xref: /kernel/linux/linux-5.10/net/xfrm/xfrm_policy.c (revision 8c2ecf20)
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/inet_ecn.h>
35#include <net/xfrm.h>
36#include <net/ip.h>
37#include <net/gre.h>
38#if IS_ENABLED(CONFIG_IPV6_MIP6)
39#include <net/mip6.h>
40#endif
41#ifdef CONFIG_XFRM_STATISTICS
42#include <net/snmp.h>
43#endif
44#ifdef CONFIG_XFRM_ESPINTCP
45#include <net/espintcp.h>
46#endif
47
48#include "xfrm_hash.h"
49
50#define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
51#define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
52#define XFRM_MAX_QUEUE_LEN	100
53
54struct xfrm_flo {
55	struct dst_entry *dst_orig;
56	u8 flags;
57};
58
59/* prefixes smaller than this are stored in lists, not trees. */
60#define INEXACT_PREFIXLEN_IPV4	16
61#define INEXACT_PREFIXLEN_IPV6	48
62
63struct xfrm_pol_inexact_node {
64	struct rb_node node;
65	union {
66		xfrm_address_t addr;
67		struct rcu_head rcu;
68	};
69	u8 prefixlen;
70
71	struct rb_root root;
72
73	/* the policies matching this node, can be empty list */
74	struct hlist_head hhead;
75};
76
77/* xfrm inexact policy search tree:
78 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
79 *  |
80 * +---- root_d: sorted by daddr:prefix
81 * |                 |
82 * |        xfrm_pol_inexact_node
83 * |                 |
84 * |                 +- root: sorted by saddr/prefix
85 * |                 |              |
86 * |                 |         xfrm_pol_inexact_node
87 * |                 |              |
88 * |                 |              + root: unused
89 * |                 |              |
90 * |                 |              + hhead: saddr:daddr policies
91 * |                 |
92 * |                 +- coarse policies and all any:daddr policies
93 * |
94 * +---- root_s: sorted by saddr:prefix
95 * |                 |
96 * |        xfrm_pol_inexact_node
97 * |                 |
98 * |                 + root: unused
99 * |                 |
100 * |                 + hhead: saddr:any policies
101 * |
102 * +---- coarse policies and all any:any policies
103 *
104 * Lookups return four candidate lists:
105 * 1. any:any list from top-level xfrm_pol_inexact_bin
106 * 2. any:daddr list from daddr tree
107 * 3. saddr:daddr list from 2nd level daddr tree
108 * 4. saddr:any list from saddr tree
109 *
110 * This result set then needs to be searched for the policy with
111 * the lowest priority.  If two results have same prio, youngest one wins.
112 */
113
114struct xfrm_pol_inexact_key {
115	possible_net_t net;
116	u32 if_id;
117	u16 family;
118	u8 dir, type;
119};
120
121struct xfrm_pol_inexact_bin {
122	struct xfrm_pol_inexact_key k;
123	struct rhash_head head;
124	/* list containing '*:*' policies */
125	struct hlist_head hhead;
126
127	seqcount_spinlock_t count;
128	/* tree sorted by daddr/prefix */
129	struct rb_root root_d;
130
131	/* tree sorted by saddr/prefix */
132	struct rb_root root_s;
133
134	/* slow path below */
135	struct list_head inexact_bins;
136	struct rcu_head rcu;
137};
138
139enum xfrm_pol_inexact_candidate_type {
140	XFRM_POL_CAND_BOTH,
141	XFRM_POL_CAND_SADDR,
142	XFRM_POL_CAND_DADDR,
143	XFRM_POL_CAND_ANY,
144
145	XFRM_POL_CAND_MAX,
146};
147
148struct xfrm_pol_inexact_candidates {
149	struct hlist_head *res[XFRM_POL_CAND_MAX];
150};
151
152static DEFINE_SPINLOCK(xfrm_if_cb_lock);
153static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
154
155static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
156static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
157						__read_mostly;
158
159static struct kmem_cache *xfrm_dst_cache __ro_after_init;
160
161static struct rhashtable xfrm_policy_inexact_table;
162static const struct rhashtable_params xfrm_pol_inexact_params;
163
164static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
165static int stale_bundle(struct dst_entry *dst);
166static int xfrm_bundle_ok(struct xfrm_dst *xdst);
167static void xfrm_policy_queue_process(struct timer_list *t);
168
169static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
170static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
171						int dir);
172
173static struct xfrm_pol_inexact_bin *
174xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
175			   u32 if_id);
176
177static struct xfrm_pol_inexact_bin *
178xfrm_policy_inexact_lookup_rcu(struct net *net,
179			       u8 type, u16 family, u8 dir, u32 if_id);
180static struct xfrm_policy *
181xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
182			bool excl);
183static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
184					    struct xfrm_policy *policy);
185
186static bool
187xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
188				    struct xfrm_pol_inexact_bin *b,
189				    const xfrm_address_t *saddr,
190				    const xfrm_address_t *daddr);
191
192static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
193{
194	return refcount_inc_not_zero(&policy->refcnt);
195}
196
197static inline bool
198__xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
199{
200	const struct flowi4 *fl4 = &fl->u.ip4;
201
202	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
203		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
204		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
205		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
206		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
207		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
208}
209
210static inline bool
211__xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
212{
213	const struct flowi6 *fl6 = &fl->u.ip6;
214
215	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
216		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
217		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
218		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
219		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
220		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
221}
222
223bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
224			 unsigned short family)
225{
226	switch (family) {
227	case AF_INET:
228		return __xfrm4_selector_match(sel, fl);
229	case AF_INET6:
230		return __xfrm6_selector_match(sel, fl);
231	}
232	return false;
233}
234
235static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
236{
237	const struct xfrm_policy_afinfo *afinfo;
238
239	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
240		return NULL;
241	rcu_read_lock();
242	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
243	if (unlikely(!afinfo))
244		rcu_read_unlock();
245	return afinfo;
246}
247
248/* Called with rcu_read_lock(). */
249static const struct xfrm_if_cb *xfrm_if_get_cb(void)
250{
251	return rcu_dereference(xfrm_if_cb);
252}
253
254struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
255				    const xfrm_address_t *saddr,
256				    const xfrm_address_t *daddr,
257				    int family, u32 mark)
258{
259	const struct xfrm_policy_afinfo *afinfo;
260	struct dst_entry *dst;
261
262	afinfo = xfrm_policy_get_afinfo(family);
263	if (unlikely(afinfo == NULL))
264		return ERR_PTR(-EAFNOSUPPORT);
265
266	dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
267
268	rcu_read_unlock();
269
270	return dst;
271}
272EXPORT_SYMBOL(__xfrm_dst_lookup);
273
274static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
275						int tos, int oif,
276						xfrm_address_t *prev_saddr,
277						xfrm_address_t *prev_daddr,
278						int family, u32 mark)
279{
280	struct net *net = xs_net(x);
281	xfrm_address_t *saddr = &x->props.saddr;
282	xfrm_address_t *daddr = &x->id.daddr;
283	struct dst_entry *dst;
284
285	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
286		saddr = x->coaddr;
287		daddr = prev_daddr;
288	}
289	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
290		saddr = prev_saddr;
291		daddr = x->coaddr;
292	}
293
294	dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
295
296	if (!IS_ERR(dst)) {
297		if (prev_saddr != saddr)
298			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
299		if (prev_daddr != daddr)
300			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
301	}
302
303	return dst;
304}
305
306static inline unsigned long make_jiffies(long secs)
307{
308	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
309		return MAX_SCHEDULE_TIMEOUT-1;
310	else
311		return secs*HZ;
312}
313
314static void xfrm_policy_timer(struct timer_list *t)
315{
316	struct xfrm_policy *xp = from_timer(xp, t, timer);
317	time64_t now = ktime_get_real_seconds();
318	time64_t next = TIME64_MAX;
319	int warn = 0;
320	int dir;
321
322	read_lock(&xp->lock);
323
324	if (unlikely(xp->walk.dead))
325		goto out;
326
327	dir = xfrm_policy_id2dir(xp->index);
328
329	if (xp->lft.hard_add_expires_seconds) {
330		time64_t tmo = xp->lft.hard_add_expires_seconds +
331			xp->curlft.add_time - now;
332		if (tmo <= 0)
333			goto expired;
334		if (tmo < next)
335			next = tmo;
336	}
337	if (xp->lft.hard_use_expires_seconds) {
338		time64_t tmo = xp->lft.hard_use_expires_seconds +
339			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
340		if (tmo <= 0)
341			goto expired;
342		if (tmo < next)
343			next = tmo;
344	}
345	if (xp->lft.soft_add_expires_seconds) {
346		time64_t tmo = xp->lft.soft_add_expires_seconds +
347			xp->curlft.add_time - now;
348		if (tmo <= 0) {
349			warn = 1;
350			tmo = XFRM_KM_TIMEOUT;
351		}
352		if (tmo < next)
353			next = tmo;
354	}
355	if (xp->lft.soft_use_expires_seconds) {
356		time64_t tmo = xp->lft.soft_use_expires_seconds +
357			(xp->curlft.use_time ? : xp->curlft.add_time) - now;
358		if (tmo <= 0) {
359			warn = 1;
360			tmo = XFRM_KM_TIMEOUT;
361		}
362		if (tmo < next)
363			next = tmo;
364	}
365
366	if (warn)
367		km_policy_expired(xp, dir, 0, 0);
368	if (next != TIME64_MAX &&
369	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
370		xfrm_pol_hold(xp);
371
372out:
373	read_unlock(&xp->lock);
374	xfrm_pol_put(xp);
375	return;
376
377expired:
378	read_unlock(&xp->lock);
379	if (!xfrm_policy_delete(xp, dir))
380		km_policy_expired(xp, dir, 1, 0);
381	xfrm_pol_put(xp);
382}
383
384/* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
385 * SPD calls.
386 */
387
388struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
389{
390	struct xfrm_policy *policy;
391
392	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
393
394	if (policy) {
395		write_pnet(&policy->xp_net, net);
396		INIT_LIST_HEAD(&policy->walk.all);
397		INIT_HLIST_NODE(&policy->bydst_inexact_list);
398		INIT_HLIST_NODE(&policy->bydst);
399		INIT_HLIST_NODE(&policy->byidx);
400		rwlock_init(&policy->lock);
401		refcount_set(&policy->refcnt, 1);
402		skb_queue_head_init(&policy->polq.hold_queue);
403		timer_setup(&policy->timer, xfrm_policy_timer, 0);
404		timer_setup(&policy->polq.hold_timer,
405			    xfrm_policy_queue_process, 0);
406	}
407	return policy;
408}
409EXPORT_SYMBOL(xfrm_policy_alloc);
410
411static void xfrm_policy_destroy_rcu(struct rcu_head *head)
412{
413	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
414
415	security_xfrm_policy_free(policy->security);
416	kfree(policy);
417}
418
419/* Destroy xfrm_policy: descendant resources must be released to this moment. */
420
421void xfrm_policy_destroy(struct xfrm_policy *policy)
422{
423	BUG_ON(!policy->walk.dead);
424
425	if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
426		BUG();
427
428	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
429}
430EXPORT_SYMBOL(xfrm_policy_destroy);
431
432/* Rule must be locked. Release descendant resources, announce
433 * entry dead. The rule must be unlinked from lists to the moment.
434 */
435
436static void xfrm_policy_kill(struct xfrm_policy *policy)
437{
438	write_lock_bh(&policy->lock);
439	policy->walk.dead = 1;
440	write_unlock_bh(&policy->lock);
441
442	atomic_inc(&policy->genid);
443
444	if (del_timer(&policy->polq.hold_timer))
445		xfrm_pol_put(policy);
446	skb_queue_purge(&policy->polq.hold_queue);
447
448	if (del_timer(&policy->timer))
449		xfrm_pol_put(policy);
450
451	xfrm_pol_put(policy);
452}
453
454static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
455
456static inline unsigned int idx_hash(struct net *net, u32 index)
457{
458	return __idx_hash(index, net->xfrm.policy_idx_hmask);
459}
460
461/* calculate policy hash thresholds */
462static void __get_hash_thresh(struct net *net,
463			      unsigned short family, int dir,
464			      u8 *dbits, u8 *sbits)
465{
466	switch (family) {
467	case AF_INET:
468		*dbits = net->xfrm.policy_bydst[dir].dbits4;
469		*sbits = net->xfrm.policy_bydst[dir].sbits4;
470		break;
471
472	case AF_INET6:
473		*dbits = net->xfrm.policy_bydst[dir].dbits6;
474		*sbits = net->xfrm.policy_bydst[dir].sbits6;
475		break;
476
477	default:
478		*dbits = 0;
479		*sbits = 0;
480	}
481}
482
483static struct hlist_head *policy_hash_bysel(struct net *net,
484					    const struct xfrm_selector *sel,
485					    unsigned short family, int dir)
486{
487	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
488	unsigned int hash;
489	u8 dbits;
490	u8 sbits;
491
492	__get_hash_thresh(net, family, dir, &dbits, &sbits);
493	hash = __sel_hash(sel, family, hmask, dbits, sbits);
494
495	if (hash == hmask + 1)
496		return NULL;
497
498	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
499		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
500}
501
502static struct hlist_head *policy_hash_direct(struct net *net,
503					     const xfrm_address_t *daddr,
504					     const xfrm_address_t *saddr,
505					     unsigned short family, int dir)
506{
507	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
508	unsigned int hash;
509	u8 dbits;
510	u8 sbits;
511
512	__get_hash_thresh(net, family, dir, &dbits, &sbits);
513	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
514
515	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
516		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
517}
518
519static void xfrm_dst_hash_transfer(struct net *net,
520				   struct hlist_head *list,
521				   struct hlist_head *ndsttable,
522				   unsigned int nhashmask,
523				   int dir)
524{
525	struct hlist_node *tmp, *entry0 = NULL;
526	struct xfrm_policy *pol;
527	unsigned int h0 = 0;
528	u8 dbits;
529	u8 sbits;
530
531redo:
532	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
533		unsigned int h;
534
535		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
536		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
537				pol->family, nhashmask, dbits, sbits);
538		if (!entry0) {
539			hlist_del_rcu(&pol->bydst);
540			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
541			h0 = h;
542		} else {
543			if (h != h0)
544				continue;
545			hlist_del_rcu(&pol->bydst);
546			hlist_add_behind_rcu(&pol->bydst, entry0);
547		}
548		entry0 = &pol->bydst;
549	}
550	if (!hlist_empty(list)) {
551		entry0 = NULL;
552		goto redo;
553	}
554}
555
556static void xfrm_idx_hash_transfer(struct hlist_head *list,
557				   struct hlist_head *nidxtable,
558				   unsigned int nhashmask)
559{
560	struct hlist_node *tmp;
561	struct xfrm_policy *pol;
562
563	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
564		unsigned int h;
565
566		h = __idx_hash(pol->index, nhashmask);
567		hlist_add_head(&pol->byidx, nidxtable+h);
568	}
569}
570
571static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
572{
573	return ((old_hmask + 1) << 1) - 1;
574}
575
576static void xfrm_bydst_resize(struct net *net, int dir)
577{
578	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
579	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
580	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
581	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
582	struct hlist_head *odst;
583	int i;
584
585	if (!ndst)
586		return;
587
588	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
589	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
590
591	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
592				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
593
594	for (i = hmask; i >= 0; i--)
595		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
596
597	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
598	net->xfrm.policy_bydst[dir].hmask = nhashmask;
599
600	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
601	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
602
603	synchronize_rcu();
604
605	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
606}
607
608static void xfrm_byidx_resize(struct net *net, int total)
609{
610	unsigned int hmask = net->xfrm.policy_idx_hmask;
611	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
612	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
613	struct hlist_head *oidx = net->xfrm.policy_byidx;
614	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
615	int i;
616
617	if (!nidx)
618		return;
619
620	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
621
622	for (i = hmask; i >= 0; i--)
623		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
624
625	net->xfrm.policy_byidx = nidx;
626	net->xfrm.policy_idx_hmask = nhashmask;
627
628	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
629
630	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
631}
632
633static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
634{
635	unsigned int cnt = net->xfrm.policy_count[dir];
636	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
637
638	if (total)
639		*total += cnt;
640
641	if ((hmask + 1) < xfrm_policy_hashmax &&
642	    cnt > hmask)
643		return 1;
644
645	return 0;
646}
647
648static inline int xfrm_byidx_should_resize(struct net *net, int total)
649{
650	unsigned int hmask = net->xfrm.policy_idx_hmask;
651
652	if ((hmask + 1) < xfrm_policy_hashmax &&
653	    total > hmask)
654		return 1;
655
656	return 0;
657}
658
659void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
660{
661	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
662	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
663	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
664	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
665	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
666	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
667	si->spdhcnt = net->xfrm.policy_idx_hmask;
668	si->spdhmcnt = xfrm_policy_hashmax;
669}
670EXPORT_SYMBOL(xfrm_spd_getinfo);
671
672static DEFINE_MUTEX(hash_resize_mutex);
673static void xfrm_hash_resize(struct work_struct *work)
674{
675	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
676	int dir, total;
677
678	mutex_lock(&hash_resize_mutex);
679
680	total = 0;
681	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
682		if (xfrm_bydst_should_resize(net, dir, &total))
683			xfrm_bydst_resize(net, dir);
684	}
685	if (xfrm_byidx_should_resize(net, total))
686		xfrm_byidx_resize(net, total);
687
688	mutex_unlock(&hash_resize_mutex);
689}
690
691/* Make sure *pol can be inserted into fastbin.
