err.no Git - linux-2.6/blob - net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/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 <linux/bootmem.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/cache.h>
  28 #include <net/xfrm.h>
  29 #include <net/ip.h>
  30
  31 DEFINE_MUTEX(xfrm_cfg_mutex);
  32 EXPORT_SYMBOL(xfrm_cfg_mutex);
  33
  34 static DEFINE_RWLOCK(xfrm_policy_lock);
  35
  36 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  37 EXPORT_SYMBOL(xfrm_policy_count);
  38
  39 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  40 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  41
  42 static kmem_cache_t *xfrm_dst_cache __read_mostly;
  43
  44 static struct work_struct xfrm_policy_gc_work;
  45 static HLIST_HEAD(xfrm_policy_gc_list);
  46 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  47
  48 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  49 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  50 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
  51 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
  52
  53 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
  54 {
  55         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  56         struct xfrm_type **typemap;
  57         int err = 0;
  58
  59         if (unlikely(afinfo == NULL))
  60                 return -EAFNOSUPPORT;
  61         typemap = afinfo->type_map;
  62
  63         if (likely(typemap[type->proto] == NULL))
  64                 typemap[type->proto] = type;
  65         else
  66                 err = -EEXIST;
  67         xfrm_policy_unlock_afinfo(afinfo);
  68         return err;
  69 }
  70 EXPORT_SYMBOL(xfrm_register_type);
  71
  72 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
  73 {
  74         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  75         struct xfrm_type **typemap;
  76         int err = 0;
  77
  78         if (unlikely(afinfo == NULL))
  79                 return -EAFNOSUPPORT;
  80         typemap = afinfo->type_map;
  81
  82         if (unlikely(typemap[type->proto] != type))
  83                 err = -ENOENT;
  84         else
  85                 typemap[type->proto] = NULL;
  86         xfrm_policy_unlock_afinfo(afinfo);
  87         return err;
  88 }
  89 EXPORT_SYMBOL(xfrm_unregister_type);
  90
  91 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
  92 {
  93         struct xfrm_policy_afinfo *afinfo;
  94         struct xfrm_type **typemap;
  95         struct xfrm_type *type;
  96         int modload_attempted = 0;
  97
  98 retry:
  99         afinfo = xfrm_policy_get_afinfo(family);
 100         if (unlikely(afinfo == NULL))
 101                 return NULL;
 102         typemap = afinfo->type_map;
 103
 104         type = typemap[proto];
 105         if (unlikely(type && !try_module_get(type->owner)))
 106                 type = NULL;
 107         if (!type && !modload_attempted) {
 108                 xfrm_policy_put_afinfo(afinfo);
 109                 request_module("xfrm-type-%d-%d",
 110                                (int) family, (int) proto);
 111                 modload_attempted = 1;
 112                 goto retry;
 113         }
 114
 115         xfrm_policy_put_afinfo(afinfo);
 116         return type;
 117 }
 118
 119 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
 120                     unsigned short family)
 121 {
 122         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 123         int err = 0;
 124
 125         if (unlikely(afinfo == NULL))
 126                 return -EAFNOSUPPORT;
 127
 128         if (likely(afinfo->dst_lookup != NULL))
 129                 err = afinfo->dst_lookup(dst, fl);
 130         else
 131                 err = -EINVAL;
 132         xfrm_policy_put_afinfo(afinfo);
 133         return err;
 134 }
 135 EXPORT_SYMBOL(xfrm_dst_lookup);
 136
 137 void xfrm_put_type(struct xfrm_type *type)
 138 {
 139         module_put(type->owner);
 140 }
 141
 142 int xfrm_register_mode(struct xfrm_mode *mode, int family)
 143 {
 144         struct xfrm_policy_afinfo *afinfo;
 145         struct xfrm_mode **modemap;
 146         int err;
 147
 148         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 149                 return -EINVAL;
 150
 151         afinfo = xfrm_policy_lock_afinfo(family);
 152         if (unlikely(afinfo == NULL))
 153                 return -EAFNOSUPPORT;
 154
 155         err = -EEXIST;
 156         modemap = afinfo->mode_map;
 157         if (likely(modemap[mode->encap] == NULL)) {
 158                 modemap[mode->encap] = mode;
 159                 err = 0;
 160         }
 161
 162         xfrm_policy_unlock_afinfo(afinfo);
 163         return err;
 164 }
 165 EXPORT_SYMBOL(xfrm_register_mode);
 166
 167 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
 168 {
 169         struct xfrm_policy_afinfo *afinfo;
 170         struct xfrm_mode **modemap;
 171         int err;
 172
 173         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 174                 return -EINVAL;
 175
 176         afinfo = xfrm_policy_lock_afinfo(family);
 177         if (unlikely(afinfo == NULL))
 178                 return -EAFNOSUPPORT;
 179
 180         err = -ENOENT;
 181         modemap = afinfo->mode_map;
 182         if (likely(modemap[mode->encap] == mode)) {
 183                 modemap[mode->encap] = NULL;
 184                 err = 0;
 185         }
 186
 187         xfrm_policy_unlock_afinfo(afinfo);
 188         return err;
 189 }
 190 EXPORT_SYMBOL(xfrm_unregister_mode);
 191
 192 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
 193 {
 194         struct xfrm_policy_afinfo *afinfo;
 195         struct xfrm_mode *mode;
 196         int modload_attempted = 0;
 197
 198         if (unlikely(encap >= XFRM_MODE_MAX))
 199                 return NULL;
 200
 201 retry:
 202         afinfo = xfrm_policy_get_afinfo(family);
