?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[firewire-audio.git] / net / xfrm / xfrm_policy.c
blob0ecb16a9a8831b9c145a3db4d2e142331061d776
1 /*
2 * xfrm_policy.c
3 *
4 * Changes:
5 * Mitsuru KANDA @USAGI
6 * Kazunori MIYAZAWA @USAGI
7 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8 * IPv6 support
9 * Kazunori MIYAZAWA @USAGI
10 * YOSHIFUJI Hideaki
11 * Split up af-specific portion
12 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
13 *
14 */
15
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/kmod.h>
19 #include <linux/list.h>
20 #include <linux/spinlock.h>
21 #include <linux/workqueue.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
26 #include <linux/cache.h>
27 #include <linux/audit.h>
28 #include <net/dst.h>
29 #include <net/xfrm.h>
30 #include <net/ip.h>
31 #ifdef CONFIG_XFRM_STATISTICS
32 #include <net/snmp.h>
33 #endif
34
35 #include "xfrm_hash.h"
36
37 DEFINE_MUTEX(xfrm_cfg_mutex);
38 EXPORT_SYMBOL(xfrm_cfg_mutex);
39
40 static DEFINE_RWLOCK(xfrm_policy_lock);
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 HLIST_HEAD(xfrm_policy_gc_list);
48 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
49
50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
52 static void xfrm_init_pmtu(struct dst_entry *dst);
53
54 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
55 int dir);
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 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
92 xfrm_address_t *saddr,
93 xfrm_address_t *daddr,
94 int family)
95 {
96 struct xfrm_policy_afinfo *afinfo;
97 struct dst_entry *dst;
98
99 afinfo = xfrm_policy_get_afinfo(family);
100 if (unlikely(afinfo == NULL))
101 return ERR_PTR(-EAFNOSUPPORT);
102
103 dst = afinfo->dst_lookup(net, tos, saddr, daddr);
104
105 xfrm_policy_put_afinfo(afinfo);
106
107 return dst;
108 }
109
110 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
111 xfrm_address_t *prev_saddr,
112 xfrm_address_t *prev_daddr,
113 int family)
114 {
115 struct net *net = xs_net(x);
116 xfrm_address_t *saddr = &x->props.saddr;
117 xfrm_address_t *daddr = &x->id.daddr;
118 struct dst_entry *dst;
119
120 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
121 saddr = x->coaddr;
122 daddr = prev_daddr;
123 }
124 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
125 saddr = prev_saddr;
126 daddr = x->coaddr;
127 }
128
129 dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
130
131 if (!IS_ERR(dst)) {
132 if (prev_saddr != saddr)
133 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
134 if (prev_daddr != daddr)
135 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
136 }
137
138 return dst;
139 }
140
141 static inline unsigned long make_jiffies(long secs)
142 {
143 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
144 return MAX_SCHEDULE_TIMEOUT-1;
145 else
146 return secs*HZ;
147 }
148
149 static void xfrm_policy_timer(unsigned long data)
150 {
151 struct xfrm_policy *xp = (struct xfrm_policy*)data;
152 unsigned long now = get_seconds();
153 long next = LONG_MAX;
154 int warn = 0;
155 int dir;
156
157 read_lock(&xp->lock);
158
159 if (xp->walk.dead)
160 goto out;
161
162 dir = xfrm_policy_id2dir(xp->index);
163
164 if (xp->lft.hard_add_expires_seconds) {
165 long tmo = xp->lft.hard_add_expires_seconds +
166 xp->curlft.add_time - now;
167 if (tmo <= 0)
168 goto expired;
169 if (tmo < next)
170 next = tmo;
171 }
172 if (xp->lft.hard_use_expires_seconds) {
173 long tmo = xp->lft.hard_use_expires_seconds +
174 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
175 if (tmo <= 0)
176 goto expired;
177 if (tmo < next)
178 next = tmo;
179 }
180 if (xp->lft.soft_add_expires_seconds) {
181 long tmo = xp->lft.soft_add_expires_seconds +
182 xp->curlft.add_time - now;
183 if (tmo <= 0) {
184 warn = 1;
185 tmo = XFRM_KM_TIMEOUT;
186 }
187 if (tmo < next)
188 next = tmo;
189 }
190 if (xp->lft.soft_use_expires_seconds) {
191 long tmo = xp->lft.soft_use_expires_seconds +
192 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
193 if (tmo <= 0) {
194 warn = 1;
195 tmo = XFRM_KM_TIMEOUT;
196 }
197 if (tmo < next)
198 next = tmo;
199 }
200
201 if (warn)
202 km_policy_expired(xp, dir, 0, 0);
203 if (next != LONG_MAX &&
204 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
205 xfrm_pol_hold(xp);
206
207 out:
208 read_unlock(&xp->lock);
209 xfrm_pol_put(xp);
210 return;
211
212 expired:
213 read_unlock(&xp->lock);
214 if (!xfrm_policy_delete(xp, dir))
215 km_policy_expired(xp, dir, 1, 0);
216 xfrm_pol_put(xp);
217 }
218
219
220 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
221 * SPD calls.
222 */
223
224 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
225 {
226 struct xfrm_policy *policy;
227
228 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
229
230 if (policy) {
231 write_pnet(&policy->xp_net, net);
232 INIT_LIST_HEAD(&policy->walk.all);
233 INIT_HLIST_NODE(&policy->bydst);
234 INIT_HLIST_NODE(&policy->byidx);
235 rwlock_init(&policy->lock);
236 atomic_set(&policy->refcnt, 1);
237 setup_timer(&policy->timer, xfrm_policy_timer,
238 (unsigned long)policy);
239 }
240 return policy;
241 }
242 EXPORT_SYMBOL(xfrm_policy_alloc);
243
244 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
245
246 void xfrm_policy_destroy(struct xfrm_policy *policy)
247 {
248 BUG_ON(!policy->walk.dead);
249
250 BUG_ON(policy->bundles);
251
252 if (del_timer(&policy->timer))
253 BUG();
254
255 security_xfrm_policy_free(policy->security);
256 kfree(policy);
257 }
258 EXPORT_SYMBOL(xfrm_policy_destroy);
259
260 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
261 {
262 struct dst_entry *dst;
263
264 while ((dst = policy->bundles) != NULL) {
265 policy->bundles = dst->next;
266 dst_free(dst);
267 }
268
269 if (del_timer(&policy->timer))
270 atomic_dec(&policy->refcnt);
271
272 if (atomic_read(&policy->refcnt) > 1)
273 flow_cache_flush();
274
275 xfrm_pol_put(policy);
276 }
277
278 static void xfrm_policy_gc_task(struct work_struct *work)
279 {
280 struct xfrm_policy *policy;
281 struct hlist_node *entry, *tmp;
282 struct hlist_head gc_list;
283
284 spin_lock_bh(&xfrm_policy_gc_lock);
285 gc_list.first = xfrm_policy_gc_list.first;
286 INIT_HLIST_HEAD(&xfrm_policy_gc_list);
287 spin_unlock_bh(&xfrm_policy_gc_lock);
288
289 hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
290 xfrm_policy_gc_kill(policy);
291 }
292 static DECLARE_WORK(xfrm_policy_gc_work, xfrm_policy_gc_task);
293
294 /* Rule must be locked. Release descentant resources, announce
295 * entry dead. The rule must be unlinked from lists to the moment.
296 */
297
298 static void xfrm_policy_kill(struct xfrm_policy *policy)
299 {
300 int dead;
301
302 write_lock_bh(&policy->lock);
303 dead = policy->walk.dead;
304 policy->walk.dead = 1;
305 write_unlock_bh(&policy->lock);
306
307 if (unlikely(dead)) {
308 WARN_ON(1);
309 return;
310 }
311
312 spin_lock_bh(&xfrm_policy_gc_lock);
313 hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
314 spin_unlock_bh(&xfrm_policy_gc_lock);
315
316 schedule_work(&xfrm_policy_gc_work);
317 }
318
319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
320
321 static inline unsigned int idx_hash(struct net *net, u32 index)
322 {
323 return __idx_hash(index, net->xfrm.policy_idx_hmask);
324 }
325
326 static struct hlist_head *policy_hash_bysel(struct net *net, struct xfrm_selector *sel, unsigned short family, int dir)
327 {
328 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
329 unsigned int hash = __sel_hash(sel, family, hmask);
330
331 return (hash == hmask + 1 ?
