net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/flow.h>
  30 #include <net/xfrm.h>
  31 #include <net/ip.h>
  32 #ifdef CONFIG_XFRM_STATISTICS
  33 #include <net/snmp.h>
  34 #endif
  35
  36 #include "xfrm_hash.h"
  37
  38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  40 #define XFRM_MAX_QUEUE_LEN      100
  41
  42 struct xfrm_flo {
  43         struct dst_entry *dst_orig;
  44         u8 flags;
  45 };
  46
  47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  48 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
  49                                                 __read_mostly;
  50
  51 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  52 static __read_mostly seqcount_t xfrm_policy_hash_generation;
  53
  54 static void xfrm_init_pmtu(struct dst_entry *dst);
  55 static int stale_bundle(struct dst_entry *dst);
  56 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  57 static void xfrm_policy_queue_process(unsigned long arg);
  58
  59 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
  60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  61                                                 int dir);
  62
  63 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
  64 {
  65         return atomic_inc_not_zero(&policy->refcnt);
  66 }
  67
  68 static inline bool
  69 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  70 {
  71         const struct flowi4 *fl4 = &fl->u.ip4;
  72
  73         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  74                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  75                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  76                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  77                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  78                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  79 }
  80
  81 static inline bool
  82 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  83 {
  84         const struct flowi6 *fl6 = &fl->u.ip6;
  85
  86         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  87                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  88                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  89                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  90                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  91                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  92 }
  93
  94 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  95                          unsigned short family)
  96 {
  97         switch (family) {
  98         case AF_INET:
  99                 return __xfrm4_selector_match(sel, fl);
 100         case AF_INET6:
 101                 return __xfrm6_selector_match(sel, fl);
 102         }
 103         return false;
 104 }
 105
 106 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 107 {
 108         const struct xfrm_policy_afinfo *afinfo;
 109
 110         if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 111                 return NULL;
 112         rcu_read_lock();
 113         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 114         if (unlikely(!afinfo))
 115                 rcu_read_unlock();
 116         return afinfo;
 117 }
 118
 119 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
 120                                     const xfrm_address_t *saddr,
 121                                     const xfrm_address_t *daddr,
 122                                     int family)
 123 {
 124         const struct xfrm_policy_afinfo *afinfo;
 125         struct dst_entry *dst;
 126
 127         afinfo = xfrm_policy_get_afinfo(family);
 128         if (unlikely(afinfo == NULL))
 129                 return ERR_PTR(-EAFNOSUPPORT);
 130
 131         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr);
 132
 133         rcu_read_unlock();
 134
 135         return dst;
 136 }
 137 EXPORT_SYMBOL(__xfrm_dst_lookup);
 138
 139 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 140                                                 int tos, int oif,
 141                                                 xfrm_address_t *prev_saddr,
 142                                                 xfrm_address_t *prev_daddr,
 143                                                 int family)
 144 {
 145         struct net *net = xs_net(x);
 146         xfrm_address_t *saddr = &x->props.saddr;
 147         xfrm_address_t *daddr = &x->id.daddr;
 148         struct dst_entry *dst;
 149
 150         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 151                 saddr = x->coaddr;
 152                 daddr = prev_daddr;
 153         }
 154         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 155                 saddr = prev_saddr;
 156                 daddr = x->coaddr;
 157         }
 158
 159         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family);
 160
 161         if (!IS_ERR(dst)) {
 162                 if (prev_saddr != saddr)
 163                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 164                 if (prev_daddr != daddr)
 165                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 166         }
 167
 168         return dst;
 169 }
 170
 171 static inline unsigned long make_jiffies(long secs)
 172 {
 173         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 174                 return MAX_SCHEDULE_TIMEOUT-1;
 175         else
 176                 return secs*HZ;
 177 }
 178
 179 static void xfrm_policy_timer(unsigned long data)
 180 {
 181         struct xfrm_policy *xp = (struct xfrm_policy *)data;
 182         unsigned long now = get_seconds();
 183         long next = LONG_MAX;
 184         int warn = 0;
 185         int dir;
 186
 187         read_lock(&xp->lock);
 188
 189         if (unlikely(xp->walk.dead))
 190                 goto out;
 191
 192         dir = xfrm_policy_id2dir(xp->index);
 193
 194         if (xp->lft.hard_add_expires_seconds) {
 195                 long tmo = xp->lft.hard_add_expires_seconds +
 196                         xp->curlft.add_time - now;
 197                 if (tmo <= 0)
 198                         goto expired;
 199                 if (tmo < next)
 200                         next = tmo;
 201         }
 202         if (xp->lft.hard_use_expires_seconds) {
 203                 long tmo = xp->lft.hard_use_expires_seconds +
 204                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 205                 if (tmo <= 0)
 206                         goto expired;
 207                 if (tmo < next)
 208                         next = tmo;
 209         }
 210         if (xp->lft.soft_add_expires_seconds) {
 211                 long tmo = xp->lft.soft_add_expires_seconds +
 212                         xp->curlft.add_time - now;
 213                 if (tmo <= 0) {
 214                         warn = 1;
 215                         tmo = XFRM_KM_TIMEOUT;
 216                 }
 217                 if (tmo < next)
 218                         next = tmo;
 219         }
 220         if (xp->lft.soft_use_expires_seconds) {
 221                 long tmo = xp->lft.soft_use_expires_seconds +
 222                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 223                 if (tmo <= 0) {
 224                         warn = 1;
 225                         tmo = XFRM_KM_TIMEOUT;
 226                 }
 227                 if (tmo < next)
 228                         next = tmo;
 229         }
 230
 231         if (warn)
 232                 km_policy_expired(xp, dir, 0, 0);
 233         if (next != LONG_MAX &&
 234             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 235                 xfrm_pol_hold(xp);
 236
 237 out:
 238         read_unlock(&xp->lock);
 239         xfrm_pol_put(xp);
 240         return;
 241
 242 expired:
 243         read_unlock(&xp->lock);
 244         if (!xfrm_policy_delete(xp, dir))
 245                 km_policy_expired(xp, dir, 1, 0);
 246         xfrm_pol_put(xp);
 247 }
 248
 249 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 250 {
 251         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 252
 253         if (unlikely(pol->walk.dead))
 254                 flo = NULL;
 255         else
 256                 xfrm_pol_hold(pol);
 257
 258         return flo;
 259 }
 260
 261 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 262 {
 263         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 264
 265         return !pol->walk.dead;
 266 }
 267
 268 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 269 {
 270         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 271 }
 272
 273 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 274         .get = xfrm_policy_flo_get,
 275         .check = xfrm_policy_flo_check,
 276         .delete = xfrm_policy_flo_delete,
 277 };
 278
 279 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 280  * SPD calls.
 281  */
 282
 283 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 284 {
 285         struct xfrm_policy *policy;
 286
 287         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 288
 289         if (policy) {
 290                 write_pnet(&policy->xp_net, net);
 291                 INIT_LIST_HEAD(&policy->walk.all);
 292                 INIT_HLIST_NODE(&policy->bydst);
 293                 INIT_HLIST_NODE(&policy->byidx);
 294                 rwlock_init(&policy->lock);
 295                 atomic_set(&policy->refcnt, 1);
 296                 skb_queue_head_init(&policy->polq.hold_queue);
 297                 setup_timer(&policy->timer, xfrm_policy_timer,
 298                                 (unsigned long)policy);
 299                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 300                             (unsigned long)policy);
 301                 policy->flo.ops = &xfrm_policy_fc_ops;
 302         }
 303         return policy;
 304 }
 305 EXPORT_SYMBOL(xfrm_policy_alloc);
 306
 307 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 308 {
 309         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 310
 311         security_xfrm_policy_free(policy->security);
 312         kfree(policy);
 313 }
 314
 315 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 316
 317 void xfrm_policy_destroy(struct xfrm_policy *policy)
 318 {
 319         BUG_ON(!policy->walk.dead);
 320
 321         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 322                 BUG();
 323
 324         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 325 }
 326 EXPORT_SYMBOL(xfrm_policy_destroy);
 327
 328 /* Rule must be locked. Release descendant resources, announce
 329  * entry dead. The rule must be unlinked from lists to the moment.
 330  */
 331
 332 static void xfrm_policy_kill(struct xfrm_policy *policy)
 333 {
 334         policy->walk.dead = 1;
 335
 336         atomic_inc(&policy->genid);
 337
 338         if (del_timer(&policy->polq.hold_timer))
 339                 xfrm_pol_put(policy);
 340         skb_queue_purge(&policy->polq.hold_queue);
 341
 342         if (del_timer(&policy->timer))
 343                 xfrm_pol_put(policy);
 344
 345         xfrm_pol_put(policy);
 346 }
 347
 348 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 349
 350 static inline unsigned int idx_hash(struct net *net, u32 index)
 351 {
 352         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 353 }
 354
 355 /* calculate policy hash thresholds */
 356 static void __get_hash_thresh(struct net *net,
 357                               unsigned short family, int dir,
 358                               u8 *dbits, u8 *sbits)
 359 {
 360         switch (family) {
 361         case AF_INET:
 362                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 363                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 364                 break;
 365
 366         case AF_INET6:
 367                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 368                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 369                 break;
 370
 371         default:
 372                 *dbits = 0;
 373                 *sbits = 0;
 374         }
 375 }
 376
 377 static struct hlist_head *policy_hash_bysel(struct net *net,
 378                                             const struct xfrm_selector *sel,
 379                                             unsigned short family, int dir)
 380 {
 381         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 382         unsigned int hash;
 383         u8 dbits;
 384         u8 sbits;
 385
 386         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 387         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 388
 389         if (hash == hmask + 1)
 390                 return &net->xfrm.policy_inexact[dir];
 391
 392         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 393                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 394 }
 395
 396 static struct hlist_head *policy_hash_direct(struct net *net,
 397                                              const xfrm_address_t *daddr,
 398                                              const xfrm_address_t *saddr,
 399                                              unsigned short family, int dir)
 400 {
 401         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 402         unsigned int hash;
 403         u8 dbits;
 404         u8 sbits;
 405
 406         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 407         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 408
 409         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 410                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 411 }
 412
 413 static void xfrm_dst_hash_transfer(struct net *net,
 414                                    struct hlist_head *list,
 415                                    struct hlist_head *ndsttable,
 416                                    unsigned int nhashmask,
 417                                    int dir)
 418 {
 419         struct hlist_node *tmp, *entry0 = NULL;
 420         struct xfrm_policy *pol;
 421         unsigned int h0 = 0;
 422         u8 dbits;
 423         u8 sbits;
 424
 425 redo:
 426         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 427                 unsigned int h;
 428
 429                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 430                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 431                                 pol->family, nhashmask, dbits, sbits);
 432                 if (!entry0) {
 433                         hlist_del_rcu(&pol->bydst);
 434                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 435                         h0 = h;
 436                 } else {
 437                         if (h != h0)
 438                                 continue;
 439                         hlist_del_rcu(&pol->bydst);
 440                         hlist_add_behind_rcu(&pol->bydst, entry0);
 441                 }
 442                 entry0 = &pol->bydst;
 443         }
 444         if (!hlist_empty(list)) {
 445                 entry0 = NULL;
 446                 goto redo;
 447         }
 448 }
 449
 450 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 451                                    struct hlist_head *nidxtable,
 452                                    unsigned int nhashmask)
 453 {
 454         struct hlist_node *tmp;
 455         struct xfrm_policy *pol;
 456
 457         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 458                 unsigned int h;
 459
 460                 h = __idx_hash(pol->index, nhashmask);
 461                 hlist_add_head(&pol->byidx, nidxtable+h);
 462         }
 463 }
 464
 465 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 466 {
 467         return ((old_hmask + 1) << 1) - 1;
 468 }
 469
 470 static void xfrm_bydst_resize(struct net *net, int dir)
 471 {
 472         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 473         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 474         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 475         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 476         struct hlist_head *odst;
 477         int i;
 478
 479         if (!ndst)
 480                 return;
 481
 482         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 483         write_seqcount_begin(&xfrm_policy_hash_generation);
 484
 485         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 486                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 487
 488         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 489                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 490
