err.no Git - linux-2.6/blob - 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/config.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 <net/xfrm.h>
  27 #include <net/ip.h>
  28
  29 DEFINE_MUTEX(xfrm_cfg_mutex);
  30 EXPORT_SYMBOL(xfrm_cfg_mutex);
  31
  32 static DEFINE_RWLOCK(xfrm_policy_lock);
  33
  34 struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2];
  35 EXPORT_SYMBOL(xfrm_policy_list);
  36
  37 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  38 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  39
  40 static kmem_cache_t *xfrm_dst_cache __read_mostly;
  41
  42 static struct work_struct xfrm_policy_gc_work;
  43 static struct list_head xfrm_policy_gc_list =
  44         LIST_HEAD_INIT(xfrm_policy_gc_list);
  45 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  46
  47 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  48 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  49 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
  50 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
  51
  52 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
  53 {
  54         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  55         struct xfrm_type **typemap;
  56         int err = 0;
  57
  58         if (unlikely(afinfo == NULL))
  59                 return -EAFNOSUPPORT;
  60         typemap = afinfo->type_map;
  61
  62         if (likely(typemap[type->proto] == NULL))
  63                 typemap[type->proto] = type;
  64         else
  65                 err = -EEXIST;
  66         xfrm_policy_unlock_afinfo(afinfo);
  67         return err;
  68 }
  69 EXPORT_SYMBOL(xfrm_register_type);
  70
  71 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
  72 {
  73         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  74         struct xfrm_type **typemap;
  75         int err = 0;
  76
  77         if (unlikely(afinfo == NULL))
  78                 return -EAFNOSUPPORT;
  79         typemap = afinfo->type_map;
  80
  81         if (unlikely(typemap[type->proto] != type))
  82                 err = -ENOENT;
  83         else
  84                 typemap[type->proto] = NULL;
  85         xfrm_policy_unlock_afinfo(afinfo);
  86         return err;
  87 }
  88 EXPORT_SYMBOL(xfrm_unregister_type);
  89
  90 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
  91 {
  92         struct xfrm_policy_afinfo *afinfo;
  93         struct xfrm_type **typemap;
  94         struct xfrm_type *type;
  95         int modload_attempted = 0;
  96
  97 retry:
  98         afinfo = xfrm_policy_get_afinfo(family);
  99         if (unlikely(afinfo == NULL))
 100                 return NULL;
 101         typemap = afinfo->type_map;
 102
 103         type = typemap[proto];
 104         if (unlikely(type && !try_module_get(type->owner)))
 105                 type = NULL;
 106         if (!type && !modload_attempted) {
 107                 xfrm_policy_put_afinfo(afinfo);
 108                 request_module("xfrm-type-%d-%d",
 109                                (int) family, (int) proto);
 110                 modload_attempted = 1;
 111                 goto retry;
 112         }
 113
 114         xfrm_policy_put_afinfo(afinfo);
 115         return type;
 116 }
 117
 118 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
 119                     unsigned short family)
 120 {
 121         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 122         int err = 0;
 123
 124         if (unlikely(afinfo == NULL))
 125                 return -EAFNOSUPPORT;
 126
 127         if (likely(afinfo->dst_lookup != NULL))
 128                 err = afinfo->dst_lookup(dst, fl);
 129         else
 130                 err = -EINVAL;
 131         xfrm_policy_put_afinfo(afinfo);
 132         return err;
 133 }
 134 EXPORT_SYMBOL(xfrm_dst_lookup);
 135
 136 void xfrm_put_type(struct xfrm_type *type)
 137 {
 138         module_put(type->owner);
 139 }
 140
 141 static inline unsigned long make_jiffies(long secs)
 142 {
 143         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 144                 return MAX_SCHEDULE_TIMEOUT-1;
 145         else
 146                 return secs*HZ;
 147 }
 148
 149 static void xfrm_policy_timer(unsigned long data)
 150 {
 151         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 152         unsigned long now = (unsigned long)xtime.tv_sec;
 153         long next = LONG_MAX;
 154         int warn = 0;
 155         int dir;
 156
 157         read_lock(&xp->lock);
 158
 159         if (xp->dead)
 160                 goto out;
 161
 162         dir = xfrm_policy_id2dir(xp->index);
 163
 164         if (xp->lft.hard_add_expires_seconds) {
 165                 long tmo = xp->lft.hard_add_expires_seconds +
 166                         xp->curlft.add_time - now;
 167                 if (tmo <= 0)
 168                         goto expired;
 169                 if (tmo < next)
 170                         next = tmo;
 171         }
 172         if (xp->lft.hard_use_expires_seconds) {
 173                 long tmo = xp->lft.hard_use_expires_seconds +
 174                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 175                 if (tmo <= 0)
 176                         goto expired;
 177                 if (tmo < next)
 178                         next = tmo;
 179         }
 180         if (xp->lft.soft_add_expires_seconds) {
 181                 long tmo = xp->lft.soft_add_expires_seconds +
 182                         xp->curlft.add_time - now;
 183                 if (tmo <= 0) {
 184                         warn = 1;
 185                         tmo = XFRM_KM_TIMEOUT;
 186                 }
 187                 if (tmo < next)
 188                         next = tmo;
 189         }
 190         if (xp->lft.soft_use_expires_seconds) {
 191                 long tmo = xp->lft.soft_use_expires_seconds +
 192                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 193                 if (tmo <= 0) {
 194                         warn = 1;
 195                         tmo = XFRM_KM_TIMEOUT;
 196                 }
 197                 if (tmo < next)
 198                         next = tmo;
 199         }
 200
 201         if (warn)
 202                 km_policy_expired(xp, dir, 0, 0);
 203         if (next != LONG_MAX &&
 204             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 205                 xfrm_pol_hold(xp);
 206
 207 out:
 208         read_unlock(&xp->lock);
 209         xfrm_pol_put(xp);
 210         return;
 211
 212 expired:
 213         read_unlock(&xp->lock);
 214         if (!xfrm_policy_delete(xp, dir))
 215                 km_policy_expired(xp, dir, 1, 0);
 216         xfrm_pol_put(xp);
 217 }
 218
 219
 220 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 221  * SPD calls.
