projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[SCSI] Merge up to linux-2.6 head
[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
30 #include "xfrm_hash.h"
31
32 int sysctl_xfrm_larval_drop;
33
34 DEFINE_MUTEX(xfrm_cfg_mutex);
35 EXPORT_SYMBOL(xfrm_cfg_mutex);
36
37 static DEFINE_RWLOCK(xfrm_policy_lock);
38
39 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
40 EXPORT_SYMBOL(xfrm_policy_count);
41
42 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
43 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
44
45 static struct kmem_cache *xfrm_dst_cache __read_mostly;
46
47 static struct work_struct xfrm_policy_gc_work;
48 static HLIST_HEAD(xfrm_policy_gc_list);
49 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
50
51 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
52 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
53 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
54 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
55
56 static inline int
57 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
58 {
59 return addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
60 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
61 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
62 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
63 (fl->proto == sel->proto || !sel->proto) &&
64 (fl->oif == sel->ifindex || !sel->ifindex);
65 }
66
67 static inline int
68 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
69 {
70 return addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
71 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
72 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
73 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
74 (fl->proto == sel->proto || !sel->proto) &&
75 (fl->oif == sel->ifindex || !sel->ifindex);
76 }
77
78 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
79 unsigned short family)
80 {
81 switch (family) {
82 case AF_INET:
83 return __xfrm4_selector_match(sel, fl);
84 case AF_INET6:
85 return __xfrm6_selector_match(sel, fl);
86 }
87 return 0;
88 }
89
90 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
91 {
92 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
93 struct xfrm_type **typemap;
94 int err = 0;
95
96 if (unlikely(afinfo == NULL))
97 return -EAFNOSUPPORT;
98 typemap = afinfo->type_map;
99
100 if (likely(typemap[type->proto] == NULL))
101 typemap[type->proto] = type;
102 else
103 err = -EEXIST;
104 xfrm_policy_unlock_afinfo(afinfo);
105 return err;
106 }
107 EXPORT_SYMBOL(xfrm_register_type);
108
109 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
110 {
111 struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
112 struct xfrm_type **typemap;
113 int err = 0;
114
115 if (unlikely(afinfo == NULL))
116 return -EAFNOSUPPORT;
117 typemap = afinfo->type_map;
118
119 if (unlikely(typemap[type->proto] != type))
120 err = -ENOENT;
121 else
122 typemap[type->proto] = NULL;
123 xfrm_policy_unlock_afinfo(afinfo);
124 return err;
125 }
126 EXPORT_SYMBOL(xfrm_unregister_type);
127
128 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
129 {
130 struct xfrm_policy_afinfo *afinfo;
131 struct xfrm_type **typemap;
132 struct xfrm_type *type;
133 int modload_attempted = 0;
134
135 retry:
136 afinfo = xfrm_policy_get_afinfo(family);
137 if (unlikely(afinfo == NULL))
138 return NULL;
139 typemap = afinfo->type_map;
140
141 type = typemap[proto];
142 if (unlikely(type && !try_module_get(type->owner)))
143 type = NULL;
144 if (!type && !modload_attempted) {
145 xfrm_policy_put_afinfo(afinfo);
146 request_module("xfrm-type-%d-%d",
147 (int) family, (int) proto);
148 modload_attempted = 1;
149 goto retry;
150 }
151
152 xfrm_policy_put_afinfo(afinfo);
153 return type;
154 }
155
156 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
157 unsigned short family)
158 {
159 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
160 int err = 0;
161
162 if (unlikely(afinfo == NULL))
163 return -EAFNOSUPPORT;
164
165 if (likely(afinfo->dst_lookup != NULL))
166 err = afinfo->dst_lookup(dst, fl);
167 else
168 err = -EINVAL;
169 xfrm_policy_put_afinfo(afinfo);
170 return err;
171 }
172 EXPORT_SYMBOL(xfrm_dst_lookup);
173
174 void xfrm_put_type(struct xfrm_type *type)
175 {
176 module_put(type->owner);
177 }
178
179 int xfrm_register_mode(struct xfrm_mode *mode, int family)
180 {
181 struct xfrm_policy_afinfo *afinfo;
182 struct xfrm_mode **modemap;
183 int err;
184
185 if (unlikely(mode->encap >= XFRM_MODE_MAX))
186 return -EINVAL;
187
188 afinfo = xfrm_policy_lock_afinfo(family);
189 if (unlikely(afinfo == NULL))
190 return -EAFNOSUPPORT;
191
192 err = -EEXIST;
193 modemap = afinfo->mode_map;
194 if (likely(modemap[mode->encap] == NULL)) {
195 modemap[mode->encap] = mode;
196 err = 0;
197 }
198
199 xfrm_policy_unlock_afinfo(afinfo);
200 return err;
201 }
202 EXPORT_SYMBOL(xfrm_register_mode);
203
204 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
205 {
206 struct xfrm_policy_afinfo *afinfo;
207 struct xfrm_mode **modemap;
208 int err;
209
210 if (unlikely(mode->encap >= XFRM_MODE_MAX))
211 return -EINVAL;
212
213 afinfo = xfrm_policy_lock_afinfo(family);
214 if (unlikely(afinfo == NULL))
215 return -EAFNOSUPPORT;
216
217 err = -ENOENT;
218 modemap = afinfo->mode_map;
219 if (likely(modemap[mode->encap] == mode)) {
220 modemap[mode->encap] = NULL;
221 err = 0;
222 }
223
224 xfrm_policy_unlock_afinfo(afinfo);
225 return err;
226 }
227 EXPORT_SYMBOL(xfrm_unregister_mode);
228
229 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
230 {
231 struct xfrm_policy_afinfo *afinfo;
232 struct xfrm_mode *mode;
233 int modload_attempted = 0;
234
235 if (unlikely(encap >= XFRM_MODE_MAX))
236 return NULL;
237
238 retry:
239 afinfo = xfrm_policy_get_afinfo(family);
240 if (unlikely(afinfo == NULL))
241 return NULL;
242
243 mode = afinfo->mode_map[encap];
244 if (unlikely(mode && !try_module_get(mode->owner)))
245 mode = NULL;
246 if (!mode && !modload_attempted) {
247 xfrm_policy_put_afinfo(afinfo);
248 request_module("xfrm-mode-%d-%d", family, encap);
249 modload_attempted = 1;
250 goto retry;
251 }
252
253 xfrm_policy_put_afinfo(afinfo);
254 return mode;
255 }
256
257 void xfrm_put_mode(struct xfrm_mode *mode)
258 {
259 module_put(mode->owner);
260 }
261
262 static inline unsigned long make_jiffies(long secs)
263 {
264 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
265 return MAX_SCHEDULE_TIMEOUT-1;
266 else
267 return secs*HZ;
268 }
269
270 static void xfrm_policy_timer(unsigned long data)
271 {
272 struct xfrm_policy *xp = (struct xfrm_policy*)data;
273 unsigned long now = get_seconds();
274 long next = LONG_MAX;
275 int warn = 0;
276 int dir;
277
278 read_lock(&xp->lock);
279
280 if (xp->dead)
281 goto out;
282
283 dir = xfrm_policy_id2dir(xp->index);
284
285 if (xp->lft.hard_add_expires_seconds) {
286 long tmo = xp->lft.hard_add_expires_seconds +
287 xp->curlft.add_time - now;
288 if (tmo <= 0)
289 goto expired;
290 if (tmo < next)
291 next = tmo;
292 }
293 if (xp->lft.hard_use_expires_seconds) {
294 long tmo = xp->lft.hard_use_expires_seconds +
295 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
296 if (tmo <= 0)
297 goto expired;
298 if (tmo < next)
299 next = tmo;
300 }
301 if (xp->lft.soft_add_expires_seconds) {
302 long tmo = xp->lft.soft_add_expires_seconds +
303 xp->curlft.add_time - now;
304 if (tmo <= 0) {
305 warn = 1;
306 tmo = XFRM_KM_TIMEOUT;
307 }
308 if (tmo < next)
309 next = tmo;
310 }
311 if (xp->lft.soft_use_expires_seconds) {
312 long tmo = xp->lft.soft_use_expires_seconds +
313 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
314 if (tmo <= 0) {
315 warn = 1;
316 tmo = XFRM_KM_TIMEOUT;
317 }
318 if (tmo < next)
319 next = tmo;
320 }
321
322 if (warn)
323 km_policy_expired(xp, dir, 0, 0);
324 if (next != LONG_MAX &&
325 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
326 xfrm_pol_hold(xp);
327
328 out:
329 read_unlock(&xp->lock);
330 xfrm_pol_put(xp);
331 return;
332
333 expired:
334 read_unlock(&xp->lock);
335 if (!xfrm_policy_delete(xp, dir))
336 km_policy_expired(xp, dir, 1, 0);
337 xfrm_pol_put(xp);
338 }
339
340
341 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
342 * SPD calls.
