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                         (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                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 358                 if (tmo <= 0) {
 359                         warn = 1;
 360                         tmo = XFRM_KM_TIMEOUT;
 361                 }
 362                 if (tmo < next)
 363                         next = tmo;
 364         }
 365
 366         if (warn)
 367                 km_policy_expired(xp, dir, 0, 0);
 368         if (next != TIME64_MAX &&
 369             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 370                 xfrm_pol_hold(xp);
 371
 372 out:
 373         read_unlock(&xp->lock);
 374         xfrm_pol_put(xp);
 375         return;
 376
 377 expired:
 378         read_unlock(&xp->lock);
 379         if (!xfrm_policy_delete(xp, dir))
 380                 km_policy_expired(xp, dir, 1, 0);
 381         xfrm_pol_put(xp);
 382 }
 383
 384 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 385  * SPD calls.
 386  */
 387
 388 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 389 {
 390         struct xfrm_policy *policy;
 391
 392         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 393
 394         if (policy) {
 395                 write_pnet(&policy->xp_net, net);
 396                 INIT_LIST_HEAD(&policy->walk.all);
 397                 INIT_HLIST_NODE(&policy->bydst_inexact_list);
 398                 INIT_HLIST_NODE(&policy->bydst);
 399                 INIT_HLIST_NODE(&policy->byidx);
 400                 rwlock_init(&policy->lock);
 401                 refcount_set(&policy->refcnt, 1);
 402                 skb_queue_head_init(&policy->polq.hold_queue);
 403                 timer_setup(&policy->timer, xfrm_policy_timer, 0);
 404                 timer_setup(&policy->polq.hold_timer,
 405                             xfrm_policy_queue_process, 0);
 406         }
 407         return policy;
 408 }
 409 EXPORT_SYMBOL(xfrm_policy_alloc);
 410
 411 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 412 {
 413         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 414
 415         security_xfrm_policy_free(policy->security);
 416         kfree(policy);
 417 }
 418
 419 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 420
 421 void xfrm_policy_destroy(struct xfrm_policy *policy)
 422 {
 423         BUG_ON(!policy->walk.dead);
 424
 425         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 426                 BUG();
 427
 428         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 429 }
 430 EXPORT_SYMBOL(xfrm_policy_destroy);
 431
 432 /* Rule must be locked. Release descendant resources, announce
 433  * entry dead. The rule must be unlinked from lists to the moment.
 434  */
 435
 436 static void xfrm_policy_kill(struct xfrm_policy *policy)
 437 {
 438         write_lock_bh(&policy->lock);
 439         policy->walk.dead = 1;
 440         write_unlock_bh(&policy->lock);
 441
 442         atomic_inc(&policy->genid);
 443
 444         if (del_timer(&policy->polq.hold_timer))
 445                 xfrm_pol_put(policy);
 446         skb_queue_purge(&policy->polq.hold_queue);
 447
 448         if (del_timer(&policy->timer))
 449                 xfrm_pol_put(policy);
 450
 451         xfrm_pol_put(policy);
 452 }
 453
 454 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 455
 456 static inline unsigned int idx_hash(struct net *net, u32 index)
 457 {
 458         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 459 }
 460
 461 /* calculate policy hash thresholds */
 462 static void __get_hash_thresh(struct net *net,
 463                               unsigned short family, int dir,
 464                               u8 *dbits, u8 *sbits)
 465 {
 466         switch (family) {
 467         case AF_INET:
 468                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 469                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 470                 break;
 471
 472         case AF_INET6:
 473                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 474                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 475                 break;
 476
 477         default:
 478                 *dbits = 0;
 479                 *sbits = 0;
 480         }
 481 }
 482
 483 static struct hlist_head *policy_hash_bysel(struct net *net,
 484                                             const struct xfrm_selector *sel,
 485                                             unsigned short family, int dir)
 486 {
 487         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 488         unsigned int hash;
 489         u8 dbits;
 490         u8 sbits;
 491
 492         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 493         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 494
 495         if (hash == hmask + 1)
 496                 return NULL;
 497
 498         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 499                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 500 }
 501
 502 static struct hlist_head *policy_hash_direct(struct net *net,
 503                                              const xfrm_address_t *daddr,
 504                                              const xfrm_address_t *saddr,
 505                                              unsigned short family, int dir)
 506 {
 507         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 508         unsigned int hash;
 509         u8 dbits;
 510         u8 sbits;
 511
 512         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 513         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 514
 515         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 516                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 517 }
 518
 519 static void xfrm_dst_hash_transfer(struct net *net,
 520                                    struct hlist_head *list,
 521                                    struct hlist_head *ndsttable,
 522                                    unsigned int nhashmask,
 523                                    int dir)
 524 {
 525         struct hlist_node *tmp, *entry0 = NULL;
 526         struct xfrm_policy *pol;
 527         unsigned int h0 = 0;
 528         u8 dbits;
 529         u8 sbits;
 530
 531 redo:
 532         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 533                 unsigned int h;
 534
 535                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 536                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 537                                 pol->family, nhashmask, dbits, sbits);
 538                 if (!entry0) {
 539                         hlist_del_rcu(&pol->bydst);
 540                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 541                         h0 = h;
 542                 } else {
 543                         if (h != h0)
 544                                 continue;
 545                         hlist_del_rcu(&pol->bydst);
 546                         hlist_add_behind_rcu(&pol->bydst, entry0);
 547                 }
 548                 entry0 = &pol->bydst;
 549         }
 550         if (!hlist_empty(list)) {
 551                 entry0 = NULL;
 552                 goto redo;
 553         }
 554 }
 555
 556 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 557                                    struct hlist_head *nidxtable,
 558                                    unsigned int nhashmask)
 559 {
 560         struct hlist_node *tmp;
 561         struct xfrm_policy *pol;
 562
 563         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 564                 unsigned int h;
 565
 566                 h = __idx_hash(pol->index, nhashmask);
 567                 hlist_add_head(&pol->byidx, nidxtable+h);
 568         }
 569 }
 570
 571 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 572 {
 573         return ((old_hmask + 1) << 1) - 1;
 574 }
 575
 576 static void xfrm_bydst_resize(struct net *net, int dir)
 577 {
 578         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 579         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 580         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 581         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 582         struct hlist_head *odst;
 583         int i;
 584
 585         if (!ndst)
 586                 return;
 587
 588         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 589         write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
 590
 591         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 592                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 593
 594         for (i = hmask; i >= 0; i--)
 595                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 596
 597         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 598         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 599
 600         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
 601         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 602
 603         synchronize_rcu();
 604
 605         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 606 }
 607
 608 static void xfrm_byidx_resize(struct net *net, int total)
 609 {
 610         unsigned int hmask = net->xfrm.policy_idx_hmask;
 611         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 612         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 613         struct hlist_head *oidx = net->xfrm.policy_byidx;
 614         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 615         int i;
 616
 617         if (!nidx)
 618                 return;
 619
 620         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 621
 622         for (i = hmask; i >= 0; i--)
 623                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 624
 625         net->xfrm.policy_byidx = nidx;
 626         net->xfrm.policy_idx_hmask = nhashmask;
 627
 628         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 629
 630         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 631 }
 632
 633 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 634 {
 635         unsigned int cnt = net->xfrm.policy_count[dir];
 636         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 637
 638         if (total)
 639                 *total += cnt;
 640
 641         if ((hmask + 1) < xfrm_policy_hashmax &&
 642             cnt > hmask)
 643                 return 1;
 644
 645         return 0;
 646 }
 647
 648 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 649 {
 650         unsigned int hmask = net->xfrm.policy_idx_hmask;
 651
 652         if ((hmask + 1) < xfrm_policy_hashmax &&
 653             total > hmask)
 654                 return 1;
 655
 656         return 0;
 657 }
 658
 659 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 660 {
 661         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 662         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 663         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 664         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 665         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 666         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 667         si->spdhcnt = net->xfrm.policy_idx_hmask;
 668         si->spdhmcnt = xfrm_policy_hashmax;
 669 }
 670 EXPORT_SYMBOL(xfrm_spd_getinfo);
 671
 672 static DEFINE_MUTEX(hash_resize_mutex);
 673 static void xfrm_hash_resize(struct work_struct *work)
 674 {
 675         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 676         int dir, total;
 677
 678         mutex_lock(&hash_resize_mutex);
 679
 680         total = 0;
 681         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 682                 if (xfrm_bydst_should_resize(net, dir, &total))
 683                         xfrm_bydst_resize(net, dir);
 684         }
 685         if (xfrm_byidx_should_resize(net, total))
 686                 xfrm_byidx_resize(net, total);
 687
 688         mutex_unlock(&hash_resize_mutex);
 689 }
 690
 691 /* Make sure *pol can be inserted into fastbin.
 692  * Useful to check that later insert requests will be successful
 693  * (provided xfrm_policy_lock is held throughout).
 694  */
 695 static struct xfrm_pol_inexact_bin *
 696 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
 697 {
 698         struct xfrm_pol_inexact_bin *bin, *prev;
 699         struct xfrm_pol_inexact_key k = {
 700                 .family = pol->family,
 701                 .type = pol->type,
 702                 .dir = dir,
 703                 .if_id = pol->if_id,
 704         };
 705         struct net *net = xp_net(pol);
 706
 707         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 708
 709         write_pnet(&k.net, net);
 710         bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
 711                                      xfrm_pol_inexact_params);
 712         if (bin)
 713                 return bin;
 714
 715         bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
 716         if (!bin)
 717                 return NULL;
 718
 719         bin->k = k;
 720         INIT_HLIST_HEAD(&bin->hhead);
 721         bin->root_d = RB_ROOT;
 722         bin->root_s = RB_ROOT;
 723         seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
 724
 725         prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
 726                                                 &bin->k, &bin->head,
 727                                                 xfrm_pol_inexact_params);
 728         if (!prev) {
 729                 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
 730                 return bin;
 731         }
 732
 733         kfree(bin);
 734
 735         return IS_ERR(prev) ? NULL : prev;
 736 }
 737
 738 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
 739                                                int family, u8 prefixlen)
 740 {
 741         if (xfrm_addr_any(addr, family))
 742                 return true;
 743
 744         if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
 745                 return true;
 746
 747         if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
 748                 return true;
 749
 750         return false;
 751 }
 752
 753 static bool
 754 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
 755 {
 756         const xfrm_address_t *addr;
 757         bool saddr_any, daddr_any;
 758         u8 prefixlen;
 759
 760         addr = &policy->selector.saddr;
 761         prefixlen = policy->selector.prefixlen_s;
 762
 763         saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 764                                                        policy->family,
 765                                                        prefixlen);
 766         addr = &policy->selector.daddr;
 767         prefixlen = policy->selector.prefixlen_d;
 768         daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 769                                                        policy->family,
 770                                                        prefixlen);
 771         return saddr_any && daddr_any;
 772 }
 773
 774 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
 775                                        const xfrm_address_t *addr, u8 prefixlen)
 776 {
 777         node->addr = *addr;
 778         node->prefixlen = prefixlen;
 779 }
 780
 781 static struct xfrm_pol_inexact_node *
 782 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
 783 {
 784         struct xfrm_pol_inexact_node *node;
 785
 786         node = kzalloc(sizeof(*node), GFP_ATOMIC);
 787         if (node)
 788                 xfrm_pol_inexact_node_init(node, addr, prefixlen);
 789
 790         return node;
 791 }
 792
 793 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
 794                                   const xfrm_address_t *b,
 795                                   u8 prefixlen, u16 family)
 796 {
 797         u32 ma, mb, mask;
 798         unsigned int pdw, pbi;
 799         int delta = 0;
 800
 801         switch (family) {
 802         case AF_INET:
 803                 if (prefixlen == 0)
 804                         return 0;
 805                 mask = ~0U << (32 - prefixlen);
 806                 ma = ntohl(a->a4) & mask;
 807                 mb = ntohl(b->a4) & mask;
 808                 if (ma < mb)
 809                         delta = -1;
 810                 else if (ma > mb)
 811                         delta = 1;
 812                 break;
 813         case AF_INET6:
 814                 pdw = prefixlen >> 5;
 815                 pbi = prefixlen & 0x1f;
 816
 817                 if (pdw) {
 818                         delta = memcmp(a->a6, b->a6, pdw << 2);
 819                         if (delta)
 820                                 return delta;
 821                 }
 822                 if (pbi) {
 823                         mask = ~0U << (32 - pbi);
 824                         ma = ntohl(a->a6[pdw]) & mask;
 825                         mb = ntohl(b->a6[pdw]) & mask;
 826                         if (ma < mb)
 827                                 delta = -1;
 828                         else if (ma > mb)
 829                                 delta = 1;
 830                 }
 831                 break;
 832         default:
 833                 break;
 834         }
 835
 836         return delta;
 837 }
 838
 839 static void xfrm_policy_inexact_list_reinsert(struct net *net,
 840                                               struct xfrm_pol_inexact_node *n,
 841                                               u16 family)
 842 {
 843         unsigned int matched_s, matched_d;
 844         struct xfrm_policy *policy, *p;
 845
 846         matched_s = 0;
 847         matched_d = 0;
 848
 849         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 850                 struct hlist_node *newpos = NULL;
 851                 bool matches_s, matches_d;
 852
 853                 if (!policy->bydst_reinsert)
 854                         continue;
 855
 856                 WARN_ON_ONCE(policy->family != family);
 857
 858                 policy->bydst_reinsert = false;
 859                 hlist_for_each_entry(p, &n->hhead, bydst) {
 860                         if (policy->priority > p->priority)
 861                                 newpos = &p->bydst;
 862                         else if (policy->priority == p->priority &&
 863                                  policy->pos > p->pos)
 864                                 newpos = &p->bydst;
 865                         else
 866                                 break;
 867                 }
 868
 869                 if (newpos)
 870                         hlist_add_behind_rcu(&policy->bydst, newpos);
 871                 else
 872                         hlist_add_head_rcu(&policy->bydst, &n->hhead);
 873
 874                 /* paranoia checks follow.
 875                  * Check that the reinserted policy matches at least
 876                  * saddr or daddr for current node prefix.
 877                  *
 878                  * Matching both is fine, matching saddr in one policy
 879                  * (but not daddr) and then matching only daddr in another
 880                  * is a bug.