 203         if (unlikely(afinfo == NULL))
 204                 return NULL;
 205
 206         mode = afinfo->mode_map[encap];
 207         if (unlikely(mode && !try_module_get(mode->owner)))
 208                 mode = NULL;
 209         if (!mode && !modload_attempted) {
 210                 xfrm_policy_put_afinfo(afinfo);
 211                 request_module("xfrm-mode-%d-%d", family, encap);
 212                 modload_attempted = 1;
 213                 goto retry;
 214         }
 215
 216         xfrm_policy_put_afinfo(afinfo);
 217         return mode;
 218 }
 219
 220 void xfrm_put_mode(struct xfrm_mode *mode)
 221 {
 222         module_put(mode->owner);
 223 }
 224
 225 static inline unsigned long make_jiffies(long secs)
 226 {
 227         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 228                 return MAX_SCHEDULE_TIMEOUT-1;
 229         else
 230                 return secs*HZ;
 231 }
 232
 233 static void xfrm_policy_timer(unsigned long data)
 234 {
 235         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 236         unsigned long now = (unsigned long)xtime.tv_sec;
 237         long next = LONG_MAX;
 238         int warn = 0;
 239         int dir;
 240
 241         read_lock(&xp->lock);
 242
 243         if (xp->dead)
 244                 goto out;
 245
 246         dir = xfrm_policy_id2dir(xp->index);
 247
 248         if (xp->lft.hard_add_expires_seconds) {
 249                 long tmo = xp->lft.hard_add_expires_seconds +
 250                         xp->curlft.add_time - now;
 251                 if (tmo <= 0)
 252                         goto expired;
 253                 if (tmo < next)
 254                         next = tmo;
 255         }
 256         if (xp->lft.hard_use_expires_seconds) {
 257                 long tmo = xp->lft.hard_use_expires_seconds +
 258                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 259                 if (tmo <= 0)
 260                         goto expired;
 261                 if (tmo < next)
 262                         next = tmo;
 263         }
 264         if (xp->lft.soft_add_expires_seconds) {
 265                 long tmo = xp->lft.soft_add_expires_seconds +
 266                         xp->curlft.add_time - now;
 267                 if (tmo <= 0) {
 268                         warn = 1;
 269                         tmo = XFRM_KM_TIMEOUT;
 270                 }
 271                 if (tmo < next)
 272                         next = tmo;
 273         }
 274         if (xp->lft.soft_use_expires_seconds) {
 275                 long tmo = xp->lft.soft_use_expires_seconds +
 276                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 277                 if (tmo <= 0) {
 278                         warn = 1;
 279                         tmo = XFRM_KM_TIMEOUT;
 280                 }
 281                 if (tmo < next)
 282                         next = tmo;
 283         }
 284
 285         if (warn)
 286                 km_policy_expired(xp, dir, 0, 0);
 287         if (next != LONG_MAX &&
 288             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 289                 xfrm_pol_hold(xp);
 290
 291 out:
 292         read_unlock(&xp->lock);
 293         xfrm_pol_put(xp);
 294         return;
 295
 296 expired:
 297         read_unlock(&xp->lock);
 298         if (!xfrm_policy_delete(xp, dir))
 299                 km_policy_expired(xp, dir, 1, 0);
 300         xfrm_pol_put(xp);
 301 }
 302
 303
 304 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 305  * SPD calls.
 306  */
 307
 308 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 309 {
 310         struct xfrm_policy *policy;
 311
 312         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 313
 314         if (policy) {
 315                 INIT_HLIST_NODE(&policy->bydst);
 316                 INIT_HLIST_NODE(&policy->byidx);
 317                 rwlock_init(&policy->lock);
 318                 atomic_set(&policy->refcnt, 1);
 319                 init_timer(&policy->timer);
 320                 policy->timer.data = (unsigned long)policy;
 321                 policy->timer.function = xfrm_policy_timer;
 322         }
 323         return policy;
 324 }
 325 EXPORT_SYMBOL(xfrm_policy_alloc);
 326
 327 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 328
 329 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 330 {
 331         BUG_ON(!policy->dead);
 332
 333         BUG_ON(policy->bundles);
 334
 335         if (del_timer(&policy->timer))
 336                 BUG();
 337
 338         security_xfrm_policy_free(policy);
 339         kfree(policy);
 340 }
 341 EXPORT_SYMBOL(__xfrm_policy_destroy);
 342
 343 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 344 {
 345         struct dst_entry *dst;
 346
 347         while ((dst = policy->bundles) != NULL) {
 348                 policy->bundles = dst->next;
 349                 dst_free(dst);
 350         }
 351
 352         if (del_timer(&policy->timer))
 353                 atomic_dec(&policy->refcnt);
 354
 355         if (atomic_read(&policy->refcnt) > 1)
 356                 flow_cache_flush();
 357
 358         xfrm_pol_put(policy);
 359 }
 360
 361 static void xfrm_policy_gc_task(void *data)
 362 {
 363         struct xfrm_policy *policy;
 364         struct hlist_node *entry, *tmp;
 365         struct hlist_head gc_list;
 366
 367         spin_lock_bh(&xfrm_policy_gc_lock);
 368         gc_list.first = xfrm_policy_gc_list.first;
 369         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 370         spin_unlock_bh(&xfrm_policy_gc_lock);
 371
 372         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 373                 xfrm_policy_gc_kill(policy);
 374 }
 375
 376 /* Rule must be locked. Release descentant resources, announce
 377  * entry dead. The rule must be unlinked from lists to the moment.
