?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[NETFILTER]: ipt_TCPMSS: reformat
[hh.org.git] / net / xfrm / xfrm_policy.c
blobb6e2e79d72612be4f99d82d3b8eb7027010c386b
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
29 #include "xfrm_hash.h"
30
31 DEFINE_MUTEX(xfrm_cfg_mutex);
32 EXPORT_SYMBOL(xfrm_cfg_mutex);
33
34 static DEFINE_RWLOCK(xfrm_policy_lock);
35
36 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
37 EXPORT_SYMBOL(xfrm_policy_count);
38
39 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
40 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
41
42 static kmem_cache_t *xfrm_dst_cache __read_mostly;
43
44 static struct work_struct xfrm_policy_gc_work;
45 static HLIST_HEAD(xfrm_policy_gc_list);
46 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
47
48 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
49 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
50 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
51 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
52
53 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
54 {
55 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
56 struct xfrm_type **typemap;
57 int err = 0;
58
59 if (unlikely(afinfo == NULL))
60 return -EAFNOSUPPORT;
61 typemap = afinfo->type_map;
62
63 if (likely(typemap[type->proto] == NULL))
64 typemap[type->proto] = type;
65 else
66 err = -EEXIST;
67 xfrm_policy_unlock_afinfo(afinfo);
68 return err;
69 }
70 EXPORT_SYMBOL(xfrm_register_type);
71
72 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
73 {
74 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
75 struct xfrm_type **typemap;
76 int err = 0;
77
78 if (unlikely(afinfo == NULL))
79 return -EAFNOSUPPORT;
80 typemap = afinfo->type_map;
81
82 if (unlikely(typemap[type->proto] != type))
83 err = -ENOENT;
84 else
85 typemap[type->proto] = NULL;
86 xfrm_policy_unlock_afinfo(afinfo);
87 return err;
88 }
89 EXPORT_SYMBOL(xfrm_unregister_type);
90
91 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
92 {
93 struct xfrm_policy_afinfo *afinfo;
94 struct xfrm_type **typemap;
95 struct xfrm_type *type;
96 int modload_attempted = 0;
97
98 retry:
99 afinfo = xfrm_policy_get_afinfo(family);
100 if (unlikely(afinfo == NULL))
101 return NULL;
102 typemap = afinfo->type_map;
103
104 type = typemap[proto];
105 if (unlikely(type && !try_module_get(type->owner)))
106 type = NULL;
107 if (!type && !modload_attempted) {
108 xfrm_policy_put_afinfo(afinfo);
109 request_module("xfrm-type-%d-%d",
110 (int) family, (int) proto);
111 modload_attempted = 1;
112 goto retry;
113 }
114
115 xfrm_policy_put_afinfo(afinfo);
116 return type;
117 }
118
119 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
120 unsigned short family)
121 {
122 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
123 int err = 0;
124
125 if (unlikely(afinfo == NULL))
126 return -EAFNOSUPPORT;
127
128 if (likely(afinfo->dst_lookup != NULL))
129 err = afinfo->dst_lookup(dst, fl);
130 else
131 err = -EINVAL;
132 xfrm_policy_put_afinfo(afinfo);
133 return err;
134 }
135 EXPORT_SYMBOL(xfrm_dst_lookup);
136
137 void xfrm_put_type(struct xfrm_type *type)
138 {
139 module_put(type->owner);
140 }
141
142 int xfrm_register_mode(struct xfrm_mode *mode, int family)
143 {
144 struct xfrm_policy_afinfo *afinfo;
145 struct xfrm_mode **modemap;
146 int err;
147
148 if (unlikely(mode->encap >= XFRM_MODE_MAX))
149 return -EINVAL;
150
151 afinfo = xfrm_policy_lock_afinfo(family);
152 if (unlikely(afinfo == NULL))
153 return -EAFNOSUPPORT;
154
155 err = -EEXIST;
156 modemap = afinfo->mode_map;
157 if (likely(modemap[mode->encap] == NULL)) {
158 modemap[mode->encap] = mode;
159 err = 0;
160 }
161
162 xfrm_policy_unlock_afinfo(afinfo);
163 return err;
164 }
165 EXPORT_SYMBOL(xfrm_register_mode);
166
167 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
168 {
169 struct xfrm_policy_afinfo *afinfo;
170 struct xfrm_mode **modemap;
171 int err;
172
173 if (unlikely(mode->encap >= XFRM_MODE_MAX))
174 return -EINVAL;
175
176 afinfo = xfrm_policy_lock_afinfo(family);
177 if (unlikely(afinfo == NULL))
178 return -EAFNOSUPPORT;
179
180 err = -ENOENT;
181 modemap = afinfo->mode_map;
182 if (likely(modemap[mode->encap] == mode)) {
183 modemap[mode->encap] = NULL;
184 err = 0;
185 }
186
187 xfrm_policy_unlock_afinfo(afinfo);
188 return err;
189 }
190 EXPORT_SYMBOL(xfrm_unregister_mode);
191
192 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
193 {
194 struct xfrm_policy_afinfo *afinfo;
195 struct xfrm_mode *mode;
196 int modload_attempted = 0;
197
198 if (unlikely(encap >= XFRM_MODE_MAX))
199 return NULL;
200
201 retry:
202 afinfo = xfrm_policy_get_afinfo(family);
203 if (unlikely(afinfo == NULL))
204 return NULL;
205
206 mode = afinfo->mode_map[encap];
207 if (unlikely(mode && !try_module_get(mode->owner)))
208 mode = NULL;
209 if (!mode && !modload_attempted) {
210 xfrm_policy_put_afinfo(afinfo);
211 request_module("xfrm-mode-%d-%d", family, encap);
212 modload_attempted = 1;
213 goto retry;
214 }
215
216 xfrm_policy_put_afinfo(afinfo);
217 return mode;
218 }
219
220 void xfrm_put_mode(struct xfrm_mode *mode)
221 {
222 module_put(mode->owner);
223 }
224
225 static inline unsigned long make_jiffies(long secs)
226 {
227 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
228 return MAX_SCHEDULE_TIMEOUT-1;
229 else
230 return secs*HZ;
231 }
232
233 static void xfrm_policy_timer(unsigned long data)
234 {
235 struct xfrm_policy *xp = (struct xfrm_policy*)data;
236 unsigned long now = (unsigned long)xtime.tv_sec;
237 long next = LONG_MAX;
238 int warn = 0;
239 int dir;
240
241 read_lock(&xp->lock);
242
243 if (xp->dead)
244 goto out;
245
246 dir = xfrm_policy_id2dir(xp->index);
247
248 if (xp->lft.hard_add_expires_seconds) {
249 long tmo = xp->lft.hard_add_expires_seconds +
250 xp->curlft.add_time - now;
251 if (tmo <= 0)
252 goto expired;
253 if (tmo < next)
254 next = tmo;
255 }
256 if (xp->lft.hard_use_expires_seconds) {
257 long tmo = xp->lft.hard_use_expires_seconds +
258 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
259 if (tmo <= 0)
260 goto expired;
261 if (tmo < next)
262 next = tmo;
263 }
264 if (xp->lft.soft_add_expires_seconds) {
265 long tmo = xp->lft.soft_add_expires_seconds +
266 xp->curlft.add_time - now;
267 if (tmo <= 0) {
268 warn = 1;
269 tmo = XFRM_KM_TIMEOUT;
270 }
271 if (tmo < next)
272 next = tmo;
273 }
274 if (xp->lft.soft_use_expires_seconds) {
275 long tmo = xp->lft.soft_use_expires_seconds +
276 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
277 if (tmo <= 0) {
278 warn = 1;
279 tmo = XFRM_KM_TIMEOUT;
280 }
281 if (tmo < next)
282 next = tmo;
283 }
284
285 if (warn)
286 km_policy_expired(xp, dir, 0, 0);
287 if (next != LONG_MAX &&
288 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
289 xfrm_pol_hold(xp);
290
291 out:
292 read_unlock(&xp->lock);
293 xfrm_pol_put(xp);
294 return;
295
296 expired:
297 read_unlock(&xp->lock);
298 if (!xfrm_policy_delete(xp, dir))
299 km_policy_expired(xp, dir, 1, 0);
300 xfrm_pol_put(xp);
301 }
302
303
304 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
305 * SPD calls.