692 * Useful to check that later insert requests will be sucessful
693 * (provided xfrm_policy_lock is held throughout).
694 */
695static struct xfrm_pol_inexact_bin *
696xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
697{
698	struct xfrm_pol_inexact_bin *bin, *prev;
699	struct xfrm_pol_inexact_key k = {
700		.family = pol->family,
701		.type = pol->type,
702		.dir = dir,
703		.if_id = pol->if_id,
704	};
705	struct net *net = xp_net(pol);
706
707	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
708
709	write_pnet(&k.net, net);
710	bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
711				     xfrm_pol_inexact_params);
712	if (bin)
713		return bin;
714
715	bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
716	if (!bin)
717		return NULL;
718
719	bin->k = k;
720	INIT_HLIST_HEAD(&bin->hhead);
721	bin->root_d = RB_ROOT;
722	bin->root_s = RB_ROOT;
723	seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
724
725	prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
726						&bin->k, &bin->head,
727						xfrm_pol_inexact_params);
728	if (!prev) {
729		list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
730		return bin;
731	}
732
733	kfree(bin);
734
735	return IS_ERR(prev) ? NULL : prev;
736}
737
738static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
739					       int family, u8 prefixlen)
740{
741	if (xfrm_addr_any(addr, family))
742		return true;
743
744	if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
745		return true;
746
747	if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
748		return true;
749
750	return false;
751}
752
753static bool
754xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
755{
756	const xfrm_address_t *addr;
757	bool saddr_any, daddr_any;
758	u8 prefixlen;
759
760	addr = &policy->selector.saddr;
761	prefixlen = policy->selector.prefixlen_s;
762
763	saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
764						       policy->family,
765						       prefixlen);
766	addr = &policy->selector.daddr;
767	prefixlen = policy->selector.prefixlen_d;
768	daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
769						       policy->family,
770						       prefixlen);
771	return saddr_any && daddr_any;
772}
773
774static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
775				       const xfrm_address_t *addr, u8 prefixlen)
776{
777	node->addr = *addr;
778	node->prefixlen = prefixlen;
779}
780
781static struct xfrm_pol_inexact_node *
782xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
783{
784	struct xfrm_pol_inexact_node *node;
785
786	node = kzalloc(sizeof(*node), GFP_ATOMIC);
787	if (node)
788		xfrm_pol_inexact_node_init(node, addr, prefixlen);
789
790	return node;
791}
792
793static int xfrm_policy_addr_delta(const xfrm_address_t *a,
794				  const xfrm_address_t *b,
795				  u8 prefixlen, u16 family)
796{
797	u32 ma, mb, mask;
798	unsigned int pdw, pbi;
799	int delta = 0;
800
801	switch (family) {
802	case AF_INET:
803		if (prefixlen == 0)
804			return 0;
805		mask = ~0U << (32 - prefixlen);
806		ma = ntohl(a->a4) & mask;
807		mb = ntohl(b->a4) & mask;
808		if (ma < mb)
809			delta = -1;
810		else if (ma > mb)
811			delta = 1;
812		break;
813	case AF_INET6:
814		pdw = prefixlen >> 5;
815		pbi = prefixlen & 0x1f;
816
817		if (pdw) {
818			delta = memcmp(a->a6, b->a6, pdw << 2);
819			if (delta)
820				return delta;
821		}
822		if (pbi) {
823			mask = ~0U << (32 - pbi);
824			ma = ntohl(a->a6[pdw]) & mask;
825			mb = ntohl(b->a6[pdw]) & mask;
826			if (ma < mb)
827				delta = -1;
828			else if (ma > mb)
829				delta = 1;
830		}
831		break;
832	default:
833		break;
834	}
835
836	return delta;
837}
838
839static void xfrm_policy_inexact_list_reinsert(struct net *net,
840					      struct xfrm_pol_inexact_node *n,
841					      u16 family)
842{
843	unsigned int matched_s, matched_d;
844	struct xfrm_policy *policy, *p;
845
846	matched_s = 0;
847	matched_d = 0;
848
849	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
850		struct hlist_node *newpos = NULL;
851		bool matches_s, matches_d;
852
853		if (!policy->bydst_reinsert)
854			continue;
855
856		WARN_ON_ONCE(policy->family != family);
857
858		policy->bydst_reinsert = false;
859		hlist_for_each_entry(p, &n->hhead, bydst) {
860			if (policy->priority > p->priority)
861				newpos = &p->bydst;
862			else if (policy->priority == p->priority &&
863				 policy->pos > p->pos)
864				newpos = &p->bydst;
865			else
866				break;
867		}
868
869		if (newpos)
870			hlist_add_behind_rcu(&policy->bydst, newpos);
871		else
872			hlist_add_head_rcu(&policy->bydst, &n->hhead);
873
874		/* paranoia checks follow.
875		 * Check that the reinserted policy matches at least
876		 * saddr or daddr for current node prefix.
877		 *
878		 * Matching both is fine, matching saddr in one policy
879		 * (but not daddr) and then matching only daddr in another
880		 * is a bug.
881		 */
882		matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
883						   &n->addr,
884						   n->prefixlen,
885						   family) == 0;
886		matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
887						   &n->addr,
888						   n->prefixlen,
889						   family) == 0;
890		if (matches_s && matches_d)
891			continue;
892
893		WARN_ON_ONCE(!matches_s && !matches_d);
894		if (matches_s)
895			matched_s++;
896		if (matches_d)
897			matched_d++;
898		WARN_ON_ONCE(matched_s && matched_d);
899	}
900}
901
902static void xfrm_policy_inexact_node_reinsert(struct net *net,
903					      struct xfrm_pol_inexact_node *n,
904					      struct rb_root *new,
905					      u16 family)
906{
907	struct xfrm_pol_inexact_node *node;
908	struct rb_node **p, *parent;
909
910	/* we should not have another subtree here */
911	WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
912restart:
913	parent = NULL;
914	p = &new->rb_node;
915	while (*p) {
916		u8 prefixlen;
917		int delta;
918
919		parent = *p;
920		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
921
922		prefixlen = min(node->prefixlen, n->prefixlen);
923
924		delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
925					       prefixlen, family);
926		if (delta < 0) {
927			p = &parent->rb_left;
928		} else if (delta > 0) {
929			p = &parent->rb_right;
930		} else {
931			bool same_prefixlen = node->prefixlen == n->prefixlen;
932			struct xfrm_policy *tmp;
933
934			hlist_for_each_entry(tmp, &n->hhead, bydst) {
935				tmp->bydst_reinsert = true;
936				hlist_del_rcu(&tmp->bydst);
937			}
938
939			node->prefixlen = prefixlen;
940
941			xfrm_policy_inexact_list_reinsert(net, node, family);
942
943			if (same_prefixlen) {
944				kfree_rcu(n, rcu);
945				return;
946			}
947
948			rb_erase(*p, new);
949			kfree_rcu(n, rcu);
950			n = node;
951			goto restart;
952		}
953	}
954
955	rb_link_node_rcu(&n->node, parent, p);
956	rb_insert_color(&n->node, new);
957}
958
959/* merge nodes v and n */
960static void xfrm_policy_inexact_node_merge(struct net *net,
961					   struct xfrm_pol_inexact_node *v,
962					   struct xfrm_pol_inexact_node *n,
963					   u16 family)
964{
965	struct xfrm_pol_inexact_node *node;
966	struct xfrm_policy *tmp;
967	struct rb_node *rnode;
968
969	/* To-be-merged node v has a subtree.
970	 *
971	 * Dismantle it and insert its nodes to n->root.
972	 */
973	while ((rnode = rb_first(&v->root)) != NULL) {
974		node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
975		rb_erase(&node->node, &v->root);
976		xfrm_policy_inexact_node_reinsert(net, node, &n->root,
977						  family);
978	}
979
980	hlist_for_each_entry(tmp, &v->hhead, bydst) {
981		tmp->bydst_reinsert = true;
982		hlist_del_rcu(&tmp->bydst);
983	}
984
985	xfrm_policy_inexact_list_reinsert(net, n, family);
986}
987
988static struct xfrm_pol_inexact_node *
989xfrm_policy_inexact_insert_node(struct net *net,
990				struct rb_root *root,
991				xfrm_address_t *addr,
992				u16 family, u8 prefixlen, u8 dir)
993{
994	struct xfrm_pol_inexact_node *cached = NULL;
995	struct rb_node **p, *parent = NULL;
996	struct xfrm_pol_inexact_node *node;
997
998	p = &root->rb_node;
999	while (*p) {
1000		int delta;
1001
1002		parent = *p;
1003		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1004
1005		delta = xfrm_policy_addr_delta(addr, &node->addr,
1006					       node->prefixlen,
1007					       family);
1008		if (delta == 0 && prefixlen >= node->prefixlen) {
1009			WARN_ON_ONCE(cached); /* ipsec policies got lost */
1010			return node;
1011		}
1012
1013		if (delta < 0)
1014			p = &parent->rb_left;
1015		else
1016			p = &parent->rb_right;
1017
1018		if (prefixlen < node->prefixlen) {
1019			delta = xfrm_policy_addr_delta(addr, &node->addr,
1020						       prefixlen,
1021						       family);
1022			if (delta)
1023				continue;
1024
1025			/* This node is a subnet of the new prefix. It needs
1026			 * to be removed and re-inserted with the smaller
1027			 * prefix and all nodes that are now also covered
1028			 * by the reduced prefixlen.
1029			 */
1030			rb_erase(&node->node, root);
1031
1032			if (!cached) {
1033				xfrm_pol_inexact_node_init(node, addr,
1034							   prefixlen);
1035				cached = node;
1036			} else {
1037				/* This node also falls within the new
1038				 * prefixlen. Merge the to-be-reinserted
1039				 * node and this one.
1040				 */
1041				xfrm_policy_inexact_node_merge(net, node,
1042							       cached, family);
1043				kfree_rcu(node, rcu);
1044			}
1045
1046			/* restart */
1047			p = &root->rb_node;
1048			parent = NULL;
1049		}
1050	}
1051
1052	node = cached;
1053	if (!node) {
1054		node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1055		if (!node)
1056			return NULL;
1057	}
1058
1059	rb_link_node_rcu(&node->node, parent, p);
1060	rb_insert_color(&node->node, root);
1061
1062	return node;
1063}
1064
1065static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1066{
1067	struct xfrm_pol_inexact_node *node;
1068	struct rb_node *rn = rb_first(r);
1069
1070	while (rn) {
1071		node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1072
1073		xfrm_policy_inexact_gc_tree(&node->root, rm);
1074		rn = rb_next(rn);
1075
1076		if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1077			WARN_ON_ONCE(rm);
1078			continue;
1079		}
1080
1081		rb_erase(&node->node, r);
1082		kfree_rcu(node, rcu);
1083	}
1084}
1085
1086static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1087{
1088	write_seqcount_begin(&b->count);
1089	xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1090	xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1091	write_seqcount_end(&b->count);
1092
1093	if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1094	    !hlist_empty(&b->hhead)) {
1095		WARN_ON_ONCE(net_exit);
1096		return;
1097	}
1098
1099	if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1100				   xfrm_pol_inexact_params) == 0) {
1101		list_del(&b->inexact_bins);
1102		kfree_rcu(b, rcu);
1103	}
1104}
1105
1106static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1107{
1108	struct net *net = read_pnet(&b->k.net);
1109
1110	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1111	__xfrm_policy_inexact_prune_bin(b, false);
1112	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1113}
1114
1115static void __xfrm_policy_inexact_flush(struct net *net)
1116{
1117	struct xfrm_pol_inexact_bin *bin, *t;
1118
1119	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1120
1121	list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1122		__xfrm_policy_inexact_prune_bin(bin, false);
1123}
1124
1125static struct hlist_head *
1126xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1127				struct xfrm_policy *policy, u8 dir)
1128{
1129	struct xfrm_pol_inexact_node *n;
1130	struct net *net;
1131
1132	net = xp_net(policy);
1133	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1134
1135	if (xfrm_policy_inexact_insert_use_any_list(policy))
1136		return &bin->hhead;
1137
1138	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1139					       policy->family,
1140					       policy->selector.prefixlen_d)) {
1141		write_seqcount_begin(&bin->count);
1142		n = xfrm_policy_inexact_insert_node(net,
1143						    &bin->root_s,
1144						    &policy->selector.saddr,
1145						    policy->family,
1146						    policy->selector.prefixlen_s,
1147						    dir);
1148		write_seqcount_end(&bin->count);
1149		if (!n)
1150			return NULL;
1151
1152		return &n->hhead;
1153	}
1154
1155	/* daddr is fixed */
1156	write_seqcount_begin(&bin->count);
1157	n = xfrm_policy_inexact_insert_node(net,
1158					    &bin->root_d,
1159					    &policy->selector.daddr,
1160					    policy->family,
1161					    policy->selector.prefixlen_d, dir);
1162	write_seqcount_end(&bin->count);
1163	if (!n)
1164		return NULL;
1165
1166	/* saddr is wildcard */
1167	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1168					       policy->family,
1169					       policy->selector.prefixlen_s))
1170		return &n->hhead;
1171
1172	write_seqcount_begin(&bin->count);
1173	n = xfrm_policy_inexact_insert_node(net,
1174					    &n->root,
1175					    &policy->selector.saddr,
1176					    policy->family,
1177					    policy->selector.prefixlen_s, dir);
1178	write_seqcount_end(&bin->count);
1179	if (!n)
1180		return NULL;
1181
1182	return &n->hhead;
1183}
1184
1185static struct xfrm_policy *
1186xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1187{
1188	struct xfrm_pol_inexact_bin *bin;
1189	struct xfrm_policy *delpol;
1190	struct hlist_head *chain;
1191	struct net *net;
1192
1193	bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1194	if (!bin)
1195		return ERR_PTR(-ENOMEM);
1196
1197	net = xp_net(policy);
1198	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1199
1200	chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1201	if (!chain) {
1202		__xfrm_policy_inexact_prune_bin(bin, false);
1203		return ERR_PTR(-ENOMEM);
1204	}
1205
1206	delpol = xfrm_policy_insert_list(chain, policy, excl);
1207	if (delpol && excl) {
1208		__xfrm_policy_inexact_prune_bin(bin, false);
1209		return ERR_PTR(-EEXIST);
1210	}
1211
1212	chain = &net->xfrm.policy_inexact[dir];
1213	xfrm_policy_insert_inexact_list(chain, policy);
1214
1215	if (delpol)
1216		__xfrm_policy_inexact_prune_bin(bin, false);
1217
1218	return delpol;
1219}
1220
1221static void xfrm_hash_rebuild(struct work_struct *work)
1222{
1223	struct net *net = container_of(work, struct net,
1224				       xfrm.policy_hthresh.work);
1225	unsigned int hmask;
1226	struct xfrm_policy *pol;
1227	struct xfrm_policy *policy;
1228	struct hlist_head *chain;
1229	struct hlist_head *odst;
1230	struct hlist_node *newpos;
1231	int i;
1232	int dir;
1233	unsigned seq;
1234	u8 lbits4, rbits4, lbits6, rbits6;
1235
1236	mutex_lock(&hash_resize_mutex);
1237
1238	/* read selector prefixlen thresholds */
1239	do {
1240		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1241
1242		lbits4 = net->xfrm.policy_hthresh.lbits4;
1243		rbits4 = net->xfrm.policy_hthresh.rbits4;
1244		lbits6 = net->xfrm.policy_hthresh.lbits6;
1245		rbits6 = net->xfrm.policy_hthresh.rbits6;
1246	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1247
1248	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1249	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1250
1251	/* make sure that we can insert the indirect policies again before
1252	 * we start with destructive action.
