projects / GitHub / MotorolaMobilityLLC / kernel-slsi.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[RAMEN9610-20741]UPSTREAM: xfrm: Extend the output_mark to support input direction...
[GitHub/MotorolaMobilityLLC/kernel-slsi.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/cpu.h>
28 #include <linux/audit.h>
29 #include <net/dst.h>
30 #include <net/flow.h>
31 #include <net/xfrm.h>
32 #include <net/ip.h>
33 #ifdef CONFIG_XFRM_STATISTICS
34 #include <net/snmp.h>
35 #endif
36
37 #include "xfrm_hash.h"
38
39 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
40 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
41 #define XFRM_MAX_QUEUE_LEN 100
42
43 struct xfrm_flo {
44 struct dst_entry *dst_orig;
45 u8 flags;
46 };
47
48 static DEFINE_PER_CPU(struct xfrm_dst *, xfrm_last_dst);
49 static struct work_struct *xfrm_pcpu_work __read_mostly;
50 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
51 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
52 __read_mostly;
53
54 static struct kmem_cache *xfrm_dst_cache __read_mostly;
55 static __read_mostly seqcount_t xfrm_policy_hash_generation;
56
57 static void xfrm_init_pmtu(struct dst_entry *dst);
58 static int stale_bundle(struct dst_entry *dst);
59 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
60 static void xfrm_policy_queue_process(unsigned long arg);
61
62 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
63 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
64 int dir);
65
66 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
67 {
68 return refcount_inc_not_zero(&policy->refcnt);
69 }
70
71 static inline bool
72 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
73 {
74 const struct flowi4 *fl4 = &fl->u.ip4;
75
76 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
77 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
78 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
79 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
80 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
81 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
82 }
83
84 static inline bool
85 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
86 {
87 const struct flowi6 *fl6 = &fl->u.ip6;
88
89 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
90 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
91 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
92 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
93 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
94 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
95 }
96
97 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
98 unsigned short family)
99 {
100 switch (family) {
101 case AF_INET:
102 return __xfrm4_selector_match(sel, fl);
103 case AF_INET6:
104 return __xfrm6_selector_match(sel, fl);
105 }
106 return false;
107 }
108
109 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
110 {
111 const struct xfrm_policy_afinfo *afinfo;
112
113 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
114 return NULL;
115 rcu_read_lock();
116 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
117 if (unlikely(!afinfo))
118 rcu_read_unlock();
119 return afinfo;
120 }
121
122 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
123 const xfrm_address_t *saddr,
124 const xfrm_address_t *daddr,
125 int family, u32 mark)
126 {
127 const struct xfrm_policy_afinfo *afinfo;
128 struct dst_entry *dst;
129
130 afinfo = xfrm_policy_get_afinfo(family);
131 if (unlikely(afinfo == NULL))
132 return ERR_PTR(-EAFNOSUPPORT);
133
134 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
135
136 rcu_read_unlock();
137
138 return dst;
139 }
140 EXPORT_SYMBOL(__xfrm_dst_lookup);
141
142 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
143 int tos, int oif,
144 xfrm_address_t *prev_saddr,
145 xfrm_address_t *prev_daddr,
146 int family, u32 mark)
147 {
148 struct net *net = xs_net(x);
149 xfrm_address_t *saddr = &x->props.saddr;
150 xfrm_address_t *daddr = &x->id.daddr;
151 struct dst_entry *dst;
152
153 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
154 saddr = x->coaddr;
155 daddr = prev_daddr;
156 }
157 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
158 saddr = prev_saddr;
159 daddr = x->coaddr;
160 }
161
162 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
163
164 if (!IS_ERR(dst)) {
165 if (prev_saddr != saddr)
166 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
167 if (prev_daddr != daddr)
168 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
169 }
170
171 return dst;
172 }
173
174 static inline unsigned long make_jiffies(long secs)
175 {
176 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
177 return MAX_SCHEDULE_TIMEOUT-1;
178 else
179 return secs*HZ;
180 }
181
182 static void xfrm_policy_timer(unsigned long data)
183 {
184 struct xfrm_policy *xp = (struct xfrm_policy *)data;
185 unsigned long now = get_seconds();
186 long next = LONG_MAX;
187 int warn = 0;
188 int dir;
189
190 read_lock(&xp->lock);
191
192 if (unlikely(xp->walk.dead))
193 goto out;
194
195 dir = xfrm_policy_id2dir(xp->index);
196
197 if (xp->lft.hard_add_expires_seconds) {
198 long tmo = xp->lft.hard_add_expires_seconds +
199 xp->curlft.add_time - now;
200 if (tmo <= 0)
201 goto expired;
202 if (tmo < next)
203 next = tmo;
204 }
205 if (xp->lft.hard_use_expires_seconds) {
206 long tmo = xp->lft.hard_use_expires_seconds +
207 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
208 if (tmo <= 0)
209 goto expired;
210 if (tmo < next)
211 next = tmo;
212 }
213 if (xp->lft.soft_add_expires_seconds) {
214 long tmo = xp->lft.soft_add_expires_seconds +
215 xp->curlft.add_time - now;
216 if (tmo <= 0) {
217 warn = 1;
218 tmo = XFRM_KM_TIMEOUT;
219 }
220 if (tmo < next)
221 next = tmo;
222 }
223 if (xp->lft.soft_use_expires_seconds) {
224 long tmo = xp->lft.soft_use_expires_seconds +
225 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
226 if (tmo <= 0) {
227 warn = 1;
228 tmo = XFRM_KM_TIMEOUT;
229 }
230 if (tmo < next)
231 next = tmo;
232 }
233
234 if (warn)
235 km_policy_expired(xp, dir, 0, 0);
236 if (next != LONG_MAX &&
237 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
238 xfrm_pol_hold(xp);
239
240 out:
241 read_unlock(&xp->lock);
242 xfrm_pol_put(xp);
243 return;
244
245 expired:
246 read_unlock(&xp->lock);
247 if (!xfrm_policy_delete(xp, dir))
248 km_policy_expired(xp, dir, 1, 0);
249 xfrm_pol_put(xp);
250 }
251
252 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
253 * SPD calls.
254 */
255
256 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
257 {
258 struct xfrm_policy *policy;
259
260 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
261
262 if (policy) {
263 write_pnet(&policy->xp_net, net);
264 INIT_LIST_HEAD(&policy->walk.all);
265 INIT_HLIST_NODE(&policy->bydst);
266 INIT_HLIST_NODE(&policy->byidx);
267 rwlock_init(&policy->lock);
268 refcount_set(&policy->refcnt, 1);
269 skb_queue_head_init(&policy->polq.hold_queue);
270 setup_timer(&policy->timer, xfrm_policy_timer,
271 (unsigned long)policy);
272 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
273 (unsigned long)policy);
274 }
275 return policy;
276 }
277 EXPORT_SYMBOL(xfrm_policy_alloc);
278
279 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
280 {
281 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
282
283 security_xfrm_policy_free(policy->security);
284 kfree(policy);
285 }
286
287 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
288
289 void xfrm_policy_destroy(struct xfrm_policy *policy)
290 {
291 BUG_ON(!policy->walk.dead);
292
293 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
294 BUG();
295
296 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
297 }
298 EXPORT_SYMBOL(xfrm_policy_destroy);
299
300 /* Rule must be locked. Release descendant resources, announce
301 * entry dead. The rule must be unlinked from lists to the moment.
302 */
303
304 static void xfrm_policy_kill(struct xfrm_policy *policy)
305 {
306 policy->walk.dead = 1;
307
308 atomic_inc(&policy->genid);
309
310 if (del_timer(&policy->polq.hold_timer))
311 xfrm_pol_put(policy);
312 skb_queue_purge(&policy->polq.hold_queue);
313
314 if (del_timer(&policy->timer))
315 xfrm_pol_put(policy);
316
317 xfrm_pol_put(policy);
318 }
319
320 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
321
322 static inline unsigned int idx_hash(struct net *net, u32 index)
323 {
324 return __idx_hash(index, net->xfrm.policy_idx_hmask);
325 }
326
327 /* calculate policy hash thresholds */
328 static void __get_hash_thresh(struct net *net,
329 unsigned short family, int dir,
330 u8 *dbits, u8 *sbits)
331 {
332 switch (family) {
333 case AF_INET:
334 *dbits = net->xfrm.policy_bydst[dir].dbits4;
335 *sbits = net->xfrm.policy_bydst[dir].sbits4;
336 break;
337
338 case AF_INET6:
339 *dbits = net->xfrm.policy_bydst[dir].dbits6;
340 *sbits = net->xfrm.policy_bydst[dir].sbits6;
341 break;
342
343 default:
344 *dbits = 0;
345 *sbits = 0;
346 }
347 }
348
349 static struct hlist_head *policy_hash_bysel(struct net *net,
350 const struct xfrm_selector *sel,
351 unsigned short family, int dir)
352 {
353 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
354 unsigned int hash;
355 u8 dbits;
356 u8 sbits;
357
358 __get_hash_thresh(net, family, dir, &dbits, &sbits);
359 hash = __sel_hash(sel, family, hmask, dbits, sbits);
360
361 if (hash == hmask + 1)
362 return &net->xfrm.policy_inexact[dir];
363
364 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
365 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
366 }
367
368 static struct hlist_head *policy_hash_direct(struct net *net,
369 const xfrm_address_t *daddr,
370 const xfrm_address_t *saddr,
371 unsigned short family, int dir)
372 {
373 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
374 unsigned int hash;
375 u8 dbits;
376 u8 sbits;
377
378 __get_hash_thresh(net, family, dir, &dbits, &sbits);
379 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
380
381 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
382 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
383 }
384
385 static void xfrm_dst_hash_transfer(struct net *net,
386 struct hlist_head *list,
387 struct hlist_head *ndsttable,
388 unsigned int nhashmask,
389 int dir)
390 {
391 struct hlist_node *tmp, *entry0 = NULL;
392 struct xfrm_policy *pol;
393 unsigned int h0 = 0;
394 u8 dbits;
395 u8 sbits;
396
397 redo:
398 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
399 unsigned int h;
400
401 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
402 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
403 pol->family, nhashmask, dbits, sbits);
404 if (!entry0) {
405 hlist_del_rcu(&pol->bydst);
406 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
407 h0 = h;
408 } else {
409 if (h != h0)
410 continue;
411 hlist_del_rcu(&pol->bydst);
412 hlist_add_behind_rcu(&pol->bydst, entry0);
413 }
414 entry0 = &pol->bydst;
415 }
416 if (!hlist_empty(list)) {
417 entry0 = NULL;
418 goto redo;
419 }
420 }
421
422 static void xfrm_idx_hash_transfer(struct hlist_head *list,
423 struct hlist_head *nidxtable,
424 unsigned int nhashmask)
425 {
426 struct hlist_node *tmp;
427 struct xfrm_policy *pol;
428
429 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
430 unsigned int h;
431
432 h = __idx_hash(pol->index, nhashmask);
433 hlist_add_head(&pol->byidx, nidxtable+h);
434 }
435 }
436