332 &net->xfrm.policy_inexact[dir] :
333 net->xfrm.policy_bydst[dir].table + hash);
334 }
335
336 static struct hlist_head *policy_hash_direct(struct net *net, xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
337 {
338 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
339 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
340
341 return net->xfrm.policy_bydst[dir].table + hash;
342 }
343
344 static void xfrm_dst_hash_transfer(struct hlist_head *list,
345 struct hlist_head *ndsttable,
346 unsigned int nhashmask)
347 {
348 struct hlist_node *entry, *tmp, *entry0 = NULL;
349 struct xfrm_policy *pol;
350 unsigned int h0 = 0;
351
352 redo:
353 hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
354 unsigned int h;
355
356 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
357 pol->family, nhashmask);
358 if (!entry0) {
359 hlist_del(entry);
360 hlist_add_head(&pol->bydst, ndsttable+h);
361 h0 = h;
362 } else {
363 if (h != h0)
364 continue;
365 hlist_del(entry);
366 hlist_add_after(entry0, &pol->bydst);
367 }
368 entry0 = entry;
369 }
370 if (!hlist_empty(list)) {
371 entry0 = NULL;
372 goto redo;
373 }
374 }
375
376 static void xfrm_idx_hash_transfer(struct hlist_head *list,
377 struct hlist_head *nidxtable,
378 unsigned int nhashmask)
379 {
380 struct hlist_node *entry, *tmp;
381 struct xfrm_policy *pol;
382
383 hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
384 unsigned int h;
385
386 h = __idx_hash(pol->index, nhashmask);
387 hlist_add_head(&pol->byidx, nidxtable+h);
388 }
389 }
390
391 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
392 {
393 return ((old_hmask + 1) << 1) - 1;
394 }
395
396 static void xfrm_bydst_resize(struct net *net, int dir)
397 {
398 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
399 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
400 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
401 struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
402 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
403 int i;
404
405 if (!ndst)
406 return;
407
408 write_lock_bh(&xfrm_policy_lock);
409
410 for (i = hmask; i >= 0; i--)
411 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
412
413 net->xfrm.policy_bydst[dir].table = ndst;
414 net->xfrm.policy_bydst[dir].hmask = nhashmask;
415
416 write_unlock_bh(&xfrm_policy_lock);
417
418 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
419 }
420
421 static void xfrm_byidx_resize(struct net *net, int total)
422 {
423 unsigned int hmask = net->xfrm.policy_idx_hmask;
424 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
425 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
426 struct hlist_head *oidx = net->xfrm.policy_byidx;
427 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
428 int i;
429
430 if (!nidx)
431 return;
432
433 write_lock_bh(&xfrm_policy_lock);
434
435 for (i = hmask; i >= 0; i--)
436 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
437
438 net->xfrm.policy_byidx = nidx;
439 net->xfrm.policy_idx_hmask = nhashmask;
440
441 write_unlock_bh(&xfrm_policy_lock);
442
443 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
444 }
445
446 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
447 {
448 unsigned int cnt = net->xfrm.policy_count[dir];
449 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
450
451 if (total)
452 *total += cnt;
453
454 if ((hmask + 1) < xfrm_policy_hashmax &&
455 cnt > hmask)
456 return 1;
457
458 return 0;
459 }
460
461 static inline int xfrm_byidx_should_resize(struct net *net, int total)
462 {
463 unsigned int hmask = net->xfrm.policy_idx_hmask;
464
465 if ((hmask + 1) < xfrm_policy_hashmax &&
466 total > hmask)
467 return 1;
468
469 return 0;
470 }
471
472 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
473 {
474 read_lock_bh(&xfrm_policy_lock);
475 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
476 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
477 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
478 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
479 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
480 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
481 si->spdhcnt = net->xfrm.policy_idx_hmask;
482 si->spdhmcnt = xfrm_policy_hashmax;
483 read_unlock_bh(&xfrm_policy_lock);
484 }
485 EXPORT_SYMBOL(xfrm_spd_getinfo);
486
487 static DEFINE_MUTEX(hash_resize_mutex);
488 static void xfrm_hash_resize(struct work_struct *work)
489 {
490 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
491 int dir, total;
492
493 mutex_lock(&hash_resize_mutex);
494
495 total = 0;
496 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
497 if (xfrm_bydst_should_resize(net, dir, &total))
498 xfrm_bydst_resize(net, dir);
499 }
500 if (xfrm_byidx_should_resize(net, total))
501 xfrm_byidx_resize(net, total);
502
503 mutex_unlock(&hash_resize_mutex);
504 }
505
506 /* Generate new index... KAME seems to generate them ordered by cost
507 * of an absolute inpredictability of ordering of rules. This will not pass. */
508 static u32 xfrm_gen_index(struct net *net, int dir)
509 {
510 static u32 idx_generator;
511
512 for (;;) {
513 struct hlist_node *entry;
514 struct hlist_head *list;
515 struct xfrm_policy *p;
516 u32 idx;
517 int found;
518
519 idx = (idx_generator | dir);
520 idx_generator += 8;
521 if (idx == 0)
522 idx = 8;
523 list = net->xfrm.policy_byidx + idx_hash(net, idx);
524 found = 0;
525 hlist_for_each_entry(p, entry, list, byidx) {
526 if (p->index == idx) {
527 found = 1;
528 break;
529 }
530 }
531 if (!found)
532 return idx;
533 }
534 }
535
536 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
537 {
538 u32 *p1 = (u32 *) s1;
539 u32 *p2 = (u32 *) s2;
540 int len = sizeof(struct xfrm_selector) / sizeof(u32);
541 int i;
542
543 for (i = 0; i < len; i++) {
544 if (p1[i] != p2[i])
545 return 1;
546 }
547
548 return 0;
549 }
550
551 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
552 {
553 struct net *net = xp_net(policy);
554 struct xfrm_policy *pol;
555 struct xfrm_policy *delpol;
556 struct hlist_head *chain;
557 struct hlist_node *entry, *newpos;
558 struct dst_entry *gc_list;
559
560 write_lock_bh(&xfrm_policy_lock);
561 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
562 delpol = NULL;
563 newpos = NULL;
564 hlist_for_each_entry(pol, entry, chain, bydst) {
565 if (pol->type == policy->type &&
566 !selector_cmp(&pol->selector, &policy->selector) &&
567 xfrm_sec_ctx_match(pol->security, policy->security) &&
568 !WARN_ON(delpol)) {
569 if (excl) {
570 write_unlock_bh(&xfrm_policy_lock);
571 return -EEXIST;
572 }
573 delpol = pol;
574 if (policy->priority > pol->priority)
575 continue;
576 } else if (policy->priority >= pol->priority) {
577 newpos = &pol->bydst;
578 continue;
579 }
580 if (delpol)
581 break;
582 }
583 if (newpos)
584 hlist_add_after(newpos, &policy->bydst);
585 else
586 hlist_add_head(&policy->bydst, chain);
587 xfrm_pol_hold(policy);
588 net->xfrm.policy_count[dir]++;
589 atomic_inc(&flow_cache_genid);
590 if (delpol)
591 __xfrm_policy_unlink(delpol, dir);
592 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
593 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
594 policy->curlft.add_time = get_seconds();
595 policy->curlft.use_time = 0;
596 if (!mod_timer(&policy->timer, jiffies + HZ))
597 xfrm_pol_hold(policy);
598 list_add(&policy->walk.all, &net->xfrm.policy_all);
599 write_unlock_bh(&xfrm_policy_lock);
600
601 if (delpol)
602 xfrm_policy_kill(delpol);
603 else if (xfrm_bydst_should_resize(net, dir, NULL))
604 schedule_work(&net->xfrm.policy_hash_work);
605
606 read_lock_bh(&xfrm_policy_lock);
607 gc_list = NULL;
608 entry = &policy->bydst;
609 hlist_for_each_entry_continue(policy, entry, bydst) {
610 struct dst_entry *dst;
611
612 write_lock(&policy->lock);
613 dst = policy->bundles;
614 if (dst) {
615 struct dst_entry *tail = dst;
616 while (tail->next)
617 tail = tail->next;
618 tail->next = gc_list;
619 gc_list = dst;
620
621 policy->bundles = NULL;
622 }
623 write_unlock(&policy->lock);
624 }
625 read_unlock_bh(&xfrm_policy_lock);
626
627 while (gc_list) {
628 struct dst_entry *dst = gc_list;
629
630 gc_list = dst->next;
631 dst_free(dst);
632 }
633
634 return 0;
635 }
636 EXPORT_SYMBOL(xfrm_policy_insert);
637
638 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u8 type, int dir,
639 struct xfrm_selector *sel,
640 struct xfrm_sec_ctx *ctx, int delete,
641 int *err)
642 {
643 struct xfrm_policy *pol, *ret;
644 struct hlist_head *chain;
645 struct hlist_node *entry;
646
647 *err = 0;
648 write_lock_bh(&xfrm_policy_lock);
649 chain = policy_hash_bysel(net, sel, sel->family, dir);
650 ret = NULL;
651 hlist_for_each_entry(pol, entry, chain, bydst) {
652 if (pol->type == type &&
653 !selector_cmp(sel, &pol->selector) &&
654 xfrm_sec_ctx_match(ctx, pol->security)) {
655 xfrm_pol_hold(pol);
656 if (delete) {
657 *err = security_xfrm_policy_delete(
658 pol->security);
659 if (*err) {
660 write_unlock_bh(&xfrm_policy_lock);
661 return pol;
662 }
663 __xfrm_policy_unlink(pol, dir);
664 }
665 ret = pol;
666 break;
667 }
668 }
669 write_unlock_bh(&xfrm_policy_lock);
670
671 if (ret && delete) {
672 atomic_inc(&flow_cache_genid);
673 xfrm_policy_kill(ret);
674 }
675 return ret;
676 }
677 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
678
679 struct xfrm_policy *xfrm_policy_byid(struct net *net, u8 type, int dir, u32 id,
680 int delete, int *err)
681 {
682 struct xfrm_policy *pol, *ret;
683 struct hlist_head *chain;
684 struct hlist_node *entry;
685
686 *err = -ENOENT;
687 if (xfrm_policy_id2dir(id) != dir)
688 return NULL;
689
690 *err = 0;
691 write_lock_bh(&xfrm_policy_lock);
692 chain = net->xfrm.policy_byidx + idx_hash(net, id);
693 ret = NULL;
694 hlist_for_each_entry(pol, entry, chain, byidx) {
695 if (pol->type == type && pol->index == id) {
696 xfrm_pol_hold(pol);
697 if (delete) {
698 *err = security_xfrm_policy_delete(
699 pol->security);
700 if (*err) {
701 write_unlock_bh(&xfrm_policy_lock);
702 return pol;
703 }
704 __xfrm_policy_unlink(pol, dir);
705 }
706 ret = pol;
707 break;
708 }
709 }
710 write_unlock_bh(&xfrm_policy_lock);
711
712 if (ret && delete) {
