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/inet_ecn.h>
  35 #include <net/xfrm.h>
  36 #include <net/ip.h>
  37 #include <net/gre.h>
  38 #if IS_ENABLED(CONFIG_IPV6_MIP6)
  39 #include <net/mip6.h>
  40 #endif
  41 #ifdef CONFIG_XFRM_STATISTICS
  42 #include <net/snmp.h>
  43 #endif
  44 #ifdef CONFIG_XFRM_ESPINTCP
  45 #include <net/espintcp.h>
  46 #endif
  47
  48 #include "xfrm_hash.h"
  49
  50 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  51 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  52 #define XFRM_MAX_QUEUE_LEN      100
  53
  54 struct xfrm_flo {
  55         struct dst_entry *dst_orig;
  56         u8 flags;
  57 };
  58
  59 /* prefixes smaller than this are stored in lists, not trees. */
  60 #define INEXACT_PREFIXLEN_IPV4  16
  61 #define INEXACT_PREFIXLEN_IPV6  48
  62
  63 struct xfrm_pol_inexact_node {
  64         struct rb_node node;
  65         union {
  66                 xfrm_address_t addr;
  67                 struct rcu_head rcu;
  68         };
  69         u8 prefixlen;
  70
  71         struct rb_root root;
  72
  73         /* the policies matching this node, can be empty list */
  74         struct hlist_head hhead;
  75 };
  76
  77 /* xfrm inexact policy search tree:
  78  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
  79  *  |
  80  * +---- root_d: sorted by daddr:prefix
  81  * |                 |
  82  * |        xfrm_pol_inexact_node
  83  * |                 |
  84  * |                 +- root: sorted by saddr/prefix
  85  * |                 |              |
  86  * |                 |         xfrm_pol_inexact_node
  87  * |                 |              |
  88  * |                 |              + root: unused
  89  * |                 |              |
  90  * |                 |              + hhead: saddr:daddr policies
  91  * |                 |
  92  * |                 +- coarse policies and all any:daddr policies
  93  * |
  94  * +---- root_s: sorted by saddr:prefix
  95  * |                 |
  96  * |        xfrm_pol_inexact_node
  97  * |                 |
  98  * |                 + root: unused
  99  * |                 |
 100  * |                 + hhead: saddr:any policies
 101  * |
 102  * +---- coarse policies and all any:any policies
 103  *
 104  * Lookups return four candidate lists:
 105  * 1. any:any list from top-level xfrm_pol_inexact_bin
 106  * 2. any:daddr list from daddr tree
 107  * 3. saddr:daddr list from 2nd level daddr tree
 108  * 4. saddr:any list from saddr tree
 109  *
 110  * This result set then needs to be searched for the policy with
 111  * the lowest priority.  If two results have same prio, youngest one wins.
 112  */
 113
 114 struct xfrm_pol_inexact_key {
 115         possible_net_t net;
 116         u32 if_id;
 117         u16 family;
 118         u8 dir, type;
 119 };
 120
 121 struct xfrm_pol_inexact_bin {
 122         struct xfrm_pol_inexact_key k;
 123         struct rhash_head head;
 124         /* list containing '*:*' policies */
 125         struct hlist_head hhead;
 126
 127         seqcount_spinlock_t count;
 128         /* tree sorted by daddr/prefix */
 129         struct rb_root root_d;
 130
 131         /* tree sorted by saddr/prefix */
 132         struct rb_root root_s;
 133
 134         /* slow path below */
 135         struct list_head inexact_bins;
 136         struct rcu_head rcu;
 137 };
 138
 139 enum xfrm_pol_inexact_candidate_type {
 140         XFRM_POL_CAND_BOTH,
 141         XFRM_POL_CAND_SADDR,
 142         XFRM_POL_CAND_DADDR,
 143         XFRM_POL_CAND_ANY,
 144
 145         XFRM_POL_CAND_MAX,
 146 };
 147
 148 struct xfrm_pol_inexact_candidates {
 149         struct hlist_head *res[XFRM_POL_CAND_MAX];
 150 };
 151
 152 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
 153 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
 154
 155 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
 156 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
 157                                                 __read_mostly;
 158
 159 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
 160
 161 static struct rhashtable xfrm_policy_inexact_table;
 162 static const struct rhashtable_params xfrm_pol_inexact_params;
 163
 164 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
 165 static int stale_bundle(struct dst_entry *dst);
 166 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
 167 static void xfrm_policy_queue_process(struct timer_list *t);
 168
 169 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
 170 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 171                                                 int dir);
 172
 173 static struct xfrm_pol_inexact_bin *
 174 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
 175                            u32 if_id);
 176
 177 static struct xfrm_pol_inexact_bin *
 178 xfrm_policy_inexact_lookup_rcu(struct net *net,
 179                                u8 type, u16 family, u8 dir, u32 if_id);
 180 static struct xfrm_policy *
 181 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
 182                         bool excl);
 183 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
 184                                             struct xfrm_policy *policy);
 185
 186 static bool
 187 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
 188                                     struct xfrm_pol_inexact_bin *b,
 189                                     const xfrm_address_t *saddr,
 190                                     const xfrm_address_t *daddr);
 191
 192 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
 193 {
 194         return refcount_inc_not_zero(&policy->refcnt);
 195 }
 196
 197 static inline bool
 198 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 199 {
 200         const struct flowi4 *fl4 = &fl->u.ip4;
 201
 202         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
 203                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
 204                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
 205                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
 206                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
 207                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
 208 }
 209
 210 static inline bool
 211 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 212 {
 213         const struct flowi6 *fl6 = &fl->u.ip6;
 214
 215         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
 216                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
 217                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
 218                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
 219                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
 220                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
 221 }
 222
 223 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
 224                          unsigned short family)
 225 {
 226         switch (family) {
 227         case AF_INET:
 228                 return __xfrm4_selector_match(sel, fl);
 229         case AF_INET6:
 230                 return __xfrm6_selector_match(sel, fl);
 231         }
 232         return false;
 233 }
 234
 235 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 236 {
 237         const struct xfrm_policy_afinfo *afinfo;
 238
 239         if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 240                 return NULL;
 241         rcu_read_lock();
 242         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 243         if (unlikely(!afinfo))
 244                 rcu_read_unlock();
 245         return afinfo;
 246 }
 247
 248 /* Called with rcu_read_lock(). */
 249 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
 250 {
 251         return rcu_dereference(xfrm_if_cb);
 252 }
 253
 254 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
 255                                     const xfrm_address_t *saddr,
 256                                     const xfrm_address_t *daddr,
 257                                     int family, u32 mark)
 258 {
 259         const struct xfrm_policy_afinfo *afinfo;
 260         struct dst_entry *dst;
 261
 262         afinfo = xfrm_policy_get_afinfo(family);
 263         if (unlikely(afinfo == NULL))
 264                 return ERR_PTR(-EAFNOSUPPORT);
 265
 266         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
 267
 268         rcu_read_unlock();
 269
 270         return dst;
 271 }
 272 EXPORT_SYMBOL(__xfrm_dst_lookup);
 273
 274 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 275                                                 int tos, int oif,
 276                                                 xfrm_address_t *prev_saddr,
 277                                                 xfrm_address_t *prev_daddr,
 278                                                 int family, u32 mark)
 279 {
 280         struct net *net = xs_net(x);
 281         xfrm_address_t *saddr = &x->props.saddr;
 282         xfrm_address_t *daddr = &x->id.daddr;
 283         struct dst_entry *dst;
 284
 285         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 286                 saddr = x->coaddr;
 287                 daddr = prev_daddr;
 288         }
 289         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 290                 saddr = prev_saddr;
 291                 daddr = x->coaddr;
 292         }
 293
 294         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
 295
 296         if (!IS_ERR(dst)) {
 297                 if (prev_saddr != saddr)
 298                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 299                 if (prev_daddr != daddr)
 300                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 301         }
 302
 303         return dst;
 304 }
 305
 306 static inline unsigned long make_jiffies(long secs)
 307 {
 308         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 309                 return MAX_SCHEDULE_TIMEOUT-1;
 310         else
 311                 return secs*HZ;
 312 }
 313
 314 static void xfrm_policy_timer(struct timer_list *t)
 315 {
 316         struct xfrm_policy *xp = from_timer(xp, t, timer);
 317         time64_t now = ktime_get_real_seconds();
 318         time64_t next = TIME64_MAX;
 319         int warn = 0;
 320         int dir;
 321
 322         read_lock(&xp->lock);
 323
 324         if (unlikely(xp->walk.dead))
 325                 goto out;
 326
 327         dir = xfrm_policy_id2dir(xp->index);
 328
 329         if (xp->lft.hard_add_expires_seconds) {
 330                 time64_t tmo = xp->lft.hard_add_expires_seconds +
 331                         xp->curlft.add_time - now;
 332                 if (tmo <= 0)
 333                         goto expired;
 334                 if (tmo < next)
 335                         next = tmo;
 336         }
 337         if (xp->lft.hard_use_expires_seconds) {
 338                 time64_t tmo = xp->lft.hard_use_expires_seconds +
 339                         (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
 340                 if (tmo <= 0)
 341                         goto expired;
 342                 if (tmo < next)
 343                         next = tmo;
 344         }
 345         if (xp->lft.soft_add_expires_seconds) {
 346                 time64_t tmo = xp->lft.soft_add_expires_seconds +
 347                         xp->curlft.add_time - now;
 348                 if (tmo <= 0) {
 349                         warn = 1;
 350                         tmo = XFRM_KM_TIMEOUT;
 351                 }
 352                 if (tmo < next)
 353                         next = tmo;
 354         }
 355         if (xp->lft.soft_use_expires_seconds) {
 356                 time64_t tmo = xp->lft.soft_use_expires_seconds +
 357                         (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
 358                 if (tmo <= 0) {
 359                         warn = 1;
 360                         tmo = XFRM_KM_TIMEOUT;
 361                 }
 362                 if (tmo < next)
 363                         next = tmo;
 364         }
 365
 366         if (warn)
 367                 km_policy_expired(xp, dir, 0, 0);
 368         if (next != TIME64_MAX &&
 369             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 370                 xfrm_pol_hold(xp);
 371
 372 out:
 373         read_unlock(&xp->lock);
 374         xfrm_pol_put(xp);
 375         return;
 376
 377 expired:
 378         read_unlock(&xp->lock);
 379         if (!xfrm_policy_delete(xp, dir))
 380                 km_policy_expired(xp, dir, 1, 0);
 381         xfrm_pol_put(xp);
 382 }
 383
 384 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 385  * SPD calls.
 386  */
 387
 388 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 389 {
 390         struct xfrm_policy *policy;
 391
 392         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 393
 394         if (policy) {
 395                 write_pnet(&policy->xp_net, net);
 396                 INIT_LIST_HEAD(&policy->walk.all);
 397                 INIT_HLIST_NODE(&policy->bydst_inexact_list);
 398                 INIT_HLIST_NODE(&policy->bydst);
 399                 INIT_HLIST_NODE(&policy->byidx);
 400                 rwlock_init(&policy->lock);
 401                 refcount_set(&policy->refcnt, 1);
 402                 skb_queue_head_init(&policy->polq.hold_queue);
 403                 timer_setup(&policy->timer, xfrm_policy_timer, 0);
 404                 timer_setup(&policy->polq.hold_timer,
 405                             xfrm_policy_queue_process, 0);
 406         }
 407         return policy;
 408 }
 409 EXPORT_SYMBOL(xfrm_policy_alloc);
 410
 411 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 412 {
 413         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 414
 415         security_xfrm_policy_free(policy->security);
 416         kfree(policy);
 417 }
 418
 419 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 420
 421 void xfrm_policy_destroy(struct xfrm_policy *policy)
 422 {
 423         BUG_ON(!policy->walk.dead);
 424
 425         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 426                 BUG();
 427
 428         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 429 }
 430 EXPORT_SYMBOL(xfrm_policy_destroy);
 431
 432 /* Rule must be locked. Release descendant resources, announce
 433  * entry dead. The rule must be unlinked from lists to the moment.
 434  */
 435
 436 static void xfrm_policy_kill(struct xfrm_policy *policy)
 437 {
 438         write_lock_bh(&policy->lock);
 439         policy->walk.dead = 1;
 440         write_unlock_bh(&policy->lock);
 441
 442         atomic_inc(&policy->genid);
 443
 444         if (del_timer(&policy->polq.hold_timer))
 445                 xfrm_pol_put(policy);
 446         skb_queue_purge(&policy->polq.hold_queue);
 447
 448         if (del_timer(&policy->timer))
 449                 xfrm_pol_put(policy);
 450
 451         xfrm_pol_put(policy);
 452 }
 453
 454 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 455
 456 static inline unsigned int idx_hash(struct net *net, u32 index)
 457 {
 458         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 459 }
 460
 461 /* calculate policy hash thresholds */
 462 static void __get_hash_thresh(struct net *net,
 463                               unsigned short family, int dir,
 464                               u8 *dbits, u8 *sbits)
 465 {
 466         switch (family) {
 467         case AF_INET:
 468                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 469                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 470                 break;
 471
 472         case AF_INET6:
 473                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 474                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 475                 break;
 476
 477         default:
 478                 *dbits = 0;
 479                 *sbits = 0;
 480         }
 481 }
 482
 483 static struct hlist_head *policy_hash_bysel(struct net *net,
 484                                             const struct xfrm_selector *sel,
 485                                             unsigned short family, int dir)
 486 {
 487         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 488         unsigned int hash;
 489         u8 dbits;
 490         u8 sbits;
 491
 492         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 493         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 494
 495         if (hash == hmask + 1)
 496                 return NULL;
 497
 498         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 499                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 500 }
 501
 502 static struct hlist_head *policy_hash_direct(struct net *net,
 503                                              const xfrm_address_t *daddr,
 504                                              const xfrm_address_t *saddr,
 505                                              unsigned short family, int dir)
 506 {
 507         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 508         unsigned int hash;
 509         u8 dbits;
 510         u8 sbits;
 511
 512         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 513         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 514
 515         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 516                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 517 }
 518
 519 static void xfrm_dst_hash_transfer(struct net *net,
 520                                    struct hlist_head *list,
 521                                    struct hlist_head *ndsttable,
 522                                    unsigned int nhashmask,
 523                                    int dir)
 524 {
 525         struct hlist_node *tmp, *entry0 = NULL;
 526         struct xfrm_policy *pol;
 527         unsigned int h0 = 0;
 528         u8 dbits;
 529         u8 sbits;
 530
 531 redo:
 532         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 533                 unsigned int h;
 534
 535                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 536                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 537                                 pol->family, nhashmask, dbits, sbits);
 538                 if (!entry0) {
 539                         hlist_del_rcu(&pol->bydst);
 540                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 541                         h0 = h;
 542                 } else {
 543                         if (h != h0)
 544                                 continue;
 545                         hlist_del_rcu(&pol->bydst);
 546                         hlist_add_behind_rcu(&pol->bydst, entry0);
 547                 }
 548                 entry0 = &pol->bydst;
 549         }
 550         if (!hlist_empty(list)) {
 551                 entry0 = NULL;
 552                 goto redo;
 553         }
 554 }
 555
 556 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 557                                    struct hlist_head *nidxtable,
 558                                    unsigned int nhashmask)
 559 {
 560         struct hlist_node *tmp;
 561         struct xfrm_policy *pol;
 562
 563         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 564                 unsigned int h;
 565
 566                 h = __idx_hash(pol->index, nhashmask);
 567                 hlist_add_head(&pol->byidx, nidxtable+h);
 568         }
 569 }
 570
 571 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 572 {
 573         return ((old_hmask + 1) << 1) - 1;
 574 }
 575
 576 static void xfrm_bydst_resize(struct net *net, int dir)
 577 {
 578         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 579         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 580         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 581         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 582         struct hlist_head *odst;
 583         int i;
 584
 585         if (!ndst)
 586                 return;
 587
 588         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 589         write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
 590
 591         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 592                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 593
 594         for (i = hmask; i >= 0; i--)
 595                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 596
 597         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 598         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 599
 600         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
 601         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 602
 603         synchronize_rcu();
 604
 605         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 606 }
 607
 608 static void xfrm_byidx_resize(struct net *net, int total)
 609 {
 610         unsigned int hmask = net->xfrm.policy_idx_hmask;
 611         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 612         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 613         struct hlist_head *oidx = net->xfrm.policy_byidx;
 614         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 615         int i;
 616
 617         if (!nidx)
 618                 return;
 619
 620         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 621
 622         for (i = hmask; i >= 0; i--)
 623                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 624
 625         net->xfrm.policy_byidx = nidx;
 626         net->xfrm.policy_idx_hmask = nhashmask;
 627
 628         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 629
 630         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 631 }
 632
 633 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 634 {
 635         unsigned int cnt = net->xfrm.policy_count[dir];
 636         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 637
 638         if (total)
 639                 *total += cnt;
 640
 641         if ((hmask + 1) < xfrm_policy_hashmax &&
 642             cnt > hmask)
 643                 return 1;
 644
 645         return 0;
 646 }
 647
 648 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 649 {
 650         unsigned int hmask = net->xfrm.policy_idx_hmask;
 651
 652         if ((hmask + 1) < xfrm_policy_hashmax &&
 653             total > hmask)
 654                 return 1;
 655
 656         return 0;
 657 }
 658
 659 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 660 {
 661         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 662         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 663         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 664         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 665         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 666         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 667         si->spdhcnt = net->xfrm.policy_idx_hmask;
 668         si->spdhmcnt = xfrm_policy_hashmax;
 669 }
 670 EXPORT_SYMBOL(xfrm_spd_getinfo);
 671
 672 static DEFINE_MUTEX(hash_resize_mutex);
 673 static void xfrm_hash_resize(struct work_struct *work)
 674 {
 675         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 676         int dir, total;
 677
 678         mutex_lock(&hash_resize_mutex);
 679
 680         total = 0;
 681         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 682                 if (xfrm_bydst_should_resize(net, dir, &total))
 683                         xfrm_bydst_resize(net, dir);
 684         }
 685         if (xfrm_byidx_should_resize(net, total))
 686                 xfrm_byidx_resize(net, total);
 687
 688         mutex_unlock(&hash_resize_mutex);
 689 }
 690
 691 /* Make sure *pol can be inserted into fastbin.
