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/slab.h>
  17 #include <linux/kmod.h>
  18 #include <linux/list.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/workqueue.h>
  21 #include <linux/notifier.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/netfilter.h>
  24 #include <linux/module.h>
  25 #include <net/xfrm.h>
  26 #include <net/ip.h>
  27
  28 DEFINE_MUTEX(xfrm_cfg_mutex);
  29 EXPORT_SYMBOL(xfrm_cfg_mutex);
  30
  31 static DEFINE_RWLOCK(xfrm_policy_lock);
  32
  33 struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2];
  34 EXPORT_SYMBOL(xfrm_policy_list);
  35
  36 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  37 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  38
  39 static kmem_cache_t *xfrm_dst_cache __read_mostly;
  40
  41 static struct work_struct xfrm_policy_gc_work;
  42 static struct list_head xfrm_policy_gc_list =
  43         LIST_HEAD_INIT(xfrm_policy_gc_list);
  44 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  45
  46 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  47 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  48 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
  49 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
  50
  51 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
  52 {
  53         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  54         struct xfrm_type **typemap;
  55         int err = 0;
  56
  57         if (unlikely(afinfo == NULL))
  58                 return -EAFNOSUPPORT;
  59         typemap = afinfo->type_map;
  60
  61         if (likely(typemap[type->proto] == NULL))
  62                 typemap[type->proto] = type;
  63         else
  64                 err = -EEXIST;
  65         xfrm_policy_unlock_afinfo(afinfo);
  66         return err;
  67 }
  68 EXPORT_SYMBOL(xfrm_register_type);
  69
  70 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
  71 {
  72         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  73         struct xfrm_type **typemap;
  74         int err = 0;
  75
  76         if (unlikely(afinfo == NULL))
  77                 return -EAFNOSUPPORT;
  78         typemap = afinfo->type_map;
  79
  80         if (unlikely(typemap[type->proto] != type))
  81                 err = -ENOENT;
  82         else
  83                 typemap[type->proto] = NULL;
  84         xfrm_policy_unlock_afinfo(afinfo);
  85         return err;
  86 }
  87 EXPORT_SYMBOL(xfrm_unregister_type);
  88
  89 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
  90 {
  91         struct xfrm_policy_afinfo *afinfo;
  92         struct xfrm_type **typemap;
  93         struct xfrm_type *type;
  94         int modload_attempted = 0;
  95
  96 retry:
  97         afinfo = xfrm_policy_get_afinfo(family);
  98         if (unlikely(afinfo == NULL))
  99                 return NULL;
 100         typemap = afinfo->type_map;
 101
 102         type = typemap[proto];
 103         if (unlikely(type && !try_module_get(type->owner)))
 104                 type = NULL;
 105         if (!type && !modload_attempted) {
 106                 xfrm_policy_put_afinfo(afinfo);
 107                 request_module("xfrm-type-%d-%d",
 108                                (int) family, (int) proto);
 109                 modload_attempted = 1;
 110                 goto retry;
 111         }
 112
 113         xfrm_policy_put_afinfo(afinfo);
 114         return type;
 115 }
 116
 117 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
 118                     unsigned short family)
 119 {
 120         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 121         int err = 0;
 122
 123         if (unlikely(afinfo == NULL))
 124                 return -EAFNOSUPPORT;
 125
 126         if (likely(afinfo->dst_lookup != NULL))
 127                 err = afinfo->dst_lookup(dst, fl);
 128         else
 129                 err = -EINVAL;
 130         xfrm_policy_put_afinfo(afinfo);
 131         return err;
 132 }
 133 EXPORT_SYMBOL(xfrm_dst_lookup);
 134
 135 void xfrm_put_type(struct xfrm_type *type)
 136 {
 137         module_put(type->owner);
 138 }
 139
 140 int xfrm_register_mode(struct xfrm_mode *mode, int family)
 141 {
 142         struct xfrm_policy_afinfo *afinfo;
 143         struct xfrm_mode **modemap;
 144         int err;
 145
 146         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 147                 return -EINVAL;
 148
 149         afinfo = xfrm_policy_lock_afinfo(family);
 150         if (unlikely(afinfo == NULL))
 151                 return -EAFNOSUPPORT;
 152
 153         err = -EEXIST;
 154         modemap = afinfo->mode_map;
 155         if (likely(modemap[mode->encap] == NULL)) {
 156                 modemap[mode->encap] = mode;
 157                 err = 0;
 158         }
 159
 160         xfrm_policy_unlock_afinfo(afinfo);
 161         return err;
 162 }
 163 EXPORT_SYMBOL(xfrm_register_mode);
 164
 165 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
 166 {
 167         struct xfrm_policy_afinfo *afinfo;
 168         struct xfrm_mode **modemap;
 169         int err;
 170
 171         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 172                 return -EINVAL;
 173
 174         afinfo = xfrm_policy_lock_afinfo(family);
 175         if (unlikely(afinfo == NULL))
 176                 return -EAFNOSUPPORT;
 177
 178         err = -ENOENT;
 179         modemap = afinfo->mode_map;
 180         if (likely(modemap[mode->encap] == mode)) {
 181                 modemap[mode->encap] = NULL;
 182                 err = 0;
 183         }
 184
 185         xfrm_policy_unlock_afinfo(afinfo);
 186         return err;
 187 }
 188 EXPORT_SYMBOL(xfrm_unregister_mode);
 189
 190 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
 191 {
 192         struct xfrm_policy_afinfo *afinfo;
 193         struct xfrm_mode *mode;
 194         int modload_attempted = 0;
 195
 196         if (unlikely(encap >= XFRM_MODE_MAX))
