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_XFRM_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_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_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_init(&bin->count);
 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 inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1437                                           struct xfrm_policy *pol)
1438 {
1439         return mark->v == pol->mark.v && mark->m == pol->mark.m;
1440 }
1441
1442 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1443 {
1444         const struct xfrm_pol_inexact_key *k = data;
1445         u32 a = k->type << 24 | k->dir << 16 | k->family;
1446
1447         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1448                             seed);
1449 }
1450
1451 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1452 {
1453         const struct xfrm_pol_inexact_bin *b = data;
1454
1455         return xfrm_pol_bin_key(&b->k, 0, seed);
1456 }
1457
1458 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1459                             const void *ptr)
1460 {
1461         const struct xfrm_pol_inexact_key *key = arg->key;
1462         const struct xfrm_pol_inexact_bin *b = ptr;
1463         int ret;
1464
1465         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1466                 return -1;
1467
1468         ret = b->k.dir ^ key->dir;
1469         if (ret)
1470                 return ret;
1471
1472         ret = b->k.type ^ key->type;
1473         if (ret)
1474                 return ret;
1475
1476         ret = b->k.family ^ key->family;
1477         if (ret)
1478                 return ret;
1479
1480         return b->k.if_id ^ key->if_id;
1481 }
1482
1483 static const struct rhashtable_params xfrm_pol_inexact_params = {
1484         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1485         .hashfn                 = xfrm_pol_bin_key,
1486         .obj_hashfn             = xfrm_pol_bin_obj,
1487         .obj_cmpfn              = xfrm_pol_bin_cmp,
1488         .automatic_shrinking    = true,
1489 };
1490
1491 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1492                                             struct xfrm_policy *policy)
1493 {
1494         struct xfrm_policy *pol, *delpol = NULL;
1495         struct hlist_node *newpos = NULL;
1496         int i = 0;
1497
1498         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1499                 if (pol->type == policy->type &&
1500                     pol->if_id == policy->if_id &&
1501                     !selector_cmp(&pol->selector, &policy->selector) &&
1502                     xfrm_policy_mark_match(&policy->mark, pol) &&
1503                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1504                     !WARN_ON(delpol)) {
1505                         delpol = pol;
1506                         if (policy->priority > pol->priority)
1507                                 continue;
1508                 } else if (policy->priority >= pol->priority) {
1509                         newpos = &pol->bydst_inexact_list;
1510                         continue;
1511                 }
1512                 if (delpol)
1513                         break;
1514         }
1515
1516         if (newpos)
1517                 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1518         else
1519                 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1520
1521         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1522                 pol->pos = i;
1523                 i++;
1524         }
1525 }
1526
1527 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1528                                                    struct xfrm_policy *policy,
1529                                                    bool excl)
1530 {
1531         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1532
1533         hlist_for_each_entry(pol, chain, bydst) {
1534                 if (pol->type == policy->type &&
1535                     pol->if_id == policy->if_id &&
1536                     !selector_cmp(&pol->selector, &policy->selector) &&
1537                     xfrm_policy_mark_match(&policy->mark, pol) &&
1538                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1539                     !WARN_ON(delpol)) {
1540                         if (excl)
1541                                 return ERR_PTR(-EEXIST);
1542                         delpol = pol;
1543                         if (policy->priority > pol->priority)
1544                                 continue;
1545                 } else if (policy->priority >= pol->priority) {
1546                         newpos = pol;
1547                         continue;
1548                 }
1549                 if (delpol)
1550                         break;
1551         }
1552
1553         if (newpos)
1554                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1555         else
1556                 hlist_add_head_rcu(&policy->bydst, chain);
1557
1558         return delpol;
1559 }
1560
1561 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1562 {
1563         struct net *net = xp_net(policy);
1564         struct xfrm_policy *delpol;
1565         struct hlist_head *chain;
1566
1567         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1568         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1569         if (chain)
1570                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1571         else
1572                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1573
1574         if (IS_ERR(delpol)) {
1575                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1576                 return PTR_ERR(delpol);
1577         }
1578
1579         __xfrm_policy_link(policy, dir);
1580
1581         /* After previous checking, family can either be AF_INET or AF_INET6 */
1582         if (policy->family == AF_INET)
1583                 rt_genid_bump_ipv4(net);
1584         else
1585                 rt_genid_bump_ipv6(net);
1586
1587         if (delpol) {
1588                 xfrm_policy_requeue(delpol, policy);
1589                 __xfrm_policy_unlink(delpol, dir);
1590         }
1591         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1592         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1593         policy->curlft.add_time = ktime_get_real_seconds();
1594         policy->curlft.use_time = 0;
1595         if (!mod_timer(&policy->timer, jiffies + HZ))
1596                 xfrm_pol_hold(policy);
1597         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1598
1599         if (delpol)
1600                 xfrm_policy_kill(delpol);
1601         else if (xfrm_bydst_should_resize(net, dir, NULL))
1602                 schedule_work(&net->xfrm.policy_hash_work);
1603
1604         return 0;
1605 }
1606 EXPORT_SYMBOL(xfrm_policy_insert);
1607
1608 static struct xfrm_policy *
1609 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1610                         u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1611                         struct xfrm_sec_ctx *ctx)
1612 {
1613         struct xfrm_policy *pol;
1614
1615         if (!chain)
1616                 return NULL;
1617
1618         hlist_for_each_entry(pol, chain, bydst) {
1619                 if (pol->type == type &&
1620                     pol->if_id == if_id &&
1621                     xfrm_policy_mark_match(mark, pol) &&
1622                     !selector_cmp(sel, &pol->selector) &&
1623                     xfrm_sec_ctx_match(ctx, pol->security))
1624                         return pol;
1625         }
1626
1627         return NULL;
1628 }
1629
1630 struct xfrm_policy *
1631 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1632                       u8 type, int dir, struct xfrm_selector *sel,
1633                       struct xfrm_sec_ctx *ctx, int delete, int *err)
1634 {
1635         struct xfrm_pol_inexact_bin *bin = NULL;
1636         struct xfrm_policy *pol, *ret = NULL;
1637         struct hlist_head *chain;
1638
1639         *err = 0;
1640         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1641         chain = policy_hash_bysel(net, sel, sel->family, dir);
1642         if (!chain) {
1643                 struct xfrm_pol_inexact_candidates cand;
1644                 int i;
1645
1646                 bin = xfrm_policy_inexact_lookup(net, type,
1647                                                  sel->family, dir, if_id);
1648                 if (!bin) {
1649                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1650                         return NULL;
1651                 }
1652
1653                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1654                                                          &sel->saddr,
1655                                                          &sel->daddr)) {
1656                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1657                         return NULL;
1658                 }
1659
1660                 pol = NULL;
1661                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1662                         struct xfrm_policy *tmp;
1663
1664                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1665                                                       if_id, type, dir,
1666                                                       sel, ctx);
1667                         if (!tmp)
1668                                 continue;
1669
1670                         if (!pol || tmp->pos < pol->pos)
1671                                 pol = tmp;
1672                 }
1673         } else {
1674                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1675                                               sel, ctx);
1676         }
1677
1678         if (pol) {
1679                 xfrm_pol_hold(pol);
1680                 if (delete) {
1681                         *err = security_xfrm_policy_delete(pol->security);
1682                         if (*err) {
1683                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1684                                 return pol;
1685                         }
1686                         __xfrm_policy_unlink(pol, dir);
1687                 }
1688                 ret = pol;
1689         }
1690         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1691
1692         if (ret && delete)
1693                 xfrm_policy_kill(ret);
1694         if (bin && delete)
1695                 xfrm_policy_inexact_prune_bin(bin);
1696         return ret;
1697 }
1698 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1699
1700 struct xfrm_policy *
1701 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1702                  u8 type, int dir, u32 id, int delete, int *err)
1703 {
1704         struct xfrm_policy *pol, *ret;
1705         struct hlist_head *chain;
1706
1707         *err = -ENOENT;
1708         if (xfrm_policy_id2dir(id) != dir)
1709                 return NULL;
1710
1711         *err = 0;
1712         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1713         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1714         ret = NULL;
1715         hlist_for_each_entry(pol, chain, byidx) {
1716                 if (pol->type == type && pol->index == id &&
1717                     pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1718                         xfrm_pol_hold(pol);
1719                         if (delete) {
1720                                 *err = security_xfrm_policy_delete(
1721                                                                 pol->security);
1722                                 if (*err) {
1723                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1724                                         return pol;
1725                                 }
1726                                 __xfrm_policy_unlink(pol, dir);
1727                         }
1728                         ret = pol;
1729                         break;
1730                 }
1731         }
1732         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1733
1734         if (ret && delete)
1735                 xfrm_policy_kill(ret);
1736         return ret;
1737 }
1738 EXPORT_SYMBOL(xfrm_policy_byid);
1739
1740 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1741 static inline int
1742 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1743 {
1744         struct xfrm_policy *pol;
1745         int err = 0;
1746
1747         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1748                 if (pol->walk.dead ||
1749                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1750                     pol->type != type)
1751                         continue;
1752
1753                 err = security_xfrm_policy_delete(pol->security);
1754                 if (err) {
1755                         xfrm_audit_policy_delete(pol, 0, task_valid);
1756                         return err;
1757                 }
1758         }
1759         return err;
1760 }
1761 #else
1762 static inline int
1763 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1764 {
1765         return 0;
1766 }
1767 #endif
1768
1769 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1770 {
1771         int dir, err = 0, cnt = 0;
1772         struct xfrm_policy *pol;
1773
1774         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1775
1776         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1777         if (err)
1778                 goto out;
1779
1780 again:
1781         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1782                 dir = xfrm_policy_id2dir(pol->index);
1783                 if (pol->walk.dead ||
1784                     dir >= XFRM_POLICY_MAX ||
1785                     pol->type != type)
1786                         continue;
1787
1788                 __xfrm_policy_unlink(pol, dir);
1789                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1790                 cnt++;
1791                 xfrm_audit_policy_delete(pol, 1, task_valid);
1792                 xfrm_policy_kill(pol);
1793                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1794                 goto again;
1795         }
1796         if (cnt)
1797                 __xfrm_policy_inexact_flush(net);
1798         else
1799                 err = -ESRCH;
1800 out:
1801         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1802         return err;
1803 }
1804 EXPORT_SYMBOL(xfrm_policy_flush);
1805
1806 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1807                      int (*func)(struct xfrm_policy *, int, int, void*),
1808                      void *data)
1809 {
1810         struct xfrm_policy *pol;
1811         struct xfrm_policy_walk_entry *x;
1812         int error = 0;
1813
1814         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1815             walk->type != XFRM_POLICY_TYPE_ANY)
1816                 return -EINVAL;
1817
1818         if (list_empty(&walk->walk.all) && walk->seq != 0)
1819                 return 0;
1820
1821         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1822         if (list_empty(&walk->walk.all))
1823                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1824         else
1825                 x = list_first_entry(&walk->walk.all,
1826                                      struct xfrm_policy_walk_entry, all);
1827
1828         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1829                 if (x->dead)
1830                         continue;
1831                 pol = container_of(x, struct xfrm_policy, walk);
1832                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1833                     walk->type != pol->type)
1834                         continue;
1835                 error = func(pol, xfrm_policy_id2dir(pol->index),
1836                              walk->seq, data);
1837                 if (error) {
1838                         list_move_tail(&walk->walk.all, &x->all);
1839                         goto out;
1840                 }
1841                 walk->seq++;
1842         }
1843         if (walk->seq == 0) {
1844                 error = -ENOENT;
1845                 goto out;
1846         }
1847         list_del_init(&walk->walk.all);
1848 out:
1849         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1850         return error;
1851 }
1852 EXPORT_SYMBOL(xfrm_policy_walk);
1853
1854 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1855 {
1856         INIT_LIST_HEAD(&walk->walk.all);
1857         walk->walk.dead = 1;
1858         walk->type = type;
1859         walk->seq = 0;
1860 }
1861 EXPORT_SYMBOL(xfrm_policy_walk_init);
1862
1863 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1864 {
1865         if (list_empty(&walk->walk.all))
1866                 return;
1867
1868         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1869         list_del(&walk->walk.all);
1870         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1871 }
1872 EXPORT_SYMBOL(xfrm_policy_walk_done);
1873
1874 /*
1875  * Find policy to apply to this flow.
