xref: /openbmc/linux/net/xfrm/xfrm_policy.c (revision 5a3e55d68ec5baac578bf32ba67607088c763657)
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 <net/dst.h>
28 #include <net/xfrm.h>
29 #include <net/ip.h>
30 
31 #include "xfrm_hash.h"
32 
33 int sysctl_xfrm_larval_drop __read_mostly;
34 
35 DEFINE_MUTEX(xfrm_cfg_mutex);
36 EXPORT_SYMBOL(xfrm_cfg_mutex);
37 
38 static DEFINE_RWLOCK(xfrm_policy_lock);
39 
40 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
41 EXPORT_SYMBOL(xfrm_policy_count);
42 
43 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
44 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
45 
46 static struct kmem_cache *xfrm_dst_cache __read_mostly;
47 
48 static struct work_struct xfrm_policy_gc_work;
49 static HLIST_HEAD(xfrm_policy_gc_list);
50 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
51 
52 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
53 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
54 static void xfrm_init_pmtu(struct dst_entry *dst);
55 
56 static inline int
57 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
58 {
59 	return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
60 		addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
61 		!((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
62 		!((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
63 		(fl->proto == sel->proto || !sel->proto) &&
64 		(fl->oif == sel->ifindex || !sel->ifindex);
65 }
66 
67 static inline int
68 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
69 {
70 	return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
71 		addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
72 		!((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
73 		!((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
74 		(fl->proto == sel->proto || !sel->proto) &&
75 		(fl->oif == sel->ifindex || !sel->ifindex);
76 }
77 
78 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
79 		    unsigned short family)
80 {
81 	switch (family) {
82 	case AF_INET:
83 		return __xfrm4_selector_match(sel, fl);
84 	case AF_INET6:
85 		return __xfrm6_selector_match(sel, fl);
86 	}
87 	return 0;
88 }
89 
90 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
91 						int family)
92 {
93 	xfrm_address_t *saddr = &x->props.saddr;
94 	xfrm_address_t *daddr = &x->id.daddr;
95 	struct xfrm_policy_afinfo *afinfo;
96 	struct dst_entry *dst;
97 
98 	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR)
99 		saddr = x->coaddr;
100 	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR)
101 		daddr = x->coaddr;
102 
103 	afinfo = xfrm_policy_get_afinfo(family);
104 	if (unlikely(afinfo == NULL))
105 		return ERR_PTR(-EAFNOSUPPORT);
106 
107 	dst = afinfo->dst_lookup(tos, saddr, daddr);
108 	xfrm_policy_put_afinfo(afinfo);
109 	return dst;
110 }
111 
112 static inline unsigned long make_jiffies(long secs)
113 {
114 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
115 		return MAX_SCHEDULE_TIMEOUT-1;
116 	else
117 		return secs*HZ;
118 }
119 
120 static void xfrm_policy_timer(unsigned long data)
121 {
122 	struct xfrm_policy *xp = (struct xfrm_policy*)data;
123 	unsigned long now = get_seconds();
124 	long next = LONG_MAX;
125 	int warn = 0;
126 	int dir;
127 
128 	read_lock(&xp->lock);
129 
130 	if (xp->dead)
131 		goto out;
132 
133 	dir = xfrm_policy_id2dir(xp->index);
134 
135 	if (xp->lft.hard_add_expires_seconds) {
136 		long tmo = xp->lft.hard_add_expires_seconds +
137 			xp->curlft.add_time - now;
138 		if (tmo <= 0)
139 			goto expired;
140 		if (tmo < next)
141 			next = tmo;
142 	}
143 	if (xp->lft.hard_use_expires_seconds) {
144 		long tmo = xp->lft.hard_use_expires_seconds +
145 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
146 		if (tmo <= 0)
147 			goto expired;
148 		if (tmo < next)
149 			next = tmo;
150 	}
151 	if (xp->lft.soft_add_expires_seconds) {
152 		long tmo = xp->lft.soft_add_expires_seconds +
153 			xp->curlft.add_time - now;
154 		if (tmo <= 0) {
155 			warn = 1;
156 			tmo = XFRM_KM_TIMEOUT;
157 		}
158 		if (tmo < next)
159 			next = tmo;
160 	}
161 	if (xp->lft.soft_use_expires_seconds) {
162 		long tmo = xp->lft.soft_use_expires_seconds +
163 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
164 		if (tmo <= 0) {
165 			warn = 1;
166 			tmo = XFRM_KM_TIMEOUT;
167 		}
168 		if (tmo < next)
169 			next = tmo;
170 	}
171 
172 	if (warn)
173 		km_policy_expired(xp, dir, 0, 0);
174 	if (next != LONG_MAX &&
175 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
176 		xfrm_pol_hold(xp);
177 
178 out:
179 	read_unlock(&xp->lock);
180 	xfrm_pol_put(xp);
181 	return;
182 
183 expired:
184 	read_unlock(&xp->lock);
185 	if (!xfrm_policy_delete(xp, dir))
186 		km_policy_expired(xp, dir, 1, 0);
187 	xfrm_pol_put(xp);
188 }
189 
190 
191 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
192  * SPD calls.
193  */
194 
195 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
196 {
197 	struct xfrm_policy *policy;
198 
199 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
200 
201 	if (policy) {
202 		INIT_HLIST_NODE(&policy->bydst);
203 		INIT_HLIST_NODE(&policy->byidx);
204 		rwlock_init(&policy->lock);
205 		atomic_set(&policy->refcnt, 1);
206 		setup_timer(&policy->timer, xfrm_policy_timer,
207 				(unsigned long)policy);
208 	}
209 	return policy;
210 }
211 EXPORT_SYMBOL(xfrm_policy_alloc);
212 
213 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
214 
215 void __xfrm_policy_destroy(struct xfrm_policy *policy)
216 {
217 	BUG_ON(!policy->dead);
218 
219 	BUG_ON(policy->bundles);
220 
221 	if (del_timer(&policy->timer))
222 		BUG();
223 
224 	security_xfrm_policy_free(policy);
225 	kfree(policy);
226 }
227 EXPORT_SYMBOL(__xfrm_policy_destroy);
228 
229 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
230 {
231 	struct dst_entry *dst;
232 
233 	while ((dst = policy->bundles) != NULL) {
234 		policy->bundles = dst->next;
235 		dst_free(dst);
236 	}
237 
238 	if (del_timer(&policy->timer))
239 		atomic_dec(&policy->refcnt);
240 
241 	if (atomic_read(&policy->refcnt) > 1)
242 		flow_cache_flush();
243 
244 	xfrm_pol_put(policy);
245 }
246 
247 static void xfrm_policy_gc_task(struct work_struct *work)
248 {
249 	struct xfrm_policy *policy;
250 	struct hlist_node *entry, *tmp;
251 	struct hlist_head gc_list;
252 
253 	spin_lock_bh(&xfrm_policy_gc_lock);
254 	gc_list.first = xfrm_policy_gc_list.first;
255 	INIT_HLIST_HEAD(&xfrm_policy_gc_list);
256 	spin_unlock_bh(&xfrm_policy_gc_lock);
257 
258 	hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
259 		xfrm_policy_gc_kill(policy);
260 }
261 
262 /* Rule must be locked. Release descentant resources, announce
263  * entry dead. The rule must be unlinked from lists to the moment.
264  */
265 
266 static void xfrm_policy_kill(struct xfrm_policy *policy)
267 {
268 	int dead;
269 
270 	write_lock_bh(&policy->lock);
271 	dead = policy->dead;
272 	policy->dead = 1;
273 	write_unlock_bh(&policy->lock);
274 
275 	if (unlikely(dead)) {
276 		WARN_ON(1);
277 		return;
278 	}
279 
280 	spin_lock(&xfrm_policy_gc_lock);
281 	hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
282 	spin_unlock(&xfrm_policy_gc_lock);
283 
284 	schedule_work(&xfrm_policy_gc_work);
285 }
286 
287 struct xfrm_policy_hash {
288 	struct hlist_head	*table;
289 	unsigned int		hmask;
290 };
291 
292 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
293 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
294 static struct hlist_head *xfrm_policy_byidx __read_mostly;
295 static unsigned int xfrm_idx_hmask __read_mostly;
296 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
297 
298 static inline unsigned int idx_hash(u32 index)
299 {
300 	return __idx_hash(index, xfrm_idx_hmask);
301 }
302 
303 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
304 {
305 	unsigned int hmask = xfrm_policy_bydst[dir].hmask;
306 	unsigned int hash = __sel_hash(sel, family, hmask);
307 
308 	return (hash == hmask + 1 ?
