net/xfrm/xfrm_policy.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * xfrm_policy.c
   4  *
   5  * Changes:
   6  *      Mitsuru KANDA @USAGI
   7  *      Kazunori MIYAZAWA @USAGI
   8  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   9  *              IPv6 support
  10  *      Kazunori MIYAZAWA @USAGI
  11  *      YOSHIFUJI Hideaki
  12  *              Split up af-specific portion
  13  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  14  *
  15  */
  16
  17 #include <linux/err.h>
  18 #include <linux/slab.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/spinlock.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/notifier.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/netfilter.h>
  26 #include <linux/module.h>
  27 #include <linux/cache.h>
  28 #include <linux/cpu.h>
  29 #include <linux/audit.h>
  30 #include <linux/rhashtable.h>
  31 #include <linux/if_tunnel.h>
  32 #include <net/dst.h>
  33 #include <net/flow.h>
  34 #include <net/xfrm.h>
  35 #include <net/ip.h>
  36 #if IS_ENABLED(CONFIG_IPV6_MIP6)
  37 #include <net/mip6.h>
  38 #endif
  39 #ifdef CONFIG_XFRM_STATISTICS
  40 #include <net/snmp.h>
  41 #endif
  42 #ifdef CONFIG_INET_ESPINTCP
  43 #include <net/espintcp.h>
  44 #endif
  45
  46 #include "xfrm_hash.h"
  47
  48 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  49 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  50 #define XFRM_MAX_QUEUE_LEN      100
  51
  52 struct xfrm_flo {
  53         struct dst_entry *dst_orig;
  54         u8 flags;
  55 };
  56
  57 /* prefixes smaller than this are stored in lists, not trees. */
  58 #define INEXACT_PREFIXLEN_IPV4  16
  59 #define INEXACT_PREFIXLEN_IPV6  48
  60
  61 struct xfrm_pol_inexact_node {
  62         struct rb_node node;
  63         union {
  64                 xfrm_address_t addr;
  65                 struct rcu_head rcu;
  66         };
  67         u8 prefixlen;
  68
  69         struct rb_root root;
  70
  71         /* the policies matching this node, can be empty list */
  72         struct hlist_head hhead;
  73 };
  74
  75 /* xfrm inexact policy search tree:
  76  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
  77  *  |
  78  * +---- root_d: sorted by daddr:prefix
  79  * |                 |
  80  * |        xfrm_pol_inexact_node
  81  * |                 |
  82  * |                 +- root: sorted by saddr/prefix
  83  * |                 |              |
  84  * |                 |         xfrm_pol_inexact_node
  85  * |                 |              |
  86  * |                 |              + root: unused
  87  * |                 |              |
  88  * |                 |              + hhead: saddr:daddr policies
  89  * |                 |
  90  * |                 +- coarse policies and all any:daddr policies
  91  * |
  92  * +---- root_s: sorted by saddr:prefix
  93  * |                 |
  94  * |        xfrm_pol_inexact_node
  95  * |                 |
  96  * |                 + root: unused
  97  * |                 |
  98  * |                 + hhead: saddr:any policies
  99  * |
 100  * +---- coarse policies and all any:any policies
 101  *
 102  * Lookups return four candidate lists:
 103  * 1. any:any list from top-level xfrm_pol_inexact_bin
 104  * 2. any:daddr list from daddr tree
 105  * 3. saddr:daddr list from 2nd level daddr tree
 106  * 4. saddr:any list from saddr tree
 107  *
 108  * This result set then needs to be searched for the policy with
 109  * the lowest priority.  If two results have same prio, youngest one wins.
 110  */
 111
 112 struct xfrm_pol_inexact_key {
 113         possible_net_t net;
 114         u32 if_id;
 115         u16 family;
 116         u8 dir, type;
 117 };
 118
 119 struct xfrm_pol_inexact_bin {
 120         struct xfrm_pol_inexact_key k;
 121         struct rhash_head head;
 122         /* list containing '*:*' policies */
 123         struct hlist_head hhead;
 124
 125         seqcount_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         policy->walk.dead = 1;
 438
 439         atomic_inc(&policy->genid);
 440
 441         if (del_timer(&policy->polq.hold_timer))
 442                 xfrm_pol_put(policy);
 443         skb_queue_purge(&policy->polq.hold_queue);
 444
 445         if (del_timer(&policy->timer))
 446                 xfrm_pol_put(policy);
 447
 448         xfrm_pol_put(policy);
 449 }
 450
 451 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 452
 453 static inline unsigned int idx_hash(struct net *net, u32 index)
 454 {
 455         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 456 }
 457
 458 /* calculate policy hash thresholds */
 459 static void __get_hash_thresh(struct net *net,
 460                               unsigned short family, int dir,
 461                               u8 *dbits, u8 *sbits)
 462 {
 463         switch (family) {
 464         case AF_INET:
 465                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 466                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 467                 break;
 468
 469         case AF_INET6:
 470                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 471                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 472                 break;
 473
 474         default:
 475                 *dbits = 0;
 476                 *sbits = 0;
 477         }
 478 }
 479
 480 static struct hlist_head *policy_hash_bysel(struct net *net,
 481                                             const struct xfrm_selector *sel,
 482                                             unsigned short family, int dir)
 483 {
 484         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 485         unsigned int hash;
 486         u8 dbits;
 487         u8 sbits;
 488
 489         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 490         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 491
 492         if (hash == hmask + 1)
 493                 return NULL;
 494
 495         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 496                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 497 }
 498
 499 static struct hlist_head *policy_hash_direct(struct net *net,
 500                                              const xfrm_address_t *daddr,
 501                                              const xfrm_address_t *saddr,
 502                                              unsigned short family, int dir)
 503 {
 504         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 505         unsigned int hash;
 506         u8 dbits;
 507         u8 sbits;
 508
 509         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 510         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 511
 512         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 513                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 514 }
 515
 516 static void xfrm_dst_hash_transfer(struct net *net,
 517                                    struct hlist_head *list,
 518                                    struct hlist_head *ndsttable,
 519                                    unsigned int nhashmask,
 520                                    int dir)
 521 {
 522         struct hlist_node *tmp, *entry0 = NULL;
 523         struct xfrm_policy *pol;
 524         unsigned int h0 = 0;
 525         u8 dbits;
 526         u8 sbits;
 527
 528 redo:
 529         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 530                 unsigned int h;
 531
 532                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 533                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 534                                 pol->family, nhashmask, dbits, sbits);
 535                 if (!entry0) {
 536                         hlist_del_rcu(&pol->bydst);
 537                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 538                         h0 = h;
 539                 } else {
 540                         if (h != h0)
 541                                 continue;
 542                         hlist_del_rcu(&pol->bydst);
 543                         hlist_add_behind_rcu(&pol->bydst, entry0);
 544                 }
 545                 entry0 = &pol->bydst;
 546         }
 547         if (!hlist_empty(list)) {
 548                 entry0 = NULL;
 549                 goto redo;
 550         }
 551 }
 552
 553 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 554                                    struct hlist_head *nidxtable,
 555                                    unsigned int nhashmask)
 556 {
 557         struct hlist_node *tmp;
 558         struct xfrm_policy *pol;
 559
 560         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 561                 unsigned int h;
 562
 563                 h = __idx_hash(pol->index, nhashmask);
 564                 hlist_add_head(&pol->byidx, nidxtable+h);
 565         }
 566 }
 567
 568 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 569 {
 570         return ((old_hmask + 1) << 1) - 1;
 571 }
 572
 573 static void xfrm_bydst_resize(struct net *net, int dir)
 574 {
 575         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 576         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 577         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 578         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 579         struct hlist_head *odst;
 580         int i;
 581
 582         if (!ndst)
 583                 return;
 584
 585         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 586         write_seqcount_begin(&xfrm_policy_hash_generation);
 587
 588         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 589                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 590
 591         for (i = hmask; i >= 0; i--)
 592                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 593
 594         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 595         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 596
 597         write_seqcount_end(&xfrm_policy_hash_generation);
 598         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 599
 600         synchronize_rcu();
 601
 602         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 603 }
 604
 605 static void xfrm_byidx_resize(struct net *net, int total)
 606 {
 607         unsigned int hmask = net->xfrm.policy_idx_hmask;
 608         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 609         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 610         struct hlist_head *oidx = net->xfrm.policy_byidx;
 611         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 612         int i;
 613
 614         if (!nidx)
 615                 return;
 616
 617         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 618
 619         for (i = hmask; i >= 0; i--)
 620                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 621
 622         net->xfrm.policy_byidx = nidx;
 623         net->xfrm.policy_idx_hmask = nhashmask;
 624
 625         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 626
 627         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 628 }
 629
 630 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 631 {
 632         unsigned int cnt = net->xfrm.policy_count[dir];
 633         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 634
 635         if (total)
 636                 *total += cnt;
 637
 638         if ((hmask + 1) < xfrm_policy_hashmax &&
 639             cnt > hmask)
 640                 return 1;
 641
 642         return 0;
 643 }
 644
 645 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 646 {
 647         unsigned int hmask = net->xfrm.policy_idx_hmask;
 648
 649         if ((hmask + 1) < xfrm_policy_hashmax &&
 650             total > hmask)
 651                 return 1;
 652
 653         return 0;
 654 }
 655
 656 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 657 {
 658         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 659         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 660         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 661         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 662         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 663         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 664         si->spdhcnt = net->xfrm.policy_idx_hmask;
 665         si->spdhmcnt = xfrm_policy_hashmax;
 666 }
 667 EXPORT_SYMBOL(xfrm_spd_getinfo);
 668
 669 static DEFINE_MUTEX(hash_resize_mutex);
 670 static void xfrm_hash_resize(struct work_struct *work)
 671 {
 672         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 673         int dir, total;
 674
 675         mutex_lock(&hash_resize_mutex);
 676
 677         total = 0;
 678         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 679                 if (xfrm_bydst_should_resize(net, dir, &total))
 680                         xfrm_bydst_resize(net, dir);
 681         }
 682         if (xfrm_byidx_should_resize(net, total))
 683                 xfrm_byidx_resize(net, total);
 684
 685         mutex_unlock(&hash_resize_mutex);
 686 }
 687
 688 /* Make sure *pol can be inserted into fastbin.
