xref: /linux/net/xfrm/xfrm_policy.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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/slab.h>
17 #include <linux/kmod.h>
18 #include <linux/list.h>
19 #include <linux/spinlock.h>
20 #include <linux/workqueue.h>
21 #include <linux/notifier.h>
22 #include <linux/netdevice.h>
23 #include <linux/netfilter.h>
24 #include <linux/module.h>
25 #include <linux/cache.h>
26 #include <net/xfrm.h>
27 #include <net/ip.h>
28 #include <linux/audit.h>
29 
30 #include "xfrm_hash.h"
31 
32 DEFINE_MUTEX(xfrm_cfg_mutex);
33 EXPORT_SYMBOL(xfrm_cfg_mutex);
34 
35 static DEFINE_RWLOCK(xfrm_policy_lock);
36 
37 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
38 EXPORT_SYMBOL(xfrm_policy_count);
39 
40 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
41 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
42 
43 static struct kmem_cache *xfrm_dst_cache __read_mostly;
44 
45 static struct work_struct xfrm_policy_gc_work;
46 static HLIST_HEAD(xfrm_policy_gc_list);
47 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
48 
49 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
50 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
51 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
52 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
53 
54 static inline int
55 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
56 {
57 	return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
58 		addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
59 		!((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
60 		!((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
61 		(fl->proto == sel->proto || !sel->proto) &&
62 		(fl->oif == sel->ifindex || !sel->ifindex);
63 }
64 
65 static inline int
66 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
67 {
68 	return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
69 		addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
70 		!((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
71 		!((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
72 		(fl->proto == sel->proto || !sel->proto) &&
73 		(fl->oif == sel->ifindex || !sel->ifindex);
74 }
75 
76 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
77 		    unsigned short family)
78 {
79 	switch (family) {
80 	case AF_INET:
81 		return __xfrm4_selector_match(sel, fl);
82 	case AF_INET6:
83 		return __xfrm6_selector_match(sel, fl);
84 	}
85 	return 0;
86 }
87 
88 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
89 {
90 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
91 	struct xfrm_type **typemap;
92 	int err = 0;
93 
94 	if (unlikely(afinfo == NULL))
95 		return -EAFNOSUPPORT;
96 	typemap = afinfo->type_map;
97 
98 	if (likely(typemap[type->proto] == NULL))
99 		typemap[type->proto] = type;
100 	else
101 		err = -EEXIST;
102 	xfrm_policy_unlock_afinfo(afinfo);
103 	return err;
104 }
105 EXPORT_SYMBOL(xfrm_register_type);
106 
107 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
108 {
109 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
110 	struct xfrm_type **typemap;
111 	int err = 0;
112 
113 	if (unlikely(afinfo == NULL))
114 		return -EAFNOSUPPORT;
115 	typemap = afinfo->type_map;
116 
117 	if (unlikely(typemap[type->proto] != type))
118 		err = -ENOENT;
119 	else
120 		typemap[type->proto] = NULL;
121 	xfrm_policy_unlock_afinfo(afinfo);
122 	return err;
123 }
124 EXPORT_SYMBOL(xfrm_unregister_type);
125 
126 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
127 {
128 	struct xfrm_policy_afinfo *afinfo;
129 	struct xfrm_type **typemap;
130 	struct xfrm_type *type;
131 	int modload_attempted = 0;
132 
133 retry:
134 	afinfo = xfrm_policy_get_afinfo(family);
135 	if (unlikely(afinfo == NULL))
136 		return NULL;
137 	typemap = afinfo->type_map;
138 
139 	type = typemap[proto];
140 	if (unlikely(type && !try_module_get(type->owner)))
141 		type = NULL;
142 	if (!type && !modload_attempted) {
143 		xfrm_policy_put_afinfo(afinfo);
144 		request_module("xfrm-type-%d-%d",
145 			       (int) family, (int) proto);
146 		modload_attempted = 1;
147 		goto retry;
148 	}
149 
150 	xfrm_policy_put_afinfo(afinfo);
151 	return type;
152 }
153 
154 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
155 		    unsigned short family)
156 {
157 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
158 	int err = 0;
159 
160 	if (unlikely(afinfo == NULL))
161 		return -EAFNOSUPPORT;
162 
163 	if (likely(afinfo->dst_lookup != NULL))
164 		err = afinfo->dst_lookup(dst, fl);
165 	else
166 		err = -EINVAL;
167 	xfrm_policy_put_afinfo(afinfo);
168 	return err;
169 }
170 EXPORT_SYMBOL(xfrm_dst_lookup);
171 
172 void xfrm_put_type(struct xfrm_type *type)
173 {
174 	module_put(type->owner);
175 }
176 
177 int xfrm_register_mode(struct xfrm_mode *mode, int family)
178 {
179 	struct xfrm_policy_afinfo *afinfo;
180 	struct xfrm_mode **modemap;
181 	int err;
182 
183 	if (unlikely(mode->encap >= XFRM_MODE_MAX))
184 		return -EINVAL;
185 
186 	afinfo = xfrm_policy_lock_afinfo(family);
187 	if (unlikely(afinfo == NULL))
188 		return -EAFNOSUPPORT;
189 
190 	err = -EEXIST;
191 	modemap = afinfo->mode_map;
192 	if (likely(modemap[mode->encap] == NULL)) {
193 		modemap[mode->encap] = mode;
194 		err = 0;
195 	}
196 
197 	xfrm_policy_unlock_afinfo(afinfo);
198 	return err;
199 }
200 EXPORT_SYMBOL(xfrm_register_mode);
201 
202 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
203 {
204 	struct xfrm_policy_afinfo *afinfo;
205 	struct xfrm_mode **modemap;
206 	int err;
207 
208 	if (unlikely(mode->encap >= XFRM_MODE_MAX))
209 		return -EINVAL;
210 
211 	afinfo = xfrm_policy_lock_afinfo(family);
212 	if (unlikely(afinfo == NULL))
213 		return -EAFNOSUPPORT;
214 
215 	err = -ENOENT;
216 	modemap = afinfo->mode_map;
217 	if (likely(modemap[mode->encap] == mode)) {
218 		modemap[mode->encap] = NULL;
219 		err = 0;
220 	}
221 
222 	xfrm_policy_unlock_afinfo(afinfo);
223 	return err;
224 }
225 EXPORT_SYMBOL(xfrm_unregister_mode);
226 
227 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
228 {
229 	struct xfrm_policy_afinfo *afinfo;
230 	struct xfrm_mode *mode;
231 	int modload_attempted = 0;
232 
233 	if (unlikely(encap >= XFRM_MODE_MAX))
234 		return NULL;
235 
236 retry:
237 	afinfo = xfrm_policy_get_afinfo(family);
238 	if (unlikely(afinfo == NULL))
239 		return NULL;
240 
241 	mode = afinfo->mode_map[encap];
242 	if (unlikely(mode && !try_module_get(mode->owner)))
243 		mode = NULL;
244 	if (!mode && !modload_attempted) {
245 		xfrm_policy_put_afinfo(afinfo);
246 		request_module("xfrm-mode-%d-%d", family, encap);
247 		modload_attempted = 1;
248 		goto retry;
249 	}
250 
251 	xfrm_policy_put_afinfo(afinfo);
252 	return mode;
253 }
254 
255 void xfrm_put_mode(struct xfrm_mode *mode)
256 {
257 	module_put(mode->owner);
258 }
259 
260 static inline unsigned long make_jiffies(long secs)
261 {
262 	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
263 		return MAX_SCHEDULE_TIMEOUT-1;
264 	else
265 		return secs*HZ;
266 }
267 
268 static void xfrm_policy_timer(unsigned long data)
269 {
270 	struct xfrm_policy *xp = (struct xfrm_policy*)data;
271 	unsigned long now = (unsigned long)xtime.tv_sec;
272 	long next = LONG_MAX;
273 	int warn = 0;
274 	int dir;
275 
276 	read_lock(&xp->lock);
277 
278 	if (xp->dead)
279 		goto out;
280 
281 	dir = xfrm_policy_id2dir(xp->index);
282 
283 	if (xp->lft.hard_add_expires_seconds) {
284 		long tmo = xp->lft.hard_add_expires_seconds +
285 			xp->curlft.add_time - now;
286 		if (tmo <= 0)
287 			goto expired;
288 		if (tmo < next)
289 			next = tmo;
290 	}
291 	if (xp->lft.hard_use_expires_seconds) {
292 		long tmo = xp->lft.hard_use_expires_seconds +
293 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
294 		if (tmo <= 0)
295 			goto expired;
296 		if (tmo < next)
297 			next = tmo;
298 	}
299 	if (xp->lft.soft_add_expires_seconds) {
300 		long tmo = xp->lft.soft_add_expires_seconds +
301 			xp->curlft.add_time - now;
302 		if (tmo <= 0) {
303 			warn = 1;
304 			tmo = XFRM_KM_TIMEOUT;
305 		}
306 		if (tmo < next)
307 			next = tmo;
308 	}
309 	if (xp->lft.soft_use_expires_seconds) {
310 		long tmo = xp->lft.soft_use_expires_seconds +
311 			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
312 		if (tmo <= 0) {
313 			warn = 1;
314 			tmo = XFRM_KM_TIMEOUT;
315 		}
316 		if (tmo < next)
317 			next = tmo;
318 	}
319 
320 	if (warn)
321 		km_policy_expired(xp, dir, 0, 0);
322 	if (next != LONG_MAX &&
323 	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
324 		xfrm_pol_hold(xp);
325 
326 out:
327 	read_unlock(&xp->lock);
328 	xfrm_pol_put(xp);
329 	return;
330 
331 expired:
332 	read_unlock(&xp->lock);
333 	if (!xfrm_policy_delete(xp, dir))
334 		km_policy_expired(xp, dir, 1, 0);
335 	xfrm_pol_put(xp);
336 }
337 
338 
339 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
340  * SPD calls.