 692  * Useful to check that later insert requests will be successful
 693  * (provided xfrm_policy_lock is held throughout).
 694  */
 695 static struct xfrm_pol_inexact_bin *
 696 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
 697 {
 698         struct xfrm_pol_inexact_bin *bin, *prev;
 699         struct xfrm_pol_inexact_key k = {
 700                 .family = pol->family,
 701                 .type = pol->type,
 702                 .dir = dir,
 703                 .if_id = pol->if_id,
 704         };
 705         struct net *net = xp_net(pol);
 706
 707         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 708
 709         write_pnet(&k.net, net);
 710         bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
 711                                      xfrm_pol_inexact_params);
 712         if (bin)
 713                 return bin;
 714
 715         bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
 716         if (!bin)
 717                 return NULL;
 718
 719         bin->k = k;
 720         INIT_HLIST_HEAD(&bin->hhead);
 721         bin->root_d = RB_ROOT;
 722         bin->root_s = RB_ROOT;
 723         seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
 724
 725         prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
 726                                                 &bin->k, &bin->head,
 727                                                 xfrm_pol_inexact_params);
 728         if (!prev) {
 729                 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
 730                 return bin;
 731         }
 732
 733         kfree(bin);
 734
 735         return IS_ERR(prev) ? NULL : prev;
 736 }
 737
 738 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
 739                                                int family, u8 prefixlen)
 740 {
 741         if (xfrm_addr_any(addr, family))
 742                 return true;
 743
 744         if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
 745                 return true;
 746
 747         if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
 748                 return true;
 749
 750         return false;
 751 }
 752
 753 static bool
 754 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
 755 {
 756         const xfrm_address_t *addr;
 757         bool saddr_any, daddr_any;
 758         u8 prefixlen;
 759
 760         addr = &policy->selector.saddr;
 761         prefixlen = policy->selector.prefixlen_s;
 762
 763         saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 764                                                        policy->family,
 765                                                        prefixlen);
 766         addr = &policy->selector.daddr;
 767         prefixlen = policy->selector.prefixlen_d;
 768         daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 769                                                        policy->family,
 770                                                        prefixlen);
 771         return saddr_any && daddr_any;
 772 }
 773
 774 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
 775                                        const xfrm_address_t *addr, u8 prefixlen)
 776 {
 777         node->addr = *addr;
 778         node->prefixlen = prefixlen;
 779 }
 780
 781 static struct xfrm_pol_inexact_node *
 782 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
 783 {
 784         struct xfrm_pol_inexact_node *node;
 785
 786         node = kzalloc(sizeof(*node), GFP_ATOMIC);
 787         if (node)
 788                 xfrm_pol_inexact_node_init(node, addr, prefixlen);
 789
 790         return node;
 791 }
 792
 793 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
 794                                   const xfrm_address_t *b,
 795                                   u8 prefixlen, u16 family)
 796 {
 797         u32 ma, mb, mask;
 798         unsigned int pdw, pbi;
 799         int delta = 0;
 800
 801         switch (family) {
 802         case AF_INET:
 803                 if (prefixlen == 0)
 804                         return 0;
 805                 mask = ~0U << (32 - prefixlen);
 806                 ma = ntohl(a->a4) & mask;
 807                 mb = ntohl(b->a4) & mask;
 808                 if (ma < mb)
 809                         delta = -1;
 810                 else if (ma > mb)
 811                         delta = 1;
 812                 break;
 813         case AF_INET6:
 814                 pdw = prefixlen >> 5;
 815                 pbi = prefixlen & 0x1f;
 816
 817                 if (pdw) {
 818                         delta = memcmp(a->a6, b->a6, pdw << 2);
 819                         if (delta)
 820                                 return delta;
 821                 }
 822                 if (pbi) {
 823                         mask = ~0U << (32 - pbi);
 824                         ma = ntohl(a->a6[pdw]) & mask;
 825                         mb = ntohl(b->a6[pdw]) & mask;
 826                         if (ma < mb)
 827                                 delta = -1;
 828                         else if (ma > mb)
 829                                 delta = 1;
 830                 }
 831                 break;
 832         default:
 833                 break;
 834         }
 835
 836         return delta;
 837 }
 838
 839 static void xfrm_policy_inexact_list_reinsert(struct net *net,
 840                                               struct xfrm_pol_inexact_node *n,
 841                                               u16 family)
 842 {
 843         unsigned int matched_s, matched_d;
 844         struct xfrm_policy *policy, *p;
 845
 846         matched_s = 0;
 847         matched_d = 0;
 848
 849         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 850                 struct hlist_node *newpos = NULL;
 851                 bool matches_s, matches_d;
 852
 853                 if (!policy->bydst_reinsert)
 854                         continue;
 855
 856                 WARN_ON_ONCE(policy->family != family);
 857
 858                 policy->bydst_reinsert = false;
 859                 hlist_for_each_entry(p, &n->hhead, bydst) {
 860                         if (policy->priority > p->priority)
 861                                 newpos = &p->bydst;
 862                         else if (policy->priority == p->priority &&
 863                                  policy->pos > p->pos)
 864                                 newpos = &p->bydst;
 865                         else
 866                                 break;
 867                 }
 868
 869                 if (newpos)
 870                         hlist_add_behind_rcu(&policy->bydst, newpos);
 871                 else
 872                         hlist_add_head_rcu(&policy->bydst, &n->hhead);
 873
 874                 /* paranoia checks follow.
 875                  * Check that the reinserted policy matches at least
 876                  * saddr or daddr for current node prefix.
 877                  *
 878                  * Matching both is fine, matching saddr in one policy
 879                  * (but not daddr) and then matching only daddr in another
 880                  * is a bug.
 881                  */
 882                 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
 883                                                    &n->addr,
 884                                                    n->prefixlen,
 885                                                    family) == 0;
 886                 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
 887                                                    &n->addr,
 888                                                    n->prefixlen,
 889                                                    family) == 0;
 890                 if (matches_s && matches_d)
 891                         continue;
 892
 893                 WARN_ON_ONCE(!matches_s && !matches_d);
 894                 if (matches_s)
 895                         matched_s++;
 896                 if (matches_d)
 897                         matched_d++;
 898                 WARN_ON_ONCE(matched_s && matched_d);
 899         }
 900 }
 901
 902 static void xfrm_policy_inexact_node_reinsert(struct net *net,
 903                                               struct xfrm_pol_inexact_node *n,
 904                                               struct rb_root *new,
 905                                               u16 family)
 906 {
 907         struct xfrm_pol_inexact_node *node;
 908         struct rb_node **p, *parent;
 909
 910         /* we should not have another subtree here */
 911         WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
 912 restart:
 913         parent = NULL;
 914         p = &new->rb_node;
 915         while (*p) {
 916                 u8 prefixlen;
 917                 int delta;
 918
 919                 parent = *p;
 920                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 921
 922                 prefixlen = min(node->prefixlen, n->prefixlen);
 923
 924                 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
 925                                                prefixlen, family);
 926                 if (delta < 0) {
 927                         p = &parent->rb_left;
 928                 } else if (delta > 0) {
 929                         p = &parent->rb_right;
 930                 } else {
 931                         bool same_prefixlen = node->prefixlen == n->prefixlen;
 932                         struct xfrm_policy *tmp;
 933
 934                         hlist_for_each_entry(tmp, &n->hhead, bydst) {
 935                                 tmp->bydst_reinsert = true;
 936                                 hlist_del_rcu(&tmp->bydst);
 937                         }
 938
 939                         node->prefixlen = prefixlen;
 940
 941                         xfrm_policy_inexact_list_reinsert(net, node, family);
 942
 943                         if (same_prefixlen) {
 944                                 kfree_rcu(n, rcu);
 945                                 return;
 946                         }
 947
 948                         rb_erase(*p, new);
 949                         kfree_rcu(n, rcu);
 950                         n = node;
 951                         goto restart;
 952                 }
 953         }
 954
 955         rb_link_node_rcu(&n->node, parent, p);
 956         rb_insert_color(&n->node, new);
 957 }
 958
 959 /* merge nodes v and n */
 960 static void xfrm_policy_inexact_node_merge(struct net *net,
 961                                            struct xfrm_pol_inexact_node *v,
 962                                            struct xfrm_pol_inexact_node *n,
 963                                            u16 family)
 964 {
 965         struct xfrm_pol_inexact_node *node;
 966         struct xfrm_policy *tmp;
 967         struct rb_node *rnode;
 968
 969         /* To-be-merged node v has a subtree.
 970          *
 971          * Dismantle it and insert its nodes to n->root.
 972          */
 973         while ((rnode = rb_first(&v->root)) != NULL) {
 974                 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
 975                 rb_erase(&node->node, &v->root);
 976                 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
 977                                                   family);
 978         }
 979
 980         hlist_for_each_entry(tmp, &v->hhead, bydst) {
 981                 tmp->bydst_reinsert = true;
 982                 hlist_del_rcu(&tmp->bydst);
 983         }
 984
 985         xfrm_policy_inexact_list_reinsert(net, n, family);
 986 }
 987
 988 static struct xfrm_pol_inexact_node *
 989 xfrm_policy_inexact_insert_node(struct net *net,
 990                                 struct rb_root *root,
 991                                 xfrm_address_t *addr,
 992                                 u16 family, u8 prefixlen, u8 dir)
 993 {
 994         struct xfrm_pol_inexact_node *cached = NULL;
 995         struct rb_node **p, *parent = NULL;
 996         struct xfrm_pol_inexact_node *node;
 997
 998         p = &root->rb_node;
 999         while (*p) {
1000                 int delta;
1001
1002                 parent = *p;
1003                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1004
1005                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1006                                                node->prefixlen,
1007                                                family);
1008                 if (delta == 0 && prefixlen >= node->prefixlen) {
1009                         WARN_ON_ONCE(cached); /* ipsec policies got lost */
1010                         return node;
1011                 }
1012
1013                 if (delta < 0)
1014                         p = &parent->rb_left;
1015                 else
1016                         p = &parent->rb_right;
1017
1018                 if (prefixlen < node->prefixlen) {
1019                         delta = xfrm_policy_addr_delta(addr, &node->addr,
1020                                                        prefixlen,
1021                                                        family);
1022                         if (delta)
1023                                 continue;
1024
1025                         /* This node is a subnet of the new prefix. It needs
1026                          * to be removed and re-inserted with the smaller
1027                          * prefix and all nodes that are now also covered
1028                          * by the reduced prefixlen.
1029                          */
1030                         rb_erase(&node->node, root);
1031
1032                         if (!cached) {
1033                                 xfrm_pol_inexact_node_init(node, addr,
1034                                                            prefixlen);
1035                                 cached = node;
1036                         } else {
1037                                 /* This node also falls within the new
1038                                  * prefixlen. Merge the to-be-reinserted
1039                                  * node and this one.