343 */
344
345 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
346 {
347 struct xfrm_policy *policy;
348
349 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
350
351 if (policy) {
352 INIT_HLIST_NODE(&policy->bydst);
353 INIT_HLIST_NODE(&policy->byidx);
354 rwlock_init(&policy->lock);
355 atomic_set(&policy->refcnt, 1);
356 init_timer(&policy->timer);
357 policy->timer.data = (unsigned long)policy;
358 policy->timer.function = xfrm_policy_timer;
359 }
360 return policy;
361 }
362 EXPORT_SYMBOL(xfrm_policy_alloc);
363
364 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
365
366 void __xfrm_policy_destroy(struct xfrm_policy *policy)
367 {
368 BUG_ON(!policy->dead);
369
370 BUG_ON(policy->bundles);
371
372 if (del_timer(&policy->timer))
373 BUG();
374
375 security_xfrm_policy_free(policy);
376 kfree(policy);
377 }
378 EXPORT_SYMBOL(__xfrm_policy_destroy);
379
380 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
381 {
382 struct dst_entry *dst;
383
384 while ((dst = policy->bundles) != NULL) {
385 policy->bundles = dst->next;
386 dst_free(dst);
387 }
388
389 if (del_timer(&policy->timer))
390 atomic_dec(&policy->refcnt);
391
392 if (atomic_read(&policy->refcnt) > 1)
393 flow_cache_flush();
394
395 xfrm_pol_put(policy);
396 }
397
398 static void xfrm_policy_gc_task(struct work_struct *work)
399 {
400 struct xfrm_policy *policy;
401 struct hlist_node *entry, *tmp;
402 struct hlist_head gc_list;
403
404 spin_lock_bh(&xfrm_policy_gc_lock);
405 gc_list.first = xfrm_policy_gc_list.first;
406 INIT_HLIST_HEAD(&xfrm_policy_gc_list);
407 spin_unlock_bh(&xfrm_policy_gc_lock);
408
409 hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
410 xfrm_policy_gc_kill(policy);
411 }
412
413 /* Rule must be locked. Release descentant resources, announce
414 * entry dead. The rule must be unlinked from lists to the moment.
415 */
416
417 static void xfrm_policy_kill(struct xfrm_policy *policy)
418 {
419 int dead;
420
421 write_lock_bh(&policy->lock);
422 dead = policy->dead;
423 policy->dead = 1;
424 write_unlock_bh(&policy->lock);
425
426 if (unlikely(dead)) {
427 WARN_ON(1);
428 return;
429 }
430
431 spin_lock(&xfrm_policy_gc_lock);
432 hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
433 spin_unlock(&xfrm_policy_gc_lock);
434
435 schedule_work(&xfrm_policy_gc_work);
436 }
437
438 struct xfrm_policy_hash {
439 struct hlist_head *table;
440 unsigned int hmask;
441 };
442
443 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
444 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
445 static struct hlist_head *xfrm_policy_byidx __read_mostly;
446 static unsigned int xfrm_idx_hmask __read_mostly;
447 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
448
449 static inline unsigned int idx_hash(u32 index)
450 {
451 return __idx_hash(index, xfrm_idx_hmask);
452 }
453
454 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
455 {
456 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
457 unsigned int hash = __sel_hash(sel, family, hmask);
458
459 return (hash == hmask + 1 ?
460 &xfrm_policy_inexact[dir] :
461 xfrm_policy_bydst[dir].table + hash);
462 }
463
464 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
465 {
466 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
467 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
468
469 return xfrm_policy_bydst[dir].table + hash;
470 }
471
472 static void xfrm_dst_hash_transfer(struct hlist_head *list,
473 struct hlist_head *ndsttable,
474 unsigned int nhashmask)
475 {
476 struct hlist_node *entry, *tmp;
477 struct xfrm_policy *pol;
478
479 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
480 unsigned int h;
481
482 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
483 pol->family, nhashmask);
484 hlist_add_head(&pol->bydst, ndsttable+h);
485 }
486 }
487
488 static void xfrm_idx_hash_transfer(struct hlist_head *list,
489 struct hlist_head *nidxtable,
490 unsigned int nhashmask)
491 {
492 struct hlist_node *entry, *tmp;
493 struct xfrm_policy *pol;
494
495 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
496 unsigned int h;
497
498 h = __idx_hash(pol->index, nhashmask);
499 hlist_add_head(&pol->byidx, nidxtable+h);
500 }
501 }
502
503 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
504 {
505 return ((old_hmask + 1) << 1) - 1;
506 }
507
508 static void xfrm_bydst_resize(int dir)
509 {
510 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
511 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
512 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
513 struct hlist_head *odst = xfrm_policy_bydst[dir].table;
514 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
515 int i;
516
517 if (!ndst)
518 return;
519
520 write_lock_bh(&xfrm_policy_lock);
521
522 for (i = hmask; i >= 0; i--)
523 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
524
525 xfrm_policy_bydst[dir].table = ndst;
526 xfrm_policy_bydst[dir].hmask = nhashmask;
527
528 write_unlock_bh(&xfrm_policy_lock);
529
530 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
531 }
532
533 static void xfrm_byidx_resize(int total)
534 {
535 unsigned int hmask = xfrm_idx_hmask;
536 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
537 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
538 struct hlist_head *oidx = xfrm_policy_byidx;
539 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
540 int i;
541
542 if (!nidx)
543 return;
544
545 write_lock_bh(&xfrm_policy_lock);
546
547 for (i = hmask; i >= 0; i--)
548 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
549
550 xfrm_policy_byidx = nidx;
551 xfrm_idx_hmask = nhashmask;
552
553 write_unlock_bh(&xfrm_policy_lock);
554
555 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
556 }
557
558 static inline int xfrm_bydst_should_resize(int dir, int *total)
559 {
560 unsigned int cnt = xfrm_policy_count[dir];
561 unsigned int hmask = xfrm_policy_bydst[dir].hmask;
562
563 if (total)
564 *total += cnt;
565
566 if ((hmask + 1) < xfrm_policy_hashmax &&
567 cnt > hmask)
568 return 1;
569
570 return 0;
571 }
572
573 static inline int xfrm_byidx_should_resize(int total)
574 {
575 unsigned int hmask = xfrm_idx_hmask;
576
577 if ((hmask + 1) < xfrm_policy_hashmax &&
578 total > hmask)
579 return 1;
580
581 return 0;
582 }
583
584 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
585 {
586 read_lock_bh(&xfrm_policy_lock);
587 si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
588 si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
589 si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
590 si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
591 si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
592 si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
593 si->spdhcnt = xfrm_idx_hmask;
594 si->spdhmcnt = xfrm_policy_hashmax;
595 read_unlock_bh(&xfrm_policy_lock);
596 }
597 EXPORT_SYMBOL(xfrm_spd_getinfo);
598
599 static DEFINE_MUTEX(hash_resize_mutex);
600 static void xfrm_hash_resize(struct work_struct *__unused)
601 {
602 int dir, total;
603
604 mutex_lock(&hash_resize_mutex);
605
606 total = 0;
607 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
608 if (xfrm_bydst_should_resize(dir, &total))
609 xfrm_bydst_resize(dir);
610 }
611 if (xfrm_byidx_should_resize(total))
612 xfrm_byidx_resize(total);
613
614 mutex_unlock(&hash_resize_mutex);
615 }
616
617 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
618
619 /* Generate new index... KAME seems to generate them ordered by cost
620 * of an absolute inpredictability of ordering of rules. This will not pass. */
621 static u32 xfrm_gen_index(u8 type, int dir)
622 {
623 static u32 idx_generator;
624
625 for (;;) {
626 struct hlist_node *entry;
627 struct hlist_head *list;
628 struct xfrm_policy *p;
629 u32 idx;
630 int found;
631