437 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
438 {
439 return ((old_hmask + 1) << 1) - 1;
440 }
441
442 static void xfrm_bydst_resize(struct net *net, int dir)
443 {
444 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
445 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
446 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
447 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
448 struct hlist_head *odst;
449 int i;
450
451 if (!ndst)
452 return;
453
454 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
455 write_seqcount_begin(&xfrm_policy_hash_generation);
456
457 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
458 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
459
460 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
461 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
462
463 for (i = hmask; i >= 0; i--)
464 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
465
466 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
467 net->xfrm.policy_bydst[dir].hmask = nhashmask;
468
469 write_seqcount_end(&xfrm_policy_hash_generation);
470 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
471
472 synchronize_rcu();
473
474 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
475 }
476
477 static void xfrm_byidx_resize(struct net *net, int total)
478 {
479 unsigned int hmask = net->xfrm.policy_idx_hmask;
480 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
481 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
482 struct hlist_head *oidx = net->xfrm.policy_byidx;
483 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
484 int i;
485
486 if (!nidx)
487 return;
488
489 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
490
491 for (i = hmask; i >= 0; i--)
492 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
493
494 net->xfrm.policy_byidx = nidx;
495 net->xfrm.policy_idx_hmask = nhashmask;
496
497 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
498
499 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
500 }
501
502 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
503 {
504 unsigned int cnt = net->xfrm.policy_count[dir];
505 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
506
507 if (total)
508 *total += cnt;
509
510 if ((hmask + 1) < xfrm_policy_hashmax &&
511 cnt > hmask)
512 return 1;
513
514 return 0;
515 }
516
517 static inline int xfrm_byidx_should_resize(struct net *net, int total)
518 {
519 unsigned int hmask = net->xfrm.policy_idx_hmask;
520
521 if ((hmask + 1) < xfrm_policy_hashmax &&
522 total > hmask)
523 return 1;
524
525 return 0;
526 }
527
528 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
529 {
530 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
531 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
532 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
533 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
534 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
535 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
536 si->spdhcnt = net->xfrm.policy_idx_hmask;
537 si->spdhmcnt = xfrm_policy_hashmax;
538 }
539 EXPORT_SYMBOL(xfrm_spd_getinfo);
540
541 static DEFINE_MUTEX(hash_resize_mutex);
542 static void xfrm_hash_resize(struct work_struct *work)
543 {
544 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
545 int dir, total;
546
547 mutex_lock(&hash_resize_mutex);
548
549 total = 0;
550 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
551 if (xfrm_bydst_should_resize(net, dir, &total))
552 xfrm_bydst_resize(net, dir);
553 }
554 if (xfrm_byidx_should_resize(net, total))
555 xfrm_byidx_resize(net, total);
556
557 mutex_unlock(&hash_resize_mutex);
558 }
559
560 static void xfrm_hash_rebuild(struct work_struct *work)
561 {
562 struct net *net = container_of(work, struct net,
563 xfrm.policy_hthresh.work);
564 unsigned int hmask;
565 struct xfrm_policy *pol;
566 struct xfrm_policy *policy;
567 struct hlist_head *chain;
568 struct hlist_head *odst;
569 struct hlist_node *newpos;
570 int i;
571 int dir;
572 unsigned seq;
573 u8 lbits4, rbits4, lbits6, rbits6;
574
575 mutex_lock(&hash_resize_mutex);
576
577 /* read selector prefixlen thresholds */
578 do {
579 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
580
581 lbits4 = net->xfrm.policy_hthresh.lbits4;
582 rbits4 = net->xfrm.policy_hthresh.rbits4;
583 lbits6 = net->xfrm.policy_hthresh.lbits6;
584 rbits6 = net->xfrm.policy_hthresh.rbits6;
585 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
586
587 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
588
589 /* reset the bydst and inexact table in all directions */
590 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
591 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
592 hmask = net->xfrm.policy_bydst[dir].hmask;
593 odst = net->xfrm.policy_bydst[dir].table;
594 for (i = hmask; i >= 0; i--)
595 INIT_HLIST_HEAD(odst + i);
596 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
597 /* dir out => dst = remote, src = local */
598 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
599 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
600 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
601 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
602 } else {
603 /* dir in/fwd => dst = local, src = remote */
604 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
605 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
606 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
607 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
608 }
609 }
610
611 /* re-insert all policies by order of creation */
612 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
613 if (policy->walk.dead ||
614 xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
615 /* skip socket policies */
616 continue;
617 }
618 newpos = NULL;
619 chain = policy_hash_bysel(net, &policy->selector,
620 policy->family,
621 xfrm_policy_id2dir(policy->index));
622 hlist_for_each_entry(pol, chain, bydst) {
623 if (policy->priority >= pol->priority)
624 newpos = &pol->bydst;
625 else
626 break;
627 }
628 if (newpos)
629 hlist_add_behind_rcu(&policy->bydst, newpos);
630 else
631 hlist_add_head_rcu(&policy->bydst, chain);
632 }
633
634 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
635
636 mutex_unlock(&hash_resize_mutex);
637 }
638
639 void xfrm_policy_hash_rebuild(struct net *net)
640 {
641 schedule_work(&net->xfrm.policy_hthresh.work);
642 }
643 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
644
645 /* Generate new index... KAME seems to generate them ordered by cost
646 * of an absolute inpredictability of ordering of rules. This will not pass. */
647 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
648 {
649 static u32 idx_generator;
650
651 for (;;) {
652 struct hlist_head *list;
653 struct xfrm_policy *p;
654 u32 idx;
655 int found;
656
657 if (!index) {
658 idx = (idx_generator | dir);
659 idx_generator += 8;
660 } else {
661 idx = index;
662 index = 0;
663 }
664
665 if (idx == 0)
666 idx = 8;
667 list = net->xfrm.policy_byidx + idx_hash(net, idx);
668 found = 0;
669 hlist_for_each_entry(p, list, byidx) {
670 if (p->index == idx) {
671 found = 1;
672 break;
673 }
674 }
675 if (!found)
676 return idx;
677 }
678 }
679
680 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
681 {
682 u32 *p1 = (u32 *) s1;
683 u32 *p2 = (u32 *) s2;
684 int len = sizeof(struct xfrm_selector) / sizeof(u32);
685 int i;
686
687 for (i = 0; i < len; i++) {
688 if (p1[i] != p2[i])
689 return 1;
690 }
691
692 return 0;
693 }
694
695 static void xfrm_policy_requeue(struct xfrm_policy *old,
696 struct xfrm_policy *new)
697 {
698 struct xfrm_policy_queue *pq = &old->polq;
699 struct sk_buff_head list;
700
701 if (skb_queue_empty(&pq->hold_queue))
702 return;
703
704 __skb_queue_head_init(&list);
705
706 spin_lock_bh(&pq->hold_queue.lock);
707 skb_queue_splice_init(&pq->hold_queue, &list);
708 if (del_timer(&pq->hold_timer))
709 xfrm_pol_put(old);
710 spin_unlock_bh(&pq->hold_queue.lock);
711
712 pq = &new->polq;
713
714 spin_lock_bh(&pq->hold_queue.lock);
715 skb_queue_splice(&list, &pq->hold_queue);
716 pq->timeout = XFRM_QUEUE_TMO_MIN;
717 if (!mod_timer(&pq->hold_timer, jiffies))
718 xfrm_pol_hold(new);
719 spin_unlock_bh(&pq->hold_queue.lock);
720 }
721
722 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
723 struct xfrm_policy *pol)
724 {
725 u32 mark = policy->mark.v & policy->mark.m;
726
727 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
728 return true;
729
730 if ((mark & pol->mark.m) == pol->mark.v &&
731 policy->priority == pol->priority)
732 return true;
733
734 return false;
735 }
736
737 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
738 {
739 struct net *net = xp_net(policy);
740 struct xfrm_policy *pol;
741 struct xfrm_policy *delpol;
742 struct hlist_head *chain;
743 struct hlist_node *newpos;
744
745 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
746 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
747 delpol = NULL;
748 newpos = NULL;
749 hlist_for_each_entry(pol, chain, bydst) {
750 if (pol->type == policy->type &&
751 !selector_cmp(&pol->selector, &policy->selector) &&
752 xfrm_policy_mark_match(policy, pol) &&
753 xfrm_sec_ctx_match(pol->security, policy->security) &&
754 !WARN_ON(delpol)) {
755 if (excl) {
756 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
757 return -EEXIST;
758 }
759 delpol = pol;
760 if (policy->priority > pol->priority)
761 continue;
762 } else if (policy->priority >= pol->priority) {
763 newpos = &pol->bydst;
764 continue;
765 }
766 if (delpol)
767 break;
768 }
769 if (newpos)
770 hlist_add_behind_rcu(&policy->bydst, newpos);
771 else
772 hlist_add_head_rcu(&policy->bydst, chain);
773 __xfrm_policy_link(policy, dir);
774
775 /* After previous checking, family can either be AF_INET or AF_INET6 */
776 if (policy->family == AF_INET)
777 rt_genid_bump_ipv4(net);
778 else
779 rt_genid_bump_ipv6(net);
780
781 if (delpol) {
782 xfrm_policy_requeue(delpol, policy);
783 __xfrm_policy_unlink(delpol, dir);
784 }
785 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
786 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
787 policy->curlft.add_time = get_seconds();
788 policy->curlft.use_time = 0;
789 if (!mod_timer(&policy->timer, jiffies + HZ))
790 xfrm_pol_hold(policy);
791 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
792
793 if (delpol)
794 xfrm_policy_kill(delpol);
795 else if (xfrm_bydst_should_resize(net, dir, NULL))
796 schedule_work(&net->xfrm.policy_hash_work);
797
798 return 0;
799 }
800 EXPORT_SYMBOL(xfrm_policy_insert);
801
802 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
803 int dir, struct xfrm_selector *sel,
804 struct xfrm_sec_ctx *ctx, int delete,
805 int *err)
806 {
807 struct xfrm_policy *pol, *ret;
808 struct hlist_head *chain;
809
810 *err = 0;
811 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
812 chain = policy_hash_bysel(net, sel, sel->family, dir);
813 ret = NULL;
814 hlist_for_each_entry(pol, chain, bydst) {
815 if (pol->type == type &&
816 (mark & pol->mark.m) == pol->mark.v &&
817 !selector_cmp(sel, &pol->selector) &&
818 xfrm_sec_ctx_match(ctx, pol->security)) {
819 xfrm_pol_hold(pol);
820 if (delete) {
821 *err = security_xfrm_policy_delete(
822 pol->security);
823 if (*err) {
824 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
825 return pol;