1040                                  */
1041                                 xfrm_policy_inexact_node_merge(net, node,
1042                                                                cached, family);
1043                                 kfree_rcu(node, rcu);
1044                         }
1045
1046                         /* restart */
1047                         p = &root->rb_node;
1048                         parent = NULL;
1049                 }
1050         }
1051
1052         node = cached;
1053         if (!node) {
1054                 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1055                 if (!node)
1056                         return NULL;
1057         }
1058
1059         rb_link_node_rcu(&node->node, parent, p);
1060         rb_insert_color(&node->node, root);
1061
1062         return node;
1063 }
1064
1065 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1066 {
1067         struct xfrm_pol_inexact_node *node;
1068         struct rb_node *rn = rb_first(r);
1069
1070         while (rn) {
1071                 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1072
1073                 xfrm_policy_inexact_gc_tree(&node->root, rm);
1074                 rn = rb_next(rn);
1075
1076                 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1077                         WARN_ON_ONCE(rm);
1078                         continue;
1079                 }
1080
1081                 rb_erase(&node->node, r);
1082                 kfree_rcu(node, rcu);
1083         }
1084 }
1085
1086 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1087 {
1088         write_seqcount_begin(&b->count);
1089         xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1090         xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1091         write_seqcount_end(&b->count);
1092
1093         if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1094             !hlist_empty(&b->hhead)) {
1095                 WARN_ON_ONCE(net_exit);
1096                 return;
1097         }
1098
1099         if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1100                                    xfrm_pol_inexact_params) == 0) {
1101                 list_del(&b->inexact_bins);
1102                 kfree_rcu(b, rcu);
1103         }
1104 }
1105
1106 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1107 {
1108         struct net *net = read_pnet(&b->k.net);
1109
1110         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1111         __xfrm_policy_inexact_prune_bin(b, false);
1112         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1113 }
1114
1115 static void __xfrm_policy_inexact_flush(struct net *net)
1116 {
1117         struct xfrm_pol_inexact_bin *bin, *t;
1118
1119         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1120
1121         list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1122                 __xfrm_policy_inexact_prune_bin(bin, false);
1123 }
1124
1125 static struct hlist_head *
1126 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1127                                 struct xfrm_policy *policy, u8 dir)
1128 {
1129         struct xfrm_pol_inexact_node *n;
1130         struct net *net;
1131
1132         net = xp_net(policy);
1133         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1134
1135         if (xfrm_policy_inexact_insert_use_any_list(policy))
1136                 return &bin->hhead;
1137
1138         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1139                                                policy->family,
1140                                                policy->selector.prefixlen_d)) {
1141                 write_seqcount_begin(&bin->count);
1142                 n = xfrm_policy_inexact_insert_node(net,
1143                                                     &bin->root_s,
1144                                                     &policy->selector.saddr,
1145                                                     policy->family,
1146                                                     policy->selector.prefixlen_s,
1147                                                     dir);
1148                 write_seqcount_end(&bin->count);
1149                 if (!n)
1150                         return NULL;
1151
1152                 return &n->hhead;
1153         }
1154
1155         /* daddr is fixed */
1156         write_seqcount_begin(&bin->count);
1157         n = xfrm_policy_inexact_insert_node(net,
1158                                             &bin->root_d,
1159                                             &policy->selector.daddr,
1160                                             policy->family,
1161                                             policy->selector.prefixlen_d, dir);
1162         write_seqcount_end(&bin->count);
1163         if (!n)
1164                 return NULL;
1165
1166         /* saddr is wildcard */
1167         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1168                                                policy->family,
1169                                                policy->selector.prefixlen_s))
1170                 return &n->hhead;
1171
1172         write_seqcount_begin(&bin->count);
1173         n = xfrm_policy_inexact_insert_node(net,
1174                                             &n->root,
1175                                             &policy->selector.saddr,
1176                                             policy->family,
1177                                             policy->selector.prefixlen_s, dir);
1178         write_seqcount_end(&bin->count);
1179         if (!n)
1180                 return NULL;
1181
1182         return &n->hhead;
1183 }
1184
1185 static struct xfrm_policy *
1186 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1187 {
1188         struct xfrm_pol_inexact_bin *bin;
1189         struct xfrm_policy *delpol;
1190         struct hlist_head *chain;
1191         struct net *net;
1192
1193         bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1194         if (!bin)
1195                 return ERR_PTR(-ENOMEM);
1196
1197         net = xp_net(policy);
1198         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1199
1200         chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1201         if (!chain) {
1202                 __xfrm_policy_inexact_prune_bin(bin, false);
1203                 return ERR_PTR(-ENOMEM);
1204         }
1205
1206         delpol = xfrm_policy_insert_list(chain, policy, excl);
1207         if (delpol && excl) {
1208                 __xfrm_policy_inexact_prune_bin(bin, false);
1209                 return ERR_PTR(-EEXIST);
1210         }
1211
1212         chain = &net->xfrm.policy_inexact[dir];
1213         xfrm_policy_insert_inexact_list(chain, policy);
1214
1215         if (delpol)
1216                 __xfrm_policy_inexact_prune_bin(bin, false);
1217
1218         return delpol;
1219 }
1220
1221 static void xfrm_hash_rebuild(struct work_struct *work)
1222 {
1223         struct net *net = container_of(work, struct net,
1224                                        xfrm.policy_hthresh.work);
1225         unsigned int hmask;
1226         struct xfrm_policy *pol;
1227         struct xfrm_policy *policy;
1228         struct hlist_head *chain;
1229         struct hlist_head *odst;
1230         struct hlist_node *newpos;
1231         int i;
1232         int dir;
1233         unsigned seq;
1234         u8 lbits4, rbits4, lbits6, rbits6;
1235
1236         mutex_lock(&hash_resize_mutex);
1237
1238         /* read selector prefixlen thresholds */
1239         do {
1240                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1241
1242                 lbits4 = net->xfrm.policy_hthresh.lbits4;
1243                 rbits4 = net->xfrm.policy_hthresh.rbits4;
1244                 lbits6 = net->xfrm.policy_hthresh.lbits6;
1245                 rbits6 = net->xfrm.policy_hthresh.rbits6;
1246         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1247
1248         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1249         write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1250
1251         /* make sure that we can insert the indirect policies again before
1252          * we start with destructive action.
1253          */
1254         list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1255                 struct xfrm_pol_inexact_bin *bin;
1256                 u8 dbits, sbits;
1257
1258                 dir = xfrm_policy_id2dir(policy->index);
1259                 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1260                         continue;
1261
1262                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1263                         if (policy->family == AF_INET) {
1264                                 dbits = rbits4;
1265                                 sbits = lbits4;
1266                         } else {
1267                                 dbits = rbits6;
1268                                 sbits = lbits6;
1269                         }
1270                 } else {
1271                         if (policy->family == AF_INET) {
1272                                 dbits = lbits4;
1273                                 sbits = rbits4;
1274                         } else {
1275                                 dbits = lbits6;
1276                                 sbits = rbits6;
1277                         }
1278                 }
1279
1280                 if (policy->selector.prefixlen_d < dbits ||
1281                     policy->selector.prefixlen_s < sbits)
1282                         continue;
1283
1284                 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1285                 if (!bin)
1286                         goto out_unlock;
1287
1288                 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1289                         goto out_unlock;
1290         }
1291
1292         /* reset the bydst and inexact table in all directions */
1293         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1294                 struct hlist_node *n;
1295
1296                 hlist_for_each_entry_safe(policy, n,
1297                                           &net->xfrm.policy_inexact[dir],
1298                                           bydst_inexact_list) {
1299                         hlist_del_rcu(&policy->bydst);
1300                         hlist_del_init(&policy->bydst_inexact_list);
1301                 }
1302
1303                 hmask = net->xfrm.policy_bydst[dir].hmask;
1304                 odst = net->xfrm.policy_bydst[dir].table;
1305                 for (i = hmask; i >= 0; i--) {
1306                         hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1307                                 hlist_del_rcu(&policy->bydst);
1308                 }
1309                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1310                         /* dir out => dst = remote, src = local */
1311                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1312                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1313                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1314                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1315                 } else {
1316                         /* dir in/fwd => dst = local, src = remote */
1317                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1318                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1319                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1320                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1321                 }
1322         }
1323
1324         /* re-insert all policies by order of creation */
1325         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1326                 if (policy->walk.dead)
1327                         continue;
1328                 dir = xfrm_policy_id2dir(policy->index);
1329                 if (dir >= XFRM_POLICY_MAX) {
1330                         /* skip socket policies */
1331                         continue;
1332                 }
1333                 newpos = NULL;
1334                 chain = policy_hash_bysel(net, &policy->selector,
1335                                           policy->family, dir);
1336
1337                 if (!chain) {
1338                         void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1339
1340                         WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1341                         continue;
1342                 }
1343
1344                 hlist_for_each_entry(pol, chain, bydst) {
1345                         if (policy->priority >= pol->priority)
1346                                 newpos = &pol->bydst;
1347                         else
1348                                 break;
1349                 }
1350                 if (newpos)
1351                         hlist_add_behind_rcu(&policy->bydst, newpos);
1352                 else
1353                         hlist_add_head_rcu(&policy->bydst, chain);
1354         }
1355
1356 out_unlock:
1357         __xfrm_policy_inexact_flush(net);
1358         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1359         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1360
1361         mutex_unlock(&hash_resize_mutex);
1362 }
1363
1364 void xfrm_policy_hash_rebuild(struct net *net)
1365 {
1366         schedule_work(&net->xfrm.policy_hthresh.work);
1367 }
1368 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1369
1370 /* Generate new index... KAME seems to generate them ordered by cost
1371  * of an absolute inpredictability of ordering of rules. This will not pass. */
1372 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1373 {
1374         static u32 idx_generator;
1375
1376         for (;;) {
1377                 struct hlist_head *list;
1378                 struct xfrm_policy *p;
1379                 u32 idx;
1380                 int found;
1381
1382                 if (!index) {
1383                         idx = (idx_generator | dir);
1384                         idx_generator += 8;
1385                 } else {
1386                         idx = index;
1387                         index = 0;
1388                 }
1389
1390                 if (idx == 0)
1391                         idx = 8;
1392                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1393                 found = 0;
1394                 hlist_for_each_entry(p, list, byidx) {
1395                         if (p->index == idx) {
1396                                 found = 1;
1397                                 break;
1398                         }
1399                 }
1400                 if (!found)
1401                         return idx;
1402         }
1403 }
1404
1405 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1406 {
1407         u32 *p1 = (u32 *) s1;
1408         u32 *p2 = (u32 *) s2;
1409         int len = sizeof(struct xfrm_selector) / sizeof(u32);
1410         int i;
1411
1412         for (i = 0; i < len; i++) {
1413                 if (p1[i] != p2[i])
1414                         return 1;
1415         }
1416
1417         return 0;
1418 }
1419
1420 static void xfrm_policy_requeue(struct xfrm_policy *old,
1421                                 struct xfrm_policy *new)
1422 {
1423         struct xfrm_policy_queue *pq = &old->polq;
1424         struct sk_buff_head list;
1425
1426         if (skb_queue_empty(&pq->hold_queue))
1427                 return;
1428
1429         __skb_queue_head_init(&list);
1430
1431         spin_lock_bh(&pq->hold_queue.lock);
1432         skb_queue_splice_init(&pq->hold_queue, &list);
1433         if (del_timer(&pq->hold_timer))
1434                 xfrm_pol_put(old);
1435         spin_unlock_bh(&pq->hold_queue.lock);
1436
1437         pq = &new->polq;
1438
1439         spin_lock_bh(&pq->hold_queue.lock);
1440         skb_queue_splice(&list, &pq->hold_queue);
1441         pq->timeout = XFRM_QUEUE_TMO_MIN;
1442         if (!mod_timer(&pq->hold_timer, jiffies))
1443                 xfrm_pol_hold(new);
1444         spin_unlock_bh(&pq->hold_queue.lock);
1445 }
1446
1447 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1448                                           struct xfrm_policy *pol)
1449 {
1450         return mark->v == pol->mark.v && mark->m == pol->mark.m;
1451 }
1452
1453 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1454 {
1455         const struct xfrm_pol_inexact_key *k = data;
1456         u32 a = k->type << 24 | k->dir << 16 | k->family;
1457
1458         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1459                             seed);
1460 }
1461
1462 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1463 {
1464         const struct xfrm_pol_inexact_bin *b = data;
1465
1466         return xfrm_pol_bin_key(&b->k, 0, seed);
1467 }
1468
1469 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1470                             const void *ptr)
1471 {
1472         const struct xfrm_pol_inexact_key *key = arg->key;
1473         const struct xfrm_pol_inexact_bin *b = ptr;
1474         int ret;
1475
1476         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1477                 return -1;
1478
1479         ret = b->k.dir ^ key->dir;
1480         if (ret)
1481                 return ret;
1482
1483         ret = b->k.type ^ key->type;
1484         if (ret)
1485                 return ret;
1486
1487         ret = b->k.family ^ key->family;
1488         if (ret)
1489                 return ret;
1490
1491         return b->k.if_id ^ key->if_id;
1492 }
1493
1494 static const struct rhashtable_params xfrm_pol_inexact_params = {
1495         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1496         .hashfn                 = xfrm_pol_bin_key,
1497         .obj_hashfn             = xfrm_pol_bin_obj,
1498         .obj_cmpfn              = xfrm_pol_bin_cmp,
1499         .automatic_shrinking    = true,
1500 };
1501
1502 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1503                                             struct xfrm_policy *policy)
1504 {
1505         struct xfrm_policy *pol, *delpol = NULL;
1506         struct hlist_node *newpos = NULL;
1507         int i = 0;
1508
1509         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1510                 if (pol->type == policy->type &&
1511                     pol->if_id == policy->if_id &&
1512                     !selector_cmp(&pol->selector, &policy->selector) &&
1513                     xfrm_policy_mark_match(&policy->mark, pol) &&
1514                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1515                     !WARN_ON(delpol)) {
1516                         delpol = pol;
1517                         if (policy->priority > pol->priority)
1518                                 continue;
1519                 } else if (policy->priority >= pol->priority) {
1520                         newpos = &pol->bydst_inexact_list;
1521                         continue;
1522                 }
1523                 if (delpol)
1524                         break;
1525         }
1526
1527         if (newpos)
1528                 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1529         else
1530                 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1531
1532         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1533                 pol->pos = i;
1534                 i++;
1535         }
1536 }
1537
1538 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1539                                                    struct xfrm_policy *policy,
1540                                                    bool excl)
1541 {
1542         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1543
1544         hlist_for_each_entry(pol, chain, bydst) {
1545                 if (pol->type == policy->type &&
1546                     pol->if_id == policy->if_id &&
1547                     !selector_cmp(&pol->selector, &policy->selector) &&
1548                     xfrm_policy_mark_match(&policy->mark, pol) &&
1549                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1550                     !WARN_ON(delpol)) {
1551                         if (excl)
1552                                 return ERR_PTR(-EEXIST);
1553                         delpol = pol;
1554                         if (policy->priority > pol->priority)
1555                                 continue;
1556                 } else if (policy->priority >= pol->priority) {
1557                         newpos = pol;
1558                         continue;
1559                 }
1560                 if (delpol)
1561                         break;
1562         }
1563
1564         if (newpos)
1565                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1566         else
1567                 hlist_add_head_rcu(&policy->bydst, chain);
1568
1569         return delpol;
1570 }
1571
