?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Revert "parisc: Use ldcw instruction for SMP spinlock release barrier"
[linux/fpc-iii.git] / net / xfrm / xfrm_policy.c
blob2917711ff8ab6bb3c6b67cf7e83435296e1303a2
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * xfrm_policy.c
4 *
5 * Changes:
6 * Mitsuru KANDA @USAGI
7 * Kazunori MIYAZAWA @USAGI
8 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
9 * IPv6 support
10 * Kazunori MIYAZAWA @USAGI
11 * YOSHIFUJI Hideaki
12 * Split up af-specific portion
13 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
14 *
15 */
16
17 #include <linux/err.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/workqueue.h>
23 #include <linux/notifier.h>
24 #include <linux/netdevice.h>
25 #include <linux/netfilter.h>
26 #include <linux/module.h>
27 #include <linux/cache.h>
28 #include <linux/cpu.h>
29 #include <linux/audit.h>
30 #include <linux/rhashtable.h>
31 #include <linux/if_tunnel.h>
32 #include <net/dst.h>
33 #include <net/flow.h>
34 #include <net/xfrm.h>
35 #include <net/ip.h>
36 #if IS_ENABLED(CONFIG_IPV6_MIP6)
37 #include <net/mip6.h>
38 #endif
39 #ifdef CONFIG_XFRM_STATISTICS
40 #include <net/snmp.h>
41 #endif
42
43 #include "xfrm_hash.h"
44
45 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
46 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
47 #define XFRM_MAX_QUEUE_LEN 100
48
49 struct xfrm_flo {
50 struct dst_entry *dst_orig;
51 u8 flags;
52 };
53
54 /* prefixes smaller than this are stored in lists, not trees. */
55 #define INEXACT_PREFIXLEN_IPV4 16
56 #define INEXACT_PREFIXLEN_IPV6 48
57
58 struct xfrm_pol_inexact_node {
59 struct rb_node node;
60 union {
61 xfrm_address_t addr;
62 struct rcu_head rcu;
63 };
64 u8 prefixlen;
65
66 struct rb_root root;
67
68 /* the policies matching this node, can be empty list */
69 struct hlist_head hhead;
70 };
71
72 /* xfrm inexact policy search tree:
73 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
74 * |
75 * +---- root_d: sorted by daddr:prefix
76 * | |
77 * | xfrm_pol_inexact_node
78 * | |
79 * | +- root: sorted by saddr/prefix
80 * | | |
81 * | | xfrm_pol_inexact_node
82 * | | |
83 * | | + root: unused
84 * | | |
85 * | | + hhead: saddr:daddr policies
86 * | |
87 * | +- coarse policies and all any:daddr policies
88 * |
89 * +---- root_s: sorted by saddr:prefix
90 * | |
91 * | xfrm_pol_inexact_node
92 * | |
93 * | + root: unused
94 * | |
95 * | + hhead: saddr:any policies
96 * |
97 * +---- coarse policies and all any:any policies
98 *
99 * Lookups return four candidate lists:
100 * 1. any:any list from top-level xfrm_pol_inexact_bin
101 * 2. any:daddr list from daddr tree
102 * 3. saddr:daddr list from 2nd level daddr tree
103 * 4. saddr:any list from saddr tree
104 *
105 * This result set then needs to be searched for the policy with
106 * the lowest priority. If two results have same prio, youngest one wins.
107 */
108
109 struct xfrm_pol_inexact_key {
110 possible_net_t net;
111 u32 if_id;
112 u16 family;
113 u8 dir, type;
114 };
115
116 struct xfrm_pol_inexact_bin {
117 struct xfrm_pol_inexact_key k;
118 struct rhash_head head;
119 /* list containing '*:*' policies */
120 struct hlist_head hhead;
121
122 seqcount_t count;
123 /* tree sorted by daddr/prefix */
124 struct rb_root root_d;
125
126 /* tree sorted by saddr/prefix */
127 struct rb_root root_s;
128
129 /* slow path below */
130 struct list_head inexact_bins;
131 struct rcu_head rcu;
132 };
133
134 enum xfrm_pol_inexact_candidate_type {
135 XFRM_POL_CAND_BOTH,
136 XFRM_POL_CAND_SADDR,
137 XFRM_POL_CAND_DADDR,
138 XFRM_POL_CAND_ANY,
139
140 XFRM_POL_CAND_MAX,
141 };
142
143 struct xfrm_pol_inexact_candidates {
144 struct hlist_head *res[XFRM_POL_CAND_MAX];
145 };
146
147 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
148 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
149
150 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
151 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
152 __read_mostly;
153
154 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
155 static __read_mostly seqcount_t xfrm_policy_hash_generation;
156
157 static struct rhashtable xfrm_policy_inexact_table;
158 static const struct rhashtable_params xfrm_pol_inexact_params;
159
160 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
161 static int stale_bundle(struct dst_entry *dst);
162 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
163 static void xfrm_policy_queue_process(struct timer_list *t);
164
165 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
166 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
167 int dir);
168
169 static struct xfrm_pol_inexact_bin *
170 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
171 u32 if_id);
172
173 static struct xfrm_pol_inexact_bin *
174 xfrm_policy_inexact_lookup_rcu(struct net *net,
175 u8 type, u16 family, u8 dir, u32 if_id);
176 static struct xfrm_policy *
177 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
178 bool excl);
179 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
180 struct xfrm_policy *policy);
181
182 static bool
183 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
184 struct xfrm_pol_inexact_bin *b,
185 const xfrm_address_t *saddr,
186 const xfrm_address_t *daddr);
187
188 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
189 {
190 return refcount_inc_not_zero(&policy->refcnt);
191 }
192
193 static inline bool
194 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
195 {
196 const struct flowi4 *fl4 = &fl->u.ip4;
197
198 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
199 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
200 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
201 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
202 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
203 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
204 }
205
206 static inline bool
207 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
208 {
209 const struct flowi6 *fl6 = &fl->u.ip6;
210
211 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
212 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
213 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
214 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
215 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
216 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
217 }
218
219 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
220 unsigned short family)
221 {
222 switch (family) {
223 case AF_INET:
224 return __xfrm4_selector_match(sel, fl);
225 case AF_INET6:
226 return __xfrm6_selector_match(sel, fl);
227 }
228 return false;
229 }
230
231 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
232 {
233 const struct xfrm_policy_afinfo *afinfo;
234
235 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
236 return NULL;
237 rcu_read_lock();
238 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
239 if (unlikely(!afinfo))
240 rcu_read_unlock();
241 return afinfo;
242 }
243
244 /* Called with rcu_read_lock(). */
245 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
246 {
247 return rcu_dereference(xfrm_if_cb);
248 }
249
250 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
251 const xfrm_address_t *saddr,
252 const xfrm_address_t *daddr,
253 int family, u32 mark)
254 {
255 const struct xfrm_policy_afinfo *afinfo;
256 struct dst_entry *dst;
257
258 afinfo = xfrm_policy_get_afinfo(family);
259 if (unlikely(afinfo == NULL))
260 return ERR_PTR(-EAFNOSUPPORT);
261
262 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
263
264 rcu_read_unlock();
265
266 return dst;
267 }
268 EXPORT_SYMBOL(__xfrm_dst_lookup);
269
270 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
271 int tos, int oif,
272 xfrm_address_t *prev_saddr,
273 xfrm_address_t *prev_daddr,
274 int family, u32 mark)
275 {
276 struct net *net = xs_net(x);
277 xfrm_address_t *saddr = &x->props.saddr;
278 xfrm_address_t *daddr = &x->id.daddr;
279 struct dst_entry *dst;
280
281 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
282 saddr = x->coaddr;
283 daddr = prev_daddr;
284 }
285 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
286 saddr = prev_saddr;
287 daddr = x->coaddr;
288 }
289
290 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
291
292 if (!IS_ERR(dst)) {
293 if (prev_saddr != saddr)
294 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
295 if (prev_daddr != daddr)
296 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
297 }
298
299 return dst;
300 }
301
302 static inline unsigned long make_jiffies(long secs)
303 {
304 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
305 return MAX_SCHEDULE_TIMEOUT-1;
306 else
307 return secs*HZ;
308 }
309
310 static void xfrm_policy_timer(struct timer_list *t)
311 {
312 struct xfrm_policy *xp = from_timer(xp, t, timer);
313 time64_t now = ktime_get_real_seconds();
314 time64_t next = TIME64_MAX;
315 int warn = 0;
316 int dir;
317
318 read_lock(&xp->lock);
319
320 if (unlikely(xp->walk.dead))
321 goto out;
322
323 dir = xfrm_policy_id2dir(xp->index);
324
325 if (xp->lft.hard_add_expires_seconds) {
326 time64_t tmo = xp->lft.hard_add_expires_seconds +
327 xp->curlft.add_time - now;
328 if (tmo <= 0)
329 goto expired;
330 if (tmo < next)
331 next = tmo;
332 }
333 if (xp->lft.hard_use_expires_seconds) {
334 time64_t tmo = xp->lft.hard_use_expires_seconds +
335 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
336 if (tmo <= 0)
337 goto expired;
338 if (tmo < next)
339 next = tmo;
340 }
341 if (xp->lft.soft_add_expires_seconds) {
342 time64_t tmo = xp->lft.soft_add_expires_seconds +
343 xp->curlft.add_time - now;
344 if (tmo <= 0) {
345 warn = 1;
346 tmo = XFRM_KM_TIMEOUT;
347 }
348 if (tmo < next)
349 next = tmo;
350 }
351 if (xp->lft.soft_use_expires_seconds) {
352 time64_t tmo = xp->lft.soft_use_expires_seconds +
353 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
354 if (tmo <= 0) {
355 warn = 1;
356 tmo = XFRM_KM_TIMEOUT;
357 }
358 if (tmo < next)
359 next = tmo;
360 }
361
362 if (warn)
363 km_policy_expired(xp, dir, 0, 0);
364 if (next != TIME64_MAX &&
365 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
366 xfrm_pol_hold(xp);
367
368 out:
369 read_unlock(&xp->lock);
370 xfrm_pol_put(xp);
371 return;
372
373 expired:
374 read_unlock(&xp->lock);
375 if (!xfrm_policy_delete(xp, dir))
376 km_policy_expired(xp, dir, 1, 0);
377 xfrm_pol_put(xp);
378 }
379
380 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
381 * SPD calls.
382 */
383
384 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
385 {
386 struct xfrm_policy *policy;
387
388 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
389
390 if (policy) {
391 write_pnet(&policy->xp_net, net);
392 INIT_LIST_HEAD(&policy->walk.all);
393 INIT_HLIST_NODE(&policy->bydst_inexact_list);
394 INIT_HLIST_NODE(&policy->bydst);
395 INIT_HLIST_NODE(&policy->byidx);
396 rwlock_init(&policy->lock);
397 refcount_set(&policy->refcnt, 1);
398 skb_queue_head_init(&policy->polq.hold_queue);
399 timer_setup(&policy->timer, xfrm_policy_timer, 0);
400 timer_setup(&policy->polq.hold_timer,
401 xfrm_policy_queue_process, 0);
402 }
403 return policy;
404 }
405 EXPORT_SYMBOL(xfrm_policy_alloc);
406
407 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
408 {
409 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
410
411 security_xfrm_policy_free(policy->security);
412 kfree(policy);
413 }
414
415 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
416
417 void xfrm_policy_destroy(struct xfrm_policy *policy)
418 {
419 BUG_ON(!policy->walk.dead);
420
421 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
422 BUG();
423
424 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
425 }
426 EXPORT_SYMBOL(xfrm_policy_destroy);
427
428 /* Rule must be locked. Release descendant resources, announce
429 * entry dead. The rule must be unlinked from lists to the moment.
430 */
431
432 static void xfrm_policy_kill(struct xfrm_policy *policy)
433 {
434 write_lock_bh(&policy->lock);
435 policy->walk.dead = 1;
436 write_unlock_bh(&policy->lock);
437
438 atomic_inc(&policy->genid);
439
440 if (del_timer(&policy->polq.hold_timer))
441 xfrm_pol_put(policy);
442 skb_queue_purge(&policy->polq.hold_queue);
443
444 if (del_timer(&policy->timer))
445 xfrm_pol_put(policy);
446
447 xfrm_pol_put(policy);
448 }
449
450 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
451
452 static inline unsigned int idx_hash(struct net *net, u32 index)
453 {
454 return __idx_hash(index, net->xfrm.policy_idx_hmask);
455 }
456
457 /* calculate policy hash thresholds */
458 static void __get_hash_thresh(struct net *net,
459 unsigned short family, int dir,
460 u8 *dbits, u8 *sbits)
461 {
462 switch (family) {
463 case AF_INET:
464 *dbits = net->xfrm.policy_bydst[dir].dbits4;
465 *sbits = net->xfrm.policy_bydst[dir].sbits4;
466 break;
467
468 case AF_INET6:
469 *dbits = net->xfrm.policy_bydst[dir].dbits6;
470 *sbits = net->xfrm.policy_bydst[dir].sbits6;
471 break;
472
473 default:
474 *dbits = 0;
475 *sbits = 0;
476 }
477 }
478
479 static struct hlist_head *policy_hash_bysel(struct net *net,
480 const struct xfrm_selector *sel,
481 unsigned short family, int dir)
482 {
483 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
484 unsigned int hash;
485 u8 dbits;
486 u8 sbits;
487
488 __get_hash_thresh(net, family, dir, &dbits, &sbits);
489 hash = __sel_hash(sel, family, hmask, dbits, sbits);
490
491 if (hash == hmask + 1)
492 return NULL;
493
494 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
495 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
496 }
497
498 static struct hlist_head *policy_hash_direct(struct net *net,
499 const xfrm_address_t *daddr,
500 const xfrm_address_t *saddr,
501 unsigned short family, int dir)
502 {
503 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
504 unsigned int hash;
505 u8 dbits;
506 u8 sbits;
507
508 __get_hash_thresh(net, family, dir, &dbits, &sbits);
509 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
510
511 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
512 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
513 }
514
515 static void xfrm_dst_hash_transfer(struct net *net,
516 struct hlist_head *list,
517 struct hlist_head *ndsttable,
518 unsigned int nhashmask,
519 int dir)
520 {
521 struct hlist_node *tmp, *entry0 = NULL;
522 struct xfrm_policy *pol;
523 unsigned int h0 = 0;
524 u8 dbits;
525 u8 sbits;
526
527 redo:
528 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
529 unsigned int h;
530
531 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
532 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
533 pol->family, nhashmask, dbits, sbits);
534 if (!entry0) {
535 hlist_del_rcu(&pol->bydst);
536 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
537 h0 = h;
538 } else {
539 if (h != h0)
540 continue;
541 hlist_del_rcu(&pol->bydst);
542 hlist_add_behind_rcu(&pol->bydst, entry0);
543 }
544 entry0 = &pol->bydst;
545 }
546 if (!hlist_empty(list)) {
547 entry0 = NULL;
548 goto redo;
549 }
550 }
551
552 static void xfrm_idx_hash_transfer(struct hlist_head *list,
553 struct hlist_head *nidxtable,
554 unsigned int nhashmask)
555 {
556 struct hlist_node *tmp;
557 struct xfrm_policy *pol;
558
559 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
560 unsigned int h;
561
562 h = __idx_hash(pol->index, nhashmask);
563 hlist_add_head(&pol->byidx, nidxtable+h);
564 }
565 }
566
567 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
568 {
569 return ((old_hmask + 1) << 1) - 1;
570 }
571
572 static void xfrm_bydst_resize(struct net *net, int dir)
573 {
574 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
575 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
576 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
577 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
578 struct hlist_head *odst;
579 int i;
580
581 if (!ndst)
582 return;
583
584 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
585 write_seqcount_begin(&xfrm_policy_hash_generation);
586
587 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
588 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
589
590 for (i = hmask; i >= 0; i--)
591 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
592
593 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
594 net->xfrm.policy_bydst[dir].hmask = nhashmask;
595
596 write_seqcount_end(&xfrm_policy_hash_generation);
597 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
598
599 synchronize_rcu();
600
601 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
602 }
603
604 static void xfrm_byidx_resize(struct net *net, int total)
605 {
606 unsigned int hmask = net->xfrm.policy_idx_hmask;
607 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
608 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
609 struct hlist_head *oidx = net->xfrm.policy_byidx;
610 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
611 int i;
612
613 if (!nidx)
614 return;
615
616 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
617
618 for (i = hmask; i >= 0; i--)
619 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
620
621 net->xfrm.policy_byidx = nidx;
622 net->xfrm.policy_idx_hmask = nhashmask;
623
624 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
625
626 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
627 }
628
629 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
630 {
631 unsigned int cnt = net->xfrm.policy_count[dir];
632 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
633
634 if (total)
635 *total += cnt;
636
637 if ((hmask + 1) < xfrm_policy_hashmax &&
638 cnt > hmask)
639 return 1;
640
641 return 0;
642 }
643
644 static inline int xfrm_byidx_should_resize(struct net *net, int total)
645 {
646 unsigned int hmask = net->xfrm.policy_idx_hmask;
647
648 if ((hmask + 1) < xfrm_policy_hashmax &&
649 total > hmask)
650 return 1;
651
652 return 0;
653 }
654
655 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
656 {
657 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
658 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
659 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
660 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
661 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
662 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
663 si->spdhcnt = net->xfrm.policy_idx_hmask;
664 si->spdhmcnt = xfrm_policy_hashmax;
665 }
666 EXPORT_SYMBOL(xfrm_spd_getinfo);
667
668 static DEFINE_MUTEX(hash_resize_mutex);
669 static void xfrm_hash_resize(struct work_struct *work)
670 {
671 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
672 int dir, total;
673
674 mutex_lock(&hash_resize_mutex);
675
676 total = 0;
677 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
678 if (xfrm_bydst_should_resize(net, dir, &total))
679 xfrm_bydst_resize(net, dir);
680 }
681 if (xfrm_byidx_should_resize(net, total))
682 xfrm_byidx_resize(net, total);
683
684 mutex_unlock(&hash_resize_mutex);
685 }
686
687 /* Make sure *pol can be inserted into fastbin.
