projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: split rt_genid for ipv4 and ipv6
[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/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/flow.h>
30 #include <net/xfrm.h>
31 #include <net/ip.h>
32 #ifdef CONFIG_XFRM_STATISTICS
33 #include <net/snmp.h>
34 #endif
35
36 #include "xfrm_hash.h"
37
38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
40 #define XFRM_MAX_QUEUE_LEN 100
41
42 DEFINE_MUTEX(xfrm_cfg_mutex);
43 EXPORT_SYMBOL(xfrm_cfg_mutex);
44
45 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
46 static struct dst_entry *xfrm_policy_sk_bundles;
47 static DEFINE_RWLOCK(xfrm_policy_lock);
48
49 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
50 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
51 __read_mostly;
52
53 static struct kmem_cache *xfrm_dst_cache __read_mostly;
54
55 static void xfrm_init_pmtu(struct dst_entry *dst);
56 static int stale_bundle(struct dst_entry *dst);
57 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
58 static void xfrm_policy_queue_process(unsigned long arg);
59
60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
61 int dir);
62
63 static inline bool
64 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
65 {
66 const struct flowi4 *fl4 = &fl->u.ip4;
67
68 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
69 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
70 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
71 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
72 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
73 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
74 }
75
76 static inline bool
77 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
78 {
79 const struct flowi6 *fl6 = &fl->u.ip6;
80
81 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
82 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
83 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
84 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
85 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
86 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
87 }
88
89 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
90 unsigned short family)
91 {
92 switch (family) {
93 case AF_INET:
94 return __xfrm4_selector_match(sel, fl);
95 case AF_INET6:
96 return __xfrm6_selector_match(sel, fl);
97 }
98 return false;
99 }
100
101 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
102 {
103 struct xfrm_policy_afinfo *afinfo;
104
105 if (unlikely(family >= NPROTO))
106 return NULL;
107 rcu_read_lock();
108 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
109 if (unlikely(!afinfo))
110 rcu_read_unlock();
111 return afinfo;
112 }
113
114 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
115 {
116 rcu_read_unlock();
117 }
118
119 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
120 const xfrm_address_t *saddr,
121 const xfrm_address_t *daddr,
122 int family)
123 {
124 struct xfrm_policy_afinfo *afinfo;
125 struct dst_entry *dst;
126
127 afinfo = xfrm_policy_get_afinfo(family);
128 if (unlikely(afinfo == NULL))
129 return ERR_PTR(-EAFNOSUPPORT);
130
131 dst = afinfo->dst_lookup(net, tos, saddr, daddr);
132
133 xfrm_policy_put_afinfo(afinfo);
134
135 return dst;
136 }
137
138 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
139 xfrm_address_t *prev_saddr,
140 xfrm_address_t *prev_daddr,
141 int family)
142 {
143 struct net *net = xs_net(x);
144 xfrm_address_t *saddr = &x->props.saddr;
145 xfrm_address_t *daddr = &x->id.daddr;
146 struct dst_entry *dst;
147
148 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
149 saddr = x->coaddr;
150 daddr = prev_daddr;
151 }
152 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
153 saddr = prev_saddr;
154 daddr = x->coaddr;
155 }
156
157 dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
158
159 if (!IS_ERR(dst)) {
160 if (prev_saddr != saddr)
161 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
162 if (prev_daddr != daddr)
163 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
164 }
165
166 return dst;
167 }
168
169 static inline unsigned long make_jiffies(long secs)
170 {
171 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
172 return MAX_SCHEDULE_TIMEOUT-1;
173 else
174 return secs*HZ;
175 }
176
177 static void xfrm_policy_timer(unsigned long data)
178 {
179 struct xfrm_policy *xp = (struct xfrm_policy*)data;
180 unsigned long now = get_seconds();
181 long next = LONG_MAX;
182 int warn = 0;
183 int dir;
184
185 read_lock(&xp->lock);
186
187 if (unlikely(xp->walk.dead))
188 goto out;
189
190 dir = xfrm_policy_id2dir(xp->index);
191
192 if (xp->lft.hard_add_expires_seconds) {
193 long tmo = xp->lft.hard_add_expires_seconds +
194 xp->curlft.add_time - now;
195 if (tmo <= 0)
196 goto expired;
197 if (tmo < next)
198 next = tmo;
199 }
200 if (xp->lft.hard_use_expires_seconds) {
201 long tmo = xp->lft.hard_use_expires_seconds +
202 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
203 if (tmo <= 0)
204 goto expired;
205 if (tmo < next)
206 next = tmo;
207 }
208 if (xp->lft.soft_add_expires_seconds) {
209 long tmo = xp->lft.soft_add_expires_seconds +
210 xp->curlft.add_time - now;
211 if (tmo <= 0) {
212 warn = 1;
213 tmo = XFRM_KM_TIMEOUT;
214 }
215 if (tmo < next)
216 next = tmo;
217 }
218 if (xp->lft.soft_use_expires_seconds) {
219 long tmo = xp->lft.soft_use_expires_seconds +
220 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
221 if (tmo <= 0) {
222 warn = 1;
223 tmo = XFRM_KM_TIMEOUT;
224 }
225 if (tmo < next)
226 next = tmo;
227 }
228
229 if (warn)
230 km_policy_expired(xp, dir, 0, 0);
231 if (next != LONG_MAX &&
232 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
233 xfrm_pol_hold(xp);
234
235 out:
236 read_unlock(&xp->lock);
237 xfrm_pol_put(xp);
238 return;
239
240 expired:
241 read_unlock(&xp->lock);
242 if (!xfrm_policy_delete(xp, dir))
243 km_policy_expired(xp, dir, 1, 0);
244 xfrm_pol_put(xp);
245 }
246
247 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
248 {
249 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
250
251 if (unlikely(pol->walk.dead))
252 flo = NULL;
253 else
254 xfrm_pol_hold(pol);
255
256 return flo;
257 }
258
259 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
260 {
261 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
262
263 return !pol->walk.dead;
264 }
265
266 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
267 {
268 xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
269 }
270
271 static const struct flow_cache_ops xfrm_policy_fc_ops = {
272 .get = xfrm_policy_flo_get,
273 .check = xfrm_policy_flo_check,
274 .delete = xfrm_policy_flo_delete,
275 };
276
277 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
278 * SPD calls.
279 */
280
281 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
282 {
283 struct xfrm_policy *policy;
284
285 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
286
287 if (policy) {
288 write_pnet(&policy->xp_net, net);
289 INIT_LIST_HEAD(&policy->walk.all);
290 INIT_HLIST_NODE(&policy->bydst);
291 INIT_HLIST_NODE(&policy->byidx);
292 rwlock_init(&policy->lock);
293 atomic_set(&policy->refcnt, 1);
294 skb_queue_head_init(&policy->polq.hold_queue);
295 setup_timer(&policy->timer, xfrm_policy_timer,
296 (unsigned long)policy);
297 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
298 (unsigned long)policy);
299 policy->flo.ops = &xfrm_policy_fc_ops;
300 }
301 return policy;
302 }
303 EXPORT_SYMBOL(xfrm_policy_alloc);
304
305 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
306
307 void xfrm_policy_destroy(struct xfrm_policy *policy)
308 {
309 BUG_ON(!policy->walk.dead);
310
311 if (del_timer(&policy->timer))
312 BUG();
313
314 security_xfrm_policy_free(policy->security);
315 kfree(policy);
316 }
317 EXPORT_SYMBOL(xfrm_policy_destroy);
318
319 static void xfrm_queue_purge(struct sk_buff_head *list)
320 {
321 struct sk_buff *skb;
322
323 while ((skb = skb_dequeue(list)) != NULL) {
324 dev_put(skb->dev);
325 kfree_skb(skb);
326 }
327 }
328
329 /* Rule must be locked. Release descentant resources, announce
330 * entry dead. The rule must be unlinked from lists to the moment.
331 */
332
333 static void xfrm_policy_kill(struct xfrm_policy *policy)
334 {
335 policy->walk.dead = 1;
336
337 atomic_inc(&policy->genid);
338
339 del_timer(&policy->polq.hold_timer);
340 xfrm_queue_purge(&policy->polq.hold_queue);
341
342 if (del_timer(&policy->timer))
343 xfrm_pol_put(policy);
344
345 xfrm_pol_put(policy);
346 }
347
348 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
349
350 static inline unsigned int idx_hash(struct net *net, u32 index)
351 {
352 return __idx_hash(index, net->xfrm.policy_idx_hmask);
353 }
354
355 static struct hlist_head *policy_hash_bysel(struct net *net,
356 const struct xfrm_selector *sel,
357 unsigned short family, int dir)
358 {
359 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
360 unsigned int hash = __sel_hash(sel, family, hmask);
361
362 return (hash == hmask + 1 ?
