git.oblomov.eu 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 int xfrm_register_mode(struct xfrm_mode *mode, int family)
 142 {
 143         struct xfrm_policy_afinfo *afinfo;
 144         struct xfrm_mode **modemap;
 145         int err;
 146
 147         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 148                 return -EINVAL;
 149
 150         afinfo = xfrm_policy_lock_afinfo(family);
 151         if (unlikely(afinfo == NULL))
 152                 return -EAFNOSUPPORT;
 153
 154         err = -EEXIST;
 155         modemap = afinfo->mode_map;
 156         if (likely(modemap[mode->encap] == NULL)) {
 157                 modemap[mode->encap] = mode;
 158                 err = 0;
 159         }
 160
 161         xfrm_policy_unlock_afinfo(afinfo);
 162         return err;
 163 }
 164 EXPORT_SYMBOL(xfrm_register_mode);
 165
 166 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
 167 {
 168         struct xfrm_policy_afinfo *afinfo;
 169         struct xfrm_mode **modemap;
 170         int err;
 171
 172         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 173                 return -EINVAL;
 174
 175         afinfo = xfrm_policy_lock_afinfo(family);
 176         if (unlikely(afinfo == NULL))
 177                 return -EAFNOSUPPORT;
 178
 179         err = -ENOENT;
 180         modemap = afinfo->mode_map;
 181         if (likely(modemap[mode->encap] == mode)) {
 182                 modemap[mode->encap] = NULL;
 183                 err = 0;
 184         }
 185
 186         xfrm_policy_unlock_afinfo(afinfo);
 187         return err;
 188 }
 189 EXPORT_SYMBOL(xfrm_unregister_mode);
 190
 191 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
 192 {
 193         struct xfrm_policy_afinfo *afinfo;
 194         struct xfrm_mode *mode;
 195         int modload_attempted = 0;
 196
 197         if (unlikely(encap >= XFRM_MODE_MAX))
 198                 return NULL;
 199
 200 retry:
 201         afinfo = xfrm_policy_get_afinfo(family);
 202         if (unlikely(afinfo == NULL))
 203                 return NULL;
 204
 205         mode = afinfo->mode_map[encap];
 206         if (unlikely(mode && !try_module_get(mode->owner)))
 207                 mode = NULL;
 208         if (!mode && !modload_attempted) {
 209                 xfrm_policy_put_afinfo(afinfo);
 210                 request_module("xfrm-mode-%d-%d", family, encap);
 211                 modload_attempted = 1;
 212                 goto retry;
 213         }
 214
 215         xfrm_policy_put_afinfo(afinfo);
 216         return mode;
 217 }
 218
 219 void xfrm_put_mode(struct xfrm_mode *mode)
 220 {
 221         module_put(mode->owner);
 222 }
 223
 224 static inline unsigned long make_jiffies(long secs)
 225 {
 226         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 227                 return MAX_SCHEDULE_TIMEOUT-1;
 228         else
 229                 return secs*HZ;
 230 }
 231
 232 static void xfrm_policy_timer(unsigned long data)
 233 {
 234         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 235         unsigned long now = (unsigned long)xtime.tv_sec;
 236         long next = LONG_MAX;
 237         int warn = 0;
 238         int dir;
 239
 240         read_lock(&xp->lock);
 241
 242         if (xp->dead)
 243                 goto out;
 244
 245         dir = xfrm_policy_id2dir(xp->index);
 246
 247         if (xp->lft.hard_add_expires_seconds) {
 248                 long tmo = xp->lft.hard_add_expires_seconds +
 249                         xp->curlft.add_time - now;
 250                 if (tmo <= 0)
 251                         goto expired;
 252                 if (tmo < next)
 253                         next = tmo;
 254         }
 255         if (xp->lft.hard_use_expires_seconds) {
 256                 long tmo = xp->lft.hard_use_expires_seconds +
 257                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 258                 if (tmo <= 0)
 259                         goto expired;
 260                 if (tmo < next)
 261                         next = tmo;
 262         }
 263         if (xp->lft.soft_add_expires_seconds) {
 264                 long tmo = xp->lft.soft_add_expires_seconds +
 265                         xp->curlft.add_time - now;
 266                 if (tmo <= 0) {
 267                         warn = 1;
 268                         tmo = XFRM_KM_TIMEOUT;
 269                 }
 270                 if (tmo < next)
 271                         next = tmo;
 272         }
 273         if (xp->lft.soft_use_expires_seconds) {
 274                 long tmo = xp->lft.soft_use_expires_seconds +
 275                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 276                 if (tmo <= 0) {
 277                         warn = 1;
 278                         tmo = XFRM_KM_TIMEOUT;
 279                 }
 280                 if (tmo < next)
 281                         next = tmo;
 282         }
 283
 284         if (warn)
 285                 km_policy_expired(xp, dir, 0, 0);
 286         if (next != LONG_MAX &&
 287             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 288                 xfrm_pol_hold(xp);
 289
 290 out:
 291         read_unlock(&xp->lock);
 292         xfrm_pol_put(xp);
 293         return;
 294
 295 expired:
 296         read_unlock(&xp->lock);
 297         if (!xfrm_policy_delete(xp, dir))
 298                 km_policy_expired(xp, dir, 1, 0);
 299         xfrm_pol_put(xp);
 300 }
 301
 302
 303 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 304  * SPD calls.
