Loading...
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 <linux/icmp.h>
33#include <net/dst.h>
34#include <net/flow.h>
35#include <net/inet_ecn.h>
36#include <net/xfrm.h>
37#include <net/ip.h>
38#include <net/gre.h>
39#if IS_ENABLED(CONFIG_IPV6_MIP6)
40#include <net/mip6.h>
41#endif
42#ifdef CONFIG_XFRM_STATISTICS
43#include <net/snmp.h>
44#endif
45#ifdef CONFIG_XFRM_ESPINTCP
46#include <net/espintcp.h>
47#endif
48#include <net/inet_dscp.h>
49
50#include "xfrm_hash.h"
51
52#define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
53#define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
54#define XFRM_MAX_QUEUE_LEN 100
55
56struct xfrm_flo {
57 struct dst_entry *dst_orig;
58 u8 flags;
59};
60
61/* prefixes smaller than this are stored in lists, not trees. */
62#define INEXACT_PREFIXLEN_IPV4 16
63#define INEXACT_PREFIXLEN_IPV6 48
64
65struct xfrm_pol_inexact_node {
66 struct rb_node node;
67 union {
68 xfrm_address_t addr;
69 struct rcu_head rcu;
70 };
71 u8 prefixlen;
72
73 struct rb_root root;
74
75 /* the policies matching this node, can be empty list */
76 struct hlist_head hhead;
77};
78
79/* xfrm inexact policy search tree:
80 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
81 * |
82 * +---- root_d: sorted by daddr:prefix
83 * | |
84 * | xfrm_pol_inexact_node
85 * | |
86 * | +- root: sorted by saddr/prefix
87 * | | |
88 * | | xfrm_pol_inexact_node
89 * | | |
90 * | | + root: unused
91 * | | |
92 * | | + hhead: saddr:daddr policies
93 * | |
94 * | +- coarse policies and all any:daddr policies
95 * |
96 * +---- root_s: sorted by saddr:prefix
97 * | |
98 * | xfrm_pol_inexact_node
99 * | |
100 * | + root: unused
101 * | |
102 * | + hhead: saddr:any policies
103 * |
104 * +---- coarse policies and all any:any policies
105 *
106 * Lookups return four candidate lists:
107 * 1. any:any list from top-level xfrm_pol_inexact_bin
108 * 2. any:daddr list from daddr tree
109 * 3. saddr:daddr list from 2nd level daddr tree
110 * 4. saddr:any list from saddr tree
111 *
112 * This result set then needs to be searched for the policy with
113 * the lowest priority. If two candidates have the same priority, the
114 * struct xfrm_policy pos member with the lower number is used.
115 *
116 * This replicates previous single-list-search algorithm which would
117 * return first matching policy in the (ordered-by-priority) list.
118 */
119
120struct xfrm_pol_inexact_key {
121 possible_net_t net;
122 u32 if_id;
123 u16 family;
124 u8 dir, type;
125};
126
127struct xfrm_pol_inexact_bin {
128 struct xfrm_pol_inexact_key k;
129 struct rhash_head head;
130 /* list containing '*:*' policies */
131 struct hlist_head hhead;
132
133 seqcount_spinlock_t count;
134 /* tree sorted by daddr/prefix */
135 struct rb_root root_d;
136
137 /* tree sorted by saddr/prefix */
138 struct rb_root root_s;
139
140 /* slow path below */
141 struct list_head inexact_bins;
142 struct rcu_head rcu;
143};
144
145enum xfrm_pol_inexact_candidate_type {
146 XFRM_POL_CAND_BOTH,
147 XFRM_POL_CAND_SADDR,
148 XFRM_POL_CAND_DADDR,
149 XFRM_POL_CAND_ANY,
150
151 XFRM_POL_CAND_MAX,
152};
153
154struct xfrm_pol_inexact_candidates {
155 struct hlist_head *res[XFRM_POL_CAND_MAX];
156};
157
158struct xfrm_flow_keys {
159 struct flow_dissector_key_basic basic;
160 struct flow_dissector_key_control control;
161 union {
162 struct flow_dissector_key_ipv4_addrs ipv4;
163 struct flow_dissector_key_ipv6_addrs ipv6;
164 } addrs;
165 struct flow_dissector_key_ip ip;
166 struct flow_dissector_key_icmp icmp;
167 struct flow_dissector_key_ports ports;
168 struct flow_dissector_key_keyid gre;
169};
170
171static struct flow_dissector xfrm_session_dissector __ro_after_init;
172
173static DEFINE_SPINLOCK(xfrm_if_cb_lock);
174static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
175
176static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
177static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
178 __read_mostly;
179
180static struct kmem_cache *xfrm_dst_cache __ro_after_init;
181
182static struct rhashtable xfrm_policy_inexact_table;
183static const struct rhashtable_params xfrm_pol_inexact_params;
184
185static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
186static int stale_bundle(struct dst_entry *dst);
187static int xfrm_bundle_ok(struct xfrm_dst *xdst);
188static void xfrm_policy_queue_process(struct timer_list *t);
189
190static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
191static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
192 int dir);
193
194static struct xfrm_pol_inexact_bin *
195xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
196 u32 if_id);
197
198static struct xfrm_pol_inexact_bin *
199xfrm_policy_inexact_lookup_rcu(struct net *net,
200 u8 type, u16 family, u8 dir, u32 if_id);
201static struct xfrm_policy *
202xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
203 bool excl);
204
205static bool
206xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
207 struct xfrm_pol_inexact_bin *b,
208 const xfrm_address_t *saddr,
209 const xfrm_address_t *daddr);
210
211static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
212{
213 return refcount_inc_not_zero(&policy->refcnt);
214}
215
216static inline bool
217__xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
218{
219 const struct flowi4 *fl4 = &fl->u.ip4;
220
221 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
222 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
223 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
224 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
225 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
226 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
227}
228
229static inline bool
230__xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
231{
232 const struct flowi6 *fl6 = &fl->u.ip6;
233
234 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
235 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
236 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
237 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
238 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
239 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
240}
241
242bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
243 unsigned short family)
244{
245 switch (family) {
246 case AF_INET:
247 return __xfrm4_selector_match(sel, fl);
248 case AF_INET6:
249 return __xfrm6_selector_match(sel, fl);
250 }
251 return false;
252}
253
254static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
255{
256 const struct xfrm_policy_afinfo *afinfo;
257
258 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
259 return NULL;
260 rcu_read_lock();
261 afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
262 if (unlikely(!afinfo))
263 rcu_read_unlock();
264 return afinfo;
265}
266
267/* Called with rcu_read_lock(). */
268static const struct xfrm_if_cb *xfrm_if_get_cb(void)
269{
270 return rcu_dereference(xfrm_if_cb);
271}
272
273struct dst_entry *__xfrm_dst_lookup(int family,
274 const struct xfrm_dst_lookup_params *params)
275{
276 const struct xfrm_policy_afinfo *afinfo;
277 struct dst_entry *dst;
278
279 afinfo = xfrm_policy_get_afinfo(family);
280 if (unlikely(afinfo == NULL))
281 return ERR_PTR(-EAFNOSUPPORT);
282
283 dst = afinfo->dst_lookup(params);
284
285 rcu_read_unlock();
286
287 return dst;
288}
289EXPORT_SYMBOL(__xfrm_dst_lookup);
290
291static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
292 dscp_t dscp, int oif,
293 xfrm_address_t *prev_saddr,
294 xfrm_address_t *prev_daddr,
295 int family, u32 mark)
296{
297 struct xfrm_dst_lookup_params params;
298 struct net *net = xs_net(x);
299 xfrm_address_t *saddr = &x->props.saddr;
300 xfrm_address_t *daddr = &x->id.daddr;
301 struct dst_entry *dst;
302
303 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
304 saddr = x->coaddr;
305 daddr = prev_daddr;
306 }
307 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
308 saddr = prev_saddr;
309 daddr = x->coaddr;
310 }
311
312 params.net = net;
313 params.saddr = saddr;
314 params.daddr = daddr;
315 params.dscp = dscp;
316 params.oif = oif;
317 params.mark = mark;
318 params.ipproto = x->id.proto;
319 if (x->encap) {
320 switch (x->encap->encap_type) {
321 case UDP_ENCAP_ESPINUDP:
322 params.ipproto = IPPROTO_UDP;
323 params.uli.ports.sport = x->encap->encap_sport;
324 params.uli.ports.dport = x->encap->encap_dport;
325 break;
326 case TCP_ENCAP_ESPINTCP:
327 params.ipproto = IPPROTO_TCP;
328 params.uli.ports.sport = x->encap->encap_sport;
329 params.uli.ports.dport = x->encap->encap_dport;
330 break;
331 }
332 }
333
334 dst = __xfrm_dst_lookup(family, ¶ms);
335
336 if (!IS_ERR(dst)) {
337 if (prev_saddr != saddr)
338 memcpy(prev_saddr, saddr, sizeof(*prev_saddr));
339 if (prev_daddr != daddr)
340 memcpy(prev_daddr, daddr, sizeof(*prev_daddr));
341 }
342
343 return dst;
344}
345
346static inline unsigned long make_jiffies(long secs)
347{
348 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
349 return MAX_SCHEDULE_TIMEOUT-1;
350 else
351 return secs*HZ;
352}
353
354static void xfrm_policy_timer(struct timer_list *t)
355{
356 struct xfrm_policy *xp = from_timer(xp, t, timer);
357 time64_t now = ktime_get_real_seconds();
358 time64_t next = TIME64_MAX;
359 int warn = 0;
360 int dir;
361
362 read_lock(&xp->lock);
363
364 if (unlikely(xp->walk.dead))
365 goto out;
366
367 dir = xfrm_policy_id2dir(xp->index);
368
369 if (xp->lft.hard_add_expires_seconds) {
370 time64_t tmo = xp->lft.hard_add_expires_seconds +
371 xp->curlft.add_time - now;
372 if (tmo <= 0)
373 goto expired;
374 if (tmo < next)
375 next = tmo;
376 }
377 if (xp->lft.hard_use_expires_seconds) {
378 time64_t tmo = xp->lft.hard_use_expires_seconds +
379 (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
380 if (tmo <= 0)
381 goto expired;
382 if (tmo < next)
383 next = tmo;
384 }
385 if (xp->lft.soft_add_expires_seconds) {
386 time64_t tmo = xp->lft.soft_add_expires_seconds +
387 xp->curlft.add_time - now;
388 if (tmo <= 0) {
389 warn = 1;
390 tmo = XFRM_KM_TIMEOUT;
391 }
392 if (tmo < next)
393 next = tmo;
394 }
395 if (xp->lft.soft_use_expires_seconds) {
396 time64_t tmo = xp->lft.soft_use_expires_seconds +
397 (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
398 if (tmo <= 0) {
399 warn = 1;
400 tmo = XFRM_KM_TIMEOUT;
401 }
402 if (tmo < next)
403 next = tmo;
404 }
405
406 if (warn)
407 km_policy_expired(xp, dir, 0, 0);
408 if (next != TIME64_MAX &&
409 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
410 xfrm_pol_hold(xp);
411
412out:
413 read_unlock(&xp->lock);
414 xfrm_pol_put(xp);
415 return;
416
417expired:
418 read_unlock(&xp->lock);
419 if (!xfrm_policy_delete(xp, dir))
420 km_policy_expired(xp, dir, 1, 0);
421 xfrm_pol_put(xp);
422}
423
424/* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
425 * SPD calls.
426 */
427
428struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
429{
430 struct xfrm_policy *policy;
431
432 policy = kzalloc(sizeof(struct xfrm_policy), gfp);
433
434 if (policy) {
435 write_pnet(&policy->xp_net, net);
436 INIT_LIST_HEAD(&policy->walk.all);
437 INIT_HLIST_HEAD(&policy->state_cache_list);
438 INIT_HLIST_NODE(&policy->bydst);
439 INIT_HLIST_NODE(&policy->byidx);
440 rwlock_init(&policy->lock);
441 refcount_set(&policy->refcnt, 1);
442 skb_queue_head_init(&policy->polq.hold_queue);
443 timer_setup(&policy->timer, xfrm_policy_timer, 0);
444 timer_setup(&policy->polq.hold_timer,
445 xfrm_policy_queue_process, 0);
446 }
447 return policy;
448}
449EXPORT_SYMBOL(xfrm_policy_alloc);
450
451static void xfrm_policy_destroy_rcu(struct rcu_head *head)
452{
453 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
454
455 security_xfrm_policy_free(policy->security);
456 kfree(policy);
457}
458
459/* Destroy xfrm_policy: descendant resources must be released to this moment. */
460
461void xfrm_policy_destroy(struct xfrm_policy *policy)
462{
463 BUG_ON(!policy->walk.dead);
464
465 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
466 BUG();
467
468 xfrm_dev_policy_free(policy);
469 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
470}
471EXPORT_SYMBOL(xfrm_policy_destroy);
472
473/* Rule must be locked. Release descendant resources, announce
474 * entry dead. The rule must be unlinked from lists to the moment.
475 */
476
477static void xfrm_policy_kill(struct xfrm_policy *policy)
478{
479 struct net *net = xp_net(policy);
480 struct xfrm_state *x;
481
482 xfrm_dev_policy_delete(policy);
483
484 write_lock_bh(&policy->lock);
485 policy->walk.dead = 1;
486 write_unlock_bh(&policy->lock);
487
488 atomic_inc(&policy->genid);
489
490 if (del_timer(&policy->polq.hold_timer))
491 xfrm_pol_put(policy);
492 skb_queue_purge(&policy->polq.hold_queue);
493
494 if (del_timer(&policy->timer))
495 xfrm_pol_put(policy);
496
497 /* XXX: Flush state cache */
498 spin_lock_bh(&net->xfrm.xfrm_state_lock);
499 hlist_for_each_entry_rcu(x, &policy->state_cache_list, state_cache) {
500 hlist_del_init_rcu(&x->state_cache);
501 }
502 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
503
504 xfrm_pol_put(policy);
505}
506
507static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
508
509static inline unsigned int idx_hash(struct net *net, u32 index)
510{
511 return __idx_hash(index, net->xfrm.policy_idx_hmask);
512}
513
514/* calculate policy hash thresholds */
515static void __get_hash_thresh(struct net *net,
516 unsigned short family, int dir,
517 u8 *dbits, u8 *sbits)
518{
519 switch (family) {
520 case AF_INET:
521 *dbits = net->xfrm.policy_bydst[dir].dbits4;
522 *sbits = net->xfrm.policy_bydst[dir].sbits4;
523 break;
524
525 case AF_INET6:
526 *dbits = net->xfrm.policy_bydst[dir].dbits6;
527 *sbits = net->xfrm.policy_bydst[dir].sbits6;
528 break;
529
530 default:
531 *dbits = 0;
532 *sbits = 0;
533 }
534}
535
536static struct hlist_head *policy_hash_bysel(struct net *net,
537 const struct xfrm_selector *sel,
538 unsigned short family, int dir)
539{
540 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
541 unsigned int hash;
542 u8 dbits;
543 u8 sbits;
544
545 __get_hash_thresh(net, family, dir, &dbits, &sbits);
546 hash = __sel_hash(sel, family, hmask, dbits, sbits);
547
548 if (hash == hmask + 1)
549 return NULL;
550
551 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
552 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
553}
554
555static struct hlist_head *policy_hash_direct(struct net *net,
556 const xfrm_address_t *daddr,
557 const xfrm_address_t *saddr,
558 unsigned short family, int dir)
559{
560 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
561 unsigned int hash;
562 u8 dbits;
563 u8 sbits;
564
565 __get_hash_thresh(net, family, dir, &dbits, &sbits);
566 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
567
568 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
569 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
570}
571
572static void xfrm_dst_hash_transfer(struct net *net,
573 struct hlist_head *list,
574 struct hlist_head *ndsttable,
575 unsigned int nhashmask,
576 int dir)
577{
578 struct hlist_node *tmp, *entry0 = NULL;
579 struct xfrm_policy *pol;
580 unsigned int h0 = 0;
581 u8 dbits;
582 u8 sbits;
583
584redo:
585 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
586 unsigned int h;
587
588 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
589 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
590 pol->family, nhashmask, dbits, sbits);
591 if (!entry0 || pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET) {
592 hlist_del_rcu(&pol->bydst);
593 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
594 h0 = h;
595 } else {
596 if (h != h0)
597 continue;
598 hlist_del_rcu(&pol->bydst);
599 hlist_add_behind_rcu(&pol->bydst, entry0);
600 }
601 entry0 = &pol->bydst;
602 }
603 if (!hlist_empty(list)) {
604 entry0 = NULL;
605 goto redo;
606 }
607}
608
609static void xfrm_idx_hash_transfer(struct hlist_head *list,
610 struct hlist_head *nidxtable,
611 unsigned int nhashmask)
612{
613 struct hlist_node *tmp;
614 struct xfrm_policy *pol;
615
616 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
617 unsigned int h;
618
619 h = __idx_hash(pol->index, nhashmask);
620 hlist_add_head(&pol->byidx, nidxtable+h);
621 }
622}
623
624static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
625{
626 return ((old_hmask + 1) << 1) - 1;
627}
628
629static void xfrm_bydst_resize(struct net *net, int dir)
630{
631 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
632 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
633 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
634 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
635 struct hlist_head *odst;
636 int i;
637
638 if (!ndst)
639 return;
640
641 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
642 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
643
644 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
645 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
646
647 for (i = hmask; i >= 0; i--)
648 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
649
650 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
651 net->xfrm.policy_bydst[dir].hmask = nhashmask;
652
653 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
654 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
655
656 synchronize_rcu();
657
658 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
659}
660
661static void xfrm_byidx_resize(struct net *net)
662{
663 unsigned int hmask = net->xfrm.policy_idx_hmask;
664 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
665 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
666 struct hlist_head *oidx = net->xfrm.policy_byidx;
667 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
668 int i;
669
670 if (!nidx)
671 return;
672
673 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
674
675 for (i = hmask; i >= 0; i--)
676 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
677
678 net->xfrm.policy_byidx = nidx;
679 net->xfrm.policy_idx_hmask = nhashmask;
680
681 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
682
683 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
684}
685
686static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
687{
688 unsigned int cnt = net->xfrm.policy_count[dir];
689 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
690
691 if (total)
692 *total += cnt;
693
694 if ((hmask + 1) < xfrm_policy_hashmax &&
695 cnt > hmask)
696 return 1;
697
698 return 0;
699}
700
701static inline int xfrm_byidx_should_resize(struct net *net, int total)
702{
703 unsigned int hmask = net->xfrm.policy_idx_hmask;
704
705 if ((hmask + 1) < xfrm_policy_hashmax &&
706 total > hmask)
707 return 1;
708
709 return 0;
710}
711
712void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
713{
714 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
715 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
716 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
717 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
718 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
719 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
720 si->spdhcnt = net->xfrm.policy_idx_hmask;
721 si->spdhmcnt = xfrm_policy_hashmax;
722}
723EXPORT_SYMBOL(xfrm_spd_getinfo);
724
725static DEFINE_MUTEX(hash_resize_mutex);
726static void xfrm_hash_resize(struct work_struct *work)
727{
728 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
729 int dir, total;
730
731 mutex_lock(&hash_resize_mutex);
732
733 total = 0;
734 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
735 if (xfrm_bydst_should_resize(net, dir, &total))
736 xfrm_bydst_resize(net, dir);
737 }
738 if (xfrm_byidx_should_resize(net, total))
739 xfrm_byidx_resize(net);
740
741 mutex_unlock(&hash_resize_mutex);
742}
743
744/* Make sure *pol can be inserted into fastbin.
745 * Useful to check that later insert requests will be successful
746 * (provided xfrm_policy_lock is held throughout).
747 */
748static struct xfrm_pol_inexact_bin *
749xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
750{
751 struct xfrm_pol_inexact_bin *bin, *prev;
752 struct xfrm_pol_inexact_key k = {
753 .family = pol->family,
754 .type = pol->type,
755 .dir = dir,
756 .if_id = pol->if_id,
757 };
758 struct net *net = xp_net(pol);
759
760 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
761
762 write_pnet(&k.net, net);
763 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
764 xfrm_pol_inexact_params);
765 if (bin)
766 return bin;
767
768 bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
769 if (!bin)
770 return NULL;
771
772 bin->k = k;
773 INIT_HLIST_HEAD(&bin->hhead);
774 bin->root_d = RB_ROOT;
775 bin->root_s = RB_ROOT;
776 seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
777
778 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
779 &bin->k, &bin->head,
780 xfrm_pol_inexact_params);
781 if (!prev) {
782 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
783 return bin;
784 }
785
786 kfree(bin);
787
788 return IS_ERR(prev) ? NULL : prev;
789}
790
791static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
792 int family, u8 prefixlen)
793{
794 if (xfrm_addr_any(addr, family))
795 return true;
796
797 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
798 return true;
799
800 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
801 return true;
802
803 return false;
804}
805
806static bool
807xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
808{
809 const xfrm_address_t *addr;
810 bool saddr_any, daddr_any;
811 u8 prefixlen;
812
813 addr = &policy->selector.saddr;
814 prefixlen = policy->selector.prefixlen_s;
815
816 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
817 policy->family,
818 prefixlen);
819 addr = &policy->selector.daddr;
820 prefixlen = policy->selector.prefixlen_d;
821 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
822 policy->family,
823 prefixlen);
824 return saddr_any && daddr_any;
825}
826
827static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
828 const xfrm_address_t *addr, u8 prefixlen)
829{
830 node->addr = *addr;
831 node->prefixlen = prefixlen;
832}
833
834static struct xfrm_pol_inexact_node *
835xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
836{
837 struct xfrm_pol_inexact_node *node;
838
839 node = kzalloc(sizeof(*node), GFP_ATOMIC);
840 if (node)
841 xfrm_pol_inexact_node_init(node, addr, prefixlen);
842
843 return node;
844}
845
846static int xfrm_policy_addr_delta(const xfrm_address_t *a,
847 const xfrm_address_t *b,
848 u8 prefixlen, u16 family)
849{
850 u32 ma, mb, mask;
851 unsigned int pdw, pbi;
852 int delta = 0;
853
854 switch (family) {
855 case AF_INET:
856 if (prefixlen == 0)
857 return 0;
858 mask = ~0U << (32 - prefixlen);
859 ma = ntohl(a->a4) & mask;
860 mb = ntohl(b->a4) & mask;
861 if (ma < mb)
862 delta = -1;
863 else if (ma > mb)
864 delta = 1;
865 break;
866 case AF_INET6:
867 pdw = prefixlen >> 5;
868 pbi = prefixlen & 0x1f;
869
870 if (pdw) {
871 delta = memcmp(a->a6, b->a6, pdw << 2);
872 if (delta)
873 return delta;
874 }
875 if (pbi) {
876 mask = ~0U << (32 - pbi);
877 ma = ntohl(a->a6[pdw]) & mask;
878 mb = ntohl(b->a6[pdw]) & mask;
879 if (ma < mb)
880 delta = -1;
881 else if (ma > mb)
882 delta = 1;
883 }
884 break;
885 default:
886 break;
887 }
888
889 return delta;
890}
891
892static void xfrm_policy_inexact_list_reinsert(struct net *net,
893 struct xfrm_pol_inexact_node *n,
894 u16 family)
895{
896 unsigned int matched_s, matched_d;
897 struct xfrm_policy *policy, *p;
898
899 matched_s = 0;
900 matched_d = 0;
901
902 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
903 struct hlist_node *newpos = NULL;
904 bool matches_s, matches_d;
905
906 if (policy->walk.dead || !policy->bydst_reinsert)
907 continue;
908
909 WARN_ON_ONCE(policy->family != family);
910
911 policy->bydst_reinsert = false;
912 hlist_for_each_entry(p, &n->hhead, bydst) {
913 if (policy->priority > p->priority)
914 newpos = &p->bydst;
915 else if (policy->priority == p->priority &&
916 policy->pos > p->pos)
917 newpos = &p->bydst;
918 else
919 break;
920 }
921
922 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
923 hlist_add_behind_rcu(&policy->bydst, newpos);
924 else
925 hlist_add_head_rcu(&policy->bydst, &n->hhead);
926
927 /* paranoia checks follow.
928 * Check that the reinserted policy matches at least
929 * saddr or daddr for current node prefix.
930 *
931 * Matching both is fine, matching saddr in one policy
932 * (but not daddr) and then matching only daddr in another
933 * is a bug.
934 */
935 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
936 &n->addr,
937 n->prefixlen,
938 family) == 0;
939 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
940 &n->addr,
941 n->prefixlen,
942 family) == 0;
943 if (matches_s && matches_d)
944 continue;
945
946 WARN_ON_ONCE(!matches_s && !matches_d);
947 if (matches_s)
948 matched_s++;
949 if (matches_d)
950 matched_d++;
951 WARN_ON_ONCE(matched_s && matched_d);
952 }
953}
954
955static void xfrm_policy_inexact_node_reinsert(struct net *net,
956 struct xfrm_pol_inexact_node *n,
957 struct rb_root *new,
958 u16 family)
959{
960 struct xfrm_pol_inexact_node *node;
961 struct rb_node **p, *parent;
962
963 /* we should not have another subtree here */
964 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
965restart:
966 parent = NULL;
967 p = &new->rb_node;
968 while (*p) {
969 u8 prefixlen;
970 int delta;
971
972 parent = *p;
973 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
974
975 prefixlen = min(node->prefixlen, n->prefixlen);
976
977 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
978 prefixlen, family);
979 if (delta < 0) {
980 p = &parent->rb_left;
981 } else if (delta > 0) {
982 p = &parent->rb_right;
983 } else {
984 bool same_prefixlen = node->prefixlen == n->prefixlen;
985 struct xfrm_policy *tmp;
986
987 hlist_for_each_entry(tmp, &n->hhead, bydst) {
988 tmp->bydst_reinsert = true;
989 hlist_del_rcu(&tmp->bydst);
990 }
991
992 node->prefixlen = prefixlen;
993
994 xfrm_policy_inexact_list_reinsert(net, node, family);
995
996 if (same_prefixlen) {
997 kfree_rcu(n, rcu);
998 return;
999 }
1000
1001 rb_erase(*p, new);
1002 kfree_rcu(n, rcu);
1003 n = node;
1004 goto restart;
1005 }
1006 }
1007
1008 rb_link_node_rcu(&n->node, parent, p);
1009 rb_insert_color(&n->node, new);
1010}
1011
1012/* merge nodes v and n */
1013static void xfrm_policy_inexact_node_merge(struct net *net,
1014 struct xfrm_pol_inexact_node *v,
1015 struct xfrm_pol_inexact_node *n,
1016 u16 family)
1017{
1018 struct xfrm_pol_inexact_node *node;
1019 struct xfrm_policy *tmp;
1020 struct rb_node *rnode;
1021
1022 /* To-be-merged node v has a subtree.
1023 *
1024 * Dismantle it and insert its nodes to n->root.
1025 */
1026 while ((rnode = rb_first(&v->root)) != NULL) {
1027 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
1028 rb_erase(&node->node, &v->root);
1029 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
1030 family);
1031 }
1032
1033 hlist_for_each_entry(tmp, &v->hhead, bydst) {
1034 tmp->bydst_reinsert = true;
1035 hlist_del_rcu(&tmp->bydst);
1036 }
1037
1038 xfrm_policy_inexact_list_reinsert(net, n, family);
1039}
1040
1041static struct xfrm_pol_inexact_node *
1042xfrm_policy_inexact_insert_node(struct net *net,
1043 struct rb_root *root,
1044 xfrm_address_t *addr,
1045 u16 family, u8 prefixlen, u8 dir)
1046{
1047 struct xfrm_pol_inexact_node *cached = NULL;
1048 struct rb_node **p, *parent = NULL;
1049 struct xfrm_pol_inexact_node *node;
1050
1051 p = &root->rb_node;
1052 while (*p) {
1053 int delta;
1054
1055 parent = *p;
1056 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1057
1058 delta = xfrm_policy_addr_delta(addr, &node->addr,
1059 node->prefixlen,
1060 family);
1061 if (delta == 0 && prefixlen >= node->prefixlen) {
1062 WARN_ON_ONCE(cached); /* ipsec policies got lost */
1063 return node;
1064 }
1065
1066 if (delta < 0)
1067 p = &parent->rb_left;
1068 else
1069 p = &parent->rb_right;
1070
1071 if (prefixlen < node->prefixlen) {
1072 delta = xfrm_policy_addr_delta(addr, &node->addr,
1073 prefixlen,
1074 family);
1075 if (delta)
1076 continue;
1077
1078 /* This node is a subnet of the new prefix. It needs
1079 * to be removed and re-inserted with the smaller
1080 * prefix and all nodes that are now also covered
1081 * by the reduced prefixlen.
1082 */
1083 rb_erase(&node->node, root);
1084
1085 if (!cached) {
1086 xfrm_pol_inexact_node_init(node, addr,
1087 prefixlen);
1088 cached = node;
1089 } else {
1090 /* This node also falls within the new
1091 * prefixlen. Merge the to-be-reinserted
1092 * node and this one.