1876  *
1877  * Returns 0 if policy found, else an -errno.
1878  */
1879 static int xfrm_policy_match(const struct xfrm_policy *pol,
1880                              const struct flowi *fl,
1881                              u8 type, u16 family, int dir, u32 if_id)
1882 {
1883         const struct xfrm_selector *sel = &pol->selector;
1884         int ret = -ESRCH;
1885         bool match;
1886
1887         if (pol->family != family ||
1888             pol->if_id != if_id ||
1889             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1890             pol->type != type)
1891                 return ret;
1892
1893         match = xfrm_selector_match(sel, fl, family);
1894         if (match)
1895                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1896                                                   dir);
1897         return ret;
1898 }
1899
1900 static struct xfrm_pol_inexact_node *
1901 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1902                                 seqcount_t *count,
1903                                 const xfrm_address_t *addr, u16 family)
1904 {
1905         const struct rb_node *parent;
1906         int seq;
1907
1908 again:
1909         seq = read_seqcount_begin(count);
1910
1911         parent = rcu_dereference_raw(r->rb_node);
1912         while (parent) {
1913                 struct xfrm_pol_inexact_node *node;
1914                 int delta;
1915
1916                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1917
1918                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1919                                                node->prefixlen, family);
1920                 if (delta < 0) {
1921                         parent = rcu_dereference_raw(parent->rb_left);
1922                         continue;
1923                 } else if (delta > 0) {
1924                         parent = rcu_dereference_raw(parent->rb_right);
1925                         continue;
1926                 }
1927
1928                 return node;
1929         }
1930
1931         if (read_seqcount_retry(count, seq))
1932                 goto again;
1933
1934         return NULL;
1935 }
1936
1937 static bool
1938 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1939                                     struct xfrm_pol_inexact_bin *b,
1940                                     const xfrm_address_t *saddr,
1941                                     const xfrm_address_t *daddr)
1942 {
1943         struct xfrm_pol_inexact_node *n;
1944         u16 family;
1945
1946         if (!b)
1947                 return false;
1948
1949         family = b->k.family;
1950         memset(cand, 0, sizeof(*cand));
1951         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1952
1953         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1954                                             family);
1955         if (n) {
1956                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1957                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1958                                                     family);
1959                 if (n)
1960                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1961         }
1962
1963         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1964                                             family);
1965         if (n)
1966                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1967
1968         return true;
1969 }
1970
1971 static struct xfrm_pol_inexact_bin *
1972 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1973                                u8 dir, u32 if_id)
1974 {
1975         struct xfrm_pol_inexact_key k = {
1976                 .family = family,
1977                 .type = type,
1978                 .dir = dir,
1979                 .if_id = if_id,
1980         };
1981
1982         write_pnet(&k.net, net);
1983
1984         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1985                                  xfrm_pol_inexact_params);
1986 }
1987
1988 static struct xfrm_pol_inexact_bin *
1989 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1990                            u8 dir, u32 if_id)
1991 {
1992         struct xfrm_pol_inexact_bin *bin;
1993
1994         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1995
1996         rcu_read_lock();
1997         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
1998         rcu_read_unlock();
1999
2000         return bin;
2001 }
2002
2003 static struct xfrm_policy *
2004 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2005                               struct xfrm_policy *prefer,
2006                               const struct flowi *fl,
2007                               u8 type, u16 family, int dir, u32 if_id)
2008 {
2009         u32 priority = prefer ? prefer->priority : ~0u;
2010         struct xfrm_policy *pol;
2011
2012         if (!chain)
2013                 return NULL;
2014
2015         hlist_for_each_entry_rcu(pol, chain, bydst) {
2016                 int err;
2017
2018                 if (pol->priority > priority)
2019                         break;
2020
2021                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2022                 if (err) {
2023                         if (err != -ESRCH)
2024                                 return ERR_PTR(err);
2025
2026                         continue;
2027                 }
2028
2029                 if (prefer) {
2030                         /* matches.  Is it older than *prefer? */
2031                         if (pol->priority == priority &&
2032                             prefer->pos < pol->pos)
2033                                 return prefer;
2034                 }
2035
2036                 return pol;
2037         }
2038
2039         return NULL;
2040 }
2041
2042 static struct xfrm_policy *
2043 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2044                             struct xfrm_policy *prefer,
2045                             const struct flowi *fl,
2046                             u8 type, u16 family, int dir, u32 if_id)
2047 {
2048         struct xfrm_policy *tmp;
2049         int i;
2050
2051         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2052                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2053                                                     prefer,
2054                                                     fl, type, family, dir,
2055                                                     if_id);
2056                 if (!tmp)
2057                         continue;
2058
2059                 if (IS_ERR(tmp))
2060                         return tmp;
2061                 prefer = tmp;
2062         }
2063
2064         return prefer;
2065 }
2066
2067 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2068                                                      const struct flowi *fl,
2069                                                      u16 family, u8 dir,
2070                                                      u32 if_id)
2071 {
2072         struct xfrm_pol_inexact_candidates cand;
2073         const xfrm_address_t *daddr, *saddr;
2074         struct xfrm_pol_inexact_bin *bin;
2075         struct xfrm_policy *pol, *ret;
2076         struct hlist_head *chain;
2077         unsigned int sequence;
2078         int err;
2079
2080         daddr = xfrm_flowi_daddr(fl, family);
2081         saddr = xfrm_flowi_saddr(fl, family);
2082         if (unlikely(!daddr || !saddr))
2083                 return NULL;
2084
2085         rcu_read_lock();
2086  retry:
2087         do {
2088                 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2089                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2090         } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2091
2092         ret = NULL;
2093         hlist_for_each_entry_rcu(pol, chain, bydst) {
2094                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2095                 if (err) {
2096                         if (err == -ESRCH)
2097                                 continue;
2098                         else {
2099                                 ret = ERR_PTR(err);
2100                                 goto fail;
2101                         }
2102                 } else {
2103                         ret = pol;
2104                         break;
2105                 }
2106         }
2107         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2108         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2109                                                          daddr))
2110                 goto skip_inexact;
2111
2112         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2113                                           family, dir, if_id);
2114         if (pol) {
2115                 ret = pol;
2116                 if (IS_ERR(pol))
2117                         goto fail;
2118         }
2119
2120 skip_inexact:
2121         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2122                 goto retry;
2123
2124         if (ret && !xfrm_pol_hold_rcu(ret))
2125                 goto retry;
2126 fail:
2127         rcu_read_unlock();
2128
2129         return ret;
2130 }
2131
2132 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2133                                               const struct flowi *fl,
2134                                               u16 family, u8 dir, u32 if_id)
2135 {
2136 #ifdef CONFIG_XFRM_SUB_POLICY
2137         struct xfrm_policy *pol;
2138
2139         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2140                                         dir, if_id);
2141         if (pol != NULL)
2142                 return pol;
2143 #endif
2144         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2145                                          dir, if_id);
2146 }
2147
2148 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2149                                                  const struct flowi *fl,
2150                                                  u16 family, u32 if_id)
2151 {
2152         struct xfrm_policy *pol;
2153
2154         rcu_read_lock();
2155  again:
2156         pol = rcu_dereference(sk->sk_policy[dir]);
2157         if (pol != NULL) {
2158                 bool match;
2159                 int err = 0;
2160
2161                 if (pol->family != family) {
2162                         pol = NULL;
2163                         goto out;
2164                 }
2165
2166                 match = xfrm_selector_match(&pol->selector, fl, family);
2167                 if (match) {
2168                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2169                             pol->if_id != if_id) {
2170                                 pol = NULL;
2171                                 goto out;
2172                         }
2173                         err = security_xfrm_policy_lookup(pol->security,
2174                                                       fl->flowi_secid,
2175                                                       dir);
2176                         if (!err) {
2177                                 if (!xfrm_pol_hold_rcu(pol))
2178                                         goto again;
2179                         } else if (err == -ESRCH) {
2180                                 pol = NULL;
2181                         } else {
2182                                 pol = ERR_PTR(err);
2183                         }
2184                 } else
2185                         pol = NULL;
2186         }
2187 out:
2188         rcu_read_unlock();
2189         return pol;
2190 }
2191
2192 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2193 {
2194         struct net *net = xp_net(pol);
2195
2196         list_add(&pol->walk.all, &net->xfrm.policy_all);
2197         net->xfrm.policy_count[dir]++;
2198         xfrm_pol_hold(pol);
2199 }
2200
2201 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2202                                                 int dir)
2203 {
2204         struct net *net = xp_net(pol);
2205
2206         if (list_empty(&pol->walk.all))
2207                 return NULL;
2208
2209         /* Socket policies are not hashed. */
2210         if (!hlist_unhashed(&pol->bydst)) {
2211                 hlist_del_rcu(&pol->bydst);
2212                 hlist_del_init(&pol->bydst_inexact_list);
2213                 hlist_del(&pol->byidx);
2214         }
2215
2216         list_del_init(&pol->walk.all);
2217         net->xfrm.policy_count[dir]--;
2218
2219         return pol;
2220 }
2221
2222 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2223 {
2224         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2225 }
2226
2227 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2228 {
2229         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2230 }
2231
2232 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2233 {
2234         struct net *net = xp_net(pol);
2235
2236         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2237         pol = __xfrm_policy_unlink(pol, dir);
2238         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2239         if (pol) {
2240                 xfrm_policy_kill(pol);
2241                 return 0;
2242         }
2243         return -ENOENT;
2244 }
2245 EXPORT_SYMBOL(xfrm_policy_delete);
2246
2247 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2248 {
2249         struct net *net = sock_net(sk);
2250         struct xfrm_policy *old_pol;
2251
2252 #ifdef CONFIG_XFRM_SUB_POLICY
2253         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2254                 return -EINVAL;
2255 #endif
2256
2257         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2258         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2259                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2260         if (pol) {
2261                 pol->curlft.add_time = ktime_get_real_seconds();
2262                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2263                 xfrm_sk_policy_link(pol, dir);
2264         }
2265         rcu_assign_pointer(sk->sk_policy[dir], pol);
2266         if (old_pol) {
2267                 if (pol)
2268                         xfrm_policy_requeue(old_pol, pol);
2269
2270                 /* Unlinking succeeds always. This is the only function
2271                  * allowed to delete or replace socket policy.