713 atomic_inc(&flow_cache_genid);
714 xfrm_policy_kill(ret);
715 }
716 return ret;
717 }
718 EXPORT_SYMBOL(xfrm_policy_byid);
719
720 #ifdef CONFIG_SECURITY_NETWORK_XFRM
721 static inline int
722 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
723 {
724 int dir, err = 0;
725
726 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
727 struct xfrm_policy *pol;
728 struct hlist_node *entry;
729 int i;
730
731 hlist_for_each_entry(pol, entry,
732 &net->xfrm.policy_inexact[dir], bydst) {
733 if (pol->type != type)
734 continue;
735 err = security_xfrm_policy_delete(pol->security);
736 if (err) {
737 xfrm_audit_policy_delete(pol, 0,
738 audit_info->loginuid,
739 audit_info->sessionid,
740 audit_info->secid);
741 return err;
742 }
743 }
744 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
745 hlist_for_each_entry(pol, entry,
746 net->xfrm.policy_bydst[dir].table + i,
747 bydst) {
748 if (pol->type != type)
749 continue;
750 err = security_xfrm_policy_delete(
751 pol->security);
752 if (err) {
753 xfrm_audit_policy_delete(pol, 0,
754 audit_info->loginuid,
755 audit_info->sessionid,
756 audit_info->secid);
757 return err;
758 }
759 }
760 }
761 }
762 return err;
763 }
764 #else
765 static inline int
766 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
767 {
768 return 0;
769 }
770 #endif
771
772 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
773 {
774 int dir, err = 0;
775
776 write_lock_bh(&xfrm_policy_lock);
777
778 err = xfrm_policy_flush_secctx_check(net, type, audit_info);
779 if (err)
780 goto out;
781
782 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
783 struct xfrm_policy *pol;
784 struct hlist_node *entry;
785 int i;
786
787 again1:
788 hlist_for_each_entry(pol, entry,
789 &net->xfrm.policy_inexact[dir], bydst) {
790 if (pol->type != type)
791 continue;
792 __xfrm_policy_unlink(pol, dir);
793 write_unlock_bh(&xfrm_policy_lock);
794
795 xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
796 audit_info->sessionid,
797 audit_info->secid);
798
799 xfrm_policy_kill(pol);
800
801 write_lock_bh(&xfrm_policy_lock);
802 goto again1;
803 }
804
805 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
806 again2:
807 hlist_for_each_entry(pol, entry,
808 net->xfrm.policy_bydst[dir].table + i,
809 bydst) {
810 if (pol->type != type)
811 continue;
812 __xfrm_policy_unlink(pol, dir);
813 write_unlock_bh(&xfrm_policy_lock);
814
815 xfrm_audit_policy_delete(pol, 1,
816 audit_info->loginuid,
817 audit_info->sessionid,
818 audit_info->secid);
819 xfrm_policy_kill(pol);
820
821 write_lock_bh(&xfrm_policy_lock);
822 goto again2;
823 }
824 }
825
826 }
827 atomic_inc(&flow_cache_genid);
828 out:
829 write_unlock_bh(&xfrm_policy_lock);
830 return err;
831 }
832 EXPORT_SYMBOL(xfrm_policy_flush);
833
834 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
835 int (*func)(struct xfrm_policy *, int, int, void*),
836 void *data)
837 {
838 struct xfrm_policy *pol;
839 struct xfrm_policy_walk_entry *x;
840 int error = 0;
841
842 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
843 walk->type != XFRM_POLICY_TYPE_ANY)
844 return -EINVAL;
845
846 if (list_empty(&walk->walk.all) && walk->seq != 0)
847 return 0;
848
849 write_lock_bh(&xfrm_policy_lock);
850 if (list_empty(&walk->walk.all))
851 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
852 else
853 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
854 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
855 if (x->dead)
856 continue;
857 pol = container_of(x, struct xfrm_policy, walk);
858 if (walk->type != XFRM_POLICY_TYPE_ANY &&
859 walk->type != pol->type)
860 continue;
861 error = func(pol, xfrm_policy_id2dir(pol->index),
862 walk->seq, data);
863 if (error) {
864 list_move_tail(&walk->walk.all, &x->all);
865 goto out;
866 }
867 walk->seq++;
868 }
869 if (walk->seq == 0) {
870 error = -ENOENT;
871 goto out;
872 }
873 list_del_init(&walk->walk.all);
874 out:
875 write_unlock_bh(&xfrm_policy_lock);
876 return error;
877 }
878 EXPORT_SYMBOL(xfrm_policy_walk);
879
880 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
881 {
882 INIT_LIST_HEAD(&walk->walk.all);
883 walk->walk.dead = 1;
884 walk->type = type;
885 walk->seq = 0;
886 }
887 EXPORT_SYMBOL(xfrm_policy_walk_init);
888
889 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
890 {
891 if (list_empty(&walk->walk.all))
892 return;
893
894 write_lock_bh(&xfrm_policy_lock);
895 list_del(&walk->walk.all);
896 write_unlock_bh(&xfrm_policy_lock);
897 }
898 EXPORT_SYMBOL(xfrm_policy_walk_done);
899
900 /*
901 * Find policy to apply to this flow.
902 *
903 * Returns 0 if policy found, else an -errno.
904 */
905 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
906 u8 type, u16 family, int dir)
907 {
908 struct xfrm_selector *sel = &pol->selector;
909 int match, ret = -ESRCH;
910
911 if (pol->family != family ||
912 pol->type != type)
913 return ret;
914
915 match = xfrm_selector_match(sel, fl, family);
916 if (match)
917 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
918 dir);
919
920 return ret;
921 }
922
923 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
924 struct flowi *fl,
925 u16 family, u8 dir)
926 {
927 int err;
928 struct xfrm_policy *pol, *ret;
929 xfrm_address_t *daddr, *saddr;
930 struct hlist_node *entry;
931 struct hlist_head *chain;
932 u32 priority = ~0U;
933
934 daddr = xfrm_flowi_daddr(fl, family);
935 saddr = xfrm_flowi_saddr(fl, family);
936 if (unlikely(!daddr || !saddr))
937 return NULL;
938
939 read_lock_bh(&xfrm_policy_lock);
940 chain = policy_hash_direct(net, daddr, saddr, family, dir);
941 ret = NULL;
942 hlist_for_each_entry(pol, entry, chain, bydst) {
943 err = xfrm_policy_match(pol, fl, type, family, dir);
944 if (err) {
945 if (err == -ESRCH)
946 continue;
947 else {
948 ret = ERR_PTR(err);
949 goto fail;
950 }
951 } else {
952 ret = pol;
953 priority = ret->priority;
954 break;
955 }
956 }
957 chain = &net->xfrm.policy_inexact[dir];
958 hlist_for_each_entry(pol, entry, chain, bydst) {
959 err = xfrm_policy_match(pol, fl, type, family, dir);
960 if (err) {
961 if (err == -ESRCH)
962 continue;
963 else {
964 ret = ERR_PTR(err);
965 goto fail;
966 }
967 } else if (pol->priority < priority) {
968 ret = pol;
969 break;
970 }
971 }
972 if (ret)
973 xfrm_pol_hold(ret);
974 fail:
975 read_unlock_bh(&xfrm_policy_lock);
976
977 return ret;
978 }
979
980 static int xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
981 u8 dir, void **objp, atomic_t **obj_refp)
982 {
983 struct xfrm_policy *pol;
984 int err = 0;
985
986 #ifdef CONFIG_XFRM_SUB_POLICY
987 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
988 if (IS_ERR(pol)) {
989 err = PTR_ERR(pol);
990 pol = NULL;
991 }
992 if (pol || err)
993 goto end;
994 #endif
995 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
996 if (IS_ERR(pol)) {
997 err = PTR_ERR(pol);
998 pol = NULL;
999 }
1000 #ifdef CONFIG_XFRM_SUB_POLICY
1001 end:
1002 #endif
1003 if ((*objp = (void *) pol) != NULL)
1004 *obj_refp = &pol->refcnt;
1005 return err;
1006 }
1007
1008 static inline int policy_to_flow_dir(int dir)
1009 {
1010 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1011 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1012 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1013 return dir;
1014 switch (dir) {
1015 default:
1016 case XFRM_POLICY_IN:
1017 return FLOW_DIR_IN;
1018 case XFRM_POLICY_OUT:
1019 return FLOW_DIR_OUT;
1020 case XFRM_POLICY_FWD:
1021 return FLOW_DIR_FWD;
1022 }
1023 }
1024
1025 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1026 {
1027 struct xfrm_policy *pol;
1028
1029 read_lock_bh(&xfrm_policy_lock);
1030 if ((pol = sk->sk_policy[dir]) != NULL) {
1031 int match = xfrm_selector_match(&pol->selector, fl,
1032 sk->sk_family);
1033 int err = 0;
1034
1035 if (match) {
1036 err = security_xfrm_policy_lookup(pol->security,
1037 fl->secid,
1038 policy_to_flow_dir(dir));
1039 if (!err)
1040 xfrm_pol_hold(pol);
1041 else if (err == -ESRCH)
1042 pol = NULL;
1043 else
1044 pol = ERR_PTR(err);
1045 } else
1046 pol = NULL;
1047 }
1048 read_unlock_bh(&xfrm_policy_lock);
1049 return pol;
1050 }
1051
1052 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1053 {
1054 struct net *net = xp_net(pol);
1055 struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1056 pol->family, dir);
1057
1058 list_add(&pol->walk.all, &net->xfrm.policy_all);
1059 hlist_add_head(&pol->bydst, chain);
1060 hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1061 net->xfrm.policy_count[dir]++;
1062 xfrm_pol_hold(pol);
1063
1064 if (xfrm_bydst_should_resize(net, dir, NULL))
1065 schedule_work(&net->xfrm.policy_hash_work);
1066 }
1067
1068 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1069 int dir)
1070 {
1071 struct net *net = xp_net(pol);
1072
1073 if (hlist_unhashed(&pol->bydst))
1074 return NULL;
1075
1076 hlist_del(&pol->bydst);
1077 hlist_del(&pol->byidx);
1078 list_del(&pol->walk.all);
1079 net->xfrm.policy_count[dir]--;
1080
1081 return pol;
1082 }
1083
1084 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1085 {
1086 write_lock_bh(&xfrm_policy_lock);
1087 pol = __xfrm_policy_unlink(pol, dir);
1088 write_unlock_bh(&xfrm_policy_lock);
1089 if (pol) {
1090 if (dir < XFRM_POLICY_MAX)
1091 atomic_inc(&flow_cache_genid);
1092 xfrm_policy_kill(pol);
1093 return 0;
1094 }
1095 return -ENOENT;
1096 }
1097 EXPORT_SYMBOL(xfrm_policy_delete);
1098
1099 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1100 {
1101 struct net *net = xp_net(pol);
1102 struct xfrm_policy *old_pol;
1103
1104 #ifdef CONFIG_XFRM_SUB_POLICY
1105 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1106 return -EINVAL;
1107 #endif
1108
1109 write_lock_bh(&xfrm_policy_lock);
1110 old_pol = sk->sk_policy[dir];
1111 sk->sk_policy[dir] = pol;
1112 if (pol) {
1113 pol->curlft.add_time = get_seconds();
1114 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1115 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1116 }
1117 if (old_pol)
1118 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1119 write_unlock_bh(&xfrm_policy_lock);
1120
1121 if (old_pol) {
1122 xfrm_policy_kill(old_pol);
1123 }
1124 return 0;
1125 }
1126
