinfo);
1388 	return err;
1389 }
1390 
1391 /* Resolve list of templates for the flow, given policy. */
1392 
1393 static int
1394 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1395 		      struct xfrm_state **xfrm, unsigned short family)
1396 {
1397 	struct net *net = xp_net(policy);
1398 	int nx;
1399 	int i, error;
1400 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1401 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1402 	xfrm_address_t tmp;
1403 
1404 	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1405 		struct xfrm_state *x;
1406 		xfrm_address_t *remote = daddr;
1407 		xfrm_address_t *local  = saddr;
1408 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1409 
1410 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1411 		    tmpl->mode == XFRM_MODE_BEET) {
1412 			remote = &tmpl->id.daddr;
1413 			local = &tmpl->saddr;
1414 			if (xfrm_addr_any(local, tmpl->encap_family)) {
1415 				error = xfrm_get_saddr(net, fl->flowi_oif,
1416 						       &tmp, remote,
1417 						       tmpl->encap_family);
1418 				if (error)
1419 					goto fail;
1420 				local = &tmp;
1421 			}
1422 		}
1423 
1424 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1425 
1426 		if (x && x->km.state == XFRM_STATE_VALID) {
1427 			xfrm[nx++] = x;
1428 			daddr = remote;
1429 			saddr = local;
1430 			continue;
1431 		}
1432 		if (x) {
1433 			error = (x->km.state == XFRM_STATE_ERROR ?
1434 				 -EINVAL : -EAGAIN);
1435 			xfrm_state_put(x);
1436 		} else if (error == -ESRCH) {
1437 			error = -EAGAIN;
1438 		}
1439 
1440 		if (!tmpl->optional)
1441 			goto fail;
1442 	}
1443 	return nx;
1444 
1445 fail:
1446 	for (nx--; nx >= 0; nx--)
1447 		xfrm_state_put(xfrm[nx]);
1448 	return error;
1449 }
1450 
1451 static int
1452 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1453 		  struct xfrm_state **xfrm, unsigned short family)
1454 {
1455 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1456 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1457 	int cnx = 0;
1458 	int error;
1459 	int ret;
1460 	int i;
1461 
1462 	for (i = 0; i < npols; i++) {
1463 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1464 			error = -ENOBUFS;
1465 			goto fail;
1466 		}
1467 
1468 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1469 		if (ret < 0) {
1470 			error = ret;
1471 			goto fail;
1472 		} else
1473 			cnx += ret;
1474 	}
1475 
1476 	/* found states are sorted for outbound processing */
1477 	if (npols > 1)
1478 		xfrm_state_sort(xfrm, tpp, cnx, family);
1479 
1480 	return cnx;
1481 
1482  fail:
1483 	for (cnx--; cnx >= 0; cnx--)
1484 		xfrm_state_put(tpp[cnx]);
1485 	return error;
1486 
1487 }
1488 
1489 /* Check that the bundle accepts the flow and its components are
1490  * still valid.
1491  */
1492 
1493 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1494 {
1495 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1496 	int tos;
1497 
1498 	if (!afinfo)
1499 		return -EINVAL;
1500 
1501 	tos = afinfo->get_tos(fl);
1502 
1503 	xfrm_policy_put_afinfo(afinfo);
1504 
1505 	return tos;
1506 }
1507 
1508 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1509 {
1510 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1511 	struct dst_entry *dst = &xdst->u.dst;
1512 
1513 	if (xdst->route == NULL) {
1514 		/* Dummy bundle - if it has xfrms we were not
1515 		 * able to build bundle as template resolution failed.
1516 		 * It means we need to try again resolving. */
1517 		if (xdst->num_xfrms > 0)
1518 			return NULL;
1519 	} else if (dst->flags & DST_XFRM_QUEUE) {
1520 		return NULL;
1521 	} else {
1522 		/* Real bundle */
1523 		if (stale_bundle(dst))
1524 			return NULL;
1525 	}
1526 
1527 	dst_hold(dst);
1528 	return flo;
1529 }
1530 
1531 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1532 {
1533 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1534 	struct dst_entry *dst = &xdst->u.dst;
1535 
1536 	if (!xdst->route)
1537 		return 0;
1538 	if (stale_bundle(dst))
1539 		return 0;
1540 
1541 	return 1;
1542 }
1543 
1544 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1545 {
1546 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1547 	struct dst_entry *dst = &xdst->u.dst;
1548 
1549 	dst_free(dst);
1550 }
1551 
1552 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1553 	.get = xfrm_bundle_flo_get,
1554 	.check = xfrm_bundle_flo_check,
1555 	.delete = xfrm_bundle_flo_delete,
1556 };
1557 
1558 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1559 {
1560 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1561 	struct dst_ops *dst_ops;
1562 	struct xfrm_dst *xdst;
1563 
1564 	if (!afinfo)
1565 		return ERR_PTR(-EINVAL);
1566 
1567 	switch (family) {
1568 	case AF_INET:
1569 		dst_ops = &net->xfrm.xfrm4_dst_ops;
1570 		break;
1571 #if IS_ENABLED(CONFIG_IPV6)
1572 	case AF_INET6:
1573 		dst_ops = &net->xfrm.xfrm6_dst_ops;
1574 		break;
1575 #endif
1576 	default:
1577 		BUG();
1578 	}
1579 	xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1580 
1581 	if (likely(xdst)) {
1582 		struct dst_entry *dst = &xdst->u.dst;
1583 
1584 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1585 		xdst->flo.ops = &xfrm_bundle_fc_ops;
1586 	} else
1587 		xdst = ERR_PTR(-ENOBUFS);
1588 
1589 	xfrm_policy_put_afinfo(afinfo);
1590 
1591 	return xdst;
1592 }
1593 
1594 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1595 				 int nfheader_len)
1596 {
1597 	struct xfrm_policy_afinfo *afinfo =
1598 		xfrm_policy_get_afinfo(dst->ops->family);
1599 	int err;
1600 
1601 	if (!afinfo)
1602 		return -EINVAL;
1603 
1604 	err = afinfo->init_path(path, dst, nfheader_len);
1605 
1606 	xfrm_policy_put_afinfo(afinfo);
1607 
1608 	return err;
1609 }
1610 
1611 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1612 				const struct flowi *fl)
1613 {
1614 	struct xfrm_policy_afinfo *afinfo =
1615 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1616 	int err;
1617 
1618 	if (!afinfo)
1619 		return -EINVAL;
1620 
1621 	err = afinfo->fill_dst(xdst, dev, fl);
1622 
1623 	xfrm_policy_put_afinfo(afinfo);
1624 
1625 	return err;
1626 }
1627 
1628 
1629 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1630  * all the metrics... Shortly, bundle a bundle.