309 		&xfrm_policy_inexact[dir] :
310 		xfrm_policy_bydst[dir].table + hash);
311 }
312 
313 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
314 {
315 	unsigned int hmask = xfrm_policy_bydst[dir].hmask;
316 	unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
317 
318 	return xfrm_policy_bydst[dir].table + hash;
319 }
320 
321 static void xfrm_dst_hash_transfer(struct hlist_head *list,
322 				   struct hlist_head *ndsttable,
323 				   unsigned int nhashmask)
324 {
325 	struct hlist_node *entry, *tmp;
326 	struct xfrm_policy *pol;
327 
328 	hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
329 		unsigned int h;
330 
331 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
332 				pol->family, nhashmask);
333 		hlist_add_head(&pol->bydst, ndsttable+h);
334 	}
335 }
336 
337 static void xfrm_idx_hash_transfer(struct hlist_head *list,
338 				   struct hlist_head *nidxtable,
339 				   unsigned int nhashmask)
340 {
341 	struct hlist_node *entry, *tmp;
342 	struct xfrm_policy *pol;
343 
344 	hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
345 		unsigned int h;
346 
347 		h = __idx_hash(pol->index, nhashmask);
348 		hlist_add_head(&pol->byidx, nidxtable+h);
349 	}
350 }
351 
352 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
353 {
354 	return ((old_hmask + 1) << 1) - 1;
355 }
356 
357 static void xfrm_bydst_resize(int dir)
358 {
359 	unsigned int hmask = xfrm_policy_bydst[dir].hmask;
360 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
361 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
362 	struct hlist_head *odst = xfrm_policy_bydst[dir].table;
363 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
364 	int i;
365 
366 	if (!ndst)
367 		return;
368 
369 	write_lock_bh(&xfrm_policy_lock);
370 
371 	for (i = hmask; i >= 0; i--)
372 		xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
373 
374 	xfrm_policy_bydst[dir].table = ndst;
375 	xfrm_policy_bydst[dir].hmask = nhashmask;
376 
377 	write_unlock_bh(&xfrm_policy_lock);
378 
379 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
380 }
381 
382 static void xfrm_byidx_resize(int total)
383 {
384 	unsigned int hmask = xfrm_idx_hmask;
385 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
386 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
387 	struct hlist_head *oidx = xfrm_policy_byidx;
388 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
389 	int i;
390 
391 	if (!nidx)
392 		return;
393 
394 	write_lock_bh(&xfrm_policy_lock);
395 
396 	for (i = hmask; i >= 0; i--)
397 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
398 
399 	xfrm_policy_byidx = nidx;
400 	xfrm_idx_hmask = nhashmask;
401 
402 	write_unlock_bh(&xfrm_policy_lock);
403 
404 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
405 }
406 
407 static inline int xfrm_bydst_should_resize(int dir, int *total)
408 {
409 	unsigned int cnt = xfrm_policy_count[dir];
410 	unsigned int hmask = xfrm_policy_bydst[dir].hmask;
411 
412 	if (total)
413 		*total += cnt;
414 
415 	if ((hmask + 1) < xfrm_policy_hashmax &&
416 	    cnt > hmask)
417 		return 1;
418 
419 	return 0;
420 }
421 
422 static inline int xfrm_byidx_should_resize(int total)
423 {
424 	unsigned int hmask = xfrm_idx_hmask;
425 
426 	if ((hmask + 1) < xfrm_policy_hashmax &&
427 	    total > hmask)
428 		return 1;
429 
430 	return 0;
431 }
432 
433 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
434 {
435 	read_lock_bh(&xfrm_policy_lock);
436 	si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
437 	si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
438 	si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
439 	si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
440 	si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
441 	si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
442 	si->spdhcnt = xfrm_idx_hmask;
443 	si->spdhmcnt = xfrm_policy_hashmax;
444 	read_unlock_bh(&xfrm_policy_lock);
445 }
446 EXPORT_SYMBOL(xfrm_spd_getinfo);
447 
448 static DEFINE_MUTEX(hash_resize_mutex);
449 static void xfrm_hash_resize(struct work_struct *__unused)
450 {
451 	int dir, total;
452 
453 	mutex_lock(&hash_resize_mutex);
454 
455 	total = 0;
456 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
457 		if (xfrm_bydst_should_resize(dir, &total))
458 			xfrm_bydst_resize(dir);
459 	}
460 	if (xfrm_byidx_should_resize(total))
461 		xfrm_byidx_resize(total);
462 
463 	mutex_unlock(&hash_resize_mutex);
464 }
465 
466 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
467 
468 /* Generate new index... KAME seems to generate them ordered by cost
469  * of an absolute inpredictability of ordering of rules. This will not pass. */
470 static u32 xfrm_gen_index(u8 type, int dir)
471 {
472 	static u32 idx_generator;
473 
474 	for (;;) {
475 		struct hlist_node *entry;
476 		struct hlist_head *list;
477 		struct xfrm_policy *p;
478 		u32 idx;
479 		int found;
480 
481 		idx = (idx_generator | dir);
482 		idx_generator += 8;
483 		if (idx == 0)
484 			idx = 8;
485 		list = xfrm_policy_byidx + idx_hash(idx);
486 		found = 0;
487 		hlist_for_each_entry(p, entry, list, byidx) {
488 			if (p->index == idx) {
489 				found = 1;
490 				break;
491 			}
492 		}
493 		if (!found)
494 			return idx;
495 	}
496 }
497 
498 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
499 {
500 	u32 *p1 = (u32 *) s1;
501 	u32 *p2 = (u32 *) s2;
502 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
503 	int i;
504 
505 	for (i = 0; i < len; i++) {
506 		if (p1[i] != p2[i])
507 			return 1;
508 	}
509 
510 	return 0;
511 }
512 
513 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
514 {
515 	struct xfrm_policy *pol;
516 	struct xfrm_policy *delpol;
517 	struct hlist_head *chain;
518 	struct hlist_node *entry, *newpos;
519 	struct dst_entry *gc_list;
520 
521 	write_lock_bh(&xfrm_policy_lock);
522 	chain = policy_hash_bysel(&policy->selector, policy->family, dir);
523 	delpol = NULL;
524 	newpos = NULL;
525 	hlist_for_each_entry(pol, entry, chain, bydst) {
526 		if (pol->type == policy->type &&
527 		    !selector_cmp(&pol->selector, &policy->selector) &&
528 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
529 		    !WARN_ON(delpol)) {
530 			if (excl) {
531 				write_unlock_bh(&xfrm_policy_lock);
532 				return -EEXIST;
533 			}
534 			delpol = pol;
535 			if (policy->priority > pol->priority)
536 				continue;
537 		} else if (policy->priority >= pol->priority) {
538 			newpos = &pol->bydst;
539 			continue;
540 		}
541 		if (delpol)
542 			break;
543 	}
544 	if (newpos)
545 		hlist_add_after(newpos, &policy->bydst);
546 	else
547 		hlist_add_head(&policy->bydst, chain);
548 	xfrm_pol_hold(policy);
549 	xfrm_policy_count[dir]++;
550 	atomic_inc(&flow_cache_genid);
551 	if (delpol) {
552 		hlist_del(&delpol->bydst);
553 		hlist_del(&delpol->byidx);
554 		xfrm_policy_count[dir]--;
555 	}
556 	policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
557 	hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
558 	policy->curlft.add_time = get_seconds();
559 	policy->curlft.use_time = 0;
560 	if (!mod_timer(&policy->timer, jiffies + HZ))
561 		xfrm_pol_hold(policy);
562 	write_unlock_bh(&xfrm_policy_lock);
563 
564 	if (delpol)
565 		xfrm_policy_kill(delpol);
566 	else if (xfrm_bydst_should_resize(dir, NULL))
567 		schedule_work(&xfrm_hash_work);
568 
569 	read_lock_bh(&xfrm_policy_lock);
570 	gc_list = NULL;
571 	entry = &policy->bydst;
572 	hlist_for_each_entry_continue(policy, entry, bydst) {
573 		struct dst_entry *dst;
574 
575 		write_lock(&policy->lock);
576 		dst = policy->bundles;
577 		if (dst) {
578 			struct dst_entry *tail = dst;
579 			while (tail->next)
580 				tail = tail->next;
581 			tail->next = gc_list;
582 			gc_list = dst;
583 
584 			policy->bundles = NULL;
585 		}
586 		write_unlock(&policy->lock);
587 	}
588 	read_unlock_bh(&xfrm_policy_lock);
589 
590 	while (gc_list) {
591 		struct dst_entry *dst = gc_list;
592 
593 		gc_list = dst->next;
594 		dst_free(dst);
595 	}
596 
597 	return 0;
598 }
599 EXPORT_SYMBOL(xfrm_policy_insert);
600 
601 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
602 					  struct xfrm_selector *sel,
603 					  struct xfrm_sec_ctx *ctx, int delete,
604 					  int *err)
605 {
606 	struct xfrm_policy *pol, *ret;
607 	struct hlist_head *chain;
608 	struct hlist_node *entry;
609 
610 	*err = 0;
611 	write_lock_bh(&xfrm_policy_lock);
612 	chain = policy_hash_bysel(sel, sel->family, dir);
613 	ret = NULL;
614 	hlist_for_each_entry(pol, entry, chain, bydst) {
615 		if (pol->type == type &&
616 		    !selector_cmp(sel, &pol->selector) &&
617 		    xfrm_sec_ctx_match(ctx, pol->security)) {
618 			xfrm_pol_hold(pol);
