xref: /linux/net/xfrm/xfrm_policy.c (revision 25aee3debe0464f6c680173041fa3de30ec9ff54)
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_MUTEX(xfrm_cfg_mutex);
39 EXPORT_SYMBOL(xfrm_cfg_mutex);
40 
41 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
42 static struct dst_entry *xfrm_policy_sk_bundles;
43 static DEFINE_RWLOCK(xfrm_policy_lock);
44 
45 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
46 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
47 
48 static struct kmem_cache *xfrm_dst_cache __read_mostly;
49 
50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
52 static void xfrm_init_pmtu(struct dst_entry *dst);
53 static int stale_bundle(struct dst_entry *dst);
54 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
55 
56 
57 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
58 						int dir);
59 
60 static inline bool
61 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
62 {
63 	const struct flowi4 *fl4 = &fl->u.ip4;
64 
65 	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
66 		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
67 		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
68 		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
69 		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
70 		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
71 }
72 
73 static inline bool
74 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
75 {
76 	const struct flowi6 *fl6 = &fl->u.ip6;
77 
78 	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
79 		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
80 		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
81 		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
82 		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
83 		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
84 }
85 
86 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
87 			 unsigned short family)
88 {
89 	switch (family) {
90 	case AF_INET:
91 		return __xfrm4_selector_match(sel, fl);
92 	case AF_INET6:
93 		return __xfrm6_selector_match(sel, fl);
94 	}
95 	return false;
96 }
97 
98 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
99 						  const xfrm_address_t *saddr,
100 						  const xfrm_address_t *daddr,
101 						  int family)
102 {
103 	struct xfrm_policy_afinfo *afinfo;
104 	struct dst_entry *dst;
105 
106 	afinfo = xfrm_policy_get_afinfo(family);
107 	if (unlikely(afinfo == NULL))
108 		return ERR_PTR(-EAFNOSUPPORT);
109 
110 	dst = afinfo->dst_lookup(net, tos, saddr, daddr);
111 
112 	xfrm_policy_put_afinfo(afinfo);
113 
114 	return dst;
115 }
116 
117 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
118 						xfrm_address_t *prev_saddr,
119 						xfrm_address_t *prev_daddr,
120 						int family)
121 {
122 	struct net *net = xs_net(x);
123 	xfrm_address_t *saddr = &x->props.saddr;
124 	xfrm_address_t *daddr = &x->id.daddr;
125 	struct dst_entry *dst;
126 
127 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
128 		saddr = x->coaddr;
129 		daddr = prev_daddr;
130 	}
131 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
132 		saddr = prev_saddr;
133 		daddr = x->coaddr;
134 	}
135 
136 	dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
137 
138 	if (!IS_ERR(dst)) {
139 		if (prev_saddr != saddr)
140 			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
141 		if (prev_daddr != daddr)
142 			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
143 	}
144 
145 	return dst;
146 }
147 
148 static inline unsigned long make_jiffies(long secs)
149 {
150 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
151 		return MAX_SCHEDULE_TIMEOUT-1;
152 	else
153 		return secs*HZ;
154 }
155 
156 static void xfrm_policy_timer(unsigned long data)
157 {
158 	struct xfrm_policy *xp = (struct xfrm_policy*)data;
159 	unsigned long now = get_seconds();
160 	long next = LONG_MAX;
161 	int warn = 0;
162 	int dir;
163 
164 	read_lock(&xp->lock);
165 
166 	if (unlikely(xp->walk.dead))
167 		goto out;
168 
169 	dir = xfrm_policy_id2dir(xp->index);
170 
171 	if (xp->lft.hard_add_expires_seconds) {
172 		long tmo = xp->lft.hard_add_expires_seconds +
173 			xp->curlft.add_time - now;
174 		if (tmo <= 0)
175 			goto expired;
176 		if (tmo < next)
177 			next = tmo;
178 	}
179 	if (xp->lft.hard_use_expires_seconds) {
180 		long tmo = xp->lft.hard_use_expires_seconds +
181 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
182 		if (tmo <= 0)
183 			goto expired;
184 		if (tmo < next)
185 			next = tmo;
186 	}
187 	if (xp->lft.soft_add_expires_seconds) {
188 		long tmo = xp->lft.soft_add_expires_seconds +
189 			xp->curlft.add_time - now;
190 		if (tmo <= 0) {
191 			warn = 1;
192 			tmo = XFRM_KM_TIMEOUT;
193 		}
194 		if (tmo < next)
195 			next = tmo;
196 	}
197 	if (xp->lft.soft_use_expires_seconds) {
198 		long tmo = xp->lft.soft_use_expires_seconds +
199 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
200 		if (tmo <= 0) {
201 			warn = 1;
202 			tmo = XFRM_KM_TIMEOUT;
203 		}
204 		if (tmo < next)
205 			next = tmo;
206 	}
207 
208 	if (warn)
209 		km_policy_expired(xp, dir, 0, 0);
210 	if (next != LONG_MAX &&
211 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
212 		xfrm_pol_hold(xp);
213 
214 out:
215 	read_unlock(&xp->lock);
216 	xfrm_pol_put(xp);
217 	return;
218 
219 expired:
220 	read_unlock(&xp->lock);
221 	if (!xfrm_policy_delete(xp, dir))
222 		km_policy_expired(xp, dir, 1, 0);
223 	xfrm_pol_put(xp);
224 }
225 
226 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
227 {
228 	struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
229 
230 	if (unlikely(pol->walk.dead))
231 		flo = NULL;
232 	else
233 		xfrm_pol_hold(pol);
234 
235 	return flo;
236 }
237 
238 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
239 {
240 	struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
241 
242 	return !pol->walk.dead;
243 }
244 
245 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
246 {
247 	xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
248 }
249 
250 static const struct flow_cache_ops xfrm_policy_fc_ops = {
251 	.get = xfrm_policy_flo_get,
252 	.check = xfrm_policy_flo_check,
253 	.delete = xfrm_policy_flo_delete,
254 };
255 
256 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
257  * SPD calls.
258  */
259 
260 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
261 {
262 	struct xfrm_policy *policy;
263 
264 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
265 
266 	if (policy) {
267 		write_pnet(&policy->xp_net, net);
268 		INIT_LIST_HEAD(&policy->walk.all);
269 		INIT_HLIST_NODE(&policy->bydst);
270 		INIT_HLIST_NODE(&policy->byidx);
271 		rwlock_init(&policy->lock);
272 		atomic_set(&policy->refcnt, 1);
273 		setup_timer(&policy->timer, xfrm_policy_timer,
274 				(unsigned long)policy);
275 		policy->flo.ops = &xfrm_policy_fc_ops;
276 	}
277 	return policy;
278 }
279 EXPORT_SYMBOL(xfrm_policy_alloc);
280 
281 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
282 
283 void xfrm_policy_destroy(struct xfrm_policy *policy)
284 {
285 	BUG_ON(!policy->walk.dead);
286 
287 	if (del_timer(&policy->timer))
288 		BUG();
289 
290 	security_xfrm_policy_free(policy->security);
291 	kfree(policy);
292 }
293 EXPORT_SYMBOL(xfrm_policy_destroy);
294 
295 /* Rule must be locked. Release descentant resources, announce
296  * entry dead. The rule must be unlinked from lists to the moment.
297  */
298 
299 static void xfrm_policy_kill(struct xfrm_policy *policy)
300 {
301 	policy->walk.dead = 1;
302 
303 	atomic_inc(&policy->genid);
304 
305 	if (del_timer(&policy->timer))
306 		xfrm_pol_put(policy);
307 
308 	xfrm_pol_put(policy);
309 }
310 
311 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
312 
313 static inline unsigned int idx_hash(struct net *net, u32 index)
314 {
315 	return __idx_hash(index, net->xfrm.policy_idx_hmask);
316 }
317 
318 static struct hlist_head *policy_hash_bysel(struct net *net,
319 					    const struct xfrm_selector *sel,
320 					    unsigned short family, int dir)
321 {
322 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
323 	unsigned int hash = __sel_hash(sel, family, hmask);
324 
325 	return (hash == hmask + 1 ?