363 &net->xfrm.policy_inexact[dir] :
364 net->xfrm.policy_bydst[dir].table + hash);
365 }
366
367 static struct hlist_head *policy_hash_direct(struct net *net,
368 const xfrm_address_t *daddr,
369 const xfrm_address_t *saddr,
370 unsigned short family, int dir)
371 {
372 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
373 unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
374
375 return net->xfrm.policy_bydst[dir].table + hash;
376 }
377
378 static void xfrm_dst_hash_transfer(struct hlist_head *list,
379 struct hlist_head *ndsttable,
380 unsigned int nhashmask)
381 {
382 struct hlist_node *tmp, *entry0 = NULL;
383 struct xfrm_policy *pol;
384 unsigned int h0 = 0;
385
386 redo:
387 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
388 unsigned int h;
389
390 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
391 pol->family, nhashmask);
392 if (!entry0) {
393 hlist_del(&pol->bydst);
394 hlist_add_head(&pol->bydst, ndsttable+h);
395 h0 = h;
396 } else {
397 if (h != h0)
398 continue;
399 hlist_del(&pol->bydst);
400 hlist_add_after(entry0, &pol->bydst);
401 }
402 entry0 = &pol->bydst;
403 }
404 if (!hlist_empty(list)) {
405 entry0 = NULL;
406 goto redo;
407 }
408 }
409
410 static void xfrm_idx_hash_transfer(struct hlist_head *list,
411 struct hlist_head *nidxtable,
412 unsigned int nhashmask)
413 {
414 struct hlist_node *tmp;
415 struct xfrm_policy *pol;
416
417 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
418 unsigned int h;
419
420 h = __idx_hash(pol->index, nhashmask);
421 hlist_add_head(&pol->byidx, nidxtable+h);
422 }
423 }
424
425 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
426 {
427 return ((old_hmask + 1) << 1) - 1;
428 }
429
430 static void xfrm_bydst_resize(struct net *net, int dir)
431 {
432 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
433 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
434 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
435 struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
436 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
437 int i;
438
439 if (!ndst)
440 return;
441
442 write_lock_bh(&xfrm_policy_lock);
443
444 for (i = hmask; i >= 0; i--)
445 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
446
447 net->xfrm.policy_bydst[dir].table = ndst;
448 net->xfrm.policy_bydst[dir].hmask = nhashmask;
449
450 write_unlock_bh(&xfrm_policy_lock);
451
452 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
453 }
454
455 static void xfrm_byidx_resize(struct net *net, int total)
456 {
457 unsigned int hmask = net->xfrm.policy_idx_hmask;
458 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
459 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
460 struct hlist_head *oidx = net->xfrm.policy_byidx;
461 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
462 int i;
463
464 if (!nidx)
465 return;
466
467 write_lock_bh(&xfrm_policy_lock);
468
469 for (i = hmask; i >= 0; i--)
470 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
471
472 net->xfrm.policy_byidx = nidx;
473 net->xfrm.policy_idx_hmask = nhashmask;
474
475 write_unlock_bh(&xfrm_policy_lock);
476
477 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
478 }
479
480 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
481 {
482 unsigned int cnt = net->xfrm.policy_count[dir];
483 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
484
485 if (total)
486 *total += cnt;
487
488 if ((hmask + 1) < xfrm_policy_hashmax &&
489 cnt > hmask)
490 return 1;
491
492 return 0;
493 }
494
495 static inline int xfrm_byidx_should_resize(struct net *net, int total)
496 {
497 unsigned int hmask = net->xfrm.policy_idx_hmask;
498
499 if ((hmask + 1) < xfrm_policy_hashmax &&
500 total > hmask)
501 return 1;
502
503 return 0;
504 }
505
506 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
507 {
508 read_lock_bh(&xfrm_policy_lock);
509 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
510 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
511 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
512 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
513 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
514 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
515 si->spdhcnt = net->xfrm.policy_idx_hmask;
516 si->spdhmcnt = xfrm_policy_hashmax;
517 read_unlock_bh(&xfrm_policy_lock);
518 }
519 EXPORT_SYMBOL(xfrm_spd_getinfo);
520
521 static DEFINE_MUTEX(hash_resize_mutex);
522 static void xfrm_hash_resize(struct work_struct *work)
523 {
524 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
525 int dir, total;
526
527 mutex_lock(&hash_resize_mutex);
528
529 total = 0;
530 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
531 if (xfrm_bydst_should_resize(net, dir, &total))
532 xfrm_bydst_resize(net, dir);
533 }
534 if (xfrm_byidx_should_resize(net, total))
535 xfrm_byidx_resize(net, total);
536
537 mutex_unlock(&hash_resize_mutex);
538 }
539
540 /* Generate new index... KAME seems to generate them ordered by cost
541 * of an absolute inpredictability of ordering of rules. This will not pass. */
542 static u32 xfrm_gen_index(struct net *net, int dir)
543 {
544 static u32 idx_generator;
545
546 for (;;) {
547 struct hlist_head *list;
548 struct xfrm_policy *p;
549 u32 idx;
550 int found;
551
552 idx = (idx_generator | dir);
553 idx_generator += 8;
554 if (idx == 0)
555 idx = 8;
556 list = net->xfrm.policy_byidx + idx_hash(net, idx);
557 found = 0;
558 hlist_for_each_entry(p, list, byidx) {
559 if (p->index == idx) {
560 found = 1;
561 break;
562 }
563 }
564 if (!found)
565 return idx;
566 }
567 }
568
569 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
570 {
571 u32 *p1 = (u32 *) s1;
572 u32 *p2 = (u32 *) s2;
573 int len = sizeof(struct xfrm_selector) / sizeof(u32);
574 int i;
575
576 for (i = 0; i < len; i++) {
577 if (p1[i] != p2[i])
578 return 1;
579 }
580
581 return 0;
582 }
583
584 static void xfrm_policy_requeue(struct xfrm_policy *old,
585 struct xfrm_policy *new)
586 {
587 struct xfrm_policy_queue *pq = &old->polq;
588 struct sk_buff_head list;
589
590 __skb_queue_head_init(&list);
591
592 spin_lock_bh(&pq->hold_queue.lock);
593 skb_queue_splice_init(&pq->hold_queue, &list);
594 del_timer(&pq->hold_timer);
595 spin_unlock_bh(&pq->hold_queue.lock);
596
597 if (skb_queue_empty(&list))
598 return;
599
600 pq = &new->polq;
601
602 spin_lock_bh(&pq->hold_queue.lock);
603 skb_queue_splice(&list, &pq->hold_queue);
604 pq->timeout = XFRM_QUEUE_TMO_MIN;
605 mod_timer(&pq->hold_timer, jiffies);
606 spin_unlock_bh(&pq->hold_queue.lock);
607 }
608
609 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
610 struct xfrm_policy *pol)
611 {
612 u32 mark = policy->mark.v & policy->mark.m;
613
614 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
615 return true;
616
617 if ((mark & pol->mark.m) == pol->mark.v &&
618 policy->priority == pol->priority)
619 return true;
620
621 return false;
622 }
623
624 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
625 {
626 struct net *net = xp_net(policy);
627 struct xfrm_policy *pol;
628 struct xfrm_policy *delpol;
629 struct hlist_head *chain;
630 struct hlist_node *newpos;
631
632 write_lock_bh(&xfrm_policy_lock);
633 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
634 delpol = NULL;
635 newpos = NULL;
636 hlist_for_each_entry(pol, chain, bydst) {
637 if (pol->type == policy->type &&
638 !selector_cmp(&pol->selector, &policy->selector) &&
639 xfrm_policy_mark_match(policy, pol) &&
640 xfrm_sec_ctx_match(pol->security, policy->security) &&
641 !WARN_ON(delpol)) {
642 if (excl) {
643 write_unlock_bh(&xfrm_policy_lock);
644 return -EEXIST;
645 }
646 delpol = pol;
647 if (policy->priority > pol->priority)
648 continue;
649 } else if (policy->priority >= pol->priority) {
650 newpos = &pol->bydst;
651 continue;
652 }
653 if (delpol)
654 break;
655 }
656 if (newpos)
657 hlist_add_after(newpos, &policy->bydst);
658 else
659 hlist_add_head(&policy->bydst, chain);
660 xfrm_pol_hold(policy);
661 net->xfrm.policy_count[dir]++;
662 atomic_inc(&flow_cache_genid);
663
664 /* After previous checking, family can either be AF_INET or AF_INET6 */
665 if (policy->family == AF_INET)
666 rt_genid_bump_ipv4(net);
667 else
668 rt_genid_bump_ipv6(net);
669
670 if (delpol) {
671 xfrm_policy_requeue(delpol, policy);
672 __xfrm_policy_unlink(delpol, dir);
673 }
674 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
675 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
676 policy->curlft.add_time = get_seconds();
677 policy->curlft.use_time = 0;
678 if (!mod_timer(&policy->timer, jiffies + HZ))
679 xfrm_pol_hold(policy);
680 list_add(&policy->walk.all, &net->xfrm.policy_all);
681 write_unlock_bh(&xfrm_policy_lock);
682
683 if (delpol)
684 xfrm_policy_kill(delpol);
685 else if (xfrm_bydst_should_resize(net, dir, NULL))
686 schedule_work(&net->xfrm.policy_hash_work);
687
688 return 0;
689 }
690 EXPORT_SYMBOL(xfrm_policy_insert);
691
692 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
693 int dir, struct xfrm_selector *sel,
694 struct xfrm_sec_ctx *ctx, int delete,
695 int *err)
696 {
697 struct xfrm_policy *pol, *ret;
698 struct hlist_head *chain;
699
700 *err = 0;
701 write_lock_bh(&xfrm_policy_lock);
702 chain = policy_hash_bysel(net, sel, sel->family, dir);
703 ret = NULL;
704 hlist_for_each_entry(pol, chain, bydst) {
705 if (pol->type == type &&
706 (mark & pol->mark.m) == pol->mark.v &&
707 !selector_cmp(sel, &pol->selector) &&
708 xfrm_sec_ctx_match(ctx, pol->security)) {
709 xfrm_pol_hold(pol);
710 if (delete) {
711 *err = security_xfrm_policy_delete(
712 pol->security);
713 if (*err) {
714 write_unlock_bh(&xfrm_policy_lock);
715 return pol;
716 }
717 __xfrm_policy_unlink(pol, dir);
718 }
719 ret = pol;
720 break;
721 }
722 }
723 write_unlock_bh(&xfrm_policy_lock);
724
725 if (ret && delete)
726 xfrm_policy_kill(ret);
727 return ret;
728 }
729 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
730
731 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
732 int dir, u32 id, int delete, int *err)
733 {
734 struct xfrm_policy *pol, *ret;
735 struct hlist_head *chain;
736
737 *err = -ENOENT;
738 if (xfrm_policy_id2dir(id) != dir)
739 return NULL;
740
741 *err = 0;
742 write_lock_bh(&xfrm_policy_lock);
743 chain = net->xfrm.policy_byidx + idx_hash(net, id);
744 ret = NULL;
745 hlist_for_each_entry(pol, chain, byidx) {
746 if (pol->type == type && pol->index == id &&
747 (mark & pol->mark.m) == pol->mark.v) {
748 xfrm_pol_hold(pol);
749 if (delete) {
750 *err = security_xfrm_policy_delete(
751 pol->security);
752 if (*err) {
753 write_unlock_bh(&xfrm_policy_lock);
754 return pol;
755 }
756 __xfrm_policy_unlink(pol, dir);
757 }
758 ret = pol;
759 break;
760 }
761 }
762 write_unlock_bh(&xfrm_policy_lock);
763
764 if (ret && delete)
765 xfrm_policy_kill(ret);
766 return ret;
767 }
768 EXPORT_SYMBOL(xfrm_policy_byid);
769
770 #ifdef CONFIG_SECURITY_NETWORK_XFRM
771 static inline int
772 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
773 {
774 int dir, err = 0;
775
776 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
777 struct xfrm_policy *pol;
778 int i;
779
780 hlist_for_each_entry(pol,
781 &net->xfrm.policy_inexact[dir], bydst) {
782 if (pol->type != type)
783 continue;
784 err = security_xfrm_policy_delete(pol->security);
785 if (err) {
786 xfrm_audit_policy_delete(pol, 0,
787 audit_info->loginuid,
788 audit_info->sessionid,
789 audit_info->secid);
790 return err;
791 }
792 }
793 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
794 hlist_for_each_entry(pol,
795 net->xfrm.policy_bydst[dir].table + i,
796 bydst) {
797 if (pol->type != type)
798 continue;
799 err = security_xfrm_policy_delete(
800 pol->security);
801 if (err) {
802 xfrm_audit_policy_delete(pol, 0,
803 audit_info->loginuid,
804 audit_info->sessionid,
805 audit_info->secid);
806 return err;
807 }
808 }
809 }
810 }
811 return err;
812 }
813 #else
814 static inline int
815 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
816 {
817 return 0;
818 }
819 #endif
820
821 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
822 {
823 int dir, err = 0, cnt = 0;
824
825 write_lock_bh(&xfrm_policy_lock);
826
827 err = xfrm_policy_flush_secctx_check(net, type, audit_info);
828 if (err)
829 goto out;
830
831 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
832 struct xfrm_policy *pol;
833 int i;
834
835 again1:
836 hlist_for_each_entry(pol,
837 &net->xfrm.policy_inexact[dir], bydst) {
838 if (pol->type != type)
839 continue;
840 __xfrm_policy_unlink(pol, dir);
841 write_unlock_bh(&xfrm_policy_lock);
842 cnt++;
843
844 xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
845 audit_info->sessionid,
846 audit_info->secid);
847
848 xfrm_policy_kill(pol);
849
850 write_lock_bh(&xfrm_policy_lock);
851 goto again1;
852 }
853
854 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
855 again2:
856 hlist_for_each_entry(pol,
857 net->xfrm.policy_bydst[dir].table + i,
858 bydst) {
859 if (pol->type != type)
860 continue;
861 __xfrm_policy_unlink(pol, dir);
862 write_unlock_bh(&xfrm_policy_lock);
863 cnt++;
864
865 xfrm_audit_policy_delete(pol, 1,
866 audit_info->loginuid,
867 audit_info->sessionid,
868 audit_info->secid);
869 xfrm_policy_kill(pol);
870
871 write_lock_bh(&xfrm_policy_lock);
872 goto again2;
873 }
874 }
875
876 }
877 if (!cnt)
878 err = -ESRCH;
879 out:
880 write_unlock_bh(&xfrm_policy_lock);
881 return err;
882 }
883 EXPORT_SYMBOL(xfrm_policy_flush);
884
885 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
886 int (*func)(struct xfrm_policy *, int, int, void*),
887 void *data)
888 {
889 struct xfrm_policy *pol;
890 struct xfrm_policy_walk_entry *x;
891 int error = 0;
892
893 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
894 walk->type != XFRM_POLICY_TYPE_ANY)
895 return -EINVAL;
896
897 if (list_empty(&walk->walk.all) && walk->seq != 0)
898 return 0;
899
900 write_lock_bh(&xfrm_policy_lock);
901 if (list_empty(&walk->walk.all))
902 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
903 else
904 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
905 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
906 if (x->dead)
907 continue;
908 pol = container_of(x, struct xfrm_policy, walk);
909 if (walk->type != XFRM_POLICY_TYPE_ANY &&
910 walk->type != pol->type)
911 continue;
912 error = func(pol, xfrm_policy_id2dir(pol->index),
913 walk->seq, data);
914 if (error) {
915 list_move_tail(&walk->walk.all, &x->all);
916 goto out;
917 }
918 walk->seq++;
919 }
920 if (walk->seq == 0) {
921 error = -ENOENT;
922 goto out;
923 }
924 list_del_init(&walk->walk.all);
925 out:
926 write_unlock_bh(&xfrm_policy_lock);
927 return error;
928 }
929 EXPORT_SYMBOL(xfrm_policy_walk);
930
931 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
932 {
933 INIT_LIST_HEAD(&walk->walk.all);
934 walk->walk.dead = 1;
935 walk->type = type;
936 walk->seq = 0;
937 }
938 EXPORT_SYMBOL(xfrm_policy_walk_init);
939
940 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
941 {
942 if (list_empty(&walk->walk.all))
943 return;
944
945 write_lock_bh(&xfrm_policy_lock);
946 list_del(&walk->walk.all);
947 write_unlock_bh(&xfrm_policy_lock);
948 }
949 EXPORT_SYMBOL(xfrm_policy_walk_done);
950
951 /*
952 * Find policy to apply to this flow.