632 idx = (idx_generator | dir);
633 idx_generator += 8;
634 if (idx == 0)
635 idx = 8;
636 list = xfrm_policy_byidx + idx_hash(idx);
637 found = 0;
638 hlist_for_each_entry(p, entry, list, byidx) {
639 if (p->index == idx) {
640 found = 1;
641 break;
642 }
643 }
644 if (!found)
645 return idx;
646 }
647 }
648
649 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
650 {
651 u32 *p1 = (u32 *) s1;
652 u32 *p2 = (u32 *) s2;
653 int len = sizeof(struct xfrm_selector) / sizeof(u32);
654 int i;
655
656 for (i = 0; i < len; i++) {
657 if (p1[i] != p2[i])
658 return 1;
659 }
660
661 return 0;
662 }
663
664 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
665 {
666 struct xfrm_policy *pol;
667 struct xfrm_policy *delpol;
668 struct hlist_head *chain;
669 struct hlist_node *entry, *newpos;
670 struct dst_entry *gc_list;
671
672 write_lock_bh(&xfrm_policy_lock);
673 chain = policy_hash_bysel(&policy->selector, policy->family, dir);
674 delpol = NULL;
675 newpos = NULL;
676 hlist_for_each_entry(pol, entry, chain, bydst) {
677 if (pol->type == policy->type &&
678 !selector_cmp(&pol->selector, &policy->selector) &&
679 xfrm_sec_ctx_match(pol->security, policy->security) &&
680 !WARN_ON(delpol)) {
681 if (excl) {
682 write_unlock_bh(&xfrm_policy_lock);
683 return -EEXIST;
684 }
685 delpol = pol;
686 if (policy->priority > pol->priority)
687 continue;
688 } else if (policy->priority >= pol->priority) {
689 newpos = &pol->bydst;
690 continue;
691 }
692 if (delpol)
693 break;
694 }
695 if (newpos)
696 hlist_add_after(newpos, &policy->bydst);
697 else
698 hlist_add_head(&policy->bydst, chain);
699 xfrm_pol_hold(policy);
700 xfrm_policy_count[dir]++;
701 atomic_inc(&flow_cache_genid);
702 if (delpol) {
703 hlist_del(&delpol->bydst);
704 hlist_del(&delpol->byidx);
705 xfrm_policy_count[dir]--;
706 }
707 policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
708 hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
709 policy->curlft.add_time = get_seconds();
710 policy->curlft.use_time = 0;
711 if (!mod_timer(&policy->timer, jiffies + HZ))
712 xfrm_pol_hold(policy);
713 write_unlock_bh(&xfrm_policy_lock);
714
715 if (delpol)
716 xfrm_policy_kill(delpol);
717 else if (xfrm_bydst_should_resize(dir, NULL))
718 schedule_work(&xfrm_hash_work);
719
720 read_lock_bh(&xfrm_policy_lock);
721 gc_list = NULL;
722 entry = &policy->bydst;
723 hlist_for_each_entry_continue(policy, entry, bydst) {
724 struct dst_entry *dst;
725
726 write_lock(&policy->lock);
727 dst = policy->bundles;
728 if (dst) {
729 struct dst_entry *tail = dst;
730 while (tail->next)
731 tail = tail->next;
732 tail->next = gc_list;
733 gc_list = dst;
734
735 policy->bundles = NULL;
736 }
737 write_unlock(&policy->lock);
738 }
739 read_unlock_bh(&xfrm_policy_lock);
740
741 while (gc_list) {
742 struct dst_entry *dst = gc_list;
743
744 gc_list = dst->next;
745 dst_free(dst);
746 }
747
748 return 0;
749 }
750 EXPORT_SYMBOL(xfrm_policy_insert);
751
752 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
753 struct xfrm_selector *sel,
754 struct xfrm_sec_ctx *ctx, int delete,
755 int *err)
756 {
757 struct xfrm_policy *pol, *ret;
758 struct hlist_head *chain;
759 struct hlist_node *entry;
760
761 *err = 0;
762 write_lock_bh(&xfrm_policy_lock);
763 chain = policy_hash_bysel(sel, sel->family, dir);
764 ret = NULL;
765 hlist_for_each_entry(pol, entry, chain, bydst) {
766 if (pol->type == type &&
767 !selector_cmp(sel, &pol->selector) &&
768 xfrm_sec_ctx_match(ctx, pol->security)) {
769 xfrm_pol_hold(pol);
770 if (delete) {
771 *err = security_xfrm_policy_delete(pol);
772 if (*err) {
773 write_unlock_bh(&xfrm_policy_lock);
774 return pol;
775 }
776 hlist_del(&pol->bydst);
777 hlist_del(&pol->byidx);
778 xfrm_policy_count[dir]--;
779 }
780 ret = pol;
781 break;
782 }
783 }
784 write_unlock_bh(&xfrm_policy_lock);
785
786 if (ret && delete) {
787 atomic_inc(&flow_cache_genid);
788 xfrm_policy_kill(ret);
789 }
790 return ret;
791 }
792 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
793
794 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
795 int *err)
796 {
797 struct xfrm_policy *pol, *ret;
798 struct hlist_head *chain;
799 struct hlist_node *entry;
800
801 *err = -ENOENT;
802 if (xfrm_policy_id2dir(id) != dir)
803 return NULL;
804
805 *err = 0;
806 write_lock_bh(&xfrm_policy_lock);
807 chain = xfrm_policy_byidx + idx_hash(id);
808 ret = NULL;
809 hlist_for_each_entry(pol, entry, chain, byidx) {
810 if (pol->type == type && pol->index == id) {
811 xfrm_pol_hold(pol);
812 if (delete) {
813 *err = security_xfrm_policy_delete(pol);
814 if (*err) {
815 write_unlock_bh(&xfrm_policy_lock);
816 return pol;
817 }
818 hlist_del(&pol->bydst);
819 hlist_del(&pol->byidx);
820 xfrm_policy_count[dir]--;
821 }
822 ret = pol;
823 break;
824 }
825 }
826 write_unlock_bh(&xfrm_policy_lock);
827
828 if (ret && delete) {
829 atomic_inc(&flow_cache_genid);
830 xfrm_policy_kill(ret);
831 }
832 return ret;
833 }
834 EXPORT_SYMBOL(xfrm_policy_byid);
835
836 void xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
837 {
838 int dir;
839
840 write_lock_bh(&xfrm_policy_lock);
841 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
842 struct xfrm_policy *pol;
843 struct hlist_node *entry;
844 int i, killed;
845
846 killed = 0;
847 again1:
848 hlist_for_each_entry(pol, entry,
849 &xfrm_policy_inexact[dir], bydst) {
850 if (pol->type != type)
851 continue;
852 hlist_del(&pol->bydst);
853 hlist_del(&pol->byidx);
854 write_unlock_bh(&xfrm_policy_lock);
855
856 xfrm_audit_log(audit_info->loginuid, audit_info->secid,
857 AUDIT_MAC_IPSEC_DELSPD, 1, pol, NULL);
858
859 xfrm_policy_kill(pol);
860 killed++;
861
862 write_lock_bh(&xfrm_policy_lock);
863 goto again1;
864 }
865
866 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
867 again2:
868 hlist_for_each_entry(pol, entry,
869 xfrm_policy_bydst[dir].table + i,
870 bydst) {
871 if (pol->type != type)
872 continue;
873 hlist_del(&pol->bydst);
874 hlist_del(&pol->byidx);
875 write_unlock_bh(&xfrm_policy_lock);
876
877 xfrm_audit_log(audit_info->loginuid,
878 audit_info->secid,
879 AUDIT_MAC_IPSEC_DELSPD, 1,
880 pol, NULL);
881
882 xfrm_policy_kill(pol);
883 killed++;
884
885 write_lock_bh(&xfrm_policy_lock);
886 goto again2;
887 }
888 }
889
890 xfrm_policy_count[dir] -= killed;
891 }
892 atomic_inc(&flow_cache_genid);
893 write_unlock_bh(&xfrm_policy_lock);
894 }
895 EXPORT_SYMBOL(xfrm_policy_flush);
896
897 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
898 void *data)
899 {
900 struct xfrm_policy *pol, *last = NULL;
901 struct hlist_node *entry;
902 int dir, last_dir = 0, count, error;
903
904 read_lock_bh(&xfrm_policy_lock);
905 count = 0;
906
907 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
908 struct hlist_head *table = xfrm_policy_bydst[dir].table;
909 int i;
910
911 hlist_for_each_entry(pol, entry,
912 &xfrm_policy_inexact[dir], bydst) {
913 if (pol->type != type)
914 continue;
915 if (last) {
916 error = func(last, last_dir % XFRM_POLICY_MAX,
917 count, data);
918 if (error)
919 goto out;
920 }
921 last = pol;
922 last_dir = dir;
923 count++;
924 }
925 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
926 hlist_for_each_entry(pol, entry, table + i, bydst) {
927 if (pol->type != type)
928 continue;
929 if (last) {
930 error = func(last, last_dir % XFRM_POLICY_MAX,
931 count, data);
932 if (error)
933 goto out;
934 }
935 last = pol;
936 last_dir = dir;
937 count++;
938 }
939 }
940 }
941 if (count == 0) {
942 error = -ENOENT;
943 goto out;
944 }
945 error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
946 out:
947 read_unlock_bh(&xfrm_policy_lock);
948 return error;
949 }
950 EXPORT_SYMBOL(xfrm_policy_walk);
951
952 /*
953 * Find policy to apply to this flow.
954 *
955 * Returns 0 if policy found, else an -errno.