341  */
342 
343 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
344 {
345 	struct xfrm_policy *policy;
346 
347 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
348 
349 	if (policy) {
350 		INIT_HLIST_NODE(&policy->bydst);
351 		INIT_HLIST_NODE(&policy->byidx);
352 		rwlock_init(&policy->lock);
353 		atomic_set(&policy->refcnt, 1);
354 		init_timer(&policy->timer);
355 		policy->timer.data = (unsigned long)policy;
356 		policy->timer.function = xfrm_policy_timer;
357 	}
358 	return policy;
359 }
360 EXPORT_SYMBOL(xfrm_policy_alloc);
361 
362 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
363 
364 void __xfrm_policy_destroy(struct xfrm_policy *policy)
365 {
366 	BUG_ON(!policy->dead);
367 
368 	BUG_ON(policy->bundles);
369 
370 	if (del_timer(&policy->timer))
371 		BUG();
372 
373 	security_xfrm_policy_free(policy);
374 	kfree(policy);
375 }
376 EXPORT_SYMBOL(__xfrm_policy_destroy);
377 
378 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
379 {
380 	struct dst_entry *dst;
381 
382 	while ((dst = policy->bundles) != NULL) {
383 		policy->bundles = dst->next;
384 		dst_free(dst);
385 	}
386 
387 	if (del_timer(&policy->timer))
388 		atomic_dec(&policy->refcnt);
389 
390 	if (atomic_read(&policy->refcnt) > 1)
391 		flow_cache_flush();
392 
393 	xfrm_pol_put(policy);
394 }
395 
396 static void xfrm_policy_gc_task(struct work_struct *work)
397 {
398 	struct xfrm_policy *policy;
399 	struct hlist_node *entry, *tmp;
400 	struct hlist_head gc_list;
401 
402 	spin_lock_bh(&xfrm_policy_gc_lock);
403 	gc_list.first = xfrm_policy_gc_list.first;
404 	INIT_HLIST_HEAD(&xfrm_policy_gc_list);
405 	spin_unlock_bh(&xfrm_policy_gc_lock);
406 
407 	hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
408 		xfrm_policy_gc_kill(policy);
409 }
410 
411 /* Rule must be locked. Release descentant resources, announce
412  * entry dead. The rule must be unlinked from lists to the moment.
413  */
414 
415 static void xfrm_policy_kill(struct xfrm_policy *policy)
416 {
417 	int dead;
418 
419 	write_lock_bh(&policy->lock);
420 	dead = policy->dead;
421 	policy->dead = 1;
422 	write_unlock_bh(&policy->lock);
423 
424 	if (unlikely(dead)) {
425 		WARN_ON(1);
426 		return;
427 	}
428 
429 	spin_lock(&xfrm_policy_gc_lock);
430 	hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
431 	spin_unlock(&xfrm_policy_gc_lock);
432 
433 	schedule_work(&xfrm_policy_gc_work);
434 }
435 
436 struct xfrm_policy_hash {
437 	struct hlist_head	*table;
438 	unsigned int		hmask;
439 };
440 
441 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
442 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
443 static struct hlist_head *xfrm_policy_byidx __read_mostly;
444 static unsigned int xfrm_idx_hmask __read_mostly;
445 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
446 
447 static inline unsigned int idx_hash(u32 index)
448 {
449 	return __idx_hash(index, xfrm_idx_hmask);
450 }
451 
452 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
453 {
454 	unsigned int hmask = xfrm_policy_bydst[dir].hmask;
455 	unsigned int hash = __sel_hash(sel, family, hmask);
456 
457 	return (hash == hmask + 1 ?
458 		&xfrm_policy_inexact[dir] :
459 		xfrm_policy_bydst[dir].table + hash);
460 }
461 
462 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
463 {
464 	unsigned int hmask = xfrm_policy_bydst[dir].hmask;
465 	unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
466 
467 	return xfrm_policy_bydst[dir].table + hash;
468 }
469 
470 static void xfrm_dst_hash_transfer(struct hlist_head *list,
471 				   struct hlist_head *ndsttable,
472 				   unsigned int nhashmask)
473 {
474 	struct hlist_node *entry, *tmp;
475 	struct xfrm_policy *pol;
476 
477 	hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
478 		unsigned int h;
479 
480 		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
481 				pol->family, nhashmask);
482 		hlist_add_head(&pol->bydst, ndsttable+h);
483 	}
484 }
485 
486 static void xfrm_idx_hash_transfer(struct hlist_head *list,
487 				   struct hlist_head *nidxtable,
488 				   unsigned int nhashmask)
489 {
490 	struct hlist_node *entry, *tmp;
491 	struct xfrm_policy *pol;
492 
493 	hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
494 		unsigned int h;
495 
496 		h = __idx_hash(pol->index, nhashmask);
497 		hlist_add_head(&pol->byidx, nidxtable+h);
498 	}
499 }
500 
501 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
502 {
503 	return ((old_hmask + 1) << 1) - 1;
504 }
505 
506 static void xfrm_bydst_resize(int dir)
507 {
508 	unsigned int hmask = xfrm_policy_bydst[dir].hmask;
509 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
510 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
511 	struct hlist_head *odst = xfrm_policy_bydst[dir].table;
512 	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
513 	int i;
514 
515 	if (!ndst)
516 		return;
517 
518 	write_lock_bh(&xfrm_policy_lock);
519 
520 	for (i = hmask; i >= 0; i--)
521 		xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
522 
523 	xfrm_policy_bydst[dir].table = ndst;
524 	xfrm_policy_bydst[dir].hmask = nhashmask;
525 
526 	write_unlock_bh(&xfrm_policy_lock);
527 
528 	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
529 }
530 
531 static void xfrm_byidx_resize(int total)
532 {
533 	unsigned int hmask = xfrm_idx_hmask;
534 	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
535 	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
536 	struct hlist_head *oidx = xfrm_policy_byidx;
537 	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
538 	int i;
539 
540 	if (!nidx)
541 		return;
542 
543 	write_lock_bh(&xfrm_policy_lock);
544 
545 	for (i = hmask; i >= 0; i--)
546 		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
547 
548 	xfrm_policy_byidx = nidx;
549 	xfrm_idx_hmask = nhashmask;
550 
551 	write_unlock_bh(&xfrm_policy_lock);
552 
553 	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
554 }
555 
556 static inline int xfrm_bydst_should_resize(int dir, int *total)
557 {
558 	unsigned int cnt = xfrm_policy_count[dir];
559 	unsigned int hmask = xfrm_policy_bydst[dir].hmask;
560 
561 	if (total)
562 		*total += cnt;
563 
564 	if ((hmask + 1) < xfrm_policy_hashmax &&
565 	    cnt > hmask)
566 		return 1;
567 
568 	return 0;
569 }
570 
571 static inline int xfrm_byidx_should_resize(int total)
572 {
573 	unsigned int hmask = xfrm_idx_hmask;
574 
575 	if ((hmask + 1) < xfrm_policy_hashmax &&
576 	    total > hmask)
577 		return 1;
578 
579 	return 0;
580 }
581 
582 static DEFINE_MUTEX(hash_resize_mutex);
583 
584 static void xfrm_hash_resize(struct work_struct *__unused)
585 {
586 	int dir, total;
587 
588 	mutex_lock(&hash_resize_mutex);
589 
590 	total = 0;
591 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
592 		if (xfrm_bydst_should_resize(dir, &total))
593 			xfrm_bydst_resize(dir);
594 	}
595 	if (xfrm_byidx_should_resize(total))
596 		xfrm_byidx_resize(total);
597 
598 	mutex_unlock(&hash_resize_mutex);
599 }
600 
601 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
602 
603 /* Generate new index... KAME seems to generate them ordered by cost
604  * of an absolute inpredictability of ordering of rules. This will not pass. */
605 static u32 xfrm_gen_index(u8 type, int dir)
606 {
607 	static u32 idx_generator;
608 
609 	for (;;) {
610 		struct hlist_node *entry;
611 		struct hlist_head *list;
612 		struct xfrm_policy *p;
613 		u32 idx;
614 		int found;
615 
616 		idx = (idx_generator | dir);
617 		idx_generator += 8;
618 		if (idx == 0)
619 			idx = 8;
620 		list = xfrm_policy_byidx + idx_hash(idx);
621 		found = 0;
622 		hlist_for_each_entry(p, entry, list, byidx) {
623 			if (p->index == idx) {
624 				found = 1;
625 				break;
626 			}
627 		}
628 		if (!found)
629 			return idx;
630 	}
631 }
632 
633 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
634 {
635 	u32 *p1 = (u32 *) s1;
636 	u32 *p2 = (u32 *) s2;
637 	int len = sizeof(struct xfrm_selector) / sizeof(u32);