1127 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1128 {
1129 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1130
1131 if (newp) {
1132 newp->selector = old->selector;
1133 if (security_xfrm_policy_clone(old->security,
1134 &newp->security)) {
1135 kfree(newp);
1136 return NULL; /* ENOMEM */
1137 }
1138 newp->lft = old->lft;
1139 newp->curlft = old->curlft;
1140 newp->action = old->action;
1141 newp->flags = old->flags;
1142 newp->xfrm_nr = old->xfrm_nr;
1143 newp->index = old->index;
1144 newp->type = old->type;
1145 memcpy(newp->xfrm_vec, old->xfrm_vec,
1146 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1147 write_lock_bh(&xfrm_policy_lock);
1148 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1149 write_unlock_bh(&xfrm_policy_lock);
1150 xfrm_pol_put(newp);
1151 }
1152 return newp;
1153 }
1154
1155 int __xfrm_sk_clone_policy(struct sock *sk)
1156 {
1157 struct xfrm_policy *p0 = sk->sk_policy[0],
1158 *p1 = sk->sk_policy[1];
1159
1160 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1161 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1162 return -ENOMEM;
1163 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1164 return -ENOMEM;
1165 return 0;
1166 }
1167
1168 static int
1169 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1170 unsigned short family)
1171 {
1172 int err;
1173 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1174
1175 if (unlikely(afinfo == NULL))
1176 return -EINVAL;
1177 err = afinfo->get_saddr(net, local, remote);
1178 xfrm_policy_put_afinfo(afinfo);
1179 return err;
1180 }
1181
1182 /* Resolve list of templates for the flow, given policy. */
1183
1184 static int
1185 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1186 struct xfrm_state **xfrm,
1187 unsigned short family)
1188 {
1189 struct net *net = xp_net(policy);
1190 int nx;
1191 int i, error;
1192 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1193 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1194 xfrm_address_t tmp;
1195
1196 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1197 struct xfrm_state *x;
1198 xfrm_address_t *remote = daddr;
1199 xfrm_address_t *local = saddr;
1200 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1201
1202 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1203 tmpl->mode == XFRM_MODE_BEET) {
1204 remote = &tmpl->id.daddr;
1205 local = &tmpl->saddr;
1206 family = tmpl->encap_family;
1207 if (xfrm_addr_any(local, family)) {
1208 error = xfrm_get_saddr(net, &tmp, remote, family);
1209 if (error)
1210 goto fail;
1211 local = &tmp;
1212 }
1213 }
1214
1215 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1216
1217 if (x && x->km.state == XFRM_STATE_VALID) {
1218 xfrm[nx++] = x;
1219 daddr = remote;
1220 saddr = local;
1221 continue;
1222 }
1223 if (x) {
1224 error = (x->km.state == XFRM_STATE_ERROR ?
1225 -EINVAL : -EAGAIN);
1226 xfrm_state_put(x);
1227 }
1228 else if (error == -ESRCH)
1229 error = -EAGAIN;
1230
1231 if (!tmpl->optional)
1232 goto fail;
1233 }
1234 return nx;
1235
1236 fail:
1237 for (nx--; nx>=0; nx--)
1238 xfrm_state_put(xfrm[nx]);
1239 return error;
1240 }
1241
1242 static int
1243 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1244 struct xfrm_state **xfrm,
1245 unsigned short family)
1246 {
1247 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1248 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1249 int cnx = 0;
1250 int error;
1251 int ret;
1252 int i;
1253
1254 for (i = 0; i < npols; i++) {
1255 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1256 error = -ENOBUFS;
1257 goto fail;
1258 }
1259
1260 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1261 if (ret < 0) {
1262 error = ret;
1263 goto fail;
1264 } else
1265 cnx += ret;
1266 }
1267
1268 /* found states are sorted for outbound processing */
1269 if (npols > 1)
1270 xfrm_state_sort(xfrm, tpp, cnx, family);
1271
1272 return cnx;
1273
1274 fail:
1275 for (cnx--; cnx>=0; cnx--)
1276 xfrm_state_put(tpp[cnx]);
1277 return error;
1278
1279 }
1280
1281 /* Check that the bundle accepts the flow and its components are
1282 * still valid.
1283 */
1284
1285 static struct dst_entry *
1286 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1287 {
1288 struct dst_entry *x;
1289 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1290 if (unlikely(afinfo == NULL))
1291 return ERR_PTR(-EINVAL);
1292 x = afinfo->find_bundle(fl, policy);
1293 xfrm_policy_put_afinfo(afinfo);
1294 return x;
1295 }
1296
1297 static inline int xfrm_get_tos(struct flowi *fl, int family)
1298 {
1299 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1300 int tos;
1301
1302 if (!afinfo)
1303 return -EINVAL;
1304
1305 tos = afinfo->get_tos(fl);
1306
1307 xfrm_policy_put_afinfo(afinfo);
1308
1309 return tos;
1310 }
1311
1312 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1313 {
1314 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1315 struct dst_ops *dst_ops;
1316 struct xfrm_dst *xdst;
1317
1318 if (!afinfo)
1319 return ERR_PTR(-EINVAL);
1320
1321 switch (family) {
1322 case AF_INET:
1323 dst_ops = &net->xfrm.xfrm4_dst_ops;
1324 break;
1325 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1326 case AF_INET6:
1327 dst_ops = &net->xfrm.xfrm6_dst_ops;
1328 break;
1329 #endif
1330 default:
1331 BUG();
1332 }
1333 xdst = dst_alloc(dst_ops) ?: ERR_PTR(-ENOBUFS);
1334
1335 xfrm_policy_put_afinfo(afinfo);
1336
1337 return xdst;
1338 }
1339
1340 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1341 int nfheader_len)
1342 {
1343 struct xfrm_policy_afinfo *afinfo =
1344 xfrm_policy_get_afinfo(dst->ops->family);
1345 int err;
1346
1347 if (!afinfo)
1348 return -EINVAL;
1349
1350 err = afinfo->init_path(path, dst, nfheader_len);
1351
1352 xfrm_policy_put_afinfo(afinfo);
1353
1354 return err;
1355 }
1356
1357 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1358 {
1359 struct xfrm_policy_afinfo *afinfo =
1360 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1361 int err;
1362
1363 if (!afinfo)
1364 return -EINVAL;
1365
1366 err = afinfo->fill_dst(xdst, dev);
1367
1368 xfrm_policy_put_afinfo(afinfo);
1369
1370 return err;
1371 }
1372
1373 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1374 * all the metrics... Shortly, bundle a bundle.
1375 */
1376
1377 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1378 struct xfrm_state **xfrm, int nx,
1379 struct flowi *fl,
1380 struct dst_entry *dst)
1381 {
1382 struct net *net = xp_net(policy);
1383 unsigned long now = jiffies;
1384 struct net_device *dev;
1385 struct dst_entry *dst_prev = NULL;
1386 struct dst_entry *dst0 = NULL;
1387 int i = 0;
1388 int err;
1389 int header_len = 0;
1390 int nfheader_len = 0;
1391 int trailer_len = 0;
1392 int tos;
1393 int family = policy->selector.family;
1394 xfrm_address_t saddr, daddr;
1395
1396 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1397
1398 tos = xfrm_get_tos(fl, family);
1399 err = tos;
1400 if (tos < 0)
1401 goto put_states;
1402
1403 dst_hold(dst);
1404
1405 for (; i < nx; i++) {
1406 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1407 struct dst_entry *dst1 = &xdst->u.dst;
1408
1409 err = PTR_ERR(xdst);
1410 if (IS_ERR(xdst)) {
1411 dst_release(dst);
1412 goto put_states;
1413 }
1414
1415 if (!dst_prev)
1416 dst0 = dst1;
1417 else {
1418 dst_prev->child = dst_clone(dst1);
1419 dst1->flags |= DST_NOHASH;
1420 }
1421
1422 xdst->route = dst;
1423 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1424
1425 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1426 family = xfrm[i]->props.family;
1427 dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1428 family);
1429 err = PTR_ERR(dst);
1430 if (IS_ERR(dst))
1431 goto put_states;
1432 } else
1433 dst_hold(dst);
1434
1435 dst1->xfrm = xfrm[i];
1436 xdst->genid = xfrm[i]->genid;
1437
1438 dst1->obsolete = -1;
1439 dst1->flags |= DST_HOST;
1440 dst1->lastuse = now;
1441
1442 dst1->input = dst_discard;
1443 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1444
1445 dst1->next = dst_prev;
1446 dst_prev = dst1;
1447
1448 header_len += xfrm[i]->props.header_len;
1449 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1450 nfheader_len += xfrm[i]->props.header_len;
1451 trailer_len += xfrm[i]->props.trailer_len;
1452 }
1453
1454 dst_prev->child = dst;
1455 dst0->path = dst;
1456
1457 err = -ENODEV;
1458 dev = dst->dev;
1459 if (!dev)
1460 goto free_dst;
1461
1462 /* Copy neighbour for reachability confirmation */
1463 dst0->neighbour = neigh_clone(dst->neighbour);
1464
1465 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1466 xfrm_init_pmtu(dst_prev);
1467
1468 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1469 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1470
1471 err = xfrm_fill_dst(xdst, dev);
1472 if (err)
1473 goto free_dst;
1474
1475 dst_prev->header_len = header_len;
1476 dst_prev->trailer_len = trailer_len;
1477 header_len -= xdst->u.dst.xfrm->props.header_len;
1478 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1479 }
1480
1481 out:
1482 return dst0;
1483
1484 put_states:
1485 for (; i < nx; i++)
1486 xfrm_state_put(xfrm[i]);
1487 free_dst:
1488 if (dst0)
1489 dst_free(dst0);
1490 dst0 = ERR_PTR(err);
1491 goto out;
1492 }
1493
1494 static int inline
1495 xfrm_dst_alloc_copy(void **target, void *src, int size)
1496 {
1497 if (!*target) {
1498 *target = kmalloc(size, GFP_ATOMIC);
1499 if (!*target)
1500 return -ENOMEM;
1501 }
1502 memcpy(*target, src, size);
1503 return 0;
1504 }
1505
1506 static int inline
1507 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1508 {
1509 #ifdef CONFIG_XFRM_SUB_POLICY
1510 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1511 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1512 sel, sizeof(*sel));
1513 #else
1514 return 0;
1515 #endif
1516 }
1517
1518 static int inline
1519 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1520 {
1521 #ifdef CONFIG_XFRM_SUB_POLICY
1522 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1523 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1524 #else
1525 return 0;
1526 #endif
1527 }
1528
1529 static int stale_bundle(struct dst_entry *dst);
1530
1531 /* Main function: finds/creates a bundle for given flow.