 491         for (i = hmask; i >= 0; i--)
 492                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 493
 494         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 495         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 496
 497         write_seqcount_end(&xfrm_policy_hash_generation);
 498         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 499
 500         synchronize_rcu();
 501
 502         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 503 }
 504
 505 static void xfrm_byidx_resize(struct net *net, int total)
 506 {
 507         unsigned int hmask = net->xfrm.policy_idx_hmask;
 508         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 509         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 510         struct hlist_head *oidx = net->xfrm.policy_byidx;
 511         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 512         int i;
 513
 514         if (!nidx)
 515                 return;
 516
 517         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 518
 519         for (i = hmask; i >= 0; i--)
 520                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 521
 522         net->xfrm.policy_byidx = nidx;
 523         net->xfrm.policy_idx_hmask = nhashmask;
 524
 525         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 526
 527         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 528 }
 529
 530 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 531 {
 532         unsigned int cnt = net->xfrm.policy_count[dir];
 533         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 534
 535         if (total)
 536                 *total += cnt;
 537
 538         if ((hmask + 1) < xfrm_policy_hashmax &&
 539             cnt > hmask)
 540                 return 1;
 541
 542         return 0;
 543 }
 544
 545 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 546 {
 547         unsigned int hmask = net->xfrm.policy_idx_hmask;
 548
 549         if ((hmask + 1) < xfrm_policy_hashmax &&
 550             total > hmask)
 551                 return 1;
 552
 553         return 0;
 554 }
 555
 556 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 557 {
 558         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 559         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 560         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 561         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 562         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 563         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 564         si->spdhcnt = net->xfrm.policy_idx_hmask;
 565         si->spdhmcnt = xfrm_policy_hashmax;
 566 }
 567 EXPORT_SYMBOL(xfrm_spd_getinfo);
 568
 569 static DEFINE_MUTEX(hash_resize_mutex);
 570 static void xfrm_hash_resize(struct work_struct *work)
 571 {
 572         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 573         int dir, total;
 574
 575         mutex_lock(&hash_resize_mutex);
 576
 577         total = 0;
 578         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 579                 if (xfrm_bydst_should_resize(net, dir, &total))
 580                         xfrm_bydst_resize(net, dir);
 581         }
 582         if (xfrm_byidx_should_resize(net, total))
 583                 xfrm_byidx_resize(net, total);
 584
 585         mutex_unlock(&hash_resize_mutex);
 586 }
 587
 588 static void xfrm_hash_rebuild(struct work_struct *work)
 589 {
 590         struct net *net = container_of(work, struct net,
 591                                        xfrm.policy_hthresh.work);
 592         unsigned int hmask;
 593         struct xfrm_policy *pol;
 594         struct xfrm_policy *policy;
 595         struct hlist_head *chain;
 596         struct hlist_head *odst;
 597         struct hlist_node *newpos;
 598         int i;
 599         int dir;
 600         unsigned seq;
 601         u8 lbits4, rbits4, lbits6, rbits6;
 602
 603         mutex_lock(&hash_resize_mutex);
 604
 605         /* read selector prefixlen thresholds */
 606         do {
 607                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
 608
 609                 lbits4 = net->xfrm.policy_hthresh.lbits4;
 610                 rbits4 = net->xfrm.policy_hthresh.rbits4;
 611                 lbits6 = net->xfrm.policy_hthresh.lbits6;
 612                 rbits6 = net->xfrm.policy_hthresh.rbits6;
 613         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
 614
 615         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 616
 617         /* reset the bydst and inexact table in all directions */
 618         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 619                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
 620                 hmask = net->xfrm.policy_bydst[dir].hmask;
 621                 odst = net->xfrm.policy_bydst[dir].table;
 622                 for (i = hmask; i >= 0; i--)
 623                         INIT_HLIST_HEAD(odst + i);
 624                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
 625                         /* dir out => dst = remote, src = local */
 626                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
 627                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
 628                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
 629                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
 630                 } else {
 631                         /* dir in/fwd => dst = local, src = remote */
 632                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
 633                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
 634                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
 635                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
 636                 }
 637         }
 638
 639         /* re-insert all policies by order of creation */
 640         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 641                 if (xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
 642                         /* skip socket policies */
 643                         continue;
 644                 }
 645                 newpos = NULL;
 646                 chain = policy_hash_bysel(net, &policy->selector,
 647                                           policy->family,
 648                                           xfrm_policy_id2dir(policy->index));
 649                 hlist_for_each_entry(pol, chain, bydst) {
 650                         if (policy->priority >= pol->priority)
 651                                 newpos = &pol->bydst;
 652                         else
 653                                 break;
 654                 }
 655                 if (newpos)
 656                         hlist_add_behind(&policy->bydst, newpos);
 657                 else
 658                         hlist_add_head(&policy->bydst, chain);
 659         }
 660
 661         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 662
 663         mutex_unlock(&hash_resize_mutex);
 664 }
 665
 666 void xfrm_policy_hash_rebuild(struct net *net)
 667 {
 668         schedule_work(&net->xfrm.policy_hthresh.work);
 669 }
 670 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 671
 672 /* Generate new index... KAME seems to generate them ordered by cost
 673  * of an absolute inpredictability of ordering of rules. This will not pass. */
 674 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 675 {
 676         static u32 idx_generator;
 677
 678         for (;;) {
 679                 struct hlist_head *list;
 680                 struct xfrm_policy *p;
 681                 u32 idx;
 682                 int found;
 683
 684                 if (!index) {
 685                         idx = (idx_generator | dir);
 686                         idx_generator += 8;
 687                 } else {
 688                         idx = index;
 689                         index = 0;
 690                 }
 691
 692                 if (idx == 0)
 693                         idx = 8;
 694                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 695                 found = 0;
 696                 hlist_for_each_entry(p, list, byidx) {
 697                         if (p->index == idx) {
 698                                 found = 1;
 699                                 break;
 700                         }
 701                 }
 702                 if (!found)
 703                         return idx;
 704         }
 705 }
 706
 707 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 708 {
 709         u32 *p1 = (u32 *) s1;
 710         u32 *p2 = (u32 *) s2;
 711         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 712         int i;
 713
 714         for (i = 0; i < len; i++) {
 715                 if (p1[i] != p2[i])
 716                         return 1;
 717         }
 718
 719         return 0;
 720 }
 721
 722 static void xfrm_policy_requeue(struct xfrm_policy *old,
 723                                 struct xfrm_policy *new)
 724 {
 725         struct xfrm_policy_queue *pq = &old->polq;
 726         struct sk_buff_head list;
 727
 728         if (skb_queue_empty(&pq->hold_queue))
 729                 return;
 730
 731         __skb_queue_head_init(&list);
 732
 733         spin_lock_bh(&pq->hold_queue.lock);
 734         skb_queue_splice_init(&pq->hold_queue, &list);
 735         if (del_timer(&pq->hold_timer))
 736                 xfrm_pol_put(old);
 737         spin_unlock_bh(&pq->hold_queue.lock);
 738
 739         pq = &new->polq;
 740
 741         spin_lock_bh(&pq->hold_queue.lock);
 742         skb_queue_splice(&list, &pq->hold_queue);
 743         pq->timeout = XFRM_QUEUE_TMO_MIN;
 744         if (!mod_timer(&pq->hold_timer, jiffies))
 745                 xfrm_pol_hold(new);
 746         spin_unlock_bh(&pq->hold_queue.lock);
 747 }
 748
 749 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 750                                    struct xfrm_policy *pol)
 751 {
 752         u32 mark = policy->mark.v & policy->mark.m;
 753
 754         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 755                 return true;
 756
 757         if ((mark & pol->mark.m) == pol->mark.v &&
 758             policy->priority == pol->priority)
 759                 return true;
 760
 761         return false;
 762 }
 763
 764 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 765 {
 766         struct net *net = xp_net(policy);
 767         struct xfrm_policy *pol;
 768         struct xfrm_policy *delpol;
 769         struct hlist_head *chain;
 770         struct hlist_node *newpos;
 771
 772         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 773         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 774         delpol = NULL;
 775         newpos = NULL;
 776         hlist_for_each_entry(pol, chain, bydst) {
 777                 if (pol->type == policy->type &&
 778                     !selector_cmp(&pol->selector, &policy->selector) &&
 779                     xfrm_policy_mark_match(policy, pol) &&
 780                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 781                     !WARN_ON(delpol)) {
 782                         if (excl) {
 783                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 784                                 return -EEXIST;
 785                         }
 786                         delpol = pol;
 787                         if (policy->priority > pol->priority)
 788                                 continue;
 789                 } else if (policy->priority >= pol->priority) {
 790                         newpos = &pol->bydst;
 791                         continue;
 792                 }
 793                 if (delpol)
 794                         break;
 795         }
 796         if (newpos)
 797                 hlist_add_behind(&policy->bydst, newpos);
 798         else
 799                 hlist_add_head(&policy->bydst, chain);
 800         __xfrm_policy_link(policy, dir);
 801         atomic_inc(&net->xfrm.flow_cache_genid);
 802
 803         /* After previous checking, family can either be AF_INET or AF_INET6 */
 804         if (policy->family == AF_INET)
 805                 rt_genid_bump_ipv4(net);
 806         else
 807                 rt_genid_bump_ipv6(net);
 808
 809         if (delpol) {
 810                 xfrm_policy_requeue(delpol, policy);
 811                 __xfrm_policy_unlink(delpol, dir);
 812         }
 813         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 814         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 815         policy->curlft.add_time = get_seconds();
 816         policy->curlft.use_time = 0;
 817         if (!mod_timer(&policy->timer, jiffies + HZ))
 818                 xfrm_pol_hold(policy);
 819         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 820
 821         if (delpol)
 822                 xfrm_policy_kill(delpol);
 823         else if (xfrm_bydst_should_resize(net, dir, NULL))
 824                 schedule_work(&net->xfrm.policy_hash_work);
 825
 826         return 0;
 827 }
 828 EXPORT_SYMBOL(xfrm_policy_insert);
 829
 830 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 831                                           int dir, struct xfrm_selector *sel,
 832                                           struct xfrm_sec_ctx *ctx, int delete,
 833                                           int *err)
 834 {
 835         struct xfrm_policy *pol, *ret;
 836         struct hlist_head *chain;
 837
 838         *err = 0;
 839         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 840         chain = policy_hash_bysel(net, sel, sel->family, dir);
 841         ret = NULL;
 842         hlist_for_each_entry(pol, chain, bydst) {
 843                 if (pol->type == type &&
 844                     (mark & pol->mark.m) == pol->mark.v &&
 845                     !selector_cmp(sel, &pol->selector) &&
 846                     xfrm_sec_ctx_match(ctx, pol->security)) {
 847                         xfrm_pol_hold(pol);
 848                         if (delete) {
 849                                 *err = security_xfrm_policy_delete(
 850                                                                 pol->security);
 851                                 if (*err) {
 852                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 853                                         return pol;
 854                                 }
 855                                 __xfrm_policy_unlink(pol, dir);
 856                         }
 857                         ret = pol;
 858                         break;
 859                 }
 860         }
 861         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 862
 863         if (ret && delete)
 864                 xfrm_policy_kill(ret);
 865         return ret;
 866 }
 867 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 868
 869 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 870                                      int dir, u32 id, int delete, int *err)
 871 {
 872         struct xfrm_policy *pol, *ret;
 873         struct hlist_head *chain;
 874
 875         *err = -ENOENT;
 876         if (xfrm_policy_id2dir(id) != dir)
 877                 return NULL;
 878
 879         *err = 0;
 880         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 881         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 882         ret = NULL;
 883         hlist_for_each_entry(pol, chain, byidx) {
 884                 if (pol->type == type && pol->index == id &&
 885                     (mark & pol->mark.m) == pol->mark.v) {
 886                         xfrm_pol_hold(pol);
 887                         if (delete) {
 888                                 *err = security_xfrm_policy_delete(
 889                                                                 pol->security);
 890                                 if (*err) {
 891                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 892                                         return pol;
 893                                 }
 894                                 __xfrm_policy_unlink(pol, dir);
 895                         }
 896                         ret = pol;
 897                         break;
 898                 }
 899         }
 900         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 901
 902         if (ret && delete)
 903                 xfrm_policy_kill(ret);
 904         return ret;
 905 }
 906 EXPORT_SYMBOL(xfrm_policy_byid);
 907
 908 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 909 static inline int
 910 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 911 {
