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->walk.dead || !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                 if (policy->walk.dead)
1259                         continue;
1260
1261                 dir = xfrm_policy_id2dir(policy->index);
1262                 if (dir >= XFRM_POLICY_MAX)
1263                         continue;
1264
1265                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1266                         if (policy->family == AF_INET) {
1267                                 dbits = rbits4;
1268                                 sbits = lbits4;
1269                         } else {
1270                                 dbits = rbits6;
1271                                 sbits = lbits6;
1272                         }
1273                 } else {
1274                         if (policy->family == AF_INET) {
1275                                 dbits = lbits4;
1276                                 sbits = rbits4;
1277                         } else {
1278                                 dbits = lbits6;
1279                                 sbits = rbits6;
1280                         }
1281                 }
1282
1283                 if (policy->selector.prefixlen_d < dbits ||
1284                     policy->selector.prefixlen_s < sbits)
1285                         continue;
1286
1287                 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1288                 if (!bin)
1289                         goto out_unlock;
1290
1291                 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1292                         goto out_unlock;
1293         }
1294
1295         /* reset the bydst and inexact table in all directions */
1296         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1297                 struct hlist_node *n;
1298
1299                 hlist_for_each_entry_safe(policy, n,
1300                                           &net->xfrm.policy_inexact[dir],
1301                                           bydst_inexact_list) {
1302                         hlist_del_rcu(&policy->bydst);
1303                         hlist_del_init(&policy->bydst_inexact_list);
1304                 }
1305
1306                 hmask = net->xfrm.policy_bydst[dir].hmask;
1307                 odst = net->xfrm.policy_bydst[dir].table;
1308                 for (i = hmask; i >= 0; i--) {
1309                         hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1310                                 hlist_del_rcu(&policy->bydst);
1311                 }
1312                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1313                         /* dir out => dst = remote, src = local */
1314                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1315                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1316                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1317                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1318                 } else {
1319                         /* dir in/fwd => dst = local, src = remote */
1320                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1321                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1322                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1323                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1324                 }
1325         }
1326
1327         /* re-insert all policies by order of creation */
1328         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1329                 if (policy->walk.dead)
1330                         continue;
1331                 dir = xfrm_policy_id2dir(policy->index);
1332                 if (dir >= XFRM_POLICY_MAX) {
1333                         /* skip socket policies */
1334                         continue;
1335                 }
1336                 newpos = NULL;
1337                 chain = policy_hash_bysel(net, &policy->selector,
1338                                           policy->family, dir);
1339
1340                 if (!chain) {
1341                         void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1342
1343                         WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1344                         continue;
1345                 }
1346
1347                 hlist_for_each_entry(pol, chain, bydst) {
1348                         if (policy->priority >= pol->priority)
1349                                 newpos = &pol->bydst;
1350                         else
1351                                 break;
1352                 }
1353                 if (newpos)
1354                         hlist_add_behind_rcu(&policy->bydst, newpos);
1355                 else
1356                         hlist_add_head_rcu(&policy->bydst, chain);
1357         }
1358
1359 out_unlock:
1360         __xfrm_policy_inexact_flush(net);
1361         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1362         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1363
1364         mutex_unlock(&hash_resize_mutex);
1365 }
1366
1367 void xfrm_policy_hash_rebuild(struct net *net)
1368 {
1369         schedule_work(&net->xfrm.policy_hthresh.work);
1370 }
1371 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1372
1373 /* Generate new index... KAME seems to generate them ordered by cost
1374  * of an absolute inpredictability of ordering of rules. This will not pass. */
1375 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1376 {
1377         for (;;) {
1378                 struct hlist_head *list;
1379                 struct xfrm_policy *p;
1380                 u32 idx;
1381                 int found;
1382
1383                 if (!index) {
1384                         idx = (net->xfrm.idx_generator | dir);
1385                         net->xfrm.idx_generator += 8;
1386                 } else {
1387                         idx = index;
1388                         index = 0;
1389                 }
1390
1391                 if (idx == 0)
1392                         idx = 8;
1393                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1394                 found = 0;
1395                 hlist_for_each_entry(p, list, byidx) {
1396                         if (p->index == idx) {
1397                                 found = 1;
1398                                 break;
1399                         }
1400                 }
1401                 if (!found)
1402                         return idx;
1403         }
1404 }
1405
1406 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1407 {
1408         u32 *p1 = (u32 *) s1;
1409         u32 *p2 = (u32 *) s2;
1410         int len = sizeof(struct xfrm_selector) / sizeof(u32);
1411         int i;
1412
1413         for (i = 0; i < len; i++) {
1414                 if (p1[i] != p2[i])
1415                         return 1;
1416         }
1417
1418         return 0;
1419 }
1420
1421 static void xfrm_policy_requeue(struct xfrm_policy *old,
1422                                 struct xfrm_policy *new)
1423 {
1424         struct xfrm_policy_queue *pq = &old->polq;
1425         struct sk_buff_head list;
1426
1427         if (skb_queue_empty(&pq->hold_queue))
1428                 return;
1429
1430         __skb_queue_head_init(&list);
1431
1432         spin_lock_bh(&pq->hold_queue.lock);
1433         skb_queue_splice_init(&pq->hold_queue, &list);
1434         if (del_timer(&pq->hold_timer))
1435                 xfrm_pol_put(old);
1436         spin_unlock_bh(&pq->hold_queue.lock);
1437
1438         pq = &new->polq;
1439
1440         spin_lock_bh(&pq->hold_queue.lock);
1441         skb_queue_splice(&list, &pq->hold_queue);
1442         pq->timeout = XFRM_QUEUE_TMO_MIN;
1443         if (!mod_timer(&pq->hold_timer, jiffies))
1444                 xfrm_pol_hold(new);
1445         spin_unlock_bh(&pq->hold_queue.lock);
1446 }
1447
1448 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1449                                           struct xfrm_policy *pol)
1450 {
1451         return mark->v == pol->mark.v && mark->m == pol->mark.m;
1452 }
1453
1454 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1455 {
1456         const struct xfrm_pol_inexact_key *k = data;
1457         u32 a = k->type << 24 | k->dir << 16 | k->family;
1458
1459         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1460                             seed);
1461 }
1462
1463 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1464 {
1465         const struct xfrm_pol_inexact_bin *b = data;
1466
1467         return xfrm_pol_bin_key(&b->k, 0, seed);
1468 }
1469
1470 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1471                             const void *ptr)
1472 {
1473         const struct xfrm_pol_inexact_key *key = arg->key;
1474         const struct xfrm_pol_inexact_bin *b = ptr;
1475         int ret;
1476
1477         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1478                 return -1;
1479
1480         ret = b->k.dir ^ key->dir;
1481         if (ret)
1482                 return ret;
1483
1484         ret = b->k.type ^ key->type;
1485         if (ret)
1486                 return ret;
1487
1488         ret = b->k.family ^ key->family;
1489         if (ret)
1490                 return ret;
1491
1492         return b->k.if_id ^ key->if_id;
1493 }
1494
1495 static const struct rhashtable_params xfrm_pol_inexact_params = {
1496         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1497         .hashfn                 = xfrm_pol_bin_key,
1498         .obj_hashfn             = xfrm_pol_bin_obj,
1499         .obj_cmpfn              = xfrm_pol_bin_cmp,
1500         .automatic_shrinking    = true,
1501 };
1502
1503 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1504                                             struct xfrm_policy *policy)
1505 {
1506         struct xfrm_policy *pol, *delpol = NULL;
1507         struct hlist_node *newpos = NULL;
1508         int i = 0;
1509
1510         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1511                 if (pol->type == policy->type &&
1512                     pol->if_id == policy->if_id &&
1513                     !selector_cmp(&pol->selector, &policy->selector) &&
1514                     xfrm_policy_mark_match(&policy->mark, pol) &&
1515                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1516                     !WARN_ON(delpol)) {
1517                         delpol = pol;
1518                         if (policy->priority > pol->priority)
1519                                 continue;
1520                 } else if (policy->priority >= pol->priority) {
1521                         newpos = &pol->bydst_inexact_list;
1522                         continue;
1523                 }
1524                 if (delpol)
1525                         break;
1526         }
1527
1528         if (newpos)
1529                 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1530         else
1531                 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1532
1533         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1534                 pol->pos = i;
1535                 i++;
1536         }
1537 }
1538
1539 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1540                                                    struct xfrm_policy *policy,
1541                                                    bool excl)
1542 {
1543         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1544
1545         hlist_for_each_entry(pol, chain, bydst) {
1546                 if (pol->type == policy->type &&
1547                     pol->if_id == policy->if_id &&
1548                     !selector_cmp(&pol->selector, &policy->selector) &&
1549                     xfrm_policy_mark_match(&policy->mark, pol) &&
1550                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1551                     !WARN_ON(delpol)) {
1552                         if (excl)
1553                                 return ERR_PTR(-EEXIST);
1554                         delpol = pol;
1555                         if (policy->priority > pol->priority)
1556                                 continue;
1557                 } else if (policy->priority >= pol->priority) {
1558                         newpos = pol;
1559                         continue;
1560                 }
1561                 if (delpol)
1562                         break;
1563         }
1564
1565         if (newpos)
1566                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1567         else
1568                 hlist_add_head_rcu(&policy->bydst, chain);
1569
1570         return delpol;
1571 }
1572
1573 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1574 {
1575         struct net *net = xp_net(policy);
1576         struct xfrm_policy *delpol;
1577         struct hlist_head *chain;
1578
1579         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1580         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1581         if (chain)
1582                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1583         else
1584                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1585
1586         if (IS_ERR(delpol)) {
1587                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1588                 return PTR_ERR(delpol);
1589         }
1590
1591         __xfrm_policy_link(policy, dir);
1592
1593         /* After previous checking, family can either be AF_INET or AF_INET6 */
1594         if (policy->family == AF_INET)
1595                 rt_genid_bump_ipv4(net);
1596         else
1597                 rt_genid_bump_ipv6(net);
1598
1599         if (delpol) {
1600                 xfrm_policy_requeue(delpol, policy);
1601                 __xfrm_policy_unlink(delpol, dir);
1602         }
1603         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1604         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1605         policy->curlft.add_time = ktime_get_real_seconds();
1606         policy->curlft.use_time = 0;
1607         if (!mod_timer(&policy->timer, jiffies + HZ))
1608                 xfrm_pol_hold(policy);
1609         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1610
1611         if (delpol)
1612                 xfrm_policy_kill(delpol);
1613         else if (xfrm_bydst_should_resize(net, dir, NULL))
1614                 schedule_work(&net->xfrm.policy_hash_work);
1615
1616         return 0;
1617 }
1618 EXPORT_SYMBOL(xfrm_policy_insert);
1619
1620 static struct xfrm_policy *
1621 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1622                         u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1623                         struct xfrm_sec_ctx *ctx)
1624 {
1625         struct xfrm_policy *pol;
1626
1627         if (!chain)
1628                 return NULL;
1629
1630         hlist_for_each_entry(pol, chain, bydst) {
1631                 if (pol->type == type &&
1632                     pol->if_id == if_id &&
1633                     xfrm_policy_mark_match(mark, pol) &&
1634                     !selector_cmp(sel, &pol->selector) &&
1635                     xfrm_sec_ctx_match(ctx, pol->security))
1636                         return pol;
1637         }
1638
1639         return NULL;
1640 }
1641
1642 struct xfrm_policy *
1643 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1644                       u8 type, int dir, struct xfrm_selector *sel,
1645                       struct xfrm_sec_ctx *ctx, int delete, int *err)
1646 {
1647         struct xfrm_pol_inexact_bin *bin = NULL;
1648         struct xfrm_policy *pol, *ret = NULL;
1649         struct hlist_head *chain;
1650
1651         *err = 0;
1652         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1653         chain = policy_hash_bysel(net, sel, sel->family, dir);
1654         if (!chain) {
1655                 struct xfrm_pol_inexact_candidates cand;
1656                 int i;
1657
1658                 bin = xfrm_policy_inexact_lookup(net, type,