1093 */
1094 xfrm_policy_inexact_node_merge(net, node,
1095 cached, family);
1096 kfree_rcu(node, rcu);
1097 }
1098
1099 /* restart */
1100 p = &root->rb_node;
1101 parent = NULL;
1102 }
1103 }
1104
1105 node = cached;
1106 if (!node) {
1107 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1108 if (!node)
1109 return NULL;
1110 }
1111
1112 rb_link_node_rcu(&node->node, parent, p);
1113 rb_insert_color(&node->node, root);
1114
1115 return node;
1116}
1117
1118static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1119{
1120 struct xfrm_pol_inexact_node *node;
1121 struct rb_node *rn = rb_first(r);
1122
1123 while (rn) {
1124 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1125
1126 xfrm_policy_inexact_gc_tree(&node->root, rm);
1127 rn = rb_next(rn);
1128
1129 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1130 WARN_ON_ONCE(rm);
1131 continue;
1132 }
1133
1134 rb_erase(&node->node, r);
1135 kfree_rcu(node, rcu);
1136 }
1137}
1138
1139static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1140{
1141 write_seqcount_begin(&b->count);
1142 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1143 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1144 write_seqcount_end(&b->count);
1145
1146 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1147 !hlist_empty(&b->hhead)) {
1148 WARN_ON_ONCE(net_exit);
1149 return;
1150 }
1151
1152 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1153 xfrm_pol_inexact_params) == 0) {
1154 list_del(&b->inexact_bins);
1155 kfree_rcu(b, rcu);
1156 }
1157}
1158
1159static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1160{
1161 struct net *net = read_pnet(&b->k.net);
1162
1163 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1164 __xfrm_policy_inexact_prune_bin(b, false);
1165 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1166}
1167
1168static void __xfrm_policy_inexact_flush(struct net *net)
1169{
1170 struct xfrm_pol_inexact_bin *bin, *t;
1171
1172 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1173
1174 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1175 __xfrm_policy_inexact_prune_bin(bin, false);
1176}
1177
1178static struct hlist_head *
1179xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1180 struct xfrm_policy *policy, u8 dir)
1181{
1182 struct xfrm_pol_inexact_node *n;
1183 struct net *net;
1184
1185 net = xp_net(policy);
1186 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1187
1188 if (xfrm_policy_inexact_insert_use_any_list(policy))
1189 return &bin->hhead;
1190
1191 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1192 policy->family,
1193 policy->selector.prefixlen_d)) {
1194 write_seqcount_begin(&bin->count);
1195 n = xfrm_policy_inexact_insert_node(net,
1196 &bin->root_s,
1197 &policy->selector.saddr,
1198 policy->family,
1199 policy->selector.prefixlen_s,
1200 dir);
1201 write_seqcount_end(&bin->count);
1202 if (!n)
1203 return NULL;
1204
1205 return &n->hhead;
1206 }
1207
1208 /* daddr is fixed */
1209 write_seqcount_begin(&bin->count);
1210 n = xfrm_policy_inexact_insert_node(net,
1211 &bin->root_d,
1212 &policy->selector.daddr,
1213 policy->family,
1214 policy->selector.prefixlen_d, dir);
1215 write_seqcount_end(&bin->count);
1216 if (!n)
1217 return NULL;
1218
1219 /* saddr is wildcard */
1220 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1221 policy->family,
1222 policy->selector.prefixlen_s))
1223 return &n->hhead;
1224
1225 write_seqcount_begin(&bin->count);
1226 n = xfrm_policy_inexact_insert_node(net,
1227 &n->root,
1228 &policy->selector.saddr,
1229 policy->family,
1230 policy->selector.prefixlen_s, dir);
1231 write_seqcount_end(&bin->count);
1232 if (!n)
1233 return NULL;
1234
1235 return &n->hhead;
1236}
1237
1238static struct xfrm_policy *
1239xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1240{
1241 struct xfrm_pol_inexact_bin *bin;
1242 struct xfrm_policy *delpol;
1243 struct hlist_head *chain;
1244 struct net *net;
1245
1246 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1247 if (!bin)
1248 return ERR_PTR(-ENOMEM);
1249
1250 net = xp_net(policy);
1251 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1252
1253 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1254 if (!chain) {
1255 __xfrm_policy_inexact_prune_bin(bin, false);
1256 return ERR_PTR(-ENOMEM);
1257 }
1258
1259 delpol = xfrm_policy_insert_list(chain, policy, excl);
1260 if (delpol && excl) {
1261 __xfrm_policy_inexact_prune_bin(bin, false);
1262 return ERR_PTR(-EEXIST);
1263 }
1264
1265 if (delpol)
1266 __xfrm_policy_inexact_prune_bin(bin, false);
1267
1268 return delpol;
1269}
1270
1271static bool xfrm_policy_is_dead_or_sk(const struct xfrm_policy *policy)
1272{
1273 int dir;
1274
1275 if (policy->walk.dead)
1276 return true;
1277
1278 dir = xfrm_policy_id2dir(policy->index);
1279 return dir >= XFRM_POLICY_MAX;
1280}
1281
1282static void xfrm_hash_rebuild(struct work_struct *work)
1283{
1284 struct net *net = container_of(work, struct net,
1285 xfrm.policy_hthresh.work);
1286 struct xfrm_policy *pol;
1287 struct xfrm_policy *policy;
1288 struct hlist_head *chain;
1289 struct hlist_node *newpos;
1290 int dir;
1291 unsigned seq;
1292 u8 lbits4, rbits4, lbits6, rbits6;
1293
1294 mutex_lock(&hash_resize_mutex);
1295
1296 /* read selector prefixlen thresholds */
1297 do {
1298 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1299
1300 lbits4 = net->xfrm.policy_hthresh.lbits4;
1301 rbits4 = net->xfrm.policy_hthresh.rbits4;
1302 lbits6 = net->xfrm.policy_hthresh.lbits6;
1303 rbits6 = net->xfrm.policy_hthresh.rbits6;
1304 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1305
1306 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1307 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1308
1309 /* make sure that we can insert the indirect policies again before
1310 * we start with destructive action.
1311 */
1312 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1313 struct xfrm_pol_inexact_bin *bin;
1314 u8 dbits, sbits;
1315
1316 if (xfrm_policy_is_dead_or_sk(policy))
1317 continue;
1318
1319 dir = xfrm_policy_id2dir(policy->index);
1320 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1321 if (policy->family == AF_INET) {
1322 dbits = rbits4;
1323 sbits = lbits4;
1324 } else {
1325 dbits = rbits6;
1326 sbits = lbits6;
1327 }
1328 } else {
1329 if (policy->family == AF_INET) {
1330 dbits = lbits4;
1331 sbits = rbits4;
1332 } else {
1333 dbits = lbits6;
1334 sbits = rbits6;
1335 }
1336 }
1337
1338 if (policy->selector.prefixlen_d < dbits ||
1339 policy->selector.prefixlen_s < sbits)
1340 continue;
1341
1342 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1343 if (!bin)
1344 goto out_unlock;
1345
1346 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1347 goto out_unlock;
1348 }
1349
1350 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1351 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1352 /* dir out => dst = remote, src = local */
1353 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1354 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1355 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1356 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1357 } else {
1358 /* dir in/fwd => dst = local, src = remote */
1359 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1360 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1361 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1362 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1363 }
1364 }
1365
1366 /* re-insert all policies by order of creation */
1367 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1368 if (xfrm_policy_is_dead_or_sk(policy))
1369 continue;
1370
1371 hlist_del_rcu(&policy->bydst);
1372
1373 newpos = NULL;
1374 dir = xfrm_policy_id2dir(policy->index);
1375 chain = policy_hash_bysel(net, &policy->selector,
1376 policy->family, dir);
1377
1378 if (!chain) {
1379 void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1380
1381 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1382 continue;
1383 }
1384
1385 hlist_for_each_entry(pol, chain, bydst) {
1386 if (policy->priority >= pol->priority)
1387 newpos = &pol->bydst;
1388 else
1389 break;
1390 }
1391 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1392 hlist_add_behind_rcu(&policy->bydst, newpos);
1393 else
1394 hlist_add_head_rcu(&policy->bydst, chain);
1395 }
1396
1397out_unlock:
1398 __xfrm_policy_inexact_flush(net);
1399 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1400 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1401
1402 mutex_unlock(&hash_resize_mutex);
1403}
1404
1405void xfrm_policy_hash_rebuild(struct net *net)
1406{
1407 schedule_work(&net->xfrm.policy_hthresh.work);
1408}
1409EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1410
1411/* Generate new index... KAME seems to generate them ordered by cost
1412 * of an absolute inpredictability of ordering of rules. This will not pass. */
1413static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1414{
1415 for (;;) {
1416 struct hlist_head *list;
1417 struct xfrm_policy *p;
1418 u32 idx;
1419 int found;
1420
1421 if (!index) {
1422 idx = (net->xfrm.idx_generator | dir);
1423 net->xfrm.idx_generator += 8;
1424 } else {
1425 idx = index;
1426 index = 0;
1427 }
1428
1429 if (idx == 0)
1430 idx = 8;
1431 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1432 found = 0;
1433 hlist_for_each_entry(p, list, byidx) {
1434 if (p->index == idx) {
1435 found = 1;
1436 break;
1437 }
1438 }
1439 if (!found)
1440 return idx;
1441 }
1442}
1443
1444static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1445{
1446 u32 *p1 = (u32 *) s1;
1447 u32 *p2 = (u32 *) s2;
1448 int len = sizeof(struct xfrm_selector) / sizeof(u32);
1449 int i;
1450
1451 for (i = 0; i < len; i++) {
1452 if (p1[i] != p2[i])
1453 return 1;
1454 }
1455
1456 return 0;
1457}
1458
1459static void xfrm_policy_requeue(struct xfrm_policy *old,
1460 struct xfrm_policy *new)
1461{
1462 struct xfrm_policy_queue *pq = &old->polq;
1463 struct sk_buff_head list;
1464
1465 if (skb_queue_empty(&pq->hold_queue))
1466 return;
1467
1468 __skb_queue_head_init(&list);
1469
1470 spin_lock_bh(&pq->hold_queue.lock);
1471 skb_queue_splice_init(&pq->hold_queue, &list);
1472 if (del_timer(&pq->hold_timer))
1473 xfrm_pol_put(old);
1474 spin_unlock_bh(&pq->hold_queue.lock);
1475
1476 pq = &new->polq;
1477
1478 spin_lock_bh(&pq->hold_queue.lock);
1479 skb_queue_splice(&list, &pq->hold_queue);
1480 pq->timeout = XFRM_QUEUE_TMO_MIN;
1481 if (!mod_timer(&pq->hold_timer, jiffies))
1482 xfrm_pol_hold(new);
1483 spin_unlock_bh(&pq->hold_queue.lock);
1484}
1485
1486static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1487 struct xfrm_policy *pol)
1488{
1489 return mark->v == pol->mark.v && mark->m == pol->mark.m;
1490}
1491
1492static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1493{
1494 const struct xfrm_pol_inexact_key *k = data;
1495 u32 a = k->type << 24 | k->dir << 16 | k->family;
1496
1497 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1498 seed);
1499}
1500
1501static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1502{
1503 const struct xfrm_pol_inexact_bin *b = data;
1504
1505 return xfrm_pol_bin_key(&b->k, 0, seed);
1506}
1507
1508static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1509 const void *ptr)
1510{
1511 const struct xfrm_pol_inexact_key *key = arg->key;
1512 const struct xfrm_pol_inexact_bin *b = ptr;
1513 int ret;
1514
1515 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1516 return -1;
1517
1518 ret = b->k.dir ^ key->dir;
1519 if (ret)
1520 return ret;
1521
1522 ret = b->k.type ^ key->type;
1523 if (ret)
1524 return ret;
1525
1526 ret = b->k.family ^ key->family;
1527 if (ret)
1528 return ret;
1529
1530 return b->k.if_id ^ key->if_id;
1531}
1532
1533static const struct rhashtable_params xfrm_pol_inexact_params = {
1534 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head),
1535 .hashfn = xfrm_pol_bin_key,
1536 .obj_hashfn = xfrm_pol_bin_obj,
1537 .obj_cmpfn = xfrm_pol_bin_cmp,
1538 .automatic_shrinking = true,
1539};
1540
1541static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1542 struct xfrm_policy *policy,
1543 bool excl)
1544{
1545 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1546
1547 hlist_for_each_entry(pol, chain, bydst) {
1548 if (pol->type == policy->type &&
1549 pol->if_id == policy->if_id &&
1550 !selector_cmp(&pol->selector, &policy->selector) &&
1551 xfrm_policy_mark_match(&policy->mark, pol) &&
1552 xfrm_sec_ctx_match(pol->security, policy->security) &&
1553 !WARN_ON(delpol)) {
1554 if (excl)
1555 return ERR_PTR(-EEXIST);
1556 delpol = pol;
1557 if (policy->priority > pol->priority)
1558 continue;
1559 } else if (policy->priority >= pol->priority) {
1560 newpos = pol;
1561 continue;
1562 }
1563 if (delpol)
1564 break;
1565 }
1566
1567 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1568 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1569 else
1570 /* Packet offload policies enter to the head
1571 * to speed-up lookups.
1572 */
1573 hlist_add_head_rcu(&policy->bydst, chain);
1574
1575 return delpol;
1576}
1577
1578int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1579{
1580 struct net *net = xp_net(policy);
1581 struct xfrm_policy *delpol;
1582 struct hlist_head *chain;
1583
1584 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1585 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1586 if (chain)
1587 delpol = xfrm_policy_insert_list(chain, policy, excl);
1588 else
1589 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1590
1591 if (IS_ERR(delpol)) {
1592 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1593 return PTR_ERR(delpol);
1594 }
1595
1596 __xfrm_policy_link(policy, dir);
1597
1598 /* After previous checking, family can either be AF_INET or AF_INET6 */
1599 if (policy->family == AF_INET)
1600 rt_genid_bump_ipv4(net);
1601 else
1602 rt_genid_bump_ipv6(net);
1603
1604 if (delpol) {
1605 xfrm_policy_requeue(delpol, policy);
1606 __xfrm_policy_unlink(delpol, dir);
1607 }
1608 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1609 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1610 policy->curlft.add_time = ktime_get_real_seconds();
1611 policy->curlft.use_time = 0;
1612 if (!mod_timer(&policy->timer, jiffies + HZ))
1613 xfrm_pol_hold(policy);
1614 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1615
1616 if (delpol)
1617 xfrm_policy_kill(delpol);
1618 else if (xfrm_bydst_should_resize(net, dir, NULL))
1619 schedule_work(&net->xfrm.policy_hash_work);
1620
1621 return 0;
1622}
1623EXPORT_SYMBOL(xfrm_policy_insert);
1624
1625static struct xfrm_policy *
1626__xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1627 u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1628 struct xfrm_sec_ctx *ctx)
1629{
1630 struct xfrm_policy *pol;
1631
1632 if (!chain)
1633 return NULL;
1634
1635 hlist_for_each_entry(pol, chain, bydst) {
1636 if (pol->type == type &&
1637 pol->if_id == if_id &&
1638 xfrm_policy_mark_match(mark, pol) &&
1639 !selector_cmp(sel, &pol->selector) &&
1640 xfrm_sec_ctx_match(ctx, pol->security))
1641 return pol;
1642 }
1643
1644 return NULL;
1645}
1646
1647struct xfrm_policy *
1648xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1649 u8 type, int dir, struct xfrm_selector *sel,
1650 struct xfrm_sec_ctx *ctx, int delete, int *err)
1651{
1652 struct xfrm_pol_inexact_bin *bin = NULL;
1653 struct xfrm_policy *pol, *ret = NULL;
1654 struct hlist_head *chain;
1655
1656 *err = 0;
1657 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1658 chain = policy_hash_bysel(net, sel, sel->family, dir);
1659 if (!chain) {
1660 struct xfrm_pol_inexact_candidates cand;
1661 int i;
1662
1663 bin = xfrm_policy_inexact_lookup(net, type,
1664 sel->family, dir, if_id);
1665 if (!bin) {
1666 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1667 return NULL;
1668 }
1669
1670 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1671 &sel->saddr,
1672 &sel->daddr)) {
1673 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1674 return NULL;
1675 }
1676
1677 pol = NULL;
1678 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1679 struct xfrm_policy *tmp;
1680
1681 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1682 if_id, type, dir,
1683 sel, ctx);
1684 if (!tmp)
1685 continue;
1686
1687 if (!pol || tmp->pos < pol->pos)
1688 pol = tmp;
1689 }
1690 } else {
1691 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1692 sel, ctx);
1693 }
1694
1695 if (pol) {
1696 xfrm_pol_hold(pol);
1697 if (delete) {
1698 *err = security_xfrm_policy_delete(pol->security);
1699 if (*err) {
1700 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1701 return pol;
1702 }
1703 __xfrm_policy_unlink(pol, dir);
1704 }
1705 ret = pol;
1706 }
1707 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1708
1709 if (ret && delete)
1710 xfrm_policy_kill(ret);
1711 if (bin && delete)
1712 xfrm_policy_inexact_prune_bin(bin);
1713 return ret;
1714}
1715EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1716
1717struct xfrm_policy *
1718xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1719 u8 type, int dir, u32 id, int delete, int *err)
1720{
1721 struct xfrm_policy *pol, *ret;
1722 struct hlist_head *chain;
1723
1724 *err = -ENOENT;
1725 if (xfrm_policy_id2dir(id) != dir)
1726 return NULL;
1727
1728 *err = 0;
1729 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1730 chain = net->xfrm.policy_byidx + idx_hash(net, id);
1731 ret = NULL;
1732 hlist_for_each_entry(pol, chain, byidx) {
1733 if (pol->type == type && pol->index == id &&
1734 pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1735 xfrm_pol_hold(pol);
1736 if (delete) {
1737 *err = security_xfrm_policy_delete(
1738 pol->security);
1739 if (*err) {
1740 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1741 return pol;
1742 }
1743 __xfrm_policy_unlink(pol, dir);
1744 }
1745 ret = pol;
1746 break;
1747 }
1748 }
1749 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1750
1751 if (ret && delete)
1752 xfrm_policy_kill(ret);
1753 return ret;
1754}
1755EXPORT_SYMBOL(xfrm_policy_byid);
1756
1757#ifdef CONFIG_SECURITY_NETWORK_XFRM
1758static inline int
1759xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1760{
1761 struct xfrm_policy *pol;
1762 int err = 0;
1763
1764 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1765 if (pol->walk.dead ||
1766 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1767 pol->type != type)
1768 continue;
1769
1770 err = security_xfrm_policy_delete(pol->security);
1771 if (err) {
1772 xfrm_audit_policy_delete(pol, 0, task_valid);
1773 return err;
1774 }
1775 }
1776 return err;
1777}
1778
1779static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1780 struct net_device *dev,
1781 bool task_valid)
1782{
1783 struct xfrm_policy *pol;
1784 int err = 0;
1785
1786 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1787 if (pol->walk.dead ||
1788 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1789 pol->xdo.dev != dev)
1790 continue;
1791
1792 err = security_xfrm_policy_delete(pol->security);
1793 if (err) {
1794 xfrm_audit_policy_delete(pol, 0, task_valid);
1795 return err;
1796 }
1797 }
1798 return err;
1799}
1800#else
1801static inline int
1802xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1803{
1804 return 0;
1805}
1806
1807static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1808 struct net_device *dev,
1809 bool task_valid)
1810{
1811 return 0;
1812}
1813#endif
1814
1815int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1816{
1817 int dir, err = 0, cnt = 0;
1818 struct xfrm_policy *pol;
1819
1820 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1821
1822 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1823 if (err)
1824 goto out;
1825
1826again:
1827 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1828 if (pol->walk.dead)
1829 continue;
1830
1831 dir = xfrm_policy_id2dir(pol->index);
1832 if (dir >= XFRM_POLICY_MAX ||
1833 pol->type != type)
1834 continue;
1835
1836 __xfrm_policy_unlink(pol, dir);
1837 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1838 cnt++;
1839 xfrm_audit_policy_delete(pol, 1, task_valid);
1840 xfrm_policy_kill(pol);
1841 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1842 goto again;
1843 }
1844 if (cnt)
1845 __xfrm_policy_inexact_flush(net);
1846 else
1847 err = -ESRCH;
1848out:
1849 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1850 return err;
1851}
1852EXPORT_SYMBOL(xfrm_policy_flush);
1853
1854int xfrm_dev_policy_flush(struct net *net, struct net_device *dev,
1855 bool task_valid)
1856{
1857 int dir, err = 0, cnt = 0;
1858 struct xfrm_policy *pol;
1859
1860 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1861
1862 err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid);
1863 if (err)
1864 goto out;
1865
1866again:
1867 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1868 if (pol->walk.dead)
1869 continue;
1870
1871 dir = xfrm_policy_id2dir(pol->index);
1872 if (dir >= XFRM_POLICY_MAX ||
1873 pol->xdo.dev != dev)
1874 continue;
1875
1876 __xfrm_policy_unlink(pol, dir);
1877 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1878 cnt++;
1879 xfrm_audit_policy_delete(pol, 1, task_valid);
1880 xfrm_policy_kill(pol);
1881 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1882 goto again;
1883 }
1884 if (cnt)
1885 __xfrm_policy_inexact_flush(net);
1886 else
1887 err = -ESRCH;
1888out:
1889 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1890 return err;
1891}
1892EXPORT_SYMBOL(xfrm_dev_policy_flush);
1893
1894int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1895 int (*func)(struct xfrm_policy *, int, int, void*),
1896 void *data)
1897{
1898 struct xfrm_policy *pol;
1899 struct xfrm_policy_walk_entry *x;
1900 int error = 0;
1901
1902 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1903 walk->type != XFRM_POLICY_TYPE_ANY)
1904 return -EINVAL;
1905
1906 if (list_empty(&walk->walk.all) && walk->seq != 0)
1907 return 0;
1908
1909 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1910 if (list_empty(&walk->walk.all))
1911 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1912 else
1913 x = list_first_entry(&walk->walk.all,
1914 struct xfrm_policy_walk_entry, all);
1915
1916 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1917 if (x->dead)
1918 continue;
1919 pol = container_of(x, struct xfrm_policy, walk);
1920 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1921 walk->type != pol->type)
1922 continue;
1923 error = func(pol, xfrm_policy_id2dir(pol->index),
1924 walk->seq, data);
1925 if (error) {
1926 list_move_tail(&walk->walk.all, &x->all);
1927 goto out;
1928 }
1929 walk->seq++;
1930 }
1931 if (walk->seq == 0) {
1932 error = -ENOENT;
1933 goto out;
1934 }
1935 list_del_init(&walk->walk.all);
1936out:
1937 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1938 return error;
1939}
1940EXPORT_SYMBOL(xfrm_policy_walk);
1941
1942void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1943{
1944 INIT_LIST_HEAD(&walk->walk.all);
1945 walk->walk.dead = 1;
1946 walk->type = type;
1947 walk->seq = 0;
1948}
1949EXPORT_SYMBOL(xfrm_policy_walk_init);
1950
1951void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1952{
1953 if (list_empty(&walk->walk.all))
1954 return;
1955
1956 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1957 list_del(&walk->walk.all);
1958 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1959}
1960EXPORT_SYMBOL(xfrm_policy_walk_done);
1961
1962/*
1963 * Find policy to apply to this flow.
1964 *
1965 * Returns 0 if policy found, else an -errno.
1966 */
1967static int xfrm_policy_match(const struct xfrm_policy *pol,
1968 const struct flowi *fl,
1969 u8 type, u16 family, u32 if_id)
1970{
1971 const struct xfrm_selector *sel = &pol->selector;
1972 int ret = -ESRCH;
1973 bool match;
1974
1975 if (pol->family != family ||
1976 pol->if_id != if_id ||
1977 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1978 pol->type != type)
1979 return ret;
1980
1981 match = xfrm_selector_match(sel, fl, family);
1982 if (match)
1983 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1984 return ret;
1985}
1986
1987static struct xfrm_pol_inexact_node *
1988xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1989 seqcount_spinlock_t *count,
1990 const xfrm_address_t *addr, u16 family)
1991{
1992 const struct rb_node *parent;
1993 int seq;
1994
1995again:
1996 seq = read_seqcount_begin(count);
1997
1998 parent = rcu_dereference_raw(r->rb_node);
1999 while (parent) {
2000 struct xfrm_pol_inexact_node *node;
2001 int delta;
2002
2003 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
2004
2005 delta = xfrm_policy_addr_delta(addr, &node->addr,
2006 node->prefixlen, family);
2007 if (delta < 0) {
2008 parent = rcu_dereference_raw(parent->rb_left);
2009 continue;
2010 } else if (delta > 0) {
2011 parent = rcu_dereference_raw(parent->rb_right);
2012 continue;
2013 }
2014
2015 return node;
2016 }
2017
2018 if (read_seqcount_retry(count, seq))
2019 goto again;
2020
2021 return NULL;
2022}
2023
2024static bool
2025xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
2026 struct xfrm_pol_inexact_bin *b,
2027 const xfrm_address_t *saddr,
2028 const xfrm_address_t *daddr)
2029{
2030 struct xfrm_pol_inexact_node *n;
2031 u16 family;
2032
2033 if (!b)
2034 return false;
2035
2036 family = b->k.family;
2037 memset(cand, 0, sizeof(*cand));
2038 cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
2039
2040 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
2041 family);
2042 if (n) {
2043 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
2044 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
2045 family);
2046 if (n)
2047 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
2048 }
2049
2050 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
2051 family);
2052 if (n)
2053 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
2054
2055 return true;
2056}
2057
2058static struct xfrm_pol_inexact_bin *
2059xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
2060 u8 dir, u32 if_id)
2061{
2062 struct xfrm_pol_inexact_key k = {
2063 .family = family,
2064 .type = type,
2065 .dir = dir,
2066 .if_id = if_id,
2067 };
2068
2069 write_pnet(&k.net, net);
2070
2071 return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
2072 xfrm_pol_inexact_params);
2073}
2074
2075static struct xfrm_pol_inexact_bin *
2076xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2077 u8 dir, u32 if_id)
2078{
2079 struct xfrm_pol_inexact_bin *bin;
2080
2081 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2082
2083 rcu_read_lock();
2084 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2085 rcu_read_unlock();
2086
2087 return bin;
2088}
2089
2090static struct xfrm_policy *
2091__xfrm_policy_eval_candidates(struct hlist_head *chain,
2092 struct xfrm_policy *prefer,
2093 const struct flowi *fl,
2094 u8 type, u16 family, u32 if_id)
2095{
2096 u32 priority = prefer ? prefer->priority : ~0u;
2097 struct xfrm_policy *pol;
2098
2099 if (!chain)
2100 return NULL;
2101
2102 hlist_for_each_entry_rcu(pol, chain, bydst) {
2103 int err;
2104
2105 if (pol->priority > priority)
2106 break;
2107
2108 err = xfrm_policy_match(pol, fl, type, family, if_id);
2109 if (err) {
2110 if (err != -ESRCH)
2111 return ERR_PTR(err);
2112
2113 continue;
2114 }
2115
2116 if (prefer) {
2117 /* matches. Is it older than *prefer? */
2118 if (pol->priority == priority &&
2119 prefer->pos < pol->pos)
2120 return prefer;
2121 }
2122
2123 return pol;
2124 }
2125
2126 return NULL;
2127}
2128
2129static struct xfrm_policy *
2130xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2131 struct xfrm_policy *prefer,
2132 const struct flowi *fl,
2133 u8 type, u16 family, u32 if_id)
2134{
2135 struct xfrm_policy *tmp;
2136 int i;
2137
2138 for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2139 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2140 prefer,
2141 fl, type, family, if_id);
2142 if (!tmp)
2143 continue;
2144
2145 if (IS_ERR(tmp))
2146 return tmp;
2147 prefer = tmp;
2148 }
2149
2150 return prefer;
2151}
2152
2153static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2154 const struct flowi *fl,
2155 u16 family, u8 dir,
2156 u32 if_id)
2157{
2158 struct xfrm_pol_inexact_candidates cand;
2159 const xfrm_address_t *daddr, *saddr;
2160 struct xfrm_pol_inexact_bin *bin;
2161 struct xfrm_policy *pol, *ret;
2162 struct hlist_head *chain;
2163 unsigned int sequence;
2164 int err;
2165
2166 daddr = xfrm_flowi_daddr(fl, family);
2167 saddr = xfrm_flowi_saddr(fl, family);
2168 if (unlikely(!daddr || !saddr))
2169 return NULL;
2170
2171 rcu_read_lock();
2172 retry:
2173 do {
2174 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2175 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2176 } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2177
2178 ret = NULL;
2179 hlist_for_each_entry_rcu(pol, chain, bydst) {
2180 err = xfrm_policy_match(pol, fl, type, family, if_id);
2181 if (err) {
2182 if (err == -ESRCH)
2183 continue;
2184 else {
2185 ret = ERR_PTR(err);
2186 goto fail;
2187 }
2188 } else {
2189 ret = pol;
2190 break;
2191 }
2192 }
2193 if (ret && ret->xdo.type == XFRM_DEV_OFFLOAD_PACKET)
2194 goto skip_inexact;
2195
2196 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2197 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2198 daddr))
2199 goto skip_inexact;
2200
2201 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2202 family, if_id);
2203 if (pol) {
2204 ret = pol;
2205 if (IS_ERR(pol))
2206 goto fail;
2207 }
2208
2209skip_inexact:
2210 if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2211 goto retry;
2212
2213 if (ret && !xfrm_pol_hold_rcu(ret))
2214 goto retry;
2215fail:
2216 rcu_read_unlock();
2217
2218 return ret;
2219}
2220
2221static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2222 const struct flowi *fl,
2223 u16 family, u8 dir, u32 if_id)
2224{
2225#ifdef CONFIG_XFRM_SUB_POLICY
2226 struct xfrm_policy *pol;
2227
2228 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2229 dir, if_id);
2230 if (pol != NULL)
2231 return pol;
2232#endif
2233 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2234 dir, if_id);
2235}
2236
2237static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2238 const struct flowi *fl,
2239 u16 family, u32 if_id)
2240{
2241 struct xfrm_policy *pol;
2242
2243 rcu_read_lock();
2244 again:
2245 pol = rcu_dereference(sk->sk_policy[dir]);
2246 if (pol != NULL) {
2247 bool match;
2248 int err = 0;
2249
2250 if (pol->family != family) {
2251 pol = NULL;
2252 goto out;
2253 }
2254
2255 match = xfrm_selector_match(&pol->selector, fl, family);
2256 if (match) {
2257 if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v ||
2258 pol->if_id != if_id) {
2259 pol = NULL;
2260 goto out;
2261 }
2262 err = security_xfrm_policy_lookup(pol->security,
2263 fl->flowi_secid);
2264 if (!err) {
2265 if (!xfrm_pol_hold_rcu(pol))
2266 goto again;
2267 } else if (err == -ESRCH) {
2268 pol = NULL;
2269 } else {
2270 pol = ERR_PTR(err);
2271 }
2272 } else
2273 pol = NULL;
2274 }
2275out:
2276 rcu_read_unlock();
2277 return pol;
2278}
2279
2280static u32 xfrm_gen_pos_slow(struct net *net)
2281{
2282 struct xfrm_policy *policy;
2283 u32 i = 0;
2284
2285 /* oldest entry is last in list */
2286 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
2287 if (!xfrm_policy_is_dead_or_sk(policy))
2288 policy->pos = ++i;
2289 }
2290
2291 return i;
2292}
2293
2294static u32 xfrm_gen_pos(struct net *net)
2295{
2296 const struct xfrm_policy *policy;
2297 u32 i = 0;
2298
2299 /* most recently added policy is at the head of the list */
2300 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
2301 if (xfrm_policy_is_dead_or_sk(policy))
2302 continue;
2303
2304 if (policy->pos == UINT_MAX)
2305 return xfrm_gen_pos_slow(net);
2306
2307 i = policy->pos + 1;
2308 break;
2309 }
2310
2311 return i;
2312}
2313
2314static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2315{
2316 struct net *net = xp_net(pol);
2317
2318 switch (dir) {
2319 case XFRM_POLICY_IN:
2320 case XFRM_POLICY_FWD:
2321 case XFRM_POLICY_OUT:
2322 pol->pos = xfrm_gen_pos(net);
2323 break;
2324 }
2325
2326 list_add(&pol->walk.all, &net->xfrm.policy_all);
2327 net->xfrm.policy_count[dir]++;
2328 xfrm_pol_hold(pol);
2329}
2330
2331static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2332 int dir)
2333{
2334 struct net *net = xp_net(pol);
2335
2336 if (list_empty(&pol->walk.all))
2337 return NULL;
2338
2339 /* Socket policies are not hashed. */
2340 if (!hlist_unhashed(&pol->bydst)) {
2341 hlist_del_rcu(&pol->bydst);
2342 hlist_del(&pol->byidx);
2343 }
2344
2345 list_del_init(&pol->walk.all);
2346 net->xfrm.policy_count[dir]--;
2347
2348 return pol;
2349}
2350
2351static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2352{
2353 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2354}
2355
2356static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2357{
2358 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2359}
2360
2361int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2362{
2363 struct net *net = xp_net(pol);
2364
2365 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2366 pol = __xfrm_policy_unlink(pol, dir);
2367 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2368 if (pol) {
2369 xfrm_policy_kill(pol);
2370 return 0;
2371 }
2372 return -ENOENT;
2373}
2374EXPORT_SYMBOL(xfrm_policy_delete);
2375
2376int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2377{
2378 struct net *net = sock_net(sk);
2379 struct xfrm_policy *old_pol;
2380
2381#ifdef CONFIG_XFRM_SUB_POLICY
2382 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2383 return -EINVAL;
2384#endif
2385
2386 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2387 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2388 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2389 if (pol) {
2390 pol->curlft.add_time = ktime_get_real_seconds();
2391 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2392 xfrm_sk_policy_link(pol, dir);
2393 }
2394 rcu_assign_pointer(sk->sk_policy[dir], pol);
2395 if (old_pol) {
2396 if (pol)
2397 xfrm_policy_requeue(old_pol, pol);
2398
2399 /* Unlinking succeeds always. This is the only function
2400 * allowed to delete or replace socket policy.