 881                  */
 882                 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
 883                                                    &n->addr,
 884                                                    n->prefixlen,
 885                                                    family) == 0;
 886                 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
 887                                                    &n->addr,
 888                                                    n->prefixlen,
 889                                                    family) == 0;
 890                 if (matches_s && matches_d)
 891                         continue;
 892
 893                 WARN_ON_ONCE(!matches_s && !matches_d);
 894                 if (matches_s)
 895                         matched_s++;
 896                 if (matches_d)
 897                         matched_d++;
 898                 WARN_ON_ONCE(matched_s && matched_d);
 899         }
 900 }
 901
 902 static void xfrm_policy_inexact_node_reinsert(struct net *net,
 903                                               struct xfrm_pol_inexact_node *n,
 904                                               struct rb_root *new,
 905                                               u16 family)
 906 {
 907         struct xfrm_pol_inexact_node *node;
 908         struct rb_node **p, *parent;
 909
 910         /* we should not have another subtree here */
 911         WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
 912 restart:
 913         parent = NULL;
 914         p = &new->rb_node;
 915         while (*p) {
 916                 u8 prefixlen;
 917                 int delta;
 918
 919                 parent = *p;
 920                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 921
 922                 prefixlen = min(node->prefixlen, n->prefixlen);
 923
 924                 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
 925                                                prefixlen, family);
 926                 if (delta < 0) {
 927                         p = &parent->rb_left;
 928                 } else if (delta > 0) {
 929                         p = &parent->rb_right;
 930                 } else {
 931                         bool same_prefixlen = node->prefixlen == n->prefixlen;
 932                         struct xfrm_policy *tmp;
 933
 934                         hlist_for_each_entry(tmp, &n->hhead, bydst) {
 935                                 tmp->bydst_reinsert = true;
 936                                 hlist_del_rcu(&tmp->bydst);
 937                         }
 938
 939                         node->prefixlen = prefixlen;
 940
 941                         xfrm_policy_inexact_list_reinsert(net, node, family);
 942
 943                         if (same_prefixlen) {
 944                                 kfree_rcu(n, rcu);
 945                                 return;
 946                         }
 947
 948                         rb_erase(*p, new);
 949                         kfree_rcu(n, rcu);
 950                         n = node;
 951                         goto restart;
 952                 }
 953         }
 954
 955         rb_link_node_rcu(&n->node, parent, p);
 956         rb_insert_color(&n->node, new);
 957 }
 958
 959 /* merge nodes v and n */
 960 static void xfrm_policy_inexact_node_merge(struct net *net,
 961                                            struct xfrm_pol_inexact_node *v,
 962                                            struct xfrm_pol_inexact_node *n,
 963                                            u16 family)
 964 {
 965         struct xfrm_pol_inexact_node *node;
 966         struct xfrm_policy *tmp;
 967         struct rb_node *rnode;
 968
 969         /* To-be-merged node v has a subtree.
 970          *
 971          * Dismantle it and insert its nodes to n->root.
 972          */
 973         while ((rnode = rb_first(&v->root)) != NULL) {
 974                 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
 975                 rb_erase(&node->node, &v->root);
 976                 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
 977                                                   family);
 978         }
 979
 980         hlist_for_each_entry(tmp, &v->hhead, bydst) {
 981                 tmp->bydst_reinsert = true;
 982                 hlist_del_rcu(&tmp->bydst);
 983         }
 984
 985         xfrm_policy_inexact_list_reinsert(net, n, family);
 986 }
 987
 988 static struct xfrm_pol_inexact_node *
 989 xfrm_policy_inexact_insert_node(struct net *net,
 990                                 struct rb_root *root,
 991                                 xfrm_address_t *addr,
 992                                 u16 family, u8 prefixlen, u8 dir)
 993 {
 994         struct xfrm_pol_inexact_node *cached = NULL;
 995         struct rb_node **p, *parent = NULL;
 996         struct xfrm_pol_inexact_node *node;
 997
 998         p = &root->rb_node;
 999         while (*p) {
1000                 int delta;
1001
1002                 parent = *p;
1003                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1004
1005                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1006                                                node->prefixlen,
1007                                                family);
1008                 if (delta == 0 && prefixlen >= node->prefixlen) {
1009                         WARN_ON_ONCE(cached); /* ipsec policies got lost */
1010                         return node;
1011                 }
1012
1013                 if (delta < 0)
1014                         p = &parent->rb_left;
1015                 else
1016                         p = &parent->rb_right;
1017
1018                 if (prefixlen < node->prefixlen) {
1019                         delta = xfrm_policy_addr_delta(addr, &node->addr,
1020                                                        prefixlen,
1021                                                        family);
1022                         if (delta)
1023                                 continue;
1024
1025                         /* This node is a subnet of the new prefix. It needs
1026                          * to be removed and re-inserted with the smaller
1027                          * prefix and all nodes that are now also covered
1028                          * by the reduced prefixlen.
1029                          */
1030                         rb_erase(&node->node, root);
1031
1032                         if (!cached) {
1033                                 xfrm_pol_inexact_node_init(node, addr,
1034                                                            prefixlen);
1035                                 cached = node;
1036                         } else {
1037                                 /* This node also falls within the new
1038                                  * prefixlen. Merge the to-be-reinserted
1039                                  * node and this one.
1040                                  */
1041                                 xfrm_policy_inexact_node_merge(net, node,
1042                                                                cached, family);
1043                                 kfree_rcu(node, rcu);
1044                         }
1045
1046                         /* restart */
1047                         p = &root->rb_node;
1048                         parent = NULL;
1049                 }
1050         }
1051
1052         node = cached;
1053         if (!node) {
1054                 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1055                 if (!node)
1056                         return NULL;
1057         }
1058
1059         rb_link_node_rcu(&node->node, parent, p);
1060         rb_insert_color(&node->node, root);
1061
1062         return node;
1063 }
1064
1065 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1066 {
1067         struct xfrm_pol_inexact_node *node;
1068         struct rb_node *rn = rb_first(r);
1069
1070         while (rn) {
1071                 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1072
1073                 xfrm_policy_inexact_gc_tree(&node->root, rm);
1074                 rn = rb_next(rn);
1075
1076                 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1077                         WARN_ON_ONCE(rm);
1078                         continue;
1079                 }
1080
1081                 rb_erase(&node->node, r);
1082                 kfree_rcu(node, rcu);
1083         }
1084 }
1085
1086 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1087 {
1088         write_seqcount_begin(&b->count);
1089         xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1090         xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1091         write_seqcount_end(&b->count);
1092
1093         if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1094             !hlist_empty(&b->hhead)) {
1095                 WARN_ON_ONCE(net_exit);
1096                 return;
1097         }
1098
1099         if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1100                                    xfrm_pol_inexact_params) == 0) {
1101                 list_del(&b->inexact_bins);
1102                 kfree_rcu(b, rcu);
1103         }
1104 }
1105
1106 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1107 {
1108         struct net *net = read_pnet(&b->k.net);
1109
1110         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1111         __xfrm_policy_inexact_prune_bin(b, false);
1112         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1113 }
1114
1115 static void __xfrm_policy_inexact_flush(struct net *net)
1116 {
1117         struct xfrm_pol_inexact_bin *bin, *t;
1118
1119         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1120
1121         list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1122                 __xfrm_policy_inexact_prune_bin(bin, false);
1123 }
1124
1125 static struct hlist_head *
1126 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1127                                 struct xfrm_policy *policy, u8 dir)
1128 {
1129         struct xfrm_pol_inexact_node *n;
1130         struct net *net;
1131
1132         net = xp_net(policy);
1133         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1134
1135         if (xfrm_policy_inexact_insert_use_any_list(policy))
1136                 return &bin->hhead;
1137
1138         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1139                                                policy->family,
1140                                                policy->selector.prefixlen_d)) {
1141                 write_seqcount_begin(&bin->count);
1142                 n = xfrm_policy_inexact_insert_node(net,
1143                                                     &bin->root_s,
1144                                                     &policy->selector.saddr,
1145                                                     policy->family,
1146                                                     policy->selector.prefixlen_s,
1147                                                     dir);
1148                 write_seqcount_end(&bin->count);
1149                 if (!n)
1150                         return NULL;
1151
1152                 return &n->hhead;
1153         }
1154
1155         /* daddr is fixed */
1156         write_seqcount_begin(&bin->count);
1157         n = xfrm_policy_inexact_insert_node(net,
1158                                             &bin->root_d,
1159                                             &policy->selector.daddr,
1160                                             policy->family,
1161                                             policy->selector.prefixlen_d, dir);
1162         write_seqcount_end(&bin->count);
1163         if (!n)
1164                 return NULL;
1165
1166         /* saddr is wildcard */
1167         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1168                                                policy->family,
1169                                                policy->selector.prefixlen_s))
1170                 return &n->hhead;
1171
1172         write_seqcount_begin(&bin->count);
1173         n = xfrm_policy_inexact_insert_node(net,
1174                                             &n->root,
1175                                             &policy->selector.saddr,
1176                                             policy->family,
1177                                             policy->selector.prefixlen_s, dir);
1178         write_seqcount_end(&bin->count);
1179         if (!n)
1180                 return NULL;
1181
1182         return &n->hhead;
1183 }
1184
1185 static struct xfrm_policy *
1186 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1187 {
1188         struct xfrm_pol_inexact_bin *bin;
1189         struct xfrm_policy *delpol;
1190         struct hlist_head *chain;
1191         struct net *net;
1192
1193         bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1194         if (!bin)
1195                 return ERR_PTR(-ENOMEM);
1196
1197         net = xp_net(policy);
1198         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1199
1200         chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1201         if (!chain) {
1202                 __xfrm_policy_inexact_prune_bin(bin, false);
1203                 return ERR_PTR(-ENOMEM);
1204         }
1205
1206         delpol = xfrm_policy_insert_list(chain, policy, excl);
1207         if (delpol && excl) {
1208                 __xfrm_policy_inexact_prune_bin(bin, false);
1209                 return ERR_PTR(-EEXIST);
1210         }
1211
1212         chain = &net->xfrm.policy_inexact[dir];
1213         xfrm_policy_insert_inexact_list(chain, policy);
1214
1215         if (delpol)
1216                 __xfrm_policy_inexact_prune_bin(bin, false);
1217
1218         return delpol;
1219 }
1220
1221 static void xfrm_hash_rebuild(struct work_struct *work)
1222 {
1223         struct net *net = container_of(work, struct net,
1224                                        xfrm.policy_hthresh.work);
1225         unsigned int hmask;
1226         struct xfrm_policy *pol;
1227         struct xfrm_policy *policy;
1228         struct hlist_head *chain;
1229         struct hlist_head *odst;
1230         struct hlist_node *newpos;
1231         int i;
1232         int dir;
1233         unsigned seq;
1234         u8 lbits4, rbits4, lbits6, rbits6;
1235
1236         mutex_lock(&hash_resize_mutex);
1237
1238         /* read selector prefixlen thresholds */
1239         do {
1240                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1241
1242                 lbits4 = net->xfrm.policy_hthresh.lbits4;
1243                 rbits4 = net->xfrm.policy_hthresh.rbits4;
1244                 lbits6 = net->xfrm.policy_hthresh.lbits6;
1245                 rbits6 = net->xfrm.policy_hthresh.rbits6;
1246         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1247
1248         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1249         write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1250
1251         /* make sure that we can insert the indirect policies again before
1252          * we start with destructive action.
