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