1631  */
1632 
1633 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1634 					    struct xfrm_state **xfrm, int nx,
1635 					    const struct flowi *fl,
1636 					    struct dst_entry *dst)
1637 {
1638 	struct net *net = xp_net(policy);
1639 	unsigned long now = jiffies;
1640 	struct net_device *dev;
1641 	struct xfrm_mode *inner_mode;
1642 	struct dst_entry *dst_prev = NULL;
1643 	struct dst_entry *dst0 = NULL;
1644 	int i = 0;
1645 	int err;
1646 	int header_len = 0;
1647 	int nfheader_len = 0;
1648 	int trailer_len = 0;
1649 	int tos;
1650 	int family = policy->selector.family;
1651 	xfrm_address_t saddr, daddr;
1652 
1653 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1654 
1655 	tos = xfrm_get_tos(fl, family);
1656 	err = tos;
1657 	if (tos < 0)
1658 		goto put_states;
1659 
1660 	dst_hold(dst);
1661 
1662 	for (; i < nx; i++) {
1663 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1664 		struct dst_entry *dst1 = &xdst->u.dst;
1665 
1666 		err = PTR_ERR(xdst);
1667 		if (IS_ERR(xdst)) {
1668 			dst_release(dst);
1669 			goto put_states;
1670 		}
1671 
1672 		if (xfrm[i]->sel.family == AF_UNSPEC) {
1673 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
1674 							xfrm_af2proto(family));
1675 			if (!inner_mode) {
1676 				err = -EAFNOSUPPORT;
1677 				dst_release(dst);
1678 				goto put_states;
1679 			}
1680 		} else
1681 			inner_mode = xfrm[i]->inner_mode;
1682 
1683 		if (!dst_prev)
1684 			dst0 = dst1;
1685 		else {
1686 			dst_prev->child = dst_clone(dst1);
1687 			dst1->flags |= DST_NOHASH;
1688 		}
1689 
1690 		xdst->route = dst;
1691 		dst_copy_metrics(dst1, dst);
1692 
1693 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1694 			family = xfrm[i]->props.family;
1695 			dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1696 					      &saddr, &daddr, family);
1697 			err = PTR_ERR(dst);
1698 			if (IS_ERR(dst))
1699 				goto put_states;
1700 		} else
1701 			dst_hold(dst);
1702 
1703 		dst1->xfrm = xfrm[i];
1704 		xdst->xfrm_genid = xfrm[i]->genid;
1705 
1706 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1707 		dst1->flags |= DST_HOST;
1708 		dst1->lastuse = now;
1709 
1710 		dst1->input = dst_discard;
1711 		dst1->output = inner_mode->afinfo->output;
1712 
1713 		dst1->next = dst_prev;
1714 		dst_prev = dst1;
1715 
1716 		header_len += xfrm[i]->props.header_len;
1717 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1718 			nfheader_len += xfrm[i]->props.header_len;
1719 		trailer_len += xfrm[i]->props.trailer_len;
1720 	}
1721 
1722 	dst_prev->child = dst;
1723 	dst0->path = dst;
1724 
1725 	err = -ENODEV;
1726 	dev = dst->dev;
1727 	if (!dev)
1728 		goto free_dst;
1729 
1730 	xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1731 	xfrm_init_pmtu(dst_prev);
1732 
1733 	for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1734 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1735 
1736 		err = xfrm_fill_dst(xdst, dev, fl);
1737 		if (err)
1738 			goto free_dst;
1739 
1740 		dst_prev->header_len = header_len;
1741 		dst_prev->trailer_len = trailer_len;
1742 		header_len -= xdst->u.dst.xfrm->props.header_len;
1743 		trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1744 	}
1745 
1746 out:
1747 	return dst0;
1748 
1749 put_states:
1750 	for (; i < nx; i++)
1751 		xfrm_state_put(xfrm[i]);
1752 free_dst:
1753 	if (dst0)
1754 		dst_free(dst0);
1755 	dst0 = ERR_PTR(err);
1756 	goto out;
1757 }
1758 
1759 #ifdef CONFIG_XFRM_SUB_POLICY
1760 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1761 {
1762 	if (!*target) {
1763 		*target = kmalloc(size, GFP_ATOMIC);
1764 		if (!*target)
1765 			return -ENOMEM;
1766 	}
1767 
1768 	memcpy(*target, src, size);
1769 	return 0;
1770 }
1771 #endif
1772 
1773 static int xfrm_dst_update_parent(struct dst_entry *dst,
1774 				  const struct xfrm_selector *sel)
1775 {
1776 #ifdef CONFIG_XFRM_SUB_POLICY
1777 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1778 	return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1779 				   sel, sizeof(*sel));
1780 #else
1781 	return 0;
1782 #endif
1783 }
1784 
1785 static int xfrm_dst_update_origin(struct dst_entry *dst,
1786 				  const struct flowi *fl)
1787 {
1788 #ifdef CONFIG_XFRM_SUB_POLICY
1789 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1790 	return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1791 #else
1792 	return 0;
1793 #endif
1794 }
1795 
1796 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1797 				struct xfrm_policy **pols,
1798 				int *num_pols, int *num_xfrms)
1799 {
1800 	int i;
1801 
1802 	if (*num_pols == 0 || !pols[0]) {
1803 		*num_pols = 0;
1804 		*num_xfrms = 0;
1805 		return 0;
1806 	}
1807 	if (IS_ERR(pols[0]))
1808 		return PTR_ERR(pols[0]);
1809 
1810 	*num_xfrms = pols[0]->xfrm_nr;
1811 
1812 #ifdef CONFIG_XFRM_SUB_POLICY
1813 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1814 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1815 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1816 						    XFRM_POLICY_TYPE_MAIN,
1817 						    fl, family,
1818 						    XFRM_POLICY_OUT);
1819 		if (pols[1]) {
1820 			if (IS_ERR(pols[1])) {
1821 				xfrm_pols_put(pols, *num_pols);
1822 				return PTR_ERR(pols[1]);
1823 			}
1824 			(*num_pols)++;
1825 			(*num_xfrms) += pols[1]->xfrm_nr;
1826 		}
1827 	}
1828 #endif
1829 	for (i = 0; i < *num_pols; i++) {
1830 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
1831 			*num_xfrms = -1;
1832 			break;
1833 		}
1834 	}
1835 
1836 	return 0;
1837 
1838 }
1839 
1840 static struct xfrm_dst *
1841 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1842 			       const struct flowi *fl, u16 family,
1843 			       struct dst_entry *dst_orig)
1844 {
1845 	struct net *net = xp_net(pols[0]);
1846 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1847 	struct dst_entry *dst;
1848 	struct xfrm_dst *xdst;
1849 	int err;
1850 
1851 	/* Try to instantiate a bundle */
1852 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1853 	if (err <= 0) {
1854 		if (err != 0 && err != -EAGAIN)
1855 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1856 		return ERR_PTR(err);
1857 	}
1858 
1859 	dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1860 	if (IS_ERR(dst)) {
1861 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1862 		return ERR_CAST(dst);
1863 	}
1864 
1865 	xdst = (struct xfrm_dst *)dst;
1866 	xdst->num_xfrms = err;
1867 	if (num_pols > 1)
1868 		err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1869 	else
1870 		err = xfrm_dst_update_origin(dst, fl);
1871 	if (unlikely(err)) {
1872 		dst_free(dst);
1873 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1874 		return ERR_PTR(err);
1875 	}
1876 
1877 	xdst->num_pols = num_pols;
1878 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1879 	xdst->policy_genid = atomic_read(&pols[0]->genid);
1880 
1881 	return xdst;
1882 }
1883 
1884 static void xfrm_policy_queue_process(unsigned long arg)
1885 {
1886 	struct sk_buff *skb;
1887 	struct sock *sk;
1888 	struct dst_entry *dst;
1889 	struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1890 	struct xfrm_policy_queue *pq = &pol->polq;
1891 	struct flowi fl;
1892 	struct sk_buff_head list;
1893 
1894 	spin_lock(&pq->hold_queue.lock);
1895 	skb = skb_peek(&pq->hold_queue);
1896 	if (!skb) {
1897 		spin_unlock(&pq->hold_queue.lock);
1898 		goto out;
1899 	}
1900 	dst = skb_dst(skb);
1901 	sk = skb->sk;
1902 	xfrm_decode_session(skb, &fl, dst->ops->family);
1903 	spin_unlock(&pq->hold_queue.lock);
1904 
1905 	dst_hold(dst->path);
