xfrm_policy.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * xfrm_policy.c
   4 *
   5 * Changes:
   6 *	Mitsuru KANDA @USAGI
   7 * 	Kazunori MIYAZAWA @USAGI
   8 * 	Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   9 * 		IPv6 support
  10 * 	Kazunori MIYAZAWA @USAGI
  11 * 	YOSHIFUJI Hideaki
  12 * 		Split up af-specific portion
  13 *	Derek Atkins <derek@ihtfp.com>		Add the post_input processor
  14 *
  15 */
  16
  17#include <linux/err.h>
  18#include <linux/slab.h>
  19#include <linux/kmod.h>
  20#include <linux/list.h>
  21#include <linux/spinlock.h>
  22#include <linux/workqueue.h>
  23#include <linux/notifier.h>
  24#include <linux/netdevice.h>
  25#include <linux/netfilter.h>
  26#include <linux/module.h>
  27#include <linux/cache.h>
  28#include <linux/cpu.h>
  29#include <linux/audit.h>
  30#include <linux/rhashtable.h>
  31#include <linux/if_tunnel.h>
  32#include <linux/icmp.h>
  33#include <net/dst.h>
  34#include <net/flow.h>
  35#include <net/inet_ecn.h>
  36#include <net/xfrm.h>
  37#include <net/ip.h>
  38#include <net/gre.h>
  39#if IS_ENABLED(CONFIG_IPV6_MIP6)
  40#include <net/mip6.h>
  41#endif
  42#ifdef CONFIG_XFRM_STATISTICS
  43#include <net/snmp.h>
  44#endif
  45#ifdef CONFIG_XFRM_ESPINTCP
  46#include <net/espintcp.h>
  47#endif
  48#include <net/inet_dscp.h>
  49
  50#include "xfrm_hash.h"
  51
  52#define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  53#define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  54#define XFRM_MAX_QUEUE_LEN	100
  55
  56struct xfrm_flo {
  57	struct dst_entry *dst_orig;
  58	u8 flags;
  59};
  60
  61/* prefixes smaller than this are stored in lists, not trees. */
  62#define INEXACT_PREFIXLEN_IPV4	16
  63#define INEXACT_PREFIXLEN_IPV6	48
  64
  65struct xfrm_pol_inexact_node {
  66	struct rb_node node;
  67	union {
  68		xfrm_address_t addr;
  69		struct rcu_head rcu;
  70	};
  71	u8 prefixlen;
  72
  73	struct rb_root root;
  74
  75	/* the policies matching this node, can be empty list */
  76	struct hlist_head hhead;
  77};
  78
  79/* xfrm inexact policy search tree:
  80 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
  81 *  |
  82 * +---- root_d: sorted by daddr:prefix
  83 * |                 |
  84 * |        xfrm_pol_inexact_node
  85 * |                 |
  86 * |                 +- root: sorted by saddr/prefix
  87 * |                 |              |
  88 * |                 |         xfrm_pol_inexact_node
  89 * |                 |              |
  90 * |                 |              + root: unused
  91 * |                 |              |
  92 * |                 |              + hhead: saddr:daddr policies
  93 * |                 |
  94 * |                 +- coarse policies and all any:daddr policies
  95 * |
  96 * +---- root_s: sorted by saddr:prefix
  97 * |                 |
  98 * |        xfrm_pol_inexact_node
  99 * |                 |
 100 * |                 + root: unused
 101 * |                 |
 102 * |                 + hhead: saddr:any policies
 103 * |
 104 * +---- coarse policies and all any:any policies
 105 *
 106 * Lookups return four candidate lists:
 107 * 1. any:any list from top-level xfrm_pol_inexact_bin
 108 * 2. any:daddr list from daddr tree
 109 * 3. saddr:daddr list from 2nd level daddr tree
 110 * 4. saddr:any list from saddr tree
 111 *
 112 * This result set then needs to be searched for the policy with
 113 * the lowest priority.  If two candidates have the same priority, the
 114 * struct xfrm_policy pos member with the lower number is used.
 115 *
 116 * This replicates previous single-list-search algorithm which would
 117 * return first matching policy in the (ordered-by-priority) list.
 118 */
 119
 120struct xfrm_pol_inexact_key {
 121	possible_net_t net;
 122	u32 if_id;
 123	u16 family;
 124	u8 dir, type;
 125};
 126
 127struct xfrm_pol_inexact_bin {
 128	struct xfrm_pol_inexact_key k;
 129	struct rhash_head head;
 130	/* list containing '*:*' policies */
 131	struct hlist_head hhead;
 132
 133	seqcount_spinlock_t count;
 134	/* tree sorted by daddr/prefix */
 135	struct rb_root root_d;
 136
 137	/* tree sorted by saddr/prefix */
 138	struct rb_root root_s;
 139
 140	/* slow path below */
 141	struct list_head inexact_bins;
 142	struct rcu_head rcu;
 143};
 144
 145enum xfrm_pol_inexact_candidate_type {
 146	XFRM_POL_CAND_BOTH,
 147	XFRM_POL_CAND_SADDR,
 148	XFRM_POL_CAND_DADDR,
 149	XFRM_POL_CAND_ANY,
 150
 151	XFRM_POL_CAND_MAX,
 152};
 153
 154struct xfrm_pol_inexact_candidates {
 155	struct hlist_head *res[XFRM_POL_CAND_MAX];
 156};
 157
 158struct xfrm_flow_keys {
 159	struct flow_dissector_key_basic basic;
 160	struct flow_dissector_key_control control;
 161	union {
 162		struct flow_dissector_key_ipv4_addrs ipv4;
 163		struct flow_dissector_key_ipv6_addrs ipv6;
 164	} addrs;
 165	struct flow_dissector_key_ip ip;
 166	struct flow_dissector_key_icmp icmp;
 167	struct flow_dissector_key_ports ports;
 168	struct flow_dissector_key_keyid gre;
 169};
 170
 171static struct flow_dissector xfrm_session_dissector __ro_after_init;
 172
 173static DEFINE_SPINLOCK(xfrm_if_cb_lock);
 174static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
 175
 176static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
 177static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
 178						__read_mostly;
 179
 180static struct kmem_cache *xfrm_dst_cache __ro_after_init;
 181
 182static struct rhashtable xfrm_policy_inexact_table;
 183static const struct rhashtable_params xfrm_pol_inexact_params;
 184
 185static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
 186static int stale_bundle(struct dst_entry *dst);
 187static int xfrm_bundle_ok(struct xfrm_dst *xdst);
 188static void xfrm_policy_queue_process(struct timer_list *t);
 189
 190static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
 191static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 192						int dir);
 193
 194static struct xfrm_pol_inexact_bin *
 195xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
 196			   u32 if_id);
 197
 198static struct xfrm_pol_inexact_bin *
 199xfrm_policy_inexact_lookup_rcu(struct net *net,
 200			       u8 type, u16 family, u8 dir, u32 if_id);
 201static struct xfrm_policy *
 202xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
 203			bool excl);
 204
 205static bool
 206xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
 207				    struct xfrm_pol_inexact_bin *b,
 208				    const xfrm_address_t *saddr,
 209				    const xfrm_address_t *daddr);
 210
 211static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
 212{
 213	return refcount_inc_not_zero(&policy->refcnt);
 214}
 215
 216static inline bool
 217__xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 218{
 219	const struct flowi4 *fl4 = &fl->u.ip4;
 220
 221	return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
 222		addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
 223		!((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
 224		!((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
 225		(fl4->flowi4_proto == sel->proto || !sel->proto) &&
 226		(fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
 227}
 228
 229static inline bool
 230__xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 231{
 232	const struct flowi6 *fl6 = &fl->u.ip6;
 233
 234	return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
 235		addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
 236		!((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
 237		!((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
 238		(fl6->flowi6_proto == sel->proto || !sel->proto) &&
 239		(fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
 240}
 241
 242bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
 243			 unsigned short family)
 244{
 245	switch (family) {
 246	case AF_INET:
 247		return __xfrm4_selector_match(sel, fl);
 248	case AF_INET6:
 249		return __xfrm6_selector_match(sel, fl);
 250	}
 251	return false;
 252}
 253
 254static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 255{
 256	const struct xfrm_policy_afinfo *afinfo;
 257
 258	if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 259		return NULL;
 260	rcu_read_lock();
 261	afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 262	if (unlikely(!afinfo))
 263		rcu_read_unlock();
 264	return afinfo;
 265}
 266
 267/* Called with rcu_read_lock(). */
 268static const struct xfrm_if_cb *xfrm_if_get_cb(void)
 269{
 270	return rcu_dereference(xfrm_if_cb);
 271}
 272
 273struct dst_entry *__xfrm_dst_lookup(int family,
 274				    const struct xfrm_dst_lookup_params *params)
 275{
 276	const struct xfrm_policy_afinfo *afinfo;
 277	struct dst_entry *dst;
 278
 279	afinfo = xfrm_policy_get_afinfo(family);
 280	if (unlikely(afinfo == NULL))
 281		return ERR_PTR(-EAFNOSUPPORT);
 282
 283	dst = afinfo->dst_lookup(params);
 284
 285	rcu_read_unlock();
 286
 287	return dst;
 288}
 289EXPORT_SYMBOL(__xfrm_dst_lookup);
 290
 291static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 292						dscp_t dscp, int oif,
 293						xfrm_address_t *prev_saddr,
 294						xfrm_address_t *prev_daddr,
 295						int family, u32 mark)
 296{
 297	struct xfrm_dst_lookup_params params;
 298	struct net *net = xs_net(x);
 299	xfrm_address_t *saddr = &x->props.saddr;
 300	xfrm_address_t *daddr = &x->id.daddr;
 301	struct dst_entry *dst;
 302
 303	if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 304		saddr = x->coaddr;
 305		daddr = prev_daddr;
 306	}
 307	if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 308		saddr = prev_saddr;
 309		daddr = x->coaddr;
 310	}
 311
 312	params.net = net;
 313	params.saddr = saddr;
 314	params.daddr = daddr;
 315	params.dscp = dscp;
 316	params.oif = oif;
 317	params.mark = mark;
 318	params.ipproto = x->id.proto;
 319	if (x->encap) {
 320		switch (x->encap->encap_type) {
 321		case UDP_ENCAP_ESPINUDP:
 322			params.ipproto = IPPROTO_UDP;
 323			params.uli.ports.sport = x->encap->encap_sport;
 324			params.uli.ports.dport = x->encap->encap_dport;
 325			break;
 326		case TCP_ENCAP_ESPINTCP:
 327			params.ipproto = IPPROTO_TCP;
 328			params.uli.ports.sport = x->encap->encap_sport;
 329			params.uli.ports.dport = x->encap->encap_dport;
 330			break;
 331		}
 332	}
 333
 334	dst = __xfrm_dst_lookup(family, &params);
 335
 336	if (!IS_ERR(dst)) {
 337		if (prev_saddr != saddr)
 338			memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 339		if (prev_daddr != daddr)
 340			memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 341	}
 342
 343	return dst;
 344}
 345
 346static inline unsigned long make_jiffies(long secs)
 347{
 348	if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 349		return MAX_SCHEDULE_TIMEOUT-1;
 350	else
 351		return secs*HZ;
 352}
 353
 354static void xfrm_policy_timer(struct timer_list *t)
 355{
 356	struct xfrm_policy *xp = timer_container_of(xp, t, timer);
 357	time64_t now = ktime_get_real_seconds();
 358	time64_t next = TIME64_MAX;
 359	int warn = 0;
 360	int dir;
 361
 362	read_lock(&xp->lock);
 363
 364	if (unlikely(xp->walk.dead))
 365		goto out;
 366
 367	dir = xfrm_policy_id2dir(xp->index);
 368
 369	if (xp->lft.hard_add_expires_seconds) {
 370		time64_t tmo = xp->lft.hard_add_expires_seconds +
 371			xp->curlft.add_time - now;
 372		if (tmo <= 0)
 373			goto expired;
 374		if (tmo < next)
 375			next = tmo;
 376	}
 377	if (xp->lft.hard_use_expires_seconds) {
 378		time64_t tmo = xp->lft.hard_use_expires_seconds +
 379			(READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
 380		if (tmo <= 0)
 381			goto expired;
 382		if (tmo < next)
 383			next = tmo;
 384	}
 385	if (xp->lft.soft_add_expires_seconds) {
 386		time64_t tmo = xp->lft.soft_add_expires_seconds +
 387			xp->curlft.add_time - now;
 388		if (tmo <= 0) {
 389			warn = 1;
 390			tmo = XFRM_KM_TIMEOUT;
 391		}
 392		if (tmo < next)
 393			next = tmo;
 394	}
 395	if (xp->lft.soft_use_expires_seconds) {
 396		time64_t tmo = xp->lft.soft_use_expires_seconds +
 397			(READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
 398		if (tmo <= 0) {
 399			warn = 1;
 400			tmo = XFRM_KM_TIMEOUT;
 401		}
 402		if (tmo < next)
 403			next = tmo;
 404	}
 405
 406	if (warn)
 407		km_policy_expired(xp, dir, 0, 0);
 408	if (next != TIME64_MAX &&
 409	    !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 410		xfrm_pol_hold(xp);
 411
 412out:
 413	read_unlock(&xp->lock);
 414	xfrm_pol_put(xp);
 415	return;
 416
 417expired:
 418	read_unlock(&xp->lock);
 419	if (!xfrm_policy_delete(xp, dir))
 420		km_policy_expired(xp, dir, 1, 0);
 421	xfrm_pol_put(xp);
 422}
 423
 424/* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 425 * SPD calls.
 426 */
 427
 428struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 429{
 430	struct xfrm_policy *policy;
 431
 432	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 433
 434	if (policy) {
 435		write_pnet(&policy->xp_net, net);
 436		INIT_LIST_HEAD(&policy->walk.all);
 437		INIT_HLIST_HEAD(&policy->state_cache_list);
 438		INIT_HLIST_NODE(&policy->bydst);
 439		INIT_HLIST_NODE(&policy->byidx);
 440		rwlock_init(&policy->lock);
 441		refcount_set(&policy->refcnt, 1);
 442		skb_queue_head_init(&policy->polq.hold_queue);
 443		timer_setup(&policy->timer, xfrm_policy_timer, 0);
 444		timer_setup(&policy->polq.hold_timer,
 445			    xfrm_policy_queue_process, 0);
 446	}
 447	return policy;
 448}
 449EXPORT_SYMBOL(xfrm_policy_alloc);
 450
 451static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 452{
 453	struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 454
 455	security_xfrm_policy_free(policy->security);
 456	kfree(policy);
 457}
 458
 459/* Destroy xfrm_policy: descendant resources must be released to this moment. */
 460
 461void xfrm_policy_destroy(struct xfrm_policy *policy)
 462{
 463	BUG_ON(!policy->walk.dead);
 464
 465	if (timer_delete(&policy->timer) || timer_delete(&policy->polq.hold_timer))
 466		BUG();
 467
 468	xfrm_dev_policy_free(policy);
 469	call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 470}
 471EXPORT_SYMBOL(xfrm_policy_destroy);
 472
 473/* Rule must be locked. Release descendant resources, announce
 474 * entry dead. The rule must be unlinked from lists to the moment.