 378  */
 379
 380 static void xfrm_policy_kill(struct xfrm_policy *policy)
 381 {
 382         int dead;
 383
 384         write_lock_bh(&policy->lock);
 385         dead = policy->dead;
 386         policy->dead = 1;
 387         write_unlock_bh(&policy->lock);
 388
 389         if (unlikely(dead)) {
 390                 WARN_ON(1);
 391                 return;
 392         }
 393
 394         spin_lock(&xfrm_policy_gc_lock);
 395         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 396         spin_unlock(&xfrm_policy_gc_lock);
 397
 398         schedule_work(&xfrm_policy_gc_work);
 399 }
 400
 401 struct xfrm_policy_hash {
 402         struct hlist_head       *table;
 403         unsigned int            hmask;
 404 };
 405
 406 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 407 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 408 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 409 static unsigned int xfrm_idx_hmask __read_mostly;
 410 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 411
 412 static inline unsigned int __idx_hash(u32 index, unsigned int hmask)
 413 {
 414         return (index ^ (index >> 8)) & hmask;
 415 }
 416
 417 static inline unsigned int idx_hash(u32 index)
 418 {
 419         return __idx_hash(index, xfrm_idx_hmask);
 420 }
 421
 422 static inline unsigned int __sel_hash(struct xfrm_selector *sel, unsigned short family, unsigned int hmask)
 423 {
 424         xfrm_address_t *daddr = &sel->daddr;
 425         xfrm_address_t *saddr = &sel->saddr;
 426         unsigned int h = 0;
 427
 428         switch (family) {
 429         case AF_INET:
 430                 if (sel->prefixlen_d != 32 ||
 431                     sel->prefixlen_s != 32)
 432                         return hmask + 1;
 433
 434                 h = ntohl(daddr->a4 ^ saddr->a4);
 435                 break;
 436
 437         case AF_INET6:
 438                 if (sel->prefixlen_d != 128 ||
 439                     sel->prefixlen_s != 128)
 440                         return hmask + 1;
 441
 442                 h = ntohl(daddr->a6[2] ^ daddr->a6[3] ^
 443                           saddr->a6[2] ^ saddr->a6[3]);
 444                 break;
 445         };
 446         h ^= (h >> 16);
 447         return h & hmask;
 448 }
 449
 450 static inline unsigned int __addr_hash(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, unsigned int hmask)
 451 {
 452         unsigned int h = 0;
 453
 454         switch (family) {
 455         case AF_INET:
 456                 h = ntohl(daddr->a4 ^ saddr->a4);
 457                 break;
 458
 459         case AF_INET6:
 460                 h = ntohl(daddr->a6[2] ^ daddr->a6[3] ^
 461                           saddr->a6[2] ^ saddr->a6[3]);
 462                 break;
 463         };
 464         h ^= (h >> 16);
 465         return h & hmask;
 466 }
 467
 468 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 469 {
 470         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 471         unsigned int hash = __sel_hash(sel, family, hmask);
 472
 473         return (hash == hmask + 1 ?
 474                 &xfrm_policy_inexact[dir] :
 475                 xfrm_policy_bydst[dir].table + hash);
 476 }
 477
 478 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 479 {
 480         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 481         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 482
 483         return xfrm_policy_bydst[dir].table + hash;
 484 }
 485
 486 static struct hlist_head *xfrm_policy_hash_alloc(unsigned int sz)
 487 {
 488         struct hlist_head *n;
 489
 490         if (sz <= PAGE_SIZE)
 491                 n = kmalloc(sz, GFP_KERNEL);
 492         else if (hashdist)
 493                 n = __vmalloc(sz, GFP_KERNEL, PAGE_KERNEL);
 494         else
 495                 n = (struct hlist_head *)
 496                         __get_free_pages(GFP_KERNEL, get_order(sz));
 497
 498         if (n)
 499                 memset(n, 0, sz);
 500
 501         return n;
 502 }
 503
 504 static void xfrm_policy_hash_free(struct hlist_head *n, unsigned int sz)
 505 {
 506         if (sz <= PAGE_SIZE)
 507                 kfree(n);
 508         else if (hashdist)
 509                 vfree(n);
 510         else
 511                 free_pages((unsigned long)n, get_order(sz));
 512 }
 513
 514 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 515                                    struct hlist_head *ndsttable,
 516                                    unsigned int nhashmask)
 517 {
 518         struct hlist_node *entry, *tmp;
 519         struct xfrm_policy *pol;
 520
 521         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 522                 unsigned int h;
 523
 524                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 525                                 pol->family, nhashmask);
 526                 hlist_add_head(&pol->bydst, ndsttable+h);
 527         }
 528 }
 529
 530 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 531                                    struct hlist_head *nidxtable,
 532                                    unsigned int nhashmask)
 533 {
 534         struct hlist_node *entry, *tmp;
 535         struct xfrm_policy *pol;
 536
 537         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 538                 unsigned int h;
 539
 540                 h = __idx_hash(pol->index, nhashmask);
 541                 hlist_add_head(&pol->byidx, nidxtable+h);
 542         }
 543 }
 544
 545 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 546 {
 547         return ((old_hmask + 1) << 1) - 1;
 548 }
 549
 550 static void xfrm_bydst_resize(int dir)
 551 {
 552         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 553         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 554         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 555         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 556         struct hlist_head *ndst = xfrm_policy_hash_alloc(nsize);
 557         int i;
 558
 559         if (!ndst)
 560                 return;
 561
 562         write_lock_bh(&xfrm_policy_lock);
 563
 564         for (i = hmask; i >= 0; i--)
 565                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 566
 567         xfrm_policy_bydst[dir].table = ndst;
 568         xfrm_policy_bydst[dir].hmask = nhashmask;
 569
 570         write_unlock_bh(&xfrm_policy_lock);
 571
 572         xfrm_policy_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 573 }
 574
 575 static void xfrm_byidx_resize(int total)
 576 {
 577         unsigned int hmask = xfrm_idx_hmask;
 578         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 579         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 580         struct hlist_head *oidx = xfrm_policy_byidx;
 581         struct hlist_head *nidx = xfrm_policy_hash_alloc(nsize);
 582         int i;
 583
 584         if (!nidx)
 585                 return;
 586
 587         write_lock_bh(&xfrm_policy_lock);
 588
 589         for (i = hmask; i >= 0; i--)
 590                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 591