826 }
827 __xfrm_policy_unlink(pol, dir);
828 }
829 ret = pol;
830 break;
831 }
832 }
833 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
834
835 if (ret && delete)
836 xfrm_policy_kill(ret);
837 return ret;
838 }
839 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
840
841 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
842 int dir, u32 id, int delete, int *err)
843 {
844 struct xfrm_policy *pol, *ret;
845 struct hlist_head *chain;
846
847 *err = -ENOENT;
848 if (xfrm_policy_id2dir(id) != dir)
849 return NULL;
850
851 *err = 0;
852 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
853 chain = net->xfrm.policy_byidx + idx_hash(net, id);
854 ret = NULL;
855 hlist_for_each_entry(pol, chain, byidx) {
856 if (pol->type == type && pol->index == id &&
857 (mark & pol->mark.m) == pol->mark.v) {
858 xfrm_pol_hold(pol);
859 if (delete) {
860 *err = security_xfrm_policy_delete(
861 pol->security);
862 if (*err) {
863 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
864 return pol;
865 }
866 __xfrm_policy_unlink(pol, dir);
867 }
868 ret = pol;
869 break;
870 }
871 }
872 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
873
874 if (ret && delete)
875 xfrm_policy_kill(ret);
876 return ret;
877 }
878 EXPORT_SYMBOL(xfrm_policy_byid);
879
880 #ifdef CONFIG_SECURITY_NETWORK_XFRM
881 static inline int
882 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
883 {
884 int dir, err = 0;
885
886 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
887 struct xfrm_policy *pol;
888 int i;
889
890 hlist_for_each_entry(pol,
891 &net->xfrm.policy_inexact[dir], bydst) {
892 if (pol->type != type)
893 continue;
894 err = security_xfrm_policy_delete(pol->security);
895 if (err) {
896 xfrm_audit_policy_delete(pol, 0, task_valid);
897 return err;
898 }
899 }
900 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
901 hlist_for_each_entry(pol,
902 net->xfrm.policy_bydst[dir].table + i,
903 bydst) {
904 if (pol->type != type)
905 continue;
906 err = security_xfrm_policy_delete(
907 pol->security);
908 if (err) {
909 xfrm_audit_policy_delete(pol, 0,
910 task_valid);
911 return err;
912 }
913 }
914 }
915 }
916 return err;
917 }
918 #else
919 static inline int
920 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
921 {
922 return 0;
923 }
924 #endif
925
926 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
927 {
928 int dir, err = 0, cnt = 0;
929
930 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
931
932 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
933 if (err)
934 goto out;
935
936 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
937 struct xfrm_policy *pol;
938 int i;
939
940 again1:
941 hlist_for_each_entry(pol,
942 &net->xfrm.policy_inexact[dir], bydst) {
943 if (pol->type != type)
944 continue;
945 __xfrm_policy_unlink(pol, dir);
946 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
947 cnt++;
948
949 xfrm_audit_policy_delete(pol, 1, task_valid);
950
951 xfrm_policy_kill(pol);
952
953 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
954 goto again1;
955 }
956
957 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
958 again2:
959 hlist_for_each_entry(pol,
960 net->xfrm.policy_bydst[dir].table + i,
961 bydst) {
962 if (pol->type != type)
963 continue;
964 __xfrm_policy_unlink(pol, dir);
965 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
966 cnt++;
967
968 xfrm_audit_policy_delete(pol, 1, task_valid);
969 xfrm_policy_kill(pol);
970
971 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
972 goto again2;
973 }
974 }
975
976 }
977 if (!cnt)
978 err = -ESRCH;
979 out:
980 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
981 return err;
982 }
983 EXPORT_SYMBOL(xfrm_policy_flush);
984
985 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
986 int (*func)(struct xfrm_policy *, int, int, void*),
987 void *data)
988 {
989 struct xfrm_policy *pol;
990 struct xfrm_policy_walk_entry *x;
991 int error = 0;
992
993 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
994 walk->type != XFRM_POLICY_TYPE_ANY)
995 return -EINVAL;
996
997 if (list_empty(&walk->walk.all) && walk->seq != 0)
998 return 0;
999
1000 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1001 if (list_empty(&walk->walk.all))
1002 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1003 else
1004 x = list_first_entry(&walk->walk.all,
1005 struct xfrm_policy_walk_entry, all);
1006
1007 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1008 if (x->dead)
1009 continue;
1010 pol = container_of(x, struct xfrm_policy, walk);
1011 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1012 walk->type != pol->type)
1013 continue;
1014 error = func(pol, xfrm_policy_id2dir(pol->index),
1015 walk->seq, data);
1016 if (error) {
1017 list_move_tail(&walk->walk.all, &x->all);
1018 goto out;
1019 }
1020 walk->seq++;
1021 }
1022 if (walk->seq == 0) {
1023 error = -ENOENT;
1024 goto out;
1025 }
1026 list_del_init(&walk->walk.all);
1027 out:
1028 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1029 return error;
1030 }
1031 EXPORT_SYMBOL(xfrm_policy_walk);
1032
1033 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1034 {
1035 INIT_LIST_HEAD(&walk->walk.all);
1036 walk->walk.dead = 1;
1037 walk->type = type;
1038 walk->seq = 0;
1039 }
1040 EXPORT_SYMBOL(xfrm_policy_walk_init);
1041
1042 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1043 {
1044 if (list_empty(&walk->walk.all))
1045 return;
1046
1047 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1048 list_del(&walk->walk.all);
1049 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1050 }
1051 EXPORT_SYMBOL(xfrm_policy_walk_done);
1052
1053 /*
1054 * Find policy to apply to this flow.
1055 *
1056 * Returns 0 if policy found, else an -errno.
1057 */
1058 static int xfrm_policy_match(const struct xfrm_policy *pol,
1059 const struct flowi *fl,
1060 u8 type, u16 family, int dir)
1061 {
1062 const struct xfrm_selector *sel = &pol->selector;
1063 int ret = -ESRCH;
1064 bool match;
1065
1066 if (pol->family != family ||
1067 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1068 pol->type != type)
1069 return ret;
1070
1071 match = xfrm_selector_match(sel, fl, family);
1072 if (match)
1073 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1074 dir);
1075
1076 return ret;
1077 }
1078
1079 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1080 const struct flowi *fl,
1081 u16 family, u8 dir)
1082 {
1083 int err;
1084 struct xfrm_policy *pol, *ret;
1085 const xfrm_address_t *daddr, *saddr;
1086 struct hlist_head *chain;
1087 unsigned int sequence;
1088 u32 priority;
1089
1090 daddr = xfrm_flowi_daddr(fl, family);
1091 saddr = xfrm_flowi_saddr(fl, family);
1092 if (unlikely(!daddr || !saddr))
1093 return NULL;
1094
1095 rcu_read_lock();
1096 retry:
1097 do {
1098 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1099 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1100 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1101
1102 priority = ~0U;
1103 ret = NULL;
1104 hlist_for_each_entry_rcu(pol, chain, bydst) {
1105 err = xfrm_policy_match(pol, fl, type, family, dir);
1106 if (err) {
1107 if (err == -ESRCH)
1108 continue;
1109 else {
1110 ret = ERR_PTR(err);
1111 goto fail;
1112 }
1113 } else {
1114 ret = pol;
1115 priority = ret->priority;
1116 break;
1117 }
1118 }
1119 chain = &net->xfrm.policy_inexact[dir];
1120 hlist_for_each_entry_rcu(pol, chain, bydst) {
1121 if ((pol->priority >= priority) && ret)
1122 break;
1123
1124 err = xfrm_policy_match(pol, fl, type, family, dir);
1125 if (err) {
1126 if (err == -ESRCH)
1127 continue;
1128 else {
1129 ret = ERR_PTR(err);
1130 goto fail;
1131 }
1132 } else {
1133 ret = pol;
1134 break;
1135 }
1136 }
1137
1138 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1139 goto retry;
1140
1141 if (ret && !xfrm_pol_hold_rcu(ret))
1142 goto retry;
1143 fail:
1144 rcu_read_unlock();
1145
1146 return ret;
1147 }
1148
1149 static struct xfrm_policy *
1150 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1151 {
1152 #ifdef CONFIG_XFRM_SUB_POLICY
1153 struct xfrm_policy *pol;
1154
1155 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1156 if (pol != NULL)
1157 return pol;
1158 #endif
1159 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1160 }
1161
1162 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1163 const struct flowi *fl, u16 family)
1164 {
1165 struct xfrm_policy *pol;
1166
1167 rcu_read_lock();
1168 again:
1169 pol = rcu_dereference(sk->sk_policy[dir]);
1170 if (pol != NULL) {
1171 bool match;
1172 int err = 0;
1173
1174 if (pol->family != family) {
1175 pol = NULL;
1176 goto out;
1177 }
1178
1179 match = xfrm_selector_match(&pol->selector, fl, family);
1180 if (match) {
1181 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1182 pol = NULL;
1183 goto out;
1184 }
1185 err = security_xfrm_policy_lookup(pol->security,
1186 fl->flowi_secid,
1187 dir);
1188 if (!err) {
1189 if (!xfrm_pol_hold_rcu(pol))
1190 goto again;
1191 } else if (err == -ESRCH) {
1192 pol = NULL;
1193 } else {
1194 pol = ERR_PTR(err);
1195 }
1196 } else
1197 pol = NULL;
1198 }
1199 out:
1200 rcu_read_unlock();
1201 return pol;
1202 }
1203
1204 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1205 {
1206 struct net *net = xp_net(pol);
1207
1208 list_add(&pol->walk.all, &net->xfrm.policy_all);
1209 net->xfrm.policy_count[dir]++;
1210 xfrm_pol_hold(pol);
1211 }
1212
1213 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1214 int dir)
1215 {
1216 struct net *net = xp_net(pol);
1217
1218 if (list_empty(&pol->walk.all))
1219 return NULL;
1220
1221 /* Socket policies are not hashed. */
1222 if (!hlist_unhashed(&pol->bydst)) {
1223 hlist_del_rcu(&pol->bydst);
1224 hlist_del(&pol->byidx);
1225 }
1226
1227 list_del_init(&pol->walk.all);
1228 net->xfrm.policy_count[dir]--;
1229
1230 return pol;
1231 }
1232
1233 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1234 {
1235 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1236 }
1237
1238 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1239 {
1240 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1241 }
1242
1243 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1244 {
1245 struct net *net = xp_net(pol);
1246
1247 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1248 pol = __xfrm_policy_unlink(pol, dir);
1249 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1250 if (pol) {
1251 xfrm_policy_kill(pol);
1252 return 0;
1253 }
1254 return -ENOENT;
1255 }
1256 EXPORT_SYMBOL(xfrm_policy_delete);
1257
1258 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1259 {
1260 struct net *net = sock_net(sk);
1261 struct xfrm_policy *old_pol;
1262
1263 #ifdef CONFIG_XFRM_SUB_POLICY
1264 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1265 return -EINVAL;
1266 #endif
1267
1268 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1269 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1270 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1271 if (pol) {
1272 pol->curlft.add_time = get_seconds();
1273 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1274 xfrm_sk_policy_link(pol, dir);
1275 }
1276 rcu_assign_pointer(sk->sk_policy[dir], pol);
1277 if (old_pol) {
1278 if (pol)
1279 xfrm_policy_requeue(old_pol, pol);
1280
1281 /* Unlinking succeeds always. This is the only function
1282 * allowed to delete or replace socket policy.