1532 *
1533 * At the moment we eat a raw IP route. Mostly to speed up lookups
1534 * on interfaces with disabled IPsec.
1535 */
1536 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1537 struct sock *sk, int flags)
1538 {
1539 struct xfrm_policy *policy;
1540 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1541 int npols;
1542 int pol_dead;
1543 int xfrm_nr;
1544 int pi;
1545 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1546 struct dst_entry *dst, *dst_orig = *dst_p;
1547 int nx = 0;
1548 int err;
1549 u32 genid;
1550 u16 family;
1551 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1552
1553 restart:
1554 genid = atomic_read(&flow_cache_genid);
1555 policy = NULL;
1556 for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1557 pols[pi] = NULL;
1558 npols = 0;
1559 pol_dead = 0;
1560 xfrm_nr = 0;
1561
1562 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1563 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1564 err = PTR_ERR(policy);
1565 if (IS_ERR(policy)) {
1566 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1567 goto dropdst;
1568 }
1569 }
1570
1571 if (!policy) {
1572 /* To accelerate a bit... */
1573 if ((dst_orig->flags & DST_NOXFRM) ||
1574 !net->xfrm.policy_count[XFRM_POLICY_OUT])
1575 goto nopol;
1576
1577 policy = flow_cache_lookup(net, fl, dst_orig->ops->family,
1578 dir, xfrm_policy_lookup);
1579 err = PTR_ERR(policy);
1580 if (IS_ERR(policy)) {
1581 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1582 goto dropdst;
1583 }
1584 }
1585
1586 if (!policy)
1587 goto nopol;
1588
1589 family = dst_orig->ops->family;
1590 pols[0] = policy;
1591 npols ++;
1592 xfrm_nr += pols[0]->xfrm_nr;
1593
1594 err = -ENOENT;
1595 if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1596 goto error;
1597
1598 policy->curlft.use_time = get_seconds();
1599
1600 switch (policy->action) {
1601 default:
1602 case XFRM_POLICY_BLOCK:
1603 /* Prohibit the flow */
1604 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1605 err = -EPERM;
1606 goto error;
1607
1608 case XFRM_POLICY_ALLOW:
1609 #ifndef CONFIG_XFRM_SUB_POLICY
1610 if (policy->xfrm_nr == 0) {
1611 /* Flow passes not transformed. */
1612 xfrm_pol_put(policy);
1613 return 0;
1614 }
1615 #endif
1616
1617 /* Try to find matching bundle.
1618 *
1619 * LATER: help from flow cache. It is optional, this
1620 * is required only for output policy.
1621 */
1622 dst = xfrm_find_bundle(fl, policy, family);
1623 if (IS_ERR(dst)) {
1624 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1625 err = PTR_ERR(dst);
1626 goto error;
1627 }
1628
1629 if (dst)
1630 break;
1631
1632 #ifdef CONFIG_XFRM_SUB_POLICY
1633 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1634 pols[1] = xfrm_policy_lookup_bytype(net,
1635 XFRM_POLICY_TYPE_MAIN,
1636 fl, family,
1637 XFRM_POLICY_OUT);
1638 if (pols[1]) {
1639 if (IS_ERR(pols[1])) {
1640 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1641 err = PTR_ERR(pols[1]);
1642 goto error;
1643 }
1644 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1645 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1646 err = -EPERM;
1647 goto error;
1648 }
1649 npols ++;
1650 xfrm_nr += pols[1]->xfrm_nr;
1651 }
1652 }
1653
1654 /*
1655 * Because neither flowi nor bundle information knows about
1656 * transformation template size. On more than one policy usage
1657 * we can realize whether all of them is bypass or not after
1658 * they are searched. See above not-transformed bypass
1659 * is surrounded by non-sub policy configuration, too.
1660 */
1661 if (xfrm_nr == 0) {
1662 /* Flow passes not transformed. */
1663 xfrm_pols_put(pols, npols);
1664 return 0;
1665 }
1666
1667 #endif
1668 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1669
1670 if (unlikely(nx<0)) {
1671 err = nx;
1672 if (err == -EAGAIN && net->xfrm.sysctl_larval_drop) {
1673 /* EREMOTE tells the caller to generate
1674 * a one-shot blackhole route.
1675 */
1676 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1677 xfrm_pol_put(policy);
1678 return -EREMOTE;
1679 }
1680 if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1681 DECLARE_WAITQUEUE(wait, current);
1682
1683 add_wait_queue(&net->xfrm.km_waitq, &wait);
1684 set_current_state(TASK_INTERRUPTIBLE);
1685 schedule();
1686 set_current_state(TASK_RUNNING);
1687 remove_wait_queue(&net->xfrm.km_waitq, &wait);
1688
1689 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1690
1691 if (nx == -EAGAIN && signal_pending(current)) {
1692 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1693 err = -ERESTART;
1694 goto error;
1695 }
1696 if (nx == -EAGAIN ||
1697 genid != atomic_read(&flow_cache_genid)) {
1698 xfrm_pols_put(pols, npols);
1699 goto restart;
1700 }
1701 err = nx;
1702 }
1703 if (err < 0) {
1704 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1705 goto error;
1706 }
1707 }
1708 if (nx == 0) {
1709 /* Flow passes not transformed. */
1710 xfrm_pols_put(pols, npols);
1711 return 0;
1712 }
1713
1714 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1715 err = PTR_ERR(dst);
1716 if (IS_ERR(dst)) {
1717 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1718 goto error;
1719 }
1720
1721 for (pi = 0; pi < npols; pi++) {
1722 read_lock_bh(&pols[pi]->lock);
1723 pol_dead |= pols[pi]->walk.dead;
1724 read_unlock_bh(&pols[pi]->lock);
1725 }
1726
1727 write_lock_bh(&policy->lock);
1728 if (unlikely(pol_dead || stale_bundle(dst))) {
1729 /* Wow! While we worked on resolving, this
1730 * policy has gone. Retry. It is not paranoia,
1731 * we just cannot enlist new bundle to dead object.
1732 * We can't enlist stable bundles either.
1733 */
1734 write_unlock_bh(&policy->lock);
1735 dst_free(dst);
1736
1737 if (pol_dead)
1738 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLDEAD);
1739 else
1740 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1741 err = -EHOSTUNREACH;
1742 goto error;
1743 }
1744
1745 if (npols > 1)
1746 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1747 else
1748 err = xfrm_dst_update_origin(dst, fl);
1749 if (unlikely(err)) {
1750 write_unlock_bh(&policy->lock);
1751 dst_free(dst);
1752 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1753 goto error;
1754 }
1755
1756 dst->next = policy->bundles;
1757 policy->bundles = dst;
1758 dst_hold(dst);
1759 write_unlock_bh(&policy->lock);
1760 }
1761 *dst_p = dst;
1762 dst_release(dst_orig);
1763 xfrm_pols_put(pols, npols);
1764 return 0;
1765
1766 error:
1767 xfrm_pols_put(pols, npols);
1768 dropdst:
1769 dst_release(dst_orig);
1770 *dst_p = NULL;
1771 return err;
1772
1773 nopol:
1774 err = -ENOENT;
1775 if (flags & XFRM_LOOKUP_ICMP)
1776 goto dropdst;
1777 return 0;
1778 }
1779 EXPORT_SYMBOL(__xfrm_lookup);
1780
1781 int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1782 struct sock *sk, int flags)
1783 {
1784 int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1785
1786 if (err == -EREMOTE) {
1787 dst_release(*dst_p);
1788 *dst_p = NULL;
1789 err = -EAGAIN;
1790 }
1791
1792 return err;
1793 }
1794 EXPORT_SYMBOL(xfrm_lookup);
1795
1796 static inline int
1797 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1798 {
1799 struct xfrm_state *x;
1800
1801 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1802 return 0;
1803 x = skb->sp->xvec[idx];
1804 if (!x->type->reject)
1805 return 0;
1806 return x->type->reject(x, skb, fl);
1807 }
1808
1809 /* When skb is transformed back to its "native" form, we have to
1810 * check policy restrictions. At the moment we make this in maximally
1811 * stupid way. Shame on me. :-) Of course, connected sockets must
1812 * have policy cached at them.
1813 */
1814
1815 static inline int
1816 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1817 unsigned short family)
1818 {
1819 if (xfrm_state_kern(x))
1820 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1821 return x->id.proto == tmpl->id.proto &&
1822 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1823 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1824 x->props.mode == tmpl->mode &&
1825 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1826 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1827 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1828 xfrm_state_addr_cmp(tmpl, x, family));
1829 }
1830
1831 /*
1832 * 0 or more than 0 is returned when validation is succeeded (either bypass
1833 * because of optional transport mode, or next index of the mathced secpath
1834 * state with the template.
1835 * -1 is returned when no matching template is found.
1836 * Otherwise "-2 - errored_index" is returned.
