projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394...
[deliverable/linux.git] / net / xfrm / xfrm_policy.c
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 #include <linux/cache.h>
30
31 #include "xfrm_hash.h"
32
33 int sysctl_xfrm_larval_drop __read_mostly;
34
35 DEFINE_MUTEX(xfrm_cfg_mutex);
36 EXPORT_SYMBOL(xfrm_cfg_mutex);
37
38 static DEFINE_RWLOCK(xfrm_policy_lock);
39
40 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
41 EXPORT_SYMBOL(xfrm_policy_count);
42
43 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
44 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
45
46 static struct kmem_cache *xfrm_dst_cache __read_mostly;
47
48 static struct work_struct xfrm_policy_gc_work;
49 static HLIST_HEAD(xfrm_policy_gc_list);
50 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
51
52 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
53 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
54 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
55 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
56
57 static inline int
58 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
59 {
60 return addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
61 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
62 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
63 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
64 (fl->proto == sel->proto || !sel->proto) &&
65 (fl->oif == sel->ifindex || !sel->ifindex);
66 }
67
68 static inline int
69 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
70 {
71 return addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
72 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
73 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
74 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
75 (fl->proto == sel->proto || !sel->proto) &&
76 (fl->oif == sel->ifindex || !sel->ifindex);
77 }
78
79 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
80 unsigned short family)
81 {
82 switch (family) {
83 case AF_INET:
84 return __xfrm4_selector_match(sel, fl);
85 case AF_INET6:
86 return __xfrm6_selector_match(sel, fl);
87 }
88 return 0;
89 }
90
91 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
92 {
93 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
94 struct xfrm_type **typemap;
95 int err = 0;
96
97 if (unlikely(afinfo == NULL))
98 return -EAFNOSUPPORT;
99 typemap = afinfo->type_map;
100
101 if (likely(typemap[type->proto] == NULL))
102 typemap[type->proto] = type;
103 else
104 err = -EEXIST;
105 xfrm_policy_unlock_afinfo(afinfo);
106 return err;
107 }
108 EXPORT_SYMBOL(xfrm_register_type);
109
110 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
111 {
112 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
113 struct xfrm_type **typemap;
114 int err = 0;
115
116 if (unlikely(afinfo == NULL))
117 return -EAFNOSUPPORT;
118 typemap = afinfo->type_map;
119
120 if (unlikely(typemap[type->proto] != type))
121 err = -ENOENT;
122 else
123 typemap[type->proto] = NULL;
124 xfrm_policy_unlock_afinfo(afinfo);
125 return err;
126 }
127 EXPORT_SYMBOL(xfrm_unregister_type);
128
129 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
130 {
131 struct xfrm_policy_afinfo *afinfo;
132 struct xfrm_type **typemap;
133 struct xfrm_type *type;
134 int modload_attempted = 0;
135
136 retry:
137 afinfo = xfrm_policy_get_afinfo(family);
138 if (unlikely(afinfo == NULL))
139 return NULL;
140 typemap = afinfo->type_map;
141
142 type = typemap[proto];
143 if (unlikely(type && !try_module_get(type->owner)))
144 type = NULL;
145 if (!type && !modload_attempted) {
146 xfrm_policy_put_afinfo(afinfo);
147 request_module("xfrm-type-%d-%d",
148 (int) family, (int) proto);
149 modload_attempted = 1;
150 goto retry;
151 }
152
153 xfrm_policy_put_afinfo(afinfo);
154 return type;
155 }
156
157 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
158 unsigned short family)
159 {
160 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
161 int err = 0;
162
163 if (unlikely(afinfo == NULL))
164 return -EAFNOSUPPORT;
165
166 if (likely(afinfo->dst_lookup != NULL))
167 err = afinfo->dst_lookup(dst, fl);
168 else
169 err = -EINVAL;
170 xfrm_policy_put_afinfo(afinfo);
171 return err;
172 }
173 EXPORT_SYMBOL(xfrm_dst_lookup);
174
175 void xfrm_put_type(struct xfrm_type *type)
176 {
177 module_put(type->owner);
178 }
179
180 int xfrm_register_mode(struct xfrm_mode *mode, int family)
181 {
182 struct xfrm_policy_afinfo *afinfo;
183 struct xfrm_mode **modemap;
184 int err;
185
186 if (unlikely(mode->encap >= XFRM_MODE_MAX))
187 return -EINVAL;
188
189 afinfo = xfrm_policy_lock_afinfo(family);
190 if (unlikely(afinfo == NULL))
191 return -EAFNOSUPPORT;
192
193 err = -EEXIST;
194 modemap = afinfo->mode_map;
195 if (likely(modemap[mode->encap] == NULL)) {
196 modemap[mode->encap] = mode;
197 err = 0;
198 }
199
200 xfrm_policy_unlock_afinfo(afinfo);
201 return err;
202 }
203 EXPORT_SYMBOL(xfrm_register_mode);
204
205 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
206 {
207 struct xfrm_policy_afinfo *afinfo;
208 struct xfrm_mode **modemap;
209 int err;
210
211 if (unlikely(mode->encap >= XFRM_MODE_MAX))
212 return -EINVAL;
213
214 afinfo = xfrm_policy_lock_afinfo(family);
215 if (unlikely(afinfo == NULL))
216 return -EAFNOSUPPORT;
217
218 err = -ENOENT;
219 modemap = afinfo->mode_map;
220 if (likely(modemap[mode->encap] == mode)) {
221 modemap[mode->encap] = NULL;
222 err = 0;
223 }
224
225 xfrm_policy_unlock_afinfo(afinfo);
226 return err;
227 }
228 EXPORT_SYMBOL(xfrm_unregister_mode);
229
230 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
231 {
232 struct xfrm_policy_afinfo *afinfo;
233 struct xfrm_mode *mode;
234 int modload_attempted = 0;
235
236 if (unlikely(encap >= XFRM_MODE_MAX))
237 return NULL;
238
239 retry:
240 afinfo = xfrm_policy_get_afinfo(family);
241 if (unlikely(afinfo == NULL))
242 return NULL;
243
244 mode = afinfo->mode_map[encap];
245 if (unlikely(mode && !try_module_get(mode->owner)))
246 mode = NULL;
247 if (!mode && !modload_attempted) {
248 xfrm_policy_put_afinfo(afinfo);
249 request_module("xfrm-mode-%d-%d", family, encap);
250 modload_attempted = 1;
251 goto retry;
252 }
253
254 xfrm_policy_put_afinfo(afinfo);
255 return mode;
256 }
257
258 void xfrm_put_mode(struct xfrm_mode *mode)
259 {
260 module_put(mode->owner);
261 }
262
263 static inline unsigned long make_jiffies(long secs)
264 {
265 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
266 return MAX_SCHEDULE_TIMEOUT-1;
267 else
268 return secs*HZ;
269 }
270
271 static void xfrm_policy_timer(unsigned long data)
272 {
273 struct xfrm_policy *xp = (struct xfrm_policy*)data;
274 unsigned long now = get_seconds();
275 long next = LONG_MAX;
276 int warn = 0;
277 int dir;
278
279 read_lock(&xp->lock);
280
281 if (xp->dead)
282 goto out;
283
284 dir = xfrm_policy_id2dir(xp->index);
285
286 if (xp->lft.hard_add_expires_seconds) {
287 long tmo = xp->lft.hard_add_expires_seconds +
288 xp->curlft.add_time - now;
289 if (tmo <= 0)
290 goto expired;
291 if (tmo < next)
292 next = tmo;
293 }
294 if (xp->lft.hard_use_expires_seconds) {
295 long tmo = xp->lft.hard_use_expires_seconds +
296 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
297 if (tmo <= 0)
298 goto expired;
299 if (tmo < next)
300 next = tmo;
301 }
302 if (xp->lft.soft_add_expires_seconds) {
303 long tmo = xp->lft.soft_add_expires_seconds +
304 xp->curlft.add_time - now;
305 if (tmo <= 0) {
306 warn = 1;
307 tmo = XFRM_KM_TIMEOUT;
308 }
309 if (tmo < next)
310 next = tmo;
311 }
312 if (xp->lft.soft_use_expires_seconds) {
313 long tmo = xp->lft.soft_use_expires_seconds +
314 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
315 if (tmo <= 0) {
316 warn = 1;
317 tmo = XFRM_KM_TIMEOUT;
318 }
319 if (tmo < next)
320 next = tmo;
321 }
322
323 if (warn)
324 km_policy_expired(xp, dir, 0, 0);
325 if (next != LONG_MAX &&
326 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
327 xfrm_pol_hold(xp);
328
329 out:
330 read_unlock(&xp->lock);
331 xfrm_pol_put(xp);
332 return;
333
334 expired:
335 read_unlock(&xp->lock);
336 if (!xfrm_policy_delete(xp, dir))
337 km_policy_expired(xp, dir, 1, 0);
338 xfrm_pol_put(xp);
339 }
340
341
342 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
343 * SPD calls.