1906 	dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1907 			  sk, 0);
1908 	if (IS_ERR(dst))
1909 		goto purge_queue;
1910 
1911 	if (dst->flags & DST_XFRM_QUEUE) {
1912 		dst_release(dst);
1913 
1914 		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1915 			goto purge_queue;
1916 
1917 		pq->timeout = pq->timeout << 1;
1918 		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1919 			xfrm_pol_hold(pol);
1920 	goto out;
1921 	}
1922 
1923 	dst_release(dst);
1924 
1925 	__skb_queue_head_init(&list);
1926 
1927 	spin_lock(&pq->hold_queue.lock);
1928 	pq->timeout = 0;
1929 	skb_queue_splice_init(&pq->hold_queue, &list);
1930 	spin_unlock(&pq->hold_queue.lock);
1931 
1932 	while (!skb_queue_empty(&list)) {
1933 		skb = __skb_dequeue(&list);
1934 
1935 		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1936 		dst_hold(skb_dst(skb)->path);
1937 		dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1938 				  &fl, skb->sk, 0);
1939 		if (IS_ERR(dst)) {
1940 			kfree_skb(skb);
1941 			continue;
1942 		}
1943 
1944 		nf_reset(skb);
1945 		skb_dst_drop(skb);
1946 		skb_dst_set(skb, dst);
1947 
1948 		dst_output(skb->sk, skb);
1949 	}
1950 
1951 out:
1952 	xfrm_pol_put(pol);
1953 	return;
1954 
1955 purge_queue:
1956 	pq->timeout = 0;
1957 	skb_queue_purge(&pq->hold_queue);
1958 	xfrm_pol_put(pol);
1959 }
1960 
1961 static int xdst_queue_output(struct sock *sk, struct sk_buff *skb)
1962 {
1963 	unsigned long sched_next;
1964 	struct dst_entry *dst = skb_dst(skb);
1965 	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1966 	struct xfrm_policy *pol = xdst->pols[0];
1967 	struct xfrm_policy_queue *pq = &pol->polq;
1968 
1969 	if (unlikely(skb_fclone_busy(sk, skb))) {
1970 		kfree_skb(skb);
1971 		return 0;
1972 	}
1973 
1974 	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1975 		kfree_skb(skb);
1976 		return -EAGAIN;
1977 	}
1978 
1979 	skb_dst_force(skb);
1980 
1981 	spin_lock_bh(&pq->hold_queue.lock);
1982 
1983 	if (!pq->timeout)
1984 		pq->timeout = XFRM_QUEUE_TMO_MIN;
1985 
1986 	sched_next = jiffies + pq->timeout;
1987 
1988 	if (del_timer(&pq->hold_timer)) {
1989 		if (time_before(pq->hold_timer.expires, sched_next))
1990 			sched_next = pq->hold_timer.expires;
1991 		xfrm_pol_put(pol);
1992 	}
1993 
1994 	__skb_queue_tail(&pq->hold_queue, skb);
1995 	if (!mod_timer(&pq->hold_timer, sched_next))
1996 		xfrm_pol_hold(pol);
1997 
1998 	spin_unlock_bh(&pq->hold_queue.lock);
1999 
2000 	return 0;
2001 }
2002 
2003 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2004 						 struct xfrm_flo *xflo,
2005 						 const struct flowi *fl,
2006 						 int num_xfrms,
2007 						 u16 family)
2008 {
2009 	int err;
2010 	struct net_device *dev;
2011 	struct dst_entry *dst;
2012 	struct dst_entry *dst1;
2013 	struct xfrm_dst *xdst;
2014 
2015 	xdst = xfrm_alloc_dst(net, family);
2016 	if (IS_ERR(xdst))
2017 		return xdst;
2018 
2019 	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2020 	    net->xfrm.sysctl_larval_drop ||
2021 	    num_xfrms <= 0)
2022 		return xdst;
2023 
2024 	dst = xflo->dst_orig;
2025 	dst1 = &xdst->u.dst;
2026 	dst_hold(dst);
2027 	xdst->route = dst;
2028 
2029 	dst_copy_metrics(dst1, dst);
2030 
2031 	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2032 	dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2033 	dst1->lastuse = jiffies;
2034 
2035 	dst1->input = dst_discard;
2036 	dst1->output = xdst_queue_output;
2037 
2038 	dst_hold(dst);
2039 	dst1->child = dst;
2040 	dst1->path = dst;
2041 
2042 	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2043 
2044 	err = -ENODEV;
2045 	dev = dst->dev;
2046 	if (!dev)
2047 		goto free_dst;
2048 
2049 	err = xfrm_fill_dst(xdst, dev, fl);
2050 	if (err)
2051 		goto free_dst;
2052 
2053 out:
2054 	return xdst;
2055 
2056 free_dst:
2057 	dst_release(dst1);
2058 	xdst = ERR_PTR(err);
2059 	goto out;
2060 }
2061 
2062 static struct flow_cache_object *
2063 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2064 		   struct flow_cache_object *oldflo, void *ctx)
2065 {
2066 	struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2067 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2068 	struct xfrm_dst *xdst, *new_xdst;
2069 	int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2070 
2071 	/* Check if the policies from old bundle are usable */
2072 	xdst = NULL;
2073 	if (oldflo) {
2074 		xdst = container_of(oldflo, struct xfrm_dst, flo);
2075 		num_pols = xdst->num_pols;
2076 		num_xfrms = xdst->num_xfrms;
2077 		pol_dead = 0;
2078 		for (i = 0; i < num_pols; i++) {
2079 			pols[i] = xdst->pols[i];
2080 			pol_dead |= pols[i]->walk.dead;
2081 		}
2082 		if (pol_dead) {
2083 			dst_free(&xdst->u.dst);
2084 			xdst = NULL;
2085 			num_pols = 0;
2086 			num_xfrms = 0;
2087 			oldflo = NULL;
2088 		}
2089 	}
2090 
2091 	/* Resolve policies to use if we couldn't get them from
2092 	 * previous cache entry */
2093 	if (xdst == NULL) {
2094 		num_pols = 1;
2095 		pols[0] = __xfrm_policy_lookup(net, fl, family,
2096 					       flow_to_policy_dir(dir));
2097 		err = xfrm_expand_policies(fl, family, pols,
2098 					   &num_pols, &num_xfrms);
2099 		if (err < 0)
2100 			goto inc_error;
2101 		if (num_pols == 0)
2102 			return NULL;
2103 		if (num_xfrms <= 0)
2104 			goto make_dummy_bundle;
2105 	}
2106 
2107 	new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2108 						  xflo->dst_orig);
2109 	if (IS_ERR(new_xdst)) {
2110 		err = PTR_ERR(new_xdst);
2111 		if (err != -EAGAIN)
2112 			goto error;
2113 		if (oldflo == NULL)
2114 			goto make_dummy_bundle;
2115 		dst_hold(&xdst->u.dst);
2116 		return oldflo;
2117 	} else if (new_xdst == NULL) {
2118 		num_xfrms = 0;
2119 		if (oldflo == NULL)
2120 			goto make_dummy_bundle;
2121 		xdst->num_xfrms = 0;
2122 		dst_hold(&xdst->u.dst);
2123 		return oldflo;
2124 	}
2125 
2126 	/* Kill the previous bundle */
2127 	if (xdst) {
2128 		/* The policies were stolen for newly generated bundle */
2129 		xdst->num_pols = 0;
2130 		dst_free(&xdst->u.dst);
2131 	}
2132 
2133 	/* Flow cache does not have reference, it dst_free()'s,
2134 	 * but we do need to return one reference for original caller */
2135 	dst_hold(&new_xdst->u.dst);
2136 	return &new_xdst->flo;
2137 
2138 make_dummy_bundle:
2139 	/* We found policies, but there's no bundles to instantiate:
2140 	 * either because the policy blocks, has no transformations or
2141 	 * we could not build template (no xfrm_states).*/
2142 	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2143 	if (IS_ERR(xdst)) {
2144 		xfrm_pols_put(pols, num_pols);
2145 		return ERR_CAST(xdst);
2146 	}
2147 	xdst->num_pols = num_pols;
2148 	xdst->num_xfrms = num_xfrms;
2149 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2150 
2151 	dst_hold(&xdst->u.dst);
2152 	return &xdst->flo;
2153 
2154 inc_error:
2155 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2156 error:
2157 	if (xdst != NULL)
2158 		dst_free(&xdst->u.dst);
2159 	else
2160 		xfrm_pols_put(pols, num_pols);
2161 	return ERR_PTR(err);
2162 }
2163 
2164 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2165 					struct dst_entry *dst_orig)
2166 {
2167 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2168 	struct dst_entry *ret;
2169 
2170 	if (!afinfo) {
2171 		dst_release(dst_orig);
2172 		return ERR_PTR(-EINVAL);
2173 	} else {
2174 		ret = afinfo->blackhole_route(net, dst_orig);
2175 	}
2176 	xfrm_policy_put_afinfo(afinfo);
2177 
2178 	return ret;
2179 }
2180 
2181 /* Main function: finds/creates a bundle for given flow.