688 * Useful to check that later insert requests will be sucessful
689 * (provided xfrm_policy_lock is held throughout).
690 */
691 static struct xfrm_pol_inexact_bin *
692 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
693 {
694 struct xfrm_pol_inexact_bin *bin, *prev;
695 struct xfrm_pol_inexact_key k = {
696 .family = pol->family,
697 .type = pol->type,
698 .dir = dir,
699 .if_id = pol->if_id,
700 };
701 struct net *net = xp_net(pol);
702
703 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
704
705 write_pnet(&k.net, net);
706 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
707 xfrm_pol_inexact_params);
708 if (bin)
709 return bin;
710
711 bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
712 if (!bin)
713 return NULL;
714
715 bin->k = k;
716 INIT_HLIST_HEAD(&bin->hhead);
717 bin->root_d = RB_ROOT;
718 bin->root_s = RB_ROOT;
719 seqcount_init(&bin->count);
720
721 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
722 &bin->k, &bin->head,
723 xfrm_pol_inexact_params);
724 if (!prev) {
725 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
726 return bin;
727 }
728
729 kfree(bin);
730
731 return IS_ERR(prev) ? NULL : prev;
732 }
733
734 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
735 int family, u8 prefixlen)
736 {
737 if (xfrm_addr_any(addr, family))
738 return true;
739
740 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
741 return true;
742
743 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
744 return true;
745
746 return false;
747 }
748
749 static bool
750 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
751 {
752 const xfrm_address_t *addr;
753 bool saddr_any, daddr_any;
754 u8 prefixlen;
755
756 addr = &policy->selector.saddr;
757 prefixlen = policy->selector.prefixlen_s;
758
759 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
760 policy->family,
761 prefixlen);
762 addr = &policy->selector.daddr;
763 prefixlen = policy->selector.prefixlen_d;
764 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
765 policy->family,
766 prefixlen);
767 return saddr_any && daddr_any;
768 }
769
770 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
771 const xfrm_address_t *addr, u8 prefixlen)
772 {
773 node->addr = *addr;
774 node->prefixlen = prefixlen;
775 }
776
777 static struct xfrm_pol_inexact_node *
778 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
779 {
780 struct xfrm_pol_inexact_node *node;
781
782 node = kzalloc(sizeof(*node), GFP_ATOMIC);
783 if (node)
784 xfrm_pol_inexact_node_init(node, addr, prefixlen);
785
786 return node;
787 }
788
789 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
790 const xfrm_address_t *b,
791 u8 prefixlen, u16 family)
792 {
793 unsigned int pdw, pbi;
794 int delta = 0;
795
796 switch (family) {
797 case AF_INET:
798 if (sizeof(long) == 4 && prefixlen == 0)
799 return ntohl(a->a4) - ntohl(b->a4);
800 return (ntohl(a->a4) & ((~0UL << (32 - prefixlen)))) -
801 (ntohl(b->a4) & ((~0UL << (32 - prefixlen))));
802 case AF_INET6:
803 pdw = prefixlen >> 5;
804 pbi = prefixlen & 0x1f;
805
806 if (pdw) {
807 delta = memcmp(a->a6, b->a6, pdw << 2);
808 if (delta)
809 return delta;
810 }
811 if (pbi) {
812 u32 mask = ~0u << (32 - pbi);
813
814 delta = (ntohl(a->a6[pdw]) & mask) -
815 (ntohl(b->a6[pdw]) & mask);
816 }
817 break;
818 default:
819 break;
820 }
821
822 return delta;
823 }
824
825 static void xfrm_policy_inexact_list_reinsert(struct net *net,
826 struct xfrm_pol_inexact_node *n,
827 u16 family)
828 {
829 unsigned int matched_s, matched_d;
830 struct xfrm_policy *policy, *p;
831
832 matched_s = 0;
833 matched_d = 0;
834
835 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
836 struct hlist_node *newpos = NULL;
837 bool matches_s, matches_d;
838
839 if (!policy->bydst_reinsert)
840 continue;
841
842 WARN_ON_ONCE(policy->family != family);
843
844 policy->bydst_reinsert = false;
845 hlist_for_each_entry(p, &n->hhead, bydst) {
846 if (policy->priority > p->priority)
847 newpos = &p->bydst;
848 else if (policy->priority == p->priority &&
849 policy->pos > p->pos)
850 newpos = &p->bydst;
851 else
852 break;
853 }
854
855 if (newpos)
856 hlist_add_behind_rcu(&policy->bydst, newpos);
857 else
858 hlist_add_head_rcu(&policy->bydst, &n->hhead);
859
860 /* paranoia checks follow.
861 * Check that the reinserted policy matches at least
862 * saddr or daddr for current node prefix.
863 *
864 * Matching both is fine, matching saddr in one policy
865 * (but not daddr) and then matching only daddr in another
866 * is a bug.
867 */
868 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
869 &n->addr,
870 n->prefixlen,
871 family) == 0;
872 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
873 &n->addr,
874 n->prefixlen,
875 family) == 0;
876 if (matches_s && matches_d)
877 continue;
878
879 WARN_ON_ONCE(!matches_s && !matches_d);
880 if (matches_s)
881 matched_s++;
882 if (matches_d)
883 matched_d++;
884 WARN_ON_ONCE(matched_s && matched_d);
885 }
886 }
887
888 static void xfrm_policy_inexact_node_reinsert(struct net *net,
889 struct xfrm_pol_inexact_node *n,
890 struct rb_root *new,
891 u16 family)
892 {
893 struct xfrm_pol_inexact_node *node;
894 struct rb_node **p, *parent;
895
896 /* we should not have another subtree here */
897 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
898 restart:
899 parent = NULL;
900 p = &new->rb_node;
901 while (*p) {
902 u8 prefixlen;
903 int delta;
904
905 parent = *p;
906 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
907
908 prefixlen = min(node->prefixlen, n->prefixlen);
909
910 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
911 prefixlen, family);
912 if (delta < 0) {
913 p = &parent->rb_left;
914 } else if (delta > 0) {
915 p = &parent->rb_right;
916 } else {
917 bool same_prefixlen = node->prefixlen == n->prefixlen;
918 struct xfrm_policy *tmp;
919
920 hlist_for_each_entry(tmp, &n->hhead, bydst) {
921 tmp->bydst_reinsert = true;
922 hlist_del_rcu(&tmp->bydst);
923 }
924
925 node->prefixlen = prefixlen;
926
927 xfrm_policy_inexact_list_reinsert(net, node, family);
928
929 if (same_prefixlen) {
930 kfree_rcu(n, rcu);
931 return;
932 }
933
934 rb_erase(*p, new);
935 kfree_rcu(n, rcu);
936 n = node;
937 goto restart;
938 }
939 }
940
941 rb_link_node_rcu(&n->node, parent, p);
942 rb_insert_color(&n->node, new);
943 }
944
945 /* merge nodes v and n */
946 static void xfrm_policy_inexact_node_merge(struct net *net,
947 struct xfrm_pol_inexact_node *v,
948 struct xfrm_pol_inexact_node *n,
949 u16 family)
950 {
951 struct xfrm_pol_inexact_node *node;
952 struct xfrm_policy *tmp;
953 struct rb_node *rnode;
954
955 /* To-be-merged node v has a subtree.
956 *
957 * Dismantle it and insert its nodes to n->root.
958 */
959 while ((rnode = rb_first(&v->root)) != NULL) {
960 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
961 rb_erase(&node->node, &v->root);
962 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
963 family);
964 }
965
966 hlist_for_each_entry(tmp, &v->hhead, bydst) {
967 tmp->bydst_reinsert = true;
968 hlist_del_rcu(&tmp->bydst);
969 }
970
971 xfrm_policy_inexact_list_reinsert(net, n, family);
972 }
973
974 static struct xfrm_pol_inexact_node *
975 xfrm_policy_inexact_insert_node(struct net *net,
976 struct rb_root *root,
977 xfrm_address_t *addr,
978 u16 family, u8 prefixlen, u8 dir)
979 {
980 struct xfrm_pol_inexact_node *cached = NULL;
981 struct rb_node **p, *parent = NULL;
982 struct xfrm_pol_inexact_node *node;
983
984 p = &root->rb_node;
985 while (*p) {
986 int delta;
987
988 parent = *p;
989 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
990
991 delta = xfrm_policy_addr_delta(addr, &node->addr,
992 node->prefixlen,
993 family);
994 if (delta == 0 && prefixlen >= node->prefixlen) {
995 WARN_ON_ONCE(cached); /* ipsec policies got lost */
996 return node;
997 }
998
999 if (delta < 0)
1000 p = &parent->rb_left;
1001 else
1002 p = &parent->rb_right;
1003
1004 if (prefixlen < node->prefixlen) {
1005 delta = xfrm_policy_addr_delta(addr, &node->addr,
1006 prefixlen,
1007 family);
1008 if (delta)
1009 continue;
1010
1011 /* This node is a subnet of the new prefix. It needs
1012 * to be removed and re-inserted with the smaller
1013 * prefix and all nodes that are now also covered
1014 * by the reduced prefixlen.
1015 */
1016 rb_erase(&node->node, root);
1017
1018 if (!cached) {
1019 xfrm_pol_inexact_node_init(node, addr,
1020 prefixlen);
1021 cached = node;
1022 } else {
1023 /* This node also falls within the new
1024 * prefixlen. Merge the to-be-reinserted
1025 * node and this one.
1026 */
1027 xfrm_policy_inexact_node_merge(net, node,
1028 cached, family);
1029 kfree_rcu(node, rcu);
1030 }
1031
1032 /* restart */
1033 p = &root->rb_node;
1034 parent = NULL;
1035 }
1036 }
1037
1038 node = cached;
1039 if (!node) {
1040 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1041 if (!node)
1042 return NULL;
1043 }
1044
1045 rb_link_node_rcu(&node->node, parent, p);
1046 rb_insert_color(&node->node, root);
1047
1048 return node;
1049 }
1050
1051 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1052 {
1053 struct xfrm_pol_inexact_node *node;
1054 struct rb_node *rn = rb_first(r);
1055
1056 while (rn) {
1057 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1058
1059 xfrm_policy_inexact_gc_tree(&node->root, rm);
1060 rn = rb_next(rn);
1061
1062 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1063 WARN_ON_ONCE(rm);
1064 continue;
1065 }
1066
1067 rb_erase(&node->node, r);
1068 kfree_rcu(node, rcu);
1069 }
1070 }
1071
1072 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1073 {
1074 write_seqcount_begin(&b->count);
1075 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1076 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1077 write_seqcount_end(&b->count);
1078
1079 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1080 !hlist_empty(&b->hhead)) {
1081 WARN_ON_ONCE(net_exit);
1082 return;
1083 }
1084
1085 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1086 xfrm_pol_inexact_params) == 0) {
1087 list_del(&b->inexact_bins);
1088 kfree_rcu(b, rcu);
1089 }
1090 }
1091
1092 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1093 {
1094 struct net *net = read_pnet(&b->k.net);
1095
1096 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1097 __xfrm_policy_inexact_prune_bin(b, false);
1098 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1099 }
1100
1101 static void __xfrm_policy_inexact_flush(struct net *net)
1102 {
1103 struct xfrm_pol_inexact_bin *bin, *t;
1104
1105 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1106
1107 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1108 __xfrm_policy_inexact_prune_bin(bin, false);
1109 }
1110
1111 static struct hlist_head *
1112 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1113 struct xfrm_policy *policy, u8 dir)
1114 {
1115 struct xfrm_pol_inexact_node *n;
1116 struct net *net;
1117
1118 net = xp_net(policy);
1119 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1120
1121 if (xfrm_policy_inexact_insert_use_any_list(policy))
1122 return &bin->hhead;
1123
1124 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1125 policy->family,
1126 policy->selector.prefixlen_d)) {
1127 write_seqcount_begin(&bin->count);
1128 n = xfrm_policy_inexact_insert_node(net,
1129 &bin->root_s,
1130 &policy->selector.saddr,
1131 policy->family,
1132 policy->selector.prefixlen_s,
1133 dir);
1134 write_seqcount_end(&bin->count);
1135 if (!n)
1136 return NULL;
1137
1138 return &n->hhead;
1139 }
1140
1141 /* daddr is fixed */
1142 write_seqcount_begin(&bin->count);
1143 n = xfrm_policy_inexact_insert_node(net,
1144 &bin->root_d,
1145 &policy->selector.daddr,
1146 policy->family,
1147 policy->selector.prefixlen_d, dir);
1148 write_seqcount_end(&bin->count);
1149 if (!n)
1150 return NULL;
1151
1152 /* saddr is wildcard */
1153 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1154 policy->family,
1155 policy->selector.prefixlen_s))
1156 return &n->hhead;
1157
1158 write_seqcount_begin(&bin->count);
1159 n = xfrm_policy_inexact_insert_node(net,
1160 &n->root,
1161 &policy->selector.saddr,
1162 policy->family,
1163 policy->selector.prefixlen_s, dir);
1164 write_seqcount_end(&bin->count);
1165 if (!n)
1166 return NULL;
1167
1168 return &n->hhead;
1169 }
1170
1171 static struct xfrm_policy *
1172 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1173 {
1174 struct xfrm_pol_inexact_bin *bin;
1175 struct xfrm_policy *delpol;
1176 struct hlist_head *chain;
1177 struct net *net;
1178
1179 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1180 if (!bin)
1181 return ERR_PTR(-ENOMEM);
1182
1183 net = xp_net(policy);
1184 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1185
1186 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1187 if (!chain) {
1188 __xfrm_policy_inexact_prune_bin(bin, false);
1189 return ERR_PTR(-ENOMEM);
1190 }
1191
1192 delpol = xfrm_policy_insert_list(chain, policy, excl);
1193 if (delpol && excl) {
1194 __xfrm_policy_inexact_prune_bin(bin, false);
1195 return ERR_PTR(-EEXIST);
1196 }
1197
1198 chain = &net->xfrm.policy_inexact[dir];
1199 xfrm_policy_insert_inexact_list(chain, policy);
1200
1201 if (delpol)
1202 __xfrm_policy_inexact_prune_bin(bin, false);
1203
1204 return delpol;
1205 }
1206
1207 static void xfrm_hash_rebuild(struct work_struct *work)
1208 {
1209 struct net *net = container_of(work, struct net,
1210 xfrm.policy_hthresh.work);
1211 unsigned int hmask;
1212 struct xfrm_policy *pol;
1213 struct xfrm_policy *policy;
1214 struct hlist_head *chain;
1215 struct hlist_head *odst;
1216 struct hlist_node *newpos;
1217 int i;
1218 int dir;
1219 unsigned seq;
1220 u8 lbits4, rbits4, lbits6, rbits6;
1221
1222 mutex_lock(&hash_resize_mutex);
1223
1224 /* read selector prefixlen thresholds */
1225 do {
1226 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1227
1228 lbits4 = net->xfrm.policy_hthresh.lbits4;
1229 rbits4 = net->xfrm.policy_hthresh.rbits4;
1230 lbits6 = net->xfrm.policy_hthresh.lbits6;
1231 rbits6 = net->xfrm.policy_hthresh.rbits6;
1232 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1233
1234 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1235 write_seqcount_begin(&xfrm_policy_hash_generation);
1236
1237 /* make sure that we can insert the indirect policies again before
1238 * we start with destructive action.