953 *
954 * Returns 0 if policy found, else an -errno.
955 */
956 static int xfrm_policy_match(const struct xfrm_policy *pol,
957 const struct flowi *fl,
958 u8 type, u16 family, int dir)
959 {
960 const struct xfrm_selector *sel = &pol->selector;
961 int ret = -ESRCH;
962 bool match;
963
964 if (pol->family != family ||
965 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
966 pol->type != type)
967 return ret;
968
969 match = xfrm_selector_match(sel, fl, family);
970 if (match)
971 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
972 dir);
973
974 return ret;
975 }
976
977 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
978 const struct flowi *fl,
979 u16 family, u8 dir)
980 {
981 int err;
982 struct xfrm_policy *pol, *ret;
983 const xfrm_address_t *daddr, *saddr;
984 struct hlist_head *chain;
985 u32 priority = ~0U;
986
987 daddr = xfrm_flowi_daddr(fl, family);
988 saddr = xfrm_flowi_saddr(fl, family);
989 if (unlikely(!daddr || !saddr))
990 return NULL;
991
992 read_lock_bh(&xfrm_policy_lock);
993 chain = policy_hash_direct(net, daddr, saddr, family, dir);
994 ret = NULL;
995 hlist_for_each_entry(pol, chain, bydst) {
996 err = xfrm_policy_match(pol, fl, type, family, dir);
997 if (err) {
998 if (err == -ESRCH)
999 continue;
1000 else {
1001 ret = ERR_PTR(err);
1002 goto fail;
1003 }
1004 } else {
1005 ret = pol;
1006 priority = ret->priority;
1007 break;
1008 }
1009 }
1010 chain = &net->xfrm.policy_inexact[dir];
1011 hlist_for_each_entry(pol, chain, bydst) {
1012 err = xfrm_policy_match(pol, fl, type, family, dir);
1013 if (err) {
1014 if (err == -ESRCH)
1015 continue;
1016 else {
1017 ret = ERR_PTR(err);
1018 goto fail;
1019 }
1020 } else if (pol->priority < priority) {
1021 ret = pol;
1022 break;
1023 }
1024 }
1025 if (ret)
1026 xfrm_pol_hold(ret);
1027 fail:
1028 read_unlock_bh(&xfrm_policy_lock);
1029
1030 return ret;
1031 }
1032
1033 static struct xfrm_policy *
1034 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1035 {
1036 #ifdef CONFIG_XFRM_SUB_POLICY
1037 struct xfrm_policy *pol;
1038
1039 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1040 if (pol != NULL)
1041 return pol;
1042 #endif
1043 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1044 }
1045
1046 static int flow_to_policy_dir(int dir)
1047 {
1048 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1049 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1050 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1051 return dir;
1052
1053 switch (dir) {
1054 default:
1055 case FLOW_DIR_IN:
1056 return XFRM_POLICY_IN;
1057 case FLOW_DIR_OUT:
1058 return XFRM_POLICY_OUT;
1059 case FLOW_DIR_FWD:
1060 return XFRM_POLICY_FWD;
1061 }
1062 }
1063
1064 static struct flow_cache_object *
1065 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1066 u8 dir, struct flow_cache_object *old_obj, void *ctx)
1067 {
1068 struct xfrm_policy *pol;
1069
1070 if (old_obj)
1071 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1072
1073 pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1074 if (IS_ERR_OR_NULL(pol))
1075 return ERR_CAST(pol);
1076
1077 /* Resolver returns two references:
1078 * one for cache and one for caller of flow_cache_lookup() */
1079 xfrm_pol_hold(pol);
1080
1081 return &pol->flo;
1082 }
1083
1084 static inline int policy_to_flow_dir(int dir)
1085 {
1086 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1087 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1088 XFRM_POLICY_FWD == FLOW_DIR_FWD)
1089 return dir;
1090 switch (dir) {
1091 default:
1092 case XFRM_POLICY_IN:
1093 return FLOW_DIR_IN;
1094 case XFRM_POLICY_OUT:
1095 return FLOW_DIR_OUT;
1096 case XFRM_POLICY_FWD:
1097 return FLOW_DIR_FWD;
1098 }
1099 }
1100
1101 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1102 const struct flowi *fl)
1103 {
1104 struct xfrm_policy *pol;
1105
1106 read_lock_bh(&xfrm_policy_lock);
1107 if ((pol = sk->sk_policy[dir]) != NULL) {
1108 bool match = xfrm_selector_match(&pol->selector, fl,
1109 sk->sk_family);
1110 int err = 0;
1111
1112 if (match) {
1113 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1114 pol = NULL;
1115 goto out;
1116 }
1117 err = security_xfrm_policy_lookup(pol->security,
1118 fl->flowi_secid,
1119 policy_to_flow_dir(dir));
1120 if (!err)
1121 xfrm_pol_hold(pol);
1122 else if (err == -ESRCH)
1123 pol = NULL;
1124 else
1125 pol = ERR_PTR(err);
1126 } else
1127 pol = NULL;
1128 }
1129 out:
1130 read_unlock_bh(&xfrm_policy_lock);
1131 return pol;
1132 }
1133
1134 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1135 {
1136 struct net *net = xp_net(pol);
1137 struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1138 pol->family, dir);
1139
1140 list_add(&pol->walk.all, &net->xfrm.policy_all);
1141 hlist_add_head(&pol->bydst, chain);
1142 hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1143 net->xfrm.policy_count[dir]++;
1144 xfrm_pol_hold(pol);
1145
1146 if (xfrm_bydst_should_resize(net, dir, NULL))
1147 schedule_work(&net->xfrm.policy_hash_work);
1148 }
1149
1150 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1151 int dir)
1152 {
1153 struct net *net = xp_net(pol);
1154
1155 if (hlist_unhashed(&pol->bydst))
1156 return NULL;
1157
1158 hlist_del(&pol->bydst);
1159 hlist_del(&pol->byidx);
1160 list_del(&pol->walk.all);
1161 net->xfrm.policy_count[dir]--;
1162
1163 return pol;
1164 }
1165
1166 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1167 {
1168 write_lock_bh(&xfrm_policy_lock);
1169 pol = __xfrm_policy_unlink(pol, dir);
1170 write_unlock_bh(&xfrm_policy_lock);
1171 if (pol) {
1172 xfrm_policy_kill(pol);
1173 return 0;
1174 }
1175 return -ENOENT;
1176 }
1177 EXPORT_SYMBOL(xfrm_policy_delete);
1178
1179 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1180 {
1181 struct net *net = xp_net(pol);
1182 struct xfrm_policy *old_pol;
1183
1184 #ifdef CONFIG_XFRM_SUB_POLICY
1185 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1186 return -EINVAL;
1187 #endif
1188
1189 write_lock_bh(&xfrm_policy_lock);
1190 old_pol = sk->sk_policy[dir];
1191 sk->sk_policy[dir] = pol;
1192 if (pol) {
1193 pol->curlft.add_time = get_seconds();
1194 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1195 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1196 }
1197 if (old_pol) {
1198 if (pol)
1199 xfrm_policy_requeue(old_pol, pol);
1200
1201 /* Unlinking succeeds always. This is the only function
1202 * allowed to delete or replace socket policy.