1253          */
1254         list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1255                 struct xfrm_pol_inexact_bin *bin;
1256                 u8 dbits, sbits;
1257
1258                 dir = xfrm_policy_id2dir(policy->index);
1259                 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1260                         continue;
1261
1262                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1263                         if (policy->family == AF_INET) {
1264                                 dbits = rbits4;
1265                                 sbits = lbits4;
1266                         } else {
1267                                 dbits = rbits6;
1268                                 sbits = lbits6;
1269                         }
1270                 } else {
1271                         if (policy->family == AF_INET) {
1272                                 dbits = lbits4;
1273                                 sbits = rbits4;
1274                         } else {
1275                                 dbits = lbits6;
1276                                 sbits = rbits6;
1277                         }
1278                 }
1279
1280                 if (policy->selector.prefixlen_d < dbits ||
1281                     policy->selector.prefixlen_s < sbits)
1282                         continue;
1283
1284                 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1285                 if (!bin)
1286                         goto out_unlock;
1287
1288                 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1289                         goto out_unlock;
1290         }
1291
1292         /* reset the bydst and inexact table in all directions */
1293         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1294                 struct hlist_node *n;
1295
1296                 hlist_for_each_entry_safe(policy, n,
1297                                           &net->xfrm.policy_inexact[dir],
1298                                           bydst_inexact_list) {
1299                         hlist_del_rcu(&policy->bydst);
1300                         hlist_del_init(&policy->bydst_inexact_list);
1301                 }
1302
1303                 hmask = net->xfrm.policy_bydst[dir].hmask;
1304                 odst = net->xfrm.policy_bydst[dir].table;
1305                 for (i = hmask; i >= 0; i--) {
1306                         hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1307                                 hlist_del_rcu(&policy->bydst);
1308                 }
1309                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1310                         /* dir out => dst = remote, src = local */
1311                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1312                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1313                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1314                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1315                 } else {
1316                         /* dir in/fwd => dst = local, src = remote */
1317                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1318                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1319                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1320                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1321                 }
1322         }
1323
1324         /* re-insert all policies by order of creation */
1325         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1326                 if (policy->walk.dead)
1327                         continue;
1328                 dir = xfrm_policy_id2dir(policy->index);
1329                 if (dir >= XFRM_POLICY_MAX) {
1330                         /* skip socket policies */
1331                         continue;
1332                 }
1333                 newpos = NULL;
1334                 chain = policy_hash_bysel(net, &policy->selector,
1335                                           policy->family, dir);
1336
1337                 if (!chain) {
1338                         void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1339
1340                         WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1341                         continue;
1342                 }
1343
1344                 hlist_for_each_entry(pol, chain, bydst) {
1345                         if (policy->priority >= pol->priority)
1346                                 newpos = &pol->bydst;
1347                         else
1348                                 break;
1349                 }
1350                 if (newpos)
1351                         hlist_add_behind_rcu(&policy->bydst, newpos);
1352                 else
1353                         hlist_add_head_rcu(&policy->bydst, chain);
1354         }
1355
1356 out_unlock:
1357         __xfrm_policy_inexact_flush(net);
1358         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1359         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1360
1361         mutex_unlock(&hash_resize_mutex);
1362 }
1363
1364 void xfrm_policy_hash_rebuild(struct net *net)
1365 {
1366         schedule_work(&net->xfrm.policy_hthresh.work);
1367 }
1368 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1369
1370 /* Generate new index... KAME seems to generate them ordered by cost
1371  * of an absolute inpredictability of ordering of rules. This will not pass. */
1372 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1373 {
1374         for (;;) {
1375                 struct hlist_head *list;
1376                 struct xfrm_policy *p;
1377                 u32 idx;
1378                 int found;
1379
1380                 if (!index) {
1381                         idx = (net->xfrm.idx_generator | dir);
1382                         net->xfrm.idx_generator += 8;
1383                 } else {
1384                         idx = index;
1385                         index = 0;
1386                 }
1387
1388                 if (idx == 0)
1389                         idx = 8;
1390                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1391                 found = 0;
1392                 hlist_for_each_entry(p, list, byidx) {
1393                         if (p->index == idx) {
1394                                 found = 1;
1395                                 break;
1396                         }
1397                 }
1398                 if (!found)
1399                         return idx;
1400         }
1401 }
1402
1403 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1404 {
1405         u32 *p1 = (u32 *) s1;
1406         u32 *p2 = (u32 *) s2;
1407         int len = sizeof(struct xfrm_selector) / sizeof(u32);
1408         int i;
1409
1410         for (i = 0; i < len; i++) {
1411                 if (p1[i] != p2[i])
1412                         return 1;
1413         }
1414
1415         return 0;
1416 }
1417
1418 static void xfrm_policy_requeue(struct xfrm_policy *old,
1419                                 struct xfrm_policy *new)
1420 {
1421         struct xfrm_policy_queue *pq = &old->polq;
1422         struct sk_buff_head list;
1423
1424         if (skb_queue_empty(&pq->hold_queue))
1425                 return;
1426
1427         __skb_queue_head_init(&list);
1428
1429         spin_lock_bh(&pq->hold_queue.lock);
1430         skb_queue_splice_init(&pq->hold_queue, &list);
1431         if (del_timer(&pq->hold_timer))
1432                 xfrm_pol_put(old);
1433         spin_unlock_bh(&pq->hold_queue.lock);
1434
1435         pq = &new->polq;
1436
1437         spin_lock_bh(&pq->hold_queue.lock);
1438         skb_queue_splice(&list, &pq->hold_queue);
1439         pq->timeout = XFRM_QUEUE_TMO_MIN;
1440         if (!mod_timer(&pq->hold_timer, jiffies))
1441                 xfrm_pol_hold(new);
1442         spin_unlock_bh(&pq->hold_queue.lock);
1443 }
1444
1445 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1446                                           struct xfrm_policy *pol)
1447 {
1448         return mark->v == pol->mark.v && mark->m == pol->mark.m;
1449 }
1450
1451 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1452 {
1453         const struct xfrm_pol_inexact_key *k = data;
1454         u32 a = k->type << 24 | k->dir << 16 | k->family;
1455
1456         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1457                             seed);
1458 }
1459
1460 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1461 {
1462         const struct xfrm_pol_inexact_bin *b = data;
1463
1464         return xfrm_pol_bin_key(&b->k, 0, seed);
1465 }
1466
1467 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1468                             const void *ptr)
1469 {
1470         const struct xfrm_pol_inexact_key *key = arg->key;
1471         const struct xfrm_pol_inexact_bin *b = ptr;
1472         int ret;
1473
1474         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1475                 return -1;
1476
1477         ret = b->k.dir ^ key->dir;
1478         if (ret)
1479                 return ret;
1480
1481         ret = b->k.type ^ key->type;
1482         if (ret)
1483                 return ret;
1484
1485         ret = b->k.family ^ key->family;
1486         if (ret)
1487                 return ret;
1488
1489         return b->k.if_id ^ key->if_id;
1490 }
1491
1492 static const struct rhashtable_params xfrm_pol_inexact_params = {
1493         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1494         .hashfn                 = xfrm_pol_bin_key,
1495         .obj_hashfn             = xfrm_pol_bin_obj,
1496         .obj_cmpfn              = xfrm_pol_bin_cmp,
1497         .automatic_shrinking    = true,
1498 };
1499
1500 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1501                                             struct xfrm_policy *policy)
1502 {
1503         struct xfrm_policy *pol, *delpol = NULL;
1504         struct hlist_node *newpos = NULL;
1505         int i = 0;
1506
1507         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1508                 if (pol->type == policy->type &&
1509                     pol->if_id == policy->if_id &&
1510                     !selector_cmp(&pol->selector, &policy->selector) &&
1511                     xfrm_policy_mark_match(&policy->mark, pol) &&
1512                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1513                     !WARN_ON(delpol)) {
1514                         delpol = pol;
1515                         if (policy->priority > pol->priority)
1516                                 continue;
1517                 } else if (policy->priority >= pol->priority) {
1518                         newpos = &pol->bydst_inexact_list;
1519                         continue;
1520                 }
1521                 if (delpol)
1522                         break;
1523         }
1524
1525         if (newpos)
1526                 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1527         else
1528                 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1529
1530         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1531                 pol->pos = i;
1532                 i++;
1533         }
1534 }
1535
1536 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1537                                                    struct xfrm_policy *policy,
1538                                                    bool excl)
1539 {
1540         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1541
1542         hlist_for_each_entry(pol, chain, bydst) {
1543                 if (pol->type == policy->type &&
1544                     pol->if_id == policy->if_id &&
1545                     !selector_cmp(&pol->selector, &policy->selector) &&
1546                     xfrm_policy_mark_match(&policy->mark, pol) &&
1547                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1548                     !WARN_ON(delpol)) {
1549                         if (excl)
1550                                 return ERR_PTR(-EEXIST);
1551                         delpol = pol;
1552                         if (policy->priority > pol->priority)
1553                                 continue;
1554                 } else if (policy->priority >= pol->priority) {
1555                         newpos = pol;
1556                         continue;
1557                 }
1558                 if (delpol)
1559                         break;
1560         }
1561
1562         if (newpos)
1563                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1564         else
1565                 hlist_add_head_rcu(&policy->bydst, chain);
1566
1567         return delpol;
1568 }
1569
1570 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1571 {
1572         struct net *net = xp_net(policy);
1573         struct xfrm_policy *delpol;
1574         struct hlist_head *chain;
1575
1576         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1577         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1578         if (chain)
1579                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1580         else
1581                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1582
1583         if (IS_ERR(delpol)) {
1584                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1585                 return PTR_ERR(delpol);
1586         }
1587
1588         __xfrm_policy_link(policy, dir);
1589
1590         /* After previous checking, family can either be AF_INET or AF_INET6 */
1591         if (policy->family == AF_INET)
1592                 rt_genid_bump_ipv4(net);
1593         else
1594                 rt_genid_bump_ipv6(net);
1595
1596         if (delpol) {
1597                 xfrm_policy_requeue(delpol, policy);
1598                 __xfrm_policy_unlink(delpol, dir);
1599         }
1600         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1601         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1602         policy->curlft.add_time = ktime_get_real_seconds();
1603         policy->curlft.use_time = 0;
1604         if (!mod_timer(&policy->timer, jiffies + HZ))
1605                 xfrm_pol_hold(policy);
1606         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1607
1608         if (delpol)
1609                 xfrm_policy_kill(delpol);
1610         else if (xfrm_bydst_should_resize(net, dir, NULL))
1611                 schedule_work(&net->xfrm.policy_hash_work);
1612
1613         return 0;
1614 }
1615 EXPORT_SYMBOL(xfrm_policy_insert);
1616
1617 static struct xfrm_policy *
1618 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1619                         u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1620                         struct xfrm_sec_ctx *ctx)
1621 {
1622         struct xfrm_policy *pol;
1623
1624         if (!chain)
1625                 return NULL;
1626
1627         hlist_for_each_entry(pol, chain, bydst) {
1628                 if (pol->type == type &&
1629                     pol->if_id == if_id &&
1630                     xfrm_policy_mark_match(mark, pol) &&
1631                     !selector_cmp(sel, &pol->selector) &&
1632                     xfrm_sec_ctx_match(ctx, pol->security))
1633                         return pol;
1634         }
1635
1636         return NULL;
1637 }
1638
1639 struct xfrm_policy *
1640 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1641                       u8 type, int dir, struct xfrm_selector *sel,
1642                       struct xfrm_sec_ctx *ctx, int delete, int *err)
1643 {
1644         struct xfrm_pol_inexact_bin *bin = NULL;
1645         struct xfrm_policy *pol, *ret = NULL;
1646         struct hlist_head *chain;
1647
1648         *err = 0;
1649         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1650         chain = policy_hash_bysel(net, sel, sel->family, dir);
1651         if (!chain) {
1652                 struct xfrm_pol_inexact_candidates cand;
1653                 int i;
1654
1655                 bin = xfrm_policy_inexact_lookup(net, type,
1656                                                  sel->family, dir, if_id);
1657                 if (!bin) {
1658                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1659                         return NULL;
1660                 }
1661
1662                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1663                                                          &sel->saddr,
1664                                                          &sel->daddr)) {
1665                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1666                         return NULL;
1667                 }
1668
1669                 pol = NULL;
1670                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1671                         struct xfrm_policy *tmp;
1672
1673                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1674                                                       if_id, type, dir,
1675                                                       sel, ctx);
1676                         if (!tmp)
1677                                 continue;
1678
1679                         if (!pol || tmp->pos < pol->pos)
1680                                 pol = tmp;
1681                 }
1682         } else {
1683                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1684                                               sel, ctx);
1685         }
1686
1687         if (pol) {
1688                 xfrm_pol_hold(pol);
1689                 if (delete) {
1690                         *err = security_xfrm_policy_delete(pol->security);
1691                         if (*err) {
1692                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1693                                 return pol;
1694                         }
1695                         __xfrm_policy_unlink(pol, dir);
1696                 }
1697                 ret = pol;
1698         }
1699         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1700
1701         if (ret && delete)
1702                 xfrm_policy_kill(ret);
1703         if (bin && delete)
1704                 xfrm_policy_inexact_prune_bin(bin);
1705         return ret;
1706 }
1707 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1708
1709 struct xfrm_policy *
1710 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1711                  u8 type, int dir, u32 id, int delete, int *err)
1712 {
1713         struct xfrm_policy *pol, *ret;
1714         struct hlist_head *chain;
1715
1716         *err = -ENOENT;
1717         if (xfrm_policy_id2dir(id) != dir)
1718                 return NULL;
1719
1720         *err = 0;
1721         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1722         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1723         ret = NULL;
1724         hlist_for_each_entry(pol, chain, byidx) {
1725                 if (pol->type == type && pol->index == id &&
1726                     pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1727                         xfrm_pol_hold(pol);
1728                         if (delete) {
1729                                 *err = security_xfrm_policy_delete(
1730                                                                 pol->security);
1731                                 if (*err) {
1732                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1733                                         return pol;
1734                                 }
1735                                 __xfrm_policy_unlink(pol, dir);
1736                         }
1737                         ret = pol;
1738                         break;
1739                 }
1740         }
1741         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1742
1743         if (ret && delete)
1744                 xfrm_policy_kill(ret);
1745         return ret;
1746 }
1747 EXPORT_SYMBOL(xfrm_policy_byid);
1748
1749 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1750 static inline int
1751 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1752 {
1753         struct xfrm_policy *pol;
1754         int err = 0;
1755
1756         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1757                 if (pol->walk.dead ||
1758                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1759                     pol->type != type)
1760                         continue;
1761
1762                 err = security_xfrm_policy_delete(pol->security);
1763                 if (err) {
1764                         xfrm_audit_policy_delete(pol, 0, task_valid);
1765                         return err;
1766                 }
1767         }
1768         return err;
1769 }
1770 #else
1771 static inline int
1772 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1773 {
1774         return 0;
1775 }
1776 #endif
1777
1778 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1779 {
1780         int dir, err = 0, cnt = 0;
1781         struct xfrm_policy *pol;
1782
1783         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1784
1785         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1786         if (err)
1787                 goto out;
1788
1789 again:
1790         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1791                 dir = xfrm_policy_id2dir(pol->index);
1792                 if (pol->walk.dead ||
1793                     dir >= XFRM_POLICY_MAX ||
1794                     pol->type != type)
1795                         continue;
1796
1797                 __xfrm_policy_unlink(pol, dir);
1798                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1799                 cnt++;
1800                 xfrm_audit_policy_delete(pol, 1, task_valid);
1801                 xfrm_policy_kill(pol);
1802                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1803                 goto again;
1804         }
1805         if (cnt)
1806                 __xfrm_policy_inexact_flush(net);
1807         else
1808                 err = -ESRCH;
1809 out:
1810         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1811         return err;
1812 }
1813 EXPORT_SYMBOL(xfrm_policy_flush);
1814
1815 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1816                      int (*func)(struct xfrm_policy *, int, int, void*),
1817                      void *data)
1818 {
1819         struct xfrm_policy *pol;
1820         struct xfrm_policy_walk_entry *x;
1821         int error = 0;
1822
1823         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1824             walk->type != XFRM_POLICY_TYPE_ANY)
1825                 return -EINVAL;
1826
1827         if (list_empty(&walk->walk.all) && walk->seq != 0)
1828                 return 0;
1829
1830         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1831         if (list_empty(&walk->walk.all))
1832                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1833         else
1834                 x = list_first_entry(&walk->walk.all,
1835                                      struct xfrm_policy_walk_entry, all);
1836
1837         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1838                 if (x->dead)
1839                         continue;
1840                 pol = container_of(x, struct xfrm_policy, walk);
1841                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1842                     walk->type != pol->type)
1843                         continue;
1844                 error = func(pol, xfrm_policy_id2dir(pol->index),
1845                              walk->seq, data);
1846                 if (error) {
1847                         list_move_tail(&walk->walk.all, &x->all);
1848                         goto out;
1849                 }
1850                 walk->seq++;
1851         }
1852         if (walk->seq == 0) {
1853                 error = -ENOENT;
1854                 goto out;
1855         }
1856         list_del_init(&walk->walk.all);
1857 out:
1858         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1859         return error;
1860 }
1861 EXPORT_SYMBOL(xfrm_policy_walk);
1862
1863 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1864 {
1865         INIT_LIST_HEAD(&walk->walk.all);
1866         walk->walk.dead = 1;
1867         walk->type = type;
1868         walk->seq = 0;
1869 }
1870 EXPORT_SYMBOL(xfrm_policy_walk_init);
1871
1872 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1873 {
1874         if (list_empty(&walk->walk.all))
1875                 return;
1876
1877         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1878         list_del(&walk->walk.all);
1879         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1880 }
1881 EXPORT_SYMBOL(xfrm_policy_walk_done);
1882
1883 /*
1884  * Find policy to apply to this flow.
1885  *
1886  * Returns 0 if policy found, else an -errno.