 592         xfrm_policy_byidx = nidx;
 593         xfrm_idx_hmask = nhashmask;
 594
 595         write_unlock_bh(&xfrm_policy_lock);
 596
 597         xfrm_policy_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 598 }
 599
 600 static inline int xfrm_bydst_should_resize(int dir, int *total)
 601 {
 602         unsigned int cnt = xfrm_policy_count[dir];
 603         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 604
 605         if (total)
 606                 *total += cnt;
 607
 608         if ((hmask + 1) < xfrm_policy_hashmax &&
 609             cnt > hmask)
 610                 return 1;
 611
 612         return 0;
 613 }
 614
 615 static inline int xfrm_byidx_should_resize(int total)
 616 {
 617         unsigned int hmask = xfrm_idx_hmask;
 618
 619         if ((hmask + 1) < xfrm_policy_hashmax &&
 620             total > hmask)
 621                 return 1;
 622
 623         return 0;
 624 }
 625
 626 static DEFINE_MUTEX(hash_resize_mutex);
 627
 628 static void xfrm_hash_resize(void *__unused)
 629 {
 630         int dir, total;
 631
 632         mutex_lock(&hash_resize_mutex);
 633
 634         total = 0;
 635         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 636                 if (xfrm_bydst_should_resize(dir, &total))
 637                         xfrm_bydst_resize(dir);
 638         }
 639         if (xfrm_byidx_should_resize(total))
 640                 xfrm_byidx_resize(total);
 641
 642         mutex_unlock(&hash_resize_mutex);
 643 }
 644
 645 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize, NULL);
 646
 647 /* Generate new index... KAME seems to generate them ordered by cost
 648  * of an absolute inpredictability of ordering of rules. This will not pass. */
 649 static u32 xfrm_gen_index(u8 type, int dir)
 650 {
 651         static u32 idx_generator;
 652
 653         for (;;) {
 654                 struct hlist_node *entry;
 655                 struct hlist_head *list;
 656                 struct xfrm_policy *p;
 657                 u32 idx;
 658                 int found;
 659
 660                 idx = (idx_generator | dir);
 661                 idx_generator += 8;
 662                 if (idx == 0)
 663                         idx = 8;
 664                 list = xfrm_policy_byidx + idx_hash(idx);
 665                 found = 0;
 666                 hlist_for_each_entry(p, entry, list, byidx) {
 667                         if (p->index == idx) {
 668                                 found = 1;
 669                                 break;
 670                         }
 671                 }
 672                 if (!found)
 673                         return idx;
 674         }
 675 }
 676
 677 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 678 {
 679         u32 *p1 = (u32 *) s1;
 680         u32 *p2 = (u32 *) s2;
 681         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 682         int i;
 683
 684         for (i = 0; i < len; i++) {
 685                 if (p1[i] != p2[i])
 686                         return 1;
 687         }
 688
 689         return 0;
 690 }
 691
 692 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 693 {
 694         struct xfrm_policy *pol;
 695         struct xfrm_policy *delpol;
 696         struct hlist_head *chain;
 697         struct hlist_node *entry, *newpos, *last;
 698         struct dst_entry *gc_list;
 699
 700         write_lock_bh(&xfrm_policy_lock);
 701         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 702         delpol = NULL;
 703         newpos = NULL;
 704         last = NULL;
 705         hlist_for_each_entry(pol, entry, chain, bydst) {
 706                 if (!delpol &&
 707                     pol->type == policy->type &&
 708                     !selector_cmp(&pol->selector, &policy->selector) &&
 709                     xfrm_sec_ctx_match(pol->security, policy->security)) {
 710                         if (excl) {
 711                                 write_unlock_bh(&xfrm_policy_lock);
 712                                 return -EEXIST;
 713                         }
 714                         delpol = pol;
 715                         if (policy->priority > pol->priority)
 716                                 continue;
 717                 } else if (policy->priority >= pol->priority) {
 718                         last = &pol->bydst;
 719                         continue;
 720                 }
 721                 if (!newpos)
 722                         newpos = &pol->bydst;
 723                 if (delpol)
 724                         break;
 725                 last = &pol->bydst;
 726         }
 727         if (!newpos)
 728                 newpos = last;
 729         if (newpos)
 730                 hlist_add_after(newpos, &policy->bydst);
 731         else
 732                 hlist_add_head(&policy->bydst, chain);
 733         xfrm_pol_hold(policy);
 734         xfrm_policy_count[dir]++;
 735         atomic_inc(&flow_cache_genid);
 736         if (delpol) {
 737                 hlist_del(&delpol->bydst);
 738                 hlist_del(&delpol->byidx);
 739                 xfrm_policy_count[dir]--;
 740         }
 741         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 742         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 743         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
 744         policy->curlft.use_time = 0;
 745         if (!mod_timer(&policy->timer, jiffies + HZ))
 746                 xfrm_pol_hold(policy);
 747         write_unlock_bh(&xfrm_policy_lock);
 748
 749         if (delpol)
 750                 xfrm_policy_kill(delpol);
 751         else if (xfrm_bydst_should_resize(dir, NULL))
 752                 schedule_work(&xfrm_hash_work);
 753
 754         read_lock_bh(&xfrm_policy_lock);
 755         gc_list = NULL;
 756         entry = &policy->bydst;
 757         hlist_for_each_entry_continue(policy, entry, bydst) {
 758                 struct dst_entry *dst;
 759
 760                 write_lock(&policy->lock);
 761                 dst = policy->bundles;
 762                 if (dst) {
 763                         struct dst_entry *tail = dst;
 764                         while (tail->next)
 765                                 tail = tail->next;
 766                         tail->next = gc_list;
 767                         gc_list = dst;
 768
 769                         policy->bundles = NULL;
 770                 }
 771                 write_unlock(&policy->lock);
 772         }
 773         read_unlock_bh(&xfrm_policy_lock);
 774
 775         while (gc_list) {
 776                 struct dst_entry *dst = gc_list;
 777
 778                 gc_list = dst->next;
 779                 dst_free(dst);
 780         }
 781
 782         return 0;
 783 }
 784 EXPORT_SYMBOL(xfrm_policy_insert);
 785
 786 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 787                                           struct xfrm_selector *sel,
 788                                           struct xfrm_sec_ctx *ctx, int delete)
 789 {
 790         struct xfrm_policy *pol, *ret;
 791         struct hlist_head *chain;
 792         struct hlist_node *entry;