 912         int dir, err = 0;
 913
 914         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 915                 struct xfrm_policy *pol;
 916                 int i;
 917
 918                 hlist_for_each_entry(pol,
 919                                      &net->xfrm.policy_inexact[dir], bydst) {
 920                         if (pol->type != type)
 921                                 continue;
 922                         err = security_xfrm_policy_delete(pol->security);
 923                         if (err) {
 924                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 925                                 return err;
 926                         }
 927                 }
 928                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 929                         hlist_for_each_entry(pol,
 930                                              net->xfrm.policy_bydst[dir].table + i,
 931                                              bydst) {
 932                                 if (pol->type != type)
 933                                         continue;
 934                                 err = security_xfrm_policy_delete(
 935                                                                 pol->security);
 936                                 if (err) {
 937                                         xfrm_audit_policy_delete(pol, 0,
 938                                                                  task_valid);
 939                                         return err;
 940                                 }
 941                         }
 942                 }
 943         }
 944         return err;
 945 }
 946 #else
 947 static inline int
 948 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 949 {
 950         return 0;
 951 }
 952 #endif
 953
 954 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 955 {
 956         int dir, err = 0, cnt = 0;
 957
 958         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 959
 960         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 961         if (err)
 962                 goto out;
 963
 964         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 965                 struct xfrm_policy *pol;
 966                 int i;
 967
 968         again1:
 969                 hlist_for_each_entry(pol,
 970                                      &net->xfrm.policy_inexact[dir], bydst) {
 971                         if (pol->type != type)
 972                                 continue;
 973                         __xfrm_policy_unlink(pol, dir);
 974                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 975                         cnt++;
 976
 977                         xfrm_audit_policy_delete(pol, 1, task_valid);
 978
 979                         xfrm_policy_kill(pol);
 980
 981                         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 982                         goto again1;
 983                 }
 984
 985                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 986         again2:
 987                         hlist_for_each_entry(pol,
 988                                              net->xfrm.policy_bydst[dir].table + i,
 989                                              bydst) {
 990                                 if (pol->type != type)
 991                                         continue;
 992                                 __xfrm_policy_unlink(pol, dir);
 993                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 994                                 cnt++;
 995
 996                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 997                                 xfrm_policy_kill(pol);
 998
 999                                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1000                                 goto again2;
1001                         }
1002                 }
1003
1004         }
1005         if (!cnt)
1006                 err = -ESRCH;
1007 out:
1008         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1009
1010         if (cnt)
1011                 xfrm_garbage_collect(net);
1012
1013         return err;
1014 }
1015 EXPORT_SYMBOL(xfrm_policy_flush);
1016
1017 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1018                      int (*func)(struct xfrm_policy *, int, int, void*),
1019                      void *data)
1020 {
1021         struct xfrm_policy *pol;
1022         struct xfrm_policy_walk_entry *x;
1023         int error = 0;
1024
1025         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1026             walk->type != XFRM_POLICY_TYPE_ANY)
1027                 return -EINVAL;
1028
1029         if (list_empty(&walk->walk.all) && walk->seq != 0)
1030                 return 0;
1031
1032         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1033         if (list_empty(&walk->walk.all))
1034                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1035         else
1036                 x = list_first_entry(&walk->walk.all,
1037                                      struct xfrm_policy_walk_entry, all);
1038
1039         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1040                 if (x->dead)
1041                         continue;
1042                 pol = container_of(x, struct xfrm_policy, walk);
1043                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1044                     walk->type != pol->type)
1045                         continue;
1046                 error = func(pol, xfrm_policy_id2dir(pol->index),
1047                              walk->seq, data);
1048                 if (error) {
1049                         list_move_tail(&walk->walk.all, &x->all);
1050                         goto out;
1051                 }
1052                 walk->seq++;
1053         }
1054         if (walk->seq == 0) {
1055                 error = -ENOENT;
1056                 goto out;
1057         }
1058         list_del_init(&walk->walk.all);
1059 out:
1060         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1061         return error;
1062 }
1063 EXPORT_SYMBOL(xfrm_policy_walk);
1064
1065 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1066 {
1067         INIT_LIST_HEAD(&walk->walk.all);
1068         walk->walk.dead = 1;
1069         walk->type = type;
1070         walk->seq = 0;
1071 }
1072 EXPORT_SYMBOL(xfrm_policy_walk_init);
1073
1074 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1075 {
1076         if (list_empty(&walk->walk.all))
1077                 return;
1078
1079         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1080         list_del(&walk->walk.all);
1081         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1082 }
1083 EXPORT_SYMBOL(xfrm_policy_walk_done);
1084
1085 /*
1086  * Find policy to apply to this flow.
1087  *
1088  * Returns 0 if policy found, else an -errno.
1089  */
1090 static int xfrm_policy_match(const struct xfrm_policy *pol,
1091                              const struct flowi *fl,
1092                              u8 type, u16 family, int dir)
1093 {
1094         const struct xfrm_selector *sel = &pol->selector;
1095         int ret = -ESRCH;
1096         bool match;
1097
1098         if (pol->family != family ||
1099             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1100             pol->type != type)
1101                 return ret;
1102
1103         match = xfrm_selector_match(sel, fl, family);
1104         if (match)
1105                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1106                                                   dir);
1107
1108         return ret;
1109 }
1110
1111 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1112                                                      const struct flowi *fl,
1113                                                      u16 family, u8 dir)
1114 {
1115         int err;
1116         struct xfrm_policy *pol, *ret;
1117         const xfrm_address_t *daddr, *saddr;
1118         struct hlist_head *chain;
1119         unsigned int sequence;
1120         u32 priority;
1121
1122         daddr = xfrm_flowi_daddr(fl, family);
1123         saddr = xfrm_flowi_saddr(fl, family);
1124         if (unlikely(!daddr || !saddr))
1125                 return NULL;
1126
1127         rcu_read_lock();
1128  retry:
1129         do {
1130                 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1131                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1132         } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1133
1134         priority = ~0U;
1135         ret = NULL;
1136         hlist_for_each_entry_rcu(pol, chain, bydst) {
1137                 err = xfrm_policy_match(pol, fl, type, family, dir);
1138                 if (err) {
1139                         if (err == -ESRCH)
1140                                 continue;
1141                         else {
1142                                 ret = ERR_PTR(err);
1143                                 goto fail;
1144                         }
1145                 } else {
1146                         ret = pol;
1147                         priority = ret->priority;
1148                         break;
1149                 }
1150         }
1151         chain = &net->xfrm.policy_inexact[dir];
1152         hlist_for_each_entry_rcu(pol, chain, bydst) {
1153                 if ((pol->priority >= priority) && ret)
1154                         break;
1155
1156                 err = xfrm_policy_match(pol, fl, type, family, dir);
1157                 if (err) {
1158                         if (err == -ESRCH)
1159                                 continue;
1160                         else {
1161                                 ret = ERR_PTR(err);
1162                                 goto fail;
1163                         }
1164                 } else {
1165                         ret = pol;
1166                         break;
1167                 }
1168         }
1169
1170         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1171                 goto retry;
1172
1173         if (ret && !xfrm_pol_hold_rcu(ret))
1174                 goto retry;
1175 fail:
1176         rcu_read_unlock();
1177
1178         return ret;
1179 }
1180
1181 static struct xfrm_policy *
1182 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1183 {
1184 #ifdef CONFIG_XFRM_SUB_POLICY
1185         struct xfrm_policy *pol;
1186
1187         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1188         if (pol != NULL)
1189                 return pol;
1190 #endif
1191         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1192 }
1193
1194 static int flow_to_policy_dir(int dir)
1195 {
1196         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1197             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1198             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1199                 return dir;
1200
1201         switch (dir) {
1202         default:
1203         case FLOW_DIR_IN:
1204                 return XFRM_POLICY_IN;
1205         case FLOW_DIR_OUT:
1206                 return XFRM_POLICY_OUT;
1207         case FLOW_DIR_FWD:
1208                 return XFRM_POLICY_FWD;
1209         }
1210 }
1211
1212 static struct flow_cache_object *
1213 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1214                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1215 {
1216         struct xfrm_policy *pol;
1217
1218         if (old_obj)
1219                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1220
1221         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1222         if (IS_ERR_OR_NULL(pol))
1223                 return ERR_CAST(pol);
1224
1225         /* Resolver returns two references:
1226          * one for cache and one for caller of flow_cache_lookup() */
1227         xfrm_pol_hold(pol);
1228
1229         return &pol->flo;
1230 }
1231
1232 static inline int policy_to_flow_dir(int dir)
1233 {
1234         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1235             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1236             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1237                 return dir;
1238         switch (dir) {
1239         default:
1240         case XFRM_POLICY_IN:
1241                 return FLOW_DIR_IN;
1242         case XFRM_POLICY_OUT:
1243                 return FLOW_DIR_OUT;
1244         case XFRM_POLICY_FWD:
1245                 return FLOW_DIR_FWD;
1246         }
1247 }
1248
1249 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1250                                                  const struct flowi *fl, u16 family)
1251 {
1252         struct xfrm_policy *pol;
1253
1254         rcu_read_lock();
1255  again:
1256         pol = rcu_dereference(sk->sk_policy[dir]);
1257         if (pol != NULL) {
1258                 bool match = xfrm_selector_match(&pol->selector, fl, family);
1259                 int err = 0;
1260
1261                 if (match) {
1262                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1263                                 pol = NULL;
1264                                 goto out;
1265                         }
1266                         err = security_xfrm_policy_lookup(pol->security,
1267                                                       fl->flowi_secid,
1268                                                       policy_to_flow_dir(dir));
1269                         if (!err) {
1270                                 if (!xfrm_pol_hold_rcu(pol))
1271                                         goto again;
1272                         } else if (err == -ESRCH) {
1273                                 pol = NULL;
1274                         } else {
1275                                 pol = ERR_PTR(err);
1276                         }
1277                 } else
1278                         pol = NULL;
1279         }
1280 out:
1281         rcu_read_unlock();
1282         return pol;
1283 }
1284
1285 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1286 {
1287         struct net *net = xp_net(pol);
1288
1289         list_add(&pol->walk.all, &net->xfrm.policy_all);
1290         net->xfrm.policy_count[dir]++;
1291         xfrm_pol_hold(pol);
1292 }
1293
1294 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1295                                                 int dir)
1296 {
1297         struct net *net = xp_net(pol);
1298
1299         if (list_empty(&pol->walk.all))
1300                 return NULL;
1301
1302         /* Socket policies are not hashed. */
1303         if (!hlist_unhashed(&pol->bydst)) {
1304                 hlist_del_rcu(&pol->bydst);
1305                 hlist_del(&pol->byidx);
1306         }
1307
1308         list_del_init(&pol->walk.all);
1309         net->xfrm.policy_count[dir]--;
1310
1311         return pol;
1312 }
1313
1314 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1315 {
1316         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1317 }
1318
1319 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1320 {
1321         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1322 }
1323
1324 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1325 {
1326         struct net *net = xp_net(pol);
1327
1328         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1329         pol = __xfrm_policy_unlink(pol, dir);
1330         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1331         if (pol) {
1332                 xfrm_policy_kill(pol);
1333                 return 0;
1334         }
1335         return -ENOENT;
1336 }
1337 EXPORT_SYMBOL(xfrm_policy_delete);
1338
1339 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1340 {
1341         struct net *net = xp_net(pol);
1342         struct xfrm_policy *old_pol;
1343
1344 #ifdef CONFIG_XFRM_SUB_POLICY
1345         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1346                 return -EINVAL;
1347 #endif
1348
1349         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1350         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1351                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1352         if (pol) {
1353                 pol->curlft.add_time = get_seconds();
1354                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1355                 xfrm_sk_policy_link(pol, dir);
1356         }
1357         rcu_assign_pointer(sk->sk_policy[dir], pol);
1358         if (old_pol) {
1359                 if (pol)
1360                         xfrm_policy_requeue(old_pol, pol);
1361
1362                 /* Unlinking succeeds always. This is the only function
1363                  * allowed to delete or replace socket policy.