 475 */
 476
 477static void xfrm_policy_kill(struct xfrm_policy *policy)
 478{
 479	struct net *net = xp_net(policy);
 480	struct xfrm_state *x;
 481
 482	xfrm_dev_policy_delete(policy);
 483
 484	write_lock_bh(&policy->lock);
 485	policy->walk.dead = 1;
 486	write_unlock_bh(&policy->lock);
 487
 488	atomic_inc(&policy->genid);
 489
 490	if (timer_delete(&policy->polq.hold_timer))
 491		xfrm_pol_put(policy);
 492	skb_queue_purge(&policy->polq.hold_queue);
 493
 494	if (timer_delete(&policy->timer))
 495		xfrm_pol_put(policy);
 496
 497	/* XXX: Flush state cache */
 498	spin_lock_bh(&net->xfrm.xfrm_state_lock);
 499	hlist_for_each_entry_rcu(x, &policy->state_cache_list, state_cache) {
 500		hlist_del_init_rcu(&x->state_cache);
 501	}
 502	spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 503
 504	xfrm_pol_put(policy);
 505}
 506
 507static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 508
 509static inline unsigned int idx_hash(struct net *net, u32 index)
 510{
 511	return __idx_hash(index, net->xfrm.policy_idx_hmask);
 512}
 513
 514/* calculate policy hash thresholds */
 515static void __get_hash_thresh(struct net *net,
 516			      unsigned short family, int dir,
 517			      u8 *dbits, u8 *sbits)
 518{
 519	switch (family) {
 520	case AF_INET:
 521		*dbits = net->xfrm.policy_bydst[dir].dbits4;
 522		*sbits = net->xfrm.policy_bydst[dir].sbits4;
 523		break;
 524
 525	case AF_INET6:
 526		*dbits = net->xfrm.policy_bydst[dir].dbits6;
 527		*sbits = net->xfrm.policy_bydst[dir].sbits6;
 528		break;
 529
 530	default:
 531		*dbits = 0;
 532		*sbits = 0;
 533	}
 534}
 535
 536static struct hlist_head *policy_hash_bysel(struct net *net,
 537					    const struct xfrm_selector *sel,
 538					    unsigned short family, int dir)
 539{
 540	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 541	unsigned int hash;
 542	u8 dbits;
 543	u8 sbits;
 544
 545	__get_hash_thresh(net, family, dir, &dbits, &sbits);
 546	hash = __sel_hash(sel, family, hmask, dbits, sbits);
 547
 548	if (hash == hmask + 1)
 549		return NULL;
 550
 551	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 552		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 553}
 554
 555static struct hlist_head *policy_hash_direct(struct net *net,
 556					     const xfrm_address_t *daddr,
 557					     const xfrm_address_t *saddr,
 558					     unsigned short family, int dir)
 559{
 560	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 561	unsigned int hash;
 562	u8 dbits;
 563	u8 sbits;
 564
 565	__get_hash_thresh(net, family, dir, &dbits, &sbits);
 566	hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 567
 568	return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 569		     lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 570}
 571
 572static void xfrm_dst_hash_transfer(struct net *net,
 573				   struct hlist_head *list,
 574				   struct hlist_head *ndsttable,
 575				   unsigned int nhashmask,
 576				   int dir)
 577{
 578	struct hlist_node *tmp, *entry0 = NULL;
 579	struct xfrm_policy *pol;
 580	unsigned int h0 = 0;
 581	u8 dbits;
 582	u8 sbits;
 583
 584redo:
 585	hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 586		unsigned int h;
 587
 588		__get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 589		h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 590				pol->family, nhashmask, dbits, sbits);
 591		if (!entry0 || pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET) {
 592			hlist_del_rcu(&pol->bydst);
 593			hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 594			h0 = h;
 595		} else {
 596			if (h != h0)
 597				continue;
 598			hlist_del_rcu(&pol->bydst);
 599			hlist_add_behind_rcu(&pol->bydst, entry0);
 600		}
 601		entry0 = &pol->bydst;
 602	}
 603	if (!hlist_empty(list)) {
 604		entry0 = NULL;
 605		goto redo;
 606	}
 607}
 608
 609static void xfrm_idx_hash_transfer(struct hlist_head *list,
 610				   struct hlist_head *nidxtable,
 611				   unsigned int nhashmask)
 612{
 613	struct hlist_node *tmp;
 614	struct xfrm_policy *pol;
 615
 616	hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 617		unsigned int h;
 618
 619		h = __idx_hash(pol->index, nhashmask);
 620		hlist_add_head(&pol->byidx, nidxtable+h);
 621	}
 622}
 623
 624static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 625{
 626	return ((old_hmask + 1) << 1) - 1;
 627}
 628
 629static void xfrm_bydst_resize(struct net *net, int dir)
 630{
 631	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 632	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 633	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 634	struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 635	struct hlist_head *odst;
 636	int i;
 637
 638	if (!ndst)
 639		return;
 640
 641	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 642	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
 643
 644	odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 645				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 646
 647	for (i = hmask; i >= 0; i--)
 648		xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 649
 650	rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 651	net->xfrm.policy_bydst[dir].hmask = nhashmask;
 652
 653	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
 654	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 655
 656	synchronize_rcu();
 657
 658	xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 659}
 660
 661static void xfrm_byidx_resize(struct net *net)
 662{
 663	unsigned int hmask = net->xfrm.policy_idx_hmask;
 664	unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 665	unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 666	struct hlist_head *oidx = net->xfrm.policy_byidx;
 667	struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 668	int i;
 669
 670	if (!nidx)
 671		return;
 672
 673	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 674
 675	for (i = hmask; i >= 0; i--)
 676		xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 677
 678	net->xfrm.policy_byidx = nidx;
 679	net->xfrm.policy_idx_hmask = nhashmask;
 680
 681	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 682
 683	xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 684}
 685
 686static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 687{
 688	unsigned int cnt = net->xfrm.policy_count[dir];
 689	unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 690
 691	if (total)
 692		*total += cnt;
 693
 694	if ((hmask + 1) < xfrm_policy_hashmax &&
 695	    cnt > hmask)
 696		return 1;
 697
 698	return 0;
 699}
 700
 701static inline int xfrm_byidx_should_resize(struct net *net, int total)
 702{
 703	unsigned int hmask = net->xfrm.policy_idx_hmask;
 704
 705	if ((hmask + 1) < xfrm_policy_hashmax &&
 706	    total > hmask)
 707		return 1;
 708
 709	return 0;
 710}
 711
 712void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 713{
 714	si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 715	si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 716	si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 717	si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 718	si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 719	si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 720	si->spdhcnt = net->xfrm.policy_idx_hmask;
 721	si->spdhmcnt = xfrm_policy_hashmax;
 722}
 723EXPORT_SYMBOL(xfrm_spd_getinfo);
 724
 725static DEFINE_MUTEX(hash_resize_mutex);
 726static void xfrm_hash_resize(struct work_struct *work)
 727{
 728	struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 729	int dir, total;
 730
 731	mutex_lock(&hash_resize_mutex);
 732
 733	total = 0;
 734	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 735		if (xfrm_bydst_should_resize(net, dir, &total))
 736			xfrm_bydst_resize(net, dir);
 737	}
 738	if (xfrm_byidx_should_resize(net, total))
 739		xfrm_byidx_resize(net);
 740
 741	mutex_unlock(&hash_resize_mutex);
 742}
 743
 744/* Make sure *pol can be inserted into fastbin.
 745 * Useful to check that later insert requests will be successful
 746 * (provided xfrm_policy_lock is held throughout).
 747 */
 748static struct xfrm_pol_inexact_bin *
 749xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
 750{
 751	struct xfrm_pol_inexact_bin *bin, *prev;
 752	struct xfrm_pol_inexact_key k = {
 753		.family = pol->family,
 754		.type = pol->type,
 755		.dir = dir,
 756		.if_id = pol->if_id,
 757	};
 758	struct net *net = xp_net(pol);
 759
 760	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 761
 762	write_pnet(&k.net, net);
 763	bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
 764				     xfrm_pol_inexact_params);
 765	if (bin)
 766		return bin;
 767
 768	bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
 769	if (!bin)
 770		return NULL;
 771
 772	bin->k = k;
 773	INIT_HLIST_HEAD(&bin->hhead);
 774	bin->root_d = RB_ROOT;
 775	bin->root_s = RB_ROOT;
 776	seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
 777
 778	prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
 779						&bin->k, &bin->head,
 780						xfrm_pol_inexact_params);
 781	if (!prev) {
 782		list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
 783		return bin;
 784	}
 785
 786	kfree(bin);
 787
 788	return IS_ERR(prev) ? NULL : prev;
 789}
 790
 791static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
 792					       int family, u8 prefixlen)
 793{
 794	if (xfrm_addr_any(addr, family))
 795		return true;
 796
 797	if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
 798		return true;
 799
 800	if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
 801		return true;
 802
 803	return false;
 804}
 805
 806static bool
 807xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
 808{
 809	const xfrm_address_t *addr;
 810	bool saddr_any, daddr_any;
 811	u8 prefixlen;
 812
 813	addr = &policy->selector.saddr;
 814	prefixlen = policy->selector.prefixlen_s;
 815
 816	saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 817						       policy->family,
 818						       prefixlen);
 819	addr = &policy->selector.daddr;
 820	prefixlen = policy->selector.prefixlen_d;
 821	daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 822						       policy->family,
 823						       prefixlen);
 824	return saddr_any && daddr_any;
 825}
 826
 827static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
 828				       const xfrm_address_t *addr, u8 prefixlen)
 829{
 830	node->addr = *addr;
 831	node->prefixlen = prefixlen;
 832}
 833
 834static struct xfrm_pol_inexact_node *
 835xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
 836{
 837	struct xfrm_pol_inexact_node *node;
 838
 839	node = kzalloc(sizeof(*node), GFP_ATOMIC);
 840	if (node)
 841		xfrm_pol_inexact_node_init(node, addr, prefixlen);
 842
 843	return node;
 844}
 845
 846static int xfrm_policy_addr_delta(const xfrm_address_t *a,
 847				  const xfrm_address_t *b,
 848				  u8 prefixlen, u16 family)
 849{
 850	u32 ma, mb, mask;
 851	unsigned int pdw, pbi;
 852	int delta = 0;
 853
 854	switch (family) {
 855	case AF_INET:
 856		if (prefixlen == 0)
 857			return 0;
 858		mask = ~0U << (32 - prefixlen);
 859		ma = ntohl(a->a4) & mask;
 860		mb = ntohl(b->a4) & mask;
 861		if (ma < mb)
 862			delta = -1;
 863		else if (ma > mb)
 864			delta = 1;
 865		break;
 866	case AF_INET6:
 867		pdw = prefixlen >> 5;
 868		pbi = prefixlen & 0x1f;
 869
 870		if (pdw) {
 871			delta = memcmp(a->a6, b->a6, pdw << 2);
 872			if (delta)
 873				return delta;
 874		}
 875		if (pbi) {
 876			mask = ~0U << (32 - pbi);
 877			ma = ntohl(a->a6[pdw]) & mask;
 878			mb = ntohl(b->a6[pdw]) & mask;
 879			if (ma < mb)
 880				delta = -1;
 881			else if (ma > mb)
 882				delta = 1;
 883		}
 884		break;
 885	default:
 886		break;
 887	}
 888
 889	return delta;
 890}
 891
 892static void xfrm_policy_inexact_list_reinsert(struct net *net,
 893					      struct xfrm_pol_inexact_node *n,
 894					      u16 family)
 895{
 896	unsigned int matched_s, matched_d;
 897	struct xfrm_policy *policy, *p;
 898
 899	matched_s = 0;
 900	matched_d = 0;
 901
 902	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 903		struct hlist_node *newpos = NULL;
 904		bool matches_s, matches_d;
 905
 906		if (policy->walk.dead || !policy->bydst_reinsert)
 907			continue;
 908
 909		WARN_ON_ONCE(policy->family != family);
 910
 911		policy->bydst_reinsert = false;
 912		hlist_for_each_entry(p, &n->hhead, bydst) {
 913			if (policy->priority > p->priority)
 914				newpos = &p->bydst;
 915			else if (policy->priority == p->priority &&
 916				 policy->pos > p->pos)
 917				newpos = &p->bydst;
 918			else
 919				break;
 920		}
 921
 922		if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
 923			hlist_add_behind_rcu(&policy->bydst, newpos);
 924		else
 925			hlist_add_head_rcu(&policy->bydst, &n->hhead);
 926
 927		/* paranoia checks follow.
 928		 * Check that the reinserted policy matches at least
 929		 * saddr or daddr for current node prefix.
 930		 *
 931		 * Matching both is fine, matching saddr in one policy
 932		 * (but not daddr) and then matching only daddr in another
 933		 * is a bug.
 934		 */
 935		matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
 936						   &n->addr,
 937						   n->prefixlen,
 938						   family) == 0;
 939		matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
 940						   &n->addr,
 941						   n->prefixlen,
 942						   family) == 0;
 943		if (matches_s && matches_d)
 944			continue;
 945
 946		WARN_ON_ONCE(!matches_s && !matches_d);
 947		if (matches_s)
 948			matched_s++;
 949		if (matches_d)
 950			matched_d++;
 951		WARN_ON_ONCE(matched_s && matched_d);
 952	}
 953}
 954
 955static void xfrm_policy_inexact_node_reinsert(struct net *net,
 956					      struct xfrm_pol_inexact_node *n,
 957					      struct rb_root *new,
 958					      u16 family)
 959{
 960	struct xfrm_pol_inexact_node *node;
 961	struct rb_node **p, *parent;
 962
 963	/* we should not have another subtree here */
 964	WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
 965restart:
 966	parent = NULL;
 967	p = &new->rb_node;
 968	while (*p) {
 969		u8 prefixlen;
 970		int delta;
 971
 972		parent = *p;
 973		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 974
 975		prefixlen = min(node->prefixlen, n->prefixlen);
 976
 977		delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
 978					       prefixlen, family);
 979		if (delta < 0) {
 980			p = &parent->rb_left;
 981		} else if (delta > 0) {
 982			p = &parent->rb_right;
 983		} else {
 984			bool same_prefixlen = node->prefixlen == n->prefixlen;
 985			struct xfrm_policy *tmp;
 986
 987			hlist_for_each_entry(tmp, &n->hhead, bydst) {
 988				tmp->bydst_reinsert = true;
 989				hlist_del_rcu(&tmp->bydst);
 990			}
 991
 992			node->prefixlen = prefixlen;
 993
 994			xfrm_policy_inexact_list_reinsert(net, node, family);
 995
 996			if (same_prefixlen) {
 997				kfree_rcu(n, rcu);
 998				return;
 999			}
1000
1001			rb_erase(*p, new);
1002			kfree_rcu(n, rcu);
1003			n = node;
1004			goto restart;
1005		}
1006	}
1007
1008	rb_link_node_rcu(&n->node, parent, p);
1009	rb_insert_color(&n->node, new);
1010}
1011
1012/* merge nodes v and n */
1013static void xfrm_policy_inexact_node_merge(struct net *net,
1014					   struct xfrm_pol_inexact_node *v,
1015					   struct xfrm_pol_inexact_node *n,
1016					   u16 family)
1017{
1018	struct xfrm_pol_inexact_node *node;
1019	struct xfrm_policy *tmp;
1020	struct rb_node *rnode;
1021
1022	/* To-be-merged node v has a subtree.
1023	 *
1024	 * Dismantle it and insert its nodes to n->root.
1025	 */
1026	while ((rnode = rb_first(&v->root)) != NULL) {
1027		node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
1028		rb_erase(&node->node, &v->root);
1029		xfrm_policy_inexact_node_reinsert(net, node, &n->root,
1030						  family);
1031	}
1032
1033	hlist_for_each_entry(tmp, &v->hhead, bydst) {
1034		tmp->bydst_reinsert = true;
1035		hlist_del_rcu(&tmp->bydst);
1036	}
1037
1038	xfrm_policy_inexact_list_reinsert(net, n, family);
1039}
1040
1041static struct xfrm_pol_inexact_node *
1042xfrm_policy_inexact_insert_node(struct net *net,
1043				struct rb_root *root,
1044				xfrm_address_t *addr,
1045				u16 family, u8 prefixlen, u8 dir)
1046{
1047	struct xfrm_pol_inexact_node *cached = NULL;
1048	struct rb_node **p, *parent = NULL;
1049	struct xfrm_pol_inexact_node *node;
1050
1051	p = &root->rb_node;
1052	while (*p) {
1053		int delta;
1054
1055		parent = *p;
1056		node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1057
1058		delta = xfrm_policy_addr_delta(addr, &node->addr,
1059					       node->prefixlen,
1060					       family);
1061		if (delta == 0 && prefixlen >= node->prefixlen) {
1062			WARN_ON_ONCE(cached); /* ipsec policies got lost */
1063			return node;
1064		}
1065
1066		if (delta < 0)
1067			p = &parent->rb_left;
1068		else
1069			p = &parent->rb_right;
1070
1071		if (prefixlen < node->prefixlen) {
1072			delta = xfrm_policy_addr_delta(addr, &node->addr,
1073						       prefixlen,
1074						       family);
1075			if (delta)
1076				continue;
1077
1078			/* This node is a subnet of the new prefix. It needs
1079			 * to be removed and re-inserted with the smaller
1080			 * prefix and all nodes that are now also covered
1081			 * by the reduced prefixlen.
1082			 */
1083			rb_erase(&node->node, root);
1084
1085			if (!cached) {
1086				xfrm_pol_inexact_node_init(node, addr,
1087							   prefixlen);
1088				cached = node;
1089			} else {
1090				/* This node also falls within the new
1091				 * prefixlen. Merge the to-be-reinserted
1092				 * node and this one.