2182  *
2183  * At the moment we eat a raw IP route. Mostly to speed up lookups
2184  * on interfaces with disabled IPsec.
2185  */
2186 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2187 			      const struct flowi *fl,
2188 			      struct sock *sk, int flags)
2189 {
2190 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2191 	struct flow_cache_object *flo;
2192 	struct xfrm_dst *xdst;
2193 	struct dst_entry *dst, *route;
2194 	u16 family = dst_orig->ops->family;
2195 	u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2196 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2197 
2198 	dst = NULL;
2199 	xdst = NULL;
2200 	route = NULL;
2201 
2202 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2203 		num_pols = 1;
2204 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2205 		err = xfrm_expand_policies(fl, family, pols,
2206 					   &num_pols, &num_xfrms);
2207 		if (err < 0)
2208 			goto dropdst;
2209 
2210 		if (num_pols) {
2211 			if (num_xfrms <= 0) {
2212 				drop_pols = num_pols;
2213 				goto no_transform;
2214 			}
2215 
2216 			xdst = xfrm_resolve_and_create_bundle(
2217 					pols, num_pols, fl,
2218 					family, dst_orig);
2219 			if (IS_ERR(xdst)) {
2220 				xfrm_pols_put(pols, num_pols);
2221 				err = PTR_ERR(xdst);
2222 				goto dropdst;
2223 			} else if (xdst == NULL) {
2224 				num_xfrms = 0;
2225 				drop_pols = num_pols;
2226 				goto no_transform;
2227 			}
2228 
2229 			dst_hold(&xdst->u.dst);
2230 			xdst->u.dst.flags |= DST_NOCACHE;
2231 			route = xdst->route;
2232 		}
2233 	}
2234 
2235 	if (xdst == NULL) {
2236 		struct xfrm_flo xflo;
2237 
2238 		xflo.dst_orig = dst_orig;
2239 		xflo.flags = flags;
2240 
2241 		/* To accelerate a bit...  */
2242 		if ((dst_orig->flags & DST_NOXFRM) ||
2243 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
2244 			goto nopol;
2245 
2246 		flo = flow_cache_lookup(net, fl, family, dir,
2247 					xfrm_bundle_lookup, &xflo);
2248 		if (flo == NULL)
2249 			goto nopol;
2250 		if (IS_ERR(flo)) {
2251 			err = PTR_ERR(flo);
2252 			goto dropdst;
2253 		}
2254 		xdst = container_of(flo, struct xfrm_dst, flo);
2255 
2256 		num_pols = xdst->num_pols;
2257 		num_xfrms = xdst->num_xfrms;
2258 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2259 		route = xdst->route;
2260 	}
2261 
2262 	dst = &xdst->u.dst;
2263 	if (route == NULL && num_xfrms > 0) {
2264 		/* The only case when xfrm_bundle_lookup() returns a
2265 		 * bundle with null route, is when the template could
2266 		 * not be resolved. It means policies are there, but
2267 		 * bundle could not be created, since we don't yet
2268 		 * have the xfrm_state's. We need to wait for KM to
2269 		 * negotiate new SA's or bail out with error.*/
2270 		if (net->xfrm.sysctl_larval_drop) {
2271 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2272 			err = -EREMOTE;
2273 			goto error;
2274 		}
2275 
2276 		err = -EAGAIN;
2277 
2278 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2279 		goto error;
2280 	}
2281 
2282 no_transform:
2283 	if (num_pols == 0)
2284 		goto nopol;
2285 
2286 	if ((flags & XFRM_LOOKUP_ICMP) &&
2287 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2288 		err = -ENOENT;
2289 		goto error;
2290 	}
2291 
2292 	for (i = 0; i < num_pols; i++)
2293 		pols[i]->curlft.use_time = get_seconds();
2294 
2295 	if (num_xfrms < 0) {
2296 		/* Prohibit the flow */
2297 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2298 		err = -EPERM;
2299 		goto error;
2300 	} else if (num_xfrms > 0) {
2301 		/* Flow transformed */
2302 		dst_release(dst_orig);
2303 	} else {
2304 		/* Flow passes untransformed */
2305 		dst_release(dst);
2306 		dst = dst_orig;
2307 	}
2308 ok:
2309 	xfrm_pols_put(pols, drop_pols);
2310 	if (dst && dst->xfrm &&
2311 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2312 		dst->flags |= DST_XFRM_TUNNEL;
2313 	return dst;
2314 
2315 nopol:
2316 	if (!(flags & XFRM_LOOKUP_ICMP)) {
2317 		dst = dst_orig;
2318 		goto ok;
2319 	}
2320 	err = -ENOENT;
2321 error:
2322 	dst_release(dst);
2323 dropdst:
2324 	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2325 		dst_release(dst_orig);
2326 	xfrm_pols_put(pols, drop_pols);
2327 	return ERR_PTR(err);
2328 }
2329 EXPORT_SYMBOL(xfrm_lookup);
2330 
2331 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2332  * Otherwise we may send out blackholed packets.
2333  */
2334 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2335 				    const struct flowi *fl,
2336 				    struct sock *sk, int flags)
2337 {
2338 	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2339 					    flags | XFRM_LOOKUP_QUEUE |
2340 					    XFRM_LOOKUP_KEEP_DST_REF);
2341 
2342 	if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2343 		return make_blackhole(net, dst_orig->ops->family, dst_orig);
2344 
2345 	return dst;
2346 }
2347 EXPORT_SYMBOL(xfrm_lookup_route);
2348 
2349 static inline int
2350 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2351 {
2352 	struct xfrm_state *x;
2353 
2354 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2355 		return 0;
2356 	x = skb->sp->xvec[idx];
2357 	if (!x->type->reject)
2358 		return 0;
2359 	return x->type->reject(x, skb, fl);
2360 }
2361 
2362 /* When skb is transformed back to its "native" form, we have to
2363  * check policy restrictions. At the moment we make this in maximally
2364  * stupid way. Shame on me. :-) Of course, connected sockets must
2365  * have policy cached at them.
2366  */
2367 
2368 static inline int
2369 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2370 	      unsigned short family)
2371 {
2372 	if (xfrm_state_kern(x))
2373 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2374 	return	x->id.proto == tmpl->id.proto &&
2375 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2376 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2377 		x->props.mode == tmpl->mode &&
2378 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2379 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2380 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
2381 		  xfrm_state_addr_cmp(tmpl, x, family));
2382 }
2383 
2384 /*
2385  * 0 or more than 0 is returned when validation is succeeded (either bypass
2386  * because of optional transport mode, or next index of the mathced secpath
2387  * state with the template.