326 		&net->xfrm.policy_inexact[dir] :
327 		net->xfrm.policy_bydst[dir].table + hash);
328 }
329 
330 static struct hlist_head *policy_hash_direct(struct net *net,
331 					     const xfrm_address_t *daddr,
332 					     const xfrm_address_t *saddr,
333 					     unsigned short family, int dir)
334 {
335 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
336 	unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
337 
338 	return net->xfrm.policy_bydst[dir].table + hash;
339 }
340 
341 static void xfrm_dst_hash_transfer(struct hlist_head *list,
342 				   struct hlist_head *ndsttable,
343 				   unsigned int nhashmask)
344 {
345 	struct hlist_node *entry, *tmp, *entry0 = NULL;
346 	struct xfrm_policy *pol;
347 	unsigned int h0 = 0;
348 
349 redo:
350 	hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
351 		unsigned int h;
352 
353 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
354 				pol->family, nhashmask);
355 		if (!entry0) {
356 			hlist_del(entry);
357 			hlist_add_head(&pol->bydst, ndsttable+h);
358 			h0 = h;
359 		} else {
360 			if (h != h0)
361 				continue;
362 			hlist_del(entry);
363 			hlist_add_after(entry0, &pol->bydst);
364 		}
365 		entry0 = entry;
366 	}
367 	if (!hlist_empty(list)) {
368 		entry0 = NULL;
369 		goto redo;
370 	}
371 }
372 
373 static void xfrm_idx_hash_transfer(struct hlist_head *list,
374 				   struct hlist_head *nidxtable,
375 				   unsigned int nhashmask)
376 {
377 	struct hlist_node *entry, *tmp;
378 	struct xfrm_policy *pol;
379 
380 	hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
381 		unsigned int h;
382 
383 		h = __idx_hash(pol->index, nhashmask);
384 		hlist_add_head(&pol->byidx, nidxtable+h);
385 	}
386 }
387 
388 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
389 {
390 	return ((old_hmask + 1) << 1) - 1;
391 }
392 
393 static void xfrm_bydst_resize(struct net *net, int dir)
394 {
395 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
396 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
397 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
398 	struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
399 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
400 	int i;
401 
402 	if (!ndst)
403 		return;
404 
405 	write_lock_bh(&xfrm_policy_lock);
406 
407 	for (i = hmask; i >= 0; i--)
408 		xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
409 
410 	net->xfrm.policy_bydst[dir].table = ndst;
411 	net->xfrm.policy_bydst[dir].hmask = nhashmask;
412 
413 	write_unlock_bh(&xfrm_policy_lock);
414 
415 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
416 }
417 
418 static void xfrm_byidx_resize(struct net *net, int total)
419 {
420 	unsigned int hmask = net->xfrm.policy_idx_hmask;
421 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
422 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
423 	struct hlist_head *oidx = net->xfrm.policy_byidx;
424 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
425 	int i;
426 
427 	if (!nidx)
428 		return;
429 
430 	write_lock_bh(&xfrm_policy_lock);
431 
432 	for (i = hmask; i >= 0; i--)
433 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
434 
435 	net->xfrm.policy_byidx = nidx;
436 	net->xfrm.policy_idx_hmask = nhashmask;
437 
438 	write_unlock_bh(&xfrm_policy_lock);
439 
440 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
441 }
442 
443 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
444 {
445 	unsigned int cnt = net->xfrm.policy_count[dir];
446 	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
447 
448 	if (total)
449 		*total += cnt;
450 
451 	if ((hmask + 1) < xfrm_policy_hashmax &&
452 	    cnt > hmask)
453 		return 1;
454 
455 	return 0;
456 }
457 
458 static inline int xfrm_byidx_should_resize(struct net *net, int total)
459 {
460 	unsigned int hmask = net->xfrm.policy_idx_hmask;
461 
462 	if ((hmask + 1) < xfrm_policy_hashmax &&
463 	    total > hmask)
464 		return 1;
465 
466 	return 0;
467 }
468 
469 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
470 {
471 	read_lock_bh(&xfrm_policy_lock);
472 	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
473 	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
474 	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
475 	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
476 	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
477 	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
478 	si->spdhcnt = net->xfrm.policy_idx_hmask;
479 	si->spdhmcnt = xfrm_policy_hashmax;
480 	read_unlock_bh(&xfrm_policy_lock);
481 }
482 EXPORT_SYMBOL(xfrm_spd_getinfo);
483 
484 static DEFINE_MUTEX(hash_resize_mutex);
485 static void xfrm_hash_resize(struct work_struct *work)
486 {
487 	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
488 	int dir, total;
489 
490 	mutex_lock(&hash_resize_mutex);
491 
492 	total = 0;
493 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
494 		if (xfrm_bydst_should_resize(net, dir, &total))
495 			xfrm_bydst_resize(net, dir);
496 	}
497 	if (xfrm_byidx_should_resize(net, total))
498 		xfrm_byidx_resize(net, total);
499 
500 	mutex_unlock(&hash_resize_mutex);
501 }
502 
503 /* Generate new index... KAME seems to generate them ordered by cost
504  * of an absolute inpredictability of ordering of rules. This will not pass. */
505 static u32 xfrm_gen_index(struct net *net, int dir)
506 {
507 	static u32 idx_generator;
508 
509 	for (;;) {
510 		struct hlist_node *entry;
511 		struct hlist_head *list;
512 		struct xfrm_policy *p;
513 		u32 idx;
514 		int found;
515 
516 		idx = (idx_generator | dir);
517 		idx_generator += 8;
518 		if (idx == 0)
519 			idx = 8;
520 		list = net->xfrm.policy_byidx + idx_hash(net, idx);
521 		found = 0;
522 		hlist_for_each_entry(p, entry, list, byidx) {
523 			if (p->index == idx) {
524 				found = 1;
525 				break;
526 			}
527 		}
528 		if (!found)
529 			return idx;
530 	}
531 }
532 
533 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
534 {
535 	u32 *p1 = (u32 *) s1;
536 	u32 *p2 = (u32 *) s2;
537 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
538 	int i;
539 
540 	for (i = 0; i < len; i++) {
541 		if (p1[i] != p2[i])
542 			return 1;
543 	}
544 
545 	return 0;
546 }
547 
548 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
549 {
550 	struct net *net = xp_net(policy);
551 	struct xfrm_policy *pol;
552 	struct xfrm_policy *delpol;
553 	struct hlist_head *chain;
554 	struct hlist_node *entry, *newpos;
555 	u32 mark = policy->mark.v & policy->mark.m;
556 
557 	write_lock_bh(&xfrm_policy_lock);
558 	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
559 	delpol = NULL;
560 	newpos = NULL;
561 	hlist_for_each_entry(pol, entry, chain, bydst) {
562 		if (pol->type == policy->type &&
563 		    !selector_cmp(&pol->selector, &policy->selector) &&
564 		    (mark & pol->mark.m) == pol->mark.v &&
565 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
566 		    !WARN_ON(delpol)) {
567 			if (excl) {
568 				write_unlock_bh(&xfrm_policy_lock);
569 				return -EEXIST;
570 			}
571 			delpol = pol;
572 			if (policy->priority > pol->priority)
573 				continue;
574 		} else if (policy->priority >= pol->priority) {
575 			newpos = &pol->bydst;
576 			continue;
577 		}
578 		if (delpol)
579 			break;
580 	}
581 	if (newpos)
582 		hlist_add_after(newpos, &policy->bydst);
583 	else
584 		hlist_add_head(&policy->bydst, chain);
585 	xfrm_pol_hold(policy);
586 	net->xfrm.policy_count[dir]++;
587 	atomic_inc(&flow_cache_genid);
588 	if (delpol)
589 		__xfrm_policy_unlink(delpol, dir);
590 	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
591 	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
592 	policy->curlft.add_time = get_seconds();
593 	policy->curlft.use_time = 0;
594 	if (!mod_timer(&policy->timer, jiffies + HZ))
595 		xfrm_pol_hold(policy);
596 	list_add(&policy->walk.all, &net->xfrm.policy_all);
597 	write_unlock_bh(&xfrm_policy_lock);
598 
599 	if (delpol)
600 		xfrm_policy_kill(delpol);
601 	else if (xfrm_bydst_should_resize(net, dir, NULL))
602 		schedule_work(&net->xfrm.policy_hash_work);
603 
604 	return 0;
605 }
606 EXPORT_SYMBOL(xfrm_policy_insert);
607 
608 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
609 					  int dir, struct xfrm_selector *sel,
610 					  struct xfrm_sec_ctx *ctx, int delete,
611 					  int *err)
612 {
613 	struct xfrm_policy *pol, *ret;
614 	struct hlist_head *chain;
615 	struct hlist_node *entry;
616 
617 	*err = 0;
618 	write_lock_bh(&xfrm_policy_lock);
619 	chain = policy_hash_bysel(net, sel, sel->family, dir);
620 	ret = NULL;
621 	hlist_for_each_entry(pol, entry, chain, bydst) {
622 		if (pol->type == type &&
623 		    (mark & pol->mark.m) == pol->mark.v &&
624 		    !selector_cmp(sel, &pol->selector) &&
625 		    xfrm_sec_ctx_match(ctx, pol->security)) {
626 			xfrm_pol_hold(pol);
627 			if (delete) {
628 				*err = security_xfrm_policy_delete(
629 								pol->security);
630 				if (*err) {
631 					write_unlock_bh(&xfrm_policy_lock);
632 					return pol;
633 				}
634 				__xfrm_policy_unlink(pol, dir);
635 			}
636 			ret = pol;
637 			break;
638 		}
639 	}
640 	write_unlock_bh(&xfrm_policy_lock);
641 
642 	if (ret && delete)
643 		xfrm_policy_kill(ret);
644 	return ret;
645 }
646 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
647 
648 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
649 				     int dir, u32 id, int delete, int *err)
650 {
651 	struct xfrm_policy *pol, *ret;
652 	struct hlist_head *chain;
653 	struct hlist_node *entry;
654 
655 	*err = -ENOENT;
656 	if (xfrm_policy_id2dir(id) != dir)
657 		return NULL;
658 
659 	*err = 0;
660 	write_lock_bh(&xfrm_policy_lock);
661 	chain = net->xfrm.policy_byidx + idx_hash(net, id);
662 	ret = NULL;
663 	hlist_for_each_entry(pol, entry, chain, byidx) {
664 		if (pol->type == type && pol->index == id &&
665 		    (mark & pol->mark.m) == pol->mark.v) {
666 			xfrm_pol_hold(pol);
667 			if (delete) {
668 				*err = security_xfrm_policy_delete(
669 								pol->security);
670 				if (*err) {
671 					write_unlock_bh(&xfrm_policy_lock);
672 					return pol;
673 				}
674 				__xfrm_policy_unlink(pol, dir);
675 			}
676 			ret = pol;
677 			break;
678 		}
679 	}
680 	write_unlock_bh(&xfrm_policy_lock);
681 
682 	if (ret && delete)
683 		xfrm_policy_kill(ret);
684 	return ret;
685 }
686 EXPORT_SYMBOL(xfrm_policy_byid);
687 
688 #ifdef CONFIG_SECURITY_NETWORK_XFRM
689 static inline int
690 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
691 {
692 	int dir, err = 0;
693 
694 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
695 		struct xfrm_policy *pol;
696 		struct hlist_node *entry;
697 		int i;
698 
699 		hlist_for_each_entry(pol, entry,
700 				     &net->xfrm.policy_inexact[dir], bydst) {
701 			if (pol->type != type)
702 				continue;
703 			err = security_xfrm_policy_delete(pol->security);
704 			if (err) {
705 				xfrm_audit_policy_delete(pol, 0,
706 							 audit_info->loginuid,
707 							 audit_info->sessionid,
708 							 audit_info->secid);
709 				return err;
710 			}
711 		}
712 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
713 			hlist_for_each_entry(pol, entry,
714 					     net->xfrm.policy_bydst[dir].table + i,
715 					     bydst) {
716 				if (pol->type != type)
717 					continue;
718 				err = security_xfrm_policy_delete(
719 								pol->security);
720 				if (err) {
721 					xfrm_audit_policy_delete(pol, 0,
722 							audit_info->loginuid,
723 							audit_info->sessionid,
724 							audit_info->secid);
725 					return err;
726 				}
727 			}
728 		}
729 	}
730 	return err;
731 }
732 #else
733 static inline int
734 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
735 {
736 	return 0;
737 }
738 #endif
739 
740 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
741 {
742 	int dir, err = 0, cnt = 0;
743 
744 	write_lock_bh(&xfrm_policy_lock);
745 
746 	err = xfrm_policy_flush_secctx_check(net, type, audit_info);
747 	if (err)
748 		goto out;
749 
750 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
751 		struct xfrm_policy *pol;
752 		struct hlist_node *entry;
753 		int i;
754 
755 	again1:
756 		hlist_for_each_entry(pol, entry,
757 				     &net->xfrm.policy_inexact[dir], bydst) {
758 			if (pol->type != type)
759 				continue;
760 			__xfrm_policy_unlink(pol, dir);
761 			write_unlock_bh(&xfrm_policy_lock);
762 			cnt++;
763 
764 			xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
765 						 audit_info->sessionid,
766 						 audit_info->secid);
767 
768 			xfrm_policy_kill(pol);
769 
770 			write_lock_bh(&xfrm_policy_lock);
771 			goto again1;
772 		}
773 
774 		for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
775 	again2:
776 			hlist_for_each_entry(pol, entry,
777 					     net->xfrm.policy_bydst[dir].table + i,
778 					     bydst) {
779 				if (pol->type != type)
780 					continue;
781 				__xfrm_policy_unlink(pol, dir);
782 				write_unlock_bh(&xfrm_policy_lock);
783 				cnt++;
784 
785 				xfrm_audit_policy_delete(pol, 1,
786 							 audit_info->loginuid,
787 							 audit_info->sessionid,
788 							 audit_info->secid);
789 				xfrm_policy_kill(pol);
790 
791 				write_lock_bh(&xfrm_policy_lock);
792 				goto again2;
793 			}
794 		}
795 
796 	}
797 	if (!cnt)
798 		err = -ESRCH;
799 out:
800 	write_unlock_bh(&xfrm_policy_lock);
801 	return err;
802 }
803 EXPORT_SYMBOL(xfrm_policy_flush);
804 
805 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
806 		     int (*func)(struct xfrm_policy *, int, int, void*),
807 		     void *data)
808 {
809 	struct xfrm_policy *pol;
810 	struct xfrm_policy_walk_entry *x;
811 	int error = 0;
812 
813 	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
814 	    walk->type != XFRM_POLICY_TYPE_ANY)
815 		return -EINVAL;
816 
817 	if (list_empty(&walk->walk.all) && walk->seq != 0)
818 		return 0;
819 
820 	write_lock_bh(&xfrm_policy_lock);
821 	if (list_empty(&walk->walk.all))
822 		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
823 	else
824 		x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
825 	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
826 		if (x->dead)
827 			continue;
828 		pol = container_of(x, struct xfrm_policy, walk);
829 		if (walk->type != XFRM_POLICY_TYPE_ANY &&
830 		    walk->type != pol->type)
831 			continue;
832 		error = func(pol, xfrm_policy_id2dir(pol->index),
833 			     walk->seq, data);
834 		if (error) {
835 			list_move_tail(&walk->walk.all, &x->all);
836 			goto out;
837 		}
838 		walk->seq++;
839 	}
840 	if (walk->seq == 0) {
841 		error = -ENOENT;
842 		goto out;
843 	}
844 	list_del_init(&walk->walk.all);
845 out:
846 	write_unlock_bh(&xfrm_policy_lock);
847 	return error;
848 }
849 EXPORT_SYMBOL(xfrm_policy_walk);
850 
851 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
852 {
853 	INIT_LIST_HEAD(&walk->walk.all);
854 	walk->walk.dead = 1;
855 	walk->type = type;
856 	walk->seq = 0;
857 }
858 EXPORT_SYMBOL(xfrm_policy_walk_init);
859 
860 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
861 {
862 	if (list_empty(&walk->walk.all))
863 		return;
864 
865 	write_lock_bh(&xfrm_policy_lock);
866 	list_del(&walk->walk.all);
867 	write_unlock_bh(&xfrm_policy_lock);
868 }
869 EXPORT_SYMBOL(xfrm_policy_walk_done);
870 
871 /*
872  * Find policy to apply to this flow.