1093				 */
1094				xfrm_policy_inexact_node_merge(net, node,
1095							       cached, family);
1096				kfree_rcu(node, rcu);
1097			}
1098
1099			/* restart */
1100			p = &root->rb_node;
1101			parent = NULL;
1102		}
1103	}
1104
1105	node = cached;
1106	if (!node) {
1107		node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1108		if (!node)
1109			return NULL;
1110	}
1111
1112	rb_link_node_rcu(&node->node, parent, p);
1113	rb_insert_color(&node->node, root);
1114
1115	return node;
1116}
1117
1118static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1119{
1120	struct xfrm_pol_inexact_node *node;
1121	struct rb_node *rn = rb_first(r);
1122
1123	while (rn) {
1124		node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1125
1126		xfrm_policy_inexact_gc_tree(&node->root, rm);
1127		rn = rb_next(rn);
1128
1129		if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1130			WARN_ON_ONCE(rm);
1131			continue;
1132		}
1133
1134		rb_erase(&node->node, r);
1135		kfree_rcu(node, rcu);
1136	}
1137}
1138
1139static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1140{
1141	write_seqcount_begin(&b->count);
1142	xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1143	xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1144	write_seqcount_end(&b->count);
1145
1146	if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1147	    !hlist_empty(&b->hhead)) {
1148		WARN_ON_ONCE(net_exit);
1149		return;
1150	}
1151
1152	if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1153				   xfrm_pol_inexact_params) == 0) {
1154		list_del(&b->inexact_bins);
1155		kfree_rcu(b, rcu);
1156	}
1157}
1158
1159static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1160{
1161	struct net *net = read_pnet(&b->k.net);
1162
1163	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1164	__xfrm_policy_inexact_prune_bin(b, false);
1165	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1166}
1167
1168static void __xfrm_policy_inexact_flush(struct net *net)
1169{
1170	struct xfrm_pol_inexact_bin *bin, *t;
1171
1172	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1173
1174	list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1175		__xfrm_policy_inexact_prune_bin(bin, false);
1176}
1177
1178static struct hlist_head *
1179xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1180				struct xfrm_policy *policy, u8 dir)
1181{
1182	struct xfrm_pol_inexact_node *n;
1183	struct net *net;
1184
1185	net = xp_net(policy);
1186	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1187
1188	if (xfrm_policy_inexact_insert_use_any_list(policy))
1189		return &bin->hhead;
1190
1191	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1192					       policy->family,
1193					       policy->selector.prefixlen_d)) {
1194		write_seqcount_begin(&bin->count);
1195		n = xfrm_policy_inexact_insert_node(net,
1196						    &bin->root_s,
1197						    &policy->selector.saddr,
1198						    policy->family,
1199						    policy->selector.prefixlen_s,
1200						    dir);
1201		write_seqcount_end(&bin->count);
1202		if (!n)
1203			return NULL;
1204
1205		return &n->hhead;
1206	}
1207
1208	/* daddr is fixed */
1209	write_seqcount_begin(&bin->count);
1210	n = xfrm_policy_inexact_insert_node(net,
1211					    &bin->root_d,
1212					    &policy->selector.daddr,
1213					    policy->family,
1214					    policy->selector.prefixlen_d, dir);
1215	write_seqcount_end(&bin->count);
1216	if (!n)
1217		return NULL;
1218
1219	/* saddr is wildcard */
1220	if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1221					       policy->family,
1222					       policy->selector.prefixlen_s))
1223		return &n->hhead;
1224
1225	write_seqcount_begin(&bin->count);
1226	n = xfrm_policy_inexact_insert_node(net,
1227					    &n->root,
1228					    &policy->selector.saddr,
1229					    policy->family,
1230					    policy->selector.prefixlen_s, dir);
1231	write_seqcount_end(&bin->count);
1232	if (!n)
1233		return NULL;
1234
1235	return &n->hhead;
1236}
1237
1238static struct xfrm_policy *
1239xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1240{
1241	struct xfrm_pol_inexact_bin *bin;
1242	struct xfrm_policy *delpol;
1243	struct hlist_head *chain;
1244	struct net *net;
1245
1246	bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1247	if (!bin)
1248		return ERR_PTR(-ENOMEM);
1249
1250	net = xp_net(policy);
1251	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1252
1253	chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1254	if (!chain) {
1255		__xfrm_policy_inexact_prune_bin(bin, false);
1256		return ERR_PTR(-ENOMEM);
1257	}
1258
1259	delpol = xfrm_policy_insert_list(chain, policy, excl);
1260	if (delpol && excl) {
1261		__xfrm_policy_inexact_prune_bin(bin, false);
1262		return ERR_PTR(-EEXIST);
1263	}
1264
1265	if (delpol)
1266		__xfrm_policy_inexact_prune_bin(bin, false);
1267
1268	return delpol;
1269}
1270
1271static bool xfrm_policy_is_dead_or_sk(const struct xfrm_policy *policy)
1272{
1273	int dir;
1274
1275	if (policy->walk.dead)
1276		return true;
1277
1278	dir = xfrm_policy_id2dir(policy->index);
1279	return dir >= XFRM_POLICY_MAX;
1280}
1281
1282static void xfrm_hash_rebuild(struct work_struct *work)
1283{
1284	struct net *net = container_of(work, struct net,
1285				       xfrm.policy_hthresh.work);
1286	struct xfrm_policy *pol;
1287	struct xfrm_policy *policy;
1288	struct hlist_head *chain;
1289	struct hlist_node *newpos;
1290	int dir;
1291	unsigned seq;
1292	u8 lbits4, rbits4, lbits6, rbits6;
1293
1294	mutex_lock(&hash_resize_mutex);
1295
1296	/* read selector prefixlen thresholds */
1297	do {
1298		seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1299
1300		lbits4 = net->xfrm.policy_hthresh.lbits4;
1301		rbits4 = net->xfrm.policy_hthresh.rbits4;
1302		lbits6 = net->xfrm.policy_hthresh.lbits6;
1303		rbits6 = net->xfrm.policy_hthresh.rbits6;
1304	} while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1305
1306	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1307	write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1308
1309	/* make sure that we can insert the indirect policies again before
1310	 * we start with destructive action.
1311	 */
1312	list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1313		struct xfrm_pol_inexact_bin *bin;
1314		u8 dbits, sbits;
1315
1316		if (xfrm_policy_is_dead_or_sk(policy))
1317			continue;
1318
1319		dir = xfrm_policy_id2dir(policy->index);
1320		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1321			if (policy->family == AF_INET) {
1322				dbits = rbits4;
1323				sbits = lbits4;
1324			} else {
1325				dbits = rbits6;
1326				sbits = lbits6;
1327			}
1328		} else {
1329			if (policy->family == AF_INET) {
1330				dbits = lbits4;
1331				sbits = rbits4;
1332			} else {
1333				dbits = lbits6;
1334				sbits = rbits6;
1335			}
1336		}
1337
1338		if (policy->selector.prefixlen_d < dbits ||
1339		    policy->selector.prefixlen_s < sbits)
1340			continue;
1341
1342		bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1343		if (!bin)
1344			goto out_unlock;
1345
1346		if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1347			goto out_unlock;
1348	}
1349
1350	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1351		if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1352			/* dir out => dst = remote, src = local */
1353			net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1354			net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1355			net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1356			net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1357		} else {
1358			/* dir in/fwd => dst = local, src = remote */
1359			net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1360			net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1361			net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1362			net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1363		}
1364	}
1365
1366	/* re-insert all policies by order of creation */
1367	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1368		if (xfrm_policy_is_dead_or_sk(policy))
1369			continue;
1370
1371		hlist_del_rcu(&policy->bydst);
1372
1373		newpos = NULL;
1374		dir = xfrm_policy_id2dir(policy->index);
1375		chain = policy_hash_bysel(net, &policy->selector,
1376					  policy->family, dir);
1377
1378		if (!chain) {
1379			void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1380
1381			WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1382			continue;
1383		}
1384
1385		hlist_for_each_entry(pol, chain, bydst) {
1386			if (policy->priority >= pol->priority)
1387				newpos = &pol->bydst;
1388			else
1389				break;
1390		}
1391		if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1392			hlist_add_behind_rcu(&policy->bydst, newpos);
1393		else
1394			hlist_add_head_rcu(&policy->bydst, chain);
1395	}
1396
1397out_unlock:
1398	__xfrm_policy_inexact_flush(net);
1399	write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1400	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1401
1402	mutex_unlock(&hash_resize_mutex);
1403}
1404
1405void xfrm_policy_hash_rebuild(struct net *net)
1406{
1407	schedule_work(&net->xfrm.policy_hthresh.work);
1408}
1409EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1410
1411/* Generate new index... KAME seems to generate them ordered by cost
1412 * of an absolute inpredictability of ordering of rules. This will not pass. */
1413static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1414{
1415	for (;;) {
1416		struct hlist_head *list;
1417		struct xfrm_policy *p;
1418		u32 idx;
1419		int found;
1420
1421		if (!index) {
1422			idx = (net->xfrm.idx_generator | dir);
1423			net->xfrm.idx_generator += 8;
1424		} else {
1425			idx = index;
1426			index = 0;
1427		}
1428
1429		if (idx == 0)
1430			idx = 8;
1431		list = net->xfrm.policy_byidx + idx_hash(net, idx);
1432		found = 0;
1433		hlist_for_each_entry(p, list, byidx) {
1434			if (p->index == idx) {
1435				found = 1;
1436				break;
1437			}
1438		}
1439		if (!found)
1440			return idx;
1441	}
1442}
1443
1444static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1445{
1446	u32 *p1 = (u32 *) s1;
1447	u32 *p2 = (u32 *) s2;
1448	int len = sizeof(struct xfrm_selector) / sizeof(u32);
1449	int i;
1450
1451	for (i = 0; i < len; i++) {
1452		if (p1[i] != p2[i])
1453			return 1;
1454	}
1455
1456	return 0;
1457}
1458
1459static void xfrm_policy_requeue(struct xfrm_policy *old,
1460				struct xfrm_policy *new)
1461{
1462	struct xfrm_policy_queue *pq = &old->polq;
1463	struct sk_buff_head list;
1464
1465	if (skb_queue_empty(&pq->hold_queue))
1466		return;
1467
1468	__skb_queue_head_init(&list);
1469
1470	spin_lock_bh(&pq->hold_queue.lock);
1471	skb_queue_splice_init(&pq->hold_queue, &list);
1472	if (timer_delete(&pq->hold_timer))
1473		xfrm_pol_put(old);
1474	spin_unlock_bh(&pq->hold_queue.lock);
1475
1476	pq = &new->polq;
1477
1478	spin_lock_bh(&pq->hold_queue.lock);
1479	skb_queue_splice(&list, &pq->hold_queue);
1480	pq->timeout = XFRM_QUEUE_TMO_MIN;
1481	if (!mod_timer(&pq->hold_timer, jiffies))
1482		xfrm_pol_hold(new);
1483	spin_unlock_bh(&pq->hold_queue.lock);
1484}
1485
1486static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1487					  struct xfrm_policy *pol)
1488{
1489	return mark->v == pol->mark.v && mark->m == pol->mark.m;
1490}
1491
1492static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1493{
1494	const struct xfrm_pol_inexact_key *k = data;
1495	u32 a = k->type << 24 | k->dir << 16 | k->family;
1496
1497	return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1498			    seed);
1499}
1500
1501static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1502{
1503	const struct xfrm_pol_inexact_bin *b = data;
1504
1505	return xfrm_pol_bin_key(&b->k, 0, seed);
1506}
1507
1508static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1509			    const void *ptr)
1510{
1511	const struct xfrm_pol_inexact_key *key = arg->key;
1512	const struct xfrm_pol_inexact_bin *b = ptr;
1513	int ret;
1514
1515	if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1516		return -1;
1517
1518	ret = b->k.dir ^ key->dir;
1519	if (ret)
1520		return ret;
1521
1522	ret = b->k.type ^ key->type;
1523	if (ret)
1524		return ret;
1525
1526	ret = b->k.family ^ key->family;
1527	if (ret)
1528		return ret;
1529
1530	return b->k.if_id ^ key->if_id;
1531}
1532
1533static const struct rhashtable_params xfrm_pol_inexact_params = {
1534	.head_offset		= offsetof(struct xfrm_pol_inexact_bin, head),
1535	.hashfn			= xfrm_pol_bin_key,
1536	.obj_hashfn		= xfrm_pol_bin_obj,
1537	.obj_cmpfn		= xfrm_pol_bin_cmp,
1538	.automatic_shrinking	= true,
1539};
1540
1541static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1542						   struct xfrm_policy *policy,
1543						   bool excl)
1544{
1545	struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1546
1547	hlist_for_each_entry(pol, chain, bydst) {
1548		if (pol->type == policy->type &&
1549		    pol->if_id == policy->if_id &&
1550		    !selector_cmp(&pol->selector, &policy->selector) &&
1551		    xfrm_policy_mark_match(&policy->mark, pol) &&
1552		    xfrm_sec_ctx_match(pol->security, policy->security) &&
1553		    !WARN_ON(delpol)) {
1554			if (excl)
1555				return ERR_PTR(-EEXIST);
1556			delpol = pol;
1557			if (policy->priority > pol->priority)
1558				continue;
1559		} else if (policy->priority >= pol->priority) {
1560			newpos = pol;
1561			continue;
1562		}
1563		if (delpol)
1564			break;
1565	}
1566
1567	if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1568		hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1569	else
1570		/* Packet offload policies enter to the head
1571		 * to speed-up lookups.
1572		 */
1573		hlist_add_head_rcu(&policy->bydst, chain);
1574
1575	return delpol;
1576}
1577
1578int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1579{
1580	struct net *net = xp_net(policy);
1581	struct xfrm_policy *delpol;
1582	struct hlist_head *chain;
1583
1584	/* Sanitize mark before store */
1585	policy->mark.v &= policy->mark.m;
1586
1587	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1588	chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1589	if (chain)
1590		delpol = xfrm_policy_insert_list(chain, policy, excl);
1591	else
1592		delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1593
1594	if (IS_ERR(delpol)) {
1595		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1596		return PTR_ERR(delpol);
1597	}
1598
1599	__xfrm_policy_link(policy, dir);
1600
1601	/* After previous checking, family can either be AF_INET or AF_INET6 */
1602	if (policy->family == AF_INET)
1603		rt_genid_bump_ipv4(net);
1604	else
1605		rt_genid_bump_ipv6(net);
1606
1607	if (delpol) {
1608		xfrm_policy_requeue(delpol, policy);
1609		__xfrm_policy_unlink(delpol, dir);
1610	}
1611	policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1612	hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1613	policy->curlft.add_time = ktime_get_real_seconds();
1614	policy->curlft.use_time = 0;
1615	if (!mod_timer(&policy->timer, jiffies + HZ))
1616		xfrm_pol_hold(policy);
1617	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1618
1619	if (delpol)
1620		xfrm_policy_kill(delpol);
1621	else if (xfrm_bydst_should_resize(net, dir, NULL))
1622		schedule_work(&net->xfrm.policy_hash_work);
1623
1624	return 0;
1625}
1626EXPORT_SYMBOL(xfrm_policy_insert);
1627
1628static struct xfrm_policy *
1629__xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1630			u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1631			struct xfrm_sec_ctx *ctx)
1632{
1633	struct xfrm_policy *pol;
1634
1635	if (!chain)
1636		return NULL;
1637
1638	hlist_for_each_entry(pol, chain, bydst) {
1639		if (pol->type == type &&
1640		    pol->if_id == if_id &&
1641		    xfrm_policy_mark_match(mark, pol) &&
1642		    !selector_cmp(sel, &pol->selector) &&
1643		    xfrm_sec_ctx_match(ctx, pol->security))
1644			return pol;
1645	}
1646
1647	return NULL;
1648}
1649
1650struct xfrm_policy *
1651xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1652		      u8 type, int dir, struct xfrm_selector *sel,
1653		      struct xfrm_sec_ctx *ctx, int delete, int *err)
1654{
1655	struct xfrm_pol_inexact_bin *bin = NULL;
1656	struct xfrm_policy *pol, *ret = NULL;
1657	struct hlist_head *chain;
1658
1659	*err = 0;
1660	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1661	chain = policy_hash_bysel(net, sel, sel->family, dir);
1662	if (!chain) {
1663		struct xfrm_pol_inexact_candidates cand;
1664		int i;
1665
1666		bin = xfrm_policy_inexact_lookup(net, type,
1667						 sel->family, dir, if_id);
1668		if (!bin) {
1669			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1670			return NULL;
1671		}
1672
1673		if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1674							 &sel->saddr,
1675							 &sel->daddr)) {
1676			spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1677			return NULL;
1678		}
1679
1680		pol = NULL;
1681		for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1682			struct xfrm_policy *tmp;
1683
1684			tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1685						      if_id, type, dir,
1686						      sel, ctx);
1687			if (!tmp)
1688				continue;
1689
1690			if (!pol || tmp->pos < pol->pos)
1691				pol = tmp;
1692		}
1693	} else {
1694		pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1695					      sel, ctx);
1696	}
1697
1698	if (pol) {
1699		xfrm_pol_hold(pol);
1700		if (delete) {
1701			*err = security_xfrm_policy_delete(pol->security);
1702			if (*err) {
1703				spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1704				return pol;
1705			}
1706			__xfrm_policy_unlink(pol, dir);
1707		}
1708		ret = pol;
1709	}
1710	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1711
1712	if (ret && delete)
1713		xfrm_policy_kill(ret);
1714	if (bin && delete)
1715		xfrm_policy_inexact_prune_bin(bin);
1716	return ret;
1717}
1718EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1719
1720struct xfrm_policy *
1721xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1722		 u8 type, int dir, u32 id, int delete, int *err)
1723{
1724	struct xfrm_policy *pol, *ret;
1725	struct hlist_head *chain;
1726
1727	*err = -ENOENT;
1728	if (xfrm_policy_id2dir(id) != dir)
1729		return NULL;
1730
1731	*err = 0;
1732	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1733	chain = net->xfrm.policy_byidx + idx_hash(net, id);
1734	ret = NULL;
1735	hlist_for_each_entry(pol, chain, byidx) {
1736		if (pol->type == type && pol->index == id &&
1737		    pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1738			xfrm_pol_hold(pol);
1739			if (delete) {
1740				*err = security_xfrm_policy_delete(
1741								pol->security);
1742				if (*err) {
1743					spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1744					return pol;
1745				}
1746				__xfrm_policy_unlink(pol, dir);
1747			}
1748			ret = pol;
1749			break;
1750		}
1751	}
1752	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1753
1754	if (ret && delete)
1755		xfrm_policy_kill(ret);
1756	return ret;
1757}
1758EXPORT_SYMBOL(xfrm_policy_byid);
1759
1760#ifdef CONFIG_SECURITY_NETWORK_XFRM
1761static inline int
1762xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1763{
1764	struct xfrm_policy *pol;
1765	int err = 0;
1766
1767	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1768		if (pol->walk.dead ||
1769		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1770		    pol->type != type)
1771			continue;
1772
1773		err = security_xfrm_policy_delete(pol->security);
1774		if (err) {
1775			xfrm_audit_policy_delete(pol, 0, task_valid);
1776			return err;
1777		}
1778	}
1779	return err;
1780}
1781
1782static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1783						     struct net_device *dev,
1784						     bool task_valid)
1785{
1786	struct xfrm_policy *pol;
1787	int err = 0;
1788
1789	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1790		if (pol->walk.dead ||
1791		    xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1792		    pol->xdo.dev != dev)
1793			continue;
1794
1795		err = security_xfrm_policy_delete(pol->security);
1796		if (err) {
1797			xfrm_audit_policy_delete(pol, 0, task_valid);
1798			return err;
1799		}
1800	}
1801	return err;
1802}
1803#else
1804static inline int
1805xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1806{
1807	return 0;
1808}
1809
1810static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1811						     struct net_device *dev,
1812						     bool task_valid)
1813{
1814	return 0;
1815}
1816#endif
1817
1818int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1819{
1820	int dir, err = 0, cnt = 0;
1821	struct xfrm_policy *pol;
1822
1823	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1824
1825	err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1826	if (err)
1827		goto out;
1828
1829again:
1830	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1831		if (pol->walk.dead)
1832			continue;
1833
1834		dir = xfrm_policy_id2dir(pol->index);
1835		if (dir >= XFRM_POLICY_MAX ||
1836		    pol->type != type)
1837			continue;
1838
1839		__xfrm_policy_unlink(pol, dir);
1840		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1841		cnt++;
1842		xfrm_audit_policy_delete(pol, 1, task_valid);
1843		xfrm_policy_kill(pol);
1844		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1845		goto again;
1846	}
1847	if (cnt)
1848		__xfrm_policy_inexact_flush(net);
1849	else
1850		err = -ESRCH;
1851out:
1852	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1853	return err;
1854}
1855EXPORT_SYMBOL(xfrm_policy_flush);
1856
1857int xfrm_dev_policy_flush(struct net *net, struct net_device *dev,
1858			  bool task_valid)
1859{
1860	int dir, err = 0, cnt = 0;
1861	struct xfrm_policy *pol;
1862
1863	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1864
1865	err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid);
1866	if (err)
1867		goto out;
1868
1869again:
1870	list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1871		if (pol->walk.dead)
1872			continue;
1873
1874		dir = xfrm_policy_id2dir(pol->index);
1875		if (dir >= XFRM_POLICY_MAX ||
1876		    pol->xdo.dev != dev)
1877			continue;
1878
1879		__xfrm_policy_unlink(pol, dir);
1880		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1881		cnt++;
1882		xfrm_audit_policy_delete(pol, 1, task_valid);
1883		xfrm_policy_kill(pol);
1884		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1885		goto again;
1886	}
1887	if (cnt)
1888		__xfrm_policy_inexact_flush(net);
1889	else
1890		err = -ESRCH;
1891out:
1892	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1893	return err;
1894}
1895EXPORT_SYMBOL(xfrm_dev_policy_flush);
1896
1897int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1898		     int (*func)(struct xfrm_policy *, int, int, void*),
1899		     void *data)
1900{
1901	struct xfrm_policy *pol;
1902	struct xfrm_policy_walk_entry *x;
1903	int error = 0;
1904
1905	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1906	    walk->type != XFRM_POLICY_TYPE_ANY)
1907		return -EINVAL;
1908
1909	if (list_empty(&walk->walk.all) && walk->seq != 0)
1910		return 0;
1911
1912	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1913	if (list_empty(&walk->walk.all))
1914		x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1915	else
1916		x = list_first_entry(&walk->walk.all,
1917				     struct xfrm_policy_walk_entry, all);
1918
1919	list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1920		if (x->dead)
1921			continue;
1922		pol = container_of(x, struct xfrm_policy, walk);
1923		if (walk->type != XFRM_POLICY_TYPE_ANY &&
1924		    walk->type != pol->type)
1925			continue;
1926		error = func(pol, xfrm_policy_id2dir(pol->index),
1927			     walk->seq, data);
1928		if (error) {
1929			list_move_tail(&walk->walk.all, &x->all);
1930			goto out;
1931		}
1932		walk->seq++;
1933	}
1934	if (walk->seq == 0) {
1935		error = -ENOENT;
1936		goto out;
1937	}
1938	list_del_init(&walk->walk.all);
1939out:
1940	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1941	return error;
1942}
1943EXPORT_SYMBOL(xfrm_policy_walk);
1944
1945void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1946{
1947	INIT_LIST_HEAD(&walk->walk.all);
1948	walk->walk.dead = 1;
1949	walk->type = type;
1950	walk->seq = 0;
1951}
1952EXPORT_SYMBOL(xfrm_policy_walk_init);
1953
1954void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1955{
1956	if (list_empty(&walk->walk.all))
1957		return;
1958
1959	spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1960	list_del(&walk->walk.all);
1961	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1962}
1963EXPORT_SYMBOL(xfrm_policy_walk_done);
1964
1965/*
1966 * Find policy to apply to this flow.