 197                 return NULL;
 198
 199 retry:
 200         afinfo = xfrm_policy_get_afinfo(family);
 201         if (unlikely(afinfo == NULL))
 202                 return NULL;
 203
 204         mode = afinfo->mode_map[encap];
 205         if (unlikely(mode && !try_module_get(mode->owner)))
 206                 mode = NULL;
 207         if (!mode && !modload_attempted) {
 208                 xfrm_policy_put_afinfo(afinfo);
 209                 request_module("xfrm-mode-%d-%d", family, encap);
 210                 modload_attempted = 1;
 211                 goto retry;
 212         }
 213
 214         xfrm_policy_put_afinfo(afinfo);
 215         return mode;
 216 }
 217
 218 void xfrm_put_mode(struct xfrm_mode *mode)
 219 {
 220         module_put(mode->owner);
 221 }
 222
 223 static inline unsigned long make_jiffies(long secs)
 224 {
 225         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 226                 return MAX_SCHEDULE_TIMEOUT-1;
 227         else
 228                 return secs*HZ;
 229 }
 230
 231 static void xfrm_policy_timer(unsigned long data)
 232 {
 233         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 234         unsigned long now = (unsigned long)xtime.tv_sec;
 235         long next = LONG_MAX;
 236         int warn = 0;
 237         int dir;
 238
 239         read_lock(&xp->lock);
 240
 241         if (xp->dead)
 242                 goto out;
 243
 244         dir = xfrm_policy_id2dir(xp->index);
 245
 246         if (xp->lft.hard_add_expires_seconds) {
 247                 long tmo = xp->lft.hard_add_expires_seconds +
 248                         xp->curlft.add_time - now;
 249                 if (tmo <= 0)
 250                         goto expired;
 251                 if (tmo < next)
 252                         next = tmo;
 253         }
 254         if (xp->lft.hard_use_expires_seconds) {
 255                 long tmo = xp->lft.hard_use_expires_seconds +
 256                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 257                 if (tmo <= 0)
 258                         goto expired;
 259                 if (tmo < next)
 260                         next = tmo;
 261         }
 262         if (xp->lft.soft_add_expires_seconds) {
 263                 long tmo = xp->lft.soft_add_expires_seconds +
 264                         xp->curlft.add_time - now;
 265                 if (tmo <= 0) {
 266                         warn = 1;
 267                         tmo = XFRM_KM_TIMEOUT;
 268                 }
 269                 if (tmo < next)
 270                         next = tmo;
 271         }
 272         if (xp->lft.soft_use_expires_seconds) {
 273                 long tmo = xp->lft.soft_use_expires_seconds +
 274                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 275                 if (tmo <= 0) {
 276                         warn = 1;
 277                         tmo = XFRM_KM_TIMEOUT;
 278                 }
 279                 if (tmo < next)
 280                         next = tmo;
 281         }
 282
 283         if (warn)
 284                 km_policy_expired(xp, dir, 0, 0);
 285         if (next != LONG_MAX &&
 286             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 287                 xfrm_pol_hold(xp);
 288
 289 out:
 290         read_unlock(&xp->lock);
 291         xfrm_pol_put(xp);
 292         return;
 293
 294 expired:
 295         read_unlock(&xp->lock);
 296         if (!xfrm_policy_delete(xp, dir))
 297                 km_policy_expired(xp, dir, 1, 0);
 298         xfrm_pol_put(xp);
 299 }
 300
 301
 302 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 303  * SPD calls.
 304  */
 305
 306 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 307 {
 308         struct xfrm_policy *policy;
 309
 310         policy = kmalloc(sizeof(struct xfrm_policy), gfp);
 311
 312         if (policy) {
 313                 memset(policy, 0, sizeof(struct xfrm_policy));
 314                 atomic_set(&policy->refcnt, 1);
 315                 rwlock_init(&policy->lock);
 316                 init_timer(&policy->timer);
 317                 policy->timer.data = (unsigned long)policy;
 318                 policy->timer.function = xfrm_policy_timer;
 319         }
 320         return policy;
 321 }
 322 EXPORT_SYMBOL(xfrm_policy_alloc);
 323
 324 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 325
 326 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 327 {
 328         BUG_ON(!policy->dead);
 329
 330         BUG_ON(policy->bundles);
 331
 332         if (del_timer(&policy->timer))
 333                 BUG();
 334
 335         security_xfrm_policy_free(policy);
 336         kfree(policy);
 337 }
 338 EXPORT_SYMBOL(__xfrm_policy_destroy);
 339
 340 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 341 {
 342         struct dst_entry *dst;
 343
 344         while ((dst = policy->bundles) != NULL) {
 345                 policy->bundles = dst->next;
 346                 dst_free(dst);
 347         }
 348
 349         if (del_timer(&policy->timer))
 350                 atomic_dec(&policy->refcnt);
 351
 352         if (atomic_read(&policy->refcnt) > 1)
 353                 flow_cache_flush();
 354
 355         xfrm_pol_put(policy);
 356 }
 357
 358 static void xfrm_policy_gc_task(void *data)
 359 {
 360         struct xfrm_policy *policy;
 361         struct list_head *entry, *tmp;
 362         struct list_head gc_list = LIST_HEAD_INIT(gc_list);
 363
 364         spin_lock_bh(&xfrm_policy_gc_lock);
 365         list_splice_init(&xfrm_policy_gc_list, &gc_list);
 366         spin_unlock_bh(&xfrm_policy_gc_lock);
 367
 368         list_for_each_safe(entry, tmp, &gc_list) {
 369                 policy = list_entry(entry, struct xfrm_policy, list);
 370                 xfrm_policy_gc_kill(policy);
 371         }
 372 }
 373
 374 /* Rule must be locked. Release descentant resources, announce
 375  * entry dead. The rule must be unlinked from lists to the moment.