1659                                                  sel->family, dir, if_id);
1660                 if (!bin) {
1661                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1662                         return NULL;
1663                 }
1664
1665                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1666                                                          &sel->saddr,
1667                                                          &sel->daddr)) {
1668                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1669                         return NULL;
1670                 }
1671
1672                 pol = NULL;
1673                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1674                         struct xfrm_policy *tmp;
1675
1676                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1677                                                       if_id, type, dir,
1678                                                       sel, ctx);
1679                         if (!tmp)
1680                                 continue;
1681
1682                         if (!pol || tmp->pos < pol->pos)
1683                                 pol = tmp;
1684                 }
1685         } else {
1686                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1687                                               sel, ctx);
1688         }
1689
1690         if (pol) {
1691                 xfrm_pol_hold(pol);
1692                 if (delete) {
1693                         *err = security_xfrm_policy_delete(pol->security);
1694                         if (*err) {
1695                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1696                                 return pol;
1697                         }
1698                         __xfrm_policy_unlink(pol, dir);
1699                 }
1700                 ret = pol;
1701         }
1702         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1703
1704         if (ret && delete)
1705                 xfrm_policy_kill(ret);
1706         if (bin && delete)
1707                 xfrm_policy_inexact_prune_bin(bin);
1708         return ret;
1709 }
1710 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1711
1712 struct xfrm_policy *
1713 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1714                  u8 type, int dir, u32 id, int delete, int *err)
1715 {
1716         struct xfrm_policy *pol, *ret;
1717         struct hlist_head *chain;
1718
1719         *err = -ENOENT;
1720         if (xfrm_policy_id2dir(id) != dir)
1721                 return NULL;
1722
1723         *err = 0;
1724         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1725         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1726         ret = NULL;
1727         hlist_for_each_entry(pol, chain, byidx) {
1728                 if (pol->type == type && pol->index == id &&
1729                     pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1730                         xfrm_pol_hold(pol);
1731                         if (delete) {
1732                                 *err = security_xfrm_policy_delete(
1733                                                                 pol->security);
1734                                 if (*err) {
1735                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1736                                         return pol;
1737                                 }
1738                                 __xfrm_policy_unlink(pol, dir);
1739                         }
1740                         ret = pol;
1741                         break;
1742                 }
1743         }
1744         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1745
1746         if (ret && delete)
1747                 xfrm_policy_kill(ret);
1748         return ret;
1749 }
1750 EXPORT_SYMBOL(xfrm_policy_byid);
1751
1752 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1753 static inline int
1754 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1755 {
1756         struct xfrm_policy *pol;
1757         int err = 0;
1758
1759         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1760                 if (pol->walk.dead ||
1761                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1762                     pol->type != type)
1763                         continue;
1764
1765                 err = security_xfrm_policy_delete(pol->security);
1766                 if (err) {
1767                         xfrm_audit_policy_delete(pol, 0, task_valid);
1768                         return err;
1769                 }
1770         }
1771         return err;
1772 }
1773 #else
1774 static inline int
1775 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1776 {
1777         return 0;
1778 }
1779 #endif
1780
1781 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1782 {
1783         int dir, err = 0, cnt = 0;
1784         struct xfrm_policy *pol;
1785
1786         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1787
1788         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1789         if (err)
1790                 goto out;
1791
1792 again:
1793         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1794                 if (pol->walk.dead)
1795                         continue;
1796
1797                 dir = xfrm_policy_id2dir(pol->index);
1798                 if (dir >= XFRM_POLICY_MAX ||
1799                     pol->type != type)
1800                         continue;
1801
1802                 __xfrm_policy_unlink(pol, dir);
1803                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1804                 cnt++;
1805                 xfrm_audit_policy_delete(pol, 1, task_valid);
1806                 xfrm_policy_kill(pol);
1807                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1808                 goto again;
1809         }
1810         if (cnt)
1811                 __xfrm_policy_inexact_flush(net);
1812         else
1813                 err = -ESRCH;
1814 out:
1815         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1816         return err;
1817 }
1818 EXPORT_SYMBOL(xfrm_policy_flush);
1819
1820 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1821                      int (*func)(struct xfrm_policy *, int, int, void*),
1822                      void *data)
1823 {
1824         struct xfrm_policy *pol;
1825         struct xfrm_policy_walk_entry *x;
1826         int error = 0;
1827
1828         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1829             walk->type != XFRM_POLICY_TYPE_ANY)
1830                 return -EINVAL;
1831
1832         if (list_empty(&walk->walk.all) && walk->seq != 0)
1833                 return 0;
1834
1835         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1836         if (list_empty(&walk->walk.all))
1837                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1838         else
1839                 x = list_first_entry(&walk->walk.all,
1840                                      struct xfrm_policy_walk_entry, all);
1841
1842         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1843                 if (x->dead)
1844                         continue;
1845                 pol = container_of(x, struct xfrm_policy, walk);
1846                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1847                     walk->type != pol->type)
1848                         continue;
1849                 error = func(pol, xfrm_policy_id2dir(pol->index),
1850                              walk->seq, data);
1851                 if (error) {
1852                         list_move_tail(&walk->walk.all, &x->all);
1853                         goto out;
1854                 }
1855                 walk->seq++;
1856         }
1857         if (walk->seq == 0) {
1858                 error = -ENOENT;
1859                 goto out;
1860         }
1861         list_del_init(&walk->walk.all);
1862 out:
1863         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1864         return error;
1865 }
1866 EXPORT_SYMBOL(xfrm_policy_walk);
1867
1868 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1869 {
1870         INIT_LIST_HEAD(&walk->walk.all);
1871         walk->walk.dead = 1;
1872         walk->type = type;
1873         walk->seq = 0;
1874 }
1875 EXPORT_SYMBOL(xfrm_policy_walk_init);
1876
1877 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1878 {
1879         if (list_empty(&walk->walk.all))
1880                 return;
1881
1882         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1883         list_del(&walk->walk.all);
1884         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1885 }
1886 EXPORT_SYMBOL(xfrm_policy_walk_done);
1887
1888 /*
1889  * Find policy to apply to this flow.
1890  *
1891  * Returns 0 if policy found, else an -errno.
1892  */
1893 static int xfrm_policy_match(const struct xfrm_policy *pol,
1894                              const struct flowi *fl,
1895                              u8 type, u16 family, u32 if_id)
1896 {
1897         const struct xfrm_selector *sel = &pol->selector;
1898         int ret = -ESRCH;
1899         bool match;
1900
1901         if (pol->family != family ||
1902             pol->if_id != if_id ||
1903             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1904             pol->type != type)
1905                 return ret;
1906
1907         match = xfrm_selector_match(sel, fl, family);
1908         if (match)
1909                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1910         return ret;
1911 }
1912
1913 static struct xfrm_pol_inexact_node *
1914 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1915                                 seqcount_spinlock_t *count,
1916                                 const xfrm_address_t *addr, u16 family)
1917 {
1918         const struct rb_node *parent;
1919         int seq;
1920
1921 again:
1922         seq = read_seqcount_begin(count);
1923
1924         parent = rcu_dereference_raw(r->rb_node);
1925         while (parent) {
1926                 struct xfrm_pol_inexact_node *node;
1927                 int delta;
1928
1929                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1930
1931                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1932                                                node->prefixlen, family);
1933                 if (delta < 0) {
1934                         parent = rcu_dereference_raw(parent->rb_left);
1935                         continue;
1936                 } else if (delta > 0) {
1937                         parent = rcu_dereference_raw(parent->rb_right);
1938                         continue;
1939                 }
1940
1941                 return node;
1942         }
1943
1944         if (read_seqcount_retry(count, seq))
1945                 goto again;
1946
1947         return NULL;
1948 }
1949
1950 static bool
1951 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1952                                     struct xfrm_pol_inexact_bin *b,
1953                                     const xfrm_address_t *saddr,
1954                                     const xfrm_address_t *daddr)
1955 {
1956         struct xfrm_pol_inexact_node *n;
1957         u16 family;
1958
1959         if (!b)
1960                 return false;
1961
1962         family = b->k.family;
1963         memset(cand, 0, sizeof(*cand));
1964         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1965
1966         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1967                                             family);
1968         if (n) {
1969                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1970                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1971                                                     family);
1972                 if (n)
1973                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1974         }
1975
1976         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1977                                             family);
1978         if (n)
1979                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1980
1981         return true;
1982 }
1983
1984 static struct xfrm_pol_inexact_bin *
1985 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1986                                u8 dir, u32 if_id)
1987 {
1988         struct xfrm_pol_inexact_key k = {
1989                 .family = family,
1990                 .type = type,
1991                 .dir = dir,
1992                 .if_id = if_id,
1993         };
1994
1995         write_pnet(&k.net, net);
1996
1997         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1998                                  xfrm_pol_inexact_params);
1999 }
2000
2001 static struct xfrm_pol_inexact_bin *
2002 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2003                            u8 dir, u32 if_id)
2004 {
2005         struct xfrm_pol_inexact_bin *bin;
2006
2007         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2008
2009         rcu_read_lock();
2010         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2011         rcu_read_unlock();
2012
2013         return bin;
2014 }
2015
2016 static struct xfrm_policy *
2017 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2018                               struct xfrm_policy *prefer,
2019                               const struct flowi *fl,
2020                               u8 type, u16 family, u32 if_id)
2021 {
2022         u32 priority = prefer ? prefer->priority : ~0u;
2023         struct xfrm_policy *pol;
2024
2025         if (!chain)
2026                 return NULL;
2027
2028         hlist_for_each_entry_rcu(pol, chain, bydst) {
2029                 int err;
2030
2031                 if (pol->priority > priority)
2032                         break;
2033
2034                 err = xfrm_policy_match(pol, fl, type, family, if_id);
2035                 if (err) {
2036                         if (err != -ESRCH)
2037                                 return ERR_PTR(err);
2038
2039                         continue;
2040                 }
2041
2042                 if (prefer) {
2043                         /* matches.  Is it older than *prefer? */
2044                         if (pol->priority == priority &&
2045                             prefer->pos < pol->pos)
2046                                 return prefer;
2047                 }
2048
2049                 return pol;
2050         }
2051
2052         return NULL;
2053 }
2054
2055 static struct xfrm_policy *
2056 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2057                             struct xfrm_policy *prefer,
2058                             const struct flowi *fl,
2059                             u8 type, u16 family, u32 if_id)
2060 {
2061         struct xfrm_policy *tmp;
2062         int i;
2063
2064         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2065                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2066                                                     prefer,
2067                                                     fl, type, family, if_id);
2068                 if (!tmp)
2069                         continue;
2070
2071                 if (IS_ERR(tmp))
2072                         return tmp;
2073                 prefer = tmp;
2074         }
2075
2076         return prefer;
2077 }
2078
2079 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2080                                                      const struct flowi *fl,
2081                                                      u16 family, u8 dir,
2082                                                      u32 if_id)
2083 {
2084         struct xfrm_pol_inexact_candidates cand;
2085         const xfrm_address_t *daddr, *saddr;
2086         struct xfrm_pol_inexact_bin *bin;
2087         struct xfrm_policy *pol, *ret;
2088         struct hlist_head *chain;