1967 *
1968 * Returns 0 if policy found, else an -errno.
1969 */
1970static int xfrm_policy_match(const struct xfrm_policy *pol,
1971			     const struct flowi *fl,
1972			     u8 type, u16 family, u32 if_id)
1973{
1974	const struct xfrm_selector *sel = &pol->selector;
1975	int ret = -ESRCH;
1976	bool match;
1977
1978	if (pol->family != family ||
1979	    pol->if_id != if_id ||
1980	    (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1981	    pol->type != type)
1982		return ret;
1983
1984	match = xfrm_selector_match(sel, fl, family);
1985	if (match)
1986		ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1987	return ret;
1988}
1989
1990static struct xfrm_pol_inexact_node *
1991xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1992				seqcount_spinlock_t *count,
1993				const xfrm_address_t *addr, u16 family)
1994{
1995	const struct rb_node *parent;
1996	int seq;
1997
1998again:
1999	seq = read_seqcount_begin(count);
2000
2001	parent = rcu_dereference_raw(r->rb_node);
2002	while (parent) {
2003		struct xfrm_pol_inexact_node *node;
2004		int delta;
2005
2006		node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
2007
2008		delta = xfrm_policy_addr_delta(addr, &node->addr,
2009					       node->prefixlen, family);
2010		if (delta < 0) {
2011			parent = rcu_dereference_raw(parent->rb_left);
2012			continue;
2013		} else if (delta > 0) {
2014			parent = rcu_dereference_raw(parent->rb_right);
2015			continue;
2016		}
2017
2018		return node;
2019	}
2020
2021	if (read_seqcount_retry(count, seq))
2022		goto again;
2023
2024	return NULL;
2025}
2026
2027static bool
2028xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
2029				    struct xfrm_pol_inexact_bin *b,
2030				    const xfrm_address_t *saddr,
2031				    const xfrm_address_t *daddr)
2032{
2033	struct xfrm_pol_inexact_node *n;
2034	u16 family;
2035
2036	if (!b)
2037		return false;
2038
2039	family = b->k.family;
2040	memset(cand, 0, sizeof(*cand));
2041	cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
2042
2043	n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
2044					    family);
2045	if (n) {
2046		cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
2047		n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
2048						    family);
2049		if (n)
2050			cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
2051	}
2052
2053	n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
2054					    family);
2055	if (n)
2056		cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
2057
2058	return true;
2059}
2060
2061static struct xfrm_pol_inexact_bin *
2062xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
2063			       u8 dir, u32 if_id)
2064{
2065	struct xfrm_pol_inexact_key k = {
2066		.family = family,
2067		.type = type,
2068		.dir = dir,
2069		.if_id = if_id,
2070	};
2071
2072	write_pnet(&k.net, net);
2073
2074	return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
2075				 xfrm_pol_inexact_params);
2076}
2077
2078static struct xfrm_pol_inexact_bin *
2079xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2080			   u8 dir, u32 if_id)
2081{
2082	struct xfrm_pol_inexact_bin *bin;
2083
2084	lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2085
2086	rcu_read_lock();
2087	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2088	rcu_read_unlock();
2089
2090	return bin;
2091}
2092
2093static struct xfrm_policy *
2094__xfrm_policy_eval_candidates(struct hlist_head *chain,
2095			      struct xfrm_policy *prefer,
2096			      const struct flowi *fl,
2097			      u8 type, u16 family, u32 if_id)
2098{
2099	u32 priority = prefer ? prefer->priority : ~0u;
2100	struct xfrm_policy *pol;
2101
2102	if (!chain)
2103		return NULL;
2104
2105	hlist_for_each_entry_rcu(pol, chain, bydst) {
2106		int err;
2107
2108		if (pol->priority > priority)
2109			break;
2110
2111		err = xfrm_policy_match(pol, fl, type, family, if_id);
2112		if (err) {
2113			if (err != -ESRCH)
2114				return ERR_PTR(err);
2115
2116			continue;
2117		}
2118
2119		if (prefer) {
2120			/* matches.  Is it older than *prefer? */
2121			if (pol->priority == priority &&
2122			    prefer->pos < pol->pos)
2123				return prefer;
2124		}
2125
2126		return pol;
2127	}
2128
2129	return NULL;
2130}
2131
2132static struct xfrm_policy *
2133xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2134			    struct xfrm_policy *prefer,
2135			    const struct flowi *fl,
2136			    u8 type, u16 family, u32 if_id)
2137{
2138	struct xfrm_policy *tmp;
2139	int i;
2140
2141	for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2142		tmp = __xfrm_policy_eval_candidates(cand->res[i],
2143						    prefer,
2144						    fl, type, family, if_id);
2145		if (!tmp)
2146			continue;
2147
2148		if (IS_ERR(tmp))
2149			return tmp;
2150		prefer = tmp;
2151	}
2152
2153	return prefer;
2154}
2155
2156static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2157						     const struct flowi *fl,
2158						     u16 family, u8 dir,
2159						     u32 if_id)
2160{
2161	struct xfrm_pol_inexact_candidates cand;
2162	const xfrm_address_t *daddr, *saddr;
2163	struct xfrm_pol_inexact_bin *bin;
2164	struct xfrm_policy *pol, *ret;
2165	struct hlist_head *chain;
2166	unsigned int sequence;
2167	int err;
2168
2169	daddr = xfrm_flowi_daddr(fl, family);
2170	saddr = xfrm_flowi_saddr(fl, family);
2171	if (unlikely(!daddr || !saddr))
2172		return NULL;
2173
2174	rcu_read_lock();
2175 retry:
2176	do {
2177		sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2178		chain = policy_hash_direct(net, daddr, saddr, family, dir);
2179	} while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2180
2181	ret = NULL;
2182	hlist_for_each_entry_rcu(pol, chain, bydst) {
2183		err = xfrm_policy_match(pol, fl, type, family, if_id);
2184		if (err) {
2185			if (err == -ESRCH)
2186				continue;
2187			else {
2188				ret = ERR_PTR(err);
2189				goto fail;
2190			}
2191		} else {
2192			ret = pol;
2193			break;
2194		}
2195	}
2196	if (ret && ret->xdo.type == XFRM_DEV_OFFLOAD_PACKET)
2197		goto skip_inexact;
2198
2199	bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2200	if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2201							 daddr))
2202		goto skip_inexact;
2203
2204	pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2205					  family, if_id);
2206	if (pol) {
2207		ret = pol;
2208		if (IS_ERR(pol))
2209			goto fail;
2210	}
2211
2212skip_inexact:
2213	if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2214		goto retry;
2215
2216	if (ret && !xfrm_pol_hold_rcu(ret))
2217		goto retry;
2218fail:
2219	rcu_read_unlock();
2220
2221	return ret;
2222}
2223
2224static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2225					      const struct flowi *fl,
2226					      u16 family, u8 dir, u32 if_id)
2227{
2228#ifdef CONFIG_XFRM_SUB_POLICY
2229	struct xfrm_policy *pol;
2230
2231	pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2232					dir, if_id);
2233	if (pol != NULL)
2234		return pol;
2235#endif
2236	return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2237					 dir, if_id);
2238}
2239
2240static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2241						 const struct flowi *fl,
2242						 u16 family, u32 if_id)
2243{
2244	struct xfrm_policy *pol;
2245
2246	rcu_read_lock();
2247 again:
2248	pol = rcu_dereference(sk->sk_policy[dir]);
2249	if (pol != NULL) {
2250		bool match;
2251		int err = 0;
2252
2253		if (pol->family != family) {
2254			pol = NULL;
2255			goto out;
2256		}
2257
2258		match = xfrm_selector_match(&pol->selector, fl, family);
2259		if (match) {
2260			if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v ||
2261			    pol->if_id != if_id) {
2262				pol = NULL;
2263				goto out;
2264			}
2265			err = security_xfrm_policy_lookup(pol->security,
2266						      fl->flowi_secid);
2267			if (!err) {
2268				if (!xfrm_pol_hold_rcu(pol))
2269					goto again;
2270			} else if (err == -ESRCH) {
2271				pol = NULL;
2272			} else {
2273				pol = ERR_PTR(err);
2274			}
2275		} else
2276			pol = NULL;
2277	}
2278out:
2279	rcu_read_unlock();
2280	return pol;
2281}
2282
2283static u32 xfrm_gen_pos_slow(struct net *net)
2284{
2285	struct xfrm_policy *policy;
2286	u32 i = 0;
2287
2288	/* oldest entry is last in list */
2289	list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
2290		if (!xfrm_policy_is_dead_or_sk(policy))
2291			policy->pos = ++i;
2292	}
2293
2294	return i;
2295}
2296
2297static u32 xfrm_gen_pos(struct net *net)
2298{
2299	const struct xfrm_policy *policy;
2300	u32 i = 0;
2301
2302	/* most recently added policy is at the head of the list */
2303	list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
2304		if (xfrm_policy_is_dead_or_sk(policy))
2305			continue;
2306
2307		if (policy->pos == UINT_MAX)
2308			return xfrm_gen_pos_slow(net);
2309
2310		i = policy->pos + 1;
2311		break;
2312	}
2313
2314	return i;
2315}
2316
2317static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2318{
2319	struct net *net = xp_net(pol);
2320
2321	switch (dir) {
2322	case XFRM_POLICY_IN:
2323	case XFRM_POLICY_FWD:
2324	case XFRM_POLICY_OUT:
2325		pol->pos = xfrm_gen_pos(net);
2326		break;
2327	}
2328
2329	list_add(&pol->walk.all, &net->xfrm.policy_all);
2330	net->xfrm.policy_count[dir]++;
2331	xfrm_pol_hold(pol);
2332}
2333
2334static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2335						int dir)
2336{
2337	struct net *net = xp_net(pol);
2338
2339	if (list_empty(&pol->walk.all))
2340		return NULL;
2341
2342	/* Socket policies are not hashed. */
2343	if (!hlist_unhashed(&pol->bydst)) {
2344		hlist_del_rcu(&pol->bydst);
2345		hlist_del(&pol->byidx);
2346	}
2347
2348	list_del_init(&pol->walk.all);
2349	net->xfrm.policy_count[dir]--;
2350
2351	return pol;
2352}
2353
2354static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2355{
2356	__xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2357}
2358
2359static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2360{
2361	__xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2362}
2363
2364int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2365{
2366	struct net *net = xp_net(pol);
2367
2368	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2369	pol = __xfrm_policy_unlink(pol, dir);
2370	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2371	if (pol) {
2372		xfrm_policy_kill(pol);
2373		return 0;
2374	}
2375	return -ENOENT;
2376}
2377EXPORT_SYMBOL(xfrm_policy_delete);
2378
2379int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2380{
2381	struct net *net = sock_net(sk);
2382	struct xfrm_policy *old_pol;
2383
2384#ifdef CONFIG_XFRM_SUB_POLICY
2385	if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2386		return -EINVAL;
2387#endif
2388
2389	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2390	old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2391				lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2392	if (pol) {
2393		pol->curlft.add_time = ktime_get_real_seconds();
2394		pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2395		xfrm_sk_policy_link(pol, dir);
2396	}
2397	rcu_assign_pointer(sk->sk_policy[dir], pol);
2398	if (old_pol) {
2399		if (pol)
2400			xfrm_policy_requeue(old_pol, pol);
2401
2402		/* Unlinking succeeds always. This is the only function
2403		 * allowed to delete or replace socket policy.
2404		 */
2405		xfrm_sk_policy_unlink(old_pol, dir);
2406	}
2407	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2408
2409	if (old_pol) {
2410		xfrm_policy_kill(old_pol);
2411	}
2412	return 0;
2413}
2414
2415static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2416{
2417	struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2418	struct net *net = xp_net(old);
2419
2420	if (newp) {
2421		newp->selector = old->selector;
2422		if (security_xfrm_policy_clone(old->security,
2423					       &newp->security)) {
2424			kfree(newp);
2425			return NULL;  /* ENOMEM */
2426		}
2427		newp->lft = old->lft;
2428		newp->curlft = old->curlft;
2429		newp->mark = old->mark;
2430		newp->if_id = old->if_id;
2431		newp->action = old->action;
2432		newp->flags = old->flags;
2433		newp->xfrm_nr = old->xfrm_nr;
2434		newp->index = old->index;
2435		newp->type = old->type;
2436		newp->family = old->family;
2437		memcpy(newp->xfrm_vec, old->xfrm_vec,
2438		       newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2439		spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2440		xfrm_sk_policy_link(newp, dir);
2441		spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2442		xfrm_pol_put(newp);
2443	}
2444	return newp;
2445}
2446
2447int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2448{
2449	const struct xfrm_policy *p;
2450	struct xfrm_policy *np;
2451	int i, ret = 0;
2452
2453	rcu_read_lock();
2454	for (i = 0; i < 2; i++) {
2455		p = rcu_dereference(osk->sk_policy[i]);
2456		if (p) {
2457			np = clone_policy(p, i);
2458			if (unlikely(!np)) {
2459				ret = -ENOMEM;
2460				break;
2461			}
2462			rcu_assign_pointer(sk->sk_policy[i], np);
2463		}
2464	}
2465	rcu_read_unlock();
2466	return ret;
2467}
2468
2469static int
2470xfrm_get_saddr(unsigned short family, xfrm_address_t *saddr,
2471	       const struct xfrm_dst_lookup_params *params)
2472{
2473	int err;
2474	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2475
2476	if (unlikely(afinfo == NULL))
2477		return -EINVAL;
2478	err = afinfo->get_saddr(saddr, params);
2479	rcu_read_unlock();
2480	return err;
2481}
2482
2483/* Resolve list of templates for the flow, given policy. */
2484
2485static int
2486xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2487		      struct xfrm_state **xfrm, unsigned short family)
2488{
2489	struct net *net = xp_net(policy);
2490	int nx;
2491	int i, error;
2492	xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2493	xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2494	xfrm_address_t tmp;
2495
2496	for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2497		struct xfrm_state *x;
2498		xfrm_address_t *remote = daddr;
2499		xfrm_address_t *local  = saddr;
2500		struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2501
2502		if (tmpl->mode == XFRM_MODE_TUNNEL ||
2503		    tmpl->mode == XFRM_MODE_IPTFS ||
2504		    tmpl->mode == XFRM_MODE_BEET) {
2505			remote = &tmpl->id.daddr;
2506			local = &tmpl->saddr;
2507			if (xfrm_addr_any(local, tmpl->encap_family)) {
2508				struct xfrm_dst_lookup_params params;
2509
2510				memset(&params, 0, sizeof(params));
2511				params.net = net;
2512				params.oif = fl->flowi_oif;
2513				params.daddr = remote;
2514				error = xfrm_get_saddr(tmpl->encap_family, &tmp,
2515						       &params);
2516				if (error)
2517					goto fail;
2518				local = &tmp;
2519			}
2520		}
2521
2522		x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2523				    family, policy->if_id);
2524		if (x && x->dir && x->dir != XFRM_SA_DIR_OUT) {
2525			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEDIRERROR);
2526			xfrm_state_put(x);
2527			error = -EINVAL;
2528			goto fail;
2529		}
2530
2531		if (x && x->km.state == XFRM_STATE_VALID) {
2532			xfrm[nx++] = x;
2533			daddr = remote;
2534			saddr = local;
2535			continue;
2536		}
2537		if (x) {
2538			error = (x->km.state == XFRM_STATE_ERROR ?