 376  */
 377
 378 static void xfrm_policy_kill(struct xfrm_policy *policy)
 379 {
 380         int dead;
 381
 382         write_lock_bh(&policy->lock);
 383         dead = policy->dead;
 384         policy->dead = 1;
 385         write_unlock_bh(&policy->lock);
 386
 387         if (unlikely(dead)) {
 388                 WARN_ON(1);
 389                 return;
 390         }
 391
 392         spin_lock(&xfrm_policy_gc_lock);
 393         list_add(&policy->list, &xfrm_policy_gc_list);
 394         spin_unlock(&xfrm_policy_gc_lock);
 395
 396         schedule_work(&xfrm_policy_gc_work);
 397 }
 398
 399 /* Generate new index... KAME seems to generate them ordered by cost
 400  * of an absolute inpredictability of ordering of rules. This will not pass. */
 401 static u32 xfrm_gen_index(int dir)
 402 {
 403         u32 idx;
 404         struct xfrm_policy *p;
 405         static u32 idx_generator;
 406
 407         for (;;) {
 408                 idx = (idx_generator | dir);
 409                 idx_generator += 8;
 410                 if (idx == 0)
 411                         idx = 8;
 412                 for (p = xfrm_policy_list[dir]; p; p = p->next) {
 413                         if (p->index == idx)
 414                                 break;
 415                 }
 416                 if (!p)
 417                         return idx;
 418         }
 419 }
 420
 421 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 422 {
 423         struct xfrm_policy *pol, **p;
 424         struct xfrm_policy *delpol = NULL;
 425         struct xfrm_policy **newpos = NULL;
 426         struct dst_entry *gc_list;
 427
 428         write_lock_bh(&xfrm_policy_lock);
 429         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
 430                 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0 &&
 431                     xfrm_sec_ctx_match(pol->security, policy->security)) {
 432                         if (excl) {
 433                                 write_unlock_bh(&xfrm_policy_lock);
 434                                 return -EEXIST;
 435                         }
 436                         *p = pol->next;
 437                         delpol = pol;
 438                         if (policy->priority > pol->priority)
 439                                 continue;
 440                 } else if (policy->priority >= pol->priority) {
 441                         p = &pol->next;
 442                         continue;
 443                 }
 444                 if (!newpos)
 445                         newpos = p;
 446                 if (delpol)
 447                         break;
 448                 p = &pol->next;
 449         }
 450         if (newpos)
 451                 p = newpos;
 452         xfrm_pol_hold(policy);
 453         policy->next = *p;
 454         *p = policy;
 455         atomic_inc(&flow_cache_genid);
 456         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
 457         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
 458         policy->curlft.use_time = 0;
 459         if (!mod_timer(&policy->timer, jiffies + HZ))
 460                 xfrm_pol_hold(policy);
 461         write_unlock_bh(&xfrm_policy_lock);
 462
 463         if (delpol)
 464                 xfrm_policy_kill(delpol);
 465
 466         read_lock_bh(&xfrm_policy_lock);
 467         gc_list = NULL;
 468         for (policy = policy->next; policy; policy = policy->next) {
 469                 struct dst_entry *dst;
 470
 471                 write_lock(&policy->lock);
 472                 dst = policy->bundles;
 473                 if (dst) {
 474                         struct dst_entry *tail = dst;
 475                         while (tail->next)
 476                                 tail = tail->next;
 477                         tail->next = gc_list;
 478                         gc_list = dst;
 479
 480                         policy->bundles = NULL;
 481                 }
 482                 write_unlock(&policy->lock);
 483         }
 484         read_unlock_bh(&xfrm_policy_lock);
 485
 486         while (gc_list) {
 487                 struct dst_entry *dst = gc_list;
 488
 489                 gc_list = dst->next;
 490                 dst_free(dst);
 491         }
 492
 493         return 0;
 494 }
 495 EXPORT_SYMBOL(xfrm_policy_insert);
 496
 497 struct xfrm_policy *xfrm_policy_bysel_ctx(int dir, struct xfrm_selector *sel,
 498                                           struct xfrm_sec_ctx *ctx, int delete)
 499 {
 500         struct xfrm_policy *pol, **p;
 501
 502         write_lock_bh(&xfrm_policy_lock);
 503         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 504                 if ((memcmp(sel, &pol->selector, sizeof(*sel)) == 0) &&
 505                     (xfrm_sec_ctx_match(ctx, pol->security))) {
 506                         xfrm_pol_hold(pol);
 507                         if (delete)
 508                                 *p = pol->next;
 509                         break;
 510                 }
 511         }
 512         write_unlock_bh(&xfrm_policy_lock);
 513
 514         if (pol && delete) {
 515                 atomic_inc(&flow_cache_genid);
 516                 xfrm_policy_kill(pol);
 517         }
 518         return pol;
 519 }
 520 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 521
 522 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
 523 {
 524         struct xfrm_policy *pol, **p;
 525
 526         write_lock_bh(&xfrm_policy_lock);
 527         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
 528                 if (pol->index == id) {
 529                         xfrm_pol_hold(pol);
 530                         if (delete)
 531                                 *p = pol->next;
 532                         break;
 533                 }
 534         }
 535         write_unlock_bh(&xfrm_policy_lock);
 536
 537         if (pol && delete) {
 538                 atomic_inc(&flow_cache_genid);
 539                 xfrm_policy_kill(pol);
 540         }
 541         return pol;
 542 }
 543 EXPORT_SYMBOL(xfrm_policy_byid);
 544
 545 void xfrm_policy_flush(void)
 546 {
 547         struct xfrm_policy *xp;
 548         int dir;
 549
 550         write_lock_bh(&xfrm_policy_lock);