2089         unsigned int sequence;
2090         int err;
2091
2092         daddr = xfrm_flowi_daddr(fl, family);
2093         saddr = xfrm_flowi_saddr(fl, family);
2094         if (unlikely(!daddr || !saddr))
2095                 return NULL;
2096
2097         rcu_read_lock();
2098  retry:
2099         do {
2100                 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2101                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2102         } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2103
2104         ret = NULL;
2105         hlist_for_each_entry_rcu(pol, chain, bydst) {
2106                 err = xfrm_policy_match(pol, fl, type, family, if_id);
2107                 if (err) {
2108                         if (err == -ESRCH)
2109                                 continue;
2110                         else {
2111                                 ret = ERR_PTR(err);
2112                                 goto fail;
2113                         }
2114                 } else {
2115                         ret = pol;
2116                         break;
2117                 }
2118         }
2119         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2120         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2121                                                          daddr))
2122                 goto skip_inexact;
2123
2124         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2125                                           family, if_id);
2126         if (pol) {
2127                 ret = pol;
2128                 if (IS_ERR(pol))
2129                         goto fail;
2130         }
2131
2132 skip_inexact:
2133         if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2134                 goto retry;
2135
2136         if (ret && !xfrm_pol_hold_rcu(ret))
2137                 goto retry;
2138 fail:
2139         rcu_read_unlock();
2140
2141         return ret;
2142 }
2143
2144 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2145                                               const struct flowi *fl,
2146                                               u16 family, u8 dir, u32 if_id)
2147 {
2148 #ifdef CONFIG_XFRM_SUB_POLICY
2149         struct xfrm_policy *pol;
2150
2151         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2152                                         dir, if_id);
2153         if (pol != NULL)
2154                 return pol;
2155 #endif
2156         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2157                                          dir, if_id);
2158 }
2159
2160 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2161                                                  const struct flowi *fl,
2162                                                  u16 family, u32 if_id)
2163 {
2164         struct xfrm_policy *pol;
2165
2166         rcu_read_lock();
2167  again:
2168         pol = rcu_dereference(sk->sk_policy[dir]);
2169         if (pol != NULL) {
2170                 bool match;
2171                 int err = 0;
2172
2173                 if (pol->family != family) {
2174                         pol = NULL;
2175                         goto out;
2176                 }
2177
2178                 match = xfrm_selector_match(&pol->selector, fl, family);
2179                 if (match) {
2180                         if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v ||
2181                             pol->if_id != if_id) {
2182                                 pol = NULL;
2183                                 goto out;
2184                         }
2185                         err = security_xfrm_policy_lookup(pol->security,
2186                                                       fl->flowi_secid);
2187                         if (!err) {
2188                                 if (!xfrm_pol_hold_rcu(pol))
2189                                         goto again;
2190                         } else if (err == -ESRCH) {
2191                                 pol = NULL;
2192                         } else {
2193                                 pol = ERR_PTR(err);
2194                         }
2195                 } else
2196                         pol = NULL;
2197         }
2198 out:
2199         rcu_read_unlock();
2200         return pol;
2201 }
2202
2203 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2204 {
2205         struct net *net = xp_net(pol);
2206
2207         list_add(&pol->walk.all, &net->xfrm.policy_all);
2208         net->xfrm.policy_count[dir]++;
2209         xfrm_pol_hold(pol);
2210 }
2211
2212 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2213                                                 int dir)
2214 {
2215         struct net *net = xp_net(pol);
2216
2217         if (list_empty(&pol->walk.all))
2218                 return NULL;
2219
2220         /* Socket policies are not hashed. */
2221         if (!hlist_unhashed(&pol->bydst)) {
2222                 hlist_del_rcu(&pol->bydst);
2223                 hlist_del_init(&pol->bydst_inexact_list);
2224                 hlist_del(&pol->byidx);
2225         }
2226
2227         list_del_init(&pol->walk.all);
2228         net->xfrm.policy_count[dir]--;
2229
2230         return pol;
2231 }
2232
2233 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2234 {
2235         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2236 }
2237
2238 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2239 {
2240         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2241 }
2242
2243 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2244 {
2245         struct net *net = xp_net(pol);
2246
2247         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2248         pol = __xfrm_policy_unlink(pol, dir);
2249         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2250         if (pol) {
2251                 xfrm_policy_kill(pol);
2252                 return 0;
2253         }
2254         return -ENOENT;
2255 }
2256 EXPORT_SYMBOL(xfrm_policy_delete);
2257
2258 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2259 {
2260         struct net *net = sock_net(sk);
2261         struct xfrm_policy *old_pol;
2262
2263 #ifdef CONFIG_XFRM_SUB_POLICY
2264         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2265                 return -EINVAL;
2266 #endif
2267
2268         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2269         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2270                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2271         if (pol) {
2272                 pol->curlft.add_time = ktime_get_real_seconds();
2273                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2274                 xfrm_sk_policy_link(pol, dir);
2275         }
2276         rcu_assign_pointer(sk->sk_policy[dir], pol);
2277         if (old_pol) {
2278                 if (pol)
2279                         xfrm_policy_requeue(old_pol, pol);
2280
2281                 /* Unlinking succeeds always. This is the only function
2282                  * allowed to delete or replace socket policy.
2283                  */
2284                 xfrm_sk_policy_unlink(old_pol, dir);
2285         }
2286         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2287
2288         if (old_pol) {
2289                 xfrm_policy_kill(old_pol);
2290         }
2291         return 0;
2292 }
2293
2294 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2295 {
2296         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2297         struct net *net = xp_net(old);
2298
2299         if (newp) {
2300                 newp->selector = old->selector;
2301                 if (security_xfrm_policy_clone(old->security,
2302                                                &newp->security)) {
2303                         kfree(newp);
2304                         return NULL;  /* ENOMEM */
2305                 }
2306                 newp->lft = old->lft;
2307                 newp->curlft = old->curlft;
2308                 newp->mark = old->mark;
2309                 newp->if_id = old->if_id;
2310                 newp->action = old->action;
2311                 newp->flags = old->flags;
2312                 newp->xfrm_nr = old->xfrm_nr;
2313                 newp->index = old->index;
2314                 newp->type = old->type;
2315                 newp->family = old->family;
2316                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2317                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2318                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2319                 xfrm_sk_policy_link(newp, dir);
2320                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2321                 xfrm_pol_put(newp);
2322         }
2323         return newp;
2324 }
2325
2326 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2327 {
2328         const struct xfrm_policy *p;
2329         struct xfrm_policy *np;
2330         int i, ret = 0;
2331
2332         rcu_read_lock();
2333         for (i = 0; i < 2; i++) {
2334                 p = rcu_dereference(osk->sk_policy[i]);
2335                 if (p) {
2336                         np = clone_policy(p, i);
2337                         if (unlikely(!np)) {
2338                                 ret = -ENOMEM;
2339                                 break;
2340                         }
2341                         rcu_assign_pointer(sk->sk_policy[i], np);
2342                 }
2343         }
2344         rcu_read_unlock();
2345         return ret;
2346 }
2347
2348 static int
2349 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2350                xfrm_address_t *remote, unsigned short family, u32 mark)
2351 {
2352         int err;
2353         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2354
2355         if (unlikely(afinfo == NULL))
2356                 return -EINVAL;
2357         err = afinfo->get_saddr(net, oif, local, remote, mark);
2358         rcu_read_unlock();
2359         return err;
2360 }
2361
2362 /* Resolve list of templates for the flow, given policy. */
2363
2364 static int
2365 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2366                       struct xfrm_state **xfrm, unsigned short family)
2367 {
2368         struct net *net = xp_net(policy);
2369         int nx;
2370         int i, error;
2371         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2372         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2373         xfrm_address_t tmp;
2374
2375         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2376                 struct xfrm_state *x;
2377                 xfrm_address_t *remote = daddr;
2378                 xfrm_address_t *local  = saddr;
2379                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2380
2381                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2382                     tmpl->mode == XFRM_MODE_BEET) {
2383                         remote = &tmpl->id.daddr;
2384                         local = &tmpl->saddr;
2385                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2386                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2387                                                        &tmp, remote,
2388                                                        tmpl->encap_family, 0);
2389                                 if (error)
2390                                         goto fail;
2391                                 local = &tmp;
2392                         }
2393                 }
2394
2395                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2396                                     family, policy->if_id);
2397
2398                 if (x && x->km.state == XFRM_STATE_VALID) {
2399                         xfrm[nx++] = x;
2400                         daddr = remote;
2401                         saddr = local;
2402                         continue;
2403                 }
2404                 if (x) {
2405                         error = (x->km.state == XFRM_STATE_ERROR ?
2406                                  -EINVAL : -EAGAIN);
2407                         xfrm_state_put(x);
2408                 } else if (error == -ESRCH) {
2409                         error = -EAGAIN;
2410                 }
2411
2412                 if (!tmpl->optional)
2413                         goto fail;
2414         }
2415         return nx;
2416
2417 fail:
2418         for (nx--; nx >= 0; nx--)
2419                 xfrm_state_put(xfrm[nx]);
2420         return error;
2421 }
2422
2423 static int
2424 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2425                   struct xfrm_state **xfrm, unsigned short family)
2426 {
2427         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2428         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2429         int cnx = 0;
2430         int error;
2431         int ret;
2432         int i;
2433
2434         for (i = 0; i < npols; i++) {
2435                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2436                         error = -ENOBUFS;
2437                         goto fail;
2438                 }
2439
2440                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2441                 if (ret < 0) {
2442                         error = ret;
2443                         goto fail;
2444                 } else
2445                         cnx += ret;
2446         }
2447
2448         /* found states are sorted for outbound processing */
2449         if (npols > 1)
2450                 xfrm_state_sort(xfrm, tpp, cnx, family);
2451
2452         return cnx;
2453
2454  fail:
2455         for (cnx--; cnx >= 0; cnx--)
2456                 xfrm_state_put(tpp[cnx]);
2457         return error;
2458
2459 }
2460
2461 static int xfrm_get_tos(const struct flowi *fl, int family)
2462 {
2463         if (family == AF_INET)
2464                 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2465
2466         return 0;
2467 }
2468
2469 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2470 {
2471         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2472         struct dst_ops *dst_ops;
2473         struct xfrm_dst *xdst;
2474
2475         if (!afinfo)
2476                 return ERR_PTR(-EINVAL);
2477
2478         switch (family) {
2479         case AF_INET:
2480                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2481                 break;
2482 #if IS_ENABLED(CONFIG_IPV6)
2483         case AF_INET6:
2484                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2485                 break;
2486 #endif
2487         default:
2488                 BUG();
2489         }
2490         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2491
2492         if (likely(xdst)) {
2493                 memset_after(xdst, 0, u.dst);
2494         } else
2495                 xdst = ERR_PTR(-ENOBUFS);
2496
2497         rcu_read_unlock();
2498
2499         return xdst;
2500 }
2501
2502 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2503                            int nfheader_len)
2504 {
2505         if (dst->ops->family == AF_INET6) {
2506                 struct rt6_info *rt = (struct rt6_info *)dst;
2507                 path->path_cookie = rt6_get_cookie(rt);
2508                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2509         }
2510 }
2511
2512 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2513                                 const struct flowi *fl)
2514 {
2515         const struct xfrm_policy_afinfo *afinfo =
2516                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2517         int err;
2518
2519         if (!afinfo)
2520                 return -EINVAL;
2521
2522         err = afinfo->fill_dst(xdst, dev, fl);
2523
2524         rcu_read_unlock();
2525
2526         return err;
2527 }
2528
2529
2530 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2531  * all the metrics... Shortly, bundle a bundle.