1887  */
1888 static int xfrm_policy_match(const struct xfrm_policy *pol,
1889                              const struct flowi *fl,
1890                              u8 type, u16 family, int dir, u32 if_id)
1891 {
1892         const struct xfrm_selector *sel = &pol->selector;
1893         int ret = -ESRCH;
1894         bool match;
1895
1896         if (pol->family != family ||
1897             pol->if_id != if_id ||
1898             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1899             pol->type != type)
1900                 return ret;
1901
1902         match = xfrm_selector_match(sel, fl, family);
1903         if (match)
1904                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1905         return ret;
1906 }
1907
1908 static struct xfrm_pol_inexact_node *
1909 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1910                                 seqcount_spinlock_t *count,
1911                                 const xfrm_address_t *addr, u16 family)
1912 {
1913         const struct rb_node *parent;
1914         int seq;
1915
1916 again:
1917         seq = read_seqcount_begin(count);
1918
1919         parent = rcu_dereference_raw(r->rb_node);
1920         while (parent) {
1921                 struct xfrm_pol_inexact_node *node;
1922                 int delta;
1923
1924                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1925
1926                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1927                                                node->prefixlen, family);
1928                 if (delta < 0) {
1929                         parent = rcu_dereference_raw(parent->rb_left);
1930                         continue;
1931                 } else if (delta > 0) {
1932                         parent = rcu_dereference_raw(parent->rb_right);
1933                         continue;
1934                 }
1935
1936                 return node;
1937         }
1938
1939         if (read_seqcount_retry(count, seq))
1940                 goto again;
1941
1942         return NULL;
1943 }
1944
1945 static bool
1946 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1947                                     struct xfrm_pol_inexact_bin *b,
1948                                     const xfrm_address_t *saddr,
1949                                     const xfrm_address_t *daddr)
1950 {
1951         struct xfrm_pol_inexact_node *n;
1952         u16 family;
1953
1954         if (!b)
1955                 return false;
1956
1957         family = b->k.family;
1958         memset(cand, 0, sizeof(*cand));
1959         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1960
1961         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1962                                             family);
1963         if (n) {
1964                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1965                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1966                                                     family);
1967                 if (n)
1968                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1969         }
1970
1971         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1972                                             family);
1973         if (n)
1974                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1975
1976         return true;
1977 }
1978
1979 static struct xfrm_pol_inexact_bin *
1980 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1981                                u8 dir, u32 if_id)
1982 {
1983         struct xfrm_pol_inexact_key k = {
1984                 .family = family,
1985                 .type = type,
1986                 .dir = dir,
1987                 .if_id = if_id,
1988         };
1989
1990         write_pnet(&k.net, net);
1991
1992         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1993                                  xfrm_pol_inexact_params);
1994 }
1995
1996 static struct xfrm_pol_inexact_bin *
1997 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1998                            u8 dir, u32 if_id)
1999 {
2000         struct xfrm_pol_inexact_bin *bin;
2001
2002         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2003
2004         rcu_read_lock();
2005         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2006         rcu_read_unlock();
2007
2008         return bin;
2009 }
2010
2011 static struct xfrm_policy *
2012 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2013                               struct xfrm_policy *prefer,
2014                               const struct flowi *fl,
2015                               u8 type, u16 family, int dir, u32 if_id)
2016 {
2017         u32 priority = prefer ? prefer->priority : ~0u;
2018         struct xfrm_policy *pol;
2019
2020         if (!chain)
2021                 return NULL;
2022
2023         hlist_for_each_entry_rcu(pol, chain, bydst) {
2024                 int err;
2025
2026                 if (pol->priority > priority)
2027                         break;
2028
2029                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2030                 if (err) {
2031                         if (err != -ESRCH)
2032                                 return ERR_PTR(err);
2033
2034                         continue;
2035                 }
2036
2037                 if (prefer) {
2038                         /* matches.  Is it older than *prefer? */
2039                         if (pol->priority == priority &&
2040                             prefer->pos < pol->pos)
2041                                 return prefer;
2042                 }
2043
2044                 return pol;
2045         }
2046
2047         return NULL;
2048 }
2049
2050 static struct xfrm_policy *
2051 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2052                             struct xfrm_policy *prefer,
2053                             const struct flowi *fl,
2054                             u8 type, u16 family, int dir, u32 if_id)
2055 {
2056         struct xfrm_policy *tmp;
2057         int i;
2058
2059         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2060                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2061                                                     prefer,
2062                                                     fl, type, family, dir,
2063                                                     if_id);
2064                 if (!tmp)
2065                         continue;
2066
2067                 if (IS_ERR(tmp))
2068                         return tmp;
2069                 prefer = tmp;
2070         }
2071
2072         return prefer;
2073 }
2074
2075 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2076                                                      const struct flowi *fl,
2077                                                      u16 family, u8 dir,
2078                                                      u32 if_id)
2079 {
2080         struct xfrm_pol_inexact_candidates cand;
2081         const xfrm_address_t *daddr, *saddr;
2082         struct xfrm_pol_inexact_bin *bin;
2083         struct xfrm_policy *pol, *ret;
2084         struct hlist_head *chain;
2085         unsigned int sequence;
2086         int err;
2087
2088         daddr = xfrm_flowi_daddr(fl, family);
2089         saddr = xfrm_flowi_saddr(fl, family);
2090         if (unlikely(!daddr || !saddr))
2091                 return NULL;
2092
2093         rcu_read_lock();
2094  retry:
2095         do {
2096                 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2097                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2098         } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2099
2100         ret = NULL;
2101         hlist_for_each_entry_rcu(pol, chain, bydst) {
2102                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2103                 if (err) {
2104                         if (err == -ESRCH)
2105                                 continue;
2106                         else {
2107                                 ret = ERR_PTR(err);
2108                                 goto fail;
2109                         }
2110                 } else {
2111                         ret = pol;
2112                         break;
2113                 }
2114         }
2115         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2116         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2117                                                          daddr))
2118                 goto skip_inexact;
2119
2120         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2121                                           family, dir, if_id);
2122         if (pol) {
2123                 ret = pol;
2124                 if (IS_ERR(pol))
2125                         goto fail;
2126         }
2127
2128 skip_inexact:
2129         if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2130                 goto retry;
2131
2132         if (ret && !xfrm_pol_hold_rcu(ret))
2133                 goto retry;
2134 fail:
2135         rcu_read_unlock();
2136
2137         return ret;
2138 }
2139
2140 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2141                                               const struct flowi *fl,
2142                                               u16 family, u8 dir, u32 if_id)
2143 {
2144 #ifdef CONFIG_XFRM_SUB_POLICY
2145         struct xfrm_policy *pol;
2146
2147         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2148                                         dir, if_id);
2149         if (pol != NULL)
2150                 return pol;
2151 #endif
2152         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2153                                          dir, if_id);
2154 }
2155
2156 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2157                                                  const struct flowi *fl,
2158                                                  u16 family, u32 if_id)
2159 {
2160         struct xfrm_policy *pol;
2161
2162         rcu_read_lock();
2163  again:
2164         pol = rcu_dereference(sk->sk_policy[dir]);
2165         if (pol != NULL) {
2166                 bool match;
2167                 int err = 0;
2168
2169                 if (pol->family != family) {
2170                         pol = NULL;
2171                         goto out;
2172                 }
2173
2174                 match = xfrm_selector_match(&pol->selector, fl, family);
2175                 if (match) {
2176                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2177                             pol->if_id != if_id) {
2178                                 pol = NULL;
2179                                 goto out;
2180                         }
2181                         err = security_xfrm_policy_lookup(pol->security,
2182                                                       fl->flowi_secid);
2183                         if (!err) {
2184                                 if (!xfrm_pol_hold_rcu(pol))
2185                                         goto again;
2186                         } else if (err == -ESRCH) {
2187                                 pol = NULL;
2188                         } else {
2189                                 pol = ERR_PTR(err);
2190                         }
2191                 } else
2192                         pol = NULL;
2193         }
2194 out:
2195         rcu_read_unlock();
2196         return pol;
2197 }
2198
2199 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2200 {
2201         struct net *net = xp_net(pol);
2202
2203         list_add(&pol->walk.all, &net->xfrm.policy_all);
2204         net->xfrm.policy_count[dir]++;
2205         xfrm_pol_hold(pol);
2206 }
2207
2208 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2209                                                 int dir)
2210 {
2211         struct net *net = xp_net(pol);
2212
2213         if (list_empty(&pol->walk.all))
2214                 return NULL;
2215
2216         /* Socket policies are not hashed. */
2217         if (!hlist_unhashed(&pol->bydst)) {
2218                 hlist_del_rcu(&pol->bydst);
2219                 hlist_del_init(&pol->bydst_inexact_list);
2220                 hlist_del(&pol->byidx);
2221         }
2222
2223         list_del_init(&pol->walk.all);
2224         net->xfrm.policy_count[dir]--;
2225
2226         return pol;
2227 }
2228
2229 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2230 {
2231         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2232 }
2233
2234 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2235 {
2236         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2237 }
2238
2239 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2240 {
2241         struct net *net = xp_net(pol);
2242
2243         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2244         pol = __xfrm_policy_unlink(pol, dir);
2245         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2246         if (pol) {
2247                 xfrm_policy_kill(pol);
2248                 return 0;
2249         }
2250         return -ENOENT;
2251 }
2252 EXPORT_SYMBOL(xfrm_policy_delete);
2253
2254 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2255 {
2256         struct net *net = sock_net(sk);
2257         struct xfrm_policy *old_pol;
2258
2259 #ifdef CONFIG_XFRM_SUB_POLICY
2260         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2261                 return -EINVAL;
2262 #endif
2263
2264         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2265         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2266                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2267         if (pol) {
2268                 pol->curlft.add_time = ktime_get_real_seconds();
2269                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2270                 xfrm_sk_policy_link(pol, dir);
2271         }
2272         rcu_assign_pointer(sk->sk_policy[dir], pol);
2273         if (old_pol) {
2274                 if (pol)
2275                         xfrm_policy_requeue(old_pol, pol);
2276
2277                 /* Unlinking succeeds always. This is the only function
2278                  * allowed to delete or replace socket policy.
2279                  */
2280                 xfrm_sk_policy_unlink(old_pol, dir);
2281         }
2282         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2283
2284         if (old_pol) {
2285                 xfrm_policy_kill(old_pol);
2286         }
2287         return 0;
2288 }
2289
2290 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2291 {
2292         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2293         struct net *net = xp_net(old);
2294
2295         if (newp) {
2296                 newp->selector = old->selector;
2297                 if (security_xfrm_policy_clone(old->security,
2298                                                &newp->security)) {
2299                         kfree(newp);
2300                         return NULL;  /* ENOMEM */
2301                 }
2302                 newp->lft = old->lft;
2303                 newp->curlft = old->curlft;
2304                 newp->mark = old->mark;
2305                 newp->if_id = old->if_id;
2306                 newp->action = old->action;
2307                 newp->flags = old->flags;
2308                 newp->xfrm_nr = old->xfrm_nr;
2309                 newp->index = old->index;
2310                 newp->type = old->type;
2311                 newp->family = old->family;
2312                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2313                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2314                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2315                 xfrm_sk_policy_link(newp, dir);
2316                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2317                 xfrm_pol_put(newp);
2318         }
2319         return newp;
2320 }
2321
2322 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2323 {
2324         const struct xfrm_policy *p;
2325         struct xfrm_policy *np;
2326         int i, ret = 0;
2327
2328         rcu_read_lock();
2329         for (i = 0; i < 2; i++) {
2330                 p = rcu_dereference(osk->sk_policy[i]);
2331                 if (p) {
2332                         np = clone_policy(p, i);
2333                         if (unlikely(!np)) {
2334                                 ret = -ENOMEM;
2335                                 break;
2336                         }
2337                         rcu_assign_pointer(sk->sk_policy[i], np);
2338                 }
2339         }
2340         rcu_read_unlock();
2341         return ret;
2342 }
2343
2344 static int
2345 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2346                xfrm_address_t *remote, unsigned short family, u32 mark)
2347 {
2348         int err;
2349         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2350
2351         if (unlikely(afinfo == NULL))
2352                 return -EINVAL;
2353         err = afinfo->get_saddr(net, oif, local, remote, mark);
2354         rcu_read_unlock();
2355         return err;
2356 }
2357
2358 /* Resolve list of templates for the flow, given policy. */
2359
2360 static int
2361 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2362                       struct xfrm_state **xfrm, unsigned short family)
2363 {
2364         struct net *net = xp_net(policy);
2365         int nx;
2366         int i, error;
2367         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2368         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2369         xfrm_address_t tmp;
2370
2371         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2372                 struct xfrm_state *x;
2373                 xfrm_address_t *remote = daddr;
2374                 xfrm_address_t *local  = saddr;
2375                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2376
2377                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2378                     tmpl->mode == XFRM_MODE_BEET) {
2379                         remote = &tmpl->id.daddr;
2380                         local = &tmpl->saddr;
2381                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2382                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2383                                                        &tmp, remote,
2384                                                        tmpl->encap_family, 0);
2385                                 if (error)
2386                                         goto fail;
2387                                 local = &tmp;
2388                         }
2389                 }
2390
2391                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2392                                     family, policy->if_id);
2393
2394                 if (x && x->km.state == XFRM_STATE_VALID) {
2395                         xfrm[nx++] = x;
2396                         daddr = remote;
2397                         saddr = local;
2398                         continue;
2399                 }
2400                 if (x) {
2401                         error = (x->km.state == XFRM_STATE_ERROR ?
2402                                  -EINVAL : -EAGAIN);
2403                         xfrm_state_put(x);
2404                 } else if (error == -ESRCH) {
2405                         error = -EAGAIN;
2406                 }
2407
2408                 if (!tmpl->optional)
2409                         goto fail;
2410         }
2411         return nx;
2412
2413 fail:
2414         for (nx--; nx >= 0; nx--)
2415                 xfrm_state_put(xfrm[nx]);
2416         return error;
2417 }
2418
2419 static int
2420 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2421                   struct xfrm_state **xfrm, unsigned short family)
2422 {
2423         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2424         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2425         int cnx = 0;
2426         int error;
2427         int ret;
2428         int i;
2429
2430         for (i = 0; i < npols; i++) {
2431                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2432                         error = -ENOBUFS;
2433                         goto fail;
2434                 }
2435
2436                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2437                 if (ret < 0) {
2438                         error = ret;
2439                         goto fail;
2440                 } else
2441                         cnx += ret;
2442         }
2443
2444         /* found states are sorted for outbound processing */
2445         if (npols > 1)
2446                 xfrm_state_sort(xfrm, tpp, cnx, family);
2447
2448         return cnx;
2449
2450  fail:
2451         for (cnx--; cnx >= 0; cnx--)
2452                 xfrm_state_put(tpp[cnx]);
2453         return error;
2454
2455 }
2456
2457 static int xfrm_get_tos(const struct flowi *fl, int family)
2458 {
2459         if (family == AF_INET)
2460                 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2461
2462         return 0;
2463 }
2464
2465 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2466 {
2467         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2468         struct dst_ops *dst_ops;
2469         struct xfrm_dst *xdst;
2470
2471         if (!afinfo)
2472                 return ERR_PTR(-EINVAL);
2473
2474         switch (family) {
2475         case AF_INET:
2476                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2477                 break;
2478 #if IS_ENABLED(CONFIG_IPV6)
2479         case AF_INET6:
2480                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2481                 break;
2482 #endif
2483         default:
2484                 BUG();
2485         }
2486         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2487
2488         if (likely(xdst)) {
2489                 struct dst_entry *dst = &xdst->u.dst;
2490
2491                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2492         } else
2493                 xdst = ERR_PTR(-ENOBUFS);
2494
2495         rcu_read_unlock();
2496
2497         return xdst;
2498 }
2499
2500 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2501                            int nfheader_len)
2502 {
2503         if (dst->ops->family == AF_INET6) {
2504                 struct rt6_info *rt = (struct rt6_info *)dst;
2505                 path->path_cookie = rt6_get_cookie(rt);
2506                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2507         }
2508 }
2509
2510 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2511                                 const struct flowi *fl)
2512 {
2513         const struct xfrm_policy_afinfo *afinfo =
2514                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2515         int err;
2516
2517         if (!afinfo)
2518                 return -EINVAL;
2519
2520         err = afinfo->fill_dst(xdst, dev, fl);
2521
2522         rcu_read_unlock();
2523
2524         return err;
2525 }
2526
2527
2528 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2529  * all the metrics... Shortly, bundle a bundle.