 689  * Useful to check that later insert requests will be sucessful
 690  * (provided xfrm_policy_lock is held throughout).
 691  */
 692 static struct xfrm_pol_inexact_bin *
 693 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
 694 {
 695         struct xfrm_pol_inexact_bin *bin, *prev;
 696         struct xfrm_pol_inexact_key k = {
 697                 .family = pol->family,
 698                 .type = pol->type,
 699                 .dir = dir,
 700                 .if_id = pol->if_id,
 701         };
 702         struct net *net = xp_net(pol);
 703
 704         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 705
 706         write_pnet(&k.net, net);
 707         bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
 708                                      xfrm_pol_inexact_params);
 709         if (bin)
 710                 return bin;
 711
 712         bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
 713         if (!bin)
 714                 return NULL;
 715
 716         bin->k = k;
 717         INIT_HLIST_HEAD(&bin->hhead);
 718         bin->root_d = RB_ROOT;
 719         bin->root_s = RB_ROOT;
 720         seqcount_init(&bin->count);
 721
 722         prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
 723                                                 &bin->k, &bin->head,
 724                                                 xfrm_pol_inexact_params);
 725         if (!prev) {
 726                 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
 727                 return bin;
 728         }
 729
 730         kfree(bin);
 731
 732         return IS_ERR(prev) ? NULL : prev;
 733 }
 734
 735 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
 736                                                int family, u8 prefixlen)
 737 {
 738         if (xfrm_addr_any(addr, family))
 739                 return true;
 740
 741         if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
 742                 return true;
 743
 744         if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
 745                 return true;
 746
 747         return false;
 748 }
 749
 750 static bool
 751 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
 752 {
 753         const xfrm_address_t *addr;
 754         bool saddr_any, daddr_any;
 755         u8 prefixlen;
 756
 757         addr = &policy->selector.saddr;
 758         prefixlen = policy->selector.prefixlen_s;
 759
 760         saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 761                                                        policy->family,
 762                                                        prefixlen);
 763         addr = &policy->selector.daddr;
 764         prefixlen = policy->selector.prefixlen_d;
 765         daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 766                                                        policy->family,
 767                                                        prefixlen);
 768         return saddr_any && daddr_any;
 769 }
 770
 771 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
 772                                        const xfrm_address_t *addr, u8 prefixlen)
 773 {
 774         node->addr = *addr;
 775         node->prefixlen = prefixlen;
 776 }
 777
 778 static struct xfrm_pol_inexact_node *
 779 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
 780 {
 781         struct xfrm_pol_inexact_node *node;
 782
 783         node = kzalloc(sizeof(*node), GFP_ATOMIC);
 784         if (node)
 785                 xfrm_pol_inexact_node_init(node, addr, prefixlen);
 786
 787         return node;
 788 }
 789
 790 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
 791                                   const xfrm_address_t *b,
 792                                   u8 prefixlen, u16 family)
 793 {
 794         unsigned int pdw, pbi;
 795         int delta = 0;
 796
 797         switch (family) {
 798         case AF_INET:
 799                 if (sizeof(long) == 4 && prefixlen == 0)
 800                         return ntohl(a->a4) - ntohl(b->a4);
 801                 return (ntohl(a->a4) & ((~0UL << (32 - prefixlen)))) -
 802                        (ntohl(b->a4) & ((~0UL << (32 - prefixlen))));
 803         case AF_INET6:
 804                 pdw = prefixlen >> 5;
 805                 pbi = prefixlen & 0x1f;
 806
 807                 if (pdw) {
 808                         delta = memcmp(a->a6, b->a6, pdw << 2);
 809                         if (delta)
 810                                 return delta;
 811                 }
 812                 if (pbi) {
 813                         u32 mask = ~0u << (32 - pbi);
 814
 815                         delta = (ntohl(a->a6[pdw]) & mask) -
 816                                 (ntohl(b->a6[pdw]) & mask);
 817                 }
 818                 break;
 819         default:
 820                 break;
 821         }
 822
 823         return delta;
 824 }
 825
 826 static void xfrm_policy_inexact_list_reinsert(struct net *net,
 827                                               struct xfrm_pol_inexact_node *n,
 828                                               u16 family)
 829 {
 830         unsigned int matched_s, matched_d;
 831         struct xfrm_policy *policy, *p;
 832
 833         matched_s = 0;
 834         matched_d = 0;
 835
 836         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 837                 struct hlist_node *newpos = NULL;
 838                 bool matches_s, matches_d;
 839
 840                 if (!policy->bydst_reinsert)
 841                         continue;
 842
 843                 WARN_ON_ONCE(policy->family != family);
 844
 845                 policy->bydst_reinsert = false;
 846                 hlist_for_each_entry(p, &n->hhead, bydst) {
 847                         if (policy->priority > p->priority)
 848                                 newpos = &p->bydst;
 849                         else if (policy->priority == p->priority &&
 850                                  policy->pos > p->pos)
 851                                 newpos = &p->bydst;
 852                         else
 853                                 break;
 854                 }
 855
 856                 if (newpos)
 857                         hlist_add_behind_rcu(&policy->bydst, newpos);
 858                 else
 859                         hlist_add_head_rcu(&policy->bydst, &n->hhead);
 860
 861                 /* paranoia checks follow.
 862                  * Check that the reinserted policy matches at least
 863                  * saddr or daddr for current node prefix.
 864                  *
 865                  * Matching both is fine, matching saddr in one policy
 866                  * (but not daddr) and then matching only daddr in another
 867                  * is a bug.
 868                  */
 869                 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
 870                                                    &n->addr,
 871                                                    n->prefixlen,
 872                                                    family) == 0;
 873                 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
 874                                                    &n->addr,
 875                                                    n->prefixlen,
 876                                                    family) == 0;
 877                 if (matches_s && matches_d)
 878                         continue;
 879
 880                 WARN_ON_ONCE(!matches_s && !matches_d);
 881                 if (matches_s)
 882                         matched_s++;
 883                 if (matches_d)
 884                         matched_d++;
 885                 WARN_ON_ONCE(matched_s && matched_d);
 886         }
 887 }
 888
 889 static void xfrm_policy_inexact_node_reinsert(struct net *net,
 890                                               struct xfrm_pol_inexact_node *n,
 891                                               struct rb_root *new,
 892                                               u16 family)
 893 {
 894         struct xfrm_pol_inexact_node *node;
 895         struct rb_node **p, *parent;
 896
 897         /* we should not have another subtree here */
 898         WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
 899 restart:
 900         parent = NULL;
 901         p = &new->rb_node;
 902         while (*p) {
 903                 u8 prefixlen;
 904                 int delta;
 905
 906                 parent = *p;
 907                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 908
 909                 prefixlen = min(node->prefixlen, n->prefixlen);
 910
 911                 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
 912                                                prefixlen, family);
 913                 if (delta < 0) {
 914                         p = &parent->rb_left;
 915                 } else if (delta > 0) {
 916                         p = &parent->rb_right;
 917                 } else {
 918                         bool same_prefixlen = node->prefixlen == n->prefixlen;
 919                         struct xfrm_policy *tmp;
 920
 921                         hlist_for_each_entry(tmp, &n->hhead, bydst) {
 922                                 tmp->bydst_reinsert = true;
 923                                 hlist_del_rcu(&tmp->bydst);
 924                         }
 925
 926                         node->prefixlen = prefixlen;
 927
 928                         xfrm_policy_inexact_list_reinsert(net, node, family);
 929
 930                         if (same_prefixlen) {
 931                                 kfree_rcu(n, rcu);
 932                                 return;
 933                         }
 934
 935                         rb_erase(*p, new);
 936                         kfree_rcu(n, rcu);
 937                         n = node;
 938                         goto restart;
 939                 }
 940         }
 941
 942         rb_link_node_rcu(&n->node, parent, p);
 943         rb_insert_color(&n->node, new);
 944 }
 945
 946 /* merge nodes v and n */
 947 static void xfrm_policy_inexact_node_merge(struct net *net,
 948                                            struct xfrm_pol_inexact_node *v,
 949                                            struct xfrm_pol_inexact_node *n,
 950                                            u16 family)
 951 {
 952         struct xfrm_pol_inexact_node *node;
 953         struct xfrm_policy *tmp;
 954         struct rb_node *rnode;
 955
 956         /* To-be-merged node v has a subtree.