2532  */
2533
2534 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2535                                             struct xfrm_state **xfrm,
2536                                             struct xfrm_dst **bundle,
2537                                             int nx,
2538                                             const struct flowi *fl,
2539                                             struct dst_entry *dst)
2540 {
2541         const struct xfrm_state_afinfo *afinfo;
2542         const struct xfrm_mode *inner_mode;
2543         struct net *net = xp_net(policy);
2544         unsigned long now = jiffies;
2545         struct net_device *dev;
2546         struct xfrm_dst *xdst_prev = NULL;
2547         struct xfrm_dst *xdst0 = NULL;
2548         int i = 0;
2549         int err;
2550         int header_len = 0;
2551         int nfheader_len = 0;
2552         int trailer_len = 0;
2553         int tos;
2554         int family = policy->selector.family;
2555         xfrm_address_t saddr, daddr;
2556
2557         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2558
2559         tos = xfrm_get_tos(fl, family);
2560
2561         dst_hold(dst);
2562
2563         for (; i < nx; i++) {
2564                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2565                 struct dst_entry *dst1 = &xdst->u.dst;
2566
2567                 err = PTR_ERR(xdst);
2568                 if (IS_ERR(xdst)) {
2569                         dst_release(dst);
2570                         goto put_states;
2571                 }
2572
2573                 bundle[i] = xdst;
2574                 if (!xdst_prev)
2575                         xdst0 = xdst;
2576                 else
2577                         /* Ref count is taken during xfrm_alloc_dst()
2578                          * No need to do dst_clone() on dst1
2579                          */
2580                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2581
2582                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2583                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2584                                                         xfrm_af2proto(family));
2585                         if (!inner_mode) {
2586                                 err = -EAFNOSUPPORT;
2587                                 dst_release(dst);
2588                                 goto put_states;
2589                         }
2590                 } else
2591                         inner_mode = &xfrm[i]->inner_mode;
2592
2593                 xdst->route = dst;
2594                 dst_copy_metrics(dst1, dst);
2595
2596                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2597                         __u32 mark = 0;
2598                         int oif;
2599
2600                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2601                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2602
2603                         family = xfrm[i]->props.family;
2604                         oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2605                         dst = xfrm_dst_lookup(xfrm[i], tos, oif,
2606                                               &saddr, &daddr, family, mark);
2607                         err = PTR_ERR(dst);
2608                         if (IS_ERR(dst))
2609                                 goto put_states;
2610                 } else
2611                         dst_hold(dst);
2612
2613                 dst1->xfrm = xfrm[i];
2614                 xdst->xfrm_genid = xfrm[i]->genid;
2615
2616                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2617                 dst1->lastuse = now;
2618
2619                 dst1->input = dst_discard;
2620
2621                 rcu_read_lock();
2622                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2623                 if (likely(afinfo))
2624                         dst1->output = afinfo->output;
2625                 else
2626                         dst1->output = dst_discard_out;
2627                 rcu_read_unlock();
2628
2629                 xdst_prev = xdst;
2630
2631                 header_len += xfrm[i]->props.header_len;
2632                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2633                         nfheader_len += xfrm[i]->props.header_len;
2634                 trailer_len += xfrm[i]->props.trailer_len;
2635         }
2636
2637         xfrm_dst_set_child(xdst_prev, dst);
2638         xdst0->path = dst;
2639
2640         err = -ENODEV;
2641         dev = dst->dev;
2642         if (!dev)
2643                 goto free_dst;
2644
2645         xfrm_init_path(xdst0, dst, nfheader_len);
2646         xfrm_init_pmtu(bundle, nx);
2647
2648         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2649              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2650                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2651                 if (err)
2652                         goto free_dst;
2653
2654                 xdst_prev->u.dst.header_len = header_len;
2655                 xdst_prev->u.dst.trailer_len = trailer_len;
2656                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2657                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2658         }
2659
2660         return &xdst0->u.dst;
2661
2662 put_states:
2663         for (; i < nx; i++)
2664                 xfrm_state_put(xfrm[i]);
2665 free_dst:
2666         if (xdst0)
2667                 dst_release_immediate(&xdst0->u.dst);
2668
2669         return ERR_PTR(err);
2670 }
2671
2672 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2673                                 struct xfrm_policy **pols,
2674                                 int *num_pols, int *num_xfrms)
2675 {
2676         int i;
2677
2678         if (*num_pols == 0 || !pols[0]) {
2679                 *num_pols = 0;
2680                 *num_xfrms = 0;
2681                 return 0;
2682         }
2683         if (IS_ERR(pols[0])) {
2684                 *num_pols = 0;
2685                 return PTR_ERR(pols[0]);
2686         }
2687
2688         *num_xfrms = pols[0]->xfrm_nr;
2689
2690 #ifdef CONFIG_XFRM_SUB_POLICY
2691         if (pols[0]->action == XFRM_POLICY_ALLOW &&
2692             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2693                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2694                                                     XFRM_POLICY_TYPE_MAIN,
2695                                                     fl, family,
2696                                                     XFRM_POLICY_OUT,
2697                                                     pols[0]->if_id);
2698                 if (pols[1]) {
2699                         if (IS_ERR(pols[1])) {
2700                                 xfrm_pols_put(pols, *num_pols);
2701                                 *num_pols = 0;
2702                                 return PTR_ERR(pols[1]);
2703                         }
2704                         (*num_pols)++;
2705                         (*num_xfrms) += pols[1]->xfrm_nr;
2706                 }
2707         }
2708 #endif
2709         for (i = 0; i < *num_pols; i++) {
2710                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2711                         *num_xfrms = -1;
2712                         break;
2713                 }
2714         }
2715
2716         return 0;
2717
2718 }
2719
2720 static struct xfrm_dst *
2721 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2722                                const struct flowi *fl, u16 family,
2723                                struct dst_entry *dst_orig)
2724 {
2725         struct net *net = xp_net(pols[0]);
2726         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2727         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2728         struct xfrm_dst *xdst;
2729         struct dst_entry *dst;
2730         int err;
2731
2732         /* Try to instantiate a bundle */
2733         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2734         if (err <= 0) {
2735                 if (err == 0)
2736                         return NULL;
2737
2738                 if (err != -EAGAIN)
2739                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2740                 return ERR_PTR(err);
2741         }
2742
2743         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2744         if (IS_ERR(dst)) {
2745                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2746                 return ERR_CAST(dst);
2747         }
2748
2749         xdst = (struct xfrm_dst *)dst;
2750         xdst->num_xfrms = err;
2751         xdst->num_pols = num_pols;
2752         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2753         xdst->policy_genid = atomic_read(&pols[0]->genid);
2754
2755         return xdst;
2756 }
2757
2758 static void xfrm_policy_queue_process(struct timer_list *t)
2759 {
2760         struct sk_buff *skb;
2761         struct sock *sk;
2762         struct dst_entry *dst;
2763         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2764         struct net *net = xp_net(pol);
2765         struct xfrm_policy_queue *pq = &pol->polq;
2766         struct flowi fl;
2767         struct sk_buff_head list;
2768         __u32 skb_mark;
2769
2770         spin_lock(&pq->hold_queue.lock);
2771         skb = skb_peek(&pq->hold_queue);
2772         if (!skb) {
2773                 spin_unlock(&pq->hold_queue.lock);
2774                 goto out;
2775         }
2776         dst = skb_dst(skb);
2777         sk = skb->sk;
2778
2779         /* Fixup the mark to support VTI. */
2780         skb_mark = skb->mark;
2781         skb->mark = pol->mark.v;
2782         xfrm_decode_session(skb, &fl, dst->ops->family);
2783         skb->mark = skb_mark;
2784         spin_unlock(&pq->hold_queue.lock);
2785
2786         dst_hold(xfrm_dst_path(dst));
2787         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2788         if (IS_ERR(dst))
2789                 goto purge_queue;
2790
2791         if (dst->flags & DST_XFRM_QUEUE) {
2792                 dst_release(dst);
2793
2794                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2795                         goto purge_queue;
2796
2797                 pq->timeout = pq->timeout << 1;
2798                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2799                         xfrm_pol_hold(pol);
2800                 goto out;
2801         }
2802
2803         dst_release(dst);
2804
2805         __skb_queue_head_init(&list);
2806
2807         spin_lock(&pq->hold_queue.lock);
2808         pq->timeout = 0;
2809         skb_queue_splice_init(&pq->hold_queue, &list);
2810         spin_unlock(&pq->hold_queue.lock);
2811
2812         while (!skb_queue_empty(&list)) {
2813                 skb = __skb_dequeue(&list);
2814
2815                 /* Fixup the mark to support VTI. */
2816                 skb_mark = skb->mark;
2817                 skb->mark = pol->mark.v;
2818                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2819                 skb->mark = skb_mark;
2820
2821                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2822                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2823                 if (IS_ERR(dst)) {
2824                         kfree_skb(skb);
2825                         continue;
2826                 }
2827
2828                 nf_reset_ct(skb);
2829                 skb_dst_drop(skb);
2830                 skb_dst_set(skb, dst);
2831
2832                 dst_output(net, skb->sk, skb);
2833         }
2834
2835 out:
2836         xfrm_pol_put(pol);
2837         return;
2838
2839 purge_queue:
2840         pq->timeout = 0;
2841         skb_queue_purge(&pq->hold_queue);
2842         xfrm_pol_put(pol);
2843 }
2844
2845 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2846 {
2847         unsigned long sched_next;
2848         struct dst_entry *dst = skb_dst(skb);
2849         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2850         struct xfrm_policy *pol = xdst->pols[0];
2851         struct xfrm_policy_queue *pq = &pol->polq;
2852
2853         if (unlikely(skb_fclone_busy(sk, skb))) {
2854                 kfree_skb(skb);
2855                 return 0;
2856         }
2857
2858         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2859                 kfree_skb(skb);
2860                 return -EAGAIN;
2861         }
2862
2863         skb_dst_force(skb);
2864
2865         spin_lock_bh(&pq->hold_queue.lock);
2866
2867         if (!pq->timeout)
2868                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2869
2870         sched_next = jiffies + pq->timeout;
2871
2872         if (del_timer(&pq->hold_timer)) {
2873                 if (time_before(pq->hold_timer.expires, sched_next))
2874                         sched_next = pq->hold_timer.expires;
2875                 xfrm_pol_put(pol);
2876         }
2877
2878         __skb_queue_tail(&pq->hold_queue, skb);
2879         if (!mod_timer(&pq->hold_timer, sched_next))
2880                 xfrm_pol_hold(pol);
2881
2882         spin_unlock_bh(&pq->hold_queue.lock);
2883
2884         return 0;
2885 }
2886
2887 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2888                                                  struct xfrm_flo *xflo,
2889                                                  const struct flowi *fl,
2890                                                  int num_xfrms,
2891                                                  u16 family)
2892 {
2893         int err;
2894         struct net_device *dev;
2895         struct dst_entry *dst;
2896         struct dst_entry *dst1;
2897         struct xfrm_dst *xdst;
2898
2899         xdst = xfrm_alloc_dst(net, family);
2900         if (IS_ERR(xdst))
2901                 return xdst;
2902
2903         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2904             net->xfrm.sysctl_larval_drop ||
2905             num_xfrms <= 0)
2906                 return xdst;
2907
2908         dst = xflo->dst_orig;
2909         dst1 = &xdst->u.dst;
2910         dst_hold(dst);
2911         xdst->route = dst;
2912
2913         dst_copy_metrics(dst1, dst);
2914
2915         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2916         dst1->flags |= DST_XFRM_QUEUE;
2917         dst1->lastuse = jiffies;
2918
2919         dst1->input = dst_discard;
2920         dst1->output = xdst_queue_output;
2921
2922         dst_hold(dst);
2923         xfrm_dst_set_child(xdst, dst);
2924         xdst->path = dst;
2925
2926         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2927
2928         err = -ENODEV;
2929         dev = dst->dev;
2930         if (!dev)
2931                 goto free_dst;
2932
2933         err = xfrm_fill_dst(xdst, dev, fl);
2934         if (err)
2935                 goto free_dst;
2936
2937 out:
2938         return xdst;
2939
2940 free_dst:
2941         dst_release(dst1);
2942         xdst = ERR_PTR(err);
2943         goto out;
2944 }
2945
2946 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2947                                            const struct flowi *fl,
2948                                            u16 family, u8 dir,
2949                                            struct xfrm_flo *xflo, u32 if_id)
2950 {
2951         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2952         int num_pols = 0, num_xfrms = 0, err;
2953         struct xfrm_dst *xdst;
2954
2955         /* Resolve policies to use if we couldn't get them from
2956          * previous cache entry */
2957         num_pols = 1;
2958         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2959         err = xfrm_expand_policies(fl, family, pols,
2960                                            &num_pols, &num_xfrms);
2961         if (err < 0)
2962                 goto inc_error;
2963         if (num_pols == 0)
2964                 return NULL;
2965         if (num_xfrms <= 0)
2966                 goto make_dummy_bundle;
2967
2968         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2969                                               xflo->dst_orig);
2970         if (IS_ERR(xdst)) {
2971                 err = PTR_ERR(xdst);
2972                 if (err == -EREMOTE) {
2973                         xfrm_pols_put(pols, num_pols);
2974                         return NULL;
2975                 }
2976
2977                 if (err != -EAGAIN)
2978                         goto error;
2979                 goto make_dummy_bundle;
2980         } else if (xdst == NULL) {
2981                 num_xfrms = 0;
2982                 goto make_dummy_bundle;
2983         }
2984
2985         return xdst;
2986
2987 make_dummy_bundle:
2988         /* We found policies, but there's no bundles to instantiate:
2989          * either because the policy blocks, has no transformations or
2990          * we could not build template (no xfrm_states).*/
2991         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2992         if (IS_ERR(xdst)) {
2993                 xfrm_pols_put(pols, num_pols);
2994                 return ERR_CAST(xdst);
2995         }
2996         xdst->num_pols = num_pols;
2997         xdst->num_xfrms = num_xfrms;
2998         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2999
3000         return xdst;
3001
3002 inc_error:
3003         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3004 error:
3005         xfrm_pols_put(pols, num_pols);
3006         return ERR_PTR(err);
3007 }
3008
3009 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3010                                         struct dst_entry *dst_orig)
3011 {
3012         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3013         struct dst_entry *ret;
3014
3015         if (!afinfo) {
3016                 dst_release(dst_orig);
3017                 return ERR_PTR(-EINVAL);
3018         } else {
3019                 ret = afinfo->blackhole_route(net, dst_orig);
3020         }
3021         rcu_read_unlock();
3022
3023         return ret;
3024 }
3025
3026 /* Finds/creates a bundle for given flow and if_id
3027  *
3028  * At the moment we eat a raw IP route. Mostly to speed up lookups
3029  * on interfaces with disabled IPsec.