 551         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 552                 while ((xp = xfrm_policy_list[dir]) != NULL) {
 553                         xfrm_policy_list[dir] = xp->next;
 554                         write_unlock_bh(&xfrm_policy_lock);
 555
 556                         xfrm_policy_kill(xp);
 557
 558                         write_lock_bh(&xfrm_policy_lock);
 559                 }
 560         }
 561         atomic_inc(&flow_cache_genid);
 562         write_unlock_bh(&xfrm_policy_lock);
 563 }
 564 EXPORT_SYMBOL(xfrm_policy_flush);
 565
 566 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
 567                      void *data)
 568 {
 569         struct xfrm_policy *xp;
 570         int dir;
 571         int count = 0;
 572         int error = 0;
 573
 574         read_lock_bh(&xfrm_policy_lock);
 575         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 576                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
 577                         count++;
 578         }
 579
 580         if (count == 0) {
 581                 error = -ENOENT;
 582                 goto out;
 583         }
 584
 585         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 586                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
 587                         error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
 588                         if (error)
 589                                 goto out;
 590                 }
 591         }
 592
 593 out:
 594         read_unlock_bh(&xfrm_policy_lock);
 595         return error;
 596 }
 597 EXPORT_SYMBOL(xfrm_policy_walk);
 598
 599 /* Find policy to apply to this flow. */
 600
 601 static void xfrm_policy_lookup(struct flowi *fl, u32 sk_sid, u16 family, u8 dir,
 602                                void **objp, atomic_t **obj_refp)
 603 {
 604         struct xfrm_policy *pol;
 605
 606         read_lock_bh(&xfrm_policy_lock);
 607         for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
 608                 struct xfrm_selector *sel = &pol->selector;
 609                 int match;
 610
 611                 if (pol->family != family)
 612                         continue;
 613
 614                 match = xfrm_selector_match(sel, fl, family);
 615
 616                 if (match) {
 617                         if (!security_xfrm_policy_lookup(pol, sk_sid, dir)) {
 618                                 xfrm_pol_hold(pol);
 619                                 break;
 620                         }
 621                 }
 622         }
 623         read_unlock_bh(&xfrm_policy_lock);
 624         if ((*objp = (void *) pol) != NULL)
 625                 *obj_refp = &pol->refcnt;
 626 }
 627
 628 static inline int policy_to_flow_dir(int dir)
 629 {
 630         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 631             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 632             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 633                 return dir;
 634         switch (dir) {
 635         default:
 636         case XFRM_POLICY_IN:
 637                 return FLOW_DIR_IN;
 638         case XFRM_POLICY_OUT:
 639                 return FLOW_DIR_OUT;
 640         case XFRM_POLICY_FWD:
 641                 return FLOW_DIR_FWD;
 642         };
 643 }
 644
 645 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl, u32 sk_sid)
 646 {
 647         struct xfrm_policy *pol;
 648
 649         read_lock_bh(&xfrm_policy_lock);
 650         if ((pol = sk->sk_policy[dir]) != NULL) {
 651                 int match = xfrm_selector_match(&pol->selector, fl,
 652                                                 sk->sk_family);
 653                 int err = 0;
 654
 655                 if (match)
 656                   err = security_xfrm_policy_lookup(pol, sk_sid, policy_to_flow_dir(dir));
 657
 658                 if (match && !err)
 659                         xfrm_pol_hold(pol);
 660                 else
 661                         pol = NULL;
 662         }
 663         read_unlock_bh(&xfrm_policy_lock);
 664         return pol;
 665 }
 666
 667 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
 668 {
 669         pol->next = xfrm_policy_list[dir];
 670         xfrm_policy_list[dir] = pol;
 671         xfrm_pol_hold(pol);
 672 }
 673
 674 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 675                                                 int dir)
 676 {
 677         struct xfrm_policy **polp;
 678
 679         for (polp = &xfrm_policy_list[dir];
 680              *polp != NULL; polp = &(*polp)->next) {
 681                 if (*polp == pol) {
 682                         *polp = pol->next;
 683                         return pol;
 684                 }
 685         }
 686         return NULL;
 687 }
 688
 689 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
 690 {
 691         write_lock_bh(&xfrm_policy_lock);
 692         pol = __xfrm_policy_unlink(pol, dir);
 693         write_unlock_bh(&xfrm_policy_lock);
 694         if (pol) {
 695                 if (dir < XFRM_POLICY_MAX)
 696                         atomic_inc(&flow_cache_genid);
 697                 xfrm_policy_kill(pol);
 698                 return 0;
 699         }
 700         return -ENOENT;
 701 }
 702 EXPORT_SYMBOL(xfrm_policy_delete);
 703
 704 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
 705 {
 706         struct xfrm_policy *old_pol;
 707
 708         write_lock_bh(&xfrm_policy_lock);
 709         old_pol = sk->sk_policy[dir];
 710         sk->sk_policy[dir] = pol;
 711         if (pol) {
 712                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
 713                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
 714                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
 715         }
 716         if (old_pol)
 717                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
 718         write_unlock_bh(&xfrm_policy_lock);
 719
 720         if (old_pol) {
 721                 xfrm_policy_kill(old_pol);
 722         }
 723         return 0;
 724 }
 725
 726 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
 727 {
 728         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
 729
 730         if (newp) {
 731                 newp->selector = old->selector;