1364                  */
1365                 xfrm_sk_policy_unlink(old_pol, dir);
1366         }
1367         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1368
1369         if (old_pol) {
1370                 xfrm_policy_kill(old_pol);
1371         }
1372         return 0;
1373 }
1374
1375 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1376 {
1377         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1378         struct net *net = xp_net(old);
1379
1380         if (newp) {
1381                 newp->selector = old->selector;
1382                 if (security_xfrm_policy_clone(old->security,
1383                                                &newp->security)) {
1384                         kfree(newp);
1385                         return NULL;  /* ENOMEM */
1386                 }
1387                 newp->lft = old->lft;
1388                 newp->curlft = old->curlft;
1389                 newp->mark = old->mark;
1390                 newp->action = old->action;
1391                 newp->flags = old->flags;
1392                 newp->xfrm_nr = old->xfrm_nr;
1393                 newp->index = old->index;
1394                 newp->type = old->type;
1395                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1396                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1397                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1398                 xfrm_sk_policy_link(newp, dir);
1399                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1400                 xfrm_pol_put(newp);
1401         }
1402         return newp;
1403 }
1404
1405 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1406 {
1407         const struct xfrm_policy *p;
1408         struct xfrm_policy *np;
1409         int i, ret = 0;
1410
1411         rcu_read_lock();
1412         for (i = 0; i < 2; i++) {
1413                 p = rcu_dereference(osk->sk_policy[i]);
1414                 if (p) {
1415                         np = clone_policy(p, i);
1416                         if (unlikely(!np)) {
1417                                 ret = -ENOMEM;
1418                                 break;
1419                         }
1420                         rcu_assign_pointer(sk->sk_policy[i], np);
1421                 }
1422         }
1423         rcu_read_unlock();
1424         return ret;
1425 }
1426
1427 static int
1428 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1429                xfrm_address_t *remote, unsigned short family)
1430 {
1431         int err;
1432         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1433
1434         if (unlikely(afinfo == NULL))
1435                 return -EINVAL;
1436         err = afinfo->get_saddr(net, oif, local, remote);
1437         rcu_read_unlock();
1438         return err;
1439 }
1440
1441 /* Resolve list of templates for the flow, given policy. */
1442
1443 static int
1444 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1445                       struct xfrm_state **xfrm, unsigned short family)
1446 {
1447         struct net *net = xp_net(policy);
1448         int nx;
1449         int i, error;
1450         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1451         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1452         xfrm_address_t tmp;
1453
1454         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1455                 struct xfrm_state *x;
1456                 xfrm_address_t *remote = daddr;
1457                 xfrm_address_t *local  = saddr;
1458                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1459
1460                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1461                     tmpl->mode == XFRM_MODE_BEET) {
1462                         remote = &tmpl->id.daddr;
1463                         local = &tmpl->saddr;
1464                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1465                                 error = xfrm_get_saddr(net, fl->flowi_oif,
1466                                                        &tmp, remote,
1467                                                        tmpl->encap_family);
1468                                 if (error)
1469                                         goto fail;
1470                                 local = &tmp;
1471                         }
1472                 }
1473
1474                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1475
1476                 if (x && x->km.state == XFRM_STATE_VALID) {
1477                         xfrm[nx++] = x;
1478                         daddr = remote;
1479                         saddr = local;
1480                         continue;
1481                 }
1482                 if (x) {
1483                         error = (x->km.state == XFRM_STATE_ERROR ?
1484                                  -EINVAL : -EAGAIN);
1485                         xfrm_state_put(x);
1486                 } else if (error == -ESRCH) {
1487                         error = -EAGAIN;
1488                 }
1489
1490                 if (!tmpl->optional)
1491                         goto fail;
1492         }
1493         return nx;
1494
1495 fail:
1496         for (nx--; nx >= 0; nx--)
1497                 xfrm_state_put(xfrm[nx]);
1498         return error;
1499 }
1500
1501 static int
1502 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1503                   struct xfrm_state **xfrm, unsigned short family)
1504 {
1505         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1506         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1507         int cnx = 0;
1508         int error;
1509         int ret;
1510         int i;
1511
1512         for (i = 0; i < npols; i++) {
1513                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1514                         error = -ENOBUFS;
1515                         goto fail;
1516                 }
1517
1518                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1519                 if (ret < 0) {
1520                         error = ret;
1521                         goto fail;
1522                 } else
1523                         cnx += ret;
1524         }
1525
1526         /* found states are sorted for outbound processing */
1527         if (npols > 1)
1528                 xfrm_state_sort(xfrm, tpp, cnx, family);
1529
1530         return cnx;
1531
1532  fail:
1533         for (cnx--; cnx >= 0; cnx--)
1534                 xfrm_state_put(tpp[cnx]);
1535         return error;
1536
1537 }
1538
1539 static int xfrm_get_tos(const struct flowi *fl, int family)
1540 {
1541         const struct xfrm_policy_afinfo *afinfo;
1542         int tos = 0;
1543
1544         afinfo = xfrm_policy_get_afinfo(family);
1545         tos = afinfo ? afinfo->get_tos(fl) : 0;
1546
1547         rcu_read_unlock();
1548
1549         return tos;
1550 }
1551
1552 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1553 {
1554         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1555         struct dst_entry *dst = &xdst->u.dst;
1556
1557         if (xdst->route == NULL) {
1558                 /* Dummy bundle - if it has xfrms we were not
1559                  * able to build bundle as template resolution failed.
1560                  * It means we need to try again resolving. */
1561                 if (xdst->num_xfrms > 0)
1562                         return NULL;
1563         } else if (dst->flags & DST_XFRM_QUEUE) {
1564                 return NULL;
1565         } else {
1566                 /* Real bundle */
1567                 if (stale_bundle(dst))
1568                         return NULL;
1569         }
1570
1571         dst_hold(dst);
1572         return flo;
1573 }
1574
1575 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1576 {
1577         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1578         struct dst_entry *dst = &xdst->u.dst;
1579
1580         if (!xdst->route)
1581                 return 0;
1582         if (stale_bundle(dst))
1583                 return 0;
1584
1585         return 1;
1586 }
1587
1588 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1589 {
1590         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1591         struct dst_entry *dst = &xdst->u.dst;
1592
1593         dst_free(dst);
1594 }
1595
1596 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1597         .get = xfrm_bundle_flo_get,
1598         .check = xfrm_bundle_flo_check,
1599         .delete = xfrm_bundle_flo_delete,
1600 };
1601
1602 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1603 {
1604         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1605         struct dst_ops *dst_ops;
1606         struct xfrm_dst *xdst;
1607
1608         if (!afinfo)
1609                 return ERR_PTR(-EINVAL);
1610
1611         switch (family) {
1612         case AF_INET:
1613                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1614                 break;
1615 #if IS_ENABLED(CONFIG_IPV6)
1616         case AF_INET6:
1617                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1618                 break;
1619 #endif
1620         default:
1621                 BUG();
1622         }
1623         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1624
1625         if (likely(xdst)) {
1626                 struct dst_entry *dst = &xdst->u.dst;
1627
1628                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1629                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1630         } else
1631                 xdst = ERR_PTR(-ENOBUFS);
1632
1633         rcu_read_unlock();
1634
1635         return xdst;
1636 }
1637
1638 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1639                                  int nfheader_len)
1640 {
1641         const struct xfrm_policy_afinfo *afinfo =
1642                 xfrm_policy_get_afinfo(dst->ops->family);
1643         int err;
1644
1645         if (!afinfo)
1646                 return -EINVAL;
1647
1648         err = afinfo->init_path(path, dst, nfheader_len);
1649
1650         rcu_read_unlock();
1651
1652         return err;
1653 }
1654
1655 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1656                                 const struct flowi *fl)
1657 {
1658         const struct xfrm_policy_afinfo *afinfo =
1659                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1660         int err;
1661
1662         if (!afinfo)
1663                 return -EINVAL;
1664
1665         err = afinfo->fill_dst(xdst, dev, fl);
1666
1667         rcu_read_unlock();
1668
1669         return err;
1670 }
1671
1672
1673 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1674  * all the metrics... Shortly, bundle a bundle.