619 			if (delete) {
620 				*err = security_xfrm_policy_delete(pol);
621 				if (*err) {
622 					write_unlock_bh(&xfrm_policy_lock);
623 					return pol;
624 				}
625 				hlist_del(&pol->bydst);
626 				hlist_del(&pol->byidx);
627 				xfrm_policy_count[dir]--;
628 			}
629 			ret = pol;
630 			break;
631 		}
632 	}
633 	write_unlock_bh(&xfrm_policy_lock);
634 
635 	if (ret && delete) {
636 		atomic_inc(&flow_cache_genid);
637 		xfrm_policy_kill(ret);
638 	}
639 	return ret;
640 }
641 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
642 
643 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
644 				     int *err)
645 {
646 	struct xfrm_policy *pol, *ret;
647 	struct hlist_head *chain;
648 	struct hlist_node *entry;
649 
650 	*err = -ENOENT;
651 	if (xfrm_policy_id2dir(id) != dir)
652 		return NULL;
653 
654 	*err = 0;
655 	write_lock_bh(&xfrm_policy_lock);
656 	chain = xfrm_policy_byidx + idx_hash(id);
657 	ret = NULL;
658 	hlist_for_each_entry(pol, entry, chain, byidx) {
659 		if (pol->type == type && pol->index == id) {
660 			xfrm_pol_hold(pol);
661 			if (delete) {
662 				*err = security_xfrm_policy_delete(pol);
663 				if (*err) {
664 					write_unlock_bh(&xfrm_policy_lock);
665 					return pol;
666 				}
667 				hlist_del(&pol->bydst);
668 				hlist_del(&pol->byidx);
669 				xfrm_policy_count[dir]--;
670 			}
671 			ret = pol;
672 			break;
673 		}
674 	}
675 	write_unlock_bh(&xfrm_policy_lock);
676 
677 	if (ret && delete) {
678 		atomic_inc(&flow_cache_genid);
679 		xfrm_policy_kill(ret);
680 	}
681 	return ret;
682 }
683 EXPORT_SYMBOL(xfrm_policy_byid);
684 
685 #ifdef CONFIG_SECURITY_NETWORK_XFRM
686 static inline int
687 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
688 {
689 	int dir, err = 0;
690 
691 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
692 		struct xfrm_policy *pol;
693 		struct hlist_node *entry;
694 		int i;
695 
696 		hlist_for_each_entry(pol, entry,
697 				     &xfrm_policy_inexact[dir], bydst) {
698 			if (pol->type != type)
699 				continue;
700 			err = security_xfrm_policy_delete(pol);
701 			if (err) {
702 				xfrm_audit_policy_delete(pol, 0,
703 							 audit_info->loginuid,
704 							 audit_info->secid);
705 				return err;
706 			}
707 		}
708 		for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
709 			hlist_for_each_entry(pol, entry,
710 					     xfrm_policy_bydst[dir].table + i,
711 					     bydst) {
712 				if (pol->type != type)
713 					continue;
714 				err = security_xfrm_policy_delete(pol);
715 				if (err) {
716 					xfrm_audit_policy_delete(pol, 0,
717 							audit_info->loginuid,
718 							audit_info->secid);
719 					return err;
720 				}
721 			}
722 		}
723 	}
724 	return err;
725 }
726 #else
727 static inline int
728 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
729 {
730 	return 0;
731 }
732 #endif
733 
734 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
735 {
736 	int dir, err = 0;
737 
738 	write_lock_bh(&xfrm_policy_lock);
739 
740 	err = xfrm_policy_flush_secctx_check(type, audit_info);
741 	if (err)
742 		goto out;
743 
744 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
745 		struct xfrm_policy *pol;
746 		struct hlist_node *entry;
747 		int i, killed;
748 
749 		killed = 0;
750 	again1:
751 		hlist_for_each_entry(pol, entry,
752 				     &xfrm_policy_inexact[dir], bydst) {
753 			if (pol->type != type)
754 				continue;
755 			hlist_del(&pol->bydst);
756 			hlist_del(&pol->byidx);
757 			write_unlock_bh(&xfrm_policy_lock);
758 
759 			xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
760 						 audit_info->secid);
761 
762 			xfrm_policy_kill(pol);
763 			killed++;
764 
765 			write_lock_bh(&xfrm_policy_lock);
766 			goto again1;
767 		}
768 
769 		for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
770 	again2:
771 			hlist_for_each_entry(pol, entry,
772 					     xfrm_policy_bydst[dir].table + i,
773 					     bydst) {
774 				if (pol->type != type)
775 					continue;
776 				hlist_del(&pol->bydst);
777 				hlist_del(&pol->byidx);
778 				write_unlock_bh(&xfrm_policy_lock);
779 
780 				xfrm_audit_policy_delete(pol, 1,
781 							 audit_info->loginuid,
782 							 audit_info->secid);
783 				xfrm_policy_kill(pol);
784 				killed++;
785 
786 				write_lock_bh(&xfrm_policy_lock);
787 				goto again2;
788 			}
789 		}
790 
791 		xfrm_policy_count[dir] -= killed;
792 	}
793 	atomic_inc(&flow_cache_genid);
794 out:
795 	write_unlock_bh(&xfrm_policy_lock);
796 	return err;
797 }
798 EXPORT_SYMBOL(xfrm_policy_flush);
799 
800 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
801 		     void *data)
802 {
803 	struct xfrm_policy *pol, *last = NULL;
804 	struct hlist_node *entry;
805 	int dir, last_dir = 0, count, error;
806 
807 	read_lock_bh(&xfrm_policy_lock);
808 	count = 0;
809 
810 	for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
811 		struct hlist_head *table = xfrm_policy_bydst[dir].table;
812 		int i;
813 
814 		hlist_for_each_entry(pol, entry,
815 				     &xfrm_policy_inexact[dir], bydst) {
816 			if (pol->type != type)
817 				continue;
818 			if (last) {
819 				error = func(last, last_dir % XFRM_POLICY_MAX,
820 					     count, data);
821 				if (error)
822 					goto out;
823 			}
824 			last = pol;
825 			last_dir = dir;
826 			count++;
827 		}
828 		for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
829 			hlist_for_each_entry(pol, entry, table + i, bydst) {
830 				if (pol->type != type)
831 					continue;
832 				if (last) {
833 					error = func(last, last_dir % XFRM_POLICY_MAX,
834 						     count, data);
835 					if (error)
836 						goto out;
837 				}
838 				last = pol;
839 				last_dir = dir;
840 				count++;
841 			}
842 		}
843 	}
844 	if (count == 0) {
845 		error = -ENOENT;
846 		goto out;
847 	}
848 	error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
849 out:
850 	read_unlock_bh(&xfrm_policy_lock);
851 	return error;
852 }
853 EXPORT_SYMBOL(xfrm_policy_walk);
854 
855 /*
856  * Find policy to apply to this flow.
857  *
858  * Returns 0 if policy found, else an -errno.
859  */
860 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
861 			     u8 type, u16 family, int dir)
862 {
863 	struct xfrm_selector *sel = &pol->selector;
864 	int match, ret = -ESRCH;
865 
866 	if (pol->family != family ||
867 	    pol->type != type)
868 		return ret;
869 
870 	match = xfrm_selector_match(sel, fl, family);
871 	if (match)
872 		ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
873 
874 	return ret;
875 }
876 
877 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
878 						     u16 family, u8 dir)
879 {
880 	int err;
881 	struct xfrm_policy *pol, *ret;
882 	xfrm_address_t *daddr, *saddr;
883 	struct hlist_node *entry;
884 	struct hlist_head *chain;
885 	u32 priority = ~0U;
886 
887 	daddr = xfrm_flowi_daddr(fl, family);
888 	saddr = xfrm_flowi_saddr(fl, family);
889 	if (unlikely(!daddr || !saddr))
890 		return NULL;
891 
892 	read_lock_bh(&xfrm_policy_lock);
893 	chain = policy_hash_direct(daddr, saddr, family, dir);
894 	ret = NULL;
895 	hlist_for_each_entry(pol, entry, chain, bydst) {
896 		err = xfrm_policy_match(pol, fl, type, family, dir);
897 		if (err) {
898 			if (err == -ESRCH)
899 				continue;
900 			else {
901 				ret = ERR_PTR(err);
902 				goto fail;
903 			}
904 		} else {
905 			ret = pol;
906 			priority = ret->priority;
907 			break;
908 		}
909 	}
910 	chain = &xfrm_policy_inexact[dir];
911 	hlist_for_each_entry(pol, entry, chain, bydst) {
912 		err = xfrm_policy_match(pol, fl, type, family, dir);
913 		if (err) {
914 			if (err == -ESRCH)
915 				continue;
916 			else {
917 				ret = ERR_PTR(err);
918 				goto fail;
919 			}
920 		} else if (pol->priority < priority) {
921 			ret = pol;
922 			break;
923 		}
924 	}
925 	if (ret)
926 		xfrm_pol_hold(ret);
927 fail:
928 	read_unlock_bh(&xfrm_policy_lock);
929 
930 	return ret;
931 }
932 
933 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
934 			       void **objp, atomic_t **obj_refp)
935 {
936 	struct xfrm_policy *pol;
937 	int err = 0;
938 
939 #ifdef CONFIG_XFRM_SUB_POLICY
940 	pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
941 	if (IS_ERR(pol)) {
942 		err = PTR_ERR(pol);
943 		pol = NULL;
944 	}
945 	if (pol || err)
946 		goto end;
947 #endif
948 	pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
949 	if (IS_ERR(pol)) {
950 		err = PTR_ERR(pol);
951 		pol = NULL;
952 	}
953 #ifdef CONFIG_XFRM_SUB_POLICY
954 end:
955 #endif
956 	if ((*objp = (void *) pol) != NULL)
957 		*obj_refp = &pol->refcnt;
958 	return err;
959 }
960 
961 static inline int policy_to_flow_dir(int dir)
962 {
963 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