 793
 794         write_lock_bh(&xfrm_policy_lock);
 795         chain = policy_hash_bysel(sel, sel->family, dir);
 796         ret = NULL;
 797         hlist_for_each_entry(pol, entry, chain, bydst) {
 798                 if (pol->type == type &&
 799                     !selector_cmp(sel, &pol->selector) &&
 800                     xfrm_sec_ctx_match(ctx, pol->security)) {
 801                         xfrm_pol_hold(pol);
 802                         if (delete) {
 803                                 hlist_del(&pol->bydst);
 804                                 hlist_del(&pol->byidx);
 805                                 xfrm_policy_count[dir]--;
 806                         }
 807                         ret = pol;
 808                         break;
 809                 }
 810         }
 811         write_unlock_bh(&xfrm_policy_lock);
 812
 813         if (ret && delete) {
 814                 atomic_inc(&flow_cache_genid);
 815                 xfrm_policy_kill(ret);
 816         }
 817         return ret;
 818 }
 819 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 820
 821 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete)
 822 {
 823         struct xfrm_policy *pol, *ret;
 824         struct hlist_head *chain;
 825         struct hlist_node *entry;
 826
 827         write_lock_bh(&xfrm_policy_lock);
 828         chain = xfrm_policy_byidx + idx_hash(id);
 829         ret = NULL;
 830         hlist_for_each_entry(pol, entry, chain, byidx) {
 831                 if (pol->type == type && pol->index == id) {
 832                         xfrm_pol_hold(pol);
 833                         if (delete) {
 834                                 hlist_del(&pol->bydst);
 835                                 hlist_del(&pol->byidx);
 836                                 xfrm_policy_count[dir]--;
 837                         }
 838                         ret = pol;
 839                         break;
 840                 }
 841         }
 842         write_unlock_bh(&xfrm_policy_lock);
 843
 844         if (ret && delete) {
 845                 atomic_inc(&flow_cache_genid);
 846                 xfrm_policy_kill(ret);
 847         }
 848         return ret;
 849 }
 850 EXPORT_SYMBOL(xfrm_policy_byid);
 851
 852 void xfrm_policy_flush(u8 type)
 853 {
 854         int dir;
 855
 856         write_lock_bh(&xfrm_policy_lock);
 857         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 858                 struct xfrm_policy *pol;
 859                 struct hlist_node *entry;
 860                 int i;
 861
 862         again1:
 863                 hlist_for_each_entry(pol, entry,
 864                                      &xfrm_policy_inexact[dir], bydst) {
 865                         if (pol->type != type)
 866                                 continue;
 867                         hlist_del(&pol->bydst);
 868                         hlist_del(&pol->byidx);
 869                         write_unlock_bh(&xfrm_policy_lock);
 870
 871                         xfrm_policy_kill(pol);
 872
 873                         write_lock_bh(&xfrm_policy_lock);
 874                         goto again1;
 875                 }
 876
 877                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 878         again2:
 879                         hlist_for_each_entry(pol, entry,
 880                                              xfrm_policy_bydst[dir].table + i,
 881                                              bydst) {
 882                                 if (pol->type != type)
 883                                         continue;
 884                                 hlist_del(&pol->bydst);
 885                                 hlist_del(&pol->byidx);
 886                                 write_unlock_bh(&xfrm_policy_lock);
 887
 888                                 xfrm_policy_kill(pol);
 889
 890                                 write_lock_bh(&xfrm_policy_lock);
 891                                 goto again2;
 892                         }
 893                 }
 894
 895                 xfrm_policy_count[dir] = 0;
 896         }
 897         atomic_inc(&flow_cache_genid);
 898         write_unlock_bh(&xfrm_policy_lock);
 899 }
 900 EXPORT_SYMBOL(xfrm_policy_flush);
 901
 902 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 903                      void *data)
 904 {
 905         struct xfrm_policy *pol;
 906         struct hlist_node *entry;
 907         int dir, count, error;
 908
 909         read_lock_bh(&xfrm_policy_lock);
 910         count = 0;
 911         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 912                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 913                 int i;
 914
 915                 hlist_for_each_entry(pol, entry,
 916                                      &xfrm_policy_inexact[dir], bydst) {
 917                         if (pol->type == type)
 918                                 count++;
 919                 }
 920                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 921                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 922                                 if (pol->type == type)
 923                                         count++;
 924                         }
 925                 }
 926         }
 927
 928         if (count == 0) {
 929                 error = -ENOENT;
 930                 goto out;
 931         }
 932
 933         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 934                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 935                 int i;
 936
 937                 hlist_for_each_entry(pol, entry,
 938                                      &xfrm_policy_inexact[dir], bydst) {
 939                         if (pol->type != type)
 940                                 continue;
 941                         error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
 942                         if (error)
 943                                 goto out;
 944                 }
 945                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 946                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 947                                 if (pol->type != type)
 948                                         continue;
 949                                 error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
 950                                 if (error)
 951                                         goto out;
 952                         }
 953                 }
 954         }
 955         error = 0;
 956 out:
 957         read_unlock_bh(&xfrm_policy_lock);
 958         return error;
 959 }
 960 EXPORT_SYMBOL(xfrm_policy_walk);
 961
 962 /* Find policy to apply to this flow. */
 963
 964 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 965                              u8 type, u16 family, int dir)
 966 {
 967         struct xfrm_selector *sel = &pol->selector;
 968         int match;
 969
 970         if (pol->family != family ||
 971             pol->type != type)
 972                 return 0;
 973
 974         match = xfrm_selector_match(sel, fl, family);