2272                  */
2273                 xfrm_sk_policy_unlink(old_pol, dir);
2274         }
2275         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2276
2277         if (old_pol) {
2278                 xfrm_policy_kill(old_pol);
2279         }
2280         return 0;
2281 }
2282
2283 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2284 {
2285         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2286         struct net *net = xp_net(old);
2287
2288         if (newp) {
2289                 newp->selector = old->selector;
2290                 if (security_xfrm_policy_clone(old->security,
2291                                                &newp->security)) {
2292                         kfree(newp);
2293                         return NULL;  /* ENOMEM */
2294                 }
2295                 newp->lft = old->lft;
2296                 newp->curlft = old->curlft;
2297                 newp->mark = old->mark;
2298                 newp->if_id = old->if_id;
2299                 newp->action = old->action;
2300                 newp->flags = old->flags;
2301                 newp->xfrm_nr = old->xfrm_nr;
2302                 newp->index = old->index;
2303                 newp->type = old->type;
2304                 newp->family = old->family;
2305                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2306                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2307                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2308                 xfrm_sk_policy_link(newp, dir);
2309                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2310                 xfrm_pol_put(newp);
2311         }
2312         return newp;
2313 }
2314
2315 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2316 {
2317         const struct xfrm_policy *p;
2318         struct xfrm_policy *np;
2319         int i, ret = 0;
2320
2321         rcu_read_lock();
2322         for (i = 0; i < 2; i++) {
2323                 p = rcu_dereference(osk->sk_policy[i]);
2324                 if (p) {
2325                         np = clone_policy(p, i);
2326                         if (unlikely(!np)) {
2327                                 ret = -ENOMEM;
2328                                 break;
2329                         }
2330                         rcu_assign_pointer(sk->sk_policy[i], np);
2331                 }
2332         }
2333         rcu_read_unlock();
2334         return ret;
2335 }
2336
2337 static int
2338 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2339                xfrm_address_t *remote, unsigned short family, u32 mark)
2340 {
2341         int err;
2342         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2343
2344         if (unlikely(afinfo == NULL))
2345                 return -EINVAL;
2346         err = afinfo->get_saddr(net, oif, local, remote, mark);
2347         rcu_read_unlock();
2348         return err;
2349 }
2350
2351 /* Resolve list of templates for the flow, given policy. */
2352
2353 static int
2354 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2355                       struct xfrm_state **xfrm, unsigned short family)
2356 {
2357         struct net *net = xp_net(policy);
2358         int nx;
2359         int i, error;
2360         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2361         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2362         xfrm_address_t tmp;
2363
2364         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2365                 struct xfrm_state *x;
2366                 xfrm_address_t *remote = daddr;
2367                 xfrm_address_t *local  = saddr;
2368                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2369
2370                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2371                     tmpl->mode == XFRM_MODE_BEET) {
2372                         remote = &tmpl->id.daddr;
2373                         local = &tmpl->saddr;
2374                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2375                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2376                                                        &tmp, remote,
2377                                                        tmpl->encap_family, 0);
2378                                 if (error)
2379                                         goto fail;
2380                                 local = &tmp;
2381                         }
2382                 }
2383
2384                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2385                                     family, policy->if_id);
2386
2387                 if (x && x->km.state == XFRM_STATE_VALID) {
2388                         xfrm[nx++] = x;
2389                         daddr = remote;
2390                         saddr = local;
2391                         continue;
2392                 }
2393                 if (x) {
2394                         error = (x->km.state == XFRM_STATE_ERROR ?
2395                                  -EINVAL : -EAGAIN);
2396                         xfrm_state_put(x);
2397                 } else if (error == -ESRCH) {
2398                         error = -EAGAIN;
2399                 }
2400
2401                 if (!tmpl->optional)
2402                         goto fail;
2403         }
2404         return nx;
2405
2406 fail:
2407         for (nx--; nx >= 0; nx--)
2408                 xfrm_state_put(xfrm[nx]);
2409         return error;
2410 }
2411
2412 static int
2413 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2414                   struct xfrm_state **xfrm, unsigned short family)
2415 {
2416         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2417         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2418         int cnx = 0;
2419         int error;
2420         int ret;
2421         int i;
2422
2423         for (i = 0; i < npols; i++) {
2424                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2425                         error = -ENOBUFS;
2426                         goto fail;
2427                 }
2428
2429                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2430                 if (ret < 0) {
2431                         error = ret;
2432                         goto fail;
2433                 } else
2434                         cnx += ret;
2435         }
2436
2437         /* found states are sorted for outbound processing */
2438         if (npols > 1)
2439                 xfrm_state_sort(xfrm, tpp, cnx, family);
2440
2441         return cnx;
2442
2443  fail:
2444         for (cnx--; cnx >= 0; cnx--)
2445                 xfrm_state_put(tpp[cnx]);
2446         return error;
2447
2448 }
2449
2450 static int xfrm_get_tos(const struct flowi *fl, int family)
2451 {
2452         if (family == AF_INET)
2453                 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2454
2455         return 0;
2456 }
2457
2458 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2459 {
2460         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2461         struct dst_ops *dst_ops;
2462         struct xfrm_dst *xdst;
2463
2464         if (!afinfo)
2465                 return ERR_PTR(-EINVAL);
2466
2467         switch (family) {
2468         case AF_INET:
2469                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2470                 break;
2471 #if IS_ENABLED(CONFIG_IPV6)
2472         case AF_INET6:
2473                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2474                 break;
2475 #endif
2476         default:
2477                 BUG();
2478         }
2479         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2480
2481         if (likely(xdst)) {
2482                 struct dst_entry *dst = &xdst->u.dst;
2483
2484                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2485         } else
2486                 xdst = ERR_PTR(-ENOBUFS);
2487
2488         rcu_read_unlock();
2489
2490         return xdst;
2491 }
2492
2493 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2494                            int nfheader_len)
2495 {
2496         if (dst->ops->family == AF_INET6) {
2497                 struct rt6_info *rt = (struct rt6_info *)dst;
2498                 path->path_cookie = rt6_get_cookie(rt);
2499                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2500         }
2501 }
2502
2503 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2504                                 const struct flowi *fl)
2505 {
2506         const struct xfrm_policy_afinfo *afinfo =
2507                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2508         int err;
2509
2510         if (!afinfo)
2511                 return -EINVAL;
2512
2513         err = afinfo->fill_dst(xdst, dev, fl);
2514
2515         rcu_read_unlock();
2516
2517         return err;
2518 }
2519
2520
2521 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2522  * all the metrics... Shortly, bundle a bundle.
2523  */
2524
2525 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2526                                             struct xfrm_state **xfrm,
2527                                             struct xfrm_dst **bundle,
2528                                             int nx,
2529                                             const struct flowi *fl,
2530                                             struct dst_entry *dst)
2531 {
2532         const struct xfrm_state_afinfo *afinfo;
2533         const struct xfrm_mode *inner_mode;
2534         struct net *net = xp_net(policy);
2535         unsigned long now = jiffies;
2536         struct net_device *dev;
2537         struct xfrm_dst *xdst_prev = NULL;
2538         struct xfrm_dst *xdst0 = NULL;
2539         int i = 0;
2540         int err;
2541         int header_len = 0;
2542         int nfheader_len = 0;
2543         int trailer_len = 0;
2544         int tos;
2545         int family = policy->selector.family;
2546         xfrm_address_t saddr, daddr;
2547
2548         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2549
2550         tos = xfrm_get_tos(fl, family);
2551
2552         dst_hold(dst);
2553
2554         for (; i < nx; i++) {
2555                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2556                 struct dst_entry *dst1 = &xdst->u.dst;
2557
2558                 err = PTR_ERR(xdst);
2559                 if (IS_ERR(xdst)) {
2560                         dst_release(dst);
2561                         goto put_states;
2562                 }
2563
2564                 bundle[i] = xdst;
2565                 if (!xdst_prev)
2566                         xdst0 = xdst;
2567                 else
2568                         /* Ref count is taken during xfrm_alloc_dst()
2569                          * No need to do dst_clone() on dst1
2570                          */
2571                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2572
2573                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2574                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2575                                                         xfrm_af2proto(family));
2576                         if (!inner_mode) {
2577                                 err = -EAFNOSUPPORT;
2578                                 dst_release(dst);
2579                                 goto put_states;
2580                         }
2581                 } else
2582                         inner_mode = &xfrm[i]->inner_mode;
2583
2584                 xdst->route = dst;
2585                 dst_copy_metrics(dst1, dst);
2586
2587                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2588                         __u32 mark = 0;
2589
2590                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2591                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2592
2593                         family = xfrm[i]->props.family;
2594                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2595                                               &saddr, &daddr, family, mark);
2596                         err = PTR_ERR(dst);
2597                         if (IS_ERR(dst))
2598                                 goto put_states;
2599                 } else
2600                         dst_hold(dst);
2601
2602                 dst1->xfrm = xfrm[i];
2603                 xdst->xfrm_genid = xfrm[i]->genid;
2604
2605                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2606                 dst1->lastuse = now;
2607
2608                 dst1->input = dst_discard;
2609
2610                 rcu_read_lock();
2611                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2612                 if (likely(afinfo))
2613                         dst1->output = afinfo->output;
2614                 else
2615                         dst1->output = dst_discard_out;
2616                 rcu_read_unlock();
2617
2618                 xdst_prev = xdst;
2619
2620                 header_len += xfrm[i]->props.header_len;
2621                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2622                         nfheader_len += xfrm[i]->props.header_len;
2623                 trailer_len += xfrm[i]->props.trailer_len;
2624         }
2625
2626         xfrm_dst_set_child(xdst_prev, dst);
2627         xdst0->path = dst;
2628
2629         err = -ENODEV;
2630         dev = dst->dev;
2631         if (!dev)
2632                 goto free_dst;
2633
2634         xfrm_init_path(xdst0, dst, nfheader_len);