1253	 */
1254	list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1255		struct xfrm_pol_inexact_bin *bin;
1256		u8 dbits, sbits;
1257
1258		dir = xfrm_policy_id2dir(policy->index);
1259		if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1260			continue;
1261
1262		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1263			if (policy->family == AF_INET) {
1264				dbits = rbits4;
1265				sbits = lbits4;
1266			} else {
1267				dbits = rbits6;
1268				sbits = lbits6;
1269			}
1270		} else {
1271			if (policy->family == AF_INET) {
1272				dbits = lbits4;
1273				sbits = rbits4;
1274			} else {
1275				dbits = lbits6;
1276				sbits = rbits6;
1277			}
1278		}
1279
1280		if (policy->selector.prefixlen_d < dbits ||
1281		    policy->selector.prefixlen_s < sbits)
1282			continue;
1283
1284		bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1285		if (!bin)
1286			goto out_unlock;
1287
1288		if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1289			goto out_unlock;
1290	}
1291
1292	/* reset the bydst and inexact table in all directions */
1293	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1294		struct hlist_node *n;
1295
1296		hlist_for_each_entry_safe(policy, n,
1297					  &net->xfrm.policy_inexact[dir],
1298					  bydst_inexact_list) {
1299			hlist_del_rcu(&policy->bydst);
1300			hlist_del_init(&policy->bydst_inexact_list);
1301		}
1302
1303		hmask = net->xfrm.policy_bydst[dir].hmask;
1304		odst = net->xfrm.policy_bydst[dir].table;
1305		for (i = hmask; i >= 0; i--) {
1306			hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1307				hlist_del_rcu(&policy->bydst);
1308		}
1309		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1310			/* dir out => dst = remote, src = local */
1311			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1312			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1313			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1314			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1315		} else {
1316			/* dir in/fwd => dst = local, src = remote */
1317			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1318			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1319			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1320			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1321		}
1322	}
1323
1324	/* re-insert all policies by order of creation */
1325	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1326		if (policy->walk.dead)
1327			continue;
1328		dir = xfrm_policy_id2dir(policy->index);
1329		if (dir >= XFRM_POLICY_MAX) {
1330			/* skip socket policies */
1331			continue;
1332		}
1333		newpos = NULL;
1334		chain = policy_hash_bysel(net, &policy->selector,
1335					  policy->family, dir);
1336
1337		if (!chain) {
1338			void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1339
1340			WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1341			continue;
1342		}
1343
1344		hlist_for_each_entry(pol, chain, bydst) {
1345			if (policy->priority >= pol->priority)
1346				newpos = &pol->bydst;
1347			else
1348				break;
1349		}
1350		if (newpos)
1351			hlist_add_behind_rcu(&policy->bydst, newpos);
1352		else
1353			hlist_add_head_rcu(&policy->bydst, chain);
1354	}
1355
1356out_unlock:
1357	__xfrm_policy_inexact_flush(net);
1358	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1359	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1360
1361	mutex_unlock(&hash_resize_mutex);
1362}
1363
1364void xfrm_policy_hash_rebuild(struct net *net)
1365{
1366	schedule_work(&net->xfrm.policy_hthresh.work);
1367}
1368EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1369
1370/* Generate new index... KAME seems to generate them ordered by cost
1371 * of an absolute inpredictability of ordering of rules. This will not pass. */
1372static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1373{
1374	for (;;) {
1375		struct hlist_head *list;
1376		struct xfrm_policy *p;
1377		u32 idx;
1378		int found;
1379
1380		if (!index) {
1381			idx = (net->xfrm.idx_generator | dir);
1382			net->xfrm.idx_generator += 8;
1383		} else {
1384			idx = index;
1385			index = 0;
1386		}
1387
1388		if (idx == 0)
1389			idx = 8;
1390		list = net->xfrm.policy_byidx + idx_hash(net, idx);
1391		found = 0;
1392		hlist_for_each_entry(p, list, byidx) {
1393			if (p->index == idx) {
1394				found = 1;
1395				break;
1396			}
1397		}
1398		if (!found)
1399			return idx;
1400	}
1401}
1402
1403static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1404{
1405	u32 *p1 = (u32 *) s1;
1406	u32 *p2 = (u32 *) s2;
1407	int len = sizeof(struct xfrm_selector) / sizeof(u32);
1408	int i;
1409
1410	for (i = 0; i < len; i++) {
1411		if (p1[i] != p2[i])
1412			return 1;
1413	}
1414
1415	return 0;
1416}
1417
1418static void xfrm_policy_requeue(struct xfrm_policy *old,
1419				struct xfrm_policy *new)
1420{
1421	struct xfrm_policy_queue *pq = &old->polq;
1422	struct sk_buff_head list;
1423
1424	if (skb_queue_empty(&pq->hold_queue))
1425		return;
1426
1427	__skb_queue_head_init(&list);
1428
1429	spin_lock_bh(&pq->hold_queue.lock);
1430	skb_queue_splice_init(&pq->hold_queue, &list);
1431	if (del_timer(&pq->hold_timer))
1432		xfrm_pol_put(old);
1433	spin_unlock_bh(&pq->hold_queue.lock);
1434
1435	pq = &new->polq;
1436
1437	spin_lock_bh(&pq->hold_queue.lock);
1438	skb_queue_splice(&list, &pq->hold_queue);
1439	pq->timeout = XFRM_QUEUE_TMO_MIN;
1440	if (!mod_timer(&pq->hold_timer, jiffies))
1441		xfrm_pol_hold(new);
1442	spin_unlock_bh(&pq->hold_queue.lock);
1443}
1444
1445static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1446					  struct xfrm_policy *pol)
1447{
1448	return mark->v == pol->mark.v && mark->m == pol->mark.m;
1449}
1450
1451static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1452{
1453	const struct xfrm_pol_inexact_key *k = data;
1454	u32 a = k->type << 24 | k->dir << 16 | k->family;
1455
1456	return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1457			    seed);
1458}
1459
1460static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1461{
1462	const struct xfrm_pol_inexact_bin *b = data;
1463
1464	return xfrm_pol_bin_key(&b->k, 0, seed);
1465}
1466
1467static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1468			    const void *ptr)
1469{
1470	const struct xfrm_pol_inexact_key *key = arg->key;
1471	const struct xfrm_pol_inexact_bin *b = ptr;
1472	int ret;
1473
1474	if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1475		return -1;
1476
1477	ret = b->k.dir ^ key->dir;
1478	if (ret)
1479		return ret;
1480
1481	ret = b->k.type ^ key->type;
1482	if (ret)
1483		return ret;
1484
1485	ret = b->k.family ^ key->family;
1486	if (ret)
1487		return ret;
1488
1489	return b->k.if_id ^ key->if_id;
1490}
1491
1492static const struct rhashtable_params xfrm_pol_inexact_params = {
1493	.head_offset		= offsetof(struct xfrm_pol_inexact_bin, head),
1494	.hashfn			= xfrm_pol_bin_key,
1495	.obj_hashfn		= xfrm_pol_bin_obj,
1496	.obj_cmpfn		= xfrm_pol_bin_cmp,
1497	.automatic_shrinking	= true,
1498};
1499
1500static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1501					    struct xfrm_policy *policy)
1502{
1503	struct xfrm_policy *pol, *delpol = NULL;
1504	struct hlist_node *newpos = NULL;
1505	int i = 0;
1506
1507	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1508		if (pol->type == policy->type &&
1509		    pol->if_id == policy->if_id &&
1510		    !selector_cmp(&pol->selector, &policy->selector) &&
1511		    xfrm_policy_mark_match(&policy->mark, pol) &&
1512		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1513		    !WARN_ON(delpol)) {
1514			delpol = pol;
1515			if (policy->priority > pol->priority)
1516				continue;
1517		} else if (policy->priority >= pol->priority) {
1518			newpos = &pol->bydst_inexact_list;
1519			continue;
1520		}
1521		if (delpol)
1522			break;
1523	}
1524
1525	if (newpos)
1526		hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1527	else
1528		hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1529
1530	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1531		pol->pos = i;
1532		i++;
1533	}
1534}
1535
1536static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1537						   struct xfrm_policy *policy,
1538						   bool excl)
1539{
1540	struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1541
1542	hlist_for_each_entry(pol, chain, bydst) {
1543		if (pol->type == policy->type &&
1544		    pol->if_id == policy->if_id &&
1545		    !selector_cmp(&pol->selector, &policy->selector) &&
1546		    xfrm_policy_mark_match(&policy->mark, pol) &&
1547		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1548		    !WARN_ON(delpol)) {
1549			if (excl)
1550				return ERR_PTR(-EEXIST);
1551			delpol = pol;
1552			if (policy->priority > pol->priority)
1553				continue;
1554		} else if (policy->priority >= pol->priority) {
1555			newpos = pol;
1556			continue;
1557		}
1558		if (delpol)
1559			break;
1560	}
1561
1562	if (newpos)
1563		hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1564	else
1565		hlist_add_head_rcu(&policy->bydst, chain);
1566
1567	return delpol;
1568}
1569
1570int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1571{
1572	struct net *net = xp_net(policy);
1573	struct xfrm_policy *delpol;
1574	struct hlist_head *chain;
1575
1576	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1577	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1578	if (chain)
1579		delpol = xfrm_policy_insert_list(chain, policy, excl);
1580	else
1581		delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1582
1583	if (IS_ERR(delpol)) {
1584		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1585		return PTR_ERR(delpol);
1586	}
1587
1588	__xfrm_policy_link(policy, dir);
1589
1590	/* After previous checking, family can either be AF_INET or AF_INET6 */
1591	if (policy->family == AF_INET)
1592		rt_genid_bump_ipv4(net);
1593	else
1594		rt_genid_bump_ipv6(net);
1595
1596	if (delpol) {
1597		xfrm_policy_requeue(delpol, policy);
1598		__xfrm_policy_unlink(delpol, dir);
1599	}
1600	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1601	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1602	policy->curlft.add_time = ktime_get_real_seconds();
1603	policy->curlft.use_time = 0;
1604	if (!mod_timer(&policy->timer, jiffies + HZ))
1605		xfrm_pol_hold(policy);
1606	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1607
1608	if (delpol)
1609		xfrm_policy_kill(delpol);
1610	else if (xfrm_bydst_should_resize(net, dir, NULL))
1611		schedule_work(&net->xfrm.policy_hash_work);
1612
1613	return 0;
1614}
1615EXPORT_SYMBOL(xfrm_policy_insert);
1616
1617static struct xfrm_policy *
1618__xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1619			u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1620			struct xfrm_sec_ctx *ctx)
1621{
1622	struct xfrm_policy *pol;
1623
1624	if (!chain)
1625		return NULL;
1626
1627	hlist_for_each_entry(pol, chain, bydst) {
1628		if (pol->type == type &&
1629		    pol->if_id == if_id &&
1630		    xfrm_policy_mark_match(mark, pol) &&
1631		    !selector_cmp(sel, &pol->selector) &&
1632		    xfrm_sec_ctx_match(ctx, pol->security))
1633			return pol;
1634	}
1635
1636	return NULL;
1637}
1638
1639struct xfrm_policy *
1640xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1641		      u8 type, int dir, struct xfrm_selector *sel,
1642		      struct xfrm_sec_ctx *ctx, int delete, int *err)
1643{
1644	struct xfrm_pol_inexact_bin *bin = NULL;
1645	struct xfrm_policy *pol, *ret = NULL;
1646	struct hlist_head *chain;
1647
1648	*err = 0;
1649	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1650	chain = policy_hash_bysel(net, sel, sel->family, dir);
1651	if (!chain) {
1652		struct xfrm_pol_inexact_candidates cand;
1653		int i;
1654
1655		bin = xfrm_policy_inexact_lookup(net, type,
1656						 sel->family, dir, if_id);
1657		if (!bin) {
1658			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1659			return NULL;
1660		}
1661
1662		if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1663							 &sel->saddr,
1664							 &sel->daddr)) {
1665			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1666			return NULL;
1667		}
1668
1669		pol = NULL;
1670		for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1671			struct xfrm_policy *tmp;
1672
1673			tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1674						      if_id, type, dir,
1675						      sel, ctx);
1676			if (!tmp)
1677				continue;
1678
1679			if (!pol || tmp->pos < pol->pos)
1680				pol = tmp;
1681		}
1682	} else {
1683		pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1684					      sel, ctx);
1685	}
1686
1687	if (pol) {
1688		xfrm_pol_hold(pol);
1689		if (delete) {
1690			*err = security_xfrm_policy_delete(pol->security);
1691			if (*err) {
1692				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1693				return pol;
1694			}
1695			__xfrm_policy_unlink(pol, dir);
1696		}
1697		ret = pol;
1698	}
1699	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1700
1701	if (ret && delete)
1702		xfrm_policy_kill(ret);
1703	if (bin && delete)
1704		xfrm_policy_inexact_prune_bin(bin);
1705	return ret;
1706}
1707EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1708
1709struct xfrm_policy *
1710xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1711		 u8 type, int dir, u32 id, int delete, int *err)
1712{
1713	struct xfrm_policy *pol, *ret;
1714	struct hlist_head *chain;
1715
1716	*err = -ENOENT;
1717	if (xfrm_policy_id2dir(id) != dir)
1718		return NULL;
1719
1720	*err = 0;
1721	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1722	chain = net->xfrm.policy_byidx + idx_hash(net, id);
1723	ret = NULL;
1724	hlist_for_each_entry(pol, chain, byidx) {
1725		if (pol->type == type && pol->index == id &&
1726		    pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1727			xfrm_pol_hold(pol);
1728			if (delete) {
1729				*err = security_xfrm_policy_delete(
1730								pol->security);
1731				if (*err) {
1732					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1733					return pol;
1734				}
1735				__xfrm_policy_unlink(pol, dir);
1736			}
1737			ret = pol;
1738			break;
1739		}
1740	}
1741	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1742
1743	if (ret && delete)
1744		xfrm_policy_kill(ret);
1745	return ret;
1746}
1747EXPORT_SYMBOL(xfrm_policy_byid);
1748
1749#ifdef CONFIG_SECURITY_NETWORK_XFRM
1750static inline int
1751xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1752{
1753	struct xfrm_policy *pol;
1754	int err = 0;
1755
1756	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1757		if (pol->walk.dead ||
1758		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1759		    pol->type != type)
1760			continue;
1761
1762		err = security_xfrm_policy_delete(pol->security);
1763		if (err) {
1764			xfrm_audit_policy_delete(pol, 0, task_valid);
1765			return err;
1766		}
1767	}
1768	return err;
1769}
1770#else
1771static inline int
1772xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1773{
1774	return 0;
1775}
1776#endif
1777
1778int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1779{
1780	int dir, err = 0, cnt = 0;
1781	struct xfrm_policy *pol;
1782
1783	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1784
1785	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1786	if (err)
1787		goto out;
1788
1789again:
1790	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1791		dir = xfrm_policy_id2dir(pol->index);
1792		if (pol->walk.dead ||
1793		    dir >= XFRM_POLICY_MAX ||
1794		    pol->type != type)
1795			continue;
1796
1797		__xfrm_policy_unlink(pol, dir);
1798		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1799		cnt++;
1800		xfrm_audit_policy_delete(pol, 1, task_valid);
1801		xfrm_policy_kill(pol);
1802		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1803		goto again;
1804	}
1805	if (cnt)
1806		__xfrm_policy_inexact_flush(net);
1807	else
1808		err = -ESRCH;
1809out:
1810	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1811	return err;
1812}
1813EXPORT_SYMBOL(xfrm_policy_flush);
1814
1815int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1816		     int (*func)(struct xfrm_policy *, int, int, void*),
1817		     void *data)
1818{
1819	struct xfrm_policy *pol;
1820	struct xfrm_policy_walk_entry *x;
1821	int error = 0;
1822
1823	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1824	    walk->type != XFRM_POLICY_TYPE_ANY)
1825		return -EINVAL;
1826
1827	if (list_empty(&walk->walk.all) && walk->seq != 0)
1828		return 0;
1829
1830	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1831	if (list_empty(&walk->walk.all))
1832		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1833	else
1834		x = list_first_entry(&walk->walk.all,
1835				     struct xfrm_policy_walk_entry, all);
1836
1837	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1838		if (x->dead)
1839			continue;
1840		pol = container_of(x, struct xfrm_policy, walk);
1841		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1842		    walk->type != pol->type)
1843			continue;
1844		error = func(pol, xfrm_policy_id2dir(pol->index),
1845			     walk->seq, data);
1846		if (error) {
1847			list_move_tail(&walk->walk.all, &x->all);
1848			goto out;
1849		}
1850		walk->seq++;
1851	}
1852	if (walk->seq == 0) {
1853		error = -ENOENT;
1854		goto out;
1855	}
1856	list_del_init(&walk->walk.all);
1857out:
1858	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1859	return error;
1860}
1861EXPORT_SYMBOL(xfrm_policy_walk);
1862
1863void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1864{
1865	INIT_LIST_HEAD(&walk->walk.all);
1866	walk->walk.dead = 1;
1867	walk->type = type;
1868	walk->seq = 0;
1869}
1870EXPORT_SYMBOL(xfrm_policy_walk_init);
1871
1872void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1873{
1874	if (list_empty(&walk->walk.all))
1875		return;
1876
1877	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1878	list_del(&walk->walk.all);
1879	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1880}
1881EXPORT_SYMBOL(xfrm_policy_walk_done);
1882
1883/*
1884 * Find policy to apply to this flow.