 222  */
 223
 224 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 225 {
 226         struct xfrm_policy *policy;
 227
 228         policy = kmalloc(sizeof(struct xfrm_policy), gfp);
 229
 230         if (policy) {
 231                 memset(policy, 0, sizeof(struct xfrm_policy));
 232                 atomic_set(&policy->refcnt, 1);
 233                 rwlock_init(&policy->lock);
 234                 init_timer(&policy->timer);
 235                 policy->timer.data = (unsigned long)policy;
 236                 policy->timer.function = xfrm_policy_timer;
 237         }
 238         return policy;
 239 }
 240 EXPORT_SYMBOL(xfrm_policy_alloc);
 241
 242 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 243
 244 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 245 {
 246         BUG_ON(!policy->dead);
 247
 248         BUG_ON(policy->bundles);
 249
 250         if (del_timer(&policy->timer))
 251                 BUG();
 252
 253         security_xfrm_policy_free(policy);
 254         kfree(policy);
 255 }
 256 EXPORT_SYMBOL(__xfrm_policy_destroy);
 257
 258 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 259 {
 260         struct dst_entry *dst;
 261
 262         while ((dst = policy->bundles) != NULL) {
 263                 policy->bundles = dst->next;
 264                 dst_free(dst);
 265         }
 266
 267         if (del_timer(&policy->timer))
 268                 atomic_dec(&policy->refcnt);
 269
 270         if (atomic_read(&policy->refcnt) > 1)
 271                 flow_cache_flush();
 272
 273         xfrm_pol_put(policy);
 274 }
 275
 276 static void xfrm_policy_gc_task(void *data)
 277 {
 278         struct xfrm_policy *policy;
 279         struct list_head *entry, *tmp;
 280         struct list_head gc_list = LIST_HEAD_INIT(gc_list);
 281
 282         spin_lock_bh(&xfrm_policy_gc_lock);
 283         list_splice_init(&xfrm_policy_gc_list, &gc_list);
 284         spin_unlock_bh(&xfrm_policy_gc_lock);
 285
 286         list_for_each_safe(entry, tmp, &gc_list) {
 287                 policy = list_entry(entry, struct xfrm_policy, list);
 288                 xfrm_policy_gc_kill(policy);
 289         }
 290 }
 291
 292 /* Rule must be locked. Release descentant resources, announce
 293  * entry dead. The rule must be unlinked from lists to the moment.
 294  */
 295
 296 static void xfrm_policy_kill(struct xfrm_policy *policy)
 297 {
 298         int dead;
 299
 300         write_lock_bh(&policy->lock);
 301         dead = policy->dead;
 302         policy->dead = 1;
 303         write_unlock_bh(&policy->lock);
 304
 305         if (unlikely(dead)) {
 306                 WARN_ON(1);
 307                 return;
 308         }
 309
 310         spin_lock(&xfrm_policy_gc_lock);
 311         list_add(&policy->list, &xfrm_policy_gc_list);
 312         spin_unlock(&xfrm_policy_gc_lock);
 313
 314         schedule_work(&xfrm_policy_gc_work);
 315 }
 316
 317 /* Generate new index... KAME seems to generate them ordered by cost
 318  * of an absolute inpredictability of ordering of rules. This will not pass. */
 319 static u32 xfrm_gen_index(int dir)
 320 {
 321         u32 idx;
 322         struct xfrm_policy *p;
 323         static u32 idx_generator;
 324
 325         for (;;) {
 326                 idx = (idx_generator | dir);
 327                 idx_generator += 8;
 328                 if (idx == 0)
 329                         idx = 8;
 330                 for (p = xfrm_policy_list[dir]; p; p = p->next) {
 331                         if (p->index == idx)
 332                                 break;
 333                 }
 334                 if (!p)
 335                         return idx;
 336         }
 337 }
 338
 339 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 340 {
 341         struct xfrm_policy *pol, **p;
 342         struct xfrm_policy *delpol = NULL;
 343         struct xfrm_policy **newpos = NULL;
 344         struct dst_entry *gc_list;
 345
 346         write_lock_bh(&xfrm_policy_lock);
 347         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
 348                 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0 &&
 349                     xfrm_sec_ctx_match(pol->security, policy->security)) {
 350                         if (excl) {
 351                                 write_unlock_bh(&xfrm_policy_lock);
 352                                 return -EEXIST;
 353                         }
 354                         *p = pol->next;
 355                         delpol = pol;
 356                         if (policy->priority > pol->priority)
 357                                 continue;
 358                 } else if (policy->priority >= pol->priority) {
 359                         p = &pol->next;
 360                         continue;
 361                 }
 362                 if (!newpos)
 363                         newpos = p;
 364                 if (delpol)
 365                         break;
 366                 p = &pol->next;
 367         }
 368         if (newpos)
 369                 p = newpos;
 370         xfrm_pol_hold(policy);
 371         policy->next = *p;
 372         *p = policy;
 373         atomic_inc(&flow_cache_genid);
 374         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
 375         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
 376         policy->curlft.use_time = 0;
 377         if (!mod_timer(&policy->timer, jiffies + HZ))
 378                 xfrm_pol_hold(policy);
 379         write_unlock_bh(&xfrm_policy_lock);
 380
 381         if (delpol)
 382                 xfrm_policy_kill(delpol);
 383
 384         read_lock_bh(&xfrm_policy_lock);
 385         gc_list = NULL;
 386         for (policy = policy->next; policy; policy = policy->next) {
 387                 struct dst_entry *dst;
 388
 389                 write_lock(&policy->lock);