638 	int i;
639 
640 	for (i = 0; i < len; i++) {
641 		if (p1[i] != p2[i])
642 			return 1;
643 	}
644 
645 	return 0;
646 }
647 
648 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
649 {
650 	struct xfrm_policy *pol;
651 	struct xfrm_policy *delpol;
652 	struct hlist_head *chain;
653 	struct hlist_node *entry, *newpos;
654 	struct dst_entry *gc_list;
655 
656 	write_lock_bh(&xfrm_policy_lock);
657 	chain = policy_hash_bysel(&policy->selector, policy->family, dir);
658 	delpol = NULL;
659 	newpos = NULL;
660 	hlist_for_each_entry(pol, entry, chain, bydst) {
661 		if (pol->type == policy->type &&
662 		    !selector_cmp(&pol->selector, &policy->selector) &&
663 		    xfrm_sec_ctx_match(pol->security, policy->security) &&
664 		    !WARN_ON(delpol)) {
665 			if (excl) {
666 				write_unlock_bh(&xfrm_policy_lock);
667 				return -EEXIST;
668 			}
669 			delpol = pol;
670 			if (policy->priority > pol->priority)
671 				continue;
672 		} else if (policy->priority >= pol->priority) {
673 			newpos = &pol->bydst;
674 			continue;
675 		}
676 		if (delpol)
677 			break;
678 	}
679 	if (newpos)
680 		hlist_add_after(newpos, &policy->bydst);
681 	else
682 		hlist_add_head(&policy->bydst, chain);
683 	xfrm_pol_hold(policy);
684 	xfrm_policy_count[dir]++;
685 	atomic_inc(&flow_cache_genid);
686 	if (delpol) {
687 		hlist_del(&delpol->bydst);
688 		hlist_del(&delpol->byidx);
689 		xfrm_policy_count[dir]--;
690 	}
691 	policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
692 	hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
693 	policy->curlft.add_time = (unsigned long)xtime.tv_sec;
694 	policy->curlft.use_time = 0;
695 	if (!mod_timer(&policy->timer, jiffies + HZ))
696 		xfrm_pol_hold(policy);
697 	write_unlock_bh(&xfrm_policy_lock);
698 
699 	if (delpol)
700 		xfrm_policy_kill(delpol);
701 	else if (xfrm_bydst_should_resize(dir, NULL))
702 		schedule_work(&xfrm_hash_work);
703 
704 	read_lock_bh(&xfrm_policy_lock);
705 	gc_list = NULL;
706 	entry = &policy->bydst;
707 	hlist_for_each_entry_continue(policy, entry, bydst) {
708 		struct dst_entry *dst;
709 
710 		write_lock(&policy->lock);
711 		dst = policy->bundles;
712 		if (dst) {
713 			struct dst_entry *tail = dst;
714 			while (tail->next)
715 				tail = tail->next;
716 			tail->next = gc_list;
717 			gc_list = dst;
718 
719 			policy->bundles = NULL;
720 		}
721 		write_unlock(&policy->lock);
722 	}
723 	read_unlock_bh(&xfrm_policy_lock);
724 
725 	while (gc_list) {
726 		struct dst_entry *dst = gc_list;
727 
728 		gc_list = dst->next;
729 		dst_free(dst);
730 	}
731 
732 	return 0;
733 }
734 EXPORT_SYMBOL(xfrm_policy_insert);
735 
736 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
737 					  struct xfrm_selector *sel,
738 					  struct xfrm_sec_ctx *ctx, int delete,
739 					  int *err)
740 {
741 	struct xfrm_policy *pol, *ret;
742 	struct hlist_head *chain;
743 	struct hlist_node *entry;
744 
745 	*err = 0;
746 	write_lock_bh(&xfrm_policy_lock);
747 	chain = policy_hash_bysel(sel, sel->family, dir);
748 	ret = NULL;
749 	hlist_for_each_entry(pol, entry, chain, bydst) {
750 		if (pol->type == type &&
751 		    !selector_cmp(sel, &pol->selector) &&
752 		    xfrm_sec_ctx_match(ctx, pol->security)) {
753 			xfrm_pol_hold(pol);
754 			if (delete) {
755 				*err = security_xfrm_policy_delete(pol);
756 				if (*err) {
757 					write_unlock_bh(&xfrm_policy_lock);
758 					return pol;
759 				}
760 				hlist_del(&pol->bydst);
761 				hlist_del(&pol->byidx);
762 				xfrm_policy_count[dir]--;
763 			}
764 			ret = pol;
765 			break;
766 		}
767 	}
768 	write_unlock_bh(&xfrm_policy_lock);
769 
770 	if (ret && delete) {
771 		atomic_inc(&flow_cache_genid);
772 		xfrm_policy_kill(ret);
773 	}
774 	return ret;
775 }
776 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
777 
778 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
779 				     int *err)
780 {
781 	struct xfrm_policy *pol, *ret;
782 	struct hlist_head *chain;
783 	struct hlist_node *entry;
784 
785 	*err = 0;
786 	write_lock_bh(&xfrm_policy_lock);
787 	chain = xfrm_policy_byidx + idx_hash(id);
788 	ret = NULL;
789 	hlist_for_each_entry(pol, entry, chain, byidx) {
790 		if (pol->type == type && pol->index == id) {
791 			xfrm_pol_hold(pol);
792 			if (delete) {
793 				*err = security_xfrm_policy_delete(pol);
794 				if (*err) {
795 					write_unlock_bh(&xfrm_policy_lock);
796 					return pol;
797 				}
798 				hlist_del(&pol->bydst);
799 				hlist_del(&pol->byidx);
800 				xfrm_policy_count[dir]--;
801 			}
802 			ret = pol;
803 			break;
804 		}
805 	}
806 	write_unlock_bh(&xfrm_policy_lock);
807 
808 	if (ret && delete) {
809 		atomic_inc(&flow_cache_genid);
810 		xfrm_policy_kill(ret);
811 	}
812 	return ret;
813 }
814 EXPORT_SYMBOL(xfrm_policy_byid);
815 
816 void xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
817 {
818 	int dir;
819 
820 	write_lock_bh(&xfrm_policy_lock);
821 	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
822 		struct xfrm_policy *pol;
823 		struct hlist_node *entry;
824 		int i, killed;
825 
826 		killed = 0;
827 	again1:
828 		hlist_for_each_entry(pol, entry,
829 				     &xfrm_policy_inexact[dir], bydst) {
830 			if (pol->type != type)
831 				continue;
832 			hlist_del(&pol->bydst);
833 			hlist_del(&pol->byidx);
834 			write_unlock_bh(&xfrm_policy_lock);
835 
836 			xfrm_audit_log(audit_info->loginuid, audit_info->secid,
837 				       AUDIT_MAC_IPSEC_DELSPD, 1, pol, NULL);
838 
839 			xfrm_policy_kill(pol);
840 			killed++;
841 
842 			write_lock_bh(&xfrm_policy_lock);
843 			goto again1;
844 		}
845 
846 		for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
847 	again2:
848 			hlist_for_each_entry(pol, entry,
849 					     xfrm_policy_bydst[dir].table + i,
850 					     bydst) {
851 				if (pol->type != type)
852 					continue;
853 				hlist_del(&pol->bydst);
854 				hlist_del(&pol->byidx);
855 				write_unlock_bh(&xfrm_policy_lock);
856 
857 				xfrm_audit_log(audit_info->loginuid,
858 					       audit_info->secid,
859 					       AUDIT_MAC_IPSEC_DELSPD, 1,
860 					       pol, NULL);
861 
862 				xfrm_policy_kill(pol);
863 				killed++;
864 
865 				write_lock_bh(&xfrm_policy_lock);
866 				goto again2;
867 			}
868 		}
869 
870 		xfrm_policy_count[dir] -= killed;
871 	}
872 	atomic_inc(&flow_cache_genid);
873 	write_unlock_bh(&xfrm_policy_lock);
874 }
875 EXPORT_SYMBOL(xfrm_policy_flush);
876 
877 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
878 		     void *data)
879 {
880 	struct xfrm_policy *pol, *last = NULL;
881 	struct hlist_node *entry;
882 	int dir, last_dir = 0, count, error;
883 
884 	read_lock_bh(&xfrm_policy_lock);
885 	count = 0;
886 
887 	for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
888 		struct hlist_head *table = xfrm_policy_bydst[dir].table;
889 		int i;
890 
891 		hlist_for_each_entry(pol, entry,
892 				     &xfrm_policy_inexact[dir], bydst) {
893 			if (pol->type != type)
894 				continue;
895 			if (last) {
896 				error = func(last, last_dir % XFRM_POLICY_MAX,
897 					     count, data);
898 				if (error)
899 					goto out;
900 			}
901 			last = pol;
902 			last_dir = dir;
903 			count++;
904 		}
905 		for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
906 			hlist_for_each_entry(pol, entry, table + i, bydst) {
907 				if (pol->type != type)
908 					continue;
909 				if (last) {
910 					error = func(last, last_dir % XFRM_POLICY_MAX,
911 						     count, data);
912 					if (error)
913 						goto out;
914 				}
915 				last = pol;
916 				last_dir = dir;
917 				count++;
918 			}
919 		}
920 	}
921 	if (count == 0) {
922 		error = -ENOENT;
923 		goto out;
924 	}
925 	error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
926 out:
927 	read_unlock_bh(&xfrm_policy_lock);
928 	return error;
929 }
930 EXPORT_SYMBOL(xfrm_policy_walk);
931 
932 /*
933  * Find policy to apply to this flow.
934  *
935  * Returns 0 if policy found, else an -errno.