1283 */
1284 xfrm_sk_policy_unlink(old_pol, dir);
1285 }
1286 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1287
1288 if (old_pol) {
1289 xfrm_policy_kill(old_pol);
1290 }
1291 return 0;
1292 }
1293
1294 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1295 {
1296 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1297 struct net *net = xp_net(old);
1298
1299 if (newp) {
1300 newp->selector = old->selector;
1301 if (security_xfrm_policy_clone(old->security,
1302 &newp->security)) {
1303 kfree(newp);
1304 return NULL; /* ENOMEM */
1305 }
1306 newp->lft = old->lft;
1307 newp->curlft = old->curlft;
1308 newp->mark = old->mark;
1309 newp->action = old->action;
1310 newp->flags = old->flags;
1311 newp->xfrm_nr = old->xfrm_nr;
1312 newp->index = old->index;
1313 newp->type = old->type;
1314 newp->family = old->family;
1315 memcpy(newp->xfrm_vec, old->xfrm_vec,
1316 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1317 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1318 xfrm_sk_policy_link(newp, dir);
1319 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1320 xfrm_pol_put(newp);
1321 }
1322 return newp;
1323 }
1324
1325 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1326 {
1327 const struct xfrm_policy *p;
1328 struct xfrm_policy *np;
1329 int i, ret = 0;
1330
1331 rcu_read_lock();
1332 for (i = 0; i < 2; i++) {
1333 p = rcu_dereference(osk->sk_policy[i]);
1334 if (p) {
1335 np = clone_policy(p, i);
1336 if (unlikely(!np)) {
1337 ret = -ENOMEM;
1338 break;
1339 }
1340 rcu_assign_pointer(sk->sk_policy[i], np);
1341 }
1342 }
1343 rcu_read_unlock();
1344 return ret;
1345 }
1346
1347 static int
1348 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1349 xfrm_address_t *remote, unsigned short family, u32 mark)
1350 {
1351 int err;
1352 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1353
1354 if (unlikely(afinfo == NULL))
1355 return -EINVAL;
1356 err = afinfo->get_saddr(net, oif, local, remote, mark);
1357 rcu_read_unlock();
1358 return err;
1359 }
1360
1361 /* Resolve list of templates for the flow, given policy. */
1362
1363 static int
1364 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1365 struct xfrm_state **xfrm, unsigned short family)
1366 {
1367 struct net *net = xp_net(policy);
1368 int nx;
1369 int i, error;
1370 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1371 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1372 xfrm_address_t tmp;
1373
1374 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1375 struct xfrm_state *x;
1376 xfrm_address_t *remote = daddr;
1377 xfrm_address_t *local = saddr;
1378 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1379
1380 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1381 tmpl->mode == XFRM_MODE_BEET) {
1382 remote = &tmpl->id.daddr;
1383 local = &tmpl->saddr;
1384 if (xfrm_addr_any(local, tmpl->encap_family)) {
1385 error = xfrm_get_saddr(net, fl->flowi_oif,
1386 &tmp, remote,
1387 tmpl->encap_family, 0);
1388 if (error)
1389 goto fail;
1390 local = &tmp;
1391 }
1392 }
1393
1394 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1395
1396 if (x && x->km.state == XFRM_STATE_VALID) {
1397 xfrm[nx++] = x;
1398 daddr = remote;
1399 saddr = local;
1400 continue;
1401 }
1402 if (x) {
1403 error = (x->km.state == XFRM_STATE_ERROR ?
1404 -EINVAL : -EAGAIN);
1405 xfrm_state_put(x);
1406 } else if (error == -ESRCH) {
1407 error = -EAGAIN;
1408 }
1409
1410 if (!tmpl->optional)
1411 goto fail;
1412 }
1413 return nx;
1414
1415 fail:
1416 for (nx--; nx >= 0; nx--)
1417 xfrm_state_put(xfrm[nx]);
1418 return error;
1419 }
1420
1421 static int
1422 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1423 struct xfrm_state **xfrm, unsigned short family)
1424 {
1425 struct xfrm_state *tp[XFRM_MAX_DEPTH];
1426 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1427 int cnx = 0;
1428 int error;
1429 int ret;
1430 int i;
1431
1432 for (i = 0; i < npols; i++) {
1433 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1434 error = -ENOBUFS;
1435 goto fail;
1436 }
1437
1438 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1439 if (ret < 0) {
1440 error = ret;
1441 goto fail;
1442 } else
1443 cnx += ret;
1444 }
1445
1446 /* found states are sorted for outbound processing */
1447 if (npols > 1)
1448 xfrm_state_sort(xfrm, tpp, cnx, family);
1449
1450 return cnx;
1451
1452 fail:
1453 for (cnx--; cnx >= 0; cnx--)
1454 xfrm_state_put(tpp[cnx]);
1455 return error;
1456
1457 }
1458
1459 static int xfrm_get_tos(const struct flowi *fl, int family)
1460 {
1461 const struct xfrm_policy_afinfo *afinfo;
1462 int tos;
1463
1464 afinfo = xfrm_policy_get_afinfo(family);
1465 if (!afinfo)
1466 return 0;
1467
1468 tos = afinfo->get_tos(fl);
1469
1470 rcu_read_unlock();
1471
1472 return tos;
1473 }
1474
1475 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1476 {
1477 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1478 struct dst_ops *dst_ops;
1479 struct xfrm_dst *xdst;
1480
1481 if (!afinfo)
1482 return ERR_PTR(-EINVAL);
1483
1484 switch (family) {
1485 case AF_INET:
1486 dst_ops = &net->xfrm.xfrm4_dst_ops;
1487 break;
1488 #if IS_ENABLED(CONFIG_IPV6)
1489 case AF_INET6:
1490 dst_ops = &net->xfrm.xfrm6_dst_ops;
1491 break;
1492 #endif
1493 default:
1494 BUG();
1495 }
1496 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1497
1498 if (likely(xdst)) {
1499 struct dst_entry *dst = &xdst->u.dst;
1500
1501 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1502 } else
1503 xdst = ERR_PTR(-ENOBUFS);
1504
1505 rcu_read_unlock();
1506
1507 return xdst;
1508 }
1509
1510 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1511 int nfheader_len)
1512 {
1513 const struct xfrm_policy_afinfo *afinfo =
1514 xfrm_policy_get_afinfo(dst->ops->family);
1515 int err;
1516
1517 if (!afinfo)
1518 return -EINVAL;
1519
1520 err = afinfo->init_path(path, dst, nfheader_len);
1521
1522 rcu_read_unlock();
1523
1524 return err;
1525 }
1526
1527 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1528 const struct flowi *fl)
1529 {
1530 const struct xfrm_policy_afinfo *afinfo =
1531 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1532 int err;
1533
1534 if (!afinfo)
1535 return -EINVAL;
1536
1537 err = afinfo->fill_dst(xdst, dev, fl);
1538
1539 rcu_read_unlock();
1540
1541 return err;
1542 }
1543
1544
1545 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1546 * all the metrics... Shortly, bundle a bundle.
1547 */
1548
1549 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1550 struct xfrm_state **xfrm, int nx,
1551 const struct flowi *fl,
1552 struct dst_entry *dst)
1553 {
1554 struct net *net = xp_net(policy);
1555 unsigned long now = jiffies;
1556 struct net_device *dev;
1557 struct xfrm_mode *inner_mode;
1558 struct dst_entry *dst_prev = NULL;
1559 struct dst_entry *dst0 = NULL;
1560 int i = 0;
1561 int err;
1562 int header_len = 0;
1563 int nfheader_len = 0;
1564 int trailer_len = 0;
1565 int tos;
1566 int family = policy->selector.family;
1567 xfrm_address_t saddr, daddr;
1568
1569 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1570
1571 tos = xfrm_get_tos(fl, family);
1572
1573 dst_hold(dst);
1574
1575 for (; i < nx; i++) {
1576 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1577 struct dst_entry *dst1 = &xdst->u.dst;
1578
1579 err = PTR_ERR(xdst);
1580 if (IS_ERR(xdst)) {
1581 dst_release(dst);
1582 goto put_states;
1583 }
1584
1585 if (!dst_prev)
1586 dst0 = dst1;
1587 else
1588 /* Ref count is taken during xfrm_alloc_dst()
1589 * No need to do dst_clone() on dst1
1590 */
1591 dst_prev->child = dst1;
1592
1593 if (xfrm[i]->sel.family == AF_UNSPEC) {
1594 inner_mode = xfrm_ip2inner_mode(xfrm[i],
1595 xfrm_af2proto(family));
1596 if (!inner_mode) {
1597 err = -EAFNOSUPPORT;
1598 dst_release(dst);
1599 goto put_states;
1600 }
1601 } else
1602 inner_mode = xfrm[i]->inner_mode;
1603
1604 xdst->route = dst;
1605 dst_copy_metrics(dst1, dst);
1606
1607 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1608 __u32 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
1609
1610 family = xfrm[i]->props.family;
1611 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1612 &saddr, &daddr, family, mark);
1613 err = PTR_ERR(dst);
1614 if (IS_ERR(dst))
1615 goto put_states;
1616 } else
1617 dst_hold(dst);
1618
1619 dst1->xfrm = xfrm[i];
1620 xdst->xfrm_genid = xfrm[i]->genid;
1621
1622 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1623 dst1->flags |= DST_HOST;
1624 dst1->lastuse = now;
1625
1626 dst1->input = dst_discard;
1627 dst1->output = inner_mode->afinfo->output;
1628
1629 dst1->next = dst_prev;
1630 dst_prev = dst1;
1631
1632 header_len += xfrm[i]->props.header_len;
1633 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1634 nfheader_len += xfrm[i]->props.header_len;
1635 trailer_len += xfrm[i]->props.trailer_len;
1636 }
1637
1638 dst_prev->child = dst;
1639 dst0->path = dst;
1640
1641 err = -ENODEV;
1642 dev = dst->dev;
1643 if (!dev)
1644 goto free_dst;
1645
1646 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1647 xfrm_init_pmtu(dst_prev);
1648
1649 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1650 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1651
1652 err = xfrm_fill_dst(xdst, dev, fl);
1653 if (err)
1654 goto free_dst;
1655
1656 dst_prev->header_len = header_len;
1657 dst_prev->trailer_len = trailer_len;
1658 header_len -= xdst->u.dst.xfrm->props.header_len;
1659 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1660 }
1661
1662 out:
1663 return dst0;
1664
1665 put_states:
1666 for (; i < nx; i++)
1667 xfrm_state_put(xfrm[i]);
1668 free_dst:
1669 if (dst0)
1670 dst_release_immediate(dst0);
1671 dst0 = ERR_PTR(err);
1672 goto out;
1673 }
1674
1675 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1676 struct xfrm_policy **pols,
1677 int *num_pols, int *num_xfrms)
1678 {
1679 int i;
1680
1681 if (*num_pols == 0 || !pols[0]) {
1682 *num_pols = 0;
1683 *num_xfrms = 0;
1684 return 0;
1685 }
1686 if (IS_ERR(pols[0]))
1687 return PTR_ERR(pols[0]);
1688
1689 *num_xfrms = pols[0]->xfrm_nr;
1690
1691 #ifdef CONFIG_XFRM_SUB_POLICY
1692 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1693 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1694 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1695 XFRM_POLICY_TYPE_MAIN,
1696 fl, family,
1697 XFRM_POLICY_OUT);
1698 if (pols[1]) {
1699 if (IS_ERR(pols[1])) {
1700 xfrm_pols_put(pols, *num_pols);
1701 return PTR_ERR(pols[1]);