964 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
965 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
966 		return dir;
967 	switch (dir) {
968 	default:
969 	case XFRM_POLICY_IN:
970 		return FLOW_DIR_IN;
971 	case XFRM_POLICY_OUT:
972 		return FLOW_DIR_OUT;
973 	case XFRM_POLICY_FWD:
974 		return FLOW_DIR_FWD;
975 	}
976 }
977 
978 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
979 {
980 	struct xfrm_policy *pol;
981 
982 	read_lock_bh(&xfrm_policy_lock);
983 	if ((pol = sk->sk_policy[dir]) != NULL) {
984 		int match = xfrm_selector_match(&pol->selector, fl,
985 						sk->sk_family);
986 		int err = 0;
987 
988 		if (match) {
989 			err = security_xfrm_policy_lookup(pol, fl->secid,
990 					policy_to_flow_dir(dir));
991 			if (!err)
992 				xfrm_pol_hold(pol);
993 			else if (err == -ESRCH)
994 				pol = NULL;
995 			else
996 				pol = ERR_PTR(err);
997 		} else
998 			pol = NULL;
999 	}
1000 	read_unlock_bh(&xfrm_policy_lock);
1001 	return pol;
1002 }
1003 
1004 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1005 {
1006 	struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1007 						     pol->family, dir);
1008 
1009 	hlist_add_head(&pol->bydst, chain);
1010 	hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1011 	xfrm_policy_count[dir]++;
1012 	xfrm_pol_hold(pol);
1013 
1014 	if (xfrm_bydst_should_resize(dir, NULL))
1015 		schedule_work(&xfrm_hash_work);
1016 }
1017 
1018 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1019 						int dir)
1020 {
1021 	if (hlist_unhashed(&pol->bydst))
1022 		return NULL;
1023 
1024 	hlist_del(&pol->bydst);
1025 	hlist_del(&pol->byidx);
1026 	xfrm_policy_count[dir]--;
1027 
1028 	return pol;
1029 }
1030 
1031 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1032 {
1033 	write_lock_bh(&xfrm_policy_lock);
1034 	pol = __xfrm_policy_unlink(pol, dir);
1035 	write_unlock_bh(&xfrm_policy_lock);
1036 	if (pol) {
1037 		if (dir < XFRM_POLICY_MAX)
1038 			atomic_inc(&flow_cache_genid);
1039 		xfrm_policy_kill(pol);
1040 		return 0;
1041 	}
1042 	return -ENOENT;
1043 }
1044 EXPORT_SYMBOL(xfrm_policy_delete);
1045 
1046 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1047 {
1048 	struct xfrm_policy *old_pol;
1049 
1050 #ifdef CONFIG_XFRM_SUB_POLICY
1051 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1052 		return -EINVAL;
1053 #endif
1054 
1055 	write_lock_bh(&xfrm_policy_lock);
1056 	old_pol = sk->sk_policy[dir];
1057 	sk->sk_policy[dir] = pol;
1058 	if (pol) {
1059 		pol->curlft.add_time = get_seconds();
1060 		pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1061 		__xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1062 	}
1063 	if (old_pol)
1064 		__xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1065 	write_unlock_bh(&xfrm_policy_lock);
1066 
1067 	if (old_pol) {
1068 		xfrm_policy_kill(old_pol);
1069 	}
1070 	return 0;
1071 }
1072 
1073 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1074 {
1075 	struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1076 
1077 	if (newp) {
1078 		newp->selector = old->selector;
1079 		if (security_xfrm_policy_clone(old, newp)) {
1080 			kfree(newp);
1081 			return NULL;  /* ENOMEM */
1082 		}
1083 		newp->lft = old->lft;
1084 		newp->curlft = old->curlft;
1085 		newp->action = old->action;
1086 		newp->flags = old->flags;
1087 		newp->xfrm_nr = old->xfrm_nr;
1088 		newp->index = old->index;
1089 		newp->type = old->type;
1090 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1091 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1092 		write_lock_bh(&xfrm_policy_lock);
1093 		__xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1094 		write_unlock_bh(&xfrm_policy_lock);
1095 		xfrm_pol_put(newp);
1096 	}
1097 	return newp;
1098 }
1099 
1100 int __xfrm_sk_clone_policy(struct sock *sk)
1101 {
1102 	struct xfrm_policy *p0 = sk->sk_policy[0],
1103 			   *p1 = sk->sk_policy[1];
1104 
1105 	sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1106 	if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1107 		return -ENOMEM;
1108 	if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1109 		return -ENOMEM;
1110 	return 0;
1111 }
1112 
1113 static int
1114 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1115 	       unsigned short family)
1116 {
1117 	int err;
1118 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1119 
1120 	if (unlikely(afinfo == NULL))
1121 		return -EINVAL;
1122 	err = afinfo->get_saddr(local, remote);
1123 	xfrm_policy_put_afinfo(afinfo);
1124 	return err;
1125 }
1126 
1127 /* Resolve list of templates for the flow, given policy. */
1128 
1129 static int
1130 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1131 		      struct xfrm_state **xfrm,
1132 		      unsigned short family)
1133 {
1134 	int nx;
1135 	int i, error;
1136 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1137 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1138 	xfrm_address_t tmp;
1139 
1140 	for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1141 		struct xfrm_state *x;
1142 		xfrm_address_t *remote = daddr;
1143 		xfrm_address_t *local  = saddr;
1144 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1145 
1146 		if (tmpl->mode == XFRM_MODE_TUNNEL ||
1147 		    tmpl->mode == XFRM_MODE_BEET) {
1148 			remote = &tmpl->id.daddr;
1149 			local = &tmpl->saddr;
1150 			family = tmpl->encap_family;
1151 			if (xfrm_addr_any(local, family)) {
1152 				error = xfrm_get_saddr(&tmp, remote, family);
1153 				if (error)
1154 					goto fail;
1155 				local = &tmp;
1156 			}
1157 		}
1158 
1159 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1160 
1161 		if (x && x->km.state == XFRM_STATE_VALID) {
1162 			xfrm[nx++] = x;
1163 			daddr = remote;
1164 			saddr = local;
1165 			continue;
1166 		}
1167 		if (x) {
1168 			error = (x->km.state == XFRM_STATE_ERROR ?
1169 				 -EINVAL : -EAGAIN);
1170 			xfrm_state_put(x);
1171 		}
1172 
1173 		if (!tmpl->optional)
1174 			goto fail;
1175 	}
1176 	return nx;
1177 
1178 fail:
1179 	for (nx--; nx>=0; nx--)
1180 		xfrm_state_put(xfrm[nx]);
1181 	return error;
1182 }
1183 
1184 static int
1185 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1186 		  struct xfrm_state **xfrm,
1187 		  unsigned short family)
1188 {
1189 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1190 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1191 	int cnx = 0;
1192 	int error;
1193 	int ret;
1194 	int i;
1195 
1196 	for (i = 0; i < npols; i++) {
1197 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1198 			error = -ENOBUFS;
1199 			goto fail;
1200 		}
1201 
1202 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1203 		if (ret < 0) {
1204 			error = ret;
1205 			goto fail;
1206 		} else
1207 			cnx += ret;
1208 	}
1209 
1210 	/* found states are sorted for outbound processing */
1211 	if (npols > 1)
1212 		xfrm_state_sort(xfrm, tpp, cnx, family);
1213 
1214 	return cnx;
1215 
1216  fail:
1217 	for (cnx--; cnx>=0; cnx--)
1218 		xfrm_state_put(tpp[cnx]);
1219 	return error;
1220 
1221 }
1222 
1223 /* Check that the bundle accepts the flow and its components are
1224  * still valid.
1225  */
1226 
1227 static struct dst_entry *
1228 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1229 {
1230 	struct dst_entry *x;
1231 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1232 	if (unlikely(afinfo == NULL))
1233 		return ERR_PTR(-EINVAL);
1234 	x = afinfo->find_bundle(fl, policy);
1235 	xfrm_policy_put_afinfo(afinfo);
1236 	return x;
1237 }
1238 
1239 static inline int xfrm_get_tos(struct flowi *fl, int family)
1240 {
1241 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1242 	int tos;
1243 
1244 	if (!afinfo)
1245 		return -EINVAL;
1246 
1247 	tos = afinfo->get_tos(fl);
1248 
1249 	xfrm_policy_put_afinfo(afinfo);
1250 
1251 	return tos;
1252 }
1253 
1254 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1255 {
1256 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1257 	struct xfrm_dst *xdst;
1258 
1259 	if (!afinfo)
1260 		return ERR_PTR(-EINVAL);
1261 
1262 	xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1263 
1264 	xfrm_policy_put_afinfo(afinfo);
1265 
1266 	return xdst;
1267 }
1268 
1269 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1270 {
1271 	struct xfrm_policy_afinfo *afinfo =
1272 		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1273 	int err;
1274 
1275 	if (!afinfo)
1276 		return -EINVAL;
1277 
1278 	err = afinfo->fill_dst(xdst, dev);
1279 
1280 	xfrm_policy_put_afinfo(afinfo);
1281 
1282 	return err;
1283 }
1284 
1285 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1286  * all the metrics... Shortly, bundle a bundle.