1239 */
1240 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1241 struct xfrm_pol_inexact_bin *bin;
1242 u8 dbits, sbits;
1243
1244 dir = xfrm_policy_id2dir(policy->index);
1245 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1246 continue;
1247
1248 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1249 if (policy->family == AF_INET) {
1250 dbits = rbits4;
1251 sbits = lbits4;
1252 } else {
1253 dbits = rbits6;
1254 sbits = lbits6;
1255 }
1256 } else {
1257 if (policy->family == AF_INET) {
1258 dbits = lbits4;
1259 sbits = rbits4;
1260 } else {
1261 dbits = lbits6;
1262 sbits = rbits6;
1263 }
1264 }
1265
1266 if (policy->selector.prefixlen_d < dbits ||
1267 policy->selector.prefixlen_s < sbits)
1268 continue;
1269
1270 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1271 if (!bin)
1272 goto out_unlock;
1273
1274 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1275 goto out_unlock;
1276 }
1277
1278 /* reset the bydst and inexact table in all directions */
1279 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1280 struct hlist_node *n;
1281
1282 hlist_for_each_entry_safe(policy, n,
1283 &net->xfrm.policy_inexact[dir],
1284 bydst_inexact_list) {
1285 hlist_del_rcu(&policy->bydst);
1286 hlist_del_init(&policy->bydst_inexact_list);
1287 }
1288
1289 hmask = net->xfrm.policy_bydst[dir].hmask;
1290 odst = net->xfrm.policy_bydst[dir].table;
1291 for (i = hmask; i >= 0; i--) {
1292 hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1293 hlist_del_rcu(&policy->bydst);
1294 }
1295 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1296 /* dir out => dst = remote, src = local */
1297 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1298 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1299 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1300 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1301 } else {
1302 /* dir in/fwd => dst = local, src = remote */
1303 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1304 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1305 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1306 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1307 }
1308 }
1309
1310 /* re-insert all policies by order of creation */
1311 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1312 if (policy->walk.dead)
1313 continue;
1314 dir = xfrm_policy_id2dir(policy->index);
1315 if (dir >= XFRM_POLICY_MAX) {
1316 /* skip socket policies */
1317 continue;
1318 }
1319 newpos = NULL;
1320 chain = policy_hash_bysel(net, &policy->selector,
1321 policy->family, dir);
1322
1323 if (!chain) {
1324 void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1325
1326 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1327 continue;
1328 }
1329
1330 hlist_for_each_entry(pol, chain, bydst) {
1331 if (policy->priority >= pol->priority)
1332 newpos = &pol->bydst;
1333 else
1334 break;
1335 }
1336 if (newpos)
1337 hlist_add_behind_rcu(&policy->bydst, newpos);
1338 else
1339 hlist_add_head_rcu(&policy->bydst, chain);
1340 }
1341
1342 out_unlock:
1343 __xfrm_policy_inexact_flush(net);
1344 write_seqcount_end(&xfrm_policy_hash_generation);
1345 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1346
1347 mutex_unlock(&hash_resize_mutex);
1348 }
1349
1350 void xfrm_policy_hash_rebuild(struct net *net)
1351 {
1352 schedule_work(&net->xfrm.policy_hthresh.work);
1353 }
1354 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1355
1356 /* Generate new index... KAME seems to generate them ordered by cost
1357 * of an absolute inpredictability of ordering of rules. This will not pass. */
1358 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1359 {
1360 static u32 idx_generator;
1361
1362 for (;;) {
1363 struct hlist_head *list;
1364 struct xfrm_policy *p;
1365 u32 idx;
1366 int found;
1367
1368 if (!index) {
1369 idx = (idx_generator | dir);
1370 idx_generator += 8;
1371 } else {
1372 idx = index;
1373 index = 0;
1374 }
1375
1376 if (idx == 0)
1377 idx = 8;
1378 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1379 found = 0;
1380 hlist_for_each_entry(p, list, byidx) {
1381 if (p->index == idx) {
1382 found = 1;
1383 break;
1384 }
1385 }
1386 if (!found)
1387 return idx;
1388 }
1389 }
1390
1391 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1392 {
1393 u32 *p1 = (u32 *) s1;
1394 u32 *p2 = (u32 *) s2;
1395 int len = sizeof(struct xfrm_selector) / sizeof(u32);
1396 int i;
1397
1398 for (i = 0; i < len; i++) {
1399 if (p1[i] != p2[i])
1400 return 1;
1401 }
1402
1403 return 0;
1404 }
1405
1406 static void xfrm_policy_requeue(struct xfrm_policy *old,
1407 struct xfrm_policy *new)
1408 {
1409 struct xfrm_policy_queue *pq = &old->polq;
1410 struct sk_buff_head list;
1411
1412 if (skb_queue_empty(&pq->hold_queue))
1413 return;
1414
1415 __skb_queue_head_init(&list);
1416
1417 spin_lock_bh(&pq->hold_queue.lock);
1418 skb_queue_splice_init(&pq->hold_queue, &list);
1419 if (del_timer(&pq->hold_timer))
1420 xfrm_pol_put(old);
1421 spin_unlock_bh(&pq->hold_queue.lock);
1422
1423 pq = &new->polq;
1424
1425 spin_lock_bh(&pq->hold_queue.lock);
1426 skb_queue_splice(&list, &pq->hold_queue);
1427 pq->timeout = XFRM_QUEUE_TMO_MIN;
1428 if (!mod_timer(&pq->hold_timer, jiffies))
1429 xfrm_pol_hold(new);
1430 spin_unlock_bh(&pq->hold_queue.lock);
1431 }
1432
1433 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1434 struct xfrm_policy *pol)
1435 {
1436 return mark->v == pol->mark.v && mark->m == pol->mark.m;
1437 }
1438
1439 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1440 {
1441 const struct xfrm_pol_inexact_key *k = data;
1442 u32 a = k->type << 24 | k->dir << 16 | k->family;
1443
1444 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1445 seed);
1446 }
1447
1448 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1449 {
1450 const struct xfrm_pol_inexact_bin *b = data;
1451
1452 return xfrm_pol_bin_key(&b->k, 0, seed);
1453 }
1454
1455 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1456 const void *ptr)
1457 {
1458 const struct xfrm_pol_inexact_key *key = arg->key;
1459 const struct xfrm_pol_inexact_bin *b = ptr;
1460 int ret;
1461
1462 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1463 return -1;
1464
1465 ret = b->k.dir ^ key->dir;
1466 if (ret)
1467 return ret;
1468
1469 ret = b->k.type ^ key->type;
1470 if (ret)
1471 return ret;
1472
1473 ret = b->k.family ^ key->family;
1474 if (ret)
1475 return ret;
1476
1477 return b->k.if_id ^ key->if_id;
1478 }
1479
1480 static const struct rhashtable_params xfrm_pol_inexact_params = {
1481 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head),
1482 .hashfn = xfrm_pol_bin_key,
1483 .obj_hashfn = xfrm_pol_bin_obj,
1484 .obj_cmpfn = xfrm_pol_bin_cmp,
1485 .automatic_shrinking = true,
1486 };
1487
1488 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1489 struct xfrm_policy *policy)
1490 {
1491 struct xfrm_policy *pol, *delpol = NULL;
1492 struct hlist_node *newpos = NULL;
1493 int i = 0;
1494
1495 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1496 if (pol->type == policy->type &&
1497 pol->if_id == policy->if_id &&
1498 !selector_cmp(&pol->selector, &policy->selector) &&
1499 xfrm_policy_mark_match(&policy->mark, pol) &&
1500 xfrm_sec_ctx_match(pol->security, policy->security) &&
1501 !WARN_ON(delpol)) {
1502 delpol = pol;
1503 if (policy->priority > pol->priority)
1504 continue;
1505 } else if (policy->priority >= pol->priority) {
1506 newpos = &pol->bydst_inexact_list;
1507 continue;
1508 }
1509 if (delpol)
1510 break;
1511 }
1512
1513 if (newpos)
1514 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1515 else
1516 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1517
1518 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1519 pol->pos = i;
1520 i++;
1521 }
1522 }
1523
1524 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1525 struct xfrm_policy *policy,
1526 bool excl)
1527 {
1528 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1529
1530 hlist_for_each_entry(pol, chain, bydst) {
1531 if (pol->type == policy->type &&
1532 pol->if_id == policy->if_id &&
1533 !selector_cmp(&pol->selector, &policy->selector) &&
1534 xfrm_policy_mark_match(&policy->mark, pol) &&
1535 xfrm_sec_ctx_match(pol->security, policy->security) &&
1536 !WARN_ON(delpol)) {
1537 if (excl)
1538 return ERR_PTR(-EEXIST);
1539 delpol = pol;
1540 if (policy->priority > pol->priority)
1541 continue;
1542 } else if (policy->priority >= pol->priority) {
1543 newpos = pol;
1544 continue;
1545 }
1546 if (delpol)
1547 break;
1548 }
1549
1550 if (newpos)
1551 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1552 else
1553 hlist_add_head_rcu(&policy->bydst, chain);
1554
1555 return delpol;
1556 }
1557
1558 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1559 {
1560 struct net *net = xp_net(policy);
1561 struct xfrm_policy *delpol;
1562 struct hlist_head *chain;
1563
1564 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1565 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1566 if (chain)
1567 delpol = xfrm_policy_insert_list(chain, policy, excl);
1568 else
1569 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1570
1571 if (IS_ERR(delpol)) {
1572 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1573 return PTR_ERR(delpol);
1574 }
1575
1576 __xfrm_policy_link(policy, dir);
1577
1578 /* After previous checking, family can either be AF_INET or AF_INET6 */
1579 if (policy->family == AF_INET)
1580 rt_genid_bump_ipv4(net);
1581 else
1582 rt_genid_bump_ipv6(net);
1583
1584 if (delpol) {
1585 xfrm_policy_requeue(delpol, policy);
1586 __xfrm_policy_unlink(delpol, dir);
1587 }
1588 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1589 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1590 policy->curlft.add_time = ktime_get_real_seconds();
1591 policy->curlft.use_time = 0;
1592 if (!mod_timer(&policy->timer, jiffies + HZ))
1593 xfrm_pol_hold(policy);
1594 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1595
1596 if (delpol)
1597 xfrm_policy_kill(delpol);
1598 else if (xfrm_bydst_should_resize(net, dir, NULL))
1599 schedule_work(&net->xfrm.policy_hash_work);
1600
1601 return 0;
1602 }
1603 EXPORT_SYMBOL(xfrm_policy_insert);
1604
1605 static struct xfrm_policy *
1606 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1607 u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1608 struct xfrm_sec_ctx *ctx)
1609 {
1610 struct xfrm_policy *pol;
1611
1612 if (!chain)
1613 return NULL;
1614
1615 hlist_for_each_entry(pol, chain, bydst) {
1616 if (pol->type == type &&
1617 pol->if_id == if_id &&
1618 xfrm_policy_mark_match(mark, pol) &&
1619 !selector_cmp(sel, &pol->selector) &&
1620 xfrm_sec_ctx_match(ctx, pol->security))
1621 return pol;
1622 }
1623
1624 return NULL;
1625 }
1626
1627 struct xfrm_policy *
1628 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1629 u8 type, int dir, struct xfrm_selector *sel,
1630 struct xfrm_sec_ctx *ctx, int delete, int *err)
1631 {
1632 struct xfrm_pol_inexact_bin *bin = NULL;
1633 struct xfrm_policy *pol, *ret = NULL;
1634 struct hlist_head *chain;
1635
1636 *err = 0;
1637 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1638 chain = policy_hash_bysel(net, sel, sel->family, dir);
1639 if (!chain) {
1640 struct xfrm_pol_inexact_candidates cand;
1641 int i;
1642
1643 bin = xfrm_policy_inexact_lookup(net, type,
1644 sel->family, dir, if_id);
1645 if (!bin) {
1646 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1647 return NULL;
1648 }
1649
1650 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1651 &sel->saddr,
1652 &sel->daddr)) {
1653 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1654 return NULL;
1655 }
1656
1657 pol = NULL;
1658 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1659 struct xfrm_policy *tmp;
1660
1661 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1662 if_id, type, dir,
1663 sel, ctx);
1664 if (!tmp)
1665 continue;
1666
1667 if (!pol || tmp->pos < pol->pos)
1668 pol = tmp;
1669 }
1670 } else {
1671 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1672 sel, ctx);
1673 }
1674
1675 if (pol) {
1676 xfrm_pol_hold(pol);
1677 if (delete) {
1678 *err = security_xfrm_policy_delete(pol->security);
1679 if (*err) {
1680 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1681 return pol;
1682 }
1683 __xfrm_policy_unlink(pol, dir);
1684 }
1685 ret = pol;
1686 }
1687 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1688
1689 if (ret && delete)
1690 xfrm_policy_kill(ret);
1691 if (bin && delete)
1692 xfrm_policy_inexact_prune_bin(bin);
1693 return ret;
1694 }
1695 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1696
1697 struct xfrm_policy *
1698 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1699 u8 type, int dir, u32 id, int delete, int *err)
1700 {
1701 struct xfrm_policy *pol, *ret;
1702 struct hlist_head *chain;
1703
1704 *err = -ENOENT;
1705 if (xfrm_policy_id2dir(id) != dir)
1706 return NULL;
1707
1708 *err = 0;
1709 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1710 chain = net->xfrm.policy_byidx + idx_hash(net, id);
1711 ret = NULL;
1712 hlist_for_each_entry(pol, chain, byidx) {
1713 if (pol->type == type && pol->index == id &&
1714 pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1715 xfrm_pol_hold(pol);
1716 if (delete) {
1717 *err = security_xfrm_policy_delete(
1718 pol->security);
1719 if (*err) {
1720 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1721 return pol;
1722 }
1723 __xfrm_policy_unlink(pol, dir);
1724 }
1725 ret = pol;
1726 break;
1727 }
1728 }
1729 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1730
1731 if (ret && delete)