936  */
937 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
938 			     u8 type, u16 family, int dir)
939 {
940 	struct xfrm_selector *sel = &pol->selector;
941 	int match, ret = -ESRCH;
942 
943 	if (pol->family != family ||
944 	    pol->type != type)
945 		return ret;
946 
947 	match = xfrm_selector_match(sel, fl, family);
948 	if (match)
949 		ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
950 
951 	return ret;
952 }
953 
954 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
955 						     u16 family, u8 dir)
956 {
957 	int err;
958 	struct xfrm_policy *pol, *ret;
959 	xfrm_address_t *daddr, *saddr;
960 	struct hlist_node *entry;
961 	struct hlist_head *chain;
962 	u32 priority = ~0U;
963 
964 	daddr = xfrm_flowi_daddr(fl, family);
965 	saddr = xfrm_flowi_saddr(fl, family);
966 	if (unlikely(!daddr || !saddr))
967 		return NULL;
968 
969 	read_lock_bh(&xfrm_policy_lock);
970 	chain = policy_hash_direct(daddr, saddr, family, dir);
971 	ret = NULL;
972 	hlist_for_each_entry(pol, entry, chain, bydst) {
973 		err = xfrm_policy_match(pol, fl, type, family, dir);
974 		if (err) {
975 			if (err == -ESRCH)
976 				continue;
977 			else {
978 				ret = ERR_PTR(err);
979 				goto fail;
980 			}
981 		} else {
982 			ret = pol;
983 			priority = ret->priority;
984 			break;
985 		}
986 	}
987 	chain = &xfrm_policy_inexact[dir];
988 	hlist_for_each_entry(pol, entry, chain, bydst) {
989 		err = xfrm_policy_match(pol, fl, type, family, dir);
990 		if (err) {
991 			if (err == -ESRCH)
992 				continue;
993 			else {
994 				ret = ERR_PTR(err);
995 				goto fail;
996 			}
997 		} else if (pol->priority < priority) {
998 			ret = pol;
999 			break;
1000 		}
1001 	}
1002 	if (ret)
1003 		xfrm_pol_hold(ret);
1004 fail:
1005 	read_unlock_bh(&xfrm_policy_lock);
1006 
1007 	return ret;
1008 }
1009 
1010 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1011 			       void **objp, atomic_t **obj_refp)
1012 {
1013 	struct xfrm_policy *pol;
1014 	int err = 0;
1015 
1016 #ifdef CONFIG_XFRM_SUB_POLICY
1017 	pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1018 	if (IS_ERR(pol)) {
1019 		err = PTR_ERR(pol);
1020 		pol = NULL;
1021 	}
1022 	if (pol || err)
1023 		goto end;
1024 #endif
1025 	pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1026 	if (IS_ERR(pol)) {
1027 		err = PTR_ERR(pol);
1028 		pol = NULL;
1029 	}
1030 #ifdef CONFIG_XFRM_SUB_POLICY
1031 end:
1032 #endif
1033 	if ((*objp = (void *) pol) != NULL)
1034 		*obj_refp = &pol->refcnt;
1035 	return err;
1036 }
1037 
1038 static inline int policy_to_flow_dir(int dir)
1039 {
1040 	if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1041 	    XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1042 	    XFRM_POLICY_FWD == FLOW_DIR_FWD)
1043 		return dir;
1044 	switch (dir) {
1045 	default:
1046 	case XFRM_POLICY_IN:
1047 		return FLOW_DIR_IN;
1048 	case XFRM_POLICY_OUT:
1049 		return FLOW_DIR_OUT;
1050 	case XFRM_POLICY_FWD:
1051 		return FLOW_DIR_FWD;
1052 	};
1053 }
1054 
1055 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1056 {
1057 	struct xfrm_policy *pol;
1058 
1059 	read_lock_bh(&xfrm_policy_lock);
1060 	if ((pol = sk->sk_policy[dir]) != NULL) {
1061 		int match = xfrm_selector_match(&pol->selector, fl,
1062 						sk->sk_family);
1063 		int err = 0;
1064 
1065 		if (match) {
1066 			err = security_xfrm_policy_lookup(pol, fl->secid,
1067 					policy_to_flow_dir(dir));
1068 			if (!err)
1069 				xfrm_pol_hold(pol);
1070 			else if (err == -ESRCH)
1071 				pol = NULL;
1072 			else
1073 				pol = ERR_PTR(err);
1074 		} else
1075 			pol = NULL;
1076 	}
1077 	read_unlock_bh(&xfrm_policy_lock);
1078 	return pol;
1079 }
1080 
1081 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1082 {
1083 	struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1084 						     pol->family, dir);
1085 
1086 	hlist_add_head(&pol->bydst, chain);
1087 	hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1088 	xfrm_policy_count[dir]++;
1089 	xfrm_pol_hold(pol);
1090 
1091 	if (xfrm_bydst_should_resize(dir, NULL))
1092 		schedule_work(&xfrm_hash_work);
1093 }
1094 
1095 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1096 						int dir)
1097 {
1098 	if (hlist_unhashed(&pol->bydst))
1099 		return NULL;
1100 
1101 	hlist_del(&pol->bydst);
1102 	hlist_del(&pol->byidx);
1103 	xfrm_policy_count[dir]--;
1104 
1105 	return pol;
1106 }
1107 
1108 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1109 {
1110 	write_lock_bh(&xfrm_policy_lock);
1111 	pol = __xfrm_policy_unlink(pol, dir);
1112 	write_unlock_bh(&xfrm_policy_lock);
1113 	if (pol) {
1114 		if (dir < XFRM_POLICY_MAX)
1115 			atomic_inc(&flow_cache_genid);
1116 		xfrm_policy_kill(pol);
1117 		return 0;
1118 	}
1119 	return -ENOENT;
1120 }
1121 EXPORT_SYMBOL(xfrm_policy_delete);
1122 
1123 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1124 {
1125 	struct xfrm_policy *old_pol;
1126 
1127 #ifdef CONFIG_XFRM_SUB_POLICY
1128 	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1129 		return -EINVAL;
1130 #endif
1131 
1132 	write_lock_bh(&xfrm_policy_lock);
1133 	old_pol = sk->sk_policy[dir];
1134 	sk->sk_policy[dir] = pol;
1135 	if (pol) {
1136 		pol->curlft.add_time = (unsigned long)xtime.tv_sec;
1137 		pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1138 		__xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1139 	}
1140 	if (old_pol)
1141 		__xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1142 	write_unlock_bh(&xfrm_policy_lock);
1143 
1144 	if (old_pol) {
1145 		xfrm_policy_kill(old_pol);
1146 	}
1147 	return 0;
1148 }
1149 
1150 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1151 {
1152 	struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1153 
1154 	if (newp) {
1155 		newp->selector = old->selector;
1156 		if (security_xfrm_policy_clone(old, newp)) {
1157 			kfree(newp);
1158 			return NULL;  /* ENOMEM */
1159 		}
1160 		newp->lft = old->lft;
1161 		newp->curlft = old->curlft;
1162 		newp->action = old->action;
1163 		newp->flags = old->flags;
1164 		newp->xfrm_nr = old->xfrm_nr;
1165 		newp->index = old->index;
1166 		newp->type = old->type;
1167 		memcpy(newp->xfrm_vec, old->xfrm_vec,
1168 		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1169 		write_lock_bh(&xfrm_policy_lock);
1170 		__xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1171 		write_unlock_bh(&xfrm_policy_lock);
1172 		xfrm_pol_put(newp);
1173 	}
1174 	return newp;
1175 }
1176 
1177 int __xfrm_sk_clone_policy(struct sock *sk)
1178 {
1179 	struct xfrm_policy *p0 = sk->sk_policy[0],
1180 			   *p1 = sk->sk_policy[1];
1181 
1182 	sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1183 	if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1184 		return -ENOMEM;
1185 	if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1186 		return -ENOMEM;
1187 	return 0;
1188 }
1189 
1190 static int
1191 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1192 	       unsigned short family)
1193 {
1194 	int err;
1195 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1196 
1197 	if (unlikely(afinfo == NULL))
1198 		return -EINVAL;
1199 	err = afinfo->get_saddr(local, remote);
1200 	xfrm_policy_put_afinfo(afinfo);
1201 	return err;
1202 }
1203 
1204 /* Resolve list of templates for the flow, given policy. */
1205 
1206 static int
1207 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1208 		      struct xfrm_state **xfrm,
1209 		      unsigned short family)
1210 {
1211 	int nx;
1212 	int i, error;
1213 	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1214 	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1215 	xfrm_address_t tmp;
1216 
1217 	for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1218 		struct xfrm_state *x;
1219 		xfrm_address_t *remote = daddr;
1220 		xfrm_address_t *local  = saddr;
1221 		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1222 
1223 		if (tmpl->mode == XFRM_MODE_TUNNEL) {
1224 			remote = &tmpl->id.daddr;
1225 			local = &tmpl->saddr;
1226 			family = tmpl->encap_family;
1227 			if (xfrm_addr_any(local, family)) {
1228 				error = xfrm_get_saddr(&tmp, remote, family);
1229 				if (error)
1230 					goto fail;
1231 				local = &tmp;
1232 			}
1233 		}
1234 
1235 		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1236 
1237 		if (x && x->km.state == XFRM_STATE_VALID) {
1238 			xfrm[nx++] = x;
1239 			daddr = remote;
1240 			saddr = local;
1241 			continue;
1242 		}
1243 		if (x) {
1244 			error = (x->km.state == XFRM_STATE_ERROR ?
1245 				 -EINVAL : -EAGAIN);
1246 			xfrm_state_put(x);
1247 		}
1248 
1249 		if (!tmpl->optional)
1250 			goto fail;
1251 	}
1252 	return nx;
1253 
1254 fail:
1255 	for (nx--; nx>=0; nx--)
1256 		xfrm_state_put(xfrm[nx]);
1257 	return error;
1258 }
1259 
1260 static int
1261 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1262 		  struct xfrm_state **xfrm,
1263 		  unsigned short family)
1264 {
1265 	struct xfrm_state *tp[XFRM_MAX_DEPTH];
1266 	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1267 	int cnx = 0;
1268 	int error;
1269 	int ret;
1270 	int i;
1271 
1272 	for (i = 0; i < npols; i++) {
1273 		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1274 			error = -ENOBUFS;
1275 			goto fail;
1276 		}
1277 
1278 		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1279 		if (ret < 0) {
1280 			error = ret;
1281 			goto fail;
1282 		} else
1283 			cnx += ret;
1284 	}
1285 
1286 	/* found states are sorted for outbound processing */
1287 	if (npols > 1)
1288 		xfrm_state_sort(xfrm, tpp, cnx, family);
1289 
1290 	return cnx;
1291 
1292  fail:
1293 	for (cnx--; cnx>=0; cnx--)
1294 		xfrm_state_put(tpp[cnx]);
1295 	return error;
1296 
1297 }
1298 
1299 /* Check that the bundle accepts the flow and its components are
1300  * still valid.