2388  * -1 is returned when no matching template is found.
2389  * Otherwise "-2 - errored_index" is returned.
2390  */
2391 static inline int
2392 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2393 	       unsigned short family)
2394 {
2395 	int idx = start;
2396 
2397 	if (tmpl->optional) {
2398 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
2399 			return start;
2400 	} else
2401 		start = -1;
2402 	for (; idx < sp->len; idx++) {
2403 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2404 			return ++idx;
2405 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2406 			if (start == -1)
2407 				start = -2-idx;
2408 			break;
2409 		}
2410 	}
2411 	return start;
2412 }
2413 
2414 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2415 			  unsigned int family, int reverse)
2416 {
2417 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2418 	int err;
2419 
2420 	if (unlikely(afinfo == NULL))
2421 		return -EAFNOSUPPORT;
2422 
2423 	afinfo->decode_session(skb, fl, reverse);
2424 	err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2425 	xfrm_policy_put_afinfo(afinfo);
2426 	return err;
2427 }
2428 EXPORT_SYMBOL(__xfrm_decode_session);
2429 
2430 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2431 {
2432 	for (; k < sp->len; k++) {
2433 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2434 			*idxp = k;
2435 			return 1;
2436 		}
2437 	}
2438 
2439 	return 0;
2440 }
2441 
2442 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2443 			unsigned short family)
2444 {
2445 	struct net *net = dev_net(skb->dev);
2446 	struct xfrm_policy *pol;
2447 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2448 	int npols = 0;
2449 	int xfrm_nr;
2450 	int pi;
2451 	int reverse;
2452 	struct flowi fl;
2453 	u8 fl_dir;
2454 	int xerr_idx = -1;
2455 
2456 	reverse = dir & ~XFRM_POLICY_MASK;
2457 	dir &= XFRM_POLICY_MASK;
2458 	fl_dir = policy_to_flow_dir(dir);
2459 
2460 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2461 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2462 		return 0;
2463 	}
2464 
2465 	nf_nat_decode_session(skb, &fl, family);
2466 
2467 	/* First, check used SA against their selectors. */
2468 	if (skb->sp) {
2469 		int i;
2470 
2471 		for (i = skb->sp->len-1; i >= 0; i--) {
2472 			struct xfrm_state *x = skb->sp->xvec[i];
2473 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
2474 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2475 				return 0;
2476 			}
2477 		}
2478 	}
2479 
2480 	pol = NULL;
2481 	if (sk && sk->sk_policy[dir]) {
2482 		pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2483 		if (IS_ERR(pol)) {
2484 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2485 			return 0;
2486 		}
2487 	}
2488 
2489 	if (!pol) {
2490 		struct flow_cache_object *flo;
2491 
2492 		flo = flow_cache_lookup(net, &fl, family, fl_dir,
2493 					xfrm_policy_lookup, NULL);
2494 		if (IS_ERR_OR_NULL(flo))
2495 			pol = ERR_CAST(flo);
2496 		else
2497 			pol = container_of(flo, struct xfrm_policy, flo);
2498 	}
2499 
2500 	if (IS_ERR(pol)) {
2501 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2502 		return 0;
2503 	}
2504 
2505 	if (!pol) {
2506 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2507 			xfrm_secpath_reject(xerr_idx, skb, &fl);
2508 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2509 			return 0;
2510 		}
2511 		return 1;
2512 	}
2513 
2514 	pol->curlft.use_time = get_seconds();
2515 
2516 	pols[0] = pol;
2517 	npols++;
2518 #ifdef CONFIG_XFRM_SUB_POLICY
2519 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2520 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2521 						    &fl, family,
2522 						    XFRM_POLICY_IN);
2523 		if (pols[1]) {
2524 			if (IS_ERR(pols[1])) {
2525 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2526 				return 0;
2527 			}
2528 			pols[1]->curlft.use_time = get_seconds();
2529 			npols++;
2530 		}
2531 	}
2532 #endif
2533 
2534 	if (pol->action == XFRM_POLICY_ALLOW) {
2535 		struct sec_path *sp;
2536 		static struct sec_path dummy;
2537 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2538 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2539 		struct xfrm_tmpl **tpp = tp;
2540 		int ti = 0;
2541 		int i, k;
2542 
2543 		if ((sp = skb->sp) == NULL)
2544 			sp = &dummy;
2545 
2546 		for (pi = 0; pi < npols; pi++) {
2547 			if (pols[pi] != pol &&
2548 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
2549 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2550 				goto reject;
2551 			}
2552 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2553 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2554 				goto reject_error;
2555 			}
2556 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
2557 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
2558 		}
2559 		xfrm_nr = ti;
2560 		if (npols > 1) {
2561 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2562 			tpp = stp;
2563 		}
2564 
2565 		/* For each tunnel xfrm, find the first matching tmpl.
2566 		 * For each tmpl before that, find corresponding xfrm.
2567 		 * Order is _important_. Later we will implement
2568 		 * some barriers, but at the moment barriers
2569 		 * are implied between each two transformations.
2570 		 */
2571 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2572 			k = xfrm_policy_ok(tpp[i], sp, k, family);
2573 			if (k < 0) {
2574 				if (k < -1)
2575 					/* "-2 - errored_index" returned */
2576 					xerr_idx = -(2+k);
2577 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2578 				goto reject;
2579 			}
2580 		}
2581 
2582 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2583 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2584 			goto reject;
2585 		}
2586 
2587 		xfrm_pols_put(pols, npols);
2588 		return 1;
2589 	}
2590 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2591 
2592 reject:
2593 	xfrm_secpath_reject(xerr_idx, skb, &fl);
2594 reject_error:
2595 	xfrm_pols_put(pols, npols);
2596 	return 0;
2597 }
2598 EXPORT_SYMBOL(__xfrm_policy_check);
2599 
2600 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2601 {
2602 	struct net *net = dev_net(skb->dev);
2603 	struct flowi fl;
2604 	struct dst_entry *dst;
2605 	int res = 1;
2606 
2607 	if (xfrm_decode_session(skb, &fl, family) < 0) {
2608 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2609 		return 0;
2610 	}
2611 
2612 	skb_dst_force(skb);
2613 
2614 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2615 	if (IS_ERR(dst)) {
2616 		res = 0;
2617 		dst = NULL;
2618 	}
2619 	skb_dst_set(skb, dst);
2620 	return res;
2621 }
2622 EXPORT_SYMBOL(__xfrm_route_forward);
2623 
2624 /* Optimize later using cookies and generation ids. */
2625 
2626 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2627 {
2628 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2629 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2630 	 * get validated by dst_ops->check on every use.  We do this
2631 	 * because when a normal route referenced by an XFRM dst is
2632 	 * obsoleted we do not go looking around for all parent
2633 	 * referencing XFRM dsts so that we can invalidate them.  It
2634 	 * is just too much work.  Instead we make the checks here on
2635 	 * every use.  For example:
2636 	 *
2637 	 *	XFRM dst A --> IPv4 dst X
2638 	 *
2639 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
2640 	 * in this example).  If X is marked obsolete, "A" will not
2641 	 * notice.  That's what we are validating here via the
2642 	 * stale_bundle() check.
2643 	 *
2644 	 * When a policy's bundle is pruned, we dst_free() the XFRM
2645 	 * dst which causes it's ->obsolete field to be set to
2646 	 * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2647 	 * this, we want to force a new route lookup.