 732                 if (security_xfrm_policy_clone(old, newp)) {
 733                         kfree(newp);
 734                         return NULL;  /* ENOMEM */
 735                 }
 736                 newp->lft = old->lft;
 737                 newp->curlft = old->curlft;
 738                 newp->action = old->action;
 739                 newp->flags = old->flags;
 740                 newp->xfrm_nr = old->xfrm_nr;
 741                 newp->index = old->index;
 742                 memcpy(newp->xfrm_vec, old->xfrm_vec,
 743                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
 744                 write_lock_bh(&xfrm_policy_lock);
 745                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
 746                 write_unlock_bh(&xfrm_policy_lock);
 747                 xfrm_pol_put(newp);
 748         }
 749         return newp;
 750 }
 751
 752 int __xfrm_sk_clone_policy(struct sock *sk)
 753 {
 754         struct xfrm_policy *p0 = sk->sk_policy[0],
 755                            *p1 = sk->sk_policy[1];
 756
 757         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
 758         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
 759                 return -ENOMEM;
 760         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
 761                 return -ENOMEM;
 762         return 0;
 763 }
 764
 765 /* Resolve list of templates for the flow, given policy. */
 766
 767 static int
 768 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
 769                   struct xfrm_state **xfrm,
 770                   unsigned short family)
 771 {
 772         int nx;
 773         int i, error;
 774         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
 775         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
 776
 777         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
 778                 struct xfrm_state *x;
 779                 xfrm_address_t *remote = daddr;
 780                 xfrm_address_t *local  = saddr;
 781                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
 782
 783                 if (tmpl->mode) {
 784                         remote = &tmpl->id.daddr;
 785                         local = &tmpl->saddr;
 786                 }
 787
 788                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
 789
 790                 if (x && x->km.state == XFRM_STATE_VALID) {
 791                         xfrm[nx++] = x;
 792                         daddr = remote;
 793                         saddr = local;
 794                         continue;
 795                 }
 796                 if (x) {
 797                         error = (x->km.state == XFRM_STATE_ERROR ?
 798                                  -EINVAL : -EAGAIN);
 799                         xfrm_state_put(x);
 800                 }
 801
 802                 if (!tmpl->optional)
 803                         goto fail;
 804         }
 805         return nx;
 806
 807 fail:
 808         for (nx--; nx>=0; nx--)
 809                 xfrm_state_put(xfrm[nx]);
 810         return error;
 811 }
 812
 813 /* Check that the bundle accepts the flow and its components are
 814  * still valid.
 815  */
 816
 817 static struct dst_entry *
 818 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
 819 {
 820         struct dst_entry *x;
 821         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 822         if (unlikely(afinfo == NULL))
 823                 return ERR_PTR(-EINVAL);
 824         x = afinfo->find_bundle(fl, policy);
 825         xfrm_policy_put_afinfo(afinfo);
 826         return x;
 827 }
 828
 829 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
 830  * all the metrics... Shortly, bundle a bundle.
 831  */
 832
 833 static int
 834 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
 835                    struct flowi *fl, struct dst_entry **dst_p,
 836                    unsigned short family)
 837 {
 838         int err;
 839         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 840         if (unlikely(afinfo == NULL))
 841                 return -EINVAL;
 842         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
 843         xfrm_policy_put_afinfo(afinfo);
 844         return err;
 845 }
 846
 847
 848 static int stale_bundle(struct dst_entry *dst);
 849
 850 /* Main function: finds/creates a bundle for given flow.
 851  *
 852  * At the moment we eat a raw IP route. Mostly to speed up lookups
 853  * on interfaces with disabled IPsec.
 854  */
 855 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
 856                 struct sock *sk, int flags)
 857 {
 858         struct xfrm_policy *policy;
 859         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
 860         struct dst_entry *dst, *dst_orig = *dst_p;
 861         int nx = 0;
 862         int err;
 863         u32 genid;
 864         u16 family;
 865         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
 866         u32 sk_sid = security_sk_sid(sk, fl, dir);
 867 restart:
 868         genid = atomic_read(&flow_cache_genid);
 869         policy = NULL;
 870         if (sk && sk->sk_policy[1])
 871                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, sk_sid);
 872
 873         if (!policy) {
 874                 /* To accelerate a bit...  */
 875                 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
 876                         return 0;
 877
 878                 policy = flow_cache_lookup(fl, sk_sid, dst_orig->ops->family,
 879                                            dir, xfrm_policy_lookup);
 880         }
 881
 882         if (!policy)
 883                 return 0;
 884
 885         family = dst_orig->ops->family;
 886         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
 887
 888         switch (policy->action) {
 889         case XFRM_POLICY_BLOCK:
 890                 /* Prohibit the flow */
 891                 err = -EPERM;
 892                 goto error;
 893
 894         case XFRM_POLICY_ALLOW:
 895                 if (policy->xfrm_nr == 0) {
 896                         /* Flow passes not transformed. */
 897                         xfrm_pol_put(policy);
 898                         return 0;
 899                 }
 900
 901                 /* Try to find matching bundle.