1301  */
1302 
1303 static struct dst_entry *
1304 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1305 {
1306 	struct dst_entry *x;
1307 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1308 	if (unlikely(afinfo == NULL))
1309 		return ERR_PTR(-EINVAL);
1310 	x = afinfo->find_bundle(fl, policy);
1311 	xfrm_policy_put_afinfo(afinfo);
1312 	return x;
1313 }
1314 
1315 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1316  * all the metrics... Shortly, bundle a bundle.
1317  */
1318 
1319 static int
1320 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1321 		   struct flowi *fl, struct dst_entry **dst_p,
1322 		   unsigned short family)
1323 {
1324 	int err;
1325 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1326 	if (unlikely(afinfo == NULL))
1327 		return -EINVAL;
1328 	err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1329 	xfrm_policy_put_afinfo(afinfo);
1330 	return err;
1331 }
1332 
1333 
1334 static int stale_bundle(struct dst_entry *dst);
1335 
1336 /* Main function: finds/creates a bundle for given flow.
1337  *
1338  * At the moment we eat a raw IP route. Mostly to speed up lookups
1339  * on interfaces with disabled IPsec.
1340  */
1341 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1342 		struct sock *sk, int flags)
1343 {
1344 	struct xfrm_policy *policy;
1345 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1346 	int npols;
1347 	int pol_dead;
1348 	int xfrm_nr;
1349 	int pi;
1350 	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1351 	struct dst_entry *dst, *dst_orig = *dst_p;
1352 	int nx = 0;
1353 	int err;
1354 	u32 genid;
1355 	u16 family;
1356 	u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1357 
1358 restart:
1359 	genid = atomic_read(&flow_cache_genid);
1360 	policy = NULL;
1361 	for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1362 		pols[pi] = NULL;
1363 	npols = 0;
1364 	pol_dead = 0;
1365 	xfrm_nr = 0;
1366 
1367 	if (sk && sk->sk_policy[1]) {
1368 		policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1369 		if (IS_ERR(policy))
1370 			return PTR_ERR(policy);
1371 	}
1372 
1373 	if (!policy) {
1374 		/* To accelerate a bit...  */
1375 		if ((dst_orig->flags & DST_NOXFRM) ||
1376 		    !xfrm_policy_count[XFRM_POLICY_OUT])
1377 			return 0;
1378 
1379 		policy = flow_cache_lookup(fl, dst_orig->ops->family,
1380 					   dir, xfrm_policy_lookup);
1381 		if (IS_ERR(policy))
1382 			return PTR_ERR(policy);
1383 	}
1384 
1385 	if (!policy)
1386 		return 0;
1387 
1388 	family = dst_orig->ops->family;
1389 	policy->curlft.use_time = (unsigned long)xtime.tv_sec;
1390 	pols[0] = policy;
1391 	npols ++;
1392 	xfrm_nr += pols[0]->xfrm_nr;
1393 
1394 	switch (policy->action) {
1395 	case XFRM_POLICY_BLOCK:
1396 		/* Prohibit the flow */
1397 		err = -EPERM;
1398 		goto error;
1399 
1400 	case XFRM_POLICY_ALLOW:
1401 #ifndef CONFIG_XFRM_SUB_POLICY
1402 		if (policy->xfrm_nr == 0) {
1403 			/* Flow passes not transformed. */
1404 			xfrm_pol_put(policy);
1405 			return 0;
1406 		}
1407 #endif
1408 
1409 		/* Try to find matching bundle.
1410 		 *
1411 		 * LATER: help from flow cache. It is optional, this
1412 		 * is required only for output policy.
1413 		 */
1414 		dst = xfrm_find_bundle(fl, policy, family);
1415 		if (IS_ERR(dst)) {
1416 			err = PTR_ERR(dst);
1417 			goto error;
1418 		}
1419 
1420 		if (dst)
1421 			break;
1422 
1423 #ifdef CONFIG_XFRM_SUB_POLICY
1424 		if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1425 			pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1426 							    fl, family,
1427 							    XFRM_POLICY_OUT);
1428 			if (pols[1]) {
1429 				if (IS_ERR(pols[1])) {
1430 					err = PTR_ERR(pols[1]);
1431 					goto error;
1432 				}
1433 				if (pols[1]->action == XFRM_POLICY_BLOCK) {
1434 					err = -EPERM;
1435 					goto error;
1436 				}
1437 				npols ++;
1438 				xfrm_nr += pols[1]->xfrm_nr;
1439 			}
1440 		}
1441 
1442 		/*
1443 		 * Because neither flowi nor bundle information knows about
1444 		 * transformation template size. On more than one policy usage
1445 		 * we can realize whether all of them is bypass or not after
1446 		 * they are searched. See above not-transformed bypass
1447 		 * is surrounded by non-sub policy configuration, too.
1448 		 */
1449 		if (xfrm_nr == 0) {
1450 			/* Flow passes not transformed. */
1451 			xfrm_pols_put(pols, npols);
1452 			return 0;
1453 		}
1454 
1455 #endif
1456 		nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1457 
1458 		if (unlikely(nx<0)) {
1459 			err = nx;
1460 			if (err == -EAGAIN && flags) {
1461 				DECLARE_WAITQUEUE(wait, current);
1462 
1463 				add_wait_queue(&km_waitq, &wait);
1464 				set_current_state(TASK_INTERRUPTIBLE);
1465 				schedule();
1466 				set_current_state(TASK_RUNNING);
1467 				remove_wait_queue(&km_waitq, &wait);
1468 
1469 				nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1470 
1471 				if (nx == -EAGAIN && signal_pending(current)) {
1472 					err = -ERESTART;
1473 					goto error;
1474 				}
1475 				if (nx == -EAGAIN ||
1476 				    genid != atomic_read(&flow_cache_genid)) {
1477 					xfrm_pols_put(pols, npols);
1478 					goto restart;
1479 				}
1480 				err = nx;
1481 			}
1482 			if (err < 0)
1483 				goto error;
1484 		}
1485 		if (nx == 0) {
1486 			/* Flow passes not transformed. */
1487 			xfrm_pols_put(pols, npols);
1488 			return 0;
1489 		}
1490 
1491 		dst = dst_orig;
1492 		err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1493 
1494 		if (unlikely(err)) {
1495 			int i;
1496 			for (i=0; i<nx; i++)
1497 				xfrm_state_put(xfrm[i]);
1498 			goto error;
1499 		}
1500 
1501 		for (pi = 0; pi < npols; pi++) {
1502 			read_lock_bh(&pols[pi]->lock);
1503 			pol_dead |= pols[pi]->dead;
1504 			read_unlock_bh(&pols[pi]->lock);
1505 		}
1506 
1507 		write_lock_bh(&policy->lock);
1508 		if (unlikely(pol_dead || stale_bundle(dst))) {
1509 			/* Wow! While we worked on resolving, this
1510 			 * policy has gone. Retry. It is not paranoia,
1511 			 * we just cannot enlist new bundle to dead object.
1512 			 * We can't enlist stable bundles either.
1513 			 */
1514 			write_unlock_bh(&policy->lock);
1515 			if (dst)
1516 				dst_free(dst);
1517 
1518 			err = -EHOSTUNREACH;
1519 			goto error;
1520 		}
1521 		dst->next = policy->bundles;
1522 		policy->bundles = dst;
1523 		dst_hold(dst);
1524 		write_unlock_bh(&policy->lock);
1525 	}
1526 	*dst_p = dst;
1527 	dst_release(dst_orig);
1528 	xfrm_pols_put(pols, npols);
1529 	return 0;
1530 
1531 error:
1532 	dst_release(dst_orig);
1533 	xfrm_pols_put(pols, npols);
1534 	*dst_p = NULL;
1535 	return err;
1536 }
1537 EXPORT_SYMBOL(xfrm_lookup);
1538 
1539 static inline int
1540 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1541 {
1542 	struct xfrm_state *x;
1543 	int err;
1544 
1545 	if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1546 		return 0;
1547 	x = skb->sp->xvec[idx];
1548 	if (!x->type->reject)
1549 		return 0;
1550 	xfrm_state_hold(x);
1551 	err = x->type->reject(x, skb, fl);
1552 	xfrm_state_put(x);
1553 	return err;
1554 }
1555 
1556 /* When skb is transformed back to its "native" form, we have to
1557  * check policy restrictions. At the moment we make this in maximally
1558  * stupid way. Shame on me. :-) Of course, connected sockets must
1559  * have policy cached at them.
1560  */
1561 
1562 static inline int
1563 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1564 	      unsigned short family)
1565 {
1566 	if (xfrm_state_kern(x))
1567 		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1568 	return	x->id.proto == tmpl->id.proto &&
1569 		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1570 		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1571 		x->props.mode == tmpl->mode &&
1572 		((tmpl->aalgos & (1<<x->props.aalgo)) ||
1573 		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1574 		!(x->props.mode != XFRM_MODE_TRANSPORT &&
1575 		  xfrm_state_addr_cmp(tmpl, x, family));
1576 }
1577 
1578 /*
1579  * 0 or more than 0 is returned when validation is succeeded (either bypass
1580  * because of optional transport mode, or next index of the mathced secpath
1581  * state with the template.
1582  * -1 is returned when no matching template is found.
1583  * Otherwise "-2 - errored_index" is returned.