3030  *
3031  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3032  * compatibility
3033  */
3034 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3035                                         struct dst_entry *dst_orig,
3036                                         const struct flowi *fl,
3037                                         const struct sock *sk,
3038                                         int flags, u32 if_id)
3039 {
3040         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3041         struct xfrm_dst *xdst;
3042         struct dst_entry *dst, *route;
3043         u16 family = dst_orig->ops->family;
3044         u8 dir = XFRM_POLICY_OUT;
3045         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3046
3047         dst = NULL;
3048         xdst = NULL;
3049         route = NULL;
3050
3051         sk = sk_const_to_full_sk(sk);
3052         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3053                 num_pols = 1;
3054                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3055                                                 if_id);
3056                 err = xfrm_expand_policies(fl, family, pols,
3057                                            &num_pols, &num_xfrms);
3058                 if (err < 0)
3059                         goto dropdst;
3060
3061                 if (num_pols) {
3062                         if (num_xfrms <= 0) {
3063                                 drop_pols = num_pols;
3064                                 goto no_transform;
3065                         }
3066
3067                         xdst = xfrm_resolve_and_create_bundle(
3068                                         pols, num_pols, fl,
3069                                         family, dst_orig);
3070
3071                         if (IS_ERR(xdst)) {
3072                                 xfrm_pols_put(pols, num_pols);
3073                                 err = PTR_ERR(xdst);
3074                                 if (err == -EREMOTE)
3075                                         goto nopol;
3076
3077                                 goto dropdst;
3078                         } else if (xdst == NULL) {
3079                                 num_xfrms = 0;
3080                                 drop_pols = num_pols;
3081                                 goto no_transform;
3082                         }
3083
3084                         route = xdst->route;
3085                 }
3086         }
3087
3088         if (xdst == NULL) {
3089                 struct xfrm_flo xflo;
3090
3091                 xflo.dst_orig = dst_orig;
3092                 xflo.flags = flags;
3093
3094                 /* To accelerate a bit...  */
3095                 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3096                                !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3097                         goto nopol;
3098
3099                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3100                 if (xdst == NULL)
3101                         goto nopol;
3102                 if (IS_ERR(xdst)) {
3103                         err = PTR_ERR(xdst);
3104                         goto dropdst;
3105                 }
3106
3107                 num_pols = xdst->num_pols;
3108                 num_xfrms = xdst->num_xfrms;
3109                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3110                 route = xdst->route;
3111         }
3112
3113         dst = &xdst->u.dst;
3114         if (route == NULL && num_xfrms > 0) {
3115                 /* The only case when xfrm_bundle_lookup() returns a
3116                  * bundle with null route, is when the template could
3117                  * not be resolved. It means policies are there, but
3118                  * bundle could not be created, since we don't yet
3119                  * have the xfrm_state's. We need to wait for KM to
3120                  * negotiate new SA's or bail out with error.*/
3121                 if (net->xfrm.sysctl_larval_drop) {
3122                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3123                         err = -EREMOTE;
3124                         goto error;
3125                 }
3126
3127                 err = -EAGAIN;
3128
3129                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3130                 goto error;
3131         }
3132
3133 no_transform:
3134         if (num_pols == 0)
3135                 goto nopol;
3136
3137         if ((flags & XFRM_LOOKUP_ICMP) &&
3138             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3139                 err = -ENOENT;
3140                 goto error;
3141         }
3142
3143         for (i = 0; i < num_pols; i++)
3144                 WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds());
3145
3146         if (num_xfrms < 0) {
3147                 /* Prohibit the flow */
3148                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3149                 err = -EPERM;
3150                 goto error;
3151         } else if (num_xfrms > 0) {
3152                 /* Flow transformed */
3153                 dst_release(dst_orig);
3154         } else {
3155                 /* Flow passes untransformed */
3156                 dst_release(dst);
3157                 dst = dst_orig;
3158         }
3159 ok:
3160         xfrm_pols_put(pols, drop_pols);
3161         if (dst && dst->xfrm &&
3162             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3163                 dst->flags |= DST_XFRM_TUNNEL;
3164         return dst;
3165
3166 nopol:
3167         if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3168             net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3169                 err = -EPERM;
3170                 goto error;
3171         }
3172         if (!(flags & XFRM_LOOKUP_ICMP)) {
3173                 dst = dst_orig;
3174                 goto ok;
3175         }
3176         err = -ENOENT;
3177 error:
3178         dst_release(dst);
3179 dropdst:
3180         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3181                 dst_release(dst_orig);
3182         xfrm_pols_put(pols, drop_pols);
3183         return ERR_PTR(err);
3184 }
3185 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3186
3187 /* Main function: finds/creates a bundle for given flow.
3188  *
3189  * At the moment we eat a raw IP route. Mostly to speed up lookups
3190  * on interfaces with disabled IPsec.
3191  */
3192 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3193                               const struct flowi *fl, const struct sock *sk,
3194                               int flags)
3195 {
3196         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3197 }
3198 EXPORT_SYMBOL(xfrm_lookup);
3199
3200 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3201  * Otherwise we may send out blackholed packets.
3202  */
3203 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3204                                     const struct flowi *fl,
3205                                     const struct sock *sk, int flags)
3206 {
3207         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3208                                             flags | XFRM_LOOKUP_QUEUE |
3209                                             XFRM_LOOKUP_KEEP_DST_REF);
3210
3211         if (PTR_ERR(dst) == -EREMOTE)
3212                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3213
3214         if (IS_ERR(dst))
3215                 dst_release(dst_orig);
3216
3217         return dst;
3218 }
3219 EXPORT_SYMBOL(xfrm_lookup_route);
3220
3221 static inline int
3222 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3223 {
3224         struct sec_path *sp = skb_sec_path(skb);
3225         struct xfrm_state *x;
3226
3227         if (!sp || idx < 0 || idx >= sp->len)
3228                 return 0;
3229         x = sp->xvec[idx];
3230         if (!x->type->reject)
3231                 return 0;
3232         return x->type->reject(x, skb, fl);
3233 }
3234
3235 /* When skb is transformed back to its "native" form, we have to
3236  * check policy restrictions. At the moment we make this in maximally
3237  * stupid way. Shame on me. :-) Of course, connected sockets must
3238  * have policy cached at them.
3239  */
3240
3241 static inline int
3242 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3243               unsigned short family, u32 if_id)
3244 {
3245         if (xfrm_state_kern(x))
3246                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3247         return  x->id.proto == tmpl->id.proto &&
3248                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3249                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3250                 x->props.mode == tmpl->mode &&
3251                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3252                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3253                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3254                   xfrm_state_addr_cmp(tmpl, x, family)) &&
3255                 (if_id == 0 || if_id == x->if_id);
3256 }
3257
3258 /*
3259  * 0 or more than 0 is returned when validation is succeeded (either bypass
3260  * because of optional transport mode, or next index of the matched secpath
3261  * state with the template.
3262  * -1 is returned when no matching template is found.
3263  * Otherwise "-2 - errored_index" is returned.