1885 *
1886 * Returns 0 if policy found, else an -errno.
1887 */
1888static int xfrm_policy_match(const struct xfrm_policy *pol,
1889			     const struct flowi *fl,
1890			     u8 type, u16 family, int dir, u32 if_id)
1891{
1892	const struct xfrm_selector *sel = &pol->selector;
1893	int ret = -ESRCH;
1894	bool match;
1895
1896	if (pol->family != family ||
1897	    pol->if_id != if_id ||
1898	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1899	    pol->type != type)
1900		return ret;
1901
1902	match = xfrm_selector_match(sel, fl, family);
1903	if (match)
1904		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1905						  dir);
1906	return ret;
1907}
1908
1909static struct xfrm_pol_inexact_node *
1910xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1911				seqcount_spinlock_t *count,
1912				const xfrm_address_t *addr, u16 family)
1913{
1914	const struct rb_node *parent;
1915	int seq;
1916
1917again:
1918	seq = read_seqcount_begin(count);
1919
1920	parent = rcu_dereference_raw(r->rb_node);
1921	while (parent) {
1922		struct xfrm_pol_inexact_node *node;
1923		int delta;
1924
1925		node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1926
1927		delta = xfrm_policy_addr_delta(addr, &node->addr,
1928					       node->prefixlen, family);
1929		if (delta < 0) {
1930			parent = rcu_dereference_raw(parent->rb_left);
1931			continue;
1932		} else if (delta > 0) {
1933			parent = rcu_dereference_raw(parent->rb_right);
1934			continue;
1935		}
1936
1937		return node;
1938	}
1939
1940	if (read_seqcount_retry(count, seq))
1941		goto again;
1942
1943	return NULL;
1944}
1945
1946static bool
1947xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1948				    struct xfrm_pol_inexact_bin *b,
1949				    const xfrm_address_t *saddr,
1950				    const xfrm_address_t *daddr)
1951{
1952	struct xfrm_pol_inexact_node *n;
1953	u16 family;
1954
1955	if (!b)
1956		return false;
1957
1958	family = b->k.family;
1959	memset(cand, 0, sizeof(*cand));
1960	cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1961
1962	n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1963					    family);
1964	if (n) {
1965		cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1966		n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1967						    family);
1968		if (n)
1969			cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1970	}
1971
1972	n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1973					    family);
1974	if (n)
1975		cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1976
1977	return true;
1978}
1979
1980static struct xfrm_pol_inexact_bin *
1981xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1982			       u8 dir, u32 if_id)
1983{
1984	struct xfrm_pol_inexact_key k = {
1985		.family = family,
1986		.type = type,
1987		.dir = dir,
1988		.if_id = if_id,
1989	};
1990
1991	write_pnet(&k.net, net);
1992
1993	return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1994				 xfrm_pol_inexact_params);
1995}
1996
1997static struct xfrm_pol_inexact_bin *
1998xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1999			   u8 dir, u32 if_id)
2000{
2001	struct xfrm_pol_inexact_bin *bin;
2002
2003	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2004
2005	rcu_read_lock();
2006	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2007	rcu_read_unlock();
2008
2009	return bin;
2010}
2011
2012static struct xfrm_policy *
2013__xfrm_policy_eval_candidates(struct hlist_head *chain,
2014			      struct xfrm_policy *prefer,
2015			      const struct flowi *fl,
2016			      u8 type, u16 family, int dir, u32 if_id)
2017{
2018	u32 priority = prefer ? prefer->priority : ~0u;
2019	struct xfrm_policy *pol;
2020
2021	if (!chain)
2022		return NULL;
2023
2024	hlist_for_each_entry_rcu(pol, chain, bydst) {
2025		int err;
2026
2027		if (pol->priority > priority)
2028			break;
2029
2030		err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2031		if (err) {
2032			if (err != -ESRCH)
2033				return ERR_PTR(err);
2034
2035			continue;
2036		}
2037
2038		if (prefer) {
2039			/* matches.  Is it older than *prefer? */
2040			if (pol->priority == priority &&
2041			    prefer->pos < pol->pos)
2042				return prefer;
2043		}
2044
2045		return pol;
2046	}
2047
2048	return NULL;
2049}
2050
2051static struct xfrm_policy *
2052xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2053			    struct xfrm_policy *prefer,
2054			    const struct flowi *fl,
2055			    u8 type, u16 family, int dir, u32 if_id)
2056{
2057	struct xfrm_policy *tmp;
2058	int i;
2059
2060	for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2061		tmp = __xfrm_policy_eval_candidates(cand->res[i],
2062						    prefer,
2063						    fl, type, family, dir,
2064						    if_id);
2065		if (!tmp)
2066			continue;
2067
2068		if (IS_ERR(tmp))
2069			return tmp;
2070		prefer = tmp;
2071	}
2072
2073	return prefer;
2074}
2075
2076static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2077						     const struct flowi *fl,
2078						     u16 family, u8 dir,
2079						     u32 if_id)
2080{
2081	struct xfrm_pol_inexact_candidates cand;
2082	const xfrm_address_t *daddr, *saddr;
2083	struct xfrm_pol_inexact_bin *bin;
2084	struct xfrm_policy *pol, *ret;
2085	struct hlist_head *chain;
2086	unsigned int sequence;
2087	int err;
2088
2089	daddr = xfrm_flowi_daddr(fl, family);
2090	saddr = xfrm_flowi_saddr(fl, family);
2091	if (unlikely(!daddr || !saddr))
2092		return NULL;
2093
2094	rcu_read_lock();
2095 retry:
2096	do {
2097		sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2098		chain = policy_hash_direct(net, daddr, saddr, family, dir);
2099	} while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2100
2101	ret = NULL;
2102	hlist_for_each_entry_rcu(pol, chain, bydst) {
2103		err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2104		if (err) {
2105			if (err == -ESRCH)
2106				continue;
2107			else {
2108				ret = ERR_PTR(err);
2109				goto fail;
2110			}
2111		} else {
2112			ret = pol;
2113			break;
2114		}
2115	}
2116	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2117	if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2118							 daddr))
2119		goto skip_inexact;
2120
2121	pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2122					  family, dir, if_id);
2123	if (pol) {
2124		ret = pol;
2125		if (IS_ERR(pol))
2126			goto fail;
2127	}
2128
2129skip_inexact:
2130	if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2131		goto retry;
2132
2133	if (ret && !xfrm_pol_hold_rcu(ret))
2134		goto retry;
2135fail:
2136	rcu_read_unlock();
2137
2138	return ret;
2139}
2140
2141static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2142					      const struct flowi *fl,
2143					      u16 family, u8 dir, u32 if_id)
2144{
2145#ifdef CONFIG_XFRM_SUB_POLICY
2146	struct xfrm_policy *pol;
2147
2148	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2149					dir, if_id);
2150	if (pol != NULL)
2151		return pol;
2152#endif
2153	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2154					 dir, if_id);
2155}
2156
2157static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2158						 const struct flowi *fl,
2159						 u16 family, u32 if_id)
2160{
2161	struct xfrm_policy *pol;
2162
2163	rcu_read_lock();
2164 again:
2165	pol = rcu_dereference(sk->sk_policy[dir]);
2166	if (pol != NULL) {
2167		bool match;
2168		int err = 0;
2169
2170		if (pol->family != family) {
2171			pol = NULL;
2172			goto out;
2173		}
2174
2175		match = xfrm_selector_match(&pol->selector, fl, family);
2176		if (match) {
2177			if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2178			    pol->if_id != if_id) {
2179				pol = NULL;
2180				goto out;
2181			}
2182			err = security_xfrm_policy_lookup(pol->security,
2183						      fl->flowi_secid,
2184						      dir);
2185			if (!err) {
2186				if (!xfrm_pol_hold_rcu(pol))
2187					goto again;
2188			} else if (err == -ESRCH) {
2189				pol = NULL;
2190			} else {
2191				pol = ERR_PTR(err);
2192			}
2193		} else
2194			pol = NULL;
2195	}
2196out:
2197	rcu_read_unlock();
2198	return pol;
2199}
2200
2201static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2202{
2203	struct net *net = xp_net(pol);
2204
2205	list_add(&pol->walk.all, &net->xfrm.policy_all);
2206	net->xfrm.policy_count[dir]++;
2207	xfrm_pol_hold(pol);
2208}
2209
2210static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2211						int dir)
2212{
2213	struct net *net = xp_net(pol);
2214
2215	if (list_empty(&pol->walk.all))
2216		return NULL;
2217
2218	/* Socket policies are not hashed. */
2219	if (!hlist_unhashed(&pol->bydst)) {
2220		hlist_del_rcu(&pol->bydst);
2221		hlist_del_init(&pol->bydst_inexact_list);
2222		hlist_del(&pol->byidx);
2223	}
2224
2225	list_del_init(&pol->walk.all);
2226	net->xfrm.policy_count[dir]--;
2227
2228	return pol;
2229}
2230
2231static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2232{
2233	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2234}
2235
2236static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2237{
2238	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2239}
2240
2241int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2242{
2243	struct net *net = xp_net(pol);
2244
2245	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2246	pol = __xfrm_policy_unlink(pol, dir);
2247	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2248	if (pol) {
2249		xfrm_policy_kill(pol);
2250		return 0;
2251	}
2252	return -ENOENT;
2253}
2254EXPORT_SYMBOL(xfrm_policy_delete);
2255
2256int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2257{
2258	struct net *net = sock_net(sk);
2259	struct xfrm_policy *old_pol;
2260
2261#ifdef CONFIG_XFRM_SUB_POLICY
2262	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2263		return -EINVAL;
2264#endif
2265
2266	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2267	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2268				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2269	if (pol) {
2270		pol->curlft.add_time = ktime_get_real_seconds();
2271		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2272		xfrm_sk_policy_link(pol, dir);
2273	}
2274	rcu_assign_pointer(sk->sk_policy[dir], pol);
2275	if (old_pol) {
2276		if (pol)
2277			xfrm_policy_requeue(old_pol, pol);
2278
2279		/* Unlinking succeeds always. This is the only function
2280		 * allowed to delete or replace socket policy.
2281		 */
2282		xfrm_sk_policy_unlink(old_pol, dir);
2283	}
2284	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2285
2286	if (old_pol) {
2287		xfrm_policy_kill(old_pol);
2288	}
2289	return 0;
2290}
2291
2292static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2293{
2294	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2295	struct net *net = xp_net(old);
2296
2297	if (newp) {
2298		newp->selector = old->selector;
2299		if (security_xfrm_policy_clone(old->security,
2300					       &newp->security)) {
2301			kfree(newp);
2302			return NULL;  /* ENOMEM */
2303		}
2304		newp->lft = old->lft;
2305		newp->curlft = old->curlft;
2306		newp->mark = old->mark;
2307		newp->if_id = old->if_id;
2308		newp->action = old->action;
2309		newp->flags = old->flags;
2310		newp->xfrm_nr = old->xfrm_nr;
2311		newp->index = old->index;
2312		newp->type = old->type;
2313		newp->family = old->family;
2314		memcpy(newp->xfrm_vec, old->xfrm_vec,
2315		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2316		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2317		xfrm_sk_policy_link(newp, dir);
2318		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2319		xfrm_pol_put(newp);
2320	}
2321	return newp;
2322}
2323
2324int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2325{
2326	const struct xfrm_policy *p;
2327	struct xfrm_policy *np;
2328	int i, ret = 0;
2329
2330	rcu_read_lock();
2331	for (i = 0; i < 2; i++) {
2332		p = rcu_dereference(osk->sk_policy[i]);
2333		if (p) {
2334			np = clone_policy(p, i);
2335			if (unlikely(!np)) {
2336				ret = -ENOMEM;
2337				break;
2338			}
2339			rcu_assign_pointer(sk->sk_policy[i], np);
2340		}
2341	}
2342	rcu_read_unlock();
2343	return ret;
2344}
2345
2346static int
2347xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2348	       xfrm_address_t *remote, unsigned short family, u32 mark)
2349{
2350	int err;
2351	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2352
2353	if (unlikely(afinfo == NULL))
2354		return -EINVAL;
2355	err = afinfo->get_saddr(net, oif, local, remote, mark);
2356	rcu_read_unlock();
2357	return err;
2358}
2359
2360/* Resolve list of templates for the flow, given policy. */
2361
2362static int
2363xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2364		      struct xfrm_state **xfrm, unsigned short family)
2365{
2366	struct net *net = xp_net(policy);
2367	int nx;
2368	int i, error;
2369	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2370	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2371	xfrm_address_t tmp;
2372
2373	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2374		struct xfrm_state *x;
2375		xfrm_address_t *remote = daddr;
2376		xfrm_address_t *local  = saddr;
2377		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2378
2379		if (tmpl->mode == XFRM_MODE_TUNNEL ||
2380		    tmpl->mode == XFRM_MODE_BEET) {
2381			remote = &tmpl->id.daddr;
2382			local = &tmpl->saddr;
2383			if (xfrm_addr_any(local, tmpl->encap_family)) {
2384				error = xfrm_get_saddr(net, fl->flowi_oif,
2385						       &tmp, remote,
2386						       tmpl->encap_family, 0);
2387				if (error)
2388					goto fail;
2389				local = &tmp;
2390			}
2391		}
2392
2393		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2394				    family, policy->if_id);
2395
2396		if (x && x->km.state == XFRM_STATE_VALID) {
2397			xfrm[nx++] = x;
2398			daddr = remote;
2399			saddr = local;
2400			continue;
2401		}
2402		if (x) {
2403			error = (x->km.state == XFRM_STATE_ERROR ?