1584  */
1585 static inline int
1586 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1587 	       unsigned short family)
1588 {
1589 	int idx = start;
1590 
1591 	if (tmpl->optional) {
1592 		if (tmpl->mode == XFRM_MODE_TRANSPORT)
1593 			return start;
1594 	} else
1595 		start = -1;
1596 	for (; idx < sp->len; idx++) {
1597 		if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1598 			return ++idx;
1599 		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1600 			if (start == -1)
1601 				start = -2-idx;
1602 			break;
1603 		}
1604 	}
1605 	return start;
1606 }
1607 
1608 int
1609 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1610 {
1611 	struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1612 	int err;
1613 
1614 	if (unlikely(afinfo == NULL))
1615 		return -EAFNOSUPPORT;
1616 
1617 	afinfo->decode_session(skb, fl);
1618 	err = security_xfrm_decode_session(skb, &fl->secid);
1619 	xfrm_policy_put_afinfo(afinfo);
1620 	return err;
1621 }
1622 EXPORT_SYMBOL(xfrm_decode_session);
1623 
1624 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1625 {
1626 	for (; k < sp->len; k++) {
1627 		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1628 			*idxp = k;
1629 			return 1;
1630 		}
1631 	}
1632 
1633 	return 0;
1634 }
1635 
1636 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1637 			unsigned short family)
1638 {
1639 	struct xfrm_policy *pol;
1640 	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1641 	int npols = 0;
1642 	int xfrm_nr;
1643 	int pi;
1644 	struct flowi fl;
1645 	u8 fl_dir = policy_to_flow_dir(dir);
1646 	int xerr_idx = -1;
1647 
1648 	if (xfrm_decode_session(skb, &fl, family) < 0)
1649 		return 0;
1650 	nf_nat_decode_session(skb, &fl, family);
1651 
1652 	/* First, check used SA against their selectors. */
1653 	if (skb->sp) {
1654 		int i;
1655 
1656 		for (i=skb->sp->len-1; i>=0; i--) {
1657 			struct xfrm_state *x = skb->sp->xvec[i];
1658 			if (!xfrm_selector_match(&x->sel, &fl, family))
1659 				return 0;
1660 		}
1661 	}
1662 
1663 	pol = NULL;
1664 	if (sk && sk->sk_policy[dir]) {
1665 		pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1666 		if (IS_ERR(pol))
1667 			return 0;
1668 	}
1669 
1670 	if (!pol)
1671 		pol = flow_cache_lookup(&fl, family, fl_dir,
1672 					xfrm_policy_lookup);
1673 
1674 	if (IS_ERR(pol))
1675 		return 0;
1676 
1677 	if (!pol) {
1678 		if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1679 			xfrm_secpath_reject(xerr_idx, skb, &fl);
1680 			return 0;
1681 		}
1682 		return 1;
1683 	}
1684 
1685 	pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1686 
1687 	pols[0] = pol;
1688 	npols ++;
1689 #ifdef CONFIG_XFRM_SUB_POLICY
1690 	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1691 		pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1692 						    &fl, family,
1693 						    XFRM_POLICY_IN);
1694 		if (pols[1]) {
1695 			if (IS_ERR(pols[1]))
1696 				return 0;
1697 			pols[1]->curlft.use_time = (unsigned long)xtime.tv_sec;
1698 			npols ++;
1699 		}
1700 	}
1701 #endif
1702 
1703 	if (pol->action == XFRM_POLICY_ALLOW) {
1704 		struct sec_path *sp;
1705 		static struct sec_path dummy;
1706 		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1707 		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1708 		struct xfrm_tmpl **tpp = tp;
1709 		int ti = 0;
1710 		int i, k;
1711 
1712 		if ((sp = skb->sp) == NULL)
1713 			sp = &dummy;
1714 
1715 		for (pi = 0; pi < npols; pi++) {
1716 			if (pols[pi] != pol &&
1717 			    pols[pi]->action != XFRM_POLICY_ALLOW)
1718 				goto reject;
1719 			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1720 				goto reject_error;
1721 			for (i = 0; i < pols[pi]->xfrm_nr; i++)
1722 				tpp[ti++] = &pols[pi]->xfrm_vec[i];
1723 		}
1724 		xfrm_nr = ti;
1725 		if (npols > 1) {
1726 			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1727 			tpp = stp;
1728 		}
1729 
1730 		/* For each tunnel xfrm, find the first matching tmpl.
1731 		 * For each tmpl before that, find corresponding xfrm.
1732 		 * Order is _important_. Later we will implement
1733 		 * some barriers, but at the moment barriers
1734 		 * are implied between each two transformations.
1735 		 */
1736 		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1737 			k = xfrm_policy_ok(tpp[i], sp, k, family);
1738 			if (k < 0) {
1739 				if (k < -1)
1740 					/* "-2 - errored_index" returned */
1741 					xerr_idx = -(2+k);
1742 				goto reject;
1743 			}
1744 		}
1745 
1746 		if (secpath_has_nontransport(sp, k, &xerr_idx))
1747 			goto reject;
1748 
1749 		xfrm_pols_put(pols, npols);
1750 		return 1;
1751 	}
1752 
1753 reject:
1754 	xfrm_secpath_reject(xerr_idx, skb, &fl);
1755 reject_error:
1756 	xfrm_pols_put(pols, npols);
1757 	return 0;
1758 }
1759 EXPORT_SYMBOL(__xfrm_policy_check);
1760 
1761 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1762 {
1763 	struct flowi fl;
1764 
1765 	if (xfrm_decode_session(skb, &fl, family) < 0)
1766 		return 0;
1767 
1768 	return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1769 }
1770 EXPORT_SYMBOL(__xfrm_route_forward);
1771 
1772 /* Optimize later using cookies and generation ids. */
1773 
1774 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1775 {
1776 	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1777 	 * to "-1" to force all XFRM destinations to get validated by
1778 	 * dst_ops->check on every use.  We do this because when a
1779 	 * normal route referenced by an XFRM dst is obsoleted we do
1780 	 * not go looking around for all parent referencing XFRM dsts
1781 	 * so that we can invalidate them.  It is just too much work.
1782 	 * Instead we make the checks here on every use.  For example:
1783 	 *
1784 	 *	XFRM dst A --> IPv4 dst X
1785 	 *
1786 	 * X is the "xdst->route" of A (X is also the "dst->path" of A
1787 	 * in this example).  If X is marked obsolete, "A" will not
1788 	 * notice.  That's what we are validating here via the
1789 	 * stale_bundle() check.
1790 	 *
1791 	 * When a policy's bundle is pruned, we dst_free() the XFRM
1792 	 * dst which causes it's ->obsolete field to be set to a
1793 	 * positive non-zero integer.  If an XFRM dst has been pruned
1794 	 * like this, we want to force a new route lookup.
1795 	 */
1796 	if (dst->obsolete < 0 && !stale_bundle(dst))
1797 		return dst;
1798 
1799 	return NULL;
1800 }
1801 
1802 static int stale_bundle(struct dst_entry *dst)
1803 {
1804 	return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1805 }
1806 
1807 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1808 {
1809 	while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1810 		dst->dev = &loopback_dev;
1811 		dev_hold(&loopback_dev);
1812 		dev_put(dev);
1813 	}
1814 }
1815 EXPORT_SYMBOL(xfrm_dst_ifdown);
1816 
1817 static void xfrm_link_failure(struct sk_buff *skb)
1818 {
1819 	/* Impossible. Such dst must be popped before reaches point of failure. */
1820 	return;
1821 }
1822 
1823 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1824 {
1825 	if (dst) {
1826 		if (dst->obsolete) {
1827 			dst_release(dst);
1828 			dst = NULL;
1829 		}
1830 	}
1831 	return dst;
1832 }
1833 
1834 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1835 {
1836 	struct dst_entry *dst, **dstp;
1837 
1838 	write_lock(&pol->lock);
1839 	dstp = &pol->bundles;
1840 	while ((dst=*dstp) != NULL) {
1841 		if (func(dst)) {
1842 			*dstp = dst->next;
1843 			dst->next = *gc_list_p;
1844 			*gc_list_p = dst;
1845 		} else {
1846 			dstp = &dst->next;
1847 		}
1848 	}
1849 	write_unlock(&pol->lock);
1850 }
1851 
1852 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1853 {
1854 	struct dst_entry *gc_list = NULL;
1855 	int dir;
1856 
1857 	read_lock_bh(&xfrm_policy_lock);
1858 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1859 		struct xfrm_policy *pol;
1860 		struct hlist_node *entry;
1861 		struct hlist_head *table;
1862 		int i;
1863 
1864 		hlist_for_each_entry(pol, entry,
1865 				     &xfrm_policy_inexact[dir], bydst)
1866 			prune_one_bundle(pol, func, &gc_list);
1867 
1868 		table = xfrm_policy_bydst[dir].table;
1869 		for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1870 			hlist_for_each_entry(pol, entry, table + i, bydst)
1871 				prune_one_bundle(pol, func, &gc_list);
1872 		}
1873 	}
1874 	read_unlock_bh(&xfrm_policy_lock);
1875 
1876 	while (gc_list) {
1877 		struct dst_entry *dst = gc_list;
1878 		gc_list = dst->next;
1879 		dst_free(dst);
1880 	}
1881 }
1882 
1883 static int unused_bundle(struct dst_entry *dst)
1884 {
1885 	return !atomic_read(&dst->__refcnt);
1886 }
1887 
1888 static void __xfrm_garbage_collect(void)
1889 {
1890 	xfrm_prune_bundles(unused_bundle);
1891 }
1892 
1893 static int xfrm_flush_bundles(void)
1894 {
1895 	xfrm_prune_bundles(stale_bundle);
1896 	return 0;
1897 }
1898 
1899 void xfrm_init_pmtu(struct dst_entry *dst)
1900 {
1901 	do {
1902 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1903 		u32 pmtu, route_mtu_cached;
1904 
1905 		pmtu = dst_mtu(dst->child);
1906 		xdst->child_mtu_cached = pmtu;
1907 
1908 		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1909 
1910 		route_mtu_cached = dst_mtu(xdst->route);
1911 		xdst->route_mtu_cached = route_mtu_cached;
1912 
1913 		if (pmtu > route_mtu_cached)
1914 			pmtu = route_mtu_cached;
1915 
1916 		dst->metrics[RTAX_MTU-1] = pmtu;
1917 	} while ((dst = dst->next));
1918 }
1919 
1920 EXPORT_SYMBOL(xfrm_init_pmtu);
1921 
1922 /* Check that the bundle accepts the flow and its components are
1923  * still valid.