1675  */
1676
1677 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1678                                             struct xfrm_state **xfrm, int nx,
1679                                             const struct flowi *fl,
1680                                             struct dst_entry *dst)
1681 {
1682         struct net *net = xp_net(policy);
1683         unsigned long now = jiffies;
1684         struct net_device *dev;
1685         struct xfrm_mode *inner_mode;
1686         struct dst_entry *dst_prev = NULL;
1687         struct dst_entry *dst0 = NULL;
1688         int i = 0;
1689         int err;
1690         int header_len = 0;
1691         int nfheader_len = 0;
1692         int trailer_len = 0;
1693         int tos;
1694         int family = policy->selector.family;
1695         xfrm_address_t saddr, daddr;
1696
1697         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1698
1699         tos = xfrm_get_tos(fl, family);
1700
1701         dst_hold(dst);
1702
1703         for (; i < nx; i++) {
1704                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1705                 struct dst_entry *dst1 = &xdst->u.dst;
1706
1707                 err = PTR_ERR(xdst);
1708                 if (IS_ERR(xdst)) {
1709                         dst_release(dst);
1710                         goto put_states;
1711                 }
1712
1713                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1714                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1715                                                         xfrm_af2proto(family));
1716                         if (!inner_mode) {
1717                                 err = -EAFNOSUPPORT;
1718                                 dst_release(dst);
1719                                 goto put_states;
1720                         }
1721                 } else
1722                         inner_mode = xfrm[i]->inner_mode;
1723
1724                 if (!dst_prev)
1725                         dst0 = dst1;
1726                 else {
1727                         dst_prev->child = dst_clone(dst1);
1728                         dst1->flags |= DST_NOHASH;
1729                 }
1730
1731                 xdst->route = dst;
1732                 dst_copy_metrics(dst1, dst);
1733
1734                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1735                         family = xfrm[i]->props.family;
1736                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1737                                               &saddr, &daddr, family);
1738                         err = PTR_ERR(dst);
1739                         if (IS_ERR(dst))
1740                                 goto put_states;
1741                 } else
1742                         dst_hold(dst);
1743
1744                 dst1->xfrm = xfrm[i];
1745                 xdst->xfrm_genid = xfrm[i]->genid;
1746
1747                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1748                 dst1->flags |= DST_HOST;
1749                 dst1->lastuse = now;
1750
1751                 dst1->input = dst_discard;
1752                 dst1->output = inner_mode->afinfo->output;
1753
1754                 dst1->next = dst_prev;
1755                 dst_prev = dst1;
1756
1757                 header_len += xfrm[i]->props.header_len;
1758                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1759                         nfheader_len += xfrm[i]->props.header_len;
1760                 trailer_len += xfrm[i]->props.trailer_len;
1761         }
1762
1763         dst_prev->child = dst;
1764         dst0->path = dst;
1765
1766         err = -ENODEV;
1767         dev = dst->dev;
1768         if (!dev)
1769                 goto free_dst;
1770
1771         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1772         xfrm_init_pmtu(dst_prev);
1773
1774         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1775                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1776
1777                 err = xfrm_fill_dst(xdst, dev, fl);
1778                 if (err)
1779                         goto free_dst;
1780
1781                 dst_prev->header_len = header_len;
1782                 dst_prev->trailer_len = trailer_len;
1783                 header_len -= xdst->u.dst.xfrm->props.header_len;
1784                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1785         }
1786
1787 out:
1788         return dst0;
1789
1790 put_states:
1791         for (; i < nx; i++)
1792                 xfrm_state_put(xfrm[i]);
1793 free_dst:
1794         if (dst0)
1795                 dst_free(dst0);
1796         dst0 = ERR_PTR(err);
1797         goto out;
1798 }
1799
1800 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1801                                 struct xfrm_policy **pols,
1802                                 int *num_pols, int *num_xfrms)
1803 {
1804         int i;
1805
1806         if (*num_pols == 0 || !pols[0]) {
1807                 *num_pols = 0;
1808                 *num_xfrms = 0;
1809                 return 0;
1810         }
1811         if (IS_ERR(pols[0]))
1812                 return PTR_ERR(pols[0]);
1813
1814         *num_xfrms = pols[0]->xfrm_nr;
1815
1816 #ifdef CONFIG_XFRM_SUB_POLICY
1817         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1818             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1819                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1820                                                     XFRM_POLICY_TYPE_MAIN,
1821                                                     fl, family,
1822                                                     XFRM_POLICY_OUT);
1823                 if (pols[1]) {
1824                         if (IS_ERR(pols[1])) {
1825                                 xfrm_pols_put(pols, *num_pols);
1826                                 return PTR_ERR(pols[1]);
1827                         }
1828                         (*num_pols)++;
1829                         (*num_xfrms) += pols[1]->xfrm_nr;
1830                 }
1831         }
1832 #endif
1833         for (i = 0; i < *num_pols; i++) {
1834                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1835                         *num_xfrms = -1;
1836                         break;
1837                 }
1838         }
1839
1840         return 0;
1841
1842 }
1843
1844 static struct xfrm_dst *
1845 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1846                                const struct flowi *fl, u16 family,
1847                                struct dst_entry *dst_orig)
1848 {
1849         struct net *net = xp_net(pols[0]);
1850         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1851         struct dst_entry *dst;
1852         struct xfrm_dst *xdst;
1853         int err;
1854
1855         /* Try to instantiate a bundle */
1856         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1857         if (err <= 0) {
1858                 if (err != 0 && err != -EAGAIN)
1859                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1860                 return ERR_PTR(err);
1861         }
1862
1863         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1864         if (IS_ERR(dst)) {
1865                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1866                 return ERR_CAST(dst);
1867         }
1868
1869         xdst = (struct xfrm_dst *)dst;
1870         xdst->num_xfrms = err;
1871         xdst->num_pols = num_pols;
1872         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1873         xdst->policy_genid = atomic_read(&pols[0]->genid);
1874
1875         return xdst;
1876 }
1877
1878 static void xfrm_policy_queue_process(unsigned long arg)
1879 {
1880         struct sk_buff *skb;
1881         struct sock *sk;
1882         struct dst_entry *dst;
1883         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1884         struct net *net = xp_net(pol);
1885         struct xfrm_policy_queue *pq = &pol->polq;
1886         struct flowi fl;
1887         struct sk_buff_head list;
1888
1889         spin_lock(&pq->hold_queue.lock);
1890         skb = skb_peek(&pq->hold_queue);
1891         if (!skb) {
1892                 spin_unlock(&pq->hold_queue.lock);
1893                 goto out;
1894         }
1895         dst = skb_dst(skb);
1896         sk = skb->sk;
1897         xfrm_decode_session(skb, &fl, dst->ops->family);
1898         spin_unlock(&pq->hold_queue.lock);
1899
1900         dst_hold(dst->path);
1901         dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1902         if (IS_ERR(dst))
1903                 goto purge_queue;
1904
1905         if (dst->flags & DST_XFRM_QUEUE) {
1906                 dst_release(dst);
1907
1908                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1909                         goto purge_queue;
1910
1911                 pq->timeout = pq->timeout << 1;
1912                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1913                         xfrm_pol_hold(pol);
1914         goto out;
1915         }
1916
1917         dst_release(dst);
1918
1919         __skb_queue_head_init(&list);
1920
1921         spin_lock(&pq->hold_queue.lock);
1922         pq->timeout = 0;
1923         skb_queue_splice_init(&pq->hold_queue, &list);
1924         spin_unlock(&pq->hold_queue.lock);
1925
1926         while (!skb_queue_empty(&list)) {
1927                 skb = __skb_dequeue(&list);
1928
1929                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1930                 dst_hold(skb_dst(skb)->path);
1931                 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1932                 if (IS_ERR(dst)) {
1933                         kfree_skb(skb);
1934                         continue;
1935                 }
1936
1937                 nf_reset(skb);
1938                 skb_dst_drop(skb);
1939                 skb_dst_set(skb, dst);
1940
1941                 dst_output(net, skb->sk, skb);
1942         }
1943
1944 out:
1945         xfrm_pol_put(pol);
1946         return;
1947
1948 purge_queue:
1949         pq->timeout = 0;
1950         skb_queue_purge(&pq->hold_queue);
1951         xfrm_pol_put(pol);
1952 }
1953
1954 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1955 {
1956         unsigned long sched_next;
1957         struct dst_entry *dst = skb_dst(skb);
1958         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1959         struct xfrm_policy *pol = xdst->pols[0];
1960         struct xfrm_policy_queue *pq = &pol->polq;
1961
1962         if (unlikely(skb_fclone_busy(sk, skb))) {
1963                 kfree_skb(skb);
1964                 return 0;
1965         }
1966
1967         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1968                 kfree_skb(skb);
1969                 return -EAGAIN;
1970         }
1971
1972         skb_dst_force(skb);
1973
1974         spin_lock_bh(&pq->hold_queue.lock);
1975
1976         if (!pq->timeout)
1977                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1978
1979         sched_next = jiffies + pq->timeout;
1980
1981         if (del_timer(&pq->hold_timer)) {
1982                 if (time_before(pq->hold_timer.expires, sched_next))
1983                         sched_next = pq->hold_timer.expires;
1984                 xfrm_pol_put(pol);
1985         }
1986
1987         __skb_queue_tail(&pq->hold_queue, skb);
1988         if (!mod_timer(&pq->hold_timer, sched_next))
1989                 xfrm_pol_hold(pol);
1990
1991         spin_unlock_bh(&pq->hold_queue.lock);
1992
1993         return 0;
1994 }
1995
1996 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1997                                                  struct xfrm_flo *xflo,
1998                                                  const struct flowi *fl,
1999                                                  int num_xfrms,
2000                                                  u16 family)
2001 {
2002         int err;
2003         struct net_device *dev;
2004         struct dst_entry *dst;
2005         struct dst_entry *dst1;
2006         struct xfrm_dst *xdst;
2007
2008         xdst = xfrm_alloc_dst(net, family);
2009         if (IS_ERR(xdst))
2010                 return xdst;
2011
2012         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2013             net->xfrm.sysctl_larval_drop ||
2014             num_xfrms <= 0)
2015                 return xdst;
2016
2017         dst = xflo->dst_orig;
2018         dst1 = &xdst->u.dst;
2019         dst_hold(dst);
2020         xdst->route = dst;
2021
2022         dst_copy_metrics(dst1, dst);
2023
2024         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2025         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2026         dst1->lastuse = jiffies;
2027
2028         dst1->input = dst_discard;
2029         dst1->output = xdst_queue_output;
2030
2031         dst_hold(dst);
2032         dst1->child = dst;
2033         dst1->path = dst;
2034
2035         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2036
2037         err = -ENODEV;
2038         dev = dst->dev;
2039         if (!dev)
2040                 goto free_dst;
2041
2042         err = xfrm_fill_dst(xdst, dev, fl);
2043         if (err)
2044                 goto free_dst;
2045
2046 out:
2047         return xdst;
2048
2049 free_dst:
2050         dst_release(dst1);
2051         xdst = ERR_PTR(err);
2052         goto out;
2053 }
2054
2055 static struct flow_cache_object *
2056 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2057                    struct flow_cache_object *oldflo, void *ctx)
2058 {
2059         struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2060         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2061         struct xfrm_dst *xdst, *new_xdst;
2062         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2063
2064         /* Check if the policies from old bundle are usable */
2065         xdst = NULL;
2066         if (oldflo) {
2067                 xdst = container_of(oldflo, struct xfrm_dst, flo);
2068                 num_pols = xdst->num_pols;
2069                 num_xfrms = xdst->num_xfrms;
2070                 pol_dead = 0;
2071                 for (i = 0; i < num_pols; i++) {
2072                         pols[i] = xdst->pols[i];
2073                         pol_dead |= pols[i]->walk.dead;
2074                 }
2075                 if (pol_dead) {
2076                         dst_free(&xdst->u.dst);
2077                         xdst = NULL;
2078                         num_pols = 0;
2079                         num_xfrms = 0;
2080                         oldflo = NULL;
2081                 }
2082         }
2083
2084         /* Resolve policies to use if we couldn't get them from
2085          * previous cache entry */
2086         if (xdst == NULL) {
2087                 num_pols = 1;
2088                 pols[0] = __xfrm_policy_lookup(net, fl, family,
2089                                                flow_to_policy_dir(dir));
2090                 err = xfrm_expand_policies(fl, family, pols,
2091                                            &num_pols, &num_xfrms);
2092                 if (err < 0)
2093                         goto inc_error;
2094                 if (num_pols == 0)
2095                         return NULL;
2096                 if (num_xfrms <= 0)
2097                         goto make_dummy_bundle;
2098         }
2099
2100         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2101                                                   xflo->dst_orig);
2102         if (IS_ERR(new_xdst)) {
2103                 err = PTR_ERR(new_xdst);
2104                 if (err != -EAGAIN)
2105                         goto error;
2106                 if (oldflo == NULL)
2107                         goto make_dummy_bundle;
2108                 dst_hold(&xdst->u.dst);
2109                 return oldflo;
2110         } else if (new_xdst == NULL) {
2111                 num_xfrms = 0;
2112                 if (oldflo == NULL)
2113                         goto make_dummy_bundle;
2114                 xdst->num_xfrms = 0;
2115                 dst_hold(&xdst->u.dst);
2116                 return oldflo;
2117         }
2118
2119         /* Kill the previous bundle */
2120         if (xdst) {
2121                 /* The policies were stolen for newly generated bundle */
2122                 xdst->num_pols = 0;
2123                 dst_free(&xdst->u.dst);
2124         }
2125
2126         /* Flow cache does not have reference, it dst_free()'s,
2127          * but we do need to return one reference for original caller */
2128         dst_hold(&new_xdst->u.dst);
2129         return &new_xdst->flo;
2130
2131 make_dummy_bundle:
2132         /* We found policies, but there's no bundles to instantiate:
2133          * either because the policy blocks, has no transformations or
2134          * we could not build template (no xfrm_states).*/
2135         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2136         if (IS_ERR(xdst)) {
2137                 xfrm_pols_put(pols, num_pols);
2138                 return ERR_CAST(xdst);
2139         }
2140         xdst->num_pols = num_pols;
2141         xdst->num_xfrms = num_xfrms;
2142         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2143
2144         dst_hold(&xdst->u.dst);
2145         return &xdst->flo;
2146
2147 inc_error:
2148         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2149 error:
2150         if (xdst != NULL)
2151                 dst_free(&xdst->u.dst);
2152         else
2153                 xfrm_pols_put(pols, num_pols);
2154         return ERR_PTR(err);
2155 }
2156
2157 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2158                                         struct dst_entry *dst_orig)
2159 {
2160         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2161         struct dst_entry *ret;
2162
2163         if (!afinfo) {
2164                 dst_release(dst_orig);
2165                 return ERR_PTR(-EINVAL);
2166         } else {
2167                 ret = afinfo->blackhole_route(net, dst_orig);
2168         }
2169         rcu_read_unlock();
2170
2171         return ret;
2172 }
2173
2174 /* Main function: finds/creates a bundle for given flow.