2404				 -EINVAL : -EAGAIN);
2405			xfrm_state_put(x);
2406		} else if (error == -ESRCH) {
2407			error = -EAGAIN;
2408		}
2409
2410		if (!tmpl->optional)
2411			goto fail;
2412	}
2413	return nx;
2414
2415fail:
2416	for (nx--; nx >= 0; nx--)
2417		xfrm_state_put(xfrm[nx]);
2418	return error;
2419}
2420
2421static int
2422xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2423		  struct xfrm_state **xfrm, unsigned short family)
2424{
2425	struct xfrm_state *tp[XFRM_MAX_DEPTH];
2426	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2427	int cnx = 0;
2428	int error;
2429	int ret;
2430	int i;
2431
2432	for (i = 0; i < npols; i++) {
2433		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2434			error = -ENOBUFS;
2435			goto fail;
2436		}
2437
2438		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2439		if (ret < 0) {
2440			error = ret;
2441			goto fail;
2442		} else
2443			cnx += ret;
2444	}
2445
2446	/* found states are sorted for outbound processing */
2447	if (npols > 1)
2448		xfrm_state_sort(xfrm, tpp, cnx, family);
2449
2450	return cnx;
2451
2452 fail:
2453	for (cnx--; cnx >= 0; cnx--)
2454		xfrm_state_put(tpp[cnx]);
2455	return error;
2456
2457}
2458
2459static int xfrm_get_tos(const struct flowi *fl, int family)
2460{
2461	if (family == AF_INET)
2462		return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2463
2464	return 0;
2465}
2466
2467static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2468{
2469	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2470	struct dst_ops *dst_ops;
2471	struct xfrm_dst *xdst;
2472
2473	if (!afinfo)
2474		return ERR_PTR(-EINVAL);
2475
2476	switch (family) {
2477	case AF_INET:
2478		dst_ops = &net->xfrm.xfrm4_dst_ops;
2479		break;
2480#if IS_ENABLED(CONFIG_IPV6)
2481	case AF_INET6:
2482		dst_ops = &net->xfrm.xfrm6_dst_ops;
2483		break;
2484#endif
2485	default:
2486		BUG();
2487	}
2488	xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2489
2490	if (likely(xdst)) {
2491		struct dst_entry *dst = &xdst->u.dst;
2492
2493		memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2494	} else
2495		xdst = ERR_PTR(-ENOBUFS);
2496
2497	rcu_read_unlock();
2498
2499	return xdst;
2500}
2501
2502static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2503			   int nfheader_len)
2504{
2505	if (dst->ops->family == AF_INET6) {
2506		struct rt6_info *rt = (struct rt6_info *)dst;
2507		path->path_cookie = rt6_get_cookie(rt);
2508		path->u.rt6.rt6i_nfheader_len = nfheader_len;
2509	}
2510}
2511
2512static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2513				const struct flowi *fl)
2514{
2515	const struct xfrm_policy_afinfo *afinfo =
2516		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2517	int err;
2518
2519	if (!afinfo)
2520		return -EINVAL;
2521
2522	err = afinfo->fill_dst(xdst, dev, fl);
2523
2524	rcu_read_unlock();
2525
2526	return err;
2527}
2528
2529
2530/* Allocate chain of dst_entry's, attach known xfrm's, calculate
2531 * all the metrics... Shortly, bundle a bundle.
2532 */
2533
2534static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2535					    struct xfrm_state **xfrm,
2536					    struct xfrm_dst **bundle,
2537					    int nx,
2538					    const struct flowi *fl,
2539					    struct dst_entry *dst)
2540{
2541	const struct xfrm_state_afinfo *afinfo;
2542	const struct xfrm_mode *inner_mode;
2543	struct net *net = xp_net(policy);
2544	unsigned long now = jiffies;
2545	struct net_device *dev;
2546	struct xfrm_dst *xdst_prev = NULL;
2547	struct xfrm_dst *xdst0 = NULL;
2548	int i = 0;
2549	int err;
2550	int header_len = 0;
2551	int nfheader_len = 0;
2552	int trailer_len = 0;
2553	int tos;
2554	int family = policy->selector.family;
2555	xfrm_address_t saddr, daddr;
2556
2557	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2558
2559	tos = xfrm_get_tos(fl, family);
2560
2561	dst_hold(dst);
2562
2563	for (; i < nx; i++) {
2564		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2565		struct dst_entry *dst1 = &xdst->u.dst;
2566
2567		err = PTR_ERR(xdst);
2568		if (IS_ERR(xdst)) {
2569			dst_release(dst);
2570			goto put_states;
2571		}
2572
2573		bundle[i] = xdst;
2574		if (!xdst_prev)
2575			xdst0 = xdst;
2576		else
2577			/* Ref count is taken during xfrm_alloc_dst()
2578			 * No need to do dst_clone() on dst1
2579			 */
2580			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2581
2582		if (xfrm[i]->sel.family == AF_UNSPEC) {
2583			inner_mode = xfrm_ip2inner_mode(xfrm[i],
2584							xfrm_af2proto(family));
2585			if (!inner_mode) {
2586				err = -EAFNOSUPPORT;
2587				dst_release(dst);
2588				goto put_states;
2589			}
2590		} else
2591			inner_mode = &xfrm[i]->inner_mode;
2592
2593		xdst->route = dst;
2594		dst_copy_metrics(dst1, dst);
2595
2596		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2597			__u32 mark = 0;
2598
2599			if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2600				mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2601
2602			family = xfrm[i]->props.family;
2603			dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2604					      &saddr, &daddr, family, mark);
2605			err = PTR_ERR(dst);
2606			if (IS_ERR(dst))
2607				goto put_states;
2608		} else
2609			dst_hold(dst);
2610
2611		dst1->xfrm = xfrm[i];
2612		xdst->xfrm_genid = xfrm[i]->genid;
2613
2614		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2615		dst1->lastuse = now;
2616
2617		dst1->input = dst_discard;
2618
2619		rcu_read_lock();
2620		afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2621		if (likely(afinfo))
2622			dst1->output = afinfo->output;
2623		else
2624			dst1->output = dst_discard_out;
2625		rcu_read_unlock();
2626
2627		xdst_prev = xdst;
2628
2629		header_len += xfrm[i]->props.header_len;
2630		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2631			nfheader_len += xfrm[i]->props.header_len;
2632		trailer_len += xfrm[i]->props.trailer_len;
2633	}
2634
2635	xfrm_dst_set_child(xdst_prev, dst);
2636	xdst0->path = dst;
2637
2638	err = -ENODEV;
2639	dev = dst->dev;
2640	if (!dev)
2641		goto free_dst;
2642
2643	xfrm_init_path(xdst0, dst, nfheader_len);
2644	xfrm_init_pmtu(bundle, nx);
2645
2646	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2647	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2648		err = xfrm_fill_dst(xdst_prev, dev, fl);
2649		if (err)
2650			goto free_dst;
2651
2652		xdst_prev->u.dst.header_len = header_len;
2653		xdst_prev->u.dst.trailer_len = trailer_len;
2654		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2655		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2656	}
2657
2658	return &xdst0->u.dst;
2659
2660put_states:
2661	for (; i < nx; i++)
2662		xfrm_state_put(xfrm[i]);
2663free_dst:
2664	if (xdst0)
2665		dst_release_immediate(&xdst0->u.dst);
2666
2667	return ERR_PTR(err);
2668}
2669
2670static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2671				struct xfrm_policy **pols,
2672				int *num_pols, int *num_xfrms)
2673{
2674	int i;
2675
2676	if (*num_pols == 0 || !pols[0]) {
2677		*num_pols = 0;
2678		*num_xfrms = 0;
2679		return 0;
2680	}
2681	if (IS_ERR(pols[0])) {
2682		*num_pols = 0;
2683		return PTR_ERR(pols[0]);
2684	}
2685
2686	*num_xfrms = pols[0]->xfrm_nr;
2687
2688#ifdef CONFIG_XFRM_SUB_POLICY
2689	if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2690	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2691		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2692						    XFRM_POLICY_TYPE_MAIN,
2693						    fl, family,
2694						    XFRM_POLICY_OUT,
2695						    pols[0]->if_id);
2696		if (pols[1]) {
2697			if (IS_ERR(pols[1])) {
2698				xfrm_pols_put(pols, *num_pols);
2699				*num_pols = 0;
2700				return PTR_ERR(pols[1]);
2701			}
2702			(*num_pols)++;
2703			(*num_xfrms) += pols[1]->xfrm_nr;
2704		}
2705	}
2706#endif
2707	for (i = 0; i < *num_pols; i++) {
2708		if (pols[i]->action != XFRM_POLICY_ALLOW) {
2709			*num_xfrms = -1;
2710			break;
2711		}
2712	}
2713
2714	return 0;
2715
2716}
2717
2718static struct xfrm_dst *
2719xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2720			       const struct flowi *fl, u16 family,
2721			       struct dst_entry *dst_orig)
2722{
2723	struct net *net = xp_net(pols[0]);
2724	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2725	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2726	struct xfrm_dst *xdst;
2727	struct dst_entry *dst;
2728	int err;
2729
2730	/* Try to instantiate a bundle */
2731	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2732	if (err <= 0) {
2733		if (err == 0)
2734			return NULL;
2735
2736		if (err != -EAGAIN)
2737			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2738		return ERR_PTR(err);
2739	}
2740
2741	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2742	if (IS_ERR(dst)) {
2743		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2744		return ERR_CAST(dst);
2745	}
2746
2747	xdst = (struct xfrm_dst *)dst;
2748	xdst->num_xfrms = err;
2749	xdst->num_pols = num_pols;
2750	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2751	xdst->policy_genid = atomic_read(&pols[0]->genid);
2752
2753	return xdst;
2754}
2755
2756static void xfrm_policy_queue_process(struct timer_list *t)
2757{
2758	struct sk_buff *skb;
2759	struct sock *sk;
2760	struct dst_entry *dst;
2761	struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2762	struct net *net = xp_net(pol);
2763	struct xfrm_policy_queue *pq = &pol->polq;
2764	struct flowi fl;
2765	struct sk_buff_head list;
2766	__u32 skb_mark;
2767
2768	spin_lock(&pq->hold_queue.lock);
2769	skb = skb_peek(&pq->hold_queue);
2770	if (!skb) {
2771		spin_unlock(&pq->hold_queue.lock);
2772		goto out;
2773	}
2774	dst = skb_dst(skb);
2775	sk = skb->sk;
2776
2777	/* Fixup the mark to support VTI. */
2778	skb_mark = skb->mark;
2779	skb->mark = pol->mark.v;
2780	xfrm_decode_session(skb, &fl, dst->ops->family);
2781	skb->mark = skb_mark;
2782	spin_unlock(&pq->hold_queue.lock);
2783
2784	dst_hold(xfrm_dst_path(dst));
2785	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2786	if (IS_ERR(dst))
2787		goto purge_queue;
2788
2789	if (dst->flags & DST_XFRM_QUEUE) {
2790		dst_release(dst);
2791
2792		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2793			goto purge_queue;
2794
2795		pq->timeout = pq->timeout << 1;
2796		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2797			xfrm_pol_hold(pol);
2798		goto out;
2799	}
2800
2801	dst_release(dst);
2802
2803	__skb_queue_head_init(&list);
2804
2805	spin_lock(&pq->hold_queue.lock);
2806	pq->timeout = 0;
2807	skb_queue_splice_init(&pq->hold_queue, &list);
2808	spin_unlock(&pq->hold_queue.lock);
2809
2810	while (!skb_queue_empty(&list)) {
2811		skb = __skb_dequeue(&list);
2812
2813		/* Fixup the mark to support VTI. */
2814		skb_mark = skb->mark;
2815		skb->mark = pol->mark.v;
2816		xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2817		skb->mark = skb_mark;
2818
2819		dst_hold(xfrm_dst_path(skb_dst(skb)));
2820		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2821		if (IS_ERR(dst)) {
2822			kfree_skb(skb);
2823			continue;
2824		}
2825
2826		nf_reset_ct(skb);
2827		skb_dst_drop(skb);
2828		skb_dst_set(skb, dst);
2829
2830		dst_output(net, skb->sk, skb);
2831	}
2832
2833out:
2834	xfrm_pol_put(pol);
2835	return;
2836
2837purge_queue:
2838	pq->timeout = 0;
2839	skb_queue_purge(&pq->hold_queue);
2840	xfrm_pol_put(pol);
2841}
2842
2843static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2844{
2845	unsigned long sched_next;
2846	struct dst_entry *dst = skb_dst(skb);
2847	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2848	struct xfrm_policy *pol = xdst->pols[0];
2849	struct xfrm_policy_queue *pq = &pol->polq;
2850
2851	if (unlikely(skb_fclone_busy(sk, skb))) {
2852		kfree_skb(skb);
2853		return 0;
2854	}
2855
2856	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2857		kfree_skb(skb);
2858		return -EAGAIN;
2859	}
2860
2861	skb_dst_force(skb);
2862
2863	spin_lock_bh(&pq->hold_queue.lock);
2864
2865	if (!pq->timeout)
2866		pq->timeout = XFRM_QUEUE_TMO_MIN;
2867
2868	sched_next = jiffies + pq->timeout;
2869
2870	if (del_timer(&pq->hold_timer)) {
2871		if (time_before(pq->hold_timer.expires, sched_next))
2872			sched_next = pq->hold_timer.expires;
2873		xfrm_pol_put(pol);
2874	}
2875
2876	__skb_queue_tail(&pq->hold_queue, skb);
2877	if (!mod_timer(&pq->hold_timer, sched_next))
2878		xfrm_pol_hold(pol);
2879
2880	spin_unlock_bh(&pq->hold_queue.lock);
2881
2882	return 0;
2883}
2884
2885static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2886						 struct xfrm_flo *xflo,
2887						 const struct flowi *fl,
2888						 int num_xfrms,
2889						 u16 family)
2890{
2891	int err;
2892	struct net_device *dev;
2893	struct dst_entry *dst;
2894	struct dst_entry *dst1;
2895	struct xfrm_dst *xdst;
2896
2897	xdst = xfrm_alloc_dst(net, family);
2898	if (IS_ERR(xdst))
2899		return xdst;
2900
2901	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2902	    net->xfrm.sysctl_larval_drop ||
2903	    num_xfrms <= 0)
2904		return xdst;
2905
2906	dst = xflo->dst_orig;
2907	dst1 = &xdst->u.dst;
2908	dst_hold(dst);
2909	xdst->route = dst;
2910
2911	dst_copy_metrics(dst1, dst);
2912
2913	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2914	dst1->flags |= DST_XFRM_QUEUE;
2915	dst1->lastuse = jiffies;
2916
2917	dst1->input = dst_discard;
2918	dst1->output = xdst_queue_output;
2919
2920	dst_hold(dst);
2921	xfrm_dst_set_child(xdst, dst);
2922	xdst->path = dst;
2923
2924	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2925
2926	err = -ENODEV;
2927	dev = dst->dev;
2928	if (!dev)
2929		goto free_dst;
2930
2931	err = xfrm_fill_dst(xdst, dev, fl);
2932	if (err)
2933		goto free_dst;
2934
2935out:
2936	return xdst;
2937
2938free_dst:
2939	dst_release(dst1);
2940	xdst = ERR_PTR(err);
2941	goto out;
2942}
2943
2944static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2945					   const struct flowi *fl,
2946					   u16 family, u8 dir,
2947					   struct xfrm_flo *xflo, u32 if_id)
2948{
2949	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2950	int num_pols = 0, num_xfrms = 0, err;
2951	struct xfrm_dst *xdst;
2952
2953	/* Resolve policies to use if we couldn't get them from
2954	 * previous cache entry */
2955	num_pols = 1;
2956	pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2957	err = xfrm_expand_policies(fl, family, pols,
2958					   &num_pols, &num_xfrms);
2959	if (err < 0)
2960		goto inc_error;
2961	if (num_pols == 0)
2962		return NULL;
2963	if (num_xfrms <= 0)
2964		goto make_dummy_bundle;
2965
2966	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2967					      xflo->dst_orig);
2968	if (IS_ERR(xdst)) {
2969		err = PTR_ERR(xdst);
2970		if (err == -EREMOTE) {
2971			xfrm_pols_put(pols, num_pols);
2972			return NULL;
2973		}
2974
2975		if (err != -EAGAIN)
2976			goto error;
2977		goto make_dummy_bundle;
2978	} else if (xdst == NULL) {
2979		num_xfrms = 0;
2980		goto make_dummy_bundle;
2981	}
2982
2983	return xdst;
2984
2985make_dummy_bundle:
2986	/* We found policies, but there's no bundles to instantiate:
2987	 * either because the policy blocks, has no transformations or
2988	 * we could not build template (no xfrm_states).*/
2989	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2990	if (IS_ERR(xdst)) {
2991		xfrm_pols_put(pols, num_pols);
2992		return ERR_CAST(xdst);
2993	}
2994	xdst->num_pols = num_pols;
2995	xdst->num_xfrms = num_xfrms;
2996	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2997
2998	return xdst;
2999
3000inc_error:
3001	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3002error:
3003	xfrm_pols_put(pols, num_pols);
3004	return ERR_PTR(err);
3005}
3006
3007static struct dst_entry *make_blackhole(struct net *net, u16 family,
3008					struct dst_entry *dst_orig)
3009{
3010	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3011	struct dst_entry *ret;
3012
3013	if (!afinfo) {
3014		dst_release(dst_orig);
3015		return ERR_PTR(-EINVAL);
3016	} else {
3017		ret = afinfo->blackhole_route(net, dst_orig);
3018	}
3019	rcu_read_unlock();
3020
3021	return ret;
3022}
3023
3024/* Finds/creates a bundle for given flow and if_id
3025 *
3026 * At the moment we eat a raw IP route. Mostly to speed up lookups
3027 * on interfaces with disabled IPsec.