2530  */
2531
2532 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2533                                             struct xfrm_state **xfrm,
2534                                             struct xfrm_dst **bundle,
2535                                             int nx,
2536                                             const struct flowi *fl,
2537                                             struct dst_entry *dst)
2538 {
2539         const struct xfrm_state_afinfo *afinfo;
2540         const struct xfrm_mode *inner_mode;
2541         struct net *net = xp_net(policy);
2542         unsigned long now = jiffies;
2543         struct net_device *dev;
2544         struct xfrm_dst *xdst_prev = NULL;
2545         struct xfrm_dst *xdst0 = NULL;
2546         int i = 0;
2547         int err;
2548         int header_len = 0;
2549         int nfheader_len = 0;
2550         int trailer_len = 0;
2551         int tos;
2552         int family = policy->selector.family;
2553         xfrm_address_t saddr, daddr;
2554
2555         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2556
2557         tos = xfrm_get_tos(fl, family);
2558
2559         dst_hold(dst);
2560
2561         for (; i < nx; i++) {
2562                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2563                 struct dst_entry *dst1 = &xdst->u.dst;
2564
2565                 err = PTR_ERR(xdst);
2566                 if (IS_ERR(xdst)) {
2567                         dst_release(dst);
2568                         goto put_states;
2569                 }
2570
2571                 bundle[i] = xdst;
2572                 if (!xdst_prev)
2573                         xdst0 = xdst;
2574                 else
2575                         /* Ref count is taken during xfrm_alloc_dst()
2576                          * No need to do dst_clone() on dst1
2577                          */
2578                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2579
2580                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2581                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2582                                                         xfrm_af2proto(family));
2583                         if (!inner_mode) {
2584                                 err = -EAFNOSUPPORT;
2585                                 dst_release(dst);
2586                                 goto put_states;
2587                         }
2588                 } else
2589                         inner_mode = &xfrm[i]->inner_mode;
2590
2591                 xdst->route = dst;
2592                 dst_copy_metrics(dst1, dst);
2593
2594                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2595                         __u32 mark = 0;
2596
2597                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2598                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2599
2600                         family = xfrm[i]->props.family;
2601                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2602                                               &saddr, &daddr, family, mark);
2603                         err = PTR_ERR(dst);
2604                         if (IS_ERR(dst))
2605                                 goto put_states;
2606                 } else
2607                         dst_hold(dst);
2608
2609                 dst1->xfrm = xfrm[i];
2610                 xdst->xfrm_genid = xfrm[i]->genid;
2611
2612                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2613                 dst1->lastuse = now;
2614
2615                 dst1->input = dst_discard;
2616
2617                 rcu_read_lock();
2618                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2619                 if (likely(afinfo))
2620                         dst1->output = afinfo->output;
2621                 else
2622                         dst1->output = dst_discard_out;
2623                 rcu_read_unlock();
2624
2625                 xdst_prev = xdst;
2626
2627                 header_len += xfrm[i]->props.header_len;
2628                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2629                         nfheader_len += xfrm[i]->props.header_len;
2630                 trailer_len += xfrm[i]->props.trailer_len;
2631         }
2632
2633         xfrm_dst_set_child(xdst_prev, dst);
2634         xdst0->path = dst;
2635
2636         err = -ENODEV;
2637         dev = dst->dev;
2638         if (!dev)
2639                 goto free_dst;
2640
2641         xfrm_init_path(xdst0, dst, nfheader_len);
2642         xfrm_init_pmtu(bundle, nx);
2643
2644         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2645              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2646                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2647                 if (err)
2648                         goto free_dst;
2649
2650                 xdst_prev->u.dst.header_len = header_len;
2651                 xdst_prev->u.dst.trailer_len = trailer_len;
2652                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2653                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2654         }
2655
2656         return &xdst0->u.dst;
2657
2658 put_states:
2659         for (; i < nx; i++)
2660                 xfrm_state_put(xfrm[i]);
2661 free_dst:
2662         if (xdst0)
2663                 dst_release_immediate(&xdst0->u.dst);
2664
2665         return ERR_PTR(err);
2666 }
2667
2668 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2669                                 struct xfrm_policy **pols,
2670                                 int *num_pols, int *num_xfrms)
2671 {
2672         int i;
2673
2674         if (*num_pols == 0 || !pols[0]) {
2675                 *num_pols = 0;
2676                 *num_xfrms = 0;
2677                 return 0;
2678         }
2679         if (IS_ERR(pols[0])) {
2680                 *num_pols = 0;
2681                 return PTR_ERR(pols[0]);
2682         }
2683
2684         *num_xfrms = pols[0]->xfrm_nr;
2685
2686 #ifdef CONFIG_XFRM_SUB_POLICY
2687         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2688             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2689                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2690                                                     XFRM_POLICY_TYPE_MAIN,
2691                                                     fl, family,
2692                                                     XFRM_POLICY_OUT,
2693                                                     pols[0]->if_id);
2694                 if (pols[1]) {
2695                         if (IS_ERR(pols[1])) {
2696                                 xfrm_pols_put(pols, *num_pols);
2697                                 *num_pols = 0;
2698                                 return PTR_ERR(pols[1]);
2699                         }
2700                         (*num_pols)++;
2701                         (*num_xfrms) += pols[1]->xfrm_nr;
2702                 }
2703         }
2704 #endif
2705         for (i = 0; i < *num_pols; i++) {
2706                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2707                         *num_xfrms = -1;
2708                         break;
2709                 }
2710         }
2711
2712         return 0;
2713
2714 }
2715
2716 static struct xfrm_dst *
2717 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2718                                const struct flowi *fl, u16 family,
2719                                struct dst_entry *dst_orig)
2720 {
2721         struct net *net = xp_net(pols[0]);
2722         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2723         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2724         struct xfrm_dst *xdst;
2725         struct dst_entry *dst;
2726         int err;
2727
2728         /* Try to instantiate a bundle */
2729         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2730         if (err <= 0) {
2731                 if (err == 0)
2732                         return NULL;
2733
2734                 if (err != -EAGAIN)
2735                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2736                 return ERR_PTR(err);
2737         }
2738
2739         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2740         if (IS_ERR(dst)) {
2741                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2742                 return ERR_CAST(dst);
2743         }
2744
2745         xdst = (struct xfrm_dst *)dst;
2746         xdst->num_xfrms = err;
2747         xdst->num_pols = num_pols;
2748         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2749         xdst->policy_genid = atomic_read(&pols[0]->genid);
2750
2751         return xdst;
2752 }
2753
2754 static void xfrm_policy_queue_process(struct timer_list *t)
2755 {
2756         struct sk_buff *skb;
2757         struct sock *sk;
2758         struct dst_entry *dst;
2759         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2760         struct net *net = xp_net(pol);
2761         struct xfrm_policy_queue *pq = &pol->polq;
2762         struct flowi fl;
2763         struct sk_buff_head list;
2764         __u32 skb_mark;
2765
2766         spin_lock(&pq->hold_queue.lock);
2767         skb = skb_peek(&pq->hold_queue);
2768         if (!skb) {
2769                 spin_unlock(&pq->hold_queue.lock);
2770                 goto out;
2771         }
2772         dst = skb_dst(skb);
2773         sk = skb->sk;
2774
2775         /* Fixup the mark to support VTI. */
2776         skb_mark = skb->mark;
2777         skb->mark = pol->mark.v;
2778         xfrm_decode_session(skb, &fl, dst->ops->family);
2779         skb->mark = skb_mark;
2780         spin_unlock(&pq->hold_queue.lock);
2781
2782         dst_hold(xfrm_dst_path(dst));
2783         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2784         if (IS_ERR(dst))
2785                 goto purge_queue;
2786
2787         if (dst->flags & DST_XFRM_QUEUE) {
2788                 dst_release(dst);
2789
2790                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2791                         goto purge_queue;
2792
2793                 pq->timeout = pq->timeout << 1;
2794                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2795                         xfrm_pol_hold(pol);
2796                 goto out;
2797         }
2798
2799         dst_release(dst);
2800
2801         __skb_queue_head_init(&list);
2802
2803         spin_lock(&pq->hold_queue.lock);
2804         pq->timeout = 0;
2805         skb_queue_splice_init(&pq->hold_queue, &list);
2806         spin_unlock(&pq->hold_queue.lock);
2807
2808         while (!skb_queue_empty(&list)) {
2809                 skb = __skb_dequeue(&list);
2810
2811                 /* Fixup the mark to support VTI. */
2812                 skb_mark = skb->mark;
2813                 skb->mark = pol->mark.v;
2814                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2815                 skb->mark = skb_mark;
2816
2817                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2818                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2819                 if (IS_ERR(dst)) {
2820                         kfree_skb(skb);
2821                         continue;
2822                 }
2823
2824                 nf_reset_ct(skb);
2825                 skb_dst_drop(skb);
2826                 skb_dst_set(skb, dst);
2827
2828                 dst_output(net, skb->sk, skb);
2829         }
2830
2831 out:
2832         xfrm_pol_put(pol);
2833         return;
2834
2835 purge_queue:
2836         pq->timeout = 0;
2837         skb_queue_purge(&pq->hold_queue);
2838         xfrm_pol_put(pol);
2839 }
2840
2841 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2842 {
2843         unsigned long sched_next;
2844         struct dst_entry *dst = skb_dst(skb);
2845         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2846         struct xfrm_policy *pol = xdst->pols[0];
2847         struct xfrm_policy_queue *pq = &pol->polq;
2848
2849         if (unlikely(skb_fclone_busy(sk, skb))) {
2850                 kfree_skb(skb);
2851                 return 0;
2852         }
2853
2854         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2855                 kfree_skb(skb);
2856                 return -EAGAIN;
2857         }
2858
2859         skb_dst_force(skb);
2860
2861         spin_lock_bh(&pq->hold_queue.lock);
2862
2863         if (!pq->timeout)
2864                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2865
2866         sched_next = jiffies + pq->timeout;
2867
2868         if (del_timer(&pq->hold_timer)) {
2869                 if (time_before(pq->hold_timer.expires, sched_next))
2870                         sched_next = pq->hold_timer.expires;
2871                 xfrm_pol_put(pol);
2872         }
2873
2874         __skb_queue_tail(&pq->hold_queue, skb);
2875         if (!mod_timer(&pq->hold_timer, sched_next))
2876                 xfrm_pol_hold(pol);
2877
2878         spin_unlock_bh(&pq->hold_queue.lock);
2879
2880         return 0;
2881 }
2882
2883 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2884                                                  struct xfrm_flo *xflo,
2885                                                  const struct flowi *fl,
2886                                                  int num_xfrms,
2887                                                  u16 family)
2888 {
2889         int err;
2890         struct net_device *dev;
2891         struct dst_entry *dst;
2892         struct dst_entry *dst1;
2893         struct xfrm_dst *xdst;
2894
2895         xdst = xfrm_alloc_dst(net, family);
2896         if (IS_ERR(xdst))
2897                 return xdst;
2898
2899         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2900             net->xfrm.sysctl_larval_drop ||
2901             num_xfrms <= 0)
2902                 return xdst;
2903
2904         dst = xflo->dst_orig;
2905         dst1 = &xdst->u.dst;
2906         dst_hold(dst);
2907         xdst->route = dst;
2908
2909         dst_copy_metrics(dst1, dst);
2910
2911         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2912         dst1->flags |= DST_XFRM_QUEUE;
2913         dst1->lastuse = jiffies;
2914
2915         dst1->input = dst_discard;
2916         dst1->output = xdst_queue_output;
2917
2918         dst_hold(dst);
2919         xfrm_dst_set_child(xdst, dst);
2920         xdst->path = dst;
2921
2922         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2923
2924         err = -ENODEV;
2925         dev = dst->dev;
2926         if (!dev)
2927                 goto free_dst;
2928
2929         err = xfrm_fill_dst(xdst, dev, fl);
2930         if (err)
2931                 goto free_dst;
2932
2933 out:
2934         return xdst;
2935
2936 free_dst:
2937         dst_release(dst1);
2938         xdst = ERR_PTR(err);
2939         goto out;
2940 }
2941
2942 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2943                                            const struct flowi *fl,
2944                                            u16 family, u8 dir,
2945                                            struct xfrm_flo *xflo, u32 if_id)
2946 {
2947         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2948         int num_pols = 0, num_xfrms = 0, err;
2949         struct xfrm_dst *xdst;
2950
2951         /* Resolve policies to use if we couldn't get them from
2952          * previous cache entry */
2953         num_pols = 1;
2954         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2955         err = xfrm_expand_policies(fl, family, pols,
2956                                            &num_pols, &num_xfrms);
2957         if (err < 0)
2958                 goto inc_error;
2959         if (num_pols == 0)
2960                 return NULL;
2961         if (num_xfrms <= 0)
2962                 goto make_dummy_bundle;
2963
2964         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2965                                               xflo->dst_orig);
2966         if (IS_ERR(xdst)) {
2967                 err = PTR_ERR(xdst);
2968                 if (err == -EREMOTE) {
2969                         xfrm_pols_put(pols, num_pols);
2970                         return NULL;
2971                 }
2972
2973                 if (err != -EAGAIN)
2974                         goto error;
2975                 goto make_dummy_bundle;
2976         } else if (xdst == NULL) {
2977                 num_xfrms = 0;
2978                 goto make_dummy_bundle;
2979         }
2980
2981         return xdst;
2982
2983 make_dummy_bundle:
2984         /* We found policies, but there's no bundles to instantiate:
2985          * either because the policy blocks, has no transformations or
2986          * we could not build template (no xfrm_states).*/
2987         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2988         if (IS_ERR(xdst)) {
2989                 xfrm_pols_put(pols, num_pols);
2990                 return ERR_CAST(xdst);
2991         }
2992         xdst->num_pols = num_pols;
2993         xdst->num_xfrms = num_xfrms;
2994         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2995
2996         return xdst;
2997
2998 inc_error:
2999         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3000 error:
3001         xfrm_pols_put(pols, num_pols);
3002         return ERR_PTR(err);
3003 }
3004
3005 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3006                                         struct dst_entry *dst_orig)
3007 {
3008         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3009         struct dst_entry *ret;
3010
3011         if (!afinfo) {
3012                 dst_release(dst_orig);
3013                 return ERR_PTR(-EINVAL);
3014         } else {
3015                 ret = afinfo->blackhole_route(net, dst_orig);
3016         }
3017         rcu_read_unlock();
3018
3019         return ret;
3020 }
3021
3022 /* Finds/creates a bundle for given flow and if_id
3023  *
3024  * At the moment we eat a raw IP route. Mostly to speed up lookups
3025  * on interfaces with disabled IPsec.