1572 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1573 {
1574         struct net *net = xp_net(policy);
1575         struct xfrm_policy *delpol;
1576         struct hlist_head *chain;
1577
1578         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1579         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1580         if (chain)
1581                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1582         else
1583                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1584
1585         if (IS_ERR(delpol)) {
1586                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1587                 return PTR_ERR(delpol);
1588         }
1589
1590         __xfrm_policy_link(policy, dir);
1591
1592         /* After previous checking, family can either be AF_INET or AF_INET6 */
1593         if (policy->family == AF_INET)
1594                 rt_genid_bump_ipv4(net);
1595         else
1596                 rt_genid_bump_ipv6(net);
1597
1598         if (delpol) {
1599                 xfrm_policy_requeue(delpol, policy);
1600                 __xfrm_policy_unlink(delpol, dir);
1601         }
1602         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1603         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1604         policy->curlft.add_time = ktime_get_real_seconds();
1605         policy->curlft.use_time = 0;
1606         if (!mod_timer(&policy->timer, jiffies + HZ))
1607                 xfrm_pol_hold(policy);
1608         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1609
1610         if (delpol)
1611                 xfrm_policy_kill(delpol);
1612         else if (xfrm_bydst_should_resize(net, dir, NULL))
1613                 schedule_work(&net->xfrm.policy_hash_work);
1614
1615         return 0;
1616 }
1617 EXPORT_SYMBOL(xfrm_policy_insert);
1618
1619 static struct xfrm_policy *
1620 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1621                         u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1622                         struct xfrm_sec_ctx *ctx)
1623 {
1624         struct xfrm_policy *pol;
1625
1626         if (!chain)
1627                 return NULL;
1628
1629         hlist_for_each_entry(pol, chain, bydst) {
1630                 if (pol->type == type &&
1631                     pol->if_id == if_id &&
1632                     xfrm_policy_mark_match(mark, pol) &&
1633                     !selector_cmp(sel, &pol->selector) &&
1634                     xfrm_sec_ctx_match(ctx, pol->security))
1635                         return pol;
1636         }
1637
1638         return NULL;
1639 }
1640
1641 struct xfrm_policy *
1642 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1643                       u8 type, int dir, struct xfrm_selector *sel,
1644                       struct xfrm_sec_ctx *ctx, int delete, int *err)
1645 {
1646         struct xfrm_pol_inexact_bin *bin = NULL;
1647         struct xfrm_policy *pol, *ret = NULL;
1648         struct hlist_head *chain;
1649
1650         *err = 0;
1651         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1652         chain = policy_hash_bysel(net, sel, sel->family, dir);
1653         if (!chain) {
1654                 struct xfrm_pol_inexact_candidates cand;
1655                 int i;
1656
1657                 bin = xfrm_policy_inexact_lookup(net, type,
1658                                                  sel->family, dir, if_id);
1659                 if (!bin) {
1660                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1661                         return NULL;
1662                 }
1663
1664                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1665                                                          &sel->saddr,
1666                                                          &sel->daddr)) {
1667                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1668                         return NULL;
1669                 }
1670
1671                 pol = NULL;
1672                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1673                         struct xfrm_policy *tmp;
1674
1675                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1676                                                       if_id, type, dir,
1677                                                       sel, ctx);
1678                         if (!tmp)
1679                                 continue;
1680
1681                         if (!pol || tmp->pos < pol->pos)
1682                                 pol = tmp;
1683                 }
1684         } else {
1685                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1686                                               sel, ctx);
1687         }
1688
1689         if (pol) {
1690                 xfrm_pol_hold(pol);
1691                 if (delete) {
1692                         *err = security_xfrm_policy_delete(pol->security);
1693                         if (*err) {
1694                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1695                                 return pol;
1696                         }
1697                         __xfrm_policy_unlink(pol, dir);
1698                 }
1699                 ret = pol;
1700         }
1701         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1702
1703         if (ret && delete)
1704                 xfrm_policy_kill(ret);
1705         if (bin && delete)
1706                 xfrm_policy_inexact_prune_bin(bin);
1707         return ret;
1708 }
1709 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1710
1711 struct xfrm_policy *
1712 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1713                  u8 type, int dir, u32 id, int delete, int *err)
1714 {
1715         struct xfrm_policy *pol, *ret;
1716         struct hlist_head *chain;
1717
1718         *err = -ENOENT;
1719         if (xfrm_policy_id2dir(id) != dir)
1720                 return NULL;
1721
1722         *err = 0;
1723         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1724         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1725         ret = NULL;
1726         hlist_for_each_entry(pol, chain, byidx) {
1727                 if (pol->type == type && pol->index == id &&
1728                     pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1729                         xfrm_pol_hold(pol);
1730                         if (delete) {
1731                                 *err = security_xfrm_policy_delete(
1732                                                                 pol->security);
1733                                 if (*err) {
1734                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1735                                         return pol;
1736                                 }
1737                                 __xfrm_policy_unlink(pol, dir);
1738                         }
1739                         ret = pol;
1740                         break;
1741                 }
1742         }
1743         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1744
1745         if (ret && delete)
1746                 xfrm_policy_kill(ret);
1747         return ret;
1748 }
1749 EXPORT_SYMBOL(xfrm_policy_byid);
1750
1751 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1752 static inline int
1753 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1754 {
1755         struct xfrm_policy *pol;
1756         int err = 0;
1757
1758         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1759                 if (pol->walk.dead ||
1760                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1761                     pol->type != type)
1762                         continue;
1763
1764                 err = security_xfrm_policy_delete(pol->security);
1765                 if (err) {
1766                         xfrm_audit_policy_delete(pol, 0, task_valid);
1767                         return err;
1768                 }
1769         }
1770         return err;
1771 }
1772 #else
1773 static inline int
1774 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1775 {
1776         return 0;
1777 }
1778 #endif
1779
1780 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1781 {
1782         int dir, err = 0, cnt = 0;
1783         struct xfrm_policy *pol;
1784
1785         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1786
1787         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1788         if (err)
1789                 goto out;
1790
1791 again:
1792         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1793                 dir = xfrm_policy_id2dir(pol->index);
1794                 if (pol->walk.dead ||
1795                     dir >= XFRM_POLICY_MAX ||
1796                     pol->type != type)
1797                         continue;
1798
1799                 __xfrm_policy_unlink(pol, dir);
1800                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1801                 cnt++;
1802                 xfrm_audit_policy_delete(pol, 1, task_valid);
1803                 xfrm_policy_kill(pol);
1804                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1805                 goto again;
1806         }
1807         if (cnt)
1808                 __xfrm_policy_inexact_flush(net);
1809         else
1810                 err = -ESRCH;
1811 out:
1812         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1813         return err;
1814 }
1815 EXPORT_SYMBOL(xfrm_policy_flush);
1816
1817 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1818                      int (*func)(struct xfrm_policy *, int, int, void*),
1819                      void *data)
1820 {
1821         struct xfrm_policy *pol;
1822         struct xfrm_policy_walk_entry *x;
1823         int error = 0;
1824
1825         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1826             walk->type != XFRM_POLICY_TYPE_ANY)
1827                 return -EINVAL;
1828
1829         if (list_empty(&walk->walk.all) && walk->seq != 0)
1830                 return 0;
1831
1832         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1833         if (list_empty(&walk->walk.all))
1834                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1835         else
1836                 x = list_first_entry(&walk->walk.all,
1837                                      struct xfrm_policy_walk_entry, all);
1838
1839         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1840                 if (x->dead)
1841                         continue;
1842                 pol = container_of(x, struct xfrm_policy, walk);
1843                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1844                     walk->type != pol->type)
1845                         continue;
1846                 error = func(pol, xfrm_policy_id2dir(pol->index),
1847                              walk->seq, data);
1848                 if (error) {
1849                         list_move_tail(&walk->walk.all, &x->all);
1850                         goto out;
1851                 }
1852                 walk->seq++;
1853         }
1854         if (walk->seq == 0) {
1855                 error = -ENOENT;
1856                 goto out;
1857         }
1858         list_del_init(&walk->walk.all);
1859 out:
1860         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1861         return error;
1862 }
1863 EXPORT_SYMBOL(xfrm_policy_walk);
1864
1865 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1866 {
1867         INIT_LIST_HEAD(&walk->walk.all);
1868         walk->walk.dead = 1;
1869         walk->type = type;
1870         walk->seq = 0;
1871 }
1872 EXPORT_SYMBOL(xfrm_policy_walk_init);
1873
1874 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1875 {
1876         if (list_empty(&walk->walk.all))
1877                 return;
1878
1879         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1880         list_del(&walk->walk.all);
1881         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1882 }
1883 EXPORT_SYMBOL(xfrm_policy_walk_done);
1884
1885 /*
1886  * Find policy to apply to this flow.
1887  *
1888  * Returns 0 if policy found, else an -errno.
1889  */
1890 static int xfrm_policy_match(const struct xfrm_policy *pol,
1891                              const struct flowi *fl,
1892                              u8 type, u16 family, u32 if_id)
1893 {
1894         const struct xfrm_selector *sel = &pol->selector;
1895         int ret = -ESRCH;
1896         bool match;
1897
1898         if (pol->family != family ||
1899             pol->if_id != if_id ||
1900             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1901             pol->type != type)
1902                 return ret;
1903
1904         match = xfrm_selector_match(sel, fl, family);
1905         if (match)
1906                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
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_spinlock_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, 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, 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, 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, if_id);
2065                 if (!tmp)
2066                         continue;
2067
2068                 if (IS_ERR(tmp))
2069                         return tmp;
2070                 prefer = tmp;
2071         }
2072
2073         return prefer;
2074 }
2075
2076 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2077                                                      const struct flowi *fl,
2078                                                      u16 family, u8 dir,
2079                                                      u32 if_id)
2080 {
2081         struct xfrm_pol_inexact_candidates cand;
2082         const xfrm_address_t *daddr, *saddr;
2083         struct xfrm_pol_inexact_bin *bin;
2084         struct xfrm_policy *pol, *ret;
2085         struct hlist_head *chain;
2086         unsigned int sequence;
2087         int err;
2088
2089         daddr = xfrm_flowi_daddr(fl, family);
2090         saddr = xfrm_flowi_saddr(fl, family);
2091         if (unlikely(!daddr || !saddr))
2092                 return NULL;
2093
2094         rcu_read_lock();
2095  retry:
2096         do {
2097                 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2098                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2099         } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2100
2101         ret = NULL;
2102         hlist_for_each_entry_rcu(pol, chain, bydst) {
2103                 err = xfrm_policy_match(pol, fl, type, family, if_id);
2104                 if (err) {
2105                         if (err == -ESRCH)
2106                                 continue;
2107                         else {
2108                                 ret = ERR_PTR(err);
2109                                 goto fail;
2110                         }
2111                 } else {
2112                         ret = pol;
2113                         break;
2114                 }
2115         }
2116         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2117         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2118                                                          daddr))
2119                 goto skip_inexact;
2120
2121         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2122                                           family, if_id);
2123         if (pol) {
2124                 ret = pol;
2125                 if (IS_ERR(pol))
2126                         goto fail;
2127         }
2128
2129 skip_inexact:
2130         if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2131                 goto retry;
2132
2133         if (ret && !xfrm_pol_hold_rcu(ret))
2134                 goto retry;
2135 fail:
2136         rcu_read_unlock();
2137
2138         return ret;
2139 }
2140
2141 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2142                                               const struct flowi *fl,
2143                                               u16 family, u8 dir, u32 if_id)
2144 {
2145 #ifdef CONFIG_XFRM_SUB_POLICY
2146         struct xfrm_policy *pol;
2147
2148         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2149                                         dir, if_id);
2150         if (pol != NULL)
2151                 return pol;
2152 #endif
2153         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2154                                          dir, if_id);
2155 }
2156
2157 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2158                                                  const struct flowi *fl,
2159                                                  u16 family, u32 if_id)
2160 {
2161         struct xfrm_policy *pol;
2162
2163         rcu_read_lock();
2164  again:
2165         pol = rcu_dereference(sk->sk_policy[dir]);
2166         if (pol != NULL) {
2167                 bool match;
2168                 int err = 0;
2169
2170                 if (pol->family != family) {
2171                         pol = NULL;
2172                         goto out;
2173                 }
2174
2175                 match = xfrm_selector_match(&pol->selector, fl, family);
2176                 if (match) {
2177                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2178                             pol->if_id != if_id) {
2179                                 pol = NULL;
2180                                 goto out;
2181                         }
2182                         err = security_xfrm_policy_lookup(pol->security,
2183                                                       fl->flowi_secid);
2184                         if (!err) {
2185                                 if (!xfrm_pol_hold_rcu(pol))
2186                                         goto again;
2187                         } else if (err == -ESRCH) {
2188                                 pol = NULL;
2189                         } else {
2190                                 pol = ERR_PTR(err);
2191                         }
2192                 } else
2193                         pol = NULL;
2194         }
2195 out:
2196         rcu_read_unlock();
2197         return pol;
2198 }
2199
2200 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2201 {
2202         struct net *net = xp_net(pol);
2203
2204         list_add(&pol->walk.all, &net->xfrm.policy_all);
2205         net->xfrm.policy_count[dir]++;
2206         xfrm_pol_hold(pol);
2207 }
2208
2209 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2210                                                 int dir)
2211 {
2212         struct net *net = xp_net(pol);
2213
2214         if (list_empty(&pol->walk.all))
2215                 return NULL;
2216
2217         /* Socket policies are not hashed. */
2218         if (!hlist_unhashed(&pol->bydst)) {
2219                 hlist_del_rcu(&pol->bydst);
2220                 hlist_del_init(&pol->bydst_inexact_list);
2221                 hlist_del(&pol->byidx);
2222         }
2223
2224         list_del_init(&pol->walk.all);
2225         net->xfrm.policy_count[dir]--;
2226
2227         return pol;
2228 }
2229
2230 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2231 {
2232         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2233 }
2234
2235 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2236 {
2237         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2238 }
2239
2240 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2241 {
2242         struct net *net = xp_net(pol);
2243
2244         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2245         pol = __xfrm_policy_unlink(pol, dir);
2246         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2247         if (pol) {
2248                 xfrm_policy_kill(pol);
2249                 return 0;
2250         }
2251         return -ENOENT;
2252 }
2253 EXPORT_SYMBOL(xfrm_policy_delete);
2254
2255 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2256 {
2257         struct net *net = sock_net(sk);
2258         struct xfrm_policy *old_pol;
2259
2260 #ifdef CONFIG_XFRM_SUB_POLICY
2261         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2262                 return -EINVAL;
2263 #endif
2264
2265         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2266         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2267                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2268         if (pol) {
2269                 pol->curlft.add_time = ktime_get_real_seconds();
2270                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2271                 xfrm_sk_policy_link(pol, dir);
2272         }
2273         rcu_assign_pointer(sk->sk_policy[dir], pol);
2274         if (old_pol) {
2275                 if (pol)
2276                         xfrm_policy_requeue(old_pol, pol);
2277
2278                 /* Unlinking succeeds always. This is the only function
2279                  * allowed to delete or replace socket policy.