3028 *
3029 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3030 * compatibility
3031 */
3032struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3033					struct dst_entry *dst_orig,
3034					const struct flowi *fl,
3035					const struct sock *sk,
3036					int flags, u32 if_id)
3037{
3038	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3039	struct xfrm_dst *xdst;
3040	struct dst_entry *dst, *route;
3041	u16 family = dst_orig->ops->family;
3042	u8 dir = XFRM_POLICY_OUT;
3043	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3044
3045	dst = NULL;
3046	xdst = NULL;
3047	route = NULL;
3048
3049	sk = sk_const_to_full_sk(sk);
3050	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3051		num_pols = 1;
3052		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3053						if_id);
3054		err = xfrm_expand_policies(fl, family, pols,
3055					   &num_pols, &num_xfrms);
3056		if (err < 0)
3057			goto dropdst;
3058
3059		if (num_pols) {
3060			if (num_xfrms <= 0) {
3061				drop_pols = num_pols;
3062				goto no_transform;
3063			}
3064
3065			xdst = xfrm_resolve_and_create_bundle(
3066					pols, num_pols, fl,
3067					family, dst_orig);
3068
3069			if (IS_ERR(xdst)) {
3070				xfrm_pols_put(pols, num_pols);
3071				err = PTR_ERR(xdst);
3072				if (err == -EREMOTE)
3073					goto nopol;
3074
3075				goto dropdst;
3076			} else if (xdst == NULL) {
3077				num_xfrms = 0;
3078				drop_pols = num_pols;
3079				goto no_transform;
3080			}
3081
3082			route = xdst->route;
3083		}
3084	}
3085
3086	if (xdst == NULL) {
3087		struct xfrm_flo xflo;
3088
3089		xflo.dst_orig = dst_orig;
3090		xflo.flags = flags;
3091
3092		/* To accelerate a bit...  */
3093		if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3094			       !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3095			goto nopol;
3096
3097		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3098		if (xdst == NULL)
3099			goto nopol;
3100		if (IS_ERR(xdst)) {
3101			err = PTR_ERR(xdst);
3102			goto dropdst;
3103		}
3104
3105		num_pols = xdst->num_pols;
3106		num_xfrms = xdst->num_xfrms;
3107		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3108		route = xdst->route;
3109	}
3110
3111	dst = &xdst->u.dst;
3112	if (route == NULL && num_xfrms > 0) {
3113		/* The only case when xfrm_bundle_lookup() returns a
3114		 * bundle with null route, is when the template could
3115		 * not be resolved. It means policies are there, but
3116		 * bundle could not be created, since we don't yet
3117		 * have the xfrm_state's. We need to wait for KM to
3118		 * negotiate new SA's or bail out with error.*/
3119		if (net->xfrm.sysctl_larval_drop) {
3120			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3121			err = -EREMOTE;
3122			goto error;
3123		}
3124
3125		err = -EAGAIN;
3126
3127		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3128		goto error;
3129	}
3130
3131no_transform:
3132	if (num_pols == 0)
3133		goto nopol;
3134
3135	if ((flags & XFRM_LOOKUP_ICMP) &&
3136	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3137		err = -ENOENT;
3138		goto error;
3139	}
3140
3141	for (i = 0; i < num_pols; i++)
3142		pols[i]->curlft.use_time = ktime_get_real_seconds();
3143
3144	if (num_xfrms < 0) {
3145		/* Prohibit the flow */
3146		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3147		err = -EPERM;
3148		goto error;
3149	} else if (num_xfrms > 0) {
3150		/* Flow transformed */
3151		dst_release(dst_orig);
3152	} else {
3153		/* Flow passes untransformed */
3154		dst_release(dst);
3155		dst = dst_orig;
3156	}
3157ok:
3158	xfrm_pols_put(pols, drop_pols);
3159	if (dst && dst->xfrm &&
3160	    dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3161		dst->flags |= DST_XFRM_TUNNEL;
3162	return dst;
3163
3164nopol:
3165	if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3166	    net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3167		err = -EPERM;
3168		goto error;
3169	}
3170	if (!(flags & XFRM_LOOKUP_ICMP)) {
3171		dst = dst_orig;
3172		goto ok;
3173	}
3174	err = -ENOENT;
3175error:
3176	dst_release(dst);
3177dropdst:
3178	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3179		dst_release(dst_orig);
3180	xfrm_pols_put(pols, drop_pols);
3181	return ERR_PTR(err);
3182}
3183EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3184
3185/* Main function: finds/creates a bundle for given flow.
3186 *
3187 * At the moment we eat a raw IP route. Mostly to speed up lookups
3188 * on interfaces with disabled IPsec.
3189 */
3190struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3191			      const struct flowi *fl, const struct sock *sk,
3192			      int flags)
3193{
3194	return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3195}
3196EXPORT_SYMBOL(xfrm_lookup);
3197
3198/* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3199 * Otherwise we may send out blackholed packets.
3200 */
3201struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3202				    const struct flowi *fl,
3203				    const struct sock *sk, int flags)
3204{
3205	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3206					    flags | XFRM_LOOKUP_QUEUE |
3207					    XFRM_LOOKUP_KEEP_DST_REF);
3208
3209	if (PTR_ERR(dst) == -EREMOTE)
3210		return make_blackhole(net, dst_orig->ops->family, dst_orig);
3211
3212	if (IS_ERR(dst))
3213		dst_release(dst_orig);
3214
3215	return dst;
3216}
3217EXPORT_SYMBOL(xfrm_lookup_route);
3218
3219static inline int
3220xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3221{
3222	struct sec_path *sp = skb_sec_path(skb);
3223	struct xfrm_state *x;
3224
3225	if (!sp || idx < 0 || idx >= sp->len)
3226		return 0;
3227	x = sp->xvec[idx];
3228	if (!x->type->reject)
3229		return 0;
3230	return x->type->reject(x, skb, fl);
3231}
3232
3233/* When skb is transformed back to its "native" form, we have to
3234 * check policy restrictions. At the moment we make this in maximally
3235 * stupid way. Shame on me. :-) Of course, connected sockets must
3236 * have policy cached at them.
3237 */
3238
3239static inline int
3240xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3241	      unsigned short family, u32 if_id)
3242{
3243	if (xfrm_state_kern(x))
3244		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3245	return	x->id.proto == tmpl->id.proto &&
3246		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3247		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3248		x->props.mode == tmpl->mode &&
3249		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3250		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3251		!(x->props.mode != XFRM_MODE_TRANSPORT &&
3252		  xfrm_state_addr_cmp(tmpl, x, family)) &&
3253		(if_id == 0 || if_id == x->if_id);
3254}
3255
3256/*
3257 * 0 or more than 0 is returned when validation is succeeded (either bypass
3258 * because of optional transport mode, or next index of the mathced secpath
3259 * state with the template.
3260 * -1 is returned when no matching template is found.
3261 * Otherwise "-2 - errored_index" is returned.
3262 */
3263static inline int
3264xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3265	       unsigned short family, u32 if_id)
3266{
3267	int idx = start;
3268
3269	if (tmpl->optional) {
3270		if (tmpl->mode == XFRM_MODE_TRANSPORT)
3271			return start;
3272	} else
3273		start = -1;
3274	for (; idx < sp->len; idx++) {
3275		if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3276			return ++idx;
3277		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3278			if (idx < sp->verified_cnt) {
3279				/* Secpath entry previously verified, consider optional and
3280				 * continue searching
3281				 */
3282				continue;
3283			}
3284
3285			if (start == -1)
3286				start = -2-idx;
3287			break;
3288		}
3289	}
3290	return start;
3291}
3292
3293static void
3294decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3295{
3296	const struct iphdr *iph = ip_hdr(skb);
3297	int ihl = iph->ihl;
3298	u8 *xprth = skb_network_header(skb) + ihl * 4;
3299	struct flowi4 *fl4 = &fl->u.ip4;
3300	int oif = 0;
3301
3302	if (skb_dst(skb) && skb_dst(skb)->dev)
3303		oif = skb_dst(skb)->dev->ifindex;
3304
3305	memset(fl4, 0, sizeof(struct flowi4));
3306	fl4->flowi4_mark = skb->mark;
3307	fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3308
3309	fl4->flowi4_proto = iph->protocol;
3310	fl4->daddr = reverse ? iph->saddr : iph->daddr;
3311	fl4->saddr = reverse ? iph->daddr : iph->saddr;
3312	fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3313
3314	if (!ip_is_fragment(iph)) {
3315		switch (iph->protocol) {
3316		case IPPROTO_UDP:
3317		case IPPROTO_UDPLITE:
3318		case IPPROTO_TCP:
3319		case IPPROTO_SCTP:
3320		case IPPROTO_DCCP:
3321			if (xprth + 4 < skb->data ||
3322			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3323				__be16 *ports;
3324
3325				xprth = skb_network_header(skb) + ihl * 4;
3326				ports = (__be16 *)xprth;
3327
3328				fl4->fl4_sport = ports[!!reverse];
3329				fl4->fl4_dport = ports[!reverse];
3330			}
3331			break;
3332		case IPPROTO_ICMP:
3333			if (xprth + 2 < skb->data ||
3334			    pskb_may_pull(skb, xprth + 2 - skb->data)) {
3335				u8 *icmp;
3336
3337				xprth = skb_network_header(skb) + ihl * 4;
3338				icmp = xprth;
3339
3340				fl4->fl4_icmp_type = icmp[0];
3341				fl4->fl4_icmp_code = icmp[1];
3342			}
3343			break;
3344		case IPPROTO_ESP:
3345			if (xprth + 4 < skb->data ||
3346			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3347				__be32 *ehdr;
3348
3349				xprth = skb_network_header(skb) + ihl * 4;
3350				ehdr = (__be32 *)xprth;
3351
3352				fl4->fl4_ipsec_spi = ehdr[0];
3353			}
3354			break;
3355		case IPPROTO_AH:
3356			if (xprth + 8 < skb->data ||
3357			    pskb_may_pull(skb, xprth + 8 - skb->data)) {
3358				__be32 *ah_hdr;
3359
3360				xprth = skb_network_header(skb) + ihl * 4;
3361				ah_hdr = (__be32 *)xprth;
3362
3363				fl4->fl4_ipsec_spi = ah_hdr[1];
3364			}
3365			break;
3366		case IPPROTO_COMP:
3367			if (xprth + 4 < skb->data ||
3368			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
3369				__be16 *ipcomp_hdr;
3370
3371				xprth = skb_network_header(skb) + ihl * 4;
3372				ipcomp_hdr = (__be16 *)xprth;
3373
3374				fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
3375			}
3376			break;
3377		case IPPROTO_GRE:
3378			if (xprth + 12 < skb->data ||
3379			    pskb_may_pull(skb, xprth + 12 - skb->data)) {
3380				__be16 *greflags;
3381				__be32 *gre_hdr;
3382
3383				xprth = skb_network_header(skb) + ihl * 4;
3384				greflags = (__be16 *)xprth;
3385				gre_hdr = (__be32 *)xprth;
3386
3387				if (greflags[0] & GRE_KEY) {
3388					if (greflags[0] & GRE_CSUM)
3389						gre_hdr++;
3390					fl4->fl4_gre_key = gre_hdr[1];
3391				}
3392			}
3393			break;
3394		default:
3395			fl4->fl4_ipsec_spi = 0;
3396			break;
3397		}
3398	}
3399}
3400
3401#if IS_ENABLED(CONFIG_IPV6)
3402static void
3403decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3404{
3405	struct flowi6 *fl6 = &fl->u.ip6;
3406	int onlyproto = 0;
3407	const struct ipv6hdr *hdr = ipv6_hdr(skb);
3408	u32 offset = sizeof(*hdr);
3409	struct ipv6_opt_hdr *exthdr;
3410	const unsigned char *nh = skb_network_header(skb);
3411	u16 nhoff = IP6CB(skb)->nhoff;
3412	int oif = 0;
3413	u8 nexthdr;
3414
3415	if (!nhoff)
3416		nhoff = offsetof(struct ipv6hdr, nexthdr);
3417
3418	nexthdr = nh[nhoff];
3419
3420	if (skb_dst(skb) && skb_dst(skb)->dev)
3421		oif = skb_dst(skb)->dev->ifindex;
3422
3423	memset(fl6, 0, sizeof(struct flowi6));
3424	fl6->flowi6_mark = skb->mark;
3425	fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3426
3427	fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3428	fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3429
3430	while (nh + offset + sizeof(*exthdr) < skb->data ||