1702 }
1703 (*num_pols)++;
1704 (*num_xfrms) += pols[1]->xfrm_nr;
1705 }
1706 }
1707 #endif
1708 for (i = 0; i < *num_pols; i++) {
1709 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1710 *num_xfrms = -1;
1711 break;
1712 }
1713 }
1714
1715 return 0;
1716
1717 }
1718
1719 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1720 {
1721 this_cpu_write(xfrm_last_dst, xdst);
1722 if (old)
1723 dst_release(&old->u.dst);
1724 }
1725
1726 static void __xfrm_pcpu_work_fn(void)
1727 {
1728 struct xfrm_dst *old;
1729
1730 old = this_cpu_read(xfrm_last_dst);
1731 if (old && !xfrm_bundle_ok(old))
1732 xfrm_last_dst_update(NULL, old);
1733 }
1734
1735 static void xfrm_pcpu_work_fn(struct work_struct *work)
1736 {
1737 local_bh_disable();
1738 rcu_read_lock();
1739 __xfrm_pcpu_work_fn();
1740 rcu_read_unlock();
1741 local_bh_enable();
1742 }
1743
1744 void xfrm_policy_cache_flush(void)
1745 {
1746 struct xfrm_dst *old;
1747 bool found = 0;
1748 int cpu;
1749
1750 might_sleep();
1751
1752 local_bh_disable();
1753 rcu_read_lock();
1754 for_each_possible_cpu(cpu) {
1755 old = per_cpu(xfrm_last_dst, cpu);
1756 if (old && !xfrm_bundle_ok(old)) {
1757 if (smp_processor_id() == cpu) {
1758 __xfrm_pcpu_work_fn();
1759 continue;
1760 }
1761 found = true;
1762 break;
1763 }
1764 }
1765
1766 rcu_read_unlock();
1767 local_bh_enable();
1768
1769 if (!found)
1770 return;
1771
1772 get_online_cpus();
1773
1774 for_each_possible_cpu(cpu) {
1775 bool bundle_release;
1776
1777 rcu_read_lock();
1778 old = per_cpu(xfrm_last_dst, cpu);
1779 bundle_release = old && !xfrm_bundle_ok(old);
1780 rcu_read_unlock();
1781
1782 if (!bundle_release)
1783 continue;
1784
1785 if (cpu_online(cpu)) {
1786 schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1787 continue;
1788 }
1789
1790 rcu_read_lock();
1791 old = per_cpu(xfrm_last_dst, cpu);
1792 if (old && !xfrm_bundle_ok(old)) {
1793 per_cpu(xfrm_last_dst, cpu) = NULL;
1794 dst_release(&old->u.dst);
1795 }
1796 rcu_read_unlock();
1797 }
1798
1799 put_online_cpus();
1800 }
1801
1802 static bool xfrm_xdst_can_reuse(struct xfrm_dst *xdst,
1803 struct xfrm_state * const xfrm[],
1804 int num)
1805 {
1806 const struct dst_entry *dst = &xdst->u.dst;
1807 int i;
1808
1809 if (xdst->num_xfrms != num)
1810 return false;
1811
1812 for (i = 0; i < num; i++) {
1813 if (!dst || dst->xfrm != xfrm[i])
1814 return false;
1815 dst = dst->child;
1816 }
1817
1818 return xfrm_bundle_ok(xdst);
1819 }
1820
1821 static struct xfrm_dst *
1822 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1823 const struct flowi *fl, u16 family,
1824 struct dst_entry *dst_orig)
1825 {
1826 struct net *net = xp_net(pols[0]);
1827 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1828 struct xfrm_dst *xdst, *old;
1829 struct dst_entry *dst;
1830 int err;
1831
1832 /* Try to instantiate a bundle */
1833 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1834 if (err <= 0) {
1835 if (err == 0)
1836 return NULL;
1837
1838 if (err != -EAGAIN)
1839 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1840 return ERR_PTR(err);
1841 }
1842
1843 xdst = this_cpu_read(xfrm_last_dst);
1844 if (xdst &&
1845 xdst->u.dst.dev == dst_orig->dev &&
1846 xdst->num_pols == num_pols &&
1847 memcmp(xdst->pols, pols,
1848 sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1849 xfrm_xdst_can_reuse(xdst, xfrm, err)) {
1850 dst_hold(&xdst->u.dst);
1851 while (err > 0)
1852 xfrm_state_put(xfrm[--err]);
1853 return xdst;
1854 }
1855
1856 old = xdst;
1857
1858 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1859 if (IS_ERR(dst)) {
1860 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1861 return ERR_CAST(dst);
1862 }
1863
1864 xdst = (struct xfrm_dst *)dst;
1865 xdst->num_xfrms = err;
1866 xdst->num_pols = num_pols;
1867 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1868 xdst->policy_genid = atomic_read(&pols[0]->genid);
1869
1870 atomic_set(&xdst->u.dst.__refcnt, 2);
1871 xfrm_last_dst_update(xdst, old);
1872
1873 return xdst;
1874 }
1875
1876 static void xfrm_policy_queue_process(unsigned long arg)
1877 {
1878 struct sk_buff *skb;
1879 struct sock *sk;
1880 struct dst_entry *dst;
1881 struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1882 struct net *net = xp_net(pol);
1883 struct xfrm_policy_queue *pq = &pol->polq;
1884 struct flowi fl;
1885 struct sk_buff_head list;
1886
1887 spin_lock(&pq->hold_queue.lock);
1888 skb = skb_peek(&pq->hold_queue);
1889 if (!skb) {
1890 spin_unlock(&pq->hold_queue.lock);
1891 goto out;
1892 }
1893 dst = skb_dst(skb);
1894 sk = skb->sk;
1895 xfrm_decode_session(skb, &fl, dst->ops->family);
1896 spin_unlock(&pq->hold_queue.lock);
1897
1898 dst_hold(dst->path);
1899 dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1900 if (IS_ERR(dst))
1901 goto purge_queue;
1902
1903 if (dst->flags & DST_XFRM_QUEUE) {
1904 dst_release(dst);
1905
1906 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1907 goto purge_queue;
1908
1909 pq->timeout = pq->timeout << 1;
1910 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1911 xfrm_pol_hold(pol);
1912 goto out;
1913 }
1914
1915 dst_release(dst);
1916
1917 __skb_queue_head_init(&list);
1918
1919 spin_lock(&pq->hold_queue.lock);
1920 pq->timeout = 0;
1921 skb_queue_splice_init(&pq->hold_queue, &list);
1922 spin_unlock(&pq->hold_queue.lock);
1923
1924 while (!skb_queue_empty(&list)) {
1925 skb = __skb_dequeue(&list);
1926
1927 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1928 dst_hold(skb_dst(skb)->path);
1929 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1930 if (IS_ERR(dst)) {
1931 kfree_skb(skb);
1932 continue;
1933 }
1934
1935 nf_reset(skb);
1936 skb_dst_drop(skb);
1937 skb_dst_set(skb, dst);
1938
1939 dst_output(net, skb->sk, skb);
1940 }
1941
1942 out:
1943 xfrm_pol_put(pol);
1944 return;
1945
1946 purge_queue:
1947 pq->timeout = 0;
1948 skb_queue_purge(&pq->hold_queue);
1949 xfrm_pol_put(pol);
1950 }
1951
1952 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1953 {
1954 unsigned long sched_next;
1955 struct dst_entry *dst = skb_dst(skb);
1956 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1957 struct xfrm_policy *pol = xdst->pols[0];
1958 struct xfrm_policy_queue *pq = &pol->polq;
1959
1960 if (unlikely(skb_fclone_busy(sk, skb))) {
1961 kfree_skb(skb);
1962 return 0;
1963 }
1964
1965 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1966 kfree_skb(skb);
1967 return -EAGAIN;
1968 }
1969
1970 skb_dst_force(skb);
1971
1972 spin_lock_bh(&pq->hold_queue.lock);
1973
1974 if (!pq->timeout)
1975 pq->timeout = XFRM_QUEUE_TMO_MIN;
1976
1977 sched_next = jiffies + pq->timeout;
1978
1979 if (del_timer(&pq->hold_timer)) {
1980 if (time_before(pq->hold_timer.expires, sched_next))
1981 sched_next = pq->hold_timer.expires;
1982 xfrm_pol_put(pol);
1983 }
1984
1985 __skb_queue_tail(&pq->hold_queue, skb);
1986 if (!mod_timer(&pq->hold_timer, sched_next))
1987 xfrm_pol_hold(pol);
1988
1989 spin_unlock_bh(&pq->hold_queue.lock);
1990
1991 return 0;
1992 }
1993
1994 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1995 struct xfrm_flo *xflo,
1996 const struct flowi *fl,
1997 int num_xfrms,
1998 u16 family)
1999 {
2000 int err;
2001 struct net_device *dev;
2002 struct dst_entry *dst;
2003 struct dst_entry *dst1;
2004 struct xfrm_dst *xdst;
2005
2006 xdst = xfrm_alloc_dst(net, family);
2007 if (IS_ERR(xdst))
2008 return xdst;
2009
2010 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2011 net->xfrm.sysctl_larval_drop ||
2012 num_xfrms <= 0)
2013 return xdst;
2014
2015 dst = xflo->dst_orig;
2016 dst1 = &xdst->u.dst;
2017 dst_hold(dst);
2018 xdst->route = dst;
2019
2020 dst_copy_metrics(dst1, dst);
2021
2022 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2023 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2024 dst1->lastuse = jiffies;
2025
2026 dst1->input = dst_discard;
2027 dst1->output = xdst_queue_output;
2028
2029 dst_hold(dst);
2030 dst1->child = dst;
2031 dst1->path = dst;
2032
2033 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2034
2035 err = -ENODEV;
2036 dev = dst->dev;
2037 if (!dev)
2038 goto free_dst;
2039
2040 err = xfrm_fill_dst(xdst, dev, fl);
2041 if (err)
2042 goto free_dst;
2043
2044 out:
2045 return xdst;
2046
2047 free_dst:
2048 dst_release(dst1);
2049 xdst = ERR_PTR(err);
2050 goto out;
2051 }
2052
2053 static struct xfrm_dst *
2054 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2055 {
2056 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2057 int num_pols = 0, num_xfrms = 0, err;
2058 struct xfrm_dst *xdst;
2059
2060 /* Resolve policies to use if we couldn't get them from
2061 * previous cache entry */
2062 num_pols = 1;
2063 pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2064 err = xfrm_expand_policies(fl, family, pols,
2065 &num_pols, &num_xfrms);
2066 if (err < 0)
2067 goto inc_error;
2068 if (num_pols == 0)
2069 return NULL;
2070 if (num_xfrms <= 0)
2071 goto make_dummy_bundle;
2072
2073 local_bh_disable();
2074 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2075 xflo->dst_orig);
2076 local_bh_enable();
2077
2078 if (IS_ERR(xdst)) {
2079 err = PTR_ERR(xdst);
2080 if (err != -EAGAIN)
2081 goto error;
2082 goto make_dummy_bundle;
2083 } else if (xdst == NULL) {
2084 num_xfrms = 0;
2085 goto make_dummy_bundle;
2086 }
2087
2088 return xdst;
2089
2090 make_dummy_bundle:
2091 /* We found policies, but there's no bundles to instantiate:
2092 * either because the policy blocks, has no transformations or
2093 * we could not build template (no xfrm_states).*/
2094 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2095 if (IS_ERR(xdst)) {
2096 xfrm_pols_put(pols, num_pols);
2097 return ERR_CAST(xdst);
2098 }
2099 xdst->num_pols = num_pols;
2100 xdst->num_xfrms = num_xfrms;
2101 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2102
2103 return xdst;
2104
2105 inc_error:
2106 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2107 error:
2108 xfrm_pols_put(pols, num_pols);
2109 return ERR_PTR(err);
2110 }
2111
2112 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2113 struct dst_entry *dst_orig)
2114 {
2115 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2116 struct dst_entry *ret;
2117
2118 if (!afinfo) {
2119 dst_release(dst_orig);
2120 return ERR_PTR(-EINVAL);
2121 } else {
2122 ret = afinfo->blackhole_route(net, dst_orig);
2123 }
2124 rcu_read_unlock();
2125
2126 return ret;
2127 }
2128
2129 /* Main function: finds/creates a bundle for given flow.