2280                  */
2281                 xfrm_sk_policy_unlink(old_pol, dir);
2282         }
2283         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2284
2285         if (old_pol) {
2286                 xfrm_policy_kill(old_pol);
2287         }
2288         return 0;
2289 }
2290
2291 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2292 {
2293         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2294         struct net *net = xp_net(old);
2295
2296         if (newp) {
2297                 newp->selector = old->selector;
2298                 if (security_xfrm_policy_clone(old->security,
2299                                                &newp->security)) {
2300                         kfree(newp);
2301                         return NULL;  /* ENOMEM */
2302                 }
2303                 newp->lft = old->lft;
2304                 newp->curlft = old->curlft;
2305                 newp->mark = old->mark;
2306                 newp->if_id = old->if_id;
2307                 newp->action = old->action;
2308                 newp->flags = old->flags;
2309                 newp->xfrm_nr = old->xfrm_nr;
2310                 newp->index = old->index;
2311                 newp->type = old->type;
2312                 newp->family = old->family;
2313                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2314                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2315                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2316                 xfrm_sk_policy_link(newp, dir);
2317                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2318                 xfrm_pol_put(newp);
2319         }
2320         return newp;
2321 }
2322
2323 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2324 {
2325         const struct xfrm_policy *p;
2326         struct xfrm_policy *np;
2327         int i, ret = 0;
2328
2329         rcu_read_lock();
2330         for (i = 0; i < 2; i++) {
2331                 p = rcu_dereference(osk->sk_policy[i]);
2332                 if (p) {
2333                         np = clone_policy(p, i);
2334                         if (unlikely(!np)) {
2335                                 ret = -ENOMEM;
2336                                 break;
2337                         }
2338                         rcu_assign_pointer(sk->sk_policy[i], np);
2339                 }
2340         }
2341         rcu_read_unlock();
2342         return ret;
2343 }
2344
2345 static int
2346 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2347                xfrm_address_t *remote, unsigned short family, u32 mark)
2348 {
2349         int err;
2350         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2351
2352         if (unlikely(afinfo == NULL))
2353                 return -EINVAL;
2354         err = afinfo->get_saddr(net, oif, local, remote, mark);
2355         rcu_read_unlock();
2356         return err;
2357 }
2358
2359 /* Resolve list of templates for the flow, given policy. */
2360
2361 static int
2362 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2363                       struct xfrm_state **xfrm, unsigned short family)
2364 {
2365         struct net *net = xp_net(policy);
2366         int nx;
2367         int i, error;
2368         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2369         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2370         xfrm_address_t tmp;
2371
2372         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2373                 struct xfrm_state *x;
2374                 xfrm_address_t *remote = daddr;
2375                 xfrm_address_t *local  = saddr;
2376                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2377
2378                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2379                     tmpl->mode == XFRM_MODE_BEET) {
2380                         remote = &tmpl->id.daddr;
2381                         local = &tmpl->saddr;
2382                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2383                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2384                                                        &tmp, remote,
2385                                                        tmpl->encap_family, 0);
2386                                 if (error)
2387                                         goto fail;
2388                                 local = &tmp;
2389                         }
2390                 }
2391
2392                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2393                                     family, policy->if_id);
2394
2395                 if (x && x->km.state == XFRM_STATE_VALID) {
2396                         xfrm[nx++] = x;
2397                         daddr = remote;
2398                         saddr = local;
2399                         continue;
2400                 }
2401                 if (x) {
2402                         error = (x->km.state == XFRM_STATE_ERROR ?
2403                                  -EINVAL : -EAGAIN);
2404                         xfrm_state_put(x);
2405                 } else if (error == -ESRCH) {
2406                         error = -EAGAIN;
2407                 }
2408
2409                 if (!tmpl->optional)
2410                         goto fail;
2411         }
2412         return nx;
2413
2414 fail:
2415         for (nx--; nx >= 0; nx--)
2416                 xfrm_state_put(xfrm[nx]);
2417         return error;
2418 }
2419
2420 static int
2421 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2422                   struct xfrm_state **xfrm, unsigned short family)
2423 {
2424         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2425         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2426         int cnx = 0;
2427         int error;
2428         int ret;
2429         int i;
2430
2431         for (i = 0; i < npols; i++) {
2432                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2433                         error = -ENOBUFS;
2434                         goto fail;
2435                 }
2436
2437                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2438                 if (ret < 0) {
2439                         error = ret;
2440                         goto fail;
2441                 } else
2442                         cnx += ret;
2443         }
2444
2445         /* found states are sorted for outbound processing */
2446         if (npols > 1)
2447                 xfrm_state_sort(xfrm, tpp, cnx, family);
2448
2449         return cnx;
2450
2451  fail:
2452         for (cnx--; cnx >= 0; cnx--)
2453                 xfrm_state_put(tpp[cnx]);
2454         return error;
2455
2456 }
2457
2458 static int xfrm_get_tos(const struct flowi *fl, int family)
2459 {
2460         if (family == AF_INET)
2461                 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2462
2463         return 0;
2464 }
2465
2466 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2467 {
2468         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2469         struct dst_ops *dst_ops;
2470         struct xfrm_dst *xdst;
2471
2472         if (!afinfo)
2473                 return ERR_PTR(-EINVAL);
2474
2475         switch (family) {
2476         case AF_INET:
2477                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2478                 break;
2479 #if IS_ENABLED(CONFIG_IPV6)
2480         case AF_INET6:
2481                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2482                 break;
2483 #endif
2484         default:
2485                 BUG();
2486         }
2487         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2488
2489         if (likely(xdst)) {
2490                 memset_after(xdst, 0, u.dst);
2491         } else
2492                 xdst = ERR_PTR(-ENOBUFS);
2493
2494         rcu_read_unlock();
2495
2496         return xdst;
2497 }
2498
2499 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2500                            int nfheader_len)
2501 {
2502         if (dst->ops->family == AF_INET6) {
2503                 struct rt6_info *rt = (struct rt6_info *)dst;
2504                 path->path_cookie = rt6_get_cookie(rt);
2505                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2506         }
2507 }
2508
2509 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2510                                 const struct flowi *fl)
2511 {
2512         const struct xfrm_policy_afinfo *afinfo =
2513                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2514         int err;
2515
2516         if (!afinfo)
2517                 return -EINVAL;
2518
2519         err = afinfo->fill_dst(xdst, dev, fl);
2520
2521         rcu_read_unlock();
2522
2523         return err;
2524 }
2525
2526
2527 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2528  * all the metrics... Shortly, bundle a bundle.
2529  */
2530
2531 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2532                                             struct xfrm_state **xfrm,
2533                                             struct xfrm_dst **bundle,
2534                                             int nx,
2535                                             const struct flowi *fl,
2536                                             struct dst_entry *dst)
2537 {
2538         const struct xfrm_state_afinfo *afinfo;
2539         const struct xfrm_mode *inner_mode;
2540         struct net *net = xp_net(policy);
2541         unsigned long now = jiffies;
2542         struct net_device *dev;
2543         struct xfrm_dst *xdst_prev = NULL;
2544         struct xfrm_dst *xdst0 = NULL;
2545         int i = 0;
2546         int err;
2547         int header_len = 0;
2548         int nfheader_len = 0;
2549         int trailer_len = 0;
2550         int tos;
2551         int family = policy->selector.family;
2552         xfrm_address_t saddr, daddr;
2553
2554         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2555
2556         tos = xfrm_get_tos(fl, family);
2557
2558         dst_hold(dst);
2559
2560         for (; i < nx; i++) {
2561                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2562                 struct dst_entry *dst1 = &xdst->u.dst;
2563
2564                 err = PTR_ERR(xdst);
2565                 if (IS_ERR(xdst)) {
2566                         dst_release(dst);
2567                         goto put_states;
2568                 }
2569
2570                 bundle[i] = xdst;
2571                 if (!xdst_prev)
2572                         xdst0 = xdst;
2573                 else
2574                         /* Ref count is taken during xfrm_alloc_dst()
2575                          * No need to do dst_clone() on dst1
2576                          */
2577                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2578
2579                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2580                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2581                                                         xfrm_af2proto(family));
2582                         if (!inner_mode) {
2583                                 err = -EAFNOSUPPORT;
2584                                 dst_release(dst);
2585                                 goto put_states;
2586                         }
2587                 } else
2588                         inner_mode = &xfrm[i]->inner_mode;
2589
2590                 xdst->route = dst;
2591                 dst_copy_metrics(dst1, dst);
2592
2593                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2594                         __u32 mark = 0;
2595                         int oif;
2596
2597                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2598                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2599
2600                         family = xfrm[i]->props.family;
2601                         oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2602                         dst = xfrm_dst_lookup(xfrm[i], tos, oif,
2603                                               &saddr, &daddr, family, mark);
2604                         err = PTR_ERR(dst);
2605                         if (IS_ERR(dst))
2606                                 goto put_states;
2607                 } else
2608                         dst_hold(dst);
2609
2610                 dst1->xfrm = xfrm[i];
2611                 xdst->xfrm_genid = xfrm[i]->genid;
2612
2613                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2614                 dst1->lastuse = now;
2615
2616                 dst1->input = dst_discard;
2617
2618                 rcu_read_lock();
2619                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2620                 if (likely(afinfo))
2621                         dst1->output = afinfo->output;
2622                 else
2623                         dst1->output = dst_discard_out;
2624                 rcu_read_unlock();
2625
2626                 xdst_prev = xdst;
2627
2628                 header_len += xfrm[i]->props.header_len;
2629                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2630                         nfheader_len += xfrm[i]->props.header_len;
2631                 trailer_len += xfrm[i]->props.trailer_len;
2632         }
2633
2634         xfrm_dst_set_child(xdst_prev, dst);
2635         xdst0->path = dst;
2636
2637         err = -ENODEV;
2638         dev = dst->dev;
2639         if (!dev)
2640                 goto free_dst;
2641
2642         xfrm_init_path(xdst0, dst, nfheader_len);
2643         xfrm_init_pmtu(bundle, nx);
2644
2645         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2646              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2647                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2648                 if (err)
2649                         goto free_dst;
2650
2651                 xdst_prev->u.dst.header_len = header_len;
2652                 xdst_prev->u.dst.trailer_len = trailer_len;
2653                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2654                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2655         }
2656
2657         return &xdst0->u.dst;
2658
2659 put_states:
2660         for (; i < nx; i++)
2661                 xfrm_state_put(xfrm[i]);
2662 free_dst:
2663         if (xdst0)
2664                 dst_release_immediate(&xdst0->u.dst);
2665
2666         return ERR_PTR(err);
2667 }
2668
2669 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2670                                 struct xfrm_policy **pols,
2671                                 int *num_pols, int *num_xfrms)
2672 {
2673         int i;
2674
2675         if (*num_pols == 0 || !pols[0]) {
2676                 *num_pols = 0;
2677                 *num_xfrms = 0;
2678                 return 0;
2679         }
2680         if (IS_ERR(pols[0])) {
2681                 *num_pols = 0;
2682                 return PTR_ERR(pols[0]);
2683         }
2684
2685         *num_xfrms = pols[0]->xfrm_nr;
2686
2687 #ifdef CONFIG_XFRM_SUB_POLICY
2688         if (pols[0]->action == XFRM_POLICY_ALLOW &&
2689             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2690                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2691                                                     XFRM_POLICY_TYPE_MAIN,
2692                                                     fl, family,
2693                                                     XFRM_POLICY_OUT,
2694                                                     pols[0]->if_id);
2695                 if (pols[1]) {
2696                         if (IS_ERR(pols[1])) {
2697                                 xfrm_pols_put(pols, *num_pols);
2698                                 *num_pols = 0;
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         __u32 skb_mark;
2766
2767         spin_lock(&pq->hold_queue.lock);
2768         skb = skb_peek(&pq->hold_queue);
2769         if (!skb) {
2770                 spin_unlock(&pq->hold_queue.lock);
2771                 goto out;
2772         }
2773         dst = skb_dst(skb);
2774         sk = skb->sk;
2775
2776         /* Fixup the mark to support VTI. */
2777         skb_mark = skb->mark;
2778         skb->mark = pol->mark.v;
2779         xfrm_decode_session(skb, &fl, dst->ops->family);
2780         skb->mark = skb_mark;
2781         spin_unlock(&pq->hold_queue.lock);
2782
2783         dst_hold(xfrm_dst_path(dst));
2784         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2785         if (IS_ERR(dst))
2786                 goto purge_queue;
2787
2788         if (dst->flags & DST_XFRM_QUEUE) {
2789                 dst_release(dst);
2790
2791                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2792                         goto purge_queue;
2793
2794                 pq->timeout = pq->timeout << 1;
2795                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2796                         xfrm_pol_hold(pol);
2797                 goto out;
2798         }
2799
2800         dst_release(dst);
2801
2802         __skb_queue_head_init(&list);
2803
2804         spin_lock(&pq->hold_queue.lock);
2805         pq->timeout = 0;
2806         skb_queue_splice_init(&pq->hold_queue, &list);
2807         spin_unlock(&pq->hold_queue.lock);
2808
2809         while (!skb_queue_empty(&list)) {
2810                 skb = __skb_dequeue(&list);
2811
2812                 /* Fixup the mark to support VTI. */
2813                 skb_mark = skb->mark;
2814                 skb->mark = pol->mark.v;
2815                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2816                 skb->mark = skb_mark;
2817
2818                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2819                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2820                 if (IS_ERR(dst)) {
2821                         kfree_skb(skb);
2822                         continue;
2823                 }
2824
2825                 nf_reset_ct(skb);
2826                 skb_dst_drop(skb);
2827                 skb_dst_set(skb, dst);
2828
2829                 dst_output(net, skb->sk, skb);
2830         }
2831
2832 out:
2833         xfrm_pol_put(pol);
2834         return;
2835
2836 purge_queue:
2837         pq->timeout = 0;
2838         skb_queue_purge(&pq->hold_queue);
2839         xfrm_pol_put(pol);
2840 }
2841
2842 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2843 {
2844         unsigned long sched_next;
2845         struct dst_entry *dst = skb_dst(skb);
2846         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2847         struct xfrm_policy *pol = xdst->pols[0];
2848         struct xfrm_policy_queue *pq = &pol->polq;
2849
2850         if (unlikely(skb_fclone_busy(sk, skb))) {
2851                 kfree_skb(skb);
2852                 return 0;
2853         }
2854
2855         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2856                 kfree_skb(skb);
2857                 return -EAGAIN;
2858         }
2859
2860         skb_dst_force(skb);
2861
2862         spin_lock_bh(&pq->hold_queue.lock);
2863
2864         if (!pq->timeout)
2865                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2866
2867         sched_next = jiffies + pq->timeout;
2868
2869         if (del_timer(&pq->hold_timer)) {
2870                 if (time_before(pq->hold_timer.expires, sched_next))
2871                         sched_next = pq->hold_timer.expires;
2872                 xfrm_pol_put(pol);
2873         }
2874
2875         __skb_queue_tail(&pq->hold_queue, skb);
2876         if (!mod_timer(&pq->hold_timer, sched_next))
2877                 xfrm_pol_hold(pol);
2878
2879         spin_unlock_bh(&pq->hold_queue.lock);
2880
2881         return 0;
2882 }
2883
2884 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2885                                                  struct xfrm_flo *xflo,
2886                                                  const struct flowi *fl,
2887                                                  int num_xfrms,
2888                                                  u16 family)
2889 {
2890         int err;
2891         struct net_device *dev;
2892         struct dst_entry *dst;
2893         struct dst_entry *dst1;
2894         struct xfrm_dst *xdst;
2895
2896         xdst = xfrm_alloc_dst(net, family);
2897         if (IS_ERR(xdst))
2898                 return xdst;
2899
2900         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2901             net->xfrm.sysctl_larval_drop ||
2902             num_xfrms <= 0)
2903                 return xdst;
2904
2905         dst = xflo->dst_orig;
2906         dst1 = &xdst->u.dst;
2907         dst_hold(dst);
2908         xdst->route = dst;
2909
2910         dst_copy_metrics(dst1, dst);
2911
2912         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2913         dst1->flags |= DST_XFRM_QUEUE;
2914         dst1->lastuse = jiffies;
2915
2916         dst1->input = dst_discard;
2917         dst1->output = xdst_queue_output;
2918
2919         dst_hold(dst);
2920         xfrm_dst_set_child(xdst, dst);
2921         xdst->path = dst;
2922
2923         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2924
2925         err = -ENODEV;
2926         dev = dst->dev;
2927         if (!dev)
2928                 goto free_dst;
2929
2930         err = xfrm_fill_dst(xdst, dev, fl);
2931         if (err)
2932                 goto free_dst;
2933
2934 out:
2935         return xdst;
2936
2937 free_dst:
2938         dst_release(dst1);
2939         xdst = ERR_PTR(err);
2940         goto out;
2941 }
2942
2943 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2944                                            const struct flowi *fl,
2945                                            u16 family, u8 dir,
2946                                            struct xfrm_flo *xflo, u32 if_id)
2947 {
2948         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2949         int num_pols = 0, num_xfrms = 0, err;
2950         struct xfrm_dst *xdst;
2951
2952         /* Resolve policies to use if we couldn't get them from
2953          * previous cache entry */
2954         num_pols = 1;
2955         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2956         err = xfrm_expand_policies(fl, family, pols,
2957                                            &num_pols, &num_xfrms);
2958         if (err < 0)
2959                 goto inc_error;
2960         if (num_pols == 0)
2961                 return NULL;
2962         if (num_xfrms <= 0)
2963                 goto make_dummy_bundle;
2964
2965         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2966                                               xflo->dst_orig);
2967         if (IS_ERR(xdst)) {
2968                 err = PTR_ERR(xdst);
2969                 if (err == -EREMOTE) {
2970                         xfrm_pols_put(pols, num_pols);
2971                         return NULL;
2972                 }
2973
2974                 if (err != -EAGAIN)
2975                         goto error;
2976                 goto make_dummy_bundle;
2977         } else if (xdst == NULL) {
2978                 num_xfrms = 0;
2979                 goto make_dummy_bundle;
2980         }
2981
2982         return xdst;
2983
2984 make_dummy_bundle:
2985         /* We found policies, but there's no bundles to instantiate:
2986          * either because the policy blocks, has no transformations or
2987          * we could not build template (no xfrm_states).*/
2988         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2989         if (IS_ERR(xdst)) {
2990                 xfrm_pols_put(pols, num_pols);
2991                 return ERR_CAST(xdst);
2992         }
2993         xdst->num_pols = num_pols;
2994         xdst->num_xfrms = num_xfrms;
2995         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2996
2997         return xdst;
2998
2999 inc_error:
3000         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3001 error:
3002         xfrm_pols_put(pols, num_pols);
3003         return ERR_PTR(err);
3004 }
3005
3006 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3007                                         struct dst_entry *dst_orig)
3008 {
3009         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3010         struct dst_entry *ret;
3011
3012         if (!afinfo) {
3013                 dst_release(dst_orig);
3014                 return ERR_PTR(-EINVAL);
3015         } else {
3016                 ret = afinfo->blackhole_route(net, dst_orig);
3017         }
3018         rcu_read_unlock();
3019
3020         return ret;
3021 }
3022
3023 /* Finds/creates a bundle for given flow and if_id
3024  *
3025  * At the moment we eat a raw IP route. Mostly to speed up lookups
3026  * on interfaces with disabled IPsec.