1924  */
1925 
1926 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
1927 		struct flowi *fl, int family, int strict)
1928 {
1929 	struct dst_entry *dst = &first->u.dst;
1930 	struct xfrm_dst *last;
1931 	u32 mtu;
1932 
1933 	if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1934 	    (dst->dev && !netif_running(dst->dev)))
1935 		return 0;
1936 
1937 	last = NULL;
1938 
1939 	do {
1940 		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1941 
1942 		if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1943 			return 0;
1944 		if (fl && pol &&
1945 		    !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
1946 			return 0;
1947 		if (dst->xfrm->km.state != XFRM_STATE_VALID)
1948 			return 0;
1949 		if (xdst->genid != dst->xfrm->genid)
1950 			return 0;
1951 
1952 		if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
1953 		    !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1954 			return 0;
1955 
1956 		mtu = dst_mtu(dst->child);
1957 		if (xdst->child_mtu_cached != mtu) {
1958 			last = xdst;
1959 			xdst->child_mtu_cached = mtu;
1960 		}
1961 
1962 		if (!dst_check(xdst->route, xdst->route_cookie))
1963 			return 0;
1964 		mtu = dst_mtu(xdst->route);
1965 		if (xdst->route_mtu_cached != mtu) {
1966 			last = xdst;
1967 			xdst->route_mtu_cached = mtu;
1968 		}
1969 
1970 		dst = dst->child;
1971 	} while (dst->xfrm);
1972 
1973 	if (likely(!last))
1974 		return 1;
1975 
1976 	mtu = last->child_mtu_cached;
1977 	for (;;) {
1978 		dst = &last->u.dst;
1979 
1980 		mtu = xfrm_state_mtu(dst->xfrm, mtu);
1981 		if (mtu > last->route_mtu_cached)
1982 			mtu = last->route_mtu_cached;
1983 		dst->metrics[RTAX_MTU-1] = mtu;
1984 
1985 		if (last == first)
1986 			break;
1987 
1988 		last = last->u.next;
1989 		last->child_mtu_cached = mtu;
1990 	}
1991 
1992 	return 1;
1993 }
1994 
1995 EXPORT_SYMBOL(xfrm_bundle_ok);
1996 
1997 #ifdef CONFIG_AUDITSYSCALL
1998 /* Audit addition and deletion of SAs and ipsec policy */
1999 
2000 void xfrm_audit_log(uid_t auid, u32 sid, int type, int result,
2001 		    struct xfrm_policy *xp, struct xfrm_state *x)
2002 {
2003 
2004 	char *secctx;
2005 	u32 secctx_len;
2006 	struct xfrm_sec_ctx *sctx = NULL;
2007 	struct audit_buffer *audit_buf;
2008 	int family;
2009 	extern int audit_enabled;
2010 
2011 	if (audit_enabled == 0)
2012 		return;
2013 
2014 	BUG_ON((type == AUDIT_MAC_IPSEC_ADDSA ||
2015 		type == AUDIT_MAC_IPSEC_DELSA) && !x);
2016 	BUG_ON((type == AUDIT_MAC_IPSEC_ADDSPD ||
2017 		type == AUDIT_MAC_IPSEC_DELSPD) && !xp);
2018 
2019 	audit_buf = audit_log_start(current->audit_context, GFP_ATOMIC, type);
2020 	if (audit_buf == NULL)
2021 		return;
2022 
2023 	switch(type) {
2024 	case AUDIT_MAC_IPSEC_ADDSA:
2025 		audit_log_format(audit_buf, "SAD add: auid=%u", auid);
2026 		break;
2027 	case AUDIT_MAC_IPSEC_DELSA:
2028 		audit_log_format(audit_buf, "SAD delete: auid=%u", auid);
2029 		break;
2030 	case AUDIT_MAC_IPSEC_ADDSPD:
2031 		audit_log_format(audit_buf, "SPD add: auid=%u", auid);
2032 		break;
2033 	case AUDIT_MAC_IPSEC_DELSPD:
2034 		audit_log_format(audit_buf, "SPD delete: auid=%u", auid);
2035 		break;
2036 	default:
2037 		return;
2038 	}
2039 
2040 	if (sid != 0 &&
2041 		security_secid_to_secctx(sid, &secctx, &secctx_len) == 0)
2042 		audit_log_format(audit_buf, " subj=%s", secctx);
2043 	else
2044 		audit_log_task_context(audit_buf);
2045 
2046 	if (xp) {
2047 		family = xp->selector.family;
2048 		if (xp->security)
2049 			sctx = xp->security;
2050 	} else {
2051 		family = x->props.family;
2052 		if (x->security)
2053 			sctx = x->security;
2054 	}
2055 
2056 	if (sctx)
2057 		audit_log_format(audit_buf,
2058 				" sec_alg=%u sec_doi=%u sec_obj=%s",
2059 				sctx->ctx_alg, sctx->ctx_doi, sctx->ctx_str);
2060 
2061 	switch(family) {
2062 	case AF_INET:
2063 		{
2064 			struct in_addr saddr, daddr;
2065 			if (xp) {
2066 				saddr.s_addr = xp->selector.saddr.a4;
2067 				daddr.s_addr = xp->selector.daddr.a4;
2068 			} else {
2069 				saddr.s_addr = x->props.saddr.a4;
2070 				daddr.s_addr = x->id.daddr.a4;
2071 			}
2072 			audit_log_format(audit_buf,
2073 					 " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2074 					 NIPQUAD(saddr), NIPQUAD(daddr));
2075 		}
2076 			break;
2077 	case AF_INET6:
2078 		{
2079 			struct in6_addr saddr6, daddr6;
2080 			if (xp) {
2081 				memcpy(&saddr6, xp->selector.saddr.a6,
2082 					sizeof(struct in6_addr));
2083 				memcpy(&daddr6, xp->selector.daddr.a6,
2084 					sizeof(struct in6_addr));
2085 			} else {
2086 				memcpy(&saddr6, x->props.saddr.a6,
2087 					sizeof(struct in6_addr));
2088 				memcpy(&daddr6, x->id.daddr.a6,
2089 					sizeof(struct in6_addr));
2090 			}
2091 			audit_log_format(audit_buf,
2092 					 " src=" NIP6_FMT "dst=" NIP6_FMT,
2093 					 NIP6(saddr6), NIP6(daddr6));
2094 		}
2095 		break;
2096 	}
2097 
2098 	if (x)
2099 		audit_log_format(audit_buf, " spi=%lu(0x%lx) protocol=%s",
2100 				(unsigned long)ntohl(x->id.spi),
2101 				(unsigned long)ntohl(x->id.spi),
2102 				x->id.proto == IPPROTO_AH ? "AH" :
2103 				(x->id.proto == IPPROTO_ESP ?