873  *
874  * Returns 0 if policy found, else an -errno.
875  */
876 static int xfrm_policy_match(const struct xfrm_policy *pol,
877 			     const struct flowi *fl,
878 			     u8 type, u16 family, int dir)
879 {
880 	const struct xfrm_selector *sel = &pol->selector;
881 	int ret = -ESRCH;
882 	bool match;
883 
884 	if (pol->family != family ||
885 	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
886 	    pol->type != type)
887 		return ret;
888 
889 	match = xfrm_selector_match(sel, fl, family);
890 	if (match)
891 		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
892 						  dir);
893 
894 	return ret;
895 }
896 
897 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
898 						     const struct flowi *fl,
899 						     u16 family, u8 dir)
900 {
901 	int err;
902 	struct xfrm_policy *pol, *ret;
903 	const xfrm_address_t *daddr, *saddr;
904 	struct hlist_node *entry;
905 	struct hlist_head *chain;
906 	u32 priority = ~0U;
907 
908 	daddr = xfrm_flowi_daddr(fl, family);
909 	saddr = xfrm_flowi_saddr(fl, family);
910 	if (unlikely(!daddr || !saddr))
911 		return NULL;
912 
913 	read_lock_bh(&xfrm_policy_lock);
914 	chain = policy_hash_direct(net, daddr, saddr, family, dir);
915 	ret = NULL;
916 	hlist_for_each_entry(pol, entry, chain, bydst) {
917 		err = xfrm_policy_match(pol, fl, type, family, dir);
918 		if (err) {
919 			if (err == -ESRCH)
920 				continue;
921 			else {
922 				ret = ERR_PTR(err);
923 				goto fail;
924 			}
925 		} else {
926 			ret = pol;
927 			priority = ret->priority;
928 			break;
929 		}
930 	}
931 	chain = &net->xfrm.policy_inexact[dir];
932 	hlist_for_each_entry(pol, entry, chain, bydst) {
933 		err = xfrm_policy_match(pol, fl, type, family, dir);
934 		if (err) {
935 			if (err == -ESRCH)
936 				continue;
937 			else {
938 				ret = ERR_PTR(err);
939 				goto fail;
940 			}
941 		} else if (pol->priority < priority) {
942 			ret = pol;
943 			break;
944 		}
945 	}
946 	if (ret)
947 		xfrm_pol_hold(ret);
948 fail:
949 	read_unlock_bh(&xfrm_policy_lock);
950 
951 	return ret;
952 }
953 
954 static struct xfrm_policy *
955 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
956 {
957 #ifdef CONFIG_XFRM_SUB_POLICY
958 	struct xfrm_policy *pol;
959 
960 	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
961 	if (pol != NULL)
962 		return pol;
963 #endif
964 	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
965 }
966 
967 static struct flow_cache_object *
968 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
969 		   u8 dir, struct flow_cache_object *old_obj, void *ctx)
970 {
971 	struct xfrm_policy *pol;
972 
973 	if (old_obj)
974 		xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
975 
976 	pol = __xfrm_policy_lookup(net, fl, family, dir);
977 	if (IS_ERR_OR_NULL(pol))
978 		return ERR_CAST(pol);
979 
980 	/* Resolver returns two references:
981 	 * one for cache and one for caller of flow_cache_lookup() */
982 	xfrm_pol_hold(pol);
983 
984 	return &pol->flo;
985 }
986 
987 static inline int policy_to_flow_dir(int dir)
988 {
989 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
990 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
991 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
992 		return dir;
993 	switch (dir) {
994 	default:
995 	case XFRM_POLICY_IN:
996 		return FLOW_DIR_IN;
997 	case XFRM_POLICY_OUT:
998 		return FLOW_DIR_OUT;
999 	case XFRM_POLICY_FWD:
1000 		return FLOW_DIR_FWD;
1001 	}
1002 }
1003 
1004 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1005 						 const struct flowi *fl)
1006 {
1007 	struct xfrm_policy *pol;
1008 
1009 	read_lock_bh(&xfrm_policy_lock);
1010 	if ((pol = sk->sk_policy[dir]) != NULL) {
1011 		bool match = xfrm_selector_match(&pol->selector, fl,
1012 						 sk->sk_family);
1013 		int err = 0;
1014 
1015 		if (match) {
1016 			if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1017 				pol = NULL;
1018 				goto out;
1019 			}
1020 			err = security_xfrm_policy_lookup(pol->security,
1021 						      fl->flowi_secid,
1022 						      policy_to_flow_dir(dir));
1023 			if (!err)
1024 				xfrm_pol_hold(pol);
1025 			else if (err == -ESRCH)
1026 				pol = NULL;
1027 			else
1028 				pol = ERR_PTR(err);
1029 		} else
1030 			pol = NULL;
1031 	}
1032 out:
1033 	read_unlock_bh(&xfrm_policy_lock);
1034 	return pol;
1035 }
1036 
1037 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1038 {
1039 	struct net *net = xp_net(pol);
1040 	struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1041 						     pol->family, dir);
1042 
1043 	list_add(&pol->walk.all, &net->xfrm.policy_all);
1044 	hlist_add_head(&pol->bydst, chain);
1045 	hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1046 	net->xfrm.policy_count[dir]++;
1047 	xfrm_pol_hold(pol);
1048 
1049 	if (xfrm_bydst_should_resize(net, dir, NULL))
1050 		schedule_work(&net->xfrm.policy_hash_work);
1051 }
1052 
1053 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1054 						int dir)
1055 {
1056 	struct net *net = xp_net(pol);
1057 
1058 	if (hlist_unhashed(&pol->bydst))
1059 		return NULL;
1060 
1061 	hlist_del(&pol->bydst);
1062 	hlist_del(&pol->byidx);
1063 	list_del(&pol->walk.all);
1064 	net->xfrm.policy_count[dir]--;
1065 
1066 	return pol;
1067 }
1068 
1069 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1070 {
1071 	write_lock_bh(&xfrm_policy_lock);
1072 	pol = __xfrm_policy_unlink(pol, dir);
1073 	write_unlock_bh(&xfrm_policy_lock);
1074 	if (pol) {
1075 		xfrm_policy_kill(pol);
1076 		return 0;
1077 	}
1078 	return -ENOENT;
1079 }
1080 EXPORT_SYMBOL(xfrm_policy_delete);
1081 
1082 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1083 {
1084 	struct net *net = xp_net(pol);
1085 	struct xfrm_policy *old_pol;
1086 
1087 #ifdef CONFIG_XFRM_SUB_POLICY
1088 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1089 		return -EINVAL;
1090 #endif
1091 
1092 	write_lock_bh(&xfrm_policy_lock);
1093 	old_pol = sk->sk_policy[dir];
1094 	sk->sk_policy[dir] = pol;
1095 	if (pol) {
1096 		pol->curlft.add_time = get_seconds();
1097 		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1098 		__xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1099 	}
1100 	if (old_pol)
1101 		/* Unlinking succeeds always. This is the only function
1102 		 * allowed to delete or replace socket policy.
1103 		 */
1104 		__xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1105 	write_unlock_bh(&xfrm_policy_lock);
1106 
1107 	if (old_pol) {
1108 		xfrm_policy_kill(old_pol);
1109 	}
1110 	return 0;
1111 }
1112 
1113 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1114 {
1115 	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1116 
1117 	if (newp) {
1118 		newp->selector = old->selector;
1119 		if (security_xfrm_policy_clone(old->security,
1120 					       &newp->security)) {
1121 			kfree(newp);
1122 			return NULL;  /* ENOMEM */
1123 		}
1124 		newp->lft = old->lft;
1125 		newp->curlft = old->curlft;
1126 		newp->mark = old->mark;
1127 		newp->action = old->action;
1128 		newp->flags = old->flags;
1129 		newp->xfrm_nr = old->xfrm_nr;
1130 		newp->index = old->index;
1131 		newp->type = old->type;
1132 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1133 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1134 		write_lock_bh(&xfrm_policy_lock);
1135 		__xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1136 		write_unlock_bh(&xfrm_policy_lock);
1137 		xfrm_pol_put(newp);
1138 	}
1139 	return newp;
1140 }
1141 
1142 int __xfrm_sk_clone_policy(struct sock *sk)
1143 {
1144 	struct xfrm_policy *p0 = sk->sk_policy[0],
1145 			   *p1 = sk->sk_policy[1];
1146 
1147 	sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1148 	if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1149 		return -ENOMEM;
1150 	if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1151 		return -ENOMEM;
1152 	return 0;
1153 }
1154 
1155 static int
1156 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1157 	       unsigned short family)
1158 {
1159 	int err;
1160 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1161 
1162 	if (unlikely(afinfo == NULL))
1163 		return -EINVAL;
1164 	err = afinfo->get_saddr(net, local, remote);
1165 	xfrm_policy_put_afinfo(afinfo);
1166 	return err;
1167 }
1168 
1169 /* Resolve list of templates for the flow, given policy. */
1170 
1171 static int
1172 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1173 		      struct xfrm_state **xfrm, unsigned short family)
1174 {
1175 	struct net *net = xp_net(policy);
1176 	int nx;
1177 	int i, error;
1178 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1179 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1180 	xfrm_address_t tmp;
1181 
1182 	for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1183 		struct xfrm_state *x;
1184 		xfrm_address_t *remote = daddr;
1185 		xfrm_address_t *local  = saddr;
1186 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1187 
1188 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1189 		    tmpl->mode == XFRM_MODE_BEET) {
1190 			remote = &tmpl->id.daddr;
1191 			local = &tmpl->saddr;
1192 			if (xfrm_addr_any(local, tmpl->encap_family)) {
1193 				error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1194 				if (error)
1195 					goto fail;
1196 				local = &tmp;
1197 			}
1198 		}
1199 
1200 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1201 
1202 		if (x && x->km.state == XFRM_STATE_VALID) {
1203 			xfrm[nx++] = x;
1204 			daddr = remote;
1205 			saddr = local;
1206 			continue;
1207 		}
1208 		if (x) {
1209 			error = (x->km.state == XFRM_STATE_ERROR ?
