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