1732 xfrm_policy_kill(ret);
1733 return ret;
1734 }
1735 EXPORT_SYMBOL(xfrm_policy_byid);
1736
1737 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1738 static inline int
1739 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1740 {
1741 struct xfrm_policy *pol;
1742 int err = 0;
1743
1744 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1745 if (pol->walk.dead ||
1746 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1747 pol->type != type)
1748 continue;
1749
1750 err = security_xfrm_policy_delete(pol->security);
1751 if (err) {
1752 xfrm_audit_policy_delete(pol, 0, task_valid);
1753 return err;
1754 }
1755 }
1756 return err;
1757 }
1758 #else
1759 static inline int
1760 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1761 {
1762 return 0;
1763 }
1764 #endif
1765
1766 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1767 {
1768 int dir, err = 0, cnt = 0;
1769 struct xfrm_policy *pol;
1770
1771 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1772
1773 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1774 if (err)
1775 goto out;
1776
1777 again:
1778 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1779 dir = xfrm_policy_id2dir(pol->index);
1780 if (pol->walk.dead ||
1781 dir >= XFRM_POLICY_MAX ||
1782 pol->type != type)
1783 continue;
1784
1785 __xfrm_policy_unlink(pol, dir);
1786 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1787 cnt++;
1788 xfrm_audit_policy_delete(pol, 1, task_valid);
1789 xfrm_policy_kill(pol);
1790 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1791 goto again;
1792 }
1793 if (cnt)
1794 __xfrm_policy_inexact_flush(net);
1795 else
1796 err = -ESRCH;
1797 out:
1798 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1799 return err;
1800 }
1801 EXPORT_SYMBOL(xfrm_policy_flush);
1802
1803 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1804 int (*func)(struct xfrm_policy *, int, int, void*),
1805 void *data)
1806 {
1807 struct xfrm_policy *pol;
1808 struct xfrm_policy_walk_entry *x;
1809 int error = 0;
1810
1811 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1812 walk->type != XFRM_POLICY_TYPE_ANY)
1813 return -EINVAL;
1814
1815 if (list_empty(&walk->walk.all) && walk->seq != 0)
1816 return 0;
1817
1818 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1819 if (list_empty(&walk->walk.all))
1820 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1821 else
1822 x = list_first_entry(&walk->walk.all,
1823 struct xfrm_policy_walk_entry, all);
1824
1825 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1826 if (x->dead)
1827 continue;
1828 pol = container_of(x, struct xfrm_policy, walk);
1829 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1830 walk->type != pol->type)
1831 continue;
1832 error = func(pol, xfrm_policy_id2dir(pol->index),
1833 walk->seq, data);
1834 if (error) {
1835 list_move_tail(&walk->walk.all, &x->all);
1836 goto out;
1837 }
1838 walk->seq++;
1839 }
1840 if (walk->seq == 0) {
1841 error = -ENOENT;
1842 goto out;
1843 }
1844 list_del_init(&walk->walk.all);
1845 out:
1846 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1847 return error;
1848 }
1849 EXPORT_SYMBOL(xfrm_policy_walk);
1850
1851 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1852 {
1853 INIT_LIST_HEAD(&walk->walk.all);
1854 walk->walk.dead = 1;
1855 walk->type = type;
1856 walk->seq = 0;
1857 }
1858 EXPORT_SYMBOL(xfrm_policy_walk_init);
1859
1860 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1861 {
1862 if (list_empty(&walk->walk.all))
1863 return;
1864
1865 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1866 list_del(&walk->walk.all);
1867 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1868 }
1869 EXPORT_SYMBOL(xfrm_policy_walk_done);
1870
1871 /*
1872 * Find policy to apply to this flow.
1873 *
1874 * Returns 0 if policy found, else an -errno.
1875 */
1876 static int xfrm_policy_match(const struct xfrm_policy *pol,
1877 const struct flowi *fl,
1878 u8 type, u16 family, int dir, u32 if_id)
1879 {
1880 const struct xfrm_selector *sel = &pol->selector;
1881 int ret = -ESRCH;
1882 bool match;
1883
1884 if (pol->family != family ||
1885 pol->if_id != if_id ||
1886 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1887 pol->type != type)
1888 return ret;
1889
1890 match = xfrm_selector_match(sel, fl, family);
1891 if (match)
1892 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1893 dir);
1894 return ret;
1895 }
1896
1897 static struct xfrm_pol_inexact_node *
1898 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1899 seqcount_t *count,
1900 const xfrm_address_t *addr, u16 family)
1901 {
1902 const struct rb_node *parent;
1903 int seq;
1904
1905 again:
1906 seq = read_seqcount_begin(count);
1907
1908 parent = rcu_dereference_raw(r->rb_node);
1909 while (parent) {
1910 struct xfrm_pol_inexact_node *node;
1911 int delta;
1912
1913 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1914
1915 delta = xfrm_policy_addr_delta(addr, &node->addr,
1916 node->prefixlen, family);
1917 if (delta < 0) {
1918 parent = rcu_dereference_raw(parent->rb_left);
1919 continue;
1920 } else if (delta > 0) {
1921 parent = rcu_dereference_raw(parent->rb_right);
1922 continue;
1923 }
1924
1925 return node;
1926 }
1927
1928 if (read_seqcount_retry(count, seq))
1929 goto again;
1930
1931 return NULL;
1932 }
1933
1934 static bool
1935 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1936 struct xfrm_pol_inexact_bin *b,
1937 const xfrm_address_t *saddr,
1938 const xfrm_address_t *daddr)
1939 {
1940 struct xfrm_pol_inexact_node *n;
1941 u16 family;
1942
1943 if (!b)
1944 return false;
1945
1946 family = b->k.family;
1947 memset(cand, 0, sizeof(*cand));
1948 cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1949
1950 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1951 family);
1952 if (n) {
1953 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1954 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1955 family);
1956 if (n)
1957 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1958 }
1959
1960 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1961 family);
1962 if (n)
1963 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1964
1965 return true;
1966 }
1967
1968 static struct xfrm_pol_inexact_bin *
1969 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1970 u8 dir, u32 if_id)
1971 {
1972 struct xfrm_pol_inexact_key k = {
1973 .family = family,
1974 .type = type,
1975 .dir = dir,
1976 .if_id = if_id,
1977 };
1978
1979 write_pnet(&k.net, net);
1980
1981 return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1982 xfrm_pol_inexact_params);
1983 }
1984
1985 static struct xfrm_pol_inexact_bin *
1986 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1987 u8 dir, u32 if_id)
1988 {
1989 struct xfrm_pol_inexact_bin *bin;
1990
1991 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1992
1993 rcu_read_lock();
1994 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
1995 rcu_read_unlock();
1996
1997 return bin;
1998 }
1999
2000 static struct xfrm_policy *
2001 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2002 struct xfrm_policy *prefer,
2003 const struct flowi *fl,
2004 u8 type, u16 family, int dir, u32 if_id)
2005 {
2006 u32 priority = prefer ? prefer->priority : ~0u;
2007 struct xfrm_policy *pol;
2008
2009 if (!chain)
2010 return NULL;
2011
2012 hlist_for_each_entry_rcu(pol, chain, bydst) {
2013 int err;
2014
2015 if (pol->priority > priority)
2016 break;
2017
2018 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2019 if (err) {
2020 if (err != -ESRCH)
2021 return ERR_PTR(err);
2022
2023 continue;
2024 }
2025
2026 if (prefer) {
2027 /* matches. Is it older than *prefer? */
2028 if (pol->priority == priority &&
2029 prefer->pos < pol->pos)
2030 return prefer;
2031 }
2032
2033 return pol;
2034 }
2035
2036 return NULL;
2037 }
2038
2039 static struct xfrm_policy *
2040 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2041 struct xfrm_policy *prefer,
2042 const struct flowi *fl,
2043 u8 type, u16 family, int dir, u32 if_id)
2044 {
2045 struct xfrm_policy *tmp;
2046 int i;
2047
2048 for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2049 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2050 prefer,
2051 fl, type, family, dir,
2052 if_id);
2053 if (!tmp)
2054 continue;
2055
2056 if (IS_ERR(tmp))
2057 return tmp;
2058 prefer = tmp;
2059 }
2060
2061 return prefer;
2062 }
2063
2064 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2065 const struct flowi *fl,
2066 u16 family, u8 dir,
2067 u32 if_id)
2068 {
2069 struct xfrm_pol_inexact_candidates cand;
2070 const xfrm_address_t *daddr, *saddr;
2071 struct xfrm_pol_inexact_bin *bin;
2072 struct xfrm_policy *pol, *ret;
2073 struct hlist_head *chain;
2074 unsigned int sequence;
2075 int err;
2076
2077 daddr = xfrm_flowi_daddr(fl, family);
2078 saddr = xfrm_flowi_saddr(fl, family);
2079 if (unlikely(!daddr || !saddr))
2080 return NULL;
2081
2082 rcu_read_lock();
2083 retry:
2084 do {
2085 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2086 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2087 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2088
2089 ret = NULL;
2090 hlist_for_each_entry_rcu(pol, chain, bydst) {
2091 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2092 if (err) {
2093 if (err == -ESRCH)
2094 continue;
2095 else {
2096 ret = ERR_PTR(err);
2097 goto fail;
2098 }
2099 } else {
2100 ret = pol;
2101 break;
2102 }
2103 }
2104 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2105 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2106 daddr))
2107 goto skip_inexact;
2108
2109 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2110 family, dir, if_id);
2111 if (pol) {
2112 ret = pol;
2113 if (IS_ERR(pol))
2114 goto fail;
2115 }
2116
2117 skip_inexact:
2118 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2119 goto retry;
2120
2121 if (ret && !xfrm_pol_hold_rcu(ret))
2122 goto retry;
2123 fail:
2124 rcu_read_unlock();
2125
2126 return ret;
2127 }
2128
2129 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2130 const struct flowi *fl,
2131 u16 family, u8 dir, u32 if_id)
2132 {
2133 #ifdef CONFIG_XFRM_SUB_POLICY
2134 struct xfrm_policy *pol;
2135
2136 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2137 dir, if_id);
2138 if (pol != NULL)
2139 return pol;
2140 #endif
2141 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2142 dir, if_id);
2143 }
2144
2145 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2146 const struct flowi *fl,
2147 u16 family, u32 if_id)
2148 {
2149 struct xfrm_policy *pol;
2150
2151 rcu_read_lock();
2152 again:
2153 pol = rcu_dereference(sk->sk_policy[dir]);
2154 if (pol != NULL) {
2155 bool match;
2156 int err = 0;
2157
2158 if (pol->family != family) {
2159 pol = NULL;
2160 goto out;
2161 }
2162
2163 match = xfrm_selector_match(&pol->selector, fl, family);
2164 if (match) {
2165 if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2166 pol->if_id != if_id) {
2167 pol = NULL;
2168 goto out;
2169 }
2170 err = security_xfrm_policy_lookup(pol->security,
2171 fl->flowi_secid,
2172 dir);
2173 if (!err) {
2174 if (!xfrm_pol_hold_rcu(pol))
2175 goto again;
2176 } else if (err == -ESRCH) {
2177 pol = NULL;
2178 } else {
2179 pol = ERR_PTR(err);
2180 }
2181 } else
2182 pol = NULL;
2183 }
2184 out:
2185 rcu_read_unlock();
2186 return pol;
2187 }
2188
2189 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2190 {
2191 struct net *net = xp_net(pol);
2192
2193 list_add(&pol->walk.all, &net->xfrm.policy_all);
2194 net->xfrm.policy_count[dir]++;
2195 xfrm_pol_hold(pol);
2196 }
2197
2198 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2199 int dir)
2200 {
2201 struct net *net = xp_net(pol);
2202
2203 if (list_empty(&pol->walk.all))
2204 return NULL;
2205
2206 /* Socket policies are not hashed. */
2207 if (!hlist_unhashed(&pol->bydst)) {
2208 hlist_del_rcu(&pol->bydst);
2209 hlist_del_init(&pol->bydst_inexact_list);
2210 hlist_del(&pol->byidx);
2211 }
2212
2213 list_del_init(&pol->walk.all);
2214 net->xfrm.policy_count[dir]--;
2215
2216 return pol;
2217 }
2218
2219 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2220 {
2221 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2222 }
2223
2224 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2225 {
2226 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2227 }
2228
2229 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2230 {
2231 struct net *net = xp_net(pol);
2232
2233 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2234 pol = __xfrm_policy_unlink(pol, dir);
2235 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2236 if (pol) {
2237 xfrm_policy_kill(pol);
2238 return 0;
2239 }
2240 return -ENOENT;
2241 }
2242 EXPORT_SYMBOL(xfrm_policy_delete);
2243
2244 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2245 {
2246 struct net *net = sock_net(sk);
2247 struct xfrm_policy *old_pol;
2248
2249 #ifdef CONFIG_XFRM_SUB_POLICY
2250 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2251 return -EINVAL;
2252 #endif
2253
2254 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2255 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2256 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2257 if (pol) {
2258 pol->curlft.add_time = ktime_get_real_seconds();
2259 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2260 xfrm_sk_policy_link(pol, dir);
2261 }
2262 rcu_assign_pointer(sk->sk_policy[dir], pol);
2263 if (old_pol) {
2264 if (pol)
2265 xfrm_policy_requeue(old_pol, pol);
2266
2267 /* Unlinking succeeds always. This is the only function
2268 * allowed to delete or replace socket policy.