3026  *
3027  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3028  * compatibility
3029  */
3030 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3031                                         struct dst_entry *dst_orig,
3032                                         const struct flowi *fl,
3033                                         const struct sock *sk,
3034                                         int flags, u32 if_id)
3035 {
3036         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3037         struct xfrm_dst *xdst;
3038         struct dst_entry *dst, *route;
3039         u16 family = dst_orig->ops->family;
3040         u8 dir = XFRM_POLICY_OUT;
3041         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3042
3043         dst = NULL;
3044         xdst = NULL;
3045         route = NULL;
3046
3047         sk = sk_const_to_full_sk(sk);
3048         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3049                 num_pols = 1;
3050                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3051                                                 if_id);
3052                 err = xfrm_expand_policies(fl, family, pols,
3053                                            &num_pols, &num_xfrms);
3054                 if (err < 0)
3055                         goto dropdst;
3056
3057                 if (num_pols) {
3058                         if (num_xfrms <= 0) {
3059                                 drop_pols = num_pols;
3060                                 goto no_transform;
3061                         }
3062
3063                         xdst = xfrm_resolve_and_create_bundle(
3064                                         pols, num_pols, fl,
3065                                         family, dst_orig);
3066
3067                         if (IS_ERR(xdst)) {
3068                                 xfrm_pols_put(pols, num_pols);
3069                                 err = PTR_ERR(xdst);
3070                                 if (err == -EREMOTE)
3071                                         goto nopol;
3072
3073                                 goto dropdst;
3074                         } else if (xdst == NULL) {
3075                                 num_xfrms = 0;
3076                                 drop_pols = num_pols;
3077                                 goto no_transform;
3078                         }
3079
3080                         route = xdst->route;
3081                 }
3082         }
3083
3084         if (xdst == NULL) {
3085                 struct xfrm_flo xflo;
3086
3087                 xflo.dst_orig = dst_orig;
3088                 xflo.flags = flags;
3089
3090                 /* To accelerate a bit...  */
3091                 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3092                                !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3093                         goto nopol;
3094
3095                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3096                 if (xdst == NULL)
3097                         goto nopol;
3098                 if (IS_ERR(xdst)) {
3099                         err = PTR_ERR(xdst);
3100                         goto dropdst;
3101                 }
3102
3103                 num_pols = xdst->num_pols;
3104                 num_xfrms = xdst->num_xfrms;
3105                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3106                 route = xdst->route;
3107         }
3108
3109         dst = &xdst->u.dst;
3110         if (route == NULL && num_xfrms > 0) {
3111                 /* The only case when xfrm_bundle_lookup() returns a
3112                  * bundle with null route, is when the template could
3113                  * not be resolved. It means policies are there, but
3114                  * bundle could not be created, since we don't yet
3115                  * have the xfrm_state's. We need to wait for KM to
3116                  * negotiate new SA's or bail out with error.*/
3117                 if (net->xfrm.sysctl_larval_drop) {
3118                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3119                         err = -EREMOTE;
3120                         goto error;
3121                 }
3122
3123                 err = -EAGAIN;
3124
3125                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3126                 goto error;
3127         }
3128
3129 no_transform:
3130         if (num_pols == 0)
3131                 goto nopol;
3132
3133         if ((flags & XFRM_LOOKUP_ICMP) &&
3134             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3135                 err = -ENOENT;
3136                 goto error;
3137         }
3138
3139         for (i = 0; i < num_pols; i++)
3140                 pols[i]->curlft.use_time = ktime_get_real_seconds();
3141
3142         if (num_xfrms < 0) {
3143                 /* Prohibit the flow */
3144                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3145                 err = -EPERM;
3146                 goto error;
3147         } else if (num_xfrms > 0) {
3148                 /* Flow transformed */
3149                 dst_release(dst_orig);
3150         } else {
3151                 /* Flow passes untransformed */
3152                 dst_release(dst);
3153                 dst = dst_orig;
3154         }
3155 ok:
3156         xfrm_pols_put(pols, drop_pols);
3157         if (dst && dst->xfrm &&
3158             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3159                 dst->flags |= DST_XFRM_TUNNEL;
3160         return dst;
3161
3162 nopol:
3163         if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3164             net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3165                 err = -EPERM;
3166                 goto error;
3167         }
3168         if (!(flags & XFRM_LOOKUP_ICMP)) {
3169                 dst = dst_orig;
3170                 goto ok;
3171         }
3172         err = -ENOENT;
3173 error:
3174         dst_release(dst);
3175 dropdst:
3176         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3177                 dst_release(dst_orig);
3178         xfrm_pols_put(pols, drop_pols);
3179         return ERR_PTR(err);
3180 }
3181 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3182
3183 /* Main function: finds/creates a bundle for given flow.
3184  *
3185  * At the moment we eat a raw IP route. Mostly to speed up lookups
3186  * on interfaces with disabled IPsec.
3187  */
3188 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3189                               const struct flowi *fl, const struct sock *sk,
3190                               int flags)
3191 {
3192         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3193 }
3194 EXPORT_SYMBOL(xfrm_lookup);
3195
3196 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3197  * Otherwise we may send out blackholed packets.
3198  */
3199 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3200                                     const struct flowi *fl,
3201                                     const struct sock *sk, int flags)
3202 {
3203         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3204                                             flags | XFRM_LOOKUP_QUEUE |
3205                                             XFRM_LOOKUP_KEEP_DST_REF);
3206
3207         if (PTR_ERR(dst) == -EREMOTE)
3208                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3209
3210         if (IS_ERR(dst))
3211                 dst_release(dst_orig);
3212
3213         return dst;
3214 }
3215 EXPORT_SYMBOL(xfrm_lookup_route);
3216
3217 static inline int
3218 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3219 {
3220         struct sec_path *sp = skb_sec_path(skb);
3221         struct xfrm_state *x;
3222
3223         if (!sp || idx < 0 || idx >= sp->len)
3224                 return 0;
3225         x = sp->xvec[idx];
3226         if (!x->type->reject)
3227                 return 0;
3228         return x->type->reject(x, skb, fl);
3229 }
3230
3231 /* When skb is transformed back to its "native" form, we have to
3232  * check policy restrictions. At the moment we make this in maximally
3233  * stupid way. Shame on me. :-) Of course, connected sockets must
3234  * have policy cached at them.
3235  */
3236
3237 static inline int
3238 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3239               unsigned short family, u32 if_id)
3240 {
3241         if (xfrm_state_kern(x))
3242                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3243         return  x->id.proto == tmpl->id.proto &&
3244                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3245                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3246                 x->props.mode == tmpl->mode &&
3247                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3248                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3249                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3250                   xfrm_state_addr_cmp(tmpl, x, family)) &&
3251                 (if_id == 0 || if_id == x->if_id);
3252 }
3253
3254 /*
3255  * 0 or more than 0 is returned when validation is succeeded (either bypass
3256  * because of optional transport mode, or next index of the matched secpath
3257  * state with the template.
3258  * -1 is returned when no matching template is found.
3259  * Otherwise "-2 - errored_index" is returned.
3260  */
3261 static inline int
3262 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3263                unsigned short family, u32 if_id)
3264 {
3265         int idx = start;
3266
3267         if (tmpl->optional) {
3268                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3269                         return start;
3270         } else
3271                 start = -1;
3272         for (; idx < sp->len; idx++) {
3273                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3274                         return ++idx;
3275                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3276                         if (idx < sp->verified_cnt) {
3277                                 /* Secpath entry previously verified, consider optional and
3278                                  * continue searching
3279                                  */
3280                                 continue;
3281                         }
3282
3283                         if (start == -1)
3284                                 start = -2-idx;
3285                         break;
3286                 }
3287         }
3288         return start;
3289 }
3290
3291 static void
3292 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3293 {
3294         const struct iphdr *iph = ip_hdr(skb);
3295         int ihl = iph->ihl;
3296         u8 *xprth = skb_network_header(skb) + ihl * 4;
3297         struct flowi4 *fl4 = &fl->u.ip4;
3298         int oif = 0;
3299
3300         if (skb_dst(skb) && skb_dst(skb)->dev)
3301                 oif = skb_dst(skb)->dev->ifindex;
3302
3303         memset(fl4, 0, sizeof(struct flowi4));
3304         fl4->flowi4_mark = skb->mark;
3305         fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3306
3307         fl4->flowi4_proto = iph->protocol;
3308         fl4->daddr = reverse ? iph->saddr : iph->daddr;
3309         fl4->saddr = reverse ? iph->daddr : iph->saddr;
3310         fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK;
3311
3312         if (!ip_is_fragment(iph)) {
3313                 switch (iph->protocol) {
3314                 case IPPROTO_UDP:
3315                 case IPPROTO_UDPLITE:
3316                 case IPPROTO_TCP:
3317                 case IPPROTO_SCTP:
3318                 case IPPROTO_DCCP:
3319                         if (xprth + 4 < skb->data ||
3320                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3321                                 __be16 *ports;
3322
3323                                 xprth = skb_network_header(skb) + ihl * 4;
3324                                 ports = (__be16 *)xprth;
3325
3326                                 fl4->fl4_sport = ports[!!reverse];
3327                                 fl4->fl4_dport = ports[!reverse];
3328                         }
3329                         break;
3330                 case IPPROTO_ICMP:
3331                         if (xprth + 2 < skb->data ||
3332                             pskb_may_pull(skb, xprth + 2 - skb->data)) {
3333                                 u8 *icmp;
3334
3335                                 xprth = skb_network_header(skb) + ihl * 4;
3336                                 icmp = xprth;
3337
3338                                 fl4->fl4_icmp_type = icmp[0];
3339                                 fl4->fl4_icmp_code = icmp[1];
3340                         }
3341                         break;
3342                 case IPPROTO_GRE:
3343                         if (xprth + 12 < skb->data ||
3344                             pskb_may_pull(skb, xprth + 12 - skb->data)) {
3345                                 __be16 *greflags;
3346                                 __be32 *gre_hdr;
3347
3348                                 xprth = skb_network_header(skb) + ihl * 4;
3349                                 greflags = (__be16 *)xprth;
3350                                 gre_hdr = (__be32 *)xprth;
3351
3352                                 if (greflags[0] & GRE_KEY) {
3353                                         if (greflags[0] & GRE_CSUM)
3354                                                 gre_hdr++;
3355                                         fl4->fl4_gre_key = gre_hdr[1];
3356                                 }
3357                         }
3358                         break;
3359                 default:
3360                         break;
3361                 }
3362         }
3363 }
3364
3365 #if IS_ENABLED(CONFIG_IPV6)
3366 static void
3367 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3368 {
3369         struct flowi6 *fl6 = &fl->u.ip6;
3370         int onlyproto = 0;
3371         const struct ipv6hdr *hdr = ipv6_hdr(skb);
3372         u32 offset = sizeof(*hdr);
3373         struct ipv6_opt_hdr *exthdr;
3374         const unsigned char *nh = skb_network_header(skb);
3375         u16 nhoff = IP6CB(skb)->nhoff;
3376         int oif = 0;
3377         u8 nexthdr;
3378
3379         if (!nhoff)
3380                 nhoff = offsetof(struct ipv6hdr, nexthdr);
3381
3382         nexthdr = nh[nhoff];
3383
3384         if (skb_dst(skb) && skb_dst(skb)->dev)
3385                 oif = skb_dst(skb)->dev->ifindex;
3386
3387         memset(fl6, 0, sizeof(struct flowi6));
3388         fl6->flowi6_mark = skb->mark;
3389         fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3390
3391         fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3392         fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3393
3394         while (nh + offset + sizeof(*exthdr) < skb->data ||
3395                pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3396                 nh = skb_network_header(skb);
3397                 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3398
3399                 switch (nexthdr) {
3400                 case NEXTHDR_FRAGMENT:
3401                         onlyproto = 1;
3402                         fallthrough;
3403                 case NEXTHDR_ROUTING:
3404                 case NEXTHDR_HOP:
3405                 case NEXTHDR_DEST:
3406                         offset += ipv6_optlen(exthdr);
3407                         nexthdr = exthdr->nexthdr;
3408                         exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3409                         break;
3410                 case IPPROTO_UDP:
3411                 case IPPROTO_UDPLITE:
3412                 case IPPROTO_TCP:
3413                 case IPPROTO_SCTP:
3414                 case IPPROTO_DCCP:
3415                         if (!onlyproto && (nh + offset + 4 < skb->data ||
3416                              pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3417                                 __be16 *ports;
3418
3419                                 nh = skb_network_header(skb);
3420                                 ports = (__be16 *)(nh + offset);
3421                                 fl6->fl6_sport = ports[!!reverse];
3422                                 fl6->fl6_dport = ports[!reverse];
3423                         }
3424                         fl6->flowi6_proto = nexthdr;
3425                         return;
3426                 case IPPROTO_ICMPV6:
3427                         if (!onlyproto && (nh + offset + 2 < skb->data ||
3428                             pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3429                                 u8 *icmp;
3430
3431                                 nh = skb_network_header(skb);
3432                                 icmp = (u8 *)(nh + offset);
3433                                 fl6->fl6_icmp_type = icmp[0];
3434                                 fl6->fl6_icmp_code = icmp[1];
3435                         }
3436                         fl6->flowi6_proto = nexthdr;
3437                         return;
3438                 case IPPROTO_GRE:
3439                         if (!onlyproto &&
3440                             (nh + offset + 12 < skb->data ||
3441                              pskb_may_pull(skb, nh + offset + 12 - skb->data))) {
3442                                 struct gre_base_hdr *gre_hdr;
3443                                 __be32 *gre_key;
3444
3445                                 nh = skb_network_header(skb);
3446                                 gre_hdr = (struct gre_base_hdr *)(nh + offset);
3447                                 gre_key = (__be32 *)(gre_hdr + 1);
3448
3449                                 if (gre_hdr->flags & GRE_KEY) {
3450                                         if (gre_hdr->flags & GRE_CSUM)
3451                                                 gre_key++;
3452                                         fl6->fl6_gre_key = *gre_key;
3453                                 }
3454                         }
3455                         fl6->flowi6_proto = nexthdr;
3456                         return;
3457
3458 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3459                 case IPPROTO_MH:
3460                         offset += ipv6_optlen(exthdr);
3461                         if (!onlyproto && (nh + offset + 3 < skb->data ||
3462                             pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3463                                 struct ip6_mh *mh;
3464
3465                                 nh = skb_network_header(skb);
3466                                 mh = (struct ip6_mh *)(nh + offset);
3467                                 fl6->fl6_mh_type = mh->ip6mh_type;
3468                         }
3469                         fl6->flowi6_proto = nexthdr;
3470                         return;
3471 #endif
3472                 default:
3473                         fl6->flowi6_proto = nexthdr;
3474                         return;
3475                 }
3476         }