2401 */
2402 xfrm_sk_policy_unlink(old_pol, dir);
2403 }
2404 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2405
2406 if (old_pol) {
2407 xfrm_policy_kill(old_pol);
2408 }
2409 return 0;
2410}
2411
2412static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2413{
2414 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2415 struct net *net = xp_net(old);
2416
2417 if (newp) {
2418 newp->selector = old->selector;
2419 if (security_xfrm_policy_clone(old->security,
2420 &newp->security)) {
2421 kfree(newp);
2422 return NULL; /* ENOMEM */
2423 }
2424 newp->lft = old->lft;
2425 newp->curlft = old->curlft;
2426 newp->mark = old->mark;
2427 newp->if_id = old->if_id;
2428 newp->action = old->action;
2429 newp->flags = old->flags;
2430 newp->xfrm_nr = old->xfrm_nr;
2431 newp->index = old->index;
2432 newp->type = old->type;
2433 newp->family = old->family;
2434 memcpy(newp->xfrm_vec, old->xfrm_vec,
2435 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2436 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2437 xfrm_sk_policy_link(newp, dir);
2438 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2439 xfrm_pol_put(newp);
2440 }
2441 return newp;
2442}
2443
2444int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2445{
2446 const struct xfrm_policy *p;
2447 struct xfrm_policy *np;
2448 int i, ret = 0;
2449
2450 rcu_read_lock();
2451 for (i = 0; i < 2; i++) {
2452 p = rcu_dereference(osk->sk_policy[i]);
2453 if (p) {
2454 np = clone_policy(p, i);
2455 if (unlikely(!np)) {
2456 ret = -ENOMEM;
2457 break;
2458 }
2459 rcu_assign_pointer(sk->sk_policy[i], np);
2460 }
2461 }
2462 rcu_read_unlock();
2463 return ret;
2464}
2465
2466static int
2467xfrm_get_saddr(unsigned short family, xfrm_address_t *saddr,
2468 const struct xfrm_dst_lookup_params *params)
2469{
2470 int err;
2471 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2472
2473 if (unlikely(afinfo == NULL))
2474 return -EINVAL;
2475 err = afinfo->get_saddr(saddr, params);
2476 rcu_read_unlock();
2477 return err;
2478}
2479
2480/* Resolve list of templates for the flow, given policy. */
2481
2482static int
2483xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2484 struct xfrm_state **xfrm, unsigned short family)
2485{
2486 struct net *net = xp_net(policy);
2487 int nx;
2488 int i, error;
2489 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2490 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2491 xfrm_address_t tmp;
2492
2493 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2494 struct xfrm_state *x;
2495 xfrm_address_t *remote = daddr;
2496 xfrm_address_t *local = saddr;
2497 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2498
2499 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2500 tmpl->mode == XFRM_MODE_BEET) {
2501 remote = &tmpl->id.daddr;
2502 local = &tmpl->saddr;
2503 if (xfrm_addr_any(local, tmpl->encap_family)) {
2504 struct xfrm_dst_lookup_params params;
2505
2506 memset(¶ms, 0, sizeof(params));
2507 params.net = net;
2508 params.oif = fl->flowi_oif;
2509 params.daddr = remote;
2510 error = xfrm_get_saddr(tmpl->encap_family, &tmp,
2511 ¶ms);
2512 if (error)
2513 goto fail;
2514 local = &tmp;
2515 }
2516 }
2517
2518 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2519 family, policy->if_id);
2520 if (x && x->dir && x->dir != XFRM_SA_DIR_OUT) {
2521 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEDIRERROR);
2522 xfrm_state_put(x);
2523 error = -EINVAL;
2524 goto fail;
2525 }
2526
2527 if (x && x->km.state == XFRM_STATE_VALID) {
2528 xfrm[nx++] = x;
2529 daddr = remote;
2530 saddr = local;
2531 continue;
2532 }
2533 if (x) {
2534 error = (x->km.state == XFRM_STATE_ERROR ?
2535 -EINVAL : -EAGAIN);
2536 xfrm_state_put(x);
2537 } else if (error == -ESRCH) {
2538 error = -EAGAIN;
2539 }
2540
2541 if (!tmpl->optional)
2542 goto fail;
2543 }
2544 return nx;
2545
2546fail:
2547 for (nx--; nx >= 0; nx--)
2548 xfrm_state_put(xfrm[nx]);
2549 return error;
2550}
2551
2552static int
2553xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2554 struct xfrm_state **xfrm, unsigned short family)
2555{
2556 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2557 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2558 int cnx = 0;
2559 int error;
2560 int ret;
2561 int i;
2562
2563 for (i = 0; i < npols; i++) {
2564 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2565 error = -ENOBUFS;
2566 goto fail;
2567 }
2568
2569 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2570 if (ret < 0) {
2571 error = ret;
2572 goto fail;
2573 } else
2574 cnx += ret;
2575 }
2576
2577 /* found states are sorted for outbound processing */
2578 if (npols > 1)
2579 xfrm_state_sort(xfrm, tpp, cnx, family);
2580
2581 return cnx;
2582
2583 fail:
2584 for (cnx--; cnx >= 0; cnx--)
2585 xfrm_state_put(tpp[cnx]);
2586 return error;
2587
2588}
2589
2590static dscp_t xfrm_get_dscp(const struct flowi *fl, int family)
2591{
2592 if (family == AF_INET)
2593 return inet_dsfield_to_dscp(fl->u.ip4.flowi4_tos);
2594
2595 return 0;
2596}
2597
2598static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2599{
2600 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2601 struct dst_ops *dst_ops;
2602 struct xfrm_dst *xdst;
2603
2604 if (!afinfo)
2605 return ERR_PTR(-EINVAL);
2606
2607 switch (family) {
2608 case AF_INET:
2609 dst_ops = &net->xfrm.xfrm4_dst_ops;
2610 break;
2611#if IS_ENABLED(CONFIG_IPV6)
2612 case AF_INET6:
2613 dst_ops = &net->xfrm.xfrm6_dst_ops;
2614 break;
2615#endif
2616 default:
2617 BUG();
2618 }
2619 xdst = dst_alloc(dst_ops, NULL, DST_OBSOLETE_NONE, 0);
2620
2621 if (likely(xdst)) {
2622 memset_after(xdst, 0, u.dst);
2623 } else
2624 xdst = ERR_PTR(-ENOBUFS);
2625
2626 rcu_read_unlock();
2627
2628 return xdst;
2629}
2630
2631static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2632 int nfheader_len)
2633{
2634 if (dst->ops->family == AF_INET6) {
2635 path->path_cookie = rt6_get_cookie(dst_rt6_info(dst));
2636 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2637 }
2638}
2639
2640static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2641 const struct flowi *fl)
2642{
2643 const struct xfrm_policy_afinfo *afinfo =
2644 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2645 int err;
2646
2647 if (!afinfo)
2648 return -EINVAL;
2649
2650 err = afinfo->fill_dst(xdst, dev, fl);
2651
2652 rcu_read_unlock();
2653
2654 return err;
2655}
2656
2657
2658/* Allocate chain of dst_entry's, attach known xfrm's, calculate
2659 * all the metrics... Shortly, bundle a bundle.
2660 */
2661
2662static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2663 struct xfrm_state **xfrm,
2664 struct xfrm_dst **bundle,
2665 int nx,
2666 const struct flowi *fl,
2667 struct dst_entry *dst)
2668{
2669 const struct xfrm_state_afinfo *afinfo;
2670 const struct xfrm_mode *inner_mode;
2671 struct net *net = xp_net(policy);
2672 unsigned long now = jiffies;
2673 struct net_device *dev;
2674 struct xfrm_dst *xdst_prev = NULL;
2675 struct xfrm_dst *xdst0 = NULL;
2676 int i = 0;
2677 int err;
2678 int header_len = 0;
2679 int nfheader_len = 0;
2680 int trailer_len = 0;
2681 int family = policy->selector.family;
2682 xfrm_address_t saddr, daddr;
2683 dscp_t dscp;
2684
2685 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2686
2687 dscp = xfrm_get_dscp(fl, family);
2688
2689 dst_hold(dst);
2690
2691 for (; i < nx; i++) {
2692 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2693 struct dst_entry *dst1 = &xdst->u.dst;
2694
2695 err = PTR_ERR(xdst);
2696 if (IS_ERR(xdst)) {
2697 dst_release(dst);
2698 goto put_states;
2699 }
2700
2701 bundle[i] = xdst;
2702 if (!xdst_prev)
2703 xdst0 = xdst;
2704 else
2705 /* Ref count is taken during xfrm_alloc_dst()
2706 * No need to do dst_clone() on dst1
2707 */
2708 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2709
2710 if (xfrm[i]->sel.family == AF_UNSPEC) {
2711 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2712 xfrm_af2proto(family));
2713 if (!inner_mode) {
2714 err = -EAFNOSUPPORT;
2715 dst_release(dst);
2716 goto put_states;
2717 }
2718 } else
2719 inner_mode = &xfrm[i]->inner_mode;
2720
2721 xdst->route = dst;
2722 dst_copy_metrics(dst1, dst);
2723
2724 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2725 __u32 mark = 0;
2726 int oif;
2727
2728 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2729 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2730
2731 if (xfrm[i]->xso.type != XFRM_DEV_OFFLOAD_PACKET)
2732 family = xfrm[i]->props.family;
2733
2734 oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2735 dst = xfrm_dst_lookup(xfrm[i], dscp, oif, &saddr,
2736 &daddr, family, mark);
2737 err = PTR_ERR(dst);
2738 if (IS_ERR(dst))
2739 goto put_states;
2740 } else
2741 dst_hold(dst);
2742
2743 dst1->xfrm = xfrm[i];
2744 xdst->xfrm_genid = xfrm[i]->genid;
2745
2746 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2747 dst1->lastuse = now;
2748
2749 dst1->input = dst_discard;
2750
2751 rcu_read_lock();
2752 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2753 if (likely(afinfo))
2754 dst1->output = afinfo->output;
2755 else
2756 dst1->output = dst_discard_out;
2757 rcu_read_unlock();
2758
2759 xdst_prev = xdst;
2760
2761 header_len += xfrm[i]->props.header_len;
2762 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2763 nfheader_len += xfrm[i]->props.header_len;
2764 trailer_len += xfrm[i]->props.trailer_len;
2765 }
2766
2767 xfrm_dst_set_child(xdst_prev, dst);
2768 xdst0->path = dst;
2769
2770 err = -ENODEV;
2771 dev = dst->dev;
2772 if (!dev)
2773 goto free_dst;
2774
2775 xfrm_init_path(xdst0, dst, nfheader_len);
2776 xfrm_init_pmtu(bundle, nx);
2777
2778 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2779 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2780 err = xfrm_fill_dst(xdst_prev, dev, fl);
2781 if (err)
2782 goto free_dst;
2783
2784 xdst_prev->u.dst.header_len = header_len;
2785 xdst_prev->u.dst.trailer_len = trailer_len;
2786 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2787 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2788 }
2789
2790 return &xdst0->u.dst;
2791
2792put_states:
2793 for (; i < nx; i++)
2794 xfrm_state_put(xfrm[i]);
2795free_dst:
2796 if (xdst0)
2797 dst_release_immediate(&xdst0->u.dst);
2798
2799 return ERR_PTR(err);
2800}
2801
2802static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2803 struct xfrm_policy **pols,
2804 int *num_pols, int *num_xfrms)
2805{
2806 int i;
2807
2808 if (*num_pols == 0 || !pols[0]) {
2809 *num_pols = 0;
2810 *num_xfrms = 0;
2811 return 0;
2812 }
2813 if (IS_ERR(pols[0])) {
2814 *num_pols = 0;
2815 return PTR_ERR(pols[0]);
2816 }
2817
2818 *num_xfrms = pols[0]->xfrm_nr;
2819
2820#ifdef CONFIG_XFRM_SUB_POLICY
2821 if (pols[0]->action == XFRM_POLICY_ALLOW &&
2822 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2823 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2824 XFRM_POLICY_TYPE_MAIN,
2825 fl, family,
2826 XFRM_POLICY_OUT,
2827 pols[0]->if_id);
2828 if (pols[1]) {
2829 if (IS_ERR(pols[1])) {
2830 xfrm_pols_put(pols, *num_pols);
2831 *num_pols = 0;
2832 return PTR_ERR(pols[1]);
2833 }
2834 (*num_pols)++;
2835 (*num_xfrms) += pols[1]->xfrm_nr;
2836 }
2837 }
2838#endif
2839 for (i = 0; i < *num_pols; i++) {
2840 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2841 *num_xfrms = -1;
2842 break;
2843 }
2844 }
2845
2846 return 0;
2847
2848}
2849
2850static struct xfrm_dst *
2851xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2852 const struct flowi *fl, u16 family,
2853 struct dst_entry *dst_orig)
2854{
2855 struct net *net = xp_net(pols[0]);
2856 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2857 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2858 struct xfrm_dst *xdst;
2859 struct dst_entry *dst;
2860 int err;
2861
2862 /* Try to instantiate a bundle */
2863 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2864 if (err <= 0) {
2865 if (err == 0)
2866 return NULL;
2867
2868 if (err != -EAGAIN)
2869 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2870 return ERR_PTR(err);
2871 }
2872
2873 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2874 if (IS_ERR(dst)) {
2875 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2876 return ERR_CAST(dst);
2877 }
2878
2879 xdst = (struct xfrm_dst *)dst;
2880 xdst->num_xfrms = err;
2881 xdst->num_pols = num_pols;
2882 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2883 xdst->policy_genid = atomic_read(&pols[0]->genid);
2884
2885 return xdst;
2886}
2887
2888static void xfrm_policy_queue_process(struct timer_list *t)
2889{
2890 struct sk_buff *skb;
2891 struct sock *sk;
2892 struct dst_entry *dst;
2893 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2894 struct net *net = xp_net(pol);
2895 struct xfrm_policy_queue *pq = &pol->polq;
2896 struct flowi fl;
2897 struct sk_buff_head list;
2898 __u32 skb_mark;
2899
2900 spin_lock(&pq->hold_queue.lock);
2901 skb = skb_peek(&pq->hold_queue);
2902 if (!skb) {
2903 spin_unlock(&pq->hold_queue.lock);
2904 goto out;
2905 }
2906 dst = skb_dst(skb);
2907 sk = skb->sk;
2908
2909 /* Fixup the mark to support VTI. */
2910 skb_mark = skb->mark;
2911 skb->mark = pol->mark.v;
2912 xfrm_decode_session(net, skb, &fl, dst->ops->family);
2913 skb->mark = skb_mark;
2914 spin_unlock(&pq->hold_queue.lock);
2915
2916 dst_hold(xfrm_dst_path(dst));
2917 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2918 if (IS_ERR(dst))
2919 goto purge_queue;
2920
2921 if (dst->flags & DST_XFRM_QUEUE) {
2922 dst_release(dst);
2923
2924 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2925 goto purge_queue;
2926
2927 pq->timeout = pq->timeout << 1;
2928 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2929 xfrm_pol_hold(pol);
2930 goto out;
2931 }
2932
2933 dst_release(dst);
2934
2935 __skb_queue_head_init(&list);
2936
2937 spin_lock(&pq->hold_queue.lock);
2938 pq->timeout = 0;
2939 skb_queue_splice_init(&pq->hold_queue, &list);
2940 spin_unlock(&pq->hold_queue.lock);
2941
2942 while (!skb_queue_empty(&list)) {
2943 skb = __skb_dequeue(&list);
2944
2945 /* Fixup the mark to support VTI. */
2946 skb_mark = skb->mark;
2947 skb->mark = pol->mark.v;
2948 xfrm_decode_session(net, skb, &fl, skb_dst(skb)->ops->family);
2949 skb->mark = skb_mark;
2950
2951 dst_hold(xfrm_dst_path(skb_dst(skb)));
2952 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2953 if (IS_ERR(dst)) {
2954 kfree_skb(skb);
2955 continue;
2956 }
2957
2958 nf_reset_ct(skb);
2959 skb_dst_drop(skb);
2960 skb_dst_set(skb, dst);
2961
2962 dst_output(net, skb->sk, skb);
2963 }
2964
2965out:
2966 xfrm_pol_put(pol);
2967 return;
2968
2969purge_queue:
2970 pq->timeout = 0;
2971 skb_queue_purge(&pq->hold_queue);
2972 xfrm_pol_put(pol);
2973}
2974
2975static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2976{
2977 unsigned long sched_next;
2978 struct dst_entry *dst = skb_dst(skb);
2979 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2980 struct xfrm_policy *pol = xdst->pols[0];
2981 struct xfrm_policy_queue *pq = &pol->polq;
2982
2983 if (unlikely(skb_fclone_busy(sk, skb))) {
2984 kfree_skb(skb);
2985 return 0;
2986 }
2987
2988 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2989 kfree_skb(skb);
2990 return -EAGAIN;
2991 }
2992
2993 skb_dst_force(skb);
2994
2995 spin_lock_bh(&pq->hold_queue.lock);
2996
2997 if (!pq->timeout)
2998 pq->timeout = XFRM_QUEUE_TMO_MIN;
2999
3000 sched_next = jiffies + pq->timeout;
3001
3002 if (del_timer(&pq->hold_timer)) {
3003 if (time_before(pq->hold_timer.expires, sched_next))
3004 sched_next = pq->hold_timer.expires;
3005 xfrm_pol_put(pol);
3006 }
3007
3008 __skb_queue_tail(&pq->hold_queue, skb);
3009 if (!mod_timer(&pq->hold_timer, sched_next))
3010 xfrm_pol_hold(pol);
3011
3012 spin_unlock_bh(&pq->hold_queue.lock);
3013
3014 return 0;
3015}
3016
3017static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
3018 struct xfrm_flo *xflo,
3019 const struct flowi *fl,
3020 int num_xfrms,
3021 u16 family)
3022{
3023 int err;
3024 struct net_device *dev;
3025 struct dst_entry *dst;
3026 struct dst_entry *dst1;
3027 struct xfrm_dst *xdst;
3028
3029 xdst = xfrm_alloc_dst(net, family);
3030 if (IS_ERR(xdst))
3031 return xdst;
3032
3033 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
3034 net->xfrm.sysctl_larval_drop ||
3035 num_xfrms <= 0)
3036 return xdst;
3037
3038 dst = xflo->dst_orig;
3039 dst1 = &xdst->u.dst;
3040 dst_hold(dst);
3041 xdst->route = dst;
3042
3043 dst_copy_metrics(dst1, dst);
3044
3045 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
3046 dst1->flags |= DST_XFRM_QUEUE;
3047 dst1->lastuse = jiffies;
3048
3049 dst1->input = dst_discard;
3050 dst1->output = xdst_queue_output;
3051
3052 dst_hold(dst);
3053 xfrm_dst_set_child(xdst, dst);
3054 xdst->path = dst;
3055
3056 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
3057
3058 err = -ENODEV;
3059 dev = dst->dev;
3060 if (!dev)
3061 goto free_dst;
3062
3063 err = xfrm_fill_dst(xdst, dev, fl);
3064 if (err)
3065 goto free_dst;
3066
3067out:
3068 return xdst;
3069
3070free_dst:
3071 dst_release(dst1);
3072 xdst = ERR_PTR(err);
3073 goto out;
3074}
3075
3076static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
3077 const struct flowi *fl,
3078 u16 family, u8 dir,
3079 struct xfrm_flo *xflo, u32 if_id)
3080{
3081 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3082 int num_pols = 0, num_xfrms = 0, err;
3083 struct xfrm_dst *xdst;
3084
3085 /* Resolve policies to use if we couldn't get them from
3086 * previous cache entry */
3087 num_pols = 1;
3088 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
3089 err = xfrm_expand_policies(fl, family, pols,
3090 &num_pols, &num_xfrms);
3091 if (err < 0)
3092 goto inc_error;
3093 if (num_pols == 0)
3094 return NULL;
3095 if (num_xfrms <= 0)
3096 goto make_dummy_bundle;
3097
3098 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
3099 xflo->dst_orig);
3100 if (IS_ERR(xdst)) {
3101 err = PTR_ERR(xdst);
3102 if (err == -EREMOTE) {
3103 xfrm_pols_put(pols, num_pols);
3104 return NULL;
3105 }
3106
3107 if (err != -EAGAIN)
3108 goto error;
3109 goto make_dummy_bundle;
3110 } else if (xdst == NULL) {
3111 num_xfrms = 0;
3112 goto make_dummy_bundle;
3113 }
3114
3115 return xdst;
3116
3117make_dummy_bundle:
3118 /* We found policies, but there's no bundles to instantiate:
3119 * either because the policy blocks, has no transformations or
3120 * we could not build template (no xfrm_states).*/
3121 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
3122 if (IS_ERR(xdst)) {
3123 xfrm_pols_put(pols, num_pols);
3124 return ERR_CAST(xdst);
3125 }
3126 xdst->num_pols = num_pols;
3127 xdst->num_xfrms = num_xfrms;
3128 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
3129
3130 return xdst;
3131
3132inc_error:
3133 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3134error:
3135 xfrm_pols_put(pols, num_pols);
3136 return ERR_PTR(err);
3137}
3138
3139static struct dst_entry *make_blackhole(struct net *net, u16 family,
3140 struct dst_entry *dst_orig)
3141{
3142 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3143 struct dst_entry *ret;
3144
3145 if (!afinfo) {
3146 dst_release(dst_orig);
3147 return ERR_PTR(-EINVAL);
3148 } else {
3149 ret = afinfo->blackhole_route(net, dst_orig);
3150 }
3151 rcu_read_unlock();
3152
3153 return ret;
3154}
3155
3156/* Finds/creates a bundle for given flow and if_id
3157 *
3158 * At the moment we eat a raw IP route. Mostly to speed up lookups
3159 * on interfaces with disabled IPsec.
3160 *
3161 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3162 * compatibility
3163 */
3164struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3165 struct dst_entry *dst_orig,
3166 const struct flowi *fl,
3167 const struct sock *sk,
3168 int flags, u32 if_id)
3169{
3170 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3171 struct xfrm_dst *xdst;
3172 struct dst_entry *dst, *route;
3173 u16 family = dst_orig->ops->family;
3174 u8 dir = XFRM_POLICY_OUT;
3175 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3176
3177 dst = NULL;
3178 xdst = NULL;
3179 route = NULL;
3180
3181 sk = sk_const_to_full_sk(sk);
3182 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3183 num_pols = 1;
3184 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3185 if_id);
3186 err = xfrm_expand_policies(fl, family, pols,
3187 &num_pols, &num_xfrms);
3188 if (err < 0)
3189 goto dropdst;
3190
3191 if (num_pols) {
3192 if (num_xfrms <= 0) {
3193 drop_pols = num_pols;
3194 goto no_transform;
3195 }
3196
3197 xdst = xfrm_resolve_and_create_bundle(
3198 pols, num_pols, fl,
3199 family, dst_orig);
3200
3201 if (IS_ERR(xdst)) {
3202 xfrm_pols_put(pols, num_pols);
3203 err = PTR_ERR(xdst);
3204 if (err == -EREMOTE)
3205 goto nopol;
3206
3207 goto dropdst;
3208 } else if (xdst == NULL) {
3209 num_xfrms = 0;
3210 drop_pols = num_pols;
3211 goto no_transform;
3212 }
3213
3214 route = xdst->route;
3215 }
3216 }
3217
3218 if (xdst == NULL) {
3219 struct xfrm_flo xflo;
3220
3221 xflo.dst_orig = dst_orig;
3222 xflo.flags = flags;
3223
3224 /* To accelerate a bit... */
3225 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3226 !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3227 goto nopol;
3228
3229 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3230 if (xdst == NULL)
3231 goto nopol;
3232 if (IS_ERR(xdst)) {
3233 err = PTR_ERR(xdst);
3234 goto dropdst;
3235 }
3236
3237 num_pols = xdst->num_pols;
3238 num_xfrms = xdst->num_xfrms;
3239 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3240 route = xdst->route;
3241 }
3242
3243 dst = &xdst->u.dst;
3244 if (route == NULL && num_xfrms > 0) {
3245 /* The only case when xfrm_bundle_lookup() returns a
3246 * bundle with null route, is when the template could
3247 * not be resolved. It means policies are there, but
3248 * bundle could not be created, since we don't yet
3249 * have the xfrm_state's. We need to wait for KM to
3250 * negotiate new SA's or bail out with error.*/
3251 if (net->xfrm.sysctl_larval_drop) {
3252 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3253 err = -EREMOTE;
3254 goto error;
3255 }
3256
3257 err = -EAGAIN;
3258
3259 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3260 goto error;
3261 }
3262
3263no_transform:
3264 if (num_pols == 0)
3265 goto nopol;
3266
3267 if ((flags & XFRM_LOOKUP_ICMP) &&
3268 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3269 err = -ENOENT;
3270 goto error;
3271 }
3272
3273 for (i = 0; i < num_pols; i++)
3274 WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds());
3275
3276 if (num_xfrms < 0) {
3277 /* Prohibit the flow */
3278 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3279 err = -EPERM;
3280 goto error;
3281 } else if (num_xfrms > 0) {
3282 /* Flow transformed */
3283 dst_release(dst_orig);
3284 } else {
3285 /* Flow passes untransformed */
3286 dst_release(dst);
3287 dst = dst_orig;
3288 }
3289
3290ok:
3291 xfrm_pols_put(pols, drop_pols);
3292 if (dst && dst->xfrm &&
3293 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3294 dst->flags |= DST_XFRM_TUNNEL;
3295 return dst;
3296
3297nopol:
3298 if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3299 net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3300 err = -EPERM;
3301 goto error;
3302 }
3303 if (!(flags & XFRM_LOOKUP_ICMP)) {
3304 dst = dst_orig;
3305 goto ok;
3306 }
3307 err = -ENOENT;
3308error:
3309 dst_release(dst);
3310dropdst:
3311 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3312 dst_release(dst_orig);
3313 xfrm_pols_put(pols, drop_pols);
3314 return ERR_PTR(err);
3315}
3316EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3317
3318/* Main function: finds/creates a bundle for given flow.
3319 *
3320 * At the moment we eat a raw IP route. Mostly to speed up lookups
3321 * on interfaces with disabled IPsec.
3322 */
3323struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3324 const struct flowi *fl, const struct sock *sk,
3325 int flags)
3326{
3327 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3328}
3329EXPORT_SYMBOL(xfrm_lookup);
3330
3331/* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3332 * Otherwise we may send out blackholed packets.
3333 */
3334struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3335 const struct flowi *fl,
3336 const struct sock *sk, int flags)
3337{
3338 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3339 flags | XFRM_LOOKUP_QUEUE |
3340 XFRM_LOOKUP_KEEP_DST_REF);
3341
3342 if (PTR_ERR(dst) == -EREMOTE)
3343 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3344
3345 if (IS_ERR(dst))
3346 dst_release(dst_orig);
3347
3348 return dst;
3349}
3350EXPORT_SYMBOL(xfrm_lookup_route);
3351
3352static inline int
3353xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3354{
3355 struct sec_path *sp = skb_sec_path(skb);
3356 struct xfrm_state *x;
3357
3358 if (!sp || idx < 0 || idx >= sp->len)
3359 return 0;
3360 x = sp->xvec[idx];
3361 if (!x->type->reject)
3362 return 0;
3363 return x->type->reject(x, skb, fl);
3364}
3365
3366/* When skb is transformed back to its "native" form, we have to
3367 * check policy restrictions. At the moment we make this in maximally
3368 * stupid way. Shame on me. :-) Of course, connected sockets must
3369 * have policy cached at them.
3370 */
3371
3372static inline int
3373xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3374 unsigned short family, u32 if_id)
3375{
3376 if (xfrm_state_kern(x))
3377 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3378 return x->id.proto == tmpl->id.proto &&
3379 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3380 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3381 x->props.mode == tmpl->mode &&
3382 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3383 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3384 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3385 xfrm_state_addr_cmp(tmpl, x, family)) &&
3386 (if_id == 0 || if_id == x->if_id);
3387}
3388
3389/*
3390 * 0 or more than 0 is returned when validation is succeeded (either bypass
3391 * because of optional transport mode, or next index of the matched secpath
3392 * state with the template.
3393 * -1 is returned when no matching template is found.
3394 * Otherwise "-2 - errored_index" is returned.