956 */
957 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
958 u8 type, u16 family, int dir)
959 {
960 struct xfrm_selector *sel = &pol->selector;
961 int match, ret = -ESRCH;
962
963 if (pol->family != family ||
964 pol->type != type)
965 return ret;
966
967 match = xfrm_selector_match(sel, fl, family);
968 if (match)
969 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
970
971 return ret;
972 }
973
974 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
975 u16 family, u8 dir)
976 {
977 int err;
978 struct xfrm_policy *pol, *ret;
979 xfrm_address_t *daddr, *saddr;
980 struct hlist_node *entry;
981 struct hlist_head *chain;
982 u32 priority = ~0U;
983
984 daddr = xfrm_flowi_daddr(fl, family);
985 saddr = xfrm_flowi_saddr(fl, family);
986 if (unlikely(!daddr || !saddr))
987 return NULL;
988
989 read_lock_bh(&xfrm_policy_lock);
990 chain = policy_hash_direct(daddr, saddr, family, dir);
991 ret = NULL;
992 hlist_for_each_entry(pol, entry, chain, bydst) {
993 err = xfrm_policy_match(pol, fl, type, family, dir);
994 if (err) {
995 if (err == -ESRCH)
996 continue;
997 else {
998 ret = ERR_PTR(err);
999 goto fail;
1000 }
1001 } else {
1002 ret = pol;
1003 priority = ret->priority;
1004 break;
1005 }
1006 }
1007 chain = &xfrm_policy_inexact[dir];
1008 hlist_for_each_entry(pol, entry, chain, bydst) {
1009 err = xfrm_policy_match(pol, fl, type, family, dir);
1010 if (err) {
1011 if (err == -ESRCH)
1012 continue;
1013 else {
1014 ret = ERR_PTR(err);
1015 goto fail;
1016 }
1017 } else if (pol->priority < priority) {
1018 ret = pol;
1019 break;
1020 }
1021 }
1022 if (ret)
1023 xfrm_pol_hold(ret);
1024 fail:
1025 read_unlock_bh(&xfrm_policy_lock);
1026
1027 return ret;
1028 }
1029
1030 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1031 void **objp, atomic_t **obj_refp)
1032 {
1033 struct xfrm_policy *pol;
1034 int err = 0;
1035
1036 #ifdef CONFIG_XFRM_SUB_POLICY
1037 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1038 if (IS_ERR(pol)) {
1039 err = PTR_ERR(pol);
1040 pol = NULL;
1041 }
1042 if (pol || err)
1043 goto end;
1044 #endif
1045 pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1046 if (IS_ERR(pol)) {
1047 err = PTR_ERR(pol);
1048 pol = NULL;
1049 }
1050 #ifdef CONFIG_XFRM_SUB_POLICY
1051 end:
1052 #endif
1053 if ((*objp = (void *) pol) != NULL)
1054 *obj_refp = &pol->refcnt;
1055 return err;
1056 }
1057
1058 static inline int policy_to_flow_dir(int dir)
1059 {
1060 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1061 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1062 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1063 return dir;
1064 switch (dir) {
1065 default:
1066 case XFRM_POLICY_IN:
1067 return FLOW_DIR_IN;
1068 case XFRM_POLICY_OUT:
1069 return FLOW_DIR_OUT;
1070 case XFRM_POLICY_FWD:
1071 return FLOW_DIR_FWD;
1072 }
1073 }
1074
1075 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1076 {
1077 struct xfrm_policy *pol;
1078
1079 read_lock_bh(&xfrm_policy_lock);
1080 if ((pol = sk->sk_policy[dir]) != NULL) {
1081 int match = xfrm_selector_match(&pol->selector, fl,
1082 sk->sk_family);
1083 int err = 0;
1084
1085 if (match) {
1086 err = security_xfrm_policy_lookup(pol, fl->secid,
1087 policy_to_flow_dir(dir));
1088 if (!err)
1089 xfrm_pol_hold(pol);
1090 else if (err == -ESRCH)
1091 pol = NULL;
1092 else
1093 pol = ERR_PTR(err);
1094 } else
1095 pol = NULL;
1096 }
1097 read_unlock_bh(&xfrm_policy_lock);
1098 return pol;
1099 }
1100
1101 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1102 {
1103 struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1104 pol->family, dir);
1105
1106 hlist_add_head(&pol->bydst, chain);
1107 hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1108 xfrm_policy_count[dir]++;
1109 xfrm_pol_hold(pol);
1110
1111 if (xfrm_bydst_should_resize(dir, NULL))
1112 schedule_work(&xfrm_hash_work);
1113 }
1114
1115 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1116 int dir)
1117 {
1118 if (hlist_unhashed(&pol->bydst))
1119 return NULL;
1120
1121 hlist_del(&pol->bydst);
1122 hlist_del(&pol->byidx);
1123 xfrm_policy_count[dir]--;
1124
1125 return pol;
1126 }
1127
1128 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1129 {
1130 write_lock_bh(&xfrm_policy_lock);
1131 pol = __xfrm_policy_unlink(pol, dir);
1132 write_unlock_bh(&xfrm_policy_lock);
1133 if (pol) {
1134 if (dir < XFRM_POLICY_MAX)
1135 atomic_inc(&flow_cache_genid);
1136 xfrm_policy_kill(pol);
1137 return 0;
1138 }
1139 return -ENOENT;
1140 }
1141 EXPORT_SYMBOL(xfrm_policy_delete);
1142
1143 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1144 {
1145 struct xfrm_policy *old_pol;
1146
1147 #ifdef CONFIG_XFRM_SUB_POLICY
1148 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1149 return -EINVAL;
1150 #endif
1151
1152 write_lock_bh(&xfrm_policy_lock);
1153 old_pol = sk->sk_policy[dir];
1154 sk->sk_policy[dir] = pol;
1155 if (pol) {
1156 pol->curlft.add_time = get_seconds();
1157 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1158 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1159 }
1160 if (old_pol)
1161 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1162 write_unlock_bh(&xfrm_policy_lock);
1163
1164 if (old_pol) {
1165 xfrm_policy_kill(old_pol);
1166 }
1167 return 0;
1168 }
1169
1170 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1171 {
1172 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1173
1174 if (newp) {
1175 newp->selector = old->selector;
1176 if (security_xfrm_policy_clone(old, newp)) {
1177 kfree(newp);
1178 return NULL; /* ENOMEM */
1179 }
1180 newp->lft = old->lft;
1181 newp->curlft = old->curlft;
1182 newp->action = old->action;
1183 newp->flags = old->flags;
1184 newp->xfrm_nr = old->xfrm_nr;
1185 newp->index = old->index;
1186 newp->type = old->type;
1187 memcpy(newp->xfrm_vec, old->xfrm_vec,
1188 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1189 write_lock_bh(&xfrm_policy_lock);
1190 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1191 write_unlock_bh(&xfrm_policy_lock);
1192 xfrm_pol_put(newp);
1193 }
1194 return newp;
1195 }
1196
1197 int __xfrm_sk_clone_policy(struct sock *sk)
1198 {
1199 struct xfrm_policy *p0 = sk->sk_policy[0],
1200 *p1 = sk->sk_policy[1];
1201
1202 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1203 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1204 return -ENOMEM;
1205 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1206 return -ENOMEM;
1207 return 0;
1208 }
1209
1210 static int
1211 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1212 unsigned short family)
1213 {
1214 int err;
1215 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1216
1217 if (unlikely(afinfo == NULL))
1218 return -EINVAL;
1219 err = afinfo->get_saddr(local, remote);
1220 xfrm_policy_put_afinfo(afinfo);
1221 return err;
1222 }
1223
1224 /* Resolve list of templates for the flow, given policy. */
1225
1226 static int
1227 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1228 struct xfrm_state **xfrm,
1229 unsigned short family)
1230 {
1231 int nx;
1232 int i, error;
1233 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1234 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1235 xfrm_address_t tmp;
1236
1237 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1238 struct xfrm_state *x;
1239 xfrm_address_t *remote = daddr;
1240 xfrm_address_t *local = saddr;
1241 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1242
1243 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1244 remote = &tmpl->id.daddr;
1245 local = &tmpl->saddr;
1246 family = tmpl->encap_family;
1247 if (xfrm_addr_any(local, family)) {
1248 error = xfrm_get_saddr(&tmp, remote, family);
1249 if (error)
1250 goto fail;
1251 local = &tmp;
1252 }
1253 }
1254
1255 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1256
1257 if (x && x->km.state == XFRM_STATE_VALID) {
1258 xfrm[nx++] = x;
1259 daddr = remote;
1260 saddr = local;
1261 continue;
1262 }
1263 if (x) {
1264 error = (x->km.state == XFRM_STATE_ERROR ?
1265 -EINVAL : -EAGAIN);
1266 xfrm_state_put(x);
1267 }
1268
1269 if (!tmpl->optional)
1270 goto fail;
1271 }
1272 return nx;
1273
1274 fail:
1275 for (nx--; nx>=0; nx--)
1276 xfrm_state_put(xfrm[nx]);
1277 return error;
1278 }
1279
1280 static int
1281 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1282 struct xfrm_state **xfrm,
1283 unsigned short family)
1284 {
1285 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1286 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1287 int cnx = 0;
1288 int error;
1289 int ret;
1290 int i;
1291
1292 for (i = 0; i < npols; i++) {
1293 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1294 error = -ENOBUFS;
1295 goto fail;
1296 }
1297
1298 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1299 if (ret < 0) {
1300 error = ret;
1301 goto fail;
1302 } else
1303 cnx += ret;
1304 }
1305
1306 /* found states are sorted for outbound processing */
1307 if (npols > 1)
1308 xfrm_state_sort(xfrm, tpp, cnx, family);
1309
1310 return cnx;
1311
1312 fail:
1313 for (cnx--; cnx>=0; cnx--)
1314 xfrm_state_put(tpp[cnx]);
1315 return error;
1316
1317 }
1318
1319 /* Check that the bundle accepts the flow and its components are
1320 * still valid.
1321 */
1322
1323 static struct dst_entry *
1324 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1325 {
1326 struct dst_entry *x;
1327 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1328 if (unlikely(afinfo == NULL))
1329 return ERR_PTR(-EINVAL);
1330 x = afinfo->find_bundle(fl, policy);
1331 xfrm_policy_put_afinfo(afinfo);
1332 return x;
1333 }
1334
1335 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1336 * all the metrics... Shortly, bundle a bundle.