1203 */
1204 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1205 }
1206 write_unlock_bh(&xfrm_policy_lock);
1207
1208 if (old_pol) {
1209 xfrm_policy_kill(old_pol);
1210 }
1211 return 0;
1212 }
1213
1214 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1215 {
1216 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1217
1218 if (newp) {
1219 newp->selector = old->selector;
1220 if (security_xfrm_policy_clone(old->security,
1221 &newp->security)) {
1222 kfree(newp);
1223 return NULL; /* ENOMEM */
1224 }
1225 newp->lft = old->lft;
1226 newp->curlft = old->curlft;
1227 newp->mark = old->mark;
1228 newp->action = old->action;
1229 newp->flags = old->flags;
1230 newp->xfrm_nr = old->xfrm_nr;
1231 newp->index = old->index;
1232 newp->type = old->type;
1233 memcpy(newp->xfrm_vec, old->xfrm_vec,
1234 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1235 write_lock_bh(&xfrm_policy_lock);
1236 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1237 write_unlock_bh(&xfrm_policy_lock);
1238 xfrm_pol_put(newp);
1239 }
1240 return newp;
1241 }
1242
1243 int __xfrm_sk_clone_policy(struct sock *sk)
1244 {
1245 struct xfrm_policy *p0 = sk->sk_policy[0],
1246 *p1 = sk->sk_policy[1];
1247
1248 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1249 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1250 return -ENOMEM;
1251 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1252 return -ENOMEM;
1253 return 0;
1254 }
1255
1256 static int
1257 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1258 unsigned short family)
1259 {
1260 int err;
1261 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1262
1263 if (unlikely(afinfo == NULL))
1264 return -EINVAL;
1265 err = afinfo->get_saddr(net, local, remote);
1266 xfrm_policy_put_afinfo(afinfo);
1267 return err;
1268 }
1269
1270 /* Resolve list of templates for the flow, given policy. */
1271
1272 static int
1273 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1274 struct xfrm_state **xfrm, unsigned short family)
1275 {
1276 struct net *net = xp_net(policy);
1277 int nx;
1278 int i, error;
1279 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1280 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1281 xfrm_address_t tmp;
1282
1283 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1284 struct xfrm_state *x;
1285 xfrm_address_t *remote = daddr;
1286 xfrm_address_t *local = saddr;
1287 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1288
1289 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1290 tmpl->mode == XFRM_MODE_BEET) {
1291 remote = &tmpl->id.daddr;
1292 local = &tmpl->saddr;
1293 if (xfrm_addr_any(local, tmpl->encap_family)) {
1294 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1295 if (error)
1296 goto fail;
1297 local = &tmp;
1298 }
1299 }
1300
1301 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1302
1303 if (x && x->km.state == XFRM_STATE_VALID) {
1304 xfrm[nx++] = x;
1305 daddr = remote;
1306 saddr = local;
1307 continue;
1308 }
1309 if (x) {
1310 error = (x->km.state == XFRM_STATE_ERROR ?
1311 -EINVAL : -EAGAIN);
1312 xfrm_state_put(x);
1313 }
1314 else if (error == -ESRCH)
1315 error = -EAGAIN;
1316
1317 if (!tmpl->optional)
1318 goto fail;
1319 }
1320 return nx;
1321
1322 fail:
1323 for (nx--; nx>=0; nx--)
1324 xfrm_state_put(xfrm[nx]);
1325 return error;
1326 }
1327
1328 static int
1329 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1330 struct xfrm_state **xfrm, unsigned short family)
1331 {
1332 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1333 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1334 int cnx = 0;
1335 int error;
1336 int ret;
1337 int i;
1338
1339 for (i = 0; i < npols; i++) {
1340 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1341 error = -ENOBUFS;
1342 goto fail;
1343 }
1344
1345 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1346 if (ret < 0) {
1347 error = ret;
1348 goto fail;
1349 } else
1350 cnx += ret;
1351 }
1352
1353 /* found states are sorted for outbound processing */
1354 if (npols > 1)
1355 xfrm_state_sort(xfrm, tpp, cnx, family);
1356
1357 return cnx;
1358
1359 fail:
1360 for (cnx--; cnx>=0; cnx--)
1361 xfrm_state_put(tpp[cnx]);
1362 return error;
1363
1364 }
1365
1366 /* Check that the bundle accepts the flow and its components are
1367 * still valid.
1368 */
1369
1370 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1371 {
1372 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1373 int tos;
1374
1375 if (!afinfo)
1376 return -EINVAL;
1377
1378 tos = afinfo->get_tos(fl);
1379
1380 xfrm_policy_put_afinfo(afinfo);
1381
1382 return tos;
1383 }
1384
1385 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1386 {
1387 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1388 struct dst_entry *dst = &xdst->u.dst;
1389
1390 if (xdst->route == NULL) {
1391 /* Dummy bundle - if it has xfrms we were not
1392 * able to build bundle as template resolution failed.
1393 * It means we need to try again resolving. */
1394 if (xdst->num_xfrms > 0)
1395 return NULL;
1396 } else if (dst->flags & DST_XFRM_QUEUE) {
1397 return NULL;
1398 } else {
1399 /* Real bundle */
1400 if (stale_bundle(dst))
1401 return NULL;
1402 }
1403
1404 dst_hold(dst);
1405 return flo;
1406 }
1407
1408 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1409 {
1410 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1411 struct dst_entry *dst = &xdst->u.dst;
1412
1413 if (!xdst->route)
1414 return 0;
1415 if (stale_bundle(dst))
1416 return 0;
1417
1418 return 1;
1419 }
1420
1421 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1422 {
1423 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1424 struct dst_entry *dst = &xdst->u.dst;
1425
1426 dst_free(dst);
1427 }
1428
1429 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1430 .get = xfrm_bundle_flo_get,
1431 .check = xfrm_bundle_flo_check,
1432 .delete = xfrm_bundle_flo_delete,
1433 };
1434
1435 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1436 {
1437 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1438 struct dst_ops *dst_ops;
1439 struct xfrm_dst *xdst;
1440
1441 if (!afinfo)
1442 return ERR_PTR(-EINVAL);
1443
1444 switch (family) {
1445 case AF_INET:
1446 dst_ops = &net->xfrm.xfrm4_dst_ops;
1447 break;
1448 #if IS_ENABLED(CONFIG_IPV6)
1449 case AF_INET6:
1450 dst_ops = &net->xfrm.xfrm6_dst_ops;
1451 break;
1452 #endif
1453 default:
1454 BUG();
1455 }
1456 xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1457
1458 if (likely(xdst)) {
1459 struct dst_entry *dst = &xdst->u.dst;
1460
1461 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1462 xdst->flo.ops = &xfrm_bundle_fc_ops;
1463 if (afinfo->init_dst)
1464 afinfo->init_dst(net, xdst);
1465 } else
1466 xdst = ERR_PTR(-ENOBUFS);
1467
1468 xfrm_policy_put_afinfo(afinfo);
1469
1470 return xdst;
1471 }
1472
1473 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1474 int nfheader_len)
1475 {
1476 struct xfrm_policy_afinfo *afinfo =
1477 xfrm_policy_get_afinfo(dst->ops->family);
1478 int err;
1479
1480 if (!afinfo)
1481 return -EINVAL;
1482
1483 err = afinfo->init_path(path, dst, nfheader_len);
1484
1485 xfrm_policy_put_afinfo(afinfo);
1486
1487 return err;
1488 }
1489
1490 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1491 const struct flowi *fl)
1492 {
1493 struct xfrm_policy_afinfo *afinfo =
1494 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1495 int err;
1496
1497 if (!afinfo)
1498 return -EINVAL;
1499
1500 err = afinfo->fill_dst(xdst, dev, fl);
1501
1502 xfrm_policy_put_afinfo(afinfo);
1503
1504 return err;
1505 }
1506
1507
1508 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1509 * all the metrics... Shortly, bundle a bundle.
1510 */
1511
1512 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1513 struct xfrm_state **xfrm, int nx,
1514 const struct flowi *fl,
1515 struct dst_entry *dst)
1516 {
1517 struct net *net = xp_net(policy);
1518 unsigned long now = jiffies;
1519 struct net_device *dev;
1520 struct xfrm_mode *inner_mode;
1521 struct dst_entry *dst_prev = NULL;
1522 struct dst_entry *dst0 = NULL;
1523 int i = 0;
1524 int err;
1525 int header_len = 0;
1526 int nfheader_len = 0;
1527 int trailer_len = 0;
1528 int tos;
1529 int family = policy->selector.family;
1530 xfrm_address_t saddr, daddr;
1531
1532 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1533
1534 tos = xfrm_get_tos(fl, family);
1535 err = tos;
1536 if (tos < 0)
1537 goto put_states;
1538
1539 dst_hold(dst);
1540
1541 for (; i < nx; i++) {
1542 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1543 struct dst_entry *dst1 = &xdst->u.dst;
1544
1545 err = PTR_ERR(xdst);
1546 if (IS_ERR(xdst)) {
1547 dst_release(dst);
1548 goto put_states;
1549 }
1550
1551 if (xfrm[i]->sel.family == AF_UNSPEC) {
1552 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1553 xfrm_af2proto(family));
1554 if (!inner_mode) {
1555 err = -EAFNOSUPPORT;
1556 dst_release(dst);
1557 goto put_states;
1558 }
1559 } else
1560 inner_mode = xfrm[i]->inner_mode;
1561
1562 if (!dst_prev)
1563 dst0 = dst1;
1564 else {
1565 dst_prev->child = dst_clone(dst1);
1566 dst1->flags |= DST_NOHASH;
1567 }
1568
1569 xdst->route = dst;
1570 dst_copy_metrics(dst1, dst);
1571
1572 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1573 family = xfrm[i]->props.family;
1574 dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1575 family);
1576 err = PTR_ERR(dst);
1577 if (IS_ERR(dst))
1578 goto put_states;
1579 } else
1580 dst_hold(dst);
1581
1582 dst1->xfrm = xfrm[i];
1583 xdst->xfrm_genid = xfrm[i]->genid;
1584
1585 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1586 dst1->flags |= DST_HOST;
1587 dst1->lastuse = now;
1588
1589 dst1->input = dst_discard;
1590 dst1->output = inner_mode->afinfo->output;
1591
1592 dst1->next = dst_prev;
1593 dst_prev = dst1;
1594
1595 header_len += xfrm[i]->props.header_len;
1596 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1597 nfheader_len += xfrm[i]->props.header_len;
1598 trailer_len += xfrm[i]->props.trailer_len;
1599 }
1600
1601 dst_prev->child = dst;
1602 dst0->path = dst;
1603
1604 err = -ENODEV;
1605 dev = dst->dev;
1606 if (!dev)
1607 goto free_dst;
1608
1609 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1610 xfrm_init_pmtu(dst_prev);
1611
1612 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1613 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1614
1615 err = xfrm_fill_dst(xdst, dev, fl);
1616 if (err)
1617 goto free_dst;
1618
1619 dst_prev->header_len = header_len;
1620 dst_prev->trailer_len = trailer_len;
1621 header_len -= xdst->u.dst.xfrm->props.header_len;
1622 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1623 }
1624
1625 out:
1626 return dst0;
1627
1628 put_states:
1629 for (; i < nx; i++)
1630 xfrm_state_put(xfrm[i]);
1631 free_dst:
1632 if (dst0)
1633 dst_free(dst0);