 902                  *
 903                  * LATER: help from flow cache. It is optional, this
 904                  * is required only for output policy.
 905                  */
 906                 dst = xfrm_find_bundle(fl, policy, family);
 907                 if (IS_ERR(dst)) {
 908                         err = PTR_ERR(dst);
 909                         goto error;
 910                 }
 911
 912                 if (dst)
 913                         break;
 914
 915                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 916
 917                 if (unlikely(nx<0)) {
 918                         err = nx;
 919                         if (err == -EAGAIN && flags) {
 920                                 DECLARE_WAITQUEUE(wait, current);
 921
 922                                 add_wait_queue(&km_waitq, &wait);
 923                                 set_current_state(TASK_INTERRUPTIBLE);
 924                                 schedule();
 925                                 set_current_state(TASK_RUNNING);
 926                                 remove_wait_queue(&km_waitq, &wait);
 927
 928                                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
 929
 930                                 if (nx == -EAGAIN && signal_pending(current)) {
 931                                         err = -ERESTART;
 932                                         goto error;
 933                                 }
 934                                 if (nx == -EAGAIN ||
 935                                     genid != atomic_read(&flow_cache_genid)) {
 936                                         xfrm_pol_put(policy);
 937                                         goto restart;
 938                                 }
 939                                 err = nx;
 940                         }
 941                         if (err < 0)
 942                                 goto error;
 943                 }
 944                 if (nx == 0) {
 945                         /* Flow passes not transformed. */
 946                         xfrm_pol_put(policy);
 947                         return 0;
 948                 }
 949
 950                 dst = dst_orig;
 951                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
 952
 953                 if (unlikely(err)) {
 954                         int i;
 955                         for (i=0; i<nx; i++)
 956                                 xfrm_state_put(xfrm[i]);
 957                         goto error;
 958                 }
 959
 960                 write_lock_bh(&policy->lock);
 961                 if (unlikely(policy->dead || stale_bundle(dst))) {
 962                         /* Wow! While we worked on resolving, this
 963                          * policy has gone. Retry. It is not paranoia,
 964                          * we just cannot enlist new bundle to dead object.
 965                          * We can't enlist stable bundles either.
 966                          */
 967                         write_unlock_bh(&policy->lock);
 968                         if (dst)
 969                                 dst_free(dst);
 970
 971                         err = -EHOSTUNREACH;
 972                         goto error;
 973                 }
 974                 dst->next = policy->bundles;
 975                 policy->bundles = dst;
 976                 dst_hold(dst);
 977                 write_unlock_bh(&policy->lock);
 978         }
 979         *dst_p = dst;
 980         dst_release(dst_orig);
 981         xfrm_pol_put(policy);
 982         return 0;
 983
 984 error:
 985         dst_release(dst_orig);
 986         xfrm_pol_put(policy);
 987         *dst_p = NULL;
 988         return err;
 989 }
 990 EXPORT_SYMBOL(xfrm_lookup);
 991
 992 /* When skb is transformed back to its "native" form, we have to
 993  * check policy restrictions. At the moment we make this in maximally
 994  * stupid way. Shame on me. :-) Of course, connected sockets must
 995  * have policy cached at them.
 996  */
 997
 998 static inline int
 999 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1000               unsigned short family)
1001 {
1002         if (xfrm_state_kern(x))
1003                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1004         return  x->id.proto == tmpl->id.proto &&
1005                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1006                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1007                 x->props.mode == tmpl->mode &&
1008                 (tmpl->aalgos & (1<<x->props.aalgo)) &&
1009                 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
1010 }
1011
1012 static inline int
1013 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1014                unsigned short family)
1015 {
1016         int idx = start;
1017
1018         if (tmpl->optional) {
1019                 if (!tmpl->mode)
1020                         return start;
1021         } else
1022                 start = -1;
1023         for (; idx < sp->len; idx++) {
1024                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1025                         return ++idx;
1026                 if (sp->xvec[idx]->props.mode)
1027                         break;
1028         }
1029         return start;
1030 }
1031
1032 int
1033 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1034 {
1035         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1036
1037         if (unlikely(afinfo == NULL))
1038                 return -EAFNOSUPPORT;
1039
1040         afinfo->decode_session(skb, fl);
1041         xfrm_policy_put_afinfo(afinfo);
1042         return 0;
1043 }
1044 EXPORT_SYMBOL(xfrm_decode_session);
1045
1046 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
1047 {
1048         for (; k < sp->len; k++) {
1049                 if (sp->xvec[k]->props.mode)
1050                         return 1;
1051         }
1052
1053         return 0;
1054 }
1055
1056 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1057                         unsigned short family)
1058 {
1059         struct xfrm_policy *pol;
1060         struct flowi fl;
1061         u8 fl_dir = policy_to_flow_dir(dir);
1062         u32 sk_sid;
1063
1064         if (xfrm_decode_session(skb, &fl, family) < 0)
1065                 return 0;
1066         nf_nat_decode_session(skb, &fl, family);
1067
1068         sk_sid = security_sk_sid(sk, &fl, fl_dir);
1069
1070         /* First, check used SA against their selectors. */
1071         if (skb->sp) {
1072                 int i;
1073
1074                 for (i=skb->sp->len-1; i>=0; i--) {
1075                         struct xfrm_state *x = skb->sp->xvec[i];
1076                         if (!xfrm_selector_match(&x->sel, &fl, family))
1077                                 return 0;
1078                 }
1079         }
1080
1081         pol = NULL;
1082         if (sk && sk->sk_policy[dir])
1083                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, sk_sid);
1084
1085         if (!pol)
1086                 pol = flow_cache_lookup(&fl, sk_sid, family, fl_dir,
1087                                         xfrm_policy_lookup);
1088
1089         if (!pol)
1090                 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
1091
1092         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1093
1094         if (pol->action == XFRM_POLICY_ALLOW) {
1095                 struct sec_path *sp;
1096                 static struct sec_path dummy;
1097                 int i, k;
1098
1099                 if ((sp = skb->sp) == NULL)
1100                         sp = &dummy;
1101
1102                 /* For each tunnel xfrm, find the first matching tmpl.