1337 */
1338
1339 static int
1340 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1341 struct flowi *fl, struct dst_entry **dst_p,
1342 unsigned short family)
1343 {
1344 int err;
1345 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1346 if (unlikely(afinfo == NULL))
1347 return -EINVAL;
1348 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1349 xfrm_policy_put_afinfo(afinfo);
1350 return err;
1351 }
1352
1353 static int inline
1354 xfrm_dst_alloc_copy(void **target, void *src, int size)
1355 {
1356 if (!*target) {
1357 *target = kmalloc(size, GFP_ATOMIC);
1358 if (!*target)
1359 return -ENOMEM;
1360 }
1361 memcpy(*target, src, size);
1362 return 0;
1363 }
1364
1365 static int inline
1366 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1367 {
1368 #ifdef CONFIG_XFRM_SUB_POLICY
1369 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1370 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1371 sel, sizeof(*sel));
1372 #else
1373 return 0;
1374 #endif
1375 }
1376
1377 static int inline
1378 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1379 {
1380 #ifdef CONFIG_XFRM_SUB_POLICY
1381 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1382 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1383 #else
1384 return 0;
1385 #endif
1386 }
1387
1388 static int stale_bundle(struct dst_entry *dst);
1389
1390 /* Main function: finds/creates a bundle for given flow.
1391 *
1392 * At the moment we eat a raw IP route. Mostly to speed up lookups
1393 * on interfaces with disabled IPsec.
1394 */
1395 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1396 struct sock *sk, int flags)
1397 {
1398 struct xfrm_policy *policy;
1399 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1400 int npols;
1401 int pol_dead;
1402 int xfrm_nr;
1403 int pi;
1404 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1405 struct dst_entry *dst, *dst_orig = *dst_p;
1406 int nx = 0;
1407 int err;
1408 u32 genid;
1409 u16 family;
1410 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1411
1412 restart:
1413 genid = atomic_read(&flow_cache_genid);
1414 policy = NULL;
1415 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1416 pols[pi] = NULL;
1417 npols = 0;
1418 pol_dead = 0;
1419 xfrm_nr = 0;
1420
1421 if (sk && sk->sk_policy[1]) {
1422 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1423 if (IS_ERR(policy))
1424 return PTR_ERR(policy);
1425 }
1426
1427 if (!policy) {
1428 /* To accelerate a bit... */
1429 if ((dst_orig->flags & DST_NOXFRM) ||
1430 !xfrm_policy_count[XFRM_POLICY_OUT])
1431 return 0;
1432
1433 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1434 dir, xfrm_policy_lookup);
1435 if (IS_ERR(policy))
1436 return PTR_ERR(policy);
1437 }
1438
1439 if (!policy)
1440 return 0;
1441
1442 family = dst_orig->ops->family;
1443 policy->curlft.use_time = get_seconds();
1444 pols[0] = policy;
1445 npols ++;
1446 xfrm_nr += pols[0]->xfrm_nr;
1447
1448 switch (policy->action) {
1449 case XFRM_POLICY_BLOCK:
1450 /* Prohibit the flow */
1451 err = -EPERM;
1452 goto error;
1453
1454 case XFRM_POLICY_ALLOW:
1455 #ifndef CONFIG_XFRM_SUB_POLICY
1456 if (policy->xfrm_nr == 0) {
1457 /* Flow passes not transformed. */
1458 xfrm_pol_put(policy);
1459 return 0;
1460 }
1461 #endif
1462
1463 /* Try to find matching bundle.
1464 *
1465 * LATER: help from flow cache. It is optional, this
1466 * is required only for output policy.
1467 */
1468 dst = xfrm_find_bundle(fl, policy, family);
1469 if (IS_ERR(dst)) {
1470 err = PTR_ERR(dst);
1471 goto error;
1472 }
1473
1474 if (dst)
1475 break;
1476
1477 #ifdef CONFIG_XFRM_SUB_POLICY
1478 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1479 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1480 fl, family,
1481 XFRM_POLICY_OUT);
1482 if (pols[1]) {
1483 if (IS_ERR(pols[1])) {
1484 err = PTR_ERR(pols[1]);
1485 goto error;
1486 }
1487 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1488 err = -EPERM;
1489 goto error;
1490 }
1491 npols ++;
1492 xfrm_nr += pols[1]->xfrm_nr;
1493 }
1494 }
1495
1496 /*
1497 * Because neither flowi nor bundle information knows about
1498 * transformation template size. On more than one policy usage
1499 * we can realize whether all of them is bypass or not after
1500 * they are searched. See above not-transformed bypass
1501 * is surrounded by non-sub policy configuration, too.
1502 */
1503 if (xfrm_nr == 0) {
1504 /* Flow passes not transformed. */
1505 xfrm_pols_put(pols, npols);
1506 return 0;
1507 }
1508
1509 #endif
1510 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1511
1512 if (unlikely(nx<0)) {
1513 err = nx;
1514 if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1515 /* EREMOTE tells the caller to generate
1516 * a one-shot blackhole route.
1517 */
1518 xfrm_pol_put(policy);
1519 return -EREMOTE;
1520 }
1521 if (err == -EAGAIN && flags) {
1522 DECLARE_WAITQUEUE(wait, current);
1523
1524 add_wait_queue(&km_waitq, &wait);
1525 set_current_state(TASK_INTERRUPTIBLE);
1526 schedule();
1527 set_current_state(TASK_RUNNING);
1528 remove_wait_queue(&km_waitq, &wait);
1529
1530 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1531
1532 if (nx == -EAGAIN && signal_pending(current)) {
1533 err = -ERESTART;
1534 goto error;
1535 }
1536 if (nx == -EAGAIN ||
1537 genid != atomic_read(&flow_cache_genid)) {
1538 xfrm_pols_put(pols, npols);
1539 goto restart;
1540 }
1541 err = nx;
1542 }
1543 if (err < 0)
1544 goto error;
1545 }
1546 if (nx == 0) {
1547 /* Flow passes not transformed. */
1548 xfrm_pols_put(pols, npols);
1549 return 0;
1550 }
1551
1552 dst = dst_orig;
1553 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1554
1555 if (unlikely(err)) {
1556 int i;
1557 for (i=0; i<nx; i++)
1558 xfrm_state_put(xfrm[i]);
1559 goto error;
1560 }
1561
1562 for (pi = 0; pi < npols; pi++) {
1563 read_lock_bh(&pols[pi]->lock);
1564 pol_dead |= pols[pi]->dead;
1565 read_unlock_bh(&pols[pi]->lock);
1566 }
1567
1568 write_lock_bh(&policy->lock);
1569 if (unlikely(pol_dead || stale_bundle(dst))) {
1570 /* Wow! While we worked on resolving, this
1571 * policy has gone. Retry. It is not paranoia,
1572 * we just cannot enlist new bundle to dead object.
1573 * We can't enlist stable bundles either.
1574 */
1575 write_unlock_bh(&policy->lock);
1576 if (dst)
1577 dst_free(dst);
1578
1579 err = -EHOSTUNREACH;
1580 goto error;
1581 }
1582
1583 if (npols > 1)
1584 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1585 else
1586 err = xfrm_dst_update_origin(dst, fl);
1587 if (unlikely(err)) {
1588 write_unlock_bh(&policy->lock);
1589 if (dst)
1590 dst_free(dst);
1591 goto error;
1592 }
1593
1594 dst->next = policy->bundles;
1595 policy->bundles = dst;
1596 dst_hold(dst);
1597 write_unlock_bh(&policy->lock);
1598 }
1599 *dst_p = dst;
1600 dst_release(dst_orig);
1601 xfrm_pols_put(pols, npols);
1602 return 0;
1603
1604 error:
1605 dst_release(dst_orig);
1606 xfrm_pols_put(pols, npols);
1607 *dst_p = NULL;
1608 return err;
1609 }
1610 EXPORT_SYMBOL(__xfrm_lookup);
1611
1612 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1613 struct sock *sk, int flags)
1614 {
1615 int err = __xfrm_lookup(dst_p, fl, sk, flags);
1616
1617 if (err == -EREMOTE) {
1618 dst_release(*dst_p);
1619 *dst_p = NULL;
1620 err = -EAGAIN;
1621 }
1622
1623 return err;
1624 }
1625 EXPORT_SYMBOL(xfrm_lookup);
1626
1627 static inline int
1628 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1629 {
1630 struct xfrm_state *x;
1631 int err;
1632
1633 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1634 return 0;
1635 x = skb->sp->xvec[idx];
1636 if (!x->type->reject)
1637 return 0;
1638 xfrm_state_hold(x);
1639 err = x->type->reject(x, skb, fl);
1640 xfrm_state_put(x);
1641 return err;
1642 }
1643
1644 /* When skb is transformed back to its "native" form, we have to
1645 * check policy restrictions. At the moment we make this in maximally
1646 * stupid way. Shame on me. :-) Of course, connected sockets must
1647 * have policy cached at them.
1648 */
1649
1650 static inline int
1651 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1652 unsigned short family)
1653 {
1654 if (xfrm_state_kern(x))
1655 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1656 return x->id.proto == tmpl->id.proto &&
1657 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1658 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1659 x->props.mode == tmpl->mode &&
1660 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1661 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1662 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1663 xfrm_state_addr_cmp(tmpl, x, family));
1664 }
1665
1666 /*
1667 * 0 or more than 0 is returned when validation is succeeded (either bypass
1668 * because of optional transport mode, or next index of the mathced secpath
1669 * state with the template.