1287  */
1288 
1289 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1290 					    struct xfrm_state **xfrm, int nx,
1291 					    struct flowi *fl,
1292 					    struct dst_entry *dst)
1293 {
1294 	unsigned long now = jiffies;
1295 	struct net_device *dev;
1296 	struct dst_entry *dst_prev = NULL;
1297 	struct dst_entry *dst0 = NULL;
1298 	int i = 0;
1299 	int err;
1300 	int header_len = 0;
1301 	int trailer_len = 0;
1302 	int tos;
1303 	int family = policy->selector.family;
1304 
1305 	tos = xfrm_get_tos(fl, family);
1306 	err = tos;
1307 	if (tos < 0)
1308 		goto put_states;
1309 
1310 	dst_hold(dst);
1311 
1312 	for (; i < nx; i++) {
1313 		struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1314 		struct dst_entry *dst1 = &xdst->u.dst;
1315 
1316 		err = PTR_ERR(xdst);
1317 		if (IS_ERR(xdst)) {
1318 			dst_release(dst);
1319 			goto put_states;
1320 		}
1321 
1322 		if (!dst_prev)
1323 			dst0 = dst1;
1324 		else {
1325 			dst_prev->child = dst_clone(dst1);
1326 			dst1->flags |= DST_NOHASH;
1327 		}
1328 
1329 		xdst->route = dst;
1330 		memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1331 
1332 		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1333 			family = xfrm[i]->props.family;
1334 			dst = xfrm_dst_lookup(xfrm[i], tos, family);
1335 			err = PTR_ERR(dst);
1336 			if (IS_ERR(dst))
1337 				goto put_states;
1338 		} else
1339 			dst_hold(dst);
1340 
1341 		dst1->xfrm = xfrm[i];
1342 		xdst->genid = xfrm[i]->genid;
1343 
1344 		dst1->obsolete = -1;
1345 		dst1->flags |= DST_HOST;
1346 		dst1->lastuse = now;
1347 
1348 		dst1->input = dst_discard;
1349 		dst1->output = xfrm[i]->outer_mode->afinfo->output;
1350 
1351 		dst1->next = dst_prev;
1352 		dst_prev = dst1;
1353 
1354 		header_len += xfrm[i]->props.header_len;
1355 		trailer_len += xfrm[i]->props.trailer_len;
1356 	}
1357 
1358 	dst_prev->child = dst;
1359 	dst0->path = dst;
1360 
1361 	err = -ENODEV;
1362 	dev = dst->dev;
1363 	if (!dev)
1364 		goto free_dst;
1365 
1366 	/* Copy neighbout for reachability confirmation */
1367 	dst0->neighbour = neigh_clone(dst->neighbour);
1368 
1369 	xfrm_init_pmtu(dst_prev);
1370 
1371 	for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1372 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1373 
1374 		err = xfrm_fill_dst(xdst, dev);
1375 		if (err)
1376 			goto free_dst;
1377 
1378 		dst_prev->header_len = header_len;
1379 		dst_prev->trailer_len = trailer_len;
1380 		header_len -= xdst->u.dst.xfrm->props.header_len;
1381 		trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1382 	}
1383 
1384 out:
1385 	return dst0;
1386 
1387 put_states:
1388 	for (; i < nx; i++)
1389 		xfrm_state_put(xfrm[i]);
1390 free_dst:
1391 	if (dst0)
1392 		dst_free(dst0);
1393 	dst0 = ERR_PTR(err);
1394 	goto out;
1395 }
1396 
1397 static int inline
1398 xfrm_dst_alloc_copy(void **target, void *src, int size)
1399 {
1400 	if (!*target) {
1401 		*target = kmalloc(size, GFP_ATOMIC);
1402 		if (!*target)
1403 			return -ENOMEM;
1404 	}
1405 	memcpy(*target, src, size);
1406 	return 0;
1407 }
1408 
1409 static int inline
1410 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1411 {
1412 #ifdef CONFIG_XFRM_SUB_POLICY
1413 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1414 	return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1415 				   sel, sizeof(*sel));
1416 #else
1417 	return 0;
1418 #endif
1419 }
1420 
1421 static int inline
1422 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1423 {
1424 #ifdef CONFIG_XFRM_SUB_POLICY
1425 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1426 	return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1427 #else
1428 	return 0;
1429 #endif
1430 }
1431 
1432 static int stale_bundle(struct dst_entry *dst);
1433 
1434 /* Main function: finds/creates a bundle for given flow.
1435  *
1436  * At the moment we eat a raw IP route. Mostly to speed up lookups
1437  * on interfaces with disabled IPsec.
1438  */
1439 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1440 		  struct sock *sk, int flags)
1441 {
1442 	struct xfrm_policy *policy;
1443 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1444 	int npols;
1445 	int pol_dead;
1446 	int xfrm_nr;
1447 	int pi;
1448 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1449 	struct dst_entry *dst, *dst_orig = *dst_p;
1450 	int nx = 0;
1451 	int err;
1452 	u32 genid;
1453 	u16 family;
1454 	u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1455 
1456 restart:
1457 	genid = atomic_read(&flow_cache_genid);
1458 	policy = NULL;
1459 	for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1460 		pols[pi] = NULL;
1461 	npols = 0;
1462 	pol_dead = 0;
1463 	xfrm_nr = 0;
1464 
1465 	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1466 		policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1467 		err = PTR_ERR(policy);
1468 		if (IS_ERR(policy))
1469 			goto dropdst;
1470 	}
1471 
1472 	if (!policy) {
1473 		/* To accelerate a bit...  */
1474 		if ((dst_orig->flags & DST_NOXFRM) ||
1475 		    !xfrm_policy_count[XFRM_POLICY_OUT])
1476 			return 0;
1477 
1478 		policy = flow_cache_lookup(fl, dst_orig->ops->family,
1479 					   dir, xfrm_policy_lookup);
1480 		err = PTR_ERR(policy);
1481 		if (IS_ERR(policy))
1482 			goto dropdst;
1483 	}
1484 
1485 	if (!policy)
1486 		return 0;
1487 
1488 	family = dst_orig->ops->family;
1489 	policy->curlft.use_time = get_seconds();
1490 	pols[0] = policy;
1491 	npols ++;
1492 	xfrm_nr += pols[0]->xfrm_nr;
1493 
1494 	switch (policy->action) {
1495 	default:
1496 	case XFRM_POLICY_BLOCK:
1497 		/* Prohibit the flow */
1498 		err = -EPERM;
1499 		goto error;
1500 
1501 	case XFRM_POLICY_ALLOW:
1502 #ifndef CONFIG_XFRM_SUB_POLICY
1503 		if (policy->xfrm_nr == 0) {
1504 			/* Flow passes not transformed. */
1505 			xfrm_pol_put(policy);
1506 			return 0;
1507 		}
1508 #endif
1509 
1510 		/* Try to find matching bundle.
1511 		 *
1512 		 * LATER: help from flow cache. It is optional, this
1513 		 * is required only for output policy.
1514 		 */
1515 		dst = xfrm_find_bundle(fl, policy, family);
1516 		if (IS_ERR(dst)) {
1517 			err = PTR_ERR(dst);
1518 			goto error;
1519 		}
1520 
1521 		if (dst)
1522 			break;
1523 
1524 #ifdef CONFIG_XFRM_SUB_POLICY
1525 		if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1526 			pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1527 							    fl, family,
1528 							    XFRM_POLICY_OUT);
1529 			if (pols[1]) {
1530 				if (IS_ERR(pols[1])) {
1531 					err = PTR_ERR(pols[1]);
1532 					goto error;
1533 				}
1534 				if (pols[1]->action == XFRM_POLICY_BLOCK) {
1535 					err = -EPERM;
1536 					goto error;
1537 				}
1538 				npols ++;
1539 				xfrm_nr += pols[1]->xfrm_nr;
1540 			}
1541 		}
1542 
1543 		/*
1544 		 * Because neither flowi nor bundle information knows about
1545 		 * transformation template size. On more than one policy usage
1546 		 * we can realize whether all of them is bypass or not after
1547 		 * they are searched. See above not-transformed bypass
1548 		 * is surrounded by non-sub policy configuration, too.
1549 		 */
1550 		if (xfrm_nr == 0) {
1551 			/* Flow passes not transformed. */
1552 			xfrm_pols_put(pols, npols);
1553 			return 0;
1554 		}
1555 
1556 #endif
1557 		nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1558 
1559 		if (unlikely(nx<0)) {
1560 			err = nx;
1561 			if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1562 				/* EREMOTE tells the caller to generate
1563 				 * a one-shot blackhole route.
1564 				 */
1565 				xfrm_pol_put(policy);
1566 				return -EREMOTE;
1567 			}
1568 			if (err == -EAGAIN && flags) {
1569 				DECLARE_WAITQUEUE(wait, current);
1570 
1571 				add_wait_queue(&km_waitq, &wait);
1572 				set_current_state(TASK_INTERRUPTIBLE);
1573 				schedule();
1574 				set_current_state(TASK_RUNNING);
1575 				remove_wait_queue(&km_waitq, &wait);
1576 
1577 				nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1578 
1579 				if (nx == -EAGAIN && signal_pending(current)) {
1580 					err = -ERESTART;
1581 					goto error;
1582 				}
1583 				if (nx == -EAGAIN ||
1584 				    genid != atomic_read(&flow_cache_genid)) {
1585 					xfrm_pols_put(pols, npols);
1586 					goto restart;
1587 				}
1588 				err = nx;
1589 			}
1590 			if (err < 0)
1591 				goto error;
1592 		}
1593 		if (nx == 0) {
1594 			/* Flow passes not transformed. */
1595 			xfrm_pols_put(pols, npols);
1596 			return 0;
1597 		}
1598 
1599 		dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1600 		err = PTR_ERR(dst);
1601 		if (IS_ERR(dst))
1602 			goto error;
1603 
1604 		for (pi = 0; pi < npols; pi++) {
1605 			read_lock_bh(&pols[pi]->lock);
1606 			pol_dead |= pols[pi]->dead;
1607 			read_unlock_bh(&pols[pi]->lock);
1608 		}
1609 
1610 		write_lock_bh(&policy->lock);
1611 		if (unlikely(pol_dead || stale_bundle(dst))) {
1612 			/* Wow! While we worked on resolving, this
1613 			 * policy has gone. Retry. It is not paranoia,
1614 			 * we just cannot enlist new bundle to dead object.
1615 			 * We can't enlist stable bundles either.