1103                  * For each tmpl before that, find corresponding xfrm.
1104                  * Order is _important_. Later we will implement
1105                  * some barriers, but at the moment barriers
1106                  * are implied between each two transformations.
1107                  */
1108                 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
1109                         k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
1110                         if (k < 0)
1111                                 goto reject;
1112                 }
1113
1114                 if (secpath_has_tunnel(sp, k))
1115                         goto reject;
1116
1117                 xfrm_pol_put(pol);
1118                 return 1;
1119         }
1120
1121 reject:
1122         xfrm_pol_put(pol);
1123         return 0;
1124 }
1125 EXPORT_SYMBOL(__xfrm_policy_check);
1126
1127 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1128 {
1129         struct flowi fl;
1130
1131         if (xfrm_decode_session(skb, &fl, family) < 0)
1132                 return 0;
1133
1134         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1135 }
1136 EXPORT_SYMBOL(__xfrm_route_forward);
1137
1138 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1139 {
1140         /* If it is marked obsolete, which is how we even get here,
1141          * then we have purged it from the policy bundle list and we
1142          * did that for a good reason.
1143          */
1144         return NULL;
1145 }
1146
1147 static int stale_bundle(struct dst_entry *dst)
1148 {
1149         return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1150 }
1151
1152 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1153 {
1154         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1155                 dst->dev = &loopback_dev;
1156                 dev_hold(&loopback_dev);
1157                 dev_put(dev);
1158         }
1159 }
1160 EXPORT_SYMBOL(xfrm_dst_ifdown);
1161
1162 static void xfrm_link_failure(struct sk_buff *skb)
1163 {
1164         /* Impossible. Such dst must be popped before reaches point of failure. */
1165         return;
1166 }
1167
1168 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1169 {
1170         if (dst) {
1171                 if (dst->obsolete) {
1172                         dst_release(dst);
1173                         dst = NULL;
1174                 }
1175         }
1176         return dst;
1177 }
1178
1179 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1180 {
1181         int i;
1182         struct xfrm_policy *pol;
1183         struct dst_entry *dst, **dstp, *gc_list = NULL;
1184
1185         read_lock_bh(&xfrm_policy_lock);
1186         for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1187                 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1188                         write_lock(&pol->lock);
1189                         dstp = &pol->bundles;
1190                         while ((dst=*dstp) != NULL) {
1191                                 if (func(dst)) {
1192                                         *dstp = dst->next;
1193                                         dst->next = gc_list;
1194                                         gc_list = dst;
1195                                 } else {
1196                                         dstp = &dst->next;
1197                                 }
1198                         }
1199                         write_unlock(&pol->lock);
1200                 }
1201         }
1202         read_unlock_bh(&xfrm_policy_lock);
1203
1204         while (gc_list) {
1205                 dst = gc_list;
1206                 gc_list = dst->next;
1207                 dst_free(dst);
1208         }
1209 }
1210
1211 static int unused_bundle(struct dst_entry *dst)
1212 {
1213         return !atomic_read(&dst->__refcnt);
1214 }
1215
1216 static void __xfrm_garbage_collect(void)
1217 {
1218         xfrm_prune_bundles(unused_bundle);
1219 }
1220
1221 int xfrm_flush_bundles(void)
1222 {
1223         xfrm_prune_bundles(stale_bundle);
1224         return 0;
1225 }
1226
1227 static int always_true(struct dst_entry *dst)
1228 {
1229         return 1;
1230 }
1231
1232 void xfrm_flush_all_bundles(void)
1233 {
1234         xfrm_prune_bundles(always_true);
1235 }
1236
1237 void xfrm_init_pmtu(struct dst_entry *dst)
1238 {
1239         do {
1240                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1241                 u32 pmtu, route_mtu_cached;
1242
1243                 pmtu = dst_mtu(dst->child);
1244                 xdst->child_mtu_cached = pmtu;
1245
1246                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1247
1248                 route_mtu_cached = dst_mtu(xdst->route);
1249                 xdst->route_mtu_cached = route_mtu_cached;
1250
1251                 if (pmtu > route_mtu_cached)
1252                         pmtu = route_mtu_cached;
1253
1254                 dst->metrics[RTAX_MTU-1] = pmtu;
1255         } while ((dst = dst->next));
1256 }
1257
1258 EXPORT_SYMBOL(xfrm_init_pmtu);
1259
1260 /* Check that the bundle accepts the flow and its components are
1261  * still valid.