306 */
307
308 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
309 {
310 struct xfrm_policy *policy;
311
312 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
313
314 if (policy) {
315 INIT_HLIST_NODE(&policy->bydst);
316 INIT_HLIST_NODE(&policy->byidx);
317 rwlock_init(&policy->lock);
318 atomic_set(&policy->refcnt, 1);
319 init_timer(&policy->timer);
320 policy->timer.data = (unsigned long)policy;
321 policy->timer.function = xfrm_policy_timer;
322 }
323 return policy;
324 }
325 EXPORT_SYMBOL(xfrm_policy_alloc);
326
327 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
328
329 void __xfrm_policy_destroy(struct xfrm_policy *policy)
330 {
331 BUG_ON(!policy->dead);
332
333 BUG_ON(policy->bundles);
334
335 if (del_timer(&policy->timer))
336 BUG();
337
338 security_xfrm_policy_free(policy);
339 kfree(policy);
340 }
341 EXPORT_SYMBOL(__xfrm_policy_destroy);
342
343 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
344 {
345 struct dst_entry *dst;
346
347 while ((dst = policy->bundles) != NULL) {
348 policy->bundles = dst->next;
349 dst_free(dst);
350 }
351
352 if (del_timer(&policy->timer))
353 atomic_dec(&policy->refcnt);
354
355 if (atomic_read(&policy->refcnt) > 1)
356 flow_cache_flush();
357
358 xfrm_pol_put(policy);
359 }
360
361 static void xfrm_policy_gc_task(void *data)
362 {
363 struct xfrm_policy *policy;
364 struct hlist_node *entry, *tmp;
365 struct hlist_head gc_list;
366
367 spin_lock_bh(&xfrm_policy_gc_lock);
368 gc_list.first = xfrm_policy_gc_list.first;
369 INIT_HLIST_HEAD(&xfrm_policy_gc_list);
370 spin_unlock_bh(&xfrm_policy_gc_lock);
371
372 hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
373 xfrm_policy_gc_kill(policy);
374 }
375
376 /* Rule must be locked. Release descentant resources, announce
377 * entry dead. The rule must be unlinked from lists to the moment.
378 */
379
380 static void xfrm_policy_kill(struct xfrm_policy *policy)
381 {
382 int dead;
383
384 write_lock_bh(&policy->lock);
385 dead = policy->dead;
386 policy->dead = 1;
387 write_unlock_bh(&policy->lock);
388
389 if (unlikely(dead)) {
390 WARN_ON(1);
391 return;
392 }
393
394 spin_lock(&xfrm_policy_gc_lock);
395 hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
396 spin_unlock(&xfrm_policy_gc_lock);
397
398 schedule_work(&xfrm_policy_gc_work);
399 }
400
401 struct xfrm_policy_hash {
402 struct hlist_head *table;
403 unsigned int hmask;
404 };
405
406 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
407 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
408 static struct hlist_head *xfrm_policy_byidx __read_mostly;
409 static unsigned int xfrm_idx_hmask __read_mostly;
410 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
411
412 static inline unsigned int idx_hash(u32 index)
413 {
414 return __idx_hash(index, xfrm_idx_hmask);
415 }
416
417 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
418 {
419 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
420 unsigned int hash = __sel_hash(sel, family, hmask);
421
422 return (hash == hmask + 1 ?
423 &xfrm_policy_inexact[dir] :
424 xfrm_policy_bydst[dir].table + hash);
425 }
426
427 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
428 {
429 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
430 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
431
432 return xfrm_policy_bydst[dir].table + hash;
433 }
434
435 static void xfrm_dst_hash_transfer(struct hlist_head *list,
436 struct hlist_head *ndsttable,
437 unsigned int nhashmask)
438 {
439 struct hlist_node *entry, *tmp;
440 struct xfrm_policy *pol;
441
442 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
443 unsigned int h;
444
445 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
446 pol->family, nhashmask);
447 hlist_add_head(&pol->bydst, ndsttable+h);
448 }
449 }
450
451 static void xfrm_idx_hash_transfer(struct hlist_head *list,
452 struct hlist_head *nidxtable,
453 unsigned int nhashmask)
454 {
455 struct hlist_node *entry, *tmp;
456 struct xfrm_policy *pol;
457
458 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
459 unsigned int h;
460
461 h = __idx_hash(pol->index, nhashmask);
462 hlist_add_head(&pol->byidx, nidxtable+h);
463 }
464 }
465
466 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
467 {
468 return ((old_hmask + 1) << 1) - 1;
469 }
470
471 static void xfrm_bydst_resize(int dir)
472 {
473 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
474 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
475 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
476 struct hlist_head *odst = xfrm_policy_bydst[dir].table;
477 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
478 int i;
479
480 if (!ndst)
481 return;
482
483 write_lock_bh(&xfrm_policy_lock);
484
485 for (i = hmask; i >= 0; i--)
486 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
487
488 xfrm_policy_bydst[dir].table = ndst;
489 xfrm_policy_bydst[dir].hmask = nhashmask;
490
491 write_unlock_bh(&xfrm_policy_lock);
492
493 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
494 }
495
496 static void xfrm_byidx_resize(int total)
497 {
498 unsigned int hmask = xfrm_idx_hmask;
499 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
500 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
501 struct hlist_head *oidx = xfrm_policy_byidx;
502 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
503 int i;
504
505 if (!nidx)
506 return;
507
508 write_lock_bh(&xfrm_policy_lock);
509
510 for (i = hmask; i >= 0; i--)
511 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
512
513 xfrm_policy_byidx = nidx;
514 xfrm_idx_hmask = nhashmask;
515
516 write_unlock_bh(&xfrm_policy_lock);
517
518 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
519 }
520
521 static inline int xfrm_bydst_should_resize(int dir, int *total)
522 {
523 unsigned int cnt = xfrm_policy_count[dir];
524 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
525
526 if (total)
527 *total += cnt;
528
529 if ((hmask + 1) < xfrm_policy_hashmax &&
530 cnt > hmask)
531 return 1;
532
533 return 0;
534 }
535
536 static inline int xfrm_byidx_should_resize(int total)
537 {
538 unsigned int hmask = xfrm_idx_hmask;
539
540 if ((hmask + 1) < xfrm_policy_hashmax &&
541 total > hmask)
542 return 1;
543
544 return 0;
545 }
546
547 static DEFINE_MUTEX(hash_resize_mutex);
548
549 static void xfrm_hash_resize(void *__unused)
550 {
551 int dir, total;
552
553 mutex_lock(&hash_resize_mutex);
554
555 total = 0;
556 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