1670 * -1 is returned when no matching template is found.
1671 * Otherwise "-2 - errored_index" is returned.
1672 */
1673 static inline int
1674 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1675 unsigned short family)
1676 {
1677 int idx = start;
1678
1679 if (tmpl->optional) {
1680 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1681 return start;
1682 } else
1683 start = -1;
1684 for (; idx < sp->len; idx++) {
1685 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1686 return ++idx;
1687 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1688 if (start == -1)
1689 start = -2-idx;
1690 break;
1691 }
1692 }
1693 return start;
1694 }
1695
1696 int
1697 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1698 {
1699 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1700 int err;
1701
1702 if (unlikely(afinfo == NULL))
1703 return -EAFNOSUPPORT;
1704
1705 afinfo->decode_session(skb, fl);
1706 err = security_xfrm_decode_session(skb, &fl->secid);
1707 xfrm_policy_put_afinfo(afinfo);
1708 return err;
1709 }
1710 EXPORT_SYMBOL(xfrm_decode_session);
1711
1712 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1713 {
1714 for (; k < sp->len; k++) {
1715 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1716 *idxp = k;
1717 return 1;
1718 }
1719 }
1720
1721 return 0;
1722 }
1723
1724 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1725 unsigned short family)
1726 {
1727 struct xfrm_policy *pol;
1728 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1729 int npols = 0;
1730 int xfrm_nr;
1731 int pi;
1732 struct flowi fl;
1733 u8 fl_dir = policy_to_flow_dir(dir);
1734 int xerr_idx = -1;
1735
1736 if (xfrm_decode_session(skb, &fl, family) < 0)
1737 return 0;
1738 nf_nat_decode_session(skb, &fl, family);
1739
1740 /* First, check used SA against their selectors. */
1741 if (skb->sp) {
1742 int i;
1743
1744 for (i=skb->sp->len-1; i>=0; i--) {
1745 struct xfrm_state *x = skb->sp->xvec[i];
1746 if (!xfrm_selector_match(&x->sel, &fl, family))
1747 return 0;
1748 }
1749 }
1750
1751 pol = NULL;
1752 if (sk && sk->sk_policy[dir]) {
1753 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1754 if (IS_ERR(pol))
1755 return 0;
1756 }
1757
1758 if (!pol)
1759 pol = flow_cache_lookup(&fl, family, fl_dir,
1760 xfrm_policy_lookup);
1761
1762 if (IS_ERR(pol))
1763 return 0;
1764
1765 if (!pol) {
1766 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1767 xfrm_secpath_reject(xerr_idx, skb, &fl);
1768 return 0;
1769 }
1770 return 1;
1771 }
1772
1773 pol->curlft.use_time = get_seconds();
1774
1775 pols[0] = pol;
1776 npols ++;
1777 #ifdef CONFIG_XFRM_SUB_POLICY
1778 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1779 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1780 &fl, family,
1781 XFRM_POLICY_IN);
1782 if (pols[1]) {
1783 if (IS_ERR(pols[1]))
1784 return 0;
1785 pols[1]->curlft.use_time = get_seconds();
1786 npols ++;
1787 }
1788 }
1789 #endif
1790
1791 if (pol->action == XFRM_POLICY_ALLOW) {
1792 struct sec_path *sp;
1793 static struct sec_path dummy;
1794 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1795 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1796 struct xfrm_tmpl **tpp = tp;
1797 int ti = 0;
1798 int i, k;
1799
1800 if ((sp = skb->sp) == NULL)
1801 sp = &dummy;
1802
1803 for (pi = 0; pi < npols; pi++) {
1804 if (pols[pi] != pol &&
1805 pols[pi]->action != XFRM_POLICY_ALLOW)
1806 goto reject;
1807 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1808 goto reject_error;
1809 for (i = 0; i < pols[pi]->xfrm_nr; i++)
1810 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1811 }
1812 xfrm_nr = ti;
1813 if (npols > 1) {
1814 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1815 tpp = stp;
1816 }
1817
1818 /* For each tunnel xfrm, find the first matching tmpl.
1819 * For each tmpl before that, find corresponding xfrm.
1820 * Order is _important_. Later we will implement
1821 * some barriers, but at the moment barriers
1822 * are implied between each two transformations.
1823 */
1824 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1825 k = xfrm_policy_ok(tpp[i], sp, k, family);
1826 if (k < 0) {
1827 if (k < -1)
1828 /* "-2 - errored_index" returned */
1829 xerr_idx = -(2+k);
1830 goto reject;
1831 }
1832 }
1833
1834 if (secpath_has_nontransport(sp, k, &xerr_idx))
1835 goto reject;
1836
1837 xfrm_pols_put(pols, npols);
1838 return 1;
1839 }
1840
1841 reject:
1842 xfrm_secpath_reject(xerr_idx, skb, &fl);
1843 reject_error:
1844 xfrm_pols_put(pols, npols);
1845 return 0;
1846 }
1847 EXPORT_SYMBOL(__xfrm_policy_check);
1848
1849 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1850 {
1851 struct flowi fl;
1852
1853 if (xfrm_decode_session(skb, &fl, family) < 0)
1854 return 0;
1855
1856 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1857 }
1858 EXPORT_SYMBOL(__xfrm_route_forward);
1859
1860 /* Optimize later using cookies and generation ids. */
1861
1862 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1863 {
1864 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1865 * to "-1" to force all XFRM destinations to get validated by
1866 * dst_ops->check on every use. We do this because when a
1867 * normal route referenced by an XFRM dst is obsoleted we do
1868 * not go looking around for all parent referencing XFRM dsts
1869 * so that we can invalidate them. It is just too much work.
1870 * Instead we make the checks here on every use. For example:
1871 *
1872 * XFRM dst A --> IPv4 dst X
1873 *
1874 * X is the "xdst->route" of A (X is also the "dst->path" of A
1875 * in this example). If X is marked obsolete, "A" will not
1876 * notice. That's what we are validating here via the
1877 * stale_bundle() check.
1878 *
1879 * When a policy's bundle is pruned, we dst_free() the XFRM
1880 * dst which causes it's ->obsolete field to be set to a
1881 * positive non-zero integer. If an XFRM dst has been pruned
1882 * like this, we want to force a new route lookup.
1883 */
1884 if (dst->obsolete < 0 && !stale_bundle(dst))
1885 return dst;
1886
1887 return NULL;
1888 }
1889
1890 static int stale_bundle(struct dst_entry *dst)
1891 {
1892 return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1893 }
1894
1895 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1896 {
1897 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1898 dst->dev = &loopback_dev;
1899 dev_hold(&loopback_dev);
1900 dev_put(dev);
1901 }
1902 }
1903 EXPORT_SYMBOL(xfrm_dst_ifdown);
1904
1905 static void xfrm_link_failure(struct sk_buff *skb)
1906 {
1907 /* Impossible. Such dst must be popped before reaches point of failure. */
1908 return;
1909 }
1910
1911 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1912 {
1913 if (dst) {
1914 if (dst->obsolete) {
1915 dst_release(dst);
1916 dst = NULL;
1917 }
1918 }
1919 return dst;
1920 }
1921
1922 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1923 {
1924 struct dst_entry *dst, **dstp;
1925
1926 write_lock(&pol->lock);
1927 dstp = &pol->bundles;
1928 while ((dst=*dstp) != NULL) {
1929 if (func(dst)) {
1930 *dstp = dst->next;
1931 dst->next = *gc_list_p;
1932 *gc_list_p = dst;
1933 } else {
1934 dstp = &dst->next;
1935 }
1936 }
1937 write_unlock(&pol->lock);
1938 }
1939
1940 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1941 {
1942 struct dst_entry *gc_list = NULL;
1943 int dir;
1944
1945 read_lock_bh(&xfrm_policy_lock);
1946 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1947 struct xfrm_policy *pol;
1948 struct hlist_node *entry;
1949 struct hlist_head *table;
1950 int i;
1951
1952 hlist_for_each_entry(pol, entry,
1953 &xfrm_policy_inexact[dir], bydst)
1954 prune_one_bundle(pol, func, &gc_list);
1955
1956 table = xfrm_policy_bydst[dir].table;
1957 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1958 hlist_for_each_entry(pol, entry, table + i, bydst)
1959 prune_one_bundle(pol, func, &gc_list);
1960 }
1961 }
1962 read_unlock_bh(&xfrm_policy_lock);
1963
1964 while (gc_list) {
1965 struct dst_entry *dst = gc_list;
1966 gc_list = dst->next;
1967 dst_free(dst);
1968 }
1969 }
1970
1971 static int unused_bundle(struct dst_entry *dst)
1972 {
1973 return !atomic_read(&dst->__refcnt);
1974 }
1975
1976 static void __xfrm_garbage_collect(void)
1977 {
1978 xfrm_prune_bundles(unused_bundle);
1979 }
1980
1981 static int xfrm_flush_bundles(void)
1982 {
1983 xfrm_prune_bundles(stale_bundle);
1984 return 0;
1985 }
1986
1987 void xfrm_init_pmtu(struct dst_entry *dst)
1988 {
1989 do {
1990 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1991 u32 pmtu, route_mtu_cached;
1992
1993 pmtu = dst_mtu(dst->child);
1994 xdst->child_mtu_cached = pmtu;
1995
1996 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1997
1998 route_mtu_cached = dst_mtu(xdst->route);
1999 xdst->route_mtu_cached = route_mtu_cached;
2000
2001 if (pmtu > route_mtu_cached)
2002 pmtu = route_mtu_cached;
2003
2004 dst->metrics[RTAX_MTU-1] = pmtu;
2005 } while ((dst = dst->next));
2006 }
2007
2008 EXPORT_SYMBOL(xfrm_init_pmtu);
2009
2010 /* Check that the bundle accepts the flow and its components are
2011 * still valid.