3264  */
3265 static inline int
3266 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3267                unsigned short family, u32 if_id)
3268 {
3269         int idx = start;
3270
3271         if (tmpl->optional) {
3272                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3273                         return start;
3274         } else
3275                 start = -1;
3276         for (; idx < sp->len; idx++) {
3277                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3278                         return ++idx;
3279                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3280                         if (idx < sp->verified_cnt) {
3281                                 /* Secpath entry previously verified, consider optional and
3282                                  * continue searching
3283                                  */
3284                                 continue;
3285                         }
3286
3287                         if (start == -1)
3288                                 start = -2-idx;
3289                         break;
3290                 }
3291         }
3292         return start;
3293 }
3294
3295 static void
3296 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3297 {
3298         const struct iphdr *iph = ip_hdr(skb);
3299         int ihl = iph->ihl;
3300         u8 *xprth = skb_network_header(skb) + ihl * 4;
3301         struct flowi4 *fl4 = &fl->u.ip4;
3302         int oif = 0;
3303
3304         if (skb_dst(skb) && skb_dst(skb)->dev)
3305                 oif = skb_dst(skb)->dev->ifindex;
3306
3307         memset(fl4, 0, sizeof(struct flowi4));
3308         fl4->flowi4_mark = skb->mark;
3309         fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3310
3311         fl4->flowi4_proto = iph->protocol;
3312         fl4->daddr = reverse ? iph->saddr : iph->daddr;
3313         fl4->saddr = reverse ? iph->daddr : iph->saddr;
3314         fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3315
3316         if (!ip_is_fragment(iph)) {
3317                 switch (iph->protocol) {
3318                 case IPPROTO_UDP:
3319                 case IPPROTO_UDPLITE:
3320                 case IPPROTO_TCP:
3321                 case IPPROTO_SCTP:
3322                 case IPPROTO_DCCP:
3323                         if (xprth + 4 < skb->data ||
3324                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3325                                 __be16 *ports;
3326
3327                                 xprth = skb_network_header(skb) + ihl * 4;
3328                                 ports = (__be16 *)xprth;
3329
3330                                 fl4->fl4_sport = ports[!!reverse];
3331                                 fl4->fl4_dport = ports[!reverse];
3332                         }
3333                         break;
3334                 case IPPROTO_ICMP:
3335                         if (xprth + 2 < skb->data ||
3336                             pskb_may_pull(skb, xprth + 2 - skb->data)) {
3337                                 u8 *icmp;
3338
3339                                 xprth = skb_network_header(skb) + ihl * 4;
3340                                 icmp = xprth;
3341
3342                                 fl4->fl4_icmp_type = icmp[0];
3343                                 fl4->fl4_icmp_code = icmp[1];
3344                         }
3345                         break;
3346                 case IPPROTO_GRE:
3347                         if (xprth + 12 < skb->data ||
3348                             pskb_may_pull(skb, xprth + 12 - skb->data)) {
3349                                 __be16 *greflags;
3350                                 __be32 *gre_hdr;
3351
3352                                 xprth = skb_network_header(skb) + ihl * 4;
3353                                 greflags = (__be16 *)xprth;
3354                                 gre_hdr = (__be32 *)xprth;
3355
3356                                 if (greflags[0] & GRE_KEY) {
3357                                         if (greflags[0] & GRE_CSUM)
3358                                                 gre_hdr++;
3359                                         fl4->fl4_gre_key = gre_hdr[1];
3360                                 }
3361                         }
3362                         break;
3363                 default:
3364                         break;
3365                 }
3366         }
3367 }
3368
3369 #if IS_ENABLED(CONFIG_IPV6)
3370 static void
3371 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3372 {
3373         struct flowi6 *fl6 = &fl->u.ip6;
3374         int onlyproto = 0;
3375         const struct ipv6hdr *hdr = ipv6_hdr(skb);
3376         u32 offset = sizeof(*hdr);
3377         struct ipv6_opt_hdr *exthdr;
3378         const unsigned char *nh = skb_network_header(skb);
3379         u16 nhoff = IP6CB(skb)->nhoff;
3380         int oif = 0;
3381         u8 nexthdr;
3382
3383         if (!nhoff)
3384                 nhoff = offsetof(struct ipv6hdr, nexthdr);
3385
3386         nexthdr = nh[nhoff];
3387
3388         if (skb_dst(skb) && skb_dst(skb)->dev)
3389                 oif = skb_dst(skb)->dev->ifindex;
3390
3391         memset(fl6, 0, sizeof(struct flowi6));
3392         fl6->flowi6_mark = skb->mark;
3393         fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3394
3395         fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3396         fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3397
3398         while (nh + offset + sizeof(*exthdr) < skb->data ||
3399                pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3400                 nh = skb_network_header(skb);
3401                 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3402
3403                 switch (nexthdr) {
3404                 case NEXTHDR_FRAGMENT:
3405                         onlyproto = 1;
3406                         fallthrough;
3407                 case NEXTHDR_ROUTING:
3408                 case NEXTHDR_HOP:
3409                 case NEXTHDR_DEST:
3410                         offset += ipv6_optlen(exthdr);
3411                         nexthdr = exthdr->nexthdr;
3412                         break;
3413                 case IPPROTO_UDP:
3414                 case IPPROTO_UDPLITE:
3415                 case IPPROTO_TCP:
3416                 case IPPROTO_SCTP:
3417                 case IPPROTO_DCCP:
3418                         if (!onlyproto && (nh + offset + 4 < skb->data ||
3419                              pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3420                                 __be16 *ports;
3421
3422                                 nh = skb_network_header(skb);
3423                                 ports = (__be16 *)(nh + offset);
3424                                 fl6->fl6_sport = ports[!!reverse];
3425                                 fl6->fl6_dport = ports[!reverse];
3426                         }
3427                         fl6->flowi6_proto = nexthdr;
3428                         return;
3429                 case IPPROTO_ICMPV6:
3430                         if (!onlyproto && (nh + offset + 2 < skb->data ||
3431                             pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3432                                 u8 *icmp;
3433
3434                                 nh = skb_network_header(skb);
3435                                 icmp = (u8 *)(nh + offset);
3436                                 fl6->fl6_icmp_type = icmp[0];
3437                                 fl6->fl6_icmp_code = icmp[1];
3438                         }
3439                         fl6->flowi6_proto = nexthdr;
3440                         return;
3441                 case IPPROTO_GRE:
3442                         if (!onlyproto &&
3443                             (nh + offset + 12 < skb->data ||
3444                              pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3445                                 struct gre_base_hdr *gre_hdr;
3446                                 __be32 *gre_key;
3447
3448                                 nh = skb_network_header(skb);
3449                                 gre_hdr = (struct gre_base_hdr *)(nh + offset);
3450                                 gre_key = (__be32 *)(gre_hdr + 1);
3451
3452                                 if (gre_hdr->flags & GRE_KEY) {
3453                                         if (gre_hdr->flags & GRE_CSUM)
3454                                                 gre_key++;
3455                                         fl6->fl6_gre_key = *gre_key;
3456                                 }
3457                         }
3458                         fl6->flowi6_proto = nexthdr;
3459                         return;
3460
3461 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3462                 case IPPROTO_MH:
3463                         offset += ipv6_optlen(exthdr);
3464                         if (!onlyproto && (nh + offset + 3 < skb->data ||
3465                             pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3466                                 struct ip6_mh *mh;
3467
3468                                 nh = skb_network_header(skb);
3469                                 mh = (struct ip6_mh *)(nh + offset);
3470                                 fl6->fl6_mh_type = mh->ip6mh_type;
3471                         }
3472                         fl6->flowi6_proto = nexthdr;
3473                         return;
3474 #endif
3475                 default:
3476                         fl6->flowi6_proto = nexthdr;
3477                         return;
3478                 }
3479         }
3480 }
3481 #endif
3482
3483 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3484                           unsigned int family, int reverse)
3485 {
3486         switch (family) {
3487         case AF_INET:
3488                 decode_session4(skb, fl, reverse);
3489                 break;
3490 #if IS_ENABLED(CONFIG_IPV6)
3491         case AF_INET6:
3492                 decode_session6(skb, fl, reverse);
3493                 break;
3494 #endif
3495         default:
3496                 return -EAFNOSUPPORT;
3497         }
3498
3499         return security_xfrm_decode_session(skb, &fl->flowi_secid);
3500 }
3501 EXPORT_SYMBOL(__xfrm_decode_session);
3502
3503 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3504 {
3505         for (; k < sp->len; k++) {
3506                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3507                         *idxp = k;
3508                         return 1;
3509                 }
3510         }
3511
3512         return 0;
3513 }
3514
3515 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3516                         unsigned short family)
3517 {
3518         struct net *net = dev_net(skb->dev);
3519         struct xfrm_policy *pol;
3520         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3521         int npols = 0;
3522         int xfrm_nr;
3523         int pi;
3524         int reverse;
3525         struct flowi fl;
3526         int xerr_idx = -1;
3527         const struct xfrm_if_cb *ifcb;
3528         struct sec_path *sp;
3529         u32 if_id = 0;
3530
3531         rcu_read_lock();
3532         ifcb = xfrm_if_get_cb();
3533
3534         if (ifcb) {
3535                 struct xfrm_if_decode_session_result r;
3536
3537                 if (ifcb->decode_session(skb, family, &r)) {
3538                         if_id = r.if_id;
3539                         net = r.net;
3540                 }
3541         }
3542         rcu_read_unlock();
3543
3544         reverse = dir & ~XFRM_POLICY_MASK;
3545         dir &= XFRM_POLICY_MASK;
3546
3547         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3548                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3549                 return 0;
3550         }
3551
3552         nf_nat_decode_session(skb, &fl, family);
3553
3554         /* First, check used SA against their selectors. */
3555         sp = skb_sec_path(skb);
3556         if (sp) {
3557                 int i;
3558
3559                 for (i = sp->len - 1; i >= 0; i--) {
3560                         struct xfrm_state *x = sp->xvec[i];
3561                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3562                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3563                                 return 0;
3564                         }
3565                 }
3566         }
3567
3568         pol = NULL;
3569         sk = sk_to_full_sk(sk);
3570         if (sk && sk->sk_policy[dir]) {
3571                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3572                 if (IS_ERR(pol)) {
3573                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3574                         return 0;
3575                 }
3576         }
3577
3578         if (!pol)
3579                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3580
3581         if (IS_ERR(pol)) {
3582                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3583                 return 0;
3584         }
3585
3586         if (!pol) {
3587                 if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3588                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3589                         return 0;
3590                 }
3591
3592                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3593                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3594                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3595                         return 0;
3596                 }
3597                 return 1;
3598         }
3599
3600         /* This lockless write can happen from different cpus. */
3601         WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3602
3603         pols[0] = pol;
3604         npols++;
3605 #ifdef CONFIG_XFRM_SUB_POLICY
3606         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3607                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3608                                                     &fl, family,
3609                                                     XFRM_POLICY_IN, if_id);
3610                 if (pols[1]) {
3611                         if (IS_ERR(pols[1])) {
3612                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3613                                 xfrm_pol_put(pols[0]);
3614                                 return 0;
3615                         }
3616                         /* This write can happen from different cpus. */
3617                         WRITE_ONCE(pols[1]->curlft.use_time,
3618                                    ktime_get_real_seconds());
3619                         npols++;
3620                 }
3621         }
3622 #endif
3623
3624         if (pol->action == XFRM_POLICY_ALLOW) {
3625                 static struct sec_path dummy;
3626                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3627                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3628                 struct xfrm_tmpl **tpp = tp;
3629                 int ti = 0;
3630                 int i, k;
3631
3632                 sp = skb_sec_path(skb);
3633                 if (!sp)
3634                         sp = &dummy;
3635
3636                 for (pi = 0; pi < npols; pi++) {
3637                         if (pols[pi] != pol &&
3638                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3639                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3640                                 goto reject;
3641                         }
3642                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3643                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3644                                 goto reject_error;
3645                         }
3646                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3647                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3648                 }
3649                 xfrm_nr = ti;
3650
3651                 if (npols > 1) {
3652                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3653                         tpp = stp;
3654                 }
3655
3656                 /* For each tunnel xfrm, find the first matching tmpl.
3657                  * For each tmpl before that, find corresponding xfrm.
3658                  * Order is _important_. Later we will implement
3659                  * some barriers, but at the moment barriers
3660                  * are implied between each two transformations.
3661                  * Upon success, marks secpath entries as having been
3662                  * verified to allow them to be skipped in future policy
3663                  * checks (e.g. nested tunnels).