3431	       pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3432		nh = skb_network_header(skb);
3433		exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3434
3435		switch (nexthdr) {
3436		case NEXTHDR_FRAGMENT:
3437			onlyproto = 1;
3438			fallthrough;
3439		case NEXTHDR_ROUTING:
3440		case NEXTHDR_HOP:
3441		case NEXTHDR_DEST:
3442			offset += ipv6_optlen(exthdr);
3443			nexthdr = exthdr->nexthdr;
3444			exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3445			break;
3446		case IPPROTO_UDP:
3447		case IPPROTO_UDPLITE:
3448		case IPPROTO_TCP:
3449		case IPPROTO_SCTP:
3450		case IPPROTO_DCCP:
3451			if (!onlyproto && (nh + offset + 4 < skb->data ||
3452			     pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3453				__be16 *ports;
3454
3455				nh = skb_network_header(skb);
3456				ports = (__be16 *)(nh + offset);
3457				fl6->fl6_sport = ports[!!reverse];
3458				fl6->fl6_dport = ports[!reverse];
3459			}
3460			fl6->flowi6_proto = nexthdr;
3461			return;
3462		case IPPROTO_ICMPV6:
3463			if (!onlyproto && (nh + offset + 2 < skb->data ||
3464			    pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3465				u8 *icmp;
3466
3467				nh = skb_network_header(skb);
3468				icmp = (u8 *)(nh + offset);
3469				fl6->fl6_icmp_type = icmp[0];
3470				fl6->fl6_icmp_code = icmp[1];
3471			}
3472			fl6->flowi6_proto = nexthdr;
3473			return;
3474		case IPPROTO_GRE:
3475			if (!onlyproto &&
3476			    (nh + offset + 12 < skb->data ||
3477			     pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3478				struct gre_base_hdr *gre_hdr;
3479				__be32 *gre_key;
3480
3481				nh = skb_network_header(skb);
3482				gre_hdr = (struct gre_base_hdr *)(nh + offset);
3483				gre_key = (__be32 *)(gre_hdr + 1);
3484
3485				if (gre_hdr->flags & GRE_KEY) {
3486					if (gre_hdr->flags & GRE_CSUM)
3487						gre_key++;
3488					fl6->fl6_gre_key = *gre_key;
3489				}
3490			}
3491			fl6->flowi6_proto = nexthdr;
3492			return;
3493
3494#if IS_ENABLED(CONFIG_IPV6_MIP6)
3495		case IPPROTO_MH:
3496			offset += ipv6_optlen(exthdr);
3497			if (!onlyproto && (nh + offset + 3 < skb->data ||
3498			    pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3499				struct ip6_mh *mh;
3500
3501				nh = skb_network_header(skb);
3502				mh = (struct ip6_mh *)(nh + offset);
3503				fl6->fl6_mh_type = mh->ip6mh_type;
3504			}
3505			fl6->flowi6_proto = nexthdr;
3506			return;
3507#endif
3508		/* XXX Why are there these headers? */
3509		case IPPROTO_AH:
3510		case IPPROTO_ESP:
3511		case IPPROTO_COMP:
3512		default:
3513			fl6->fl6_ipsec_spi = 0;
3514			fl6->flowi6_proto = nexthdr;
3515			return;
3516		}
3517	}
3518}
3519#endif
3520
3521int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3522			  unsigned int family, int reverse)
3523{
3524	switch (family) {
3525	case AF_INET:
3526		decode_session4(skb, fl, reverse);
3527		break;
3528#if IS_ENABLED(CONFIG_IPV6)
3529	case AF_INET6:
3530		decode_session6(skb, fl, reverse);
3531		break;
3532#endif
3533	default:
3534		return -EAFNOSUPPORT;
3535	}
3536
3537	return security_xfrm_decode_session(skb, &fl->flowi_secid);
3538}
3539EXPORT_SYMBOL(__xfrm_decode_session);
3540
3541static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3542{
3543	for (; k < sp->len; k++) {
3544		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3545			*idxp = k;
3546			return 1;
3547		}
3548	}
3549
3550	return 0;
3551}
3552
3553int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3554			unsigned short family)
3555{
3556	struct net *net = dev_net(skb->dev);
3557	struct xfrm_policy *pol;
3558	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3559	int npols = 0;
3560	int xfrm_nr;
3561	int pi;
3562	int reverse;
3563	struct flowi fl;
3564	int xerr_idx = -1;
3565	const struct xfrm_if_cb *ifcb;
3566	struct sec_path *sp;
3567	struct xfrm_if *xi;
3568	u32 if_id = 0;
3569
3570	rcu_read_lock();
3571	ifcb = xfrm_if_get_cb();
3572
3573	if (ifcb) {
3574		xi = ifcb->decode_session(skb, family);
3575		if (xi) {
3576			if_id = xi->p.if_id;
3577			net = xi->net;
3578		}
3579	}
3580	rcu_read_unlock();
3581
3582	reverse = dir & ~XFRM_POLICY_MASK;
3583	dir &= XFRM_POLICY_MASK;
3584
3585	if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3586		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3587		return 0;
3588	}
3589
3590	nf_nat_decode_session(skb, &fl, family);
3591
3592	/* First, check used SA against their selectors. */
3593	sp = skb_sec_path(skb);
3594	if (sp) {
3595		int i;
3596
3597		for (i = sp->len - 1; i >= 0; i--) {
3598			struct xfrm_state *x = sp->xvec[i];
3599			if (!xfrm_selector_match(&x->sel, &fl, family)) {
3600				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3601				return 0;
3602			}
3603		}
3604	}
3605
3606	pol = NULL;
3607	sk = sk_to_full_sk(sk);
3608	if (sk && sk->sk_policy[dir]) {
3609		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3610		if (IS_ERR(pol)) {
3611			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3612			return 0;
3613		}
3614	}
3615
3616	if (!pol)
3617		pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3618
3619	if (IS_ERR(pol)) {
3620		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3621		return 0;
3622	}
3623
3624	if (!pol) {
3625		if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3626			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3627			return 0;
3628		}
3629
3630		if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3631			xfrm_secpath_reject(xerr_idx, skb, &fl);
3632			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3633			return 0;
3634		}
3635		return 1;
3636	}
3637
3638	pol->curlft.use_time = ktime_get_real_seconds();
3639
3640	pols[0] = pol;
3641	npols++;
3642#ifdef CONFIG_XFRM_SUB_POLICY
3643	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3644		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3645						    &fl, family,
3646						    XFRM_POLICY_IN, if_id);
3647		if (pols[1]) {
3648			if (IS_ERR(pols[1])) {
3649				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3650				xfrm_pol_put(pols[0]);
3651				return 0;
3652			}
3653			pols[1]->curlft.use_time = ktime_get_real_seconds();
3654			npols++;
3655		}
3656	}
3657#endif
3658
3659	if (pol->action == XFRM_POLICY_ALLOW) {
3660		static struct sec_path dummy;
3661		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3662		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3663		struct xfrm_tmpl **tpp = tp;
3664		int ti = 0;
3665		int i, k;
3666
3667		sp = skb_sec_path(skb);
3668		if (!sp)
3669			sp = &dummy;
3670
3671		for (pi = 0; pi < npols; pi++) {
3672			if (pols[pi] != pol &&
3673			    pols[pi]->action != XFRM_POLICY_ALLOW) {
3674				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3675				goto reject;
3676			}
3677			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3678				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3679				goto reject_error;
3680			}
3681			for (i = 0; i < pols[pi]->xfrm_nr; i++)
3682				tpp[ti++] = &pols[pi]->xfrm_vec[i];
3683		}
3684		xfrm_nr = ti;
3685
3686		if (npols > 1) {
3687			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3688			tpp = stp;
3689		}
3690
3691		/* For each tunnel xfrm, find the first matching tmpl.
3692		 * For each tmpl before that, find corresponding xfrm.
3693		 * Order is _important_. Later we will implement
3694		 * some barriers, but at the moment barriers
3695		 * are implied between each two transformations.
3696		 * Upon success, marks secpath entries as having been
3697		 * verified to allow them to be skipped in future policy
3698		 * checks (e.g. nested tunnels).
3699		 */
3700		for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3701			k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3702			if (k < 0) {
3703				if (k < -1)
3704					/* "-2 - errored_index" returned */
3705					xerr_idx = -(2+k);
3706				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3707				goto reject;
3708			}
3709		}
3710
3711		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3712			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3713			goto reject;
3714		}
3715
3716		xfrm_pols_put(pols, npols);
3717		sp->verified_cnt = k;
3718
3719		return 1;
3720	}
3721	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3722
3723reject:
3724	xfrm_secpath_reject(xerr_idx, skb, &fl);
3725reject_error:
3726	xfrm_pols_put(pols, npols);
3727	return 0;
3728}
3729EXPORT_SYMBOL(__xfrm_policy_check);
3730
3731int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3732{
3733	struct net *net = dev_net(skb->dev);
3734	struct flowi fl;
3735	struct dst_entry *dst;
3736	int res = 1;
3737
3738	if (xfrm_decode_session(skb, &fl, family) < 0) {
3739		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3740		return 0;
3741	}
3742
3743	skb_dst_force(skb);
3744	if (!skb_dst(skb)) {
3745		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3746		return 0;
3747	}
3748
3749	dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3750	if (IS_ERR(dst)) {
3751		res = 0;
3752		dst = NULL;
3753	}
3754	skb_dst_set(skb, dst);
3755	return res;
3756}
3757EXPORT_SYMBOL(__xfrm_route_forward);
3758
3759/* Optimize later using cookies and generation ids. */
3760
3761static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3762{
3763	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3764	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3765	 * get validated by dst_ops->check on every use.  We do this
3766	 * because when a normal route referenced by an XFRM dst is
3767	 * obsoleted we do not go looking around for all parent
3768	 * referencing XFRM dsts so that we can invalidate them.  It
3769	 * is just too much work.  Instead we make the checks here on
3770	 * every use.  For example:
3771	 *
3772	 *	XFRM dst A --> IPv4 dst X
3773	 *
3774	 * X is the "xdst->route" of A (X is also the "dst->path" of A
3775	 * in this example).  If X is marked obsolete, "A" will not
3776	 * notice.  That's what we are validating here via the
3777	 * stale_bundle() check.
3778	 *
3779	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3780	 * be marked on it.
3781	 * This will force stale_bundle() to fail on any xdst bundle with
3782	 * this dst linked in it.
3783	 */
3784	if (dst->obsolete < 0 && !stale_bundle(dst))
3785		return dst;
3786
3787	return NULL;
3788}
3789
3790static int stale_bundle(struct dst_entry *dst)
3791{
3792	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3793}
3794
3795void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3796{
3797	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3798		dst->dev = dev_net(dev)->loopback_dev;
3799		dev_hold(dst->dev);
3800		dev_put(dev);
3801	}
3802}
3803EXPORT_SYMBOL(xfrm_dst_ifdown);
3804
3805static void xfrm_link_failure(struct sk_buff *skb)
3806{
3807	/* Impossible. Such dst must be popped before reaches point of failure. */
3808}
3809
3810static void xfrm_negative_advice(struct sock *sk, struct dst_entry *dst)
3811{
3812	if (dst->obsolete)
3813		sk_dst_reset(sk);
3814}
3815
3816static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3817{
3818	while (nr--) {
3819		struct xfrm_dst *xdst = bundle[nr];
3820		u32 pmtu, route_mtu_cached;
3821		struct dst_entry *dst;
3822
3823		dst = &xdst->u.dst;
3824		pmtu = dst_mtu(xfrm_dst_child(dst));
3825		xdst->child_mtu_cached = pmtu;
3826
3827		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3828
3829		route_mtu_cached = dst_mtu(xdst->route);
3830		xdst->route_mtu_cached = route_mtu_cached;
3831
3832		if (pmtu > route_mtu_cached)
3833			pmtu = route_mtu_cached;
3834
3835		dst_metric_set(dst, RTAX_MTU, pmtu);
3836	}
3837}
3838
3839/* Check that the bundle accepts the flow and its components are
3840 * still valid.