557 if (xfrm_bydst_should_resize(dir, &total))
558 xfrm_bydst_resize(dir);
559 }
560 if (xfrm_byidx_should_resize(total))
561 xfrm_byidx_resize(total);
562
563 mutex_unlock(&hash_resize_mutex);
564 }
565
566 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize, NULL);
567
568 /* Generate new index... KAME seems to generate them ordered by cost
569 * of an absolute inpredictability of ordering of rules. This will not pass. */
570 static u32 xfrm_gen_index(u8 type, int dir)
571 {
572 static u32 idx_generator;
573
574 for (;;) {
575 struct hlist_node *entry;
576 struct hlist_head *list;
577 struct xfrm_policy *p;
578 u32 idx;
579 int found;
580
581 idx = (idx_generator | dir);
582 idx_generator += 8;
583 if (idx == 0)
584 idx = 8;
585 list = xfrm_policy_byidx + idx_hash(idx);
586 found = 0;
587 hlist_for_each_entry(p, entry, list, byidx) {
588 if (p->index == idx) {
589 found = 1;
590 break;
591 }
592 }
593 if (!found)
594 return idx;
595 }
596 }
597
598 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
599 {
600 u32 *p1 = (u32 *) s1;
601 u32 *p2 = (u32 *) s2;
602 int len = sizeof(struct xfrm_selector) / sizeof(u32);
603 int i;
604
605 for (i = 0; i < len; i++) {
606 if (p1[i] != p2[i])
607 return 1;
608 }
609
610 return 0;
611 }
612
613 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
614 {
615 struct xfrm_policy *pol;
616 struct xfrm_policy *delpol;
617 struct hlist_head *chain;
618 struct hlist_node *entry, *newpos, *last;
619 struct dst_entry *gc_list;
620
621 write_lock_bh(&xfrm_policy_lock);
622 chain = policy_hash_bysel(&policy->selector, policy->family, dir);
623 delpol = NULL;
624 newpos = NULL;
625 last = NULL;
626 hlist_for_each_entry(pol, entry, chain, bydst) {
627 if (!delpol &&
628 pol->type == policy->type &&
629 !selector_cmp(&pol->selector, &policy->selector) &&
630 xfrm_sec_ctx_match(pol->security, policy->security)) {
631 if (excl) {
632 write_unlock_bh(&xfrm_policy_lock);
633 return -EEXIST;
634 }
635 delpol = pol;
636 if (policy->priority > pol->priority)
637 continue;
638 } else if (policy->priority >= pol->priority) {
639 last = &pol->bydst;
640 continue;
641 }
642 if (!newpos)
643 newpos = &pol->bydst;
644 if (delpol)
645 break;
646 last = &pol->bydst;
647 }
648 if (!newpos)
649 newpos = last;
650 if (newpos)
651 hlist_add_after(newpos, &policy->bydst);
652 else
653 hlist_add_head(&policy->bydst, chain);
654 xfrm_pol_hold(policy);
655 xfrm_policy_count[dir]++;
656 atomic_inc(&flow_cache_genid);
657 if (delpol) {
658 hlist_del(&delpol->bydst);
659 hlist_del(&delpol->byidx);
660 xfrm_policy_count[dir]--;
661 }
662 policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
663 hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
664 policy->curlft.add_time = (unsigned long)xtime.tv_sec;
665 policy->curlft.use_time = 0;
666 if (!mod_timer(&policy->timer, jiffies + HZ))
667 xfrm_pol_hold(policy);
668 write_unlock_bh(&xfrm_policy_lock);
669
670 if (delpol)
671 xfrm_policy_kill(delpol);
672 else if (xfrm_bydst_should_resize(dir, NULL))
673 schedule_work(&xfrm_hash_work);
674
675 read_lock_bh(&xfrm_policy_lock);
676 gc_list = NULL;
677 entry = &policy->bydst;
678 hlist_for_each_entry_continue(policy, entry, bydst) {
679 struct dst_entry *dst;
680
681 write_lock(&policy->lock);
682 dst = policy->bundles;
683 if (dst) {
684 struct dst_entry *tail = dst;
685 while (tail->next)
686 tail = tail->next;
687 tail->next = gc_list;
688 gc_list = dst;
689
690 policy->bundles = NULL;
691 }
692 write_unlock(&policy->lock);
693 }
694 read_unlock_bh(&xfrm_policy_lock);
695
696 while (gc_list) {
697 struct dst_entry *dst = gc_list;
698
699 gc_list = dst->next;
700 dst_free(dst);
701 }
702
703 return 0;
704 }
705 EXPORT_SYMBOL(xfrm_policy_insert);
706
707 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
708 struct xfrm_selector *sel,
709 struct xfrm_sec_ctx *ctx, int delete)
710 {
711 struct xfrm_policy *pol, *ret;
712 struct hlist_head *chain;
713 struct hlist_node *entry;
714
715 write_lock_bh(&xfrm_policy_lock);
716 chain = policy_hash_bysel(sel, sel->family, dir);
717 ret = NULL;
718 hlist_for_each_entry(pol, entry, chain, bydst) {
719 if (pol->type == type &&
720 !selector_cmp(sel, &pol->selector) &&
721 xfrm_sec_ctx_match(ctx, pol->security)) {
722 xfrm_pol_hold(pol);
723 if (delete) {
724 hlist_del(&pol->bydst);
725 hlist_del(&pol->byidx);
726 xfrm_policy_count[dir]--;
727 }
728 ret = pol;
729 break;
730 }
731 }
732 write_unlock_bh(&xfrm_policy_lock);
733
734 if (ret && delete) {
735 atomic_inc(&flow_cache_genid);
736 xfrm_policy_kill(ret);
737 }
738 return ret;
739 }
740 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
741
742 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete)
743 {
744 struct xfrm_policy *pol, *ret;
745 struct hlist_head *chain;
746 struct hlist_node *entry;
747
748 write_lock_bh(&xfrm_policy_lock);
749 chain = xfrm_policy_byidx + idx_hash(id);
750 ret = NULL;
751 hlist_for_each_entry(pol, entry, chain, byidx) {
752 if (pol->type == type && pol->index == id) {
753 xfrm_pol_hold(pol);
754 if (delete) {
755 hlist_del(&pol->bydst);
756 hlist_del(&pol->byidx);
757 xfrm_policy_count[dir]--;
758 }
759 ret = pol;
760 break;
761 }
762 }
763 write_unlock_bh(&xfrm_policy_lock);
764
765 if (ret && delete) {
766 atomic_inc(&flow_cache_genid);
767 xfrm_policy_kill(ret);
768 }
769 return ret;
770 }
771 EXPORT_SYMBOL(xfrm_policy_byid);
772
773 void xfrm_policy_flush(u8 type)
774 {
775 int dir;
776
777 write_lock_bh(&xfrm_policy_lock);
778 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
779 struct xfrm_policy *pol;
780 struct hlist_node *entry;
781 int i;
782
783 again1:
784 hlist_for_each_entry(pol, entry,
785 &xfrm_policy_inexact[dir], bydst) {
786 if (pol->type != type)
787 continue;
788 hlist_del(&pol->bydst);
789 hlist_del(&pol->byidx);
790 write_unlock_bh(&xfrm_policy_lock);
791
792 xfrm_policy_kill(pol);
793
794 write_lock_bh(&xfrm_policy_lock);
795 goto again1;
796 }
797
798 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
799 again2:
800 hlist_for_each_entry(pol, entry,
801 xfrm_policy_bydst[dir].table + i,
802 bydst) {
803 if (pol->type != type)
804 continue;
805 hlist_del(&pol->bydst);
806 hlist_del(&pol->byidx);
807 write_unlock_bh(&xfrm_policy_lock);
808
809 xfrm_policy_kill(pol);
810
811 write_lock_bh(&xfrm_policy_lock);
812 goto again2;
813 }
814 }
815
816 xfrm_policy_count[dir] = 0;
817 }
818 atomic_inc(&flow_cache_genid);
819 write_unlock_bh(&xfrm_policy_lock);