2648 	 */
2649 	if (dst->obsolete < 0 && !stale_bundle(dst))
2650 		return dst;
2651 
2652 	return NULL;
2653 }
2654 
2655 static int stale_bundle(struct dst_entry *dst)
2656 {
2657 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2658 }
2659 
2660 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2661 {
2662 	while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2663 		dst->dev = dev_net(dev)->loopback_dev;
2664 		dev_hold(dst->dev);
2665 		dev_put(dev);
2666 	}
2667 }
2668 EXPORT_SYMBOL(xfrm_dst_ifdown);
2669 
2670 static void xfrm_link_failure(struct sk_buff *skb)
2671 {
2672 	/* Impossible. Such dst must be popped before reaches point of failure. */
2673 }
2674 
2675 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2676 {
2677 	if (dst) {
2678 		if (dst->obsolete) {
2679 			dst_release(dst);
2680 			dst = NULL;
2681 		}
2682 	}
2683 	return dst;
2684 }
2685 
2686 void xfrm_garbage_collect(struct net *net)
2687 {
2688 	flow_cache_flush(net);
2689 }
2690 EXPORT_SYMBOL(xfrm_garbage_collect);
2691 
2692 static void xfrm_garbage_collect_deferred(struct net *net)
2693 {
2694 	flow_cache_flush_deferred(net);
2695 }
2696 
2697 static void xfrm_init_pmtu(struct dst_entry *dst)
2698 {
2699 	do {
2700 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2701 		u32 pmtu, route_mtu_cached;
2702 
2703 		pmtu = dst_mtu(dst->child);
2704 		xdst->child_mtu_cached = pmtu;
2705 
2706 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2707 
2708 		route_mtu_cached = dst_mtu(xdst->route);
2709 		xdst->route_mtu_cached = route_mtu_cached;
2710 
2711 		if (pmtu > route_mtu_cached)
2712 			pmtu = route_mtu_cached;
2713 
2714 		dst_metric_set(dst, RTAX_MTU, pmtu);
2715 	} while ((dst = dst->next));
2716 }
2717 
2718 /* Check that the bundle accepts the flow and its components are
2719  * still valid.
2720  */
2721 
2722 static int xfrm_bundle_ok(struct xfrm_dst *first)
2723 {
2724 	struct dst_entry *dst = &first->u.dst;
2725 	struct xfrm_dst *last;
2726 	u32 mtu;
2727 
2728 	if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2729 	    (dst->dev && !netif_running(dst->dev)))
2730 		return 0;
2731 
2732 	if (dst->flags & DST_XFRM_QUEUE)
2733 		return 1;
2734 
2735 	last = NULL;
2736 
2737 	do {
2738 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2739 
2740 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2741 			return 0;
2742 		if (xdst->xfrm_genid != dst->xfrm->genid)
2743 			return 0;
2744 		if (xdst->num_pols > 0 &&
2745 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2746 			return 0;
2747 
2748 		mtu = dst_mtu(dst->child);
2749 		if (xdst->child_mtu_cached != mtu) {
2750 			last = xdst;
2751 			xdst->child_mtu_cached = mtu;
2752 		}
2753 
2754 		if (!dst_check(xdst->route, xdst->route_cookie))
2755 			return 0;
2756 		mtu = dst_mtu(xdst->route);
2757 		if (xdst->route_mtu_cached != mtu) {
2758 			last = xdst;
2759 			xdst->route_mtu_cached = mtu;
2760 		}
2761 
2762 		dst = dst->child;
2763 	} while (dst->xfrm);
2764 
2765 	if (likely(!last))
2766 		return 1;
2767 
2768 	mtu = last->child_mtu_cached;
2769 	for (;;) {
2770 		dst = &last->u.dst;
2771 
2772 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2773 		if (mtu > last->route_mtu_cached)
2774 			mtu = last->route_mtu_cached;
2775 		dst_metric_set(dst, RTAX_MTU, mtu);
2776 
2777 		if (last == first)
2778 			break;
2779 
2780 		last = (struct xfrm_dst *)last->u.dst.next;
2781 		last->child_mtu_cached = mtu;
2782 	}
2783 
2784 	return 1;
2785 }
2786 
2787 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2788 {
2789 	return dst_metric_advmss(dst->path);
2790 }
2791 
2792 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2793 {
2794 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2795 
2796 	return mtu ? : dst_mtu(dst->path);
2797 }
2798 
2799 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2800 					   struct sk_buff *skb,
2801 					   const void *daddr)
2802 {
2803 	return dst->path->ops->neigh_lookup(dst, skb, daddr);
2804 }
2805 
2806 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2807 {
2808 	struct net *net;
2809 	int err = 0;
2810 	if (unlikely(afinfo == NULL))
2811 		return -EINVAL;
2812 	if (unlikely(afinfo->family >= NPROTO))
2813 		return -EAFNOSUPPORT;
2814 	spin_lock(&xfrm_policy_afinfo_lock);
2815 	if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2816 		err = -EEXIST;
2817 	else {
2818 		struct dst_ops *dst_ops = afinfo->dst_ops;
2819 		if (likely(dst_ops->kmem_cachep == NULL))
2820 			dst_ops->kmem_cachep = xfrm_dst_cache;
2821 		if (likely(dst_ops->check == NULL))
2822 			dst_ops->check = xfrm_dst_check;
2823 		if (likely(dst_ops->default_advmss == NULL))
2824 			dst_ops->default_advmss = xfrm_default_advmss;
2825 		if (likely(dst_ops->mtu == NULL))
2826 			dst_ops->mtu = xfrm_mtu;
2827 		if (likely(dst_ops->negative_advice == NULL))
2828 			dst_ops->negative_advice = xfrm_negative_advice;
2829 		if (likely(dst_ops->link_failure == NULL))
2830 			dst_ops->link_failure = xfrm_link_failure;
2831 		if (likely(dst_ops->neigh_lookup == NULL))
2832 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
2833 		if (likely(afinfo->garbage_collect == NULL))
2834 			afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2835 		rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2836 	}
2837 	spin_unlock(&xfrm_policy_afinfo_lock);
2838 
2839 	rtnl_lock();
2840 	for_each_net(net) {
2841 		struct dst_ops *xfrm_dst_ops;
2842 
2843 		switch (afinfo->family) {
2844 		case AF_INET:
2845 			xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2846 			break;
2847 #if IS_ENABLED(CONFIG_IPV6)
2848 		case AF_INET6:
2849 			xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2850 			break;
2851 #endif
2852 		default:
2853 			BUG();
2854 		}
2855 		*xfrm_dst_ops = *afinfo->dst_ops;
2856 	}
2857 	rtnl_unlock();
2858 
2859 	return err;
2860 }
2861 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2862 
2863 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2864 {
2865 	int err = 0;
2866 	if (unlikely(afinfo == NULL))
2867 		return -EINVAL;
2868 	if (unlikely(afinfo->family >= NPROTO))
2869 		return -EAFNOSUPPORT;
2870 	spin_lock(&xfrm_policy_afinfo_lock);
2871 	if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2872 		if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2873 			err = -EINVAL;
2874 		else
2875 			RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2876 					 NULL);
2877 	}
2878 	spin_unlock(&xfrm_policy_afinfo_lock);
2879 	if (!err) {
2880 		struct dst_ops *dst_ops = afinfo->dst_ops;
2881 
2882 		synchronize_rcu();
2883 
2884 		dst_ops->kmem_cachep = NULL;
2885 		dst_ops->check = NULL;
2886 		dst_ops->negative_advice = NULL;
2887 		dst_ops->link_failure = NULL;
2888 		afinfo->garbage_collect = NULL;
2889 	}
2890 	return err;
2891 }
2892 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2893 
2894 static void __net_init xfrm_dst_ops_init(struct net *net)
2895 {
2896 	struct xfrm_policy_afinfo *afinfo;
2897 
2898 	rcu_read_lock();
2899 	afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2900 	if (afinfo)
2901 		net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2902 #if IS_ENABLED(CONFIG_IPV6)
2903 	afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2904 	if (afinfo)