3664                  */
3665                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3666                         k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3667                         if (k < 0) {
3668                                 if (k < -1)
3669                                         /* "-2 - errored_index" returned */
3670                                         xerr_idx = -(2+k);
3671                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3672                                 goto reject;
3673                         }
3674                 }
3675
3676                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3677                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3678                         goto reject;
3679                 }
3680
3681                 xfrm_pols_put(pols, npols);
3682                 sp->verified_cnt = k;
3683
3684                 return 1;
3685         }
3686         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3687
3688 reject:
3689         xfrm_secpath_reject(xerr_idx, skb, &fl);
3690 reject_error:
3691         xfrm_pols_put(pols, npols);
3692         return 0;
3693 }
3694 EXPORT_SYMBOL(__xfrm_policy_check);
3695
3696 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3697 {
3698         struct net *net = dev_net(skb->dev);
3699         struct flowi fl;
3700         struct dst_entry *dst;
3701         int res = 1;
3702
3703         if (xfrm_decode_session(skb, &fl, family) < 0) {
3704                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3705                 return 0;
3706         }
3707
3708         skb_dst_force(skb);
3709         if (!skb_dst(skb)) {
3710                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3711                 return 0;
3712         }
3713
3714         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3715         if (IS_ERR(dst)) {
3716                 res = 0;
3717                 dst = NULL;
3718         }
3719         skb_dst_set(skb, dst);
3720         return res;
3721 }
3722 EXPORT_SYMBOL(__xfrm_route_forward);
3723
3724 /* Optimize later using cookies and generation ids. */
3725
3726 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3727 {
3728         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3729          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3730          * get validated by dst_ops->check on every use.  We do this
3731          * because when a normal route referenced by an XFRM dst is
3732          * obsoleted we do not go looking around for all parent
3733          * referencing XFRM dsts so that we can invalidate them.  It
3734          * is just too much work.  Instead we make the checks here on
3735          * every use.  For example:
3736          *
3737          *      XFRM dst A --> IPv4 dst X
3738          *
3739          * X is the "xdst->route" of A (X is also the "dst->path" of A
3740          * in this example).  If X is marked obsolete, "A" will not
3741          * notice.  That's what we are validating here via the
3742          * stale_bundle() check.
3743          *
3744          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3745          * be marked on it.
3746          * This will force stale_bundle() to fail on any xdst bundle with
3747          * this dst linked in it.
3748          */
3749         if (dst->obsolete < 0 && !stale_bundle(dst))
3750                 return dst;
3751
3752         return NULL;
3753 }
3754
3755 static int stale_bundle(struct dst_entry *dst)
3756 {
3757         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3758 }
3759
3760 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3761 {
3762         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3763                 dst->dev = blackhole_netdev;
3764                 dev_hold(dst->dev);
3765                 dev_put(dev);
3766         }
3767 }
3768 EXPORT_SYMBOL(xfrm_dst_ifdown);
3769
3770 static void xfrm_link_failure(struct sk_buff *skb)
3771 {
3772         /* Impossible. Such dst must be popped before reaches point of failure. */
3773 }
3774
3775 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3776 {
3777         if (dst) {
3778                 if (dst->obsolete) {
3779                         dst_release(dst);
3780                         dst = NULL;
3781                 }
3782         }
3783         return dst;
3784 }
3785
3786 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3787 {
3788         while (nr--) {
3789                 struct xfrm_dst *xdst = bundle[nr];
3790                 u32 pmtu, route_mtu_cached;
3791                 struct dst_entry *dst;
3792
3793                 dst = &xdst->u.dst;
3794                 pmtu = dst_mtu(xfrm_dst_child(dst));
3795                 xdst->child_mtu_cached = pmtu;
3796
3797                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3798
3799                 route_mtu_cached = dst_mtu(xdst->route);
3800                 xdst->route_mtu_cached = route_mtu_cached;
3801
3802                 if (pmtu > route_mtu_cached)
3803                         pmtu = route_mtu_cached;
3804
3805                 dst_metric_set(dst, RTAX_MTU, pmtu);
3806         }
3807 }
3808
3809 /* Check that the bundle accepts the flow and its components are
3810  * still valid.
3811  */
3812
3813 static int xfrm_bundle_ok(struct xfrm_dst *first)
3814 {
3815         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3816         struct dst_entry *dst = &first->u.dst;
3817         struct xfrm_dst *xdst;
3818         int start_from, nr;
3819         u32 mtu;
3820
3821         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3822             (dst->dev && !netif_running(dst->dev)))
3823                 return 0;
3824
3825         if (dst->flags & DST_XFRM_QUEUE)
3826                 return 1;
3827
3828         start_from = nr = 0;
3829         do {
3830                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3831
3832                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3833                         return 0;
3834                 if (xdst->xfrm_genid != dst->xfrm->genid)
3835                         return 0;
3836                 if (xdst->num_pols > 0 &&
3837                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3838                         return 0;
3839
3840                 bundle[nr++] = xdst;
3841
3842                 mtu = dst_mtu(xfrm_dst_child(dst));
3843                 if (xdst->child_mtu_cached != mtu) {
3844                         start_from = nr;
3845                         xdst->child_mtu_cached = mtu;
3846                 }
3847
3848                 if (!dst_check(xdst->route, xdst->route_cookie))
3849                         return 0;
3850                 mtu = dst_mtu(xdst->route);
3851                 if (xdst->route_mtu_cached != mtu) {
3852                         start_from = nr;
3853                         xdst->route_mtu_cached = mtu;
3854                 }
3855
3856                 dst = xfrm_dst_child(dst);
3857         } while (dst->xfrm);
3858
3859         if (likely(!start_from))
3860                 return 1;
3861
3862         xdst = bundle[start_from - 1];
3863         mtu = xdst->child_mtu_cached;
3864         while (start_from--) {
3865                 dst = &xdst->u.dst;
3866
3867                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3868                 if (mtu > xdst->route_mtu_cached)
3869                         mtu = xdst->route_mtu_cached;
3870                 dst_metric_set(dst, RTAX_MTU, mtu);
3871                 if (!start_from)
3872                         break;
3873
3874                 xdst = bundle[start_from - 1];
3875                 xdst->child_mtu_cached = mtu;
3876         }
3877
3878         return 1;
3879 }
3880
3881 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3882 {
3883         return dst_metric_advmss(xfrm_dst_path(dst));
3884 }
3885
3886 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3887 {
3888         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3889
3890         return mtu ? : dst_mtu(xfrm_dst_path(dst));
3891 }
3892
3893 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3894                                         const void *daddr)
3895 {
3896         while (dst->xfrm) {
3897                 const struct xfrm_state *xfrm = dst->xfrm;
3898
3899                 dst = xfrm_dst_child(dst);
3900
3901                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3902                         continue;
3903                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3904                         daddr = xfrm->coaddr;
3905                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3906                         daddr = &xfrm->id.daddr;
3907         }
3908         return daddr;
3909 }
3910
3911 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3912                                            struct sk_buff *skb,
3913                                            const void *daddr)
3914 {
3915         const struct dst_entry *path = xfrm_dst_path(dst);
3916
3917         if (!skb)
3918                 daddr = xfrm_get_dst_nexthop(dst, daddr);
3919         return path->ops->neigh_lookup(path, skb, daddr);
3920 }
3921
3922 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3923 {
3924         const struct dst_entry *path = xfrm_dst_path(dst);
3925
3926         daddr = xfrm_get_dst_nexthop(dst, daddr);
3927         path->ops->confirm_neigh(path, daddr);
3928 }
3929
3930 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3931 {
3932         int err = 0;
3933
3934         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3935                 return -EAFNOSUPPORT;
3936
3937         spin_lock(&xfrm_policy_afinfo_lock);
3938         if (unlikely(xfrm_policy_afinfo[family] != NULL))
3939                 err = -EEXIST;
3940         else {
3941                 struct dst_ops *dst_ops = afinfo->dst_ops;
3942                 if (likely(dst_ops->kmem_cachep == NULL))
3943                         dst_ops->kmem_cachep = xfrm_dst_cache;
3944                 if (likely(dst_ops->check == NULL))
3945                         dst_ops->check = xfrm_dst_check;
3946                 if (likely(dst_ops->default_advmss == NULL))
3947                         dst_ops->default_advmss = xfrm_default_advmss;
3948                 if (likely(dst_ops->mtu == NULL))
3949                         dst_ops->mtu = xfrm_mtu;
3950                 if (likely(dst_ops->negative_advice == NULL))
3951                         dst_ops->negative_advice = xfrm_negative_advice;
3952                 if (likely(dst_ops->link_failure == NULL))
3953                         dst_ops->link_failure = xfrm_link_failure;
3954                 if (likely(dst_ops->neigh_lookup == NULL))
3955                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
3956                 if (likely(!dst_ops->confirm_neigh))
3957                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
3958                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3959         }
3960         spin_unlock(&xfrm_policy_afinfo_lock);
3961
3962         return err;
3963 }
3964 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3965
3966 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3967 {
3968         struct dst_ops *dst_ops = afinfo->dst_ops;
3969         int i;
3970
3971         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3972                 if (xfrm_policy_afinfo[i] != afinfo)
3973                         continue;
3974                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3975                 break;
3976         }
3977
3978         synchronize_rcu();
3979
3980         dst_ops->kmem_cachep = NULL;
3981         dst_ops->check = NULL;
3982         dst_ops->negative_advice = NULL;
3983         dst_ops->link_failure = NULL;
3984 }
3985 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3986
3987 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3988 {
3989         spin_lock(&xfrm_if_cb_lock);
3990         rcu_assign_pointer(xfrm_if_cb, ifcb);
3991         spin_unlock(&xfrm_if_cb_lock);
3992 }
3993 EXPORT_SYMBOL(xfrm_if_register_cb);
3994
3995 void xfrm_if_unregister_cb(void)
3996 {
3997         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3998         synchronize_rcu();
3999 }
4000 EXPORT_SYMBOL(xfrm_if_unregister_cb);
4001
4002 #ifdef CONFIG_XFRM_STATISTICS
4003 static int __net_init xfrm_statistics_init(struct net *net)
4004 {
4005         int rv;
4006         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4007         if (!net->mib.xfrm_statistics)
4008                 return -ENOMEM;
4009         rv = xfrm_proc_init(net);
4010         if (rv < 0)
4011                 free_percpu(net->mib.xfrm_statistics);
4012         return rv;
4013 }
4014
4015 static void xfrm_statistics_fini(struct net *net)
4016 {
4017         xfrm_proc_fini(net);
4018         free_percpu(net->mib.xfrm_statistics);
4019 }
4020 #else
4021 static int __net_init xfrm_statistics_init(struct net *net)
4022 {
4023         return 0;
4024 }
4025
4026 static void xfrm_statistics_fini(struct net *net)
4027 {
4028 }
4029 #endif
4030
4031 static int __net_init xfrm_policy_init(struct net *net)
4032 {
4033         unsigned int hmask, sz;
4034         int dir, err;
4035
4036         if (net_eq(net, &init_net)) {
4037                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4038                                            sizeof(struct xfrm_dst),
4039                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4040                                            NULL);
4041                 err = rhashtable_init(&xfrm_policy_inexact_table,
4042                                       &xfrm_pol_inexact_params);
4043                 BUG_ON(err);
4044         }
4045
4046         hmask = 8 - 1;
4047         sz = (hmask+1) * sizeof(struct hlist_head);
4048
4049         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4050         if (!net->xfrm.policy_byidx)
4051                 goto out_byidx;
4052         net->xfrm.policy_idx_hmask = hmask;
4053
4054         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4055                 struct xfrm_policy_hash *htab;
4056
4057                 net->xfrm.policy_count[dir] = 0;
4058                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4059                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4060
4061                 htab = &net->xfrm.policy_bydst[dir];
4062                 htab->table = xfrm_hash_alloc(sz);
4063                 if (!htab->table)
4064                         goto out_bydst;
4065                 htab->hmask = hmask;
4066                 htab->dbits4 = 32;
4067                 htab->sbits4 = 32;
4068                 htab->dbits6 = 128;
4069                 htab->sbits6 = 128;
4070         }
4071         net->xfrm.policy_hthresh.lbits4 = 32;
4072         net->xfrm.policy_hthresh.rbits4 = 32;
4073         net->xfrm.policy_hthresh.lbits6 = 128;
4074         net->xfrm.policy_hthresh.rbits6 = 128;
4075
4076         seqlock_init(&net->xfrm.policy_hthresh.lock);
4077
4078         INIT_LIST_HEAD(&net->xfrm.policy_all);
4079         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4080         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4081         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4082         return 0;
4083
4084 out_bydst:
4085         for (dir--; dir >= 0; dir--) {
4086                 struct xfrm_policy_hash *htab;
4087
4088                 htab = &net->xfrm.policy_bydst[dir];