820 }
821 EXPORT_SYMBOL(xfrm_policy_flush);
822
823 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
824 void *data)
825 {
826 struct xfrm_policy *pol;
827 struct hlist_node *entry;
828 int dir, count, error;
829
830 read_lock_bh(&xfrm_policy_lock);
831 count = 0;
832 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
833 struct hlist_head *table = xfrm_policy_bydst[dir].table;
834 int i;
835
836 hlist_for_each_entry(pol, entry,
837 &xfrm_policy_inexact[dir], bydst) {
838 if (pol->type == type)
839 count++;
840 }
841 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
842 hlist_for_each_entry(pol, entry, table + i, bydst) {
843 if (pol->type == type)
844 count++;
845 }
846 }
847 }
848
849 if (count == 0) {
850 error = -ENOENT;
851 goto out;
852 }
853
854 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
855 struct hlist_head *table = xfrm_policy_bydst[dir].table;
856 int i;
857
858 hlist_for_each_entry(pol, entry,
859 &xfrm_policy_inexact[dir], bydst) {
860 if (pol->type != type)
861 continue;
862 error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
863 if (error)
864 goto out;
865 }
866 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
867 hlist_for_each_entry(pol, entry, table + i, bydst) {
868 if (pol->type != type)
869 continue;
870 error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
871 if (error)
872 goto out;
873 }
874 }
875 }
876 error = 0;
877 out:
878 read_unlock_bh(&xfrm_policy_lock);
879 return error;
880 }
881 EXPORT_SYMBOL(xfrm_policy_walk);
882
883 /* Find policy to apply to this flow. */
884
885 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
886 u8 type, u16 family, int dir)
887 {
888 struct xfrm_selector *sel = &pol->selector;
889 int match;
890
891 if (pol->family != family ||
892 pol->type != type)
893 return 0;
894
895 match = xfrm_selector_match(sel, fl, family);
896 if (match) {
897 if (!security_xfrm_policy_lookup(pol, fl->secid, dir))
898 return 1;
899 }
900
901 return 0;
902 }
903
904 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
905 u16 family, u8 dir)
906 {
907 struct xfrm_policy *pol, *ret;
908 xfrm_address_t *daddr, *saddr;
909 struct hlist_node *entry;
910 struct hlist_head *chain;
911 u32 priority = ~0U;
912
913 daddr = xfrm_flowi_daddr(fl, family);
914 saddr = xfrm_flowi_saddr(fl, family);
915 if (unlikely(!daddr || !saddr))
916 return NULL;
917
918 read_lock_bh(&xfrm_policy_lock);
919 chain = policy_hash_direct(daddr, saddr, family, dir);
920 ret = NULL;
921 hlist_for_each_entry(pol, entry, chain, bydst) {
922 if (xfrm_policy_match(pol, fl, type, family, dir)) {
923 ret = pol;
924 priority = ret->priority;
925 break;
926 }
927 }
928 chain = &xfrm_policy_inexact[dir];
929 hlist_for_each_entry(pol, entry, chain, bydst) {
930 if (xfrm_policy_match(pol, fl, type, family, dir) &&
931 pol->priority < priority) {
932 ret = pol;
933 break;
934 }
935 }
936 if (ret)
937 xfrm_pol_hold(ret);
938 read_unlock_bh(&xfrm_policy_lock);
939
940 return ret;
941 }
942
943 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
944 void **objp, atomic_t **obj_refp)
945 {
946 struct xfrm_policy *pol;
947
948 #ifdef CONFIG_XFRM_SUB_POLICY
949 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
950 if (pol)
951 goto end;
952 #endif
953 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
954
955 #ifdef CONFIG_XFRM_SUB_POLICY
956 end:
957 #endif
958 if ((*objp = (void *) pol) != NULL)
959 *obj_refp = &pol->refcnt;
960 }
961
962 static inline int policy_to_flow_dir(int dir)
963 {
964 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
965 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
966 XFRM_POLICY_FWD == FLOW_DIR_FWD)
967 return dir;
968 switch (dir) {
969 default:
970 case XFRM_POLICY_IN:
971 return FLOW_DIR_IN;
972 case XFRM_POLICY_OUT:
973 return FLOW_DIR_OUT;
974 case XFRM_POLICY_FWD:
975 return FLOW_DIR_FWD;
976 };
977 }
978
979 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
980 {
981 struct xfrm_policy *pol;
982
983 read_lock_bh(&xfrm_policy_lock);
984 if ((pol = sk->sk_policy[dir]) != NULL) {
985 int match = xfrm_selector_match(&pol->selector, fl,
986 sk->sk_family);
987 int err = 0;
988
989 if (match)
990 err = security_xfrm_policy_lookup(pol, fl->secid, policy_to_flow_dir(dir));
991
992 if (match && !err)
993 xfrm_pol_hold(pol);
994 else
995 pol = NULL;
996 }
997 read_unlock_bh(&xfrm_policy_lock);
998 return pol;
999 }
1000
1001 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1002 {
1003 struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1004 pol->family, dir);
1005
1006 hlist_add_head(&pol->bydst, chain);
1007 hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1008 xfrm_policy_count[dir]++;
1009 xfrm_pol_hold(pol);
1010
1011 if (xfrm_bydst_should_resize(dir, NULL))
1012 schedule_work(&xfrm_hash_work);
1013 }
1014
1015 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1016 int dir)
1017 {
1018 if (hlist_unhashed(&pol->bydst))
1019 return NULL;
1020
1021 hlist_del(&pol->bydst);
1022 hlist_del(&pol->byidx);
1023 xfrm_policy_count[dir]--;
1024
1025 return pol;
1026 }
1027
1028 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1029 {
1030 write_lock_bh(&xfrm_policy_lock);
1031 pol = __xfrm_policy_unlink(pol, dir);
1032 write_unlock_bh(&xfrm_policy_lock);
1033 if (pol) {
1034 if (dir < XFRM_POLICY_MAX)
1035 atomic_inc(&flow_cache_genid);
1036 xfrm_policy_kill(pol);
1037 return 0;
1038 }
1039 return -ENOENT;
1040 }
1041 EXPORT_SYMBOL(xfrm_policy_delete);
1042
1043 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1044 {
1045 struct xfrm_policy *old_pol;
1046
1047 #ifdef CONFIG_XFRM_SUB_POLICY
1048 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1049 return -EINVAL;
1050 #endif
1051
1052 write_lock_bh(&xfrm_policy_lock);
1053 old_pol = sk->sk_policy[dir];
1054 sk->sk_policy[dir] = pol;
1055 if (pol) {
1056 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
1057 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1058 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1059 }
1060 if (old_pol)
1061 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1062 write_unlock_bh(&xfrm_policy_lock);
1063
1064 if (old_pol) {
1065 xfrm_policy_kill(old_pol);
1066 }
1067 return 0;
1068 }
1069
1070 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1071 {
1072 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1073
1074 if (newp) {
1075 newp->selector = old->selector;
1076 if (security_xfrm_policy_clone(old, newp)) {