3395 */
3396static inline int
3397xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3398 unsigned short family, u32 if_id)
3399{
3400 int idx = start;
3401
3402 if (tmpl->optional) {
3403 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3404 return start;
3405 } else
3406 start = -1;
3407 for (; idx < sp->len; idx++) {
3408 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3409 return ++idx;
3410 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3411 if (idx < sp->verified_cnt) {
3412 /* Secpath entry previously verified, consider optional and
3413 * continue searching
3414 */
3415 continue;
3416 }
3417
3418 if (start == -1)
3419 start = -2-idx;
3420 break;
3421 }
3422 }
3423 return start;
3424}
3425
3426static void
3427decode_session4(const struct xfrm_flow_keys *flkeys, struct flowi *fl, bool reverse)
3428{
3429 struct flowi4 *fl4 = &fl->u.ip4;
3430
3431 memset(fl4, 0, sizeof(struct flowi4));
3432
3433 if (reverse) {
3434 fl4->saddr = flkeys->addrs.ipv4.dst;
3435 fl4->daddr = flkeys->addrs.ipv4.src;
3436 fl4->fl4_sport = flkeys->ports.dst;
3437 fl4->fl4_dport = flkeys->ports.src;
3438 } else {
3439 fl4->saddr = flkeys->addrs.ipv4.src;
3440 fl4->daddr = flkeys->addrs.ipv4.dst;
3441 fl4->fl4_sport = flkeys->ports.src;
3442 fl4->fl4_dport = flkeys->ports.dst;
3443 }
3444
3445 switch (flkeys->basic.ip_proto) {
3446 case IPPROTO_GRE:
3447 fl4->fl4_gre_key = flkeys->gre.keyid;
3448 break;
3449 case IPPROTO_ICMP:
3450 fl4->fl4_icmp_type = flkeys->icmp.type;
3451 fl4->fl4_icmp_code = flkeys->icmp.code;
3452 break;
3453 }
3454
3455 fl4->flowi4_proto = flkeys->basic.ip_proto;
3456 fl4->flowi4_tos = flkeys->ip.tos & ~INET_ECN_MASK;
3457}
3458
3459#if IS_ENABLED(CONFIG_IPV6)
3460static void
3461decode_session6(const struct xfrm_flow_keys *flkeys, struct flowi *fl, bool reverse)
3462{
3463 struct flowi6 *fl6 = &fl->u.ip6;
3464
3465 memset(fl6, 0, sizeof(struct flowi6));
3466
3467 if (reverse) {
3468 fl6->saddr = flkeys->addrs.ipv6.dst;
3469 fl6->daddr = flkeys->addrs.ipv6.src;
3470 fl6->fl6_sport = flkeys->ports.dst;
3471 fl6->fl6_dport = flkeys->ports.src;
3472 } else {
3473 fl6->saddr = flkeys->addrs.ipv6.src;
3474 fl6->daddr = flkeys->addrs.ipv6.dst;
3475 fl6->fl6_sport = flkeys->ports.src;
3476 fl6->fl6_dport = flkeys->ports.dst;
3477 }
3478
3479 switch (flkeys->basic.ip_proto) {
3480 case IPPROTO_GRE:
3481 fl6->fl6_gre_key = flkeys->gre.keyid;
3482 break;
3483 case IPPROTO_ICMPV6:
3484 fl6->fl6_icmp_type = flkeys->icmp.type;
3485 fl6->fl6_icmp_code = flkeys->icmp.code;
3486 break;
3487 }
3488
3489 fl6->flowi6_proto = flkeys->basic.ip_proto;
3490}
3491#endif
3492
3493int __xfrm_decode_session(struct net *net, struct sk_buff *skb, struct flowi *fl,
3494 unsigned int family, int reverse)
3495{
3496 struct xfrm_flow_keys flkeys;
3497
3498 memset(&flkeys, 0, sizeof(flkeys));
3499 __skb_flow_dissect(net, skb, &xfrm_session_dissector, &flkeys,
3500 NULL, 0, 0, 0, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
3501
3502 switch (family) {
3503 case AF_INET:
3504 decode_session4(&flkeys, fl, reverse);
3505 break;
3506#if IS_ENABLED(CONFIG_IPV6)
3507 case AF_INET6:
3508 decode_session6(&flkeys, fl, reverse);
3509 break;
3510#endif
3511 default:
3512 return -EAFNOSUPPORT;
3513 }
3514
3515 fl->flowi_mark = skb->mark;
3516 if (reverse) {
3517 fl->flowi_oif = skb->skb_iif;
3518 } else {
3519 int oif = 0;
3520
3521 if (skb_dst(skb) && skb_dst(skb)->dev)
3522 oif = skb_dst(skb)->dev->ifindex;
3523
3524 fl->flowi_oif = oif;
3525 }
3526
3527 return security_xfrm_decode_session(skb, &fl->flowi_secid);
3528}
3529EXPORT_SYMBOL(__xfrm_decode_session);
3530
3531static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3532{
3533 for (; k < sp->len; k++) {
3534 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3535 *idxp = k;
3536 return 1;
3537 }
3538 }
3539
3540 return 0;
3541}
3542
3543static bool icmp_err_packet(const struct flowi *fl, unsigned short family)
3544{
3545 const struct flowi4 *fl4 = &fl->u.ip4;
3546
3547 if (family == AF_INET &&
3548 fl4->flowi4_proto == IPPROTO_ICMP &&
3549 (fl4->fl4_icmp_type == ICMP_DEST_UNREACH ||
3550 fl4->fl4_icmp_type == ICMP_TIME_EXCEEDED))
3551 return true;
3552
3553#if IS_ENABLED(CONFIG_IPV6)
3554 if (family == AF_INET6) {
3555 const struct flowi6 *fl6 = &fl->u.ip6;
3556
3557 if (fl6->flowi6_proto == IPPROTO_ICMPV6 &&
3558 (fl6->fl6_icmp_type == ICMPV6_DEST_UNREACH ||
3559 fl6->fl6_icmp_type == ICMPV6_PKT_TOOBIG ||
3560 fl6->fl6_icmp_type == ICMPV6_TIME_EXCEED))
3561 return true;
3562 }
3563#endif
3564 return false;
3565}
3566
3567static bool xfrm_icmp_flow_decode(struct sk_buff *skb, unsigned short family,
3568 const struct flowi *fl, struct flowi *fl1)
3569{
3570 bool ret = true;
3571 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
3572 int hl = family == AF_INET ? (sizeof(struct iphdr) + sizeof(struct icmphdr)) :
3573 (sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr));
3574
3575 if (!newskb)
3576 return true;
3577
3578 if (!pskb_pull(newskb, hl))
3579 goto out;
3580
3581 skb_reset_network_header(newskb);
3582
3583 if (xfrm_decode_session_reverse(dev_net(skb->dev), newskb, fl1, family) < 0)
3584 goto out;
3585
3586 fl1->flowi_oif = fl->flowi_oif;
3587 fl1->flowi_mark = fl->flowi_mark;
3588 fl1->flowi_tos = fl->flowi_tos;
3589 nf_nat_decode_session(newskb, fl1, family);
3590 ret = false;
3591
3592out:
3593 consume_skb(newskb);
3594 return ret;
3595}
3596
3597static bool xfrm_selector_inner_icmp_match(struct sk_buff *skb, unsigned short family,
3598 const struct xfrm_selector *sel,
3599 const struct flowi *fl)
3600{
3601 bool ret = false;
3602
3603 if (icmp_err_packet(fl, family)) {
3604 struct flowi fl1;
3605
3606 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3607 return ret;
3608
3609 ret = xfrm_selector_match(sel, &fl1, family);
3610 }
3611
3612 return ret;
3613}
3614
3615static inline struct
3616xfrm_policy *xfrm_in_fwd_icmp(struct sk_buff *skb,
3617 const struct flowi *fl, unsigned short family,
3618 u32 if_id)
3619{
3620 struct xfrm_policy *pol = NULL;
3621
3622 if (icmp_err_packet(fl, family)) {
3623 struct flowi fl1;
3624 struct net *net = dev_net(skb->dev);
3625
3626 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3627 return pol;
3628
3629 pol = xfrm_policy_lookup(net, &fl1, family, XFRM_POLICY_FWD, if_id);
3630 if (IS_ERR(pol))
3631 pol = NULL;
3632 }
3633
3634 return pol;
3635}
3636
3637static inline struct
3638dst_entry *xfrm_out_fwd_icmp(struct sk_buff *skb, struct flowi *fl,
3639 unsigned short family, struct dst_entry *dst)
3640{
3641 if (icmp_err_packet(fl, family)) {
3642 struct net *net = dev_net(skb->dev);
3643 struct dst_entry *dst2;
3644 struct flowi fl1;
3645
3646 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3647 return dst;
3648
3649 dst_hold(dst);
3650
3651 dst2 = xfrm_lookup(net, dst, &fl1, NULL, (XFRM_LOOKUP_QUEUE | XFRM_LOOKUP_ICMP));
3652
3653 if (IS_ERR(dst2))
3654 return dst;
3655
3656 if (dst2->xfrm) {
3657 dst_release(dst);
3658 dst = dst2;
3659 } else {
3660 dst_release(dst2);
3661 }
3662 }
3663
3664 return dst;
3665}
3666
3667int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3668 unsigned short family)
3669{
3670 struct net *net = dev_net(skb->dev);
3671 struct xfrm_policy *pol;
3672 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3673 int npols = 0;
3674 int xfrm_nr;
3675 int pi;
3676 int reverse;
3677 struct flowi fl;
3678 int xerr_idx = -1;
3679 const struct xfrm_if_cb *ifcb;
3680 struct sec_path *sp;
3681 u32 if_id = 0;
3682
3683 rcu_read_lock();
3684 ifcb = xfrm_if_get_cb();
3685
3686 if (ifcb) {
3687 struct xfrm_if_decode_session_result r;
3688
3689 if (ifcb->decode_session(skb, family, &r)) {
3690 if_id = r.if_id;
3691 net = r.net;
3692 }
3693 }
3694 rcu_read_unlock();
3695
3696 reverse = dir & ~XFRM_POLICY_MASK;
3697 dir &= XFRM_POLICY_MASK;
3698
3699 if (__xfrm_decode_session(net, skb, &fl, family, reverse) < 0) {
3700 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3701 return 0;
3702 }
3703
3704 nf_nat_decode_session(skb, &fl, family);
3705
3706 /* First, check used SA against their selectors. */
3707 sp = skb_sec_path(skb);
3708 if (sp) {
3709 int i;
3710
3711 for (i = sp->len - 1; i >= 0; i--) {
3712 struct xfrm_state *x = sp->xvec[i];
3713 int ret = 0;
3714
3715 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3716 ret = 1;
3717 if (x->props.flags & XFRM_STATE_ICMP &&
3718 xfrm_selector_inner_icmp_match(skb, family, &x->sel, &fl))
3719 ret = 0;
3720 if (ret) {
3721 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3722 return 0;
3723 }
3724 }
3725 }
3726 }
3727
3728 pol = NULL;
3729 sk = sk_to_full_sk(sk);
3730 if (sk && sk->sk_policy[dir]) {
3731 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3732 if (IS_ERR(pol)) {
3733 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3734 return 0;
3735 }
3736 }
3737
3738 if (!pol)
3739 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3740
3741 if (IS_ERR(pol)) {
3742 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3743 return 0;
3744 }
3745
3746 if (!pol && dir == XFRM_POLICY_FWD)
3747 pol = xfrm_in_fwd_icmp(skb, &fl, family, if_id);
3748
3749 if (!pol) {
3750 const bool is_crypto_offload = sp &&
3751 (xfrm_input_state(skb)->xso.type == XFRM_DEV_OFFLOAD_CRYPTO);
3752
3753 if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3754 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3755 return 0;
3756 }
3757
3758 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx) && !is_crypto_offload) {
3759 xfrm_secpath_reject(xerr_idx, skb, &fl);
3760 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3761 return 0;
3762 }
3763 return 1;
3764 }
3765
3766 /* This lockless write can happen from different cpus. */
3767 WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3768
3769 pols[0] = pol;
3770 npols++;
3771#ifdef CONFIG_XFRM_SUB_POLICY
3772 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3773 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3774 &fl, family,
3775 XFRM_POLICY_IN, if_id);
3776 if (pols[1]) {
3777 if (IS_ERR(pols[1])) {
3778 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3779 xfrm_pol_put(pols[0]);
3780 return 0;
3781 }
3782 /* This write can happen from different cpus. */
3783 WRITE_ONCE(pols[1]->curlft.use_time,
3784 ktime_get_real_seconds());
3785 npols++;
3786 }
3787 }
3788#endif
3789
3790 if (pol->action == XFRM_POLICY_ALLOW) {
3791 static struct sec_path dummy;
3792 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3793 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3794 struct xfrm_tmpl **tpp = tp;
3795 int ti = 0;
3796 int i, k;
3797
3798 sp = skb_sec_path(skb);
3799 if (!sp)
3800 sp = &dummy;
3801
3802 for (pi = 0; pi < npols; pi++) {
3803 if (pols[pi] != pol &&
3804 pols[pi]->action != XFRM_POLICY_ALLOW) {
3805 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3806 goto reject;
3807 }
3808 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3809 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3810 goto reject_error;
3811 }
3812 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3813 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3814 }
3815 xfrm_nr = ti;
3816
3817 if (npols > 1) {
3818 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3819 tpp = stp;
3820 }
3821
3822 /* For each tunnel xfrm, find the first matching tmpl.
3823 * For each tmpl before that, find corresponding xfrm.
3824 * Order is _important_. Later we will implement
3825 * some barriers, but at the moment barriers
3826 * are implied between each two transformations.
3827 * Upon success, marks secpath entries as having been
3828 * verified to allow them to be skipped in future policy
3829 * checks (e.g. nested tunnels).
3830 */
3831 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3832 k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3833 if (k < 0) {
3834 if (k < -1)
3835 /* "-2 - errored_index" returned */
3836 xerr_idx = -(2+k);
3837 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3838 goto reject;
3839 }
3840 }
3841
3842 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3843 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3844 goto reject;
3845 }
3846
3847 xfrm_pols_put(pols, npols);
3848 sp->verified_cnt = k;
3849
3850 return 1;
3851 }
3852 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3853
3854reject:
3855 xfrm_secpath_reject(xerr_idx, skb, &fl);
3856reject_error:
3857 xfrm_pols_put(pols, npols);
3858 return 0;
3859}
3860EXPORT_SYMBOL(__xfrm_policy_check);
3861
3862int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3863{
3864 struct net *net = dev_net(skb->dev);
3865 struct flowi fl;
3866 struct dst_entry *dst;
3867 int res = 1;
3868
3869 if (xfrm_decode_session(net, skb, &fl, family) < 0) {
3870 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3871 return 0;
3872 }
3873
3874 skb_dst_force(skb);
3875 if (!skb_dst(skb)) {
3876 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3877 return 0;
3878 }
3879
3880 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3881 if (IS_ERR(dst)) {
3882 res = 0;
3883 dst = NULL;
3884 }
3885
3886 if (dst && !dst->xfrm)
3887 dst = xfrm_out_fwd_icmp(skb, &fl, family, dst);
3888
3889 skb_dst_set(skb, dst);
3890 return res;
3891}
3892EXPORT_SYMBOL(__xfrm_route_forward);
3893
3894/* Optimize later using cookies and generation ids. */
3895
3896static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3897{
3898 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3899 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3900 * get validated by dst_ops->check on every use. We do this
3901 * because when a normal route referenced by an XFRM dst is
3902 * obsoleted we do not go looking around for all parent
3903 * referencing XFRM dsts so that we can invalidate them. It
3904 * is just too much work. Instead we make the checks here on
3905 * every use. For example:
3906 *
3907 * XFRM dst A --> IPv4 dst X
3908 *
3909 * X is the "xdst->route" of A (X is also the "dst->path" of A
3910 * in this example). If X is marked obsolete, "A" will not
3911 * notice. That's what we are validating here via the
3912 * stale_bundle() check.
3913 *
3914 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3915 * be marked on it.
3916 * This will force stale_bundle() to fail on any xdst bundle with
3917 * this dst linked in it.
3918 */
3919 if (dst->obsolete < 0 && !stale_bundle(dst))
3920 return dst;
3921
3922 return NULL;
3923}
3924
3925static int stale_bundle(struct dst_entry *dst)
3926{
3927 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3928}
3929
3930void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3931{
3932 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3933 dst->dev = blackhole_netdev;
3934 dev_hold(dst->dev);
3935 dev_put(dev);
3936 }
3937}
3938EXPORT_SYMBOL(xfrm_dst_ifdown);
3939
3940static void xfrm_link_failure(struct sk_buff *skb)
3941{
3942 /* Impossible. Such dst must be popped before reaches point of failure. */
3943}
3944
3945static void xfrm_negative_advice(struct sock *sk, struct dst_entry *dst)
3946{
3947 if (dst->obsolete)
3948 sk_dst_reset(sk);
3949}
3950
3951static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3952{
3953 while (nr--) {
3954 struct xfrm_dst *xdst = bundle[nr];
3955 u32 pmtu, route_mtu_cached;
3956 struct dst_entry *dst;
3957
3958 dst = &xdst->u.dst;
3959 pmtu = dst_mtu(xfrm_dst_child(dst));
3960 xdst->child_mtu_cached = pmtu;
3961
3962 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3963
3964 route_mtu_cached = dst_mtu(xdst->route);
3965 xdst->route_mtu_cached = route_mtu_cached;
3966
3967 if (pmtu > route_mtu_cached)
3968 pmtu = route_mtu_cached;
3969
3970 dst_metric_set(dst, RTAX_MTU, pmtu);
3971 }
3972}
3973
3974/* Check that the bundle accepts the flow and its components are
3975 * still valid.
3976 */
3977
3978static int xfrm_bundle_ok(struct xfrm_dst *first)
3979{
3980 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3981 struct dst_entry *dst = &first->u.dst;
3982 struct xfrm_dst *xdst;
3983 int start_from, nr;
3984 u32 mtu;
3985
3986 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3987 (dst->dev && !netif_running(dst->dev)))
3988 return 0;
3989
3990 if (dst->flags & DST_XFRM_QUEUE)
3991 return 1;
3992
3993 start_from = nr = 0;
3994 do {
3995 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3996
3997 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3998 return 0;
3999 if (xdst->xfrm_genid != dst->xfrm->genid)
4000 return 0;
4001 if (xdst->num_pols > 0 &&
4002 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
4003 return 0;
4004
4005 bundle[nr++] = xdst;
4006
4007 mtu = dst_mtu(xfrm_dst_child(dst));
4008 if (xdst->child_mtu_cached != mtu) {
4009 start_from = nr;
4010 xdst->child_mtu_cached = mtu;
4011 }
4012
4013 if (!dst_check(xdst->route, xdst->route_cookie))
4014 return 0;
4015 mtu = dst_mtu(xdst->route);
4016 if (xdst->route_mtu_cached != mtu) {
4017 start_from = nr;
4018 xdst->route_mtu_cached = mtu;
4019 }
4020
4021 dst = xfrm_dst_child(dst);
4022 } while (dst->xfrm);
4023
4024 if (likely(!start_from))
4025 return 1;
4026
4027 xdst = bundle[start_from - 1];
4028 mtu = xdst->child_mtu_cached;
4029 while (start_from--) {
4030 dst = &xdst->u.dst;
4031
4032 mtu = xfrm_state_mtu(dst->xfrm, mtu);
4033 if (mtu > xdst->route_mtu_cached)
4034 mtu = xdst->route_mtu_cached;
4035 dst_metric_set(dst, RTAX_MTU, mtu);
4036 if (!start_from)
4037 break;
4038
4039 xdst = bundle[start_from - 1];
4040 xdst->child_mtu_cached = mtu;
4041 }
4042
4043 return 1;
4044}
4045
4046static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
4047{
4048 return dst_metric_advmss(xfrm_dst_path(dst));
4049}
4050
4051static unsigned int xfrm_mtu(const struct dst_entry *dst)
4052{
4053 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
4054
4055 return mtu ? : dst_mtu(xfrm_dst_path(dst));
4056}
4057
4058static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
4059 const void *daddr)
4060{
4061 while (dst->xfrm) {
4062 const struct xfrm_state *xfrm = dst->xfrm;
4063
4064 dst = xfrm_dst_child(dst);
4065
4066 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
4067 continue;
4068 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
4069 daddr = xfrm->coaddr;
4070 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
4071 daddr = &xfrm->id.daddr;
4072 }
4073 return daddr;
4074}
4075
4076static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
4077 struct sk_buff *skb,
4078 const void *daddr)
4079{
4080 const struct dst_entry *path = xfrm_dst_path(dst);
4081
4082 if (!skb)
4083 daddr = xfrm_get_dst_nexthop(dst, daddr);
4084 return path->ops->neigh_lookup(path, skb, daddr);
4085}
4086
4087static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
4088{
4089 const struct dst_entry *path = xfrm_dst_path(dst);
4090
4091 daddr = xfrm_get_dst_nexthop(dst, daddr);
4092 path->ops->confirm_neigh(path, daddr);
4093}
4094
4095int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
4096{
4097 int err = 0;
4098
4099 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
4100 return -EAFNOSUPPORT;
4101
4102 spin_lock(&xfrm_policy_afinfo_lock);
4103 if (unlikely(xfrm_policy_afinfo[family] != NULL))
4104 err = -EEXIST;
4105 else {
4106 struct dst_ops *dst_ops = afinfo->dst_ops;
4107 if (likely(dst_ops->kmem_cachep == NULL))
4108 dst_ops->kmem_cachep = xfrm_dst_cache;
4109 if (likely(dst_ops->check == NULL))
4110 dst_ops->check = xfrm_dst_check;
4111 if (likely(dst_ops->default_advmss == NULL))
4112 dst_ops->default_advmss = xfrm_default_advmss;
4113 if (likely(dst_ops->mtu == NULL))
4114 dst_ops->mtu = xfrm_mtu;
4115 if (likely(dst_ops->negative_advice == NULL))
4116 dst_ops->negative_advice = xfrm_negative_advice;
4117 if (likely(dst_ops->link_failure == NULL))
4118 dst_ops->link_failure = xfrm_link_failure;
4119 if (likely(dst_ops->neigh_lookup == NULL))
4120 dst_ops->neigh_lookup = xfrm_neigh_lookup;
4121 if (likely(!dst_ops->confirm_neigh))
4122 dst_ops->confirm_neigh = xfrm_confirm_neigh;
4123 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
4124 }
4125 spin_unlock(&xfrm_policy_afinfo_lock);
4126
4127 return err;
4128}
4129EXPORT_SYMBOL(xfrm_policy_register_afinfo);
4130
4131void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
4132{
4133 struct dst_ops *dst_ops = afinfo->dst_ops;
4134 int i;
4135
4136 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
4137 if (xfrm_policy_afinfo[i] != afinfo)
4138 continue;
4139 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
4140 break;
4141 }
4142
4143 synchronize_rcu();
4144
4145 dst_ops->kmem_cachep = NULL;
4146 dst_ops->check = NULL;
4147 dst_ops->negative_advice = NULL;
4148 dst_ops->link_failure = NULL;
4149}
4150EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
4151
4152void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
4153{
4154 spin_lock(&xfrm_if_cb_lock);
4155 rcu_assign_pointer(xfrm_if_cb, ifcb);
4156 spin_unlock(&xfrm_if_cb_lock);
4157}
4158EXPORT_SYMBOL(xfrm_if_register_cb);
4159
4160void xfrm_if_unregister_cb(void)
4161{
4162 RCU_INIT_POINTER(xfrm_if_cb, NULL);
4163 synchronize_rcu();
4164}
4165EXPORT_SYMBOL(xfrm_if_unregister_cb);
4166
4167#ifdef CONFIG_XFRM_STATISTICS
4168static int __net_init xfrm_statistics_init(struct net *net)
4169{
4170 int rv;
4171 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4172 if (!net->mib.xfrm_statistics)
4173 return -ENOMEM;
4174 rv = xfrm_proc_init(net);
4175 if (rv < 0)
4176 free_percpu(net->mib.xfrm_statistics);
4177 return rv;
4178}
4179
4180static void xfrm_statistics_fini(struct net *net)
4181{
4182 xfrm_proc_fini(net);
4183 free_percpu(net->mib.xfrm_statistics);
4184}
4185#else
4186static int __net_init xfrm_statistics_init(struct net *net)
4187{
4188 return 0;
4189}
4190
4191static void xfrm_statistics_fini(struct net *net)
4192{
4193}
4194#endif
4195
4196static int __net_init xfrm_policy_init(struct net *net)
4197{
4198 unsigned int hmask, sz;
4199 int dir, err;
4200
4201 if (net_eq(net, &init_net)) {
4202 xfrm_dst_cache = KMEM_CACHE(xfrm_dst, SLAB_HWCACHE_ALIGN | SLAB_PANIC);
4203 err = rhashtable_init(&xfrm_policy_inexact_table,
4204 &xfrm_pol_inexact_params);
4205 BUG_ON(err);
4206 }
4207
4208 hmask = 8 - 1;
4209 sz = (hmask+1) * sizeof(struct hlist_head);
4210
4211 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4212 if (!net->xfrm.policy_byidx)
4213 goto out_byidx;
4214 net->xfrm.policy_idx_hmask = hmask;
4215
4216 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4217 struct xfrm_policy_hash *htab;
4218
4219 net->xfrm.policy_count[dir] = 0;
4220 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4221
4222 htab = &net->xfrm.policy_bydst[dir];
4223 htab->table = xfrm_hash_alloc(sz);
4224 if (!htab->table)
4225 goto out_bydst;
4226 htab->hmask = hmask;
4227 htab->dbits4 = 32;
4228 htab->sbits4 = 32;
4229 htab->dbits6 = 128;
4230 htab->sbits6 = 128;
4231 }
4232 net->xfrm.policy_hthresh.lbits4 = 32;
4233 net->xfrm.policy_hthresh.rbits4 = 32;
4234 net->xfrm.policy_hthresh.lbits6 = 128;
4235 net->xfrm.policy_hthresh.rbits6 = 128;
4236
4237 seqlock_init(&net->xfrm.policy_hthresh.lock);
4238
4239 INIT_LIST_HEAD(&net->xfrm.policy_all);
4240 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4241 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4242 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4243 return 0;
4244
4245out_bydst:
4246 for (dir--; dir >= 0; dir--) {
4247 struct xfrm_policy_hash *htab;
4248
4249 htab = &net->xfrm.policy_bydst[dir];
4250 xfrm_hash_free(htab->table, sz);
4251 }
4252 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4253out_byidx:
4254 return -ENOMEM;
4255}
4256
4257static void xfrm_policy_fini(struct net *net)
4258{
4259 struct xfrm_pol_inexact_bin *b, *t;
4260 unsigned int sz;
4261 int dir;
4262
4263 flush_work(&net->xfrm.policy_hash_work);
4264#ifdef CONFIG_XFRM_SUB_POLICY
4265 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4266#endif
4267 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4268
4269 WARN_ON(!list_empty(&net->xfrm.policy_all));
4270
4271 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4272 struct xfrm_policy_hash *htab;
4273
4274 htab = &net->xfrm.policy_bydst[dir];
4275 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4276 WARN_ON(!hlist_empty(htab->table));
4277 xfrm_hash_free(htab->table, sz);
4278 }
4279
4280 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4281 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4282 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4283
4284 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4285 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4286 __xfrm_policy_inexact_prune_bin(b, true);
4287 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4288}
4289
4290static int __net_init xfrm_net_init(struct net *net)
4291{
4292 int rv;
4293
4294 /* Initialize the per-net locks here */
4295 spin_lock_init(&net->xfrm.xfrm_state_lock);
4296 spin_lock_init(&net->xfrm.xfrm_policy_lock);
4297 seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4298 mutex_init(&net->xfrm.xfrm_cfg_mutex);
4299 net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4300 net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4301 net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4302
4303 rv = xfrm_statistics_init(net);
4304 if (rv < 0)
4305 goto out_statistics;
4306 rv = xfrm_state_init(net);
4307 if (rv < 0)
4308 goto out_state;
4309 rv = xfrm_policy_init(net);
4310 if (rv < 0)
4311 goto out_policy;
4312 rv = xfrm_sysctl_init(net);
4313 if (rv < 0)
4314 goto out_sysctl;
4315
4316 rv = xfrm_nat_keepalive_net_init(net);
4317 if (rv < 0)
4318 goto out_nat_keepalive;
4319
4320 return 0;
4321
4322out_nat_keepalive:
4323 xfrm_sysctl_fini(net);
4324out_sysctl:
4325 xfrm_policy_fini(net);
4326out_policy:
4327 xfrm_state_fini(net);
4328out_state:
4329 xfrm_statistics_fini(net);
4330out_statistics:
4331 return rv;
4332}
4333
4334static void __net_exit xfrm_net_exit(struct net *net)
4335{
4336 xfrm_nat_keepalive_net_fini(net);
4337 xfrm_sysctl_fini(net);
4338 xfrm_policy_fini(net);
4339 xfrm_state_fini(net);
4340 xfrm_statistics_fini(net);
4341}
4342
4343static struct pernet_operations __net_initdata xfrm_net_ops = {
4344 .init = xfrm_net_init,
4345 .exit = xfrm_net_exit,
4346};
4347
4348static const struct flow_dissector_key xfrm_flow_dissector_keys[] = {
4349 {
4350 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
4351 .offset = offsetof(struct xfrm_flow_keys, control),
4352 },
4353 {
4354 .key_id = FLOW_DISSECTOR_KEY_BASIC,
4355 .offset = offsetof(struct xfrm_flow_keys, basic),
4356 },
4357 {
4358 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
4359 .offset = offsetof(struct xfrm_flow_keys, addrs.ipv4),
4360 },
4361 {
4362 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
4363 .offset = offsetof(struct xfrm_flow_keys, addrs.ipv6),
4364 },
4365 {
4366 .key_id = FLOW_DISSECTOR_KEY_PORTS,
4367 .offset = offsetof(struct xfrm_flow_keys, ports),
4368 },
4369 {
4370 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
4371 .offset = offsetof(struct xfrm_flow_keys, gre),
4372 },
4373 {
4374 .key_id = FLOW_DISSECTOR_KEY_IP,
4375 .offset = offsetof(struct xfrm_flow_keys, ip),
4376 },
4377 {
4378 .key_id = FLOW_DISSECTOR_KEY_ICMP,
4379 .offset = offsetof(struct xfrm_flow_keys, icmp),
4380 },
4381};
4382
4383void __init xfrm_init(void)
4384{
4385 skb_flow_dissector_init(&xfrm_session_dissector,
4386 xfrm_flow_dissector_keys,
4387 ARRAY_SIZE(xfrm_flow_dissector_keys));
4388
4389 register_pernet_subsys(&xfrm_net_ops);
4390 xfrm_dev_init();
4391 xfrm_input_init();
4392
4393#ifdef CONFIG_XFRM_ESPINTCP
4394 espintcp_init();
4395#endif
4396
4397 register_xfrm_state_bpf();
4398 xfrm_nat_keepalive_init(AF_INET);
4399}
4400
4401#ifdef CONFIG_AUDITSYSCALL
4402static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4403 struct audit_buffer *audit_buf)
4404{
4405 struct xfrm_sec_ctx *ctx = xp->security;
4406 struct xfrm_selector *sel = &xp->selector;
4407
4408 if (ctx)
4409 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4410 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4411
4412 switch (sel->family) {
4413 case AF_INET:
4414 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4415 if (sel->prefixlen_s != 32)
4416 audit_log_format(audit_buf, " src_prefixlen=%d",
4417 sel->prefixlen_s);
4418 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4419 if (sel->prefixlen_d != 32)
4420 audit_log_format(audit_buf, " dst_prefixlen=%d",
4421 sel->prefixlen_d);
4422 break;
4423 case AF_INET6:
4424 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4425 if (sel->prefixlen_s != 128)
4426 audit_log_format(audit_buf, " src_prefixlen=%d",
4427 sel->prefixlen_s);
4428 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4429 if (sel->prefixlen_d != 128)
4430 audit_log_format(audit_buf, " dst_prefixlen=%d",
4431 sel->prefixlen_d);
4432 break;
4433 }
4434}
4435
4436void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4437{
4438 struct audit_buffer *audit_buf;
4439
4440 audit_buf = xfrm_audit_start("SPD-add");
4441 if (audit_buf == NULL)
4442 return;
4443 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4444 audit_log_format(audit_buf, " res=%u", result);
4445 xfrm_audit_common_policyinfo(xp, audit_buf);
4446 audit_log_end(audit_buf);
4447}
4448EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4449
4450void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4451 bool task_valid)
4452{
4453 struct audit_buffer *audit_buf;
4454
4455 audit_buf = xfrm_audit_start("SPD-delete");
4456 if (audit_buf == NULL)
4457 return;
4458 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4459 audit_log_format(audit_buf, " res=%u", result);
4460 xfrm_audit_common_policyinfo(xp, audit_buf);
4461 audit_log_end(audit_buf);
4462}
4463EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4464#endif
4465
4466#ifdef CONFIG_XFRM_MIGRATE
4467static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4468 u8 dir, u8 type, struct net *net, u32 if_id)
4469{
4470 struct xfrm_policy *pol;
4471 struct flowi fl;
4472
4473 memset(&fl, 0, sizeof(fl));
4474
4475 fl.flowi_proto = sel->proto;
4476
4477 switch (sel->family) {
4478 case AF_INET:
4479 fl.u.ip4.saddr = sel->saddr.a4;
4480 fl.u.ip4.daddr = sel->daddr.a4;
4481 if (sel->proto == IPSEC_ULPROTO_ANY)
4482 break;
4483 fl.u.flowi4_oif = sel->ifindex;
4484 fl.u.ip4.fl4_sport = sel->sport;
4485 fl.u.ip4.fl4_dport = sel->dport;