2130 *
2131 * At the moment we eat a raw IP route. Mostly to speed up lookups
2132 * on interfaces with disabled IPsec.
2133 */
2134 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2135 const struct flowi *fl,
2136 const struct sock *sk, int flags)
2137 {
2138 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2139 struct xfrm_dst *xdst;
2140 struct dst_entry *dst, *route;
2141 u16 family = dst_orig->ops->family;
2142 u8 dir = XFRM_POLICY_OUT;
2143 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2144
2145 dst = NULL;
2146 xdst = NULL;
2147 route = NULL;
2148
2149 sk = sk_const_to_full_sk(sk);
2150 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2151 num_pols = 1;
2152 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2153 err = xfrm_expand_policies(fl, family, pols,
2154 &num_pols, &num_xfrms);
2155 if (err < 0)
2156 goto dropdst;
2157
2158 if (num_pols) {
2159 if (num_xfrms <= 0) {
2160 drop_pols = num_pols;
2161 goto no_transform;
2162 }
2163
2164 local_bh_disable();
2165 xdst = xfrm_resolve_and_create_bundle(
2166 pols, num_pols, fl,
2167 family, dst_orig);
2168 local_bh_enable();
2169
2170 if (IS_ERR(xdst)) {
2171 xfrm_pols_put(pols, num_pols);
2172 err = PTR_ERR(xdst);
2173 goto dropdst;
2174 } else if (xdst == NULL) {
2175 num_xfrms = 0;
2176 drop_pols = num_pols;
2177 goto no_transform;
2178 }
2179
2180 route = xdst->route;
2181 }
2182 }
2183
2184 if (xdst == NULL) {
2185 struct xfrm_flo xflo;
2186
2187 xflo.dst_orig = dst_orig;
2188 xflo.flags = flags;
2189
2190 /* To accelerate a bit... */
2191 if ((dst_orig->flags & DST_NOXFRM) ||
2192 !net->xfrm.policy_count[XFRM_POLICY_OUT])
2193 goto nopol;
2194
2195 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2196 if (xdst == NULL)
2197 goto nopol;
2198 if (IS_ERR(xdst)) {
2199 err = PTR_ERR(xdst);
2200 goto dropdst;
2201 }
2202
2203 num_pols = xdst->num_pols;
2204 num_xfrms = xdst->num_xfrms;
2205 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2206 route = xdst->route;
2207 }
2208
2209 dst = &xdst->u.dst;
2210 if (route == NULL && num_xfrms > 0) {
2211 /* The only case when xfrm_bundle_lookup() returns a
2212 * bundle with null route, is when the template could
2213 * not be resolved. It means policies are there, but
2214 * bundle could not be created, since we don't yet
2215 * have the xfrm_state's. We need to wait for KM to
2216 * negotiate new SA's or bail out with error.*/
2217 if (net->xfrm.sysctl_larval_drop) {
2218 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2219 err = -EREMOTE;
2220 goto error;
2221 }
2222
2223 err = -EAGAIN;
2224
2225 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2226 goto error;
2227 }
2228
2229 no_transform:
2230 if (num_pols == 0)
2231 goto nopol;
2232
2233 if ((flags & XFRM_LOOKUP_ICMP) &&
2234 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2235 err = -ENOENT;
2236 goto error;
2237 }
2238
2239 for (i = 0; i < num_pols; i++)
2240 pols[i]->curlft.use_time = get_seconds();
2241
2242 if (num_xfrms < 0) {
2243 /* Prohibit the flow */
2244 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2245 err = -EPERM;
2246 goto error;
2247 } else if (num_xfrms > 0) {
2248 /* Flow transformed */
2249 dst_release(dst_orig);
2250 } else {
2251 /* Flow passes untransformed */
2252 dst_release(dst);
2253 dst = dst_orig;
2254 }
2255 ok:
2256 xfrm_pols_put(pols, drop_pols);
2257 if (dst && dst->xfrm &&
2258 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2259 dst->flags |= DST_XFRM_TUNNEL;
2260 return dst;
2261
2262 nopol:
2263 if (!(flags & XFRM_LOOKUP_ICMP)) {
2264 dst = dst_orig;
2265 goto ok;
2266 }
2267 err = -ENOENT;
2268 error:
2269 dst_release(dst);
2270 dropdst:
2271 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2272 dst_release(dst_orig);
2273 xfrm_pols_put(pols, drop_pols);
2274 return ERR_PTR(err);
2275 }
2276 EXPORT_SYMBOL(xfrm_lookup);
2277
2278 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2279 * Otherwise we may send out blackholed packets.
2280 */
2281 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2282 const struct flowi *fl,
2283 const struct sock *sk, int flags)
2284 {
2285 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2286 flags | XFRM_LOOKUP_QUEUE |
2287 XFRM_LOOKUP_KEEP_DST_REF);
2288
2289 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2290 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2291
2292 if (IS_ERR(dst))
2293 dst_release(dst_orig);
2294
2295 return dst;
2296 }
2297 EXPORT_SYMBOL(xfrm_lookup_route);
2298
2299 static inline int
2300 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2301 {
2302 struct xfrm_state *x;
2303
2304 if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2305 return 0;
2306 x = skb->sp->xvec[idx];
2307 if (!x->type->reject)
2308 return 0;
2309 return x->type->reject(x, skb, fl);
2310 }
2311
2312 /* When skb is transformed back to its "native" form, we have to
2313 * check policy restrictions. At the moment we make this in maximally
2314 * stupid way. Shame on me. :-) Of course, connected sockets must
2315 * have policy cached at them.
2316 */
2317
2318 static inline int
2319 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2320 unsigned short family)
2321 {
2322 if (xfrm_state_kern(x))
2323 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2324 return x->id.proto == tmpl->id.proto &&
2325 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2326 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2327 x->props.mode == tmpl->mode &&
2328 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2329 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2330 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2331 xfrm_state_addr_cmp(tmpl, x, family));
2332 }
2333
2334 /*
2335 * 0 or more than 0 is returned when validation is succeeded (either bypass
2336 * because of optional transport mode, or next index of the mathced secpath
2337 * state with the template.
2338 * -1 is returned when no matching template is found.
2339 * Otherwise "-2 - errored_index" is returned.
2340 */
2341 static inline int
2342 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2343 unsigned short family)
2344 {
2345 int idx = start;
2346
2347 if (tmpl->optional) {
2348 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2349 return start;
2350 } else
2351 start = -1;
2352 for (; idx < sp->len; idx++) {
2353 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2354 return ++idx;
2355 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2356 if (start == -1)
2357 start = -2-idx;
2358 break;
2359 }
2360 }
2361 return start;
2362 }
2363
2364 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2365 unsigned int family, int reverse)
2366 {
2367 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2368 int err;
2369
2370 if (unlikely(afinfo == NULL))
2371 return -EAFNOSUPPORT;
2372
2373 afinfo->decode_session(skb, fl, reverse);
2374 err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2375 rcu_read_unlock();
2376 return err;
2377 }
2378 EXPORT_SYMBOL(__xfrm_decode_session);
2379
2380 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2381 {
2382 for (; k < sp->len; k++) {
2383 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2384 *idxp = k;
2385 return 1;
2386 }
2387 }
2388
2389 return 0;
2390 }
2391
2392 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2393 unsigned short family)
2394 {
2395 struct net *net = dev_net(skb->dev);
2396 struct xfrm_policy *pol;
2397 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2398 int npols = 0;
2399 int xfrm_nr;
2400 int pi;
2401 int reverse;
2402 struct flowi fl;
2403 int xerr_idx = -1;
2404
2405 reverse = dir & ~XFRM_POLICY_MASK;
2406 dir &= XFRM_POLICY_MASK;
2407
2408 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2409 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2410 return 0;
2411 }
2412
2413 nf_nat_decode_session(skb, &fl, family);
2414
2415 /* First, check used SA against their selectors. */
2416 if (skb->sp) {
2417 int i;
2418
2419 for (i = skb->sp->len-1; i >= 0; i--) {
2420 struct xfrm_state *x = skb->sp->xvec[i];
2421 if (!xfrm_selector_match(&x->sel, &fl, family)) {
2422 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2423 return 0;
2424 }
2425 }
2426 }
2427
2428 pol = NULL;
2429 sk = sk_to_full_sk(sk);
2430 if (sk && sk->sk_policy[dir]) {
2431 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2432 if (IS_ERR(pol)) {
2433 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2434 return 0;
2435 }
2436 }
2437
2438 if (!pol)
2439 pol = xfrm_policy_lookup(net, &fl, family, dir);
2440
2441 if (IS_ERR(pol)) {
2442 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2443 return 0;
2444 }
2445
2446 if (!pol) {
2447 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2448 xfrm_secpath_reject(xerr_idx, skb, &fl);
2449 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2450 return 0;
2451 }
2452 return 1;
2453 }
2454
2455 pol->curlft.use_time = get_seconds();
2456
2457 pols[0] = pol;
2458 npols++;
2459 #ifdef CONFIG_XFRM_SUB_POLICY
2460 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2461 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2462 &fl, family,
2463 XFRM_POLICY_IN);
2464 if (pols[1]) {
2465 if (IS_ERR(pols[1])) {
2466 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2467 return 0;
2468 }
2469 pols[1]->curlft.use_time = get_seconds();
2470 npols++;
2471 }
2472 }
2473 #endif
2474
2475 if (pol->action == XFRM_POLICY_ALLOW) {
2476 struct sec_path *sp;
2477 static struct sec_path dummy;
2478 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2479 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2480 struct xfrm_tmpl **tpp = tp;
2481 int ti = 0;
2482 int i, k;
2483
2484 if ((sp = skb->sp) == NULL)
2485 sp = &dummy;
2486
2487 for (pi = 0; pi < npols; pi++) {
2488 if (pols[pi] != pol &&
2489 pols[pi]->action != XFRM_POLICY_ALLOW) {
2490 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2491 goto reject;
2492 }
2493 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2494 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2495 goto reject_error;
2496 }
2497 for (i = 0; i < pols[pi]->xfrm_nr; i++)
2498 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2499 }
2500 xfrm_nr = ti;
2501 if (npols > 1) {
2502 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2503 tpp = stp;
2504 }
2505
2506 /* For each tunnel xfrm, find the first matching tmpl.
2507 * For each tmpl before that, find corresponding xfrm.
2508 * Order is _important_. Later we will implement
2509 * some barriers, but at the moment barriers
2510 * are implied between each two transformations.