344 */
345
346 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
347 {
348 struct xfrm_policy *policy;
349
350 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
351
352 if (policy) {
353 INIT_HLIST_NODE(&policy->bydst);
354 INIT_HLIST_NODE(&policy->byidx);
355 rwlock_init(&policy->lock);
356 atomic_set(&policy->refcnt, 1);
357 init_timer(&policy->timer);
358 policy->timer.data = (unsigned long)policy;
359 policy->timer.function = xfrm_policy_timer;
360 }
361 return policy;
362 }
363 EXPORT_SYMBOL(xfrm_policy_alloc);
364
365 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
366
367 void __xfrm_policy_destroy(struct xfrm_policy *policy)
368 {
369 BUG_ON(!policy->dead);
370
371 BUG_ON(policy->bundles);
372
373 if (del_timer(&policy->timer))
374 BUG();
375
376 security_xfrm_policy_free(policy);
377 kfree(policy);
378 }
379 EXPORT_SYMBOL(__xfrm_policy_destroy);
380
381 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
382 {
383 struct dst_entry *dst;
384
385 while ((dst = policy->bundles) != NULL) {
386 policy->bundles = dst->next;
387 dst_free(dst);
388 }
389
390 if (del_timer(&policy->timer))
391 atomic_dec(&policy->refcnt);
392
393 if (atomic_read(&policy->refcnt) > 1)
394 flow_cache_flush();
395
396 xfrm_pol_put(policy);
397 }
398
399 static void xfrm_policy_gc_task(struct work_struct *work)
400 {
401 struct xfrm_policy *policy;
402 struct hlist_node *entry, *tmp;
403 struct hlist_head gc_list;
404
405 spin_lock_bh(&xfrm_policy_gc_lock);
406 gc_list.first = xfrm_policy_gc_list.first;
407 INIT_HLIST_HEAD(&xfrm_policy_gc_list);
408 spin_unlock_bh(&xfrm_policy_gc_lock);
409
410 hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
411 xfrm_policy_gc_kill(policy);
412 }
413
414 /* Rule must be locked. Release descentant resources, announce
415 * entry dead. The rule must be unlinked from lists to the moment.
416 */
417
418 static void xfrm_policy_kill(struct xfrm_policy *policy)
419 {
420 int dead;
421
422 write_lock_bh(&policy->lock);
423 dead = policy->dead;
424 policy->dead = 1;
425 write_unlock_bh(&policy->lock);
426
427 if (unlikely(dead)) {
428 WARN_ON(1);
429 return;
430 }
431
432 spin_lock(&xfrm_policy_gc_lock);
433 hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
434 spin_unlock(&xfrm_policy_gc_lock);
435
436 schedule_work(&xfrm_policy_gc_work);
437 }
438
439 struct xfrm_policy_hash {
440 struct hlist_head *table;
441 unsigned int hmask;
442 };
443
444 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
445 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
446 static struct hlist_head *xfrm_policy_byidx __read_mostly;
447 static unsigned int xfrm_idx_hmask __read_mostly;
448 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
449
450 static inline unsigned int idx_hash(u32 index)
451 {
452 return __idx_hash(index, xfrm_idx_hmask);
453 }
454
455 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
456 {
457 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
458 unsigned int hash = __sel_hash(sel, family, hmask);
459
460 return (hash == hmask + 1 ?
461 &xfrm_policy_inexact[dir] :
462 xfrm_policy_bydst[dir].table + hash);
463 }
464
465 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
466 {
467 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
468 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
469
470 return xfrm_policy_bydst[dir].table + hash;
471 }
472
473 static void xfrm_dst_hash_transfer(struct hlist_head *list,
474 struct hlist_head *ndsttable,
475 unsigned int nhashmask)
476 {
477 struct hlist_node *entry, *tmp;
478 struct xfrm_policy *pol;
479
480 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
481 unsigned int h;
482
483 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
484 pol->family, nhashmask);
485 hlist_add_head(&pol->bydst, ndsttable+h);
486 }
487 }
488
489 static void xfrm_idx_hash_transfer(struct hlist_head *list,
490 struct hlist_head *nidxtable,
491 unsigned int nhashmask)
492 {
493 struct hlist_node *entry, *tmp;
494 struct xfrm_policy *pol;
495
496 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
497 unsigned int h;
498
499 h = __idx_hash(pol->index, nhashmask);
500 hlist_add_head(&pol->byidx, nidxtable+h);
501 }
502 }
503
504 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
505 {
506 return ((old_hmask + 1) << 1) - 1;
507 }
508
509 static void xfrm_bydst_resize(int dir)
510 {
511 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
512 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
513 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
514 struct hlist_head *odst = xfrm_policy_bydst[dir].table;
515 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
516 int i;
517
518 if (!ndst)
519 return;
520
521 write_lock_bh(&xfrm_policy_lock);
522
523 for (i = hmask; i >= 0; i--)
524 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
525
526 xfrm_policy_bydst[dir].table = ndst;
527 xfrm_policy_bydst[dir].hmask = nhashmask;
528
529 write_unlock_bh(&xfrm_policy_lock);
530
531 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
532 }
533
534 static void xfrm_byidx_resize(int total)
535 {
536 unsigned int hmask = xfrm_idx_hmask;
537 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
538 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
539 struct hlist_head *oidx = xfrm_policy_byidx;
540 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
541 int i;
542
543 if (!nidx)
544 return;
545
546 write_lock_bh(&xfrm_policy_lock);
547
548 for (i = hmask; i >= 0; i--)
549 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
550
551 xfrm_policy_byidx = nidx;
552 xfrm_idx_hmask = nhashmask;
553
554 write_unlock_bh(&xfrm_policy_lock);
555
556 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
557 }
558
559 static inline int xfrm_bydst_should_resize(int dir, int *total)
560 {
561 unsigned int cnt = xfrm_policy_count[dir];
562 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
563
564 if (total)
565 *total += cnt;
566
567 if ((hmask + 1) < xfrm_policy_hashmax &&
568 cnt > hmask)
569 return 1;
570
571 return 0;
572 }
573
574 static inline int xfrm_byidx_should_resize(int total)
575 {
576 unsigned int hmask = xfrm_idx_hmask;
577
578 if ((hmask + 1) < xfrm_policy_hashmax &&
579 total > hmask)
580 return 1;
581
582 return 0;
583 }
584
585 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
586 {
587 read_lock_bh(&xfrm_policy_lock);
588 si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
589 si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
590 si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
591 si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
592 si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
593 si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
594 si->spdhcnt = xfrm_idx_hmask;
595 si->spdhmcnt = xfrm_policy_hashmax;
596 read_unlock_bh(&xfrm_policy_lock);
597 }
598 EXPORT_SYMBOL(xfrm_spd_getinfo);
599
600 static DEFINE_MUTEX(hash_resize_mutex);
601 static void xfrm_hash_resize(struct work_struct *__unused)
602 {
603 int dir, total;
604
605 mutex_lock(&hash_resize_mutex);
606
607 total = 0;
608 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
609 if (xfrm_bydst_should_resize(dir, &total))
610 xfrm_bydst_resize(dir);
611 }
612 if (xfrm_byidx_should_resize(total))
613 xfrm_byidx_resize(total);
614
615 mutex_unlock(&hash_resize_mutex);
616 }
617
618 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
619
620 /* Generate new index... KAME seems to generate them ordered by cost
621 * of an absolute inpredictability of ordering of rules. This will not pass. */
622 static u32 xfrm_gen_index(u8 type, int dir)
623 {
624 static u32 idx_generator;
625
626 for (;;) {
627 struct hlist_node *entry;
628 struct hlist_head *list;
629 struct xfrm_policy *p;
630 u32 idx;
631 int found;
632
633 idx = (idx_generator | dir);
634 idx_generator += 8;
635 if (idx == 0)
636 idx = 8;
637 list = xfrm_policy_byidx + idx_hash(idx);
638 found = 0;
639 hlist_for_each_entry(p, entry, list, byidx) {
640 if (p->index == idx) {
641 found = 1;
642 break;
643 }
644 }
645 if (!found)
646 return idx;
647 }
648 }
649
650 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
651 {
652 u32 *p1 = (u32 *) s1;
653 u32 *p2 = (u32 *) s2;
654 int len = sizeof(struct xfrm_selector) / sizeof(u32);
655 int i;
656
657 for (i = 0; i < len; i++) {
658 if (p1[i] != p2[i])
659 return 1;
660 }
661
662 return 0;
663 }
664
665 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