2905 		net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2906 #endif
2907 	rcu_read_unlock();
2908 }
2909 
2910 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2911 {
2912 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2913 
2914 	switch (event) {
2915 	case NETDEV_DOWN:
2916 		xfrm_garbage_collect(dev_net(dev));
2917 	}
2918 	return NOTIFY_DONE;
2919 }
2920 
2921 static struct notifier_block xfrm_dev_notifier = {
2922 	.notifier_call	= xfrm_dev_event,
2923 };
2924 
2925 #ifdef CONFIG_XFRM_STATISTICS
2926 static int __net_init xfrm_statistics_init(struct net *net)
2927 {
2928 	int rv;
2929 	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2930 	if (!net->mib.xfrm_statistics)
2931 		return -ENOMEM;
2932 	rv = xfrm_proc_init(net);
2933 	if (rv < 0)
2934 		free_percpu(net->mib.xfrm_statistics);
2935 	return rv;
2936 }
2937 
2938 static void xfrm_statistics_fini(struct net *net)
2939 {
2940 	xfrm_proc_fini(net);
2941 	free_percpu(net->mib.xfrm_statistics);
2942 }
2943 #else
2944 static int __net_init xfrm_statistics_init(struct net *net)
2945 {
2946 	return 0;
2947 }
2948 
2949 static void xfrm_statistics_fini(struct net *net)
2950 {
2951 }
2952 #endif
2953 
2954 static int __net_init xfrm_policy_init(struct net *net)
2955 {
2956 	unsigned int hmask, sz;
2957 	int dir;
2958 
2959 	if (net_eq(net, &init_net))
2960 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2961 					   sizeof(struct xfrm_dst),
2962 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2963 					   NULL);
2964 
2965 	hmask = 8 - 1;
2966 	sz = (hmask+1) * sizeof(struct hlist_head);
2967 
2968 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2969 	if (!net->xfrm.policy_byidx)
2970 		goto out_byidx;
2971 	net->xfrm.policy_idx_hmask = hmask;
2972 
2973 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2974 		struct xfrm_policy_hash *htab;
2975 
2976 		net->xfrm.policy_count[dir] = 0;
2977 		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2978 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2979 
2980 		htab = &net->xfrm.policy_bydst[dir];
2981 		htab->table = xfrm_hash_alloc(sz);
2982 		if (!htab->table)
2983 			goto out_bydst;
2984 		htab->hmask = hmask;
2985 		htab->dbits4 = 32;
2986 		htab->sbits4 = 32;
2987 		htab->dbits6 = 128;
2988 		htab->sbits6 = 128;
2989 	}
2990 	net->xfrm.policy_hthresh.lbits4 = 32;
2991 	net->xfrm.policy_hthresh.rbits4 = 32;
2992 	net->xfrm.policy_hthresh.lbits6 = 128;
2993 	net->xfrm.policy_hthresh.rbits6 = 128;
2994 
2995 	seqlock_init(&net->xfrm.policy_hthresh.lock);
2996 
2997 	INIT_LIST_HEAD(&net->xfrm.policy_all);
2998 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2999 	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
3000 	if (net_eq(net, &init_net))
3001 		register_netdevice_notifier(&xfrm_dev_notifier);
3002 	return 0;
3003 
3004 out_bydst:
3005 	for (dir--; dir >= 0; dir--) {
3006 		struct xfrm_policy_hash *htab;
3007 
3008 		htab = &net->xfrm.policy_bydst[dir];
3009 		xfrm_hash_free(htab->table, sz);
3010 	}
3011 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
3012 out_byidx:
3013 	return -ENOMEM;
3014 }
3015 
3016 static void xfrm_policy_fini(struct net *net)
3017 {
3018 	unsigned int sz;
3019 	int dir;
3020 
3021 	flush_work(&net->xfrm.policy_hash_work);
3022 #ifdef CONFIG_XFRM_SUB_POLICY
3023 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3024 #endif
3025 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3026 
3027 	WARN_ON(!list_empty(&net->xfrm.policy_all));
3028 
3029 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3030 		struct xfrm_policy_hash *htab;
3031 
3032 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3033 
3034 		htab = &net->xfrm.policy_bydst[dir];
3035 		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3036 		WARN_ON(!hlist_empty(htab->table));
3037 		xfrm_hash_free(htab->table, sz);
3038 	}
3039 
3040 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3041 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3042 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
3043 }
3044 
3045 static int __net_init xfrm_net_init(struct net *net)
3046 {
3047 	int rv;
3048 
3049 	rv = xfrm_statistics_init(net);
3050 	if (rv < 0)
3051 		goto out_statistics;
3052 	rv = xfrm_state_init(net);
3053 	if (rv < 0)
3054 		goto out_state;
3055 	rv = xfrm_policy_init(net);
3056 	if (rv < 0)
3057 		goto out_policy;
3058 	xfrm_dst_ops_init(net);
3059 	rv = xfrm_sysctl_init(net);
3060 	if (rv < 0)
3061 		goto out_sysctl;
3062 	rv = flow_cache_init(net);
3063 	if (rv < 0)
3064 		goto out;
3065 
3066 	/* Initialize the per-net locks here */
3067 	spin_lock_init(&net->xfrm.xfrm_state_lock);
3068 	rwlock_init(&net->xfrm.xfrm_policy_lock);
3069 	mutex_init(&net->xfrm.xfrm_cfg_mutex);
3070 
3071 	return 0;
3072 
3073 out:
3074 	xfrm_sysctl_fini(net);
3075 out_sysctl:
3076 	xfrm_policy_fini(net);
3077 out_policy:
3078 	xfrm_state_fini(net);
3079 out_state:
3080 	xfrm_statistics_fini(net);
3081 out_statistics:
3082 	return rv;
3083 }
3084 
3085 static void __net_exit xfrm_net_exit(struct net *net)
3086 {
3087 	flow_cache_fini(net);
3088 	xfrm_sysctl_fini(net);
3089 	xfrm_policy_fini(net);
3090 	xfrm_state_fini(net);
3091 	xfrm_statistics_fini(net);
3092 }
3093 
3094 static struct pernet_operations __net_initdata xfrm_net_ops = {
3095 	.init = xfrm_net_init,
3096 	.exit = xfrm_net_exit,
3097 };
3098 
3099 void __init xfrm_init(void)
3100 {
3101 	register_pernet_subsys(&xfrm_net_ops);
3102 	xfrm_input_init();
3103 }
3104 
3105 #ifdef CONFIG_AUDITSYSCALL
3106 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3107 					 struct audit_buffer *audit_buf)
3108 {
3109 	struct xfrm_sec_ctx *ctx = xp->security;
3110 	struct xfrm_selector *sel = &xp->selector;
3111 
3112 	if (ctx)
3113 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3114 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3115 
3116 	switch (sel->family) {
3117 	case AF_INET:
3118 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3119 		if (sel->prefixlen_s != 32)
3120 			audit_log_format(audit_buf, " src_prefixlen=%d",
3121 					 sel->prefixlen_s);
3122 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3123 		if (sel->prefixlen_d != 32)
3124 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3125 					 sel->prefixlen_d);
3126 		break;
3127 	case AF_INET6:
3128 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3129 		if (sel->prefixlen_s != 128)
3130 			audit_log_format(audit_buf, " src_prefixlen=%d",
3131 					 sel->prefixlen_s);
3132 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3133 		if (sel->prefixlen_d != 128)
3134 			audit_log_format(audit_buf, " dst_prefixlen=%d",
3135 					 sel->prefixlen_d);
3136 		break;
3137 	}
3138 }
3139 
3140 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3141 {
3142 	struct audit_buffer *audit_buf;
3143 
3144 	audit_buf = xfrm_audit_start("SPD-add");
3145 	if (audit_buf == NULL)
3146 		return;
3147 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3148 	audit_log_format(audit_buf, " res=%u", result);
3149 	xfrm_audit_common_policyinfo(xp, audit_buf);
3150 	audit_log_end(audit_buf);
3151 }
3152 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3153 