1634 dst0 = ERR_PTR(err);
1635 goto out;
1636 }
1637
1638 static int inline
1639 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1640 {
1641 if (!*target) {
1642 *target = kmalloc(size, GFP_ATOMIC);
1643 if (!*target)
1644 return -ENOMEM;
1645 }
1646 memcpy(*target, src, size);
1647 return 0;
1648 }
1649
1650 static int inline
1651 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1652 {
1653 #ifdef CONFIG_XFRM_SUB_POLICY
1654 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1655 return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1656 sel, sizeof(*sel));
1657 #else
1658 return 0;
1659 #endif
1660 }
1661
1662 static int inline
1663 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1664 {
1665 #ifdef CONFIG_XFRM_SUB_POLICY
1666 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1667 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1668 #else
1669 return 0;
1670 #endif
1671 }
1672
1673 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1674 struct xfrm_policy **pols,
1675 int *num_pols, int *num_xfrms)
1676 {
1677 int i;
1678
1679 if (*num_pols == 0 || !pols[0]) {
1680 *num_pols = 0;
1681 *num_xfrms = 0;
1682 return 0;
1683 }
1684 if (IS_ERR(pols[0]))
1685 return PTR_ERR(pols[0]);
1686
1687 *num_xfrms = pols[0]->xfrm_nr;
1688
1689 #ifdef CONFIG_XFRM_SUB_POLICY
1690 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1691 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1692 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1693 XFRM_POLICY_TYPE_MAIN,
1694 fl, family,
1695 XFRM_POLICY_OUT);
1696 if (pols[1]) {
1697 if (IS_ERR(pols[1])) {
1698 xfrm_pols_put(pols, *num_pols);
1699 return PTR_ERR(pols[1]);
1700 }
1701 (*num_pols) ++;
1702 (*num_xfrms) += pols[1]->xfrm_nr;
1703 }
1704 }
1705 #endif
1706 for (i = 0; i < *num_pols; i++) {
1707 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1708 *num_xfrms = -1;
1709 break;
1710 }
1711 }
1712
1713 return 0;
1714
1715 }
1716
1717 static struct xfrm_dst *
1718 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1719 const struct flowi *fl, u16 family,
1720 struct dst_entry *dst_orig)
1721 {
1722 struct net *net = xp_net(pols[0]);
1723 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1724 struct dst_entry *dst;
1725 struct xfrm_dst *xdst;
1726 int err;
1727
1728 /* Try to instantiate a bundle */
1729 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1730 if (err <= 0) {
1731 if (err != 0 && err != -EAGAIN)
1732 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1733 return ERR_PTR(err);
1734 }
1735
1736 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1737 if (IS_ERR(dst)) {
1738 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1739 return ERR_CAST(dst);
1740 }
1741
1742 xdst = (struct xfrm_dst *)dst;
1743 xdst->num_xfrms = err;
1744 if (num_pols > 1)
1745 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1746 else
1747 err = xfrm_dst_update_origin(dst, fl);
1748 if (unlikely(err)) {
1749 dst_free(dst);
1750 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1751 return ERR_PTR(err);
1752 }
1753
1754 xdst->num_pols = num_pols;
1755 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1756 xdst->policy_genid = atomic_read(&pols[0]->genid);
1757
1758 return xdst;
1759 }
1760
1761 static void xfrm_policy_queue_process(unsigned long arg)
1762 {
1763 int err = 0;
1764 struct sk_buff *skb;
1765 struct sock *sk;
1766 struct dst_entry *dst;
1767 struct net_device *dev;
1768 struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1769 struct xfrm_policy_queue *pq = &pol->polq;
1770 struct flowi fl;
1771 struct sk_buff_head list;
1772
1773 spin_lock(&pq->hold_queue.lock);
1774 skb = skb_peek(&pq->hold_queue);
1775 dst = skb_dst(skb);
1776 sk = skb->sk;
1777 xfrm_decode_session(skb, &fl, dst->ops->family);
1778 spin_unlock(&pq->hold_queue.lock);
1779
1780 dst_hold(dst->path);
1781 dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1782 sk, 0);
1783 if (IS_ERR(dst))
1784 goto purge_queue;
1785
1786 if (dst->flags & DST_XFRM_QUEUE) {
1787 dst_release(dst);
1788
1789 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1790 goto purge_queue;
1791
1792 pq->timeout = pq->timeout << 1;
1793 mod_timer(&pq->hold_timer, jiffies + pq->timeout);
1794 return;
1795 }
1796
1797 dst_release(dst);
1798
1799 __skb_queue_head_init(&list);
1800
1801 spin_lock(&pq->hold_queue.lock);
1802 pq->timeout = 0;
1803 skb_queue_splice_init(&pq->hold_queue, &list);
1804 spin_unlock(&pq->hold_queue.lock);
1805
1806 while (!skb_queue_empty(&list)) {
1807 skb = __skb_dequeue(&list);
1808
1809 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1810 dst_hold(skb_dst(skb)->path);
1811 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1812 &fl, skb->sk, 0);
1813 if (IS_ERR(dst)) {
1814 dev_put(skb->dev);
1815 kfree_skb(skb);
1816 continue;
1817 }
1818
1819 nf_reset(skb);
1820 skb_dst_drop(skb);
1821 skb_dst_set(skb, dst);
1822
1823 dev = skb->dev;
1824 err = dst_output(skb);
1825 dev_put(dev);
1826 }
1827
1828 return;
1829
1830 purge_queue:
1831 pq->timeout = 0;
1832 xfrm_queue_purge(&pq->hold_queue);
1833 }
1834
1835 static int xdst_queue_output(struct sk_buff *skb)
1836 {
1837 unsigned long sched_next;
1838 struct dst_entry *dst = skb_dst(skb);
1839 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1840 struct xfrm_policy_queue *pq = &xdst->pols[0]->polq;
1841
1842 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1843 kfree_skb(skb);
1844 return -EAGAIN;
1845 }
1846
1847 skb_dst_force(skb);
1848 dev_hold(skb->dev);
1849
1850 spin_lock_bh(&pq->hold_queue.lock);
1851
1852 if (!pq->timeout)
1853 pq->timeout = XFRM_QUEUE_TMO_MIN;
1854
1855 sched_next = jiffies + pq->timeout;
1856
1857 if (del_timer(&pq->hold_timer)) {
1858 if (time_before(pq->hold_timer.expires, sched_next))
1859 sched_next = pq->hold_timer.expires;
1860 }
1861
1862 __skb_queue_tail(&pq->hold_queue, skb);
1863 mod_timer(&pq->hold_timer, sched_next);
1864
1865 spin_unlock_bh(&pq->hold_queue.lock);
1866
1867 return 0;
1868 }
1869
1870 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1871 struct dst_entry *dst,
1872 const struct flowi *fl,
1873 int num_xfrms,
1874 u16 family)
1875 {
1876 int err;
1877 struct net_device *dev;
1878 struct dst_entry *dst1;
1879 struct xfrm_dst *xdst;
1880
1881 xdst = xfrm_alloc_dst(net, family);
1882 if (IS_ERR(xdst))
1883 return xdst;
1884
1885 if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0 ||
1886 (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP))
1887 return xdst;
1888
1889 dst1 = &xdst->u.dst;
1890 dst_hold(dst);
1891 xdst->route = dst;
1892
1893 dst_copy_metrics(dst1, dst);
1894
1895 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1896 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1897 dst1->lastuse = jiffies;
1898
1899 dst1->input = dst_discard;
1900 dst1->output = xdst_queue_output;
1901
1902 dst_hold(dst);
1903 dst1->child = dst;
1904 dst1->path = dst;
1905
1906 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1907
1908 err = -ENODEV;
1909 dev = dst->dev;
1910 if (!dev)
1911 goto free_dst;
1912
1913 err = xfrm_fill_dst(xdst, dev, fl);
1914 if (err)
1915 goto free_dst;
1916
1917 out:
1918 return xdst;
1919
1920 free_dst:
1921 dst_release(dst1);
1922 xdst = ERR_PTR(err);
1923 goto out;
1924 }
1925
1926 static struct flow_cache_object *
1927 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1928 struct flow_cache_object *oldflo, void *ctx)
1929 {
1930 struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1931 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1932 struct xfrm_dst *xdst, *new_xdst;
1933 int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1934
1935 /* Check if the policies from old bundle are usable */
1936 xdst = NULL;
1937 if (oldflo) {
1938 xdst = container_of(oldflo, struct xfrm_dst, flo);
1939 num_pols = xdst->num_pols;
1940 num_xfrms = xdst->num_xfrms;
1941 pol_dead = 0;
1942 for (i = 0; i < num_pols; i++) {
1943 pols[i] = xdst->pols[i];
1944 pol_dead |= pols[i]->walk.dead;
1945 }
1946 if (pol_dead) {
1947 dst_free(&xdst->u.dst);
1948 xdst = NULL;
1949 num_pols = 0;
1950 num_xfrms = 0;
1951 oldflo = NULL;
1952 }
1953 }
1954
1955 /* Resolve policies to use if we couldn't get them from
1956 * previous cache entry */
1957 if (xdst == NULL) {
1958 num_pols = 1;
1959 pols[0] = __xfrm_policy_lookup(net, fl, family,
1960 flow_to_policy_dir(dir));
1961 err = xfrm_expand_policies(fl, family, pols,
1962 &num_pols, &num_xfrms);
1963 if (err < 0)
1964 goto inc_error;
1965 if (num_pols == 0)
1966 return NULL;
1967 if (num_xfrms <= 0)
1968 goto make_dummy_bundle;
1969 }
1970
1971 new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1972 if (IS_ERR(new_xdst)) {
1973 err = PTR_ERR(new_xdst);
1974 if (err != -EAGAIN)
1975 goto error;
1976 if (oldflo == NULL)
1977 goto make_dummy_bundle;
1978 dst_hold(&xdst->u.dst);
1979 return oldflo;
1980 } else if (new_xdst == NULL) {
1981 num_xfrms = 0;
1982 if (oldflo == NULL)
1983 goto make_dummy_bundle;
1984 xdst->num_xfrms = 0;
1985 dst_hold(&xdst->u.dst);
1986 return oldflo;
1987 }
1988
1989 /* Kill the previous bundle */
1990 if (xdst) {
1991 /* The policies were stolen for newly generated bundle */
1992 xdst->num_pols = 0;
1993 dst_free(&xdst->u.dst);
1994 }
1995
1996 /* Flow cache does not have reference, it dst_free()'s,
1997 * but we do need to return one reference for original caller */
1998 dst_hold(&new_xdst->u.dst);
1999 return &new_xdst->flo;
2000
2001 make_dummy_bundle:
2002 /* We found policies, but there's no bundles to instantiate:
2003 * either because the policy blocks, has no transformations or
2004 * we could not build template (no xfrm_states).*/
2005 xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
2006 if (IS_ERR(xdst)) {
2007 xfrm_pols_put(pols, num_pols);
2008 return ERR_CAST(xdst);
2009 }
2010 xdst->num_pols = num_pols;
2011 xdst->num_xfrms = num_xfrms;
2012 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
2013
2014 dst_hold(&xdst->u.dst);
2015 return &xdst->flo;
2016
2017 inc_error:
2018 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2019 error:
2020 if (xdst != NULL)
2021 dst_free(&xdst->u.dst);
2022 else
2023 xfrm_pols_put(pols, num_pols);
2024 return ERR_PTR(err);
2025 }
2026
2027 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2028 struct dst_entry *dst_orig)
2029 {
2030 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2031 struct dst_entry *ret;
2032
2033 if (!afinfo) {
2034 dst_release(dst_orig);
2035 return ERR_PTR(-EINVAL);
2036 } else {
2037 ret = afinfo->blackhole_route(net, dst_orig);
2038 }
2039 xfrm_policy_put_afinfo(afinfo);
2040
2041 return ret;
2042 }
2043
2044 /* Main function: finds/creates a bundle for given flow.
2045 *
2046 * At the moment we eat a raw IP route. Mostly to speed up lookups
2047 * on interfaces with disabled IPsec.