2635         xfrm_init_pmtu(bundle, nx);
2636
2637         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2638              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2639                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2640                 if (err)
2641                         goto free_dst;
2642
2643                 xdst_prev->u.dst.header_len = header_len;
2644                 xdst_prev->u.dst.trailer_len = trailer_len;
2645                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2646                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2647         }
2648
2649         return &xdst0->u.dst;
2650
2651 put_states:
2652         for (; i < nx; i++)
2653                 xfrm_state_put(xfrm[i]);
2654 free_dst:
2655         if (xdst0)
2656                 dst_release_immediate(&xdst0->u.dst);
2657
2658         return ERR_PTR(err);
2659 }
2660
2661 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2662                                 struct xfrm_policy **pols,
2663                                 int *num_pols, int *num_xfrms)
2664 {
2665         int i;
2666
2667         if (*num_pols == 0 || !pols[0]) {
2668                 *num_pols = 0;
2669                 *num_xfrms = 0;
2670                 return 0;
2671         }
2672         if (IS_ERR(pols[0]))
2673                 return PTR_ERR(pols[0]);
2674
2675         *num_xfrms = pols[0]->xfrm_nr;
2676
2677 #ifdef CONFIG_XFRM_SUB_POLICY
2678         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2679             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2680                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2681                                                     XFRM_POLICY_TYPE_MAIN,
2682                                                     fl, family,
2683                                                     XFRM_POLICY_OUT,
2684                                                     pols[0]->if_id);
2685                 if (pols[1]) {
2686                         if (IS_ERR(pols[1])) {
2687                                 xfrm_pols_put(pols, *num_pols);
2688                                 return PTR_ERR(pols[1]);
2689                         }
2690                         (*num_pols)++;
2691                         (*num_xfrms) += pols[1]->xfrm_nr;
2692                 }
2693         }
2694 #endif
2695         for (i = 0; i < *num_pols; i++) {
2696                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2697                         *num_xfrms = -1;
2698                         break;
2699                 }
2700         }
2701
2702         return 0;
2703
2704 }
2705
2706 static struct xfrm_dst *
2707 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2708                                const struct flowi *fl, u16 family,
2709                                struct dst_entry *dst_orig)
2710 {
2711         struct net *net = xp_net(pols[0]);
2712         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2713         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2714         struct xfrm_dst *xdst;
2715         struct dst_entry *dst;
2716         int err;
2717
2718         /* Try to instantiate a bundle */
2719         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2720         if (err <= 0) {
2721                 if (err == 0)
2722                         return NULL;
2723
2724                 if (err != -EAGAIN)
2725                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2726                 return ERR_PTR(err);
2727         }
2728
2729         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2730         if (IS_ERR(dst)) {
2731                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2732                 return ERR_CAST(dst);
2733         }
2734
2735         xdst = (struct xfrm_dst *)dst;
2736         xdst->num_xfrms = err;
2737         xdst->num_pols = num_pols;
2738         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2739         xdst->policy_genid = atomic_read(&pols[0]->genid);
2740
2741         return xdst;
2742 }
2743
2744 static void xfrm_policy_queue_process(struct timer_list *t)
2745 {
2746         struct sk_buff *skb;
2747         struct sock *sk;
2748         struct dst_entry *dst;
2749         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2750         struct net *net = xp_net(pol);
2751         struct xfrm_policy_queue *pq = &pol->polq;
2752         struct flowi fl;
2753         struct sk_buff_head list;
2754         __u32 skb_mark;
2755
2756         spin_lock(&pq->hold_queue.lock);
2757         skb = skb_peek(&pq->hold_queue);
2758         if (!skb) {
2759                 spin_unlock(&pq->hold_queue.lock);
2760                 goto out;
2761         }
2762         dst = skb_dst(skb);
2763         sk = skb->sk;
2764
2765         /* Fixup the mark to support VTI. */
2766         skb_mark = skb->mark;
2767         skb->mark = pol->mark.v;
2768         xfrm_decode_session(skb, &fl, dst->ops->family);
2769         skb->mark = skb_mark;
2770         spin_unlock(&pq->hold_queue.lock);
2771
2772         dst_hold(xfrm_dst_path(dst));
2773         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2774         if (IS_ERR(dst))
2775                 goto purge_queue;
2776
2777         if (dst->flags & DST_XFRM_QUEUE) {
2778                 dst_release(dst);
2779
2780                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2781                         goto purge_queue;
2782
2783                 pq->timeout = pq->timeout << 1;
2784                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2785                         xfrm_pol_hold(pol);
2786                 goto out;
2787         }
2788
2789         dst_release(dst);
2790
2791         __skb_queue_head_init(&list);
2792
2793         spin_lock(&pq->hold_queue.lock);
2794         pq->timeout = 0;
2795         skb_queue_splice_init(&pq->hold_queue, &list);
2796         spin_unlock(&pq->hold_queue.lock);
2797
2798         while (!skb_queue_empty(&list)) {
2799                 skb = __skb_dequeue(&list);
2800
2801                 /* Fixup the mark to support VTI. */
2802                 skb_mark = skb->mark;
2803                 skb->mark = pol->mark.v;
2804                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2805                 skb->mark = skb_mark;
2806
2807                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2808                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2809                 if (IS_ERR(dst)) {
2810                         kfree_skb(skb);
2811                         continue;
2812                 }
2813
2814                 nf_reset_ct(skb);
2815                 skb_dst_drop(skb);
2816                 skb_dst_set(skb, dst);
2817
2818                 dst_output(net, skb->sk, skb);
2819         }
2820
2821 out:
2822         xfrm_pol_put(pol);
2823         return;
2824
2825 purge_queue:
2826         pq->timeout = 0;
2827         skb_queue_purge(&pq->hold_queue);
2828         xfrm_pol_put(pol);
2829 }
2830
2831 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2832 {
2833         unsigned long sched_next;
2834         struct dst_entry *dst = skb_dst(skb);
2835         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2836         struct xfrm_policy *pol = xdst->pols[0];
2837         struct xfrm_policy_queue *pq = &pol->polq;
2838
2839         if (unlikely(skb_fclone_busy(sk, skb))) {
2840                 kfree_skb(skb);
2841                 return 0;
2842         }
2843
2844         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2845                 kfree_skb(skb);
2846                 return -EAGAIN;
2847         }
2848
2849         skb_dst_force(skb);
2850
2851         spin_lock_bh(&pq->hold_queue.lock);
2852
2853         if (!pq->timeout)
2854                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2855
2856         sched_next = jiffies + pq->timeout;
2857
2858         if (del_timer(&pq->hold_timer)) {
2859                 if (time_before(pq->hold_timer.expires, sched_next))
2860                         sched_next = pq->hold_timer.expires;
2861                 xfrm_pol_put(pol);
2862         }
2863
2864         __skb_queue_tail(&pq->hold_queue, skb);
2865         if (!mod_timer(&pq->hold_timer, sched_next))
2866                 xfrm_pol_hold(pol);
2867
2868         spin_unlock_bh(&pq->hold_queue.lock);
2869
2870         return 0;
2871 }
2872
2873 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2874                                                  struct xfrm_flo *xflo,
2875                                                  const struct flowi *fl,
2876                                                  int num_xfrms,
2877                                                  u16 family)
2878 {
2879         int err;
2880         struct net_device *dev;
2881         struct dst_entry *dst;
2882         struct dst_entry *dst1;
2883         struct xfrm_dst *xdst;
2884
2885         xdst = xfrm_alloc_dst(net, family);
2886         if (IS_ERR(xdst))
2887                 return xdst;
2888
2889         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2890             net->xfrm.sysctl_larval_drop ||
2891             num_xfrms <= 0)
2892                 return xdst;
2893
2894         dst = xflo->dst_orig;
2895         dst1 = &xdst->u.dst;
2896         dst_hold(dst);
2897         xdst->route = dst;
2898
2899         dst_copy_metrics(dst1, dst);
2900
2901         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2902         dst1->flags |= DST_XFRM_QUEUE;
2903         dst1->lastuse = jiffies;
2904
2905         dst1->input = dst_discard;
2906         dst1->output = xdst_queue_output;
2907
2908         dst_hold(dst);
2909         xfrm_dst_set_child(xdst, dst);
2910         xdst->path = dst;
2911
2912         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2913
2914         err = -ENODEV;
2915         dev = dst->dev;
2916         if (!dev)
2917                 goto free_dst;
2918
2919         err = xfrm_fill_dst(xdst, dev, fl);
2920         if (err)
2921                 goto free_dst;
2922
2923 out:
2924         return xdst;
2925
2926 free_dst:
2927         dst_release(dst1);
2928         xdst = ERR_PTR(err);
2929         goto out;
2930 }
2931
2932 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2933                                            const struct flowi *fl,
2934                                            u16 family, u8 dir,
2935                                            struct xfrm_flo *xflo, u32 if_id)
2936 {
2937         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2938         int num_pols = 0, num_xfrms = 0, err;
2939         struct xfrm_dst *xdst;
2940
2941         /* Resolve policies to use if we couldn't get them from
2942          * previous cache entry */
2943         num_pols = 1;
2944         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2945         err = xfrm_expand_policies(fl, family, pols,
2946                                            &num_pols, &num_xfrms);
2947         if (err < 0)
2948                 goto inc_error;
2949         if (num_pols == 0)
2950                 return NULL;
2951         if (num_xfrms <= 0)
2952                 goto make_dummy_bundle;
2953
2954         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2955                                               xflo->dst_orig);
2956         if (IS_ERR(xdst)) {
2957                 err = PTR_ERR(xdst);
2958                 if (err == -EREMOTE) {
2959                         xfrm_pols_put(pols, num_pols);
2960                         return NULL;
2961                 }
2962
2963                 if (err != -EAGAIN)
2964                         goto error;
2965                 goto make_dummy_bundle;
2966         } else if (xdst == NULL) {
2967                 num_xfrms = 0;
2968                 goto make_dummy_bundle;
2969         }
2970
2971         return xdst;
2972
2973 make_dummy_bundle:
2974         /* We found policies, but there's no bundles to instantiate:
2975          * either because the policy blocks, has no transformations or
2976          * we could not build template (no xfrm_states).*/
2977         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2978         if (IS_ERR(xdst)) {
2979                 xfrm_pols_put(pols, num_pols);
2980                 return ERR_CAST(xdst);
2981         }
2982         xdst->num_pols = num_pols;
2983         xdst->num_xfrms = num_xfrms;
2984         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2985
2986         return xdst;
2987
2988 inc_error:
2989         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2990 error:
2991         xfrm_pols_put(pols, num_pols);
2992         return ERR_PTR(err);
2993 }
2994
2995 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2996                                         struct dst_entry *dst_orig)
2997 {
2998         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2999         struct dst_entry *ret;
3000
3001         if (!afinfo) {
3002                 dst_release(dst_orig);
3003                 return ERR_PTR(-EINVAL);
3004         } else {
3005                 ret = afinfo->blackhole_route(net, dst_orig);
3006         }
3007         rcu_read_unlock();
3008
3009         return ret;
3010 }
3011
3012 /* Finds/creates a bundle for given flow and if_id
3013  *
3014  * At the moment we eat a raw IP route. Mostly to speed up lookups
3015  * on interfaces with disabled IPsec.