2269 */
2270 xfrm_sk_policy_unlink(old_pol, dir);
2271 }
2272 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2273
2274 if (old_pol) {
2275 xfrm_policy_kill(old_pol);
2276 }
2277 return 0;
2278 }
2279
2280 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2281 {
2282 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2283 struct net *net = xp_net(old);
2284
2285 if (newp) {
2286 newp->selector = old->selector;
2287 if (security_xfrm_policy_clone(old->security,
2288 &newp->security)) {
2289 kfree(newp);
2290 return NULL; /* ENOMEM */
2291 }
2292 newp->lft = old->lft;
2293 newp->curlft = old->curlft;
2294 newp->mark = old->mark;
2295 newp->if_id = old->if_id;
2296 newp->action = old->action;
2297 newp->flags = old->flags;
2298 newp->xfrm_nr = old->xfrm_nr;
2299 newp->index = old->index;
2300 newp->type = old->type;
2301 newp->family = old->family;
2302 memcpy(newp->xfrm_vec, old->xfrm_vec,
2303 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2304 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2305 xfrm_sk_policy_link(newp, dir);
2306 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2307 xfrm_pol_put(newp);
2308 }
2309 return newp;
2310 }
2311
2312 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2313 {
2314 const struct xfrm_policy *p;
2315 struct xfrm_policy *np;
2316 int i, ret = 0;
2317
2318 rcu_read_lock();
2319 for (i = 0; i < 2; i++) {
2320 p = rcu_dereference(osk->sk_policy[i]);
2321 if (p) {
2322 np = clone_policy(p, i);
2323 if (unlikely(!np)) {
2324 ret = -ENOMEM;
2325 break;
2326 }
2327 rcu_assign_pointer(sk->sk_policy[i], np);
2328 }
2329 }
2330 rcu_read_unlock();
2331 return ret;
2332 }
2333
2334 static int
2335 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2336 xfrm_address_t *remote, unsigned short family, u32 mark)
2337 {
2338 int err;
2339 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2340
2341 if (unlikely(afinfo == NULL))
2342 return -EINVAL;
2343 err = afinfo->get_saddr(net, oif, local, remote, mark);
2344 rcu_read_unlock();
2345 return err;
2346 }
2347
2348 /* Resolve list of templates for the flow, given policy. */
2349
2350 static int
2351 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2352 struct xfrm_state **xfrm, unsigned short family)
2353 {
2354 struct net *net = xp_net(policy);
2355 int nx;
2356 int i, error;
2357 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2358 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2359 xfrm_address_t tmp;
2360
2361 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2362 struct xfrm_state *x;
2363 xfrm_address_t *remote = daddr;
2364 xfrm_address_t *local = saddr;
2365 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2366
2367 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2368 tmpl->mode == XFRM_MODE_BEET) {
2369 remote = &tmpl->id.daddr;
2370 local = &tmpl->saddr;
2371 if (xfrm_addr_any(local, tmpl->encap_family)) {
2372 error = xfrm_get_saddr(net, fl->flowi_oif,
2373 &tmp, remote,
2374 tmpl->encap_family, 0);
2375 if (error)
2376 goto fail;
2377 local = &tmp;
2378 }
2379 }
2380
2381 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2382 family, policy->if_id);
2383
2384 if (x && x->km.state == XFRM_STATE_VALID) {
2385 xfrm[nx++] = x;
2386 daddr = remote;
2387 saddr = local;
2388 continue;
2389 }
2390 if (x) {
2391 error = (x->km.state == XFRM_STATE_ERROR ?
2392 -EINVAL : -EAGAIN);
2393 xfrm_state_put(x);
2394 } else if (error == -ESRCH) {
2395 error = -EAGAIN;
2396 }
2397
2398 if (!tmpl->optional)
2399 goto fail;
2400 }
2401 return nx;
2402
2403 fail:
2404 for (nx--; nx >= 0; nx--)
2405 xfrm_state_put(xfrm[nx]);
2406 return error;
2407 }
2408
2409 static int
2410 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2411 struct xfrm_state **xfrm, unsigned short family)
2412 {
2413 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2414 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2415 int cnx = 0;
2416 int error;
2417 int ret;
2418 int i;
2419
2420 for (i = 0; i < npols; i++) {
2421 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2422 error = -ENOBUFS;
2423 goto fail;
2424 }
2425
2426 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2427 if (ret < 0) {
2428 error = ret;
2429 goto fail;
2430 } else
2431 cnx += ret;
2432 }
2433
2434 /* found states are sorted for outbound processing */
2435 if (npols > 1)
2436 xfrm_state_sort(xfrm, tpp, cnx, family);
2437
2438 return cnx;
2439
2440 fail:
2441 for (cnx--; cnx >= 0; cnx--)
2442 xfrm_state_put(tpp[cnx]);
2443 return error;
2444
2445 }
2446
2447 static int xfrm_get_tos(const struct flowi *fl, int family)
2448 {
2449 if (family == AF_INET)
2450 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2451
2452 return 0;
2453 }
2454
2455 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2456 {
2457 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2458 struct dst_ops *dst_ops;
2459 struct xfrm_dst *xdst;
2460
2461 if (!afinfo)
2462 return ERR_PTR(-EINVAL);
2463
2464 switch (family) {
2465 case AF_INET:
2466 dst_ops = &net->xfrm.xfrm4_dst_ops;
2467 break;
2468 #if IS_ENABLED(CONFIG_IPV6)
2469 case AF_INET6:
2470 dst_ops = &net->xfrm.xfrm6_dst_ops;
2471 break;
2472 #endif
2473 default:
2474 BUG();
2475 }
2476 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2477
2478 if (likely(xdst)) {
2479 struct dst_entry *dst = &xdst->u.dst;
2480
2481 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2482 } else
2483 xdst = ERR_PTR(-ENOBUFS);
2484
2485 rcu_read_unlock();
2486
2487 return xdst;
2488 }
2489
2490 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2491 int nfheader_len)
2492 {
2493 if (dst->ops->family == AF_INET6) {
2494 struct rt6_info *rt = (struct rt6_info *)dst;
2495 path->path_cookie = rt6_get_cookie(rt);
2496 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2497 }
2498 }
2499
2500 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2501 const struct flowi *fl)
2502 {
2503 const struct xfrm_policy_afinfo *afinfo =
2504 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2505 int err;
2506
2507 if (!afinfo)
2508 return -EINVAL;
2509
2510 err = afinfo->fill_dst(xdst, dev, fl);
2511
2512 rcu_read_unlock();
2513
2514 return err;
2515 }
2516
2517
2518 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2519 * all the metrics... Shortly, bundle a bundle.
2520 */
2521
2522 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2523 struct xfrm_state **xfrm,
2524 struct xfrm_dst **bundle,
2525 int nx,
2526 const struct flowi *fl,
2527 struct dst_entry *dst)
2528 {
2529 const struct xfrm_state_afinfo *afinfo;
2530 const struct xfrm_mode *inner_mode;
2531 struct net *net = xp_net(policy);
2532 unsigned long now = jiffies;
2533 struct net_device *dev;
2534 struct xfrm_dst *xdst_prev = NULL;
2535 struct xfrm_dst *xdst0 = NULL;
2536 int i = 0;
2537 int err;
2538 int header_len = 0;
2539 int nfheader_len = 0;
2540 int trailer_len = 0;
2541 int tos;
2542 int family = policy->selector.family;
2543 xfrm_address_t saddr, daddr;
2544
2545 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2546
2547 tos = xfrm_get_tos(fl, family);
2548
2549 dst_hold(dst);
2550
2551 for (; i < nx; i++) {
2552 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2553 struct dst_entry *dst1 = &xdst->u.dst;
2554
2555 err = PTR_ERR(xdst);
2556 if (IS_ERR(xdst)) {
2557 dst_release(dst);
2558 goto put_states;
2559 }
2560
2561 bundle[i] = xdst;
2562 if (!xdst_prev)
2563 xdst0 = xdst;
2564 else
2565 /* Ref count is taken during xfrm_alloc_dst()
2566 * No need to do dst_clone() on dst1
2567 */
2568 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2569
2570 if (xfrm[i]->sel.family == AF_UNSPEC) {
2571 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2572 xfrm_af2proto(family));
2573 if (!inner_mode) {
2574 err = -EAFNOSUPPORT;
2575 dst_release(dst);
2576 goto put_states;
2577 }
2578 } else
2579 inner_mode = &xfrm[i]->inner_mode;
2580
2581 xdst->route = dst;
2582 dst_copy_metrics(dst1, dst);
2583
2584 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2585 __u32 mark = 0;
2586
2587 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2588 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2589
2590 family = xfrm[i]->props.family;
2591 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2592 &saddr, &daddr, family, mark);
2593 err = PTR_ERR(dst);
2594 if (IS_ERR(dst))
2595 goto put_states;
2596 } else
2597 dst_hold(dst);
2598
2599 dst1->xfrm = xfrm[i];
2600 xdst->xfrm_genid = xfrm[i]->genid;
2601
2602 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2603 dst1->flags |= DST_HOST;
2604 dst1->lastuse = now;
2605
2606 dst1->input = dst_discard;
2607
2608 rcu_read_lock();
2609 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2610 if (likely(afinfo))
2611 dst1->output = afinfo->output;
2612 else
2613 dst1->output = dst_discard_out;
2614 rcu_read_unlock();
2615
2616 xdst_prev = xdst;
2617
2618 header_len += xfrm[i]->props.header_len;
2619 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2620 nfheader_len += xfrm[i]->props.header_len;
2621 trailer_len += xfrm[i]->props.trailer_len;
2622 }
2623
2624 xfrm_dst_set_child(xdst_prev, dst);
2625 xdst0->path = dst;
2626
2627 err = -ENODEV;
2628 dev = dst->dev;
2629 if (!dev)
2630 goto free_dst;
2631
2632 xfrm_init_path(xdst0, dst, nfheader_len);
2633 xfrm_init_pmtu(bundle, nx);
2634
2635 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2636 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2637 err = xfrm_fill_dst(xdst_prev, dev, fl);
2638 if (err)
2639 goto free_dst;
2640
2641 xdst_prev->u.dst.header_len = header_len;
2642 xdst_prev->u.dst.trailer_len = trailer_len;
2643 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2644 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2645 }
2646
2647 return &xdst0->u.dst;
2648
2649 put_states:
2650 for (; i < nx; i++)
2651 xfrm_state_put(xfrm[i]);
2652 free_dst:
2653 if (xdst0)
2654 dst_release_immediate(&xdst0->u.dst);
2655
2656 return ERR_PTR(err);
2657 }
2658
2659 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2660 struct xfrm_policy **pols,
2661 int *num_pols, int *num_xfrms)
2662 {
2663 int i;
2664
2665 if (*num_pols == 0 || !pols[0]) {
2666 *num_pols = 0;
2667 *num_xfrms = 0;
2668 return 0;
2669 }
2670 if (IS_ERR(pols[0]))
2671 return PTR_ERR(pols[0]);
2672
2673 *num_xfrms = pols[0]->xfrm_nr;
2674
2675 #ifdef CONFIG_XFRM_SUB_POLICY
2676 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2677 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2678 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2679 XFRM_POLICY_TYPE_MAIN,
2680 fl, family,
2681 XFRM_POLICY_OUT,
2682 pols[0]->if_id);
2683 if (pols[1]) {
2684 if (IS_ERR(pols[1])) {
2685 xfrm_pols_put(pols, *num_pols);
2686 return PTR_ERR(pols[1]);
2687 }
2688 (*num_pols)++;
2689 (*num_xfrms) += pols[1]->xfrm_nr;
2690 }
2691 }
2692 #endif
2693 for (i = 0; i < *num_pols; i++) {
2694 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2695 *num_xfrms = -1;
2696 break;
2697 }
2698 }
2699
2700 return 0;
2701
2702 }
2703
2704 static struct xfrm_dst *
2705 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2706 const struct flowi *fl, u16 family,
2707 struct dst_entry *dst_orig)
2708 {
2709 struct net *net = xp_net(pols[0]);
2710 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2711 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2712 struct xfrm_dst *xdst;
2713 struct dst_entry *dst;
2714 int err;
2715
2716 /* Try to instantiate a bundle */
2717 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2718 if (err <= 0) {
2719 if (err == 0)
2720 return NULL;
2721
2722 if (err != -EAGAIN)
2723 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2724 return ERR_PTR(err);
2725 }
2726
2727 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2728 if (IS_ERR(dst)) {
2729 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2730 return ERR_CAST(dst);
2731 }
2732
2733 xdst = (struct xfrm_dst *)dst;
2734 xdst->num_xfrms = err;
2735 xdst->num_pols = num_pols;
2736 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2737 xdst->policy_genid = atomic_read(&pols[0]->genid);
2738
2739 return xdst;
2740 }
2741
2742 static void xfrm_policy_queue_process(struct timer_list *t)
2743 {
2744 struct sk_buff *skb;
2745 struct sock *sk;
2746 struct dst_entry *dst;
2747 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2748 struct net *net = xp_net(pol);
2749 struct xfrm_policy_queue *pq = &pol->polq;
2750 struct flowi fl;
2751 struct sk_buff_head list;
2752
2753 spin_lock(&pq->hold_queue.lock);
2754 skb = skb_peek(&pq->hold_queue);
2755 if (!skb) {
2756 spin_unlock(&pq->hold_queue.lock);
2757 goto out;
2758 }
2759 dst = skb_dst(skb);
2760 sk = skb->sk;
2761 xfrm_decode_session(skb, &fl, dst->ops->family);
2762 spin_unlock(&pq->hold_queue.lock);
2763
2764 dst_hold(xfrm_dst_path(dst));
2765 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2766 if (IS_ERR(dst))
2767 goto purge_queue;
2768
2769 if (dst->flags & DST_XFRM_QUEUE) {
2770 dst_release(dst);
2771
2772 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2773 goto purge_queue;
2774
2775 pq->timeout = pq->timeout << 1;
2776 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2777 xfrm_pol_hold(pol);
2778 goto out;
2779 }
2780
2781 dst_release(dst);
2782
2783 __skb_queue_head_init(&list);
2784
2785 spin_lock(&pq->hold_queue.lock);
2786 pq->timeout = 0;
2787 skb_queue_splice_init(&pq->hold_queue, &list);
2788 spin_unlock(&pq->hold_queue.lock);
2789
2790 while (!skb_queue_empty(&list)) {
2791 skb = __skb_dequeue(&list);
2792
2793 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2794 dst_hold(xfrm_dst_path(skb_dst(skb)));
2795 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2796 if (IS_ERR(dst)) {
2797 kfree_skb(skb);
2798 continue;
2799 }
2800
2801 nf_reset_ct(skb);
2802 skb_dst_drop(skb);
2803 skb_dst_set(skb, dst);
2804
2805 dst_output(net, skb->sk, skb);
2806 }
2807
2808 out:
2809 xfrm_pol_put(pol);
2810 return;
2811
2812 purge_queue:
2813 pq->timeout = 0;
2814 skb_queue_purge(&pq->hold_queue);
2815 xfrm_pol_put(pol);
2816 }
2817
2818 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2819 {
2820 unsigned long sched_next;
2821 struct dst_entry *dst = skb_dst(skb);
2822 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2823 struct xfrm_policy *pol = xdst->pols[0];
2824 struct xfrm_policy_queue *pq = &pol->polq;
2825
2826 if (unlikely(skb_fclone_busy(sk, skb))) {
2827 kfree_skb(skb);
2828 return 0;
2829 }
2830
2831 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2832 kfree_skb(skb);
2833 return -EAGAIN;
2834 }
2835
2836 skb_dst_force(skb);
2837
2838 spin_lock_bh(&pq->hold_queue.lock);
2839
2840 if (!pq->timeout)
2841 pq->timeout = XFRM_QUEUE_TMO_MIN;
2842
2843 sched_next = jiffies + pq->timeout;
2844
2845 if (del_timer(&pq->hold_timer)) {
2846 if (time_before(pq->hold_timer.expires, sched_next))
2847 sched_next = pq->hold_timer.expires;
2848 xfrm_pol_put(pol);
2849 }
2850
2851 __skb_queue_tail(&pq->hold_queue, skb);
2852 if (!mod_timer(&pq->hold_timer, sched_next))
2853 xfrm_pol_hold(pol);
2854
2855 spin_unlock_bh(&pq->hold_queue.lock);
2856
2857 return 0;
2858 }
2859
2860 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2861 struct xfrm_flo *xflo,
2862 const struct flowi *fl,
2863 int num_xfrms,
2864 u16 family)
2865 {
2866 int err;
2867 struct net_device *dev;
2868 struct dst_entry *dst;
2869 struct dst_entry *dst1;
2870 struct xfrm_dst *xdst;
2871
2872 xdst = xfrm_alloc_dst(net, family);
2873 if (IS_ERR(xdst))
2874 return xdst;
2875
2876 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2877 net->xfrm.sysctl_larval_drop ||
2878 num_xfrms <= 0)
2879 return xdst;
2880
2881 dst = xflo->dst_orig;
2882 dst1 = &xdst->u.dst;
2883 dst_hold(dst);
2884 xdst->route = dst;
2885
2886 dst_copy_metrics(dst1, dst);
2887
2888 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2889 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2890 dst1->lastuse = jiffies;
2891
2892 dst1->input = dst_discard;
2893 dst1->output = xdst_queue_output;
2894
2895 dst_hold(dst);
2896 xfrm_dst_set_child(xdst, dst);
2897 xdst->path = dst;
2898
2899 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2900
2901 err = -ENODEV;
2902 dev = dst->dev;
2903 if (!dev)
2904 goto free_dst;
2905
2906 err = xfrm_fill_dst(xdst, dev, fl);
2907 if (err)
2908 goto free_dst;
2909
2910 out:
2911 return xdst;
2912
2913 free_dst:
2914 dst_release(dst1);
2915 xdst = ERR_PTR(err);
2916 goto out;
2917 }
2918
2919 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2920 const struct flowi *fl,
2921 u16 family, u8 dir,
2922 struct xfrm_flo *xflo, u32 if_id)
2923 {
2924 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2925 int num_pols = 0, num_xfrms = 0, err;
2926 struct xfrm_dst *xdst;
2927
2928 /* Resolve policies to use if we couldn't get them from
2929 * previous cache entry */
2930 num_pols = 1;
2931 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2932 err = xfrm_expand_policies(fl, family, pols,
2933 &num_pols, &num_xfrms);
2934 if (err < 0)
2935 goto inc_error;
2936 if (num_pols == 0)
2937 return NULL;
2938 if (num_xfrms <= 0)
2939 goto make_dummy_bundle;
2940
2941 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2942 xflo->dst_orig);
2943 if (IS_ERR(xdst)) {
2944 err = PTR_ERR(xdst);
2945 if (err == -EREMOTE) {
2946 xfrm_pols_put(pols, num_pols);
2947 return NULL;
2948 }
2949
2950 if (err != -EAGAIN)
2951 goto error;
2952 goto make_dummy_bundle;
2953 } else if (xdst == NULL) {
2954 num_xfrms = 0;
2955 goto make_dummy_bundle;
2956 }
2957
2958 return xdst;
2959
2960 make_dummy_bundle:
2961 /* We found policies, but there's no bundles to instantiate:
2962 * either because the policy blocks, has no transformations or
2963 * we could not build template (no xfrm_states).*/
2964 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2965 if (IS_ERR(xdst)) {
2966 xfrm_pols_put(pols, num_pols);
2967 return ERR_CAST(xdst);
2968 }
2969 xdst->num_pols = num_pols;
2970 xdst->num_xfrms = num_xfrms;
2971 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2972
2973 return xdst;
2974
2975 inc_error:
2976 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2977 error:
2978 xfrm_pols_put(pols, num_pols);
2979 return ERR_PTR(err);
2980 }
2981
2982 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2983 struct dst_entry *dst_orig)
2984 {
2985 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2986 struct dst_entry *ret;
2987
2988 if (!afinfo) {
2989 dst_release(dst_orig);
2990 return ERR_PTR(-EINVAL);
2991 } else {
2992 ret = afinfo->blackhole_route(net, dst_orig);
2993 }
2994 rcu_read_unlock();
2995
2996 return ret;
2997 }
2998
2999 /* Finds/creates a bundle for given flow and if_id
3000 *
3001 * At the moment we eat a raw IP route. Mostly to speed up lookups
3002 * on interfaces with disabled IPsec.