 957          *
 958          * Dismantle it and insert its nodes to n->root.
 959          */
 960         while ((rnode = rb_first(&v->root)) != NULL) {
 961                 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
 962                 rb_erase(&node->node, &v->root);
 963                 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
 964                                                   family);
 965         }
 966
 967         hlist_for_each_entry(tmp, &v->hhead, bydst) {
 968                 tmp->bydst_reinsert = true;
 969                 hlist_del_rcu(&tmp->bydst);
 970         }
 971
 972         xfrm_policy_inexact_list_reinsert(net, n, family);
 973 }
 974
 975 static struct xfrm_pol_inexact_node *
 976 xfrm_policy_inexact_insert_node(struct net *net,
 977                                 struct rb_root *root,
 978                                 xfrm_address_t *addr,
 979                                 u16 family, u8 prefixlen, u8 dir)
 980 {
 981         struct xfrm_pol_inexact_node *cached = NULL;
 982         struct rb_node **p, *parent = NULL;
 983         struct xfrm_pol_inexact_node *node;
 984
 985         p = &root->rb_node;
 986         while (*p) {
 987                 int delta;
 988
 989                 parent = *p;
 990                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 991
 992                 delta = xfrm_policy_addr_delta(addr, &node->addr,
 993                                                node->prefixlen,
 994                                                family);
 995                 if (delta == 0 && prefixlen >= node->prefixlen) {
 996                         WARN_ON_ONCE(cached); /* ipsec policies got lost */
 997                         return node;
 998                 }
 999
1000                 if (delta < 0)
1001                         p = &parent->rb_left;
1002                 else
1003                         p = &parent->rb_right;
1004
1005                 if (prefixlen < node->prefixlen) {
1006                         delta = xfrm_policy_addr_delta(addr, &node->addr,
1007                                                        prefixlen,
1008                                                        family);
1009                         if (delta)
1010                                 continue;
1011
1012                         /* This node is a subnet of the new prefix. It needs
1013                          * to be removed and re-inserted with the smaller
1014                          * prefix and all nodes that are now also covered
1015                          * by the reduced prefixlen.
1016                          */
1017                         rb_erase(&node->node, root);
1018
1019                         if (!cached) {
1020                                 xfrm_pol_inexact_node_init(node, addr,
1021                                                            prefixlen);
1022                                 cached = node;
1023                         } else {
1024                                 /* This node also falls within the new
1025                                  * prefixlen. Merge the to-be-reinserted
1026                                  * node and this one.
1027                                  */
1028                                 xfrm_policy_inexact_node_merge(net, node,
1029                                                                cached, family);
1030                                 kfree_rcu(node, rcu);
1031                         }
1032
1033                         /* restart */
1034                         p = &root->rb_node;
1035                         parent = NULL;
1036                 }
1037         }
1038
1039         node = cached;
1040         if (!node) {
1041                 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1042                 if (!node)
1043                         return NULL;
1044         }
1045
1046         rb_link_node_rcu(&node->node, parent, p);
1047         rb_insert_color(&node->node, root);
1048
1049         return node;
1050 }
1051
1052 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1053 {
1054         struct xfrm_pol_inexact_node *node;
1055         struct rb_node *rn = rb_first(r);
1056
1057         while (rn) {
1058                 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1059
1060                 xfrm_policy_inexact_gc_tree(&node->root, rm);
1061                 rn = rb_next(rn);
1062
1063                 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1064                         WARN_ON_ONCE(rm);
1065                         continue;
1066                 }
1067
1068                 rb_erase(&node->node, r);
1069                 kfree_rcu(node, rcu);
1070         }
1071 }
1072
1073 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1074 {
1075         write_seqcount_begin(&b->count);
1076         xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1077         xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1078         write_seqcount_end(&b->count);
1079
1080         if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1081             !hlist_empty(&b->hhead)) {
1082                 WARN_ON_ONCE(net_exit);
1083                 return;
1084         }
1085
1086         if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1087                                    xfrm_pol_inexact_params) == 0) {
1088                 list_del(&b->inexact_bins);
1089                 kfree_rcu(b, rcu);
1090         }
1091 }
1092
1093 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1094 {
1095         struct net *net = read_pnet(&b->k.net);
1096
1097         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1098         __xfrm_policy_inexact_prune_bin(b, false);
1099         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1100 }
1101
1102 static void __xfrm_policy_inexact_flush(struct net *net)
1103 {
1104         struct xfrm_pol_inexact_bin *bin, *t;
1105
1106         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1107
1108         list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1109                 __xfrm_policy_inexact_prune_bin(bin, false);
1110 }
1111
1112 static struct hlist_head *
1113 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1114                                 struct xfrm_policy *policy, u8 dir)
1115 {
1116         struct xfrm_pol_inexact_node *n;
1117         struct net *net;
1118
1119         net = xp_net(policy);
1120         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1121
1122         if (xfrm_policy_inexact_insert_use_any_list(policy))
1123                 return &bin->hhead;
1124
1125         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1126                                                policy->family,
1127                                                policy->selector.prefixlen_d)) {
1128                 write_seqcount_begin(&bin->count);
1129                 n = xfrm_policy_inexact_insert_node(net,
1130                                                     &bin->root_s,
1131                                                     &policy->selector.saddr,
1132                                                     policy->family,
1133                                                     policy->selector.prefixlen_s,
1134                                                     dir);
1135                 write_seqcount_end(&bin->count);
1136                 if (!n)
1137                         return NULL;
1138
1139                 return &n->hhead;
1140         }
1141
1142         /* daddr is fixed */
1143         write_seqcount_begin(&bin->count);
1144         n = xfrm_policy_inexact_insert_node(net,
1145                                             &bin->root_d,
1146                                             &policy->selector.daddr,
1147                                             policy->family,
1148                                             policy->selector.prefixlen_d, dir);
1149         write_seqcount_end(&bin->count);
1150         if (!n)
1151                 return NULL;
1152
1153         /* saddr is wildcard */
1154         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1155                                                policy->family,
1156                                                policy->selector.prefixlen_s))
1157                 return &n->hhead;
1158
1159         write_seqcount_begin(&bin->count);
1160         n = xfrm_policy_inexact_insert_node(net,
1161                                             &n->root,
1162                                             &policy->selector.saddr,
1163                                             policy->family,
1164                                             policy->selector.prefixlen_s, dir);
1165         write_seqcount_end(&bin->count);
1166         if (!n)
1167                 return NULL;
1168
1169         return &n->hhead;
1170 }
1171
1172 static struct xfrm_policy *
1173 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1174 {
1175         struct xfrm_pol_inexact_bin *bin;
1176         struct xfrm_policy *delpol;
1177         struct hlist_head *chain;
1178         struct net *net;
1179
1180         bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1181         if (!bin)
1182                 return ERR_PTR(-ENOMEM);
1183
1184         net = xp_net(policy);
1185         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1186
1187         chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1188         if (!chain) {
1189                 __xfrm_policy_inexact_prune_bin(bin, false);
1190                 return ERR_PTR(-ENOMEM);
1191         }
1192
1193         delpol = xfrm_policy_insert_list(chain, policy, excl);
1194         if (delpol && excl) {
1195                 __xfrm_policy_inexact_prune_bin(bin, false);
1196                 return ERR_PTR(-EEXIST);
1197         }
1198
1199         chain = &net->xfrm.policy_inexact[dir];
1200         xfrm_policy_insert_inexact_list(chain, policy);
1201
1202         if (delpol)
1203                 __xfrm_policy_inexact_prune_bin(bin, false);
1204
1205         return delpol;
1206 }
1207
1208 static void xfrm_hash_rebuild(struct work_struct *work)
1209 {
1210         struct net *net = container_of(work, struct net,
1211                                        xfrm.policy_hthresh.work);
1212         unsigned int hmask;
1213         struct xfrm_policy *pol;
1214         struct xfrm_policy *policy;
1215         struct hlist_head *chain;
1216         struct hlist_head *odst;
1217         struct hlist_node *newpos;
1218         int i;
1219         int dir;
1220         unsigned seq;
1221         u8 lbits4, rbits4, lbits6, rbits6;
1222
1223         mutex_lock(&hash_resize_mutex);
1224
1225         /* read selector prefixlen thresholds */
1226         do {
1227                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1228
1229                 lbits4 = net->xfrm.policy_hthresh.lbits4;
1230                 rbits4 = net->xfrm.policy_hthresh.rbits4;
1231                 lbits6 = net->xfrm.policy_hthresh.lbits6;
1232                 rbits6 = net->xfrm.policy_hthresh.rbits6;
1233         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1234
1235         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1236         write_seqcount_begin(&xfrm_policy_hash_generation);
1237
1238         /* make sure that we can insert the indirect policies again before
1239          * we start with destructive action.