3016  *
3017  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3018  * compatibility
3019  */
3020 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3021                                         struct dst_entry *dst_orig,
3022                                         const struct flowi *fl,
3023                                         const struct sock *sk,
3024                                         int flags, u32 if_id)
3025 {
3026         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3027         struct xfrm_dst *xdst;
3028         struct dst_entry *dst, *route;
3029         u16 family = dst_orig->ops->family;
3030         u8 dir = XFRM_POLICY_OUT;
3031         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3032
3033         dst = NULL;
3034         xdst = NULL;
3035         route = NULL;
3036
3037         sk = sk_const_to_full_sk(sk);
3038         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3039                 num_pols = 1;
3040                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3041                                                 if_id);
3042                 err = xfrm_expand_policies(fl, family, pols,
3043                                            &num_pols, &num_xfrms);
3044                 if (err < 0)
3045                         goto dropdst;
3046
3047                 if (num_pols) {
3048                         if (num_xfrms <= 0) {
3049                                 drop_pols = num_pols;
3050                                 goto no_transform;
3051                         }
3052
3053                         xdst = xfrm_resolve_and_create_bundle(
3054                                         pols, num_pols, fl,
3055                                         family, dst_orig);
3056
3057                         if (IS_ERR(xdst)) {
3058                                 xfrm_pols_put(pols, num_pols);
3059                                 err = PTR_ERR(xdst);
3060                                 if (err == -EREMOTE)
3061                                         goto nopol;
3062
3063                                 goto dropdst;
3064                         } else if (xdst == NULL) {
3065                                 num_xfrms = 0;
3066                                 drop_pols = num_pols;
3067                                 goto no_transform;
3068                         }
3069
3070                         route = xdst->route;
3071                 }
3072         }
3073
3074         if (xdst == NULL) {
3075                 struct xfrm_flo xflo;
3076
3077                 xflo.dst_orig = dst_orig;
3078                 xflo.flags = flags;
3079
3080                 /* To accelerate a bit...  */
3081                 if ((dst_orig->flags & DST_NOXFRM) ||
3082                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
3083                         goto nopol;
3084
3085                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3086                 if (xdst == NULL)
3087                         goto nopol;
3088                 if (IS_ERR(xdst)) {
3089                         err = PTR_ERR(xdst);
3090                         goto dropdst;
3091                 }
3092
3093                 num_pols = xdst->num_pols;
3094                 num_xfrms = xdst->num_xfrms;
3095                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3096                 route = xdst->route;
3097         }
3098
3099         dst = &xdst->u.dst;
3100         if (route == NULL && num_xfrms > 0) {
3101                 /* The only case when xfrm_bundle_lookup() returns a
3102                  * bundle with null route, is when the template could
3103                  * not be resolved. It means policies are there, but
3104                  * bundle could not be created, since we don't yet
3105                  * have the xfrm_state's. We need to wait for KM to
3106                  * negotiate new SA's or bail out with error.*/
3107                 if (net->xfrm.sysctl_larval_drop) {
3108                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3109                         err = -EREMOTE;
3110                         goto error;
3111                 }
3112
3113                 err = -EAGAIN;
3114
3115                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3116                 goto error;
3117         }
3118
3119 no_transform:
3120         if (num_pols == 0)
3121                 goto nopol;
3122
3123         if ((flags & XFRM_LOOKUP_ICMP) &&
3124             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3125                 err = -ENOENT;
3126                 goto error;
3127         }
3128
3129         for (i = 0; i < num_pols; i++)
3130                 pols[i]->curlft.use_time = ktime_get_real_seconds();
3131
3132         if (num_xfrms < 0) {
3133                 /* Prohibit the flow */
3134                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3135                 err = -EPERM;
3136                 goto error;
3137         } else if (num_xfrms > 0) {
3138                 /* Flow transformed */
3139                 dst_release(dst_orig);
3140         } else {
3141                 /* Flow passes untransformed */
3142                 dst_release(dst);
3143                 dst = dst_orig;
3144         }
3145 ok:
3146         xfrm_pols_put(pols, drop_pols);
3147         if (dst && dst->xfrm &&
3148             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3149                 dst->flags |= DST_XFRM_TUNNEL;
3150         return dst;
3151
3152 nopol:
3153         if (!(flags & XFRM_LOOKUP_ICMP)) {
3154                 dst = dst_orig;
3155                 goto ok;
3156         }
3157         err = -ENOENT;
3158 error:
3159         dst_release(dst);
3160 dropdst:
3161         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3162                 dst_release(dst_orig);
3163         xfrm_pols_put(pols, drop_pols);
3164         return ERR_PTR(err);
3165 }
3166 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3167
3168 /* Main function: finds/creates a bundle for given flow.
3169  *
3170  * At the moment we eat a raw IP route. Mostly to speed up lookups
3171  * on interfaces with disabled IPsec.
3172  */
3173 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3174                               const struct flowi *fl, const struct sock *sk,
3175                               int flags)
3176 {
3177         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3178 }
3179 EXPORT_SYMBOL(xfrm_lookup);
3180
3181 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3182  * Otherwise we may send out blackholed packets.
3183  */
3184 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3185                                     const struct flowi *fl,
3186                                     const struct sock *sk, int flags)
3187 {
3188         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3189                                             flags | XFRM_LOOKUP_QUEUE |
3190                                             XFRM_LOOKUP_KEEP_DST_REF);
3191
3192         if (PTR_ERR(dst) == -EREMOTE)
3193                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3194
3195         if (IS_ERR(dst))
3196                 dst_release(dst_orig);
3197
3198         return dst;
3199 }
3200 EXPORT_SYMBOL(xfrm_lookup_route);
3201
3202 static inline int
3203 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3204 {
3205         struct sec_path *sp = skb_sec_path(skb);
3206         struct xfrm_state *x;
3207
3208         if (!sp || idx < 0 || idx >= sp->len)
3209                 return 0;
3210         x = sp->xvec[idx];
3211         if (!x->type->reject)
3212                 return 0;
3213         return x->type->reject(x, skb, fl);
3214 }
3215
3216 /* When skb is transformed back to its "native" form, we have to
3217  * check policy restrictions. At the moment we make this in maximally
3218  * stupid way. Shame on me. :-) Of course, connected sockets must
3219  * have policy cached at them.
3220  */
3221
3222 static inline int
3223 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3224               unsigned short family)
3225 {
3226         if (xfrm_state_kern(x))
3227                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3228         return  x->id.proto == tmpl->id.proto &&
3229                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3230                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3231                 x->props.mode == tmpl->mode &&
3232                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3233                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3234                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3235                   xfrm_state_addr_cmp(tmpl, x, family));
3236 }
3237
3238 /*
3239  * 0 or more than 0 is returned when validation is succeeded (either bypass
3240  * because of optional transport mode, or next index of the mathced secpath
3241  * state with the template.
3242  * -1 is returned when no matching template is found.
3243  * Otherwise "-2 - errored_index" is returned.
3244  */
3245 static inline int
3246 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3247                unsigned short family)
3248 {
3249         int idx = start;
3250
3251         if (tmpl->optional) {
3252                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3253                         return start;
3254         } else
3255                 start = -1;
3256         for (; idx < sp->len; idx++) {
3257                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3258                         return ++idx;
3259                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3260                         if (start == -1)
3261                                 start = -2-idx;
3262                         break;
3263                 }
3264         }
3265         return start;
3266 }
3267
3268 static void
3269 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3270 {
3271         const struct iphdr *iph = ip_hdr(skb);
3272         int ihl = iph->ihl;
3273         u8 *xprth = skb_network_header(skb) + ihl * 4;
3274         struct flowi4 *fl4 = &fl->u.ip4;
3275         int oif = 0;
3276
3277         if (skb_dst(skb) && skb_dst(skb)->dev)
3278                 oif = skb_dst(skb)->dev->ifindex;
3279
3280         memset(fl4, 0, sizeof(struct flowi4));
3281         fl4->flowi4_mark = skb->mark;
3282         fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3283
3284         fl4->flowi4_proto = iph->protocol;
3285         fl4->daddr = reverse ? iph->saddr : iph->daddr;
3286         fl4->saddr = reverse ? iph->daddr : iph->saddr;
3287         fl4->flowi4_tos = iph->tos;
3288
3289         if (!ip_is_fragment(iph)) {
3290                 switch (iph->protocol) {
3291                 case IPPROTO_UDP:
3292                 case IPPROTO_UDPLITE:
3293                 case IPPROTO_TCP:
3294                 case IPPROTO_SCTP:
3295                 case IPPROTO_DCCP:
3296                         if (xprth + 4 < skb->data ||
3297                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3298                                 __be16 *ports;
3299
3300                                 xprth = skb_network_header(skb) + ihl * 4;
3301                                 ports = (__be16 *)xprth;
3302
3303                                 fl4->fl4_sport = ports[!!reverse];
3304                                 fl4->fl4_dport = ports[!reverse];
3305                         }
3306                         break;
3307                 case IPPROTO_ICMP:
3308                         if (xprth + 2 < skb->data ||
3309                             pskb_may_pull(skb, xprth + 2 - skb->data)) {
3310                                 u8 *icmp;
3311
3312                                 xprth = skb_network_header(skb) + ihl * 4;
3313                                 icmp = xprth;
3314
3315                                 fl4->fl4_icmp_type = icmp[0];
3316                                 fl4->fl4_icmp_code = icmp[1];
3317                         }
3318                         break;
3319                 case IPPROTO_ESP:
3320                         if (xprth + 4 < skb->data ||
3321                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3322                                 __be32 *ehdr;
3323
3324                                 xprth = skb_network_header(skb) + ihl * 4;
3325                                 ehdr = (__be32 *)xprth;
3326
3327                                 fl4->fl4_ipsec_spi = ehdr[0];
3328                         }
3329                         break;
3330                 case IPPROTO_AH:
3331                         if (xprth + 8 < skb->data ||
3332                             pskb_may_pull(skb, xprth + 8 - skb->data)) {
3333                                 __be32 *ah_hdr;
3334
3335                                 xprth = skb_network_header(skb) + ihl * 4;
3336                                 ah_hdr = (__be32 *)xprth;
3337
3338                                 fl4->fl4_ipsec_spi = ah_hdr[1];
3339                         }
3340                         break;
3341                 case IPPROTO_COMP:
3342                         if (xprth + 4 < skb->data ||
3343                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3344                                 __be16 *ipcomp_hdr;
3345
3346                                 xprth = skb_network_header(skb) + ihl * 4;
3347                                 ipcomp_hdr = (__be16 *)xprth;
3348
3349                                 fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
3350                         }
3351                         break;
3352                 case IPPROTO_GRE:
3353                         if (xprth + 12 < skb->data ||
3354                             pskb_may_pull(skb, xprth + 12 - skb->data)) {
3355                                 __be16 *greflags;
3356                                 __be32 *gre_hdr;
3357
3358                                 xprth = skb_network_header(skb) + ihl * 4;
3359                                 greflags = (__be16 *)xprth;
3360                                 gre_hdr = (__be32 *)xprth;
3361
3362                                 if (greflags[0] & GRE_KEY) {
3363                                         if (greflags[0] & GRE_CSUM)
3364                                                 gre_hdr++;