1616 			 */
1617 			write_unlock_bh(&policy->lock);
1618 			if (dst)
1619 				dst_free(dst);
1620 
1621 			err = -EHOSTUNREACH;
1622 			goto error;
1623 		}
1624 
1625 		if (npols > 1)
1626 			err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1627 		else
1628 			err = xfrm_dst_update_origin(dst, fl);
1629 		if (unlikely(err)) {
1630 			write_unlock_bh(&policy->lock);
1631 			if (dst)
1632 				dst_free(dst);
1633 			goto error;
1634 		}
1635 
1636 		dst->next = policy->bundles;
1637 		policy->bundles = dst;
1638 		dst_hold(dst);
1639 		write_unlock_bh(&policy->lock);
1640 	}
1641 	*dst_p = dst;
1642 	dst_release(dst_orig);
1643 	xfrm_pols_put(pols, npols);
1644 	return 0;
1645 
1646 error:
1647 	xfrm_pols_put(pols, npols);
1648 dropdst:
1649 	dst_release(dst_orig);
1650 	*dst_p = NULL;
1651 	return err;
1652 }
1653 EXPORT_SYMBOL(__xfrm_lookup);
1654 
1655 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1656 		struct sock *sk, int flags)
1657 {
1658 	int err = __xfrm_lookup(dst_p, fl, sk, flags);
1659 
1660 	if (err == -EREMOTE) {
1661 		dst_release(*dst_p);
1662 		*dst_p = NULL;
1663 		err = -EAGAIN;
1664 	}
1665 
1666 	return err;
1667 }
1668 EXPORT_SYMBOL(xfrm_lookup);
1669 
1670 static inline int
1671 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1672 {
1673 	struct xfrm_state *x;
1674 
1675 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1676 		return 0;
1677 	x = skb->sp->xvec[idx];
1678 	if (!x->type->reject)
1679 		return 0;
1680 	return x->type->reject(x, skb, fl);
1681 }
1682 
1683 /* When skb is transformed back to its "native" form, we have to
1684  * check policy restrictions. At the moment we make this in maximally
1685  * stupid way. Shame on me. :-) Of course, connected sockets must
1686  * have policy cached at them.
1687  */
1688 
1689 static inline int
1690 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1691 	      unsigned short family)
1692 {
1693 	if (xfrm_state_kern(x))
1694 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1695 	return	x->id.proto == tmpl->id.proto &&
1696 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1697 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1698 		x->props.mode == tmpl->mode &&
1699 		((tmpl->aalgos & (1<<x->props.aalgo)) ||
1700 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1701 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
1702 		  xfrm_state_addr_cmp(tmpl, x, family));
1703 }
1704 
1705 /*
1706  * 0 or more than 0 is returned when validation is succeeded (either bypass
1707  * because of optional transport mode, or next index of the mathced secpath
1708  * state with the template.
1709  * -1 is returned when no matching template is found.
1710  * Otherwise "-2 - errored_index" is returned.
1711  */
1712 static inline int
1713 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1714 	       unsigned short family)
1715 {
1716 	int idx = start;
1717 
1718 	if (tmpl->optional) {
1719 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
1720 			return start;
1721 	} else
1722 		start = -1;
1723 	for (; idx < sp->len; idx++) {
1724 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1725 			return ++idx;
1726 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1727 			if (start == -1)
1728 				start = -2-idx;
1729 			break;
1730 		}
1731 	}
1732 	return start;
1733 }
1734 
1735 int
1736 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1737 {
1738 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1739 	int err;
1740 
1741 	if (unlikely(afinfo == NULL))
1742 		return -EAFNOSUPPORT;
1743 
1744 	afinfo->decode_session(skb, fl);
1745 	err = security_xfrm_decode_session(skb, &fl->secid);
1746 	xfrm_policy_put_afinfo(afinfo);
1747 	return err;
1748 }
1749 EXPORT_SYMBOL(xfrm_decode_session);
1750 
1751 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1752 {
1753 	for (; k < sp->len; k++) {
1754 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1755 			*idxp = k;
1756 			return 1;
1757 		}
1758 	}
1759 
1760 	return 0;
1761 }
1762 
1763 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1764 			unsigned short family)
1765 {
1766 	struct xfrm_policy *pol;
1767 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1768 	int npols = 0;
1769 	int xfrm_nr;
1770 	int pi;
1771 	struct flowi fl;
1772 	u8 fl_dir = policy_to_flow_dir(dir);
1773 	int xerr_idx = -1;
1774 
1775 	if (xfrm_decode_session(skb, &fl, family) < 0)
1776 		return 0;
1777 	nf_nat_decode_session(skb, &fl, family);
1778 
1779 	/* First, check used SA against their selectors. */
1780 	if (skb->sp) {
1781 		int i;
1782 
1783 		for (i=skb->sp->len-1; i>=0; i--) {
1784 			struct xfrm_state *x = skb->sp->xvec[i];
1785 			if (!xfrm_selector_match(&x->sel, &fl, family))
1786 				return 0;
1787 		}
1788 	}
1789 
1790 	pol = NULL;
1791 	if (sk && sk->sk_policy[dir]) {
1792 		pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1793 		if (IS_ERR(pol))
1794 			return 0;
1795 	}
1796 
1797 	if (!pol)
1798 		pol = flow_cache_lookup(&fl, family, fl_dir,
1799 					xfrm_policy_lookup);
1800 
1801 	if (IS_ERR(pol))
1802 		return 0;
1803 
1804 	if (!pol) {
1805 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1806 			xfrm_secpath_reject(xerr_idx, skb, &fl);
1807 			return 0;
1808 		}
1809 		return 1;
1810 	}
1811 
1812 	pol->curlft.use_time = get_seconds();
1813 
1814 	pols[0] = pol;
1815 	npols ++;
1816 #ifdef CONFIG_XFRM_SUB_POLICY
1817 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1818 		pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1819 						    &fl, family,
1820 						    XFRM_POLICY_IN);
1821 		if (pols[1]) {
1822 			if (IS_ERR(pols[1]))
1823 				return 0;
1824 			pols[1]->curlft.use_time = get_seconds();
1825 			npols ++;
1826 		}
1827 	}
1828 #endif
1829 
1830 	if (pol->action == XFRM_POLICY_ALLOW) {
1831 		struct sec_path *sp;
1832 		static struct sec_path dummy;
1833 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1834 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1835 		struct xfrm_tmpl **tpp = tp;
1836 		int ti = 0;
1837 		int i, k;
1838 
1839 		if ((sp = skb->sp) == NULL)
1840 			sp = &dummy;
1841 
1842 		for (pi = 0; pi < npols; pi++) {
1843 			if (pols[pi] != pol &&
1844 			    pols[pi]->action != XFRM_POLICY_ALLOW)
1845 				goto reject;
1846 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1847 				goto reject_error;
1848 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
1849 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
1850 		}
1851 		xfrm_nr = ti;
1852 		if (npols > 1) {
1853 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1854 			tpp = stp;
1855 		}
1856 
1857 		/* For each tunnel xfrm, find the first matching tmpl.
1858 		 * For each tmpl before that, find corresponding xfrm.
1859 		 * Order is _important_. Later we will implement
1860 		 * some barriers, but at the moment barriers
1861 		 * are implied between each two transformations.
1862 		 */
1863 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1864 			k = xfrm_policy_ok(tpp[i], sp, k, family);
1865 			if (k < 0) {
1866 				if (k < -1)
1867 					/* "-2 - errored_index" returned */
1868 					xerr_idx = -(2+k);
1869 				goto reject;
1870 			}
1871 		}
1872 
1873 		if (secpath_has_nontransport(sp, k, &xerr_idx))
1874 			goto reject;
1875 
1876 		xfrm_pols_put(pols, npols);
1877 		return 1;
1878 	}
1879 
1880 reject:
1881 	xfrm_secpath_reject(xerr_idx, skb, &fl);
1882 reject_error:
1883 	xfrm_pols_put(pols, npols);
1884 	return 0;
1885 }
1886 EXPORT_SYMBOL(__xfrm_policy_check);
1887 
1888 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1889 {
1890 	struct flowi fl;
1891 
1892 	if (xfrm_decode_session(skb, &fl, family) < 0)
1893 		return 0;
1894 
1895 	return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1896 }
1897 EXPORT_SYMBOL(__xfrm_route_forward);
1898 
1899 /* Optimize later using cookies and generation ids. */
1900 
1901 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1902 {
1903 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1904 	 * to "-1" to force all XFRM destinations to get validated by
1905 	 * dst_ops->check on every use.  We do this because when a
1906 	 * normal route referenced by an XFRM dst is obsoleted we do
1907 	 * not go looking around for all parent referencing XFRM dsts
1908 	 * so that we can invalidate them.  It is just too much work.
1909 	 * Instead we make the checks here on every use.  For example:
1910 	 *
1911 	 *	XFRM dst A --> IPv4 dst X
1912 	 *
1913 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
1914 	 * in this example).  If X is marked obsolete, "A" will not
1915 	 * notice.  That's what we are validating here via the
1916 	 * stale_bundle() check.
1917 	 *
1918 	 * When a policy's bundle is pruned, we dst_free() the XFRM
1919 	 * dst which causes it's ->obsolete field to be set to a
1920 	 * positive non-zero integer.  If an XFRM dst has been pruned
1921 	 * like this, we want to force a new route lookup.