2175  *
2176  * At the moment we eat a raw IP route. Mostly to speed up lookups
2177  * on interfaces with disabled IPsec.
2178  */
2179 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2180                               const struct flowi *fl,
2181                               const struct sock *sk, int flags)
2182 {
2183         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2184         struct flow_cache_object *flo;
2185         struct xfrm_dst *xdst;
2186         struct dst_entry *dst, *route;
2187         u16 family = dst_orig->ops->family;
2188         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2189         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2190
2191         dst = NULL;
2192         xdst = NULL;
2193         route = NULL;
2194
2195         sk = sk_const_to_full_sk(sk);
2196         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2197                 num_pols = 1;
2198                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2199                 err = xfrm_expand_policies(fl, family, pols,
2200                                            &num_pols, &num_xfrms);
2201                 if (err < 0)
2202                         goto dropdst;
2203
2204                 if (num_pols) {
2205                         if (num_xfrms <= 0) {
2206                                 drop_pols = num_pols;
2207                                 goto no_transform;
2208                         }
2209
2210                         xdst = xfrm_resolve_and_create_bundle(
2211                                         pols, num_pols, fl,
2212                                         family, dst_orig);
2213                         if (IS_ERR(xdst)) {
2214                                 xfrm_pols_put(pols, num_pols);
2215                                 err = PTR_ERR(xdst);
2216                                 goto dropdst;
2217                         } else if (xdst == NULL) {
2218                                 num_xfrms = 0;
2219                                 drop_pols = num_pols;
2220                                 goto no_transform;
2221                         }
2222
2223                         dst_hold(&xdst->u.dst);
2224                         xdst->u.dst.flags |= DST_NOCACHE;
2225                         route = xdst->route;
2226                 }
2227         }
2228
2229         if (xdst == NULL) {
2230                 struct xfrm_flo xflo;
2231
2232                 xflo.dst_orig = dst_orig;
2233                 xflo.flags = flags;
2234
2235                 /* To accelerate a bit...  */
2236                 if ((dst_orig->flags & DST_NOXFRM) ||
2237                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2238                         goto nopol;
2239
2240                 flo = flow_cache_lookup(net, fl, family, dir,
2241                                         xfrm_bundle_lookup, &xflo);
2242                 if (flo == NULL)
2243                         goto nopol;
2244                 if (IS_ERR(flo)) {
2245                         err = PTR_ERR(flo);
2246                         goto dropdst;
2247                 }
2248                 xdst = container_of(flo, struct xfrm_dst, flo);
2249
2250                 num_pols = xdst->num_pols;
2251                 num_xfrms = xdst->num_xfrms;
2252                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2253                 route = xdst->route;
2254         }
2255
2256         dst = &xdst->u.dst;
2257         if (route == NULL && num_xfrms > 0) {
2258                 /* The only case when xfrm_bundle_lookup() returns a
2259                  * bundle with null route, is when the template could
2260                  * not be resolved. It means policies are there, but
2261                  * bundle could not be created, since we don't yet
2262                  * have the xfrm_state's. We need to wait for KM to
2263                  * negotiate new SA's or bail out with error.*/
2264                 if (net->xfrm.sysctl_larval_drop) {
2265                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2266                         err = -EREMOTE;
2267                         goto error;
2268                 }
2269
2270                 err = -EAGAIN;
2271
2272                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2273                 goto error;
2274         }
2275
2276 no_transform:
2277         if (num_pols == 0)
2278                 goto nopol;
2279
2280         if ((flags & XFRM_LOOKUP_ICMP) &&
2281             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2282                 err = -ENOENT;
2283                 goto error;
2284         }
2285
2286         for (i = 0; i < num_pols; i++)
2287                 pols[i]->curlft.use_time = get_seconds();
2288
2289         if (num_xfrms < 0) {
2290                 /* Prohibit the flow */
2291                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2292                 err = -EPERM;
2293                 goto error;
2294         } else if (num_xfrms > 0) {
2295                 /* Flow transformed */
2296                 dst_release(dst_orig);
2297         } else {
2298                 /* Flow passes untransformed */
2299                 dst_release(dst);
2300                 dst = dst_orig;
2301         }
2302 ok:
2303         xfrm_pols_put(pols, drop_pols);
2304         if (dst && dst->xfrm &&
2305             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2306                 dst->flags |= DST_XFRM_TUNNEL;
2307         return dst;
2308
2309 nopol:
2310         if (!(flags & XFRM_LOOKUP_ICMP)) {
2311                 dst = dst_orig;
2312                 goto ok;
2313         }
2314         err = -ENOENT;
2315 error:
2316         dst_release(dst);
2317 dropdst:
2318         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2319                 dst_release(dst_orig);
2320         xfrm_pols_put(pols, drop_pols);
2321         return ERR_PTR(err);
2322 }
2323 EXPORT_SYMBOL(xfrm_lookup);
2324
2325 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2326  * Otherwise we may send out blackholed packets.
2327  */
2328 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2329                                     const struct flowi *fl,
2330                                     const struct sock *sk, int flags)
2331 {
2332         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2333                                             flags | XFRM_LOOKUP_QUEUE |
2334                                             XFRM_LOOKUP_KEEP_DST_REF);
2335
2336         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2337                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2338
2339         return dst;
2340 }
2341 EXPORT_SYMBOL(xfrm_lookup_route);
2342
2343 static inline int
2344 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2345 {
2346         struct xfrm_state *x;
2347
2348         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2349                 return 0;
2350         x = skb->sp->xvec[idx];
2351         if (!x->type->reject)
2352                 return 0;
2353         return x->type->reject(x, skb, fl);
2354 }
2355
2356 /* When skb is transformed back to its "native" form, we have to
2357  * check policy restrictions. At the moment we make this in maximally
2358  * stupid way. Shame on me. :-) Of course, connected sockets must
2359  * have policy cached at them.
2360  */
2361
2362 static inline int
2363 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2364               unsigned short family)
2365 {
2366         if (xfrm_state_kern(x))
2367                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2368         return  x->id.proto == tmpl->id.proto &&
2369                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2370                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2371                 x->props.mode == tmpl->mode &&
2372                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2373                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2374                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2375                   xfrm_state_addr_cmp(tmpl, x, family));
2376 }
2377
2378 /*
2379  * 0 or more than 0 is returned when validation is succeeded (either bypass
2380  * because of optional transport mode, or next index of the mathced secpath
2381  * state with the template.
2382  * -1 is returned when no matching template is found.
2383  * Otherwise "-2 - errored_index" is returned.
2384  */
2385 static inline int
2386 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2387                unsigned short family)
2388 {
2389         int idx = start;
2390
2391         if (tmpl->optional) {
2392                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2393                         return start;
2394         } else
2395                 start = -1;
2396         for (; idx < sp->len; idx++) {
2397                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2398                         return ++idx;
2399                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2400                         if (start == -1)
2401                                 start = -2-idx;
2402                         break;
2403                 }
2404         }
2405         return start;
2406 }
2407
2408 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2409                           unsigned int family, int reverse)
2410 {
2411         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2412         int err;
2413
2414         if (unlikely(afinfo == NULL))
2415                 return -EAFNOSUPPORT;
2416
2417         afinfo->decode_session(skb, fl, reverse);
2418         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2419         rcu_read_unlock();
2420         return err;
2421 }
2422 EXPORT_SYMBOL(__xfrm_decode_session);
2423
2424 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2425 {
2426         for (; k < sp->len; k++) {
2427                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2428                         *idxp = k;
2429                         return 1;
2430                 }
2431         }
2432
2433         return 0;
2434 }
2435
2436 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2437                         unsigned short family)
2438 {
2439         struct net *net = dev_net(skb->dev);
2440         struct xfrm_policy *pol;
2441         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2442         int npols = 0;
2443         int xfrm_nr;
2444         int pi;
2445         int reverse;
2446         struct flowi fl;
2447         u8 fl_dir;
2448         int xerr_idx = -1;
2449
2450         reverse = dir & ~XFRM_POLICY_MASK;
2451         dir &= XFRM_POLICY_MASK;
2452         fl_dir = policy_to_flow_dir(dir);
2453
2454         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2455                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2456                 return 0;
2457         }
2458
2459         nf_nat_decode_session(skb, &fl, family);
2460
2461         /* First, check used SA against their selectors. */
2462         if (skb->sp) {
2463                 int i;
2464
2465                 for (i = skb->sp->len-1; i >= 0; i--) {
2466                         struct xfrm_state *x = skb->sp->xvec[i];
2467                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2468                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2469                                 return 0;
2470                         }
2471                 }
2472         }
2473
2474         pol = NULL;
2475         sk = sk_to_full_sk(sk);
2476         if (sk && sk->sk_policy[dir]) {
2477                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2478                 if (IS_ERR(pol)) {
2479                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2480                         return 0;
2481                 }
2482         }
2483
2484         if (!pol) {
2485                 struct flow_cache_object *flo;
2486
2487                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2488                                         xfrm_policy_lookup, NULL);
2489                 if (IS_ERR_OR_NULL(flo))
2490                         pol = ERR_CAST(flo);
2491                 else
2492                         pol = container_of(flo, struct xfrm_policy, flo);
2493         }
2494
2495         if (IS_ERR(pol)) {
2496                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2497                 return 0;
2498         }
2499
2500         if (!pol) {
2501                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2502                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2503                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2504                         return 0;
2505                 }
2506                 return 1;
2507         }
2508
2509         pol->curlft.use_time = get_seconds();
2510
2511         pols[0] = pol;
2512         npols++;
2513 #ifdef CONFIG_XFRM_SUB_POLICY
2514         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2515                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2516                                                     &fl, family,
2517                                                     XFRM_POLICY_IN);
2518                 if (pols[1]) {
2519                         if (IS_ERR(pols[1])) {
2520                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2521                                 return 0;
2522                         }
2523                         pols[1]->curlft.use_time = get_seconds();
2524                         npols++;
2525                 }
2526         }
2527 #endif
2528
2529         if (pol->action == XFRM_POLICY_ALLOW) {
2530                 struct sec_path *sp;
2531                 static struct sec_path dummy;
2532                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2533                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2534                 struct xfrm_tmpl **tpp = tp;
2535                 int ti = 0;
2536                 int i, k;
2537
2538                 if ((sp = skb->sp) == NULL)
2539                         sp = &dummy;
2540
2541                 for (pi = 0; pi < npols; pi++) {
2542                         if (pols[pi] != pol &&
2543                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2544                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2545                                 goto reject;
2546                         }
2547                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2548                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2549                                 goto reject_error;
2550                         }
2551                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2552                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2553                 }
2554                 xfrm_nr = ti;
2555                 if (npols > 1) {
2556                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2557                         tpp = stp;
2558                 }
2559
2560                 /* For each tunnel xfrm, find the first matching tmpl.