1837 */
1838 static inline int
1839 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1840 unsigned short family)
1841 {
1842 int idx = start;
1843
1844 if (tmpl->optional) {
1845 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1846 return start;
1847 } else
1848 start = -1;
1849 for (; idx < sp->len; idx++) {
1850 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1851 return ++idx;
1852 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1853 if (start == -1)
1854 start = -2-idx;
1855 break;
1856 }
1857 }
1858 return start;
1859 }
1860
1861 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1862 unsigned int family, int reverse)
1863 {
1864 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1865 int err;
1866
1867 if (unlikely(afinfo == NULL))
1868 return -EAFNOSUPPORT;
1869
1870 afinfo->decode_session(skb, fl, reverse);
1871 err = security_xfrm_decode_session(skb, &fl->secid);
1872 xfrm_policy_put_afinfo(afinfo);
1873 return err;
1874 }
1875 EXPORT_SYMBOL(__xfrm_decode_session);
1876
1877 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1878 {
1879 for (; k < sp->len; k++) {
1880 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1881 *idxp = k;
1882 return 1;
1883 }
1884 }
1885
1886 return 0;
1887 }
1888
1889 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1890 unsigned short family)
1891 {
1892 struct net *net = dev_net(skb->dev);
1893 struct xfrm_policy *pol;
1894 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1895 int npols = 0;
1896 int xfrm_nr;
1897 int pi;
1898 int reverse;
1899 struct flowi fl;
1900 u8 fl_dir;
1901 int xerr_idx = -1;
1902
1903 reverse = dir & ~XFRM_POLICY_MASK;
1904 dir &= XFRM_POLICY_MASK;
1905 fl_dir = policy_to_flow_dir(dir);
1906
1907 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1908 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
1909 return 0;
1910 }
1911
1912 nf_nat_decode_session(skb, &fl, family);
1913
1914 /* First, check used SA against their selectors. */
1915 if (skb->sp) {
1916 int i;
1917
1918 for (i=skb->sp->len-1; i>=0; i--) {
1919 struct xfrm_state *x = skb->sp->xvec[i];
1920 if (!xfrm_selector_match(&x->sel, &fl, family)) {
1921 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
1922 return 0;
1923 }
1924 }
1925 }
1926
1927 pol = NULL;
1928 if (sk && sk->sk_policy[dir]) {
1929 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1930 if (IS_ERR(pol)) {
1931 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1932 return 0;
1933 }
1934 }
1935
1936 if (!pol)
1937 pol = flow_cache_lookup(net, &fl, family, fl_dir,
1938 xfrm_policy_lookup);
1939
1940 if (IS_ERR(pol)) {
1941 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1942 return 0;
1943 }
1944
1945 if (!pol) {
1946 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1947 xfrm_secpath_reject(xerr_idx, skb, &fl);
1948 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
1949 return 0;
1950 }
1951 return 1;
1952 }
1953
1954 pol->curlft.use_time = get_seconds();
1955
1956 pols[0] = pol;
1957 npols ++;
1958 #ifdef CONFIG_XFRM_SUB_POLICY
1959 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1960 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
1961 &fl, family,
1962 XFRM_POLICY_IN);
1963 if (pols[1]) {
1964 if (IS_ERR(pols[1])) {
1965 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1966 return 0;
1967 }
1968 pols[1]->curlft.use_time = get_seconds();
1969 npols ++;
1970 }
1971 }
1972 #endif
1973
1974 if (pol->action == XFRM_POLICY_ALLOW) {
1975 struct sec_path *sp;
1976 static struct sec_path dummy;
1977 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1978 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1979 struct xfrm_tmpl **tpp = tp;
1980 int ti = 0;
1981 int i, k;
1982
1983 if ((sp = skb->sp) == NULL)
1984 sp = &dummy;
1985
1986 for (pi = 0; pi < npols; pi++) {
1987 if (pols[pi] != pol &&
1988 pols[pi]->action != XFRM_POLICY_ALLOW) {
1989 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
1990 goto reject;
1991 }
1992 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1993 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
1994 goto reject_error;
1995 }
1996 for (i = 0; i < pols[pi]->xfrm_nr; i++)
1997 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1998 }
1999 xfrm_nr = ti;
2000 if (npols > 1) {
2001 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2002 tpp = stp;
2003 }
2004
2005 /* For each tunnel xfrm, find the first matching tmpl.
2006 * For each tmpl before that, find corresponding xfrm.
2007 * Order is _important_. Later we will implement
2008 * some barriers, but at the moment barriers
2009 * are implied between each two transformations.
2010 */
2011 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2012 k = xfrm_policy_ok(tpp[i], sp, k, family);
2013 if (k < 0) {
2014 if (k < -1)
2015 /* "-2 - errored_index" returned */
2016 xerr_idx = -(2+k);
2017 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2018 goto reject;
2019 }
2020 }
2021
2022 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2023 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2024 goto reject;
2025 }
2026
2027 xfrm_pols_put(pols, npols);
2028 return 1;
2029 }
2030 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2031
2032 reject:
2033 xfrm_secpath_reject(xerr_idx, skb, &fl);
2034 reject_error:
2035 xfrm_pols_put(pols, npols);
2036 return 0;
2037 }
2038 EXPORT_SYMBOL(__xfrm_policy_check);
2039
2040 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2041 {
2042 struct net *net = dev_net(skb->dev);
2043 struct flowi fl;
2044 struct dst_entry *dst;
2045 int res;
2046
2047 if (xfrm_decode_session(skb, &fl, family) < 0) {
2048 /* XXX: we should have something like FWDHDRERROR here. */
2049 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2050 return 0;
2051 }
2052
2053 dst = skb_dst(skb);
2054
2055 res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
2056 skb_dst_set(skb, dst);
2057 return res;
2058 }
2059 EXPORT_SYMBOL(__xfrm_route_forward);
2060
2061 /* Optimize later using cookies and generation ids. */
2062
2063 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2064 {
2065 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2066 * to "-1" to force all XFRM destinations to get validated by
2067 * dst_ops->check on every use. We do this because when a
2068 * normal route referenced by an XFRM dst is obsoleted we do
2069 * not go looking around for all parent referencing XFRM dsts
2070 * so that we can invalidate them. It is just too much work.
2071 * Instead we make the checks here on every use. For example:
2072 *
2073 * XFRM dst A --> IPv4 dst X
2074 *
2075 * X is the "xdst->route" of A (X is also the "dst->path" of A
2076 * in this example). If X is marked obsolete, "A" will not
2077 * notice. That's what we are validating here via the
2078 * stale_bundle() check.
2079 *
2080 * When a policy's bundle is pruned, we dst_free() the XFRM
2081 * dst which causes it's ->obsolete field to be set to a
2082 * positive non-zero integer. If an XFRM dst has been pruned
2083 * like this, we want to force a new route lookup.
2084 */
2085 if (dst->obsolete < 0 && !stale_bundle(dst))
2086 return dst;
2087
2088 return NULL;
2089 }
2090
2091 static int stale_bundle(struct dst_entry *dst)
2092 {
2093 return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2094 }
2095
2096 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2097 {
2098 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2099 dst->dev = dev_net(dev)->loopback_dev;
2100 dev_hold(dst->dev);
2101 dev_put(dev);
2102 }
2103 }
2104 EXPORT_SYMBOL(xfrm_dst_ifdown);
2105
2106 static void xfrm_link_failure(struct sk_buff *skb)
2107 {
2108 /* Impossible. Such dst must be popped before reaches point of failure. */
2109 return;
2110 }
2111
2112 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2113 {
2114 if (dst) {
2115 if (dst->obsolete) {
2116 dst_release(dst);
2117 dst = NULL;
2118 }
2119 }
2120 return dst;
2121 }
2122
2123 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2124 {
2125 struct dst_entry *dst, **dstp;
2126
2127 write_lock(&pol->lock);
2128 dstp = &pol->bundles;
2129 while ((dst=*dstp) != NULL) {
2130 if (func(dst)) {
2131 *dstp = dst->next;
2132 dst->next = *gc_list_p;
2133 *gc_list_p = dst;
2134 } else {
2135 dstp = &dst->next;
2136 }
2137 }
2138 write_unlock(&pol->lock);
2139 }
2140
2141 static void xfrm_prune_bundles(struct net *net, int (*func)(struct dst_entry *))
2142 {
2143 struct dst_entry *gc_list = NULL;
2144 int dir;
2145
2146 read_lock_bh(&xfrm_policy_lock);
2147 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2148 struct xfrm_policy *pol;
2149 struct hlist_node *entry;
2150 struct hlist_head *table;
2151 int i;
2152
2153 hlist_for_each_entry(pol, entry,
2154 &net->xfrm.policy_inexact[dir], bydst)
2155 prune_one_bundle(pol, func, &gc_list);
2156
2157 table = net->xfrm.policy_bydst[dir].table;
2158 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2159 hlist_for_each_entry(pol, entry, table + i, bydst)
2160 prune_one_bundle(pol, func, &gc_list);
2161 }
2162 }
2163 read_unlock_bh(&xfrm_policy_lock);
2164
2165 while (gc_list) {
2166 struct dst_entry *dst = gc_list;
2167 gc_list = dst->next;
2168 dst_free(dst);
2169 }
2170 }
2171
2172 static int unused_bundle(struct dst_entry *dst)
2173 {
2174 return !atomic_read(&dst->__refcnt);
2175 }
2176
2177 static void __xfrm_garbage_collect(struct net *net)
2178 {
2179 xfrm_prune_bundles(net, unused_bundle);
2180 }
2181
2182 static int xfrm_flush_bundles(struct net *net)
2183 {
2184 xfrm_prune_bundles(net, stale_bundle);
2185 return 0;
2186 }
2187
2188 static void xfrm_init_pmtu(struct dst_entry *dst)
2189 {
2190 do {
2191 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2192 u32 pmtu, route_mtu_cached;
2193
2194 pmtu = dst_mtu(dst->child);
2195 xdst->child_mtu_cached = pmtu;
2196
2197 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2198
2199 route_mtu_cached = dst_mtu(xdst->route);
2200 xdst->route_mtu_cached = route_mtu_cached;
2201
2202 if (pmtu > route_mtu_cached)
2203 pmtu = route_mtu_cached;
2204
2205 dst->metrics[RTAX_MTU-1] = pmtu;
2206 } while ((dst = dst->next));
2207 }
2208
2209 /* Check that the bundle accepts the flow and its components are
2210 * still valid.