3027  *
3028  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3029  * compatibility
3030  */
3031 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3032                                         struct dst_entry *dst_orig,
3033                                         const struct flowi *fl,
3034                                         const struct sock *sk,
3035                                         int flags, u32 if_id)
3036 {
3037         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3038         struct xfrm_dst *xdst;
3039         struct dst_entry *dst, *route;
3040         u16 family = dst_orig->ops->family;
3041         u8 dir = XFRM_POLICY_OUT;
3042         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3043
3044         dst = NULL;
3045         xdst = NULL;
3046         route = NULL;
3047
3048         sk = sk_const_to_full_sk(sk);
3049         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3050                 num_pols = 1;
3051                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3052                                                 if_id);
3053                 err = xfrm_expand_policies(fl, family, pols,
3054                                            &num_pols, &num_xfrms);
3055                 if (err < 0)
3056                         goto dropdst;
3057
3058                 if (num_pols) {
3059                         if (num_xfrms <= 0) {
3060                                 drop_pols = num_pols;
3061                                 goto no_transform;
3062                         }
3063
3064                         xdst = xfrm_resolve_and_create_bundle(
3065                                         pols, num_pols, fl,
3066                                         family, dst_orig);
3067
3068                         if (IS_ERR(xdst)) {
3069                                 xfrm_pols_put(pols, num_pols);
3070                                 err = PTR_ERR(xdst);
3071                                 if (err == -EREMOTE)
3072                                         goto nopol;
3073
3074                                 goto dropdst;
3075                         } else if (xdst == NULL) {
3076                                 num_xfrms = 0;
3077                                 drop_pols = num_pols;
3078                                 goto no_transform;
3079                         }
3080
3081                         route = xdst->route;
3082                 }
3083         }
3084
3085         if (xdst == NULL) {
3086                 struct xfrm_flo xflo;
3087
3088                 xflo.dst_orig = dst_orig;
3089                 xflo.flags = flags;
3090
3091                 /* To accelerate a bit...  */
3092                 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3093                                !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3094                         goto nopol;
3095
3096                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3097                 if (xdst == NULL)
3098                         goto nopol;
3099                 if (IS_ERR(xdst)) {
3100                         err = PTR_ERR(xdst);
3101                         goto dropdst;
3102                 }
3103
3104                 num_pols = xdst->num_pols;
3105                 num_xfrms = xdst->num_xfrms;
3106                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3107                 route = xdst->route;
3108         }
3109
3110         dst = &xdst->u.dst;
3111         if (route == NULL && num_xfrms > 0) {
3112                 /* The only case when xfrm_bundle_lookup() returns a
3113                  * bundle with null route, is when the template could
3114                  * not be resolved. It means policies are there, but
3115                  * bundle could not be created, since we don't yet
3116                  * have the xfrm_state's. We need to wait for KM to
3117                  * negotiate new SA's or bail out with error.*/
3118                 if (net->xfrm.sysctl_larval_drop) {
3119                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3120                         err = -EREMOTE;
3121                         goto error;
3122                 }
3123
3124                 err = -EAGAIN;
3125
3126                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3127                 goto error;
3128         }
3129
3130 no_transform:
3131         if (num_pols == 0)
3132                 goto nopol;
3133
3134         if ((flags & XFRM_LOOKUP_ICMP) &&
3135             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3136                 err = -ENOENT;
3137                 goto error;
3138         }
3139
3140         for (i = 0; i < num_pols; i++)
3141                 pols[i]->curlft.use_time = ktime_get_real_seconds();
3142
3143         if (num_xfrms < 0) {
3144                 /* Prohibit the flow */
3145                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3146                 err = -EPERM;
3147                 goto error;
3148         } else if (num_xfrms > 0) {
3149                 /* Flow transformed */
3150                 dst_release(dst_orig);
3151         } else {
3152                 /* Flow passes untransformed */
3153                 dst_release(dst);
3154                 dst = dst_orig;
3155         }
3156 ok:
3157         xfrm_pols_put(pols, drop_pols);
3158         if (dst && dst->xfrm &&
3159             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3160                 dst->flags |= DST_XFRM_TUNNEL;
3161         return dst;
3162
3163 nopol:
3164         if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3165             net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3166                 err = -EPERM;
3167                 goto error;
3168         }
3169         if (!(flags & XFRM_LOOKUP_ICMP)) {
3170                 dst = dst_orig;
3171                 goto ok;
3172         }
3173         err = -ENOENT;
3174 error:
3175         dst_release(dst);
3176 dropdst:
3177         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3178                 dst_release(dst_orig);
3179         xfrm_pols_put(pols, drop_pols);
3180         return ERR_PTR(err);
3181 }
3182 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3183
3184 /* Main function: finds/creates a bundle for given flow.
3185  *
3186  * At the moment we eat a raw IP route. Mostly to speed up lookups
3187  * on interfaces with disabled IPsec.
3188  */
3189 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3190                               const struct flowi *fl, const struct sock *sk,
3191                               int flags)
3192 {
3193         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3194 }
3195 EXPORT_SYMBOL(xfrm_lookup);
3196
3197 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3198  * Otherwise we may send out blackholed packets.
3199  */
3200 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3201                                     const struct flowi *fl,
3202                                     const struct sock *sk, int flags)
3203 {
3204         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3205                                             flags | XFRM_LOOKUP_QUEUE |
3206                                             XFRM_LOOKUP_KEEP_DST_REF);
3207
3208         if (PTR_ERR(dst) == -EREMOTE)
3209                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3210
3211         if (IS_ERR(dst))
3212                 dst_release(dst_orig);
3213
3214         return dst;
3215 }
3216 EXPORT_SYMBOL(xfrm_lookup_route);
3217
3218 static inline int
3219 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3220 {
3221         struct sec_path *sp = skb_sec_path(skb);
3222         struct xfrm_state *x;
3223
3224         if (!sp || idx < 0 || idx >= sp->len)
3225                 return 0;
3226         x = sp->xvec[idx];
3227         if (!x->type->reject)
3228                 return 0;
3229         return x->type->reject(x, skb, fl);
3230 }
3231
3232 /* When skb is transformed back to its "native" form, we have to
3233  * check policy restrictions. At the moment we make this in maximally
3234  * stupid way. Shame on me. :-) Of course, connected sockets must
3235  * have policy cached at them.
3236  */
3237
3238 static inline int
3239 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3240               unsigned short family, u32 if_id)
3241 {
3242         if (xfrm_state_kern(x))
3243                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3244         return  x->id.proto == tmpl->id.proto &&
3245                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3246                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3247                 x->props.mode == tmpl->mode &&
3248                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3249                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3250                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3251                   xfrm_state_addr_cmp(tmpl, x, family)) &&
3252                 (if_id == 0 || if_id == x->if_id);
3253 }
3254
3255 /*
3256  * 0 or more than 0 is returned when validation is succeeded (either bypass
3257  * because of optional transport mode, or next index of the matched secpath
3258  * state with the template.
3259  * -1 is returned when no matching template is found.
3260  * Otherwise "-2 - errored_index" is returned.
3261  */
3262 static inline int
3263 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3264                unsigned short family, u32 if_id)
3265 {
3266         int idx = start;
3267
3268         if (tmpl->optional) {
3269                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3270                         return start;
3271         } else
3272                 start = -1;
3273         for (; idx < sp->len; idx++) {
3274                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3275                         return ++idx;
3276                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3277                         if (idx < sp->verified_cnt) {
3278                                 /* Secpath entry previously verified, consider optional and
3279                                  * continue searching
3280                                  */
3281                                 continue;
3282                         }
3283
3284                         if (start == -1)
3285                                 start = -2-idx;
3286                         break;
3287                 }
3288         }
3289         return start;
3290 }
3291
3292 static void
3293 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3294 {
3295         const struct iphdr *iph = ip_hdr(skb);
3296         int ihl = iph->ihl;
3297         u8 *xprth = skb_network_header(skb) + ihl * 4;
3298         struct flowi4 *fl4 = &fl->u.ip4;
3299         int oif = 0;
3300
3301         if (skb_dst(skb) && skb_dst(skb)->dev)
3302                 oif = skb_dst(skb)->dev->ifindex;
3303
3304         memset(fl4, 0, sizeof(struct flowi4));
3305         fl4->flowi4_mark = skb->mark;
3306         fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3307
3308         fl4->flowi4_proto = iph->protocol;
3309         fl4->daddr = reverse ? iph->saddr : iph->daddr;
3310         fl4->saddr = reverse ? iph->daddr : iph->saddr;
3311         fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3312
3313         if (!ip_is_fragment(iph)) {
3314                 switch (iph->protocol) {
3315                 case IPPROTO_UDP:
3316                 case IPPROTO_UDPLITE:
3317                 case IPPROTO_TCP:
3318                 case IPPROTO_SCTP:
3319                 case IPPROTO_DCCP:
3320                         if (xprth + 4 < skb->data ||
3321                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3322                                 __be16 *ports;
3323
3324                                 xprth = skb_network_header(skb) + ihl * 4;
3325                                 ports = (__be16 *)xprth;
3326
3327                                 fl4->fl4_sport = ports[!!reverse];
3328                                 fl4->fl4_dport = ports[!reverse];
3329                         }
3330                         break;
3331                 case IPPROTO_ICMP:
3332                         if (xprth + 2 < skb->data ||
3333                             pskb_may_pull(skb, xprth + 2 - skb->data)) {
3334                                 u8 *icmp;
3335
3336                                 xprth = skb_network_header(skb) + ihl * 4;
3337                                 icmp = xprth;
3338
3339                                 fl4->fl4_icmp_type = icmp[0];
3340                                 fl4->fl4_icmp_code = icmp[1];
3341                         }
3342                         break;
3343                 case IPPROTO_GRE:
3344                         if (xprth + 12 < skb->data ||
3345                             pskb_may_pull(skb, xprth + 12 - skb->data)) {
3346                                 __be16 *greflags;
3347                                 __be32 *gre_hdr;
3348
3349                                 xprth = skb_network_header(skb) + ihl * 4;
3350                                 greflags = (__be16 *)xprth;
3351                                 gre_hdr = (__be32 *)xprth;
3352
3353                                 if (greflags[0] & GRE_KEY) {
3354                                         if (greflags[0] & GRE_CSUM)
3355                                                 gre_hdr++;
3356                                         fl4->fl4_gre_key = gre_hdr[1];
3357                                 }
3358                         }
3359                         break;
3360                 default:
3361                         break;
3362                 }
3363         }
3364 }
3365
3366 #if IS_ENABLED(CONFIG_IPV6)
3367 static void
3368 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3369 {
3370         struct flowi6 *fl6 = &fl->u.ip6;
3371         int onlyproto = 0;
3372         const struct ipv6hdr *hdr = ipv6_hdr(skb);
3373         u32 offset = sizeof(*hdr);
3374         struct ipv6_opt_hdr *exthdr;
3375         const unsigned char *nh = skb_network_header(skb);
3376         u16 nhoff = IP6CB(skb)->nhoff;
3377         int oif = 0;
3378         u8 nexthdr;
3379
3380         if (!nhoff)
3381                 nhoff = offsetof(struct ipv6hdr, nexthdr);
3382
3383         nexthdr = nh[nhoff];
3384
3385         if (skb_dst(skb) && skb_dst(skb)->dev)
3386                 oif = skb_dst(skb)->dev->ifindex;
3387
3388         memset(fl6, 0, sizeof(struct flowi6));
3389         fl6->flowi6_mark = skb->mark;
3390         fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3391
3392         fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3393         fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3394
3395         while (nh + offset + sizeof(*exthdr) < skb->data ||
3396                pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3397                 nh = skb_network_header(skb);
3398                 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3399
3400                 switch (nexthdr) {
3401                 case NEXTHDR_FRAGMENT:
3402                         onlyproto = 1;
3403                         fallthrough;
3404                 case NEXTHDR_ROUTING:
3405                 case NEXTHDR_HOP:
3406                 case NEXTHDR_DEST:
3407                         offset += ipv6_optlen(exthdr);
3408                         nexthdr = exthdr->nexthdr;
3409                         break;
3410                 case IPPROTO_UDP:
3411                 case IPPROTO_UDPLITE:
3412                 case IPPROTO_TCP:
3413                 case IPPROTO_SCTP:
3414                 case IPPROTO_DCCP:
3415                         if (!onlyproto && (nh + offset + 4 < skb->data ||
3416                              pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3417                                 __be16 *ports;
3418
3419                                 nh = skb_network_header(skb);
3420                                 ports = (__be16 *)(nh + offset);
3421                                 fl6->fl6_sport = ports[!!reverse];
3422                                 fl6->fl6_dport = ports[!reverse];
3423                         }
3424                         fl6->flowi6_proto = nexthdr;
3425                         return;
3426                 case IPPROTO_ICMPV6:
3427                         if (!onlyproto && (nh + offset + 2 < skb->data ||
3428                             pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3429                                 u8 *icmp;
3430
3431                                 nh = skb_network_header(skb);
3432                                 icmp = (u8 *)(nh + offset);
3433                                 fl6->fl6_icmp_type = icmp[0];
3434                                 fl6->fl6_icmp_code = icmp[1];
3435                         }
3436                         fl6->flowi6_proto = nexthdr;
3437                         return;
3438                 case IPPROTO_GRE:
3439                         if (!onlyproto &&
3440                             (nh + offset + 12 < skb->data ||
3441                              pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3442                                 struct gre_base_hdr *gre_hdr;
3443                                 __be32 *gre_key;
3444
3445                                 nh = skb_network_header(skb);
3446                                 gre_hdr = (struct gre_base_hdr *)(nh + offset);
3447                                 gre_key = (__be32 *)(gre_hdr + 1);
3448
3449                                 if (gre_hdr->flags & GRE_KEY) {
3450                                         if (gre_hdr->flags & GRE_CSUM)
3451                                                 gre_key++;
3452                                         fl6->fl6_gre_key = *gre_key;
3453                                 }
3454                         }
3455                         fl6->flowi6_proto = nexthdr;
3456                         return;
3457
3458 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3459                 case IPPROTO_MH:
3460                         offset += ipv6_optlen(exthdr);
3461                         if (!onlyproto && (nh + offset + 3 < skb->data ||
3462                             pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3463                                 struct ip6_mh *mh;
3464
3465                                 nh = skb_network_header(skb);
3466                                 mh = (struct ip6_mh *)(nh + offset);
3467                                 fl6->fl6_mh_type = mh->ip6mh_type;
3468                         }
3469                         fl6->flowi6_proto = nexthdr;
3470                         return;
3471 #endif
3472                 default:
3473                         fl6->flowi6_proto = nexthdr;
3474                         return;
3475                 }
3476         }
3477 }
3478 #endif
3479
3480 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3481                           unsigned int family, int reverse)