3003 *
3004 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3005 * compatibility
3006 */
3007 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3008 struct dst_entry *dst_orig,
3009 const struct flowi *fl,
3010 const struct sock *sk,
3011 int flags, u32 if_id)
3012 {
3013 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3014 struct xfrm_dst *xdst;
3015 struct dst_entry *dst, *route;
3016 u16 family = dst_orig->ops->family;
3017 u8 dir = XFRM_POLICY_OUT;
3018 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3019
3020 dst = NULL;
3021 xdst = NULL;
3022 route = NULL;
3023
3024 sk = sk_const_to_full_sk(sk);
3025 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3026 num_pols = 1;
3027 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3028 if_id);
3029 err = xfrm_expand_policies(fl, family, pols,
3030 &num_pols, &num_xfrms);
3031 if (err < 0)
3032 goto dropdst;
3033
3034 if (num_pols) {
3035 if (num_xfrms <= 0) {
3036 drop_pols = num_pols;
3037 goto no_transform;
3038 }
3039
3040 xdst = xfrm_resolve_and_create_bundle(
3041 pols, num_pols, fl,
3042 family, dst_orig);
3043
3044 if (IS_ERR(xdst)) {
3045 xfrm_pols_put(pols, num_pols);
3046 err = PTR_ERR(xdst);
3047 if (err == -EREMOTE)
3048 goto nopol;
3049
3050 goto dropdst;
3051 } else if (xdst == NULL) {
3052 num_xfrms = 0;
3053 drop_pols = num_pols;
3054 goto no_transform;
3055 }
3056
3057 route = xdst->route;
3058 }
3059 }
3060
3061 if (xdst == NULL) {
3062 struct xfrm_flo xflo;
3063
3064 xflo.dst_orig = dst_orig;
3065 xflo.flags = flags;
3066
3067 /* To accelerate a bit... */
3068 if ((dst_orig->flags & DST_NOXFRM) ||
3069 !net->xfrm.policy_count[XFRM_POLICY_OUT])
3070 goto nopol;
3071
3072 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3073 if (xdst == NULL)
3074 goto nopol;
3075 if (IS_ERR(xdst)) {
3076 err = PTR_ERR(xdst);
3077 goto dropdst;
3078 }
3079
3080 num_pols = xdst->num_pols;
3081 num_xfrms = xdst->num_xfrms;
3082 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3083 route = xdst->route;
3084 }
3085
3086 dst = &xdst->u.dst;
3087 if (route == NULL && num_xfrms > 0) {
3088 /* The only case when xfrm_bundle_lookup() returns a
3089 * bundle with null route, is when the template could
3090 * not be resolved. It means policies are there, but
3091 * bundle could not be created, since we don't yet
3092 * have the xfrm_state's. We need to wait for KM to
3093 * negotiate new SA's or bail out with error.*/
3094 if (net->xfrm.sysctl_larval_drop) {
3095 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3096 err = -EREMOTE;
3097 goto error;
3098 }
3099
3100 err = -EAGAIN;
3101
3102 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3103 goto error;
3104 }
3105
3106 no_transform:
3107 if (num_pols == 0)
3108 goto nopol;
3109
3110 if ((flags & XFRM_LOOKUP_ICMP) &&
3111 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3112 err = -ENOENT;
3113 goto error;
3114 }
3115
3116 for (i = 0; i < num_pols; i++)
3117 pols[i]->curlft.use_time = ktime_get_real_seconds();
3118
3119 if (num_xfrms < 0) {
3120 /* Prohibit the flow */
3121 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3122 err = -EPERM;
3123 goto error;
3124 } else if (num_xfrms > 0) {
3125 /* Flow transformed */
3126 dst_release(dst_orig);
3127 } else {
3128 /* Flow passes untransformed */
3129 dst_release(dst);
3130 dst = dst_orig;
3131 }
3132 ok:
3133 xfrm_pols_put(pols, drop_pols);
3134 if (dst && dst->xfrm &&
3135 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3136 dst->flags |= DST_XFRM_TUNNEL;
3137 return dst;
3138
3139 nopol:
3140 if (!(flags & XFRM_LOOKUP_ICMP)) {
3141 dst = dst_orig;
3142 goto ok;
3143 }
3144 err = -ENOENT;
3145 error:
3146 dst_release(dst);
3147 dropdst:
3148 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3149 dst_release(dst_orig);
3150 xfrm_pols_put(pols, drop_pols);
3151 return ERR_PTR(err);
3152 }
3153 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3154
3155 /* Main function: finds/creates a bundle for given flow.
3156 *
3157 * At the moment we eat a raw IP route. Mostly to speed up lookups
3158 * on interfaces with disabled IPsec.
3159 */
3160 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3161 const struct flowi *fl, const struct sock *sk,
3162 int flags)
3163 {
3164 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3165 }
3166 EXPORT_SYMBOL(xfrm_lookup);
3167
3168 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3169 * Otherwise we may send out blackholed packets.
3170 */
3171 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3172 const struct flowi *fl,
3173 const struct sock *sk, int flags)
3174 {
3175 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3176 flags | XFRM_LOOKUP_QUEUE |
3177 XFRM_LOOKUP_KEEP_DST_REF);
3178
3179 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
3180 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3181
3182 if (IS_ERR(dst))
3183 dst_release(dst_orig);
3184
3185 return dst;
3186 }
3187 EXPORT_SYMBOL(xfrm_lookup_route);
3188
3189 static inline int
3190 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3191 {
3192 struct sec_path *sp = skb_sec_path(skb);
3193 struct xfrm_state *x;
3194
3195 if (!sp || idx < 0 || idx >= sp->len)
3196 return 0;
3197 x = sp->xvec[idx];
3198 if (!x->type->reject)
3199 return 0;
3200 return x->type->reject(x, skb, fl);
3201 }
3202
3203 /* When skb is transformed back to its "native" form, we have to
3204 * check policy restrictions. At the moment we make this in maximally
3205 * stupid way. Shame on me. :-) Of course, connected sockets must
3206 * have policy cached at them.
3207 */
3208
3209 static inline int
3210 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3211 unsigned short family)
3212 {
3213 if (xfrm_state_kern(x))
3214 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3215 return x->id.proto == tmpl->id.proto &&
3216 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3217 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3218 x->props.mode == tmpl->mode &&
3219 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3220 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3221 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3222 xfrm_state_addr_cmp(tmpl, x, family));
3223 }
3224
3225 /*
3226 * 0 or more than 0 is returned when validation is succeeded (either bypass
3227 * because of optional transport mode, or next index of the mathced secpath
3228 * state with the template.
3229 * -1 is returned when no matching template is found.
3230 * Otherwise "-2 - errored_index" is returned.
3231 */
3232 static inline int
3233 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3234 unsigned short family)
3235 {
3236 int idx = start;
3237
3238 if (tmpl->optional) {
3239 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3240 return start;
3241 } else
3242 start = -1;
3243 for (; idx < sp->len; idx++) {
3244 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3245 return ++idx;
3246 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3247 if (start == -1)
3248 start = -2-idx;
3249 break;
3250 }
3251 }
3252 return start;
3253 }
3254
3255 static void
3256 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3257 {
3258 const struct iphdr *iph = ip_hdr(skb);
3259 int ihl = iph->ihl;
3260 u8 *xprth = skb_network_header(skb) + ihl * 4;
3261 struct flowi4 *fl4 = &fl->u.ip4;
3262 int oif = 0;
3263
3264 if (skb_dst(skb) && skb_dst(skb)->dev)
3265 oif = skb_dst(skb)->dev->ifindex;
3266
3267 memset(fl4, 0, sizeof(struct flowi4));
3268 fl4->flowi4_mark = skb->mark;
3269 fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3270
3271 fl4->flowi4_proto = iph->protocol;
3272 fl4->daddr = reverse ? iph->saddr : iph->daddr;
3273 fl4->saddr = reverse ? iph->daddr : iph->saddr;
3274 fl4->flowi4_tos = iph->tos;
3275
3276 if (!ip_is_fragment(iph)) {
3277 switch (iph->protocol) {
3278 case IPPROTO_UDP:
3279 case IPPROTO_UDPLITE:
3280 case IPPROTO_TCP:
3281 case IPPROTO_SCTP:
3282 case IPPROTO_DCCP:
3283 if (xprth + 4 < skb->data ||
3284 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3285 __be16 *ports;
3286
3287 xprth = skb_network_header(skb) + ihl * 4;
3288 ports = (__be16 *)xprth;
3289
3290 fl4->fl4_sport = ports[!!reverse];
3291 fl4->fl4_dport = ports[!reverse];
3292 }
3293 break;
3294 case IPPROTO_ICMP:
3295 if (xprth + 2 < skb->data ||
3296 pskb_may_pull(skb, xprth + 2 - skb->data)) {
3297 u8 *icmp;
3298
3299 xprth = skb_network_header(skb) + ihl * 4;
3300 icmp = xprth;
3301
3302 fl4->fl4_icmp_type = icmp[0];
3303 fl4->fl4_icmp_code = icmp[1];
3304 }
3305 break;
3306 case IPPROTO_ESP:
3307 if (xprth + 4 < skb->data ||
3308 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3309 __be32 *ehdr;
3310
3311 xprth = skb_network_header(skb) + ihl * 4;
3312 ehdr = (__be32 *)xprth;
3313
3314 fl4->fl4_ipsec_spi = ehdr[0];
3315 }
3316 break;
3317 case IPPROTO_AH:
3318 if (xprth + 8 < skb->data ||
3319 pskb_may_pull(skb, xprth + 8 - skb->data)) {
3320 __be32 *ah_hdr;
3321
3322 xprth = skb_network_header(skb) + ihl * 4;
3323 ah_hdr = (__be32 *)xprth;
3324
3325 fl4->fl4_ipsec_spi = ah_hdr[1];
3326 }
3327 break;
3328 case IPPROTO_COMP:
3329 if (xprth + 4 < skb->data ||
3330 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3331 __be16 *ipcomp_hdr;
3332
3333 xprth = skb_network_header(skb) + ihl * 4;
3334 ipcomp_hdr = (__be16 *)xprth;
3335
3336 fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
3337 }
3338 break;
3339 case IPPROTO_GRE:
3340 if (xprth + 12 < skb->data ||
3341 pskb_may_pull(skb, xprth + 12 - skb->data)) {
3342 __be16 *greflags;
3343 __be32 *gre_hdr;
3344
3345 xprth = skb_network_header(skb) + ihl * 4;
3346 greflags = (__be16 *)xprth;
3347 gre_hdr = (__be32 *)xprth;
3348
3349 if (greflags[0] & GRE_KEY) {
3350 if (greflags[0] & GRE_CSUM)
3351 gre_hdr++;
3352 fl4->fl4_gre_key = gre_hdr[1];
3353 }
3354 }
3355 break;
3356 default:
3357 fl4->fl4_ipsec_spi = 0;
3358 break;
3359 }
3360 }
3361 }
3362
3363 #if IS_ENABLED(CONFIG_IPV6)
3364 static void
3365 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3366 {
3367 struct flowi6 *fl6 = &fl->u.ip6;
3368 int onlyproto = 0;
3369 const struct ipv6hdr *hdr = ipv6_hdr(skb);
3370 u32 offset = sizeof(*hdr);
3371 struct ipv6_opt_hdr *exthdr;
3372 const unsigned char *nh = skb_network_header(skb);
3373 u16 nhoff = IP6CB(skb)->nhoff;
3374 int oif = 0;
3375 u8 nexthdr;
3376
3377 if (!nhoff)
3378 nhoff = offsetof(struct ipv6hdr, nexthdr);
3379
3380 nexthdr = nh[nhoff];
3381
3382 if (skb_dst(skb) && skb_dst(skb)->dev)
3383 oif = skb_dst(skb)->dev->ifindex;
3384
3385 memset(fl6, 0, sizeof(struct flowi6));
3386 fl6->flowi6_mark = skb->mark;