 305  */
 306
 307 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 308 {
 309         struct xfrm_policy *policy;
 310
 311         policy = kmalloc(sizeof(struct xfrm_policy), gfp);
 312
 313         if (policy) {
 314                 memset(policy, 0, sizeof(struct xfrm_policy));
 315                 atomic_set(&policy->refcnt, 1);
 316                 rwlock_init(&policy->lock);
 317                 init_timer(&policy->timer);
 318                 policy->timer.data = (unsigned long)policy;
 319                 policy->timer.function = xfrm_policy_timer;
 320         }
 321         return policy;
 322 }
 323 EXPORT_SYMBOL(xfrm_policy_alloc);
 324
 325 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 326
 327 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 328 {
 329         BUG_ON(!policy->dead);
 330
 331         BUG_ON(policy->bundles);
 332
 333         if (del_timer(&policy->timer))
 334                 BUG();
 335
 336         security_xfrm_policy_free(policy);
 337         kfree(policy);
 338 }
 339 EXPORT_SYMBOL(__xfrm_policy_destroy);
 340
 341 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 342 {
 343         struct dst_entry *dst;
 344
 345         while ((dst = policy->bundles) != NULL) {
 346                 policy->bundles = dst->next;
 347                 dst_free(dst);
 348         }
 349
 350         if (del_timer(&policy->timer))
 351                 atomic_dec(&policy->refcnt);
 352
 353         if (atomic_read(&policy->refcnt) > 1)
 354                 flow_cache_flush();
 355
 356         xfrm_pol_put(policy);
 357 }
 358
 359 static void xfrm_policy_gc_task(void *data)
 360 {
 361         struct xfrm_policy *policy;
 362         struct list_head *entry, *tmp;
 363         struct list_head gc_list = LIST_HEAD_INIT(gc_list);
 364
 365         spin_lock_bh(&xfrm_policy_gc_lock);
 366         list_splice_init(&xfrm_policy_gc_list, &gc_list);
 367         spin_unlock_bh(&xfrm_policy_gc_lock);
 368
 369         list_for_each_safe(entry, tmp, &gc_list) {
 370                 policy = list_entry(entry, struct xfrm_policy, list);
 371                 xfrm_policy_gc_kill(policy);
 372         }
 373 }
 374
 375 /* Rule must be locked. Release descentant resources, announce
 376  * entry dead. The rule must be unlinked from lists to the moment.
 377  */
 378
 379 static void xfrm_policy_kill(struct xfrm_policy *policy)
 380 {
 381         int dead;
 382
 383         write_lock_bh(&policy->lock);
 384         dead = policy->dead;
 385         policy->dead = 1;
 386         write_unlock_bh(&policy->lock);
 387
 388         if (unlikely(dead)) {
 389                 WARN_ON(1);
 390                 return;
 391         }
 392
 393         spin_lock(&xfrm_policy_gc_lock);
 394         list_add(&policy->list, &xfrm_policy_gc_list);
 395         spin_unlock(&xfrm_policy_gc_lock);
 396
 397         schedule_work(&xfrm_policy_gc_work);
 398 }
 399
 400 /* Generate new index... KAME seems to generate them ordered by cost
 401  * of an absolute inpredictability of ordering of rules. This will not pass. */
 402 static u32 xfrm_gen_index(int dir)
 403 {
 404         u32 idx;
 405         struct xfrm_policy *p;
 406         static u32 idx_generator;
 407
 408         for (;;) {
 409                 idx = (idx_generator | dir);
 410                 idx_generator += 8;
 411                 if (idx == 0)
 412                         idx = 8;
 413                 for (p = xfrm_policy_list[dir]; p; p = p->next) {
 414                         if (p->index == idx)
 415                                 break;
 416                 }
 417                 if (!p)
 418                         return idx;
 419         }
 420 }
 421
 422 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 423 {
 424         struct xfrm_policy *pol, **p;
 425         struct xfrm_policy *delpol = NULL;
 426         struct xfrm_policy **newpos = NULL;
 427         struct dst_entry *gc_list;
 428
 429         write_lock_bh(&xfrm_policy_lock);
 430         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
 431                 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0 &&
 432                     xfrm_sec_ctx_match(pol->security, policy->security)) {
 433                         if (excl) {
 434                                 write_unlock_bh(&xfrm_policy_lock);
 435                                 return -EEXIST;
 436                         }
 437                         *p = pol->next;
 438                         delpol = pol;
 439                         if (policy->priority > pol->priority)
 440                                 continue;
 441                 } else if (policy->priority >= pol->priority) {
 442                         p = &pol->next;
 443                         continue;
 444                 }
 445                 if (!newpos)
 446                         newpos = p;
 447                 if (delpol)
 448                         break;
 449                 p = &pol->next;
 450         }
 451         if (newpos)
 452                 p = newpos;
 453         xfrm_pol_hold(policy);
 454         policy->next = *p;
 455         *p = policy;
 456         atomic_inc(&flow_cache_genid);
 457         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
 458         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
 459         policy->curlft.use_time = 0;
 460         if (!mod_timer(&policy->timer, jiffies + HZ))
 461                 xfrm_pol_hold(policy);
 462         write_unlock_bh(&xfrm_policy_lock);
 463
 464         if (delpol)
 465                 xfrm_policy_kill(delpol);
 466
 467         read_lock_bh(&xfrm_policy_lock);
 468         gc_list = NULL;
 469         for (policy = policy->next; policy; policy = policy->next) {
 470                 struct dst_entry *dst;
 471
 472                 write_lock(&policy->lock);