666 {
667 struct xfrm_policy *pol;
668 struct xfrm_policy *delpol;
669 struct hlist_head *chain;
670 struct hlist_node *entry, *newpos;
671 struct dst_entry *gc_list;
672
673 write_lock_bh(&xfrm_policy_lock);
674 chain = policy_hash_bysel(&policy->selector, policy->family, dir);
675 delpol = NULL;
676 newpos = NULL;
677 hlist_for_each_entry(pol, entry, chain, bydst) {
678 if (pol->type == policy->type &&
679 !selector_cmp(&pol->selector, &policy->selector) &&
680 xfrm_sec_ctx_match(pol->security, policy->security) &&
681 !WARN_ON(delpol)) {
682 if (excl) {
683 write_unlock_bh(&xfrm_policy_lock);
684 return -EEXIST;
685 }
686 delpol = pol;
687 if (policy->priority > pol->priority)
688 continue;
689 } else if (policy->priority >= pol->priority) {
690 newpos = &pol->bydst;
691 continue;
692 }
693 if (delpol)
694 break;
695 }
696 if (newpos)
697 hlist_add_after(newpos, &policy->bydst);
698 else
699 hlist_add_head(&policy->bydst, chain);
700 xfrm_pol_hold(policy);
701 xfrm_policy_count[dir]++;
702 atomic_inc(&flow_cache_genid);
703 if (delpol) {
704 hlist_del(&delpol->bydst);
705 hlist_del(&delpol->byidx);
706 xfrm_policy_count[dir]--;
707 }
708 policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
709 hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
710 policy->curlft.add_time = get_seconds();
711 policy->curlft.use_time = 0;
712 if (!mod_timer(&policy->timer, jiffies + HZ))
713 xfrm_pol_hold(policy);
714 write_unlock_bh(&xfrm_policy_lock);
715
716 if (delpol)
717 xfrm_policy_kill(delpol);
718 else if (xfrm_bydst_should_resize(dir, NULL))
719 schedule_work(&xfrm_hash_work);
720
721 read_lock_bh(&xfrm_policy_lock);
722 gc_list = NULL;
723 entry = &policy->bydst;
724 hlist_for_each_entry_continue(policy, entry, bydst) {
725 struct dst_entry *dst;
726
727 write_lock(&policy->lock);
728 dst = policy->bundles;
729 if (dst) {
730 struct dst_entry *tail = dst;
731 while (tail->next)
732 tail = tail->next;
733 tail->next = gc_list;
734 gc_list = dst;
735
736 policy->bundles = NULL;
737 }
738 write_unlock(&policy->lock);
739 }
740 read_unlock_bh(&xfrm_policy_lock);
741
742 while (gc_list) {
743 struct dst_entry *dst = gc_list;
744
745 gc_list = dst->next;
746 dst_free(dst);
747 }
748
749 return 0;
750 }
751 EXPORT_SYMBOL(xfrm_policy_insert);
752
753 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
754 struct xfrm_selector *sel,
755 struct xfrm_sec_ctx *ctx, int delete,
756 int *err)
757 {
758 struct xfrm_policy *pol, *ret;
759 struct hlist_head *chain;
760 struct hlist_node *entry;
761
762 *err = 0;
763 write_lock_bh(&xfrm_policy_lock);
764 chain = policy_hash_bysel(sel, sel->family, dir);
765 ret = NULL;
766 hlist_for_each_entry(pol, entry, chain, bydst) {
767 if (pol->type == type &&
768 !selector_cmp(sel, &pol->selector) &&
769 xfrm_sec_ctx_match(ctx, pol->security)) {
770 xfrm_pol_hold(pol);
771 if (delete) {
772 *err = security_xfrm_policy_delete(pol);
773 if (*err) {
774 write_unlock_bh(&xfrm_policy_lock);
775 return pol;
776 }
777 hlist_del(&pol->bydst);
778 hlist_del(&pol->byidx);
779 xfrm_policy_count[dir]--;
780 }
781 ret = pol;
782 break;
783 }
784 }
785 write_unlock_bh(&xfrm_policy_lock);
786
787 if (ret && delete) {
788 atomic_inc(&flow_cache_genid);
789 xfrm_policy_kill(ret);
790 }
791 return ret;
792 }
793 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
794
795 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
796 int *err)
797 {
798 struct xfrm_policy *pol, *ret;
799 struct hlist_head *chain;
800 struct hlist_node *entry;
801
802 *err = -ENOENT;
803 if (xfrm_policy_id2dir(id) != dir)
804 return NULL;
805
806 *err = 0;
807 write_lock_bh(&xfrm_policy_lock);
808 chain = xfrm_policy_byidx + idx_hash(id);
809 ret = NULL;
810 hlist_for_each_entry(pol, entry, chain, byidx) {
811 if (pol->type == type && pol->index == id) {
812 xfrm_pol_hold(pol);
813 if (delete) {
814 *err = security_xfrm_policy_delete(pol);
815 if (*err) {
816 write_unlock_bh(&xfrm_policy_lock);
817 return pol;
818 }
819 hlist_del(&pol->bydst);
820 hlist_del(&pol->byidx);
821 xfrm_policy_count[dir]--;
822 }
823 ret = pol;
824 break;
825 }
826 }
827 write_unlock_bh(&xfrm_policy_lock);
828
829 if (ret && delete) {
830 atomic_inc(&flow_cache_genid);
831 xfrm_policy_kill(ret);
832 }
833 return ret;
834 }
835 EXPORT_SYMBOL(xfrm_policy_byid);
836
837 #ifdef CONFIG_SECURITY_NETWORK_XFRM
838 static inline int
839 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
840 {
841 int dir, err = 0;
842
843 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
844 struct xfrm_policy *pol;
845 struct hlist_node *entry;
846 int i;
847
848 hlist_for_each_entry(pol, entry,
849 &xfrm_policy_inexact[dir], bydst) {
850 if (pol->type != type)
851 continue;
852 err = security_xfrm_policy_delete(pol);
853 if (err) {
854 xfrm_audit_log(audit_info->loginuid,
855 audit_info->secid,
856 AUDIT_MAC_IPSEC_DELSPD, 0,
857 pol, NULL);
858 return err;
859 }
860 }
861 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
862 hlist_for_each_entry(pol, entry,
863 xfrm_policy_bydst[dir].table + i,
864 bydst) {
865 if (pol->type != type)
866 continue;
867 err = security_xfrm_policy_delete(pol);
868 if (err) {
869 xfrm_audit_log(audit_info->loginuid,
870 audit_info->secid,
871 AUDIT_MAC_IPSEC_DELSPD,
872 0, pol, NULL);
873 return err;
874 }
875 }
876 }
877 }
878 return err;
879 }
880 #else
881 static inline int
882 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
883 {
884 return 0;
885 }
886 #endif
887
888 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
889 {
890 int dir, err = 0;
891
892 write_lock_bh(&xfrm_policy_lock);
893
894 err = xfrm_policy_flush_secctx_check(type, audit_info);
895 if (err)
896 goto out;
897
898 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
899 struct xfrm_policy *pol;
900 struct hlist_node *entry;
901 int i, killed;
902
903 killed = 0;
904 again1:
905 hlist_for_each_entry(pol, entry,
906 &xfrm_policy_inexact[dir], bydst) {
907 if (pol->type != type)
908 continue;
909 hlist_del(&pol->bydst);
910 hlist_del(&pol->byidx);
911 write_unlock_bh(&xfrm_policy_lock);
912
913 xfrm_audit_log(audit_info->loginuid, audit_info->secid,
914 AUDIT_MAC_IPSEC_DELSPD, 1, pol, NULL);
915
916 xfrm_policy_kill(pol);
917 killed++;
918
919 write_lock_bh(&xfrm_policy_lock);
920 goto again1;
921 }
922
923 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
924 again2:
925 hlist_for_each_entry(pol, entry,
926 xfrm_policy_bydst[dir].table + i,
927 bydst) {
928 if (pol->type != type)
929 continue;
930 hlist_del(&pol->bydst);
931 hlist_del(&pol->byidx);
932 write_unlock_bh(&xfrm_policy_lock);
933
934 xfrm_audit_log(audit_info->loginuid,
935 audit_info->secid,
936 AUDIT_MAC_IPSEC_DELSPD, 1,
937 pol, NULL);
938
939 xfrm_policy_kill(pol);
940 killed++;
941
942 write_lock_bh(&xfrm_policy_lock);
943 goto again2;
944 }
945 }
946
947 xfrm_policy_count[dir] -= killed;
948 }
949 atomic_inc(&flow_cache_genid);
950 out:
951 write_unlock_bh(&xfrm_policy_lock);
952 return err;
953 }
954 EXPORT_SYMBOL(xfrm_policy_flush);
955
956 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
957 void *data)
958 {
959 struct xfrm_policy *pol, *last = NULL;
960 struct hlist_node *entry;
961 int dir, last_dir = 0, count, error;
962
963 read_lock_bh(&xfrm_policy_lock);
964 count = 0;
965
966 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
967 struct hlist_head *table = xfrm_policy_bydst[dir].table;
968 int i;
969
970 hlist_for_each_entry(pol, entry,
971 &xfrm_policy_inexact[dir], bydst) {
972 if (pol->type != type)
973 continue;
974 if (last) {
975 error = func(last, last_dir % XFRM_POLICY_MAX,
976 count, data);
977 if (error)
978 goto out;
979 }
980 last = pol;
981 last_dir = dir;
982 count++;
983 }
984 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
985 hlist_for_each_entry(pol, entry, table + i, bydst) {
986 if (pol->type != type)
987 continue;
988 if (last) {
989 error = func(last, last_dir % XFRM_POLICY_MAX,
990 count, data);
991 if (error)
992 goto out;
993 }
994 last = pol;
995 last_dir = dir;
996 count++;
997 }
998 }
999 }
1000 if (count == 0) {
1001 error = -ENOENT;
1002 goto out;
1003 }
1004 error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
1005 out:
1006 read_unlock_bh(&xfrm_policy_lock);
1007 return error;
1008 }
1009 EXPORT_SYMBOL(xfrm_policy_walk);
1010
1011 /*
1012 * Find policy to apply to this flow.
1013 *
1014 * Returns 0 if policy found, else an -errno.