3387 fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3388
3389 fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3390 fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3391
3392 while (nh + offset + sizeof(*exthdr) < skb->data ||
3393 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3394 nh = skb_network_header(skb);
3395 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3396
3397 switch (nexthdr) {
3398 case NEXTHDR_FRAGMENT:
3399 onlyproto = 1;
3400 /* fall through */
3401 case NEXTHDR_ROUTING:
3402 case NEXTHDR_HOP:
3403 case NEXTHDR_DEST:
3404 offset += ipv6_optlen(exthdr);
3405 nexthdr = exthdr->nexthdr;
3406 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3407 break;
3408 case IPPROTO_UDP:
3409 case IPPROTO_UDPLITE:
3410 case IPPROTO_TCP:
3411 case IPPROTO_SCTP:
3412 case IPPROTO_DCCP:
3413 if (!onlyproto && (nh + offset + 4 < skb->data ||
3414 pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3415 __be16 *ports;
3416
3417 nh = skb_network_header(skb);
3418 ports = (__be16 *)(nh + offset);
3419 fl6->fl6_sport = ports[!!reverse];
3420 fl6->fl6_dport = ports[!reverse];
3421 }
3422 fl6->flowi6_proto = nexthdr;
3423 return;
3424 case IPPROTO_ICMPV6:
3425 if (!onlyproto && (nh + offset + 2 < skb->data ||
3426 pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3427 u8 *icmp;
3428
3429 nh = skb_network_header(skb);
3430 icmp = (u8 *)(nh + offset);
3431 fl6->fl6_icmp_type = icmp[0];
3432 fl6->fl6_icmp_code = icmp[1];
3433 }
3434 fl6->flowi6_proto = nexthdr;
3435 return;
3436 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3437 case IPPROTO_MH:
3438 offset += ipv6_optlen(exthdr);
3439 if (!onlyproto && (nh + offset + 3 < skb->data ||
3440 pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3441 struct ip6_mh *mh;
3442
3443 nh = skb_network_header(skb);
3444 mh = (struct ip6_mh *)(nh + offset);
3445 fl6->fl6_mh_type = mh->ip6mh_type;
3446 }
3447 fl6->flowi6_proto = nexthdr;
3448 return;
3449 #endif
3450 /* XXX Why are there these headers? */
3451 case IPPROTO_AH:
3452 case IPPROTO_ESP:
3453 case IPPROTO_COMP:
3454 default:
3455 fl6->fl6_ipsec_spi = 0;
3456 fl6->flowi6_proto = nexthdr;
3457 return;
3458 }
3459 }
3460 }
3461 #endif
3462
3463 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3464 unsigned int family, int reverse)
3465 {
3466 switch (family) {
3467 case AF_INET:
3468 decode_session4(skb, fl, reverse);
3469 break;
3470 #if IS_ENABLED(CONFIG_IPV6)
3471 case AF_INET6:
3472 decode_session6(skb, fl, reverse);
3473 break;
3474 #endif
3475 default:
3476 return -EAFNOSUPPORT;
3477 }
3478
3479 return security_xfrm_decode_session(skb, &fl->flowi_secid);
3480 }
3481 EXPORT_SYMBOL(__xfrm_decode_session);
3482
3483 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3484 {
3485 for (; k < sp->len; k++) {
3486 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3487 *idxp = k;
3488 return 1;
3489 }
3490 }
3491
3492 return 0;
3493 }
3494
3495 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3496 unsigned short family)
3497 {
3498 struct net *net = dev_net(skb->dev);
3499 struct xfrm_policy *pol;
3500 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3501 int npols = 0;
3502 int xfrm_nr;
3503 int pi;
3504 int reverse;
3505 struct flowi fl;
3506 int xerr_idx = -1;
3507 const struct xfrm_if_cb *ifcb;
3508 struct sec_path *sp;
3509 struct xfrm_if *xi;
3510 u32 if_id = 0;
3511
3512 rcu_read_lock();
3513 ifcb = xfrm_if_get_cb();
3514
3515 if (ifcb) {
3516 xi = ifcb->decode_session(skb, family);
3517 if (xi) {
3518 if_id = xi->p.if_id;
3519 net = xi->net;
3520 }
3521 }
3522 rcu_read_unlock();
3523
3524 reverse = dir & ~XFRM_POLICY_MASK;
3525 dir &= XFRM_POLICY_MASK;
3526
3527 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3528 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3529 return 0;
3530 }
3531
3532 nf_nat_decode_session(skb, &fl, family);
3533
3534 /* First, check used SA against their selectors. */
3535 sp = skb_sec_path(skb);
3536 if (sp) {
3537 int i;
3538
3539 for (i = sp->len - 1; i >= 0; i--) {
3540 struct xfrm_state *x = sp->xvec[i];
3541 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3542 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3543 return 0;
3544 }
3545 }
3546 }
3547
3548 pol = NULL;
3549 sk = sk_to_full_sk(sk);
3550 if (sk && sk->sk_policy[dir]) {
3551 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3552 if (IS_ERR(pol)) {
3553 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3554 return 0;
3555 }
3556 }
3557
3558 if (!pol)
3559 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3560
3561 if (IS_ERR(pol)) {
3562 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3563 return 0;
3564 }
3565
3566 if (!pol) {
3567 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3568 xfrm_secpath_reject(xerr_idx, skb, &fl);
3569 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3570 return 0;
3571 }
3572 return 1;
3573 }
3574
3575 pol->curlft.use_time = ktime_get_real_seconds();
3576
3577 pols[0] = pol;
3578 npols++;
3579 #ifdef CONFIG_XFRM_SUB_POLICY
3580 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3581 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3582 &fl, family,
3583 XFRM_POLICY_IN, if_id);
3584 if (pols[1]) {
3585 if (IS_ERR(pols[1])) {
3586 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3587 return 0;
3588 }
3589 pols[1]->curlft.use_time = ktime_get_real_seconds();
3590 npols++;
3591 }
3592 }
3593 #endif
3594
3595 if (pol->action == XFRM_POLICY_ALLOW) {
3596 static struct sec_path dummy;
3597 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3598 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3599 struct xfrm_tmpl **tpp = tp;
3600 int ti = 0;
3601 int i, k;
3602
3603 sp = skb_sec_path(skb);
3604 if (!sp)
3605 sp = &dummy;
3606
3607 for (pi = 0; pi < npols; pi++) {
3608 if (pols[pi] != pol &&
3609 pols[pi]->action != XFRM_POLICY_ALLOW) {
3610 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3611 goto reject;
3612 }
3613 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3614 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3615 goto reject_error;
3616 }
3617 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3618 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3619 }
3620 xfrm_nr = ti;
3621 if (npols > 1) {
3622 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3623 tpp = stp;
3624 }
3625
3626 /* For each tunnel xfrm, find the first matching tmpl.
3627 * For each tmpl before that, find corresponding xfrm.
3628 * Order is _important_. Later we will implement
3629 * some barriers, but at the moment barriers
3630 * are implied between each two transformations.
3631 */
3632 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3633 k = xfrm_policy_ok(tpp[i], sp, k, family);
3634 if (k < 0) {
3635 if (k < -1)
3636 /* "-2 - errored_index" returned */
3637 xerr_idx = -(2+k);
3638 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3639 goto reject;
3640 }
3641 }
3642
3643 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3644 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3645 goto reject;
3646 }
3647
3648 xfrm_pols_put(pols, npols);
3649 return 1;
3650 }
3651 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3652
3653 reject:
3654 xfrm_secpath_reject(xerr_idx, skb, &fl);
3655 reject_error:
3656 xfrm_pols_put(pols, npols);
3657 return 0;
3658 }
3659 EXPORT_SYMBOL(__xfrm_policy_check);
3660
3661 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3662 {
3663 struct net *net = dev_net(skb->dev);
3664 struct flowi fl;
3665 struct dst_entry *dst;
3666 int res = 1;
3667
3668 if (xfrm_decode_session(skb, &fl, family) < 0) {
3669 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3670 return 0;
3671 }
3672
3673 skb_dst_force(skb);
3674 if (!skb_dst(skb)) {
3675 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3676 return 0;
3677 }
3678
3679 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3680 if (IS_ERR(dst)) {
3681 res = 0;
3682 dst = NULL;
3683 }
3684 skb_dst_set(skb, dst);
3685 return res;
3686 }
3687 EXPORT_SYMBOL(__xfrm_route_forward);
3688
3689 /* Optimize later using cookies and generation ids. */
3690
3691 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3692 {
3693 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3694 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3695 * get validated by dst_ops->check on every use. We do this
3696 * because when a normal route referenced by an XFRM dst is
3697 * obsoleted we do not go looking around for all parent
3698 * referencing XFRM dsts so that we can invalidate them. It
3699 * is just too much work. Instead we make the checks here on
3700 * every use. For example:
3701 *
3702 * XFRM dst A --> IPv4 dst X
3703 *
3704 * X is the "xdst->route" of A (X is also the "dst->path" of A
3705 * in this example). If X is marked obsolete, "A" will not
3706 * notice. That's what we are validating here via the
3707 * stale_bundle() check.
3708 *
3709 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3710 * be marked on it.
3711 * This will force stale_bundle() to fail on any xdst bundle with
3712 * this dst linked in it.
3713 */
3714 if (dst->obsolete < 0 && !stale_bundle(dst))
3715 return dst;
3716
3717 return NULL;
3718 }
3719
3720 static int stale_bundle(struct dst_entry *dst)
3721 {
3722 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3723 }
3724
3725 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3726 {
3727 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3728 dst->dev = dev_net(dev)->loopback_dev;
3729 dev_hold(dst->dev);
3730 dev_put(dev);
3731 }
3732 }
3733 EXPORT_SYMBOL(xfrm_dst_ifdown);
3734
3735 static void xfrm_link_failure(struct sk_buff *skb)
3736 {
3737 /* Impossible. Such dst must be popped before reaches point of failure. */
3738 }
3739
3740 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3741 {
3742 if (dst) {
3743 if (dst->obsolete) {
3744 dst_release(dst);
3745 dst = NULL;
3746 }
3747 }
3748 return dst;
3749 }
3750
3751 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3752 {
3753 while (nr--) {
3754 struct xfrm_dst *xdst = bundle[nr];
3755 u32 pmtu, route_mtu_cached;
3756 struct dst_entry *dst;
3757
3758 dst = &xdst->u.dst;
3759 pmtu = dst_mtu(xfrm_dst_child(dst));
3760 xdst->child_mtu_cached = pmtu;
3761
3762 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3763
3764 route_mtu_cached = dst_mtu(xdst->route);
3765 xdst->route_mtu_cached = route_mtu_cached;
3766
3767 if (pmtu > route_mtu_cached)
3768 pmtu = route_mtu_cached;
3769
3770 dst_metric_set(dst, RTAX_MTU, pmtu);
3771 }
3772 }
3773
3774 /* Check that the bundle accepts the flow and its components are
3775 * still valid.