2539				 -EINVAL : -EAGAIN);
2540			xfrm_state_put(x);
2541		} else if (error == -ESRCH) {
2542			error = -EAGAIN;
2543		}
2544
2545		if (!tmpl->optional)
2546			goto fail;
2547	}
2548	return nx;
2549
2550fail:
2551	for (nx--; nx >= 0; nx--)
2552		xfrm_state_put(xfrm[nx]);
2553	return error;
2554}
2555
2556static int
2557xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2558		  struct xfrm_state **xfrm, unsigned short family)
2559{
2560	struct xfrm_state *tp[XFRM_MAX_DEPTH];
2561	struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2562	int cnx = 0;
2563	int error;
2564	int ret;
2565	int i;
2566
2567	for (i = 0; i < npols; i++) {
2568		if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2569			error = -ENOBUFS;
2570			goto fail;
2571		}
2572
2573		ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2574		if (ret < 0) {
2575			error = ret;
2576			goto fail;
2577		} else
2578			cnx += ret;
2579	}
2580
2581	/* found states are sorted for outbound processing */
2582	if (npols > 1)
2583		xfrm_state_sort(xfrm, tpp, cnx, family);
2584
2585	return cnx;
2586
2587 fail:
2588	for (cnx--; cnx >= 0; cnx--)
2589		xfrm_state_put(tpp[cnx]);
2590	return error;
2591
2592}
2593
2594static dscp_t xfrm_get_dscp(const struct flowi *fl, int family)
2595{
2596	if (family == AF_INET)
2597		return fl->u.ip4.flowi4_dscp;
2598
2599	return 0;
2600}
2601
2602static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2603{
2604	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2605	struct dst_ops *dst_ops;
2606	struct xfrm_dst *xdst;
2607
2608	if (!afinfo)
2609		return ERR_PTR(-EINVAL);
2610
2611	switch (family) {
2612	case AF_INET:
2613		dst_ops = &net->xfrm.xfrm4_dst_ops;
2614		break;
2615#if IS_ENABLED(CONFIG_IPV6)
2616	case AF_INET6:
2617		dst_ops = &net->xfrm.xfrm6_dst_ops;
2618		break;
2619#endif
2620	default:
2621		BUG();
2622	}
2623	xdst = dst_alloc(dst_ops, NULL, DST_OBSOLETE_NONE, 0);
2624
2625	if (likely(xdst)) {
2626		memset_after(xdst, 0, u.dst);
2627	} else
2628		xdst = ERR_PTR(-ENOBUFS);
2629
2630	rcu_read_unlock();
2631
2632	return xdst;
2633}
2634
2635static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2636			   int nfheader_len)
2637{
2638	if (dst->ops->family == AF_INET6) {
2639		path->path_cookie = rt6_get_cookie(dst_rt6_info(dst));
2640		path->u.rt6.rt6i_nfheader_len = nfheader_len;
2641	}
2642}
2643
2644static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2645				const struct flowi *fl)
2646{
2647	const struct xfrm_policy_afinfo *afinfo =
2648		xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2649	int err;
2650
2651	if (!afinfo)
2652		return -EINVAL;
2653
2654	err = afinfo->fill_dst(xdst, dev, fl);
2655
2656	rcu_read_unlock();
2657
2658	return err;
2659}
2660
2661
2662/* Allocate chain of dst_entry's, attach known xfrm's, calculate
2663 * all the metrics... Shortly, bundle a bundle.
2664 */
2665
2666static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2667					    struct xfrm_state **xfrm,
2668					    struct xfrm_dst **bundle,
2669					    int nx,
2670					    const struct flowi *fl,
2671					    struct dst_entry *dst)
2672{
2673	const struct xfrm_state_afinfo *afinfo;
2674	const struct xfrm_mode *inner_mode;
2675	struct net *net = xp_net(policy);
2676	unsigned long now = jiffies;
2677	struct net_device *dev;
2678	struct xfrm_dst *xdst_prev = NULL;
2679	struct xfrm_dst *xdst0 = NULL;
2680	int i = 0;
2681	int err;
2682	int header_len = 0;
2683	int nfheader_len = 0;
2684	int trailer_len = 0;
2685	int family = policy->selector.family;
2686	xfrm_address_t saddr, daddr;
2687	dscp_t dscp;
2688
2689	xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2690
2691	dscp = xfrm_get_dscp(fl, family);
2692
2693	dst_hold(dst);
2694
2695	for (; i < nx; i++) {
2696		struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2697		struct dst_entry *dst1 = &xdst->u.dst;
2698
2699		err = PTR_ERR(xdst);
2700		if (IS_ERR(xdst)) {
2701			dst_release(dst);
2702			goto put_states;
2703		}
2704
2705		bundle[i] = xdst;
2706		if (!xdst_prev)
2707			xdst0 = xdst;
2708		else
2709			/* Ref count is taken during xfrm_alloc_dst()
2710			 * No need to do dst_clone() on dst1
2711			 */
2712			xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2713
2714		if (xfrm[i]->sel.family == AF_UNSPEC) {
2715			inner_mode = xfrm_ip2inner_mode(xfrm[i],
2716							xfrm_af2proto(family));
2717			if (!inner_mode) {
2718				err = -EAFNOSUPPORT;
2719				dst_release(dst);
2720				goto put_states;
2721			}
2722		} else
2723			inner_mode = &xfrm[i]->inner_mode;
2724
2725		xdst->route = dst;
2726		dst_copy_metrics(dst1, dst);
2727
2728		if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2729			__u32 mark = 0;
2730			int oif;
2731
2732			if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2733				mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2734
2735			if (xfrm[i]->xso.type != XFRM_DEV_OFFLOAD_PACKET)
2736				family = xfrm[i]->props.family;
2737
2738			oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2739			dst = xfrm_dst_lookup(xfrm[i], dscp, oif, &saddr,
2740					      &daddr, family, mark);
2741			err = PTR_ERR(dst);
2742			if (IS_ERR(dst))
2743				goto put_states;
2744		} else
2745			dst_hold(dst);
2746
2747		dst1->xfrm = xfrm[i];
2748		xdst->xfrm_genid = xfrm[i]->genid;
2749
2750		dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2751		dst1->lastuse = now;
2752
2753		dst1->input = dst_discard;
2754
2755		if (xfrm[i]->mode_cbs && xfrm[i]->mode_cbs->output) {
2756			dst1->output = xfrm[i]->mode_cbs->output;
2757		} else {
2758			rcu_read_lock();
2759			afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2760			if (likely(afinfo))
2761				dst1->output = afinfo->output;
2762			else
2763				dst1->output = dst_discard_out;
2764			rcu_read_unlock();
2765		}
2766
2767		xdst_prev = xdst;
2768
2769		header_len += xfrm[i]->props.header_len;
2770		if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2771			nfheader_len += xfrm[i]->props.header_len;
2772		trailer_len += xfrm[i]->props.trailer_len;
2773	}
2774
2775	xfrm_dst_set_child(xdst_prev, dst);
2776	xdst0->path = dst;
2777
2778	err = -ENODEV;
2779	dev = dst->dev;
2780	if (!dev)
2781		goto free_dst;
2782
2783	xfrm_init_path(xdst0, dst, nfheader_len);
2784	xfrm_init_pmtu(bundle, nx);
2785
2786	for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2787	     xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2788		err = xfrm_fill_dst(xdst_prev, dev, fl);
2789		if (err)
2790			goto free_dst;
2791
2792		xdst_prev->u.dst.header_len = header_len;
2793		xdst_prev->u.dst.trailer_len = trailer_len;
2794		header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2795		trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2796	}
2797
2798	return &xdst0->u.dst;
2799
2800put_states:
2801	for (; i < nx; i++)
2802		xfrm_state_put(xfrm[i]);
2803free_dst:
2804	if (xdst0)
2805		dst_release_immediate(&xdst0->u.dst);
2806
2807	return ERR_PTR(err);
2808}
2809
2810static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2811				struct xfrm_policy **pols,
2812				int *num_pols, int *num_xfrms)
2813{
2814	int i;
2815
2816	if (*num_pols == 0 || !pols[0]) {
2817		*num_pols = 0;
2818		*num_xfrms = 0;
2819		return 0;
2820	}
2821	if (IS_ERR(pols[0])) {
2822		*num_pols = 0;
2823		return PTR_ERR(pols[0]);
2824	}
2825
2826	*num_xfrms = pols[0]->xfrm_nr;
2827
2828#ifdef CONFIG_XFRM_SUB_POLICY
2829	if (pols[0]->action == XFRM_POLICY_ALLOW &&
2830	    pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2831		pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2832						    XFRM_POLICY_TYPE_MAIN,
2833						    fl, family,
2834						    XFRM_POLICY_OUT,
2835						    pols[0]->if_id);
2836		if (pols[1]) {
2837			if (IS_ERR(pols[1])) {
2838				xfrm_pols_put(pols, *num_pols);
2839				*num_pols = 0;
2840				return PTR_ERR(pols[1]);
2841			}
2842			(*num_pols)++;
2843			(*num_xfrms) += pols[1]->xfrm_nr;
2844		}
2845	}
2846#endif
2847	for (i = 0; i < *num_pols; i++) {
2848		if (pols[i]->action != XFRM_POLICY_ALLOW) {
2849			*num_xfrms = -1;
2850			break;
2851		}
2852	}
2853
2854	return 0;
2855
2856}
2857
2858static struct xfrm_dst *
2859xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2860			       const struct flowi *fl, u16 family,
2861			       struct dst_entry *dst_orig)
2862{
2863	struct net *net = xp_net(pols[0]);
2864	struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2865	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2866	struct xfrm_dst *xdst;
2867	struct dst_entry *dst;
2868	int err;
2869
2870	/* Try to instantiate a bundle */
2871	err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2872	if (err <= 0) {
2873		if (err == 0)
2874			return NULL;
2875
2876		if (err != -EAGAIN)
2877			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2878		return ERR_PTR(err);
2879	}
2880
2881	dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2882	if (IS_ERR(dst)) {
2883		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2884		return ERR_CAST(dst);
2885	}
2886
2887	xdst = (struct xfrm_dst *)dst;
2888	xdst->num_xfrms = err;
2889	xdst->num_pols = num_pols;
2890	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2891	xdst->policy_genid = atomic_read(&pols[0]->genid);
2892
2893	return xdst;
2894}
2895
2896static void xfrm_policy_queue_process(struct timer_list *t)
2897{
2898	struct sk_buff *skb;
2899	struct sock *sk;
2900	struct dst_entry *dst;
2901	struct xfrm_policy *pol = timer_container_of(pol, t, polq.hold_timer);
2902	struct net *net = xp_net(pol);
2903	struct xfrm_policy_queue *pq = &pol->polq;
2904	struct flowi fl;
2905	struct sk_buff_head list;
2906	__u32 skb_mark;
2907
2908	spin_lock(&pq->hold_queue.lock);
2909	skb = skb_peek(&pq->hold_queue);
2910	if (!skb) {
2911		spin_unlock(&pq->hold_queue.lock);
2912		goto out;
2913	}
2914	dst = skb_dst(skb);
2915	sk = skb->sk;
2916
2917	/* Fixup the mark to support VTI. */
2918	skb_mark = skb->mark;
2919	skb->mark = pol->mark.v;
2920	xfrm_decode_session(net, skb, &fl, dst->ops->family);
2921	skb->mark = skb_mark;
2922	spin_unlock(&pq->hold_queue.lock);
2923
2924	dst_hold(xfrm_dst_path(dst));
2925	dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2926	if (IS_ERR(dst))
2927		goto purge_queue;
2928
2929	if (dst->flags & DST_XFRM_QUEUE) {
2930		dst_release(dst);
2931
2932		if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2933			goto purge_queue;
2934
2935		pq->timeout = pq->timeout << 1;
2936		if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2937			xfrm_pol_hold(pol);
2938		goto out;
2939	}
2940
2941	dst_release(dst);
2942
2943	__skb_queue_head_init(&list);
2944
2945	spin_lock(&pq->hold_queue.lock);
2946	pq->timeout = 0;
2947	skb_queue_splice_init(&pq->hold_queue, &list);
2948	spin_unlock(&pq->hold_queue.lock);
2949
2950	while (!skb_queue_empty(&list)) {
2951		skb = __skb_dequeue(&list);
2952
2953		/* Fixup the mark to support VTI. */
2954		skb_mark = skb->mark;
2955		skb->mark = pol->mark.v;
2956		xfrm_decode_session(net, skb, &fl, skb_dst(skb)->ops->family);
2957		skb->mark = skb_mark;
2958
2959		dst_hold(xfrm_dst_path(skb_dst(skb)));
2960		dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2961		if (IS_ERR(dst)) {
2962			kfree_skb(skb);
2963			continue;
2964		}
2965
2966		nf_reset_ct(skb);
2967		skb_dst_drop(skb);
2968		skb_dst_set(skb, dst);
2969
2970		dst_output(net, skb_to_full_sk(skb), skb);
2971	}
2972
2973out:
2974	xfrm_pol_put(pol);
2975	return;
2976
2977purge_queue:
2978	pq->timeout = 0;
2979	skb_queue_purge(&pq->hold_queue);
2980	xfrm_pol_put(pol);
2981}
2982
2983static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2984{
2985	unsigned long sched_next;
2986	struct dst_entry *dst = skb_dst(skb);
2987	struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2988	struct xfrm_policy *pol = xdst->pols[0];
2989	struct xfrm_policy_queue *pq = &pol->polq;
2990
2991	if (unlikely(skb_fclone_busy(sk, skb))) {
2992		kfree_skb(skb);
2993		return 0;
2994	}
2995
2996	if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2997		kfree_skb(skb);
2998		return -EAGAIN;
2999	}
3000
3001	skb_dst_force(skb);
3002
3003	spin_lock_bh(&pq->hold_queue.lock);
3004
3005	if (!pq->timeout)
3006		pq->timeout = XFRM_QUEUE_TMO_MIN;
3007
3008	sched_next = jiffies + pq->timeout;
3009
3010	if (timer_delete(&pq->hold_timer)) {
3011		if (time_before(pq->hold_timer.expires, sched_next))
3012			sched_next = pq->hold_timer.expires;
3013		xfrm_pol_put(pol);
3014	}
3015
3016	__skb_queue_tail(&pq->hold_queue, skb);
3017	if (!mod_timer(&pq->hold_timer, sched_next))
3018		xfrm_pol_hold(pol);
3019
3020	spin_unlock_bh(&pq->hold_queue.lock);
3021
3022	return 0;
3023}
3024
3025static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
3026						 struct xfrm_flo *xflo,
3027						 const struct flowi *fl,
3028						 int num_xfrms,
3029						 u16 family)
3030{
3031	int err;
3032	struct net_device *dev;
3033	struct dst_entry *dst;
3034	struct dst_entry *dst1;
3035	struct xfrm_dst *xdst;
3036
3037	xdst = xfrm_alloc_dst(net, family);
3038	if (IS_ERR(xdst))
3039		return xdst;
3040
3041	if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
3042	    net->xfrm.sysctl_larval_drop ||
3043	    num_xfrms <= 0)
3044		return xdst;
3045
3046	dst = xflo->dst_orig;
3047	dst1 = &xdst->u.dst;
3048	dst_hold(dst);
3049	xdst->route = dst;
3050
3051	dst_copy_metrics(dst1, dst);
3052
3053	dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
3054	dst1->flags |= DST_XFRM_QUEUE;
3055	dst1->lastuse = jiffies;
3056
3057	dst1->input = dst_discard;
3058	dst1->output = xdst_queue_output;
3059
3060	dst_hold(dst);
3061	xfrm_dst_set_child(xdst, dst);
3062	xdst->path = dst;
3063
3064	xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
3065
3066	err = -ENODEV;
3067	dev = dst->dev;
3068	if (!dev)
3069		goto free_dst;
3070
3071	err = xfrm_fill_dst(xdst, dev, fl);
3072	if (err)
3073		goto free_dst;
3074
3075out:
3076	return xdst;
3077
3078free_dst:
3079	dst_release(dst1);
3080	xdst = ERR_PTR(err);
3081	goto out;
3082}
3083
3084static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
3085					   const struct flowi *fl,
3086					   u16 family, u8 dir,
3087					   struct xfrm_flo *xflo, u32 if_id)
3088{
3089	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3090	int num_pols = 0, num_xfrms = 0, err;
3091	struct xfrm_dst *xdst;
3092
3093	/* Resolve policies to use if we couldn't get them from
3094	 * previous cache entry */
3095	num_pols = 1;
3096	pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
3097	err = xfrm_expand_policies(fl, family, pols,
3098					   &num_pols, &num_xfrms);
3099	if (err < 0)
3100		goto inc_error;
3101	if (num_pols == 0)
3102		return NULL;
3103	if (num_xfrms <= 0)
3104		goto make_dummy_bundle;
3105
3106	xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
3107					      xflo->dst_orig);
3108	if (IS_ERR(xdst)) {
3109		err = PTR_ERR(xdst);
3110		if (err == -EREMOTE) {
3111			xfrm_pols_put(pols, num_pols);
3112			return NULL;
3113		}
3114
3115		if (err != -EAGAIN)
3116			goto error;
3117		goto make_dummy_bundle;
3118	} else if (xdst == NULL) {
3119		num_xfrms = 0;
3120		goto make_dummy_bundle;
3121	}
3122
3123	return xdst;
3124
3125make_dummy_bundle:
3126	/* We found policies, but there's no bundles to instantiate:
3127	 * either because the policy blocks, has no transformations or
3128	 * we could not build template (no xfrm_states).*/
3129	xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
3130	if (IS_ERR(xdst)) {
3131		xfrm_pols_put(pols, num_pols);
3132		return ERR_CAST(xdst);
3133	}
3134	xdst->num_pols = num_pols;
3135	xdst->num_xfrms = num_xfrms;
3136	memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
3137
3138	return xdst;
3139
3140inc_error:
3141	XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3142error:
3143	xfrm_pols_put(pols, num_pols);
3144	return ERR_PTR(err);
3145}
3146
3147static struct dst_entry *make_blackhole(struct net *net, u16 family,
3148					struct dst_entry *dst_orig)
3149{
3150	const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3151	struct dst_entry *ret;
3152
3153	if (!afinfo) {
3154		dst_release(dst_orig);
3155		return ERR_PTR(-EINVAL);
3156	} else {
3157		ret = afinfo->blackhole_route(net, dst_orig);
3158	}
3159	rcu_read_unlock();
3160
3161	return ret;
3162}
3163
3164/* Finds/creates a bundle for given flow and if_id
3165 *
3166 * At the moment we eat a raw IP route. Mostly to speed up lookups
3167 * on interfaces with disabled IPsec.