1015 */
1016 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
1017 u8 type, u16 family, int dir)
1018 {
1019 struct xfrm_selector *sel = &pol->selector;
1020 int match, ret = -ESRCH;
1021
1022 if (pol->family != family ||
1023 pol->type != type)
1024 return ret;
1025
1026 match = xfrm_selector_match(sel, fl, family);
1027 if (match)
1028 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
1029
1030 return ret;
1031 }
1032
1033 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
1034 u16 family, u8 dir)
1035 {
1036 int err;
1037 struct xfrm_policy *pol, *ret;
1038 xfrm_address_t *daddr, *saddr;
1039 struct hlist_node *entry;
1040 struct hlist_head *chain;
1041 u32 priority = ~0U;
1042
1043 daddr = xfrm_flowi_daddr(fl, family);
1044 saddr = xfrm_flowi_saddr(fl, family);
1045 if (unlikely(!daddr || !saddr))
1046 return NULL;
1047
1048 read_lock_bh(&xfrm_policy_lock);
1049 chain = policy_hash_direct(daddr, saddr, family, dir);
1050 ret = NULL;
1051 hlist_for_each_entry(pol, entry, chain, bydst) {
1052 err = xfrm_policy_match(pol, fl, type, family, dir);
1053 if (err) {
1054 if (err == -ESRCH)
1055 continue;
1056 else {
1057 ret = ERR_PTR(err);
1058 goto fail;
1059 }
1060 } else {
1061 ret = pol;
1062 priority = ret->priority;
1063 break;
1064 }
1065 }
1066 chain = &xfrm_policy_inexact[dir];
1067 hlist_for_each_entry(pol, entry, chain, bydst) {
1068 err = xfrm_policy_match(pol, fl, type, family, dir);
1069 if (err) {
1070 if (err == -ESRCH)
1071 continue;
1072 else {
1073 ret = ERR_PTR(err);
1074 goto fail;
1075 }
1076 } else if (pol->priority < priority) {
1077 ret = pol;
1078 break;
1079 }
1080 }
1081 if (ret)
1082 xfrm_pol_hold(ret);
1083 fail:
1084 read_unlock_bh(&xfrm_policy_lock);
1085
1086 return ret;
1087 }
1088
1089 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1090 void **objp, atomic_t **obj_refp)
1091 {
1092 struct xfrm_policy *pol;
1093 int err = 0;
1094
1095 #ifdef CONFIG_XFRM_SUB_POLICY
1096 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1097 if (IS_ERR(pol)) {
1098 err = PTR_ERR(pol);
1099 pol = NULL;
1100 }
1101 if (pol || err)
1102 goto end;
1103 #endif
1104 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1105 if (IS_ERR(pol)) {
1106 err = PTR_ERR(pol);
1107 pol = NULL;
1108 }
1109 #ifdef CONFIG_XFRM_SUB_POLICY
1110 end:
1111 #endif
1112 if ((*objp = (void *) pol) != NULL)
1113 *obj_refp = &pol->refcnt;
1114 return err;
1115 }
1116
1117 static inline int policy_to_flow_dir(int dir)
1118 {
1119 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1120 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1121 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1122 return dir;
1123 switch (dir) {
1124 default:
1125 case XFRM_POLICY_IN:
1126 return FLOW_DIR_IN;
1127 case XFRM_POLICY_OUT:
1128 return FLOW_DIR_OUT;
1129 case XFRM_POLICY_FWD:
1130 return FLOW_DIR_FWD;
1131 }
1132 }
1133
1134 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1135 {
1136 struct xfrm_policy *pol;
1137
1138 read_lock_bh(&xfrm_policy_lock);
1139 if ((pol = sk->sk_policy[dir]) != NULL) {
1140 int match = xfrm_selector_match(&pol->selector, fl,
1141 sk->sk_family);
1142 int err = 0;
1143
1144 if (match) {
1145 err = security_xfrm_policy_lookup(pol, fl->secid,
1146 policy_to_flow_dir(dir));
1147 if (!err)
1148 xfrm_pol_hold(pol);
1149 else if (err == -ESRCH)
1150 pol = NULL;
1151 else
1152 pol = ERR_PTR(err);
1153 } else
1154 pol = NULL;
1155 }
1156 read_unlock_bh(&xfrm_policy_lock);
1157 return pol;
1158 }
1159
1160 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1161 {
1162 struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1163 pol->family, dir);
1164
1165 hlist_add_head(&pol->bydst, chain);
1166 hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1167 xfrm_policy_count[dir]++;
1168 xfrm_pol_hold(pol);
1169
1170 if (xfrm_bydst_should_resize(dir, NULL))
1171 schedule_work(&xfrm_hash_work);
1172 }
1173
1174 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1175 int dir)
1176 {
1177 if (hlist_unhashed(&pol->bydst))
1178 return NULL;
1179
1180 hlist_del(&pol->bydst);
1181 hlist_del(&pol->byidx);
1182 xfrm_policy_count[dir]--;
1183
1184 return pol;
1185 }
1186
1187 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1188 {
1189 write_lock_bh(&xfrm_policy_lock);
1190 pol = __xfrm_policy_unlink(pol, dir);
1191 write_unlock_bh(&xfrm_policy_lock);
1192 if (pol) {
1193 if (dir < XFRM_POLICY_MAX)
1194 atomic_inc(&flow_cache_genid);
1195 xfrm_policy_kill(pol);
1196 return 0;
1197 }
1198 return -ENOENT;
1199 }
1200 EXPORT_SYMBOL(xfrm_policy_delete);
1201
1202 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1203 {
1204 struct xfrm_policy *old_pol;
1205
1206 #ifdef CONFIG_XFRM_SUB_POLICY
1207 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1208 return -EINVAL;
1209 #endif
1210
1211 write_lock_bh(&xfrm_policy_lock);
1212 old_pol = sk->sk_policy[dir];
1213 sk->sk_policy[dir] = pol;
1214 if (pol) {
1215 pol->curlft.add_time = get_seconds();
1216 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1217 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1218 }
1219 if (old_pol)
1220 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1221 write_unlock_bh(&xfrm_policy_lock);
1222
1223 if (old_pol) {
1224 xfrm_policy_kill(old_pol);
1225 }
1226 return 0;
1227 }
1228
1229 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1230 {
1231 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1232
1233 if (newp) {
1234 newp->selector = old->selector;
1235 if (security_xfrm_policy_clone(old, newp)) {
1236 kfree(newp);
1237 return NULL; /* ENOMEM */
1238 }
1239 newp->lft = old->lft;
1240 newp->curlft = old->curlft;
1241 newp->action = old->action;
1242 newp->flags = old->flags;
1243 newp->xfrm_nr = old->xfrm_nr;
1244 newp->index = old->index;
1245 newp->type = old->type;
1246 memcpy(newp->xfrm_vec, old->xfrm_vec,
1247 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1248 write_lock_bh(&xfrm_policy_lock);
1249 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1250 write_unlock_bh(&xfrm_policy_lock);
1251 xfrm_pol_put(newp);
1252 }
1253 return newp;
1254 }
1255
1256 int __xfrm_sk_clone_policy(struct sock *sk)
1257 {
1258 struct xfrm_policy *p0 = sk->sk_policy[0],
1259 *p1 = sk->sk_policy[1];
1260
1261 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1262 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1263 return -ENOMEM;
1264 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1265 return -ENOMEM;
1266 return 0;
1267 }
1268
1269 static int
1270 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1271 unsigned short family)
1272 {
1273 int err;
1274 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1275
1276 if (unlikely(afinfo == NULL))
1277 return -EINVAL;
1278 err = afinfo->get_saddr(local, remote);
1279 xfrm_policy_put_afinfo(afinfo);
1280 return err;
1281 }
1282
1283 /* Resolve list of templates for the flow, given policy. */
1284
1285 static int
1286 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1287 struct xfrm_state **xfrm,
1288 unsigned short family)
1289 {
1290 int nx;
1291 int i, error;
1292 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1293 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1294 xfrm_address_t tmp;
1295
1296 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1297 struct xfrm_state *x;
1298 xfrm_address_t *remote = daddr;
1299 xfrm_address_t *local = saddr;
1300 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1301
1302 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1303 tmpl->mode == XFRM_MODE_BEET) {
1304 remote = &tmpl->id.daddr;
1305 local = &tmpl->saddr;
1306 family = tmpl->encap_family;
1307 if (xfrm_addr_any(local, family)) {
1308 error = xfrm_get_saddr(&tmp, remote, family);
1309 if (error)
1310 goto fail;
1311 local = &tmp;
1312 }
1313 }
1314
1315 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1316
1317 if (x && x->km.state == XFRM_STATE_VALID) {
1318 xfrm[nx++] = x;
1319 daddr = remote;
1320 saddr = local;
1321 continue;
1322 }
1323 if (x) {
1324 error = (x->km.state == XFRM_STATE_ERROR ?
1325 -EINVAL : -EAGAIN);
1326 xfrm_state_put(x);
1327 }
1328
1329 if (!tmpl->optional)
1330 goto fail;
1331 }
1332 return nx;
1333
1334 fail:
1335 for (nx--; nx>=0; nx--)
1336 xfrm_state_put(xfrm[nx]);
1337 return error;
1338 }
1339
1340 static int
1341 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1342 struct xfrm_state **xfrm,
1343 unsigned short family)
1344 {
1345 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1346 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1347 int cnx = 0;
1348 int error;
1349 int ret;
1350 int i;
1351
1352 for (i = 0; i < npols; i++) {
1353 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1354 error = -ENOBUFS;
1355 goto fail;
1356 }
1357
1358 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1359 if (ret < 0) {
1360 error = ret;
1361 goto fail;
1362 } else
1363 cnx += ret;
1364 }
1365
1366 /* found states are sorted for outbound processing */
1367 if (npols > 1)
1368 xfrm_state_sort(xfrm, tpp, cnx, family);
1369
1370 return cnx;
1371
1372 fail:
1373 for (cnx--; cnx>=0; cnx--)
1374 xfrm_state_put(tpp[cnx]);
1375 return error;
1376
1377 }
1378
1379 /* Check that the bundle accepts the flow and its components are
1380 * still valid.