 473                 dst = policy->bundles;
 474                 if (dst) {
 475                         struct dst_entry *tail = dst;
 476                         while (tail->next)
 477                                 tail = tail->next;
 478                         tail->next = gc_list;
 479                         gc_list = dst;
 480
 481                         policy->bundles = NULL;
 482                 }
 483                 write_unlock(&policy->lock);
 484         }
 485         read_unlock_bh(&xfrm_policy_lock);
 486
 487         while (gc_list) {
 488                 struct dst_entry *dst = gc_list;
 489
 490                 gc_list = dst->next;
 491                 dst_free(dst);
 492         }
 493
 494         return 0;
 495 }
 496 EXPORT_SYMBOL(xfrm_policy_insert);
 497
 498 struct xfrm_policy *xfrm_policy_bysel_ctx(int dir, struct xfrm_selector *sel,
 499                                           struct xfrm_sec_ctx *ctx, int delete)
 500 {
 501         struct xfrm_policy *pol, **p;
 502
 503         write_lock_bh(&xfrm_policy_lock);
 504         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 505                 if ((memcmp(sel, &pol->selector, sizeof(*sel)) == 0) &&
 506                     (xfrm_sec_ctx_match(ctx, pol->security))) {
 507                         xfrm_pol_hold(pol);
 508                         if (delete)
 509                                 *p = pol->next;
 510                         break;
 511                 }
 512         }
 513         write_unlock_bh(&xfrm_policy_lock);
 514
 515         if (pol && delete) {
 516                 atomic_inc(&flow_cache_genid);
 517                 xfrm_policy_kill(pol);
 518         }
 519         return pol;
 520 }
 521 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 522
 523 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
 524 {
 525         struct xfrm_policy *pol, **p;
 526
 527         write_lock_bh(&xfrm_policy_lock);
 528         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 529                 if (pol->index == id) {
 530                         xfrm_pol_hold(pol);
 531                         if (delete)
 532                                 *p = pol->next;
 533                         break;
 534                 }
 535         }
 536         write_unlock_bh(&xfrm_policy_lock);
 537
 538         if (pol && delete) {
 539                 atomic_inc(&flow_cache_genid);
 540                 xfrm_policy_kill(pol);
 541         }
 542         return pol;
 543 }
 544 EXPORT_SYMBOL(xfrm_policy_byid);
 545
 546 void xfrm_policy_flush(void)
 547 {
 548         struct xfrm_policy *xp;
 549         int dir;
 550
 551         write_lock_bh(&xfrm_policy_lock);
 552         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 553                 while ((xp = xfrm_policy_list[dir]) != NULL) {
 554                         xfrm_policy_list[dir] = xp->next;
 555                         write_unlock_bh(&xfrm_policy_lock);
 556
 557                         xfrm_policy_kill(xp);
 558
 559                         write_lock_bh(&xfrm_policy_lock);
 560                 }
 561         }
 562         atomic_inc(&flow_cache_genid);
 563         write_unlock_bh(&xfrm_policy_lock);
 564 }
 565 EXPORT_SYMBOL(xfrm_policy_flush);
 566
 567 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
 568                      void *data)
 569 {
 570         struct xfrm_policy *xp;
 571         int dir;
 572         int count = 0;
 573         int error = 0;
 574
 575         read_lock_bh(&xfrm_policy_lock);
 576         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 577                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
 578                         count++;
 579         }
 580
 581         if (count == 0) {
 582                 error = -ENOENT;
 583                 goto out;
 584         }
 585
 586         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 587                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
 588                         error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
 589                         if (error)
 590                                 goto out;
 591                 }
 592         }
 593
 594 out:
 595         read_unlock_bh(&xfrm_policy_lock);
 596         return error;
 597 }
 598 EXPORT_SYMBOL(xfrm_policy_walk);
 599
 600 /* Find policy to apply to this flow. */
 601
 602 static void xfrm_policy_lookup(struct flowi *fl, u32 sk_sid, u16 family, u8 dir,
 603                                void **objp, atomic_t **obj_refp)
 604 {
 605         struct xfrm_policy *pol;
 606
 607         read_lock_bh(&xfrm_policy_lock);
 608         for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
 609                 struct xfrm_selector *sel = &pol->selector;
 610                 int match;
 611
 612                 if (pol->family != family)
 613                         continue;
 614
 615                 match = xfrm_selector_match(sel, fl, family);
 616
 617                 if (match) {
 618                         if (!security_xfrm_policy_lookup(pol, sk_sid, dir)) {
 619                                 xfrm_pol_hold(pol);
 620                                 break;
 621                         }
 622                 }
 623         }
 624         read_unlock_bh(&xfrm_policy_lock);
 625         if ((*objp = (void *) pol) != NULL)
 626                 *obj_refp = &pol->refcnt;
 627 }
 628
 629 static inline int policy_to_flow_dir(int dir)
 630 {
 631         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 632             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 633             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 634                 return dir;
 635         switch (dir) {
 636         default:
 637         case XFRM_POLICY_IN:
 638                 return FLOW_DIR_IN;
 639         case XFRM_POLICY_OUT:
 640                 return FLOW_DIR_OUT;
 641         case XFRM_POLICY_FWD:
 642                 return FLOW_DIR_FWD;
 643         };
 644 }
 645
 646 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl, u32 sk_sid)
 647 {
 648         struct xfrm_policy *pol;
 649
 650         read_lock_bh(&xfrm_policy_lock);
 651         if ((pol = sk->sk_policy[dir]) != NULL) {
 652                 int match = xfrm_selector_match(&pol->selector, fl,
 653                                                 sk->sk_family);
 654                 int err = 0;
 655
 656                 if (match)
 657                   err = security_xfrm_policy_lookup(pol, sk_sid, policy_to_flow_dir(dir));
 658
 659                 if (match && !err)
 660                         xfrm_pol_hold(pol);
 661                 else
 662                         pol = NULL;
 663         }
 664         read_unlock_bh(&xfrm_policy_lock);
 665         return pol;
 666 }
 667
 668 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
 669 {
 670         pol->next = xfrm_policy_list[dir];
 671         xfrm_policy_list[dir] = pol;
 672         xfrm_pol_hold(pol);
 673 }
 674
 675 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 676                                                 int dir)
 677 {
 678         struct xfrm_policy **polp;
 679
 680         for (polp = &xfrm_policy_list[dir];
 681              *polp != NULL; polp = &(*polp)->next) {
 682                 if (*polp == pol) {
 683                         *polp = pol->next;
 684                         return pol;
 685                 }
 686         }
 687         return NULL;
 688 }
 689
 690 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
 691 {
 692         write_lock_bh(&xfrm_policy_lock);
 693         pol = __xfrm_policy_unlink(pol, dir);
 694         write_unlock_bh(&xfrm_policy_lock);
 695         if (pol) {
 696                 if (dir < XFRM_POLICY_MAX)
 697                         atomic_inc(&flow_cache_genid);
 698                 xfrm_policy_kill(pol);
 699                 return 0;
 700         }
 701         return -ENOENT;
 702 }
 703 EXPORT_SYMBOL(xfrm_policy_delete);
 704
 705 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
 706 {
 707         struct xfrm_policy *old_pol;
 708
 709         write_lock_bh(&xfrm_policy_lock);
 710         old_pol = sk->sk_policy[dir];
 711         sk->sk_policy[dir] = pol;
 712         if (pol) {
 713                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
 714                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
 715                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
 716         }
 717         if (old_pol)
 718                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
 719         write_unlock_bh(&xfrm_policy_lock);
 720
 721         if (old_pol) {
 722                 xfrm_policy_kill(old_pol);
 723         }
 724         return 0;
 725 }
 726
 727 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
 728 {
 729         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
 730
 731         if (newp) {
 732                 newp->selector = old->selector;
 733                 if (security_xfrm_policy_clone(old, newp)) {
 734                         kfree(newp);
 735                         return NULL;  /* ENOMEM */
 736                 }
 737                 newp->lft = old->lft;
 738                 newp->curlft = old->curlft;
 739                 newp->action = old->action;
 740                 newp->flags = old->flags;
 741                 newp->xfrm_nr = old->xfrm_nr;
 742                 newp->index = old->index;
 743                 memcpy(newp->xfrm_vec, old->xfrm_vec,
 744                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
 745                 write_lock_bh(&xfrm_policy_lock);
 746                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
 747                 write_unlock_bh(&xfrm_policy_lock);
 748                 xfrm_pol_put(newp);
 749         }
 750         return newp;
 751 }
 752
 753 int __xfrm_sk_clone_policy(struct sock *sk)
 754 {
 755         struct xfrm_policy *p0 = sk->sk_policy[0],
 756                            *p1 = sk->sk_policy[1];
 757
 758         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
 759         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
 760                 return -ENOMEM;
 761         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
 762                 return -ENOMEM;
 763         return 0;
 764 }
 765
 766 /* Resolve list of templates for the flow, given policy. */
 767
 768 static int
 769 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
 770                   struct xfrm_state **xfrm,
 771                   unsigned short family)
 772 {
 773         int nx;
 774         int i, error;
 775         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
 776         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
 777
 778         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
 779                 struct xfrm_state *x;
 780                 xfrm_address_t *remote = daddr;
 781                 xfrm_address_t *local  = saddr;
 782                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
 783
 784                 if (tmpl->mode) {
 785                         remote = &tmpl->id.daddr;
 786                         local = &tmpl->saddr;
 787                 }
 788
 789                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
 790
 791                 if (x && x->km.state == XFRM_STATE_VALID) {
 792                         xfrm[nx++] = x;
 793                         daddr = remote;
 794                         saddr = local;
 795                         continue;
 796                 }
 797                 if (x) {
 798                         error = (x->km.state == XFRM_STATE_ERROR ?