3365                                         fl4->fl4_gre_key = gre_hdr[1];
3366                                 }
3367                         }
3368                         break;
3369                 default:
3370                         fl4->fl4_ipsec_spi = 0;
3371                         break;
3372                 }
3373         }
3374 }
3375
3376 #if IS_ENABLED(CONFIG_IPV6)
3377 static void
3378 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3379 {
3380         struct flowi6 *fl6 = &fl->u.ip6;
3381         int onlyproto = 0;
3382         const struct ipv6hdr *hdr = ipv6_hdr(skb);
3383         u32 offset = sizeof(*hdr);
3384         struct ipv6_opt_hdr *exthdr;
3385         const unsigned char *nh = skb_network_header(skb);
3386         u16 nhoff = IP6CB(skb)->nhoff;
3387         int oif = 0;
3388         u8 nexthdr;
3389
3390         if (!nhoff)
3391                 nhoff = offsetof(struct ipv6hdr, nexthdr);
3392
3393         nexthdr = nh[nhoff];
3394
3395         if (skb_dst(skb) && skb_dst(skb)->dev)
3396                 oif = skb_dst(skb)->dev->ifindex;
3397
3398         memset(fl6, 0, sizeof(struct flowi6));
3399         fl6->flowi6_mark = skb->mark;
3400         fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3401
3402         fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3403         fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3404
3405         while (nh + offset + sizeof(*exthdr) < skb->data ||
3406                pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3407                 nh = skb_network_header(skb);
3408                 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3409
3410                 switch (nexthdr) {
3411                 case NEXTHDR_FRAGMENT:
3412                         onlyproto = 1;
3413                         /* fall through */
3414                 case NEXTHDR_ROUTING:
3415                 case NEXTHDR_HOP:
3416                 case NEXTHDR_DEST:
3417                         offset += ipv6_optlen(exthdr);
3418                         nexthdr = exthdr->nexthdr;
3419                         exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3420                         break;
3421                 case IPPROTO_UDP:
3422                 case IPPROTO_UDPLITE:
3423                 case IPPROTO_TCP:
3424                 case IPPROTO_SCTP:
3425                 case IPPROTO_DCCP:
3426                         if (!onlyproto && (nh + offset + 4 < skb->data ||
3427                              pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3428                                 __be16 *ports;
3429
3430                                 nh = skb_network_header(skb);
3431                                 ports = (__be16 *)(nh + offset);
3432                                 fl6->fl6_sport = ports[!!reverse];
3433                                 fl6->fl6_dport = ports[!reverse];
3434                         }
3435                         fl6->flowi6_proto = nexthdr;
3436                         return;
3437                 case IPPROTO_ICMPV6:
3438                         if (!onlyproto && (nh + offset + 2 < skb->data ||
3439                             pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3440                                 u8 *icmp;
3441
3442                                 nh = skb_network_header(skb);
3443                                 icmp = (u8 *)(nh + offset);
3444                                 fl6->fl6_icmp_type = icmp[0];
3445                                 fl6->fl6_icmp_code = icmp[1];
3446                         }
3447                         fl6->flowi6_proto = nexthdr;
3448                         return;
3449 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3450                 case IPPROTO_MH:
3451                         offset += ipv6_optlen(exthdr);
3452                         if (!onlyproto && (nh + offset + 3 < skb->data ||
3453                             pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3454                                 struct ip6_mh *mh;
3455
3456                                 nh = skb_network_header(skb);
3457                                 mh = (struct ip6_mh *)(nh + offset);
3458                                 fl6->fl6_mh_type = mh->ip6mh_type;
3459                         }
3460                         fl6->flowi6_proto = nexthdr;
3461                         return;
3462 #endif
3463                 /* XXX Why are there these headers? */
3464                 case IPPROTO_AH:
3465                 case IPPROTO_ESP:
3466                 case IPPROTO_COMP:
3467                 default:
3468                         fl6->fl6_ipsec_spi = 0;
3469                         fl6->flowi6_proto = nexthdr;
3470                         return;
3471                 }
3472         }
3473 }
3474 #endif
3475
3476 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3477                           unsigned int family, int reverse)
3478 {
3479         switch (family) {
3480         case AF_INET:
3481                 decode_session4(skb, fl, reverse);
3482                 break;
3483 #if IS_ENABLED(CONFIG_IPV6)
3484         case AF_INET6:
3485                 decode_session6(skb, fl, reverse);
3486                 break;
3487 #endif
3488         default:
3489                 return -EAFNOSUPPORT;
3490         }
3491
3492         return security_xfrm_decode_session(skb, &fl->flowi_secid);
3493 }
3494 EXPORT_SYMBOL(__xfrm_decode_session);
3495
3496 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3497 {
3498         for (; k < sp->len; k++) {
3499                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3500                         *idxp = k;
3501                         return 1;
3502                 }
3503         }
3504
3505         return 0;
3506 }
3507
3508 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3509                         unsigned short family)
3510 {
3511         struct net *net = dev_net(skb->dev);
3512         struct xfrm_policy *pol;
3513         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3514         int npols = 0;
3515         int xfrm_nr;
3516         int pi;
3517         int reverse;
3518         struct flowi fl;
3519         int xerr_idx = -1;
3520         const struct xfrm_if_cb *ifcb;
3521         struct sec_path *sp;
3522         struct xfrm_if *xi;
3523         u32 if_id = 0;
3524
3525         rcu_read_lock();
3526         ifcb = xfrm_if_get_cb();
3527
3528         if (ifcb) {
3529                 xi = ifcb->decode_session(skb, family);
3530                 if (xi) {
3531                         if_id = xi->p.if_id;
3532                         net = xi->net;
3533                 }
3534         }
3535         rcu_read_unlock();
3536
3537         reverse = dir & ~XFRM_POLICY_MASK;
3538         dir &= XFRM_POLICY_MASK;
3539
3540         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3541                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3542                 return 0;
3543         }
3544
3545         nf_nat_decode_session(skb, &fl, family);
3546
3547         /* First, check used SA against their selectors. */
3548         sp = skb_sec_path(skb);
3549         if (sp) {
3550                 int i;
3551
3552                 for (i = sp->len - 1; i >= 0; i--) {
3553                         struct xfrm_state *x = sp->xvec[i];
3554                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3555                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3556                                 return 0;
3557                         }
3558                 }
3559         }
3560
3561         pol = NULL;
3562         sk = sk_to_full_sk(sk);
3563         if (sk && sk->sk_policy[dir]) {
3564                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3565                 if (IS_ERR(pol)) {
3566                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3567                         return 0;
3568                 }
3569         }
3570
3571         if (!pol)
3572                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3573
3574         if (IS_ERR(pol)) {
3575                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3576                 return 0;
3577         }
3578
3579         if (!pol) {
3580                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3581                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3582                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3583                         return 0;
3584                 }
3585                 return 1;
3586         }
3587
3588         pol->curlft.use_time = ktime_get_real_seconds();
3589
3590         pols[0] = pol;
3591         npols++;
3592 #ifdef CONFIG_XFRM_SUB_POLICY
3593         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3594                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3595                                                     &fl, family,
3596                                                     XFRM_POLICY_IN, if_id);
3597                 if (pols[1]) {
3598                         if (IS_ERR(pols[1])) {
3599                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3600                                 return 0;
3601                         }
3602                         pols[1]->curlft.use_time = ktime_get_real_seconds();
3603                         npols++;
3604                 }
3605         }
3606 #endif
3607
3608         if (pol->action == XFRM_POLICY_ALLOW) {
3609                 static struct sec_path dummy;
3610                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3611                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3612                 struct xfrm_tmpl **tpp = tp;
3613                 int ti = 0;
3614                 int i, k;
3615
3616                 sp = skb_sec_path(skb);
3617                 if (!sp)
3618                         sp = &dummy;
3619
3620                 for (pi = 0; pi < npols; pi++) {
3621                         if (pols[pi] != pol &&
3622                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3623                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3624                                 goto reject;
3625                         }
3626                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3627                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3628                                 goto reject_error;
3629                         }
3630                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3631                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3632                 }
3633                 xfrm_nr = ti;
3634                 if (npols > 1) {
3635                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3636                         tpp = stp;
3637                 }
3638
3639                 /* For each tunnel xfrm, find the first matching tmpl.
3640                  * For each tmpl before that, find corresponding xfrm.
3641                  * Order is _important_. Later we will implement
3642                  * some barriers, but at the moment barriers
3643                  * are implied between each two transformations.
3644                  */
3645                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3646                         k = xfrm_policy_ok(tpp[i], sp, k, family);
3647                         if (k < 0) {
3648                                 if (k < -1)
3649                                         /* "-2 - errored_index" returned */
3650                                         xerr_idx = -(2+k);
3651                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3652                                 goto reject;
3653                         }
3654                 }
3655
3656                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3657                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3658                         goto reject;
3659                 }
3660
3661                 xfrm_pols_put(pols, npols);
3662                 return 1;
3663         }
3664         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3665
3666 reject:
3667         xfrm_secpath_reject(xerr_idx, skb, &fl);
3668 reject_error:
3669         xfrm_pols_put(pols, npols);
3670         return 0;
3671 }
3672 EXPORT_SYMBOL(__xfrm_policy_check);
3673
3674 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3675 {
3676         struct net *net = dev_net(skb->dev);
3677         struct flowi fl;
3678         struct dst_entry *dst;
3679         int res = 1;
3680
3681         if (xfrm_decode_session(skb, &fl, family) < 0) {
3682                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3683                 return 0;
3684         }
3685
3686         skb_dst_force(skb);
3687         if (!skb_dst(skb)) {
3688                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3689                 return 0;
3690         }
3691
3692         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3693         if (IS_ERR(dst)) {
3694                 res = 0;
3695                 dst = NULL;
3696         }
3697         skb_dst_set(skb, dst);
3698         return res;
3699 }
3700 EXPORT_SYMBOL(__xfrm_route_forward);
3701
3702 /* Optimize later using cookies and generation ids. */
3703
3704 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3705 {
3706         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3707          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3708          * get validated by dst_ops->check on every use.  We do this
3709          * because when a normal route referenced by an XFRM dst is
3710          * obsoleted we do not go looking around for all parent
3711          * referencing XFRM dsts so that we can invalidate them.  It
3712          * is just too much work.  Instead we make the checks here on
3713          * every use.  For example:
3714          *
3715          *      XFRM dst A --> IPv4 dst X
3716          *
3717          * X is the "xdst->route" of A (X is also the "dst->path" of A
3718          * in this example).  If X is marked obsolete, "A" will not
3719          * notice.  That's what we are validating here via the
3720          * stale_bundle() check.
3721          *
3722          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3723          * be marked on it.