 390                 dst = policy->bundles;
 391                 if (dst) {
 392                         struct dst_entry *tail = dst;
 393                         while (tail->next)
 394                                 tail = tail->next;
 395                         tail->next = gc_list;
 396                         gc_list = dst;
 397
 398                         policy->bundles = NULL;
 399                 }
 400                 write_unlock(&policy->lock);
 401         }
 402         read_unlock_bh(&xfrm_policy_lock);
 403
 404         while (gc_list) {
 405                 struct dst_entry *dst = gc_list;
 406
 407                 gc_list = dst->next;
 408                 dst_free(dst);
 409         }
 410
 411         return 0;
 412 }
 413 EXPORT_SYMBOL(xfrm_policy_insert);
 414
 415 struct xfrm_policy *xfrm_policy_bysel_ctx(int dir, struct xfrm_selector *sel,
 416                                           struct xfrm_sec_ctx *ctx, int delete)
 417 {
 418         struct xfrm_policy *pol, **p;
 419
 420         write_lock_bh(&xfrm_policy_lock);
 421         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 422                 if ((memcmp(sel, &pol->selector, sizeof(*sel)) == 0) &&
 423                     (xfrm_sec_ctx_match(ctx, pol->security))) {
 424                         xfrm_pol_hold(pol);
 425                         if (delete)
 426                                 *p = pol->next;
 427                         break;
 428                 }
 429         }
 430         write_unlock_bh(&xfrm_policy_lock);
 431
 432         if (pol && delete) {
 433                 atomic_inc(&flow_cache_genid);
 434                 xfrm_policy_kill(pol);
 435         }
 436         return pol;
 437 }
 438 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 439
 440 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
 441 {
 442         struct xfrm_policy *pol, **p;
 443
 444         write_lock_bh(&xfrm_policy_lock);
 445         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 446                 if (pol->index == id) {
 447                         xfrm_pol_hold(pol);
 448                         if (delete)
 449                                 *p = pol->next;
 450                         break;
 451                 }
 452         }
 453         write_unlock_bh(&xfrm_policy_lock);
 454
 455         if (pol && delete) {
 456                 atomic_inc(&flow_cache_genid);
 457                 xfrm_policy_kill(pol);
 458         }
 459         return pol;
 460 }
 461 EXPORT_SYMBOL(xfrm_policy_byid);
 462
 463 void xfrm_policy_flush(void)
 464 {
 465         struct xfrm_policy *xp;
 466         int dir;
 467
 468         write_lock_bh(&xfrm_policy_lock);
 469         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 470                 while ((xp = xfrm_policy_list[dir]) != NULL) {
 471                         xfrm_policy_list[dir] = xp->next;
 472                         write_unlock_bh(&xfrm_policy_lock);
 473
 474                         xfrm_policy_kill(xp);
 475
 476                         write_lock_bh(&xfrm_policy_lock);
 477                 }
 478         }
 479         atomic_inc(&flow_cache_genid);
 480         write_unlock_bh(&xfrm_policy_lock);
 481 }
 482 EXPORT_SYMBOL(xfrm_policy_flush);
 483
 484 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
 485                      void *data)
 486 {
 487         struct xfrm_policy *xp;
 488         int dir;
 489         int count = 0;
 490         int error = 0;
 491
 492         read_lock_bh(&xfrm_policy_lock);
 493         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 494                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
 495                         count++;
 496         }
 497
 498         if (count == 0) {
 499                 error = -ENOENT;
 500                 goto out;
 501         }
 502
 503         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 504                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
 505                         error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
 506                         if (error)
 507                                 goto out;
 508                 }
 509         }
 510
 511 out:
 512         read_unlock_bh(&xfrm_policy_lock);
 513         return error;
 514 }
 515 EXPORT_SYMBOL(xfrm_policy_walk);
 516
 517 /* Find policy to apply to this flow. */
 518
 519 static void xfrm_policy_lookup(struct flowi *fl, u32 sk_sid, u16 family, u8 dir,
 520                                void **objp, atomic_t **obj_refp)
 521 {
 522         struct xfrm_policy *pol;
 523
 524         read_lock_bh(&xfrm_policy_lock);
 525         for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
 526                 struct xfrm_selector *sel = &pol->selector;
 527                 int match;
 528
 529                 if (pol->family != family)
 530                         continue;
 531
 532                 match = xfrm_selector_match(sel, fl, family);
 533
 534                 if (match) {
 535                         if (!security_xfrm_policy_lookup(pol, sk_sid, dir)) {
 536                                 xfrm_pol_hold(pol);
 537                                 break;
 538                         }
 539                 }
 540         }
 541         read_unlock_bh(&xfrm_policy_lock);
 542         if ((*objp = (void *) pol) != NULL)
 543                 *obj_refp = &pol->refcnt;
 544 }
 545
 546 static inline int policy_to_flow_dir(int dir)
 547 {
 548         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 549             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 550             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 551                 return dir;
 552         switch (dir) {
 553         default:
 554         case XFRM_POLICY_IN:
 555                 return FLOW_DIR_IN;
 556         case XFRM_POLICY_OUT:
 557                 return FLOW_DIR_OUT;
 558         case XFRM_POLICY_FWD:
 559                 return FLOW_DIR_FWD;
 560         };
 561 }
 562
 563 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl, u32 sk_sid)
 564 {
 565         struct xfrm_policy *pol;
 566
 567         read_lock_bh(&xfrm_policy_lock);
 568         if ((pol = sk->sk_policy[dir]) != NULL) {
 569                 int match = xfrm_selector_match(&pol->selector, fl,
 570                                                 sk->sk_family);
 571                 int err = 0;
 572
 573                 if (match)
 574                   err = security_xfrm_policy_lookup(pol, sk_sid, policy_to_flow_dir(dir));
 575
 576                 if (match && !err)
 577                         xfrm_pol_hold(pol);
 578                 else
 579                         pol = NULL;
 580         }
 581         read_unlock_bh(&xfrm_policy_lock);
 582         return pol;
 583 }
 584
 585 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
 586 {
 587         pol->next = xfrm_policy_list[dir];
 588         xfrm_policy_list[dir] = pol;
 589         xfrm_pol_hold(pol);
 590 }
 591
 592 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 593                                                 int dir)
 594 {
 595         struct xfrm_policy **polp;
 596
 597         for (polp = &xfrm_policy_list[dir];
 598              *polp != NULL; polp = &(*polp)->next) {
 599                 if (*polp == pol) {
 600                         *polp = pol->next;
 601                         return pol;
 602                 }
 603         }
 604         return NULL;
 605 }
 606
 607 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
 608 {
 609         write_lock_bh(&xfrm_policy_lock);
 610         pol = __xfrm_policy_unlink(pol, dir);
 611         write_unlock_bh(&xfrm_policy_lock);
 612         if (pol) {
 613                 if (dir < XFRM_POLICY_MAX)
 614                         atomic_inc(&flow_cache_genid);
 615                 xfrm_policy_kill(pol);
 616                 return 0;
 617         }
 618         return -ENOENT;
 619 }
 620 EXPORT_SYMBOL(xfrm_policy_delete);
 621
 622 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
 623 {
 624         struct xfrm_policy *old_pol;
 625
 626         write_lock_bh(&xfrm_policy_lock);
 627         old_pol = sk->sk_policy[dir];
 628         sk->sk_policy[dir] = pol;
 629         if (pol) {
 630                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
 631                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
 632                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
 633         }
 634         if (old_pol)
 635                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
 636         write_unlock_bh(&xfrm_policy_lock);
 637
 638         if (old_pol) {
 639                 xfrm_policy_kill(old_pol);
 640         }
 641         return 0;
 642 }
 643
 644 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
 645 {
 646         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
 647
 648         if (newp) {
 649                 newp->selector = old->selector;
 650                 if (security_xfrm_policy_clone(old, newp)) {
 651                         kfree(newp);
 652                         return NULL;  /* ENOMEM */
 653                 }
 654                 newp->lft = old->lft;
 655                 newp->curlft = old->curlft;
 656                 newp->action = old->action;
 657                 newp->flags = old->flags;
 658                 newp->xfrm_nr = old->xfrm_nr;
 659                 newp->index = old->index;
 660                 memcpy(newp->xfrm_vec, old->xfrm_vec,
 661                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
 662                 write_lock_bh(&xfrm_policy_lock);
 663                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
 664                 write_unlock_bh(&xfrm_policy_lock);
 665                 xfrm_pol_put(newp);
 666         }
 667         return newp;
 668 }
 669
 670 int __xfrm_sk_clone_policy(struct sock *sk)
 671 {
 672         struct xfrm_policy *p0 = sk->sk_policy[0],
 673                            *p1 = sk->sk_policy[1];
 674
 675         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
 676         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
 677                 return -ENOMEM;
 678         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
 679                 return -ENOMEM;
 680         return 0;
 681 }
 682
 683 /* Resolve list of templates for the flow, given policy. */
 684
 685 static int
 686 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
 687                   struct xfrm_state **xfrm,
 688                   unsigned short family)
 689 {
 690         int nx;
 691         int i, error;
 692         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
 693         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
 694
 695         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
 696                 struct xfrm_state *x;
 697                 xfrm_address_t *remote = daddr;
 698                 xfrm_address_t *local  = saddr;
 699                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
 700
 701                 if (tmpl->mode) {
 702                         remote = &tmpl->id.daddr;
 703                         local = &tmpl->saddr;
 704                 }
 705
 706                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
 707
 708                 if (x && x->km.state == XFRM_STATE_VALID) {
 709                         xfrm[nx++] = x;
 710                         daddr = remote;
 711                         saddr = local;
 712                         continue;
 713                 }
 714                 if (x) {
 715                         error = (x->km.state == XFRM_STATE_ERROR ?