2561                  * For each tmpl before that, find corresponding xfrm.
2562                  * Order is _important_. Later we will implement
2563                  * some barriers, but at the moment barriers
2564                  * are implied between each two transformations.
2565                  */
2566                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2567                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2568                         if (k < 0) {
2569                                 if (k < -1)
2570                                         /* "-2 - errored_index" returned */
2571                                         xerr_idx = -(2+k);
2572                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2573                                 goto reject;
2574                         }
2575                 }
2576
2577                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2578                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2579                         goto reject;
2580                 }
2581
2582                 xfrm_pols_put(pols, npols);
2583                 return 1;
2584         }
2585         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2586
2587 reject:
2588         xfrm_secpath_reject(xerr_idx, skb, &fl);
2589 reject_error:
2590         xfrm_pols_put(pols, npols);
2591         return 0;
2592 }
2593 EXPORT_SYMBOL(__xfrm_policy_check);
2594
2595 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2596 {
2597         struct net *net = dev_net(skb->dev);
2598         struct flowi fl;
2599         struct dst_entry *dst;
2600         int res = 1;
2601
2602         if (xfrm_decode_session(skb, &fl, family) < 0) {
2603                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2604                 return 0;
2605         }
2606
2607         skb_dst_force(skb);
2608
2609         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2610         if (IS_ERR(dst)) {
2611                 res = 0;
2612                 dst = NULL;
2613         }
2614         skb_dst_set(skb, dst);
2615         return res;
2616 }
2617 EXPORT_SYMBOL(__xfrm_route_forward);
2618
2619 /* Optimize later using cookies and generation ids. */
2620
2621 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2622 {
2623         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2624          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2625          * get validated by dst_ops->check on every use.  We do this
2626          * because when a normal route referenced by an XFRM dst is
2627          * obsoleted we do not go looking around for all parent
2628          * referencing XFRM dsts so that we can invalidate them.  It
2629          * is just too much work.  Instead we make the checks here on
2630          * every use.  For example:
2631          *
2632          *      XFRM dst A --> IPv4 dst X
2633          *
2634          * X is the "xdst->route" of A (X is also the "dst->path" of A
2635          * in this example).  If X is marked obsolete, "A" will not
2636          * notice.  That's what we are validating here via the
2637          * stale_bundle() check.
2638          *
2639          * When a policy's bundle is pruned, we dst_free() the XFRM
2640          * dst which causes it's ->obsolete field to be set to
2641          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2642          * this, we want to force a new route lookup.
2643          */
2644         if (dst->obsolete < 0 && !stale_bundle(dst))
2645                 return dst;
2646
2647         return NULL;
2648 }
2649
2650 static int stale_bundle(struct dst_entry *dst)
2651 {
2652         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2653 }
2654
2655 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2656 {
2657         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2658                 dst->dev = dev_net(dev)->loopback_dev;
2659                 dev_hold(dst->dev);
2660                 dev_put(dev);
2661         }
2662 }
2663 EXPORT_SYMBOL(xfrm_dst_ifdown);
2664
2665 static void xfrm_link_failure(struct sk_buff *skb)
2666 {
2667         /* Impossible. Such dst must be popped before reaches point of failure. */
2668 }
2669
2670 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2671 {
2672         if (dst) {
2673                 if (dst->obsolete) {
2674                         dst_release(dst);
2675                         dst = NULL;
2676                 }
2677         }
2678         return dst;
2679 }
2680
2681 void xfrm_garbage_collect(struct net *net)
2682 {
2683         flow_cache_flush(net);
2684 }
2685 EXPORT_SYMBOL(xfrm_garbage_collect);
2686
2687 void xfrm_garbage_collect_deferred(struct net *net)
2688 {
2689         flow_cache_flush_deferred(net);
2690 }
2691 EXPORT_SYMBOL(xfrm_garbage_collect_deferred);
2692
2693 static void xfrm_init_pmtu(struct dst_entry *dst)
2694 {
2695         do {
2696                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2697                 u32 pmtu, route_mtu_cached;
2698
2699                 pmtu = dst_mtu(dst->child);
2700                 xdst->child_mtu_cached = pmtu;
2701
2702                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2703
2704                 route_mtu_cached = dst_mtu(xdst->route);
2705                 xdst->route_mtu_cached = route_mtu_cached;
2706
2707                 if (pmtu > route_mtu_cached)
2708                         pmtu = route_mtu_cached;
2709
2710                 dst_metric_set(dst, RTAX_MTU, pmtu);
2711         } while ((dst = dst->next));
2712 }
2713
2714 /* Check that the bundle accepts the flow and its components are
2715  * still valid.
2716  */
2717
2718 static int xfrm_bundle_ok(struct xfrm_dst *first)
2719 {
2720         struct dst_entry *dst = &first->u.dst;
2721         struct xfrm_dst *last;
2722         u32 mtu;
2723
2724         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2725             (dst->dev && !netif_running(dst->dev)))
2726                 return 0;
2727
2728         if (dst->flags & DST_XFRM_QUEUE)
2729                 return 1;
2730
2731         last = NULL;
2732
2733         do {
2734                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2735
2736                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2737                         return 0;
2738                 if (xdst->xfrm_genid != dst->xfrm->genid)
2739                         return 0;
2740                 if (xdst->num_pols > 0 &&
2741                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2742                         return 0;
2743
2744                 mtu = dst_mtu(dst->child);
2745                 if (xdst->child_mtu_cached != mtu) {
2746                         last = xdst;
2747                         xdst->child_mtu_cached = mtu;
2748                 }
2749
2750                 if (!dst_check(xdst->route, xdst->route_cookie))
2751                         return 0;
2752                 mtu = dst_mtu(xdst->route);
2753                 if (xdst->route_mtu_cached != mtu) {
2754                         last = xdst;
2755                         xdst->route_mtu_cached = mtu;
2756                 }
2757
2758                 dst = dst->child;
2759         } while (dst->xfrm);
2760
2761         if (likely(!last))
2762                 return 1;
2763
2764         mtu = last->child_mtu_cached;
2765         for (;;) {
2766                 dst = &last->u.dst;
2767
2768                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2769                 if (mtu > last->route_mtu_cached)
2770                         mtu = last->route_mtu_cached;
2771                 dst_metric_set(dst, RTAX_MTU, mtu);
2772
2773                 if (last == first)
2774                         break;
2775
2776                 last = (struct xfrm_dst *)last->u.dst.next;
2777                 last->child_mtu_cached = mtu;
2778         }
2779
2780         return 1;
2781 }
2782
2783 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2784 {
2785         return dst_metric_advmss(dst->path);
2786 }
2787
2788 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2789 {
2790         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2791
2792         return mtu ? : dst_mtu(dst->path);
2793 }
2794
2795 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2796                                         const void *daddr)
2797 {
2798         const struct dst_entry *path = dst->path;
2799
2800         for (; dst != path; dst = dst->child) {
2801                 const struct xfrm_state *xfrm = dst->xfrm;
2802
2803                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2804                         continue;
2805                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2806                         daddr = xfrm->coaddr;
2807                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2808                         daddr = &xfrm->id.daddr;
2809         }
2810         return daddr;
2811 }
2812
2813 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2814                                            struct sk_buff *skb,
2815                                            const void *daddr)
2816 {
2817         const struct dst_entry *path = dst->path;
2818
2819         if (!skb)
2820                 daddr = xfrm_get_dst_nexthop(dst, daddr);
2821         return path->ops->neigh_lookup(path, skb, daddr);
2822 }
2823
2824 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2825 {
2826         const struct dst_entry *path = dst->path;
2827
2828         daddr = xfrm_get_dst_nexthop(dst, daddr);
2829         path->ops->confirm_neigh(path, daddr);
2830 }
2831
2832 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2833 {
2834         int err = 0;
2835
2836         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2837                 return -EAFNOSUPPORT;
2838
2839         spin_lock(&xfrm_policy_afinfo_lock);
2840         if (unlikely(xfrm_policy_afinfo[family] != NULL))
2841                 err = -EEXIST;
2842         else {
2843                 struct dst_ops *dst_ops = afinfo->dst_ops;
2844                 if (likely(dst_ops->kmem_cachep == NULL))
2845                         dst_ops->kmem_cachep = xfrm_dst_cache;
2846                 if (likely(dst_ops->check == NULL))
2847                         dst_ops->check = xfrm_dst_check;
2848                 if (likely(dst_ops->default_advmss == NULL))
2849                         dst_ops->default_advmss = xfrm_default_advmss;
2850                 if (likely(dst_ops->mtu == NULL))
2851                         dst_ops->mtu = xfrm_mtu;
2852                 if (likely(dst_ops->negative_advice == NULL))
2853                         dst_ops->negative_advice = xfrm_negative_advice;
2854                 if (likely(dst_ops->link_failure == NULL))
2855                         dst_ops->link_failure = xfrm_link_failure;
2856                 if (likely(dst_ops->neigh_lookup == NULL))
2857                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2858                 if (likely(!dst_ops->confirm_neigh))
2859                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
2860                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2861         }
2862         spin_unlock(&xfrm_policy_afinfo_lock);
2863
2864         return err;
2865 }
2866 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2867
2868 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2869 {
2870         struct dst_ops *dst_ops = afinfo->dst_ops;
2871         int i;
2872
2873         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2874                 if (xfrm_policy_afinfo[i] != afinfo)
2875                         continue;
2876                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2877                 break;
2878         }
2879
2880         synchronize_rcu();
2881
2882         dst_ops->kmem_cachep = NULL;
2883         dst_ops->check = NULL;
2884         dst_ops->negative_advice = NULL;
2885         dst_ops->link_failure = NULL;
2886 }
2887 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2888
2889 #ifdef CONFIG_XFRM_STATISTICS
2890 static int __net_init xfrm_statistics_init(struct net *net)
2891 {
2892         int rv;
2893         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2894         if (!net->mib.xfrm_statistics)
2895                 return -ENOMEM;
2896         rv = xfrm_proc_init(net);
2897         if (rv < 0)
2898                 free_percpu(net->mib.xfrm_statistics);
2899         return rv;
2900 }
2901
2902 static void xfrm_statistics_fini(struct net *net)
2903 {
2904         xfrm_proc_fini(net);
2905         free_percpu(net->mib.xfrm_statistics);
2906 }
2907 #else
2908 static int __net_init xfrm_statistics_init(struct net *net)
2909 {
2910         return 0;
2911 }
2912
2913 static void xfrm_statistics_fini(struct net *net)
2914 {
2915 }
2916 #endif
2917
2918 static int __net_init xfrm_policy_init(struct net *net)
2919 {
2920         unsigned int hmask, sz;
2921         int dir;
2922
2923         if (net_eq(net, &init_net))
2924                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2925                                            sizeof(struct xfrm_dst),
2926                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2927                                            NULL);
2928
2929         hmask = 8 - 1;
2930         sz = (hmask+1) * sizeof(struct hlist_head);
2931
2932         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2933         if (!net->xfrm.policy_byidx)
2934                 goto out_byidx;
2935         net->xfrm.policy_idx_hmask = hmask;
2936
2937         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2938                 struct xfrm_policy_hash *htab;
2939
2940                 net->xfrm.policy_count[dir] = 0;
2941                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2942                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2943
2944                 htab = &net->xfrm.policy_bydst[dir];
2945                 htab->table = xfrm_hash_alloc(sz);
2946                 if (!htab->table)
2947                         goto out_bydst;
2948                 htab->hmask = hmask;
2949                 htab->dbits4 = 32;
2950                 htab->sbits4 = 32;
2951                 htab->dbits6 = 128;
2952                 htab->sbits6 = 128;
2953         }
2954         net->xfrm.policy_hthresh.lbits4 = 32;
2955         net->xfrm.policy_hthresh.rbits4 = 32;
2956         net->xfrm.policy_hthresh.lbits6 = 128;
2957         net->xfrm.policy_hthresh.rbits6 = 128;
2958
2959         seqlock_init(&net->xfrm.policy_hthresh.lock);
2960
2961         INIT_LIST_HEAD(&net->xfrm.policy_all);
2962         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2963         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2964         if (net_eq(net, &init_net))
2965                 xfrm_dev_init();
2966         return 0;
2967
2968 out_bydst:
2969         for (dir--; dir >= 0; dir--) {
2970                 struct xfrm_policy_hash *htab;
2971