 799                                  -EINVAL : -EAGAIN);
 800                         xfrm_state_put(x);
 801                 }
 802
 803                 if (!tmpl->optional)
 804                         goto fail;
 805         }
 806         return nx;
 807
 808 fail:
 809         for (nx--; nx>=0; nx--)
 810                 xfrm_state_put(xfrm[nx]);
 811         return error;
 812 }
 813
 814 /* Check that the bundle accepts the flow and its components are
 815  * still valid.
 816  */
 817
 818 static struct dst_entry *
 819 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
 820 {
 821         struct dst_entry *x;
 822         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 823         if (unlikely(afinfo == NULL))
 824                 return ERR_PTR(-EINVAL);
 825         x = afinfo->find_bundle(fl, policy);
 826         xfrm_policy_put_afinfo(afinfo);
 827         return x;
 828 }
 829
 830 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
 831  * all the metrics... Shortly, bundle a bundle.
 832  */
 833
 834 static int
 835 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
 836                    struct flowi *fl, struct dst_entry **dst_p,
 837                    unsigned short family)
 838 {
 839         int err;
 840         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 841         if (unlikely(afinfo == NULL))
 842                 return -EINVAL;
 843         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
 844         xfrm_policy_put_afinfo(afinfo);
 845         return err;
 846 }
 847
 848
 849 static int stale_bundle(struct dst_entry *dst);
 850
 851 /* Main function: finds/creates a bundle for given flow.
 852  *
 853  * At the moment we eat a raw IP route. Mostly to speed up lookups
 854  * on interfaces with disabled IPsec.
 855  */
 856 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
 857                 struct sock *sk, int flags)
 858 {
 859         struct xfrm_policy *policy;
 860         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
 861         struct dst_entry *dst, *dst_orig = *dst_p;
 862         int nx = 0;
 863         int err;
 864         u32 genid;
 865         u16 family;
 866         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
 867         u32 sk_sid = security_sk_sid(sk, fl, dir);
 868 restart:
 869         genid = atomic_read(&flow_cache_genid);
 870         policy = NULL;
 871         if (sk && sk->sk_policy[1])
 872                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, sk_sid);
 873
 874         if (!policy) {
 875                 /* To accelerate a bit...  */
 876                 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
 877                         return 0;
 878
 879                 policy = flow_cache_lookup(fl, sk_sid, dst_orig->ops->family,
 880                                            dir, xfrm_policy_lookup);
 881         }
 882
 883         if (!policy)
 884                 return 0;
 885
 886         family = dst_orig->ops->family;
 887         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
 888
 889         switch (policy->action) {
 890         case XFRM_POLICY_BLOCK:
 891                 /* Prohibit the flow */
 892                 err = -EPERM;
 893                 goto error;
 894
 895         case XFRM_POLICY_ALLOW:
 896                 if (policy->xfrm_nr == 0) {
 897                         /* Flow passes not transformed. */
 898                         xfrm_pol_put(policy);
 899                         return 0;
 900                 }
 901
 902                 /* Try to find matching bundle.
 903                  *
 904                  * LATER: help from flow cache. It is optional, this
 905                  * is required only for output policy.
 906                  */
 907                 dst = xfrm_find_bundle(fl, policy, family);
 908                 if (IS_ERR(dst)) {
 909                         err = PTR_ERR(dst);
 910                         goto error;
 911                 }
 912
 913                 if (dst)
 914                         break;
 915
 916                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 917
 918                 if (unlikely(nx<0)) {
 919                         err = nx;
 920                         if (err == -EAGAIN && flags) {
 921                                 DECLARE_WAITQUEUE(wait, current);
 922
 923                                 add_wait_queue(&km_waitq, &wait);
 924                                 set_current_state(TASK_INTERRUPTIBLE);
 925                                 schedule();
 926                                 set_current_state(TASK_RUNNING);
 927                                 remove_wait_queue(&km_waitq, &wait);
 928
 929                                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 930
 931                                 if (nx == -EAGAIN && signal_pending(current)) {
 932                                         err = -ERESTART;
 933                                         goto error;
 934                                 }
 935                                 if (nx == -EAGAIN ||
 936                                     genid != atomic_read(&flow_cache_genid)) {
 937                                         xfrm_pol_put(policy);
 938                                         goto restart;
 939                                 }
 940                                 err = nx;
 941                         }
 942                         if (err < 0)
 943                                 goto error;
 944                 }
 945                 if (nx == 0) {
 946                         /* Flow passes not transformed. */
 947                         xfrm_pol_put(policy);
 948                         return 0;
 949                 }
 950
 951                 dst = dst_orig;
 952                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
 953
 954                 if (unlikely(err)) {
 955                         int i;
 956                         for (i=0; i<nx; i++)
 957                                 xfrm_state_put(xfrm[i]);
 958                         goto error;
 959                 }
 960
 961                 write_lock_bh(&policy->lock);
 962                 if (unlikely(policy->dead || stale_bundle(dst))) {
 963                         /* Wow! While we worked on resolving, this
 964                          * policy has gone. Retry. It is not paranoia,
 965                          * we just cannot enlist new bundle to dead object.