1240          */
1241         list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1242                 struct xfrm_pol_inexact_bin *bin;
1243                 u8 dbits, sbits;
1244
1245                 dir = xfrm_policy_id2dir(policy->index);
1246                 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1247                         continue;
1248
1249                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1250                         if (policy->family == AF_INET) {
1251                                 dbits = rbits4;
1252                                 sbits = lbits4;
1253                         } else {
1254                                 dbits = rbits6;
1255                                 sbits = lbits6;
1256                         }
1257                 } else {
1258                         if (policy->family == AF_INET) {
1259                                 dbits = lbits4;
1260                                 sbits = rbits4;
1261                         } else {
1262                                 dbits = lbits6;
1263                                 sbits = rbits6;
1264                         }
1265                 }
1266
1267                 if (policy->selector.prefixlen_d < dbits ||
1268                     policy->selector.prefixlen_s < sbits)
1269                         continue;
1270
1271                 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1272                 if (!bin)
1273                         goto out_unlock;
1274
1275                 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1276                         goto out_unlock;
1277         }
1278
1279         /* reset the bydst and inexact table in all directions */
1280         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1281                 struct hlist_node *n;
1282
1283                 hlist_for_each_entry_safe(policy, n,
1284                                           &net->xfrm.policy_inexact[dir],
1285                                           bydst_inexact_list) {
1286                         hlist_del_rcu(&policy->bydst);
1287                         hlist_del_init(&policy->bydst_inexact_list);
1288                 }
1289
1290                 hmask = net->xfrm.policy_bydst[dir].hmask;
1291                 odst = net->xfrm.policy_bydst[dir].table;
1292                 for (i = hmask; i >= 0; i--) {
1293                         hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1294                                 hlist_del_rcu(&policy->bydst);
1295                 }
1296                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1297                         /* dir out => dst = remote, src = local */
1298                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1299                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1300                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1301                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1302                 } else {
1303                         /* dir in/fwd => dst = local, src = remote */
1304                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1305                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1306                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1307                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1308                 }
1309         }
1310
1311         /* re-insert all policies by order of creation */
1312         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1313                 if (policy->walk.dead)
1314                         continue;
1315                 dir = xfrm_policy_id2dir(policy->index);
1316                 if (dir >= XFRM_POLICY_MAX) {
1317                         /* skip socket policies */
1318                         continue;
1319                 }
1320                 newpos = NULL;
1321                 chain = policy_hash_bysel(net, &policy->selector,
1322                                           policy->family, dir);
1323
1324                 if (!chain) {
1325                         void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1326
1327                         WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1328                         continue;
1329                 }
1330
1331                 hlist_for_each_entry(pol, chain, bydst) {
1332                         if (policy->priority >= pol->priority)
1333                                 newpos = &pol->bydst;
1334                         else
1335                                 break;
1336                 }
1337                 if (newpos)
1338                         hlist_add_behind_rcu(&policy->bydst, newpos);
1339                 else
1340                         hlist_add_head_rcu(&policy->bydst, chain);
1341         }
1342
1343 out_unlock:
1344         __xfrm_policy_inexact_flush(net);
1345         write_seqcount_end(&xfrm_policy_hash_generation);
1346         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1347
1348         mutex_unlock(&hash_resize_mutex);
1349 }
1350
1351 void xfrm_policy_hash_rebuild(struct net *net)
1352 {
1353         schedule_work(&net->xfrm.policy_hthresh.work);
1354 }
1355 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1356
1357 /* Generate new index... KAME seems to generate them ordered by cost
1358  * of an absolute inpredictability of ordering of rules. This will not pass. */
1359 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1360 {
1361         static u32 idx_generator;
1362
1363         for (;;) {
1364                 struct hlist_head *list;
1365                 struct xfrm_policy *p;
1366                 u32 idx;
1367                 int found;
1368
1369                 if (!index) {
1370                         idx = (idx_generator | dir);
1371                         idx_generator += 8;
1372                 } else {
1373                         idx = index;
1374                         index = 0;
1375                 }
1376
1377                 if (idx == 0)
1378                         idx = 8;
1379                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1380                 found = 0;
1381                 hlist_for_each_entry(p, list, byidx) {
1382                         if (p->index == idx) {
1383                                 found = 1;
1384                                 break;
1385                         }
1386                 }
1387                 if (!found)
1388                         return idx;
1389         }
1390 }
1391
1392 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1393 {
1394         u32 *p1 = (u32 *) s1;
1395         u32 *p2 = (u32 *) s2;
1396         int len = sizeof(struct xfrm_selector) / sizeof(u32);
1397         int i;
1398
1399         for (i = 0; i < len; i++) {
1400                 if (p1[i] != p2[i])
1401                         return 1;
1402         }
1403
1404         return 0;
1405 }
1406
1407 static void xfrm_policy_requeue(struct xfrm_policy *old,
1408                                 struct xfrm_policy *new)
1409 {
1410         struct xfrm_policy_queue *pq = &old->polq;
1411         struct sk_buff_head list;
1412
1413         if (skb_queue_empty(&pq->hold_queue))
1414                 return;
1415
1416         __skb_queue_head_init(&list);
1417
1418         spin_lock_bh(&pq->hold_queue.lock);
1419         skb_queue_splice_init(&pq->hold_queue, &list);
1420         if (del_timer(&pq->hold_timer))
1421                 xfrm_pol_put(old);
1422         spin_unlock_bh(&pq->hold_queue.lock);
1423
1424         pq = &new->polq;
1425
1426         spin_lock_bh(&pq->hold_queue.lock);
1427         skb_queue_splice(&list, &pq->hold_queue);
1428         pq->timeout = XFRM_QUEUE_TMO_MIN;
1429         if (!mod_timer(&pq->hold_timer, jiffies))
1430                 xfrm_pol_hold(new);
1431         spin_unlock_bh(&pq->hold_queue.lock);
1432 }
1433
1434 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
1435                                    struct xfrm_policy *pol)
1436 {
1437         u32 mark = policy->mark.v & policy->mark.m;
1438
1439         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
1440                 return true;
1441
1442         if ((mark & pol->mark.m) == pol->mark.v &&
1443             policy->priority == pol->priority)
1444                 return true;
1445
1446         return false;
1447 }
1448
1449 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1450 {
1451         const struct xfrm_pol_inexact_key *k = data;
1452         u32 a = k->type << 24 | k->dir << 16 | k->family;
1453
1454         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1455                             seed);
1456 }
1457
1458 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1459 {
1460         const struct xfrm_pol_inexact_bin *b = data;
1461
1462         return xfrm_pol_bin_key(&b->k, 0, seed);
1463 }
1464
1465 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1466                             const void *ptr)
1467 {
1468         const struct xfrm_pol_inexact_key *key = arg->key;
1469         const struct xfrm_pol_inexact_bin *b = ptr;
1470         int ret;
1471
1472         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1473                 return -1;
1474
1475         ret = b->k.dir ^ key->dir;
1476         if (ret)
1477                 return ret;
1478
1479         ret = b->k.type ^ key->type;
1480         if (ret)
1481                 return ret;
1482
1483         ret = b->k.family ^ key->family;
1484         if (ret)
1485                 return ret;
1486
1487         return b->k.if_id ^ key->if_id;
1488 }
1489
1490 static const struct rhashtable_params xfrm_pol_inexact_params = {
1491         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1492         .hashfn                 = xfrm_pol_bin_key,
1493         .obj_hashfn             = xfrm_pol_bin_obj,
1494         .obj_cmpfn              = xfrm_pol_bin_cmp,
1495         .automatic_shrinking    = true,
1496 };
1497
1498 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1499                                             struct xfrm_policy *policy)
1500 {
1501         struct xfrm_policy *pol, *delpol = NULL;
1502         struct hlist_node *newpos = NULL;
1503         int i = 0;
1504
1505         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1506                 if (pol->type == policy->type &&
1507                     pol->if_id == policy->if_id &&
1508                     !selector_cmp(&pol->selector, &policy->selector) &&
1509                     xfrm_policy_mark_match(policy, pol) &&
1510                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1511                     !WARN_ON(delpol)) {
1512                         delpol = pol;
1513                         if (policy->priority > pol->priority)
1514                                 continue;
1515                 } else if (policy->priority >= pol->priority) {
1516                         newpos = &pol->bydst_inexact_list;
1517                         continue;
1518                 }
1519                 if (delpol)
1520                         break;
1521         }
1522
1523         if (newpos)
1524                 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1525         else
1526                 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1527
1528         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1529                 pol->pos = i;
1530                 i++;
1531         }
1532 }
1533
1534 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1535                                                    struct xfrm_policy *policy,
1536                                                    bool excl)
1537 {
1538         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1539
1540         hlist_for_each_entry(pol, chain, bydst) {
1541                 if (pol->type == policy->type &&
1542                     pol->if_id == policy->if_id &&
1543                     !selector_cmp(&pol->selector, &policy->selector) &&
1544                     xfrm_policy_mark_match(policy, pol) &&
1545                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1546                     !WARN_ON(delpol)) {
1547                         if (excl)
1548                                 return ERR_PTR(-EEXIST);
1549                         delpol = pol;
1550                         if (policy->priority > pol->priority)
1551                                 continue;
1552                 } else if (policy->priority >= pol->priority) {
1553                         newpos = pol;
1554                         continue;
1555                 }