1210 				 -EINVAL : -EAGAIN);
1211 			xfrm_state_put(x);
1212 		}
1213 		else if (error == -ESRCH)
1214 			error = -EAGAIN;
1215 
1216 		if (!tmpl->optional)
1217 			goto fail;
1218 	}
1219 	return nx;
1220 
1221 fail:
1222 	for (nx--; nx>=0; nx--)
1223 		xfrm_state_put(xfrm[nx]);
1224 	return error;
1225 }
1226 
1227 static int
1228 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1229 		  struct xfrm_state **xfrm, unsigned short family)
1230 {
1231 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1232 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1233 	int cnx = 0;
1234 	int error;
1235 	int ret;
1236 	int i;
1237 
1238 	for (i = 0; i < npols; i++) {
1239 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1240 			error = -ENOBUFS;
1241 			goto fail;
1242 		}
1243 
1244 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1245 		if (ret < 0) {
1246 			error = ret;
1247 			goto fail;
1248 		} else
1249 			cnx += ret;
1250 	}
1251 
1252 	/* found states are sorted for outbound processing */
1253 	if (npols > 1)
1254 		xfrm_state_sort(xfrm, tpp, cnx, family);
1255 
1256 	return cnx;
1257 
1258  fail:
1259 	for (cnx--; cnx>=0; cnx--)
1260 		xfrm_state_put(tpp[cnx]);
1261 	return error;
1262 
1263 }
1264 
1265 /* Check that the bundle accepts the flow and its components are
1266  * still valid.
1267  */
1268 
1269 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1270 {
1271 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1272 	int tos;
1273 
1274 	if (!afinfo)
1275 		return -EINVAL;
1276 
1277 	tos = afinfo->get_tos(fl);
1278 
1279 	xfrm_policy_put_afinfo(afinfo);
1280 
1281 	return tos;
1282 }
1283 
1284 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1285 {
1286 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1287 	struct dst_entry *dst = &xdst->u.dst;
1288 
1289 	if (xdst->route == NULL) {
1290 		/* Dummy bundle - if it has xfrms we were not
1291 		 * able to build bundle as template resolution failed.
1292 		 * It means we need to try again resolving. */
1293 		if (xdst->num_xfrms > 0)
1294 			return NULL;
1295 	} else {
1296 		/* Real bundle */
1297 		if (stale_bundle(dst))
1298 			return NULL;
1299 	}
1300 
1301 	dst_hold(dst);
1302 	return flo;
1303 }
1304 
1305 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1306 {
1307 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1308 	struct dst_entry *dst = &xdst->u.dst;
1309 
1310 	if (!xdst->route)
1311 		return 0;
1312 	if (stale_bundle(dst))
1313 		return 0;
1314 
1315 	return 1;
1316 }
1317 
1318 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1319 {
1320 	struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1321 	struct dst_entry *dst = &xdst->u.dst;
1322 
1323 	dst_free(dst);
1324 }
1325 
1326 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1327 	.get = xfrm_bundle_flo_get,
1328 	.check = xfrm_bundle_flo_check,
1329 	.delete = xfrm_bundle_flo_delete,
1330 };
1331 
1332 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1333 {
1334 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1335 	struct dst_ops *dst_ops;
1336 	struct xfrm_dst *xdst;
1337 
1338 	if (!afinfo)
1339 		return ERR_PTR(-EINVAL);
1340 
1341 	switch (family) {
1342 	case AF_INET:
1343 		dst_ops = &net->xfrm.xfrm4_dst_ops;
1344 		break;
1345 #if IS_ENABLED(CONFIG_IPV6)
1346 	case AF_INET6:
1347 		dst_ops = &net->xfrm.xfrm6_dst_ops;
1348 		break;
1349 #endif
1350 	default:
1351 		BUG();
1352 	}
1353 	xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1354 
1355 	if (likely(xdst)) {
1356 		struct dst_entry *dst = &xdst->u.dst;
1357 
1358 		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1359 		xdst->flo.ops = &xfrm_bundle_fc_ops;
1360 	} else
1361 		xdst = ERR_PTR(-ENOBUFS);
1362 
1363 	xfrm_policy_put_afinfo(afinfo);
1364 
1365 	return xdst;
1366 }
1367 
1368 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1369 				 int nfheader_len)
1370 {
1371 	struct xfrm_policy_afinfo *afinfo =
1372 		xfrm_policy_get_afinfo(dst->ops->family);
1373 	int err;
1374 
1375 	if (!afinfo)
1376 		return -EINVAL;
1377 
1378 	err = afinfo->init_path(path, dst, nfheader_len);
1379 
1380 	xfrm_policy_put_afinfo(afinfo);
1381 
1382 	return err;
1383 }
1384 
1385 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1386 				const struct flowi *fl)
1387 {
1388 	struct xfrm_policy_afinfo *afinfo =
1389 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1390 	int err;
1391 
1392 	if (!afinfo)
1393 		return -EINVAL;
1394 
1395 	err = afinfo->fill_dst(xdst, dev, fl);
1396 
1397 	xfrm_policy_put_afinfo(afinfo);
1398 
1399 	return err;
1400 }
1401 
1402 
1403 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1404  * all the metrics... Shortly, bundle a bundle.
1405  */
1406 
1407 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1408 					    struct xfrm_state **xfrm, int nx,
1409 					    const struct flowi *fl,
1410 					    struct dst_entry *dst)
1411 {
1412 	struct net *net = xp_net(policy);
1413 	unsigned long now = jiffies;
1414 	struct net_device *dev;
1415 	struct xfrm_mode *inner_mode;
1416 	struct dst_entry *dst_prev = NULL;
1417 	struct dst_entry *dst0 = NULL;
1418 	int i = 0;
1419 	int err;
1420 	int header_len = 0;
1421 	int nfheader_len = 0;
1422 	int trailer_len = 0;
1423 	int tos;
1424 	int family = policy->selector.family;
1425 	xfrm_address_t saddr, daddr;
1426 
1427 	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1428 
1429 	tos = xfrm_get_tos(fl, family);
1430 	err = tos;
1431 	if (tos < 0)
1432 		goto put_states;
1433 
1434 	dst_hold(dst);
1435 
1436 	for (; i < nx; i++) {
1437 		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1438 		struct dst_entry *dst1 = &xdst->u.dst;
1439 
1440 		err = PTR_ERR(xdst);
1441 		if (IS_ERR(xdst)) {
1442 			dst_release(dst);
1443 			goto put_states;
1444 		}
1445 
1446 		if (xfrm[i]->sel.family == AF_UNSPEC) {
1447 			inner_mode = xfrm_ip2inner_mode(xfrm[i],
1448 							xfrm_af2proto(family));
1449 			if (!inner_mode) {
1450 				err = -EAFNOSUPPORT;
1451 				dst_release(dst);
1452 				goto put_states;
1453 			}
1454 		} else
1455 			inner_mode = xfrm[i]->inner_mode;
1456 
1457 		if (!dst_prev)
1458 			dst0 = dst1;
1459 		else {
1460 			dst_prev->child = dst_clone(dst1);
1461 			dst1->flags |= DST_NOHASH;
1462 		}
1463 
1464 		xdst->route = dst;
1465 		dst_copy_metrics(dst1, dst);
1466 
1467 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1468 			family = xfrm[i]->props.family;
1469 			dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1470 					      family);
1471 			err = PTR_ERR(dst);
1472 			if (IS_ERR(dst))
1473 				goto put_states;
1474 		} else
1475 			dst_hold(dst);
1476 
1477 		dst1->xfrm = xfrm[i];
1478 		xdst->xfrm_genid = xfrm[i]->genid;
1479 
1480 		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1481 		dst1->flags |= DST_HOST;
1482 		dst1->lastuse = now;
1483 
1484 		dst1->input = dst_discard;
1485 		dst1->output = inner_mode->afinfo->output;
1486 
1487 		dst1->next = dst_prev;
1488 		dst_prev = dst1;
1489 
1490 		header_len += xfrm[i]->props.header_len;
1491 		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1492 			nfheader_len += xfrm[i]->props.header_len;
1493 		trailer_len += xfrm[i]->props.trailer_len;
1494 	}
1495 
1496 	dst_prev->child = dst;
1497 	dst0->path = dst;
1498 
1499 	err = -ENODEV;
1500 	dev = dst->dev;
1501 	if (!dev)
1502 		goto free_dst;
1503 
1504 	xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1505 	xfrm_init_pmtu(dst_prev);
1506 
1507 	for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1508 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1509 
1510 		err = xfrm_fill_dst(xdst, dev, fl);
1511 		if (err)
1512 			goto free_dst;
1513 
1514 		dst_prev->header_len = header_len;
1515 		dst_prev->trailer_len = trailer_len;
1516 		header_len -= xdst->u.dst.xfrm->props.header_len;
1517 		trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1518 	}
1519 
1520 out:
1521 	return dst0;
1522 
1523 put_states:
1524 	for (; i < nx; i++)
1525 		xfrm_state_put(xfrm[i]);
1526 free_dst:
1527 	if (dst0)
1528 		dst_free(dst0);
1529 	dst0 = ERR_PTR(err);
1530 	goto out;
1531 }
1532 
1533 static int inline
1534 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1535 {
1536 	if (!*target) {
1537 		*target = kmalloc(size, GFP_ATOMIC);
1538 		if (!*target)
1539 			return -ENOMEM;
1540 	}
1541 	memcpy(*target, src, size);
1542 	return 0;
1543 }
1544 
1545 static int inline
1546 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1547 {
1548 #ifdef CONFIG_XFRM_SUB_POLICY
1549 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1550 	return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1551 				   sel, sizeof(*sel));
1552 #else
1553 	return 0;
1554 #endif
1555 }
1556 
1557 static int inline
1558 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1559 {
1560 #ifdef CONFIG_XFRM_SUB_POLICY
1561 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1562 	return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1563 #else
1564 	return 0;
1565 #endif
1566 }
1567 
1568 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1569 				struct xfrm_policy **pols,
1570 				int *num_pols, int *num_xfrms)
1571 {
1572 	int i;
1573 
1574 	if (*num_pols == 0 || !pols[0]) {
1575 		*num_pols = 0;
1576 		*num_xfrms = 0;
1577 		return 0;
1578 	}
1579 	if (IS_ERR(pols[0]))
1580 		return PTR_ERR(pols[0]);
1581 
1582 	*num_xfrms = pols[0]->xfrm_nr;
1583 
1584 #ifdef CONFIG_XFRM_SUB_POLICY
1585 	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1586 	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1587 		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1588 						    XFRM_POLICY_TYPE_MAIN,
1589 						    fl, family,
1590 						    XFRM_POLICY_OUT);
1591 		if (pols[1]) {
1592 			if (IS_ERR(pols[1])) {
1593 				xfrm_pols_put(pols, *num_pols);
1594 				return PTR_ERR(pols[1]);
1595 			}
1596 			(*num_pols) ++;
1597 			(*num_xfrms) += pols[1]->xfrm_nr;
1598 		}
1599 	}
1600 #endif
1601 	for (i = 0; i < *num_pols; i++) {
1602 		if (pols[i]->action != XFRM_POLICY_ALLOW) {
1603 			*num_xfrms = -1;
1604 			break;
1605 		}
1606 	}
1607 
1608 	return 0;
1609 
1610 }
1611 
1612 static struct xfrm_dst *
1613 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1614 			       const struct flowi *fl, u16 family,
1615 			       struct dst_entry *dst_orig)