4089                 xfrm_hash_free(htab->table, sz);
4090         }
4091         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4092 out_byidx:
4093         return -ENOMEM;
4094 }
4095
4096 static void xfrm_policy_fini(struct net *net)
4097 {
4098         struct xfrm_pol_inexact_bin *b, *t;
4099         unsigned int sz;
4100         int dir;
4101
4102         flush_work(&net->xfrm.policy_hash_work);
4103 #ifdef CONFIG_XFRM_SUB_POLICY
4104         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4105 #endif
4106         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4107
4108         WARN_ON(!list_empty(&net->xfrm.policy_all));
4109
4110         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4111                 struct xfrm_policy_hash *htab;
4112
4113                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4114
4115                 htab = &net->xfrm.policy_bydst[dir];
4116                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4117                 WARN_ON(!hlist_empty(htab->table));
4118                 xfrm_hash_free(htab->table, sz);
4119         }
4120
4121         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4122         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4123         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4124
4125         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4126         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4127                 __xfrm_policy_inexact_prune_bin(b, true);
4128         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4129 }
4130
4131 static int __net_init xfrm_net_init(struct net *net)
4132 {
4133         int rv;
4134
4135         /* Initialize the per-net locks here */
4136         spin_lock_init(&net->xfrm.xfrm_state_lock);
4137         spin_lock_init(&net->xfrm.xfrm_policy_lock);
4138         seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4139         mutex_init(&net->xfrm.xfrm_cfg_mutex);
4140         net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4141         net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4142         net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4143
4144         rv = xfrm_statistics_init(net);
4145         if (rv < 0)
4146                 goto out_statistics;
4147         rv = xfrm_state_init(net);
4148         if (rv < 0)
4149                 goto out_state;
4150         rv = xfrm_policy_init(net);
4151         if (rv < 0)
4152                 goto out_policy;
4153         rv = xfrm_sysctl_init(net);
4154         if (rv < 0)
4155                 goto out_sysctl;
4156
4157         return 0;
4158
4159 out_sysctl:
4160         xfrm_policy_fini(net);
4161 out_policy:
4162         xfrm_state_fini(net);
4163 out_state:
4164         xfrm_statistics_fini(net);
4165 out_statistics:
4166         return rv;
4167 }
4168
4169 static void __net_exit xfrm_net_exit(struct net *net)
4170 {
4171         xfrm_sysctl_fini(net);
4172         xfrm_policy_fini(net);
4173         xfrm_state_fini(net);
4174         xfrm_statistics_fini(net);
4175 }
4176
4177 static struct pernet_operations __net_initdata xfrm_net_ops = {
4178         .init = xfrm_net_init,
4179         .exit = xfrm_net_exit,
4180 };
4181
4182 void __init xfrm_init(void)
4183 {
4184         register_pernet_subsys(&xfrm_net_ops);
4185         xfrm_dev_init();
4186         xfrm_input_init();
4187
4188 #ifdef CONFIG_XFRM_ESPINTCP
4189         espintcp_init();
4190 #endif
4191 }
4192
4193 #ifdef CONFIG_AUDITSYSCALL
4194 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4195                                          struct audit_buffer *audit_buf)
4196 {
4197         struct xfrm_sec_ctx *ctx = xp->security;
4198         struct xfrm_selector *sel = &xp->selector;
4199
4200         if (ctx)
4201                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4202                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4203
4204         switch (sel->family) {
4205         case AF_INET:
4206                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4207                 if (sel->prefixlen_s != 32)
4208                         audit_log_format(audit_buf, " src_prefixlen=%d",
4209                                          sel->prefixlen_s);
4210                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4211                 if (sel->prefixlen_d != 32)
4212                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4213                                          sel->prefixlen_d);
4214                 break;
4215         case AF_INET6:
4216                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4217                 if (sel->prefixlen_s != 128)
4218                         audit_log_format(audit_buf, " src_prefixlen=%d",
4219                                          sel->prefixlen_s);
4220                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4221                 if (sel->prefixlen_d != 128)
4222                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4223                                          sel->prefixlen_d);
4224                 break;
4225         }
4226 }
4227
4228 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4229 {
4230         struct audit_buffer *audit_buf;
4231
4232         audit_buf = xfrm_audit_start("SPD-add");
4233         if (audit_buf == NULL)
4234                 return;
4235         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4236         audit_log_format(audit_buf, " res=%u", result);
4237         xfrm_audit_common_policyinfo(xp, audit_buf);
4238         audit_log_end(audit_buf);
4239 }
4240 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4241
4242 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4243                               bool task_valid)
4244 {
4245         struct audit_buffer *audit_buf;
4246
4247         audit_buf = xfrm_audit_start("SPD-delete");
4248         if (audit_buf == NULL)
4249                 return;
4250         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4251         audit_log_format(audit_buf, " res=%u", result);
4252         xfrm_audit_common_policyinfo(xp, audit_buf);
4253         audit_log_end(audit_buf);
4254 }
4255 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4256 #endif
4257
4258 #ifdef CONFIG_XFRM_MIGRATE
4259 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4260                                         const struct xfrm_selector *sel_tgt)
4261 {
4262         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4263                 if (sel_tgt->family == sel_cmp->family &&
4264                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4265                                     sel_cmp->family) &&
4266                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4267                                     sel_cmp->family) &&
4268                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4269                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4270                         return true;
4271                 }
4272         } else {
4273                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4274                         return true;
4275                 }
4276         }
4277         return false;
4278 }
4279
4280 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4281                                                     u8 dir, u8 type, struct net *net, u32 if_id)
4282 {
4283         struct xfrm_policy *pol, *ret = NULL;
4284         struct hlist_head *chain;
4285         u32 priority = ~0U;
4286
4287         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4288         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4289         hlist_for_each_entry(pol, chain, bydst) {
4290                 if ((if_id == 0 || pol->if_id == if_id) &&
4291                     xfrm_migrate_selector_match(sel, &pol->selector) &&
4292                     pol->type == type) {
4293                         ret = pol;
4294                         priority = ret->priority;
4295                         break;
4296                 }
4297         }
4298         chain = &net->xfrm.policy_inexact[dir];
4299         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4300                 if ((pol->priority >= priority) && ret)
4301                         break;
4302
4303                 if ((if_id == 0 || pol->if_id == if_id) &&
4304                     xfrm_migrate_selector_match(sel, &pol->selector) &&
4305                     pol->type == type) {
4306                         ret = pol;
4307                         break;
4308                 }
4309         }
4310
4311         xfrm_pol_hold(ret);
4312
4313         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4314
4315         return ret;
4316 }
4317
4318 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4319 {
4320         int match = 0;
4321
4322         if (t->mode == m->mode && t->id.proto == m->proto &&
4323             (m->reqid == 0 || t->reqid == m->reqid)) {
4324                 switch (t->mode) {
4325                 case XFRM_MODE_TUNNEL:
4326                 case XFRM_MODE_BEET:
4327                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4328                                             m->old_family) &&
4329                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4330                                             m->old_family)) {
4331                                 match = 1;
4332                         }
4333                         break;
4334                 case XFRM_MODE_TRANSPORT:
4335                         /* in case of transport mode, template does not store
4336                            any IP addresses, hence we just compare mode and
4337                            protocol */
4338                         match = 1;
4339                         break;
4340                 default:
4341                         break;
4342                 }
4343         }
4344         return match;
4345 }
4346
4347 /* update endpoint address(es) of template(s) */
4348 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4349                                struct xfrm_migrate *m, int num_migrate)
4350 {
4351         struct xfrm_migrate *mp;
4352         int i, j, n = 0;
4353
4354         write_lock_bh(&pol->lock);
4355         if (unlikely(pol->walk.dead)) {
4356                 /* target policy has been deleted */
4357                 write_unlock_bh(&pol->lock);
4358                 return -ENOENT;
4359         }
4360
4361         for (i = 0; i < pol->xfrm_nr; i++) {
4362                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4363                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4364                                 continue;
4365                         n++;
4366                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4367                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4368                                 continue;
4369                         /* update endpoints */
4370                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4371                                sizeof(pol->xfrm_vec[i].id.daddr));
4372                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4373                                sizeof(pol->xfrm_vec[i].saddr));
4374                         pol->xfrm_vec[i].encap_family = mp->new_family;
4375                         /* flush bundles */
4376                         atomic_inc(&pol->genid);
4377                 }
4378         }
4379
4380         write_unlock_bh(&pol->lock);
4381
4382         if (!n)
4383                 return -ENODATA;
4384
4385         return 0;
4386 }
4387
4388 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4389 {
4390         int i, j;
4391
4392         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4393                 return -EINVAL;
4394
4395         for (i = 0; i < num_migrate; i++) {
4396                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4397                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4398                         return -EINVAL;
4399
4400                 /* check if there is any duplicated entry */
4401                 for (j = i + 1; j < num_migrate; j++) {
4402                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4403                                     sizeof(m[i].old_daddr)) &&
4404                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4405                                     sizeof(m[i].old_saddr)) &&
4406                             m[i].proto == m[j].proto &&
4407                             m[i].mode == m[j].mode &&
4408                             m[i].reqid == m[j].reqid &&
4409                             m[i].old_family == m[j].old_family)
4410                                 return -EINVAL;
4411                 }
4412         }
4413
4414         return 0;
4415 }
4416
4417 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4418                  struct xfrm_migrate *m, int num_migrate,
4419                  struct xfrm_kmaddress *k, struct net *net,
4420                  struct xfrm_encap_tmpl *encap, u32 if_id)
4421 {
4422         int i, err, nx_cur = 0, nx_new = 0;
4423         struct xfrm_policy *pol = NULL;
4424         struct xfrm_state *x, *xc;
4425         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4426         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4427         struct xfrm_migrate *mp;
4428
4429         /* Stage 0 - sanity checks */
4430         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4431                 goto out;
4432
4433         if (dir >= XFRM_POLICY_MAX) {
4434                 err = -EINVAL;
4435                 goto out;
4436         }
4437
4438         /* Stage 1 - find policy */
4439         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id)) == NULL) {
4440                 err = -ENOENT;
4441                 goto out;
4442         }
4443
4444         /* Stage 2 - find and update state(s) */
4445         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4446                 if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4447                         x_cur[nx_cur] = x;
4448                         nx_cur++;
4449                         xc = xfrm_state_migrate(x, mp, encap);
4450                         if (xc) {
4451                                 x_new[nx_new] = xc;
4452                                 nx_new++;
4453                         } else {
4454                                 err = -ENODATA;
4455                                 goto restore_state;
4456                         }
4457                 }
4458         }
4459
4460         /* Stage 3 - update policy */
4461         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4462                 goto restore_state;
4463
4464         /* Stage 4 - delete old state(s) */
4465         if (nx_cur) {
4466                 xfrm_states_put(x_cur, nx_cur);
4467                 xfrm_states_delete(x_cur, nx_cur);
4468         }
4469
4470         /* Stage 5 - announce */
4471         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4472
4473         xfrm_pol_put(pol);
4474
4475         return 0;
4476 out:
4477         return err;
4478
4479 restore_state:
4480         if (pol)
4481                 xfrm_pol_put(pol);
4482         if (nx_cur)
4483                 xfrm_states_put(x_cur, nx_cur);
4484         if (nx_new)
4485                 xfrm_states_delete(x_new, nx_new);
4486
4487         return err;
4488 }
4489 EXPORT_SYMBOL(xfrm_migrate);
4490 #endif