1556                 if (delpol)
1557                         break;
1558         }
1559
1560         if (newpos)
1561                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1562         else
1563                 hlist_add_head_rcu(&policy->bydst, chain);
1564
1565         return delpol;
1566 }
1567
1568 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1569 {
1570         struct net *net = xp_net(policy);
1571         struct xfrm_policy *delpol;
1572         struct hlist_head *chain;
1573
1574         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1575         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1576         if (chain)
1577                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1578         else
1579                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1580
1581         if (IS_ERR(delpol)) {
1582                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1583                 return PTR_ERR(delpol);
1584         }
1585
1586         __xfrm_policy_link(policy, dir);
1587
1588         /* After previous checking, family can either be AF_INET or AF_INET6 */
1589         if (policy->family == AF_INET)
1590                 rt_genid_bump_ipv4(net);
1591         else
1592                 rt_genid_bump_ipv6(net);
1593
1594         if (delpol) {
1595                 xfrm_policy_requeue(delpol, policy);
1596                 __xfrm_policy_unlink(delpol, dir);
1597         }
1598         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1599         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1600         policy->curlft.add_time = ktime_get_real_seconds();
1601         policy->curlft.use_time = 0;
1602         if (!mod_timer(&policy->timer, jiffies + HZ))
1603                 xfrm_pol_hold(policy);
1604         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1605
1606         if (delpol)
1607                 xfrm_policy_kill(delpol);
1608         else if (xfrm_bydst_should_resize(net, dir, NULL))
1609                 schedule_work(&net->xfrm.policy_hash_work);
1610
1611         return 0;
1612 }
1613 EXPORT_SYMBOL(xfrm_policy_insert);
1614
1615 static struct xfrm_policy *
1616 __xfrm_policy_bysel_ctx(struct hlist_head *chain, u32 mark, u32 if_id,
1617                         u8 type, int dir,
1618                         struct xfrm_selector *sel,
1619                         struct xfrm_sec_ctx *ctx)
1620 {
1621         struct xfrm_policy *pol;
1622
1623         if (!chain)
1624                 return NULL;
1625
1626         hlist_for_each_entry(pol, chain, bydst) {
1627                 if (pol->type == type &&
1628                     pol->if_id == if_id &&
1629                     (mark & pol->mark.m) == pol->mark.v &&
1630                     !selector_cmp(sel, &pol->selector) &&
1631                     xfrm_sec_ctx_match(ctx, pol->security))
1632                         return pol;
1633         }
1634
1635         return NULL;
1636 }
1637
1638 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u32 if_id,
1639                                           u8 type, int dir,
1640                                           struct xfrm_selector *sel,
1641                                           struct xfrm_sec_ctx *ctx, int delete,
1642                                           int *err)
1643 {
1644         struct xfrm_pol_inexact_bin *bin = NULL;
1645         struct xfrm_policy *pol, *ret = NULL;
1646         struct hlist_head *chain;
1647
1648         *err = 0;
1649         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1650         chain = policy_hash_bysel(net, sel, sel->family, dir);
1651         if (!chain) {
1652                 struct xfrm_pol_inexact_candidates cand;
1653                 int i;
1654
1655                 bin = xfrm_policy_inexact_lookup(net, type,
1656                                                  sel->family, dir, if_id);
1657                 if (!bin) {
1658                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1659                         return NULL;
1660                 }
1661
1662                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1663                                                          &sel->saddr,
1664                                                          &sel->daddr)) {
1665                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1666                         return NULL;
1667                 }
1668
1669                 pol = NULL;
1670                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1671                         struct xfrm_policy *tmp;
1672
1673                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1674                                                       if_id, type, dir,
1675                                                       sel, ctx);
1676                         if (!tmp)
1677                                 continue;
1678
1679                         if (!pol || tmp->pos < pol->pos)
1680                                 pol = tmp;
1681                 }
1682         } else {
1683                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1684                                               sel, ctx);
1685         }
1686
1687         if (pol) {
1688                 xfrm_pol_hold(pol);
1689                 if (delete) {
1690                         *err = security_xfrm_policy_delete(pol->security);
1691                         if (*err) {
1692                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1693                                 return pol;
1694                         }
1695                         __xfrm_policy_unlink(pol, dir);
1696                 }
1697                 ret = pol;
1698         }
1699         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1700
1701         if (ret && delete)
1702                 xfrm_policy_kill(ret);
1703         if (bin && delete)
1704                 xfrm_policy_inexact_prune_bin(bin);
1705         return ret;
1706 }
1707 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1708
1709 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u32 if_id,
1710                                      u8 type, int dir, u32 id, int delete,
1711                                      int *err)
1712 {
1713         struct xfrm_policy *pol, *ret;
1714         struct hlist_head *chain;
1715
1716         *err = -ENOENT;
1717         if (xfrm_policy_id2dir(id) != dir)
1718                 return NULL;
1719
1720         *err = 0;
1721         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1722         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1723         ret = NULL;
1724         hlist_for_each_entry(pol, chain, byidx) {
1725                 if (pol->type == type && pol->index == id &&
1726                     pol->if_id == if_id &&
1727                     (mark & pol->mark.m) == pol->mark.v) {
1728                         xfrm_pol_hold(pol);
1729                         if (delete) {
1730                                 *err = security_xfrm_policy_delete(
1731                                                                 pol->security);
1732                                 if (*err) {
1733                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1734                                         return pol;
1735                                 }
1736                                 __xfrm_policy_unlink(pol, dir);
1737                         }
1738                         ret = pol;
1739                         break;
1740                 }
1741         }
1742         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1743
1744         if (ret && delete)
1745                 xfrm_policy_kill(ret);
1746         return ret;
1747 }
1748 EXPORT_SYMBOL(xfrm_policy_byid);
1749
1750 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1751 static inline int
1752 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1753 {
1754         struct xfrm_policy *pol;
1755         int err = 0;
1756
1757         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1758                 if (pol->walk.dead ||
1759                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1760                     pol->type != type)
1761                         continue;
1762
1763                 err = security_xfrm_policy_delete(pol->security);
1764                 if (err) {
1765                         xfrm_audit_policy_delete(pol, 0, task_valid);
1766                         return err;
1767                 }
1768         }
1769         return err;
1770 }
1771 #else
1772 static inline int
1773 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1774 {
1775         return 0;
1776 }
1777 #endif
1778
1779 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1780 {
1781         int dir, err = 0, cnt = 0;
1782         struct xfrm_policy *pol;
1783
1784         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1785
1786         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1787         if (err)
1788                 goto out;
1789
1790 again:
1791         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1792                 dir = xfrm_policy_id2dir(pol->index);
1793                 if (pol->walk.dead ||
1794                     dir >= XFRM_POLICY_MAX ||
1795                     pol->type != type)
1796                         continue;
1797
1798                 __xfrm_policy_unlink(pol, dir);
1799                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1800                 cnt++;
1801                 xfrm_audit_policy_delete(pol, 1, task_valid);
1802                 xfrm_policy_kill(pol);
1803                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1804                 goto again;
1805         }
1806         if (cnt)
1807                 __xfrm_policy_inexact_flush(net);
1808         else
1809                 err = -ESRCH;
1810 out:
1811         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1812         return err;
1813 }
1814 EXPORT_SYMBOL(xfrm_policy_flush);
1815
1816 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1817                      int (*func)(struct xfrm_policy *, int, int, void*),
1818                      void *data)
1819 {
1820         struct xfrm_policy *pol;
1821         struct xfrm_policy_walk_entry *x;
1822         int error = 0;
1823
1824         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1825             walk->type != XFRM_POLICY_TYPE_ANY)
1826                 return -EINVAL;
1827
1828         if (list_empty(&walk->walk.all) && walk->seq != 0)
1829                 return 0;
1830
1831         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1832         if (list_empty(&walk->walk.all))
1833                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1834         else
1835                 x = list_first_entry(&walk->walk.all,
1836                                      struct xfrm_policy_walk_entry, all);
1837
1838         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1839                 if (x->dead)
1840                         continue;
1841                 pol = container_of(x, struct xfrm_policy, walk);
1842                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1843                     walk->type != pol->type)
1844                         continue;
1845                 error = func(pol, xfrm_policy_id2dir(pol->index),
1846                              walk->seq, data);
1847                 if (error) {
1848                         list_move_tail(&walk->walk.all, &x->all);
1849                         goto out;
1850                 }
1851                 walk->seq++;
1852         }
1853         if (walk->seq == 0) {
1854                 error = -ENOENT;
1855                 goto out;
1856         }
1857         list_del_init(&walk->walk.all);
1858 out:
1859         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1860         return error;
1861 }
1862 EXPORT_SYMBOL(xfrm_policy_walk);
1863
1864 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1865 {
1866         INIT_LIST_HEAD(&walk->walk.all);
1867         walk->walk.dead = 1;
1868         walk->type = type;
1869         walk->seq = 0;
1870 }
1871 EXPORT_SYMBOL(xfrm_policy_walk_init);
1872
1873 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1874 {
1875         if (list_empty(&walk->walk.all))
1876                 return;
1877
1878         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1879         list_del(&walk->walk.all);
1880         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1881 }
1882 EXPORT_SYMBOL(xfrm_policy_walk_done);
1883
1884 /*
1885  * Find policy to apply to this flow.