4486 break;
4487 case AF_INET6:
4488 fl.u.ip6.saddr = sel->saddr.in6;
4489 fl.u.ip6.daddr = sel->daddr.in6;
4490 if (sel->proto == IPSEC_ULPROTO_ANY)
4491 break;
4492 fl.u.flowi6_oif = sel->ifindex;
4493 fl.u.ip6.fl4_sport = sel->sport;
4494 fl.u.ip6.fl4_dport = sel->dport;
4495 break;
4496 default:
4497 return ERR_PTR(-EAFNOSUPPORT);
4498 }
4499
4500 rcu_read_lock();
4501
4502 pol = xfrm_policy_lookup_bytype(net, type, &fl, sel->family, dir, if_id);
4503 if (IS_ERR_OR_NULL(pol))
4504 goto out_unlock;
4505
4506 if (!xfrm_pol_hold_rcu(pol))
4507 pol = NULL;
4508out_unlock:
4509 rcu_read_unlock();
4510 return pol;
4511}
4512
4513static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4514{
4515 int match = 0;
4516
4517 if (t->mode == m->mode && t->id.proto == m->proto &&
4518 (m->reqid == 0 || t->reqid == m->reqid)) {
4519 switch (t->mode) {
4520 case XFRM_MODE_TUNNEL:
4521 case XFRM_MODE_BEET:
4522 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4523 m->old_family) &&
4524 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4525 m->old_family)) {
4526 match = 1;
4527 }
4528 break;
4529 case XFRM_MODE_TRANSPORT:
4530 /* in case of transport mode, template does not store
4531 any IP addresses, hence we just compare mode and
4532 protocol */
4533 match = 1;
4534 break;
4535 default:
4536 break;
4537 }
4538 }
4539 return match;
4540}
4541
4542/* update endpoint address(es) of template(s) */
4543static int xfrm_policy_migrate(struct xfrm_policy *pol,
4544 struct xfrm_migrate *m, int num_migrate,
4545 struct netlink_ext_ack *extack)
4546{
4547 struct xfrm_migrate *mp;
4548 int i, j, n = 0;
4549
4550 write_lock_bh(&pol->lock);
4551 if (unlikely(pol->walk.dead)) {
4552 /* target policy has been deleted */
4553 NL_SET_ERR_MSG(extack, "Target policy not found");
4554 write_unlock_bh(&pol->lock);
4555 return -ENOENT;
4556 }
4557
4558 for (i = 0; i < pol->xfrm_nr; i++) {
4559 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4560 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4561 continue;
4562 n++;
4563 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4564 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4565 continue;
4566 /* update endpoints */
4567 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4568 sizeof(pol->xfrm_vec[i].id.daddr));
4569 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4570 sizeof(pol->xfrm_vec[i].saddr));
4571 pol->xfrm_vec[i].encap_family = mp->new_family;
4572 /* flush bundles */
4573 atomic_inc(&pol->genid);
4574 }
4575 }
4576
4577 write_unlock_bh(&pol->lock);
4578
4579 if (!n)
4580 return -ENODATA;
4581
4582 return 0;
4583}
4584
4585static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate,
4586 struct netlink_ext_ack *extack)
4587{
4588 int i, j;
4589
4590 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) {
4591 NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)");
4592 return -EINVAL;
4593 }
4594
4595 for (i = 0; i < num_migrate; i++) {
4596 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4597 xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) {
4598 NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null");
4599 return -EINVAL;
4600 }
4601
4602 /* check if there is any duplicated entry */
4603 for (j = i + 1; j < num_migrate; j++) {
4604 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4605 sizeof(m[i].old_daddr)) &&
4606 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4607 sizeof(m[i].old_saddr)) &&
4608 m[i].proto == m[j].proto &&
4609 m[i].mode == m[j].mode &&
4610 m[i].reqid == m[j].reqid &&
4611 m[i].old_family == m[j].old_family) {
4612 NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique");
4613 return -EINVAL;
4614 }
4615 }
4616 }
4617
4618 return 0;
4619}
4620
4621int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4622 struct xfrm_migrate *m, int num_migrate,
4623 struct xfrm_kmaddress *k, struct net *net,
4624 struct xfrm_encap_tmpl *encap, u32 if_id,
4625 struct netlink_ext_ack *extack)
4626{
4627 int i, err, nx_cur = 0, nx_new = 0;
4628 struct xfrm_policy *pol = NULL;
4629 struct xfrm_state *x, *xc;
4630 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4631 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4632 struct xfrm_migrate *mp;
4633
4634 /* Stage 0 - sanity checks */
4635 err = xfrm_migrate_check(m, num_migrate, extack);
4636 if (err < 0)
4637 goto out;
4638
4639 if (dir >= XFRM_POLICY_MAX) {
4640 NL_SET_ERR_MSG(extack, "Invalid policy direction");
4641 err = -EINVAL;
4642 goto out;
4643 }
4644
4645 /* Stage 1 - find policy */
4646 pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id);
4647 if (IS_ERR_OR_NULL(pol)) {
4648 NL_SET_ERR_MSG(extack, "Target policy not found");
4649 err = IS_ERR(pol) ? PTR_ERR(pol) : -ENOENT;
4650 goto out;
4651 }
4652
4653 /* Stage 2 - find and update state(s) */
4654 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4655 if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4656 x_cur[nx_cur] = x;
4657 nx_cur++;
4658 xc = xfrm_state_migrate(x, mp, encap);
4659 if (xc) {
4660 x_new[nx_new] = xc;
4661 nx_new++;
4662 } else {
4663 err = -ENODATA;
4664 goto restore_state;
4665 }
4666 }
4667 }
4668
4669 /* Stage 3 - update policy */
4670 err = xfrm_policy_migrate(pol, m, num_migrate, extack);
4671 if (err < 0)
4672 goto restore_state;
4673
4674 /* Stage 4 - delete old state(s) */
4675 if (nx_cur) {
4676 xfrm_states_put(x_cur, nx_cur);
4677 xfrm_states_delete(x_cur, nx_cur);
4678 }
4679
4680 /* Stage 5 - announce */
4681 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4682
4683 xfrm_pol_put(pol);
4684
4685 return 0;
4686out:
4687 return err;
4688
4689restore_state:
4690 if (pol)
4691 xfrm_pol_put(pol);
4692 if (nx_cur)
4693 xfrm_states_put(x_cur, nx_cur);
4694 if (nx_new)
4695 xfrm_states_delete(x_new, nx_new);
4696
4697 return err;
4698}
4699EXPORT_SYMBOL(xfrm_migrate);
4700#endif
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
49struct 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
58struct 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
109struct xfrm_pol_inexact_key {
110 possible_net_t net;
111 u32 if_id;
112 u16 family;
113 u8 dir, type;
114};
115
116struct 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
134enum 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
143struct xfrm_pol_inexact_candidates {
144 struct hlist_head *res[XFRM_POL_CAND_MAX];
145};
146
147static DEFINE_SPINLOCK(xfrm_if_cb_lock);
148static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
149
150static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
151static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
152 __read_mostly;
153
154static struct kmem_cache *xfrm_dst_cache __ro_after_init;
155static __read_mostly seqcount_t xfrm_policy_hash_generation;
156
157static struct rhashtable xfrm_policy_inexact_table;
158static const struct rhashtable_params xfrm_pol_inexact_params;
159
160static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
161static int stale_bundle(struct dst_entry *dst);
162static int xfrm_bundle_ok(struct xfrm_dst *xdst);
163static void xfrm_policy_queue_process(struct timer_list *t);
164
165static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
166static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
167 int dir);
168
169static struct xfrm_pol_inexact_bin *
170xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
171 u32 if_id);
172
173static struct xfrm_pol_inexact_bin *
174xfrm_policy_inexact_lookup_rcu(struct net *net,
175 u8 type, u16 family, u8 dir, u32 if_id);
176static struct xfrm_policy *
177xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
178 bool excl);
179static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
180 struct xfrm_policy *policy);
181
182static bool
183xfrm_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
188static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
189{
190 return refcount_inc_not_zero(&policy->refcnt);
191}
192
193static 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
206static 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
219bool 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
231static 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(). */
245static const struct xfrm_if_cb *xfrm_if_get_cb(void)
246{
247 return rcu_dereference(xfrm_if_cb);
248}
249
250struct 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}
268EXPORT_SYMBOL(__xfrm_dst_lookup);
269
270static 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
302static 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
310static 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
368out:
369 read_unlock(&xp->lock);
370 xfrm_pol_put(xp);
371 return;
372
373expired:
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
384struct 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}
405EXPORT_SYMBOL(xfrm_policy_alloc);
406
407static 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
417void 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}
426EXPORT_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
432static void xfrm_policy_kill(struct xfrm_policy *policy)
433{
434 policy->walk.dead = 1;
435
436 atomic_inc(&policy->genid);
437
438 if (del_timer(&policy->polq.hold_timer))
439 xfrm_pol_put(policy);
440 skb_queue_purge(&policy->polq.hold_queue);
441
442 if (del_timer(&policy->timer))
443 xfrm_pol_put(policy);
444
445 xfrm_pol_put(policy);
446}
447
448static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
449
450static inline unsigned int idx_hash(struct net *net, u32 index)
451{
452 return __idx_hash(index, net->xfrm.policy_idx_hmask);
453}
454
455/* calculate policy hash thresholds */
456static void __get_hash_thresh(struct net *net,
457 unsigned short family, int dir,
458 u8 *dbits, u8 *sbits)
459{
460 switch (family) {
461 case AF_INET:
462 *dbits = net->xfrm.policy_bydst[dir].dbits4;
463 *sbits = net->xfrm.policy_bydst[dir].sbits4;
464 break;
465
466 case AF_INET6:
467 *dbits = net->xfrm.policy_bydst[dir].dbits6;
468 *sbits = net->xfrm.policy_bydst[dir].sbits6;
469 break;
470
471 default:
472 *dbits = 0;
473 *sbits = 0;
474 }
475}
476
477static struct hlist_head *policy_hash_bysel(struct net *net,
478 const struct xfrm_selector *sel,
479 unsigned short family, int dir)
480{
481 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
482 unsigned int hash;
483 u8 dbits;
484 u8 sbits;
485
486 __get_hash_thresh(net, family, dir, &dbits, &sbits);
487 hash = __sel_hash(sel, family, hmask, dbits, sbits);
488
489 if (hash == hmask + 1)
490 return NULL;
491
492 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
493 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
494}
495
496static struct hlist_head *policy_hash_direct(struct net *net,
497 const xfrm_address_t *daddr,
498 const xfrm_address_t *saddr,
499 unsigned short family, int dir)
500{
501 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
502 unsigned int hash;
503 u8 dbits;
504 u8 sbits;
505
506 __get_hash_thresh(net, family, dir, &dbits, &sbits);
507 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
508
509 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
510 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
511}
512
513static void xfrm_dst_hash_transfer(struct net *net,
514 struct hlist_head *list,
515 struct hlist_head *ndsttable,
516 unsigned int nhashmask,
517 int dir)
518{
519 struct hlist_node *tmp, *entry0 = NULL;
520 struct xfrm_policy *pol;
521 unsigned int h0 = 0;
522 u8 dbits;
523 u8 sbits;
524
525redo:
526 hlist_for_each_entry_safe(pol, tmp, list, bydst) {
527 unsigned int h;
528
529 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
530 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
531 pol->family, nhashmask, dbits, sbits);
532 if (!entry0) {
533 hlist_del_rcu(&pol->bydst);
534 hlist_add_head_rcu(&pol->bydst, ndsttable + h);
535 h0 = h;
536 } else {
537 if (h != h0)
538 continue;
539 hlist_del_rcu(&pol->bydst);
540 hlist_add_behind_rcu(&pol->bydst, entry0);
541 }
542 entry0 = &pol->bydst;
543 }
544 if (!hlist_empty(list)) {
545 entry0 = NULL;
546 goto redo;
547 }
548}
549
550static void xfrm_idx_hash_transfer(struct hlist_head *list,
551 struct hlist_head *nidxtable,
552 unsigned int nhashmask)
553{
554 struct hlist_node *tmp;
555 struct xfrm_policy *pol;
556
557 hlist_for_each_entry_safe(pol, tmp, list, byidx) {
558 unsigned int h;
559
560 h = __idx_hash(pol->index, nhashmask);
561 hlist_add_head(&pol->byidx, nidxtable+h);
562 }
563}
564
565static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
566{
567 return ((old_hmask + 1) << 1) - 1;
568}
569
570static void xfrm_bydst_resize(struct net *net, int dir)
571{
572 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
573 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
574 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
575 struct hlist_head *ndst = xfrm_hash_alloc(nsize);
576 struct hlist_head *odst;
577 int i;
578
579 if (!ndst)
580 return;
581
582 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
583 write_seqcount_begin(&xfrm_policy_hash_generation);
584
585 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
586 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
587
588 for (i = hmask; i >= 0; i--)
589 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
590
591 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
592 net->xfrm.policy_bydst[dir].hmask = nhashmask;
593
594 write_seqcount_end(&xfrm_policy_hash_generation);
595 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
596
597 synchronize_rcu();
598
599 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
600}
601
602static void xfrm_byidx_resize(struct net *net, int total)
603{
604 unsigned int hmask = net->xfrm.policy_idx_hmask;
605 unsigned int nhashmask = xfrm_new_hash_mask(hmask);
606 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
607 struct hlist_head *oidx = net->xfrm.policy_byidx;
608 struct hlist_head *nidx = xfrm_hash_alloc(nsize);
609 int i;
610
611 if (!nidx)
612 return;
613
614 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
615
616 for (i = hmask; i >= 0; i--)
617 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
618
619 net->xfrm.policy_byidx = nidx;
620 net->xfrm.policy_idx_hmask = nhashmask;
621
622 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
623
624 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
625}
626
627static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
628{
629 unsigned int cnt = net->xfrm.policy_count[dir];
630 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
631
632 if (total)
633 *total += cnt;
634
635 if ((hmask + 1) < xfrm_policy_hashmax &&
636 cnt > hmask)
637 return 1;
638
639 return 0;
640}
641
642static inline int xfrm_byidx_should_resize(struct net *net, int total)
643{
644 unsigned int hmask = net->xfrm.policy_idx_hmask;
645
646 if ((hmask + 1) < xfrm_policy_hashmax &&
647 total > hmask)
648 return 1;
649
650 return 0;
651}
652
653void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
654{
655 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
656 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
657 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
658 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
659 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
660 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
661 si->spdhcnt = net->xfrm.policy_idx_hmask;
662 si->spdhmcnt = xfrm_policy_hashmax;
663}
664EXPORT_SYMBOL(xfrm_spd_getinfo);
665
666static DEFINE_MUTEX(hash_resize_mutex);
667static void xfrm_hash_resize(struct work_struct *work)
668{
669 struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
670 int dir, total;
671
672 mutex_lock(&hash_resize_mutex);
673
674 total = 0;
675 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
676 if (xfrm_bydst_should_resize(net, dir, &total))
677 xfrm_bydst_resize(net, dir);
678 }
679 if (xfrm_byidx_should_resize(net, total))
680 xfrm_byidx_resize(net, total);
681
682 mutex_unlock(&hash_resize_mutex);
683}
684
685/* Make sure *pol can be inserted into fastbin.
686 * Useful to check that later insert requests will be sucessful
687 * (provided xfrm_policy_lock is held throughout).
688 */
689static struct xfrm_pol_inexact_bin *
690xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
691{
692 struct xfrm_pol_inexact_bin *bin, *prev;
693 struct xfrm_pol_inexact_key k = {
694 .family = pol->family,
695 .type = pol->type,
696 .dir = dir,
697 .if_id = pol->if_id,
698 };
699 struct net *net = xp_net(pol);
700
701 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
702
703 write_pnet(&k.net, net);
704 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
705 xfrm_pol_inexact_params);
706 if (bin)
707 return bin;
708
709 bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
710 if (!bin)
711 return NULL;
712
713 bin->k = k;
714 INIT_HLIST_HEAD(&bin->hhead);
715 bin->root_d = RB_ROOT;
716 bin->root_s = RB_ROOT;
717 seqcount_init(&bin->count);
718
719 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
720 &bin->k, &bin->head,
721 xfrm_pol_inexact_params);
722 if (!prev) {
723 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
724 return bin;
725 }
726
727 kfree(bin);
728
729 return IS_ERR(prev) ? NULL : prev;
730}
731
732static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
733 int family, u8 prefixlen)
734{
735 if (xfrm_addr_any(addr, family))
736 return true;
737
738 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
739 return true;
740
741 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
742 return true;
743
744 return false;
745}
746
747static bool
748xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
749{
750 const xfrm_address_t *addr;
751 bool saddr_any, daddr_any;
752 u8 prefixlen;
753
754 addr = &policy->selector.saddr;
755 prefixlen = policy->selector.prefixlen_s;
756
757 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
758 policy->family,
759 prefixlen);
760 addr = &policy->selector.daddr;
761 prefixlen = policy->selector.prefixlen_d;
762 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
763 policy->family,
764 prefixlen);
765 return saddr_any && daddr_any;
766}
767
768static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
769 const xfrm_address_t *addr, u8 prefixlen)
770{
771 node->addr = *addr;
772 node->prefixlen = prefixlen;
773}
774
775static struct xfrm_pol_inexact_node *
776xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
777{
778 struct xfrm_pol_inexact_node *node;
779
780 node = kzalloc(sizeof(*node), GFP_ATOMIC);
781 if (node)
782 xfrm_pol_inexact_node_init(node, addr, prefixlen);
783
784 return node;
785}
786
787static int xfrm_policy_addr_delta(const xfrm_address_t *a,
788 const xfrm_address_t *b,
789 u8 prefixlen, u16 family)
790{
791 unsigned int pdw, pbi;
792 int delta = 0;
793
794 switch (family) {
795 case AF_INET:
796 if (sizeof(long) == 4 && prefixlen == 0)
797 return ntohl(a->a4) - ntohl(b->a4);
798 return (ntohl(a->a4) & ((~0UL << (32 - prefixlen)))) -
799 (ntohl(b->a4) & ((~0UL << (32 - prefixlen))));
800 case AF_INET6:
801 pdw = prefixlen >> 5;
802 pbi = prefixlen & 0x1f;
803
804 if (pdw) {
805 delta = memcmp(a->a6, b->a6, pdw << 2);
806 if (delta)
807 return delta;
808 }
809 if (pbi) {
810 u32 mask = ~0u << (32 - pbi);
811
812 delta = (ntohl(a->a6[pdw]) & mask) -
813 (ntohl(b->a6[pdw]) & mask);
814 }
815 break;
816 default:
817 break;
818 }
819
820 return delta;
821}
822
823static void xfrm_policy_inexact_list_reinsert(struct net *net,
824 struct xfrm_pol_inexact_node *n,
825 u16 family)
826{
827 unsigned int matched_s, matched_d;
828 struct xfrm_policy *policy, *p;
829
830 matched_s = 0;
831 matched_d = 0;
832
833 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
834 struct hlist_node *newpos = NULL;
835 bool matches_s, matches_d;
836
837 if (!policy->bydst_reinsert)
838 continue;
839
840 WARN_ON_ONCE(policy->family != family);
841
842 policy->bydst_reinsert = false;
843 hlist_for_each_entry(p, &n->hhead, bydst) {
844 if (policy->priority > p->priority)
845 newpos = &p->bydst;
846 else if (policy->priority == p->priority &&
847 policy->pos > p->pos)
848 newpos = &p->bydst;
849 else
850 break;
851 }
852
853 if (newpos)
854 hlist_add_behind_rcu(&policy->bydst, newpos);
855 else
856 hlist_add_head_rcu(&policy->bydst, &n->hhead);
857
858 /* paranoia checks follow.
859 * Check that the reinserted policy matches at least
860 * saddr or daddr for current node prefix.
861 *
862 * Matching both is fine, matching saddr in one policy
863 * (but not daddr) and then matching only daddr in another
864 * is a bug.
865 */
866 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
867 &n->addr,
868 n->prefixlen,
869 family) == 0;
870 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
871 &n->addr,
872 n->prefixlen,
873 family) == 0;
874 if (matches_s && matches_d)
875 continue;
876
877 WARN_ON_ONCE(!matches_s && !matches_d);
878 if (matches_s)
879 matched_s++;
880 if (matches_d)
881 matched_d++;
882 WARN_ON_ONCE(matched_s && matched_d);
883 }
884}
885
886static void xfrm_policy_inexact_node_reinsert(struct net *net,
887 struct xfrm_pol_inexact_node *n,
888 struct rb_root *new,
889 u16 family)
890{
891 struct xfrm_pol_inexact_node *node;
892 struct rb_node **p, *parent;
893
894 /* we should not have another subtree here */
895 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
896restart:
897 parent = NULL;
898 p = &new->rb_node;
899 while (*p) {
900 u8 prefixlen;
901 int delta;
902
903 parent = *p;
904 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
905
906 prefixlen = min(node->prefixlen, n->prefixlen);
907
908 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
909 prefixlen, family);
910 if (delta < 0) {
911 p = &parent->rb_left;
912 } else if (delta > 0) {
913 p = &parent->rb_right;
914 } else {
915 bool same_prefixlen = node->prefixlen == n->prefixlen;
916 struct xfrm_policy *tmp;
917
918 hlist_for_each_entry(tmp, &n->hhead, bydst) {
919 tmp->bydst_reinsert = true;
920 hlist_del_rcu(&tmp->bydst);
921 }
922
923 node->prefixlen = prefixlen;
924
925 xfrm_policy_inexact_list_reinsert(net, node, family);
926
927 if (same_prefixlen) {
928 kfree_rcu(n, rcu);
929 return;
930 }
931
932 rb_erase(*p, new);
933 kfree_rcu(n, rcu);
934 n = node;
935 goto restart;
936 }
937 }
938
939 rb_link_node_rcu(&n->node, parent, p);
940 rb_insert_color(&n->node, new);
941}
942
943/* merge nodes v and n */
944static void xfrm_policy_inexact_node_merge(struct net *net,
945 struct xfrm_pol_inexact_node *v,
946 struct xfrm_pol_inexact_node *n,
947 u16 family)
948{
949 struct xfrm_pol_inexact_node *node;
950 struct xfrm_policy *tmp;
951 struct rb_node *rnode;
952
953 /* To-be-merged node v has a subtree.
954 *
955 * Dismantle it and insert its nodes to n->root.
956 */
957 while ((rnode = rb_first(&v->root)) != NULL) {
958 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
959 rb_erase(&node->node, &v->root);
960 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
961 family);
962 }
963
964 hlist_for_each_entry(tmp, &v->hhead, bydst) {
965 tmp->bydst_reinsert = true;
966 hlist_del_rcu(&tmp->bydst);
967 }
968
969 xfrm_policy_inexact_list_reinsert(net, n, family);
970}
971
972static struct xfrm_pol_inexact_node *
973xfrm_policy_inexact_insert_node(struct net *net,
974 struct rb_root *root,
975 xfrm_address_t *addr,
976 u16 family, u8 prefixlen, u8 dir)
977{
978 struct xfrm_pol_inexact_node *cached = NULL;
979 struct rb_node **p, *parent = NULL;
980 struct xfrm_pol_inexact_node *node;
981
982 p = &root->rb_node;
983 while (*p) {
984 int delta;
985
986 parent = *p;
987 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
988
989 delta = xfrm_policy_addr_delta(addr, &node->addr,
990 node->prefixlen,
991 family);
992 if (delta == 0 && prefixlen >= node->prefixlen) {
993 WARN_ON_ONCE(cached); /* ipsec policies got lost */
994 return node;
995 }
996
997 if (delta < 0)
998 p = &parent->rb_left;
999 else
1000 p = &parent->rb_right;
1001
1002 if (prefixlen < node->prefixlen) {
1003 delta = xfrm_policy_addr_delta(addr, &node->addr,
1004 prefixlen,
1005 family);
1006 if (delta)
1007 continue;
1008
1009 /* This node is a subnet of the new prefix. It needs
1010 * to be removed and re-inserted with the smaller
1011 * prefix and all nodes that are now also covered
1012 * by the reduced prefixlen.
1013 */
1014 rb_erase(&node->node, root);
1015
1016 if (!cached) {
1017 xfrm_pol_inexact_node_init(node, addr,
1018 prefixlen);
1019 cached = node;
1020 } else {
1021 /* This node also falls within the new
1022 * prefixlen. Merge the to-be-reinserted
1023 * node and this one.
1024 */
1025 xfrm_policy_inexact_node_merge(net, node,
1026 cached, family);
1027 kfree_rcu(node, rcu);
1028 }
1029
1030 /* restart */
1031 p = &root->rb_node;
1032 parent = NULL;
1033 }
1034 }
1035
1036 node = cached;
1037 if (!node) {
1038 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1039 if (!node)
1040 return NULL;
1041 }
1042
1043 rb_link_node_rcu(&node->node, parent, p);
1044 rb_insert_color(&node->node, root);
1045
1046 return node;
1047}
1048
1049static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1050{
1051 struct xfrm_pol_inexact_node *node;
1052 struct rb_node *rn = rb_first(r);
1053
1054 while (rn) {
1055 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1056
1057 xfrm_policy_inexact_gc_tree(&node->root, rm);
1058 rn = rb_next(rn);
1059
1060 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1061 WARN_ON_ONCE(rm);
1062 continue;
1063 }
1064
1065 rb_erase(&node->node, r);
1066 kfree_rcu(node, rcu);
1067 }
1068}
1069
1070static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1071{
1072 write_seqcount_begin(&b->count);
1073 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1074 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1075 write_seqcount_end(&b->count);
1076
1077 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1078 !hlist_empty(&b->hhead)) {
1079 WARN_ON_ONCE(net_exit);
1080 return;
1081 }
1082
1083 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1084 xfrm_pol_inexact_params) == 0) {
1085 list_del(&b->inexact_bins);
1086 kfree_rcu(b, rcu);
1087 }
1088}
1089
1090static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1091{
1092 struct net *net = read_pnet(&b->k.net);
1093
1094 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1095 __xfrm_policy_inexact_prune_bin(b, false);
1096 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1097}
1098
1099static void __xfrm_policy_inexact_flush(struct net *net)
1100{
1101 struct xfrm_pol_inexact_bin *bin, *t;
1102
1103 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1104
1105 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1106 __xfrm_policy_inexact_prune_bin(bin, false);
1107}
1108
1109static struct hlist_head *
1110xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1111 struct xfrm_policy *policy, u8 dir)
1112{
1113 struct xfrm_pol_inexact_node *n;
1114 struct net *net;
1115
1116 net = xp_net(policy);
1117 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1118
1119 if (xfrm_policy_inexact_insert_use_any_list(policy))
1120 return &bin->hhead;
1121
1122 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1123 policy->family,
1124 policy->selector.prefixlen_d)) {
1125 write_seqcount_begin(&bin->count);
1126 n = xfrm_policy_inexact_insert_node(net,
1127 &bin->root_s,
1128 &policy->selector.saddr,
1129 policy->family,
1130 policy->selector.prefixlen_s,
1131 dir);
1132 write_seqcount_end(&bin->count);
1133 if (!n)
1134 return NULL;
1135
1136 return &n->hhead;
1137 }
1138
1139 /* daddr is fixed */
1140 write_seqcount_begin(&bin->count);
1141 n = xfrm_policy_inexact_insert_node(net,
1142 &bin->root_d,
1143 &policy->selector.daddr,
1144 policy->family,
1145 policy->selector.prefixlen_d, dir);
1146 write_seqcount_end(&bin->count);
1147 if (!n)
1148 return NULL;
1149
1150 /* saddr is wildcard */
1151 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1152 policy->family,
1153 policy->selector.prefixlen_s))
1154 return &n->hhead;
1155
1156 write_seqcount_begin(&bin->count);
1157 n = xfrm_policy_inexact_insert_node(net,
1158 &n->root,
1159 &policy->selector.saddr,
1160 policy->family,
1161 policy->selector.prefixlen_s, dir);
1162 write_seqcount_end(&bin->count);
1163 if (!n)
1164 return NULL;
1165
1166 return &n->hhead;
1167}
1168
1169static struct xfrm_policy *
1170xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1171{
1172 struct xfrm_pol_inexact_bin *bin;
1173 struct xfrm_policy *delpol;
1174 struct hlist_head *chain;
1175 struct net *net;
1176
1177 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1178 if (!bin)
1179 return ERR_PTR(-ENOMEM);
1180
1181 net = xp_net(policy);
1182 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1183
1184 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1185 if (!chain) {
1186 __xfrm_policy_inexact_prune_bin(bin, false);
1187 return ERR_PTR(-ENOMEM);
1188 }
1189
1190 delpol = xfrm_policy_insert_list(chain, policy, excl);
1191 if (delpol && excl) {
1192 __xfrm_policy_inexact_prune_bin(bin, false);
1193 return ERR_PTR(-EEXIST);
1194 }
1195
1196 chain = &net->xfrm.policy_inexact[dir];
1197 xfrm_policy_insert_inexact_list(chain, policy);
1198
1199 if (delpol)
1200 __xfrm_policy_inexact_prune_bin(bin, false);
1201
1202 return delpol;
1203}
1204
1205static void xfrm_hash_rebuild(struct work_struct *work)
1206{
1207 struct net *net = container_of(work, struct net,
1208 xfrm.policy_hthresh.work);
1209 unsigned int hmask;
1210 struct xfrm_policy *pol;
1211 struct xfrm_policy *policy;
1212 struct hlist_head *chain;
1213 struct hlist_head *odst;
1214 struct hlist_node *newpos;
1215 int i;
1216 int dir;
1217 unsigned seq;
1218 u8 lbits4, rbits4, lbits6, rbits6;
1219
1220 mutex_lock(&hash_resize_mutex);
1221
1222 /* read selector prefixlen thresholds */
1223 do {
1224 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1225
1226 lbits4 = net->xfrm.policy_hthresh.lbits4;
1227 rbits4 = net->xfrm.policy_hthresh.rbits4;
1228 lbits6 = net->xfrm.policy_hthresh.lbits6;
1229 rbits6 = net->xfrm.policy_hthresh.rbits6;
1230 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1231
1232 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1233 write_seqcount_begin(&xfrm_policy_hash_generation);
1234
1235 /* make sure that we can insert the indirect policies again before
1236 * we start with destructive action.