2972                 htab = &net->xfrm.policy_bydst[dir];
2973                 xfrm_hash_free(htab->table, sz);
2974         }
2975         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2976 out_byidx:
2977         return -ENOMEM;
2978 }
2979
2980 static void xfrm_policy_fini(struct net *net)
2981 {
2982         unsigned int sz;
2983         int dir;
2984
2985         flush_work(&net->xfrm.policy_hash_work);
2986 #ifdef CONFIG_XFRM_SUB_POLICY
2987         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2988 #endif
2989         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2990
2991         WARN_ON(!list_empty(&net->xfrm.policy_all));
2992
2993         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2994                 struct xfrm_policy_hash *htab;
2995
2996                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2997
2998                 htab = &net->xfrm.policy_bydst[dir];
2999                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3000                 WARN_ON(!hlist_empty(htab->table));
3001                 xfrm_hash_free(htab->table, sz);
3002         }
3003
3004         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3005         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3006         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3007 }
3008
3009 static int __net_init xfrm_net_init(struct net *net)
3010 {
3011         int rv;
3012
3013         /* Initialize the per-net locks here */
3014         spin_lock_init(&net->xfrm.xfrm_state_lock);
3015         spin_lock_init(&net->xfrm.xfrm_policy_lock);
3016         mutex_init(&net->xfrm.xfrm_cfg_mutex);
3017
3018         rv = xfrm_statistics_init(net);
3019         if (rv < 0)
3020                 goto out_statistics;
3021         rv = xfrm_state_init(net);
3022         if (rv < 0)
3023                 goto out_state;
3024         rv = xfrm_policy_init(net);
3025         if (rv < 0)
3026                 goto out_policy;
3027         rv = xfrm_sysctl_init(net);
3028         if (rv < 0)
3029                 goto out_sysctl;
3030         rv = flow_cache_init(net);
3031         if (rv < 0)
3032                 goto out;
3033
3034         return 0;
3035
3036 out:
3037         xfrm_sysctl_fini(net);
3038 out_sysctl:
3039         xfrm_policy_fini(net);
3040 out_policy:
3041         xfrm_state_fini(net);
3042 out_state:
3043         xfrm_statistics_fini(net);
3044 out_statistics:
3045         return rv;
3046 }
3047
3048 static void __net_exit xfrm_net_exit(struct net *net)
3049 {
3050         flow_cache_fini(net);
3051         xfrm_sysctl_fini(net);
3052         xfrm_policy_fini(net);
3053         xfrm_state_fini(net);
3054         xfrm_statistics_fini(net);
3055 }
3056
3057 static struct pernet_operations __net_initdata xfrm_net_ops = {
3058         .init = xfrm_net_init,
3059         .exit = xfrm_net_exit,
3060 };
3061
3062 void __init xfrm_init(void)
3063 {
3064         flow_cache_hp_init();
3065         register_pernet_subsys(&xfrm_net_ops);
3066         seqcount_init(&xfrm_policy_hash_generation);
3067         xfrm_input_init();
3068 }
3069
3070 #ifdef CONFIG_AUDITSYSCALL
3071 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3072                                          struct audit_buffer *audit_buf)
3073 {
3074         struct xfrm_sec_ctx *ctx = xp->security;
3075         struct xfrm_selector *sel = &xp->selector;
3076
3077         if (ctx)
3078                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3079                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3080
3081         switch (sel->family) {
3082         case AF_INET:
3083                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3084                 if (sel->prefixlen_s != 32)
3085                         audit_log_format(audit_buf, " src_prefixlen=%d",
3086                                          sel->prefixlen_s);
3087                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3088                 if (sel->prefixlen_d != 32)
3089                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3090                                          sel->prefixlen_d);
3091                 break;
3092         case AF_INET6:
3093                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3094                 if (sel->prefixlen_s != 128)
3095                         audit_log_format(audit_buf, " src_prefixlen=%d",
3096                                          sel->prefixlen_s);
3097                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3098                 if (sel->prefixlen_d != 128)
3099                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3100                                          sel->prefixlen_d);
3101                 break;
3102         }
3103 }
3104
3105 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3106 {
3107         struct audit_buffer *audit_buf;
3108
3109         audit_buf = xfrm_audit_start("SPD-add");
3110         if (audit_buf == NULL)
3111                 return;
3112         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3113         audit_log_format(audit_buf, " res=%u", result);
3114         xfrm_audit_common_policyinfo(xp, audit_buf);
3115         audit_log_end(audit_buf);
3116 }
3117 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3118
3119 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3120                               bool task_valid)
3121 {
3122         struct audit_buffer *audit_buf;
3123
3124         audit_buf = xfrm_audit_start("SPD-delete");
3125         if (audit_buf == NULL)
3126                 return;
3127         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3128         audit_log_format(audit_buf, " res=%u", result);
3129         xfrm_audit_common_policyinfo(xp, audit_buf);
3130         audit_log_end(audit_buf);
3131 }
3132 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3133 #endif
3134
3135 #ifdef CONFIG_XFRM_MIGRATE
3136 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3137                                         const struct xfrm_selector *sel_tgt)
3138 {
3139         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3140                 if (sel_tgt->family == sel_cmp->family &&
3141                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3142                                     sel_cmp->family) &&
3143                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3144                                     sel_cmp->family) &&
3145                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3146                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3147                         return true;
3148                 }
3149         } else {
3150                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3151                         return true;
3152                 }
3153         }
3154         return false;
3155 }
3156
3157 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3158                                                     u8 dir, u8 type, struct net *net)
3159 {
3160         struct xfrm_policy *pol, *ret = NULL;
3161         struct hlist_head *chain;
3162         u32 priority = ~0U;
3163
3164         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3165         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3166         hlist_for_each_entry(pol, chain, bydst) {
3167                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3168                     pol->type == type) {
3169                         ret = pol;
3170                         priority = ret->priority;
3171                         break;
3172                 }
3173         }
3174         chain = &net->xfrm.policy_inexact[dir];
3175         hlist_for_each_entry(pol, chain, bydst) {
3176                 if ((pol->priority >= priority) && ret)
3177                         break;
3178
3179                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3180                     pol->type == type) {
3181                         ret = pol;
3182                         break;
3183                 }
3184         }
3185
3186         xfrm_pol_hold(ret);
3187
3188         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3189
3190         return ret;
3191 }
3192
3193 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3194 {
3195         int match = 0;
3196
3197         if (t->mode == m->mode && t->id.proto == m->proto &&
3198             (m->reqid == 0 || t->reqid == m->reqid)) {
3199                 switch (t->mode) {
3200                 case XFRM_MODE_TUNNEL:
3201                 case XFRM_MODE_BEET:
3202                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3203                                             m->old_family) &&
3204                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3205                                             m->old_family)) {
3206                                 match = 1;
3207                         }
3208                         break;
3209                 case XFRM_MODE_TRANSPORT:
3210                         /* in case of transport mode, template does not store
3211                            any IP addresses, hence we just compare mode and
3212                            protocol */
3213                         match = 1;
3214                         break;
3215                 default:
3216                         break;
3217                 }
3218         }
3219         return match;
3220 }
3221
3222 /* update endpoint address(es) of template(s) */
3223 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3224                                struct xfrm_migrate *m, int num_migrate)
3225 {
3226         struct xfrm_migrate *mp;
3227         int i, j, n = 0;
3228
3229         write_lock_bh(&pol->lock);
3230         if (unlikely(pol->walk.dead)) {
3231                 /* target policy has been deleted */
3232                 write_unlock_bh(&pol->lock);
3233                 return -ENOENT;
3234         }
3235
3236         for (i = 0; i < pol->xfrm_nr; i++) {
3237                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3238                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3239                                 continue;
3240                         n++;
3241                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3242                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3243                                 continue;
3244                         /* update endpoints */
3245                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3246                                sizeof(pol->xfrm_vec[i].id.daddr));
3247                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3248                                sizeof(pol->xfrm_vec[i].saddr));
3249                         pol->xfrm_vec[i].encap_family = mp->new_family;
3250                         /* flush bundles */
3251                         atomic_inc(&pol->genid);
3252                 }
3253         }
3254
3255         write_unlock_bh(&pol->lock);
3256
3257         if (!n)
3258                 return -ENODATA;
3259
3260         return 0;
3261 }
3262
3263 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3264 {
3265         int i, j;
3266
3267         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3268                 return -EINVAL;
3269
3270         for (i = 0; i < num_migrate; i++) {
3271                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3272                                     m[i].old_family) &&
3273                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3274                                     m[i].old_family))
3275                         return -EINVAL;
3276                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3277                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3278                         return -EINVAL;
3279
3280                 /* check if there is any duplicated entry */
3281                 for (j = i + 1; j < num_migrate; j++) {
3282                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3283                                     sizeof(m[i].old_daddr)) &&
3284                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3285                                     sizeof(m[i].old_saddr)) &&
3286                             m[i].proto == m[j].proto &&
3287                             m[i].mode == m[j].mode &&
3288                             m[i].reqid == m[j].reqid &&
3289                             m[i].old_family == m[j].old_family)
3290                                 return -EINVAL;
3291                 }
3292         }
3293
3294         return 0;
3295 }
3296
3297 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3298                  struct xfrm_migrate *m, int num_migrate,
3299                  struct xfrm_kmaddress *k, struct net *net)
3300 {
3301         int i, err, nx_cur = 0, nx_new = 0;
3302         struct xfrm_policy *pol = NULL;
3303         struct xfrm_state *x, *xc;
3304         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3305         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3306         struct xfrm_migrate *mp;
3307
3308         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3309                 goto out;
3310
3311         /* Stage 1 - find policy */
3312         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3313                 err = -ENOENT;
3314                 goto out;
3315         }
3316
3317         /* Stage 2 - find and update state(s) */
3318         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3319                 if ((x = xfrm_migrate_state_find(mp, net))) {
3320                         x_cur[nx_cur] = x;
3321                         nx_cur++;
3322                         if ((xc = xfrm_state_migrate(x, mp))) {
3323                                 x_new[nx_new] = xc;
3324                                 nx_new++;
3325                         } else {
3326                                 err = -ENODATA;
3327                                 goto restore_state;
3328                         }
3329                 }
3330         }
3331
3332         /* Stage 3 - update policy */
3333         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3334                 goto restore_state;
3335
3336         /* Stage 4 - delete old state(s) */
3337         if (nx_cur) {
3338                 xfrm_states_put(x_cur, nx_cur);
3339                 xfrm_states_delete(x_cur, nx_cur);
3340         }
3341
3342         /* Stage 5 - announce */
3343         km_migrate(sel, dir, type, m, num_migrate, k);
3344
3345         xfrm_pol_put(pol);
3346
3347         return 0;
3348 out:
3349         return err;
3350
3351 restore_state:
3352         if (pol)
3353                 xfrm_pol_put(pol);
3354         if (nx_cur)
3355                 xfrm_states_put(x_cur, nx_cur);
3356         if (nx_new)
3357                 xfrm_states_delete(x_new, nx_new);
3358
3359         return err;
3360 }
3361 EXPORT_SYMBOL(xfrm_migrate);
3362 #endif