2104 				"ESP" : "IPCOMP"));
2105 
2106 	audit_log_format(audit_buf, " res=%u", result);
2107 	audit_log_end(audit_buf);
2108 }
2109 
2110 EXPORT_SYMBOL(xfrm_audit_log);
2111 #endif /* CONFIG_AUDITSYSCALL */
2112 
2113 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2114 {
2115 	int err = 0;
2116 	if (unlikely(afinfo == NULL))
2117 		return -EINVAL;
2118 	if (unlikely(afinfo->family >= NPROTO))
2119 		return -EAFNOSUPPORT;
2120 	write_lock_bh(&xfrm_policy_afinfo_lock);
2121 	if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2122 		err = -ENOBUFS;
2123 	else {
2124 		struct dst_ops *dst_ops = afinfo->dst_ops;
2125 		if (likely(dst_ops->kmem_cachep == NULL))
2126 			dst_ops->kmem_cachep = xfrm_dst_cache;
2127 		if (likely(dst_ops->check == NULL))
2128 			dst_ops->check = xfrm_dst_check;
2129 		if (likely(dst_ops->negative_advice == NULL))
2130 			dst_ops->negative_advice = xfrm_negative_advice;
2131 		if (likely(dst_ops->link_failure == NULL))
2132 			dst_ops->link_failure = xfrm_link_failure;
2133 		if (likely(afinfo->garbage_collect == NULL))
2134 			afinfo->garbage_collect = __xfrm_garbage_collect;
2135 		xfrm_policy_afinfo[afinfo->family] = afinfo;
2136 	}
2137 	write_unlock_bh(&xfrm_policy_afinfo_lock);
2138 	return err;
2139 }
2140 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2141 
2142 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2143 {
2144 	int err = 0;
2145 	if (unlikely(afinfo == NULL))
2146 		return -EINVAL;
2147 	if (unlikely(afinfo->family >= NPROTO))
2148 		return -EAFNOSUPPORT;
2149 	write_lock_bh(&xfrm_policy_afinfo_lock);
2150 	if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2151 		if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2152 			err = -EINVAL;
2153 		else {
2154 			struct dst_ops *dst_ops = afinfo->dst_ops;
2155 			xfrm_policy_afinfo[afinfo->family] = NULL;
2156 			dst_ops->kmem_cachep = NULL;
2157 			dst_ops->check = NULL;
2158 			dst_ops->negative_advice = NULL;
2159 			dst_ops->link_failure = NULL;
2160 			afinfo->garbage_collect = NULL;
2161 		}
2162 	}
2163 	write_unlock_bh(&xfrm_policy_afinfo_lock);
2164 	return err;
2165 }
2166 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2167 
2168 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2169 {
2170 	struct xfrm_policy_afinfo *afinfo;
2171 	if (unlikely(family >= NPROTO))
2172 		return NULL;
2173 	read_lock(&xfrm_policy_afinfo_lock);
2174 	afinfo = xfrm_policy_afinfo[family];
2175 	if (unlikely(!afinfo))
2176 		read_unlock(&xfrm_policy_afinfo_lock);
2177 	return afinfo;
2178 }
2179 
2180 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2181 {
2182 	read_unlock(&xfrm_policy_afinfo_lock);
2183 }
2184 
2185 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2186 {
2187 	struct xfrm_policy_afinfo *afinfo;
2188 	if (unlikely(family >= NPROTO))
2189 		return NULL;
2190 	write_lock_bh(&xfrm_policy_afinfo_lock);
2191 	afinfo = xfrm_policy_afinfo[family];
2192 	if (unlikely(!afinfo))
2193 		write_unlock_bh(&xfrm_policy_afinfo_lock);
2194 	return afinfo;
2195 }
2196 
2197 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2198 {
2199 	write_unlock_bh(&xfrm_policy_afinfo_lock);
2200 }
2201 
2202 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2203 {
2204 	switch (event) {
2205 	case NETDEV_DOWN:
2206 		xfrm_flush_bundles();
2207 	}
2208 	return NOTIFY_DONE;
2209 }
2210 
2211 static struct notifier_block xfrm_dev_notifier = {
2212 	xfrm_dev_event,
2213 	NULL,
2214 	0
2215 };
2216 
2217 static void __init xfrm_policy_init(void)
2218 {
2219 	unsigned int hmask, sz;
2220 	int dir;
2221 
2222 	xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2223 					   sizeof(struct xfrm_dst),
2224 					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2225 					   NULL, NULL);
2226 
2227 	hmask = 8 - 1;
2228 	sz = (hmask+1) * sizeof(struct hlist_head);
2229 
2230 	xfrm_policy_byidx = xfrm_hash_alloc(sz);
2231 	xfrm_idx_hmask = hmask;
2232 	if (!xfrm_policy_byidx)
2233 		panic("XFRM: failed to allocate byidx hash\n");
2234 
2235 	for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2236 		struct xfrm_policy_hash *htab;
2237 
2238 		INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2239 
2240 		htab = &xfrm_policy_bydst[dir];
2241 		htab->table = xfrm_hash_alloc(sz);
2242 		htab->hmask = hmask;
2243 		if (!htab->table)
2244 			panic("XFRM: failed to allocate bydst hash\n");
2245 	}
2246 
2247 	INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2248 	register_netdevice_notifier(&xfrm_dev_notifier);
2249 }
2250 
2251 void __init xfrm_init(void)
2252 {
2253 	xfrm_state_init();
2254 	xfrm_policy_init();
2255 	xfrm_input_init();
2256 }
2257 
2258 #ifdef CONFIG_XFRM_MIGRATE
2259 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2260 				       struct xfrm_selector *sel_tgt)
2261 {
2262 	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2263 		if (sel_tgt->family == sel_cmp->family &&
2264 		    xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2265 				  sel_cmp->family) == 0 &&
2266 		    xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2267 				  sel_cmp->family) == 0 &&
2268 		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2269 		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2270 			return 1;
2271 		}
2272 	} else {
2273 		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2274 			return 1;
2275 		}
2276 	}
2277 	return 0;
2278 }
2279 
2280 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2281 						     u8 dir, u8 type)
2282 {
2283 	struct xfrm_policy *pol, *ret = NULL;
2284 	struct hlist_node *entry;
2285 	struct hlist_head *chain;
2286 	u32 priority = ~0U;
2287 
2288 	read_lock_bh(&xfrm_policy_lock);
2289 	chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2290 	hlist_for_each_entry(pol, entry, chain, bydst) {
2291 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2292 		    pol->type == type) {
2293 			ret = pol;
2294 			priority = ret->priority;
2295 			break;
2296 		}
2297 	}
2298 	chain = &xfrm_policy_inexact[dir];
2299 	hlist_for_each_entry(pol, entry, chain, bydst) {
2300 		if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2301 		    pol->type == type &&
2302 		    pol->priority < priority) {
2303 			ret = pol;
2304 			break;
2305 		}
2306 	}
2307 
2308 	if (ret)
2309 		xfrm_pol_hold(ret);
2310 
2311 	read_unlock_bh(&xfrm_policy_lock);
2312 
2313 	return ret;
2314 }
2315 
2316 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2317 {
2318 	int match = 0;
2319 
2320 	if (t->mode == m->mode && t->id.proto == m->proto &&
2321 	    (m->reqid == 0 || t->reqid == m->reqid)) {
2322 		switch (t->mode) {
2323 		case XFRM_MODE_TUNNEL:
2324 		case XFRM_MODE_BEET:
2325 			if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2326 					  m->old_family) == 0 &&
2327 			    xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2328 					  m->old_family) == 0) {
2329 				match = 1;
2330 			}
2331 			break;
2332 		case XFRM_MODE_TRANSPORT:
2333 			/* in case of transport mode, template does not store
2334 			   any IP addresses, hence we just compare mode and
2335 			   protocol */
2336 			match = 1;
2337 			break;
2338 		default:
2339 			break;
2340 		}
2341 	}
2342 	return match;
2343 }
2344 
2345 /* update endpoint address(es) of template(s) */
2346 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2347 			       struct xfrm_migrate *m, int num_migrate)
2348 {
2349 	struct xfrm_migrate *mp;
2350 	struct dst_entry *dst;
2351 	int i, j, n = 0;
2352 
2353 	write_lock_bh(&pol->lock);
2354 	if (unlikely(pol->dead)) {
2355 		/* target policy has been deleted */
2356 		write_unlock_bh(&pol->lock);
2357 		return -ENOENT;
2358 	}
2359 
2360 	for (i = 0; i < pol->xfrm_nr; i++) {
2361 		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2362 			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2363 				continue;
2364 			n++;
2365 			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL)
2366 				continue;
2367 			/* update endpoints */
2368 			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2369 			       sizeof(pol->xfrm_vec[i].id.daddr));
2370 			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2371 			       sizeof(pol->xfrm_vec[i].saddr));
2372 			pol->xfrm_vec[i].encap_family = mp->new_family;
2373 			/* flush bundles */
2374 			while ((dst = pol->bundles) != NULL) {
2375 				pol->bundles = dst->next;
2376 				dst_free(dst);
2377 			}
2378 		}
2379 	}
2380 
2381 	write_unlock_bh(&pol->lock);
2382 
2383 	if (!n)
2384 		return -ENODATA;
2385 
2386 	return 0;
2387 }
2388 
2389 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2390 {
2391 	int i, j;
2392 
2393 	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2394 		return -EINVAL;
2395 
2396 	for (i = 0; i < num_migrate; i++) {
2397 		if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2398 				   m[i].old_family) == 0) &&
2399 		    (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2400 				   m[i].old_family) == 0))
2401 			return -EINVAL;
2402 		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2403 		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2404 			return -EINVAL;
2405 
2406 		/* check if there is any duplicated entry */
2407 		for (j = i + 1; j < num_migrate; j++) {
2408 			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2409 				    sizeof(m[i].old_daddr)) &&
2410 			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2411 				    sizeof(m[i].old_saddr)) &&
2412 			    m[i].proto == m[j].proto &&
2413 			    m[i].mode == m[j].mode &&
2414 			    m[i].reqid == m[j].reqid &&
2415 			    m[i].old_family == m[j].old_family)
2416 				return -EINVAL;
2417 		}
2418 	}
2419 
2420 	return 0;
2421 }
2422 
2423 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2424 		 struct xfrm_migrate *m, int num_migrate)
2425 {
2426 	int i, err, nx_cur = 0, nx_new = 0;
2427 	struct xfrm_policy *pol = NULL;
2428 	struct xfrm_state *x, *xc;
2429 	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2430 	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2431 	struct xfrm_migrate *mp;
2432 
2433 	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2434 		goto out;
2435 
2436 	/* Stage 1 - find policy */
2437 	if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2438 		err = -ENOENT;
2439 		goto out;
2440 	}
2441 
2442 	/* Stage 2 - find and update state(s) */
2443 	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2444 		if ((x = xfrm_migrate_state_find(mp))) {
2445 			x_cur[nx_cur] = x;
2446 			nx_cur++;
2447 			if ((xc = xfrm_state_migrate(x, mp))) {
2448 				x_new[nx_new] = xc;
2449 				nx_new++;
2450 			} else {
2451 				err = -ENODATA;
2452 				goto restore_state;
2453 			}
2454 		}
2455 	}
2456 
2457 	/* Stage 3 - update policy */
2458 	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2459 		goto restore_state;
2460 
2461 	/* Stage 4 - delete old state(s) */
2462 	if (nx_cur) {
2463 		xfrm_states_put(x_cur, nx_cur);
2464 		xfrm_states_delete(x_cur, nx_cur);
2465 	}
2466 
2467 	/* Stage 5 - announce */
2468 	km_migrate(sel, dir, type, m, num_migrate);
2469 
2470 	xfrm_pol_put(pol);
2471 
2472 	return 0;
2473 out:
2474 	return err;
2475 
2476 restore_state:
2477 	if (pol)
2478 		xfrm_pol_put(pol);
2479 	if (nx_cur)
2480 		xfrm_states_put(x_cur, nx_cur);
2481 	if (nx_new)
2482 		xfrm_states_delete(x_new, nx_new);
2483 
2484 	return err;
2485 }
2486 EXPORT_SYMBOL(xfrm_migrate);
2487 #endif
2488 
2489