2012 */
2013
2014 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2015 struct flowi *fl, int family, int strict)
2016 {
2017 struct dst_entry *dst = &first->u.dst;
2018 struct xfrm_dst *last;
2019 u32 mtu;
2020
2021 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2022 (dst->dev && !netif_running(dst->dev)))
2023 return 0;
2024 #ifdef CONFIG_XFRM_SUB_POLICY
2025 if (fl) {
2026 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2027 return 0;
2028 if (first->partner &&
2029 !xfrm_selector_match(first->partner, fl, family))
2030 return 0;
2031 }
2032 #endif
2033
2034 last = NULL;
2035
2036 do {
2037 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2038
2039 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2040 return 0;
2041 if (fl && pol &&
2042 !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2043 return 0;
2044 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2045 return 0;
2046 if (xdst->genid != dst->xfrm->genid)
2047 return 0;
2048
2049 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
2050 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2051 return 0;
2052
2053 mtu = dst_mtu(dst->child);
2054 if (xdst->child_mtu_cached != mtu) {
2055 last = xdst;
2056 xdst->child_mtu_cached = mtu;
2057 }
2058
2059 if (!dst_check(xdst->route, xdst->route_cookie))
2060 return 0;
2061 mtu = dst_mtu(xdst->route);
2062 if (xdst->route_mtu_cached != mtu) {
2063 last = xdst;
2064 xdst->route_mtu_cached = mtu;
2065 }
2066
2067 dst = dst->child;
2068 } while (dst->xfrm);
2069
2070 if (likely(!last))
2071 return 1;
2072
2073 mtu = last->child_mtu_cached;
2074 for (;;) {
2075 dst = &last->u.dst;
2076
2077 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2078 if (mtu > last->route_mtu_cached)
2079 mtu = last->route_mtu_cached;
2080 dst->metrics[RTAX_MTU-1] = mtu;
2081
2082 if (last == first)
2083 break;
2084
2085 last = last->u.next;
2086 last->child_mtu_cached = mtu;
2087 }
2088
2089 return 1;
2090 }
2091
2092 EXPORT_SYMBOL(xfrm_bundle_ok);
2093
2094 #ifdef CONFIG_AUDITSYSCALL
2095 /* Audit addition and deletion of SAs and ipsec policy */
2096
2097 void xfrm_audit_log(uid_t auid, u32 sid, int type, int result,
2098 struct xfrm_policy *xp, struct xfrm_state *x)
2099 {
2100
2101 char *secctx;
2102 u32 secctx_len;
2103 struct xfrm_sec_ctx *sctx = NULL;
2104 struct audit_buffer *audit_buf;
2105 int family;
2106 extern int audit_enabled;
2107
2108 if (audit_enabled == 0)
2109 return;
2110
2111 BUG_ON((type == AUDIT_MAC_IPSEC_ADDSA ||
2112 type == AUDIT_MAC_IPSEC_DELSA) && !x);
2113 BUG_ON((type == AUDIT_MAC_IPSEC_ADDSPD ||
2114 type == AUDIT_MAC_IPSEC_DELSPD) && !xp);
2115
2116 audit_buf = audit_log_start(current->audit_context, GFP_ATOMIC, type);
2117 if (audit_buf == NULL)
2118 return;
2119
2120 switch(type) {
2121 case AUDIT_MAC_IPSEC_ADDSA:
2122 audit_log_format(audit_buf, "SAD add: auid=%u", auid);
2123 break;
2124 case AUDIT_MAC_IPSEC_DELSA:
2125 audit_log_format(audit_buf, "SAD delete: auid=%u", auid);
2126 break;
2127 case AUDIT_MAC_IPSEC_ADDSPD:
2128 audit_log_format(audit_buf, "SPD add: auid=%u", auid);
2129 break;
2130 case AUDIT_MAC_IPSEC_DELSPD:
2131 audit_log_format(audit_buf, "SPD delete: auid=%u", auid);
2132 break;
2133 default:
2134 return;
2135 }
2136
2137 if (sid != 0 &&
2138 security_secid_to_secctx(sid, &secctx, &secctx_len) == 0)
2139 audit_log_format(audit_buf, " subj=%s", secctx);
2140 else
2141 audit_log_task_context(audit_buf);
2142
2143 if (xp) {
2144 family = xp->selector.family;
2145 if (xp->security)
2146 sctx = xp->security;
2147 } else {
2148 family = x->props.family;
2149 if (x->security)
2150 sctx = x->security;
2151 }
2152
2153 if (sctx)
2154 audit_log_format(audit_buf,
2155 " sec_alg=%u sec_doi=%u sec_obj=%s",
2156 sctx->ctx_alg, sctx->ctx_doi, sctx->ctx_str);
2157
2158 switch(family) {
2159 case AF_INET:
2160 {
2161 struct in_addr saddr, daddr;
2162 if (xp) {
2163 saddr.s_addr = xp->selector.saddr.a4;
2164 daddr.s_addr = xp->selector.daddr.a4;
2165 } else {
2166 saddr.s_addr = x->props.saddr.a4;
2167 daddr.s_addr = x->id.daddr.a4;
2168 }
2169 audit_log_format(audit_buf,
2170 " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2171 NIPQUAD(saddr), NIPQUAD(daddr));
2172 }
2173 break;
2174 case AF_INET6:
2175 {
2176 struct in6_addr saddr6, daddr6;
2177 if (xp) {
2178 memcpy(&saddr6, xp->selector.saddr.a6,
2179 sizeof(struct in6_addr));
2180 memcpy(&daddr6, xp->selector.daddr.a6,
2181 sizeof(struct in6_addr));
2182 } else {
2183 memcpy(&saddr6, x->props.saddr.a6,
2184 sizeof(struct in6_addr));
2185 memcpy(&daddr6, x->id.daddr.a6,
2186 sizeof(struct in6_addr));
2187 }
2188 audit_log_format(audit_buf,
2189 " src=" NIP6_FMT " dst=" NIP6_FMT,
2190 NIP6(saddr6), NIP6(daddr6));
2191 }
2192 break;
2193 }
2194
2195 if (x)
2196 audit_log_format(audit_buf, " spi=%lu(0x%lx) protocol=%s",
2197 (unsigned long)ntohl(x->id.spi),
2198 (unsigned long)ntohl(x->id.spi),
2199 x->id.proto == IPPROTO_AH ? "AH" :
2200 (x->id.proto == IPPROTO_ESP ?