1922 	 */
1923 	if (dst->obsolete < 0 && !stale_bundle(dst))
1924 		return dst;
1925 
1926 	return NULL;
1927 }
1928 
1929 static int stale_bundle(struct dst_entry *dst)
1930 {
1931 	return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1932 }
1933 
1934 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1935 {
1936 	while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1937 		dst->dev = dev->nd_net->loopback_dev;
1938 		dev_hold(dst->dev);
1939 		dev_put(dev);
1940 	}
1941 }
1942 EXPORT_SYMBOL(xfrm_dst_ifdown);
1943 
1944 static void xfrm_link_failure(struct sk_buff *skb)
1945 {
1946 	/* Impossible. Such dst must be popped before reaches point of failure. */
1947 	return;
1948 }
1949 
1950 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1951 {
1952 	if (dst) {
1953 		if (dst->obsolete) {
1954 			dst_release(dst);
1955 			dst = NULL;
1956 		}
1957 	}
1958 	return dst;
1959 }
1960 
1961 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1962 {
1963 	struct dst_entry *dst, **dstp;
1964 
1965 	write_lock(&pol->lock);
1966 	dstp = &pol->bundles;
1967 	while ((dst=*dstp) != NULL) {
1968 		if (func(dst)) {
1969 			*dstp = dst->next;
1970 			dst->next = *gc_list_p;
1971 			*gc_list_p = dst;
1972 		} else {
1973 			dstp = &dst->next;
1974 		}
1975 	}
1976 	write_unlock(&pol->lock);
1977 }
1978 
1979 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1980 {
1981 	struct dst_entry *gc_list = NULL;
1982 	int dir;
1983 
1984 	read_lock_bh(&xfrm_policy_lock);
1985 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1986 		struct xfrm_policy *pol;
1987 		struct hlist_node *entry;
1988 		struct hlist_head *table;
1989 		int i;
1990 
1991 		hlist_for_each_entry(pol, entry,
1992 				     &xfrm_policy_inexact[dir], bydst)
1993 			prune_one_bundle(pol, func, &gc_list);
1994 
1995 		table = xfrm_policy_bydst[dir].table;
1996 		for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1997 			hlist_for_each_entry(pol, entry, table + i, bydst)
1998 				prune_one_bundle(pol, func, &gc_list);
1999 		}
2000 	}
2001 	read_unlock_bh(&xfrm_policy_lock);
2002 
2003 	while (gc_list) {
2004 		struct dst_entry *dst = gc_list;
2005 		gc_list = dst->next;
2006 		dst_free(dst);
2007 	}
2008 }
2009 
2010 static int unused_bundle(struct dst_entry *dst)
2011 {
2012 	return !atomic_read(&dst->__refcnt);
2013 }
2014 
2015 static void __xfrm_garbage_collect(void)
2016 {
2017 	xfrm_prune_bundles(unused_bundle);
2018 }
2019 
2020 static int xfrm_flush_bundles(void)
2021 {
2022 	xfrm_prune_bundles(stale_bundle);
2023 	return 0;
2024 }
2025 
2026 static void xfrm_init_pmtu(struct dst_entry *dst)
2027 {
2028 	do {
2029 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2030 		u32 pmtu, route_mtu_cached;
2031 
2032 		pmtu = dst_mtu(dst->child);
2033 		xdst->child_mtu_cached = pmtu;
2034 
2035 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2036 
2037 		route_mtu_cached = dst_mtu(xdst->route);
2038 		xdst->route_mtu_cached = route_mtu_cached;
2039 
2040 		if (pmtu > route_mtu_cached)
2041 			pmtu = route_mtu_cached;
2042 
2043 		dst->metrics[RTAX_MTU-1] = pmtu;
2044 	} while ((dst = dst->next));
2045 }
2046 
2047 /* Check that the bundle accepts the flow and its components are
2048  * still valid.
2049  */
2050 
2051 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2052 		struct flowi *fl, int family, int strict)
2053 {
2054 	struct dst_entry *dst = &first->u.dst;
2055 	struct xfrm_dst *last;
2056 	u32 mtu;
2057 
2058 	if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2059 	    (dst->dev && !netif_running(dst->dev)))
2060 		return 0;
2061 #ifdef CONFIG_XFRM_SUB_POLICY
2062 	if (fl) {
2063 		if (first->origin && !flow_cache_uli_match(first->origin, fl))
2064 			return 0;
2065 		if (first->partner &&
2066 		    !xfrm_selector_match(first->partner, fl, family))
2067 			return 0;
2068 	}
2069 #endif
2070 
2071 	last = NULL;
2072 
2073 	do {
2074 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2075 
2076 		if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2077 			return 0;
2078 		if (fl && pol &&
2079 		    !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2080 			return 0;
2081 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
2082 			return 0;
2083 		if (xdst->genid != dst->xfrm->genid)
2084 			return 0;
2085 
2086 		if (strict && fl &&
2087 		    !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2088 		    !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2089 			return 0;
2090 
2091 		mtu = dst_mtu(dst->child);
2092 		if (xdst->child_mtu_cached != mtu) {
2093 			last = xdst;
2094 			xdst->child_mtu_cached = mtu;
2095 		}
2096 
2097 		if (!dst_check(xdst->route, xdst->route_cookie))
2098 			return 0;
2099 		mtu = dst_mtu(xdst->route);
2100 		if (xdst->route_mtu_cached != mtu) {
2101 			last = xdst;
2102 			xdst->route_mtu_cached = mtu;
2103 		}
2104 
2105 		dst = dst->child;
2106 	} while (dst->xfrm);
2107 
2108 	if (likely(!last))
2109 		return 1;
2110 
2111 	mtu = last->child_mtu_cached;
2112 	for (;;) {
2113 		dst = &last->u.dst;
2114 
2115 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
2116 		if (mtu > last->route_mtu_cached)
2117 			mtu = last->route_mtu_cached;
2118 		dst->metrics[RTAX_MTU-1] = mtu;
2119 
2120 		if (last == first)
2121 			break;
2122 
2123 		last = (struct xfrm_dst *)last->u.dst.next;
2124 		last->child_mtu_cached = mtu;
2125 	}
2126 
2127 	return 1;
2128 }
2129 
2130 EXPORT_SYMBOL(xfrm_bundle_ok);
2131 
2132 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2133 {
2134 	int err = 0;
2135 	if (unlikely(afinfo == NULL))
2136 		return -EINVAL;
2137 	if (unlikely(afinfo->family >= NPROTO))
2138 		return -EAFNOSUPPORT;
2139 	write_lock_bh(&xfrm_policy_afinfo_lock);
2140 	if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2141 		err = -ENOBUFS;
2142 	else {
2143 		struct dst_ops *dst_ops = afinfo->dst_ops;
2144 		if (likely(dst_ops->kmem_cachep == NULL))
2145 			dst_ops->kmem_cachep = xfrm_dst_cache;
2146 		if (likely(dst_ops->check == NULL))
2147 			dst_ops->check = xfrm_dst_check;
2148 		if (likely(dst_ops->negative_advice == NULL))
2149 			dst_ops->negative_advice = xfrm_negative_advice;
2150 		if (likely(dst_ops->link_failure == NULL))
2151 			dst_ops->link_failure = xfrm_link_failure;
2152 		if (likely(afinfo->garbage_collect == NULL))
2153 			afinfo->garbage_collect = __xfrm_garbage_collect;
2154 		xfrm_policy_afinfo[afinfo->family] = afinfo;
2155 	}
2156 	write_unlock_bh(&xfrm_policy_afinfo_lock);
2157 	return err;
2158 }
2159 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2160 
2161 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2162 {
2163 	int err = 0;
2164 	if (unlikely(afinfo == NULL))
2165 		return -EINVAL;
2166 	if (unlikely(afinfo->family >= NPROTO))
2167 		return -EAFNOSUPPORT;
2168 	write_lock_bh(&xfrm_policy_afinfo_lock);
2169 	if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2170 		if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2171 			err = -EINVAL;
2172 		else {
2173 			struct dst_ops *dst_ops = afinfo->dst_ops;
2174 			xfrm_policy_afinfo[afinfo->family] = NULL;
2175 			dst_ops->kmem_cachep = NULL;
2176 			dst_ops->check = NULL;
2177 			dst_ops->negative_advice = NULL;
2178 			dst_ops->link_failure = NULL;
2179 			afinfo->garbage_collect = NULL;
2180 		}
2181 	}
2182 	write_unlock_bh(&xfrm_policy_afinfo_lock);
2183 	return err;
2184 }
2185 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2186 
2187 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2188 {
2189 	struct xfrm_policy_afinfo *afinfo;
2190 	if (unlikely(family >= NPROTO))
2191 		return NULL;
2192 	read_lock(&xfrm_policy_afinfo_lock);
2193 	afinfo = xfrm_policy_afinfo[family];
2194 	if (unlikely(!afinfo))
2195 		read_unlock(&xfrm_policy_afinfo_lock);
2196 	return afinfo;
2197 }
2198 
2199 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2200 {
2201 	read_unlock(&xfrm_policy_afinfo_lock);
2202 }
2203 
2204 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2205 {
2206 	struct net_device *dev = ptr;
2207 
2208 	if (dev->nd_net != &init_net)
2209 		return NOTIFY_DONE;
2210 
2211 	switch (event) {
2212 	case NETDEV_DOWN:
2213 		xfrm_flush_bundles();
2214 	}
2215 	return NOTIFY_DONE;
2216 }
2217 
2218 static struct notifier_block xfrm_dev_notifier = {
2219 	xfrm_dev_event,
2220 	NULL,
2221 	0
2222 };
2223 
2224 static void __init xfrm_policy_init(void)
2225 {
2226 	unsigned int hmask, sz;
2227 	int dir;
2228 
2229 	xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2230 					   sizeof(struct xfrm_dst),
2231 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2232 					   NULL);
2233 
2234 	hmask = 8 - 1;
2235 	sz = (hmask+1) * sizeof(struct hlist_head);
2236 
2237 	xfrm_policy_byidx = xfrm_hash_alloc(sz);
2238 	xfrm_idx_hmask = hmask;
2239 	if (!xfrm_policy_byidx)
2240 		panic("XFRM: failed to allocate byidx hash\n");
2241 
2242 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2243 		struct xfrm_policy_hash *htab;
2244 
2245 		INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2246 
2247 		htab = &xfrm_policy_bydst[dir];
2248 		htab->table = xfrm_hash_alloc(sz);
2249 		htab->hmask = hmask;
2250 		if (!htab->table)
2251 			panic("XFRM: failed to allocate bydst hash\n");
2252 	}
2253 
2254 	INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2255 	register_netdevice_notifier(&xfrm_dev_notifier);