3482 {
3483         switch (family) {
3484         case AF_INET:
3485                 decode_session4(skb, fl, reverse);
3486                 break;
3487 #if IS_ENABLED(CONFIG_IPV6)
3488         case AF_INET6:
3489                 decode_session6(skb, fl, reverse);
3490                 break;
3491 #endif
3492         default:
3493                 return -EAFNOSUPPORT;
3494         }
3495
3496         return security_xfrm_decode_session(skb, &fl->flowi_secid);
3497 }
3498 EXPORT_SYMBOL(__xfrm_decode_session);
3499
3500 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3501 {
3502         for (; k < sp->len; k++) {
3503                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3504                         *idxp = k;
3505                         return 1;
3506                 }
3507         }
3508
3509         return 0;
3510 }
3511
3512 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3513                         unsigned short family)
3514 {
3515         struct net *net = dev_net(skb->dev);
3516         struct xfrm_policy *pol;
3517         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3518         int npols = 0;
3519         int xfrm_nr;
3520         int pi;
3521         int reverse;
3522         struct flowi fl;
3523         int xerr_idx = -1;
3524         const struct xfrm_if_cb *ifcb;
3525         struct sec_path *sp;
3526         u32 if_id = 0;
3527
3528         rcu_read_lock();
3529         ifcb = xfrm_if_get_cb();
3530
3531         if (ifcb) {
3532                 struct xfrm_if_decode_session_result r;
3533
3534                 if (ifcb->decode_session(skb, family, &r)) {
3535                         if_id = r.if_id;
3536                         net = r.net;
3537                 }
3538         }
3539         rcu_read_unlock();
3540
3541         reverse = dir & ~XFRM_POLICY_MASK;
3542         dir &= XFRM_POLICY_MASK;
3543
3544         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3545                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3546                 return 0;
3547         }
3548
3549         nf_nat_decode_session(skb, &fl, family);
3550
3551         /* First, check used SA against their selectors. */
3552         sp = skb_sec_path(skb);
3553         if (sp) {
3554                 int i;
3555
3556                 for (i = sp->len - 1; i >= 0; i--) {
3557                         struct xfrm_state *x = sp->xvec[i];
3558                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3559                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3560                                 return 0;
3561                         }
3562                 }
3563         }
3564
3565         pol = NULL;
3566         sk = sk_to_full_sk(sk);
3567         if (sk && sk->sk_policy[dir]) {
3568                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3569                 if (IS_ERR(pol)) {
3570                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3571                         return 0;
3572                 }
3573         }
3574
3575         if (!pol)
3576                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3577
3578         if (IS_ERR(pol)) {
3579                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3580                 return 0;
3581         }
3582
3583         if (!pol) {
3584                 if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3585                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3586                         return 0;
3587                 }
3588
3589                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3590                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3591                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3592                         return 0;
3593                 }
3594                 return 1;
3595         }
3596
3597         /* This lockless write can happen from different cpus. */
3598         WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3599
3600         pols[0] = pol;
3601         npols++;
3602 #ifdef CONFIG_XFRM_SUB_POLICY
3603         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3604                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3605                                                     &fl, family,
3606                                                     XFRM_POLICY_IN, if_id);
3607                 if (pols[1]) {
3608                         if (IS_ERR(pols[1])) {
3609                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3610                                 xfrm_pol_put(pols[0]);
3611                                 return 0;
3612                         }
3613                         /* This write can happen from different cpus. */
3614                         WRITE_ONCE(pols[1]->curlft.use_time,
3615                                    ktime_get_real_seconds());
3616                         npols++;
3617                 }
3618         }
3619 #endif
3620
3621         if (pol->action == XFRM_POLICY_ALLOW) {
3622                 static struct sec_path dummy;
3623                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3624                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3625                 struct xfrm_tmpl **tpp = tp;
3626                 int ti = 0;
3627                 int i, k;
3628
3629                 sp = skb_sec_path(skb);
3630                 if (!sp)
3631                         sp = &dummy;
3632
3633                 for (pi = 0; pi < npols; pi++) {
3634                         if (pols[pi] != pol &&
3635                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3636                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3637                                 goto reject;
3638                         }
3639                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3640                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3641                                 goto reject_error;
3642                         }
3643                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3644                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3645                 }
3646                 xfrm_nr = ti;
3647
3648                 if (npols > 1) {
3649                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3650                         tpp = stp;
3651                 }
3652
3653                 /* For each tunnel xfrm, find the first matching tmpl.
3654                  * For each tmpl before that, find corresponding xfrm.
3655                  * Order is _important_. Later we will implement
3656                  * some barriers, but at the moment barriers
3657                  * are implied between each two transformations.
3658                  * Upon success, marks secpath entries as having been
3659                  * verified to allow them to be skipped in future policy
3660                  * checks (e.g. nested tunnels).
3661                  */
3662                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3663                         k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3664                         if (k < 0) {
3665                                 if (k < -1)
3666                                         /* "-2 - errored_index" returned */
3667                                         xerr_idx = -(2+k);
3668                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3669                                 goto reject;
3670                         }
3671                 }
3672
3673                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3674                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3675                         goto reject;
3676                 }
3677
3678                 xfrm_pols_put(pols, npols);
3679                 sp->verified_cnt = k;
3680
3681                 return 1;
3682         }
3683         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3684
3685 reject:
3686         xfrm_secpath_reject(xerr_idx, skb, &fl);
3687 reject_error:
3688         xfrm_pols_put(pols, npols);
3689         return 0;
3690 }
3691 EXPORT_SYMBOL(__xfrm_policy_check);
3692
3693 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3694 {
3695         struct net *net = dev_net(skb->dev);
3696         struct flowi fl;
3697         struct dst_entry *dst;
3698         int res = 1;
3699
3700         if (xfrm_decode_session(skb, &fl, family) < 0) {
3701                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3702                 return 0;
3703         }
3704
3705         skb_dst_force(skb);
3706         if (!skb_dst(skb)) {
3707                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3708                 return 0;
3709         }
3710
3711         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3712         if (IS_ERR(dst)) {
3713                 res = 0;
3714                 dst = NULL;
3715         }
3716         skb_dst_set(skb, dst);
3717         return res;
3718 }
3719 EXPORT_SYMBOL(__xfrm_route_forward);
3720
3721 /* Optimize later using cookies and generation ids. */
3722
3723 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3724 {
3725         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3726          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3727          * get validated by dst_ops->check on every use.  We do this
3728          * because when a normal route referenced by an XFRM dst is
3729          * obsoleted we do not go looking around for all parent
3730          * referencing XFRM dsts so that we can invalidate them.  It
3731          * is just too much work.  Instead we make the checks here on
3732          * every use.  For example:
3733          *
3734          *      XFRM dst A --> IPv4 dst X
3735          *
3736          * X is the "xdst->route" of A (X is also the "dst->path" of A
3737          * in this example).  If X is marked obsolete, "A" will not
3738          * notice.  That's what we are validating here via the
3739          * stale_bundle() check.
3740          *
3741          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3742          * be marked on it.
3743          * This will force stale_bundle() to fail on any xdst bundle with
3744          * this dst linked in it.
3745          */
3746         if (dst->obsolete < 0 && !stale_bundle(dst))
3747                 return dst;
3748
3749         return NULL;
3750 }
3751
3752 static int stale_bundle(struct dst_entry *dst)
3753 {
3754         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3755 }
3756
3757 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3758 {
3759         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3760                 dst->dev = blackhole_netdev;
3761                 dev_hold(dst->dev);
3762                 dev_put(dev);
3763         }
3764 }
3765 EXPORT_SYMBOL(xfrm_dst_ifdown);
3766
3767 static void xfrm_link_failure(struct sk_buff *skb)
3768 {
3769         /* Impossible. Such dst must be popped before reaches point of failure. */
3770 }
3771
3772 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3773 {
3774         if (dst) {
3775                 if (dst->obsolete) {
3776                         dst_release(dst);
3777                         dst = NULL;
3778                 }
3779         }
3780         return dst;
3781 }
3782
3783 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3784 {
3785         while (nr--) {
3786                 struct xfrm_dst *xdst = bundle[nr];
3787                 u32 pmtu, route_mtu_cached;
3788                 struct dst_entry *dst;
3789
3790                 dst = &xdst->u.dst;
3791                 pmtu = dst_mtu(xfrm_dst_child(dst));
3792                 xdst->child_mtu_cached = pmtu;
3793
3794                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3795
3796                 route_mtu_cached = dst_mtu(xdst->route);
3797                 xdst->route_mtu_cached = route_mtu_cached;
3798
3799                 if (pmtu > route_mtu_cached)
3800                         pmtu = route_mtu_cached;
3801
3802                 dst_metric_set(dst, RTAX_MTU, pmtu);
3803         }
3804 }
3805
3806 /* Check that the bundle accepts the flow and its components are
3807  * still valid.
3808  */
3809
3810 static int xfrm_bundle_ok(struct xfrm_dst *first)
3811 {
3812         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3813         struct dst_entry *dst = &first->u.dst;
3814         struct xfrm_dst *xdst;
3815         int start_from, nr;
3816         u32 mtu;
3817
3818         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3819             (dst->dev && !netif_running(dst->dev)))
3820                 return 0;
3821
3822         if (dst->flags & DST_XFRM_QUEUE)
3823                 return 1;
3824
3825         start_from = nr = 0;
3826         do {
3827                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3828
3829                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3830                         return 0;
3831                 if (xdst->xfrm_genid != dst->xfrm->genid)
3832                         return 0;
3833                 if (xdst->num_pols > 0 &&
3834                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3835                         return 0;
3836
3837                 bundle[nr++] = xdst;
3838
3839                 mtu = dst_mtu(xfrm_dst_child(dst));
3840                 if (xdst->child_mtu_cached != mtu) {
3841                         start_from = nr;
3842                         xdst->child_mtu_cached = mtu;
3843                 }
3844
3845                 if (!dst_check(xdst->route, xdst->route_cookie))
3846                         return 0;
3847                 mtu = dst_mtu(xdst->route);
3848                 if (xdst->route_mtu_cached != mtu) {
3849                         start_from = nr;
3850                         xdst->route_mtu_cached = mtu;
3851                 }
3852
3853                 dst = xfrm_dst_child(dst);
3854         } while (dst->xfrm);
3855
3856         if (likely(!start_from))
3857                 return 1;
3858
3859         xdst = bundle[start_from - 1];
3860         mtu = xdst->child_mtu_cached;
3861         while (start_from--) {
3862                 dst = &xdst->u.dst;
3863
3864                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3865                 if (mtu > xdst->route_mtu_cached)
3866                         mtu = xdst->route_mtu_cached;
3867                 dst_metric_set(dst, RTAX_MTU, mtu);
3868                 if (!start_from)
3869                         break;
3870
3871                 xdst = bundle[start_from - 1];
3872                 xdst->child_mtu_cached = mtu;
3873         }
3874
3875         return 1;
3876 }
3877
3878 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3879 {
3880         return dst_metric_advmss(xfrm_dst_path(dst));
3881 }
3882
3883 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3884 {
3885         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3886
3887         return mtu ? : dst_mtu(xfrm_dst_path(dst));
3888 }
3889
3890 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3891                                         const void *daddr)
3892 {
3893         while (dst->xfrm) {
3894                 const struct xfrm_state *xfrm = dst->xfrm;
3895
3896                 dst = xfrm_dst_child(dst);
3897
3898                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3899                         continue;
3900                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3901                         daddr = xfrm->coaddr;
3902                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3903                         daddr = &xfrm->id.daddr;
3904         }
3905         return daddr;
3906 }
3907
3908 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3909                                            struct sk_buff *skb,
3910                                            const void *daddr)
3911 {
3912         const struct dst_entry *path = xfrm_dst_path(dst);
3913
3914         if (!skb)
3915                 daddr = xfrm_get_dst_nexthop(dst, daddr);
3916         return path->ops->neigh_lookup(path, skb, daddr);
3917 }
3918
3919 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3920 {
3921         const struct dst_entry *path = xfrm_dst_path(dst);
3922
3923         daddr = xfrm_get_dst_nexthop(dst, daddr);
3924         path->ops->confirm_neigh(path, daddr);
3925 }
3926
3927 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3928 {
3929         int err = 0;
3930
3931         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3932                 return -EAFNOSUPPORT;
3933
3934         spin_lock(&xfrm_policy_afinfo_lock);
3935         if (unlikely(xfrm_policy_afinfo[family] != NULL))
3936                 err = -EEXIST;
3937         else {
3938                 struct dst_ops *dst_ops = afinfo->dst_ops;
3939                 if (likely(dst_ops->kmem_cachep == NULL))
3940                         dst_ops->kmem_cachep = xfrm_dst_cache;
3941                 if (likely(dst_ops->check == NULL))
3942                         dst_ops->check = xfrm_dst_check;
3943                 if (likely(dst_ops->default_advmss == NULL))
3944                         dst_ops->default_advmss = xfrm_default_advmss;
3945                 if (likely(dst_ops->mtu == NULL))
3946                         dst_ops->mtu = xfrm_mtu;
3947                 if (likely(dst_ops->negative_advice == NULL))
3948                         dst_ops->negative_advice = xfrm_negative_advice;
3949                 if (likely(dst_ops->link_failure == NULL))
3950                         dst_ops->link_failure = xfrm_link_failure;
3951                 if (likely(dst_ops->neigh_lookup == NULL))
3952                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
3953                 if (likely(!dst_ops->confirm_neigh))
3954                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
3955                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3956         }
3957         spin_unlock(&xfrm_policy_afinfo_lock);
3958
3959         return err;
3960 }
3961 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3962
3963 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3964 {
3965         struct dst_ops *dst_ops = afinfo->dst_ops;
3966         int i;
3967
3968         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3969                 if (xfrm_policy_afinfo[i] != afinfo)
3970                         continue;
3971                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3972                 break;
3973         }
3974
3975         synchronize_rcu();
3976
3977         dst_ops->kmem_cachep = NULL;
3978         dst_ops->check = NULL;
3979         dst_ops->negative_advice = NULL;
3980         dst_ops->link_failure = NULL;
3981 }
3982 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3983
3984 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3985 {
3986         spin_lock(&xfrm_if_cb_lock);
3987         rcu_assign_pointer(xfrm_if_cb, ifcb);