3776 */
3777
3778 static int xfrm_bundle_ok(struct xfrm_dst *first)
3779 {
3780 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3781 struct dst_entry *dst = &first->u.dst;
3782 struct xfrm_dst *xdst;
3783 int start_from, nr;
3784 u32 mtu;
3785
3786 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3787 (dst->dev && !netif_running(dst->dev)))
3788 return 0;
3789
3790 if (dst->flags & DST_XFRM_QUEUE)
3791 return 1;
3792
3793 start_from = nr = 0;
3794 do {
3795 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3796
3797 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3798 return 0;
3799 if (xdst->xfrm_genid != dst->xfrm->genid)
3800 return 0;
3801 if (xdst->num_pols > 0 &&
3802 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3803 return 0;
3804
3805 bundle[nr++] = xdst;
3806
3807 mtu = dst_mtu(xfrm_dst_child(dst));
3808 if (xdst->child_mtu_cached != mtu) {
3809 start_from = nr;
3810 xdst->child_mtu_cached = mtu;
3811 }
3812
3813 if (!dst_check(xdst->route, xdst->route_cookie))
3814 return 0;
3815 mtu = dst_mtu(xdst->route);
3816 if (xdst->route_mtu_cached != mtu) {
3817 start_from = nr;
3818 xdst->route_mtu_cached = mtu;
3819 }
3820
3821 dst = xfrm_dst_child(dst);
3822 } while (dst->xfrm);
3823
3824 if (likely(!start_from))
3825 return 1;
3826
3827 xdst = bundle[start_from - 1];
3828 mtu = xdst->child_mtu_cached;
3829 while (start_from--) {
3830 dst = &xdst->u.dst;
3831
3832 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3833 if (mtu > xdst->route_mtu_cached)
3834 mtu = xdst->route_mtu_cached;
3835 dst_metric_set(dst, RTAX_MTU, mtu);
3836 if (!start_from)
3837 break;
3838
3839 xdst = bundle[start_from - 1];
3840 xdst->child_mtu_cached = mtu;
3841 }
3842
3843 return 1;
3844 }
3845
3846 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3847 {
3848 return dst_metric_advmss(xfrm_dst_path(dst));
3849 }
3850
3851 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3852 {
3853 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3854
3855 return mtu ? : dst_mtu(xfrm_dst_path(dst));
3856 }
3857
3858 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3859 const void *daddr)
3860 {
3861 while (dst->xfrm) {
3862 const struct xfrm_state *xfrm = dst->xfrm;
3863
3864 dst = xfrm_dst_child(dst);
3865
3866 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3867 continue;
3868 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3869 daddr = xfrm->coaddr;
3870 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3871 daddr = &xfrm->id.daddr;
3872 }
3873 return daddr;
3874 }
3875
3876 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3877 struct sk_buff *skb,
3878 const void *daddr)
3879 {
3880 const struct dst_entry *path = xfrm_dst_path(dst);
3881
3882 if (!skb)
3883 daddr = xfrm_get_dst_nexthop(dst, daddr);
3884 return path->ops->neigh_lookup(path, skb, daddr);
3885 }
3886
3887 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3888 {
3889 const struct dst_entry *path = xfrm_dst_path(dst);
3890
3891 daddr = xfrm_get_dst_nexthop(dst, daddr);
3892 path->ops->confirm_neigh(path, daddr);
3893 }
3894
3895 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3896 {
3897 int err = 0;
3898
3899 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3900 return -EAFNOSUPPORT;
3901
3902 spin_lock(&xfrm_policy_afinfo_lock);
3903 if (unlikely(xfrm_policy_afinfo[family] != NULL))
3904 err = -EEXIST;
3905 else {
3906 struct dst_ops *dst_ops = afinfo->dst_ops;
3907 if (likely(dst_ops->kmem_cachep == NULL))
3908 dst_ops->kmem_cachep = xfrm_dst_cache;
3909 if (likely(dst_ops->check == NULL))
3910 dst_ops->check = xfrm_dst_check;
3911 if (likely(dst_ops->default_advmss == NULL))
3912 dst_ops->default_advmss = xfrm_default_advmss;
3913 if (likely(dst_ops->mtu == NULL))
3914 dst_ops->mtu = xfrm_mtu;
3915 if (likely(dst_ops->negative_advice == NULL))
3916 dst_ops->negative_advice = xfrm_negative_advice;
3917 if (likely(dst_ops->link_failure == NULL))
3918 dst_ops->link_failure = xfrm_link_failure;
3919 if (likely(dst_ops->neigh_lookup == NULL))
3920 dst_ops->neigh_lookup = xfrm_neigh_lookup;
3921 if (likely(!dst_ops->confirm_neigh))
3922 dst_ops->confirm_neigh = xfrm_confirm_neigh;
3923 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3924 }
3925 spin_unlock(&xfrm_policy_afinfo_lock);
3926
3927 return err;
3928 }
3929 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3930
3931 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3932 {
3933 struct dst_ops *dst_ops = afinfo->dst_ops;
3934 int i;
3935
3936 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3937 if (xfrm_policy_afinfo[i] != afinfo)
3938 continue;
3939 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3940 break;
3941 }
3942
3943 synchronize_rcu();
3944
3945 dst_ops->kmem_cachep = NULL;
3946 dst_ops->check = NULL;
3947 dst_ops->negative_advice = NULL;
3948 dst_ops->link_failure = NULL;
3949 }
3950 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3951
3952 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3953 {
3954 spin_lock(&xfrm_if_cb_lock);
3955 rcu_assign_pointer(xfrm_if_cb, ifcb);
3956 spin_unlock(&xfrm_if_cb_lock);
3957 }
3958 EXPORT_SYMBOL(xfrm_if_register_cb);
3959
3960 void xfrm_if_unregister_cb(void)
3961 {
3962 RCU_INIT_POINTER(xfrm_if_cb, NULL);
3963 synchronize_rcu();
3964 }
3965 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3966
3967 #ifdef CONFIG_XFRM_STATISTICS
3968 static int __net_init xfrm_statistics_init(struct net *net)
3969 {
3970 int rv;
3971 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3972 if (!net->mib.xfrm_statistics)
3973 return -ENOMEM;
3974 rv = xfrm_proc_init(net);
3975 if (rv < 0)
3976 free_percpu(net->mib.xfrm_statistics);
3977 return rv;
3978 }
3979
3980 static void xfrm_statistics_fini(struct net *net)
3981 {
3982 xfrm_proc_fini(net);
3983 free_percpu(net->mib.xfrm_statistics);
3984 }
3985 #else
3986 static int __net_init xfrm_statistics_init(struct net *net)
3987 {
3988 return 0;
3989 }
3990
3991 static void xfrm_statistics_fini(struct net *net)
3992 {
3993 }
3994 #endif
3995
3996 static int __net_init xfrm_policy_init(struct net *net)
3997 {
3998 unsigned int hmask, sz;
3999 int dir, err;
4000
4001 if (net_eq(net, &init_net)) {
4002 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4003 sizeof(struct xfrm_dst),
4004 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4005 NULL);
4006 err = rhashtable_init(&xfrm_policy_inexact_table,
4007 &xfrm_pol_inexact_params);
4008 BUG_ON(err);
4009 }
4010
4011 hmask = 8 - 1;
4012 sz = (hmask+1) * sizeof(struct hlist_head);
4013
4014 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4015 if (!net->xfrm.policy_byidx)
4016 goto out_byidx;
4017 net->xfrm.policy_idx_hmask = hmask;
4018
4019 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4020 struct xfrm_policy_hash *htab;
4021
4022 net->xfrm.policy_count[dir] = 0;
4023 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4024 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4025
4026 htab = &net->xfrm.policy_bydst[dir];
4027 htab->table = xfrm_hash_alloc(sz);
4028 if (!htab->table)
4029 goto out_bydst;
4030 htab->hmask = hmask;
4031 htab->dbits4 = 32;
4032 htab->sbits4 = 32;
4033 htab->dbits6 = 128;
4034 htab->sbits6 = 128;
4035 }
4036 net->xfrm.policy_hthresh.lbits4 = 32;
4037 net->xfrm.policy_hthresh.rbits4 = 32;
4038 net->xfrm.policy_hthresh.lbits6 = 128;
4039 net->xfrm.policy_hthresh.rbits6 = 128;
4040
4041 seqlock_init(&net->xfrm.policy_hthresh.lock);
4042
4043 INIT_LIST_HEAD(&net->xfrm.policy_all);
4044 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4045 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4046 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4047 return 0;
4048
4049 out_bydst:
4050 for (dir--; dir >= 0; dir--) {
4051 struct xfrm_policy_hash *htab;
4052
4053 htab = &net->xfrm.policy_bydst[dir];
4054 xfrm_hash_free(htab->table, sz);
4055 }
4056 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4057 out_byidx:
4058 return -ENOMEM;
4059 }
4060
4061 static void xfrm_policy_fini(struct net *net)
4062 {
4063 struct xfrm_pol_inexact_bin *b, *t;
4064 unsigned int sz;
4065 int dir;
4066
4067 flush_work(&net->xfrm.policy_hash_work);
4068 #ifdef CONFIG_XFRM_SUB_POLICY
4069 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4070 #endif
4071 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4072
4073 WARN_ON(!list_empty(&net->xfrm.policy_all));
4074
4075 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4076 struct xfrm_policy_hash *htab;
4077
4078 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4079
4080 htab = &net->xfrm.policy_bydst[dir];
4081 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4082 WARN_ON(!hlist_empty(htab->table));
4083 xfrm_hash_free(htab->table, sz);
4084 }
4085
4086 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4087 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4088 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4089
4090 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4091 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4092 __xfrm_policy_inexact_prune_bin(b, true);
4093 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4094 }
4095
4096 static int __net_init xfrm_net_init(struct net *net)
4097 {
4098 int rv;
4099
4100 /* Initialize the per-net locks here */
4101 spin_lock_init(&net->xfrm.xfrm_state_lock);
4102 spin_lock_init(&net->xfrm.xfrm_policy_lock);
4103 mutex_init(&net->xfrm.xfrm_cfg_mutex);
4104
4105 rv = xfrm_statistics_init(net);
4106 if (rv < 0)
4107 goto out_statistics;
4108 rv = xfrm_state_init(net);
4109 if (rv < 0)
4110 goto out_state;
4111 rv = xfrm_policy_init(net);
4112 if (rv < 0)
4113 goto out_policy;
4114 rv = xfrm_sysctl_init(net);
4115 if (rv < 0)
4116 goto out_sysctl;
4117
4118 return 0;
4119
4120 out_sysctl:
4121 xfrm_policy_fini(net);
4122 out_policy:
4123 xfrm_state_fini(net);
4124 out_state:
4125 xfrm_statistics_fini(net);
4126 out_statistics:
4127 return rv;
4128 }
4129
4130 static void __net_exit xfrm_net_exit(struct net *net)
4131 {
4132 xfrm_sysctl_fini(net);
4133 xfrm_policy_fini(net);
4134 xfrm_state_fini(net);
4135 xfrm_statistics_fini(net);
4136 }
4137
4138 static struct pernet_operations __net_initdata xfrm_net_ops = {
4139 .init = xfrm_net_init,
4140 .exit = xfrm_net_exit,
4141 };
4142
4143 void __init xfrm_init(void)
4144 {
4145 register_pernet_subsys(&xfrm_net_ops);
4146 xfrm_dev_init();
4147 seqcount_init(&xfrm_policy_hash_generation);
4148 xfrm_input_init();
4149
4150 RCU_INIT_POINTER(xfrm_if_cb, NULL);
4151 synchronize_rcu();
4152 }
4153
4154 #ifdef CONFIG_AUDITSYSCALL
4155 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4156 struct audit_buffer *audit_buf)
4157 {
4158 struct xfrm_sec_ctx *ctx = xp->security;
4159 struct xfrm_selector *sel = &xp->selector;
4160
4161 if (ctx)
4162 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4163 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4164
4165 switch (sel->family) {
4166 case AF_INET:
4167 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4168 if (sel->prefixlen_s != 32)
4169 audit_log_format(audit_buf, " src_prefixlen=%d",
4170 sel->prefixlen_s);
4171 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4172 if (sel->prefixlen_d != 32)
4173 audit_log_format(audit_buf, " dst_prefixlen=%d",
4174 sel->prefixlen_d);
4175 break;
4176 case AF_INET6:
4177 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4178 if (sel->prefixlen_s != 128)
4179 audit_log_format(audit_buf, " src_prefixlen=%d",
4180 sel->prefixlen_s);
4181 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4182 if (sel->prefixlen_d != 128)
4183 audit_log_format(audit_buf, " dst_prefixlen=%d",
4184 sel->prefixlen_d);
4185 break;
4186 }
4187 }
4188
4189 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4190 {
4191 struct audit_buffer *audit_buf;
4192
4193 audit_buf = xfrm_audit_start("SPD-add");
4194 if (audit_buf == NULL)
4195 return;
4196 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4197 audit_log_format(audit_buf, " res=%u", result);
4198 xfrm_audit_common_policyinfo(xp, audit_buf);
4199 audit_log_end(audit_buf);
4200 }
4201 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4202
4203 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4204 bool task_valid)
4205 {
4206 struct audit_buffer *audit_buf;
4207
4208 audit_buf = xfrm_audit_start("SPD-delete");
4209 if (audit_buf == NULL)
4210 return;
4211 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4212 audit_log_format(audit_buf, " res=%u", result);
4213 xfrm_audit_common_policyinfo(xp, audit_buf);
4214 audit_log_end(audit_buf);
4215 }
4216 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4217 #endif
4218
4219 #ifdef CONFIG_XFRM_MIGRATE
4220 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4221 const struct xfrm_selector *sel_tgt)
4222 {
4223 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4224 if (sel_tgt->family == sel_cmp->family &&
4225 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4226 sel_cmp->family) &&
4227 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4228 sel_cmp->family) &&
4229 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4230 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4231 return true;
4232 }
4233 } else {
4234 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4235 return true;
4236 }
4237 }
4238 return false;
4239 }
4240
4241 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4242 u8 dir, u8 type, struct net *net)
4243 {
4244 struct xfrm_policy *pol, *ret = NULL;
4245 struct hlist_head *chain;
4246 u32 priority = ~0U;
4247
4248 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4249 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4250 hlist_for_each_entry(pol, chain, bydst) {
4251 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4252 pol->type == type) {
4253 ret = pol;
4254 priority = ret->priority;
4255 break;
4256 }
4257 }
4258 chain = &net->xfrm.policy_inexact[dir];
4259 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4260 if ((pol->priority >= priority) && ret)
4261 break;
4262
4263 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4264 pol->type == type) {
4265 ret = pol;
4266 break;
4267 }
4268 }
4269
4270 xfrm_pol_hold(ret);
4271
4272 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4273
4274 return ret;
4275 }
4276
4277 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4278 {
4279 int match = 0;
4280
4281 if (t->mode == m->mode && t->id.proto == m->proto &&
4282 (m->reqid == 0 || t->reqid == m->reqid)) {
4283 switch (t->mode) {
4284 case XFRM_MODE_TUNNEL:
4285 case XFRM_MODE_BEET:
4286 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4287 m->old_family) &&
4288 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4289 m->old_family)) {
4290 match = 1;
4291 }
4292 break;
4293 case XFRM_MODE_TRANSPORT:
4294 /* in case of transport mode, template does not store
4295 any IP addresses, hence we just compare mode and
4296 protocol */
4297 match = 1;
4298 break;
4299 default:
4300 break;
4301 }
4302 }
4303 return match;
4304 }
4305
4306 /* update endpoint address(es) of template(s) */
4307 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4308 struct xfrm_migrate *m, int num_migrate)
4309 {
4310 struct xfrm_migrate *mp;
4311 int i, j, n = 0;
4312
4313 write_lock_bh(&pol->lock);
4314 if (unlikely(pol->walk.dead)) {
4315 /* target policy has been deleted */
4316 write_unlock_bh(&pol->lock);
4317 return -ENOENT;
4318 }
4319
4320 for (i = 0; i < pol->xfrm_nr; i++) {
4321 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4322 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4323 continue;
4324 n++;
4325 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4326 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4327 continue;
4328 /* update endpoints */
4329 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4330 sizeof(pol->xfrm_vec[i].id.daddr));
4331 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4332 sizeof(pol->xfrm_vec[i].saddr));
4333 pol->xfrm_vec[i].encap_family = mp->new_family;
4334 /* flush bundles */
4335 atomic_inc(&pol->genid);
4336 }
4337 }
4338
4339 write_unlock_bh(&pol->lock);
4340
4341 if (!n)
4342 return -ENODATA;
4343
4344 return 0;
4345 }
4346
4347 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4348 {
4349 int i, j;
4350
4351 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4352 return -EINVAL;
4353
4354 for (i = 0; i < num_migrate; i++) {
4355 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4356 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4357 return -EINVAL;
4358
4359 /* check if there is any duplicated entry */
4360 for (j = i + 1; j < num_migrate; j++) {
4361 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4362 sizeof(m[i].old_daddr)) &&
4363 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4364 sizeof(m[i].old_saddr)) &&
4365 m[i].proto == m[j].proto &&
4366 m[i].mode == m[j].mode &&
4367 m[i].reqid == m[j].reqid &&
4368 m[i].old_family == m[j].old_family)
4369 return -EINVAL;
4370 }
4371 }
4372
4373 return 0;
4374 }
4375
4376 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4377 struct xfrm_migrate *m, int num_migrate,
4378 struct xfrm_kmaddress *k, struct net *net,
4379 struct xfrm_encap_tmpl *encap)
4380 {
4381 int i, err, nx_cur = 0, nx_new = 0;
4382 struct xfrm_policy *pol = NULL;
4383 struct xfrm_state *x, *xc;
4384 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4385 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4386 struct xfrm_migrate *mp;
4387
4388 /* Stage 0 - sanity checks */
4389 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4390 goto out;
4391
4392 if (dir >= XFRM_POLICY_MAX) {
4393 err = -EINVAL;
4394 goto out;
4395 }
4396
4397 /* Stage 1 - find policy */
4398 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4399 err = -ENOENT;
4400 goto out;
4401 }
4402
4403 /* Stage 2 - find and update state(s) */
4404 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4405 if ((x = xfrm_migrate_state_find(mp, net))) {
4406 x_cur[nx_cur] = x;
4407 nx_cur++;
4408 xc = xfrm_state_migrate(x, mp, encap);
4409 if (xc) {
4410 x_new[nx_new] = xc;
4411 nx_new++;
4412 } else {
4413 err = -ENODATA;
4414 goto restore_state;
4415 }
4416 }
4417 }
4418
4419 /* Stage 3 - update policy */
4420 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4421 goto restore_state;
4422
4423 /* Stage 4 - delete old state(s) */
4424 if (nx_cur) {
4425 xfrm_states_put(x_cur, nx_cur);
4426 xfrm_states_delete(x_cur, nx_cur);
4427 }
4428
4429 /* Stage 5 - announce */
4430 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4431
4432 xfrm_pol_put(pol);
4433
4434 return 0;
4435 out:
4436 return err;
4437
4438 restore_state:
4439 if (pol)
4440 xfrm_pol_put(pol);
4441 if (nx_cur)
4442 xfrm_states_put(x_cur, nx_cur);
4443 if (nx_new)
4444 xfrm_states_delete(x_new, nx_new);
4445
4446 return err;
4447 }
4448 EXPORT_SYMBOL(xfrm_migrate);
4449 #endif
Linux with added FPC-III support