1237 */
1238 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1239 struct xfrm_pol_inexact_bin *bin;
1240 u8 dbits, sbits;
1241
1242 dir = xfrm_policy_id2dir(policy->index);
1243 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1244 continue;
1245
1246 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1247 if (policy->family == AF_INET) {
1248 dbits = rbits4;
1249 sbits = lbits4;
1250 } else {
1251 dbits = rbits6;
1252 sbits = lbits6;
1253 }
1254 } else {
1255 if (policy->family == AF_INET) {
1256 dbits = lbits4;
1257 sbits = rbits4;
1258 } else {
1259 dbits = lbits6;
1260 sbits = rbits6;
1261 }
1262 }
1263
1264 if (policy->selector.prefixlen_d < dbits ||
1265 policy->selector.prefixlen_s < sbits)
1266 continue;
1267
1268 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1269 if (!bin)
1270 goto out_unlock;
1271
1272 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1273 goto out_unlock;
1274 }
1275
1276 /* reset the bydst and inexact table in all directions */
1277 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1278 struct hlist_node *n;
1279
1280 hlist_for_each_entry_safe(policy, n,
1281 &net->xfrm.policy_inexact[dir],
1282 bydst_inexact_list) {
1283 hlist_del_rcu(&policy->bydst);
1284 hlist_del_init(&policy->bydst_inexact_list);
1285 }
1286
1287 hmask = net->xfrm.policy_bydst[dir].hmask;
1288 odst = net->xfrm.policy_bydst[dir].table;
1289 for (i = hmask; i >= 0; i--) {
1290 hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1291 hlist_del_rcu(&policy->bydst);
1292 }
1293 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1294 /* dir out => dst = remote, src = local */
1295 net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1296 net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1297 net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1298 net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1299 } else {
1300 /* dir in/fwd => dst = local, src = remote */
1301 net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1302 net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1303 net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1304 net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1305 }
1306 }
1307
1308 /* re-insert all policies by order of creation */
1309 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1310 if (policy->walk.dead)
1311 continue;
1312 dir = xfrm_policy_id2dir(policy->index);
1313 if (dir >= XFRM_POLICY_MAX) {
1314 /* skip socket policies */
1315 continue;
1316 }
1317 newpos = NULL;
1318 chain = policy_hash_bysel(net, &policy->selector,
1319 policy->family, dir);
1320
1321 if (!chain) {
1322 void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1323
1324 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1325 continue;
1326 }
1327
1328 hlist_for_each_entry(pol, chain, bydst) {
1329 if (policy->priority >= pol->priority)
1330 newpos = &pol->bydst;
1331 else
1332 break;
1333 }
1334 if (newpos)
1335 hlist_add_behind_rcu(&policy->bydst, newpos);
1336 else
1337 hlist_add_head_rcu(&policy->bydst, chain);
1338 }
1339
1340out_unlock:
1341 __xfrm_policy_inexact_flush(net);
1342 write_seqcount_end(&xfrm_policy_hash_generation);
1343 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1344
1345 mutex_unlock(&hash_resize_mutex);
1346}
1347
1348void xfrm_policy_hash_rebuild(struct net *net)
1349{
1350 schedule_work(&net->xfrm.policy_hthresh.work);
1351}
1352EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1353
1354/* Generate new index... KAME seems to generate them ordered by cost
1355 * of an absolute inpredictability of ordering of rules. This will not pass. */
1356static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1357{
1358 static u32 idx_generator;
1359
1360 for (;;) {
1361 struct hlist_head *list;
1362 struct xfrm_policy *p;
1363 u32 idx;
1364 int found;
1365
1366 if (!index) {
1367 idx = (idx_generator | dir);
1368 idx_generator += 8;
1369 } else {
1370 idx = index;
1371 index = 0;
1372 }
1373
1374 if (idx == 0)
1375 idx = 8;
1376 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1377 found = 0;
1378 hlist_for_each_entry(p, list, byidx) {
1379 if (p->index == idx) {
1380 found = 1;
1381 break;
1382 }
1383 }
1384 if (!found)
1385 return idx;
1386 }
1387}
1388
1389static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1390{
1391 u32 *p1 = (u32 *) s1;
1392 u32 *p2 = (u32 *) s2;
1393 int len = sizeof(struct xfrm_selector) / sizeof(u32);
1394 int i;
1395
1396 for (i = 0; i < len; i++) {
1397 if (p1[i] != p2[i])
1398 return 1;
1399 }
1400
1401 return 0;
1402}
1403
1404static void xfrm_policy_requeue(struct xfrm_policy *old,
1405 struct xfrm_policy *new)
1406{
1407 struct xfrm_policy_queue *pq = &old->polq;
1408 struct sk_buff_head list;
1409
1410 if (skb_queue_empty(&pq->hold_queue))
1411 return;
1412
1413 __skb_queue_head_init(&list);
1414
1415 spin_lock_bh(&pq->hold_queue.lock);
1416 skb_queue_splice_init(&pq->hold_queue, &list);
1417 if (del_timer(&pq->hold_timer))
1418 xfrm_pol_put(old);
1419 spin_unlock_bh(&pq->hold_queue.lock);
1420
1421 pq = &new->polq;
1422
1423 spin_lock_bh(&pq->hold_queue.lock);
1424 skb_queue_splice(&list, &pq->hold_queue);
1425 pq->timeout = XFRM_QUEUE_TMO_MIN;
1426 if (!mod_timer(&pq->hold_timer, jiffies))
1427 xfrm_pol_hold(new);
1428 spin_unlock_bh(&pq->hold_queue.lock);
1429}
1430
1431static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
1432 struct xfrm_policy *pol)
1433{
1434 u32 mark = policy->mark.v & policy->mark.m;
1435
1436 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
1437 return true;
1438
1439 if ((mark & pol->mark.m) == pol->mark.v &&
1440 policy->priority == pol->priority)
1441 return true;
1442
1443 return false;
1444}
1445
1446static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1447{
1448 const struct xfrm_pol_inexact_key *k = data;
1449 u32 a = k->type << 24 | k->dir << 16 | k->family;
1450
1451 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1452 seed);
1453}
1454
1455static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1456{
1457 const struct xfrm_pol_inexact_bin *b = data;
1458
1459 return xfrm_pol_bin_key(&b->k, 0, seed);
1460}
1461
1462static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1463 const void *ptr)
1464{
1465 const struct xfrm_pol_inexact_key *key = arg->key;
1466 const struct xfrm_pol_inexact_bin *b = ptr;
1467 int ret;
1468
1469 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1470 return -1;
1471
1472 ret = b->k.dir ^ key->dir;
1473 if (ret)
1474 return ret;
1475
1476 ret = b->k.type ^ key->type;
1477 if (ret)
1478 return ret;
1479
1480 ret = b->k.family ^ key->family;
1481 if (ret)
1482 return ret;
1483
1484 return b->k.if_id ^ key->if_id;
1485}
1486
1487static const struct rhashtable_params xfrm_pol_inexact_params = {
1488 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head),
1489 .hashfn = xfrm_pol_bin_key,
1490 .obj_hashfn = xfrm_pol_bin_obj,
1491 .obj_cmpfn = xfrm_pol_bin_cmp,
1492 .automatic_shrinking = true,
1493};
1494
1495static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1496 struct xfrm_policy *policy)
1497{
1498 struct xfrm_policy *pol, *delpol = NULL;
1499 struct hlist_node *newpos = NULL;
1500 int i = 0;
1501
1502 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1503 if (pol->type == policy->type &&
1504 pol->if_id == policy->if_id &&
1505 !selector_cmp(&pol->selector, &policy->selector) &&
1506 xfrm_policy_mark_match(policy, pol) &&
1507 xfrm_sec_ctx_match(pol->security, policy->security) &&
1508 !WARN_ON(delpol)) {
1509 delpol = pol;
1510 if (policy->priority > pol->priority)
1511 continue;
1512 } else if (policy->priority >= pol->priority) {
1513 newpos = &pol->bydst_inexact_list;
1514 continue;
1515 }
1516 if (delpol)
1517 break;
1518 }
1519
1520 if (newpos)
1521 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1522 else
1523 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1524
1525 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1526 pol->pos = i;
1527 i++;
1528 }
1529}
1530
1531static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1532 struct xfrm_policy *policy,
1533 bool excl)
1534{
1535 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1536
1537 hlist_for_each_entry(pol, chain, bydst) {
1538 if (pol->type == policy->type &&
1539 pol->if_id == policy->if_id &&
1540 !selector_cmp(&pol->selector, &policy->selector) &&
1541 xfrm_policy_mark_match(policy, pol) &&
1542 xfrm_sec_ctx_match(pol->security, policy->security) &&
1543 !WARN_ON(delpol)) {
1544 if (excl)
1545 return ERR_PTR(-EEXIST);
1546 delpol = pol;
1547 if (policy->priority > pol->priority)
1548 continue;
1549 } else if (policy->priority >= pol->priority) {
1550 newpos = pol;
1551 continue;
1552 }
1553 if (delpol)
1554 break;
1555 }
1556
1557 if (newpos)
1558 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1559 else
1560 hlist_add_head_rcu(&policy->bydst, chain);
1561
1562 return delpol;
1563}
1564
1565int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1566{
1567 struct net *net = xp_net(policy);
1568 struct xfrm_policy *delpol;
1569 struct hlist_head *chain;
1570
1571 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1572 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1573 if (chain)
1574 delpol = xfrm_policy_insert_list(chain, policy, excl);
1575 else
1576 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1577
1578 if (IS_ERR(delpol)) {
1579 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1580 return PTR_ERR(delpol);
1581 }
1582
1583 __xfrm_policy_link(policy, dir);
1584
1585 /* After previous checking, family can either be AF_INET or AF_INET6 */
1586 if (policy->family == AF_INET)
1587 rt_genid_bump_ipv4(net);
1588 else
1589 rt_genid_bump_ipv6(net);
1590
1591 if (delpol) {
1592 xfrm_policy_requeue(delpol, policy);
1593 __xfrm_policy_unlink(delpol, dir);
1594 }
1595 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1596 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1597 policy->curlft.add_time = ktime_get_real_seconds();
1598 policy->curlft.use_time = 0;
1599 if (!mod_timer(&policy->timer, jiffies + HZ))
1600 xfrm_pol_hold(policy);
1601 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1602
1603 if (delpol)
1604 xfrm_policy_kill(delpol);
1605 else if (xfrm_bydst_should_resize(net, dir, NULL))
1606 schedule_work(&net->xfrm.policy_hash_work);
1607
1608 return 0;
1609}
1610EXPORT_SYMBOL(xfrm_policy_insert);
1611
1612static struct xfrm_policy *
1613__xfrm_policy_bysel_ctx(struct hlist_head *chain, u32 mark, u32 if_id,
1614 u8 type, int dir,
1615 struct xfrm_selector *sel,
1616 struct xfrm_sec_ctx *ctx)
1617{
1618 struct xfrm_policy *pol;
1619
1620 if (!chain)
1621 return NULL;
1622
1623 hlist_for_each_entry(pol, chain, bydst) {
1624 if (pol->type == type &&
1625 pol->if_id == if_id &&
1626 (mark & pol->mark.m) == pol->mark.v &&
1627 !selector_cmp(sel, &pol->selector) &&
1628 xfrm_sec_ctx_match(ctx, pol->security))
1629 return pol;
1630 }
1631
1632 return NULL;
1633}
1634
1635struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u32 if_id,
1636 u8 type, int dir,
1637 struct xfrm_selector *sel,
1638 struct xfrm_sec_ctx *ctx, int delete,
1639 int *err)
1640{
1641 struct xfrm_pol_inexact_bin *bin = NULL;
1642 struct xfrm_policy *pol, *ret = NULL;
1643 struct hlist_head *chain;
1644
1645 *err = 0;
1646 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1647 chain = policy_hash_bysel(net, sel, sel->family, dir);
1648 if (!chain) {
1649 struct xfrm_pol_inexact_candidates cand;
1650 int i;
1651
1652 bin = xfrm_policy_inexact_lookup(net, type,
1653 sel->family, dir, if_id);
1654 if (!bin) {
1655 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1656 return NULL;
1657 }
1658
1659 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1660 &sel->saddr,
1661 &sel->daddr)) {
1662 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1663 return NULL;
1664 }
1665
1666 pol = NULL;
1667 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1668 struct xfrm_policy *tmp;
1669
1670 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1671 if_id, type, dir,
1672 sel, ctx);
1673 if (!tmp)
1674 continue;
1675
1676 if (!pol || tmp->pos < pol->pos)
1677 pol = tmp;
1678 }
1679 } else {
1680 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1681 sel, ctx);
1682 }
1683
1684 if (pol) {
1685 xfrm_pol_hold(pol);
1686 if (delete) {
1687 *err = security_xfrm_policy_delete(pol->security);
1688 if (*err) {
1689 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1690 return pol;
1691 }
1692 __xfrm_policy_unlink(pol, dir);
1693 }
1694 ret = pol;
1695 }
1696 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1697
1698 if (ret && delete)
1699 xfrm_policy_kill(ret);
1700 if (bin && delete)
1701 xfrm_policy_inexact_prune_bin(bin);
1702 return ret;
1703}
1704EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1705
1706struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u32 if_id,
1707 u8 type, int dir, u32 id, int delete,
1708 int *err)
1709{
1710 struct xfrm_policy *pol, *ret;
1711 struct hlist_head *chain;
1712
1713 *err = -ENOENT;
1714 if (xfrm_policy_id2dir(id) != dir)
1715 return NULL;
1716
1717 *err = 0;
1718 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1719 chain = net->xfrm.policy_byidx + idx_hash(net, id);
1720 ret = NULL;
1721 hlist_for_each_entry(pol, chain, byidx) {
1722 if (pol->type == type && pol->index == id &&
1723 pol->if_id == if_id &&
1724 (mark & pol->mark.m) == pol->mark.v) {
1725 xfrm_pol_hold(pol);
1726 if (delete) {
1727 *err = security_xfrm_policy_delete(
1728 pol->security);
1729 if (*err) {
1730 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1731 return pol;
1732 }
1733 __xfrm_policy_unlink(pol, dir);
1734 }
1735 ret = pol;
1736 break;
1737 }
1738 }
1739 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1740
1741 if (ret && delete)
1742 xfrm_policy_kill(ret);
1743 return ret;
1744}
1745EXPORT_SYMBOL(xfrm_policy_byid);
1746
1747#ifdef CONFIG_SECURITY_NETWORK_XFRM
1748static inline int
1749xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1750{
1751 struct xfrm_policy *pol;
1752 int err = 0;
1753
1754 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1755 if (pol->walk.dead ||
1756 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1757 pol->type != type)
1758 continue;
1759
1760 err = security_xfrm_policy_delete(pol->security);
1761 if (err) {
1762 xfrm_audit_policy_delete(pol, 0, task_valid);
1763 return err;
1764 }
1765 }
1766 return err;
1767}
1768#else
1769static inline int
1770xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1771{
1772 return 0;
1773}
1774#endif
1775
1776int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1777{
1778 int dir, err = 0, cnt = 0;
1779 struct xfrm_policy *pol;
1780
1781 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1782
1783 err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1784 if (err)
1785 goto out;
1786
1787again:
1788 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1789 dir = xfrm_policy_id2dir(pol->index);
1790 if (pol->walk.dead ||
1791 dir >= XFRM_POLICY_MAX ||
1792 pol->type != type)
1793 continue;
1794
1795 __xfrm_policy_unlink(pol, dir);
1796 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1797 cnt++;
1798 xfrm_audit_policy_delete(pol, 1, task_valid);
1799 xfrm_policy_kill(pol);
1800 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1801 goto again;
1802 }
1803 if (cnt)
1804 __xfrm_policy_inexact_flush(net);
1805 else
1806 err = -ESRCH;
1807out:
1808 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1809 return err;
1810}
1811EXPORT_SYMBOL(xfrm_policy_flush);
1812
1813int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1814 int (*func)(struct xfrm_policy *, int, int, void*),
1815 void *data)
1816{
1817 struct xfrm_policy *pol;
1818 struct xfrm_policy_walk_entry *x;
1819 int error = 0;
1820
1821 if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1822 walk->type != XFRM_POLICY_TYPE_ANY)
1823 return -EINVAL;
1824
1825 if (list_empty(&walk->walk.all) && walk->seq != 0)
1826 return 0;
1827
1828 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1829 if (list_empty(&walk->walk.all))
1830 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1831 else
1832 x = list_first_entry(&walk->walk.all,
1833 struct xfrm_policy_walk_entry, all);
1834
1835 list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1836 if (x->dead)
1837 continue;
1838 pol = container_of(x, struct xfrm_policy, walk);
1839 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1840 walk->type != pol->type)
1841 continue;
1842 error = func(pol, xfrm_policy_id2dir(pol->index),
1843 walk->seq, data);
1844 if (error) {
1845 list_move_tail(&walk->walk.all, &x->all);
1846 goto out;
1847 }
1848 walk->seq++;
1849 }
1850 if (walk->seq == 0) {
1851 error = -ENOENT;
1852 goto out;
1853 }
1854 list_del_init(&walk->walk.all);
1855out:
1856 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1857 return error;
1858}
1859EXPORT_SYMBOL(xfrm_policy_walk);
1860
1861void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1862{
1863 INIT_LIST_HEAD(&walk->walk.all);
1864 walk->walk.dead = 1;
1865 walk->type = type;
1866 walk->seq = 0;
1867}
1868EXPORT_SYMBOL(xfrm_policy_walk_init);
1869
1870void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1871{
1872 if (list_empty(&walk->walk.all))
1873 return;
1874
1875 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1876 list_del(&walk->walk.all);
1877 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1878}
1879EXPORT_SYMBOL(xfrm_policy_walk_done);
1880
1881/*
1882 * Find policy to apply to this flow.
1883 *
1884 * Returns 0 if policy found, else an -errno.
1885 */
1886static int xfrm_policy_match(const struct xfrm_policy *pol,
1887 const struct flowi *fl,
1888 u8 type, u16 family, int dir, u32 if_id)
1889{
1890 const struct xfrm_selector *sel = &pol->selector;
1891 int ret = -ESRCH;
1892 bool match;
1893
1894 if (pol->family != family ||
1895 pol->if_id != if_id ||
1896 (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1897 pol->type != type)
1898 return ret;
1899
1900 match = xfrm_selector_match(sel, fl, family);
1901 if (match)
1902 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1903 dir);
1904 return ret;
1905}
1906
1907static struct xfrm_pol_inexact_node *
1908xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1909 seqcount_t *count,
1910 const xfrm_address_t *addr, u16 family)
1911{
1912 const struct rb_node *parent;
1913 int seq;
1914
1915again:
1916 seq = read_seqcount_begin(count);
1917
1918 parent = rcu_dereference_raw(r->rb_node);
1919 while (parent) {
1920 struct xfrm_pol_inexact_node *node;
1921 int delta;
1922
1923 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1924
1925 delta = xfrm_policy_addr_delta(addr, &node->addr,
1926 node->prefixlen, family);
1927 if (delta < 0) {
1928 parent = rcu_dereference_raw(parent->rb_left);
1929 continue;
1930 } else if (delta > 0) {
1931 parent = rcu_dereference_raw(parent->rb_right);
1932 continue;
1933 }
1934
1935 return node;
1936 }
1937
1938 if (read_seqcount_retry(count, seq))
1939 goto again;
1940
1941 return NULL;
1942}
1943
1944static bool
1945xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1946 struct xfrm_pol_inexact_bin *b,
1947 const xfrm_address_t *saddr,
1948 const xfrm_address_t *daddr)
1949{
1950 struct xfrm_pol_inexact_node *n;
1951 u16 family;
1952
1953 if (!b)
1954 return false;
1955
1956 family = b->k.family;
1957 memset(cand, 0, sizeof(*cand));
1958 cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1959
1960 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1961 family);
1962 if (n) {
1963 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1964 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1965 family);
1966 if (n)
1967 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1968 }
1969
1970 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1971 family);
1972 if (n)
1973 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1974
1975 return true;
1976}
1977
1978static struct xfrm_pol_inexact_bin *
1979xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1980 u8 dir, u32 if_id)
1981{
1982 struct xfrm_pol_inexact_key k = {
1983 .family = family,
1984 .type = type,
1985 .dir = dir,
1986 .if_id = if_id,
1987 };
1988
1989 write_pnet(&k.net, net);
1990
1991 return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1992 xfrm_pol_inexact_params);
1993}
1994
1995static struct xfrm_pol_inexact_bin *
1996xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1997 u8 dir, u32 if_id)
1998{
1999 struct xfrm_pol_inexact_bin *bin;
2000
2001 lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2002
2003 rcu_read_lock();
2004 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2005 rcu_read_unlock();
2006
2007 return bin;
2008}
2009
2010static struct xfrm_policy *
2011__xfrm_policy_eval_candidates(struct hlist_head *chain,
2012 struct xfrm_policy *prefer,
2013 const struct flowi *fl,
2014 u8 type, u16 family, int dir, u32 if_id)
2015{
2016 u32 priority = prefer ? prefer->priority : ~0u;
2017 struct xfrm_policy *pol;
2018
2019 if (!chain)
2020 return NULL;
2021
2022 hlist_for_each_entry_rcu(pol, chain, bydst) {
2023 int err;
2024
2025 if (pol->priority > priority)
2026 break;
2027
2028 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2029 if (err) {
2030 if (err != -ESRCH)
2031 return ERR_PTR(err);
2032
2033 continue;
2034 }
2035
2036 if (prefer) {
2037 /* matches. Is it older than *prefer? */
2038 if (pol->priority == priority &&
2039 prefer->pos < pol->pos)
2040 return prefer;
2041 }
2042
2043 return pol;
2044 }
2045
2046 return NULL;
2047}
2048
2049static struct xfrm_policy *
2050xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2051 struct xfrm_policy *prefer,
2052 const struct flowi *fl,
2053 u8 type, u16 family, int dir, u32 if_id)
2054{
2055 struct xfrm_policy *tmp;
2056 int i;
2057
2058 for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2059 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2060 prefer,
2061 fl, type, family, dir,
2062 if_id);
2063 if (!tmp)
2064 continue;
2065
2066 if (IS_ERR(tmp))
2067 return tmp;
2068 prefer = tmp;
2069 }
2070
2071 return prefer;
2072}
2073
2074static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2075 const struct flowi *fl,
2076 u16 family, u8 dir,
2077 u32 if_id)
2078{
2079 struct xfrm_pol_inexact_candidates cand;
2080 const xfrm_address_t *daddr, *saddr;
2081 struct xfrm_pol_inexact_bin *bin;
2082 struct xfrm_policy *pol, *ret;
2083 struct hlist_head *chain;
2084 unsigned int sequence;
2085 int err;
2086
2087 daddr = xfrm_flowi_daddr(fl, family);
2088 saddr = xfrm_flowi_saddr(fl, family);
2089 if (unlikely(!daddr || !saddr))
2090 return NULL;
2091
2092 rcu_read_lock();
2093 retry:
2094 do {
2095 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2096 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2097 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2098
2099 ret = NULL;
2100 hlist_for_each_entry_rcu(pol, chain, bydst) {
2101 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2102 if (err) {
2103 if (err == -ESRCH)
2104 continue;
2105 else {
2106 ret = ERR_PTR(err);
2107 goto fail;
2108 }
2109 } else {
2110 ret = pol;
2111 break;
2112 }
2113 }
2114 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2115 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2116 daddr))
2117 goto skip_inexact;
2118
2119 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2120 family, dir, if_id);
2121 if (pol) {
2122 ret = pol;
2123 if (IS_ERR(pol))
2124 goto fail;
2125 }
2126
2127skip_inexact:
2128 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2129 goto retry;
2130
2131 if (ret && !xfrm_pol_hold_rcu(ret))
2132 goto retry;
2133fail:
2134 rcu_read_unlock();
2135
2136 return ret;
2137}
2138
2139static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2140 const struct flowi *fl,
2141 u16 family, u8 dir, u32 if_id)
2142{
2143#ifdef CONFIG_XFRM_SUB_POLICY
2144 struct xfrm_policy *pol;
2145
2146 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2147 dir, if_id);
2148 if (pol != NULL)
2149 return pol;
2150#endif
2151 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2152 dir, if_id);
2153}
2154
2155static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2156 const struct flowi *fl,
2157 u16 family, u32 if_id)
2158{
2159 struct xfrm_policy *pol;
2160
2161 rcu_read_lock();
2162 again:
2163 pol = rcu_dereference(sk->sk_policy[dir]);
2164 if (pol != NULL) {
2165 bool match;
2166 int err = 0;
2167
2168 if (pol->family != family) {
2169 pol = NULL;
2170 goto out;
2171 }
2172
2173 match = xfrm_selector_match(&pol->selector, fl, family);
2174 if (match) {
2175 if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2176 pol->if_id != if_id) {
2177 pol = NULL;
2178 goto out;
2179 }
2180 err = security_xfrm_policy_lookup(pol->security,
2181 fl->flowi_secid,
2182 dir);
2183 if (!err) {
2184 if (!xfrm_pol_hold_rcu(pol))
2185 goto again;
2186 } else if (err == -ESRCH) {
2187 pol = NULL;
2188 } else {
2189 pol = ERR_PTR(err);
2190 }
2191 } else
2192 pol = NULL;
2193 }
2194out:
2195 rcu_read_unlock();
2196 return pol;
2197}
2198
2199static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2200{
2201 struct net *net = xp_net(pol);
2202
2203 list_add(&pol->walk.all, &net->xfrm.policy_all);
2204 net->xfrm.policy_count[dir]++;
2205 xfrm_pol_hold(pol);
2206}
2207
2208static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2209 int dir)
2210{
2211 struct net *net = xp_net(pol);
2212
2213 if (list_empty(&pol->walk.all))
2214 return NULL;
2215
2216 /* Socket policies are not hashed. */
2217 if (!hlist_unhashed(&pol->bydst)) {
2218 hlist_del_rcu(&pol->bydst);
2219 hlist_del_init(&pol->bydst_inexact_list);
2220 hlist_del(&pol->byidx);
2221 }
2222
2223 list_del_init(&pol->walk.all);
2224 net->xfrm.policy_count[dir]--;
2225
2226 return pol;
2227}
2228
2229static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2230{
2231 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2232}
2233
2234static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2235{
2236 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2237}
2238
2239int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2240{
2241 struct net *net = xp_net(pol);
2242
2243 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2244 pol = __xfrm_policy_unlink(pol, dir);
2245 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2246 if (pol) {
2247 xfrm_policy_kill(pol);
2248 return 0;
2249 }
2250 return -ENOENT;
2251}
2252EXPORT_SYMBOL(xfrm_policy_delete);
2253
2254int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2255{
2256 struct net *net = sock_net(sk);
2257 struct xfrm_policy *old_pol;
2258
2259#ifdef CONFIG_XFRM_SUB_POLICY
2260 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2261 return -EINVAL;
2262#endif
2263
2264 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2265 old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2266 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2267 if (pol) {
2268 pol->curlft.add_time = ktime_get_real_seconds();
2269 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2270 xfrm_sk_policy_link(pol, dir);
2271 }
2272 rcu_assign_pointer(sk->sk_policy[dir], pol);
2273 if (old_pol) {
2274 if (pol)
2275 xfrm_policy_requeue(old_pol, pol);
2276
2277 /* Unlinking succeeds always. This is the only function
2278 * allowed to delete or replace socket policy.