1381 */
1382
1383 static struct dst_entry *
1384 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1385 {
1386 struct dst_entry *x;
1387 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1388 if (unlikely(afinfo == NULL))
1389 return ERR_PTR(-EINVAL);
1390 x = afinfo->find_bundle(fl, policy);
1391 xfrm_policy_put_afinfo(afinfo);
1392 return x;
1393 }
1394
1395 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1396 * all the metrics... Shortly, bundle a bundle.
1397 */
1398
1399 static int
1400 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1401 struct flowi *fl, struct dst_entry **dst_p,
1402 unsigned short family)
1403 {
1404 int err;
1405 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1406 if (unlikely(afinfo == NULL))
1407 return -EINVAL;
1408 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1409 xfrm_policy_put_afinfo(afinfo);
1410 return err;
1411 }
1412
1413 static int inline
1414 xfrm_dst_alloc_copy(void **target, void *src, int size)
1415 {
1416 if (!*target) {
1417 *target = kmalloc(size, GFP_ATOMIC);
1418 if (!*target)
1419 return -ENOMEM;
1420 }
1421 memcpy(*target, src, size);
1422 return 0;
1423 }
1424
1425 static int inline
1426 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1427 {
1428 #ifdef CONFIG_XFRM_SUB_POLICY
1429 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1430 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1431 sel, sizeof(*sel));
1432 #else
1433 return 0;
1434 #endif
1435 }
1436
1437 static int inline
1438 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1439 {
1440 #ifdef CONFIG_XFRM_SUB_POLICY
1441 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1442 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1443 #else
1444 return 0;
1445 #endif
1446 }
1447
1448 static int stale_bundle(struct dst_entry *dst);
1449
1450 /* Main function: finds/creates a bundle for given flow.
1451 *
1452 * At the moment we eat a raw IP route. Mostly to speed up lookups
1453 * on interfaces with disabled IPsec.
1454 */
1455 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1456 struct sock *sk, int flags)
1457 {
1458 struct xfrm_policy *policy;
1459 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1460 int npols;
1461 int pol_dead;
1462 int xfrm_nr;
1463 int pi;
1464 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1465 struct dst_entry *dst, *dst_orig = *dst_p;
1466 int nx = 0;
1467 int err;
1468 u32 genid;
1469 u16 family;
1470 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1471
1472 restart:
1473 genid = atomic_read(&flow_cache_genid);
1474 policy = NULL;
1475 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1476 pols[pi] = NULL;
1477 npols = 0;
1478 pol_dead = 0;
1479 xfrm_nr = 0;
1480
1481 if (sk && sk->sk_policy[1]) {
1482 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1483 if (IS_ERR(policy))
1484 return PTR_ERR(policy);
1485 }
1486
1487 if (!policy) {
1488 /* To accelerate a bit... */
1489 if ((dst_orig->flags & DST_NOXFRM) ||
1490 !xfrm_policy_count[XFRM_POLICY_OUT])
1491 return 0;
1492
1493 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1494 dir, xfrm_policy_lookup);
1495 if (IS_ERR(policy))
1496 return PTR_ERR(policy);
1497 }
1498
1499 if (!policy)
1500 return 0;
1501
1502 family = dst_orig->ops->family;
1503 policy->curlft.use_time = get_seconds();
1504 pols[0] = policy;
1505 npols ++;
1506 xfrm_nr += pols[0]->xfrm_nr;
1507
1508 switch (policy->action) {
1509 case XFRM_POLICY_BLOCK:
1510 /* Prohibit the flow */
1511 err = -EPERM;
1512 goto error;
1513
1514 case XFRM_POLICY_ALLOW:
1515 #ifndef CONFIG_XFRM_SUB_POLICY
1516 if (policy->xfrm_nr == 0) {
1517 /* Flow passes not transformed. */
1518 xfrm_pol_put(policy);
1519 return 0;
1520 }
1521 #endif
1522
1523 /* Try to find matching bundle.
1524 *
1525 * LATER: help from flow cache. It is optional, this
1526 * is required only for output policy.
1527 */
1528 dst = xfrm_find_bundle(fl, policy, family);
1529 if (IS_ERR(dst)) {
1530 err = PTR_ERR(dst);
1531 goto error;
1532 }
1533
1534 if (dst)
1535 break;
1536
1537 #ifdef CONFIG_XFRM_SUB_POLICY
1538 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1539 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1540 fl, family,
1541 XFRM_POLICY_OUT);
1542 if (pols[1]) {
1543 if (IS_ERR(pols[1])) {
1544 err = PTR_ERR(pols[1]);
1545 goto error;
1546 }
1547 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1548 err = -EPERM;
1549 goto error;
1550 }
1551 npols ++;
1552 xfrm_nr += pols[1]->xfrm_nr;
1553 }
1554 }
1555
1556 /*
1557 * Because neither flowi nor bundle information knows about
1558 * transformation template size. On more than one policy usage
1559 * we can realize whether all of them is bypass or not after
1560 * they are searched. See above not-transformed bypass
1561 * is surrounded by non-sub policy configuration, too.
1562 */
1563 if (xfrm_nr == 0) {
1564 /* Flow passes not transformed. */
1565 xfrm_pols_put(pols, npols);
1566 return 0;
1567 }
1568
1569 #endif
1570 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1571
1572 if (unlikely(nx<0)) {
1573 err = nx;
1574 if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1575 /* EREMOTE tells the caller to generate
1576 * a one-shot blackhole route.
1577 */
1578 xfrm_pol_put(policy);
1579 return -EREMOTE;
1580 }
1581 if (err == -EAGAIN && flags) {
1582 DECLARE_WAITQUEUE(wait, current);
1583
1584 add_wait_queue(&km_waitq, &wait);
1585 set_current_state(TASK_INTERRUPTIBLE);
1586 schedule();
1587 set_current_state(TASK_RUNNING);
1588 remove_wait_queue(&km_waitq, &wait);
1589
1590 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1591
1592 if (nx == -EAGAIN && signal_pending(current)) {
1593 err = -ERESTART;
1594 goto error;
1595 }
1596 if (nx == -EAGAIN ||
1597 genid != atomic_read(&flow_cache_genid)) {
1598 xfrm_pols_put(pols, npols);
1599 goto restart;
1600 }
1601 err = nx;
1602 }
1603 if (err < 0)
1604 goto error;
1605 }
1606 if (nx == 0) {
1607 /* Flow passes not transformed. */
1608 xfrm_pols_put(pols, npols);
1609 return 0;
1610 }
1611
1612 dst = dst_orig;
1613 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1614
1615 if (unlikely(err)) {
1616 int i;
1617 for (i=0; i<nx; i++)
1618 xfrm_state_put(xfrm[i]);
1619 goto error;
1620 }
1621
1622 for (pi = 0; pi < npols; pi++) {
1623 read_lock_bh(&pols[pi]->lock);
1624 pol_dead |= pols[pi]->dead;
1625 read_unlock_bh(&pols[pi]->lock);
1626 }
1627
1628 write_lock_bh(&policy->lock);
1629 if (unlikely(pol_dead || stale_bundle(dst))) {
1630 /* Wow! While we worked on resolving, this
1631 * policy has gone. Retry. It is not paranoia,
1632 * we just cannot enlist new bundle to dead object.
1633 * We can't enlist stable bundles either.
1634 */
1635 write_unlock_bh(&policy->lock);
1636 if (dst)
1637 dst_free(dst);
1638
1639 err = -EHOSTUNREACH;
1640 goto error;
1641 }
1642
1643 if (npols > 1)
1644 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1645 else
1646 err = xfrm_dst_update_origin(dst, fl);
1647 if (unlikely(err)) {
1648 write_unlock_bh(&policy->lock);
1649 if (dst)
1650 dst_free(dst);
1651 goto error;
1652 }
1653
1654 dst->next = policy->bundles;
1655 policy->bundles = dst;
1656 dst_hold(dst);
1657 write_unlock_bh(&policy->lock);
1658 }
1659 *dst_p = dst;
1660 dst_release(dst_orig);
1661 xfrm_pols_put(pols, npols);
1662 return 0;
1663
1664 error:
1665 dst_release(dst_orig);
1666 xfrm_pols_put(pols, npols);
1667 *dst_p = NULL;
1668 return err;
1669 }
1670 EXPORT_SYMBOL(__xfrm_lookup);
1671
1672 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1673 struct sock *sk, int flags)
1674 {
1675 int err = __xfrm_lookup(dst_p, fl, sk, flags);
1676
1677 if (err == -EREMOTE) {
1678 dst_release(*dst_p);
1679 *dst_p = NULL;
1680 err = -EAGAIN;
1681 }
1682
1683 return err;
1684 }
1685 EXPORT_SYMBOL(xfrm_lookup);
1686
1687 static inline int
1688 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1689 {
1690 struct xfrm_state *x;
1691 int err;
1692
1693 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1694 return 0;
1695 x = skb->sp->xvec[idx];
1696 if (!x->type->reject)
1697 return 0;
1698 xfrm_state_hold(x);
1699 err = x->type->reject(x, skb, fl);
1700 xfrm_state_put(x);
1701 return err;
1702 }
1703
1704 /* When skb is transformed back to its "native" form, we have to
1705 * check policy restrictions. At the moment we make this in maximally
1706 * stupid way. Shame on me. :-) Of course, connected sockets must
1707 * have policy cached at them.