2511 */
2512 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2513 k = xfrm_policy_ok(tpp[i], sp, k, family);
2514 if (k < 0) {
2515 if (k < -1)
2516 /* "-2 - errored_index" returned */
2517 xerr_idx = -(2+k);
2518 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2519 goto reject;
2520 }
2521 }
2522
2523 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2524 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2525 goto reject;
2526 }
2527
2528 xfrm_pols_put(pols, npols);
2529 return 1;
2530 }
2531 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2532
2533 reject:
2534 xfrm_secpath_reject(xerr_idx, skb, &fl);
2535 reject_error:
2536 xfrm_pols_put(pols, npols);
2537 return 0;
2538 }
2539 EXPORT_SYMBOL(__xfrm_policy_check);
2540
2541 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2542 {
2543 struct net *net = dev_net(skb->dev);
2544 struct flowi fl;
2545 struct dst_entry *dst;
2546 int res = 1;
2547
2548 if (xfrm_decode_session(skb, &fl, family) < 0) {
2549 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2550 return 0;
2551 }
2552
2553 skb_dst_force(skb);
2554 if (!skb_dst(skb)) {
2555 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2556 return 0;
2557 }
2558
2559 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2560 if (IS_ERR(dst)) {
2561 res = 0;
2562 dst = NULL;
2563 }
2564 skb_dst_set(skb, dst);
2565 return res;
2566 }
2567 EXPORT_SYMBOL(__xfrm_route_forward);
2568
2569 /* Optimize later using cookies and generation ids. */
2570
2571 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2572 {
2573 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2574 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2575 * get validated by dst_ops->check on every use. We do this
2576 * because when a normal route referenced by an XFRM dst is
2577 * obsoleted we do not go looking around for all parent
2578 * referencing XFRM dsts so that we can invalidate them. It
2579 * is just too much work. Instead we make the checks here on
2580 * every use. For example:
2581 *
2582 * XFRM dst A --> IPv4 dst X
2583 *
2584 * X is the "xdst->route" of A (X is also the "dst->path" of A
2585 * in this example). If X is marked obsolete, "A" will not
2586 * notice. That's what we are validating here via the
2587 * stale_bundle() check.
2588 *
2589 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2590 * be marked on it.
2591 * This will force stale_bundle() to fail on any xdst bundle with
2592 * this dst linked in it.
2593 */
2594 if (dst->obsolete < 0 && !stale_bundle(dst))
2595 return dst;
2596
2597 return NULL;
2598 }
2599
2600 static int stale_bundle(struct dst_entry *dst)
2601 {
2602 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2603 }
2604
2605 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2606 {
2607 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2608 dst->dev = dev_net(dev)->loopback_dev;
2609 dev_hold(dst->dev);
2610 dev_put(dev);
2611 }
2612 }
2613 EXPORT_SYMBOL(xfrm_dst_ifdown);
2614
2615 static void xfrm_link_failure(struct sk_buff *skb)
2616 {
2617 /* Impossible. Such dst must be popped before reaches point of failure. */
2618 }
2619
2620 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2621 {
2622 if (dst) {
2623 if (dst->obsolete) {
2624 dst_release(dst);
2625 dst = NULL;
2626 }
2627 }
2628 return dst;
2629 }
2630
2631 static void xfrm_init_pmtu(struct dst_entry *dst)
2632 {
2633 do {
2634 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2635 u32 pmtu, route_mtu_cached;
2636
2637 pmtu = dst_mtu(dst->child);
2638 xdst->child_mtu_cached = pmtu;
2639
2640 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2641
2642 route_mtu_cached = dst_mtu(xdst->route);
2643 xdst->route_mtu_cached = route_mtu_cached;
2644
2645 if (pmtu > route_mtu_cached)
2646 pmtu = route_mtu_cached;
2647
2648 dst_metric_set(dst, RTAX_MTU, pmtu);
2649 } while ((dst = dst->next));
2650 }
2651
2652 /* Check that the bundle accepts the flow and its components are
2653 * still valid.
2654 */
2655
2656 static int xfrm_bundle_ok(struct xfrm_dst *first)
2657 {
2658 struct dst_entry *dst = &first->u.dst;
2659 struct xfrm_dst *last;
2660 u32 mtu;
2661
2662 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2663 (dst->dev && !netif_running(dst->dev)))
2664 return 0;
2665
2666 if (dst->flags & DST_XFRM_QUEUE)
2667 return 1;
2668
2669 last = NULL;
2670
2671 do {
2672 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2673
2674 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2675 return 0;
2676 if (xdst->xfrm_genid != dst->xfrm->genid)
2677 return 0;
2678 if (xdst->num_pols > 0 &&
2679 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2680 return 0;
2681
2682 mtu = dst_mtu(dst->child);
2683 if (xdst->child_mtu_cached != mtu) {
2684 last = xdst;
2685 xdst->child_mtu_cached = mtu;
2686 }
2687
2688 if (!dst_check(xdst->route, xdst->route_cookie))
2689 return 0;
2690 mtu = dst_mtu(xdst->route);
2691 if (xdst->route_mtu_cached != mtu) {
2692 last = xdst;
2693 xdst->route_mtu_cached = mtu;
2694 }
2695
2696 dst = dst->child;
2697 } while (dst->xfrm);
2698
2699 if (likely(!last))
2700 return 1;
2701
2702 mtu = last->child_mtu_cached;
2703 for (;;) {
2704 dst = &last->u.dst;
2705
2706 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2707 if (mtu > last->route_mtu_cached)
2708 mtu = last->route_mtu_cached;
2709 dst_metric_set(dst, RTAX_MTU, mtu);
2710
2711 if (last == first)
2712 break;
2713
2714 last = (struct xfrm_dst *)last->u.dst.next;
2715 last->child_mtu_cached = mtu;
2716 }
2717
2718 return 1;
2719 }
2720
2721 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2722 {
2723 return dst_metric_advmss(dst->path);
2724 }
2725
2726 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2727 {
2728 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2729
2730 return mtu ? : dst_mtu(dst->path);
2731 }
2732
2733 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2734 const void *daddr)
2735 {
2736 const struct dst_entry *path = dst->path;
2737
2738 for (; dst != path; dst = dst->child) {
2739 const struct xfrm_state *xfrm = dst->xfrm;
2740
2741 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2742 continue;
2743 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2744 daddr = xfrm->coaddr;
2745 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2746 daddr = &xfrm->id.daddr;
2747 }
2748 return daddr;
2749 }
2750
2751 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2752 struct sk_buff *skb,
2753 const void *daddr)
2754 {
2755 const struct dst_entry *path = dst->path;
2756
2757 if (!skb)
2758 daddr = xfrm_get_dst_nexthop(dst, daddr);
2759 return path->ops->neigh_lookup(path, skb, daddr);
2760 }
2761
2762 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2763 {
2764 const struct dst_entry *path = dst->path;
2765
2766 daddr = xfrm_get_dst_nexthop(dst, daddr);
2767 path->ops->confirm_neigh(path, daddr);
2768 }
2769
2770 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2771 {
2772 int err = 0;
2773
2774 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2775 return -EAFNOSUPPORT;
2776
2777 spin_lock(&xfrm_policy_afinfo_lock);
2778 if (unlikely(xfrm_policy_afinfo[family] != NULL))
2779 err = -EEXIST;
2780 else {
2781 struct dst_ops *dst_ops = afinfo->dst_ops;
2782 if (likely(dst_ops->kmem_cachep == NULL))
2783 dst_ops->kmem_cachep = xfrm_dst_cache;
2784 if (likely(dst_ops->check == NULL))
2785 dst_ops->check = xfrm_dst_check;
2786 if (likely(dst_ops->default_advmss == NULL))
2787 dst_ops->default_advmss = xfrm_default_advmss;
2788 if (likely(dst_ops->mtu == NULL))
2789 dst_ops->mtu = xfrm_mtu;
2790 if (likely(dst_ops->negative_advice == NULL))
2791 dst_ops->negative_advice = xfrm_negative_advice;
2792 if (likely(dst_ops->link_failure == NULL))
2793 dst_ops->link_failure = xfrm_link_failure;
2794 if (likely(dst_ops->neigh_lookup == NULL))
2795 dst_ops->neigh_lookup = xfrm_neigh_lookup;
2796 if (likely(!dst_ops->confirm_neigh))
2797 dst_ops->confirm_neigh = xfrm_confirm_neigh;
2798 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2799 }
2800 spin_unlock(&xfrm_policy_afinfo_lock);
2801
2802 return err;
2803 }
2804 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2805
2806 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2807 {
2808 struct dst_ops *dst_ops = afinfo->dst_ops;
2809 int i;
2810
2811 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2812 if (xfrm_policy_afinfo[i] != afinfo)
2813 continue;
2814 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2815 break;
2816 }
2817
2818 synchronize_rcu();
2819
2820 dst_ops->kmem_cachep = NULL;
2821 dst_ops->check = NULL;
2822 dst_ops->negative_advice = NULL;
2823 dst_ops->link_failure = NULL;
2824 }
2825 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2826
2827 #ifdef CONFIG_XFRM_STATISTICS
2828 static int __net_init xfrm_statistics_init(struct net *net)
2829 {
2830 int rv;
2831 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2832 if (!net->mib.xfrm_statistics)
2833 return -ENOMEM;
2834 rv = xfrm_proc_init(net);
2835 if (rv < 0)
2836 free_percpu(net->mib.xfrm_statistics);
2837 return rv;
2838 }
2839
2840 static void xfrm_statistics_fini(struct net *net)
2841 {
2842 xfrm_proc_fini(net);
2843 free_percpu(net->mib.xfrm_statistics);
2844 }
2845 #else
2846 static int __net_init xfrm_statistics_init(struct net *net)
2847 {
2848 return 0;
2849 }
2850
2851 static void xfrm_statistics_fini(struct net *net)
2852 {
2853 }
2854 #endif
2855
2856 static int __net_init xfrm_policy_init(struct net *net)
2857 {
2858 unsigned int hmask, sz;
2859 int dir;
2860
2861 if (net_eq(net, &init_net))
2862 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2863 sizeof(struct xfrm_dst),
2864 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2865 NULL);
2866
2867 hmask = 8 - 1;
2868 sz = (hmask+1) * sizeof(struct hlist_head);
2869
2870 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2871 if (!net->xfrm.policy_byidx)
2872 goto out_byidx;
2873 net->xfrm.policy_idx_hmask = hmask;
2874
2875 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2876 struct xfrm_policy_hash *htab;
2877
2878 net->xfrm.policy_count[dir] = 0;
2879 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2880 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2881
2882 htab = &net->xfrm.policy_bydst[dir];
2883 htab->table = xfrm_hash_alloc(sz);
2884 if (!htab->table)
2885 goto out_bydst;
2886 htab->hmask = hmask;
2887 htab->dbits4 = 32;
2888 htab->sbits4 = 32;
2889 htab->dbits6 = 128;
2890 htab->sbits6 = 128;
2891 }
2892 net->xfrm.policy_hthresh.lbits4 = 32;
2893 net->xfrm.policy_hthresh.rbits4 = 32;
2894 net->xfrm.policy_hthresh.lbits6 = 128;
2895 net->xfrm.policy_hthresh.rbits6 = 128;
2896
2897 seqlock_init(&net->xfrm.policy_hthresh.lock);
2898
2899 INIT_LIST_HEAD(&net->xfrm.policy_all);
2900 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2901 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2902 if (net_eq(net, &init_net))
2903 xfrm_dev_init();
2904 return 0;
2905
2906 out_bydst:
2907 for (dir--; dir >= 0; dir--) {