2048 */
2049 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2050 const struct flowi *fl,
2051 struct sock *sk, int flags)
2052 {
2053 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2054 struct flow_cache_object *flo;
2055 struct xfrm_dst *xdst;
2056 struct dst_entry *dst, *route;
2057 u16 family = dst_orig->ops->family;
2058 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2059 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2060
2061 restart:
2062 dst = NULL;
2063 xdst = NULL;
2064 route = NULL;
2065
2066 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2067 num_pols = 1;
2068 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2069 err = xfrm_expand_policies(fl, family, pols,
2070 &num_pols, &num_xfrms);
2071 if (err < 0)
2072 goto dropdst;
2073
2074 if (num_pols) {
2075 if (num_xfrms <= 0) {
2076 drop_pols = num_pols;
2077 goto no_transform;
2078 }
2079
2080 xdst = xfrm_resolve_and_create_bundle(
2081 pols, num_pols, fl,
2082 family, dst_orig);
2083 if (IS_ERR(xdst)) {
2084 xfrm_pols_put(pols, num_pols);
2085 err = PTR_ERR(xdst);
2086 goto dropdst;
2087 } else if (xdst == NULL) {
2088 num_xfrms = 0;
2089 drop_pols = num_pols;
2090 goto no_transform;
2091 }
2092
2093 dst_hold(&xdst->u.dst);
2094
2095 spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2096 xdst->u.dst.next = xfrm_policy_sk_bundles;
2097 xfrm_policy_sk_bundles = &xdst->u.dst;
2098 spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2099
2100 route = xdst->route;
2101 }
2102 }
2103
2104 if (xdst == NULL) {
2105 /* To accelerate a bit... */
2106 if ((dst_orig->flags & DST_NOXFRM) ||
2107 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2108 goto nopol;
2109
2110 flo = flow_cache_lookup(net, fl, family, dir,
2111 xfrm_bundle_lookup, dst_orig);
2112 if (flo == NULL)
2113 goto nopol;
2114 if (IS_ERR(flo)) {
2115 err = PTR_ERR(flo);
2116 goto dropdst;
2117 }
2118 xdst = container_of(flo, struct xfrm_dst, flo);
2119
2120 num_pols = xdst->num_pols;
2121 num_xfrms = xdst->num_xfrms;
2122 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
2123 route = xdst->route;
2124 }
2125
2126 dst = &xdst->u.dst;
2127 if (route == NULL && num_xfrms > 0) {
2128 /* The only case when xfrm_bundle_lookup() returns a
2129 * bundle with null route, is when the template could
2130 * not be resolved. It means policies are there, but
2131 * bundle could not be created, since we don't yet
2132 * have the xfrm_state's. We need to wait for KM to
2133 * negotiate new SA's or bail out with error.*/
2134 if (net->xfrm.sysctl_larval_drop) {
2135 /* EREMOTE tells the caller to generate
2136 * a one-shot blackhole route. */
2137 dst_release(dst);
2138 xfrm_pols_put(pols, drop_pols);
2139 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2140
2141 return make_blackhole(net, family, dst_orig);
2142 }
2143 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
2144 DECLARE_WAITQUEUE(wait, current);
2145
2146 add_wait_queue(&net->xfrm.km_waitq, &wait);
2147 set_current_state(TASK_INTERRUPTIBLE);
2148 schedule();
2149 set_current_state(TASK_RUNNING);
2150 remove_wait_queue(&net->xfrm.km_waitq, &wait);
2151
2152 if (!signal_pending(current)) {
2153 dst_release(dst);
2154 goto restart;
2155 }
2156
2157 err = -ERESTART;
2158 } else
2159 err = -EAGAIN;
2160
2161 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2162 goto error;
2163 }
2164
2165 no_transform:
2166 if (num_pols == 0)
2167 goto nopol;
2168
2169 if ((flags & XFRM_LOOKUP_ICMP) &&
2170 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2171 err = -ENOENT;
2172 goto error;
2173 }
2174
2175 for (i = 0; i < num_pols; i++)
2176 pols[i]->curlft.use_time = get_seconds();
2177
2178 if (num_xfrms < 0) {
2179 /* Prohibit the flow */
2180 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2181 err = -EPERM;
2182 goto error;
2183 } else if (num_xfrms > 0) {
2184 /* Flow transformed */
2185 dst_release(dst_orig);
2186 } else {
2187 /* Flow passes untransformed */
2188 dst_release(dst);
2189 dst = dst_orig;
2190 }
2191 ok:
2192 xfrm_pols_put(pols, drop_pols);
2193 if (dst && dst->xfrm &&
2194 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2195 dst->flags |= DST_XFRM_TUNNEL;
2196 return dst;
2197
2198 nopol:
2199 if (!(flags & XFRM_LOOKUP_ICMP)) {
2200 dst = dst_orig;
2201 goto ok;
2202 }
2203 err = -ENOENT;
2204 error:
2205 dst_release(dst);
2206 dropdst:
2207 dst_release(dst_orig);
2208 xfrm_pols_put(pols, drop_pols);
2209 return ERR_PTR(err);
2210 }
2211 EXPORT_SYMBOL(xfrm_lookup);
2212
2213 static inline int
2214 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2215 {
2216 struct xfrm_state *x;
2217
2218 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2219 return 0;
2220 x = skb->sp->xvec[idx];
2221 if (!x->type->reject)
2222 return 0;
2223 return x->type->reject(x, skb, fl);
2224 }
2225
2226 /* When skb is transformed back to its "native" form, we have to
2227 * check policy restrictions. At the moment we make this in maximally
2228 * stupid way. Shame on me. :-) Of course, connected sockets must
2229 * have policy cached at them.
2230 */
2231
2232 static inline int
2233 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2234 unsigned short family)
2235 {
2236 if (xfrm_state_kern(x))
2237 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2238 return x->id.proto == tmpl->id.proto &&
2239 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2240 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2241 x->props.mode == tmpl->mode &&
2242 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2243 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2244 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2245 xfrm_state_addr_cmp(tmpl, x, family));
2246 }
2247
2248 /*
2249 * 0 or more than 0 is returned when validation is succeeded (either bypass
2250 * because of optional transport mode, or next index of the mathced secpath
2251 * state with the template.
2252 * -1 is returned when no matching template is found.
2253 * Otherwise "-2 - errored_index" is returned.
2254 */
2255 static inline int
2256 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2257 unsigned short family)
2258 {
2259 int idx = start;
2260
2261 if (tmpl->optional) {
2262 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2263 return start;
2264 } else
2265 start = -1;
2266 for (; idx < sp->len; idx++) {
2267 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2268 return ++idx;
2269 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2270 if (start == -1)
2271 start = -2-idx;
2272 break;
2273 }
2274 }
2275 return start;
2276 }
2277
2278 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2279 unsigned int family, int reverse)
2280 {
2281 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2282 int err;
2283
2284 if (unlikely(afinfo == NULL))
2285 return -EAFNOSUPPORT;
2286
2287 afinfo->decode_session(skb, fl, reverse);
2288 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2289 xfrm_policy_put_afinfo(afinfo);
2290 return err;
2291 }
2292 EXPORT_SYMBOL(__xfrm_decode_session);
2293
2294 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2295 {
2296 for (; k < sp->len; k++) {
2297 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2298 *idxp = k;
2299 return 1;
2300 }
2301 }
2302
2303 return 0;
2304 }
2305
2306 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2307 unsigned short family)
2308 {
2309 struct net *net = dev_net(skb->dev);
2310 struct xfrm_policy *pol;
2311 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2312 int npols = 0;
2313 int xfrm_nr;
2314 int pi;
2315 int reverse;
2316 struct flowi fl;
2317 u8 fl_dir;
2318 int xerr_idx = -1;
2319
2320 reverse = dir & ~XFRM_POLICY_MASK;
2321 dir &= XFRM_POLICY_MASK;
2322 fl_dir = policy_to_flow_dir(dir);
2323
2324 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2325 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2326 return 0;
2327 }
2328
2329 nf_nat_decode_session(skb, &fl, family);
2330
2331 /* First, check used SA against their selectors. */
2332 if (skb->sp) {
2333 int i;
2334
2335 for (i=skb->sp->len-1; i>=0; i--) {
2336 struct xfrm_state *x = skb->sp->xvec[i];
2337 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2338 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2339 return 0;
2340 }
2341 }
2342 }
2343
2344 pol = NULL;
2345 if (sk && sk->sk_policy[dir]) {
2346 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2347 if (IS_ERR(pol)) {
2348 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2349 return 0;
2350 }
2351 }
2352
2353 if (!pol) {
2354 struct flow_cache_object *flo;
2355
2356 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2357 xfrm_policy_lookup, NULL);
2358 if (IS_ERR_OR_NULL(flo))
2359 pol = ERR_CAST(flo);
2360 else
2361 pol = container_of(flo, struct xfrm_policy, flo);
2362 }
2363
2364 if (IS_ERR(pol)) {
2365 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2366 return 0;
2367 }
2368
2369 if (!pol) {
2370 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2371 xfrm_secpath_reject(xerr_idx, skb, &fl);
2372 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2373 return 0;
2374 }
2375 return 1;
2376 }
2377
2378 pol->curlft.use_time = get_seconds();
2379
2380 pols[0] = pol;
2381 npols ++;
2382 #ifdef CONFIG_XFRM_SUB_POLICY
2383 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2384 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2385 &fl, family,
2386 XFRM_POLICY_IN);
2387 if (pols[1]) {
2388 if (IS_ERR(pols[1])) {
2389 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2390 return 0;
2391 }
2392 pols[1]->curlft.use_time = get_seconds();
2393 npols ++;
2394 }
2395 }
2396 #endif
2397
2398 if (pol->action == XFRM_POLICY_ALLOW) {
2399 struct sec_path *sp;
2400 static struct sec_path dummy;
2401 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2402 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2403 struct xfrm_tmpl **tpp = tp;
2404 int ti = 0;
2405 int i, k;
2406
2407 if ((sp = skb->sp) == NULL)
2408 sp = &dummy;
2409
2410 for (pi = 0; pi < npols; pi++) {
2411 if (pols[pi] != pol &&
2412 pols[pi]->action != XFRM_POLICY_ALLOW) {
2413 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2414 goto reject;
2415 }
2416 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2417 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2418 goto reject_error;
2419 }
2420 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2421 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2422 }
2423 xfrm_nr = ti;
2424 if (npols > 1) {
2425 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2426 tpp = stp;
2427 }
2428
2429 /* For each tunnel xfrm, find the first matching tmpl.
2430 * For each tmpl before that, find corresponding xfrm.
2431 * Order is _important_. Later we will implement
2432 * some barriers, but at the moment barriers
2433 * are implied between each two transformations.
2434 */
2435 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2436 k = xfrm_policy_ok(tpp[i], sp, k, family);
2437 if (k < 0) {
2438 if (k < -1)
2439 /* "-2 - errored_index" returned */
2440 xerr_idx = -(2+k);
2441 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2442 goto reject;
2443 }
2444 }
2445
2446 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2447 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2448 goto reject;
2449 }
2450
2451 xfrm_pols_put(pols, npols);
2452 return 1;
2453 }
2454 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2455
2456 reject:
2457 xfrm_secpath_reject(xerr_idx, skb, &fl);
2458 reject_error:
2459 xfrm_pols_put(pols, npols);
2460 return 0;
2461 }
2462 EXPORT_SYMBOL(__xfrm_policy_check);
2463
2464 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2465 {
2466 struct net *net = dev_net(skb->dev);
2467 struct flowi fl;
2468 struct dst_entry *dst;
2469 int res = 1;
2470
2471 if (xfrm_decode_session(skb, &fl, family) < 0) {
2472 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2473 return 0;
2474 }
2475
2476 skb_dst_force(skb);
2477
2478 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2479 if (IS_ERR(dst)) {
2480 res = 0;
2481 dst = NULL;
2482 }
2483 skb_dst_set(skb, dst);
2484 return res;
2485 }
2486 EXPORT_SYMBOL(__xfrm_route_forward);
2487
2488 /* Optimize later using cookies and generation ids. */
2489
2490 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2491 {
2492 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2493 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2494 * get validated by dst_ops->check on every use. We do this
2495 * because when a normal route referenced by an XFRM dst is
2496 * obsoleted we do not go looking around for all parent
2497 * referencing XFRM dsts so that we can invalidate them. It
2498 * is just too much work. Instead we make the checks here on
2499 * every use. For example:
2500 *
2501 * XFRM dst A --> IPv4 dst X
2502 *
2503 * X is the "xdst->route" of A (X is also the "dst->path" of A
2504 * in this example). If X is marked obsolete, "A" will not
2505 * notice. That's what we are validating here via the
2506 * stale_bundle() check.