1077 kfree(newp);
1078 return NULL; /* ENOMEM */
1079 }
1080 newp->lft = old->lft;
1081 newp->curlft = old->curlft;
1082 newp->action = old->action;
1083 newp->flags = old->flags;
1084 newp->xfrm_nr = old->xfrm_nr;
1085 newp->index = old->index;
1086 newp->type = old->type;
1087 memcpy(newp->xfrm_vec, old->xfrm_vec,
1088 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1089 write_lock_bh(&xfrm_policy_lock);
1090 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1091 write_unlock_bh(&xfrm_policy_lock);
1092 xfrm_pol_put(newp);
1093 }
1094 return newp;
1095 }
1096
1097 int __xfrm_sk_clone_policy(struct sock *sk)
1098 {
1099 struct xfrm_policy *p0 = sk->sk_policy[0],
1100 *p1 = sk->sk_policy[1];
1101
1102 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1103 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1104 return -ENOMEM;
1105 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1106 return -ENOMEM;
1107 return 0;
1108 }
1109
1110 static int
1111 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1112 unsigned short family)
1113 {
1114 int err;
1115 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1116
1117 if (unlikely(afinfo == NULL))
1118 return -EINVAL;
1119 err = afinfo->get_saddr(local, remote);
1120 xfrm_policy_put_afinfo(afinfo);
1121 return err;
1122 }
1123
1124 /* Resolve list of templates for the flow, given policy. */
1125
1126 static int
1127 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1128 struct xfrm_state **xfrm,
1129 unsigned short family)
1130 {
1131 int nx;
1132 int i, error;
1133 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1134 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1135 xfrm_address_t tmp;
1136
1137 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1138 struct xfrm_state *x;
1139 xfrm_address_t *remote = daddr;
1140 xfrm_address_t *local = saddr;
1141 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1142
1143 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1144 remote = &tmpl->id.daddr;
1145 local = &tmpl->saddr;
1146 if (xfrm_addr_any(local, family)) {
1147 error = xfrm_get_saddr(&tmp, remote, family);
1148 if (error)
1149 goto fail;
1150 local = &tmp;
1151 }
1152 }
1153
1154 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1155
1156 if (x && x->km.state == XFRM_STATE_VALID) {
1157 xfrm[nx++] = x;
1158 daddr = remote;
1159 saddr = local;
1160 continue;
1161 }
1162 if (x) {
1163 error = (x->km.state == XFRM_STATE_ERROR ?
1164 -EINVAL : -EAGAIN);
1165 xfrm_state_put(x);
1166 }
1167
1168 if (!tmpl->optional)
1169 goto fail;
1170 }
1171 return nx;
1172
1173 fail:
1174 for (nx--; nx>=0; nx--)
1175 xfrm_state_put(xfrm[nx]);
1176 return error;
1177 }
1178
1179 static int
1180 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1181 struct xfrm_state **xfrm,
1182 unsigned short family)
1183 {
1184 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1185 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1186 int cnx = 0;
1187 int error;
1188 int ret;
1189 int i;
1190
1191 for (i = 0; i < npols; i++) {
1192 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1193 error = -ENOBUFS;
1194 goto fail;
1195 }
1196
1197 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1198 if (ret < 0) {
1199 error = ret;
1200 goto fail;
1201 } else
1202 cnx += ret;
1203 }
1204
1205 /* found states are sorted for outbound processing */
1206 if (npols > 1)
1207 xfrm_state_sort(xfrm, tpp, cnx, family);
1208
1209 return cnx;
1210
1211 fail:
1212 for (cnx--; cnx>=0; cnx--)
1213 xfrm_state_put(tpp[cnx]);
1214 return error;
1215
1216 }
1217
1218 /* Check that the bundle accepts the flow and its components are
1219 * still valid.
1220 */
1221
1222 static struct dst_entry *
1223 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1224 {
1225 struct dst_entry *x;
1226 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1227 if (unlikely(afinfo == NULL))
1228 return ERR_PTR(-EINVAL);
1229 x = afinfo->find_bundle(fl, policy);
1230 xfrm_policy_put_afinfo(afinfo);
1231 return x;
1232 }
1233
1234 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1235 * all the metrics... Shortly, bundle a bundle.
1236 */
1237
1238 static int
1239 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1240 struct flowi *fl, struct dst_entry **dst_p,
1241 unsigned short family)
1242 {
1243 int err;
1244 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1245 if (unlikely(afinfo == NULL))
1246 return -EINVAL;
1247 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1248 xfrm_policy_put_afinfo(afinfo);
1249 return err;
1250 }
1251
1252
1253 static int stale_bundle(struct dst_entry *dst);
1254
1255 /* Main function: finds/creates a bundle for given flow.
1256 *
1257 * At the moment we eat a raw IP route. Mostly to speed up lookups
1258 * on interfaces with disabled IPsec.
1259 */
1260 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1261 struct sock *sk, int flags)
1262 {
1263 struct xfrm_policy *policy;
1264 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1265 int npols;
1266 int pol_dead;
1267 int xfrm_nr;
1268 int pi;
1269 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1270 struct dst_entry *dst, *dst_orig = *dst_p;
1271 int nx = 0;
1272 int err;
1273 u32 genid;
1274 u16 family;
1275 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1276
1277 restart:
1278 genid = atomic_read(&flow_cache_genid);
1279 policy = NULL;
1280 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1281 pols[pi] = NULL;
1282 npols = 0;
1283 pol_dead = 0;
1284 xfrm_nr = 0;
1285
1286 if (sk && sk->sk_policy[1])
1287 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1288
1289 if (!policy) {
1290 /* To accelerate a bit... */
1291 if ((dst_orig->flags & DST_NOXFRM) ||
1292 !xfrm_policy_count[XFRM_POLICY_OUT])
1293 return 0;
1294
1295 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1296 dir, xfrm_policy_lookup);
1297 }
1298
1299 if (!policy)
1300 return 0;
1301
1302 family = dst_orig->ops->family;
1303 policy->curlft.use_time = (unsigned long)xtime.tv_sec;
1304 pols[0] = policy;
1305 npols ++;
1306 xfrm_nr += pols[0]->xfrm_nr;
1307
1308 switch (policy->action) {
1309 case XFRM_POLICY_BLOCK:
1310 /* Prohibit the flow */
1311 err = -EPERM;
1312 goto error;
1313
1314 case XFRM_POLICY_ALLOW:
1315 #ifndef CONFIG_XFRM_SUB_POLICY
1316 if (policy->xfrm_nr == 0) {
1317 /* Flow passes not transformed. */
1318 xfrm_pol_put(policy);
1319 return 0;
1320 }
1321 #endif
1322
1323 /* Try to find matching bundle.