3168 *
3169 * xfrm_lookup uses an if_id of 0 by default, and is provided for
3170 * compatibility
3171 */
3172struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3173					struct dst_entry *dst_orig,
3174					const struct flowi *fl,
3175					const struct sock *sk,
3176					int flags, u32 if_id)
3177{
3178	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3179	struct xfrm_dst *xdst;
3180	struct dst_entry *dst, *route;
3181	u16 family = dst_orig->ops->family;
3182	u8 dir = XFRM_POLICY_OUT;
3183	int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3184
3185	dst = NULL;
3186	xdst = NULL;
3187	route = NULL;
3188
3189	sk = sk_const_to_full_sk(sk);
3190	if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3191		num_pols = 1;
3192		pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3193						if_id);
3194		err = xfrm_expand_policies(fl, family, pols,
3195					   &num_pols, &num_xfrms);
3196		if (err < 0)
3197			goto dropdst;
3198
3199		if (num_pols) {
3200			if (num_xfrms <= 0) {
3201				drop_pols = num_pols;
3202				goto no_transform;
3203			}
3204
3205			xdst = xfrm_resolve_and_create_bundle(
3206					pols, num_pols, fl,
3207					family, dst_orig);
3208
3209			if (IS_ERR(xdst)) {
3210				xfrm_pols_put(pols, num_pols);
3211				err = PTR_ERR(xdst);
3212				if (err == -EREMOTE)
3213					goto nopol;
3214
3215				goto dropdst;
3216			} else if (xdst == NULL) {
3217				num_xfrms = 0;
3218				drop_pols = num_pols;
3219				goto no_transform;
3220			}
3221
3222			route = xdst->route;
3223		}
3224	}
3225
3226	if (xdst == NULL) {
3227		struct xfrm_flo xflo;
3228
3229		xflo.dst_orig = dst_orig;
3230		xflo.flags = flags;
3231
3232		/* To accelerate a bit...  */
3233		if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3234			       !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3235			goto nopol;
3236
3237		xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3238		if (xdst == NULL)
3239			goto nopol;
3240		if (IS_ERR(xdst)) {
3241			err = PTR_ERR(xdst);
3242			goto dropdst;
3243		}
3244
3245		num_pols = xdst->num_pols;
3246		num_xfrms = xdst->num_xfrms;
3247		memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3248		route = xdst->route;
3249	}
3250
3251	dst = &xdst->u.dst;
3252	if (route == NULL && num_xfrms > 0) {
3253		/* The only case when xfrm_bundle_lookup() returns a
3254		 * bundle with null route, is when the template could
3255		 * not be resolved. It means policies are there, but
3256		 * bundle could not be created, since we don't yet
3257		 * have the xfrm_state's. We need to wait for KM to
3258		 * negotiate new SA's or bail out with error.*/
3259		if (net->xfrm.sysctl_larval_drop) {
3260			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3261			err = -EREMOTE;
3262			goto error;
3263		}
3264
3265		err = -EAGAIN;
3266
3267		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3268		goto error;
3269	}
3270
3271no_transform:
3272	if (num_pols == 0)
3273		goto nopol;
3274
3275	if ((flags & XFRM_LOOKUP_ICMP) &&
3276	    !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3277		err = -ENOENT;
3278		goto error;
3279	}
3280
3281	for (i = 0; i < num_pols; i++)
3282		WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds());
3283
3284	if (num_xfrms < 0) {
3285		/* Prohibit the flow */
3286		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3287		err = -EPERM;
3288		goto error;
3289	} else if (num_xfrms > 0) {
3290		/* Flow transformed */
3291		dst_release(dst_orig);
3292	} else {
3293		/* Flow passes untransformed */
3294		dst_release(dst);
3295		dst = dst_orig;
3296	}
3297
3298ok:
3299	xfrm_pols_put(pols, drop_pols);
3300	if (dst->xfrm &&
3301	    (dst->xfrm->props.mode == XFRM_MODE_TUNNEL ||
3302	     dst->xfrm->props.mode == XFRM_MODE_IPTFS))
3303		dst->flags |= DST_XFRM_TUNNEL;
3304	return dst;
3305
3306nopol:
3307	if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3308	    net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3309		err = -EPERM;
3310		goto error;
3311	}
3312	if (!(flags & XFRM_LOOKUP_ICMP)) {
3313		dst = dst_orig;
3314		goto ok;
3315	}
3316	err = -ENOENT;
3317error:
3318	dst_release(dst);
3319dropdst:
3320	if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3321		dst_release(dst_orig);
3322	xfrm_pols_put(pols, drop_pols);
3323	return ERR_PTR(err);
3324}
3325EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3326
3327/* Main function: finds/creates a bundle for given flow.
3328 *
3329 * At the moment we eat a raw IP route. Mostly to speed up lookups
3330 * on interfaces with disabled IPsec.
3331 */
3332struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3333			      const struct flowi *fl, const struct sock *sk,
3334			      int flags)
3335{
3336	return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3337}
3338EXPORT_SYMBOL(xfrm_lookup);
3339
3340/* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3341 * Otherwise we may send out blackholed packets.
3342 */
3343struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3344				    const struct flowi *fl,
3345				    const struct sock *sk, int flags)
3346{
3347	struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3348					    flags | XFRM_LOOKUP_QUEUE |
3349					    XFRM_LOOKUP_KEEP_DST_REF);
3350
3351	if (PTR_ERR(dst) == -EREMOTE)
3352		return make_blackhole(net, dst_orig->ops->family, dst_orig);
3353
3354	if (IS_ERR(dst))
3355		dst_release(dst_orig);
3356
3357	return dst;
3358}
3359EXPORT_SYMBOL(xfrm_lookup_route);
3360
3361static inline int
3362xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3363{
3364	struct sec_path *sp = skb_sec_path(skb);
3365	struct xfrm_state *x;
3366
3367	if (!sp || idx < 0 || idx >= sp->len)
3368		return 0;
3369	x = sp->xvec[idx];
3370	if (!x->type->reject)
3371		return 0;
3372	return x->type->reject(x, skb, fl);
3373}
3374
3375/* When skb is transformed back to its "native" form, we have to
3376 * check policy restrictions. At the moment we make this in maximally
3377 * stupid way. Shame on me. :-) Of course, connected sockets must
3378 * have policy cached at them.
3379 */
3380
3381static inline int
3382xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3383	      unsigned short family, u32 if_id)
3384{
3385	if (xfrm_state_kern(x))
3386		return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3387	return	x->id.proto == tmpl->id.proto &&
3388		(x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3389		(x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3390		x->props.mode == tmpl->mode &&
3391		(tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3392		 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3393		!(x->props.mode != XFRM_MODE_TRANSPORT &&
3394		  xfrm_state_addr_cmp(tmpl, x, family)) &&
3395		(if_id == 0 || if_id == x->if_id);
3396}
3397
3398/*
3399 * 0 or more than 0 is returned when validation is succeeded (either bypass
3400 * because of optional transport mode, or next index of the matched secpath
3401 * state with the template.
3402 * -1 is returned when no matching template is found.
3403 * Otherwise "-2 - errored_index" is returned.
3404 */
3405static inline int
3406xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3407	       unsigned short family, u32 if_id)
3408{
3409	int idx = start;
3410
3411	if (tmpl->optional) {
3412		if (tmpl->mode == XFRM_MODE_TRANSPORT)
3413			return start;
3414	} else
3415		start = -1;
3416	for (; idx < sp->len; idx++) {
3417		if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3418			return ++idx;
3419		if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3420			if (idx < sp->verified_cnt) {
3421				/* Secpath entry previously verified, consider optional and
3422				 * continue searching
3423				 */
3424				continue;
3425			}
3426
3427			if (start == -1)
3428				start = -2-idx;
3429			break;
3430		}
3431	}
3432	return start;
3433}
3434
3435static void
3436decode_session4(const struct xfrm_flow_keys *flkeys, struct flowi *fl, bool reverse)
3437{
3438	struct flowi4 *fl4 = &fl->u.ip4;
3439
3440	memset(fl4, 0, sizeof(struct flowi4));
3441
3442	if (reverse) {
3443		fl4->saddr = flkeys->addrs.ipv4.dst;
3444		fl4->daddr = flkeys->addrs.ipv4.src;
3445		fl4->fl4_sport = flkeys->ports.dst;
3446		fl4->fl4_dport = flkeys->ports.src;
3447	} else {
3448		fl4->saddr = flkeys->addrs.ipv4.src;
3449		fl4->daddr = flkeys->addrs.ipv4.dst;
3450		fl4->fl4_sport = flkeys->ports.src;
3451		fl4->fl4_dport = flkeys->ports.dst;
3452	}
3453
3454	switch (flkeys->basic.ip_proto) {
3455	case IPPROTO_GRE:
3456		fl4->fl4_gre_key = flkeys->gre.keyid;
3457		break;
3458	case IPPROTO_ICMP:
3459		fl4->fl4_icmp_type = flkeys->icmp.type;
3460		fl4->fl4_icmp_code = flkeys->icmp.code;
3461		break;
3462	}
3463
3464	fl4->flowi4_proto = flkeys->basic.ip_proto;
3465	fl4->flowi4_dscp = inet_dsfield_to_dscp(flkeys->ip.tos);
3466}
3467
3468#if IS_ENABLED(CONFIG_IPV6)
3469static void
3470decode_session6(const struct xfrm_flow_keys *flkeys, struct flowi *fl, bool reverse)
3471{
3472	struct flowi6 *fl6 = &fl->u.ip6;
3473
3474	memset(fl6, 0, sizeof(struct flowi6));
3475
3476	if (reverse) {
3477		fl6->saddr = flkeys->addrs.ipv6.dst;
3478		fl6->daddr = flkeys->addrs.ipv6.src;
3479		fl6->fl6_sport = flkeys->ports.dst;
3480		fl6->fl6_dport = flkeys->ports.src;
3481	} else {
3482		fl6->saddr = flkeys->addrs.ipv6.src;
3483		fl6->daddr = flkeys->addrs.ipv6.dst;
3484		fl6->fl6_sport = flkeys->ports.src;
3485		fl6->fl6_dport = flkeys->ports.dst;
3486	}
3487
3488	switch (flkeys->basic.ip_proto) {
3489	case IPPROTO_GRE:
3490		fl6->fl6_gre_key = flkeys->gre.keyid;
3491		break;
3492	case IPPROTO_ICMPV6:
3493		fl6->fl6_icmp_type = flkeys->icmp.type;
3494		fl6->fl6_icmp_code = flkeys->icmp.code;
3495		break;
3496	}
3497
3498	fl6->flowi6_proto = flkeys->basic.ip_proto;
3499}
3500#endif
3501
3502int __xfrm_decode_session(struct net *net, struct sk_buff *skb, struct flowi *fl,
3503			  unsigned int family, int reverse)
3504{
3505	struct xfrm_flow_keys flkeys;
3506
3507	memset(&flkeys, 0, sizeof(flkeys));
3508	__skb_flow_dissect(net, skb, &xfrm_session_dissector, &flkeys,
3509			   NULL, 0, 0, 0, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
3510
3511	switch (family) {
3512	case AF_INET:
3513		decode_session4(&flkeys, fl, reverse);
3514		break;
3515#if IS_ENABLED(CONFIG_IPV6)
3516	case AF_INET6:
3517		decode_session6(&flkeys, fl, reverse);
3518		break;
3519#endif
3520	default:
3521		return -EAFNOSUPPORT;
3522	}
3523
3524	fl->flowi_mark = skb->mark;
3525	if (reverse) {
3526		fl->flowi_oif = skb->skb_iif;
3527	} else {
3528		int oif = 0;
3529
3530		if (skb_dst(skb) && skb_dst(skb)->dev)
3531			oif = skb_dst(skb)->dev->ifindex;
3532
3533		fl->flowi_oif = oif;
3534	}
3535
3536	return security_xfrm_decode_session(skb, &fl->flowi_secid);
3537}
3538EXPORT_SYMBOL(__xfrm_decode_session);
3539
3540static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3541{
3542	for (; k < sp->len; k++) {
3543		if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3544			*idxp = k;
3545			return 1;
3546		}
3547	}
3548
3549	return 0;
3550}
3551
3552static bool icmp_err_packet(const struct flowi *fl, unsigned short family)
3553{
3554	const struct flowi4 *fl4 = &fl->u.ip4;
3555
3556	if (family == AF_INET &&
3557	    fl4->flowi4_proto == IPPROTO_ICMP &&
3558	    (fl4->fl4_icmp_type == ICMP_DEST_UNREACH ||
3559	     fl4->fl4_icmp_type == ICMP_TIME_EXCEEDED))
3560		return true;
3561
3562#if IS_ENABLED(CONFIG_IPV6)
3563	if (family == AF_INET6) {
3564		const struct flowi6 *fl6 = &fl->u.ip6;
3565
3566		if (fl6->flowi6_proto == IPPROTO_ICMPV6 &&
3567		    (fl6->fl6_icmp_type == ICMPV6_DEST_UNREACH ||
3568		    fl6->fl6_icmp_type == ICMPV6_PKT_TOOBIG ||
3569		    fl6->fl6_icmp_type == ICMPV6_TIME_EXCEED))
3570			return true;
3571	}
3572#endif
3573	return false;
3574}
3575
3576static bool xfrm_icmp_flow_decode(struct sk_buff *skb, unsigned short family,
3577				  const struct flowi *fl, struct flowi *fl1)
3578{
3579	bool ret = true;
3580	struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
3581	int hl = family == AF_INET ? (sizeof(struct iphdr) +  sizeof(struct icmphdr)) :
3582		 (sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr));
3583
3584	if (!newskb)
3585		return true;
3586
3587	if (!pskb_pull(newskb, hl))
3588		goto out;
3589
3590	skb_reset_network_header(newskb);
3591
3592	if (xfrm_decode_session_reverse(dev_net(skb->dev), newskb, fl1, family) < 0)
3593		goto out;
3594
3595	fl1->flowi_oif = fl->flowi_oif;
3596	fl1->flowi_mark = fl->flowi_mark;
3597	fl1->flowi_dscp = fl->flowi_dscp;
3598	nf_nat_decode_session(newskb, fl1, family);
3599	ret = false;
3600
3601out:
3602	consume_skb(newskb);
3603	return ret;
3604}
3605
3606static bool xfrm_selector_inner_icmp_match(struct sk_buff *skb, unsigned short family,
3607					   const struct xfrm_selector *sel,
3608					   const struct flowi *fl)
3609{
3610	bool ret = false;
3611
3612	if (icmp_err_packet(fl, family)) {
3613		struct flowi fl1;
3614
3615		if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3616			return ret;
3617
3618		ret = xfrm_selector_match(sel, &fl1, family);
3619	}
3620
3621	return ret;
3622}
3623
3624static inline struct
3625xfrm_policy *xfrm_in_fwd_icmp(struct sk_buff *skb,
3626			      const struct flowi *fl, unsigned short family,
3627			      u32 if_id)
3628{
3629	struct xfrm_policy *pol = NULL;
3630
3631	if (icmp_err_packet(fl, family)) {
3632		struct flowi fl1;
3633		struct net *net = dev_net(skb->dev);
3634
3635		if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3636			return pol;
3637
3638		pol = xfrm_policy_lookup(net, &fl1, family, XFRM_POLICY_FWD, if_id);
3639		if (IS_ERR(pol))
3640			pol = NULL;
3641	}
3642
3643	return pol;
3644}
3645
3646static inline struct
3647dst_entry *xfrm_out_fwd_icmp(struct sk_buff *skb, struct flowi *fl,
3648			     unsigned short family, struct dst_entry *dst)
3649{
3650	if (icmp_err_packet(fl, family)) {
3651		struct net *net = dev_net(skb->dev);
3652		struct dst_entry *dst2;
3653		struct flowi fl1;
3654
3655		if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3656			return dst;
3657
3658		dst_hold(dst);
3659
3660		dst2 = xfrm_lookup(net, dst, &fl1, NULL, (XFRM_LOOKUP_QUEUE | XFRM_LOOKUP_ICMP));
3661
3662		if (IS_ERR(dst2))
3663			return dst;
3664
3665		if (dst2->xfrm) {
3666			dst_release(dst);
3667			dst = dst2;
3668		} else {
3669			dst_release(dst2);
3670		}
3671	}
3672
3673	return dst;
3674}
3675
3676int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3677			unsigned short family)
3678{
3679	struct net *net = dev_net(skb->dev);
3680	struct xfrm_policy *pol;
3681	struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3682	int npols = 0;
3683	int xfrm_nr;
3684	int pi;
3685	int reverse;
3686	struct flowi fl;
3687	int xerr_idx = -1;
3688	const struct xfrm_if_cb *ifcb;
3689	struct sec_path *sp;
3690	u32 if_id = 0;
3691
3692	rcu_read_lock();
3693	ifcb = xfrm_if_get_cb();
3694
3695	if (ifcb) {
3696		struct xfrm_if_decode_session_result r;
3697
3698		if (ifcb->decode_session(skb, family, &r)) {
3699			if_id = r.if_id;