 975         if (match) {
 976                 if (!security_xfrm_policy_lookup(pol, fl->secid, dir))
 977                         return 1;
 978         }
 979
 980         return 0;
 981 }
 982
 983 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 984                                                      u16 family, u8 dir)
 985 {
 986         struct xfrm_policy *pol, *ret;
 987         xfrm_address_t *daddr, *saddr;
 988         struct hlist_node *entry;
 989         struct hlist_head *chain;
 990
 991         daddr = xfrm_flowi_daddr(fl, family);
 992         saddr = xfrm_flowi_saddr(fl, family);
 993         if (unlikely(!daddr || !saddr))
 994                 return NULL;
 995
 996         read_lock_bh(&xfrm_policy_lock);
 997         chain = policy_hash_direct(daddr, saddr, family, dir);
 998         ret = NULL;
 999         hlist_for_each_entry(pol, entry, chain, bydst) {
1000                 if (xfrm_policy_match(pol, fl, type, family, dir)) {
1001                         xfrm_pol_hold(pol);
1002                         ret = pol;
1003                         break;
1004                 }
1005         }
1006         if (!ret) {
1007                 chain = &xfrm_policy_inexact[dir];
1008                 hlist_for_each_entry(pol, entry, chain, bydst) {
1009                         if (xfrm_policy_match(pol, fl, type, family, dir)) {
1010                                 xfrm_pol_hold(pol);
1011                                 ret = pol;
1012                                 break;
1013                         }
1014                 }
1015         }
1016         read_unlock_bh(&xfrm_policy_lock);
1017
1018         return ret;
1019 }
1020
1021 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1022                                void **objp, atomic_t **obj_refp)
1023 {
1024         struct xfrm_policy *pol;
1025
1026 #ifdef CONFIG_XFRM_SUB_POLICY
1027         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1028         if (pol)
1029                 goto end;
1030 #endif
1031         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1032
1033 #ifdef CONFIG_XFRM_SUB_POLICY
1034 end:
1035 #endif
1036         if ((*objp = (void *) pol) != NULL)
1037                 *obj_refp = &pol->refcnt;
1038 }
1039
1040 static inline int policy_to_flow_dir(int dir)
1041 {
1042         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1043             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1044             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1045                 return dir;
1046         switch (dir) {
1047         default:
1048         case XFRM_POLICY_IN:
1049                 return FLOW_DIR_IN;
1050         case XFRM_POLICY_OUT:
1051                 return FLOW_DIR_OUT;
1052         case XFRM_POLICY_FWD:
1053                 return FLOW_DIR_FWD;
1054         };
1055 }
1056
1057 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1058 {
1059         struct xfrm_policy *pol;
1060
1061         read_lock_bh(&xfrm_policy_lock);
1062         if ((pol = sk->sk_policy[dir]) != NULL) {
1063                 int match = xfrm_selector_match(&pol->selector, fl,
1064                                                 sk->sk_family);
1065                 int err = 0;
1066
1067                 if (match)
1068                   err = security_xfrm_policy_lookup(pol, fl->secid, policy_to_flow_dir(dir));
1069
1070                 if (match && !err)
1071                         xfrm_pol_hold(pol);
1072                 else
1073                         pol = NULL;
1074         }
1075         read_unlock_bh(&xfrm_policy_lock);
1076         return pol;
1077 }
1078
1079 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1080 {
1081         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1082                                                      pol->family, dir);
1083
1084         hlist_add_head(&pol->bydst, chain);
1085         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1086         xfrm_policy_count[dir]++;
1087         xfrm_pol_hold(pol);
1088
1089         if (xfrm_bydst_should_resize(dir, NULL))
1090                 schedule_work(&xfrm_hash_work);
1091 }
1092
1093 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1094                                                 int dir)
1095 {
1096         if (hlist_unhashed(&pol->bydst))
1097                 return NULL;
1098
1099         hlist_del(&pol->bydst);
1100         hlist_del(&pol->byidx);
1101         xfrm_policy_count[dir]--;
1102
1103         return pol;
1104 }
1105
1106 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1107 {
1108         write_lock_bh(&xfrm_policy_lock);
1109         pol = __xfrm_policy_unlink(pol, dir);
1110         write_unlock_bh(&xfrm_policy_lock);
1111         if (pol) {
1112                 if (dir < XFRM_POLICY_MAX)
1113                         atomic_inc(&flow_cache_genid);
1114                 xfrm_policy_kill(pol);
1115                 return 0;
1116         }
1117         return -ENOENT;
1118 }
1119 EXPORT_SYMBOL(xfrm_policy_delete);
1120
1121 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1122 {
1123         struct xfrm_policy *old_pol;
1124
1125 #ifdef CONFIG_XFRM_SUB_POLICY
1126         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1127                 return -EINVAL;
1128 #endif
1129
1130         write_lock_bh(&xfrm_policy_lock);
1131         old_pol = sk->sk_policy[dir];
1132         sk->sk_policy[dir] = pol;
1133         if (pol) {
1134                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
1135                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1136                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1137         }
1138         if (old_pol)
1139                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1140         write_unlock_bh(&xfrm_policy_lock);
1141
1142         if (old_pol) {
1143                 xfrm_policy_kill(old_pol);
1144         }
1145         return 0;
1146 }
1147
1148 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1149 {
1150         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1151
1152         if (newp) {
1153                 newp->selector = old->selector;
1154                 if (security_xfrm_policy_clone(old, newp)) {
1155                         kfree(newp);
1156                         return NULL;  /* ENOMEM */
1157                 }
1158                 newp->lft = old->lft;
1159                 newp->curlft = old->curlft;
1160                 newp->action = old->action;
1161                 newp->flags = old->flags;
1162                 newp->xfrm_nr = old->xfrm_nr;
1163                 newp->index = old->index;
1164                 newp->type = old->type;
1165                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1166                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1167                 write_lock_bh(&xfrm_policy_lock);
1168                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1169                 write_unlock_bh(&xfrm_policy_lock);
1170                 xfrm_pol_put(newp);
1171         }
1172         return newp;
1173 }
1174
1175 int __xfrm_sk_clone_policy(struct sock *sk)