1886  *
1887  * Returns 0 if policy found, else an -errno.
1888  */
1889 static int xfrm_policy_match(const struct xfrm_policy *pol,
1890                              const struct flowi *fl,
1891                              u8 type, u16 family, int dir, u32 if_id)
1892 {
1893         const struct xfrm_selector *sel = &pol->selector;
1894         int ret = -ESRCH;
1895         bool match;
1896
1897         if (pol->family != family ||
1898             pol->if_id != if_id ||
1899             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1900             pol->type != type)
1901                 return ret;
1902
1903         match = xfrm_selector_match(sel, fl, family);
1904         if (match)
1905                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1906                                                   dir);
1907         return ret;
1908 }
1909
1910 static struct xfrm_pol_inexact_node *
1911 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1912                                 seqcount_t *count,
1913                                 const xfrm_address_t *addr, u16 family)
1914 {
1915         const struct rb_node *parent;
1916         int seq;
1917
1918 again:
1919         seq = read_seqcount_begin(count);
1920
1921         parent = rcu_dereference_raw(r->rb_node);
1922         while (parent) {
1923                 struct xfrm_pol_inexact_node *node;
1924                 int delta;
1925
1926                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1927
1928                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1929                                                node->prefixlen, family);
1930                 if (delta < 0) {
1931                         parent = rcu_dereference_raw(parent->rb_left);
1932                         continue;
1933                 } else if (delta > 0) {
1934                         parent = rcu_dereference_raw(parent->rb_right);
1935                         continue;
1936                 }
1937
1938                 return node;
1939         }
1940
1941         if (read_seqcount_retry(count, seq))
1942                 goto again;
1943
1944         return NULL;
1945 }
1946
1947 static bool
1948 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1949                                     struct xfrm_pol_inexact_bin *b,
1950                                     const xfrm_address_t *saddr,
1951                                     const xfrm_address_t *daddr)
1952 {
1953         struct xfrm_pol_inexact_node *n;
1954         u16 family;
1955
1956         if (!b)
1957                 return false;
1958
1959         family = b->k.family;
1960         memset(cand, 0, sizeof(*cand));
1961         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1962
1963         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1964                                             family);
1965         if (n) {
1966                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1967                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1968                                                     family);
1969                 if (n)
1970                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1971         }
1972
1973         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1974                                             family);
1975         if (n)
1976                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1977
1978         return true;
1979 }
1980
1981 static struct xfrm_pol_inexact_bin *
1982 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1983                                u8 dir, u32 if_id)
1984 {
1985         struct xfrm_pol_inexact_key k = {
1986                 .family = family,
1987                 .type = type,
1988                 .dir = dir,
1989                 .if_id = if_id,
1990         };
1991
1992         write_pnet(&k.net, net);
1993
1994         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1995                                  xfrm_pol_inexact_params);
1996 }
1997
1998 static struct xfrm_pol_inexact_bin *
1999 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2000                            u8 dir, u32 if_id)
2001 {
2002         struct xfrm_pol_inexact_bin *bin;
2003
2004         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2005
2006         rcu_read_lock();
2007         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2008         rcu_read_unlock();
2009
2010         return bin;
2011 }
2012
2013 static struct xfrm_policy *
2014 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2015                               struct xfrm_policy *prefer,
2016                               const struct flowi *fl,
2017                               u8 type, u16 family, int dir, u32 if_id)
2018 {
2019         u32 priority = prefer ? prefer->priority : ~0u;
2020         struct xfrm_policy *pol;
2021
2022         if (!chain)
2023                 return NULL;
2024
2025         hlist_for_each_entry_rcu(pol, chain, bydst) {
2026                 int err;
2027
2028                 if (pol->priority > priority)
2029                         break;
2030
2031                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2032                 if (err) {
2033                         if (err != -ESRCH)
2034                                 return ERR_PTR(err);
2035
2036                         continue;
2037                 }
2038
2039                 if (prefer) {
2040                         /* matches.  Is it older than *prefer? */
2041                         if (pol->priority == priority &&
2042                             prefer->pos < pol->pos)
2043                                 return prefer;
2044                 }
2045
2046                 return pol;
2047         }
2048
2049         return NULL;
2050 }
2051
2052 static struct xfrm_policy *
2053 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2054                             struct xfrm_policy *prefer,
2055                             const struct flowi *fl,
2056                             u8 type, u16 family, int dir, u32 if_id)
2057 {
2058         struct xfrm_policy *tmp;
2059         int i;
2060
2061         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2062                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2063                                                     prefer,
2064                                                     fl, type, family, dir,
2065                                                     if_id);
2066                 if (!tmp)
2067                         continue;
2068
2069                 if (IS_ERR(tmp))
2070                         return tmp;
2071                 prefer = tmp;
2072         }
2073
2074         return prefer;
2075 }
2076
2077 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2078                                                      const struct flowi *fl,
2079                                                      u16 family, u8 dir,
2080                                                      u32 if_id)
2081 {
2082         struct xfrm_pol_inexact_candidates cand;
2083         const xfrm_address_t *daddr, *saddr;
2084         struct xfrm_pol_inexact_bin *bin;
2085         struct xfrm_policy *pol, *ret;
2086         struct hlist_head *chain;
2087         unsigned int sequence;
2088         int err;
2089
2090         daddr = xfrm_flowi_daddr(fl, family);
2091         saddr = xfrm_flowi_saddr(fl, family);
2092         if (unlikely(!daddr || !saddr))
2093                 return NULL;
2094
2095         rcu_read_lock();
2096  retry:
2097         do {
2098                 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2099                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2100         } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2101
2102         ret = NULL;
2103         hlist_for_each_entry_rcu(pol, chain, bydst) {
2104                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2105                 if (err) {
2106                         if (err == -ESRCH)
2107                                 continue;
2108                         else {
2109                                 ret = ERR_PTR(err);
2110                                 goto fail;
2111                         }
2112                 } else {
2113                         ret = pol;
2114                         break;
2115                 }
2116         }
2117         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2118         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2119                                                          daddr))
2120                 goto skip_inexact;
2121
2122         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2123                                           family, dir, if_id);
2124         if (pol) {
2125                 ret = pol;
2126                 if (IS_ERR(pol))
2127                         goto fail;
2128         }
2129
2130 skip_inexact:
2131         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2132                 goto retry;
2133
2134         if (ret && !xfrm_pol_hold_rcu(ret))
2135                 goto retry;
2136 fail:
2137         rcu_read_unlock();
2138
2139         return ret;
2140 }
2141
2142 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2143                                               const struct flowi *fl,
2144                                               u16 family, u8 dir, u32 if_id)
2145 {
2146 #ifdef CONFIG_XFRM_SUB_POLICY
2147         struct xfrm_policy *pol;
2148
2149         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2150                                         dir, if_id);
2151         if (pol != NULL)
2152                 return pol;
2153 #endif
2154         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2155                                          dir, if_id);
2156 }
2157
2158 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2159                                                  const struct flowi *fl,
2160                                                  u16 family, u32 if_id)
2161 {
2162         struct xfrm_policy *pol;
2163
2164         rcu_read_lock();
2165  again:
2166         pol = rcu_dereference(sk->sk_policy[dir]);
2167         if (pol != NULL) {
2168                 bool match;
2169                 int err = 0;
2170
2171                 if (pol->family != family) {
2172                         pol = NULL;
2173                         goto out;
2174                 }
2175
2176                 match = xfrm_selector_match(&pol->selector, fl, family);
2177                 if (match) {
2178                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2179                             pol->if_id != if_id) {
2180                                 pol = NULL;
2181                                 goto out;
2182                         }
2183                         err = security_xfrm_policy_lookup(pol->security,
2184                                                       fl->flowi_secid,
2185                                                       dir);
2186                         if (!err) {
2187                                 if (!xfrm_pol_hold_rcu(pol))
2188                                         goto again;
2189                         } else if (err == -ESRCH) {
2190                                 pol = NULL;
2191                         } else {
2192                                 pol = ERR_PTR(err);
2193                         }
2194                 } else
2195                         pol = NULL;
2196         }
2197 out:
2198         rcu_read_unlock();
2199         return pol;
2200 }
2201
2202 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2203 {