 716                                  -EINVAL : -EAGAIN);
 717                         xfrm_state_put(x);
 718                 }
 719
 720                 if (!tmpl->optional)
 721                         goto fail;
 722         }
 723         return nx;
 724
 725 fail:
 726         for (nx--; nx>=0; nx--)
 727                 xfrm_state_put(xfrm[nx]);
 728         return error;
 729 }
 730
 731 /* Check that the bundle accepts the flow and its components are
 732  * still valid.
 733  */
 734
 735 static struct dst_entry *
 736 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
 737 {
 738         struct dst_entry *x;
 739         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 740         if (unlikely(afinfo == NULL))
 741                 return ERR_PTR(-EINVAL);
 742         x = afinfo->find_bundle(fl, policy);
 743         xfrm_policy_put_afinfo(afinfo);
 744         return x;
 745 }
 746
 747 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
 748  * all the metrics... Shortly, bundle a bundle.
 749  */
 750
 751 static int
 752 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
 753                    struct flowi *fl, struct dst_entry **dst_p,
 754                    unsigned short family)
 755 {
 756         int err;
 757         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 758         if (unlikely(afinfo == NULL))
 759                 return -EINVAL;
 760         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
 761         xfrm_policy_put_afinfo(afinfo);
 762         return err;
 763 }
 764
 765
 766 static int stale_bundle(struct dst_entry *dst);
 767
 768 /* Main function: finds/creates a bundle for given flow.
 769  *
 770  * At the moment we eat a raw IP route. Mostly to speed up lookups
 771  * on interfaces with disabled IPsec.
 772  */
 773 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
 774                 struct sock *sk, int flags)
 775 {
 776         struct xfrm_policy *policy;
 777         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
 778         struct dst_entry *dst, *dst_orig = *dst_p;
 779         int nx = 0;
 780         int err;
 781         u32 genid;
 782         u16 family;
 783         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
 784         u32 sk_sid = security_sk_sid(sk, fl, dir);
 785 restart:
 786         genid = atomic_read(&flow_cache_genid);
 787         policy = NULL;
 788         if (sk && sk->sk_policy[1])
 789                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, sk_sid);
 790
 791         if (!policy) {
 792                 /* To accelerate a bit...  */
 793                 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
 794                         return 0;
 795
 796                 policy = flow_cache_lookup(fl, sk_sid, dst_orig->ops->family,
 797                                            dir, xfrm_policy_lookup);
 798         }
 799
 800         if (!policy)
 801                 return 0;
 802
 803         family = dst_orig->ops->family;
 804         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
 805
 806         switch (policy->action) {
 807         case XFRM_POLICY_BLOCK:
 808                 /* Prohibit the flow */
 809                 err = -EPERM;
 810                 goto error;
 811
 812         case XFRM_POLICY_ALLOW:
 813                 if (policy->xfrm_nr == 0) {
 814                         /* Flow passes not transformed. */
 815                         xfrm_pol_put(policy);
 816                         return 0;
 817                 }
 818
 819                 /* Try to find matching bundle.
 820                  *
 821                  * LATER: help from flow cache. It is optional, this
 822                  * is required only for output policy.