 966                          * We can't enlist stable bundles either.
 967                          */
 968                         write_unlock_bh(&policy->lock);
 969                         if (dst)
 970                                 dst_free(dst);
 971
 972                         err = -EHOSTUNREACH;
 973                         goto error;
 974                 }
 975                 dst->next = policy->bundles;
 976                 policy->bundles = dst;
 977                 dst_hold(dst);
 978                 write_unlock_bh(&policy->lock);
 979         }
 980         *dst_p = dst;
 981         dst_release(dst_orig);
 982         xfrm_pol_put(policy);
 983         return 0;
 984
 985 error:
 986         dst_release(dst_orig);
 987         xfrm_pol_put(policy);
 988         *dst_p = NULL;
 989         return err;
 990 }
 991 EXPORT_SYMBOL(xfrm_lookup);
 992
 993 /* When skb is transformed back to its "native" form, we have to
 994  * check policy restrictions. At the moment we make this in maximally
 995  * stupid way. Shame on me. :-) Of course, connected sockets must
 996  * have policy cached at them.
 997  */
 998
 999 static inline int
1000 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1001               unsigned short family)
1002 {
1003         if (xfrm_state_kern(x))
1004                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1005         return  x->id.proto == tmpl->id.proto &&
1006                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1007                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1008                 x->props.mode == tmpl->mode &&
1009                 (tmpl->aalgos & (1<<x->props.aalgo)) &&
1010                 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
1011 }
1012
1013 static inline int
1014 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1015                unsigned short family)
1016 {
1017         int idx = start;
1018
1019         if (tmpl->optional) {
1020                 if (!tmpl->mode)
1021                         return start;
1022         } else
1023                 start = -1;
1024         for (; idx < sp->len; idx++) {
1025                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1026                         return ++idx;
1027                 if (sp->xvec[idx]->props.mode)
1028                         break;
1029         }
1030         return start;
1031 }
1032
1033 int
1034 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1035 {
1036         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1037
1038         if (unlikely(afinfo == NULL))
1039                 return -EAFNOSUPPORT;
1040
1041         afinfo->decode_session(skb, fl);
1042         xfrm_policy_put_afinfo(afinfo);
1043         return 0;
1044 }
1045 EXPORT_SYMBOL(xfrm_decode_session);
1046
1047 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
1048 {
1049         for (; k < sp->len; k++) {
1050                 if (sp->xvec[k]->props.mode)
1051                         return 1;
1052         }
1053
1054         return 0;
1055 }
1056
1057 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1058                         unsigned short family)
1059 {
1060         struct xfrm_policy *pol;
1061         struct flowi fl;
1062         u8 fl_dir = policy_to_flow_dir(dir);
1063         u32 sk_sid;
1064
1065         if (xfrm_decode_session(skb, &fl, family) < 0)
1066                 return 0;
1067         nf_nat_decode_session(skb, &fl, family);
1068
1069         sk_sid = security_sk_sid(sk, &fl, fl_dir);
1070
1071         /* First, check used SA against their selectors. */
1072         if (skb->sp) {
1073                 int i;
1074
1075                 for (i=skb->sp->len-1; i>=0; i--) {
1076                         struct xfrm_state *x = skb->sp->xvec[i];
1077                         if (!xfrm_selector_match(&x->sel, &fl, family))
1078                                 return 0;
1079                 }
1080         }
1081
1082         pol = NULL;
1083         if (sk && sk->sk_policy[dir])
1084                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, sk_sid);
1085
1086         if (!pol)
1087                 pol = flow_cache_lookup(&fl, sk_sid, family, fl_dir,
1088                                         xfrm_policy_lookup);
1089
1090         if (!pol)
1091                 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
1092
1093         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1094
1095         if (pol->action == XFRM_POLICY_ALLOW) {
1096                 struct sec_path *sp;
1097                 static struct sec_path dummy;
1098                 int i, k;
1099
1100                 if ((sp = skb->sp) == NULL)
1101                         sp = &dummy;
1102
1103                 /* For each tunnel xfrm, find the first matching tmpl.
1104                  * For each tmpl before that, find corresponding xfrm.
1105                  * Order is _important_. Later we will implement
1106                  * some barriers, but at the moment barriers
1107                  * are implied between each two transformations.