1616 {
1617 	struct net *net = xp_net(pols[0]);
1618 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1619 	struct dst_entry *dst;
1620 	struct xfrm_dst *xdst;
1621 	int err;
1622 
1623 	/* Try to instantiate a bundle */
1624 	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1625 	if (err <= 0) {
1626 		if (err != 0 && err != -EAGAIN)
1627 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1628 		return ERR_PTR(err);
1629 	}
1630 
1631 	dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1632 	if (IS_ERR(dst)) {
1633 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1634 		return ERR_CAST(dst);
1635 	}
1636 
1637 	xdst = (struct xfrm_dst *)dst;
1638 	xdst->num_xfrms = err;
1639 	if (num_pols > 1)
1640 		err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1641 	else
1642 		err = xfrm_dst_update_origin(dst, fl);
1643 	if (unlikely(err)) {
1644 		dst_free(dst);
1645 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1646 		return ERR_PTR(err);
1647 	}
1648 
1649 	xdst->num_pols = num_pols;
1650 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1651 	xdst->policy_genid = atomic_read(&pols[0]->genid);
1652 
1653 	return xdst;
1654 }
1655 
1656 static struct flow_cache_object *
1657 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1658 		   struct flow_cache_object *oldflo, void *ctx)
1659 {
1660 	struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1661 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1662 	struct xfrm_dst *xdst, *new_xdst;
1663 	int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1664 
1665 	/* Check if the policies from old bundle are usable */
1666 	xdst = NULL;
1667 	if (oldflo) {
1668 		xdst = container_of(oldflo, struct xfrm_dst, flo);
1669 		num_pols = xdst->num_pols;
1670 		num_xfrms = xdst->num_xfrms;
1671 		pol_dead = 0;
1672 		for (i = 0; i < num_pols; i++) {
1673 			pols[i] = xdst->pols[i];
1674 			pol_dead |= pols[i]->walk.dead;
1675 		}
1676 		if (pol_dead) {
1677 			dst_free(&xdst->u.dst);
1678 			xdst = NULL;
1679 			num_pols = 0;
1680 			num_xfrms = 0;
1681 			oldflo = NULL;
1682 		}
1683 	}
1684 
1685 	/* Resolve policies to use if we couldn't get them from
1686 	 * previous cache entry */
1687 	if (xdst == NULL) {
1688 		num_pols = 1;
1689 		pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1690 		err = xfrm_expand_policies(fl, family, pols,
1691 					   &num_pols, &num_xfrms);
1692 		if (err < 0)
1693 			goto inc_error;
1694 		if (num_pols == 0)
1695 			return NULL;
1696 		if (num_xfrms <= 0)
1697 			goto make_dummy_bundle;
1698 	}
1699 
1700 	new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1701 	if (IS_ERR(new_xdst)) {
1702 		err = PTR_ERR(new_xdst);
1703 		if (err != -EAGAIN)
1704 			goto error;
1705 		if (oldflo == NULL)
1706 			goto make_dummy_bundle;
1707 		dst_hold(&xdst->u.dst);
1708 		return oldflo;
1709 	} else if (new_xdst == NULL) {
1710 		num_xfrms = 0;
1711 		if (oldflo == NULL)
1712 			goto make_dummy_bundle;
1713 		xdst->num_xfrms = 0;
1714 		dst_hold(&xdst->u.dst);
1715 		return oldflo;
1716 	}
1717 
1718 	/* Kill the previous bundle */
1719 	if (xdst) {
1720 		/* The policies were stolen for newly generated bundle */
1721 		xdst->num_pols = 0;
1722 		dst_free(&xdst->u.dst);
1723 	}
1724 
1725 	/* Flow cache does not have reference, it dst_free()'s,
1726 	 * but we do need to return one reference for original caller */
1727 	dst_hold(&new_xdst->u.dst);
1728 	return &new_xdst->flo;
1729 
1730 make_dummy_bundle:
1731 	/* We found policies, but there's no bundles to instantiate:
1732 	 * either because the policy blocks, has no transformations or
1733 	 * we could not build template (no xfrm_states).*/
1734 	xdst = xfrm_alloc_dst(net, family);
1735 	if (IS_ERR(xdst)) {
1736 		xfrm_pols_put(pols, num_pols);
1737 		return ERR_CAST(xdst);
1738 	}
1739 	xdst->num_pols = num_pols;
1740 	xdst->num_xfrms = num_xfrms;
1741 	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1742 
1743 	dst_hold(&xdst->u.dst);
1744 	return &xdst->flo;
1745 
1746 inc_error:
1747 	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1748 error:
1749 	if (xdst != NULL)
1750 		dst_free(&xdst->u.dst);
1751 	else
1752 		xfrm_pols_put(pols, num_pols);
1753 	return ERR_PTR(err);
1754 }
1755 
1756 static struct dst_entry *make_blackhole(struct net *net, u16 family,
1757 					struct dst_entry *dst_orig)
1758 {
1759 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1760 	struct dst_entry *ret;
1761 
1762 	if (!afinfo) {
1763 		dst_release(dst_orig);
1764 		ret = ERR_PTR(-EINVAL);
1765 	} else {
1766 		ret = afinfo->blackhole_route(net, dst_orig);
1767 	}
1768 	xfrm_policy_put_afinfo(afinfo);
1769 
1770 	return ret;
1771 }
1772 
1773 /* Main function: finds/creates a bundle for given flow.
1774  *
1775  * At the moment we eat a raw IP route. Mostly to speed up lookups
1776  * on interfaces with disabled IPsec.
1777  */
1778 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
1779 			      const struct flowi *fl,
1780 			      struct sock *sk, int flags)
1781 {
1782 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1783 	struct flow_cache_object *flo;
1784 	struct xfrm_dst *xdst;
1785 	struct dst_entry *dst, *route;
1786 	u16 family = dst_orig->ops->family;
1787 	u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1788 	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1789 
1790 restart:
1791 	dst = NULL;
1792 	xdst = NULL;
1793 	route = NULL;
1794 
1795 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1796 		num_pols = 1;
1797 		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1798 		err = xfrm_expand_policies(fl, family, pols,
1799 					   &num_pols, &num_xfrms);
1800 		if (err < 0)
1801 			goto dropdst;
1802 
1803 		if (num_pols) {
1804 			if (num_xfrms <= 0) {
1805 				drop_pols = num_pols;
1806 				goto no_transform;
1807 			}
1808 
1809 			xdst = xfrm_resolve_and_create_bundle(
1810 					pols, num_pols, fl,
1811 					family, dst_orig);
1812 			if (IS_ERR(xdst)) {
1813 				xfrm_pols_put(pols, num_pols);
1814 				err = PTR_ERR(xdst);
1815 				goto dropdst;
1816 			} else if (xdst == NULL) {
1817 				num_xfrms = 0;
1818 				drop_pols = num_pols;
1819 				goto no_transform;
1820 			}
1821 
1822 			dst_hold(&xdst->u.dst);
1823 
1824 			spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1825 			xdst->u.dst.next = xfrm_policy_sk_bundles;
1826 			xfrm_policy_sk_bundles = &xdst->u.dst;
1827 			spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1828 
1829 			route = xdst->route;
1830 		}
1831 	}
1832 
1833 	if (xdst == NULL) {
1834 		/* To accelerate a bit...  */
1835 		if ((dst_orig->flags & DST_NOXFRM) ||
1836 		    !net->xfrm.policy_count[XFRM_POLICY_OUT])
1837 			goto nopol;
1838 
1839 		flo = flow_cache_lookup(net, fl, family, dir,
1840 					xfrm_bundle_lookup, dst_orig);
1841 		if (flo == NULL)
1842 			goto nopol;
1843 		if (IS_ERR(flo)) {
1844 			err = PTR_ERR(flo);
1845 			goto dropdst;
1846 		}
1847 		xdst = container_of(flo, struct xfrm_dst, flo);
1848 
1849 		num_pols = xdst->num_pols;
1850 		num_xfrms = xdst->num_xfrms;
1851 		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1852 		route = xdst->route;
1853 	}
1854 
1855 	dst = &xdst->u.dst;
1856 	if (route == NULL && num_xfrms > 0) {
1857 		/* The only case when xfrm_bundle_lookup() returns a
1858 		 * bundle with null route, is when the template could
1859 		 * not be resolved. It means policies are there, but
1860 		 * bundle could not be created, since we don't yet
1861 		 * have the xfrm_state's. We need to wait for KM to
1862 		 * negotiate new SA's or bail out with error.*/
1863 		if (net->xfrm.sysctl_larval_drop) {
1864 			/* EREMOTE tells the caller to generate
1865 			 * a one-shot blackhole route. */
1866 			dst_release(dst);
1867 			xfrm_pols_put(pols, drop_pols);
1868 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1869 
1870 			return make_blackhole(net, family, dst_orig);
1871 		}
1872 		if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
1873 			DECLARE_WAITQUEUE(wait, current);
1874 
1875 			add_wait_queue(&net->xfrm.km_waitq, &wait);
1876 			set_current_state(TASK_INTERRUPTIBLE);
1877 			schedule();
1878 			set_current_state(TASK_RUNNING);
1879 			remove_wait_queue(&net->xfrm.km_waitq, &wait);
1880 
1881 			if (!signal_pending(current)) {
1882 				dst_release(dst);
1883 				goto restart;
1884 			}
1885 
1886 			err = -ERESTART;
1887 		} else
1888 			err = -EAGAIN;
1889 
1890 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1891 		goto error;
1892 	}
1893 
1894 no_transform:
1895 	if (num_pols == 0)
1896 		goto nopol;
1897 
1898 	if ((flags & XFRM_LOOKUP_ICMP) &&
1899 	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1900 		err = -ENOENT;
1901 		goto error;
1902 	}
1903 
1904 	for (i = 0; i < num_pols; i++)
1905 		pols[i]->curlft.use_time = get_seconds();
1906 
1907 	if (num_xfrms < 0) {
1908 		/* Prohibit the flow */
1909 		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1910 		err = -EPERM;
1911 		goto error;
1912 	} else if (num_xfrms > 0) {
1913 		/* Flow transformed */
1914 		dst_release(dst_orig);
1915 	} else {
1916 		/* Flow passes untransformed */
1917 		dst_release(dst);
1918 		dst = dst_orig;
1919 	}
1920 ok:
1921 	xfrm_pols_put(pols, drop_pols);
1922 	if (dst && dst->xfrm &&
1923 	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
1924 		dst->flags |= DST_XFRM_TUNNEL;
1925 	return dst;
1926 
1927 nopol:
1928 	if (!(flags & XFRM_LOOKUP_ICMP)) {
1929 		dst = dst_orig;
1930 		goto ok;
1931 	}
1932 	err = -ENOENT;
1933 error:
1934 	dst_release(dst);
1935 dropdst:
1936 	dst_release(dst_orig);
1937 	xfrm_pols_put(pols, drop_pols);
1938 	return ERR_PTR(err);
1939 }
1940 EXPORT_SYMBOL(xfrm_lookup);
1941 
1942 static inline int
1943 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
1944 {
1945 	struct xfrm_state *x;
1946 
1947 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1948 		return 0;
1949 	x = skb->sp->xvec[idx];
1950 	if (!x->type->reject)
1951 		return 0;
1952 	return x->type->reject(x, skb, fl);
1953 }
1954 
1955 /* When skb is transformed back to its "native" form, we have to
1956  * check policy restrictions. At the moment we make this in maximally
1957  * stupid way. Shame on me. :-) Of course, connected sockets must
1958  * have policy cached at them.