3700			net = r.net;
3701		}
3702	}
3703	rcu_read_unlock();
3704
3705	reverse = dir & ~XFRM_POLICY_MASK;
3706	dir &= XFRM_POLICY_MASK;
3707
3708	if (__xfrm_decode_session(net, skb, &fl, family, reverse) < 0) {
3709		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3710		return 0;
3711	}
3712
3713	nf_nat_decode_session(skb, &fl, family);
3714
3715	/* First, check used SA against their selectors. */
3716	sp = skb_sec_path(skb);
3717	if (sp) {
3718		int i;
3719
3720		for (i = sp->len - 1; i >= 0; i--) {
3721			struct xfrm_state *x = sp->xvec[i];
3722			int ret = 0;
3723
3724			if (!xfrm_selector_match(&x->sel, &fl, family)) {
3725				ret = 1;
3726				if (x->props.flags & XFRM_STATE_ICMP &&
3727				    xfrm_selector_inner_icmp_match(skb, family, &x->sel, &fl))
3728					ret = 0;
3729				if (ret) {
3730					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3731					return 0;
3732				}
3733			}
3734		}
3735	}
3736
3737	pol = NULL;
3738	sk = sk_to_full_sk(sk);
3739	if (sk && sk->sk_policy[dir]) {
3740		pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3741		if (IS_ERR(pol)) {
3742			XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3743			return 0;
3744		}
3745	}
3746
3747	if (!pol)
3748		pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3749
3750	if (IS_ERR(pol)) {
3751		XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3752		return 0;
3753	}
3754
3755	if (!pol && dir == XFRM_POLICY_FWD)
3756		pol = xfrm_in_fwd_icmp(skb, &fl, family, if_id);
3757
3758	if (!pol) {
3759		const bool is_crypto_offload = sp &&
3760			(xfrm_input_state(skb)->xso.type == XFRM_DEV_OFFLOAD_CRYPTO);
3761
3762		if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3763			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3764			return 0;
3765		}
3766
3767		if (sp && secpath_has_nontransport(sp, 0, &xerr_idx) && !is_crypto_offload) {
3768			xfrm_secpath_reject(xerr_idx, skb, &fl);
3769			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3770			return 0;
3771		}
3772		return 1;
3773	}
3774
3775	/* This lockless write can happen from different cpus. */
3776	WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3777
3778	pols[0] = pol;
3779	npols++;
3780#ifdef CONFIG_XFRM_SUB_POLICY
3781	if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3782		pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3783						    &fl, family,
3784						    XFRM_POLICY_IN, if_id);
3785		if (pols[1]) {
3786			if (IS_ERR(pols[1])) {
3787				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3788				xfrm_pol_put(pols[0]);
3789				return 0;
3790			}
3791			/* This write can happen from different cpus. */
3792			WRITE_ONCE(pols[1]->curlft.use_time,
3793				   ktime_get_real_seconds());
3794			npols++;
3795		}
3796	}
3797#endif
3798
3799	if (pol->action == XFRM_POLICY_ALLOW) {
3800		static struct sec_path dummy;
3801		struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3802		struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3803		struct xfrm_tmpl **tpp = tp;
3804		int i, k = 0;
3805		int ti = 0;
3806
3807		sp = skb_sec_path(skb);
3808		if (!sp)
3809			sp = &dummy;
3810
3811		for (pi = 0; pi < npols; pi++) {
3812			if (pols[pi] != pol &&
3813			    pols[pi]->action != XFRM_POLICY_ALLOW) {
3814				XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3815				goto reject;
3816			}
3817			if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3818				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3819				goto reject_error;
3820			}
3821			for (i = 0; i < pols[pi]->xfrm_nr; i++)
3822				tpp[ti++] = &pols[pi]->xfrm_vec[i];
3823		}
3824		xfrm_nr = ti;
3825
3826		if (npols > 1) {
3827			xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3828			tpp = stp;
3829		}
3830
3831		if (pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET && sp == &dummy)
3832			/* This policy template was already checked by HW
3833			 * and secpath was removed in __xfrm_policy_check2.
3834			 */
3835			goto out;
3836
3837		/* For each tunnel xfrm, find the first matching tmpl.
3838		 * For each tmpl before that, find corresponding xfrm.
3839		 * Order is _important_. Later we will implement
3840		 * some barriers, but at the moment barriers
3841		 * are implied between each two transformations.
3842		 * Upon success, marks secpath entries as having been
3843		 * verified to allow them to be skipped in future policy
3844		 * checks (e.g. nested tunnels).
3845		 */
3846		for (i = xfrm_nr - 1; i >= 0; i--) {
3847			k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3848			if (k < 0) {
3849				if (k < -1)
3850					/* "-2 - errored_index" returned */
3851					xerr_idx = -(2+k);
3852				XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3853				goto reject;
3854			}
3855		}
3856
3857		if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3858			XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3859			goto reject;
3860		}
3861
3862out:
3863		xfrm_pols_put(pols, npols);
3864		sp->verified_cnt = k;
3865
3866		return 1;
3867	}
3868	XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3869
3870reject:
3871	xfrm_secpath_reject(xerr_idx, skb, &fl);
3872reject_error:
3873	xfrm_pols_put(pols, npols);
3874	return 0;
3875}
3876EXPORT_SYMBOL(__xfrm_policy_check);
3877
3878int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3879{
3880	struct net *net = dev_net(skb->dev);
3881	struct flowi fl;
3882	struct dst_entry *dst;
3883	int res = 1;
3884
3885	if (xfrm_decode_session(net, skb, &fl, family) < 0) {
3886		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3887		return 0;
3888	}
3889
3890	skb_dst_force(skb);
3891	dst = skb_dst(skb);
3892	if (!dst) {
3893		XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3894		return 0;
3895	}
3896
3897	/* ignore return value from skb_dstref_steal, xfrm_lookup takes
3898	 * care of dropping the refcnt if needed.
3899	 */
3900	skb_dstref_steal(skb);
3901
3902	dst = xfrm_lookup(net, dst, &fl, NULL, XFRM_LOOKUP_QUEUE);
3903	if (IS_ERR(dst)) {
3904		res = 0;
3905		dst = NULL;
3906	}
3907
3908	if (dst && !dst->xfrm)
3909		dst = xfrm_out_fwd_icmp(skb, &fl, family, dst);
3910
3911	skb_dst_set(skb, dst);
3912	return res;
3913}
3914EXPORT_SYMBOL(__xfrm_route_forward);
3915
3916/* Optimize later using cookies and generation ids. */
3917
3918static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3919{
3920	/* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3921	 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3922	 * get validated by dst_ops->check on every use.  We do this
3923	 * because when a normal route referenced by an XFRM dst is
3924	 * obsoleted we do not go looking around for all parent
3925	 * referencing XFRM dsts so that we can invalidate them.  It
3926	 * is just too much work.  Instead we make the checks here on
3927	 * every use.  For example:
3928	 *
3929	 *	XFRM dst A --> IPv4 dst X
3930	 *
3931	 * X is the "xdst->route" of A (X is also the "dst->path" of A
3932	 * in this example).  If X is marked obsolete, "A" will not
3933	 * notice.  That's what we are validating here via the
3934	 * stale_bundle() check.
3935	 *
3936	 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3937	 * be marked on it.
3938	 * This will force stale_bundle() to fail on any xdst bundle with
3939	 * this dst linked in it.
3940	 */
3941	if (READ_ONCE(dst->obsolete) < 0 && !stale_bundle(dst))
3942		return dst;
3943
3944	return NULL;
3945}
3946
3947static int stale_bundle(struct dst_entry *dst)
3948{
3949	return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3950}
3951
3952void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3953{
3954	while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3955		dst->dev = blackhole_netdev;
3956		dev_hold(dst->dev);
3957		dev_put(dev);
3958	}
3959}
3960EXPORT_SYMBOL(xfrm_dst_ifdown);
3961
3962static void xfrm_link_failure(struct sk_buff *skb)
3963{
3964	/* Impossible. Such dst must be popped before reaches point of failure. */
3965}
3966
3967static void xfrm_negative_advice(struct sock *sk, struct dst_entry *dst)
3968{
3969	if (READ_ONCE(dst->obsolete))
3970		sk_dst_reset(sk);
3971}
3972
3973static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3974{
3975	while (nr--) {
3976		struct xfrm_dst *xdst = bundle[nr];
3977		u32 pmtu, route_mtu_cached;
3978		struct dst_entry *dst;
3979
3980		dst = &xdst->u.dst;
3981		pmtu = dst_mtu(xfrm_dst_child(dst));
3982		xdst->child_mtu_cached = pmtu;
3983
3984		pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3985
3986		route_mtu_cached = dst_mtu(xdst->route);
3987		xdst->route_mtu_cached = route_mtu_cached;
3988
3989		if (pmtu > route_mtu_cached)
3990			pmtu = route_mtu_cached;
3991
3992		dst_metric_set(dst, RTAX_MTU, pmtu);
3993	}
3994}
3995
3996/* Check that the bundle accepts the flow and its components are
3997 * still valid.
3998 */
3999
4000static int xfrm_bundle_ok(struct xfrm_dst *first)
4001{
4002	struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
4003	struct dst_entry *dst = &first->u.dst;
4004	struct xfrm_dst *xdst;
4005	int start_from, nr;
4006	u32 mtu;
4007
4008	if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
4009	    (dst->dev && !netif_running(dst->dev)))
4010		return 0;
4011
4012	if (dst->flags & DST_XFRM_QUEUE)
4013		return 1;
4014
4015	start_from = nr = 0;
4016	do {
4017		struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
4018
4019		if (dst->xfrm->km.state != XFRM_STATE_VALID)
4020			return 0;
4021		if (xdst->xfrm_genid != dst->xfrm->genid)
4022			return 0;
4023		if (xdst->num_pols > 0 &&
4024		    xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
4025			return 0;
4026
4027		bundle[nr++] = xdst;
4028
4029		mtu = dst_mtu(xfrm_dst_child(dst));
4030		if (xdst->child_mtu_cached != mtu) {
4031			start_from = nr;
4032			xdst->child_mtu_cached = mtu;
4033		}
4034
4035		if (!dst_check(xdst->route, xdst->route_cookie))
4036			return 0;
4037		mtu = dst_mtu(xdst->route);
4038		if (xdst->route_mtu_cached != mtu) {
4039			start_from = nr;
4040			xdst->route_mtu_cached = mtu;
4041		}
4042
4043		dst = xfrm_dst_child(dst);
4044	} while (dst->xfrm);
4045
4046	if (likely(!start_from))
4047		return 1;
4048
4049	xdst = bundle[start_from - 1];
4050	mtu = xdst->child_mtu_cached;
4051	while (start_from--) {
4052		dst = &xdst->u.dst;
4053
4054		mtu = xfrm_state_mtu(dst->xfrm, mtu);
4055		if (mtu > xdst->route_mtu_cached)
4056			mtu = xdst->route_mtu_cached;
4057		dst_metric_set(dst, RTAX_MTU, mtu);
4058		if (!start_from)
4059			break;
4060
4061		xdst = bundle[start_from - 1];
4062		xdst->child_mtu_cached = mtu;
4063	}
4064
4065	return 1;
4066}
4067
4068static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
4069{
4070	return dst_metric_advmss(xfrm_dst_path(dst));
4071}
4072
4073static unsigned int xfrm_mtu(const struct dst_entry *dst)
4074{
4075	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
4076
4077	return mtu ? : dst_mtu(xfrm_dst_path(dst));
4078}
4079
4080static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
4081					const void *daddr)
4082{
4083	while (dst->xfrm) {
4084		const struct xfrm_state *xfrm = dst->xfrm;
4085
4086		dst = xfrm_dst_child(dst);
4087
4088		if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
4089			continue;
4090		if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
4091			daddr = xfrm->coaddr;
4092		else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
4093			daddr = &xfrm->id.daddr;
4094	}
4095	return daddr;
4096}
4097
4098static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
4099					   struct sk_buff *skb,
4100					   const void *daddr)
4101{
4102	const struct dst_entry *path = xfrm_dst_path(dst);
4103
4104	if (!skb)
4105		daddr = xfrm_get_dst_nexthop(dst, daddr);
4106	return path->ops->neigh_lookup(path, skb, daddr);
4107}
4108
4109static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
4110{
4111	const struct dst_entry *path = xfrm_dst_path(dst);
4112
4113	daddr = xfrm_get_dst_nexthop(dst, daddr);
4114	path->ops->confirm_neigh(path, daddr);
4115}
4116
4117int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
4118{
4119	int err = 0;
4120
4121	if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
4122		return -EAFNOSUPPORT;
4123
4124	spin_lock(&xfrm_policy_afinfo_lock);
4125	if (unlikely(xfrm_policy_afinfo[family] != NULL))
4126		err = -EEXIST;
4127	else {
4128		struct dst_ops *dst_ops = afinfo->dst_ops;
4129		if (likely(dst_ops->kmem_cachep == NULL))
4130			dst_ops->kmem_cachep = xfrm_dst_cache;
4131		if (likely(dst_ops->check == NULL))
4132			dst_ops->check = xfrm_dst_check;
4133		if (likely(dst_ops->default_advmss == NULL))
4134			dst_ops->default_advmss = xfrm_default_advmss;
4135		if (likely(dst_ops->mtu == NULL))
4136			dst_ops->mtu = xfrm_mtu;
4137		if (likely(dst_ops->negative_advice == NULL))
4138			dst_ops->negative_advice = xfrm_negative_advice;
4139		if (likely(dst_ops->link_failure == NULL))
4140			dst_ops->link_failure = xfrm_link_failure;
4141		if (likely(dst_ops->neigh_lookup == NULL))
4142			dst_ops->neigh_lookup = xfrm_neigh_lookup;
4143		if (likely(!dst_ops->confirm_neigh))
4144			dst_ops->confirm_neigh = xfrm_confirm_neigh;
4145		rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
4146	}
4147	spin_unlock(&xfrm_policy_afinfo_lock);
4148
4149	return err;
4150}
4151EXPORT_SYMBOL(xfrm_policy_register_afinfo);
4152
4153void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
4154{
4155	struct dst_ops *dst_ops = afinfo->dst_ops;
4156	int i;
4157
4158	for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
4159		if (xfrm_policy_afinfo[i] != afinfo)
4160			continue;
4161		RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
4162		break;
4163	}
4164
4165	synchronize_rcu();
4166
4167	dst_ops->kmem_cachep = NULL;
4168	dst_ops->check = NULL;
4169	dst_ops->negative_advice = NULL;
4170	dst_ops->link_failure = NULL;
4171}
4172EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
4173
4174void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
4175{
4176	spin_lock(&xfrm_if_cb_lock);
4177	rcu_assign_pointer(xfrm_if_cb, ifcb);
4178	spin_unlock(&xfrm_if_cb_lock);
4179}
4180EXPORT_SYMBOL(xfrm_if_register_cb);
4181
4182void xfrm_if_unregister_cb(void)
4183{
4184	RCU_INIT_POINTER(xfrm_if_cb, NULL);
4185	synchronize_rcu();
4186}
4187EXPORT_SYMBOL(xfrm_if_unregister_cb);
4188
4189#ifdef CONFIG_XFRM_STATISTICS
4190static int __net_init xfrm_statistics_init(struct net *net)
4191{
4192	int rv;
4193	net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4194	if (!net->mib.xfrm_statistics)
4195		return -ENOMEM;
4196	rv = xfrm_proc_init(net);
4197	if (rv < 0)
4198		free_percpu(net->mib.xfrm_statistics);
4199	return rv;
4200}
4201
4202static void xfrm_statistics_fini(struct net *net)
4203{
4204	xfrm_proc_fini(net);
4205	free_percpu(net->mib.xfrm_statistics);
4206}
4207#else
4208static int __net_init xfrm_statistics_init(struct net *net)
4209{
4210	return 0;
4211}
4212
4213static void xfrm_statistics_fini(struct net *net)
4214{
4215}
4216#endif
4217
4218static int __net_init xfrm_policy_init(struct net *net)
4219{