 823                  */
 824                 dst = xfrm_find_bundle(fl, policy, family);
 825                 if (IS_ERR(dst)) {
 826                         err = PTR_ERR(dst);
 827                         goto error;
 828                 }
 829
 830                 if (dst)
 831                         break;
 832
 833                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 834
 835                 if (unlikely(nx<0)) {
 836                         err = nx;
 837                         if (err == -EAGAIN && flags) {
 838                                 DECLARE_WAITQUEUE(wait, current);
 839
 840                                 add_wait_queue(&km_waitq, &wait);
 841                                 set_current_state(TASK_INTERRUPTIBLE);
 842                                 schedule();
 843                                 set_current_state(TASK_RUNNING);
 844                                 remove_wait_queue(&km_waitq, &wait);
 845
 846                                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 847
 848                                 if (nx == -EAGAIN && signal_pending(current)) {
 849                                         err = -ERESTART;
 850                                         goto error;
 851                                 }
 852                                 if (nx == -EAGAIN ||
 853                                     genid != atomic_read(&flow_cache_genid)) {
 854                                         xfrm_pol_put(policy);
 855                                         goto restart;
 856                                 }
 857                                 err = nx;
 858                         }
 859                         if (err < 0)
 860                                 goto error;
 861                 }
 862                 if (nx == 0) {
 863                         /* Flow passes not transformed. */
 864                         xfrm_pol_put(policy);
 865                         return 0;
 866                 }
 867
 868                 dst = dst_orig;
 869                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
 870
 871                 if (unlikely(err)) {
 872                         int i;
 873                         for (i=0; i<nx; i++)
 874                                 xfrm_state_put(xfrm[i]);
 875                         goto error;
 876                 }
 877
 878                 write_lock_bh(&policy->lock);
 879                 if (unlikely(policy->dead || stale_bundle(dst))) {
 880                         /* Wow! While we worked on resolving, this
 881                          * policy has gone. Retry. It is not paranoia,
 882                          * we just cannot enlist new bundle to dead object.
 883                          * We can't enlist stable bundles either.
 884                          */
 885                         write_unlock_bh(&policy->lock);
 886                         if (dst)
 887                                 dst_free(dst);
 888
 889                         err = -EHOSTUNREACH;
 890                         goto error;
 891                 }
 892                 dst->next = policy->bundles;
 893                 policy->bundles = dst;
 894                 dst_hold(dst);
 895                 write_unlock_bh(&policy->lock);
 896         }
 897         *dst_p = dst;
 898         dst_release(dst_orig);
 899         xfrm_pol_put(policy);
 900         return 0;
 901
 902 error:
 903         dst_release(dst_orig);
 904         xfrm_pol_put(policy);
 905         *dst_p = NULL;
 906         return err;
 907 }
 908 EXPORT_SYMBOL(xfrm_lookup);
 909
 910 /* When skb is transformed back to its "native" form, we have to
 911  * check policy restrictions. At the moment we make this in maximally
 912  * stupid way. Shame on me. :-) Of course, connected sockets must
 913  * have policy cached at them.
 914  */
 915
 916 static inline int
 917 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
 918               unsigned short family)
 919 {
 920         if (xfrm_state_kern(x))
 921                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
 922         return  x->id.proto == tmpl->id.proto &&
 923                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
 924                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
 925                 x->props.mode == tmpl->mode &&
 926                 (tmpl->aalgos & (1<<x->props.aalgo)) &&
 927                 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
 928 }
 929
 930 static inline int
 931 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
 932                unsigned short family)
 933 {
 934         int idx = start;
 935
 936         if (tmpl->optional) {
 937                 if (!tmpl->mode)
 938                         return start;
 939         } else
 940                 start = -1;
 941         for (; idx < sp->len; idx++) {
 942                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
 943                         return ++idx;
 944                 if (sp->xvec[idx]->props.mode)
 945                         break;
 946         }
 947         return start;
 948 }
 949
 950 int
 951 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
 952 {
 953         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 954
 955         if (unlikely(afinfo == NULL))
 956                 return -EAFNOSUPPORT;
 957
 958         afinfo->decode_session(skb, fl);
 959         xfrm_policy_put_afinfo(afinfo);
 960         return 0;
 961 }
 962 EXPORT_SYMBOL(xfrm_decode_session);
 963
 964 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
 965 {
 966         for (; k < sp->len; k++) {
 967                 if (sp->xvec[k]->props.mode)
 968                         return 1;
 969         }
 970
 971         return 0;
 972 }
 973
 974 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
 975                         unsigned short family)
 976 {
 977         struct xfrm_policy *pol;
 978         struct flowi fl;
 979         u8 fl_dir = policy_to_flow_dir(dir);
 980         u32 sk_sid;
 981
 982         if (xfrm_decode_session(skb, &fl, family) < 0)
 983                 return 0;
 984         nf_nat_decode_session(skb, &fl, family);
 985
 986         sk_sid = security_sk_sid(sk, &fl, fl_dir);
 987
 988         /* First, check used SA against their selectors. */
 989         if (skb->sp) {
 990                 int i;
 991
 992                 for (i=skb->sp->len-1; i>=0; i--) {
 993                         struct xfrm_state *x = skb->sp->xvec[i];
 994                         if (!xfrm_selector_match(&x->sel, &fl, family))
 995                                 return 0;
 996                 }
 997         }
 998
 999         pol = NULL;
1000         if (sk && sk->sk_policy[dir])
1001                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, sk_sid);
1002
1003         if (!pol)
1004                 pol = flow_cache_lookup(&fl, sk_sid, family, fl_dir,
1005                                         xfrm_policy_lookup);
1006
1007         if (!pol)
1008                 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
1009
1010         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1011
1012         if (pol->action == XFRM_POLICY_ALLOW) {
1013                 struct sec_path *sp;
1014                 static struct sec_path dummy;
1015                 int i, k;
1016
1017                 if ((sp = skb->sp) == NULL)
1018                         sp = &dummy;
1019
1020                 /* For each tunnel xfrm, find the first matching tmpl.
1021                  * For each tmpl before that, find corresponding xfrm.