1324 *
1325 * LATER: help from flow cache. It is optional, this
1326 * is required only for output policy.
1327 */
1328 dst = xfrm_find_bundle(fl, policy, family);
1329 if (IS_ERR(dst)) {
1330 err = PTR_ERR(dst);
1331 goto error;
1332 }
1333
1334 if (dst)
1335 break;
1336
1337 #ifdef CONFIG_XFRM_SUB_POLICY
1338 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1339 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1340 fl, family,
1341 XFRM_POLICY_OUT);
1342 if (pols[1]) {
1343 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1344 err = -EPERM;
1345 goto error;
1346 }
1347 npols ++;
1348 xfrm_nr += pols[1]->xfrm_nr;
1349 }
1350 }
1351
1352 /*
1353 * Because neither flowi nor bundle information knows about
1354 * transformation template size. On more than one policy usage
1355 * we can realize whether all of them is bypass or not after
1356 * they are searched. See above not-transformed bypass
1357 * is surrounded by non-sub policy configuration, too.
1358 */
1359 if (xfrm_nr == 0) {
1360 /* Flow passes not transformed. */
1361 xfrm_pols_put(pols, npols);
1362 return 0;
1363 }
1364
1365 #endif
1366 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1367
1368 if (unlikely(nx<0)) {
1369 err = nx;
1370 if (err == -EAGAIN && flags) {
1371 DECLARE_WAITQUEUE(wait, current);
1372
1373 add_wait_queue(&km_waitq, &wait);
1374 set_current_state(TASK_INTERRUPTIBLE);
1375 schedule();
1376 set_current_state(TASK_RUNNING);
1377 remove_wait_queue(&km_waitq, &wait);
1378
1379 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1380
1381 if (nx == -EAGAIN && signal_pending(current)) {
1382 err = -ERESTART;
1383 goto error;
1384 }
1385 if (nx == -EAGAIN ||
1386 genid != atomic_read(&flow_cache_genid)) {
1387 xfrm_pols_put(pols, npols);
1388 goto restart;
1389 }
1390 err = nx;
1391 }
1392 if (err < 0)
1393 goto error;
1394 }
1395 if (nx == 0) {
1396 /* Flow passes not transformed. */
1397 xfrm_pols_put(pols, npols);
1398 return 0;
1399 }
1400
1401 dst = dst_orig;
1402 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1403
1404 if (unlikely(err)) {
1405 int i;
1406 for (i=0; i<nx; i++)
1407 xfrm_state_put(xfrm[i]);
1408 goto error;
1409 }
1410
1411 for (pi = 0; pi < npols; pi++) {
1412 read_lock_bh(&pols[pi]->lock);
1413 pol_dead |= pols[pi]->dead;
1414 read_unlock_bh(&pols[pi]->lock);
1415 }
1416
1417 write_lock_bh(&policy->lock);
1418 if (unlikely(pol_dead || stale_bundle(dst))) {
1419 /* Wow! While we worked on resolving, this
1420 * policy has gone. Retry. It is not paranoia,
1421 * we just cannot enlist new bundle to dead object.
1422 * We can't enlist stable bundles either.
1423 */
1424 write_unlock_bh(&policy->lock);
1425 if (dst)
1426 dst_free(dst);
1427
1428 err = -EHOSTUNREACH;
1429 goto error;
1430 }
1431 dst->next = policy->bundles;
1432 policy->bundles = dst;
1433 dst_hold(dst);
1434 write_unlock_bh(&policy->lock);
1435 }
1436 *dst_p = dst;
1437 dst_release(dst_orig);
1438 xfrm_pols_put(pols, npols);
1439 return 0;
1440
1441 error:
1442 dst_release(dst_orig);
1443 xfrm_pols_put(pols, npols);
1444 *dst_p = NULL;
1445 return err;
1446 }
1447 EXPORT_SYMBOL(xfrm_lookup);
1448
1449 static inline int
1450 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1451 {
1452 struct xfrm_state *x;
1453 int err;
1454
1455 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1456 return 0;
1457 x = skb->sp->xvec[idx];
1458 if (!x->type->reject)
1459 return 0;
1460 xfrm_state_hold(x);
1461 err = x->type->reject(x, skb, fl);
1462 xfrm_state_put(x);
1463 return err;
1464 }
1465
1466 /* When skb is transformed back to its "native" form, we have to
1467 * check policy restrictions. At the moment we make this in maximally
1468 * stupid way. Shame on me. :-) Of course, connected sockets must
1469 * have policy cached at them.
1470 */
1471
1472 static inline int
1473 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1474 unsigned short family)
1475 {
1476 if (xfrm_state_kern(x))
1477 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1478 return x->id.proto == tmpl->id.proto &&
1479 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1480 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1481 x->props.mode == tmpl->mode &&
1482 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1483 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1484 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1485 xfrm_state_addr_cmp(tmpl, x, family));
1486 }
1487
1488 /*
1489 * 0 or more than 0 is returned when validation is succeeded (either bypass
1490 * because of optional transport mode, or next index of the mathced secpath
1491 * state with the template.
1492 * -1 is returned when no matching template is found.
1493 * Otherwise "-2 - errored_index" is returned.