2204         struct net *net = xp_net(pol);
2205
2206         list_add(&pol->walk.all, &net->xfrm.policy_all);
2207         net->xfrm.policy_count[dir]++;
2208         xfrm_pol_hold(pol);
2209 }
2210
2211 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2212                                                 int dir)
2213 {
2214         struct net *net = xp_net(pol);
2215
2216         if (list_empty(&pol->walk.all))
2217                 return NULL;
2218
2219         /* Socket policies are not hashed. */
2220         if (!hlist_unhashed(&pol->bydst)) {
2221                 hlist_del_rcu(&pol->bydst);
2222                 hlist_del_init(&pol->bydst_inexact_list);
2223                 hlist_del(&pol->byidx);
2224         }
2225
2226         list_del_init(&pol->walk.all);
2227         net->xfrm.policy_count[dir]--;
2228
2229         return pol;
2230 }
2231
2232 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2233 {
2234         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2235 }
2236
2237 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2238 {
2239         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2240 }
2241
2242 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2243 {
2244         struct net *net = xp_net(pol);
2245
2246         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2247         pol = __xfrm_policy_unlink(pol, dir);
2248         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2249         if (pol) {
2250                 xfrm_policy_kill(pol);
2251                 return 0;
2252         }
2253         return -ENOENT;
2254 }
2255 EXPORT_SYMBOL(xfrm_policy_delete);
2256
2257 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2258 {
2259         struct net *net = sock_net(sk);
2260         struct xfrm_policy *old_pol;
2261
2262 #ifdef CONFIG_XFRM_SUB_POLICY
2263         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2264                 return -EINVAL;
2265 #endif
2266
2267         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2268         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2269                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2270         if (pol) {
2271                 pol->curlft.add_time = ktime_get_real_seconds();
2272                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2273                 xfrm_sk_policy_link(pol, dir);
2274         }
2275         rcu_assign_pointer(sk->sk_policy[dir], pol);
2276         if (old_pol) {
2277                 if (pol)
2278                         xfrm_policy_requeue(old_pol, pol);
2279
2280                 /* Unlinking succeeds always. This is the only function
2281                  * allowed to delete or replace socket policy.
2282                  */
2283                 xfrm_sk_policy_unlink(old_pol, dir);
2284         }
2285         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2286
2287         if (old_pol) {
2288                 xfrm_policy_kill(old_pol);
2289         }
2290         return 0;
2291 }
2292
2293 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2294 {
2295         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2296         struct net *net = xp_net(old);
2297
2298         if (newp) {
2299                 newp->selector = old->selector;
2300                 if (security_xfrm_policy_clone(old->security,
2301                                                &newp->security)) {
2302                         kfree(newp);
2303                         return NULL;  /* ENOMEM */
2304                 }
2305                 newp->lft = old->lft;
2306                 newp->curlft = old->curlft;
2307                 newp->mark = old->mark;
2308                 newp->if_id = old->if_id;
2309                 newp->action = old->action;
2310                 newp->flags = old->flags;
2311                 newp->xfrm_nr = old->xfrm_nr;
2312                 newp->index = old->index;
2313                 newp->type = old->type;
2314                 newp->family = old->family;
2315                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2316                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2317                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2318                 xfrm_sk_policy_link(newp, dir);
2319                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2320                 xfrm_pol_put(newp);
2321         }
2322         return newp;
2323 }
2324
2325 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2326 {
2327         const struct xfrm_policy *p;
2328         struct xfrm_policy *np;
2329         int i, ret = 0;
2330
2331         rcu_read_lock();
2332         for (i = 0; i < 2; i++) {
2333                 p = rcu_dereference(osk->sk_policy[i]);
2334                 if (p) {
2335                         np = clone_policy(p, i);
2336                         if (unlikely(!np)) {
2337                                 ret = -ENOMEM;
2338                                 break;
2339                         }
2340                         rcu_assign_pointer(sk->sk_policy[i], np);
2341                 }
2342         }
2343         rcu_read_unlock();
2344         return ret;
2345 }
2346
2347 static int
2348 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2349                xfrm_address_t *remote, unsigned short family, u32 mark)
2350 {
2351         int err;
2352         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2353
2354         if (unlikely(afinfo == NULL))
2355                 return -EINVAL;
2356         err = afinfo->get_saddr(net, oif, local, remote, mark);
2357         rcu_read_unlock();
2358         return err;
2359 }
2360
2361 /* Resolve list of templates for the flow, given policy. */
2362
2363 static int
2364 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2365                       struct xfrm_state **xfrm, unsigned short family)
2366 {
2367         struct net *net = xp_net(policy);
2368         int nx;
2369         int i, error;
2370         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2371         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2372         xfrm_address_t tmp;
2373
2374         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2375                 struct xfrm_state *x;
2376                 xfrm_address_t *remote = daddr;
2377                 xfrm_address_t *local  = saddr;
2378                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2379
2380                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2381                     tmpl->mode == XFRM_MODE_BEET) {
2382                         remote = &tmpl->id.daddr;
2383                         local = &tmpl->saddr;
2384                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2385                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2386                                                        &tmp, remote,
2387                                                        tmpl->encap_family, 0);
2388                                 if (error)
2389                                         goto fail;
2390                                 local = &tmp;
2391                         }
2392                 }
2393
2394                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2395                                     family, policy->if_id);
2396
2397                 if (x && x->km.state == XFRM_STATE_VALID) {
2398                         xfrm[nx++] = x;
2399                         daddr = remote;
2400                         saddr = local;
2401                         continue;
2402                 }
2403                 if (x) {
2404                         error = (x->km.state == XFRM_STATE_ERROR ?
2405                                  -EINVAL : -EAGAIN);
2406                         xfrm_state_put(x);
2407                 } else if (error == -ESRCH) {
2408                         error = -EAGAIN;
2409                 }
2410
2411                 if (!tmpl->optional)
2412                         goto fail;
2413         }
2414         return nx;
2415
2416 fail:
2417         for (nx--; nx >= 0; nx--)
2418                 xfrm_state_put(xfrm[nx]);
2419         return error;
2420 }
2421
2422 static int
2423 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2424                   struct xfrm_state **xfrm, unsigned short family)
2425 {
2426         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2427         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2428         int cnx = 0;
2429         int error;
2430         int ret;
2431         int i;
2432
2433         for (i = 0; i < npols; i++) {
2434                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2435                         error = -ENOBUFS;
2436                         goto fail;
2437                 }
2438
2439                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2440                 if (ret < 0) {
2441                         error = ret;
2442                         goto fail;
2443                 } else
2444                         cnx += ret;
2445         }
2446
2447         /* found states are sorted for outbound processing */
2448         if (npols > 1)
2449                 xfrm_state_sort(xfrm, tpp, cnx, family);
2450
2451         return cnx;
2452
2453  fail:
2454         for (cnx--; cnx >= 0; cnx--)
2455                 xfrm_state_put(tpp[cnx]);
2456         return error;
2457
2458 }
2459
2460 static int xfrm_get_tos(const struct flowi *fl, int family)
2461 {
2462         if (family == AF_INET)
2463                 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2464
2465         return 0;
2466 }
2467
2468 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2469 {
2470         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2471         struct dst_ops *dst_ops;
2472         struct xfrm_dst *xdst;
2473
2474         if (!afinfo)
2475                 return ERR_PTR(-EINVAL);
2476
2477         switch (family) {
2478         case AF_INET:
2479                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2480                 break;
2481 #if IS_ENABLED(CONFIG_IPV6)
2482         case AF_INET6:
2483                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2484                 break;
2485 #endif
2486         default:
2487                 BUG();
2488         }
2489         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2490
2491         if (likely(xdst)) {
2492                 struct dst_entry *dst = &xdst->u.dst;
2493
2494                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2495         } else
2496                 xdst = ERR_PTR(-ENOBUFS);
2497
2498         rcu_read_unlock();
2499
2500         return xdst;
2501 }
2502
2503 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2504                            int nfheader_len)
2505 {
2506         if (dst->ops->family == AF_INET6) {
2507                 struct rt6_info *rt = (struct rt6_info *)dst;
2508                 path->path_cookie = rt6_get_cookie(rt);
2509                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2510         }
2511 }
2512
2513 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2514                                 const struct flowi *fl)
2515 {
2516         const struct xfrm_policy_afinfo *afinfo =
2517                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2518         int err;
2519
2520         if (!afinfo)
2521                 return -EINVAL;
2522
2523         err = afinfo->fill_dst(xdst, dev, fl);
2524
2525         rcu_read_unlock();
2526
2527         return err;
2528 }
2529
2530
2531 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2532  * all the metrics... Shortly, bundle a bundle.