2279 */
2280 xfrm_sk_policy_unlink(old_pol, dir);
2281 }
2282 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2283
2284 if (old_pol) {
2285 xfrm_policy_kill(old_pol);
2286 }
2287 return 0;
2288}
2289
2290static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2291{
2292 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2293 struct net *net = xp_net(old);
2294
2295 if (newp) {
2296 newp->selector = old->selector;
2297 if (security_xfrm_policy_clone(old->security,
2298 &newp->security)) {
2299 kfree(newp);
2300 return NULL; /* ENOMEM */
2301 }
2302 newp->lft = old->lft;
2303 newp->curlft = old->curlft;
2304 newp->mark = old->mark;
2305 newp->if_id = old->if_id;
2306 newp->action = old->action;
2307 newp->flags = old->flags;
2308 newp->xfrm_nr = old->xfrm_nr;
2309 newp->index = old->index;
2310 newp->type = old->type;
2311 newp->family = old->family;
2312 memcpy(newp->xfrm_vec, old->xfrm_vec,
2313 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2314 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2315 xfrm_sk_policy_link(newp, dir);
2316 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2317 xfrm_pol_put(newp);
2318 }
2319 return newp;
2320}
2321
2322int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2323{
2324 const struct xfrm_policy *p;
2325 struct xfrm_policy *np;
2326 int i, ret = 0;
2327
2328 rcu_read_lock();
2329 for (i = 0; i < 2; i++) {
2330 p = rcu_dereference(osk->sk_policy[i]);
2331 if (p) {
2332 np = clone_policy(p, i);
2333 if (unlikely(!np)) {
2334 ret = -ENOMEM;
2335 break;
2336 }
2337 rcu_assign_pointer(sk->sk_policy[i], np);
2338 }
2339 }
2340 rcu_read_unlock();
2341 return ret;
2342}
2343
2344static int
2345xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2346 xfrm_address_t *remote, unsigned short family, u32 mark)
2347{
2348 int err;
2349 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2350
2351 if (unlikely(afinfo == NULL))
2352 return -EINVAL;
2353 err = afinfo->get_saddr(net, oif, local, remote, mark);
2354 rcu_read_unlock();
2355 return err;
2356}
2357
2358/* Resolve list of templates for the flow, given policy. */
2359
2360static int
2361xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2362 struct xfrm_state **xfrm, unsigned short family)
2363{
2364 struct net *net = xp_net(policy);
2365 int nx;
2366 int i, error;
2367 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2368 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2369 xfrm_address_t tmp;
2370
2371 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2372 struct xfrm_state *x;
2373 xfrm_address_t *remote = daddr;
2374 xfrm_address_t *local = saddr;
2375 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2376
2377 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2378 tmpl->mode == XFRM_MODE_BEET) {
2379 remote = &tmpl->id.daddr;
2380 local = &tmpl->saddr;
2381 if (xfrm_addr_any(local, tmpl->encap_family)) {
2382 error = xfrm_get_saddr(net, fl->flowi_oif,
2383 &tmp, remote,
2384 tmpl->encap_family, 0);
2385 if (error)
2386 goto fail;
2387 local = &tmp;
2388 }
2389 }
2390
2391 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2392 family, policy->if_id);
2393
2394 if (x && x->km.state == XFRM_STATE_VALID) {
2395 xfrm[nx++] = x;
2396 daddr = remote;
2397 saddr = local;
2398 continue;
2399 }
2400 if (x) {
2401 error = (x->km.state == XFRM_STATE_ERROR ?
2402 -EINVAL : -EAGAIN);
2403 xfrm_state_put(x);
2404 } else if (error == -ESRCH) {
2405 error = -EAGAIN;
2406 }
2407
2408 if (!tmpl->optional)
2409 goto fail;
2410 }
2411 return nx;
2412
2413fail:
2414 for (nx--; nx >= 0; nx--)
2415 xfrm_state_put(xfrm[nx]);
2416 return error;
2417}
2418
2419static int
2420xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2421 struct xfrm_state **xfrm, unsigned short family)
2422{
2423 struct xfrm_state *tp[XFRM_MAX_DEPTH];
2424 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2425 int cnx = 0;
2426 int error;
2427 int ret;
2428 int i;
2429
2430 for (i = 0; i < npols; i++) {
2431 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2432 error = -ENOBUFS;
2433 goto fail;
2434 }
2435
2436 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2437 if (ret < 0) {
2438 error = ret;
2439 goto fail;
2440 } else
2441 cnx += ret;
2442 }
2443
2444 /* found states are sorted for outbound processing */
2445 if (npols > 1)
2446 xfrm_state_sort(xfrm, tpp, cnx, family);
2447
2448 return cnx;
2449
2450 fail:
2451 for (cnx--; cnx >= 0; cnx--)
2452 xfrm_state_put(tpp[cnx]);
2453 return error;
2454
2455}
2456
2457static int xfrm_get_tos(const struct flowi *fl, int family)
2458{
2459 if (family == AF_INET)
2460 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2461
2462 return 0;
2463}
2464
2465static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2466{
2467 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2468 struct dst_ops *dst_ops;
2469 struct xfrm_dst *xdst;
2470
2471 if (!afinfo)
2472 return ERR_PTR(-EINVAL);
2473
2474 switch (family) {
2475 case AF_INET:
2476 dst_ops = &net->xfrm.xfrm4_dst_ops;
2477 break;
2478#if IS_ENABLED(CONFIG_IPV6)
2479 case AF_INET6:
2480 dst_ops = &net->xfrm.xfrm6_dst_ops;
2481 break;
2482#endif
2483 default:
2484 BUG();
2485 }
2486 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2487
2488 if (likely(xdst)) {
2489 struct dst_entry *dst = &xdst->u.dst;
2490
2491 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2492 } else
2493 xdst = ERR_PTR(-ENOBUFS);
2494
2495 rcu_read_unlock();
2496
2497 return xdst;
2498}
2499
2500static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2501 int nfheader_len)
2502{
2503 if (dst->ops->family == AF_INET6) {
2504 struct rt6_info *rt = (struct rt6_info *)dst;
2505 path->path_cookie = rt6_get_cookie(rt);
2506 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2507 }
2508}
2509
2510static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2511 const struct flowi *fl)
2512{
2513 const struct xfrm_policy_afinfo *afinfo =
2514 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2515 int err;
2516
2517 if (!afinfo)
2518 return -EINVAL;
2519
2520 err = afinfo->fill_dst(xdst, dev, fl);
2521
2522 rcu_read_unlock();
2523
2524 return err;
2525}
2526
2527
2528/* Allocate chain of dst_entry's, attach known xfrm's, calculate
2529 * all the metrics... Shortly, bundle a bundle.
2530 */
2531
2532static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2533 struct xfrm_state **xfrm,
2534 struct xfrm_dst **bundle,
2535 int nx,
2536 const struct flowi *fl,
2537 struct dst_entry *dst)
2538{
2539 const struct xfrm_state_afinfo *afinfo;
2540 const struct xfrm_mode *inner_mode;
2541 struct net *net = xp_net(policy);
2542 unsigned long now = jiffies;
2543 struct net_device *dev;
2544 struct xfrm_dst *xdst_prev = NULL;
2545 struct xfrm_dst *xdst0 = NULL;
2546 int i = 0;
2547 int err;
2548 int header_len = 0;
2549 int nfheader_len = 0;
2550 int trailer_len = 0;
2551 int tos;
2552 int family = policy->selector.family;
2553 xfrm_address_t saddr, daddr;
2554
2555 xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2556
2557 tos = xfrm_get_tos(fl, family);
2558
2559 dst_hold(dst);
2560
2561 for (; i < nx; i++) {
2562 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2563 struct dst_entry *dst1 = &xdst->u.dst;
2564
2565 err = PTR_ERR(xdst);
2566 if (IS_ERR(xdst)) {
2567 dst_release(dst);
2568 goto put_states;
2569 }
2570
2571 bundle[i] = xdst;
2572 if (!xdst_prev)
2573 xdst0 = xdst;
2574 else
2575 /* Ref count is taken during xfrm_alloc_dst()
2576 * No need to do dst_clone() on dst1
2577 */
2578 xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2579
2580 if (xfrm[i]->sel.family == AF_UNSPEC) {
2581 inner_mode = xfrm_ip2inner_mode(xfrm[i],
2582 xfrm_af2proto(family));
2583 if (!inner_mode) {
2584 err = -EAFNOSUPPORT;
2585 dst_release(dst);
2586 goto put_states;
2587 }
2588 } else
2589 inner_mode = &xfrm[i]->inner_mode;
2590
2591 xdst->route = dst;
2592 dst_copy_metrics(dst1, dst);
2593
2594 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2595 __u32 mark = 0;
2596
2597 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2598 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2599
2600 family = xfrm[i]->props.family;
2601 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2602 &saddr, &daddr, family, mark);
2603 err = PTR_ERR(dst);
2604 if (IS_ERR(dst))
2605 goto put_states;
2606 } else
2607 dst_hold(dst);
2608
2609 dst1->xfrm = xfrm[i];
2610 xdst->xfrm_genid = xfrm[i]->genid;
2611
2612 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2613 dst1->flags |= DST_HOST;
2614 dst1->lastuse = now;
2615
2616 dst1->input = dst_discard;
2617
2618 rcu_read_lock();
2619 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2620 if (likely(afinfo))
2621 dst1->output = afinfo->output;
2622 else
2623 dst1->output = dst_discard_out;
2624 rcu_read_unlock();
2625
2626 xdst_prev = xdst;
2627
2628 header_len += xfrm[i]->props.header_len;
2629 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2630 nfheader_len += xfrm[i]->props.header_len;
2631 trailer_len += xfrm[i]->props.trailer_len;
2632 }
2633
2634 xfrm_dst_set_child(xdst_prev, dst);
2635 xdst0->path = dst;
2636
2637 err = -ENODEV;
2638 dev = dst->dev;
2639 if (!dev)
2640 goto free_dst;
2641
2642 xfrm_init_path(xdst0, dst, nfheader_len);
2643 xfrm_init_pmtu(bundle, nx);
2644
2645 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2646 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2647 err = xfrm_fill_dst(xdst_prev, dev, fl);
2648 if (err)
2649 goto free_dst;
2650
2651 xdst_prev->u.dst.header_len = header_len;
2652 xdst_prev->u.dst.trailer_len = trailer_len;
2653 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2654 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2655 }
2656
2657 return &xdst0->u.dst;
2658
2659put_states:
2660 for (; i < nx; i++)
2661 xfrm_state_put(xfrm[i]);
2662free_dst:
2663 if (xdst0)
2664 dst_release_immediate(&xdst0->u.dst);
2665
2666 return ERR_PTR(err);
2667}
2668
2669static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2670 struct xfrm_policy **pols,
2671 int *num_pols, int *num_xfrms)
2672{
2673 int i;
2674
2675 if (*num_pols == 0 || !pols[0]) {
2676 *num_pols = 0;
2677 *num_xfrms = 0;
2678 return 0;
2679 }
2680 if (IS_ERR(pols[0]))
2681 return PTR_ERR(pols[0]);
2682
2683 *num_xfrms = pols[0]->xfrm_nr;
2684
2685#ifdef CONFIG_XFRM_SUB_POLICY
2686 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2687 pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2688 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2689 XFRM_POLICY_TYPE_MAIN,
2690 fl, family,
2691 XFRM_POLICY_OUT,
2692 pols[0]->if_id);
2693 if (pols[1]) {
2694 if (IS_ERR(pols[1])) {
2695 xfrm_pols_put(pols, *num_pols);
2696 return PTR_ERR(pols[1]);
2697 }
2698 (*num_pols)++;
2699 (*num_xfrms) += pols[1]->xfrm_nr;
2700 }
2701 }
2702#endif
2703 for (i = 0; i < *num_pols; i++) {
2704 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2705 *num_xfrms = -1;
2706 break;
2707 }
2708 }
2709
2710 return 0;
2711
2712}
2713
2714static struct xfrm_dst *
2715xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2716 const struct flowi *fl, u16 family,
2717 struct dst_entry *dst_orig)
2718{
2719 struct net *net = xp_net(pols[0]);
2720 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2721 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2722 struct xfrm_dst *xdst;
2723 struct dst_entry *dst;
2724 int err;
2725
2726 /* Try to instantiate a bundle */
2727 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2728 if (err <= 0) {
2729 if (err == 0)
2730 return NULL;
2731
2732 if (err != -EAGAIN)
2733 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2734 return ERR_PTR(err);
2735 }
2736
2737 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2738 if (IS_ERR(dst)) {
2739 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2740 return ERR_CAST(dst);
2741 }
2742
2743 xdst = (struct xfrm_dst *)dst;
2744 xdst->num_xfrms = err;
2745 xdst->num_pols = num_pols;
2746 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2747 xdst->policy_genid = atomic_read(&pols[0]->genid);
2748
2749 return xdst;
2750}
2751
2752static void xfrm_policy_queue_process(struct timer_list *t)
2753{
2754 struct sk_buff *skb;
2755 struct sock *sk;
2756 struct dst_entry *dst;
2757 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2758 struct net *net = xp_net(pol);
2759 struct xfrm_policy_queue *pq = &pol->polq;
2760 struct flowi fl;
2761 struct sk_buff_head list;
2762
2763 spin_lock(&pq->hold_queue.lock);
2764 skb = skb_peek(&pq->hold_queue);
2765 if (!skb) {
2766 spin_unlock(&pq->hold_queue.lock);
2767 goto out;
2768 }
2769 dst = skb_dst(skb);
2770 sk = skb->sk;
2771 xfrm_decode_session(skb, &fl, dst->ops->family);
2772 spin_unlock(&pq->hold_queue.lock);
2773
2774 dst_hold(xfrm_dst_path(dst));
2775 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2776 if (IS_ERR(dst))
2777 goto purge_queue;
2778
2779 if (dst->flags & DST_XFRM_QUEUE) {
2780 dst_release(dst);
2781
2782 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2783 goto purge_queue;
2784
2785 pq->timeout = pq->timeout << 1;
2786 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2787 xfrm_pol_hold(pol);
2788 goto out;
2789 }
2790
2791 dst_release(dst);
2792
2793 __skb_queue_head_init(&list);
2794
2795 spin_lock(&pq->hold_queue.lock);
2796 pq->timeout = 0;
2797 skb_queue_splice_init(&pq->hold_queue, &list);
2798 spin_unlock(&pq->hold_queue.lock);
2799
2800 while (!skb_queue_empty(&list)) {
2801 skb = __skb_dequeue(&list);
2802
2803 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2804 dst_hold(xfrm_dst_path(skb_dst(skb)));
2805 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2806 if (IS_ERR(dst)) {
2807 kfree_skb(skb);
2808 continue;
2809 }
2810
2811 nf_reset_ct(skb);
2812 skb_dst_drop(skb);
2813 skb_dst_set(skb, dst);
2814
2815 dst_output(net, skb->sk, skb);
2816 }
2817
2818out:
2819 xfrm_pol_put(pol);
2820 return;
2821
2822purge_queue:
2823 pq->timeout = 0;
2824 skb_queue_purge(&pq->hold_queue);
2825 xfrm_pol_put(pol);
2826}
2827
2828static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2829{
2830 unsigned long sched_next;
2831 struct dst_entry *dst = skb_dst(skb);
2832 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2833 struct xfrm_policy *pol = xdst->pols[0];
2834 struct xfrm_policy_queue *pq = &pol->polq;
2835
2836 if (unlikely(skb_fclone_busy(sk, skb))) {
2837 kfree_skb(skb);
2838 return 0;
2839 }
2840
2841 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2842 kfree_skb(skb);
2843 return -EAGAIN;
2844 }
2845
2846 skb_dst_force(skb);
2847
2848 spin_lock_bh(&pq->hold_queue.lock);
2849
2850 if (!pq->timeout)
2851 pq->timeout = XFRM_QUEUE_TMO_MIN;
2852
2853 sched_next = jiffies + pq->timeout;
2854
2855 if (del_timer(&pq->hold_timer)) {
2856 if (time_before(pq->hold_timer.expires, sched_next))
2857 sched_next = pq->hold_timer.expires;
2858 xfrm_pol_put(pol);
2859 }
2860
2861 __skb_queue_tail(&pq->hold_queue, skb);
2862 if (!mod_timer(&pq->hold_timer, sched_next))
2863 xfrm_pol_hold(pol);
2864
2865 spin_unlock_bh(&pq->hold_queue.lock);
2866
2867 return 0;
2868}
2869
2870static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2871 struct xfrm_flo *xflo,
2872 const struct flowi *fl,
2873 int num_xfrms,
2874 u16 family)
2875{
2876 int err;
2877 struct net_device *dev;
2878 struct dst_entry *dst;
2879 struct dst_entry *dst1;
2880 struct xfrm_dst *xdst;
2881
2882 xdst = xfrm_alloc_dst(net, family);
2883 if (IS_ERR(xdst))
2884 return xdst;
2885
2886 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2887 net->xfrm.sysctl_larval_drop ||
2888 num_xfrms <= 0)
2889 return xdst;
2890
2891 dst = xflo->dst_orig;
2892 dst1 = &xdst->u.dst;
2893 dst_hold(dst);
2894 xdst->route = dst;
2895
2896 dst_copy_metrics(dst1, dst);
2897
2898 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2899 dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2900 dst1->lastuse = jiffies;
2901
2902 dst1->input = dst_discard;
2903 dst1->output = xdst_queue_output;
2904
2905 dst_hold(dst);
2906 xfrm_dst_set_child(xdst, dst);
2907 xdst->path = dst;
2908
2909 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2910
2911 err = -ENODEV;
2912 dev = dst->dev;
2913 if (!dev)
2914 goto free_dst;
2915
2916 err = xfrm_fill_dst(xdst, dev, fl);
2917 if (err)
2918 goto free_dst;
2919
2920out:
2921 return xdst;
2922
2923free_dst:
2924 dst_release(dst1);
2925 xdst = ERR_PTR(err);
2926 goto out;
2927}
2928
2929static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2930 const struct flowi *fl,
2931 u16 family, u8 dir,
2932 struct xfrm_flo *xflo, u32 if_id)
2933{
2934 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2935 int num_pols = 0, num_xfrms = 0, err;
2936 struct xfrm_dst *xdst;
2937
2938 /* Resolve policies to use if we couldn't get them from
2939 * previous cache entry */
2940 num_pols = 1;
2941 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2942 err = xfrm_expand_policies(fl, family, pols,
2943 &num_pols, &num_xfrms);
2944 if (err < 0)
2945 goto inc_error;
2946 if (num_pols == 0)
2947 return NULL;
2948 if (num_xfrms <= 0)
2949 goto make_dummy_bundle;
2950
2951 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2952 xflo->dst_orig);
2953 if (IS_ERR(xdst)) {
2954 err = PTR_ERR(xdst);
2955 if (err == -EREMOTE) {
2956 xfrm_pols_put(pols, num_pols);
2957 return NULL;
2958 }
2959
2960 if (err != -EAGAIN)
2961 goto error;
2962 goto make_dummy_bundle;
2963 } else if (xdst == NULL) {
2964 num_xfrms = 0;
2965 goto make_dummy_bundle;
2966 }
2967
2968 return xdst;
2969
2970make_dummy_bundle:
2971 /* We found policies, but there's no bundles to instantiate:
2972 * either because the policy blocks, has no transformations or
2973 * we could not build template (no xfrm_states).*/
2974 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2975 if (IS_ERR(xdst)) {
2976 xfrm_pols_put(pols, num_pols);
2977 return ERR_CAST(xdst);
2978 }
2979 xdst->num_pols = num_pols;
2980 xdst->num_xfrms = num_xfrms;
2981 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2982
2983 return xdst;
2984
2985inc_error:
2986 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2987error:
2988 xfrm_pols_put(pols, num_pols);
2989 return ERR_PTR(err);
2990}
2991
2992static struct dst_entry *make_blackhole(struct net *net, u16 family,
2993 struct dst_entry *dst_orig)
2994{
2995 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2996 struct dst_entry *ret;
2997
2998 if (!afinfo) {
2999 dst_release(dst_orig);
3000 return ERR_PTR(-EINVAL);
3001 } else {
3002 ret = afinfo->blackhole_route(net, dst_orig);
3003 }
3004 rcu_read_unlock();
3005
3006 return ret;
3007}
3008
3009/* Finds/creates a bundle for given flow and if_id
3010 *
3011 * At the moment we eat a raw IP route. Mostly to speed up lookups
3012 * on interfaces with disabled IPsec.
3013 *
3014 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3015 * compatibility
3016 */
3017struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3018 struct dst_entry *dst_orig,
3019 const struct flowi *fl,
3020 const struct sock *sk,
3021 int flags, u32 if_id)
3022{
3023 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3024 struct xfrm_dst *xdst;
3025 struct dst_entry *dst, *route;
3026 u16 family = dst_orig->ops->family;
3027 u8 dir = XFRM_POLICY_OUT;
3028 int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3029
3030 dst = NULL;
3031 xdst = NULL;
3032 route = NULL;
3033
3034 sk = sk_const_to_full_sk(sk);
3035 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3036 num_pols = 1;
3037 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3038 if_id);
3039 err = xfrm_expand_policies(fl, family, pols,
3040 &num_pols, &num_xfrms);
3041 if (err < 0)
3042 goto dropdst;
3043
3044 if (num_pols) {
3045 if (num_xfrms <= 0) {
3046 drop_pols = num_pols;
3047 goto no_transform;
3048 }
3049
3050 xdst = xfrm_resolve_and_create_bundle(
3051 pols, num_pols, fl,
3052 family, dst_orig);
3053
3054 if (IS_ERR(xdst)) {
3055 xfrm_pols_put(pols, num_pols);
3056 err = PTR_ERR(xdst);
3057 if (err == -EREMOTE)
3058 goto nopol;
3059
3060 goto dropdst;
3061 } else if (xdst == NULL) {
3062 num_xfrms = 0;
3063 drop_pols = num_pols;
3064 goto no_transform;
3065 }
3066
3067 route = xdst->route;
3068 }
3069 }
3070
3071 if (xdst == NULL) {
3072 struct xfrm_flo xflo;
3073
3074 xflo.dst_orig = dst_orig;
3075 xflo.flags = flags;
3076
3077 /* To accelerate a bit... */
3078 if ((dst_orig->flags & DST_NOXFRM) ||
3079 !net->xfrm.policy_count[XFRM_POLICY_OUT])
3080 goto nopol;
3081
3082 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3083 if (xdst == NULL)
3084 goto nopol;
3085 if (IS_ERR(xdst)) {
3086 err = PTR_ERR(xdst);
3087 goto dropdst;
3088 }
3089
3090 num_pols = xdst->num_pols;
3091 num_xfrms = xdst->num_xfrms;
3092 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3093 route = xdst->route;
3094 }
3095
3096 dst = &xdst->u.dst;
3097 if (route == NULL && num_xfrms > 0) {
3098 /* The only case when xfrm_bundle_lookup() returns a
3099 * bundle with null route, is when the template could
3100 * not be resolved. It means policies are there, but
3101 * bundle could not be created, since we don't yet
3102 * have the xfrm_state's. We need to wait for KM to
3103 * negotiate new SA's or bail out with error.*/
3104 if (net->xfrm.sysctl_larval_drop) {
3105 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3106 err = -EREMOTE;
3107 goto error;
3108 }
3109
3110 err = -EAGAIN;
3111
3112 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3113 goto error;
3114 }
3115
3116no_transform:
3117 if (num_pols == 0)
3118 goto nopol;
3119
3120 if ((flags & XFRM_LOOKUP_ICMP) &&
3121 !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3122 err = -ENOENT;
3123 goto error;
3124 }
3125
3126 for (i = 0; i < num_pols; i++)
3127 pols[i]->curlft.use_time = ktime_get_real_seconds();
3128
3129 if (num_xfrms < 0) {
3130 /* Prohibit the flow */
3131 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3132 err = -EPERM;
3133 goto error;
3134 } else if (num_xfrms > 0) {
3135 /* Flow transformed */
3136 dst_release(dst_orig);
3137 } else {
3138 /* Flow passes untransformed */
3139 dst_release(dst);
3140 dst = dst_orig;
3141 }
3142ok:
3143 xfrm_pols_put(pols, drop_pols);
3144 if (dst && dst->xfrm &&
3145 dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3146 dst->flags |= DST_XFRM_TUNNEL;
3147 return dst;
3148
3149nopol:
3150 if (!(flags & XFRM_LOOKUP_ICMP)) {
3151 dst = dst_orig;
3152 goto ok;
3153 }
3154 err = -ENOENT;
3155error:
3156 dst_release(dst);
3157dropdst:
3158 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3159 dst_release(dst_orig);
3160 xfrm_pols_put(pols, drop_pols);
3161 return ERR_PTR(err);
3162}
3163EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3164
3165/* Main function: finds/creates a bundle for given flow.
3166 *
3167 * At the moment we eat a raw IP route. Mostly to speed up lookups
3168 * on interfaces with disabled IPsec.
3169 */
3170struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3171 const struct flowi *fl, const struct sock *sk,
3172 int flags)
3173{
3174 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3175}
3176EXPORT_SYMBOL(xfrm_lookup);
3177
3178/* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3179 * Otherwise we may send out blackholed packets.
3180 */
3181struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3182 const struct flowi *fl,
3183 const struct sock *sk, int flags)
3184{
3185 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3186 flags | XFRM_LOOKUP_QUEUE |
3187 XFRM_LOOKUP_KEEP_DST_REF);
3188
3189 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
3190 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3191
3192 if (IS_ERR(dst))
3193 dst_release(dst_orig);
3194
3195 return dst;
3196}
3197EXPORT_SYMBOL(xfrm_lookup_route);
3198
3199static inline int
3200xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3201{
3202 struct sec_path *sp = skb_sec_path(skb);
3203 struct xfrm_state *x;
3204
3205 if (!sp || idx < 0 || idx >= sp->len)
3206 return 0;
3207 x = sp->xvec[idx];
3208 if (!x->type->reject)
3209 return 0;
3210 return x->type->reject(x, skb, fl);
3211}
3212
3213/* When skb is transformed back to its "native" form, we have to
3214 * check policy restrictions. At the moment we make this in maximally
3215 * stupid way. Shame on me. :-) Of course, connected sockets must
3216 * have policy cached at them.
3217 */
3218
3219static inline int
3220xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3221 unsigned short family)
3222{
3223 if (xfrm_state_kern(x))
3224 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3225 return x->id.proto == tmpl->id.proto &&
3226 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3227 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3228 x->props.mode == tmpl->mode &&
3229 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3230 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3231 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3232 xfrm_state_addr_cmp(tmpl, x, family));
3233}
3234
3235/*
3236 * 0 or more than 0 is returned when validation is succeeded (either bypass
3237 * because of optional transport mode, or next index of the mathced secpath
3238 * state with the template.
3239 * -1 is returned when no matching template is found.
3240 * Otherwise "-2 - errored_index" is returned.