1108                  */
1109                 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
1110                         k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
1111                         if (k < 0)
1112                                 goto reject;
1113                 }
1114
1115                 if (secpath_has_tunnel(sp, k))
1116                         goto reject;
1117
1118                 xfrm_pol_put(pol);
1119                 return 1;
1120         }
1121
1122 reject:
1123         xfrm_pol_put(pol);
1124         return 0;
1125 }
1126 EXPORT_SYMBOL(__xfrm_policy_check);
1127
1128 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1129 {
1130         struct flowi fl;
1131
1132         if (xfrm_decode_session(skb, &fl, family) < 0)
1133                 return 0;
1134
1135         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1136 }
1137 EXPORT_SYMBOL(__xfrm_route_forward);
1138
1139 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1140 {
1141         /* If it is marked obsolete, which is how we even get here,
1142          * then we have purged it from the policy bundle list and we
1143          * did that for a good reason.
1144          */
1145         return NULL;
1146 }
1147
1148 static int stale_bundle(struct dst_entry *dst)
1149 {
1150         return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1151 }
1152
1153 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1154 {
1155         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1156                 dst->dev = &loopback_dev;
1157                 dev_hold(&loopback_dev);
1158                 dev_put(dev);
1159         }
1160 }
1161 EXPORT_SYMBOL(xfrm_dst_ifdown);
1162
1163 static void xfrm_link_failure(struct sk_buff *skb)
1164 {
1165         /* Impossible. Such dst must be popped before reaches point of failure. */
1166         return;
1167 }
1168
1169 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1170 {
1171         if (dst) {
1172                 if (dst->obsolete) {
1173                         dst_release(dst);
1174                         dst = NULL;
1175                 }
1176         }
1177         return dst;
1178 }
1179
1180 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1181 {
1182         int i;
1183         struct xfrm_policy *pol;
1184         struct dst_entry *dst, **dstp, *gc_list = NULL;
1185
1186         read_lock_bh(&xfrm_policy_lock);
1187         for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1188                 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1189                         write_lock(&pol->lock);
1190                         dstp = &pol->bundles;
1191                         while ((dst=*dstp) != NULL) {
1192                                 if (func(dst)) {
1193                                         *dstp = dst->next;
1194                                         dst->next = gc_list;
1195                                         gc_list = dst;
1196                                 } else {
1197                                         dstp = &dst->next;
1198                                 }
1199                         }
1200                         write_unlock(&pol->lock);
1201                 }
1202         }
1203         read_unlock_bh(&xfrm_policy_lock);
1204
1205         while (gc_list) {
1206                 dst = gc_list;
1207                 gc_list = dst->next;
1208                 dst_free(dst);
1209         }
1210 }
1211
1212 static int unused_bundle(struct dst_entry *dst)
1213 {
1214         return !atomic_read(&dst->__refcnt);
1215 }
1216
1217 static void __xfrm_garbage_collect(void)
1218 {
1219         xfrm_prune_bundles(unused_bundle);
1220 }
1221
1222 int xfrm_flush_bundles(void)
1223 {
1224         xfrm_prune_bundles(stale_bundle);
1225         return 0;
1226 }
1227
1228 static int always_true(struct dst_entry *dst)
1229 {
1230         return 1;
1231 }
1232
1233 void xfrm_flush_all_bundles(void)
1234 {
1235         xfrm_prune_bundles(always_true);
1236 }
1237
1238 void xfrm_init_pmtu(struct dst_entry *dst)
1239 {
1240         do {
1241                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1242                 u32 pmtu, route_mtu_cached;
1243
1244                 pmtu = dst_mtu(dst->child);
1245                 xdst->child_mtu_cached = pmtu;
1246
1247                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1248
1249                 route_mtu_cached = dst_mtu(xdst->route);
1250                 xdst->route_mtu_cached = route_mtu_cached;
1251
1252                 if (pmtu > route_mtu_cached)
1253                         pmtu = route_mtu_cached;
1254
1255                 dst->metrics[RTAX_MTU-1] = pmtu;
1256         } while ((dst = dst->next));
1257 }
1258
1259 EXPORT_SYMBOL(xfrm_init_pmtu);
1260
1261 /* Check that the bundle accepts the flow and its components are
1262  * still valid.