3841 */
3842
3843static int xfrm_bundle_ok(struct xfrm_dst *first)
3844{
3845	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3846	struct dst_entry *dst = &first->u.dst;
3847	struct xfrm_dst *xdst;
3848	int start_from, nr;
3849	u32 mtu;
3850
3851	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3852	    (dst->dev && !netif_running(dst->dev)))
3853		return 0;
3854
3855	if (dst->flags & DST_XFRM_QUEUE)
3856		return 1;
3857
3858	start_from = nr = 0;
3859	do {
3860		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3861
3862		if (dst->xfrm->km.state != XFRM_STATE_VALID)
3863			return 0;
3864		if (xdst->xfrm_genid != dst->xfrm->genid)
3865			return 0;
3866		if (xdst->num_pols > 0 &&
3867		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3868			return 0;
3869
3870		bundle[nr++] = xdst;
3871
3872		mtu = dst_mtu(xfrm_dst_child(dst));
3873		if (xdst->child_mtu_cached != mtu) {
3874			start_from = nr;
3875			xdst->child_mtu_cached = mtu;
3876		}
3877
3878		if (!dst_check(xdst->route, xdst->route_cookie))
3879			return 0;
3880		mtu = dst_mtu(xdst->route);
3881		if (xdst->route_mtu_cached != mtu) {
3882			start_from = nr;
3883			xdst->route_mtu_cached = mtu;
3884		}
3885
3886		dst = xfrm_dst_child(dst);
3887	} while (dst->xfrm);
3888
3889	if (likely(!start_from))
3890		return 1;
3891
3892	xdst = bundle[start_from - 1];
3893	mtu = xdst->child_mtu_cached;
3894	while (start_from--) {
3895		dst = &xdst->u.dst;
3896
3897		mtu = xfrm_state_mtu(dst->xfrm, mtu);
3898		if (mtu > xdst->route_mtu_cached)
3899			mtu = xdst->route_mtu_cached;
3900		dst_metric_set(dst, RTAX_MTU, mtu);
3901		if (!start_from)
3902			break;
3903
3904		xdst = bundle[start_from - 1];
3905		xdst->child_mtu_cached = mtu;
3906	}
3907
3908	return 1;
3909}
3910
3911static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3912{
3913	return dst_metric_advmss(xfrm_dst_path(dst));
3914}
3915
3916static unsigned int xfrm_mtu(const struct dst_entry *dst)
3917{
3918	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3919
3920	return mtu ? : dst_mtu(xfrm_dst_path(dst));
3921}
3922
3923static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3924					const void *daddr)
3925{
3926	while (dst->xfrm) {
3927		const struct xfrm_state *xfrm = dst->xfrm;
3928
3929		dst = xfrm_dst_child(dst);
3930
3931		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3932			continue;
3933		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3934			daddr = xfrm->coaddr;
3935		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3936			daddr = &xfrm->id.daddr;
3937	}
3938	return daddr;
3939}
3940
3941static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3942					   struct sk_buff *skb,
3943					   const void *daddr)
3944{
3945	const struct dst_entry *path = xfrm_dst_path(dst);
3946
3947	if (!skb)
3948		daddr = xfrm_get_dst_nexthop(dst, daddr);
3949	return path->ops->neigh_lookup(path, skb, daddr);
3950}
3951
3952static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3953{
3954	const struct dst_entry *path = xfrm_dst_path(dst);
3955
3956	daddr = xfrm_get_dst_nexthop(dst, daddr);
3957	path->ops->confirm_neigh(path, daddr);
3958}
3959
3960int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3961{
3962	int err = 0;
3963
3964	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3965		return -EAFNOSUPPORT;
3966
3967	spin_lock(&xfrm_policy_afinfo_lock);
3968	if (unlikely(xfrm_policy_afinfo[family] != NULL))
3969		err = -EEXIST;
3970	else {
3971		struct dst_ops *dst_ops = afinfo->dst_ops;
3972		if (likely(dst_ops->kmem_cachep == NULL))
3973			dst_ops->kmem_cachep = xfrm_dst_cache;
3974		if (likely(dst_ops->check == NULL))
3975			dst_ops->check = xfrm_dst_check;
3976		if (likely(dst_ops->default_advmss == NULL))
3977			dst_ops->default_advmss = xfrm_default_advmss;
3978		if (likely(dst_ops->mtu == NULL))
3979			dst_ops->mtu = xfrm_mtu;
3980		if (likely(dst_ops->negative_advice == NULL))
3981			dst_ops->negative_advice = xfrm_negative_advice;
3982		if (likely(dst_ops->link_failure == NULL))
3983			dst_ops->link_failure = xfrm_link_failure;
3984		if (likely(dst_ops->neigh_lookup == NULL))
3985			dst_ops->neigh_lookup = xfrm_neigh_lookup;
3986		if (likely(!dst_ops->confirm_neigh))
3987			dst_ops->confirm_neigh = xfrm_confirm_neigh;
3988		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3989	}
3990	spin_unlock(&xfrm_policy_afinfo_lock);
3991
3992	return err;
3993}
3994EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3995
3996void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3997{
3998	struct dst_ops *dst_ops = afinfo->dst_ops;
3999	int i;
4000
4001	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
4002		if (xfrm_policy_afinfo[i] != afinfo)
4003			continue;
4004		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
4005		break;
4006	}
4007
4008	synchronize_rcu();
4009
4010	dst_ops->kmem_cachep = NULL;
4011	dst_ops->check = NULL;
4012	dst_ops->negative_advice = NULL;
4013	dst_ops->link_failure = NULL;
4014}
4015EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
4016
4017void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
4018{
4019	spin_lock(&xfrm_if_cb_lock);
4020	rcu_assign_pointer(xfrm_if_cb, ifcb);
4021	spin_unlock(&xfrm_if_cb_lock);
4022}
4023EXPORT_SYMBOL(xfrm_if_register_cb);
4024
4025void xfrm_if_unregister_cb(void)
4026{
4027	RCU_INIT_POINTER(xfrm_if_cb, NULL);
4028	synchronize_rcu();
4029}
4030EXPORT_SYMBOL(xfrm_if_unregister_cb);
4031
4032#ifdef CONFIG_XFRM_STATISTICS
4033static int __net_init xfrm_statistics_init(struct net *net)
4034{
4035	int rv;
4036	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4037	if (!net->mib.xfrm_statistics)
4038		return -ENOMEM;
4039	rv = xfrm_proc_init(net);
4040	if (rv < 0)
4041		free_percpu(net->mib.xfrm_statistics);
4042	return rv;
4043}
4044
4045static void xfrm_statistics_fini(struct net *net)
4046{
4047	xfrm_proc_fini(net);
4048	free_percpu(net->mib.xfrm_statistics);
4049}
4050#else
4051static int __net_init xfrm_statistics_init(struct net *net)
4052{
4053	return 0;
4054}
4055
4056static void xfrm_statistics_fini(struct net *net)
4057{
4058}
4059#endif
4060
4061static int __net_init xfrm_policy_init(struct net *net)
4062{
4063	unsigned int hmask, sz;
4064	int dir, err;
4065
4066	if (net_eq(net, &init_net)) {
4067		xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4068					   sizeof(struct xfrm_dst),
4069					   0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4070					   NULL);
4071		err = rhashtable_init(&xfrm_policy_inexact_table,
4072				      &xfrm_pol_inexact_params);
4073		BUG_ON(err);
4074	}
4075
4076	hmask = 8 - 1;
4077	sz = (hmask+1) * sizeof(struct hlist_head);
4078
4079	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4080	if (!net->xfrm.policy_byidx)
4081		goto out_byidx;
4082	net->xfrm.policy_idx_hmask = hmask;
4083
4084	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4085		struct xfrm_policy_hash *htab;
4086
4087		net->xfrm.policy_count[dir] = 0;
4088		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4089		INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4090
4091		htab = &net->xfrm.policy_bydst[dir];
4092		htab->table = xfrm_hash_alloc(sz);
4093		if (!htab->table)
4094			goto out_bydst;
4095		htab->hmask = hmask;
4096		htab->dbits4 = 32;
4097		htab->sbits4 = 32;
4098		htab->dbits6 = 128;
4099		htab->sbits6 = 128;
4100	}
4101	net->xfrm.policy_hthresh.lbits4 = 32;
4102	net->xfrm.policy_hthresh.rbits4 = 32;
4103	net->xfrm.policy_hthresh.lbits6 = 128;
4104	net->xfrm.policy_hthresh.rbits6 = 128;
4105
4106	seqlock_init(&net->xfrm.policy_hthresh.lock);
4107
4108	INIT_LIST_HEAD(&net->xfrm.policy_all);
4109	INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4110	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4111	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4112	return 0;
4113
4114out_bydst:
4115	for (dir--; dir >= 0; dir--) {
4116		struct xfrm_policy_hash *htab;
4117
4118		htab = &net->xfrm.policy_bydst[dir];
4119		xfrm_hash_free(htab->table, sz);
4120	}
4121	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4122out_byidx:
4123	return -ENOMEM;
4124}
4125
4126static void xfrm_policy_fini(struct net *net)
4127{
4128	struct xfrm_pol_inexact_bin *b, *t;
4129	unsigned int sz;
4130	int dir;
4131
4132	flush_work(&net->xfrm.policy_hash_work);
4133#ifdef CONFIG_XFRM_SUB_POLICY
4134	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4135#endif
4136	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4137
4138	WARN_ON(!list_empty(&net->xfrm.policy_all));
4139
4140	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4141		struct xfrm_policy_hash *htab;
4142
4143		WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4144
4145		htab = &net->xfrm.policy_bydst[dir];
4146		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4147		WARN_ON(!hlist_empty(htab->table));
4148		xfrm_hash_free(htab->table, sz);
4149	}
4150
4151	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4152	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4153	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4154
4155	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4156	list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4157		__xfrm_policy_inexact_prune_bin(b, true);
4158	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4159}
4160
4161static int __net_init xfrm_net_init(struct net *net)
4162{
4163	int rv;
4164
4165	/* Initialize the per-net locks here */
4166	spin_lock_init(&net->xfrm.xfrm_state_lock);
4167	spin_lock_init(&net->xfrm.xfrm_policy_lock);
4168	seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4169	mutex_init(&net->xfrm.xfrm_cfg_mutex);
4170	net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4171	net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4172	net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4173
4174	rv = xfrm_statistics_init(net);
4175	if (rv < 0)
4176		goto out_statistics;
4177	rv = xfrm_state_init(net);
4178	if (rv < 0)
4179		goto out_state;
4180	rv = xfrm_policy_init(net);
4181	if (rv < 0)
4182		goto out_policy;
4183	rv = xfrm_sysctl_init(net);
4184	if (rv < 0)
4185		goto out_sysctl;
4186
4187	return 0;
4188
4189out_sysctl:
4190	xfrm_policy_fini(net);
4191out_policy:
4192	xfrm_state_fini(net);
4193out_state:
4194	xfrm_statistics_fini(net);
4195out_statistics:
4196	return rv;
4197}
4198
4199static void __net_exit xfrm_net_exit(struct net *net)
4200{
4201	xfrm_sysctl_fini(net);
4202	xfrm_policy_fini(net);
4203	xfrm_state_fini(net);
4204	xfrm_statistics_fini(net);
4205}
4206
4207static struct pernet_operations __net_initdata xfrm_net_ops = {
4208	.init = xfrm_net_init,
4209	.exit = xfrm_net_exit,
4210};
4211
4212void __init xfrm_init(void)
4213{
4214	register_pernet_subsys(&xfrm_net_ops);
4215	xfrm_dev_init();
4216	xfrm_input_init();
4217
4218#ifdef CONFIG_XFRM_ESPINTCP
4219	espintcp_init();
4220#endif
4221
4222	RCU_INIT_POINTER(xfrm_if_cb, NULL);
4223	synchronize_rcu();
4224}
4225
4226#ifdef CONFIG_AUDITSYSCALL
4227static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4228					 struct audit_buffer *audit_buf)
4229{
4230	struct xfrm_sec_ctx *ctx = xp->security;
4231	struct xfrm_selector *sel = &xp->selector;
4232
4233	if (ctx)
4234		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4235				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4236
4237	switch (sel->family) {
4238	case AF_INET:
4239		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4240		if (sel->prefixlen_s != 32)
4241			audit_log_format(audit_buf, " src_prefixlen=%d",
4242					 sel->prefixlen_s);
4243		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4244		if (sel->prefixlen_d != 32)
4245			audit_log_format(audit_buf, " dst_prefixlen=%d",
4246					 sel->prefixlen_d);
4247		break;
4248	case AF_INET6:
4249		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4250		if (sel->prefixlen_s != 128)
4251			audit_log_format(audit_buf, " src_prefixlen=%d",
4252					 sel->prefixlen_s);
4253		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4254		if (sel->prefixlen_d != 128)
4255			audit_log_format(audit_buf, " dst_prefixlen=%d",
4256					 sel->prefixlen_d);
4257		break;
4258	}
4259}
4260
4261void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4262{
4263	struct audit_buffer *audit_buf;
4264
4265	audit_buf = xfrm_audit_start("SPD-add");
4266	if (audit_buf == NULL)
4267		return;
4268	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4269	audit_log_format(audit_buf, " res=%u", result);
4270	xfrm_audit_common_policyinfo(xp, audit_buf);
4271	audit_log_end(audit_buf);
4272}
4273EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4274
4275void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4276			      bool task_valid)
4277{
4278	struct audit_buffer *audit_buf;
4279
4280	audit_buf = xfrm_audit_start("SPD-delete");
4281	if (audit_buf == NULL)
4282		return;
4283	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4284	audit_log_format(audit_buf, " res=%u", result);
4285	xfrm_audit_common_policyinfo(xp, audit_buf);
4286	audit_log_end(audit_buf);
4287}
4288EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4289#endif
4290
4291#ifdef CONFIG_XFRM_MIGRATE
4292static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4293					const struct xfrm_selector *sel_tgt)
4294{
4295	if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4296		if (sel_tgt->family == sel_cmp->family &&
4297		    xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4298				    sel_cmp->family) &&
4299		    xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4300				    sel_cmp->family) &&
4301		    sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4302		    sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4303			return true;
4304		}
4305	} else {
4306		if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4307			return true;
4308		}
4309	}
4310	return false;
4311}
4312
4313static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4314						    u8 dir, u8 type, struct net *net, u32 if_id)
4315{
4316	struct xfrm_policy *pol, *ret = NULL;
4317	struct hlist_head *chain;
4318	u32 priority = ~0U;
4319
4320	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4321	chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4322	hlist_for_each_entry(pol, chain, bydst) {
4323		if ((if_id == 0 || pol->if_id == if_id) &&
4324		    xfrm_migrate_selector_match(sel, &pol->selector) &&
4325		    pol->type == type) {
4326			ret = pol;
4327			priority = ret->priority;
4328			break;
4329		}
4330	}
4331	chain = &net->xfrm.policy_inexact[dir];
4332	hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4333		if ((pol->priority >= priority) && ret)
4334			break;
4335
4336		if ((if_id == 0 || pol->if_id == if_id) &&
4337		    xfrm_migrate_selector_match(sel, &pol->selector) &&
4338		    pol->type == type) {
4339			ret = pol;
4340			break;
4341		}
4342	}
4343
4344	xfrm_pol_hold(ret);
4345
4346	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4347
4348	return ret;
4349}
4350
4351static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4352{
4353	int match = 0;
4354
4355	if (t->mode == m->mode && t->id.proto == m->proto &&
4356	    (m->reqid == 0 || t->reqid == m->reqid)) {
4357		switch (t->mode) {
4358		case XFRM_MODE_TUNNEL:
4359		case XFRM_MODE_BEET:
4360			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4361					    m->old_family) &&
4362			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
4363					    m->old_family)) {
4364				match = 1;
4365			}
4366			break;
4367		case XFRM_MODE_TRANSPORT:
4368			/* in case of transport mode, template does not store
4369			   any IP addresses, hence we just compare mode and
4370			   protocol */
4371			match = 1;
4372			break;
4373		default:
4374			break;
4375		}
4376	}
4377	return match;
4378}
4379
4380/* update endpoint address(es) of template(s) */
4381static int xfrm_policy_migrate(struct xfrm_policy *pol,
4382			       struct xfrm_migrate *m, int num_migrate)
4383{
4384	struct xfrm_migrate *mp;
4385	int i, j, n = 0;
4386
4387	write_lock_bh(&pol->lock);
4388	if (unlikely(pol->walk.dead)) {
4389		/* target policy has been deleted */
4390		write_unlock_bh(&pol->lock);
4391		return -ENOENT;
4392	}
4393
4394	for (i = 0; i < pol->xfrm_nr; i++) {
4395		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4396			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4397				continue;
4398			n++;
4399			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4400			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4401				continue;
4402			/* update endpoints */
4403			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4404			       sizeof(pol->xfrm_vec[i].id.daddr));
4405			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4406			       sizeof(pol->xfrm_vec[i].saddr));
4407			pol->xfrm_vec[i].encap_family = mp->new_family;
4408			/* flush bundles */
4409			atomic_inc(&pol->genid);
4410		}
4411	}
4412
4413	write_unlock_bh(&pol->lock);
4414
4415	if (!n)
4416		return -ENODATA;
4417
4418	return 0;
4419}
4420
4421static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4422{
4423	int i, j;
4424
4425	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4426		return -EINVAL;
4427
4428	for (i = 0; i < num_migrate; i++) {
4429		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4430		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4431			return -EINVAL;
4432
4433		/* check if there is any duplicated entry */
4434		for (j = i + 1; j < num_migrate; j++) {
4435			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4436				    sizeof(m[i].old_daddr)) &&
4437			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4438				    sizeof(m[i].old_saddr)) &&
4439			    m[i].proto == m[j].proto &&
4440			    m[i].mode == m[j].mode &&
4441			    m[i].reqid == m[j].reqid &&
4442			    m[i].old_family == m[j].old_family)
4443				return -EINVAL;
4444		}
4445	}
4446
4447	return 0;
4448}
4449
4450int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4451		 struct xfrm_migrate *m, int num_migrate,
4452		 struct xfrm_kmaddress *k, struct net *net,
4453		 struct xfrm_encap_tmpl *encap, u32 if_id)
4454{
4455	int i, err, nx_cur = 0, nx_new = 0;
4456	struct xfrm_policy *pol = NULL;
4457	struct xfrm_state *x, *xc;
4458	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4459	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4460	struct xfrm_migrate *mp;
4461
4462	/* Stage 0 - sanity checks */
4463	if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4464		goto out;
4465
4466	if (dir >= XFRM_POLICY_MAX) {
4467		err = -EINVAL;
4468		goto out;
4469	}
4470
4471	/* Stage 1 - find policy */
4472	if ((pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id)) == NULL) {
4473		err = -ENOENT;
4474		goto out;
4475	}
4476
4477	/* Stage 2 - find and update state(s) */
4478	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4479		if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4480			x_cur[nx_cur] = x;
4481			nx_cur++;
4482			xc = xfrm_state_migrate(x, mp, encap);
4483			if (xc) {
4484				x_new[nx_new] = xc;
4485				nx_new++;
4486			} else {
4487				err = -ENODATA;
4488				goto restore_state;
4489			}
4490		}
4491	}
4492
4493	/* Stage 3 - update policy */
4494	if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4495		goto restore_state;
4496
4497	/* Stage 4 - delete old state(s) */
4498	if (nx_cur) {
4499		xfrm_states_put(x_cur, nx_cur);
4500		xfrm_states_delete(x_cur, nx_cur);
4501	}
4502
4503	/* Stage 5 - announce */
4504	km_migrate(sel, dir, type, m, num_migrate, k, encap);
4505
4506	xfrm_pol_put(pol);
4507
4508	return 0;
4509out:
4510	return err;
4511
4512restore_state:
4513	if (pol)
4514		xfrm_pol_put(pol);
4515	if (nx_cur)
4516		xfrm_states_put(x_cur, nx_cur);
4517	if (nx_new)
4518		xfrm_states_delete(x_new, nx_new);
4519
4520	return err;
4521}
4522EXPORT_SYMBOL(xfrm_migrate);
4523#endif
4524