2211 */
2212
2213 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2214 struct flowi *fl, int family, int strict)
2215 {
2216 struct dst_entry *dst = &first->u.dst;
2217 struct xfrm_dst *last;
2218 u32 mtu;
2219
2220 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2221 (dst->dev && !netif_running(dst->dev)))
2222 return 0;
2223 #ifdef CONFIG_XFRM_SUB_POLICY
2224 if (fl) {
2225 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2226 return 0;
2227 if (first->partner &&
2228 !xfrm_selector_match(first->partner, fl, family))
2229 return 0;
2230 }
2231 #endif
2232
2233 last = NULL;
2234
2235 do {
2236 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2237
2238 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2239 return 0;
2240 if (fl && pol &&
2241 !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2242 return 0;
2243 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2244 return 0;
2245 if (xdst->genid != dst->xfrm->genid)
2246 return 0;
2247
2248 if (strict && fl &&
2249 !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2250 !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2251 return 0;
2252
2253 mtu = dst_mtu(dst->child);
2254 if (xdst->child_mtu_cached != mtu) {
2255 last = xdst;
2256 xdst->child_mtu_cached = mtu;
2257 }
2258
2259 if (!dst_check(xdst->route, xdst->route_cookie))
2260 return 0;
2261 mtu = dst_mtu(xdst->route);
2262 if (xdst->route_mtu_cached != mtu) {
2263 last = xdst;
2264 xdst->route_mtu_cached = mtu;
2265 }
2266
2267 dst = dst->child;
2268 } while (dst->xfrm);
2269
2270 if (likely(!last))
2271 return 1;
2272
2273 mtu = last->child_mtu_cached;
2274 for (;;) {
2275 dst = &last->u.dst;
2276
2277 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2278 if (mtu > last->route_mtu_cached)
2279 mtu = last->route_mtu_cached;
2280 dst->metrics[RTAX_MTU-1] = mtu;
2281
2282 if (last == first)
2283 break;
2284
2285 last = (struct xfrm_dst *)last->u.dst.next;
2286 last->child_mtu_cached = mtu;
2287 }
2288
2289 return 1;
2290 }
2291
2292 EXPORT_SYMBOL(xfrm_bundle_ok);
2293
2294 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2295 {
2296 struct net *net;
2297 int err = 0;
2298 if (unlikely(afinfo == NULL))
2299 return -EINVAL;
2300 if (unlikely(afinfo->family >= NPROTO))
2301 return -EAFNOSUPPORT;
2302 write_lock_bh(&xfrm_policy_afinfo_lock);
2303 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2304 err = -ENOBUFS;
2305 else {
2306 struct dst_ops *dst_ops = afinfo->dst_ops;
2307 if (likely(dst_ops->kmem_cachep == NULL))
2308 dst_ops->kmem_cachep = xfrm_dst_cache;
2309 if (likely(dst_ops->check == NULL))
2310 dst_ops->check = xfrm_dst_check;
2311 if (likely(dst_ops->negative_advice == NULL))
2312 dst_ops->negative_advice = xfrm_negative_advice;
2313 if (likely(dst_ops->link_failure == NULL))
2314 dst_ops->link_failure = xfrm_link_failure;
2315 if (likely(afinfo->garbage_collect == NULL))
2316 afinfo->garbage_collect = __xfrm_garbage_collect;
2317 xfrm_policy_afinfo[afinfo->family] = afinfo;
2318 }
2319 write_unlock_bh(&xfrm_policy_afinfo_lock);
2320
2321 rtnl_lock();
2322 for_each_net(net) {
2323 struct dst_ops *xfrm_dst_ops;
2324
2325 switch (afinfo->family) {
2326 case AF_INET:
2327 xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2328 break;
2329 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2330 case AF_INET6:
2331 xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2332 break;
2333 #endif
2334 default:
2335 BUG();
2336 }
2337 *xfrm_dst_ops = *afinfo->dst_ops;
2338 }
2339 rtnl_unlock();
2340
2341 return err;
2342 }
2343 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2344
2345 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2346 {
2347 int err = 0;
2348 if (unlikely(afinfo == NULL))
2349 return -EINVAL;
2350 if (unlikely(afinfo->family >= NPROTO))
2351 return -EAFNOSUPPORT;
2352 write_lock_bh(&xfrm_policy_afinfo_lock);
2353 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2354 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2355 err = -EINVAL;
2356 else {
2357 struct dst_ops *dst_ops = afinfo->dst_ops;
2358 xfrm_policy_afinfo[afinfo->family] = NULL;
2359 dst_ops->kmem_cachep = NULL;
2360 dst_ops->check = NULL;
2361 dst_ops->negative_advice = NULL;
2362 dst_ops->link_failure = NULL;
2363 afinfo->garbage_collect = NULL;
2364 }
2365 }
2366 write_unlock_bh(&xfrm_policy_afinfo_lock);
2367 return err;
2368 }
2369 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2370
2371 static void __net_init xfrm_dst_ops_init(struct net *net)
2372 {
2373 struct xfrm_policy_afinfo *afinfo;
2374
2375 read_lock_bh(&xfrm_policy_afinfo_lock);
2376 afinfo = xfrm_policy_afinfo[AF_INET];
2377 if (afinfo)
2378 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2379 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2380 afinfo = xfrm_policy_afinfo[AF_INET6];
2381 if (afinfo)
2382 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2383 #endif
2384 read_unlock_bh(&xfrm_policy_afinfo_lock);
2385 }
2386
2387 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2388 {
2389 struct xfrm_policy_afinfo *afinfo;
2390 if (unlikely(family >= NPROTO))
2391 return NULL;
2392 read_lock(&xfrm_policy_afinfo_lock);
2393 afinfo = xfrm_policy_afinfo[family];
2394 if (unlikely(!afinfo))
2395 read_unlock(&xfrm_policy_afinfo_lock);
2396 return afinfo;
2397 }
2398
2399 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2400 {
2401 read_unlock(&xfrm_policy_afinfo_lock);
2402 }
2403
2404 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2405 {
2406 struct net_device *dev = ptr;
2407
2408 switch (event) {
2409 case NETDEV_DOWN:
2410 xfrm_flush_bundles(dev_net(dev));
2411 }
2412 return NOTIFY_DONE;
2413 }
2414
2415 static struct notifier_block xfrm_dev_notifier = {
2416 .notifier_call = xfrm_dev_event,
2417 };
2418
2419 #ifdef CONFIG_XFRM_STATISTICS
2420 static int __net_init xfrm_statistics_init(struct net *net)
2421 {
2422 int rv;
2423
2424 if (snmp_mib_init((void **)net->mib.xfrm_statistics,
2425 sizeof(struct linux_xfrm_mib)) < 0)
2426 return -ENOMEM;
2427 rv = xfrm_proc_init(net);
2428 if (rv < 0)
2429 snmp_mib_free((void **)net->mib.xfrm_statistics);
2430 return rv;
2431 }
2432
2433 static void xfrm_statistics_fini(struct net *net)
2434 {
2435 xfrm_proc_fini(net);
2436 snmp_mib_free((void **)net->mib.xfrm_statistics);
2437 }
2438 #else
2439 static int __net_init xfrm_statistics_init(struct net *net)
2440 {
2441 return 0;
2442 }
2443
2444 static void xfrm_statistics_fini(struct net *net)
2445 {
2446 }
2447 #endif
2448
2449 static int __net_init xfrm_policy_init(struct net *net)
2450 {
2451 unsigned int hmask, sz;
2452 int dir;
2453
2454 if (net_eq(net, &init_net))
2455 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2456 sizeof(struct xfrm_dst),
2457 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2458 NULL);
2459
2460 hmask = 8 - 1;
2461 sz = (hmask+1) * sizeof(struct hlist_head);
2462
2463 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2464 if (!net->xfrm.policy_byidx)
2465 goto out_byidx;
2466 net->xfrm.policy_idx_hmask = hmask;
2467
2468 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2469 struct xfrm_policy_hash *htab;
2470
2471 net->xfrm.policy_count[dir] = 0;
2472 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2473
2474 htab = &net->xfrm.policy_bydst[dir];
2475 htab->table = xfrm_hash_alloc(sz);
2476 if (!htab->table)
2477 goto out_bydst;
2478 htab->hmask = hmask;
2479 }
2480
2481 INIT_LIST_HEAD(&net->xfrm.policy_all);
2482 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2483 if (net_eq(net, &init_net))
2484 register_netdevice_notifier(&xfrm_dev_notifier);
2485 return 0;
2486
2487 out_bydst:
2488 for (dir--; dir >= 0; dir--) {
2489 struct xfrm_policy_hash *htab;
2490
2491 htab = &net->xfrm.policy_bydst[dir];
2492 xfrm_hash_free(htab->table, sz);
2493 }
2494 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2495 out_byidx:
2496 return -ENOMEM;
2497 }
2498
2499 static void xfrm_policy_fini(struct net *net)
2500 {
2501 struct xfrm_audit audit_info;
2502 unsigned int sz;
2503 int dir;
2504
2505 flush_work(&net->xfrm.policy_hash_work);
2506 #ifdef CONFIG_XFRM_SUB_POLICY
2507 audit_info.loginuid = -1;
2508 audit_info.sessionid = -1;
2509 audit_info.secid = 0;
2510 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2511 #endif
2512 audit_info.loginuid = -1;
2513 audit_info.sessionid = -1;
2514 audit_info.secid = 0;
2515 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2516 flush_work(&xfrm_policy_gc_work);
2517
2518 WARN_ON(!list_empty(&net->xfrm.policy_all));
2519
2520 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2521 struct xfrm_policy_hash *htab;
2522
2523 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2524
2525 htab = &net->xfrm.policy_bydst[dir];
2526 sz = (htab->hmask + 1);
2527 WARN_ON(!hlist_empty(htab->table));
2528 xfrm_hash_free(htab->table, sz);
2529 }
2530
2531 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2532 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2533 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2534 }
2535
2536 static int __net_init xfrm_net_init(struct net *net)
2537 {
2538 int rv;
2539
2540 rv = xfrm_statistics_init(net);
2541 if (rv < 0)
2542 goto out_statistics;
2543 rv = xfrm_state_init(net);
2544 if (rv < 0)
2545 goto out_state;
2546 rv = xfrm_policy_init(net);
2547 if (rv < 0)