1022                  * Order is _important_. Later we will implement
1023                  * some barriers, but at the moment barriers
1024                  * are implied between each two transformations.
1025                  */
1026                 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
1027                         k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
1028                         if (k < 0)
1029                                 goto reject;
1030                 }
1031
1032                 if (secpath_has_tunnel(sp, k))
1033                         goto reject;
1034
1035                 xfrm_pol_put(pol);
1036                 return 1;
1037         }
1038
1039 reject:
1040         xfrm_pol_put(pol);
1041         return 0;
1042 }
1043 EXPORT_SYMBOL(__xfrm_policy_check);
1044
1045 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1046 {
1047         struct flowi fl;
1048
1049         if (xfrm_decode_session(skb, &fl, family) < 0)
1050                 return 0;
1051
1052         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1053 }
1054 EXPORT_SYMBOL(__xfrm_route_forward);
1055
1056 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1057 {
1058         /* If it is marked obsolete, which is how we even get here,
1059          * then we have purged it from the policy bundle list and we
1060          * did that for a good reason.
1061          */
1062         return NULL;
1063 }
1064
1065 static int stale_bundle(struct dst_entry *dst)
1066 {
1067         return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1068 }
1069
1070 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1071 {
1072         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1073                 dst->dev = &loopback_dev;
1074                 dev_hold(&loopback_dev);
1075                 dev_put(dev);
1076         }
1077 }
1078 EXPORT_SYMBOL(xfrm_dst_ifdown);
1079
1080 static void xfrm_link_failure(struct sk_buff *skb)
1081 {
1082         /* Impossible. Such dst must be popped before reaches point of failure. */
1083         return;
1084 }
1085
1086 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1087 {
1088         if (dst) {
1089                 if (dst->obsolete) {
1090                         dst_release(dst);
1091                         dst = NULL;
1092                 }
1093         }
1094         return dst;
1095 }
1096
1097 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1098 {
1099         int i;
1100         struct xfrm_policy *pol;
1101         struct dst_entry *dst, **dstp, *gc_list = NULL;
1102
1103         read_lock_bh(&xfrm_policy_lock);
1104         for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1105                 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1106                         write_lock(&pol->lock);
1107                         dstp = &pol->bundles;
1108                         while ((dst=*dstp) != NULL) {
1109                                 if (func(dst)) {
1110                                         *dstp = dst->next;
1111                                         dst->next = gc_list;
1112                                         gc_list = dst;
1113                                 } else {
1114                                         dstp = &dst->next;
1115                                 }
1116                         }
1117                         write_unlock(&pol->lock);
1118                 }
1119         }
1120         read_unlock_bh(&xfrm_policy_lock);
1121
1122         while (gc_list) {
1123                 dst = gc_list;
1124                 gc_list = dst->next;
1125                 dst_free(dst);
1126         }
1127 }
1128
1129 static int unused_bundle(struct dst_entry *dst)
1130 {
1131         return !atomic_read(&dst->__refcnt);
1132 }
1133
1134 static void __xfrm_garbage_collect(void)
1135 {
1136         xfrm_prune_bundles(unused_bundle);
1137 }
1138
1139 int xfrm_flush_bundles(void)
1140 {
1141         xfrm_prune_bundles(stale_bundle);
1142         return 0;
1143 }
1144
1145 static int always_true(struct dst_entry *dst)
1146 {
1147         return 1;
1148 }
1149
1150 void xfrm_flush_all_bundles(void)
1151 {
1152         xfrm_prune_bundles(always_true);
1153 }
1154
1155 void xfrm_init_pmtu(struct dst_entry *dst)
1156 {
1157         do {
1158                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1159                 u32 pmtu, route_mtu_cached;
1160
1161                 pmtu = dst_mtu(dst->child);
1162                 xdst->child_mtu_cached = pmtu;
1163
1164                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1165
1166                 route_mtu_cached = dst_mtu(xdst->route);
1167                 xdst->route_mtu_cached = route_mtu_cached;
1168
1169                 if (pmtu > route_mtu_cached)
1170                         pmtu = route_mtu_cached;
1171
1172                 dst->metrics[RTAX_MTU-1] = pmtu;
1173         } while ((dst = dst->next));
1174 }
1175
1176 EXPORT_SYMBOL(xfrm_init_pmtu);
1177
1178 /* Check that the bundle accepts the flow and its components are
1179  * still valid.