1959  */
1960 
1961 static inline int
1962 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
1963 	      unsigned short family)
1964 {
1965 	if (xfrm_state_kern(x))
1966 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1967 	return	x->id.proto == tmpl->id.proto &&
1968 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1969 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1970 		x->props.mode == tmpl->mode &&
1971 		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1972 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1973 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
1974 		  xfrm_state_addr_cmp(tmpl, x, family));
1975 }
1976 
1977 /*
1978  * 0 or more than 0 is returned when validation is succeeded (either bypass
1979  * because of optional transport mode, or next index of the mathced secpath
1980  * state with the template.
1981  * -1 is returned when no matching template is found.
1982  * Otherwise "-2 - errored_index" is returned.
1983  */
1984 static inline int
1985 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
1986 	       unsigned short family)
1987 {
1988 	int idx = start;
1989 
1990 	if (tmpl->optional) {
1991 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
1992 			return start;
1993 	} else
1994 		start = -1;
1995 	for (; idx < sp->len; idx++) {
1996 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1997 			return ++idx;
1998 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1999 			if (start == -1)
2000 				start = -2-idx;
2001 			break;
2002 		}
2003 	}
2004 	return start;
2005 }
2006 
2007 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2008 			  unsigned int family, int reverse)
2009 {
2010 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2011 	int err;
2012 
2013 	if (unlikely(afinfo == NULL))
2014 		return -EAFNOSUPPORT;
2015 
2016 	afinfo->decode_session(skb, fl, reverse);
2017 	err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2018 	xfrm_policy_put_afinfo(afinfo);
2019 	return err;
2020 }
2021 EXPORT_SYMBOL(__xfrm_decode_session);
2022 
2023 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2024 {
2025 	for (; k < sp->len; k++) {
2026 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2027 			*idxp = k;
2028 			return 1;
2029 		}
2030 	}
2031 
2032 	return 0;
2033 }
2034 
2035 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2036 			unsigned short family)
2037 {
2038 	struct net *net = dev_net(skb->dev);
2039 	struct xfrm_policy *pol;
2040 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2041 	int npols = 0;
2042 	int xfrm_nr;
2043 	int pi;
2044 	int reverse;
2045 	struct flowi fl;
2046 	u8 fl_dir;
2047 	int xerr_idx = -1;
2048 
2049 	reverse = dir & ~XFRM_POLICY_MASK;
2050 	dir &= XFRM_POLICY_MASK;
2051 	fl_dir = policy_to_flow_dir(dir);
2052 
2053 	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2054 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2055 		return 0;
2056 	}
2057 
2058 	nf_nat_decode_session(skb, &fl, family);
2059 
2060 	/* First, check used SA against their selectors. */
2061 	if (skb->sp) {
2062 		int i;
2063 
2064 		for (i=skb->sp->len-1; i>=0; i--) {
2065 			struct xfrm_state *x = skb->sp->xvec[i];
2066 			if (!xfrm_selector_match(&x->sel, &fl, family)) {
2067 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2068 				return 0;
2069 			}
2070 		}
2071 	}
2072 
2073 	pol = NULL;
2074 	if (sk && sk->sk_policy[dir]) {
2075 		pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2076 		if (IS_ERR(pol)) {
2077 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2078 			return 0;
2079 		}
2080 	}
2081 
2082 	if (!pol) {
2083 		struct flow_cache_object *flo;
2084 
2085 		flo = flow_cache_lookup(net, &fl, family, fl_dir,
2086 					xfrm_policy_lookup, NULL);
2087 		if (IS_ERR_OR_NULL(flo))
2088 			pol = ERR_CAST(flo);
2089 		else
2090 			pol = container_of(flo, struct xfrm_policy, flo);
2091 	}
2092 
2093 	if (IS_ERR(pol)) {
2094 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2095 		return 0;
2096 	}
2097 
2098 	if (!pol) {
2099 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2100 			xfrm_secpath_reject(xerr_idx, skb, &fl);
2101 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2102 			return 0;
2103 		}
2104 		return 1;
2105 	}
2106 
2107 	pol->curlft.use_time = get_seconds();
2108 
2109 	pols[0] = pol;
2110 	npols ++;
2111 #ifdef CONFIG_XFRM_SUB_POLICY
2112 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2113 		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2114 						    &fl, family,
2115 						    XFRM_POLICY_IN);
2116 		if (pols[1]) {
2117 			if (IS_ERR(pols[1])) {
2118 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2119 				return 0;
2120 			}
2121 			pols[1]->curlft.use_time = get_seconds();
2122 			npols ++;
2123 		}
2124 	}
2125 #endif
2126 
2127 	if (pol->action == XFRM_POLICY_ALLOW) {
2128 		struct sec_path *sp;
2129 		static struct sec_path dummy;
2130 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2131 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2132 		struct xfrm_tmpl **tpp = tp;
2133 		int ti = 0;
2134 		int i, k;
2135 
2136 		if ((sp = skb->sp) == NULL)
2137 			sp = &dummy;
2138 
2139 		for (pi = 0; pi < npols; pi++) {
2140 			if (pols[pi] != pol &&
2141 			    pols[pi]->action != XFRM_POLICY_ALLOW) {
2142 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2143 				goto reject;
2144 			}
2145 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2146 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2147 				goto reject_error;
2148 			}
2149 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
2150 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
2151 		}
2152 		xfrm_nr = ti;
2153 		if (npols > 1) {
2154 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2155 			tpp = stp;
2156 		}
2157 
2158 		/* For each tunnel xfrm, find the first matching tmpl.
2159 		 * For each tmpl before that, find corresponding xfrm.
2160 		 * Order is _important_. Later we will implement
2161 		 * some barriers, but at the moment barriers
2162 		 * are implied between each two transformations.
2163 		 */
2164 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2165 			k = xfrm_policy_ok(tpp[i], sp, k, family);
2166 			if (k < 0) {
2167 				if (k < -1)
2168 					/* "-2 - errored_index" returned */
2169 					xerr_idx = -(2+k);
2170 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2171 				goto reject;
2172 			}
2173 		}
2174 
2175 		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2176 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2177 			goto reject;
2178 		}
2179 
2180 		xfrm_pols_put(pols, npols);
2181 		return 1;
2182 	}
2183 	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2184 
2185 reject:
2186 	xfrm_secpath_reject(xerr_idx, skb, &fl);
2187 reject_error:
2188 	xfrm_pols_put(pols, npols);
2189 	return 0;
2190 }
2191 EXPORT_SYMBOL(__xfrm_policy_check);
2192 
2193 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2194 {
2195 	struct net *net = dev_net(skb->dev);
2196 	struct flowi fl;
2197 	struct dst_entry *dst;
2198 	int res = 1;
2199 
2200 	if (xfrm_decode_session(skb, &fl, family) < 0) {
2201 		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2202 		return 0;
2203 	}
2204 
2205 	skb_dst_force(skb);
2206 
2207 	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2208 	if (IS_ERR(dst)) {
2209 		res = 0;
2210 		dst = NULL;
2211 	}
2212 	skb_dst_set(skb, dst);
2213 	return res;
2214 }
2215 EXPORT_SYMBOL(__xfrm_route_forward);
2216 
2217 /* Optimize later using cookies and generation ids. */
2218 
2219 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2220 {
2221 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2222 	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2223 	 * get validated by dst_ops->check on every use.  We do this
2224 	 * because when a normal route referenced by an XFRM dst is
2225 	 * obsoleted we do not go looking around for all parent
2226 	 * referencing XFRM dsts so that we can invalidate them.  It
2227 	 * is just too much work.  Instead we make the checks here on
2228 	 * every use.  For example:
2229 	 *
2230 	 *	XFRM dst A --> IPv4 dst X
2231 	 *
2232 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
2233 	 * in this example).  If X is marked obsolete, "A" will not
2234 	 * notice.  That's what we are validating here via the
2235 	 * stale_bundle() check.
2236 	 *
2237 	 * When a policy's bundle is pruned, we dst_free() the XFRM
2238 	 * dst which causes it's ->obsolete field to be set to
2239 	 * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2240 	 * this, we want to force a new route lookup.
2241 	 */
2242 	if (dst->obsolete < 0 && !stale_bundle(dst))
2243 		return dst;
2244 
2245 	return NULL;
2246 }
2247 
2248 static int stale_bundle(struct dst_entry *dst)
2249 {
2250 	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2251 }
2252 
2253 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2254 {
2255 	while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2256 		dst->dev = dev_net(dev)->loopback_dev;
2257 		dev_hold(dst->dev);
2258 		dev_put(dev);
2259 	}
2260 }
2261 EXPORT_SYMBOL(xfrm_dst_ifdown);
2262 
2263 static void xfrm_link_failure(struct sk_buff *skb)
2264 {
2265 	/* Impossible. Such dst must be popped before reaches point of failure. */
2266 }
2267 
2268 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2269 {
2270 	if (dst) {
2271 		if (dst->obsolete) {
2272 			dst_release(dst);
2273 			dst = NULL;
2274 		}
2275 	}
2276 	return dst;
2277 }
2278 
2279 static void __xfrm_garbage_collect(struct net *net)
2280 {
2281 	struct dst_entry *head, *next;
2282 
2283 	spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2284 	head = xfrm_policy_sk_bundles;
2285 	xfrm_policy_sk_bundles = NULL;
2286 	spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2287 
2288 	while (head) {
2289 		next = head->next;
2290 		dst_free(head);
2291 		head = next;
2292 	}
2293 }
2294 
2295 static void xfrm_garbage_collect(struct net *net)
2296 {
2297 	flow_cache_flush();
2298 	__xfrm_garbage_collect(net);
2299 }
2300 
2301 static void xfrm_garbage_collect_deferred(struct net *net)
2302 {
2303 	flow_cache_flush_deferred();
2304 	__xfrm_garbage_collect(net);
2305 }
2306 
2307 static void xfrm_init_pmtu(struct dst_entry *dst)
2308 {
2309 	do {
2310 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2311 		u32 pmtu, route_mtu_cached;
2312 
2313 		pmtu = dst_mtu(dst->child);
2314 		xdst->child_mtu_cached = pmtu;
2315 
2316 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2317 
2318 		route_mtu_cached = dst_mtu(xdst->route);
2319 		xdst->route_mtu_cached = route_mtu_cached;
2320 
2321 		if (pmtu > route_mtu_cached)
2322 			pmtu = route_mtu_cached;
2323 
2324 		dst_metric_set(dst, RTAX_MTU, pmtu);
2325 	} while ((dst = dst->next));
2326 }
2327 
2328 /* Check that the bundle accepts the flow and its components are
2329  * still valid.