3724          * This will force stale_bundle() to fail on any xdst bundle with
3725          * this dst linked in it.
3726          */
3727         if (dst->obsolete < 0 && !stale_bundle(dst))
3728                 return dst;
3729
3730         return NULL;
3731 }
3732
3733 static int stale_bundle(struct dst_entry *dst)
3734 {
3735         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3736 }
3737
3738 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3739 {
3740         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3741                 dst->dev = dev_net(dev)->loopback_dev;
3742                 dev_hold(dst->dev);
3743                 dev_put(dev);
3744         }
3745 }
3746 EXPORT_SYMBOL(xfrm_dst_ifdown);
3747
3748 static void xfrm_link_failure(struct sk_buff *skb)
3749 {
3750         /* Impossible. Such dst must be popped before reaches point of failure. */
3751 }
3752
3753 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3754 {
3755         if (dst) {
3756                 if (dst->obsolete) {
3757                         dst_release(dst);
3758                         dst = NULL;
3759                 }
3760         }
3761         return dst;
3762 }
3763
3764 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3765 {
3766         while (nr--) {
3767                 struct xfrm_dst *xdst = bundle[nr];
3768                 u32 pmtu, route_mtu_cached;
3769                 struct dst_entry *dst;
3770
3771                 dst = &xdst->u.dst;
3772                 pmtu = dst_mtu(xfrm_dst_child(dst));
3773                 xdst->child_mtu_cached = pmtu;
3774
3775                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3776
3777                 route_mtu_cached = dst_mtu(xdst->route);
3778                 xdst->route_mtu_cached = route_mtu_cached;
3779
3780                 if (pmtu > route_mtu_cached)
3781                         pmtu = route_mtu_cached;
3782
3783                 dst_metric_set(dst, RTAX_MTU, pmtu);
3784         }
3785 }
3786
3787 /* Check that the bundle accepts the flow and its components are
3788  * still valid.
3789  */
3790
3791 static int xfrm_bundle_ok(struct xfrm_dst *first)
3792 {
3793         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3794         struct dst_entry *dst = &first->u.dst;
3795         struct xfrm_dst *xdst;
3796         int start_from, nr;
3797         u32 mtu;
3798
3799         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3800             (dst->dev && !netif_running(dst->dev)))
3801                 return 0;
3802
3803         if (dst->flags & DST_XFRM_QUEUE)
3804                 return 1;
3805
3806         start_from = nr = 0;
3807         do {
3808                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3809
3810                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3811                         return 0;
3812                 if (xdst->xfrm_genid != dst->xfrm->genid)
3813                         return 0;
3814                 if (xdst->num_pols > 0 &&
3815                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3816                         return 0;
3817
3818                 bundle[nr++] = xdst;
3819
3820                 mtu = dst_mtu(xfrm_dst_child(dst));
3821                 if (xdst->child_mtu_cached != mtu) {
3822                         start_from = nr;
3823                         xdst->child_mtu_cached = mtu;
3824                 }
3825
3826                 if (!dst_check(xdst->route, xdst->route_cookie))
3827                         return 0;
3828                 mtu = dst_mtu(xdst->route);
3829                 if (xdst->route_mtu_cached != mtu) {
3830                         start_from = nr;
3831                         xdst->route_mtu_cached = mtu;
3832                 }
3833
3834                 dst = xfrm_dst_child(dst);
3835         } while (dst->xfrm);
3836
3837         if (likely(!start_from))
3838                 return 1;
3839
3840         xdst = bundle[start_from - 1];
3841         mtu = xdst->child_mtu_cached;
3842         while (start_from--) {
3843                 dst = &xdst->u.dst;
3844
3845                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3846                 if (mtu > xdst->route_mtu_cached)
3847                         mtu = xdst->route_mtu_cached;
3848                 dst_metric_set(dst, RTAX_MTU, mtu);
3849                 if (!start_from)
3850                         break;
3851
3852                 xdst = bundle[start_from - 1];
3853                 xdst->child_mtu_cached = mtu;
3854         }
3855
3856         return 1;
3857 }
3858
3859 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3860 {
3861         return dst_metric_advmss(xfrm_dst_path(dst));
3862 }
3863
3864 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3865 {
3866         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3867
3868         return mtu ? : dst_mtu(xfrm_dst_path(dst));
3869 }
3870
3871 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3872                                         const void *daddr)
3873 {
3874         while (dst->xfrm) {
3875                 const struct xfrm_state *xfrm = dst->xfrm;
3876
3877                 dst = xfrm_dst_child(dst);
3878
3879                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3880                         continue;
3881                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3882                         daddr = xfrm->coaddr;
3883                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3884                         daddr = &xfrm->id.daddr;
3885         }
3886         return daddr;
3887 }
3888
3889 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3890                                            struct sk_buff *skb,
3891                                            const void *daddr)
3892 {
3893         const struct dst_entry *path = xfrm_dst_path(dst);
3894
3895         if (!skb)
3896                 daddr = xfrm_get_dst_nexthop(dst, daddr);
3897         return path->ops->neigh_lookup(path, skb, daddr);
3898 }
3899
3900 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3901 {
3902         const struct dst_entry *path = xfrm_dst_path(dst);
3903
3904         daddr = xfrm_get_dst_nexthop(dst, daddr);
3905         path->ops->confirm_neigh(path, daddr);
3906 }
3907
3908 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3909 {
3910         int err = 0;
3911
3912         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3913                 return -EAFNOSUPPORT;
3914
3915         spin_lock(&xfrm_policy_afinfo_lock);
3916         if (unlikely(xfrm_policy_afinfo[family] != NULL))
3917                 err = -EEXIST;
3918         else {
3919                 struct dst_ops *dst_ops = afinfo->dst_ops;
3920                 if (likely(dst_ops->kmem_cachep == NULL))
3921                         dst_ops->kmem_cachep = xfrm_dst_cache;
3922                 if (likely(dst_ops->check == NULL))
3923                         dst_ops->check = xfrm_dst_check;
3924                 if (likely(dst_ops->default_advmss == NULL))
3925                         dst_ops->default_advmss = xfrm_default_advmss;
3926                 if (likely(dst_ops->mtu == NULL))
3927                         dst_ops->mtu = xfrm_mtu;
3928                 if (likely(dst_ops->negative_advice == NULL))
3929                         dst_ops->negative_advice = xfrm_negative_advice;
3930                 if (likely(dst_ops->link_failure == NULL))
3931                         dst_ops->link_failure = xfrm_link_failure;
3932                 if (likely(dst_ops->neigh_lookup == NULL))
3933                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
3934                 if (likely(!dst_ops->confirm_neigh))
3935                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
3936                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3937         }
3938         spin_unlock(&xfrm_policy_afinfo_lock);
3939
3940         return err;
3941 }
3942 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3943
3944 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3945 {
3946         struct dst_ops *dst_ops = afinfo->dst_ops;
3947         int i;
3948
3949         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3950                 if (xfrm_policy_afinfo[i] != afinfo)
3951                         continue;
3952                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3953                 break;
3954         }
3955
3956         synchronize_rcu();
3957
3958         dst_ops->kmem_cachep = NULL;
3959         dst_ops->check = NULL;
3960         dst_ops->negative_advice = NULL;
3961         dst_ops->link_failure = NULL;
3962 }
3963 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3964
3965 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3966 {
3967         spin_lock(&xfrm_if_cb_lock);
3968         rcu_assign_pointer(xfrm_if_cb, ifcb);
3969         spin_unlock(&xfrm_if_cb_lock);
3970 }
3971 EXPORT_SYMBOL(xfrm_if_register_cb);
3972
3973 void xfrm_if_unregister_cb(void)
3974 {
3975         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3976         synchronize_rcu();
3977 }
3978 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3979
3980 #ifdef CONFIG_XFRM_STATISTICS
3981 static int __net_init xfrm_statistics_init(struct net *net)
3982 {
3983         int rv;
3984         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3985         if (!net->mib.xfrm_statistics)
3986                 return -ENOMEM;
3987         rv = xfrm_proc_init(net);
3988         if (rv < 0)
3989                 free_percpu(net->mib.xfrm_statistics);
3990         return rv;
3991 }
3992
3993 static void xfrm_statistics_fini(struct net *net)
3994 {
3995         xfrm_proc_fini(net);
3996         free_percpu(net->mib.xfrm_statistics);
3997 }
3998 #else
3999 static int __net_init xfrm_statistics_init(struct net *net)
4000 {
4001         return 0;
4002 }
4003
4004 static void xfrm_statistics_fini(struct net *net)
4005 {
4006 }
4007 #endif
4008
4009 static int __net_init xfrm_policy_init(struct net *net)
4010 {
4011         unsigned int hmask, sz;
4012         int dir, err;
4013
4014         if (net_eq(net, &init_net)) {
4015                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4016                                            sizeof(struct xfrm_dst),
4017                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4018                                            NULL);
4019                 err = rhashtable_init(&xfrm_policy_inexact_table,
4020                                       &xfrm_pol_inexact_params);
4021                 BUG_ON(err);
4022         }
4023
4024         hmask = 8 - 1;
4025         sz = (hmask+1) * sizeof(struct hlist_head);
4026
4027         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4028         if (!net->xfrm.policy_byidx)
4029                 goto out_byidx;
4030         net->xfrm.policy_idx_hmask = hmask;
4031
4032         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4033                 struct xfrm_policy_hash *htab;
4034
4035                 net->xfrm.policy_count[dir] = 0;
4036                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4037                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4038
4039                 htab = &net->xfrm.policy_bydst[dir];
4040                 htab->table = xfrm_hash_alloc(sz);
4041                 if (!htab->table)
4042                         goto out_bydst;
4043                 htab->hmask = hmask;
4044                 htab->dbits4 = 32;
4045                 htab->sbits4 = 32;
4046                 htab->dbits6 = 128;
4047                 htab->sbits6 = 128;
4048         }
4049         net->xfrm.policy_hthresh.lbits4 = 32;
4050         net->xfrm.policy_hthresh.rbits4 = 32;
4051         net->xfrm.policy_hthresh.lbits6 = 128;
4052         net->xfrm.policy_hthresh.rbits6 = 128;
4053
4054         seqlock_init(&net->xfrm.policy_hthresh.lock);
4055
4056         INIT_LIST_HEAD(&net->xfrm.policy_all);
4057         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4058         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4059         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4060         return 0;
4061
4062 out_bydst:
4063         for (dir--; dir >= 0; dir--) {
4064                 struct xfrm_policy_hash *htab;
4065
4066                 htab = &net->xfrm.policy_bydst[dir];
4067                 xfrm_hash_free(htab->table, sz);
4068         }
4069         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4070 out_byidx:
4071         return -ENOMEM;
4072 }
4073
4074 static void xfrm_policy_fini(struct net *net)
4075 {
4076         struct xfrm_pol_inexact_bin *b, *t;
4077         unsigned int sz;
4078         int dir;
4079
4080         flush_work(&net->xfrm.policy_hash_work);
4081 #ifdef CONFIG_XFRM_SUB_POLICY
4082         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4083 #endif
4084         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4085
4086         WARN_ON(!list_empty(&net->xfrm.policy_all));
4087
4088         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4089                 struct xfrm_policy_hash *htab;
4090
4091                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4092
4093                 htab = &net->xfrm.policy_bydst[dir];
4094                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4095                 WARN_ON(!hlist_empty(htab->table));
4096                 xfrm_hash_free(htab->table, sz);
4097         }
4098
4099         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4100         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4101         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4102
4103         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4104         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4105                 __xfrm_policy_inexact_prune_bin(b, true);