2507 *
2508 * When a policy's bundle is pruned, we dst_free() the XFRM
2509 * dst which causes it's ->obsolete field to be set to
2510 * DST_OBSOLETE_DEAD. If an XFRM dst has been pruned like
2511 * this, we want to force a new route lookup.
2512 */
2513 if (dst->obsolete < 0 && !stale_bundle(dst))
2514 return dst;
2515
2516 return NULL;
2517 }
2518
2519 static int stale_bundle(struct dst_entry *dst)
2520 {
2521 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2522 }
2523
2524 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2525 {
2526 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2527 dst->dev = dev_net(dev)->loopback_dev;
2528 dev_hold(dst->dev);
2529 dev_put(dev);
2530 }
2531 }
2532 EXPORT_SYMBOL(xfrm_dst_ifdown);
2533
2534 static void xfrm_link_failure(struct sk_buff *skb)
2535 {
2536 /* Impossible. Such dst must be popped before reaches point of failure. */
2537 }
2538
2539 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2540 {
2541 if (dst) {
2542 if (dst->obsolete) {
2543 dst_release(dst);
2544 dst = NULL;
2545 }
2546 }
2547 return dst;
2548 }
2549
2550 static void __xfrm_garbage_collect(struct net *net)
2551 {
2552 struct dst_entry *head, *next;
2553
2554 spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2555 head = xfrm_policy_sk_bundles;
2556 xfrm_policy_sk_bundles = NULL;
2557 spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2558
2559 while (head) {
2560 next = head->next;
2561 dst_free(head);
2562 head = next;
2563 }
2564 }
2565
2566 void xfrm_garbage_collect(struct net *net)
2567 {
2568 flow_cache_flush();
2569 __xfrm_garbage_collect(net);
2570 }
2571 EXPORT_SYMBOL(xfrm_garbage_collect);
2572
2573 static void xfrm_garbage_collect_deferred(struct net *net)
2574 {
2575 flow_cache_flush_deferred();
2576 __xfrm_garbage_collect(net);
2577 }
2578
2579 static void xfrm_init_pmtu(struct dst_entry *dst)
2580 {
2581 do {
2582 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2583 u32 pmtu, route_mtu_cached;
2584
2585 pmtu = dst_mtu(dst->child);
2586 xdst->child_mtu_cached = pmtu;
2587
2588 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2589
2590 route_mtu_cached = dst_mtu(xdst->route);
2591 xdst->route_mtu_cached = route_mtu_cached;
2592
2593 if (pmtu > route_mtu_cached)
2594 pmtu = route_mtu_cached;
2595
2596 dst_metric_set(dst, RTAX_MTU, pmtu);
2597 } while ((dst = dst->next));
2598 }
2599
2600 /* Check that the bundle accepts the flow and its components are
2601 * still valid.
2602 */
2603
2604 static int xfrm_bundle_ok(struct xfrm_dst *first)
2605 {
2606 struct dst_entry *dst = &first->u.dst;
2607 struct xfrm_dst *last;
2608 u32 mtu;
2609
2610 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2611 (dst->dev && !netif_running(dst->dev)))
2612 return 0;
2613
2614 if (dst->flags & DST_XFRM_QUEUE)
2615 return 1;
2616
2617 last = NULL;
2618
2619 do {
2620 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2621
2622 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2623 return 0;
2624 if (xdst->xfrm_genid != dst->xfrm->genid)
2625 return 0;
2626 if (xdst->num_pols > 0 &&
2627 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2628 return 0;
2629
2630 mtu = dst_mtu(dst->child);
2631 if (xdst->child_mtu_cached != mtu) {
2632 last = xdst;
2633 xdst->child_mtu_cached = mtu;
2634 }
2635
2636 if (!dst_check(xdst->route, xdst->route_cookie))
2637 return 0;
2638 mtu = dst_mtu(xdst->route);
2639 if (xdst->route_mtu_cached != mtu) {
2640 last = xdst;
2641 xdst->route_mtu_cached = mtu;
2642 }
2643
2644 dst = dst->child;
2645 } while (dst->xfrm);
2646
2647 if (likely(!last))
2648 return 1;
2649
2650 mtu = last->child_mtu_cached;
2651 for (;;) {
2652 dst = &last->u.dst;
2653
2654 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2655 if (mtu > last->route_mtu_cached)
2656 mtu = last->route_mtu_cached;
2657 dst_metric_set(dst, RTAX_MTU, mtu);
2658
2659 if (last == first)
2660 break;
2661
2662 last = (struct xfrm_dst *)last->u.dst.next;
2663 last->child_mtu_cached = mtu;
2664 }
2665
2666 return 1;
2667 }
2668
2669 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2670 {
2671 return dst_metric_advmss(dst->path);
2672 }
2673
2674 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2675 {
2676 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2677
2678 return mtu ? : dst_mtu(dst->path);
2679 }
2680
2681 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2682 struct sk_buff *skb,
2683 const void *daddr)
2684 {
2685 return dst->path->ops->neigh_lookup(dst, skb, daddr);
2686 }
2687
2688 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2689 {
2690 struct net *net;
2691 int err = 0;
2692 if (unlikely(afinfo == NULL))
2693 return -EINVAL;
2694 if (unlikely(afinfo->family >= NPROTO))
2695 return -EAFNOSUPPORT;
2696 spin_lock(&xfrm_policy_afinfo_lock);
2697 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2698 err = -ENOBUFS;
2699 else {
2700 struct dst_ops *dst_ops = afinfo->dst_ops;
2701 if (likely(dst_ops->kmem_cachep == NULL))
2702 dst_ops->kmem_cachep = xfrm_dst_cache;
2703 if (likely(dst_ops->check == NULL))
2704 dst_ops->check = xfrm_dst_check;
2705 if (likely(dst_ops->default_advmss == NULL))
2706 dst_ops->default_advmss = xfrm_default_advmss;
2707 if (likely(dst_ops->mtu == NULL))
2708 dst_ops->mtu = xfrm_mtu;
2709 if (likely(dst_ops->negative_advice == NULL))
2710 dst_ops->negative_advice = xfrm_negative_advice;
2711 if (likely(dst_ops->link_failure == NULL))
2712 dst_ops->link_failure = xfrm_link_failure;
2713 if (likely(dst_ops->neigh_lookup == NULL))
2714 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2715 if (likely(afinfo->garbage_collect == NULL))
2716 afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2717 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2718 }
2719 spin_unlock(&xfrm_policy_afinfo_lock);
2720
2721 rtnl_lock();
2722 for_each_net(net) {
2723 struct dst_ops *xfrm_dst_ops;
2724
2725 switch (afinfo->family) {
2726 case AF_INET:
2727 xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2728 break;
2729 #if IS_ENABLED(CONFIG_IPV6)
2730 case AF_INET6:
2731 xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2732 break;
2733 #endif
2734 default:
2735 BUG();
2736 }
2737 *xfrm_dst_ops = *afinfo->dst_ops;
2738 }
2739 rtnl_unlock();
2740
2741 return err;
2742 }
2743 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2744
2745 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2746 {
2747 int err = 0;
2748 if (unlikely(afinfo == NULL))
2749 return -EINVAL;
2750 if (unlikely(afinfo->family >= NPROTO))
2751 return -EAFNOSUPPORT;
2752 spin_lock(&xfrm_policy_afinfo_lock);
2753 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2754 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2755 err = -EINVAL;
2756 else
2757 RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2758 NULL);
2759 }
2760 spin_unlock(&xfrm_policy_afinfo_lock);
2761 if (!err) {
2762 struct dst_ops *dst_ops = afinfo->dst_ops;
2763
2764 synchronize_rcu();
2765
2766 dst_ops->kmem_cachep = NULL;
2767 dst_ops->check = NULL;
2768 dst_ops->negative_advice = NULL;
2769 dst_ops->link_failure = NULL;
2770 afinfo->garbage_collect = NULL;
2771 }
2772 return err;
2773 }
2774 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2775
2776 static void __net_init xfrm_dst_ops_init(struct net *net)
2777 {
2778 struct xfrm_policy_afinfo *afinfo;
2779
2780 rcu_read_lock();
2781 afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2782 if (afinfo)
2783 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2784 #if IS_ENABLED(CONFIG_IPV6)
2785 afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2786 if (afinfo)
2787 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2788 #endif
2789 rcu_read_unlock();
2790 }
2791
2792 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2793 {
2794 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2795
2796 switch (event) {
2797 case NETDEV_DOWN:
2798 xfrm_garbage_collect(dev_net(dev));
2799 }
2800 return NOTIFY_DONE;
2801 }
2802
2803 static struct notifier_block xfrm_dev_notifier = {
2804 .notifier_call = xfrm_dev_event,
2805 };
2806
2807 #ifdef CONFIG_XFRM_STATISTICS
2808 static int __net_init xfrm_statistics_init(struct net *net)
2809 {
2810 int rv;
2811
2812 if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2813 sizeof(struct linux_xfrm_mib),
2814 __alignof__(struct linux_xfrm_mib)) < 0)
2815 return -ENOMEM;
2816 rv = xfrm_proc_init(net);
2817 if (rv < 0)
2818 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2819 return rv;
2820 }
2821
2822 static void xfrm_statistics_fini(struct net *net)
2823 {
2824 xfrm_proc_fini(net);
2825 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2826 }
2827 #else
2828 static int __net_init xfrm_statistics_init(struct net *net)
2829 {
2830 return 0;
2831 }
2832
2833 static void xfrm_statistics_fini(struct net *net)
2834 {
2835 }
2836 #endif
2837
2838 static int __net_init xfrm_policy_init(struct net *net)
2839 {
2840 unsigned int hmask, sz;
2841 int dir;
2842
2843 if (net_eq(net, &init_net))
2844 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2845 sizeof(struct xfrm_dst),
2846 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2847 NULL);
2848
2849 hmask = 8 - 1;
2850 sz = (hmask+1) * sizeof(struct hlist_head);
2851
2852 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2853 if (!net->xfrm.policy_byidx)
2854 goto out_byidx;
2855 net->xfrm.policy_idx_hmask = hmask;
2856
2857 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2858 struct xfrm_policy_hash *htab;
2859
2860 net->xfrm.policy_count[dir] = 0;
2861 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2862
2863 htab = &net->xfrm.policy_bydst[dir];
2864 htab->table = xfrm_hash_alloc(sz);
2865 if (!htab->table)
2866 goto out_bydst;
2867 htab->hmask = hmask;
2868 }
2869
2870 INIT_LIST_HEAD(&net->xfrm.policy_all);
2871 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2872 if (net_eq(net, &init_net))
2873 register_netdevice_notifier(&xfrm_dev_notifier);
2874 return 0;
2875
2876 out_bydst:
2877 for (dir--; dir >= 0; dir--) {
2878 struct xfrm_policy_hash *htab;
2879
2880 htab = &net->xfrm.policy_bydst[dir];
2881 xfrm_hash_free(htab->table, sz);
2882 }
2883 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2884 out_byidx:
2885 return -ENOMEM;
2886 }
2887
2888 static void xfrm_policy_fini(struct net *net)
2889 {
2890 struct xfrm_audit audit_info;
2891 unsigned int sz;
2892 int dir;
2893
2894 flush_work(&net->xfrm.policy_hash_work);