3241 */
3242static inline int
3243xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3244 unsigned short family)
3245{
3246 int idx = start;
3247
3248 if (tmpl->optional) {
3249 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3250 return start;
3251 } else
3252 start = -1;
3253 for (; idx < sp->len; idx++) {
3254 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3255 return ++idx;
3256 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3257 if (start == -1)
3258 start = -2-idx;
3259 break;
3260 }
3261 }
3262 return start;
3263}
3264
3265static void
3266decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3267{
3268 const struct iphdr *iph = ip_hdr(skb);
3269 int ihl = iph->ihl;
3270 u8 *xprth = skb_network_header(skb) + ihl * 4;
3271 struct flowi4 *fl4 = &fl->u.ip4;
3272 int oif = 0;
3273
3274 if (skb_dst(skb) && skb_dst(skb)->dev)
3275 oif = skb_dst(skb)->dev->ifindex;
3276
3277 memset(fl4, 0, sizeof(struct flowi4));
3278 fl4->flowi4_mark = skb->mark;
3279 fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3280
3281 fl4->flowi4_proto = iph->protocol;
3282 fl4->daddr = reverse ? iph->saddr : iph->daddr;
3283 fl4->saddr = reverse ? iph->daddr : iph->saddr;
3284 fl4->flowi4_tos = iph->tos;
3285
3286 if (!ip_is_fragment(iph)) {
3287 switch (iph->protocol) {
3288 case IPPROTO_UDP:
3289 case IPPROTO_UDPLITE:
3290 case IPPROTO_TCP:
3291 case IPPROTO_SCTP:
3292 case IPPROTO_DCCP:
3293 if (xprth + 4 < skb->data ||
3294 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3295 __be16 *ports;
3296
3297 xprth = skb_network_header(skb) + ihl * 4;
3298 ports = (__be16 *)xprth;
3299
3300 fl4->fl4_sport = ports[!!reverse];
3301 fl4->fl4_dport = ports[!reverse];
3302 }
3303 break;
3304 case IPPROTO_ICMP:
3305 if (xprth + 2 < skb->data ||
3306 pskb_may_pull(skb, xprth + 2 - skb->data)) {
3307 u8 *icmp;
3308
3309 xprth = skb_network_header(skb) + ihl * 4;
3310 icmp = xprth;
3311
3312 fl4->fl4_icmp_type = icmp[0];
3313 fl4->fl4_icmp_code = icmp[1];
3314 }
3315 break;
3316 case IPPROTO_ESP:
3317 if (xprth + 4 < skb->data ||
3318 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3319 __be32 *ehdr;
3320
3321 xprth = skb_network_header(skb) + ihl * 4;
3322 ehdr = (__be32 *)xprth;
3323
3324 fl4->fl4_ipsec_spi = ehdr[0];
3325 }
3326 break;
3327 case IPPROTO_AH:
3328 if (xprth + 8 < skb->data ||
3329 pskb_may_pull(skb, xprth + 8 - skb->data)) {
3330 __be32 *ah_hdr;
3331
3332 xprth = skb_network_header(skb) + ihl * 4;
3333 ah_hdr = (__be32 *)xprth;
3334
3335 fl4->fl4_ipsec_spi = ah_hdr[1];
3336 }
3337 break;
3338 case IPPROTO_COMP:
3339 if (xprth + 4 < skb->data ||
3340 pskb_may_pull(skb, xprth + 4 - skb->data)) {
3341 __be16 *ipcomp_hdr;
3342
3343 xprth = skb_network_header(skb) + ihl * 4;
3344 ipcomp_hdr = (__be16 *)xprth;
3345
3346 fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
3347 }
3348 break;
3349 case IPPROTO_GRE:
3350 if (xprth + 12 < skb->data ||
3351 pskb_may_pull(skb, xprth + 12 - skb->data)) {
3352 __be16 *greflags;
3353 __be32 *gre_hdr;
3354
3355 xprth = skb_network_header(skb) + ihl * 4;
3356 greflags = (__be16 *)xprth;
3357 gre_hdr = (__be32 *)xprth;
3358
3359 if (greflags[0] & GRE_KEY) {
3360 if (greflags[0] & GRE_CSUM)
3361 gre_hdr++;
3362 fl4->fl4_gre_key = gre_hdr[1];
3363 }
3364 }
3365 break;
3366 default:
3367 fl4->fl4_ipsec_spi = 0;
3368 break;
3369 }
3370 }
3371}
3372
3373#if IS_ENABLED(CONFIG_IPV6)
3374static void
3375decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3376{
3377 struct flowi6 *fl6 = &fl->u.ip6;
3378 int onlyproto = 0;
3379 const struct ipv6hdr *hdr = ipv6_hdr(skb);
3380 u32 offset = sizeof(*hdr);
3381 struct ipv6_opt_hdr *exthdr;
3382 const unsigned char *nh = skb_network_header(skb);
3383 u16 nhoff = IP6CB(skb)->nhoff;
3384 int oif = 0;
3385 u8 nexthdr;
3386
3387 if (!nhoff)
3388 nhoff = offsetof(struct ipv6hdr, nexthdr);
3389
3390 nexthdr = nh[nhoff];
3391
3392 if (skb_dst(skb) && skb_dst(skb)->dev)
3393 oif = skb_dst(skb)->dev->ifindex;
3394
3395 memset(fl6, 0, sizeof(struct flowi6));
3396 fl6->flowi6_mark = skb->mark;
3397 fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3398
3399 fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3400 fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3401
3402 while (nh + offset + sizeof(*exthdr) < skb->data ||
3403 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3404 nh = skb_network_header(skb);
3405 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3406
3407 switch (nexthdr) {
3408 case NEXTHDR_FRAGMENT:
3409 onlyproto = 1;
3410 /* fall through */
3411 case NEXTHDR_ROUTING:
3412 case NEXTHDR_HOP:
3413 case NEXTHDR_DEST:
3414 offset += ipv6_optlen(exthdr);
3415 nexthdr = exthdr->nexthdr;
3416 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3417 break;
3418 case IPPROTO_UDP:
3419 case IPPROTO_UDPLITE:
3420 case IPPROTO_TCP:
3421 case IPPROTO_SCTP:
3422 case IPPROTO_DCCP:
3423 if (!onlyproto && (nh + offset + 4 < skb->data ||
3424 pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3425 __be16 *ports;
3426
3427 nh = skb_network_header(skb);
3428 ports = (__be16 *)(nh + offset);
3429 fl6->fl6_sport = ports[!!reverse];
3430 fl6->fl6_dport = ports[!reverse];
3431 }
3432 fl6->flowi6_proto = nexthdr;
3433 return;
3434 case IPPROTO_ICMPV6:
3435 if (!onlyproto && (nh + offset + 2 < skb->data ||
3436 pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3437 u8 *icmp;
3438
3439 nh = skb_network_header(skb);
3440 icmp = (u8 *)(nh + offset);
3441 fl6->fl6_icmp_type = icmp[0];
3442 fl6->fl6_icmp_code = icmp[1];
3443 }
3444 fl6->flowi6_proto = nexthdr;
3445 return;
3446#if IS_ENABLED(CONFIG_IPV6_MIP6)
3447 case IPPROTO_MH:
3448 offset += ipv6_optlen(exthdr);
3449 if (!onlyproto && (nh + offset + 3 < skb->data ||
3450 pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3451 struct ip6_mh *mh;
3452
3453 nh = skb_network_header(skb);
3454 mh = (struct ip6_mh *)(nh + offset);
3455 fl6->fl6_mh_type = mh->ip6mh_type;
3456 }
3457 fl6->flowi6_proto = nexthdr;
3458 return;
3459#endif
3460 /* XXX Why are there these headers? */
3461 case IPPROTO_AH:
3462 case IPPROTO_ESP:
3463 case IPPROTO_COMP:
3464 default:
3465 fl6->fl6_ipsec_spi = 0;
3466 fl6->flowi6_proto = nexthdr;
3467 return;
3468 }
3469 }
3470}
3471#endif
3472
3473int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3474 unsigned int family, int reverse)
3475{
3476 switch (family) {
3477 case AF_INET:
3478 decode_session4(skb, fl, reverse);
3479 break;
3480#if IS_ENABLED(CONFIG_IPV6)
3481 case AF_INET6:
3482 decode_session6(skb, fl, reverse);
3483 break;
3484#endif
3485 default:
3486 return -EAFNOSUPPORT;
3487 }
3488
3489 return security_xfrm_decode_session(skb, &fl->flowi_secid);
3490}
3491EXPORT_SYMBOL(__xfrm_decode_session);
3492
3493static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3494{
3495 for (; k < sp->len; k++) {
3496 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3497 *idxp = k;
3498 return 1;
3499 }
3500 }
3501
3502 return 0;
3503}
3504
3505int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3506 unsigned short family)
3507{
3508 struct net *net = dev_net(skb->dev);
3509 struct xfrm_policy *pol;
3510 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3511 int npols = 0;
3512 int xfrm_nr;
3513 int pi;
3514 int reverse;
3515 struct flowi fl;
3516 int xerr_idx = -1;
3517 const struct xfrm_if_cb *ifcb;
3518 struct sec_path *sp;
3519 struct xfrm_if *xi;
3520 u32 if_id = 0;
3521
3522 rcu_read_lock();
3523 ifcb = xfrm_if_get_cb();
3524
3525 if (ifcb) {
3526 xi = ifcb->decode_session(skb, family);
3527 if (xi) {
3528 if_id = xi->p.if_id;
3529 net = xi->net;
3530 }
3531 }
3532 rcu_read_unlock();
3533
3534 reverse = dir & ~XFRM_POLICY_MASK;
3535 dir &= XFRM_POLICY_MASK;
3536
3537 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3538 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3539 return 0;
3540 }
3541
3542 nf_nat_decode_session(skb, &fl, family);
3543
3544 /* First, check used SA against their selectors. */
3545 sp = skb_sec_path(skb);
3546 if (sp) {
3547 int i;
3548
3549 for (i = sp->len - 1; i >= 0; i--) {
3550 struct xfrm_state *x = sp->xvec[i];
3551 if (!xfrm_selector_match(&x->sel, &fl, family)) {
3552 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3553 return 0;
3554 }
3555 }
3556 }
3557
3558 pol = NULL;
3559 sk = sk_to_full_sk(sk);
3560 if (sk && sk->sk_policy[dir]) {
3561 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3562 if (IS_ERR(pol)) {
3563 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3564 return 0;
3565 }
3566 }
3567
3568 if (!pol)
3569 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3570
3571 if (IS_ERR(pol)) {
3572 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3573 return 0;
3574 }
3575
3576 if (!pol) {
3577 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3578 xfrm_secpath_reject(xerr_idx, skb, &fl);
3579 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3580 return 0;
3581 }
3582 return 1;
3583 }
3584
3585 pol->curlft.use_time = ktime_get_real_seconds();
3586
3587 pols[0] = pol;
3588 npols++;
3589#ifdef CONFIG_XFRM_SUB_POLICY
3590 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3591 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3592 &fl, family,
3593 XFRM_POLICY_IN, if_id);
3594 if (pols[1]) {
3595 if (IS_ERR(pols[1])) {
3596 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3597 return 0;
3598 }
3599 pols[1]->curlft.use_time = ktime_get_real_seconds();
3600 npols++;
3601 }
3602 }
3603#endif
3604
3605 if (pol->action == XFRM_POLICY_ALLOW) {
3606 static struct sec_path dummy;
3607 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3608 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3609 struct xfrm_tmpl **tpp = tp;
3610 int ti = 0;
3611 int i, k;
3612
3613 sp = skb_sec_path(skb);
3614 if (!sp)
3615 sp = &dummy;
3616
3617 for (pi = 0; pi < npols; pi++) {
3618 if (pols[pi] != pol &&
3619 pols[pi]->action != XFRM_POLICY_ALLOW) {
3620 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3621 goto reject;
3622 }
3623 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3624 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3625 goto reject_error;
3626 }
3627 for (i = 0; i < pols[pi]->xfrm_nr; i++)
3628 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3629 }
3630 xfrm_nr = ti;
3631 if (npols > 1) {
3632 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3633 tpp = stp;
3634 }
3635
3636 /* For each tunnel xfrm, find the first matching tmpl.
3637 * For each tmpl before that, find corresponding xfrm.
3638 * Order is _important_. Later we will implement
3639 * some barriers, but at the moment barriers
3640 * are implied between each two transformations.
3641 */
3642 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3643 k = xfrm_policy_ok(tpp[i], sp, k, family);
3644 if (k < 0) {
3645 if (k < -1)
3646 /* "-2 - errored_index" returned */
3647 xerr_idx = -(2+k);
3648 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3649 goto reject;
3650 }
3651 }
3652
3653 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3654 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3655 goto reject;
3656 }
3657
3658 xfrm_pols_put(pols, npols);
3659 return 1;
3660 }
3661 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3662
3663reject:
3664 xfrm_secpath_reject(xerr_idx, skb, &fl);
3665reject_error:
3666 xfrm_pols_put(pols, npols);
3667 return 0;
3668}
3669EXPORT_SYMBOL(__xfrm_policy_check);
3670
3671int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3672{
3673 struct net *net = dev_net(skb->dev);
3674 struct flowi fl;
3675 struct dst_entry *dst;
3676 int res = 1;
3677
3678 if (xfrm_decode_session(skb, &fl, family) < 0) {
3679 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3680 return 0;
3681 }
3682
3683 skb_dst_force(skb);
3684 if (!skb_dst(skb)) {
3685 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3686 return 0;
3687 }
3688
3689 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3690 if (IS_ERR(dst)) {
3691 res = 0;
3692 dst = NULL;
3693 }
3694 skb_dst_set(skb, dst);
3695 return res;
3696}
3697EXPORT_SYMBOL(__xfrm_route_forward);
3698
3699/* Optimize later using cookies and generation ids. */
3700
3701static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3702{
3703 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3704 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3705 * get validated by dst_ops->check on every use. We do this
3706 * because when a normal route referenced by an XFRM dst is
3707 * obsoleted we do not go looking around for all parent
3708 * referencing XFRM dsts so that we can invalidate them. It
3709 * is just too much work. Instead we make the checks here on
3710 * every use. For example:
3711 *
3712 * XFRM dst A --> IPv4 dst X
3713 *
3714 * X is the "xdst->route" of A (X is also the "dst->path" of A
3715 * in this example). If X is marked obsolete, "A" will not
3716 * notice. That's what we are validating here via the
3717 * stale_bundle() check.
3718 *
3719 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3720 * be marked on it.
3721 * This will force stale_bundle() to fail on any xdst bundle with
3722 * this dst linked in it.
3723 */
3724 if (dst->obsolete < 0 && !stale_bundle(dst))
3725 return dst;
3726
3727 return NULL;
3728}
3729
3730static int stale_bundle(struct dst_entry *dst)
3731{
3732 return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3733}
3734
3735void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3736{
3737 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3738 dst->dev = dev_net(dev)->loopback_dev;
3739 dev_hold(dst->dev);
3740 dev_put(dev);
3741 }
3742}
3743EXPORT_SYMBOL(xfrm_dst_ifdown);
3744
3745static void xfrm_link_failure(struct sk_buff *skb)
3746{
3747 /* Impossible. Such dst must be popped before reaches point of failure. */
3748}
3749
3750static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3751{
3752 if (dst) {
3753 if (dst->obsolete) {
3754 dst_release(dst);
3755 dst = NULL;
3756 }
3757 }
3758 return dst;
3759}
3760
3761static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3762{
3763 while (nr--) {
3764 struct xfrm_dst *xdst = bundle[nr];
3765 u32 pmtu, route_mtu_cached;
3766 struct dst_entry *dst;
3767
3768 dst = &xdst->u.dst;
3769 pmtu = dst_mtu(xfrm_dst_child(dst));
3770 xdst->child_mtu_cached = pmtu;
3771
3772 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3773
3774 route_mtu_cached = dst_mtu(xdst->route);
3775 xdst->route_mtu_cached = route_mtu_cached;
3776
3777 if (pmtu > route_mtu_cached)
3778 pmtu = route_mtu_cached;
3779
3780 dst_metric_set(dst, RTAX_MTU, pmtu);
3781 }
3782}
3783
3784/* Check that the bundle accepts the flow and its components are
3785 * still valid.
3786 */
3787
3788static int xfrm_bundle_ok(struct xfrm_dst *first)
3789{
3790 struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3791 struct dst_entry *dst = &first->u.dst;
3792 struct xfrm_dst *xdst;
3793 int start_from, nr;
3794 u32 mtu;
3795
3796 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3797 (dst->dev && !netif_running(dst->dev)))
3798 return 0;
3799
3800 if (dst->flags & DST_XFRM_QUEUE)
3801 return 1;
3802
3803 start_from = nr = 0;
3804 do {
3805 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3806
3807 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3808 return 0;
3809 if (xdst->xfrm_genid != dst->xfrm->genid)
3810 return 0;
3811 if (xdst->num_pols > 0 &&
3812 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3813 return 0;
3814
3815 bundle[nr++] = xdst;
3816
3817 mtu = dst_mtu(xfrm_dst_child(dst));
3818 if (xdst->child_mtu_cached != mtu) {
3819 start_from = nr;
3820 xdst->child_mtu_cached = mtu;
3821 }
3822
3823 if (!dst_check(xdst->route, xdst->route_cookie))
3824 return 0;
3825 mtu = dst_mtu(xdst->route);
3826 if (xdst->route_mtu_cached != mtu) {
3827 start_from = nr;
3828 xdst->route_mtu_cached = mtu;
3829 }
3830
3831 dst = xfrm_dst_child(dst);
3832 } while (dst->xfrm);
3833
3834 if (likely(!start_from))
3835 return 1;
3836
3837 xdst = bundle[start_from - 1];
3838 mtu = xdst->child_mtu_cached;
3839 while (start_from--) {
3840 dst = &xdst->u.dst;
3841
3842 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3843 if (mtu > xdst->route_mtu_cached)
3844 mtu = xdst->route_mtu_cached;
3845 dst_metric_set(dst, RTAX_MTU, mtu);
3846 if (!start_from)
3847 break;
3848
3849 xdst = bundle[start_from - 1];
3850 xdst->child_mtu_cached = mtu;
3851 }
3852
3853 return 1;
3854}
3855
3856static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3857{
3858 return dst_metric_advmss(xfrm_dst_path(dst));
3859}
3860
3861static unsigned int xfrm_mtu(const struct dst_entry *dst)
3862{
3863 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3864
3865 return mtu ? : dst_mtu(xfrm_dst_path(dst));
3866}
3867
3868static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3869 const void *daddr)
3870{
3871 while (dst->xfrm) {
3872 const struct xfrm_state *xfrm = dst->xfrm;
3873
3874 dst = xfrm_dst_child(dst);
3875
3876 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3877 continue;
3878 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3879 daddr = xfrm->coaddr;
3880 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3881 daddr = &xfrm->id.daddr;
3882 }
3883 return daddr;
3884}
3885
3886static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3887 struct sk_buff *skb,
3888 const void *daddr)
3889{
3890 const struct dst_entry *path = xfrm_dst_path(dst);
3891
3892 if (!skb)
3893 daddr = xfrm_get_dst_nexthop(dst, daddr);
3894 return path->ops->neigh_lookup(path, skb, daddr);
3895}
3896
3897static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3898{
3899 const struct dst_entry *path = xfrm_dst_path(dst);
3900
3901 daddr = xfrm_get_dst_nexthop(dst, daddr);
3902 path->ops->confirm_neigh(path, daddr);
3903}
3904
3905int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3906{
3907 int err = 0;
3908
3909 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3910 return -EAFNOSUPPORT;
3911
3912 spin_lock(&xfrm_policy_afinfo_lock);
3913 if (unlikely(xfrm_policy_afinfo[family] != NULL))
3914 err = -EEXIST;
3915 else {
3916 struct dst_ops *dst_ops = afinfo->dst_ops;
3917 if (likely(dst_ops->kmem_cachep == NULL))
3918 dst_ops->kmem_cachep = xfrm_dst_cache;
3919 if (likely(dst_ops->check == NULL))
3920 dst_ops->check = xfrm_dst_check;
3921 if (likely(dst_ops->default_advmss == NULL))
3922 dst_ops->default_advmss = xfrm_default_advmss;
3923 if (likely(dst_ops->mtu == NULL))
3924 dst_ops->mtu = xfrm_mtu;
3925 if (likely(dst_ops->negative_advice == NULL))
3926 dst_ops->negative_advice = xfrm_negative_advice;
3927 if (likely(dst_ops->link_failure == NULL))
3928 dst_ops->link_failure = xfrm_link_failure;
3929 if (likely(dst_ops->neigh_lookup == NULL))
3930 dst_ops->neigh_lookup = xfrm_neigh_lookup;
3931 if (likely(!dst_ops->confirm_neigh))
3932 dst_ops->confirm_neigh = xfrm_confirm_neigh;
3933 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3934 }
3935 spin_unlock(&xfrm_policy_afinfo_lock);
3936
3937 return err;
3938}
3939EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3940
3941void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3942{
3943 struct dst_ops *dst_ops = afinfo->dst_ops;
3944 int i;
3945
3946 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3947 if (xfrm_policy_afinfo[i] != afinfo)
3948 continue;
3949 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3950 break;
3951 }
3952
3953 synchronize_rcu();
3954
3955 dst_ops->kmem_cachep = NULL;
3956 dst_ops->check = NULL;
3957 dst_ops->negative_advice = NULL;
3958 dst_ops->link_failure = NULL;
3959}
3960EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3961
3962void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3963{
3964 spin_lock(&xfrm_if_cb_lock);
3965 rcu_assign_pointer(xfrm_if_cb, ifcb);
3966 spin_unlock(&xfrm_if_cb_lock);
3967}
3968EXPORT_SYMBOL(xfrm_if_register_cb);
3969
3970void xfrm_if_unregister_cb(void)
3971{
3972 RCU_INIT_POINTER(xfrm_if_cb, NULL);
3973 synchronize_rcu();
3974}
3975EXPORT_SYMBOL(xfrm_if_unregister_cb);
3976
3977#ifdef CONFIG_XFRM_STATISTICS
3978static int __net_init xfrm_statistics_init(struct net *net)
3979{
3980 int rv;
3981 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3982 if (!net->mib.xfrm_statistics)
3983 return -ENOMEM;
3984 rv = xfrm_proc_init(net);
3985 if (rv < 0)
3986 free_percpu(net->mib.xfrm_statistics);
3987 return rv;
3988}
3989
3990static void xfrm_statistics_fini(struct net *net)
3991{
3992 xfrm_proc_fini(net);
3993 free_percpu(net->mib.xfrm_statistics);
3994}
3995#else
3996static int __net_init xfrm_statistics_init(struct net *net)
3997{
3998 return 0;
3999}
4000
4001static void xfrm_statistics_fini(struct net *net)
4002{
4003}
4004#endif
4005
4006static int __net_init xfrm_policy_init(struct net *net)
4007{
4008 unsigned int hmask, sz;
4009 int dir, err;
4010
4011 if (net_eq(net, &init_net)) {
4012 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4013 sizeof(struct xfrm_dst),
4014 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4015 NULL);
4016 err = rhashtable_init(&xfrm_policy_inexact_table,
4017 &xfrm_pol_inexact_params);
4018 BUG_ON(err);
4019 }
4020
4021 hmask = 8 - 1;
4022 sz = (hmask+1) * sizeof(struct hlist_head);
4023
4024 net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4025 if (!net->xfrm.policy_byidx)
4026 goto out_byidx;
4027 net->xfrm.policy_idx_hmask = hmask;
4028
4029 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4030 struct xfrm_policy_hash *htab;
4031
4032 net->xfrm.policy_count[dir] = 0;
4033 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4034 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4035
4036 htab = &net->xfrm.policy_bydst[dir];
4037 htab->table = xfrm_hash_alloc(sz);
4038 if (!htab->table)
4039 goto out_bydst;
4040 htab->hmask = hmask;
4041 htab->dbits4 = 32;
4042 htab->sbits4 = 32;
4043 htab->dbits6 = 128;
4044 htab->sbits6 = 128;
4045 }
4046 net->xfrm.policy_hthresh.lbits4 = 32;
4047 net->xfrm.policy_hthresh.rbits4 = 32;
4048 net->xfrm.policy_hthresh.lbits6 = 128;
4049 net->xfrm.policy_hthresh.rbits6 = 128;
4050
4051 seqlock_init(&net->xfrm.policy_hthresh.lock);
4052
4053 INIT_LIST_HEAD(&net->xfrm.policy_all);
4054 INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4055 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4056 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4057 return 0;
4058
4059out_bydst:
4060 for (dir--; dir >= 0; dir--) {
4061 struct xfrm_policy_hash *htab;
4062
4063 htab = &net->xfrm.policy_bydst[dir];
4064 xfrm_hash_free(htab->table, sz);
4065 }
4066 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4067out_byidx:
4068 return -ENOMEM;
4069}
4070
4071static void xfrm_policy_fini(struct net *net)
4072{
4073 struct xfrm_pol_inexact_bin *b, *t;
4074 unsigned int sz;
4075 int dir;
4076
4077 flush_work(&net->xfrm.policy_hash_work);
4078#ifdef CONFIG_XFRM_SUB_POLICY
4079 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4080#endif
4081 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4082
4083 WARN_ON(!list_empty(&net->xfrm.policy_all));
4084
4085 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4086 struct xfrm_policy_hash *htab;
4087
4088 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4089
4090 htab = &net->xfrm.policy_bydst[dir];
4091 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4092 WARN_ON(!hlist_empty(htab->table));
4093 xfrm_hash_free(htab->table, sz);
4094 }
4095
4096 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4097 WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4098 xfrm_hash_free(net->xfrm.policy_byidx, sz);
4099
4100 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4101 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4102 __xfrm_policy_inexact_prune_bin(b, true);
4103 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4104}
4105
4106static int __net_init xfrm_net_init(struct net *net)
4107{
4108 int rv;
4109
4110 /* Initialize the per-net locks here */
4111 spin_lock_init(&net->xfrm.xfrm_state_lock);
4112 spin_lock_init(&net->xfrm.xfrm_policy_lock);
4113 mutex_init(&net->xfrm.xfrm_cfg_mutex);
4114
4115 rv = xfrm_statistics_init(net);
4116 if (rv < 0)
4117 goto out_statistics;
4118 rv = xfrm_state_init(net);
4119 if (rv < 0)
4120 goto out_state;
4121 rv = xfrm_policy_init(net);
4122 if (rv < 0)
4123 goto out_policy;
4124 rv = xfrm_sysctl_init(net);
4125 if (rv < 0)
4126 goto out_sysctl;
4127
4128 return 0;
4129
4130out_sysctl:
4131 xfrm_policy_fini(net);
4132out_policy:
4133 xfrm_state_fini(net);
4134out_state:
4135 xfrm_statistics_fini(net);
4136out_statistics:
4137 return rv;
4138}
4139
4140static void __net_exit xfrm_net_exit(struct net *net)
4141{
4142 xfrm_sysctl_fini(net);
4143 xfrm_policy_fini(net);
4144 xfrm_state_fini(net);
4145 xfrm_statistics_fini(net);
4146}
4147
4148static struct pernet_operations __net_initdata xfrm_net_ops = {
4149 .init = xfrm_net_init,
4150 .exit = xfrm_net_exit,
4151};
4152
4153void __init xfrm_init(void)
4154{
4155 register_pernet_subsys(&xfrm_net_ops);
4156 xfrm_dev_init();
4157 seqcount_init(&xfrm_policy_hash_generation);
4158 xfrm_input_init();
4159
4160 RCU_INIT_POINTER(xfrm_if_cb, NULL);
4161 synchronize_rcu();
4162}
4163
4164#ifdef CONFIG_AUDITSYSCALL
4165static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4166 struct audit_buffer *audit_buf)
4167{
4168 struct xfrm_sec_ctx *ctx = xp->security;
4169 struct xfrm_selector *sel = &xp->selector;
4170
4171 if (ctx)
4172 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4173 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4174
4175 switch (sel->family) {
4176 case AF_INET:
4177 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4178 if (sel->prefixlen_s != 32)
4179 audit_log_format(audit_buf, " src_prefixlen=%d",
4180 sel->prefixlen_s);
4181 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4182 if (sel->prefixlen_d != 32)
4183 audit_log_format(audit_buf, " dst_prefixlen=%d",
4184 sel->prefixlen_d);
4185 break;
4186 case AF_INET6:
4187 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4188 if (sel->prefixlen_s != 128)
4189 audit_log_format(audit_buf, " src_prefixlen=%d",
4190 sel->prefixlen_s);
4191 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4192 if (sel->prefixlen_d != 128)
4193 audit_log_format(audit_buf, " dst_prefixlen=%d",
4194 sel->prefixlen_d);
4195 break;
4196 }
4197}
4198
4199void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4200{
4201 struct audit_buffer *audit_buf;
4202
4203 audit_buf = xfrm_audit_start("SPD-add");
4204 if (audit_buf == NULL)
4205 return;
4206 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4207 audit_log_format(audit_buf, " res=%u", result);
4208 xfrm_audit_common_policyinfo(xp, audit_buf);
4209 audit_log_end(audit_buf);
4210}
4211EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4212
4213void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4214 bool task_valid)
4215{
4216 struct audit_buffer *audit_buf;
4217
4218 audit_buf = xfrm_audit_start("SPD-delete");
4219 if (audit_buf == NULL)
4220 return;
4221 xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4222 audit_log_format(audit_buf, " res=%u", result);
4223 xfrm_audit_common_policyinfo(xp, audit_buf);
4224 audit_log_end(audit_buf);
4225}
4226EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4227#endif
4228
4229#ifdef CONFIG_XFRM_MIGRATE
4230static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4231 const struct xfrm_selector *sel_tgt)
4232{
4233 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4234 if (sel_tgt->family == sel_cmp->family &&
4235 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4236 sel_cmp->family) &&
4237 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4238 sel_cmp->family) &&
4239 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4240 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4241 return true;
4242 }
4243 } else {
4244 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4245 return true;
4246 }
4247 }
4248 return false;
4249}
4250
4251static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4252 u8 dir, u8 type, struct net *net)
4253{
4254 struct xfrm_policy *pol, *ret = NULL;
4255 struct hlist_head *chain;
4256 u32 priority = ~0U;
4257
4258 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4259 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4260 hlist_for_each_entry(pol, chain, bydst) {
4261 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4262 pol->type == type) {
4263 ret = pol;
4264 priority = ret->priority;
4265 break;
4266 }
4267 }
4268 chain = &net->xfrm.policy_inexact[dir];
4269 hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4270 if ((pol->priority >= priority) && ret)
4271 break;
4272
4273 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4274 pol->type == type) {
4275 ret = pol;
4276 break;
4277 }
4278 }
4279
4280 xfrm_pol_hold(ret);
4281
4282 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4283
4284 return ret;
4285}
4286
4287static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4288{
4289 int match = 0;
4290
4291 if (t->mode == m->mode && t->id.proto == m->proto &&
4292 (m->reqid == 0 || t->reqid == m->reqid)) {
4293 switch (t->mode) {
4294 case XFRM_MODE_TUNNEL:
4295 case XFRM_MODE_BEET:
4296 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4297 m->old_family) &&
4298 xfrm_addr_equal(&t->saddr, &m->old_saddr,
4299 m->old_family)) {
4300 match = 1;
4301 }
4302 break;
4303 case XFRM_MODE_TRANSPORT:
4304 /* in case of transport mode, template does not store
4305 any IP addresses, hence we just compare mode and
4306 protocol */
4307 match = 1;
4308 break;
4309 default:
4310 break;
4311 }
4312 }
4313 return match;
4314}
4315
4316/* update endpoint address(es) of template(s) */
4317static int xfrm_policy_migrate(struct xfrm_policy *pol,
4318 struct xfrm_migrate *m, int num_migrate)
4319{
4320 struct xfrm_migrate *mp;
4321 int i, j, n = 0;
4322
4323 write_lock_bh(&pol->lock);
4324 if (unlikely(pol->walk.dead)) {
4325 /* target policy has been deleted */
4326 write_unlock_bh(&pol->lock);
4327 return -ENOENT;
4328 }
4329
4330 for (i = 0; i < pol->xfrm_nr; i++) {
4331 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4332 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4333 continue;
4334 n++;
4335 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4336 pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4337 continue;
4338 /* update endpoints */
4339 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4340 sizeof(pol->xfrm_vec[i].id.daddr));
4341 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4342 sizeof(pol->xfrm_vec[i].saddr));
4343 pol->xfrm_vec[i].encap_family = mp->new_family;
4344 /* flush bundles */
4345 atomic_inc(&pol->genid);
4346 }
4347 }
4348
4349 write_unlock_bh(&pol->lock);
4350
4351 if (!n)
4352 return -ENODATA;
4353
4354 return 0;
4355}
4356
4357static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4358{
4359 int i, j;
4360
4361 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4362 return -EINVAL;
4363
4364 for (i = 0; i < num_migrate; i++) {
4365 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4366 xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4367 return -EINVAL;
4368
4369 /* check if there is any duplicated entry */
4370 for (j = i + 1; j < num_migrate; j++) {
4371 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4372 sizeof(m[i].old_daddr)) &&
4373 !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4374 sizeof(m[i].old_saddr)) &&
4375 m[i].proto == m[j].proto &&
4376 m[i].mode == m[j].mode &&
4377 m[i].reqid == m[j].reqid &&
4378 m[i].old_family == m[j].old_family)
4379 return -EINVAL;
4380 }
4381 }
4382
4383 return 0;
4384}
4385
4386int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4387 struct xfrm_migrate *m, int num_migrate,
4388 struct xfrm_kmaddress *k, struct net *net,
4389 struct xfrm_encap_tmpl *encap)
4390{
4391 int i, err, nx_cur = 0, nx_new = 0;
4392 struct xfrm_policy *pol = NULL;
4393 struct xfrm_state *x, *xc;
4394 struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4395 struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4396 struct xfrm_migrate *mp;
4397
4398 /* Stage 0 - sanity checks */
4399 if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4400 goto out;
4401
4402 if (dir >= XFRM_POLICY_MAX) {
4403 err = -EINVAL;
4404 goto out;
4405 }
4406
4407 /* Stage 1 - find policy */
4408 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4409 err = -ENOENT;
4410 goto out;
4411 }
4412
4413 /* Stage 2 - find and update state(s) */
4414 for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4415 if ((x = xfrm_migrate_state_find(mp, net))) {
4416 x_cur[nx_cur] = x;
4417 nx_cur++;
4418 xc = xfrm_state_migrate(x, mp, encap);
4419 if (xc) {
4420 x_new[nx_new] = xc;
4421 nx_new++;
4422 } else {
4423 err = -ENODATA;
4424 goto restore_state;
4425 }
4426 }
4427 }
4428
4429 /* Stage 3 - update policy */
4430 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4431 goto restore_state;
4432
4433 /* Stage 4 - delete old state(s) */
4434 if (nx_cur) {
4435 xfrm_states_put(x_cur, nx_cur);
4436 xfrm_states_delete(x_cur, nx_cur);
4437 }
4438
4439 /* Stage 5 - announce */
4440 km_migrate(sel, dir, type, m, num_migrate, k, encap);
4441
4442 xfrm_pol_put(pol);
4443
4444 return 0;
4445out:
4446 return err;
4447
4448restore_state:
4449 if (pol)
4450 xfrm_pol_put(pol);
4451 if (nx_cur)
4452 xfrm_states_put(x_cur, nx_cur);
4453 if (nx_new)
4454 xfrm_states_delete(x_new, nx_new);
4455
4456 return err;
4457}
4458EXPORT_SYMBOL(xfrm_migrate);
4459#endif