2330  */
2331 
2332 static int xfrm_bundle_ok(struct xfrm_dst *first)
2333 {
2334 	struct dst_entry *dst = &first->u.dst;
2335 	struct xfrm_dst *last;
2336 	u32 mtu;
2337 
2338 	if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2339 	    (dst->dev && !netif_running(dst->dev)))
2340 		return 0;
2341 
2342 	last = NULL;
2343 
2344 	do {
2345 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2346 
2347 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2348 			return 0;
2349 		if (xdst->xfrm_genid != dst->xfrm->genid)
2350 			return 0;
2351 		if (xdst->num_pols > 0 &&
2352 		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2353 			return 0;
2354 
2355 		mtu = dst_mtu(dst->child);
2356 		if (xdst->child_mtu_cached != mtu) {
2357 			last = xdst;
2358 			xdst->child_mtu_cached = mtu;
2359 		}
2360 
2361 		if (!dst_check(xdst->route, xdst->route_cookie))
2362 			return 0;
2363 		mtu = dst_mtu(xdst->route);
2364 		if (xdst->route_mtu_cached != mtu) {
2365 			last = xdst;
2366 			xdst->route_mtu_cached = mtu;
2367 		}
2368 
2369 		dst = dst->child;
2370 	} while (dst->xfrm);
2371 
2372 	if (likely(!last))
2373 		return 1;
2374 
2375 	mtu = last->child_mtu_cached;
2376 	for (;;) {
2377 		dst = &last->u.dst;
2378 
2379 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2380 		if (mtu > last->route_mtu_cached)
2381 			mtu = last->route_mtu_cached;
2382 		dst_metric_set(dst, RTAX_MTU, mtu);
2383 
2384 		if (last == first)
2385 			break;
2386 
2387 		last = (struct xfrm_dst *)last->u.dst.next;
2388 		last->child_mtu_cached = mtu;
2389 	}
2390 
2391 	return 1;
2392 }
2393 
2394 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2395 {
2396 	return dst_metric_advmss(dst->path);
2397 }
2398 
2399 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2400 {
2401 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2402 
2403 	return mtu ? : dst_mtu(dst->path);
2404 }
2405 
2406 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2407 					   struct sk_buff *skb,
2408 					   const void *daddr)
2409 {
2410 	return dst->path->ops->neigh_lookup(dst, skb, daddr);
2411 }
2412 
2413 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2414 {
2415 	struct net *net;
2416 	int err = 0;
2417 	if (unlikely(afinfo == NULL))
2418 		return -EINVAL;
2419 	if (unlikely(afinfo->family >= NPROTO))
2420 		return -EAFNOSUPPORT;
2421 	write_lock_bh(&xfrm_policy_afinfo_lock);
2422 	if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2423 		err = -ENOBUFS;
2424 	else {
2425 		struct dst_ops *dst_ops = afinfo->dst_ops;
2426 		if (likely(dst_ops->kmem_cachep == NULL))
2427 			dst_ops->kmem_cachep = xfrm_dst_cache;
2428 		if (likely(dst_ops->check == NULL))
2429 			dst_ops->check = xfrm_dst_check;
2430 		if (likely(dst_ops->default_advmss == NULL))
2431 			dst_ops->default_advmss = xfrm_default_advmss;
2432 		if (likely(dst_ops->mtu == NULL))
2433 			dst_ops->mtu = xfrm_mtu;
2434 		if (likely(dst_ops->negative_advice == NULL))
2435 			dst_ops->negative_advice = xfrm_negative_advice;
2436 		if (likely(dst_ops->link_failure == NULL))
2437 			dst_ops->link_failure = xfrm_link_failure;
2438 		if (likely(dst_ops->neigh_lookup == NULL))
2439 			dst_ops->neigh_lookup = xfrm_neigh_lookup;
2440 		if (likely(afinfo->garbage_collect == NULL))
2441 			afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2442 		xfrm_policy_afinfo[afinfo->family] = afinfo;
2443 	}
2444 	write_unlock_bh(&xfrm_policy_afinfo_lock);
2445 
2446 	rtnl_lock();
2447 	for_each_net(net) {
2448 		struct dst_ops *xfrm_dst_ops;
2449 
2450 		switch (afinfo->family) {
2451 		case AF_INET:
2452 			xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2453 			break;
2454 #if IS_ENABLED(CONFIG_IPV6)
2455 		case AF_INET6:
2456 			xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2457 			break;
2458 #endif
2459 		default:
2460 			BUG();
2461 		}
2462 		*xfrm_dst_ops = *afinfo->dst_ops;
2463 	}
2464 	rtnl_unlock();
2465 
2466 	return err;
2467 }
2468 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2469 
2470 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2471 {
2472 	int err = 0;
2473 	if (unlikely(afinfo == NULL))
2474 		return -EINVAL;
2475 	if (unlikely(afinfo->family >= NPROTO))
2476 		return -EAFNOSUPPORT;
2477 	write_lock_bh(&xfrm_policy_afinfo_lock);
2478 	if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2479 		if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2480 			err = -EINVAL;
2481 		else {
2482 			struct dst_ops *dst_ops = afinfo->dst_ops;
2483 			xfrm_policy_afinfo[afinfo->family] = NULL;
2484 			dst_ops->kmem_cachep = NULL;
2485 			dst_ops->check = NULL;
2486 			dst_ops->negative_advice = NULL;
2487 			dst_ops->link_failure = NULL;
2488 			afinfo->garbage_collect = NULL;
2489 		}
2490 	}
2491 	write_unlock_bh(&xfrm_policy_afinfo_lock);
2492 	return err;
2493 }
2494 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2495 
2496 static void __net_init xfrm_dst_ops_init(struct net *net)
2497 {
2498 	struct xfrm_policy_afinfo *afinfo;
2499 
2500 	read_lock_bh(&xfrm_policy_afinfo_lock);
2501 	afinfo = xfrm_policy_afinfo[AF_INET];
2502 	if (afinfo)
2503 		net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2504 #if IS_ENABLED(CONFIG_IPV6)
2505 	afinfo = xfrm_policy_afinfo[AF_INET6];
2506 	if (afinfo)
2507 		net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2508 #endif
2509 	read_unlock_bh(&xfrm_policy_afinfo_lock);
2510 }
2511 
2512 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2513 {
2514 	struct xfrm_policy_afinfo *afinfo;
2515 	if (unlikely(family >= NPROTO))
2516 		return NULL;
2517 	read_lock(&xfrm_policy_afinfo_lock);
2518 	afinfo = xfrm_policy_afinfo[family];
2519 	if (unlikely(!afinfo))
2520 		read_unlock(&xfrm_policy_afinfo_lock);
2521 	return afinfo;
2522 }
2523 
2524 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2525 {
2526 	read_unlock(&xfrm_policy_afinfo_lock);
2527 }
2528 
2529 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2530 {
2531 	struct net_device *dev = ptr;
2532 
2533 	switch (event) {
2534 	case NETDEV_DOWN:
2535 		xfrm_garbage_collect(dev_net(dev));
2536 	}
2537 	return NOTIFY_DONE;
2538 }
2539 
2540 static struct notifier_block xfrm_dev_notifier = {
2541 	.notifier_call	= xfrm_dev_event,
2542 };
2543 
2544 #ifdef CONFIG_XFRM_STATISTICS
2545 static int __net_init xfrm_statistics_init(struct net *net)
2546 {
2547 	int rv;
2548 
2549 	if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2550 			  sizeof(struct linux_xfrm_mib),
2551 			  __alignof__(struct linux_xfrm_mib)) < 0)
2552 		return -ENOMEM;
2553 	rv = xfrm_proc_init(net);
2554 	if (rv < 0)
2555 		snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2556 	return rv;
2557 }
2558 
2559 static void xfrm_statistics_fini(struct net *net)
2560 {
2561 	xfrm_proc_fini(net);
2562 	snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2563 }
2564 #else
2565 static int __net_init xfrm_statistics_init(struct net *net)
2566 {
2567 	return 0;
2568 }
2569 
2570 static void xfrm_statistics_fini(struct net *net)
2571 {
2572 }
2573 #endif
2574 
2575 static int __net_init xfrm_policy_init(struct net *net)
2576 {
2577 	unsigned int hmask, sz;
2578 	int dir;
2579 
2580 	if (net_eq(net, &init_net))
2581 		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2582 					   sizeof(struct xfrm_dst),
2583 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2584 					   NULL);
2585 
2586 	hmask = 8 - 1;
2587 	sz = (hmask+1) * sizeof(struct hlist_head);
2588 
2589 	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2590 	if (!net->xfrm.policy_byidx)
2591 		goto out_byidx;
2592 	net->xfrm.policy_idx_hmask = hmask;
2593 
2594 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2595 		struct xfrm_policy_hash *htab;
2596 
2597 		net->xfrm.policy_count[dir] = 0;
2598 		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2599 
2600 		htab = &net->xfrm.policy_bydst[dir];
2601 		htab->table = xfrm_hash_alloc(sz);
2602 		if (!htab->table)
2603 			goto out_bydst;
2604 		htab->hmask = hmask;
2605 	}
2606 
2607 	INIT_LIST_HEAD(&net->xfrm.policy_all);
2608 	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2609 	if (net_eq(net, &init_net))
2610 		register_netdevice_notifier(&xfrm_dev_notifier);
2611 	return 0;
2612 
2613 out_bydst:
2614 	for (dir--; dir >= 0; dir--) {
2615 		struct xfrm_policy_hash *htab;
2616 
2617 		htab = &net->xfrm.policy_bydst[dir];
2618 		xfrm_hash_free(htab->table, sz);
2619 	}
2620 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2621 out_byidx:
2622 	return -ENOMEM;
2623 }
2624 
2625 static void xfrm_policy_fini(struct net *net)
2626 {
2627 	struct xfrm_audit audit_info;
2628 	unsigned int sz;
2629 	int dir;
2630 
2631 	flush_work(&net->xfrm.policy_hash_work);
2632 #ifdef CONFIG_XFRM_SUB_POLICY
2633 	audit_info.loginuid = -1;
2634 	audit_info.sessionid = -1;
2635 	audit_info.secid = 0;
2636 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2637 #endif
2638 	audit_info.loginuid = -1;
2639 	audit_info.sessionid = -1;
2640 	audit_info.secid = 0;
2641 	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2642 
2643 	WARN_ON(!list_empty(&net->xfrm.policy_all));
2644 
2645 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2646 		struct xfrm_policy_hash *htab;
2647 
2648 		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2649 
2650 		htab = &net->xfrm.policy_bydst[dir];
2651 		sz = (htab->hmask + 1);
2652 		WARN_ON(!hlist_empty(htab->table));
2653 		xfrm_hash_free(htab->table, sz);
2654 	}
2655 
2656 	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2657 	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2658 	xfrm_hash_free(net->xfrm.policy_byidx, sz);
2659 }
2660 
2661 static int __net_init xfrm_net_init(struct net *net)
2662 {
2663 	int rv;
2664 
2665 	rv = xfrm_statistics_init(net);
2666 	if (rv < 0)
2667 		goto out_statistics;
2668 	rv = xfrm_state_init(net);
2669 	if (rv < 0)