2201 "ESP" : "IPCOMP"));
2202
2203 audit_log_format(audit_buf, " res=%u", result);
2204 audit_log_end(audit_buf);
2205 }
2206
2207 EXPORT_SYMBOL(xfrm_audit_log);
2208 #endif /* CONFIG_AUDITSYSCALL */
2209
2210 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2211 {
2212 int err = 0;
2213 if (unlikely(afinfo == NULL))
2214 return -EINVAL;
2215 if (unlikely(afinfo->family >= NPROTO))
2216 return -EAFNOSUPPORT;
2217 write_lock_bh(&xfrm_policy_afinfo_lock);
2218 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2219 err = -ENOBUFS;
2220 else {
2221 struct dst_ops *dst_ops = afinfo->dst_ops;
2222 if (likely(dst_ops->kmem_cachep == NULL))
2223 dst_ops->kmem_cachep = xfrm_dst_cache;
2224 if (likely(dst_ops->check == NULL))
2225 dst_ops->check = xfrm_dst_check;
2226 if (likely(dst_ops->negative_advice == NULL))
2227 dst_ops->negative_advice = xfrm_negative_advice;
2228 if (likely(dst_ops->link_failure == NULL))
2229 dst_ops->link_failure = xfrm_link_failure;
2230 if (likely(afinfo->garbage_collect == NULL))
2231 afinfo->garbage_collect = __xfrm_garbage_collect;
2232 xfrm_policy_afinfo[afinfo->family] = afinfo;
2233 }
2234 write_unlock_bh(&xfrm_policy_afinfo_lock);
2235 return err;
2236 }
2237 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2238
2239 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2240 {
2241 int err = 0;
2242 if (unlikely(afinfo == NULL))
2243 return -EINVAL;
2244 if (unlikely(afinfo->family >= NPROTO))
2245 return -EAFNOSUPPORT;
2246 write_lock_bh(&xfrm_policy_afinfo_lock);
2247 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2248 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2249 err = -EINVAL;
2250 else {
2251 struct dst_ops *dst_ops = afinfo->dst_ops;
2252 xfrm_policy_afinfo[afinfo->family] = NULL;
2253 dst_ops->kmem_cachep = NULL;
2254 dst_ops->check = NULL;
2255 dst_ops->negative_advice = NULL;
2256 dst_ops->link_failure = NULL;
2257 afinfo->garbage_collect = NULL;
2258 }
2259 }
2260 write_unlock_bh(&xfrm_policy_afinfo_lock);
2261 return err;
2262 }
2263 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2264
2265 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2266 {
2267 struct xfrm_policy_afinfo *afinfo;
2268 if (unlikely(family >= NPROTO))
2269 return NULL;
2270 read_lock(&xfrm_policy_afinfo_lock);
2271 afinfo = xfrm_policy_afinfo[family];
2272 if (unlikely(!afinfo))
2273 read_unlock(&xfrm_policy_afinfo_lock);
2274 return afinfo;
2275 }
2276
2277 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2278 {
2279 read_unlock(&xfrm_policy_afinfo_lock);
2280 }
2281
2282 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2283 {
2284 struct xfrm_policy_afinfo *afinfo;
2285 if (unlikely(family >= NPROTO))
2286 return NULL;
2287 write_lock_bh(&xfrm_policy_afinfo_lock);
2288 afinfo = xfrm_policy_afinfo[family];
2289 if (unlikely(!afinfo))
2290 write_unlock_bh(&xfrm_policy_afinfo_lock);
2291 return afinfo;
2292 }
2293
2294 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2295 {
2296 write_unlock_bh(&xfrm_policy_afinfo_lock);
2297 }
2298
2299 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2300 {
2301 switch (event) {
2302 case NETDEV_DOWN:
2303 xfrm_flush_bundles();
2304 }
2305 return NOTIFY_DONE;
2306 }
2307
2308 static struct notifier_block xfrm_dev_notifier = {
2309 xfrm_dev_event,
2310 NULL,
2311 0
2312 };
2313
2314 static void __init xfrm_policy_init(void)
2315 {
2316 unsigned int hmask, sz;
2317 int dir;
2318
2319 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2320 sizeof(struct xfrm_dst),
2321 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2322 NULL, NULL);
2323
2324 hmask = 8 - 1;
2325 sz = (hmask+1) * sizeof(struct hlist_head);
2326
2327 xfrm_policy_byidx = xfrm_hash_alloc(sz);
2328 xfrm_idx_hmask = hmask;
2329 if (!xfrm_policy_byidx)
2330 panic("XFRM: failed to allocate byidx hash\n");
2331
2332 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2333 struct xfrm_policy_hash *htab;
2334
2335 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2336
2337 htab = &xfrm_policy_bydst[dir];
2338 htab->table = xfrm_hash_alloc(sz);
2339 htab->hmask = hmask;
2340 if (!htab->table)
2341 panic("XFRM: failed to allocate bydst hash\n");
2342 }
2343
2344 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2345 register_netdevice_notifier(&xfrm_dev_notifier);
2346 }
2347
2348 void __init xfrm_init(void)
2349 {
2350 xfrm_state_init();
2351 xfrm_policy_init();
2352 xfrm_input_init();
2353 }
2354
2355 #ifdef CONFIG_XFRM_MIGRATE
2356 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2357 struct xfrm_selector *sel_tgt)
2358 {
2359 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2360 if (sel_tgt->family == sel_cmp->family &&
2361 xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2362 sel_cmp->family) == 0 &&
2363 xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2364 sel_cmp->family) == 0 &&
2365 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2366 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2367 return 1;
2368 }
2369 } else {
2370 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2371 return 1;
2372 }
2373 }
2374 return 0;
2375 }
2376
2377 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2378 u8 dir, u8 type)
2379 {
2380 struct xfrm_policy *pol, *ret = NULL;
2381 struct hlist_node *entry;
2382 struct hlist_head *chain;
2383 u32 priority = ~0U;
2384
2385 read_lock_bh(&xfrm_policy_lock);
2386 chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2387 hlist_for_each_entry(pol, entry, chain, bydst) {
2388 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2389 pol->type == type) {
2390 ret = pol;
2391 priority = ret->priority;
2392 break;
2393 }
2394 }
2395 chain = &xfrm_policy_inexact[dir];
2396 hlist_for_each_entry(pol, entry, chain, bydst) {
2397 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2398 pol->type == type &&
2399 pol->priority < priority) {
2400 ret = pol;
2401 break;
2402 }
2403 }
2404
2405 if (ret)
2406 xfrm_pol_hold(ret);
2407
2408 read_unlock_bh(&xfrm_policy_lock);
2409
2410 return ret;
2411 }
2412
2413 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2414 {
2415 int match = 0;
2416
2417 if (t->mode == m->mode && t->id.proto == m->proto &&
2418 (m->reqid == 0 || t->reqid == m->reqid)) {
2419 switch (t->mode) {
2420 case XFRM_MODE_TUNNEL:
2421 case XFRM_MODE_BEET:
2422 if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2423 m->old_family) == 0 &&
2424 xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2425 m->old_family) == 0) {
2426 match = 1;
2427 }
2428 break;
2429 case XFRM_MODE_TRANSPORT:
2430 /* in case of transport mode, template does not store
2431 any IP addresses, hence we just compare mode and
2432 protocol */
2433 match = 1;
2434 break;
2435 default:
2436 break;
2437 }
2438 }
2439 return match;
2440 }
2441
2442 /* update endpoint address(es) of template(s) */
2443 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2444 struct xfrm_migrate *m, int num_migrate)
2445 {
2446 struct xfrm_migrate *mp;
2447 struct dst_entry *dst;
2448 int i, j, n = 0;
2449
2450 write_lock_bh(&pol->lock);
2451 if (unlikely(pol->dead)) {
2452 /* target policy has been deleted */
2453 write_unlock_bh(&pol->lock);
2454 return -ENOENT;
2455 }
2456
2457 for (i = 0; i < pol->xfrm_nr; i++) {
2458 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2459 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2460 continue;
2461 n++;
2462 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL)
2463 continue;
2464 /* update endpoints */
2465 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2466 sizeof(pol->xfrm_vec[i].id.daddr));
2467 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2468 sizeof(pol->xfrm_vec[i].saddr));
2469 pol->xfrm_vec[i].encap_family = mp->new_family;
2470 /* flush bundles */
2471 while ((dst = pol->bundles) != NULL) {
2472 pol->bundles = dst->next;
2473 dst_free(dst);
2474 }
2475 }
2476 }
2477
2478 write_unlock_bh(&pol->lock);
2479
2480 if (!n)
2481 return -ENODATA;
2482
2483 return 0;
2484 }
2485
2486 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2487 {
2488 int i, j;
2489
2490 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2491 return -EINVAL;
2492
2493 for (i = 0; i < num_migrate; i++) {
2494 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2495 m[i].old_family) == 0) &&
2496 (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2497 m[i].old_family) == 0))
2498 return -EINVAL;
2499 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2500 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2501 return -EINVAL;
2502
2503 /* check if there is any duplicated entry */
2504 for (j = i + 1; j < num_migrate; j++) {
2505 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2506 sizeof(m[i].old_daddr)) &&
2507 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2508 sizeof(m[i].old_saddr)) &&
2509 m[i].proto == m[j].proto &&
2510 m[i].mode == m[j].mode &&
2511 m[i].reqid == m[j].reqid &&
2512 m[i].old_family == m[j].old_family)
2513 return -EINVAL;
2514 }
2515 }
2516
2517 return 0;
2518 }
2519
2520 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2521 struct xfrm_migrate *m, int num_migrate)
2522 {
2523 int i, err, nx_cur = 0, nx_new = 0;
2524 struct xfrm_policy *pol = NULL;
2525 struct xfrm_state *x, *xc;
2526 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2527 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2528 struct xfrm_migrate *mp;
2529
2530 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2531 goto out;
2532
2533 /* Stage 1 - find policy */
2534 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2535 err = -ENOENT;
2536 goto out;
2537 }
2538
2539 /* Stage 2 - find and update state(s) */
2540 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2541 if ((x = xfrm_migrate_state_find(mp))) {
2542 x_cur[nx_cur] = x;
2543 nx_cur++;
2544 if ((xc = xfrm_state_migrate(x, mp))) {
2545 x_new[nx_new] = xc;
2546 nx_new++;
2547 } else {
2548 err = -ENODATA;
2549 goto restore_state;
2550 }
2551 }
2552 }
2553
2554 /* Stage 3 - update policy */
2555 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2556 goto restore_state;
2557
2558 /* Stage 4 - delete old state(s) */
2559 if (nx_cur) {
2560 xfrm_states_put(x_cur, nx_cur);
2561 xfrm_states_delete(x_cur, nx_cur);
2562 }
2563
2564 /* Stage 5 - announce */
2565 km_migrate(sel, dir, type, m, num_migrate);
2566
2567 xfrm_pol_put(pol);
2568
2569 return 0;
2570 out:
2571 return err;
2572
2573 restore_state:
2574 if (pol)
2575 xfrm_pol_put(pol);
2576 if (nx_cur)
2577 xfrm_states_put(x_cur, nx_cur);
2578 if (nx_new)
2579 xfrm_states_delete(x_new, nx_new);
2580
2581 return err;
2582 }
2583 EXPORT_SYMBOL(xfrm_migrate);
2584 #endif
2585
Linux kernel - Deliverables
This page took 0.092061 seconds and 6 git commands to generate.