1708 */
1709
1710 static inline int
1711 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1712 unsigned short family)
1713 {
1714 if (xfrm_state_kern(x))
1715 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1716 return x->id.proto == tmpl->id.proto &&
1717 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1718 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1719 x->props.mode == tmpl->mode &&
1720 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1721 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1722 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1723 xfrm_state_addr_cmp(tmpl, x, family));
1724 }
1725
1726 /*
1727 * 0 or more than 0 is returned when validation is succeeded (either bypass
1728 * because of optional transport mode, or next index of the mathced secpath
1729 * state with the template.
1730 * -1 is returned when no matching template is found.
1731 * Otherwise "-2 - errored_index" is returned.
1732 */
1733 static inline int
1734 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1735 unsigned short family)
1736 {
1737 int idx = start;
1738
1739 if (tmpl->optional) {
1740 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1741 return start;
1742 } else
1743 start = -1;
1744 for (; idx < sp->len; idx++) {
1745 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1746 return ++idx;
1747 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1748 if (start == -1)
1749 start = -2-idx;
1750 break;
1751 }
1752 }
1753 return start;
1754 }
1755
1756 int
1757 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1758 {
1759 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1760 int err;
1761
1762 if (unlikely(afinfo == NULL))
1763 return -EAFNOSUPPORT;
1764
1765 afinfo->decode_session(skb, fl);
1766 err = security_xfrm_decode_session(skb, &fl->secid);
1767 xfrm_policy_put_afinfo(afinfo);
1768 return err;
1769 }
1770 EXPORT_SYMBOL(xfrm_decode_session);
1771
1772 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1773 {
1774 for (; k < sp->len; k++) {
1775 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1776 *idxp = k;
1777 return 1;
1778 }
1779 }
1780
1781 return 0;
1782 }
1783
1784 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1785 unsigned short family)
1786 {
1787 struct xfrm_policy *pol;
1788 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1789 int npols = 0;
1790 int xfrm_nr;
1791 int pi;
1792 struct flowi fl;
1793 u8 fl_dir = policy_to_flow_dir(dir);
1794 int xerr_idx = -1;
1795
1796 if (xfrm_decode_session(skb, &fl, family) < 0)
1797 return 0;
1798 nf_nat_decode_session(skb, &fl, family);
1799
1800 /* First, check used SA against their selectors. */
1801 if (skb->sp) {
1802 int i;
1803
1804 for (i=skb->sp->len-1; i>=0; i--) {
1805 struct xfrm_state *x = skb->sp->xvec[i];
1806 if (!xfrm_selector_match(&x->sel, &fl, family))
1807 return 0;
1808 }
1809 }
1810
1811 pol = NULL;
1812 if (sk && sk->sk_policy[dir]) {
1813 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1814 if (IS_ERR(pol))
1815 return 0;
1816 }
1817
1818 if (!pol)
1819 pol = flow_cache_lookup(&fl, family, fl_dir,
1820 xfrm_policy_lookup);
1821
1822 if (IS_ERR(pol))
1823 return 0;
1824
1825 if (!pol) {
1826 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1827 xfrm_secpath_reject(xerr_idx, skb, &fl);
1828 return 0;
1829 }
1830 return 1;
1831 }
1832
1833 pol->curlft.use_time = get_seconds();
1834
1835 pols[0] = pol;
1836 npols ++;
1837 #ifdef CONFIG_XFRM_SUB_POLICY
1838 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1839 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1840 &fl, family,
1841 XFRM_POLICY_IN);
1842 if (pols[1]) {
1843 if (IS_ERR(pols[1]))
1844 return 0;
1845 pols[1]->curlft.use_time = get_seconds();
1846 npols ++;
1847 }
1848 }
1849 #endif
1850
1851 if (pol->action == XFRM_POLICY_ALLOW) {
1852 struct sec_path *sp;
1853 static struct sec_path dummy;
1854 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1855 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1856 struct xfrm_tmpl **tpp = tp;
1857 int ti = 0;
1858 int i, k;
1859
1860 if ((sp = skb->sp) == NULL)
1861 sp = &dummy;
1862
1863 for (pi = 0; pi < npols; pi++) {
1864 if (pols[pi] != pol &&
1865 pols[pi]->action != XFRM_POLICY_ALLOW)
1866 goto reject;
1867 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1868 goto reject_error;
1869 for (i = 0; i < pols[pi]->xfrm_nr; i++)
1870 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1871 }
1872 xfrm_nr = ti;
1873 if (npols > 1) {
1874 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1875 tpp = stp;
1876 }
1877
1878 /* For each tunnel xfrm, find the first matching tmpl.
1879 * For each tmpl before that, find corresponding xfrm.
1880 * Order is _important_. Later we will implement
1881 * some barriers, but at the moment barriers
1882 * are implied between each two transformations.
1883 */
1884 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1885 k = xfrm_policy_ok(tpp[i], sp, k, family);
1886 if (k < 0) {
1887 if (k < -1)
1888 /* "-2 - errored_index" returned */
1889 xerr_idx = -(2+k);
1890 goto reject;
1891 }
1892 }
1893
1894 if (secpath_has_nontransport(sp, k, &xerr_idx))
1895 goto reject;
1896
1897 xfrm_pols_put(pols, npols);
1898 return 1;
1899 }
1900
1901 reject:
1902 xfrm_secpath_reject(xerr_idx, skb, &fl);
1903 reject_error:
1904 xfrm_pols_put(pols, npols);
1905 return 0;
1906 }
1907 EXPORT_SYMBOL(__xfrm_policy_check);
1908
1909 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1910 {
1911 struct flowi fl;
1912
1913 if (xfrm_decode_session(skb, &fl, family) < 0)
1914 return 0;
1915
1916 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1917 }
1918 EXPORT_SYMBOL(__xfrm_route_forward);
1919
1920 /* Optimize later using cookies and generation ids. */
1921
1922 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1923 {
1924 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1925 * to "-1" to force all XFRM destinations to get validated by
1926 * dst_ops->check on every use. We do this because when a
1927 * normal route referenced by an XFRM dst is obsoleted we do
1928 * not go looking around for all parent referencing XFRM dsts
1929 * so that we can invalidate them. It is just too much work.
1930 * Instead we make the checks here on every use. For example:
1931 *
1932 * XFRM dst A --> IPv4 dst X
1933 *
1934 * X is the "xdst->route" of A (X is also the "dst->path" of A
1935 * in this example). If X is marked obsolete, "A" will not
1936 * notice. That's what we are validating here via the
1937 * stale_bundle() check.
1938 *
1939 * When a policy's bundle is pruned, we dst_free() the XFRM
1940 * dst which causes it's ->obsolete field to be set to a
1941 * positive non-zero integer. If an XFRM dst has been pruned
1942 * like this, we want to force a new route lookup.
1943 */
1944 if (dst->obsolete < 0 && !stale_bundle(dst))
1945 return dst;
1946
1947 return NULL;
1948 }
1949
1950 static int stale_bundle(struct dst_entry *dst)
1951 {
1952 return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1953 }
1954
1955 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1956 {
1957 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1958 dst->dev = &loopback_dev;
1959 dev_hold(&loopback_dev);
1960 dev_put(dev);
1961 }
1962 }
1963 EXPORT_SYMBOL(xfrm_dst_ifdown);
1964
1965 static void xfrm_link_failure(struct sk_buff *skb)
1966 {
1967 /* Impossible. Such dst must be popped before reaches point of failure. */
1968 return;
1969 }
1970
1971 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1972 {
1973 if (dst) {
1974 if (dst->obsolete) {
1975 dst_release(dst);
1976 dst = NULL;
1977 }
1978 }
1979 return dst;
1980 }
1981
1982 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1983 {
1984 struct dst_entry *dst, **dstp;
1985
1986 write_lock(&pol->lock);
1987 dstp = &pol->bundles;
1988 while ((dst=*dstp) != NULL) {
1989 if (func(dst)) {
1990 *dstp = dst->next;
1991 dst->next = *gc_list_p;
1992 *gc_list_p = dst;
1993 } else {
1994 dstp = &dst->next;
1995 }
1996 }
1997 write_unlock(&pol->lock);
1998 }
1999
2000 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2001 {
2002 struct dst_entry *gc_list = NULL;
2003 int dir;
2004
2005 read_lock_bh(&xfrm_policy_lock);
2006 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2007 struct xfrm_policy *pol;
2008 struct hlist_node *entry;
2009 struct hlist_head *table;
2010 int i;
2011
2012 hlist_for_each_entry(pol, entry,
2013 &xfrm_policy_inexact[dir], bydst)
2014 prune_one_bundle(pol, func, &gc_list);
2015
2016 table = xfrm_policy_bydst[dir].table;
2017 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2018 hlist_for_each_entry(pol, entry, table + i, bydst)
2019 prune_one_bundle(pol, func, &gc_list);
2020 }
2021 }
2022 read_unlock_bh(&xfrm_policy_lock);
2023
2024 while (gc_list) {
2025 struct dst_entry *dst = gc_list;
2026 gc_list = dst->next;
2027 dst_free(dst);
2028 }
2029 }
2030
2031 static int unused_bundle(struct dst_entry *dst)
2032 {
2033 return !atomic_read(&dst->__refcnt);
2034 }
2035
2036 static void __xfrm_garbage_collect(void)
2037 {
2038 xfrm_prune_bundles(unused_bundle);
2039 }
2040
2041 static int xfrm_flush_bundles(void)
2042 {
2043 xfrm_prune_bundles(stale_bundle);
2044 return 0;
2045 }
2046
2047 void xfrm_init_pmtu(struct dst_entry *dst)
2048 {
2049 do {
2050 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2051 u32 pmtu, route_mtu_cached;
2052
2053 pmtu = dst_mtu(dst->child);
2054 xdst->child_mtu_cached = pmtu;
2055
2056 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2057
2058 route_mtu_cached = dst_mtu(xdst->route);
2059 xdst->route_mtu_cached = route_mtu_cached;
2060
2061 if (pmtu > route_mtu_cached)
2062 pmtu = route_mtu_cached;
2063
2064 dst->metrics[RTAX_MTU-1] = pmtu;
2065 } while ((dst = dst->next));
2066 }
2067
2068 EXPORT_SYMBOL(xfrm_init_pmtu);
2069
2070 /* Check that the bundle accepts the flow and its components are
2071 * still valid.