1494 */
1495 static inline int
1496 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1497 unsigned short family)
1498 {
1499 int idx = start;
1500
1501 if (tmpl->optional) {
1502 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1503 return start;
1504 } else
1505 start = -1;
1506 for (; idx < sp->len; idx++) {
1507 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1508 return ++idx;
1509 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1510 if (start == -1)
1511 start = -2-idx;
1512 break;
1513 }
1514 }
1515 return start;
1516 }
1517
1518 int
1519 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1520 {
1521 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1522 int err;
1523
1524 if (unlikely(afinfo == NULL))
1525 return -EAFNOSUPPORT;
1526
1527 afinfo->decode_session(skb, fl);
1528 err = security_xfrm_decode_session(skb, &fl->secid);
1529 xfrm_policy_put_afinfo(afinfo);
1530 return err;
1531 }
1532 EXPORT_SYMBOL(xfrm_decode_session);
1533
1534 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1535 {
1536 for (; k < sp->len; k++) {
1537 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1538 *idxp = k;
1539 return 1;
1540 }
1541 }
1542
1543 return 0;
1544 }
1545
1546 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1547 unsigned short family)
1548 {
1549 struct xfrm_policy *pol;
1550 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1551 int npols = 0;
1552 int xfrm_nr;
1553 int pi;
1554 struct flowi fl;
1555 u8 fl_dir = policy_to_flow_dir(dir);
1556 int xerr_idx = -1;
1557
1558 if (xfrm_decode_session(skb, &fl, family) < 0)
1559 return 0;
1560 nf_nat_decode_session(skb, &fl, family);
1561
1562 /* First, check used SA against their selectors. */
1563 if (skb->sp) {
1564 int i;
1565
1566 for (i=skb->sp->len-1; i>=0; i--) {
1567 struct xfrm_state *x = skb->sp->xvec[i];
1568 if (!xfrm_selector_match(&x->sel, &fl, family))
1569 return 0;
1570 }
1571 }
1572
1573 pol = NULL;
1574 if (sk && sk->sk_policy[dir])
1575 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1576
1577 if (!pol)
1578 pol = flow_cache_lookup(&fl, family, fl_dir,
1579 xfrm_policy_lookup);
1580
1581 if (!pol) {
1582 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1583 xfrm_secpath_reject(xerr_idx, skb, &fl);
1584 return 0;
1585 }
1586 return 1;
1587 }
1588
1589 pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1590
1591 pols[0] = pol;
1592 npols ++;
1593 #ifdef CONFIG_XFRM_SUB_POLICY
1594 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1595 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1596 &fl, family,
1597 XFRM_POLICY_IN);
1598 if (pols[1]) {
1599 pols[1]->curlft.use_time = (unsigned long)xtime.tv_sec;
1600 npols ++;
1601 }
1602 }
1603 #endif
1604
1605 if (pol->action == XFRM_POLICY_ALLOW) {
1606 struct sec_path *sp;
1607 static struct sec_path dummy;
1608 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1609 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1610 struct xfrm_tmpl **tpp = tp;
1611 int ti = 0;
1612 int i, k;
1613
1614 if ((sp = skb->sp) == NULL)
1615 sp = &dummy;
1616
1617 for (pi = 0; pi < npols; pi++) {
1618 if (pols[pi] != pol &&
1619 pols[pi]->action != XFRM_POLICY_ALLOW)
1620 goto reject;
1621 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1622 goto reject_error;
1623 for (i = 0; i < pols[pi]->xfrm_nr; i++)
1624 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1625 }
1626 xfrm_nr = ti;
1627 if (npols > 1) {
1628 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1629 tpp = stp;
1630 }
1631
1632 /* For each tunnel xfrm, find the first matching tmpl.
1633 * For each tmpl before that, find corresponding xfrm.
1634 * Order is _important_. Later we will implement
1635 * some barriers, but at the moment barriers
1636 * are implied between each two transformations.
1637 */
1638 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1639 k = xfrm_policy_ok(tpp[i], sp, k, family);
1640 if (k < 0) {
1641 if (k < -1)
1642 /* "-2 - errored_index" returned */
1643 xerr_idx = -(2+k);
1644 goto reject;
1645 }
1646 }
1647
1648 if (secpath_has_nontransport(sp, k, &xerr_idx))
1649 goto reject;
1650
1651 xfrm_pols_put(pols, npols);
1652 return 1;
1653 }
1654
1655 reject:
1656 xfrm_secpath_reject(xerr_idx, skb, &fl);
1657 reject_error:
1658 xfrm_pols_put(pols, npols);
1659 return 0;
1660 }
1661 EXPORT_SYMBOL(__xfrm_policy_check);
1662
1663 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1664 {
1665 struct flowi fl;
1666
1667 if (xfrm_decode_session(skb, &fl, family) < 0)
1668 return 0;
1669
1670 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1671 }
1672 EXPORT_SYMBOL(__xfrm_route_forward);
1673
1674 /* Optimize later using cookies and generation ids. */
1675
1676 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1677 {
1678 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1679 * to "-1" to force all XFRM destinations to get validated by
1680 * dst_ops->check on every use. We do this because when a
1681 * normal route referenced by an XFRM dst is obsoleted we do
1682 * not go looking around for all parent referencing XFRM dsts
1683 * so that we can invalidate them. It is just too much work.
1684 * Instead we make the checks here on every use. For example:
1685 *
1686 * XFRM dst A --> IPv4 dst X
1687 *
1688 * X is the "xdst->route" of A (X is also the "dst->path" of A
1689 * in this example). If X is marked obsolete, "A" will not
1690 * notice. That's what we are validating here via the
1691 * stale_bundle() check.
1692 *
1693 * When a policy's bundle is pruned, we dst_free() the XFRM
1694 * dst which causes it's ->obsolete field to be set to a
1695 * positive non-zero integer. If an XFRM dst has been pruned
1696 * like this, we want to force a new route lookup.
1697 */
1698 if (dst->obsolete < 0 && !stale_bundle(dst))
1699 return dst;
1700
1701 return NULL;
1702 }
1703
1704 static int stale_bundle(struct dst_entry *dst)
1705 {
1706 return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1707 }
1708
1709 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1710 {
1711 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1712 dst->dev = &loopback_dev;
1713 dev_hold(&loopback_dev);
1714 dev_put(dev);
1715 }
1716 }
1717 EXPORT_SYMBOL(xfrm_dst_ifdown);
1718
1719 static void xfrm_link_failure(struct sk_buff *skb)
1720 {
1721 /* Impossible. Such dst must be popped before reaches point of failure. */
1722 return;
1723 }
1724
1725 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1726 {
1727 if (dst) {
1728 if (dst->obsolete) {
1729 dst_release(dst);
1730 dst = NULL;
1731 }
1732 }
1733 return dst;
1734 }
1735
1736 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1737 {
1738 struct dst_entry *dst, **dstp;
1739
1740 write_lock(&pol->lock);
1741 dstp = &pol->bundles;
1742 while ((dst=*dstp) != NULL) {
1743 if (func(dst)) {
1744 *dstp = dst->next;
1745 dst->next = *gc_list_p;
1746 *gc_list_p = dst;
1747 } else {
1748 dstp = &dst->next;
1749 }
1750 }
1751 write_unlock(&pol->lock);
1752 }
1753
1754 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1755 {
1756 struct dst_entry *gc_list = NULL;
1757 int dir;
1758
1759 read_lock_bh(&xfrm_policy_lock);
1760 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1761 struct xfrm_policy *pol;
1762 struct hlist_node *entry;
1763 struct hlist_head *table;
1764 int i;
1765
1766 hlist_for_each_entry(pol, entry,
1767 &xfrm_policy_inexact[dir], bydst)
1768 prune_one_bundle(pol, func, &gc_list);
1769
1770 table = xfrm_policy_bydst[dir].table;
1771 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1772 hlist_for_each_entry(pol, entry, table + i, bydst)
1773 prune_one_bundle(pol, func, &gc_list);
1774 }
1775 }
1776 read_unlock_bh(&xfrm_policy_lock);
1777
1778 while (gc_list) {
1779 struct dst_entry *dst = gc_list;
1780 gc_list = dst->next;
1781 dst_free(dst);
1782 }
1783 }
1784
1785 static int unused_bundle(struct dst_entry *dst)
1786 {
1787 return !atomic_read(&dst->__refcnt);
1788 }
1789
1790 static void __xfrm_garbage_collect(void)
1791 {
1792 xfrm_prune_bundles(unused_bundle);
1793 }
1794
1795 static int xfrm_flush_bundles(void)
1796 {
1797 xfrm_prune_bundles(stale_bundle);
1798 return 0;
1799 }
1800
1801 void xfrm_init_pmtu(struct dst_entry *dst)
1802 {
1803 do {
1804 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1805 u32 pmtu, route_mtu_cached;
1806
1807 pmtu = dst_mtu(dst->child);
1808 xdst->child_mtu_cached = pmtu;
1809
1810 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1811
1812 route_mtu_cached = dst_mtu(xdst->route);
1813 xdst->route_mtu_cached = route_mtu_cached;
1814
1815 if (pmtu > route_mtu_cached)
1816 pmtu = route_mtu_cached;
1817
1818 dst->metrics[RTAX_MTU-1] = pmtu;
1819 } while ((dst = dst->next));
1820 }
1821
1822 EXPORT_SYMBOL(xfrm_init_pmtu);
1823
1824 /* Check that the bundle accepts the flow and its components are
1825 * still valid.