2256 }
2257 
2258 void __init xfrm_init(void)
2259 {
2260 	xfrm_state_init();
2261 	xfrm_policy_init();
2262 	xfrm_input_init();
2263 }
2264 
2265 #ifdef CONFIG_AUDITSYSCALL
2266 static inline void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2267 						struct audit_buffer *audit_buf)
2268 {
2269 	struct xfrm_sec_ctx *ctx = xp->security;
2270 	struct xfrm_selector *sel = &xp->selector;
2271 
2272 	if (ctx)
2273 		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2274 				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2275 
2276 	switch(sel->family) {
2277 	case AF_INET:
2278 		audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2279 				 NIPQUAD(sel->saddr.a4));
2280 		if (sel->prefixlen_s != 32)
2281 			audit_log_format(audit_buf, " src_prefixlen=%d",
2282 					 sel->prefixlen_s);
2283 		audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2284 				 NIPQUAD(sel->daddr.a4));
2285 		if (sel->prefixlen_d != 32)
2286 			audit_log_format(audit_buf, " dst_prefixlen=%d",
2287 					 sel->prefixlen_d);
2288 		break;
2289 	case AF_INET6:
2290 		audit_log_format(audit_buf, " src=" NIP6_FMT,
2291 				 NIP6(*(struct in6_addr *)sel->saddr.a6));
2292 		if (sel->prefixlen_s != 128)
2293 			audit_log_format(audit_buf, " src_prefixlen=%d",
2294 					 sel->prefixlen_s);
2295 		audit_log_format(audit_buf, " dst=" NIP6_FMT,
2296 				 NIP6(*(struct in6_addr *)sel->daddr.a6));
2297 		if (sel->prefixlen_d != 128)
2298 			audit_log_format(audit_buf, " dst_prefixlen=%d",
2299 					 sel->prefixlen_d);
2300 		break;
2301 	}
2302 }
2303 
2304 void
2305 xfrm_audit_policy_add(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2306 {
2307 	struct audit_buffer *audit_buf;
2308 	extern int audit_enabled;
2309 
2310 	if (audit_enabled == 0)
2311 		return;
2312 	audit_buf = xfrm_audit_start(auid, sid);
2313 	if (audit_buf == NULL)
2314 		return;
2315 	audit_log_format(audit_buf, " op=SPD-add res=%u", result);
2316 	xfrm_audit_common_policyinfo(xp, audit_buf);
2317 	audit_log_end(audit_buf);
2318 }
2319 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2320 
2321 void
2322 xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2323 {
2324 	struct audit_buffer *audit_buf;
2325 	extern int audit_enabled;
2326 
2327 	if (audit_enabled == 0)
2328 		return;
2329 	audit_buf = xfrm_audit_start(auid, sid);
2330 	if (audit_buf == NULL)
2331 		return;
2332 	audit_log_format(audit_buf, " op=SPD-delete res=%u", result);
2333 	xfrm_audit_common_policyinfo(xp, audit_buf);
2334 	audit_log_end(audit_buf);
2335 }
2336 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2337 #endif
2338 
2339 #ifdef CONFIG_XFRM_MIGRATE
2340 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2341 				       struct xfrm_selector *sel_tgt)
2342 {
2343 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2344 		if (sel_tgt->family == sel_cmp->family &&
2345 		    xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2346 				  sel_cmp->family) == 0 &&
2347 		    xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2348 				  sel_cmp->family) == 0 &&
2349 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2350 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2351 			return 1;
2352 		}
2353 	} else {
2354 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2355 			return 1;
2356 		}
2357 	}
2358 	return 0;
2359 }
2360 
2361 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2362 						     u8 dir, u8 type)
2363 {
2364 	struct xfrm_policy *pol, *ret = NULL;
2365 	struct hlist_node *entry;
2366 	struct hlist_head *chain;
2367 	u32 priority = ~0U;
2368 
2369 	read_lock_bh(&xfrm_policy_lock);
2370 	chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2371 	hlist_for_each_entry(pol, entry, chain, bydst) {
2372 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2373 		    pol->type == type) {
2374 			ret = pol;
2375 			priority = ret->priority;
2376 			break;
2377 		}
2378 	}
2379 	chain = &xfrm_policy_inexact[dir];
2380 	hlist_for_each_entry(pol, entry, chain, bydst) {
2381 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2382 		    pol->type == type &&
2383 		    pol->priority < priority) {
2384 			ret = pol;
2385 			break;
2386 		}
2387 	}
2388 
2389 	if (ret)
2390 		xfrm_pol_hold(ret);
2391 
2392 	read_unlock_bh(&xfrm_policy_lock);
2393 
2394 	return ret;
2395 }
2396 
2397 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2398 {
2399 	int match = 0;
2400 
2401 	if (t->mode == m->mode && t->id.proto == m->proto &&
2402 	    (m->reqid == 0 || t->reqid == m->reqid)) {
2403 		switch (t->mode) {
2404 		case XFRM_MODE_TUNNEL:
2405 		case XFRM_MODE_BEET:
2406 			if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2407 					  m->old_family) == 0 &&
2408 			    xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2409 					  m->old_family) == 0) {
2410 				match = 1;
2411 			}
2412 			break;
2413 		case XFRM_MODE_TRANSPORT:
2414 			/* in case of transport mode, template does not store
2415 			   any IP addresses, hence we just compare mode and
2416 			   protocol */
2417 			match = 1;
2418 			break;
2419 		default:
2420 			break;
2421 		}
2422 	}
2423 	return match;
2424 }
2425 
2426 /* update endpoint address(es) of template(s) */
2427 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2428 			       struct xfrm_migrate *m, int num_migrate)
2429 {
2430 	struct xfrm_migrate *mp;
2431 	struct dst_entry *dst;
2432 	int i, j, n = 0;
2433 
2434 	write_lock_bh(&pol->lock);
2435 	if (unlikely(pol->dead)) {
2436 		/* target policy has been deleted */
2437 		write_unlock_bh(&pol->lock);
2438 		return -ENOENT;
2439 	}
2440 
2441 	for (i = 0; i < pol->xfrm_nr; i++) {
2442 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2443 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2444 				continue;
2445 			n++;
2446 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2447 			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2448 				continue;
2449 			/* update endpoints */
2450 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2451 			       sizeof(pol->xfrm_vec[i].id.daddr));
2452 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2453 			       sizeof(pol->xfrm_vec[i].saddr));
2454 			pol->xfrm_vec[i].encap_family = mp->new_family;
2455 			/* flush bundles */
2456 			while ((dst = pol->bundles) != NULL) {
2457 				pol->bundles = dst->next;
2458 				dst_free(dst);
2459 			}
2460 		}
2461 	}
2462 
2463 	write_unlock_bh(&pol->lock);
2464 
2465 	if (!n)
2466 		return -ENODATA;
2467 
2468 	return 0;
2469 }
2470 
2471 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2472 {
2473 	int i, j;
2474 
2475 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2476 		return -EINVAL;
2477 
2478 	for (i = 0; i < num_migrate; i++) {
2479 		if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2480 				   m[i].old_family) == 0) &&
2481 		    (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2482 				   m[i].old_family) == 0))
2483 			return -EINVAL;
2484 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2485 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2486 			return -EINVAL;
2487 
2488 		/* check if there is any duplicated entry */
2489 		for (j = i + 1; j < num_migrate; j++) {
2490 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2491 				    sizeof(m[i].old_daddr)) &&
2492 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2493 				    sizeof(m[i].old_saddr)) &&
2494 			    m[i].proto == m[j].proto &&
2495 			    m[i].mode == m[j].mode &&
2496 			    m[i].reqid == m[j].reqid &&
2497 			    m[i].old_family == m[j].old_family)
2498 				return -EINVAL;
2499 		}
2500 	}
2501 
2502 	return 0;
2503 }
2504 
2505 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2506 		 struct xfrm_migrate *m, int num_migrate)
2507 {
2508 	int i, err, nx_cur = 0, nx_new = 0;
2509 	struct xfrm_policy *pol = NULL;
2510 	struct xfrm_state *x, *xc;
2511 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2512 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2513 	struct xfrm_migrate *mp;
2514 
2515 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2516 		goto out;
2517 
2518 	/* Stage 1 - find policy */
2519 	if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2520 		err = -ENOENT;
2521 		goto out;
2522 	}
2523 
2524 	/* Stage 2 - find and update state(s) */
2525 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2526 		if ((x = xfrm_migrate_state_find(mp))) {
2527 			x_cur[nx_cur] = x;
2528 			nx_cur++;
2529 			if ((xc = xfrm_state_migrate(x, mp))) {
2530 				x_new[nx_new] = xc;
2531 				nx_new++;
2532 			} else {
2533 				err = -ENODATA;
2534 				goto restore_state;
2535 			}
2536 		}
2537 	}
2538 
2539 	/* Stage 3 - update policy */
2540 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2541 		goto restore_state;
2542 
2543 	/* Stage 4 - delete old state(s) */
2544 	if (nx_cur) {
2545 		xfrm_states_put(x_cur, nx_cur);
2546 		xfrm_states_delete(x_cur, nx_cur);
2547 	}
2548 
2549 	/* Stage 5 - announce */
2550 	km_migrate(sel, dir, type, m, num_migrate);
2551 
2552 	xfrm_pol_put(pol);
2553 
2554 	return 0;
2555 out:
2556 	return err;
2557 
2558 restore_state:
2559 	if (pol)
2560 		xfrm_pol_put(pol);
2561 	if (nx_cur)
2562 		xfrm_states_put(x_cur, nx_cur);
2563 	if (nx_new)
2564 		xfrm_states_delete(x_new, nx_new);
2565 
2566 	return err;
2567 }
2568 EXPORT_SYMBOL(xfrm_migrate);
2569 #endif
2570