1176 {
1177         struct xfrm_policy *p0 = sk->sk_policy[0],
1178                            *p1 = sk->sk_policy[1];
1179
1180         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1181         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1182                 return -ENOMEM;
1183         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1184                 return -ENOMEM;
1185         return 0;
1186 }
1187
1188 /* Resolve list of templates for the flow, given policy. */
1189
1190 static int
1191 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1192                       struct xfrm_state **xfrm,
1193                       unsigned short family)
1194 {
1195         int nx;
1196         int i, error;
1197         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1198         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1199
1200         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1201                 struct xfrm_state *x;
1202                 xfrm_address_t *remote = daddr;
1203                 xfrm_address_t *local  = saddr;
1204                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1205
1206                 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1207                         remote = &tmpl->id.daddr;
1208                         local = &tmpl->saddr;
1209                 }
1210
1211                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1212
1213                 if (x && x->km.state == XFRM_STATE_VALID) {
1214                         xfrm[nx++] = x;
1215                         daddr = remote;
1216                         saddr = local;
1217                         continue;
1218                 }
1219                 if (x) {
1220                         error = (x->km.state == XFRM_STATE_ERROR ?
1221                                  -EINVAL : -EAGAIN);
1222                         xfrm_state_put(x);
1223                 }
1224
1225                 if (!tmpl->optional)
1226                         goto fail;
1227         }
1228         return nx;
1229
1230 fail:
1231         for (nx--; nx>=0; nx--)
1232                 xfrm_state_put(xfrm[nx]);
1233         return error;
1234 }
1235
1236 static int
1237 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1238                   struct xfrm_state **xfrm,
1239                   unsigned short family)
1240 {
1241         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1242         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1243         int cnx = 0;
1244         int error;
1245         int ret;
1246         int i;
1247
1248         for (i = 0; i < npols; i++) {
1249                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1250                         error = -ENOBUFS;
1251                         goto fail;
1252                 }
1253
1254                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1255                 if (ret < 0) {
1256                         error = ret;
1257                         goto fail;
1258                 } else
1259                         cnx += ret;
1260         }
1261
1262         /* found states are sorted for outbound processing */
1263         if (npols > 1)
1264                 xfrm_state_sort(xfrm, tpp, cnx, family);
1265
1266         return cnx;
1267
1268  fail:
1269         for (cnx--; cnx>=0; cnx--)
1270                 xfrm_state_put(tpp[cnx]);
1271         return error;
1272
1273 }
1274
1275 /* Check that the bundle accepts the flow and its components are
1276  * still valid.
1277  */
1278
1279 static struct dst_entry *
1280 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1281 {
1282         struct dst_entry *x;
1283         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1284         if (unlikely(afinfo == NULL))
1285                 return ERR_PTR(-EINVAL);
1286         x = afinfo->find_bundle(fl, policy);
1287         xfrm_policy_put_afinfo(afinfo);
1288         return x;
1289 }
1290
1291 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1292  * all the metrics... Shortly, bundle a bundle.
1293  */
1294
1295 static int
1296 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1297                    struct flowi *fl, struct dst_entry **dst_p,
1298                    unsigned short family)
1299 {
1300         int err;
1301         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1302         if (unlikely(afinfo == NULL))
1303                 return -EINVAL;
1304         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1305         xfrm_policy_put_afinfo(afinfo);
1306         return err;
1307 }
1308
1309
1310 static int stale_bundle(struct dst_entry *dst);
1311
1312 /* Main function: finds/creates a bundle for given flow.
1313  *
1314  * At the moment we eat a raw IP route. Mostly to speed up lookups
1315  * on interfaces with disabled IPsec.
1316  */
1317 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1318                 struct sock *sk, int flags)
1319 {
1320         struct xfrm_policy *policy;
1321         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1322         int npols;
1323         int pol_dead;
1324         int xfrm_nr;
1325         int pi;
1326         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1327         struct dst_entry *dst, *dst_orig = *dst_p;
1328         int nx = 0;
1329         int err;
1330         u32 genid;
1331         u16 family;
1332         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1333
1334 restart:
1335         genid = atomic_read(&flow_cache_genid);
1336         policy = NULL;
1337         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1338                 pols[pi] = NULL;
1339         npols = 0;
1340         pol_dead = 0;
1341         xfrm_nr = 0;
1342
1343         if (sk && sk->sk_policy[1])
1344                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1345
1346         if (!policy) {
1347                 /* To accelerate a bit...  */
1348                 if ((dst_orig->flags & DST_NOXFRM) ||
1349                     !xfrm_policy_count[XFRM_POLICY_OUT])
1350                         return 0;
1351
1352                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1353                                            dir, xfrm_policy_lookup);
1354         }
1355
1356         if (!policy)
1357                 return 0;
1358
1359         family = dst_orig->ops->family;
1360         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
1361         pols[0] = policy;
1362         npols ++;
1363         xfrm_nr += pols[0]->xfrm_nr;
1364
1365         switch (policy->action) {
1366         case XFRM_POLICY_BLOCK:
1367                 /* Prohibit the flow */
1368                 err = -EPERM;
1369                 goto error;
1370
1371         case XFRM_POLICY_ALLOW:
1372 #ifndef CONFIG_XFRM_SUB_POLICY
1373                 if (policy->xfrm_nr == 0) {
1374                         /* Flow passes not transformed. */
1375                         xfrm_pol_put(policy);
1376                         return 0;
1377                 }
1378 #endif
1379
1380                 /* Try to find matching bundle.