4220	unsigned int hmask, sz;
4221	int dir, err;
4222
4223	if (net_eq(net, &init_net)) {
4224		xfrm_dst_cache = KMEM_CACHE(xfrm_dst, SLAB_HWCACHE_ALIGN | SLAB_PANIC);
4225		err = rhashtable_init(&xfrm_policy_inexact_table,
4226				      &xfrm_pol_inexact_params);
4227		BUG_ON(err);
4228	}
4229
4230	hmask = 8 - 1;
4231	sz = (hmask+1) * sizeof(struct hlist_head);
4232
4233	net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4234	if (!net->xfrm.policy_byidx)
4235		goto out_byidx;
4236	net->xfrm.policy_idx_hmask = hmask;
4237
4238	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4239		struct xfrm_policy_hash *htab;
4240
4241		net->xfrm.policy_count[dir] = 0;
4242		net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4243
4244		htab = &net->xfrm.policy_bydst[dir];
4245		htab->table = xfrm_hash_alloc(sz);
4246		if (!htab->table)
4247			goto out_bydst;
4248		htab->hmask = hmask;
4249		htab->dbits4 = 32;
4250		htab->sbits4 = 32;
4251		htab->dbits6 = 128;
4252		htab->sbits6 = 128;
4253	}
4254	net->xfrm.policy_hthresh.lbits4 = 32;
4255	net->xfrm.policy_hthresh.rbits4 = 32;
4256	net->xfrm.policy_hthresh.lbits6 = 128;
4257	net->xfrm.policy_hthresh.rbits6 = 128;
4258
4259	seqlock_init(&net->xfrm.policy_hthresh.lock);
4260
4261	INIT_LIST_HEAD(&net->xfrm.policy_all);
4262	INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4263	INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4264	INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4265	return 0;
4266
4267out_bydst:
4268	for (dir--; dir >= 0; dir--) {
4269		struct xfrm_policy_hash *htab;
4270
4271		htab = &net->xfrm.policy_bydst[dir];
4272		xfrm_hash_free(htab->table, sz);
4273	}
4274	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4275out_byidx:
4276	return -ENOMEM;
4277}
4278
4279static void xfrm_policy_fini(struct net *net)
4280{
4281	struct xfrm_pol_inexact_bin *b, *t;
4282	unsigned int sz;
4283	int dir;
4284
4285	disable_work_sync(&net->xfrm.policy_hthresh.work);
4286
4287	flush_work(&net->xfrm.policy_hash_work);
4288#ifdef CONFIG_XFRM_SUB_POLICY
4289	xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4290#endif
4291	xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4292
4293	synchronize_rcu();
4294
4295	WARN_ON(!list_empty(&net->xfrm.policy_all));
4296
4297	for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4298		struct xfrm_policy_hash *htab;
4299
4300		htab = &net->xfrm.policy_bydst[dir];
4301		sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4302		WARN_ON(!hlist_empty(htab->table));
4303		xfrm_hash_free(htab->table, sz);
4304	}
4305
4306	sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4307	WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4308	xfrm_hash_free(net->xfrm.policy_byidx, sz);
4309
4310	spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4311	list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4312		__xfrm_policy_inexact_prune_bin(b, true);
4313	spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4314}
4315
4316static int __net_init xfrm_net_init(struct net *net)
4317{
4318	int rv;
4319
4320	/* Initialize the per-net locks here */
4321	spin_lock_init(&net->xfrm.xfrm_state_lock);
4322	spin_lock_init(&net->xfrm.xfrm_policy_lock);
4323	seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4324	mutex_init(&net->xfrm.xfrm_cfg_mutex);
4325	net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4326	net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4327	net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4328
4329	rv = xfrm_statistics_init(net);
4330	if (rv < 0)
4331		goto out_statistics;
4332	rv = xfrm_state_init(net);
4333	if (rv < 0)
4334		goto out_state;
4335	rv = xfrm_policy_init(net);
4336	if (rv < 0)
4337		goto out_policy;
4338	rv = xfrm_sysctl_init(net);
4339	if (rv < 0)
4340		goto out_sysctl;
4341
4342	rv = xfrm_nat_keepalive_net_init(net);
4343	if (rv < 0)
4344		goto out_nat_keepalive;
4345
4346	return 0;
4347
4348out_nat_keepalive:
4349	xfrm_sysctl_fini(net);
4350out_sysctl:
4351	xfrm_policy_fini(net);
4352out_policy:
4353	xfrm_state_fini(net);
4354out_state:
4355	xfrm_statistics_fini(net);
4356out_statistics:
4357	return rv;
4358}
4359
4360static void __net_exit xfrm_net_exit(struct net *net)
4361{
4362	xfrm_nat_keepalive_net_fini(net);
4363	xfrm_sysctl_fini(net);
4364	xfrm_policy_fini(net);
4365	xfrm_state_fini(net);
4366	xfrm_statistics_fini(net);
4367}
4368
4369static struct pernet_operations __net_initdata xfrm_net_ops = {
4370	.init = xfrm_net_init,
4371	.exit = xfrm_net_exit,
4372};
4373
4374static const struct flow_dissector_key xfrm_flow_dissector_keys[] = {
4375	{
4376		.key_id = FLOW_DISSECTOR_KEY_CONTROL,
4377		.offset = offsetof(struct xfrm_flow_keys, control),
4378	},
4379	{
4380		.key_id = FLOW_DISSECTOR_KEY_BASIC,
4381		.offset = offsetof(struct xfrm_flow_keys, basic),
4382	},
4383	{
4384		.key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
4385		.offset = offsetof(struct xfrm_flow_keys, addrs.ipv4),
4386	},
4387	{
4388		.key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
4389		.offset = offsetof(struct xfrm_flow_keys, addrs.ipv6),
4390	},
4391	{
4392		.key_id = FLOW_DISSECTOR_KEY_PORTS,
4393		.offset = offsetof(struct xfrm_flow_keys, ports),
4394	},
4395	{
4396		.key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
4397		.offset = offsetof(struct xfrm_flow_keys, gre),
4398	},
4399	{
4400		.key_id = FLOW_DISSECTOR_KEY_IP,
4401		.offset = offsetof(struct xfrm_flow_keys, ip),
4402	},
4403	{
4404		.key_id = FLOW_DISSECTOR_KEY_ICMP,
4405		.offset = offsetof(struct xfrm_flow_keys, icmp),
4406	},
4407};
4408
4409void __init xfrm_init(void)
4410{
4411	skb_flow_dissector_init(&xfrm_session_dissector,
4412				xfrm_flow_dissector_keys,
4413				ARRAY_SIZE(xfrm_flow_dissector_keys));
4414
4415	register_pernet_subsys(&xfrm_net_ops);
4416	xfrm_dev_init();
4417	xfrm_input_init();
4418
4419#ifdef CONFIG_XFRM_ESPINTCP
4420	espintcp_init();
4421#endif
4422
4423	register_xfrm_state_bpf();
4424	xfrm_nat_keepalive_init(AF_INET);
4425}
4426
4427#ifdef CONFIG_AUDITSYSCALL
4428static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4429					 struct audit_buffer *audit_buf)
4430{
4431	struct xfrm_sec_ctx *ctx = xp->security;
4432	struct xfrm_selector *sel = &xp->selector;
4433
4434	if (ctx)
4435		audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4436				 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4437
4438	switch (sel->family) {
4439	case AF_INET:
4440		audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4441		if (sel->prefixlen_s != 32)
4442			audit_log_format(audit_buf, " src_prefixlen=%d",
4443					 sel->prefixlen_s);
4444		audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4445		if (sel->prefixlen_d != 32)
4446			audit_log_format(audit_buf, " dst_prefixlen=%d",
4447					 sel->prefixlen_d);
4448		break;
4449	case AF_INET6:
4450		audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4451		if (sel->prefixlen_s != 128)
4452			audit_log_format(audit_buf, " src_prefixlen=%d",
4453					 sel->prefixlen_s);
4454		audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4455		if (sel->prefixlen_d != 128)
4456			audit_log_format(audit_buf, " dst_prefixlen=%d",
4457					 sel->prefixlen_d);
4458		break;
4459	}
4460}
4461
4462void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4463{
4464	struct audit_buffer *audit_buf;
4465
4466	audit_buf = xfrm_audit_start("SPD-add");
4467	if (audit_buf == NULL)
4468		return;
4469	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4470	audit_log_format(audit_buf, " res=%u", result);
4471	xfrm_audit_common_policyinfo(xp, audit_buf);
4472	audit_log_end(audit_buf);
4473}
4474EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4475
4476void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4477			      bool task_valid)
4478{
4479	struct audit_buffer *audit_buf;
4480
4481	audit_buf = xfrm_audit_start("SPD-delete");
4482	if (audit_buf == NULL)
4483		return;
4484	xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4485	audit_log_format(audit_buf, " res=%u", result);
4486	xfrm_audit_common_policyinfo(xp, audit_buf);
4487	audit_log_end(audit_buf);
4488}
4489EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4490#endif
4491
4492#ifdef CONFIG_XFRM_MIGRATE
4493static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4494						    u8 dir, u8 type, struct net *net, u32 if_id)
4495{
4496	struct xfrm_policy *pol;
4497	struct flowi fl;
4498
4499	memset(&fl, 0, sizeof(fl));
4500
4501	fl.flowi_proto = sel->proto;
4502
4503	switch (sel->family) {
4504	case AF_INET:
4505		fl.u.ip4.saddr = sel->saddr.a4;
4506		fl.u.ip4.daddr = sel->daddr.a4;
4507		if (sel->proto == IPSEC_ULPROTO_ANY)
4508			break;
4509		fl.u.flowi4_oif = sel->ifindex;
4510		fl.u.ip4.fl4_sport = sel->sport;
4511		fl.u.ip4.fl4_dport = sel->dport;
4512		break;
4513	case AF_INET6:
4514		fl.u.ip6.saddr = sel->saddr.in6;
4515		fl.u.ip6.daddr = sel->daddr.in6;
4516		if (sel->proto == IPSEC_ULPROTO_ANY)
4517			break;
4518		fl.u.flowi6_oif = sel->ifindex;
4519		fl.u.ip6.fl4_sport = sel->sport;
4520		fl.u.ip6.fl4_dport = sel->dport;
4521		break;
4522	default:
4523		return ERR_PTR(-EAFNOSUPPORT);
4524	}
4525
4526	rcu_read_lock();
4527
4528	pol = xfrm_policy_lookup_bytype(net, type, &fl, sel->family, dir, if_id);
4529	if (IS_ERR_OR_NULL(pol))
4530		goto out_unlock;
4531out_unlock:
4532	rcu_read_unlock();
4533	return pol;
4534}
4535
4536static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4537{
4538	int match = 0;
4539
4540	if (t->mode == m->mode && t->id.proto == m->proto &&
4541	    (m->reqid == 0 || t->reqid == m->reqid)) {
4542		switch (t->mode) {
4543		case XFRM_MODE_TUNNEL:
4544		case XFRM_MODE_BEET:
4545		case XFRM_MODE_IPTFS:
4546			if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4547					    m->old_family) &&
4548			    xfrm_addr_equal(&t->saddr, &m->old_saddr,
4549					    m->old_family)) {
4550				match = 1;
4551			}
4552			break;
4553		case XFRM_MODE_TRANSPORT:
4554			/* in case of transport mode, template does not store
4555			   any IP addresses, hence we just compare mode and
4556			   protocol */
4557			match = 1;
4558			break;
4559		default:
4560			break;
4561		}
4562	}
4563	return match;
4564}
4565
4566/* update endpoint address(es) of template(s) */
4567static int xfrm_policy_migrate(struct xfrm_policy *pol,
4568			       struct xfrm_migrate *m, int num_migrate,
4569			       struct netlink_ext_ack *extack)
4570{
4571	struct xfrm_migrate *mp;
4572	int i, j, n = 0;
4573
4574	write_lock_bh(&pol->lock);
4575	if (unlikely(pol->walk.dead)) {
4576		/* target policy has been deleted */
4577		NL_SET_ERR_MSG(extack, "Target policy not found");
4578		write_unlock_bh(&pol->lock);
4579		return -ENOENT;
4580	}
4581
4582	for (i = 0; i < pol->xfrm_nr; i++) {
4583		for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4584			if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4585				continue;
4586			n++;
4587			if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4588			    pol->xfrm_vec[i].mode != XFRM_MODE_BEET &&
4589			    pol->xfrm_vec[i].mode != XFRM_MODE_IPTFS)
4590				continue;
4591			/* update endpoints */
4592			memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4593			       sizeof(pol->xfrm_vec[i].id.daddr));
4594			memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4595			       sizeof(pol->xfrm_vec[i].saddr));
4596			pol->xfrm_vec[i].encap_family = mp->new_family;
4597			/* flush bundles */
4598			atomic_inc(&pol->genid);
4599		}
4600	}
4601
4602	write_unlock_bh(&pol->lock);
4603
4604	if (!n)
4605		return -ENODATA;
4606
4607	return 0;
4608}
4609
4610static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate,
4611			      struct netlink_ext_ack *extack)
4612{
4613	int i, j;
4614
4615	if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) {
4616		NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)");
4617		return -EINVAL;
4618	}
4619
4620	for (i = 0; i < num_migrate; i++) {
4621		if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4622		    xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) {
4623			NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null");
4624			return -EINVAL;
4625		}
4626
4627		/* check if there is any duplicated entry */
4628		for (j = i + 1; j < num_migrate; j++) {
4629			if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4630				    sizeof(m[i].old_daddr)) &&
4631			    !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4632				    sizeof(m[i].old_saddr)) &&
4633			    m[i].proto == m[j].proto &&
4634			    m[i].mode == m[j].mode &&
4635			    m[i].reqid == m[j].reqid &&
4636			    m[i].old_family == m[j].old_family) {
4637				NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique");
4638				return -EINVAL;
4639			}
4640		}
4641	}
4642
4643	return 0;
4644}
4645
4646int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4647		 struct xfrm_migrate *m, int num_migrate,
4648		 struct xfrm_kmaddress *k, struct net *net,
4649		 struct xfrm_encap_tmpl *encap, u32 if_id,
4650		 struct netlink_ext_ack *extack, struct xfrm_user_offload *xuo)
4651{
4652	int i, err, nx_cur = 0, nx_new = 0;
4653	struct xfrm_policy *pol = NULL;
4654	struct xfrm_state *x, *xc;
4655	struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4656	struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4657	struct xfrm_migrate *mp;
4658
4659	/* Stage 0 - sanity checks */
4660	err = xfrm_migrate_check(m, num_migrate, extack);
4661	if (err < 0)
4662		goto out;
4663
4664	if (dir >= XFRM_POLICY_MAX) {
4665		NL_SET_ERR_MSG(extack, "Invalid policy direction");
4666		err = -EINVAL;
4667		goto out;
4668	}
4669
4670	/* Stage 1 - find policy */
4671	pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id);
4672	if (IS_ERR_OR_NULL(pol)) {
4673		NL_SET_ERR_MSG(extack, "Target policy not found");
4674		err = IS_ERR(pol) ? PTR_ERR(pol) : -ENOENT;
4675		goto out;
4676	}
4677
4678	/* Stage 2 - find and update state(s) */
4679	for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4680		if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4681			x_cur[nx_cur] = x;
4682			nx_cur++;
4683			xc = xfrm_state_migrate(x, mp, encap, net, xuo, extack);
4684			if (xc) {
4685				x_new[nx_new] = xc;
4686				nx_new++;
4687			} else {
4688				err = -ENODATA;
4689				goto restore_state;
4690			}
4691		}
4692	}
4693
4694	/* Stage 3 - update policy */
4695	err = xfrm_policy_migrate(pol, m, num_migrate, extack);
4696	if (err < 0)
4697		goto restore_state;
4698
4699	/* Stage 4 - delete old state(s) */
4700	if (nx_cur) {
4701		xfrm_states_put(x_cur, nx_cur);
4702		xfrm_states_delete(x_cur, nx_cur);
4703	}
4704
4705	/* Stage 5 - announce */
4706	km_migrate(sel, dir, type, m, num_migrate, k, encap);
4707
4708	xfrm_pol_put(pol);
4709
4710	return 0;
4711out:
4712	return err;
4713
4714restore_state:
4715	if (pol)
4716		xfrm_pol_put(pol);
4717	if (nx_cur)
4718		xfrm_states_put(x_cur, nx_cur);
4719	if (nx_new)
4720		xfrm_states_delete(x_new, nx_new);
4721
4722	return err;
4723}
4724EXPORT_SYMBOL(xfrm_migrate);
4725#endif