1826 */
1827
1828 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family, int strict)
1829 {
1830 struct dst_entry *dst = &first->u.dst;
1831 struct xfrm_dst *last;
1832 u32 mtu;
1833
1834 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1835 (dst->dev && !netif_running(dst->dev)))
1836 return 0;
1837
1838 last = NULL;
1839
1840 do {
1841 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1842
1843 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1844 return 0;
1845 if (fl && !security_xfrm_flow_state_match(fl, dst->xfrm))
1846 return 0;
1847 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1848 return 0;
1849 if (xdst->genid != dst->xfrm->genid)
1850 return 0;
1851
1852 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
1853 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1854 return 0;
1855
1856 mtu = dst_mtu(dst->child);
1857 if (xdst->child_mtu_cached != mtu) {
1858 last = xdst;
1859 xdst->child_mtu_cached = mtu;
1860 }
1861
1862 if (!dst_check(xdst->route, xdst->route_cookie))
1863 return 0;
1864 mtu = dst_mtu(xdst->route);
1865 if (xdst->route_mtu_cached != mtu) {
1866 last = xdst;
1867 xdst->route_mtu_cached = mtu;
1868 }
1869
1870 dst = dst->child;
1871 } while (dst->xfrm);
1872
1873 if (likely(!last))
1874 return 1;
1875
1876 mtu = last->child_mtu_cached;
1877 for (;;) {
1878 dst = &last->u.dst;
1879
1880 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1881 if (mtu > last->route_mtu_cached)
1882 mtu = last->route_mtu_cached;
1883 dst->metrics[RTAX_MTU-1] = mtu;
1884
1885 if (last == first)
1886 break;
1887
1888 last = last->u.next;
1889 last->child_mtu_cached = mtu;
1890 }
1891
1892 return 1;
1893 }
1894
1895 EXPORT_SYMBOL(xfrm_bundle_ok);
1896
1897 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1898 {
1899 int err = 0;
1900 if (unlikely(afinfo == NULL))
1901 return -EINVAL;
1902 if (unlikely(afinfo->family >= NPROTO))
1903 return -EAFNOSUPPORT;
1904 write_lock_bh(&xfrm_policy_afinfo_lock);
1905 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1906 err = -ENOBUFS;
1907 else {
1908 struct dst_ops *dst_ops = afinfo->dst_ops;
1909 if (likely(dst_ops->kmem_cachep == NULL))
1910 dst_ops->kmem_cachep = xfrm_dst_cache;
1911 if (likely(dst_ops->check == NULL))
1912 dst_ops->check = xfrm_dst_check;
1913 if (likely(dst_ops->negative_advice == NULL))
1914 dst_ops->negative_advice = xfrm_negative_advice;
1915 if (likely(dst_ops->link_failure == NULL))
1916 dst_ops->link_failure = xfrm_link_failure;
1917 if (likely(afinfo->garbage_collect == NULL))
1918 afinfo->garbage_collect = __xfrm_garbage_collect;
1919 xfrm_policy_afinfo[afinfo->family] = afinfo;
1920 }
1921 write_unlock_bh(&xfrm_policy_afinfo_lock);
1922 return err;
1923 }
1924 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1925
1926 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1927 {
1928 int err = 0;
1929 if (unlikely(afinfo == NULL))
1930 return -EINVAL;
1931 if (unlikely(afinfo->family >= NPROTO))
1932 return -EAFNOSUPPORT;
1933 write_lock_bh(&xfrm_policy_afinfo_lock);
1934 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1935 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1936 err = -EINVAL;
1937 else {
1938 struct dst_ops *dst_ops = afinfo->dst_ops;
1939 xfrm_policy_afinfo[afinfo->family] = NULL;
1940 dst_ops->kmem_cachep = NULL;
1941 dst_ops->check = NULL;
1942 dst_ops->negative_advice = NULL;
1943 dst_ops->link_failure = NULL;
1944 afinfo->garbage_collect = NULL;
1945 }
1946 }
1947 write_unlock_bh(&xfrm_policy_afinfo_lock);
1948 return err;
1949 }
1950 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1951
1952 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1953 {
1954 struct xfrm_policy_afinfo *afinfo;
1955 if (unlikely(family >= NPROTO))
1956 return NULL;
1957 read_lock(&xfrm_policy_afinfo_lock);
1958 afinfo = xfrm_policy_afinfo[family];
1959 if (unlikely(!afinfo))
1960 read_unlock(&xfrm_policy_afinfo_lock);
1961 return afinfo;
1962 }
1963
1964 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1965 {
1966 read_unlock(&xfrm_policy_afinfo_lock);
1967 }
1968
1969 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
1970 {
1971 struct xfrm_policy_afinfo *afinfo;
1972 if (unlikely(family >= NPROTO))
1973 return NULL;
1974 write_lock_bh(&xfrm_policy_afinfo_lock);
1975 afinfo = xfrm_policy_afinfo[family];
1976 if (unlikely(!afinfo))
1977 write_unlock_bh(&xfrm_policy_afinfo_lock);
1978 return afinfo;
1979 }
1980
1981 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
1982 {
1983 write_unlock_bh(&xfrm_policy_afinfo_lock);
1984 }
1985
1986 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1987 {
1988 switch (event) {
1989 case NETDEV_DOWN:
1990 xfrm_flush_bundles();
1991 }
1992 return NOTIFY_DONE;
1993 }
1994
1995 static struct notifier_block xfrm_dev_notifier = {
1996 xfrm_dev_event,
1997 NULL,
1998 0
1999 };
2000
2001 static void __init xfrm_policy_init(void)
2002 {
2003 unsigned int hmask, sz;
2004 int dir;
2005
2006 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2007 sizeof(struct xfrm_dst),
2008 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2009 NULL, NULL);
2010
2011 hmask = 8 - 1;
2012 sz = (hmask+1) * sizeof(struct hlist_head);
2013
2014 xfrm_policy_byidx = xfrm_hash_alloc(sz);
2015 xfrm_idx_hmask = hmask;
2016 if (!xfrm_policy_byidx)
2017 panic("XFRM: failed to allocate byidx hash\n");
2018
2019 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2020 struct xfrm_policy_hash *htab;
2021
2022 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2023
2024 htab = &xfrm_policy_bydst[dir];
2025 htab->table = xfrm_hash_alloc(sz);
2026 htab->hmask = hmask;
2027 if (!htab->table)
2028 panic("XFRM: failed to allocate bydst hash\n");
2029 }
2030
2031 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
2032 register_netdevice_notifier(&xfrm_dev_notifier);
2033 }
2034
2035 void __init xfrm_init(void)
2036 {
2037 xfrm_state_init();
2038 xfrm_policy_init();
2039 xfrm_input_init();
2040 }
2041
hh.org GIT migration