3477 }
3478 #endif
3479
3480 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3481                           unsigned int family, int reverse)
3482 {
3483         switch (family) {
3484         case AF_INET:
3485                 decode_session4(skb, fl, reverse);
3486                 break;
3487 #if IS_ENABLED(CONFIG_IPV6)
3488         case AF_INET6:
3489                 decode_session6(skb, fl, reverse);
3490                 break;
3491 #endif
3492         default:
3493                 return -EAFNOSUPPORT;
3494         }
3495
3496         return security_xfrm_decode_session(skb, &fl->flowi_secid);
3497 }
3498 EXPORT_SYMBOL(__xfrm_decode_session);
3499
3500 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3501 {
3502         for (; k < sp->len; k++) {
3503                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3504                         *idxp = k;
3505                         return 1;
3506                 }
3507         }
3508
3509         return 0;
3510 }
3511
3512 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3513                         unsigned short family)
3514 {
3515         struct net *net = dev_net(skb->dev);
3516         struct xfrm_policy *pol;
3517         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3518         int npols = 0;
3519         int xfrm_nr;
3520         int pi;
3521         int reverse;
3522         struct flowi fl;
3523         int xerr_idx = -1;
3524         const struct xfrm_if_cb *ifcb;
3525         struct sec_path *sp;
3526         struct xfrm_if *xi;
3527         u32 if_id = 0;
3528
3529         rcu_read_lock();
3530         ifcb = xfrm_if_get_cb();
3531
3532         if (ifcb) {
3533                 xi = ifcb->decode_session(skb, family);
3534                 if (xi) {
3535                         if_id = xi->p.if_id;
3536                         net = xi->net;
3537                 }
3538         }
3539         rcu_read_unlock();
3540
3541         reverse = dir & ~XFRM_POLICY_MASK;
3542         dir &= XFRM_POLICY_MASK;
3543
3544         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3545                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3546                 return 0;
3547         }
3548
3549         nf_nat_decode_session(skb, &fl, family);
3550
3551         /* First, check used SA against their selectors. */
3552         sp = skb_sec_path(skb);
3553         if (sp) {
3554                 int i;
3555
3556                 for (i = sp->len - 1; i >= 0; i--) {
3557                         struct xfrm_state *x = sp->xvec[i];
3558                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3559                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3560                                 return 0;
3561                         }
3562                 }
3563         }
3564
3565         pol = NULL;
3566         sk = sk_to_full_sk(sk);
3567         if (sk && sk->sk_policy[dir]) {
3568                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3569                 if (IS_ERR(pol)) {
3570                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3571                         return 0;
3572                 }
3573         }
3574
3575         if (!pol)
3576                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3577
3578         if (IS_ERR(pol)) {
3579                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3580                 return 0;
3581         }
3582
3583         if (!pol) {
3584                 if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3585                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3586                         return 0;
3587                 }
3588
3589                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3590                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3591                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3592                         return 0;
3593                 }
3594                 return 1;
3595         }
3596
3597         pol->curlft.use_time = ktime_get_real_seconds();
3598
3599         pols[0] = pol;
3600         npols++;
3601 #ifdef CONFIG_XFRM_SUB_POLICY
3602         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3603                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3604                                                     &fl, family,
3605                                                     XFRM_POLICY_IN, if_id);
3606                 if (pols[1]) {
3607                         if (IS_ERR(pols[1])) {
3608                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3609                                 xfrm_pol_put(pols[0]);
3610                                 return 0;
3611                         }
3612                         pols[1]->curlft.use_time = ktime_get_real_seconds();
3613                         npols++;
3614                 }
3615         }
3616 #endif
3617
3618         if (pol->action == XFRM_POLICY_ALLOW) {
3619                 static struct sec_path dummy;
3620                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3621                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3622                 struct xfrm_tmpl **tpp = tp;
3623                 int ti = 0;
3624                 int i, k;
3625
3626                 sp = skb_sec_path(skb);
3627                 if (!sp)
3628                         sp = &dummy;
3629
3630                 for (pi = 0; pi < npols; pi++) {
3631                         if (pols[pi] != pol &&
3632                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3633                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3634                                 goto reject;
3635                         }
3636                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3637                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3638                                 goto reject_error;
3639                         }
3640                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3641                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3642                 }
3643                 xfrm_nr = ti;
3644
3645                 if (npols > 1) {
3646                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3647                         tpp = stp;
3648                 }
3649
3650                 /* For each tunnel xfrm, find the first matching tmpl.
3651                  * For each tmpl before that, find corresponding xfrm.
3652                  * Order is _important_. Later we will implement
3653                  * some barriers, but at the moment barriers
3654                  * are implied between each two transformations.
3655                  * Upon success, marks secpath entries as having been
3656                  * verified to allow them to be skipped in future policy
3657                  * checks (e.g. nested tunnels).
3658                  */
3659                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3660                         k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3661                         if (k < 0) {
3662                                 if (k < -1)
3663                                         /* "-2 - errored_index" returned */
3664                                         xerr_idx = -(2+k);
3665                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3666                                 goto reject;
3667                         }
3668                 }
3669
3670                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3671                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3672                         goto reject;
3673                 }
3674
3675                 xfrm_pols_put(pols, npols);
3676                 sp->verified_cnt = k;
3677
3678                 return 1;
3679         }
3680         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3681
3682 reject:
3683         xfrm_secpath_reject(xerr_idx, skb, &fl);
3684 reject_error:
3685         xfrm_pols_put(pols, npols);
3686         return 0;
3687 }
3688 EXPORT_SYMBOL(__xfrm_policy_check);
3689
3690 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3691 {
3692         struct net *net = dev_net(skb->dev);
3693         struct flowi fl;
3694         struct dst_entry *dst;
3695         int res = 1;
3696
3697         if (xfrm_decode_session(skb, &fl, family) < 0) {
3698                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3699                 return 0;
3700         }
3701
3702         skb_dst_force(skb);
3703         if (!skb_dst(skb)) {
3704                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3705                 return 0;
3706         }
3707
3708         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3709         if (IS_ERR(dst)) {
3710                 res = 0;
3711                 dst = NULL;
3712         }
3713         skb_dst_set(skb, dst);
3714         return res;
3715 }
3716 EXPORT_SYMBOL(__xfrm_route_forward);
3717
3718 /* Optimize later using cookies and generation ids. */
3719
3720 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3721 {
3722         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3723          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3724          * get validated by dst_ops->check on every use.  We do this
3725          * because when a normal route referenced by an XFRM dst is
3726          * obsoleted we do not go looking around for all parent
3727          * referencing XFRM dsts so that we can invalidate them.  It
3728          * is just too much work.  Instead we make the checks here on
3729          * every use.  For example:
3730          *
3731          *      XFRM dst A --> IPv4 dst X
3732          *
3733          * X is the "xdst->route" of A (X is also the "dst->path" of A
3734          * in this example).  If X is marked obsolete, "A" will not
3735          * notice.  That's what we are validating here via the
3736          * stale_bundle() check.
3737          *
3738          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3739          * be marked on it.
3740          * This will force stale_bundle() to fail on any xdst bundle with
3741          * this dst linked in it.
3742          */
3743         if (dst->obsolete < 0 && !stale_bundle(dst))
3744                 return dst;
3745
3746         return NULL;
3747 }
3748
3749 static int stale_bundle(struct dst_entry *dst)
3750 {
3751         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3752 }
3753
3754 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3755 {
3756         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3757                 dst->dev = dev_net(dev)->loopback_dev;
3758                 dev_hold(dst->dev);
3759                 dev_put(dev);
3760         }
3761 }
3762 EXPORT_SYMBOL(xfrm_dst_ifdown);
3763
3764 static void xfrm_link_failure(struct sk_buff *skb)
3765 {
3766         /* Impossible. Such dst must be popped before reaches point of failure. */
3767 }
3768
3769 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3770 {
3771         if (dst) {
3772                 if (dst->obsolete) {
3773                         dst_release(dst);
3774                         dst = NULL;
3775                 }
3776         }
3777         return dst;
3778 }
3779
3780 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3781 {
3782         while (nr--) {
3783                 struct xfrm_dst *xdst = bundle[nr];
3784                 u32 pmtu, route_mtu_cached;
3785                 struct dst_entry *dst;
3786
3787                 dst = &xdst->u.dst;
3788                 pmtu = dst_mtu(xfrm_dst_child(dst));
3789                 xdst->child_mtu_cached = pmtu;
3790
3791                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3792
3793                 route_mtu_cached = dst_mtu(xdst->route);
3794                 xdst->route_mtu_cached = route_mtu_cached;
3795
3796                 if (pmtu > route_mtu_cached)
3797                         pmtu = route_mtu_cached;
3798
3799                 dst_metric_set(dst, RTAX_MTU, pmtu);
3800         }
3801 }
3802
3803 /* Check that the bundle accepts the flow and its components are
3804  * still valid.
3805  */
3806
3807 static int xfrm_bundle_ok(struct xfrm_dst *first)
3808 {
3809         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3810         struct dst_entry *dst = &first->u.dst;
3811         struct xfrm_dst *xdst;
3812         int start_from, nr;
3813         u32 mtu;
3814
3815         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3816             (dst->dev && !netif_running(dst->dev)))
3817                 return 0;
3818
3819         if (dst->flags & DST_XFRM_QUEUE)
3820                 return 1;
3821
3822         start_from = nr = 0;
3823         do {
3824                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3825
3826                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3827                         return 0;
3828                 if (xdst->xfrm_genid != dst->xfrm->genid)
3829                         return 0;
3830                 if (xdst->num_pols > 0 &&
3831                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3832                         return 0;
3833
3834                 bundle[nr++] = xdst;
3835
3836                 mtu = dst_mtu(xfrm_dst_child(dst));
3837                 if (xdst->child_mtu_cached != mtu) {
3838                         start_from = nr;
3839                         xdst->child_mtu_cached = mtu;
3840                 }
3841
3842                 if (!dst_check(xdst->route, xdst->route_cookie))
3843                         return 0;
3844                 mtu = dst_mtu(xdst->route);
3845                 if (xdst->route_mtu_cached != mtu) {
3846                         start_from = nr;
3847                         xdst->route_mtu_cached = mtu;
3848                 }
3849
3850                 dst = xfrm_dst_child(dst);
3851         } while (dst->xfrm);
3852
3853         if (likely(!start_from))
3854                 return 1;
3855
3856         xdst = bundle[start_from - 1];
3857         mtu = xdst->child_mtu_cached;
3858         while (start_from--) {
3859                 dst = &xdst->u.dst;
3860
3861                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3862                 if (mtu > xdst->route_mtu_cached)
3863                         mtu = xdst->route_mtu_cached;
3864                 dst_metric_set(dst, RTAX_MTU, mtu);
3865                 if (!start_from)
3866                         break;
3867
3868                 xdst = bundle[start_from - 1];
3869                 xdst->child_mtu_cached = mtu;
3870         }
3871
3872         return 1;
3873 }
3874
3875 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3876 {
3877         return dst_metric_advmss(xfrm_dst_path(dst));
3878 }
3879
3880 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3881 {
3882         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3883
3884         return mtu ? : dst_mtu(xfrm_dst_path(dst));
3885 }
3886
3887 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3888                                         const void *daddr)
3889 {
3890         while (dst->xfrm) {
3891                 const struct xfrm_state *xfrm = dst->xfrm;
3892
3893                 dst = xfrm_dst_child(dst);
3894
3895                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3896                         continue;
3897                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3898                         daddr = xfrm->coaddr;
3899                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3900                         daddr = &xfrm->id.daddr;
3901         }
3902         return daddr;
3903 }
3904
3905 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3906                                            struct sk_buff *skb,
3907                                            const void *daddr)
3908 {
3909         const struct dst_entry *path = xfrm_dst_path(dst);
3910
3911         if (!skb)
3912                 daddr = xfrm_get_dst_nexthop(dst, daddr);
3913         return path->ops->neigh_lookup(path, skb, daddr);
3914 }
3915
3916 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3917 {
3918         const struct dst_entry *path = xfrm_dst_path(dst);
3919
3920         daddr = xfrm_get_dst_nexthop(dst, daddr);
3921         path->ops->confirm_neigh(path, daddr);
3922 }
3923
3924 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3925 {
3926         int err = 0;
3927
3928         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3929                 return -EAFNOSUPPORT;
3930
3931         spin_lock(&xfrm_policy_afinfo_lock);
3932         if (unlikely(xfrm_policy_afinfo[family] != NULL))
3933                 err = -EEXIST;
3934         else {
3935                 struct dst_ops *dst_ops = afinfo->dst_ops;
3936                 if (likely(dst_ops->kmem_cachep == NULL))
3937                         dst_ops->kmem_cachep = xfrm_dst_cache;
3938                 if (likely(dst_ops->check == NULL))
3939                         dst_ops->check = xfrm_dst_check;
3940                 if (likely(dst_ops->default_advmss == NULL))
3941                         dst_ops->default_advmss = xfrm_default_advmss;
3942                 if (likely(dst_ops->mtu == NULL))
3943                         dst_ops->mtu = xfrm_mtu;
3944                 if (likely(dst_ops->negative_advice == NULL))
3945                         dst_ops->negative_advice = xfrm_negative_advice;
3946                 if (likely(dst_ops->link_failure == NULL))
3947                         dst_ops->link_failure = xfrm_link_failure;
3948                 if (likely(dst_ops->neigh_lookup == NULL))
3949                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
3950                 if (likely(!dst_ops->confirm_neigh))
3951                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
3952                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3953         }
3954         spin_unlock(&xfrm_policy_afinfo_lock);
3955
3956         return err;
3957 }
3958 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3959
3960 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3961 {
3962         struct dst_ops *dst_ops = afinfo->dst_ops;
3963         int i;
3964
3965         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3966                 if (xfrm_policy_afinfo[i] != afinfo)
3967                         continue;
3968                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3969                 break;
3970         }
3971
3972         synchronize_rcu();
3973
3974         dst_ops->kmem_cachep = NULL;
3975         dst_ops->check = NULL;
3976         dst_ops->negative_advice = NULL;
3977         dst_ops->link_failure = NULL;
3978 }
3979 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3980
3981 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3982 {
3983         spin_lock(&xfrm_if_cb_lock);
3984         rcu_assign_pointer(xfrm_if_cb, ifcb);