1263  */
1264
1265 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1266 {
1267         struct dst_entry *dst = &first->u.dst;
1268         struct xfrm_dst *last;
1269         u32 mtu;
1270
1271         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1272             (dst->dev && !netif_running(dst->dev)))
1273                 return 0;
1274
1275         last = NULL;
1276
1277         do {
1278                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1279
1280                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1281                         return 0;
1282                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1283                         return 0;
1284
1285                 mtu = dst_mtu(dst->child);
1286                 if (xdst->child_mtu_cached != mtu) {
1287                         last = xdst;
1288                         xdst->child_mtu_cached = mtu;
1289                 }
1290
1291                 if (!dst_check(xdst->route, xdst->route_cookie))
1292                         return 0;
1293                 mtu = dst_mtu(xdst->route);
1294                 if (xdst->route_mtu_cached != mtu) {
1295                         last = xdst;
1296                         xdst->route_mtu_cached = mtu;
1297                 }
1298
1299                 dst = dst->child;
1300         } while (dst->xfrm);
1301
1302         if (likely(!last))
1303                 return 1;
1304
1305         mtu = last->child_mtu_cached;
1306         for (;;) {
1307                 dst = &last->u.dst;
1308
1309                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1310                 if (mtu > last->route_mtu_cached)
1311                         mtu = last->route_mtu_cached;
1312                 dst->metrics[RTAX_MTU-1] = mtu;
1313
1314                 if (last == first)
1315                         break;
1316
1317                 last = last->u.next;
1318                 last->child_mtu_cached = mtu;
1319         }
1320
1321         return 1;
1322 }
1323
1324 EXPORT_SYMBOL(xfrm_bundle_ok);
1325
1326 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1327 {
1328         int err = 0;
1329         if (unlikely(afinfo == NULL))
1330                 return -EINVAL;
1331         if (unlikely(afinfo->family >= NPROTO))
1332                 return -EAFNOSUPPORT;
1333         write_lock_bh(&xfrm_policy_afinfo_lock);
1334         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1335                 err = -ENOBUFS;
1336         else {
1337                 struct dst_ops *dst_ops = afinfo->dst_ops;
1338                 if (likely(dst_ops->kmem_cachep == NULL))
1339                         dst_ops->kmem_cachep = xfrm_dst_cache;
1340                 if (likely(dst_ops->check == NULL))
1341                         dst_ops->check = xfrm_dst_check;
1342                 if (likely(dst_ops->negative_advice == NULL))
1343                         dst_ops->negative_advice = xfrm_negative_advice;
1344                 if (likely(dst_ops->link_failure == NULL))
1345                         dst_ops->link_failure = xfrm_link_failure;
1346                 if (likely(afinfo->garbage_collect == NULL))
1347                         afinfo->garbage_collect = __xfrm_garbage_collect;
1348                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1349         }
1350         write_unlock_bh(&xfrm_policy_afinfo_lock);
1351         return err;
1352 }
1353 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1354
1355 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1356 {
1357         int err = 0;
1358         if (unlikely(afinfo == NULL))
1359                 return -EINVAL;
1360         if (unlikely(afinfo->family >= NPROTO))
1361                 return -EAFNOSUPPORT;
1362         write_lock_bh(&xfrm_policy_afinfo_lock);
1363         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1364                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1365                         err = -EINVAL;
1366                 else {
1367                         struct dst_ops *dst_ops = afinfo->dst_ops;
1368                         xfrm_policy_afinfo[afinfo->family] = NULL;
1369                         dst_ops->kmem_cachep = NULL;
1370                         dst_ops->check = NULL;
1371                         dst_ops->negative_advice = NULL;
1372                         dst_ops->link_failure = NULL;
1373                         afinfo->garbage_collect = NULL;
1374                 }
1375         }
1376         write_unlock_bh(&xfrm_policy_afinfo_lock);
1377         return err;
1378 }
1379 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1380
1381 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1382 {
1383         struct xfrm_policy_afinfo *afinfo;
1384         if (unlikely(family >= NPROTO))
1385                 return NULL;
1386         read_lock(&xfrm_policy_afinfo_lock);
1387         afinfo = xfrm_policy_afinfo[family];
1388         if (unlikely(!afinfo))
1389                 read_unlock(&xfrm_policy_afinfo_lock);
1390         return afinfo;
1391 }
1392
1393 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1394 {
1395         read_unlock(&xfrm_policy_afinfo_lock);
1396 }
1397
1398 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
1399 {
1400         struct xfrm_policy_afinfo *afinfo;
1401         if (unlikely(family >= NPROTO))
1402                 return NULL;
1403         write_lock_bh(&xfrm_policy_afinfo_lock);
1404         afinfo = xfrm_policy_afinfo[family];
1405         if (unlikely(!afinfo))
1406                 write_unlock_bh(&xfrm_policy_afinfo_lock);
1407         return afinfo;
1408 }
1409
1410 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
1411 {
1412         write_unlock_bh(&xfrm_policy_afinfo_lock);
1413 }
1414
1415 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1416 {
1417         switch (event) {
1418         case NETDEV_DOWN:
1419                 xfrm_flush_bundles();
1420         }
1421         return NOTIFY_DONE;
1422 }
1423
1424 static struct notifier_block xfrm_dev_notifier = {
1425         xfrm_dev_event,
1426         NULL,
1427         0
1428 };
1429
1430 static void __init xfrm_policy_init(void)
1431 {
1432         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1433                                            sizeof(struct xfrm_dst),
1434                                            0, SLAB_HWCACHE_ALIGN,
1435                                            NULL, NULL);
1436         if (!xfrm_dst_cache)
1437                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1438
1439         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1440         register_netdevice_notifier(&xfrm_dev_notifier);
1441 }
1442
1443 void __init xfrm_init(void)
1444 {
1445         xfrm_state_init();
1446         xfrm_policy_init();
1447         xfrm_input_init();
1448 }
1449