2908 struct xfrm_policy_hash *htab;
2909
2910 htab = &net->xfrm.policy_bydst[dir];
2911 xfrm_hash_free(htab->table, sz);
2912 }
2913 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2914 out_byidx:
2915 return -ENOMEM;
2916 }
2917
2918 static void xfrm_policy_fini(struct net *net)
2919 {
2920 unsigned int sz;
2921 int dir;
2922
2923 flush_work(&net->xfrm.policy_hash_work);
2924 #ifdef CONFIG_XFRM_SUB_POLICY
2925 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2926 #endif
2927 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2928
2929 WARN_ON(!list_empty(&net->xfrm.policy_all));
2930
2931 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2932 struct xfrm_policy_hash *htab;
2933
2934 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2935
2936 htab = &net->xfrm.policy_bydst[dir];
2937 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2938 WARN_ON(!hlist_empty(htab->table));
2939 xfrm_hash_free(htab->table, sz);
2940 }
2941
2942 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2943 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2944 xfrm_hash_free(net->xfrm.policy_byidx, sz);
2945 }
2946
2947 static int __net_init xfrm_net_init(struct net *net)
2948 {
2949 int rv;
2950
2951 /* Initialize the per-net locks here */
2952 spin_lock_init(&net->xfrm.xfrm_state_lock);
2953 spin_lock_init(&net->xfrm.xfrm_policy_lock);
2954 mutex_init(&net->xfrm.xfrm_cfg_mutex);
2955
2956 rv = xfrm_statistics_init(net);
2957 if (rv < 0)
2958 goto out_statistics;
2959 rv = xfrm_state_init(net);
2960 if (rv < 0)
2961 goto out_state;
2962 rv = xfrm_policy_init(net);
2963 if (rv < 0)
2964 goto out_policy;
2965 rv = xfrm_sysctl_init(net);
2966 if (rv < 0)
2967 goto out_sysctl;
2968
2969 return 0;
2970
2971 out_sysctl:
2972 xfrm_policy_fini(net);
2973 out_policy:
2974 xfrm_state_fini(net);
2975 out_state:
2976 xfrm_statistics_fini(net);
2977 out_statistics:
2978 return rv;
2979 }
2980
2981 static void __net_exit xfrm_net_exit(struct net *net)
2982 {
2983 xfrm_sysctl_fini(net);
2984 xfrm_policy_fini(net);
2985 xfrm_state_fini(net);
2986 xfrm_statistics_fini(net);
2987 }
2988
2989 static struct pernet_operations __net_initdata xfrm_net_ops = {
2990 .init = xfrm_net_init,
2991 .exit = xfrm_net_exit,
2992 };
2993
2994 void __init xfrm_init(void)
2995 {
2996 int i;
2997
2998 xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2999 GFP_KERNEL);
3000 BUG_ON(!xfrm_pcpu_work);
3001
3002 for (i = 0; i < NR_CPUS; i++)
3003 INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
3004
3005 register_pernet_subsys(&xfrm_net_ops);
3006 seqcount_init(&xfrm_policy_hash_generation);
3007 xfrm_input_init();
3008 }
3009
3010 #ifdef CONFIG_AUDITSYSCALL
3011 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3012 struct audit_buffer *audit_buf)
3013 {
3014 struct xfrm_sec_ctx *ctx = xp->security;
3015 struct xfrm_selector *sel = &xp->selector;
3016
3017 if (ctx)
3018 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3019 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3020
3021 switch (sel->family) {
3022 case AF_INET:
3023 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3024 if (sel->prefixlen_s != 32)
3025 audit_log_format(audit_buf, " src_prefixlen=%d",
3026 sel->prefixlen_s);
3027 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3028 if (sel->prefixlen_d != 32)
3029 audit_log_format(audit_buf, " dst_prefixlen=%d",
3030 sel->prefixlen_d);
3031 break;
3032 case AF_INET6:
3033 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3034 if (sel->prefixlen_s != 128)
3035 audit_log_format(audit_buf, " src_prefixlen=%d",
3036 sel->prefixlen_s);
3037 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3038 if (sel->prefixlen_d != 128)
3039 audit_log_format(audit_buf, " dst_prefixlen=%d",
3040 sel->prefixlen_d);
3041 break;
3042 }
3043 }
3044
3045 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3046 {
3047 struct audit_buffer *audit_buf;
3048
3049 audit_buf = xfrm_audit_start("SPD-add");
3050 if (audit_buf == NULL)
3051 return;
3052 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3053 audit_log_format(audit_buf, " res=%u", result);
3054 xfrm_audit_common_policyinfo(xp, audit_buf);
3055 audit_log_end(audit_buf);
3056 }
3057 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3058
3059 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3060 bool task_valid)
3061 {
3062 struct audit_buffer *audit_buf;
3063
3064 audit_buf = xfrm_audit_start("SPD-delete");
3065 if (audit_buf == NULL)
3066 return;
3067 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3068 audit_log_format(audit_buf, " res=%u", result);
3069 xfrm_audit_common_policyinfo(xp, audit_buf);
3070 audit_log_end(audit_buf);
3071 }
3072 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3073 #endif
3074
3075 #ifdef CONFIG_XFRM_MIGRATE
3076 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3077 const struct xfrm_selector *sel_tgt)
3078 {
3079 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3080 if (sel_tgt->family == sel_cmp->family &&
3081 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3082 sel_cmp->family) &&
3083 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3084 sel_cmp->family) &&
3085 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3086 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3087 return true;
3088 }
3089 } else {
3090 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3091 return true;
3092 }
3093 }
3094 return false;
3095 }
3096
3097 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3098 u8 dir, u8 type, struct net *net)
3099 {
3100 struct xfrm_policy *pol, *ret = NULL;
3101 struct hlist_head *chain;
3102 u32 priority = ~0U;
3103
3104 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3105 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3106 hlist_for_each_entry(pol, chain, bydst) {
3107 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3108 pol->type == type) {
3109 ret = pol;
3110 priority = ret->priority;
3111 break;
3112 }
3113 }
3114 chain = &net->xfrm.policy_inexact[dir];
3115 hlist_for_each_entry(pol, chain, bydst) {
3116 if ((pol->priority >= priority) && ret)
3117 break;
3118
3119 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3120 pol->type == type) {
3121 ret = pol;
3122 break;
3123 }
3124 }
3125
3126 xfrm_pol_hold(ret);
3127
3128 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3129
3130 return ret;
3131 }
3132
3133 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3134 {
3135 int match = 0;
3136
3137 if (t->mode == m->mode && t->id.proto == m->proto &&
3138 (m->reqid == 0 || t->reqid == m->reqid)) {
3139 switch (t->mode) {
3140 case XFRM_MODE_TUNNEL:
3141 case XFRM_MODE_BEET:
3142 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3143 m->old_family) &&
3144 xfrm_addr_equal(&t->saddr, &m->old_saddr,
3145 m->old_family)) {
3146 match = 1;
3147 }
3148 break;
3149 case XFRM_MODE_TRANSPORT:
3150 /* in case of transport mode, template does not store
3151 any IP addresses, hence we just compare mode and
3152 protocol */
3153 match = 1;
3154 break;
3155 default:
3156 break;
3157 }
3158 }
3159 return match;
3160 }
3161
3162 /* update endpoint address(es) of template(s) */
3163 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3164 struct xfrm_migrate *m, int num_migrate)
3165 {
3166 struct xfrm_migrate *mp;
3167 int i, j, n = 0;
3168
3169 write_lock_bh(&pol->lock);
3170 if (unlikely(pol->walk.dead)) {
3171 /* target policy has been deleted */
3172 write_unlock_bh(&pol->lock);
3173 return -ENOENT;
3174 }
3175
3176 for (i = 0; i < pol->xfrm_nr; i++) {
3177 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3178 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3179 continue;
3180 n++;
3181 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3182 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3183 continue;
3184 /* update endpoints */
3185 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3186 sizeof(pol->xfrm_vec[i].id.daddr));
3187 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3188 sizeof(pol->xfrm_vec[i].saddr));
3189 pol->xfrm_vec[i].encap_family = mp->new_family;
3190 /* flush bundles */
3191 atomic_inc(&pol->genid);
3192 }
3193 }
3194
3195 write_unlock_bh(&pol->lock);
3196
3197 if (!n)
3198 return -ENODATA;
3199
3200 return 0;
3201 }
3202
3203 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3204 {
3205 int i, j;
3206
3207 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3208 return -EINVAL;
3209
3210 for (i = 0; i < num_migrate; i++) {
3211 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3212 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3213 return -EINVAL;
3214
3215 /* check if there is any duplicated entry */
3216 for (j = i + 1; j < num_migrate; j++) {
3217 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3218 sizeof(m[i].old_daddr)) &&
3219 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3220 sizeof(m[i].old_saddr)) &&
3221 m[i].proto == m[j].proto &&
3222 m[i].mode == m[j].mode &&
3223 m[i].reqid == m[j].reqid &&
3224 m[i].old_family == m[j].old_family)
3225 return -EINVAL;
3226 }
3227 }
3228
3229 return 0;
3230 }
3231
3232 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3233 struct xfrm_migrate *m, int num_migrate,
3234 struct xfrm_kmaddress *k, struct net *net,
3235 struct xfrm_encap_tmpl *encap)
3236 {
3237 int i, err, nx_cur = 0, nx_new = 0;
3238 struct xfrm_policy *pol = NULL;
3239 struct xfrm_state *x, *xc;
3240 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3241 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3242 struct xfrm_migrate *mp;
3243
3244 /* Stage 0 - sanity checks */
3245 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3246 goto out;
3247
3248 if (dir >= XFRM_POLICY_MAX) {
3249 err = -EINVAL;
3250 goto out;
3251 }
3252
3253 /* Stage 1 - find policy */
3254 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3255 err = -ENOENT;
3256 goto out;
3257 }
3258
3259 /* Stage 2 - find and update state(s) */
3260 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3261 if ((x = xfrm_migrate_state_find(mp, net))) {
3262 x_cur[nx_cur] = x;
3263 nx_cur++;
3264 xc = xfrm_state_migrate(x, mp, encap);
3265 if (xc) {
3266 x_new[nx_new] = xc;
3267 nx_new++;
3268 } else {
3269 err = -ENODATA;
3270 goto restore_state;
3271 }
3272 }
3273 }
3274
3275 /* Stage 3 - update policy */
3276 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3277 goto restore_state;
3278
3279 /* Stage 4 - delete old state(s) */
3280 if (nx_cur) {
3281 xfrm_states_put(x_cur, nx_cur);
3282 xfrm_states_delete(x_cur, nx_cur);
3283 }
3284
3285 /* Stage 5 - announce */
3286 km_migrate(sel, dir, type, m, num_migrate, k, encap);
3287
3288 xfrm_pol_put(pol);
3289
3290 return 0;
3291 out:
3292 return err;
3293
3294 restore_state:
3295 if (pol)
3296 xfrm_pol_put(pol);
3297 if (nx_cur)
3298 xfrm_states_put(x_cur, nx_cur);
3299 if (nx_new)
3300 xfrm_states_delete(x_new, nx_new);
3301
3302 return err;
3303 }
3304 EXPORT_SYMBOL(xfrm_migrate);
3305 #endif
Motorola kernel-slsi