1262  */
1263
1264 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1265 {
1266         struct dst_entry *dst = &first->u.dst;
1267         struct xfrm_dst *last;
1268         u32 mtu;
1269
1270         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1271             (dst->dev && !netif_running(dst->dev)))
1272                 return 0;
1273
1274         last = NULL;
1275
1276         do {
1277                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1278
1279                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1280                         return 0;
1281                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1282                         return 0;
1283
1284                 mtu = dst_mtu(dst->child);
1285                 if (xdst->child_mtu_cached != mtu) {
1286                         last = xdst;
1287                         xdst->child_mtu_cached = mtu;
1288                 }
1289
1290                 if (!dst_check(xdst->route, xdst->route_cookie))
1291                         return 0;
1292                 mtu = dst_mtu(xdst->route);
1293                 if (xdst->route_mtu_cached != mtu) {
1294                         last = xdst;
1295                         xdst->route_mtu_cached = mtu;
1296                 }
1297
1298                 dst = dst->child;
1299         } while (dst->xfrm);
1300
1301         if (likely(!last))
1302                 return 1;
1303
1304         mtu = last->child_mtu_cached;
1305         for (;;) {
1306                 dst = &last->u.dst;
1307
1308                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1309                 if (mtu > last->route_mtu_cached)
1310                         mtu = last->route_mtu_cached;
1311                 dst->metrics[RTAX_MTU-1] = mtu;
1312
1313                 if (last == first)
1314                         break;
1315
1316                 last = last->u.next;
1317                 last->child_mtu_cached = mtu;
1318         }
1319
1320         return 1;
1321 }
1322
1323 EXPORT_SYMBOL(xfrm_bundle_ok);
1324
1325 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1326 {
1327         int err = 0;
1328         if (unlikely(afinfo == NULL))
1329                 return -EINVAL;
1330         if (unlikely(afinfo->family >= NPROTO))
1331                 return -EAFNOSUPPORT;
1332         write_lock_bh(&xfrm_policy_afinfo_lock);
1333         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1334                 err = -ENOBUFS;
1335         else {
1336                 struct dst_ops *dst_ops = afinfo->dst_ops;
1337                 if (likely(dst_ops->kmem_cachep == NULL))
1338                         dst_ops->kmem_cachep = xfrm_dst_cache;
1339                 if (likely(dst_ops->check == NULL))
1340                         dst_ops->check = xfrm_dst_check;
1341                 if (likely(dst_ops->negative_advice == NULL))
1342                         dst_ops->negative_advice = xfrm_negative_advice;
1343                 if (likely(dst_ops->link_failure == NULL))
1344                         dst_ops->link_failure = xfrm_link_failure;
1345                 if (likely(afinfo->garbage_collect == NULL))
1346                         afinfo->garbage_collect = __xfrm_garbage_collect;
1347                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1348         }
1349         write_unlock_bh(&xfrm_policy_afinfo_lock);
1350         return err;
1351 }
1352 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1353
1354 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1355 {
1356         int err = 0;
1357         if (unlikely(afinfo == NULL))
1358                 return -EINVAL;
1359         if (unlikely(afinfo->family >= NPROTO))
1360                 return -EAFNOSUPPORT;
1361         write_lock_bh(&xfrm_policy_afinfo_lock);
1362         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1363                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1364                         err = -EINVAL;
1365                 else {
1366                         struct dst_ops *dst_ops = afinfo->dst_ops;
1367                         xfrm_policy_afinfo[afinfo->family] = NULL;
1368                         dst_ops->kmem_cachep = NULL;
1369                         dst_ops->check = NULL;
1370                         dst_ops->negative_advice = NULL;
1371                         dst_ops->link_failure = NULL;
1372                         afinfo->garbage_collect = NULL;
1373                 }
1374         }
1375         write_unlock_bh(&xfrm_policy_afinfo_lock);
1376         return err;
1377 }
1378 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1379
1380 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1381 {
1382         struct xfrm_policy_afinfo *afinfo;
1383         if (unlikely(family >= NPROTO))
1384                 return NULL;
1385         read_lock(&xfrm_policy_afinfo_lock);
1386         afinfo = xfrm_policy_afinfo[family];
1387         if (unlikely(!afinfo))
1388                 read_unlock(&xfrm_policy_afinfo_lock);
1389         return afinfo;
1390 }
1391
1392 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1393 {
1394         read_unlock(&xfrm_policy_afinfo_lock);
1395 }
1396
1397 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
1398 {
1399         struct xfrm_policy_afinfo *afinfo;
1400         if (unlikely(family >= NPROTO))
1401                 return NULL;
1402         write_lock_bh(&xfrm_policy_afinfo_lock);
1403         afinfo = xfrm_policy_afinfo[family];
1404         if (unlikely(!afinfo))
1405                 write_unlock_bh(&xfrm_policy_afinfo_lock);
1406         return afinfo;
1407 }
1408
1409 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
1410 {
1411         write_unlock_bh(&xfrm_policy_afinfo_lock);
1412 }
1413
1414 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1415 {
1416         switch (event) {
1417         case NETDEV_DOWN:
1418                 xfrm_flush_bundles();
1419         }
1420         return NOTIFY_DONE;
1421 }
1422
1423 static struct notifier_block xfrm_dev_notifier = {
1424         xfrm_dev_event,
1425         NULL,
1426         0
1427 };
1428
1429 static void __init xfrm_policy_init(void)
1430 {
1431         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1432                                            sizeof(struct xfrm_dst),
1433                                            0, SLAB_HWCACHE_ALIGN,
1434                                            NULL, NULL);
1435         if (!xfrm_dst_cache)
1436                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1437
1438         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1439         register_netdevice_notifier(&xfrm_dev_notifier);
1440 }
1441
1442 void __init xfrm_init(void)
1443 {
1444         xfrm_state_init();
1445         xfrm_policy_init();
1446         xfrm_input_init();
1447 }
1448