1180  */
1181
1182 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1183 {
1184         struct dst_entry *dst = &first->u.dst;
1185         struct xfrm_dst *last;
1186         u32 mtu;
1187
1188         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1189             (dst->dev && !netif_running(dst->dev)))
1190                 return 0;
1191
1192         last = NULL;
1193
1194         do {
1195                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1196
1197                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1198                         return 0;
1199                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1200                         return 0;
1201
1202                 mtu = dst_mtu(dst->child);
1203                 if (xdst->child_mtu_cached != mtu) {
1204                         last = xdst;
1205                         xdst->child_mtu_cached = mtu;
1206                 }
1207
1208                 if (!dst_check(xdst->route, xdst->route_cookie))
1209                         return 0;
1210                 mtu = dst_mtu(xdst->route);
1211                 if (xdst->route_mtu_cached != mtu) {
1212                         last = xdst;
1213                         xdst->route_mtu_cached = mtu;
1214                 }
1215
1216                 dst = dst->child;
1217         } while (dst->xfrm);
1218
1219         if (likely(!last))
1220                 return 1;
1221
1222         mtu = last->child_mtu_cached;
1223         for (;;) {
1224                 dst = &last->u.dst;
1225
1226                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1227                 if (mtu > last->route_mtu_cached)
1228                         mtu = last->route_mtu_cached;
1229                 dst->metrics[RTAX_MTU-1] = mtu;
1230
1231                 if (last == first)
1232                         break;
1233
1234                 last = last->u.next;
1235                 last->child_mtu_cached = mtu;
1236         }
1237
1238         return 1;
1239 }
1240
1241 EXPORT_SYMBOL(xfrm_bundle_ok);
1242
1243 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1244 {
1245         int err = 0;
1246         if (unlikely(afinfo == NULL))
1247                 return -EINVAL;
1248         if (unlikely(afinfo->family >= NPROTO))
1249                 return -EAFNOSUPPORT;
1250         write_lock_bh(&xfrm_policy_afinfo_lock);
1251         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1252                 err = -ENOBUFS;
1253         else {
1254                 struct dst_ops *dst_ops = afinfo->dst_ops;
1255                 if (likely(dst_ops->kmem_cachep == NULL))
1256                         dst_ops->kmem_cachep = xfrm_dst_cache;
1257                 if (likely(dst_ops->check == NULL))
1258                         dst_ops->check = xfrm_dst_check;
1259                 if (likely(dst_ops->negative_advice == NULL))
1260                         dst_ops->negative_advice = xfrm_negative_advice;
1261                 if (likely(dst_ops->link_failure == NULL))
1262                         dst_ops->link_failure = xfrm_link_failure;
1263                 if (likely(afinfo->garbage_collect == NULL))
1264                         afinfo->garbage_collect = __xfrm_garbage_collect;
1265                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1266         }
1267         write_unlock_bh(&xfrm_policy_afinfo_lock);
1268         return err;
1269 }
1270 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1271
1272 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1273 {
1274         int err = 0;
1275         if (unlikely(afinfo == NULL))
1276                 return -EINVAL;
1277         if (unlikely(afinfo->family >= NPROTO))
1278                 return -EAFNOSUPPORT;
1279         write_lock_bh(&xfrm_policy_afinfo_lock);
1280         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1281                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1282                         err = -EINVAL;
1283                 else {
1284                         struct dst_ops *dst_ops = afinfo->dst_ops;
1285                         xfrm_policy_afinfo[afinfo->family] = NULL;
1286                         dst_ops->kmem_cachep = NULL;
1287                         dst_ops->check = NULL;
1288                         dst_ops->negative_advice = NULL;
1289                         dst_ops->link_failure = NULL;
1290                         afinfo->garbage_collect = NULL;
1291                 }
1292         }
1293         write_unlock_bh(&xfrm_policy_afinfo_lock);
1294         return err;
1295 }
1296 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1297
1298 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1299 {
1300         struct xfrm_policy_afinfo *afinfo;
1301         if (unlikely(family >= NPROTO))
1302                 return NULL;
1303         read_lock(&xfrm_policy_afinfo_lock);
1304         afinfo = xfrm_policy_afinfo[family];
1305         if (unlikely(!afinfo))
1306                 read_unlock(&xfrm_policy_afinfo_lock);
1307         return afinfo;
1308 }
1309
1310 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1311 {
1312         read_unlock(&xfrm_policy_afinfo_lock);
1313 }
1314
1315 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
1316 {
1317         struct xfrm_policy_afinfo *afinfo;
1318         if (unlikely(family >= NPROTO))
1319                 return NULL;
1320         write_lock_bh(&xfrm_policy_afinfo_lock);
1321         afinfo = xfrm_policy_afinfo[family];
1322         if (unlikely(!afinfo))
1323                 write_unlock_bh(&xfrm_policy_afinfo_lock);
1324         return afinfo;
1325 }
1326
1327 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
1328 {
1329         write_unlock_bh(&xfrm_policy_afinfo_lock);
1330 }
1331
1332 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1333 {
1334         switch (event) {
1335         case NETDEV_DOWN:
1336                 xfrm_flush_bundles();
1337         }
1338         return NOTIFY_DONE;
1339 }
1340
1341 static struct notifier_block xfrm_dev_notifier = {
1342         xfrm_dev_event,
1343         NULL,
1344         0
1345 };
1346
1347 static void __init xfrm_policy_init(void)
1348 {
1349         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1350                                            sizeof(struct xfrm_dst),
1351                                            0, SLAB_HWCACHE_ALIGN,
1352                                            NULL, NULL);
1353         if (!xfrm_dst_cache)
1354                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1355
1356         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1357         register_netdevice_notifier(&xfrm_dev_notifier);
1358 }
1359
1360 void __init xfrm_init(void)
1361 {
1362         xfrm_state_init();
1363         xfrm_policy_init();
1364         xfrm_input_init();
1365 }
1366