2670 		goto out_state;
2671 	rv = xfrm_policy_init(net);
2672 	if (rv < 0)
2673 		goto out_policy;
2674 	xfrm_dst_ops_init(net);
2675 	rv = xfrm_sysctl_init(net);
2676 	if (rv < 0)
2677 		goto out_sysctl;
2678 	return 0;
2679 
2680 out_sysctl:
2681 	xfrm_policy_fini(net);
2682 out_policy:
2683 	xfrm_state_fini(net);
2684 out_state:
2685 	xfrm_statistics_fini(net);
2686 out_statistics:
2687 	return rv;
2688 }
2689 
2690 static void __net_exit xfrm_net_exit(struct net *net)
2691 {
2692 	xfrm_sysctl_fini(net);
2693 	xfrm_policy_fini(net);
2694 	xfrm_state_fini(net);
2695 	xfrm_statistics_fini(net);
2696 }
2697 
2698 static struct pernet_operations __net_initdata xfrm_net_ops = {
2699 	.init = xfrm_net_init,
2700 	.exit = xfrm_net_exit,
2701 };
2702 
2703 void __init xfrm_init(void)
2704 {
2705 	register_pernet_subsys(&xfrm_net_ops);
2706 	xfrm_input_init();
2707 }
2708 
2709 #ifdef CONFIG_AUDITSYSCALL
2710 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2711 					 struct audit_buffer *audit_buf)
2712 {
2713 	struct xfrm_sec_ctx *ctx = xp->security;
2714 	struct xfrm_selector *sel = &xp->selector;
2715 
2716 	if (ctx)
2717 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2718 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2719 
2720 	switch(sel->family) {
2721 	case AF_INET:
2722 		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2723 		if (sel->prefixlen_s != 32)
2724 			audit_log_format(audit_buf, " src_prefixlen=%d",
2725 					 sel->prefixlen_s);
2726 		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2727 		if (sel->prefixlen_d != 32)
2728 			audit_log_format(audit_buf, " dst_prefixlen=%d",
2729 					 sel->prefixlen_d);
2730 		break;
2731 	case AF_INET6:
2732 		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2733 		if (sel->prefixlen_s != 128)
2734 			audit_log_format(audit_buf, " src_prefixlen=%d",
2735 					 sel->prefixlen_s);
2736 		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2737 		if (sel->prefixlen_d != 128)
2738 			audit_log_format(audit_buf, " dst_prefixlen=%d",
2739 					 sel->prefixlen_d);
2740 		break;
2741 	}
2742 }
2743 
2744 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2745 			   uid_t auid, u32 sessionid, u32 secid)
2746 {
2747 	struct audit_buffer *audit_buf;
2748 
2749 	audit_buf = xfrm_audit_start("SPD-add");
2750 	if (audit_buf == NULL)
2751 		return;
2752 	xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2753 	audit_log_format(audit_buf, " res=%u", result);
2754 	xfrm_audit_common_policyinfo(xp, audit_buf);
2755 	audit_log_end(audit_buf);
2756 }
2757 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2758 
2759 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2760 			      uid_t auid, u32 sessionid, u32 secid)
2761 {
2762 	struct audit_buffer *audit_buf;
2763 
2764 	audit_buf = xfrm_audit_start("SPD-delete");
2765 	if (audit_buf == NULL)
2766 		return;
2767 	xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2768 	audit_log_format(audit_buf, " res=%u", result);
2769 	xfrm_audit_common_policyinfo(xp, audit_buf);
2770 	audit_log_end(audit_buf);
2771 }
2772 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2773 #endif
2774 
2775 #ifdef CONFIG_XFRM_MIGRATE
2776 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
2777 					const struct xfrm_selector *sel_tgt)
2778 {
2779 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2780 		if (sel_tgt->family == sel_cmp->family &&
2781 		    xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2782 				  sel_cmp->family) == 0 &&
2783 		    xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2784 				  sel_cmp->family) == 0 &&
2785 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2786 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2787 			return true;
2788 		}
2789 	} else {
2790 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2791 			return true;
2792 		}
2793 	}
2794 	return false;
2795 }
2796 
2797 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
2798 						     u8 dir, u8 type)
2799 {
2800 	struct xfrm_policy *pol, *ret = NULL;
2801 	struct hlist_node *entry;
2802 	struct hlist_head *chain;
2803 	u32 priority = ~0U;
2804 
2805 	read_lock_bh(&xfrm_policy_lock);
2806 	chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2807 	hlist_for_each_entry(pol, entry, chain, bydst) {
2808 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2809 		    pol->type == type) {
2810 			ret = pol;
2811 			priority = ret->priority;
2812 			break;
2813 		}
2814 	}
2815 	chain = &init_net.xfrm.policy_inexact[dir];
2816 	hlist_for_each_entry(pol, entry, chain, bydst) {
2817 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2818 		    pol->type == type &&
2819 		    pol->priority < priority) {
2820 			ret = pol;
2821 			break;
2822 		}
2823 	}
2824 
2825 	if (ret)
2826 		xfrm_pol_hold(ret);
2827 
2828 	read_unlock_bh(&xfrm_policy_lock);
2829 
2830 	return ret;
2831 }
2832 
2833 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2834 {
2835 	int match = 0;
2836 
2837 	if (t->mode == m->mode && t->id.proto == m->proto &&
2838 	    (m->reqid == 0 || t->reqid == m->reqid)) {
2839 		switch (t->mode) {
2840 		case XFRM_MODE_TUNNEL:
2841 		case XFRM_MODE_BEET:
2842 			if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2843 					  m->old_family) == 0 &&
2844 			    xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2845 					  m->old_family) == 0) {
2846 				match = 1;
2847 			}
2848 			break;
2849 		case XFRM_MODE_TRANSPORT:
2850 			/* in case of transport mode, template does not store
2851 			   any IP addresses, hence we just compare mode and
2852 			   protocol */
2853 			match = 1;
2854 			break;
2855 		default:
2856 			break;
2857 		}
2858 	}
2859 	return match;
2860 }
2861 
2862 /* update endpoint address(es) of template(s) */
2863 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2864 			       struct xfrm_migrate *m, int num_migrate)
2865 {
2866 	struct xfrm_migrate *mp;
2867 	int i, j, n = 0;
2868 
2869 	write_lock_bh(&pol->lock);
2870 	if (unlikely(pol->walk.dead)) {
2871 		/* target policy has been deleted */
2872 		write_unlock_bh(&pol->lock);
2873 		return -ENOENT;
2874 	}
2875 
2876 	for (i = 0; i < pol->xfrm_nr; i++) {
2877 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2878 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2879 				continue;
2880 			n++;
2881 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2882 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2883 				continue;
2884 			/* update endpoints */
2885 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2886 			       sizeof(pol->xfrm_vec[i].id.daddr));
2887 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2888 			       sizeof(pol->xfrm_vec[i].saddr));
2889 			pol->xfrm_vec[i].encap_family = mp->new_family;
2890 			/* flush bundles */
2891 			atomic_inc(&pol->genid);
2892 		}
2893 	}
2894 
2895 	write_unlock_bh(&pol->lock);
2896 
2897 	if (!n)
2898 		return -ENODATA;
2899 
2900 	return 0;
2901 }
2902 
2903 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
2904 {
2905 	int i, j;
2906 
2907 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2908 		return -EINVAL;
2909 
2910 	for (i = 0; i < num_migrate; i++) {
2911 		if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2912 				   m[i].old_family) == 0) &&
2913 		    (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2914 				   m[i].old_family) == 0))
2915 			return -EINVAL;
2916 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2917 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2918 			return -EINVAL;
2919 
2920 		/* check if there is any duplicated entry */
2921 		for (j = i + 1; j < num_migrate; j++) {
2922 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2923 				    sizeof(m[i].old_daddr)) &&
2924 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2925 				    sizeof(m[i].old_saddr)) &&
2926 			    m[i].proto == m[j].proto &&
2927 			    m[i].mode == m[j].mode &&
2928 			    m[i].reqid == m[j].reqid &&
2929 			    m[i].old_family == m[j].old_family)
2930 				return -EINVAL;
2931 		}
2932 	}
2933 
2934 	return 0;
2935 }
2936 
2937 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
2938 		 struct xfrm_migrate *m, int num_migrate,
2939 		 struct xfrm_kmaddress *k)
2940 {
2941 	int i, err, nx_cur = 0, nx_new = 0;
2942 	struct xfrm_policy *pol = NULL;
2943 	struct xfrm_state *x, *xc;
2944 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2945 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2946 	struct xfrm_migrate *mp;
2947 
2948 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2949 		goto out;
2950 
2951 	/* Stage 1 - find policy */
2952 	if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2953 		err = -ENOENT;
2954 		goto out;
2955 	}
2956 
2957 	/* Stage 2 - find and update state(s) */
2958 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2959 		if ((x = xfrm_migrate_state_find(mp))) {
2960 			x_cur[nx_cur] = x;
2961 			nx_cur++;
2962 			if ((xc = xfrm_state_migrate(x, mp))) {
2963 				x_new[nx_new] = xc;
2964 				nx_new++;
2965 			} else {
2966 				err = -ENODATA;
2967 				goto restore_state;
2968 			}
2969 		}
2970 	}
2971 
2972 	/* Stage 3 - update policy */
2973 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2974 		goto restore_state;
2975 
2976 	/* Stage 4 - delete old state(s) */
2977 	if (nx_cur) {
2978 		xfrm_states_put(x_cur, nx_cur);
2979 		xfrm_states_delete(x_cur, nx_cur);
2980 	}
2981 
2982 	/* Stage 5 - announce */
2983 	km_migrate(sel, dir, type, m, num_migrate, k);
2984 
2985 	xfrm_pol_put(pol);
2986 
2987 	return 0;
2988 out:
2989 	return err;
2990 
2991 restore_state:
2992 	if (pol)
2993 		xfrm_pol_put(pol);
2994 	if (nx_cur)
2995 		xfrm_states_put(x_cur, nx_cur);
2996 	if (nx_new)
2997 		xfrm_states_delete(x_new, nx_new);
2998 
2999 	return err;
3000 }
3001 EXPORT_SYMBOL(xfrm_migrate);
3002 #endif
3003