2548 goto out_policy;
2549 xfrm_dst_ops_init(net);
2550 rv = xfrm_sysctl_init(net);
2551 if (rv < 0)
2552 goto out_sysctl;
2553 return 0;
2554
2555 out_sysctl:
2556 xfrm_policy_fini(net);
2557 out_policy:
2558 xfrm_state_fini(net);
2559 out_state:
2560 xfrm_statistics_fini(net);
2561 out_statistics:
2562 return rv;
2563 }
2564
2565 static void __net_exit xfrm_net_exit(struct net *net)
2566 {
2567 xfrm_sysctl_fini(net);
2568 xfrm_policy_fini(net);
2569 xfrm_state_fini(net);
2570 xfrm_statistics_fini(net);
2571 }
2572
2573 static struct pernet_operations __net_initdata xfrm_net_ops = {
2574 .init = xfrm_net_init,
2575 .exit = xfrm_net_exit,
2576 };
2577
2578 void __init xfrm_init(void)
2579 {
2580 register_pernet_subsys(&xfrm_net_ops);
2581 xfrm_input_init();
2582 }
2583
2584 #ifdef CONFIG_AUDITSYSCALL
2585 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2586 struct audit_buffer *audit_buf)
2587 {
2588 struct xfrm_sec_ctx *ctx = xp->security;
2589 struct xfrm_selector *sel = &xp->selector;
2590
2591 if (ctx)
2592 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2593 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2594
2595 switch(sel->family) {
2596 case AF_INET:
2597 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2598 if (sel->prefixlen_s != 32)
2599 audit_log_format(audit_buf, " src_prefixlen=%d",
2600 sel->prefixlen_s);
2601 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2602 if (sel->prefixlen_d != 32)
2603 audit_log_format(audit_buf, " dst_prefixlen=%d",
2604 sel->prefixlen_d);
2605 break;
2606 case AF_INET6:
2607 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2608 if (sel->prefixlen_s != 128)
2609 audit_log_format(audit_buf, " src_prefixlen=%d",
2610 sel->prefixlen_s);
2611 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2612 if (sel->prefixlen_d != 128)
2613 audit_log_format(audit_buf, " dst_prefixlen=%d",
2614 sel->prefixlen_d);
2615 break;
2616 }
2617 }
2618
2619 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2620 uid_t auid, u32 sessionid, u32 secid)
2621 {
2622 struct audit_buffer *audit_buf;
2623
2624 audit_buf = xfrm_audit_start("SPD-add");
2625 if (audit_buf == NULL)
2626 return;
2627 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2628 audit_log_format(audit_buf, " res=%u", result);
2629 xfrm_audit_common_policyinfo(xp, audit_buf);
2630 audit_log_end(audit_buf);
2631 }
2632 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2633
2634 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2635 uid_t auid, u32 sessionid, u32 secid)
2636 {
2637 struct audit_buffer *audit_buf;
2638
2639 audit_buf = xfrm_audit_start("SPD-delete");
2640 if (audit_buf == NULL)
2641 return;
2642 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2643 audit_log_format(audit_buf, " res=%u", result);
2644 xfrm_audit_common_policyinfo(xp, audit_buf);
2645 audit_log_end(audit_buf);
2646 }
2647 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2648 #endif
2649
2650 #ifdef CONFIG_XFRM_MIGRATE
2651 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2652 struct xfrm_selector *sel_tgt)
2653 {
2654 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2655 if (sel_tgt->family == sel_cmp->family &&
2656 xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2657 sel_cmp->family) == 0 &&
2658 xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2659 sel_cmp->family) == 0 &&
2660 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2661 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2662 return 1;
2663 }
2664 } else {
2665 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2666 return 1;
2667 }
2668 }
2669 return 0;
2670 }
2671
2672 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2673 u8 dir, u8 type)
2674 {
2675 struct xfrm_policy *pol, *ret = NULL;
2676 struct hlist_node *entry;
2677 struct hlist_head *chain;
2678 u32 priority = ~0U;
2679
2680 read_lock_bh(&xfrm_policy_lock);
2681 chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2682 hlist_for_each_entry(pol, entry, chain, bydst) {
2683 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2684 pol->type == type) {
2685 ret = pol;
2686 priority = ret->priority;
2687 break;
2688 }
2689 }
2690 chain = &init_net.xfrm.policy_inexact[dir];
2691 hlist_for_each_entry(pol, entry, chain, bydst) {
2692 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2693 pol->type == type &&
2694 pol->priority < priority) {
2695 ret = pol;
2696 break;
2697 }
2698 }
2699
2700 if (ret)
2701 xfrm_pol_hold(ret);
2702
2703 read_unlock_bh(&xfrm_policy_lock);
2704
2705 return ret;
2706 }
2707
2708 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2709 {
2710 int match = 0;
2711
2712 if (t->mode == m->mode && t->id.proto == m->proto &&
2713 (m->reqid == 0 || t->reqid == m->reqid)) {
2714 switch (t->mode) {
2715 case XFRM_MODE_TUNNEL:
2716 case XFRM_MODE_BEET:
2717 if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2718 m->old_family) == 0 &&
2719 xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2720 m->old_family) == 0) {
2721 match = 1;
2722 }
2723 break;
2724 case XFRM_MODE_TRANSPORT:
2725 /* in case of transport mode, template does not store
2726 any IP addresses, hence we just compare mode and
2727 protocol */
2728 match = 1;
2729 break;
2730 default:
2731 break;
2732 }
2733 }
2734 return match;
2735 }
2736
2737 /* update endpoint address(es) of template(s) */
2738 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2739 struct xfrm_migrate *m, int num_migrate)
2740 {
2741 struct xfrm_migrate *mp;
2742 struct dst_entry *dst;
2743 int i, j, n = 0;
2744
2745 write_lock_bh(&pol->lock);
2746 if (unlikely(pol->walk.dead)) {
2747 /* target policy has been deleted */
2748 write_unlock_bh(&pol->lock);
2749 return -ENOENT;
2750 }
2751
2752 for (i = 0; i < pol->xfrm_nr; i++) {
2753 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2754 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2755 continue;
2756 n++;
2757 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2758 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2759 continue;
2760 /* update endpoints */
2761 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2762 sizeof(pol->xfrm_vec[i].id.daddr));
2763 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2764 sizeof(pol->xfrm_vec[i].saddr));
2765 pol->xfrm_vec[i].encap_family = mp->new_family;
2766 /* flush bundles */
2767 while ((dst = pol->bundles) != NULL) {
2768 pol->bundles = dst->next;
2769 dst_free(dst);
2770 }
2771 }
2772 }
2773
2774 write_unlock_bh(&pol->lock);
2775
2776 if (!n)
2777 return -ENODATA;
2778
2779 return 0;
2780 }
2781
2782 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2783 {
2784 int i, j;
2785
2786 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2787 return -EINVAL;
2788
2789 for (i = 0; i < num_migrate; i++) {
2790 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2791 m[i].old_family) == 0) &&
2792 (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2793 m[i].old_family) == 0))
2794 return -EINVAL;
2795 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2796 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2797 return -EINVAL;
2798
2799 /* check if there is any duplicated entry */
2800 for (j = i + 1; j < num_migrate; j++) {
2801 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2802 sizeof(m[i].old_daddr)) &&
2803 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2804 sizeof(m[i].old_saddr)) &&
2805 m[i].proto == m[j].proto &&
2806 m[i].mode == m[j].mode &&
2807 m[i].reqid == m[j].reqid &&
2808 m[i].old_family == m[j].old_family)
2809 return -EINVAL;
2810 }
2811 }
2812
2813 return 0;
2814 }
2815
2816 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2817 struct xfrm_migrate *m, int num_migrate,
2818 struct xfrm_kmaddress *k)
2819 {
2820 int i, err, nx_cur = 0, nx_new = 0;
2821 struct xfrm_policy *pol = NULL;
2822 struct xfrm_state *x, *xc;
2823 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2824 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2825 struct xfrm_migrate *mp;
2826
2827 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2828 goto out;
2829
2830 /* Stage 1 - find policy */
2831 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2832 err = -ENOENT;
2833 goto out;
2834 }
2835
2836 /* Stage 2 - find and update state(s) */
2837 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2838 if ((x = xfrm_migrate_state_find(mp))) {
2839 x_cur[nx_cur] = x;
2840 nx_cur++;
2841 if ((xc = xfrm_state_migrate(x, mp))) {
2842 x_new[nx_new] = xc;
2843 nx_new++;
2844 } else {
2845 err = -ENODATA;
2846 goto restore_state;
2847 }
2848 }
2849 }
2850
2851 /* Stage 3 - update policy */
2852 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2853 goto restore_state;
2854
2855 /* Stage 4 - delete old state(s) */
2856 if (nx_cur) {
2857 xfrm_states_put(x_cur, nx_cur);
2858 xfrm_states_delete(x_cur, nx_cur);
2859 }
2860
2861 /* Stage 5 - announce */
2862 km_migrate(sel, dir, type, m, num_migrate, k);
2863
2864 xfrm_pol_put(pol);
2865
2866 return 0;
2867 out:
2868 return err;
2869
2870 restore_state:
2871 if (pol)
2872 xfrm_pol_put(pol);
2873 if (nx_cur)
2874 xfrm_states_put(x_cur, nx_cur);
2875 if (nx_new)
2876 xfrm_states_delete(x_new, nx_new);
2877
2878 return err;
2879 }
2880 EXPORT_SYMBOL(xfrm_migrate);
2881 #endif
AMDTP streaming driver for the Linux FireWire stack (Juju)