2533  */
2534
2535 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2536                                             struct xfrm_state **xfrm,
2537                                             struct xfrm_dst **bundle,
2538                                             int nx,
2539                                             const struct flowi *fl,
2540                                             struct dst_entry *dst)
2541 {
2542         const struct xfrm_state_afinfo *afinfo;
2543         const struct xfrm_mode *inner_mode;
2544         struct net *net = xp_net(policy);
2545         unsigned long now = jiffies;
2546         struct net_device *dev;
2547         struct xfrm_dst *xdst_prev = NULL;
2548         struct xfrm_dst *xdst0 = NULL;
2549         int i = 0;
2550         int err;
2551         int header_len = 0;
2552         int nfheader_len = 0;
2553         int trailer_len = 0;
2554         int tos;
2555         int family = policy->selector.family;
2556         xfrm_address_t saddr, daddr;
2557
2558         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2559
2560         tos = xfrm_get_tos(fl, family);
2561
2562         dst_hold(dst);
2563
2564         for (; i < nx; i++) {
2565                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2566                 struct dst_entry *dst1 = &xdst->u.dst;
2567
2568                 err = PTR_ERR(xdst);
2569                 if (IS_ERR(xdst)) {
2570                         dst_release(dst);
2571                         goto put_states;
2572                 }
2573
2574                 bundle[i] = xdst;
2575                 if (!xdst_prev)
2576                         xdst0 = xdst;
2577                 else
2578                         /* Ref count is taken during xfrm_alloc_dst()
2579                          * No need to do dst_clone() on dst1
2580                          */
2581                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2582
2583                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2584                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2585                                                         xfrm_af2proto(family));
2586                         if (!inner_mode) {
2587                                 err = -EAFNOSUPPORT;
2588                                 dst_release(dst);
2589                                 goto put_states;
2590                         }
2591                 } else
2592                         inner_mode = &xfrm[i]->inner_mode;
2593
2594                 xdst->route = dst;
2595                 dst_copy_metrics(dst1, dst);
2596
2597                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2598                         __u32 mark = 0;
2599
2600                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2601                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2602
2603                         family = xfrm[i]->props.family;
2604                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2605                                               &saddr, &daddr, family, mark);
2606                         err = PTR_ERR(dst);
2607                         if (IS_ERR(dst))
2608                                 goto put_states;
2609                 } else
2610                         dst_hold(dst);
2611
2612                 dst1->xfrm = xfrm[i];
2613                 xdst->xfrm_genid = xfrm[i]->genid;
2614
2615                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2616                 dst1->flags |= DST_HOST;
2617                 dst1->lastuse = now;
2618
2619                 dst1->input = dst_discard;
2620
2621                 rcu_read_lock();
2622                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2623                 if (likely(afinfo))
2624                         dst1->output = afinfo->output;
2625                 else
2626                         dst1->output = dst_discard_out;
2627                 rcu_read_unlock();
2628
2629                 xdst_prev = xdst;
2630
2631                 header_len += xfrm[i]->props.header_len;
2632                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2633                         nfheader_len += xfrm[i]->props.header_len;
2634                 trailer_len += xfrm[i]->props.trailer_len;
2635         }
2636
2637         xfrm_dst_set_child(xdst_prev, dst);
2638         xdst0->path = dst;
2639
2640         err = -ENODEV;
2641         dev = dst->dev;
2642         if (!dev)
2643                 goto free_dst;
2644
2645         xfrm_init_path(xdst0, dst, nfheader_len);
2646         xfrm_init_pmtu(bundle, nx);
2647
2648         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2649              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2650                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2651                 if (err)
2652                         goto free_dst;
2653
2654                 xdst_prev->u.dst.header_len = header_len;
2655                 xdst_prev->u.dst.trailer_len = trailer_len;
2656                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2657                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2658         }
2659
2660         return &xdst0->u.dst;
2661
2662 put_states:
2663         for (; i < nx; i++)
2664                 xfrm_state_put(xfrm[i]);
2665 free_dst:
2666         if (xdst0)
2667                 dst_release_immediate(&xdst0->u.dst);
2668
2669         return ERR_PTR(err);
2670 }
2671
2672 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2673                                 struct xfrm_policy **pols,
2674                                 int *num_pols, int *num_xfrms)
2675 {
2676         int i;
2677
2678         if (*num_pols == 0 || !pols[0]) {
2679                 *num_pols = 0;
2680                 *num_xfrms = 0;
2681                 return 0;
2682         }
2683         if (IS_ERR(pols[0]))
2684                 return PTR_ERR(pols[0]);
2685
2686         *num_xfrms = pols[0]->xfrm_nr;
2687
2688 #ifdef CONFIG_XFRM_SUB_POLICY
2689         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2690             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2691                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2692                                                     XFRM_POLICY_TYPE_MAIN,
2693                                                     fl, family,
2694                                                     XFRM_POLICY_OUT,
2695                                                     pols[0]->if_id);
2696                 if (pols[1]) {
2697                         if (IS_ERR(pols[1])) {
2698                                 xfrm_pols_put(pols, *num_pols);
2699                                 return PTR_ERR(pols[1]);
2700                         }
2701                         (*num_pols)++;
2702                         (*num_xfrms) += pols[1]->xfrm_nr;
2703                 }
2704         }
2705 #endif
2706         for (i = 0; i < *num_pols; i++) {
2707                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2708                         *num_xfrms = -1;
2709                         break;
2710                 }
2711         }
2712
2713         return 0;
2714
2715 }
2716
2717 static struct xfrm_dst *
2718 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2719                                const struct flowi *fl, u16 family,
2720                                struct dst_entry *dst_orig)
2721 {
2722         struct net *net = xp_net(pols[0]);
2723         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2724         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2725         struct xfrm_dst *xdst;
2726         struct dst_entry *dst;
2727         int err;
2728
2729         /* Try to instantiate a bundle */
2730         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2731         if (err <= 0) {
2732                 if (err == 0)
2733                         return NULL;
2734
2735                 if (err != -EAGAIN)
2736                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2737                 return ERR_PTR(err);
2738         }
2739
2740         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2741         if (IS_ERR(dst)) {
2742                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2743                 return ERR_CAST(dst);
2744         }
2745
2746         xdst = (struct xfrm_dst *)dst;
2747         xdst->num_xfrms = err;
2748         xdst->num_pols = num_pols;
2749         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2750         xdst->policy_genid = atomic_read(&pols[0]->genid);
2751
2752         return xdst;
2753 }
2754
2755 static void xfrm_policy_queue_process(struct timer_list *t)
2756 {
2757         struct sk_buff *skb;
2758         struct sock *sk;
2759         struct dst_entry *dst;
2760         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2761         struct net *net = xp_net(pol);
2762         struct xfrm_policy_queue *pq = &pol->polq;
2763         struct flowi fl;
2764         struct sk_buff_head list;
2765
2766         spin_lock(&pq->hold_queue.lock);
2767         skb = skb_peek(&pq->hold_queue);
2768         if (!skb) {
2769                 spin_unlock(&pq->hold_queue.lock);
2770                 goto out;
2771         }
2772         dst = skb_dst(skb);
2773         sk = skb->sk;
2774         xfrm_decode_session(skb, &fl, dst->ops->family);
2775         spin_unlock(&pq->hold_queue.lock);
2776
2777         dst_hold(xfrm_dst_path(dst));
2778         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2779         if (IS_ERR(dst))
2780                 goto purge_queue;
2781
2782         if (dst->flags & DST_XFRM_QUEUE) {
2783                 dst_release(dst);
2784
2785                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2786                         goto purge_queue;
2787
2788                 pq->timeout = pq->timeout << 1;
2789                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2790                         xfrm_pol_hold(pol);
2791                 goto out;
2792         }
2793
2794         dst_release(dst);
2795
2796         __skb_queue_head_init(&list);
2797
2798         spin_lock(&pq->hold_queue.lock);
2799         pq->timeout = 0;
2800         skb_queue_splice_init(&pq->hold_queue, &list);
2801         spin_unlock(&pq->hold_queue.lock);
2802
2803         while (!skb_queue_empty(&list)) {
2804                 skb = __skb_dequeue(&list);
2805
2806                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
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_HOST | 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_INET_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