3154 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3155 			      bool task_valid)
3156 {
3157 	struct audit_buffer *audit_buf;
3158 
3159 	audit_buf = xfrm_audit_start("SPD-delete");
3160 	if (audit_buf == NULL)
3161 		return;
3162 	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3163 	audit_log_format(audit_buf, " res=%u", result);
3164 	xfrm_audit_common_policyinfo(xp, audit_buf);
3165 	audit_log_end(audit_buf);
3166 }
3167 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3168 #endif
3169 
3170 #ifdef CONFIG_XFRM_MIGRATE
3171 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3172 					const struct xfrm_selector *sel_tgt)
3173 {
3174 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3175 		if (sel_tgt->family == sel_cmp->family &&
3176 		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3177 				    sel_cmp->family) &&
3178 		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3179 				    sel_cmp->family) &&
3180 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3181 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3182 			return true;
3183 		}
3184 	} else {
3185 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3186 			return true;
3187 		}
3188 	}
3189 	return false;
3190 }
3191 
3192 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3193 						    u8 dir, u8 type, struct net *net)
3194 {
3195 	struct xfrm_policy *pol, *ret = NULL;
3196 	struct hlist_head *chain;
3197 	u32 priority = ~0U;
3198 
3199 	read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3200 	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3201 	hlist_for_each_entry(pol, chain, bydst) {
3202 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3203 		    pol->type == type) {
3204 			ret = pol;
3205 			priority = ret->priority;
3206 			break;
3207 		}
3208 	}
3209 	chain = &net->xfrm.policy_inexact[dir];
3210 	hlist_for_each_entry(pol, chain, bydst) {
3211 		if ((pol->priority >= priority) && ret)
3212 			break;
3213 
3214 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3215 		    pol->type == type) {
3216 			ret = pol;
3217 			break;
3218 		}
3219 	}
3220 
3221 	xfrm_pol_hold(ret);
3222 
3223 	read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3224 
3225 	return ret;
3226 }
3227 
3228 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3229 {
3230 	int match = 0;
3231 
3232 	if (t->mode == m->mode && t->id.proto == m->proto &&
3233 	    (m->reqid == 0 || t->reqid == m->reqid)) {
3234 		switch (t->mode) {
3235 		case XFRM_MODE_TUNNEL:
3236 		case XFRM_MODE_BEET:
3237 			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3238 					    m->old_family) &&
3239 			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
3240 					    m->old_family)) {
3241 				match = 1;
3242 			}
3243 			break;
3244 		case XFRM_MODE_TRANSPORT:
3245 			/* in case of transport mode, template does not store
3246 			   any IP addresses, hence we just compare mode and
3247 			   protocol */
3248 			match = 1;
3249 			break;
3250 		default:
3251 			break;
3252 		}
3253 	}
3254 	return match;
3255 }
3256 
3257 /* update endpoint address(es) of template(s) */
3258 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3259 			       struct xfrm_migrate *m, int num_migrate)
3260 {
3261 	struct xfrm_migrate *mp;
3262 	int i, j, n = 0;
3263 
3264 	write_lock_bh(&pol->lock);
3265 	if (unlikely(pol->walk.dead)) {
3266 		/* target policy has been deleted */
3267 		write_unlock_bh(&pol->lock);
3268 		return -ENOENT;
3269 	}
3270 
3271 	for (i = 0; i < pol->xfrm_nr; i++) {
3272 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3273 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3274 				continue;
3275 			n++;
3276 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3277 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3278 				continue;
3279 			/* update endpoints */
3280 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3281 			       sizeof(pol->xfrm_vec[i].id.daddr));
3282 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3283 			       sizeof(pol->xfrm_vec[i].saddr));
3284 			pol->xfrm_vec[i].encap_family = mp->new_family;
3285 			/* flush bundles */
3286 			atomic_inc(&pol->genid);
3287 		}
3288 	}
3289 
3290 	write_unlock_bh(&pol->lock);
3291 
3292 	if (!n)
3293 		return -ENODATA;
3294 
3295 	return 0;
3296 }
3297 
3298 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3299 {
3300 	int i, j;
3301 
3302 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3303 		return -EINVAL;
3304 
3305 	for (i = 0; i < num_migrate; i++) {
3306 		if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3307 				    m[i].old_family) &&
3308 		    xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3309 				    m[i].old_family))
3310 			return -EINVAL;
3311 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3312 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3313 			return -EINVAL;
3314 
3315 		/* check if there is any duplicated entry */
3316 		for (j = i + 1; j < num_migrate; j++) {
3317 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3318 				    sizeof(m[i].old_daddr)) &&
3319 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3320 				    sizeof(m[i].old_saddr)) &&
3321 			    m[i].proto == m[j].proto &&
3322 			    m[i].mode == m[j].mode &&
3323 			    m[i].reqid == m[j].reqid &&
3324 			    m[i].old_family == m[j].old_family)
3325 				return -EINVAL;
3326 		}
3327 	}
3328 
3329 	return 0;
3330 }
3331 
3332 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3333 		 struct xfrm_migrate *m, int num_migrate,
3334 		 struct xfrm_kmaddress *k, struct net *net)
3335 {
3336 	int i, err, nx_cur = 0, nx_new = 0;
3337 	struct xfrm_policy *pol = NULL;
3338 	struct xfrm_state *x, *xc;
3339 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3340 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3341 	struct xfrm_migrate *mp;
3342 
3343 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3344 		goto out;
3345 
3346 	/* Stage 1 - find policy */
3347 	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3348 		err = -ENOENT;
3349 		goto out;
3350 	}
3351 
3352 	/* Stage 2 - find and update state(s) */
3353 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3354 		if ((x = xfrm_migrate_state_find(mp, net))) {
3355 			x_cur[nx_cur] = x;
3356 			nx_cur++;
3357 			if ((xc = xfrm_state_migrate(x, mp))) {
3358 				x_new[nx_new] = xc;
3359 				nx_new++;
3360 			} else {
3361 				err = -ENODATA;
3362 				goto restore_state;
3363 			}
3364 		}
3365 	}
3366 
3367 	/* Stage 3 - update policy */
3368 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3369 		goto restore_state;
3370 
3371 	/* Stage 4 - delete old state(s) */
3372 	if (nx_cur) {
3373 		xfrm_states_put(x_cur, nx_cur);
3374 		xfrm_states_delete(x_cur, nx_cur);
3375 	}
3376 
3377 	/* Stage 5 - announce */
3378 	km_migrate(sel, dir, type, m, num_migrate, k);
3379 
3380 	xfrm_pol_put(pol);
3381 
3382 	return 0;
3383 out:
3384 	return err;
3385 
3386 restore_state:
3387 	if (pol)
3388 		xfrm_pol_put(pol);
3389 	if (nx_cur)
3390 		xfrm_states_put(x_cur, nx_cur);
3391 	if (nx_new)
3392 		xfrm_states_delete(x_new, nx_new);
3393 
3394 	return err;
3395 }
3396 EXPORT_SYMBOL(xfrm_migrate);
3397 #endif
3398