2072 */
2073
2074 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2075 struct flowi *fl, int family, int strict)
2076 {
2077 struct dst_entry *dst = &first->u.dst;
2078 struct xfrm_dst *last;
2079 u32 mtu;
2080
2081 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2082 (dst->dev && !netif_running(dst->dev)))
2083 return 0;
2084 #ifdef CONFIG_XFRM_SUB_POLICY
2085 if (fl) {
2086 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2087 return 0;
2088 if (first->partner &&
2089 !xfrm_selector_match(first->partner, fl, family))
2090 return 0;
2091 }
2092 #endif
2093
2094 last = NULL;
2095
2096 do {
2097 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2098
2099 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2100 return 0;
2101 if (fl && pol &&
2102 !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2103 return 0;
2104 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2105 return 0;
2106 if (xdst->genid != dst->xfrm->genid)
2107 return 0;
2108
2109 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
2110 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2111 return 0;
2112
2113 mtu = dst_mtu(dst->child);
2114 if (xdst->child_mtu_cached != mtu) {
2115 last = xdst;
2116 xdst->child_mtu_cached = mtu;
2117 }
2118
2119 if (!dst_check(xdst->route, xdst->route_cookie))
2120 return 0;
2121 mtu = dst_mtu(xdst->route);
2122 if (xdst->route_mtu_cached != mtu) {
2123 last = xdst;
2124 xdst->route_mtu_cached = mtu;
2125 }
2126
2127 dst = dst->child;
2128 } while (dst->xfrm);
2129
2130 if (likely(!last))
2131 return 1;
2132
2133 mtu = last->child_mtu_cached;
2134 for (;;) {
2135 dst = &last->u.dst;
2136
2137 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2138 if (mtu > last->route_mtu_cached)
2139 mtu = last->route_mtu_cached;
2140 dst->metrics[RTAX_MTU-1] = mtu;
2141
2142 if (last == first)
2143 break;
2144
2145 last = (struct xfrm_dst *)last->u.dst.next;
2146 last->child_mtu_cached = mtu;
2147 }
2148
2149 return 1;
2150 }
2151
2152 EXPORT_SYMBOL(xfrm_bundle_ok);
2153
2154 #ifdef CONFIG_AUDITSYSCALL
2155 /* Audit addition and deletion of SAs and ipsec policy */
2156
2157 void xfrm_audit_log(uid_t auid, u32 sid, int type, int result,
2158 struct xfrm_policy *xp, struct xfrm_state *x)
2159 {
2160
2161 char *secctx;
2162 u32 secctx_len;
2163 struct xfrm_sec_ctx *sctx = NULL;
2164 struct audit_buffer *audit_buf;
2165 int family;
2166 extern int audit_enabled;
2167
2168 if (audit_enabled == 0)
2169 return;
2170
2171 BUG_ON((type == AUDIT_MAC_IPSEC_ADDSA ||
2172 type == AUDIT_MAC_IPSEC_DELSA) && !x);
2173 BUG_ON((type == AUDIT_MAC_IPSEC_ADDSPD ||
2174 type == AUDIT_MAC_IPSEC_DELSPD) && !xp);
2175
2176 audit_buf = audit_log_start(current->audit_context, GFP_ATOMIC, type);
2177 if (audit_buf == NULL)
2178 return;
2179
2180 switch(type) {
2181 case AUDIT_MAC_IPSEC_ADDSA:
2182 audit_log_format(audit_buf, "SAD add: auid=%u", auid);
2183 break;
2184 case AUDIT_MAC_IPSEC_DELSA:
2185 audit_log_format(audit_buf, "SAD delete: auid=%u", auid);
2186 break;
2187 case AUDIT_MAC_IPSEC_ADDSPD:
2188 audit_log_format(audit_buf, "SPD add: auid=%u", auid);
2189 break;
2190 case AUDIT_MAC_IPSEC_DELSPD:
2191 audit_log_format(audit_buf, "SPD delete: auid=%u", auid);
2192 break;
2193 default:
2194 return;
2195 }
2196
2197 if (sid != 0 &&
2198 security_secid_to_secctx(sid, &secctx, &secctx_len) == 0) {
2199 audit_log_format(audit_buf, " subj=%s", secctx);
2200 security_release_secctx(secctx, secctx_len);
2201 } else
2202 audit_log_task_context(audit_buf);
2203
2204 if (xp) {
2205 family = xp->selector.family;
2206 if (xp->security)
2207 sctx = xp->security;
2208 } else {
2209 family = x->props.family;
2210 if (x->security)
2211 sctx = x->security;
2212 }
2213
2214 if (sctx)
2215 audit_log_format(audit_buf,
2216 " sec_alg=%u sec_doi=%u sec_obj=%s",
2217 sctx->ctx_alg, sctx->ctx_doi, sctx->ctx_str);
2218
2219 switch(family) {
2220 case AF_INET:
2221 {
2222 struct in_addr saddr, daddr;
2223 if (xp) {
2224 saddr.s_addr = xp->selector.saddr.a4;
2225 daddr.s_addr = xp->selector.daddr.a4;
2226 } else {
2227 saddr.s_addr = x->props.saddr.a4;
2228 daddr.s_addr = x->id.daddr.a4;
2229 }
2230 audit_log_format(audit_buf,
2231 " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2232 NIPQUAD(saddr), NIPQUAD(daddr));
2233 }
2234 break;
2235 case AF_INET6:
2236 {
2237 struct in6_addr saddr6, daddr6;
2238 if (xp) {
2239 memcpy(&saddr6, xp->selector.saddr.a6,
2240 sizeof(struct in6_addr));
2241 memcpy(&daddr6, xp->selector.daddr.a6,
2242 sizeof(struct in6_addr));
2243 } else {
2244 memcpy(&saddr6, x->props.saddr.a6,
2245 sizeof(struct in6_addr));
2246 memcpy(&daddr6, x->id.daddr.a6,
2247 sizeof(struct in6_addr));
2248 }
2249 audit_log_format(audit_buf,
2250 " src=" NIP6_FMT " dst=" NIP6_FMT,
2251 NIP6(saddr6), NIP6(daddr6));
2252 }
2253 break;
2254 }
2255
2256 if (x)
2257 audit_log_format(audit_buf, " spi=%lu(0x%lx) protocol=%s",
2258 (unsigned long)ntohl(x->id.spi),
2259 (unsigned long)ntohl(x->id.spi),
2260 x->id.proto == IPPROTO_AH ? "AH" :
2261 (x->id.proto == IPPROTO_ESP ?