3985         spin_unlock(&xfrm_if_cb_lock);
3986 }
3987 EXPORT_SYMBOL(xfrm_if_register_cb);
3988
3989 void xfrm_if_unregister_cb(void)
3990 {
3991         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3992         synchronize_rcu();
3993 }
3994 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3995
3996 #ifdef CONFIG_XFRM_STATISTICS
3997 static int __net_init xfrm_statistics_init(struct net *net)
3998 {
3999         int rv;
4000         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4001         if (!net->mib.xfrm_statistics)
4002                 return -ENOMEM;
4003         rv = xfrm_proc_init(net);
4004         if (rv < 0)
4005                 free_percpu(net->mib.xfrm_statistics);
4006         return rv;
4007 }
4008
4009 static void xfrm_statistics_fini(struct net *net)
4010 {
4011         xfrm_proc_fini(net);
4012         free_percpu(net->mib.xfrm_statistics);
4013 }
4014 #else
4015 static int __net_init xfrm_statistics_init(struct net *net)
4016 {
4017         return 0;
4018 }
4019
4020 static void xfrm_statistics_fini(struct net *net)
4021 {
4022 }
4023 #endif
4024
4025 static int __net_init xfrm_policy_init(struct net *net)
4026 {
4027         unsigned int hmask, sz;
4028         int dir, err;
4029
4030         if (net_eq(net, &init_net)) {
4031                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4032                                            sizeof(struct xfrm_dst),
4033                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4034                                            NULL);
4035                 err = rhashtable_init(&xfrm_policy_inexact_table,
4036                                       &xfrm_pol_inexact_params);
4037                 BUG_ON(err);
4038         }
4039
4040         hmask = 8 - 1;
4041         sz = (hmask+1) * sizeof(struct hlist_head);
4042
4043         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4044         if (!net->xfrm.policy_byidx)
4045                 goto out_byidx;
4046         net->xfrm.policy_idx_hmask = hmask;
4047
4048         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4049                 struct xfrm_policy_hash *htab;
4050
4051                 net->xfrm.policy_count[dir] = 0;
4052                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4053                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4054
4055                 htab = &net->xfrm.policy_bydst[dir];
4056                 htab->table = xfrm_hash_alloc(sz);
4057                 if (!htab->table)
4058                         goto out_bydst;
4059                 htab->hmask = hmask;
4060                 htab->dbits4 = 32;
4061                 htab->sbits4 = 32;
4062                 htab->dbits6 = 128;
4063                 htab->sbits6 = 128;
4064         }
4065         net->xfrm.policy_hthresh.lbits4 = 32;
4066         net->xfrm.policy_hthresh.rbits4 = 32;
4067         net->xfrm.policy_hthresh.lbits6 = 128;
4068         net->xfrm.policy_hthresh.rbits6 = 128;
4069
4070         seqlock_init(&net->xfrm.policy_hthresh.lock);
4071
4072         INIT_LIST_HEAD(&net->xfrm.policy_all);
4073         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4074         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4075         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4076         return 0;
4077
4078 out_bydst:
4079         for (dir--; dir >= 0; dir--) {
4080                 struct xfrm_policy_hash *htab;
4081
4082                 htab = &net->xfrm.policy_bydst[dir];
4083                 xfrm_hash_free(htab->table, sz);
4084         }
4085         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4086 out_byidx:
4087         return -ENOMEM;
4088 }
4089
4090 static void xfrm_policy_fini(struct net *net)
4091 {
4092         struct xfrm_pol_inexact_bin *b, *t;
4093         unsigned int sz;
4094         int dir;
4095
4096         flush_work(&net->xfrm.policy_hash_work);
4097 #ifdef CONFIG_XFRM_SUB_POLICY
4098         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4099 #endif
4100         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4101
4102         WARN_ON(!list_empty(&net->xfrm.policy_all));
4103
4104         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4105                 struct xfrm_policy_hash *htab;
4106
4107                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4108
4109                 htab = &net->xfrm.policy_bydst[dir];
4110                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4111                 WARN_ON(!hlist_empty(htab->table));
4112                 xfrm_hash_free(htab->table, sz);
4113         }
4114
4115         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4116         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4117         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4118
4119         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4120         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4121                 __xfrm_policy_inexact_prune_bin(b, true);
4122         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4123 }
4124
4125 static int __net_init xfrm_net_init(struct net *net)
4126 {
4127         int rv;
4128
4129         /* Initialize the per-net locks here */
4130         spin_lock_init(&net->xfrm.xfrm_state_lock);
4131         spin_lock_init(&net->xfrm.xfrm_policy_lock);
4132         seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4133         mutex_init(&net->xfrm.xfrm_cfg_mutex);
4134         net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4135         net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4136         net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4137
4138         rv = xfrm_statistics_init(net);
4139         if (rv < 0)
4140                 goto out_statistics;
4141         rv = xfrm_state_init(net);
4142         if (rv < 0)
4143                 goto out_state;
4144         rv = xfrm_policy_init(net);
4145         if (rv < 0)
4146                 goto out_policy;
4147         rv = xfrm_sysctl_init(net);
4148         if (rv < 0)
4149                 goto out_sysctl;
4150
4151         return 0;
4152
4153 out_sysctl:
4154         xfrm_policy_fini(net);
4155 out_policy:
4156         xfrm_state_fini(net);
4157 out_state:
4158         xfrm_statistics_fini(net);
4159 out_statistics:
4160         return rv;
4161 }
4162
4163 static void __net_exit xfrm_net_exit(struct net *net)
4164 {
4165         xfrm_sysctl_fini(net);
4166         xfrm_policy_fini(net);
4167         xfrm_state_fini(net);
4168         xfrm_statistics_fini(net);
4169 }
4170
4171 static struct pernet_operations __net_initdata xfrm_net_ops = {
4172         .init = xfrm_net_init,
4173         .exit = xfrm_net_exit,
4174 };
4175
4176 void __init xfrm_init(void)
4177 {
4178         register_pernet_subsys(&xfrm_net_ops);
4179         xfrm_dev_init();
4180         xfrm_input_init();
4181
4182 #ifdef CONFIG_XFRM_ESPINTCP
4183         espintcp_init();
4184 #endif
4185 }
4186
4187 #ifdef CONFIG_AUDITSYSCALL
4188 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4189                                          struct audit_buffer *audit_buf)
4190 {
4191         struct xfrm_sec_ctx *ctx = xp->security;
4192         struct xfrm_selector *sel = &xp->selector;
4193
4194         if (ctx)
4195                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4196                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4197
4198         switch (sel->family) {
4199         case AF_INET:
4200                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4201                 if (sel->prefixlen_s != 32)
4202                         audit_log_format(audit_buf, " src_prefixlen=%d",
4203                                          sel->prefixlen_s);
4204                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4205                 if (sel->prefixlen_d != 32)
4206                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4207                                          sel->prefixlen_d);
4208                 break;
4209         case AF_INET6:
4210                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4211                 if (sel->prefixlen_s != 128)
4212                         audit_log_format(audit_buf, " src_prefixlen=%d",
4213                                          sel->prefixlen_s);
4214                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4215                 if (sel->prefixlen_d != 128)
4216                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4217                                          sel->prefixlen_d);
4218                 break;
4219         }
4220 }
4221
4222 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4223 {
4224         struct audit_buffer *audit_buf;
4225
4226         audit_buf = xfrm_audit_start("SPD-add");
4227         if (audit_buf == NULL)
4228                 return;
4229         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4230         audit_log_format(audit_buf, " res=%u", result);
4231         xfrm_audit_common_policyinfo(xp, audit_buf);
4232         audit_log_end(audit_buf);
4233 }
4234 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4235
4236 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4237                               bool task_valid)
4238 {
4239         struct audit_buffer *audit_buf;
4240
4241         audit_buf = xfrm_audit_start("SPD-delete");
4242         if (audit_buf == NULL)
4243                 return;
4244         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4245         audit_log_format(audit_buf, " res=%u", result);
4246         xfrm_audit_common_policyinfo(xp, audit_buf);
4247         audit_log_end(audit_buf);
4248 }
4249 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4250 #endif
4251
4252 #ifdef CONFIG_XFRM_MIGRATE
4253 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4254                                         const struct xfrm_selector *sel_tgt)
4255 {
4256         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4257                 if (sel_tgt->family == sel_cmp->family &&
4258                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4259                                     sel_cmp->family) &&
4260                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4261                                     sel_cmp->family) &&
4262                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4263                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4264                         return true;
4265                 }
4266         } else {
4267                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4268                         return true;
4269                 }
4270         }
4271         return false;
4272 }
4273
4274 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4275                                                     u8 dir, u8 type, struct net *net, u32 if_id)
4276 {
4277         struct xfrm_policy *pol, *ret = NULL;
4278         struct hlist_head *chain;
4279         u32 priority = ~0U;
4280
4281         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4282         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4283         hlist_for_each_entry(pol, chain, bydst) {
4284                 if ((if_id == 0 || pol->if_id == if_id) &&
4285                     xfrm_migrate_selector_match(sel, &pol->selector) &&
4286                     pol->type == type) {
4287                         ret = pol;
4288                         priority = ret->priority;
4289                         break;
4290                 }
4291         }
4292         chain = &net->xfrm.policy_inexact[dir];
4293         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4294                 if ((pol->priority >= priority) && ret)
4295                         break;
4296
4297                 if ((if_id == 0 || pol->if_id == if_id) &&
4298                     xfrm_migrate_selector_match(sel, &pol->selector) &&
4299                     pol->type == type) {
4300                         ret = pol;
4301                         break;
4302                 }
4303         }
4304
4305         xfrm_pol_hold(ret);
4306
4307         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4308
4309         return ret;
4310 }
4311
4312 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4313 {
4314         int match = 0;
4315
4316         if (t->mode == m->mode && t->id.proto == m->proto &&
4317             (m->reqid == 0 || t->reqid == m->reqid)) {
4318                 switch (t->mode) {
4319                 case XFRM_MODE_TUNNEL:
4320                 case XFRM_MODE_BEET:
4321                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4322                                             m->old_family) &&
4323                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4324                                             m->old_family)) {
4325                                 match = 1;
4326                         }
4327                         break;
4328                 case XFRM_MODE_TRANSPORT:
4329                         /* in case of transport mode, template does not store
4330                            any IP addresses, hence we just compare mode and
4331                            protocol */
4332                         match = 1;
4333                         break;
4334                 default:
4335                         break;
4336                 }
4337         }
4338         return match;
4339 }
4340
4341 /* update endpoint address(es) of template(s) */
4342 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4343                                struct xfrm_migrate *m, int num_migrate)
4344 {
4345         struct xfrm_migrate *mp;
4346         int i, j, n = 0;
4347
4348         write_lock_bh(&pol->lock);
4349         if (unlikely(pol->walk.dead)) {
4350                 /* target policy has been deleted */
4351                 write_unlock_bh(&pol->lock);
4352                 return -ENOENT;
4353         }
4354
4355         for (i = 0; i < pol->xfrm_nr; i++) {
4356                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4357                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4358                                 continue;
4359                         n++;
4360                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4361                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4362                                 continue;
4363                         /* update endpoints */
4364                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4365                                sizeof(pol->xfrm_vec[i].id.daddr));
4366                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4367                                sizeof(pol->xfrm_vec[i].saddr));
4368                         pol->xfrm_vec[i].encap_family = mp->new_family;
4369                         /* flush bundles */
4370                         atomic_inc(&pol->genid);
4371                 }
4372         }
4373
4374         write_unlock_bh(&pol->lock);
4375
4376         if (!n)
4377                 return -ENODATA;
4378
4379         return 0;
4380 }
4381
4382 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4383 {
4384         int i, j;
4385
4386         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4387                 return -EINVAL;
4388
4389         for (i = 0; i < num_migrate; i++) {
4390                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4391                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4392                         return -EINVAL;
4393
4394                 /* check if there is any duplicated entry */
4395                 for (j = i + 1; j < num_migrate; j++) {
4396                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4397                                     sizeof(m[i].old_daddr)) &&
4398                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4399                                     sizeof(m[i].old_saddr)) &&
4400                             m[i].proto == m[j].proto &&
4401                             m[i].mode == m[j].mode &&
4402                             m[i].reqid == m[j].reqid &&
4403                             m[i].old_family == m[j].old_family)
4404                                 return -EINVAL;
4405                 }
4406         }
4407
4408         return 0;
4409 }
4410
4411 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4412                  struct xfrm_migrate *m, int num_migrate,
4413                  struct xfrm_kmaddress *k, struct net *net,
4414                  struct xfrm_encap_tmpl *encap, u32 if_id)
4415 {
4416         int i, err, nx_cur = 0, nx_new = 0;
4417         struct xfrm_policy *pol = NULL;
4418         struct xfrm_state *x, *xc;
4419         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4420         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4421         struct xfrm_migrate *mp;
4422
4423         /* Stage 0 - sanity checks */
4424         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4425                 goto out;
4426
4427         if (dir >= XFRM_POLICY_MAX) {
4428                 err = -EINVAL;
4429                 goto out;
4430         }
4431
4432         /* Stage 1 - find policy */
4433         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id)) == NULL) {
4434                 err = -ENOENT;
4435                 goto out;
4436         }
4437
4438         /* Stage 2 - find and update state(s) */
4439         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4440                 if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4441                         x_cur[nx_cur] = x;
4442                         nx_cur++;
4443                         xc = xfrm_state_migrate(x, mp, encap);
4444                         if (xc) {
4445                                 x_new[nx_new] = xc;
4446                                 nx_new++;
4447                         } else {
4448                                 err = -ENODATA;
4449                                 goto restore_state;
4450                         }
4451                 }
4452         }
4453
4454         /* Stage 3 - update policy */
4455         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4456                 goto restore_state;
4457
4458         /* Stage 4 - delete old state(s) */
4459         if (nx_cur) {
4460                 xfrm_states_put(x_cur, nx_cur);
4461                 xfrm_states_delete(x_cur, nx_cur);
4462         }
4463
4464         /* Stage 5 - announce */
4465         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4466
4467         xfrm_pol_put(pol);
4468
4469         return 0;
4470 out:
4471         return err;
4472
4473 restore_state:
4474         if (pol)
4475                 xfrm_pol_put(pol);
4476         if (nx_cur)
4477                 xfrm_states_put(x_cur, nx_cur);
4478         if (nx_new)
4479                 xfrm_states_delete(x_new, nx_new);
4480
4481         return err;
4482 }
4483 EXPORT_SYMBOL(xfrm_migrate);
4484 #endif