3988         spin_unlock(&xfrm_if_cb_lock);
3989 }
3990 EXPORT_SYMBOL(xfrm_if_register_cb);
3991
3992 void xfrm_if_unregister_cb(void)
3993 {
3994         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3995         synchronize_rcu();
3996 }
3997 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3998
3999 #ifdef CONFIG_XFRM_STATISTICS
4000 static int __net_init xfrm_statistics_init(struct net *net)
4001 {
4002         int rv;
4003         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4004         if (!net->mib.xfrm_statistics)
4005                 return -ENOMEM;
4006         rv = xfrm_proc_init(net);
4007         if (rv < 0)
4008                 free_percpu(net->mib.xfrm_statistics);
4009         return rv;
4010 }
4011
4012 static void xfrm_statistics_fini(struct net *net)
4013 {
4014         xfrm_proc_fini(net);
4015         free_percpu(net->mib.xfrm_statistics);
4016 }
4017 #else
4018 static int __net_init xfrm_statistics_init(struct net *net)
4019 {
4020         return 0;
4021 }
4022
4023 static void xfrm_statistics_fini(struct net *net)
4024 {
4025 }
4026 #endif
4027
4028 static int __net_init xfrm_policy_init(struct net *net)
4029 {
4030         unsigned int hmask, sz;
4031         int dir, err;
4032
4033         if (net_eq(net, &init_net)) {
4034                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4035                                            sizeof(struct xfrm_dst),
4036                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4037                                            NULL);
4038                 err = rhashtable_init(&xfrm_policy_inexact_table,
4039                                       &xfrm_pol_inexact_params);
4040                 BUG_ON(err);
4041         }
4042
4043         hmask = 8 - 1;
4044         sz = (hmask+1) * sizeof(struct hlist_head);
4045
4046         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4047         if (!net->xfrm.policy_byidx)
4048                 goto out_byidx;
4049         net->xfrm.policy_idx_hmask = hmask;
4050
4051         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4052                 struct xfrm_policy_hash *htab;
4053
4054                 net->xfrm.policy_count[dir] = 0;
4055                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4056                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4057
4058                 htab = &net->xfrm.policy_bydst[dir];
4059                 htab->table = xfrm_hash_alloc(sz);
4060                 if (!htab->table)
4061                         goto out_bydst;
4062                 htab->hmask = hmask;
4063                 htab->dbits4 = 32;
4064                 htab->sbits4 = 32;
4065                 htab->dbits6 = 128;
4066                 htab->sbits6 = 128;
4067         }
4068         net->xfrm.policy_hthresh.lbits4 = 32;
4069         net->xfrm.policy_hthresh.rbits4 = 32;
4070         net->xfrm.policy_hthresh.lbits6 = 128;
4071         net->xfrm.policy_hthresh.rbits6 = 128;
4072
4073         seqlock_init(&net->xfrm.policy_hthresh.lock);
4074
4075         INIT_LIST_HEAD(&net->xfrm.policy_all);
4076         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4077         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4078         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4079         return 0;
4080
4081 out_bydst:
4082         for (dir--; dir >= 0; dir--) {
4083                 struct xfrm_policy_hash *htab;
4084
4085                 htab = &net->xfrm.policy_bydst[dir];
4086                 xfrm_hash_free(htab->table, sz);
4087         }
4088         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4089 out_byidx:
4090         return -ENOMEM;
4091 }
4092
4093 static void xfrm_policy_fini(struct net *net)
4094 {
4095         struct xfrm_pol_inexact_bin *b, *t;
4096         unsigned int sz;
4097         int dir;
4098
4099         flush_work(&net->xfrm.policy_hash_work);
4100 #ifdef CONFIG_XFRM_SUB_POLICY
4101         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4102 #endif
4103         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4104
4105         WARN_ON(!list_empty(&net->xfrm.policy_all));
4106
4107         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4108                 struct xfrm_policy_hash *htab;
4109
4110                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4111
4112                 htab = &net->xfrm.policy_bydst[dir];
4113                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4114                 WARN_ON(!hlist_empty(htab->table));
4115                 xfrm_hash_free(htab->table, sz);
4116         }
4117
4118         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4119         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4120         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4121
4122         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4123         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4124                 __xfrm_policy_inexact_prune_bin(b, true);
4125         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4126 }
4127
4128 static int __net_init xfrm_net_init(struct net *net)
4129 {
4130         int rv;
4131
4132         /* Initialize the per-net locks here */
4133         spin_lock_init(&net->xfrm.xfrm_state_lock);
4134         spin_lock_init(&net->xfrm.xfrm_policy_lock);
4135         seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4136         mutex_init(&net->xfrm.xfrm_cfg_mutex);
4137         net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4138         net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4139         net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4140
4141         rv = xfrm_statistics_init(net);
4142         if (rv < 0)
4143                 goto out_statistics;
4144         rv = xfrm_state_init(net);
4145         if (rv < 0)
4146                 goto out_state;
4147         rv = xfrm_policy_init(net);
4148         if (rv < 0)
4149                 goto out_policy;
4150         rv = xfrm_sysctl_init(net);
4151         if (rv < 0)
4152                 goto out_sysctl;
4153
4154         return 0;
4155
4156 out_sysctl:
4157         xfrm_policy_fini(net);
4158 out_policy:
4159         xfrm_state_fini(net);
4160 out_state:
4161         xfrm_statistics_fini(net);
4162 out_statistics:
4163         return rv;
4164 }
4165
4166 static void __net_exit xfrm_net_exit(struct net *net)
4167 {
4168         xfrm_sysctl_fini(net);
4169         xfrm_policy_fini(net);
4170         xfrm_state_fini(net);
4171         xfrm_statistics_fini(net);
4172 }
4173
4174 static struct pernet_operations __net_initdata xfrm_net_ops = {
4175         .init = xfrm_net_init,
4176         .exit = xfrm_net_exit,
4177 };
4178
4179 void __init xfrm_init(void)
4180 {
4181         register_pernet_subsys(&xfrm_net_ops);
4182         xfrm_dev_init();
4183         xfrm_input_init();
4184
4185 #ifdef CONFIG_XFRM_ESPINTCP
4186         espintcp_init();
4187 #endif
4188 }
4189
4190 #ifdef CONFIG_AUDITSYSCALL
4191 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4192                                          struct audit_buffer *audit_buf)
4193 {
4194         struct xfrm_sec_ctx *ctx = xp->security;
4195         struct xfrm_selector *sel = &xp->selector;
4196
4197         if (ctx)
4198                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4199                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4200
4201         switch (sel->family) {
4202         case AF_INET:
4203                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4204                 if (sel->prefixlen_s != 32)
4205                         audit_log_format(audit_buf, " src_prefixlen=%d",
4206                                          sel->prefixlen_s);
4207                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4208                 if (sel->prefixlen_d != 32)
4209                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4210                                          sel->prefixlen_d);
4211                 break;
4212         case AF_INET6:
4213                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4214                 if (sel->prefixlen_s != 128)
4215                         audit_log_format(audit_buf, " src_prefixlen=%d",
4216                                          sel->prefixlen_s);
4217                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4218                 if (sel->prefixlen_d != 128)
4219                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4220                                          sel->prefixlen_d);
4221                 break;
4222         }
4223 }
4224
4225 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4226 {
4227         struct audit_buffer *audit_buf;
4228
4229         audit_buf = xfrm_audit_start("SPD-add");
4230         if (audit_buf == NULL)
4231                 return;
4232         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4233         audit_log_format(audit_buf, " res=%u", result);
4234         xfrm_audit_common_policyinfo(xp, audit_buf);
4235         audit_log_end(audit_buf);
4236 }
4237 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4238
4239 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4240                               bool task_valid)
4241 {
4242         struct audit_buffer *audit_buf;
4243
4244         audit_buf = xfrm_audit_start("SPD-delete");
4245         if (audit_buf == NULL)
4246                 return;
4247         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4248         audit_log_format(audit_buf, " res=%u", result);
4249         xfrm_audit_common_policyinfo(xp, audit_buf);
4250         audit_log_end(audit_buf);
4251 }
4252 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4253 #endif
4254
4255 #ifdef CONFIG_XFRM_MIGRATE
4256 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4257                                         const struct xfrm_selector *sel_tgt)
4258 {
4259         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4260                 if (sel_tgt->family == sel_cmp->family &&
4261                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4262                                     sel_cmp->family) &&
4263                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4264                                     sel_cmp->family) &&
4265                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4266                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4267                         return true;
4268                 }
4269         } else {
4270                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4271                         return true;
4272                 }
4273         }
4274         return false;
4275 }
4276
4277 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4278                                                     u8 dir, u8 type, struct net *net, u32 if_id)
4279 {
4280         struct xfrm_policy *pol, *ret = NULL;
4281         struct hlist_head *chain;
4282         u32 priority = ~0U;
4283
4284         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4285         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4286         hlist_for_each_entry(pol, chain, bydst) {
4287                 if ((if_id == 0 || pol->if_id == if_id) &&
4288                     xfrm_migrate_selector_match(sel, &pol->selector) &&
4289                     pol->type == type) {
4290                         ret = pol;
4291                         priority = ret->priority;
4292                         break;
4293                 }
4294         }
4295         chain = &net->xfrm.policy_inexact[dir];
4296         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4297                 if ((pol->priority >= priority) && ret)
4298                         break;
4299
4300                 if ((if_id == 0 || pol->if_id == if_id) &&
4301                     xfrm_migrate_selector_match(sel, &pol->selector) &&
4302                     pol->type == type) {
4303                         ret = pol;
4304                         break;
4305                 }
4306         }
4307
4308         xfrm_pol_hold(ret);
4309
4310         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4311
4312         return ret;
4313 }
4314
4315 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4316 {
4317         int match = 0;
4318
4319         if (t->mode == m->mode && t->id.proto == m->proto &&
4320             (m->reqid == 0 || t->reqid == m->reqid)) {
4321                 switch (t->mode) {
4322                 case XFRM_MODE_TUNNEL:
4323                 case XFRM_MODE_BEET:
4324                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4325                                             m->old_family) &&
4326                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4327                                             m->old_family)) {
4328                                 match = 1;
4329                         }
4330                         break;
4331                 case XFRM_MODE_TRANSPORT:
4332                         /* in case of transport mode, template does not store
4333                            any IP addresses, hence we just compare mode and
4334                            protocol */
4335                         match = 1;
4336                         break;
4337                 default:
4338                         break;
4339                 }
4340         }
4341         return match;
4342 }
4343
4344 /* update endpoint address(es) of template(s) */
4345 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4346                                struct xfrm_migrate *m, int num_migrate)
4347 {
4348         struct xfrm_migrate *mp;
4349         int i, j, n = 0;
4350
4351         write_lock_bh(&pol->lock);
4352         if (unlikely(pol->walk.dead)) {
4353                 /* target policy has been deleted */
4354                 write_unlock_bh(&pol->lock);
4355                 return -ENOENT;
4356         }
4357
4358         for (i = 0; i < pol->xfrm_nr; i++) {
4359                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4360                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4361                                 continue;
4362                         n++;
4363                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4364                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4365                                 continue;
4366                         /* update endpoints */
4367                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4368                                sizeof(pol->xfrm_vec[i].id.daddr));
4369                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4370                                sizeof(pol->xfrm_vec[i].saddr));
4371                         pol->xfrm_vec[i].encap_family = mp->new_family;
4372                         /* flush bundles */
4373                         atomic_inc(&pol->genid);
4374                 }
4375         }
4376
4377         write_unlock_bh(&pol->lock);
4378
4379         if (!n)
4380                 return -ENODATA;
4381
4382         return 0;
4383 }
4384
4385 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4386 {
4387         int i, j;
4388
4389         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4390                 return -EINVAL;
4391
4392         for (i = 0; i < num_migrate; i++) {
4393                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4394                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4395                         return -EINVAL;
4396
4397                 /* check if there is any duplicated entry */
4398                 for (j = i + 1; j < num_migrate; j++) {
4399                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4400                                     sizeof(m[i].old_daddr)) &&
4401                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4402                                     sizeof(m[i].old_saddr)) &&
4403                             m[i].proto == m[j].proto &&
4404                             m[i].mode == m[j].mode &&
4405                             m[i].reqid == m[j].reqid &&
4406                             m[i].old_family == m[j].old_family)
4407                                 return -EINVAL;
4408                 }
4409         }
4410
4411         return 0;
4412 }
4413
4414 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4415                  struct xfrm_migrate *m, int num_migrate,
4416                  struct xfrm_kmaddress *k, struct net *net,
4417                  struct xfrm_encap_tmpl *encap, u32 if_id)
4418 {
4419         int i, err, nx_cur = 0, nx_new = 0;
4420         struct xfrm_policy *pol = NULL;
4421         struct xfrm_state *x, *xc;
4422         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4423         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4424         struct xfrm_migrate *mp;
4425
4426         /* Stage 0 - sanity checks */
4427         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4428                 goto out;
4429
4430         if (dir >= XFRM_POLICY_MAX) {
4431                 err = -EINVAL;
4432                 goto out;
4433         }
4434
4435         /* Stage 1 - find policy */
4436         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id)) == NULL) {
4437                 err = -ENOENT;
4438                 goto out;
4439         }
4440
4441         /* Stage 2 - find and update state(s) */
4442         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4443                 if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4444                         x_cur[nx_cur] = x;
4445                         nx_cur++;
4446                         xc = xfrm_state_migrate(x, mp, encap);
4447                         if (xc) {
4448                                 x_new[nx_new] = xc;
4449                                 nx_new++;
4450                         } else {
4451                                 err = -ENODATA;
4452                                 goto restore_state;
4453                         }
4454                 }
4455         }
4456
4457         /* Stage 3 - update policy */
4458         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4459                 goto restore_state;
4460
4461         /* Stage 4 - delete old state(s) */
4462         if (nx_cur) {
4463                 xfrm_states_put(x_cur, nx_cur);
4464                 xfrm_states_delete(x_cur, nx_cur);
4465         }
4466
4467         /* Stage 5 - announce */
4468         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4469
4470         xfrm_pol_put(pol);
4471
4472         return 0;
4473 out:
4474         return err;
4475
4476 restore_state:
4477         if (pol)
4478                 xfrm_pol_put(pol);
4479         if (nx_cur)
4480                 xfrm_states_put(x_cur, nx_cur);
4481         if (nx_new)
4482                 xfrm_states_delete(x_new, nx_new);
4483
4484         return err;
4485 }
4486 EXPORT_SYMBOL(xfrm_migrate);
4487 #endif