2262 "ESP" : "IPCOMP"));
2263
2264 audit_log_format(audit_buf, " res=%u", result);
2265 audit_log_end(audit_buf);
2266 }
2267
2268 EXPORT_SYMBOL(xfrm_audit_log);
2269 #endif /* CONFIG_AUDITSYSCALL */
2270
2271 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2272 {
2273 int err = 0;
2274 if (unlikely(afinfo == NULL))
2275 return -EINVAL;
2276 if (unlikely(afinfo->family >= NPROTO))
2277 return -EAFNOSUPPORT;
2278 write_lock_bh(&xfrm_policy_afinfo_lock);
2279 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2280 err = -ENOBUFS;
2281 else {
2282 struct dst_ops *dst_ops = afinfo->dst_ops;
2283 if (likely(dst_ops->kmem_cachep == NULL))
2284 dst_ops->kmem_cachep = xfrm_dst_cache;
2285 if (likely(dst_ops->check == NULL))
2286 dst_ops->check = xfrm_dst_check;
2287 if (likely(dst_ops->negative_advice == NULL))
2288 dst_ops->negative_advice = xfrm_negative_advice;
2289 if (likely(dst_ops->link_failure == NULL))
2290 dst_ops->link_failure = xfrm_link_failure;
2291 if (likely(afinfo->garbage_collect == NULL))
2292 afinfo->garbage_collect = __xfrm_garbage_collect;
2293 xfrm_policy_afinfo[afinfo->family] = afinfo;
2294 }
2295 write_unlock_bh(&xfrm_policy_afinfo_lock);
2296 return err;
2297 }
2298 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2299
2300 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2301 {
2302 int err = 0;
2303 if (unlikely(afinfo == NULL))
2304 return -EINVAL;
2305 if (unlikely(afinfo->family >= NPROTO))
2306 return -EAFNOSUPPORT;
2307 write_lock_bh(&xfrm_policy_afinfo_lock);
2308 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2309 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2310 err = -EINVAL;
2311 else {
2312 struct dst_ops *dst_ops = afinfo->dst_ops;
2313 xfrm_policy_afinfo[afinfo->family] = NULL;
2314 dst_ops->kmem_cachep = NULL;
2315 dst_ops->check = NULL;
2316 dst_ops->negative_advice = NULL;
2317 dst_ops->link_failure = NULL;
2318 afinfo->garbage_collect = NULL;
2319 }
2320 }
2321 write_unlock_bh(&xfrm_policy_afinfo_lock);
2322 return err;
2323 }
2324 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2325
2326 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2327 {
2328 struct xfrm_policy_afinfo *afinfo;
2329 if (unlikely(family >= NPROTO))
2330 return NULL;
2331 read_lock(&xfrm_policy_afinfo_lock);
2332 afinfo = xfrm_policy_afinfo[family];
2333 if (unlikely(!afinfo))
2334 read_unlock(&xfrm_policy_afinfo_lock);
2335 return afinfo;
2336 }
2337
2338 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2339 {
2340 read_unlock(&xfrm_policy_afinfo_lock);
2341 }
2342
2343 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2344 {
2345 struct xfrm_policy_afinfo *afinfo;
2346 if (unlikely(family >= NPROTO))
2347 return NULL;
2348 write_lock_bh(&xfrm_policy_afinfo_lock);
2349 afinfo = xfrm_policy_afinfo[family];
2350 if (unlikely(!afinfo))
2351 write_unlock_bh(&xfrm_policy_afinfo_lock);
2352 return afinfo;
2353 }
2354
2355 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2356 {
2357 write_unlock_bh(&xfrm_policy_afinfo_lock);
2358 }
2359
2360 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2361 {
2362 switch (event) {
2363 case NETDEV_DOWN:
2364 xfrm_flush_bundles();
2365 }
2366 return NOTIFY_DONE;
2367 }
2368
2369 static struct notifier_block xfrm_dev_notifier = {
2370 xfrm_dev_event,
2371 NULL,
2372 0
2373 };
2374
2375 static void __init xfrm_policy_init(void)
2376 {
2377 unsigned int hmask, sz;
2378 int dir;
2379
2380 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2381 sizeof(struct xfrm_dst),
2382 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2383 NULL);
2384
2385 hmask = 8 - 1;
2386 sz = (hmask+1) * sizeof(struct hlist_head);
2387
2388 xfrm_policy_byidx = xfrm_hash_alloc(sz);
2389 xfrm_idx_hmask = hmask;
2390 if (!xfrm_policy_byidx)
2391 panic("XFRM: failed to allocate byidx hash\n");
2392
2393 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2394 struct xfrm_policy_hash *htab;
2395
2396 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2397
2398 htab = &xfrm_policy_bydst[dir];
2399 htab->table = xfrm_hash_alloc(sz);
2400 htab->hmask = hmask;
2401 if (!htab->table)
2402 panic("XFRM: failed to allocate bydst hash\n");
2403 }
2404
2405 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2406 register_netdevice_notifier(&xfrm_dev_notifier);
2407 }
2408
2409 void __init xfrm_init(void)
2410 {
2411 xfrm_state_init();
2412 xfrm_policy_init();
2413 xfrm_input_init();
2414 }
2415
2416 #ifdef CONFIG_XFRM_MIGRATE
2417 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2418 struct xfrm_selector *sel_tgt)
2419 {
2420 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2421 if (sel_tgt->family == sel_cmp->family &&
2422 xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2423 sel_cmp->family) == 0 &&
2424 xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2425 sel_cmp->family) == 0 &&
2426 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2427 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2428 return 1;
2429 }
2430 } else {
2431 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2432 return 1;
2433 }
2434 }
2435 return 0;
2436 }
2437
2438 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2439 u8 dir, u8 type)
2440 {
2441 struct xfrm_policy *pol, *ret = NULL;
2442 struct hlist_node *entry;
2443 struct hlist_head *chain;
2444 u32 priority = ~0U;
2445
2446 read_lock_bh(&xfrm_policy_lock);
2447 chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2448 hlist_for_each_entry(pol, entry, chain, bydst) {
2449 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2450 pol->type == type) {
2451 ret = pol;
2452 priority = ret->priority;
2453 break;
2454 }
2455 }
2456 chain = &xfrm_policy_inexact[dir];
2457 hlist_for_each_entry(pol, entry, chain, bydst) {
2458 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2459 pol->type == type &&
2460 pol->priority < priority) {
2461 ret = pol;
2462 break;
2463 }
2464 }
2465
2466 if (ret)
2467 xfrm_pol_hold(ret);
2468
2469 read_unlock_bh(&xfrm_policy_lock);
2470
2471 return ret;
2472 }
2473
2474 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2475 {
2476 int match = 0;
2477
2478 if (t->mode == m->mode && t->id.proto == m->proto &&
2479 (m->reqid == 0 || t->reqid == m->reqid)) {
2480 switch (t->mode) {
2481 case XFRM_MODE_TUNNEL:
2482 case XFRM_MODE_BEET:
2483 if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2484 m->old_family) == 0 &&
2485 xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2486 m->old_family) == 0) {
2487 match = 1;
2488 }
2489 break;
2490 case XFRM_MODE_TRANSPORT:
2491 /* in case of transport mode, template does not store
2492 any IP addresses, hence we just compare mode and
2493 protocol */
2494 match = 1;
2495 break;
2496 default:
2497 break;
2498 }
2499 }
2500 return match;
2501 }
2502
2503 /* update endpoint address(es) of template(s) */
2504 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2505 struct xfrm_migrate *m, int num_migrate)
2506 {
2507 struct xfrm_migrate *mp;
2508 struct dst_entry *dst;
2509 int i, j, n = 0;
2510
2511 write_lock_bh(&pol->lock);
2512 if (unlikely(pol->dead)) {
2513 /* target policy has been deleted */
2514 write_unlock_bh(&pol->lock);
2515 return -ENOENT;
2516 }
2517
2518 for (i = 0; i < pol->xfrm_nr; i++) {
2519 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2520 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2521 continue;
2522 n++;
2523 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL)
2524 continue;
2525 /* update endpoints */
2526 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2527 sizeof(pol->xfrm_vec[i].id.daddr));
2528 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2529 sizeof(pol->xfrm_vec[i].saddr));
2530 pol->xfrm_vec[i].encap_family = mp->new_family;
2531 /* flush bundles */
2532 while ((dst = pol->bundles) != NULL) {
2533 pol->bundles = dst->next;
2534 dst_free(dst);
2535 }
2536 }
2537 }
2538
2539 write_unlock_bh(&pol->lock);
2540
2541 if (!n)
2542 return -ENODATA;
2543
2544 return 0;
2545 }
2546
2547 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2548 {
2549 int i, j;
2550
2551 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2552 return -EINVAL;
2553
2554 for (i = 0; i < num_migrate; i++) {
2555 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2556 m[i].old_family) == 0) &&
2557 (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2558 m[i].old_family) == 0))
2559 return -EINVAL;
2560 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2561 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2562 return -EINVAL;
2563
2564 /* check if there is any duplicated entry */
2565 for (j = i + 1; j < num_migrate; j++) {
2566 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2567 sizeof(m[i].old_daddr)) &&
2568 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2569 sizeof(m[i].old_saddr)) &&
2570 m[i].proto == m[j].proto &&
2571 m[i].mode == m[j].mode &&
2572 m[i].reqid == m[j].reqid &&
2573 m[i].old_family == m[j].old_family)
2574 return -EINVAL;
2575 }
2576 }
2577
2578 return 0;
2579 }
2580
2581 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2582 struct xfrm_migrate *m, int num_migrate)
2583 {
2584 int i, err, nx_cur = 0, nx_new = 0;
2585 struct xfrm_policy *pol = NULL;
2586 struct xfrm_state *x, *xc;
2587 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2588 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2589 struct xfrm_migrate *mp;
2590
2591 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2592 goto out;
2593
2594 /* Stage 1 - find policy */
2595 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2596 err = -ENOENT;
2597 goto out;
2598 }
2599
2600 /* Stage 2 - find and update state(s) */
2601 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2602 if ((x = xfrm_migrate_state_find(mp))) {
2603 x_cur[nx_cur] = x;
2604 nx_cur++;
2605 if ((xc = xfrm_state_migrate(x, mp))) {
2606 x_new[nx_new] = xc;
2607 nx_new++;
2608 } else {
2609 err = -ENODATA;
2610 goto restore_state;
2611 }
2612 }
2613 }
2614
2615 /* Stage 3 - update policy */
2616 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2617 goto restore_state;
2618
2619 /* Stage 4 - delete old state(s) */
2620 if (nx_cur) {
2621 xfrm_states_put(x_cur, nx_cur);
2622 xfrm_states_delete(x_cur, nx_cur);
2623 }
2624
2625 /* Stage 5 - announce */
2626 km_migrate(sel, dir, type, m, num_migrate);
2627
2628 xfrm_pol_put(pol);
2629
2630 return 0;
2631 out:
2632 return err;
2633
2634 restore_state:
2635 if (pol)
2636 xfrm_pol_put(pol);
2637 if (nx_cur)
2638 xfrm_states_put(x_cur, nx_cur);
2639 if (nx_new)
2640 xfrm_states_delete(x_new, nx_new);
2641
2642 return err;
2643 }
2644 EXPORT_SYMBOL(xfrm_migrate);
2645 #endif
Linux kernel - Deliverables
This page took 0.08671 seconds and 6 git commands to generate.