4106         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4107 }
4108
4109 static int __net_init xfrm_net_init(struct net *net)
4110 {
4111         int rv;
4112
4113         /* Initialize the per-net locks here */
4114         spin_lock_init(&net->xfrm.xfrm_state_lock);
4115         spin_lock_init(&net->xfrm.xfrm_policy_lock);
4116         mutex_init(&net->xfrm.xfrm_cfg_mutex);
4117
4118         rv = xfrm_statistics_init(net);
4119         if (rv < 0)
4120                 goto out_statistics;
4121         rv = xfrm_state_init(net);
4122         if (rv < 0)
4123                 goto out_state;
4124         rv = xfrm_policy_init(net);
4125         if (rv < 0)
4126                 goto out_policy;
4127         rv = xfrm_sysctl_init(net);
4128         if (rv < 0)
4129                 goto out_sysctl;
4130
4131         return 0;
4132
4133 out_sysctl:
4134         xfrm_policy_fini(net);
4135 out_policy:
4136         xfrm_state_fini(net);
4137 out_state:
4138         xfrm_statistics_fini(net);
4139 out_statistics:
4140         return rv;
4141 }
4142
4143 static void __net_exit xfrm_net_exit(struct net *net)
4144 {
4145         xfrm_sysctl_fini(net);
4146         xfrm_policy_fini(net);
4147         xfrm_state_fini(net);
4148         xfrm_statistics_fini(net);
4149 }
4150
4151 static struct pernet_operations __net_initdata xfrm_net_ops = {
4152         .init = xfrm_net_init,
4153         .exit = xfrm_net_exit,
4154 };
4155
4156 void __init xfrm_init(void)
4157 {
4158         register_pernet_subsys(&xfrm_net_ops);
4159         xfrm_dev_init();
4160         seqcount_init(&xfrm_policy_hash_generation);
4161         xfrm_input_init();
4162
4163 #ifdef CONFIG_XFRM_ESPINTCP
4164         espintcp_init();
4165 #endif
4166
4167         RCU_INIT_POINTER(xfrm_if_cb, NULL);
4168         synchronize_rcu();
4169 }
4170
4171 #ifdef CONFIG_AUDITSYSCALL
4172 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4173                                          struct audit_buffer *audit_buf)
4174 {
4175         struct xfrm_sec_ctx *ctx = xp->security;
4176         struct xfrm_selector *sel = &xp->selector;
4177
4178         if (ctx)
4179                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4180                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4181
4182         switch (sel->family) {
4183         case AF_INET:
4184                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4185                 if (sel->prefixlen_s != 32)
4186                         audit_log_format(audit_buf, " src_prefixlen=%d",
4187                                          sel->prefixlen_s);
4188                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4189                 if (sel->prefixlen_d != 32)
4190                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4191                                          sel->prefixlen_d);
4192                 break;
4193         case AF_INET6:
4194                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4195                 if (sel->prefixlen_s != 128)
4196                         audit_log_format(audit_buf, " src_prefixlen=%d",
4197                                          sel->prefixlen_s);
4198                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4199                 if (sel->prefixlen_d != 128)
4200                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4201                                          sel->prefixlen_d);
4202                 break;
4203         }
4204 }
4205
4206 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4207 {
4208         struct audit_buffer *audit_buf;
4209
4210         audit_buf = xfrm_audit_start("SPD-add");
4211         if (audit_buf == NULL)
4212                 return;
4213         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4214         audit_log_format(audit_buf, " res=%u", result);
4215         xfrm_audit_common_policyinfo(xp, audit_buf);
4216         audit_log_end(audit_buf);
4217 }
4218 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4219
4220 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4221                               bool task_valid)
4222 {
4223         struct audit_buffer *audit_buf;
4224
4225         audit_buf = xfrm_audit_start("SPD-delete");
4226         if (audit_buf == NULL)
4227                 return;
4228         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4229         audit_log_format(audit_buf, " res=%u", result);
4230         xfrm_audit_common_policyinfo(xp, audit_buf);
4231         audit_log_end(audit_buf);
4232 }
4233 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4234 #endif
4235
4236 #ifdef CONFIG_XFRM_MIGRATE
4237 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4238                                         const struct xfrm_selector *sel_tgt)
4239 {
4240         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4241                 if (sel_tgt->family == sel_cmp->family &&
4242                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4243                                     sel_cmp->family) &&
4244                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4245                                     sel_cmp->family) &&
4246                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4247                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4248                         return true;
4249                 }
4250         } else {
4251                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4252                         return true;
4253                 }
4254         }
4255         return false;
4256 }
4257
4258 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4259                                                     u8 dir, u8 type, struct net *net)
4260 {
4261         struct xfrm_policy *pol, *ret = NULL;
4262         struct hlist_head *chain;
4263         u32 priority = ~0U;
4264
4265         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4266         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4267         hlist_for_each_entry(pol, chain, bydst) {
4268                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4269                     pol->type == type) {
4270                         ret = pol;
4271                         priority = ret->priority;
4272                         break;
4273                 }
4274         }
4275         chain = &net->xfrm.policy_inexact[dir];
4276         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4277                 if ((pol->priority >= priority) && ret)
4278                         break;
4279
4280                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4281                     pol->type == type) {
4282                         ret = pol;
4283                         break;
4284                 }
4285         }
4286
4287         xfrm_pol_hold(ret);
4288
4289         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4290
4291         return ret;
4292 }
4293
4294 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4295 {
4296         int match = 0;
4297
4298         if (t->mode == m->mode && t->id.proto == m->proto &&
4299             (m->reqid == 0 || t->reqid == m->reqid)) {
4300                 switch (t->mode) {
4301                 case XFRM_MODE_TUNNEL:
4302                 case XFRM_MODE_BEET:
4303                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4304                                             m->old_family) &&
4305                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4306                                             m->old_family)) {
4307                                 match = 1;
4308                         }
4309                         break;
4310                 case XFRM_MODE_TRANSPORT:
4311                         /* in case of transport mode, template does not store
4312                            any IP addresses, hence we just compare mode and
4313                            protocol */
4314                         match = 1;
4315                         break;
4316                 default:
4317                         break;
4318                 }
4319         }
4320         return match;
4321 }
4322
4323 /* update endpoint address(es) of template(s) */
4324 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4325                                struct xfrm_migrate *m, int num_migrate)
4326 {
4327         struct xfrm_migrate *mp;
4328         int i, j, n = 0;
4329
4330         write_lock_bh(&pol->lock);
4331         if (unlikely(pol->walk.dead)) {
4332                 /* target policy has been deleted */
4333                 write_unlock_bh(&pol->lock);
4334                 return -ENOENT;
4335         }
4336
4337         for (i = 0; i < pol->xfrm_nr; i++) {
4338                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4339                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4340                                 continue;
4341                         n++;
4342                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4343                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4344                                 continue;
4345                         /* update endpoints */
4346                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4347                                sizeof(pol->xfrm_vec[i].id.daddr));
4348                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4349                                sizeof(pol->xfrm_vec[i].saddr));
4350                         pol->xfrm_vec[i].encap_family = mp->new_family;
4351                         /* flush bundles */
4352                         atomic_inc(&pol->genid);
4353                 }
4354         }
4355
4356         write_unlock_bh(&pol->lock);
4357
4358         if (!n)
4359                 return -ENODATA;
4360
4361         return 0;
4362 }
4363
4364 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4365 {
4366         int i, j;
4367
4368         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4369                 return -EINVAL;
4370
4371         for (i = 0; i < num_migrate; i++) {
4372                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4373                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4374                         return -EINVAL;
4375
4376                 /* check if there is any duplicated entry */
4377                 for (j = i + 1; j < num_migrate; j++) {
4378                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4379                                     sizeof(m[i].old_daddr)) &&
4380                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4381                                     sizeof(m[i].old_saddr)) &&
4382                             m[i].proto == m[j].proto &&
4383                             m[i].mode == m[j].mode &&
4384                             m[i].reqid == m[j].reqid &&
4385                             m[i].old_family == m[j].old_family)
4386                                 return -EINVAL;
4387                 }
4388         }
4389
4390         return 0;
4391 }
4392
4393 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4394                  struct xfrm_migrate *m, int num_migrate,
4395                  struct xfrm_kmaddress *k, struct net *net,
4396                  struct xfrm_encap_tmpl *encap)
4397 {
4398         int i, err, nx_cur = 0, nx_new = 0;
4399         struct xfrm_policy *pol = NULL;
4400         struct xfrm_state *x, *xc;
4401         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4402         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4403         struct xfrm_migrate *mp;
4404
4405         /* Stage 0 - sanity checks */
4406         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4407                 goto out;
4408
4409         if (dir >= XFRM_POLICY_MAX) {
4410                 err = -EINVAL;
4411                 goto out;
4412         }
4413
4414         /* Stage 1 - find policy */
4415         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4416                 err = -ENOENT;
4417                 goto out;
4418         }
4419
4420         /* Stage 2 - find and update state(s) */
4421         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4422                 if ((x = xfrm_migrate_state_find(mp, net))) {
4423                         x_cur[nx_cur] = x;
4424                         nx_cur++;
4425                         xc = xfrm_state_migrate(x, mp, encap);
4426                         if (xc) {
4427                                 x_new[nx_new] = xc;
4428                                 nx_new++;
4429                         } else {
4430                                 err = -ENODATA;
4431                                 goto restore_state;
4432                         }
4433                 }
4434         }
4435
4436         /* Stage 3 - update policy */
4437         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4438                 goto restore_state;
4439
4440         /* Stage 4 - delete old state(s) */
4441         if (nx_cur) {
4442                 xfrm_states_put(x_cur, nx_cur);
4443                 xfrm_states_delete(x_cur, nx_cur);
4444         }
4445
4446         /* Stage 5 - announce */
4447         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4448
4449         xfrm_pol_put(pol);
4450
4451         return 0;
4452 out:
4453         return err;
4454
4455 restore_state:
4456         if (pol)
4457                 xfrm_pol_put(pol);
4458         if (nx_cur)
4459                 xfrm_states_put(x_cur, nx_cur);
4460         if (nx_new)
4461                 xfrm_states_delete(x_new, nx_new);
4462
4463         return err;
4464 }
4465 EXPORT_SYMBOL(xfrm_migrate);
4466 #endif