2895 #ifdef CONFIG_XFRM_SUB_POLICY
2896 audit_info.loginuid = INVALID_UID;
2897 audit_info.sessionid = -1;
2898 audit_info.secid = 0;
2899 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2900 #endif
2901 audit_info.loginuid = INVALID_UID;
2902 audit_info.sessionid = -1;
2903 audit_info.secid = 0;
2904 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2905
2906 WARN_ON(!list_empty(&net->xfrm.policy_all));
2907
2908 for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2909 struct xfrm_policy_hash *htab;
2910
2911 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2912
2913 htab = &net->xfrm.policy_bydst[dir];
2914 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2915 WARN_ON(!hlist_empty(htab->table));
2916 xfrm_hash_free(htab->table, sz);
2917 }
2918
2919 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2920 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2921 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2922 }
2923
2924 static int __net_init xfrm_net_init(struct net *net)
2925 {
2926 int rv;
2927
2928 rv = xfrm_statistics_init(net);
2929 if (rv < 0)
2930 goto out_statistics;
2931 rv = xfrm_state_init(net);
2932 if (rv < 0)
2933 goto out_state;
2934 rv = xfrm_policy_init(net);
2935 if (rv < 0)
2936 goto out_policy;
2937 xfrm_dst_ops_init(net);
2938 rv = xfrm_sysctl_init(net);
2939 if (rv < 0)
2940 goto out_sysctl;
2941 return 0;
2942
2943 out_sysctl:
2944 xfrm_policy_fini(net);
2945 out_policy:
2946 xfrm_state_fini(net);
2947 out_state:
2948 xfrm_statistics_fini(net);
2949 out_statistics:
2950 return rv;
2951 }
2952
2953 static void __net_exit xfrm_net_exit(struct net *net)
2954 {
2955 xfrm_sysctl_fini(net);
2956 xfrm_policy_fini(net);
2957 xfrm_state_fini(net);
2958 xfrm_statistics_fini(net);
2959 }
2960
2961 static struct pernet_operations __net_initdata xfrm_net_ops = {
2962 .init = xfrm_net_init,
2963 .exit = xfrm_net_exit,
2964 };
2965
2966 void __init xfrm_init(void)
2967 {
2968 register_pernet_subsys(&xfrm_net_ops);
2969 xfrm_input_init();
2970 }
2971
2972 #ifdef CONFIG_AUDITSYSCALL
2973 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2974 struct audit_buffer *audit_buf)
2975 {
2976 struct xfrm_sec_ctx *ctx = xp->security;
2977 struct xfrm_selector *sel = &xp->selector;
2978
2979 if (ctx)
2980 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2981 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2982
2983 switch(sel->family) {
2984 case AF_INET:
2985 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2986 if (sel->prefixlen_s != 32)
2987 audit_log_format(audit_buf, " src_prefixlen=%d",
2988 sel->prefixlen_s);
2989 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2990 if (sel->prefixlen_d != 32)
2991 audit_log_format(audit_buf, " dst_prefixlen=%d",
2992 sel->prefixlen_d);
2993 break;
2994 case AF_INET6:
2995 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2996 if (sel->prefixlen_s != 128)
2997 audit_log_format(audit_buf, " src_prefixlen=%d",
2998 sel->prefixlen_s);
2999 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3000 if (sel->prefixlen_d != 128)
3001 audit_log_format(audit_buf, " dst_prefixlen=%d",
3002 sel->prefixlen_d);
3003 break;
3004 }
3005 }
3006
3007 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
3008 kuid_t auid, u32 sessionid, u32 secid)
3009 {
3010 struct audit_buffer *audit_buf;
3011
3012 audit_buf = xfrm_audit_start("SPD-add");
3013 if (audit_buf == NULL)
3014 return;
3015 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3016 audit_log_format(audit_buf, " res=%u", result);
3017 xfrm_audit_common_policyinfo(xp, audit_buf);
3018 audit_log_end(audit_buf);
3019 }
3020 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3021
3022 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3023 kuid_t auid, u32 sessionid, u32 secid)
3024 {
3025 struct audit_buffer *audit_buf;
3026
3027 audit_buf = xfrm_audit_start("SPD-delete");
3028 if (audit_buf == NULL)
3029 return;
3030 xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3031 audit_log_format(audit_buf, " res=%u", result);
3032 xfrm_audit_common_policyinfo(xp, audit_buf);
3033 audit_log_end(audit_buf);
3034 }
3035 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3036 #endif
3037
3038 #ifdef CONFIG_XFRM_MIGRATE
3039 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3040 const struct xfrm_selector *sel_tgt)
3041 {
3042 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3043 if (sel_tgt->family == sel_cmp->family &&
3044 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3045 sel_cmp->family) &&
3046 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3047 sel_cmp->family) &&
3048 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3049 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3050 return true;
3051 }
3052 } else {
3053 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3054 return true;
3055 }
3056 }
3057 return false;
3058 }
3059
3060 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3061 u8 dir, u8 type)
3062 {
3063 struct xfrm_policy *pol, *ret = NULL;
3064 struct hlist_head *chain;
3065 u32 priority = ~0U;
3066
3067 read_lock_bh(&xfrm_policy_lock);
3068 chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
3069 hlist_for_each_entry(pol, chain, bydst) {
3070 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3071 pol->type == type) {
3072 ret = pol;
3073 priority = ret->priority;
3074 break;
3075 }
3076 }
3077 chain = &init_net.xfrm.policy_inexact[dir];
3078 hlist_for_each_entry(pol, chain, bydst) {
3079 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3080 pol->type == type &&
3081 pol->priority < priority) {
3082 ret = pol;
3083 break;
3084 }
3085 }
3086
3087 if (ret)
3088 xfrm_pol_hold(ret);
3089
3090 read_unlock_bh(&xfrm_policy_lock);
3091
3092 return ret;
3093 }
3094
3095 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3096 {
3097 int match = 0;
3098
3099 if (t->mode == m->mode && t->id.proto == m->proto &&
3100 (m->reqid == 0 || t->reqid == m->reqid)) {
3101 switch (t->mode) {
3102 case XFRM_MODE_TUNNEL:
3103 case XFRM_MODE_BEET:
3104 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3105 m->old_family) &&
3106 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3107 m->old_family)) {
3108 match = 1;
3109 }
3110 break;
3111 case XFRM_MODE_TRANSPORT:
3112 /* in case of transport mode, template does not store
3113 any IP addresses, hence we just compare mode and
3114 protocol */
3115 match = 1;
3116 break;
3117 default:
3118 break;
3119 }
3120 }
3121 return match;
3122 }
3123
3124 /* update endpoint address(es) of template(s) */
3125 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3126 struct xfrm_migrate *m, int num_migrate)
3127 {
3128 struct xfrm_migrate *mp;
3129 int i, j, n = 0;
3130
3131 write_lock_bh(&pol->lock);
3132 if (unlikely(pol->walk.dead)) {
3133 /* target policy has been deleted */
3134 write_unlock_bh(&pol->lock);
3135 return -ENOENT;
3136 }
3137
3138 for (i = 0; i < pol->xfrm_nr; i++) {
3139 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3140 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3141 continue;
3142 n++;
3143 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3144 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3145 continue;
3146 /* update endpoints */
3147 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3148 sizeof(pol->xfrm_vec[i].id.daddr));
3149 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3150 sizeof(pol->xfrm_vec[i].saddr));
3151 pol->xfrm_vec[i].encap_family = mp->new_family;
3152 /* flush bundles */
3153 atomic_inc(&pol->genid);
3154 }
3155 }
3156
3157 write_unlock_bh(&pol->lock);
3158
3159 if (!n)
3160 return -ENODATA;
3161
3162 return 0;
3163 }
3164
3165 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3166 {
3167 int i, j;
3168
3169 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3170 return -EINVAL;
3171
3172 for (i = 0; i < num_migrate; i++) {
3173 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3174 m[i].old_family) &&
3175 xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3176 m[i].old_family))
3177 return -EINVAL;
3178 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3179 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3180 return -EINVAL;
3181
3182 /* check if there is any duplicated entry */
3183 for (j = i + 1; j < num_migrate; j++) {
3184 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3185 sizeof(m[i].old_daddr)) &&
3186 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3187 sizeof(m[i].old_saddr)) &&
3188 m[i].proto == m[j].proto &&
3189 m[i].mode == m[j].mode &&
3190 m[i].reqid == m[j].reqid &&
3191 m[i].old_family == m[j].old_family)
3192 return -EINVAL;
3193 }
3194 }
3195
3196 return 0;
3197 }
3198
3199 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3200 struct xfrm_migrate *m, int num_migrate,
3201 struct xfrm_kmaddress *k)
3202 {
3203 int i, err, nx_cur = 0, nx_new = 0;
3204 struct xfrm_policy *pol = NULL;
3205 struct xfrm_state *x, *xc;
3206 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3207 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3208 struct xfrm_migrate *mp;
3209
3210 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3211 goto out;
3212
3213 /* Stage 1 - find policy */
3214 if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
3215 err = -ENOENT;
3216 goto out;
3217 }
3218
3219 /* Stage 2 - find and update state(s) */
3220 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3221 if ((x = xfrm_migrate_state_find(mp))) {
3222 x_cur[nx_cur] = x;
3223 nx_cur++;
3224 if ((xc = xfrm_state_migrate(x, mp))) {
3225 x_new[nx_new] = xc;
3226 nx_new++;
3227 } else {
3228 err = -ENODATA;
3229 goto restore_state;
3230 }
3231 }
3232 }
3233
3234 /* Stage 3 - update policy */
3235 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3236 goto restore_state;
3237
3238 /* Stage 4 - delete old state(s) */
3239 if (nx_cur) {
3240 xfrm_states_put(x_cur, nx_cur);
3241 xfrm_states_delete(x_cur, nx_cur);
3242 }
3243
3244 /* Stage 5 - announce */
3245 km_migrate(sel, dir, type, m, num_migrate, k);
3246
3247 xfrm_pol_put(pol);
3248
3249 return 0;
3250 out:
3251 return err;
3252
3253 restore_state:
3254 if (pol)
3255 xfrm_pol_put(pol);
3256 if (nx_cur)
3257 xfrm_states_put(x_cur, nx_cur);
3258 if (nx_new)
3259 xfrm_states_delete(x_new, nx_new);
3260
3261 return err;
3262 }
3263 EXPORT_SYMBOL(xfrm_migrate);
3264 #endif
Linux kernel - Deliverables
This page took 0.107638 seconds and 5 git commands to generate.