1381                  *
1382                  * LATER: help from flow cache. It is optional, this
1383                  * is required only for output policy.
1384                  */
1385                 dst = xfrm_find_bundle(fl, policy, family);
1386                 if (IS_ERR(dst)) {
1387                         err = PTR_ERR(dst);
1388                         goto error;
1389                 }
1390
1391                 if (dst)
1392                         break;
1393
1394 #ifdef CONFIG_XFRM_SUB_POLICY
1395                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1396                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1397                                                             fl, family,
1398                                                             XFRM_POLICY_OUT);
1399                         if (pols[1]) {
1400                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1401                                         err = -EPERM;
1402                                         goto error;
1403                                 }
1404                                 npols ++;
1405                                 xfrm_nr += pols[1]->xfrm_nr;
1406                         }
1407                 }
1408
1409                 /*
1410                  * Because neither flowi nor bundle information knows about
1411                  * transformation template size. On more than one policy usage
1412                  * we can realize whether all of them is bypass or not after
1413                  * they are searched. See above not-transformed bypass
1414                  * is surrounded by non-sub policy configuration, too.
1415                  */
1416                 if (xfrm_nr == 0) {
1417                         /* Flow passes not transformed. */
1418                         xfrm_pols_put(pols, npols);
1419                         return 0;
1420                 }
1421
1422 #endif
1423                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1424
1425                 if (unlikely(nx<0)) {
1426                         err = nx;
1427                         if (err == -EAGAIN && flags) {
1428                                 DECLARE_WAITQUEUE(wait, current);
1429
1430                                 add_wait_queue(&km_waitq, &wait);
1431                                 set_current_state(TASK_INTERRUPTIBLE);
1432                                 schedule();
1433                                 set_current_state(TASK_RUNNING);
1434                                 remove_wait_queue(&km_waitq, &wait);
1435
1436                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1437
1438                                 if (nx == -EAGAIN && signal_pending(current)) {
1439                                         err = -ERESTART;
1440                                         goto error;
1441                                 }
1442                                 if (nx == -EAGAIN ||
1443                                     genid != atomic_read(&flow_cache_genid)) {
1444                                         xfrm_pols_put(pols, npols);
1445                                         goto restart;
1446                                 }
1447                                 err = nx;
1448                         }
1449                         if (err < 0)
1450                                 goto error;
1451                 }
1452                 if (nx == 0) {
1453                         /* Flow passes not transformed. */
1454                         xfrm_pols_put(pols, npols);
1455                         return 0;
1456                 }
1457
1458                 dst = dst_orig;
1459                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1460
1461                 if (unlikely(err)) {
1462                         int i;
1463                         for (i=0; i<nx; i++)
1464                                 xfrm_state_put(xfrm[i]);
1465                         goto error;
1466                 }
1467
1468                 for (pi = 0; pi < npols; pi++) {
1469                         read_lock_bh(&pols[pi]->lock);
1470                         pol_dead |= pols[pi]->dead;
1471                         read_unlock_bh(&pols[pi]->lock);
1472                 }
1473
1474                 write_lock_bh(&policy->lock);
1475                 if (unlikely(pol_dead || stale_bundle(dst))) {
1476                         /* Wow! While we worked on resolving, this
1477                          * policy has gone. Retry. It is not paranoia,
1478                          * we just cannot enlist new bundle to dead object.
1479                          * We can't enlist stable bundles either.
1480                          */
1481                         write_unlock_bh(&policy->lock);
1482                         if (dst)
1483                                 dst_free(dst);
1484
1485                         err = -EHOSTUNREACH;
1486                         goto error;
1487                 }
1488                 dst->next = policy->bundles;
1489                 policy->bundles = dst;
1490                 dst_hold(dst);
1491                 write_unlock_bh(&policy->lock);
1492         }
1493         *dst_p = dst;
1494         dst_release(dst_orig);
1495         xfrm_pols_put(pols, npols);
1496         return 0;
1497
1498 error:
1499         dst_release(dst_orig);
1500         xfrm_pols_put(pols, npols);
1501         *dst_p = NULL;
1502         return err;
1503 }
1504 EXPORT_SYMBOL(xfrm_lookup);
1505
1506 static inline int
1507 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1508 {
1509         struct xfrm_state *x;
1510         int err;
1511
1512         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1513                 return 0;
1514         x = skb->sp->xvec[idx];
1515         if (!x->type->reject)
1516                 return 0;
1517         xfrm_state_hold(x);
1518         err = x->type->reject(x, skb, fl);
1519         xfrm_state_put(x);
1520         return err;
1521 }
1522
1523 /* When skb is transformed back to its "native" form, we have to
1524  * check policy restrictions. At the moment we make this in maximally
1525  * stupid way. Shame on me. :-) Of course, connected sockets must
1526  * have policy cached at them.
1527  */
1528
1529 static inline int
1530 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1531               unsigned short family)
1532 {
1533         if (xfrm_state_kern(x))
1534                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1535         return  x->id.proto == tmpl->id.proto &&
1536                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1537                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1538                 x->props.mode == tmpl->mode &&
1539                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1540                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1541                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1542                   xfrm_state_addr_cmp(tmpl, x, family));
1543 }
1544
1545 /*
1546  * 0 or more than 0 is returned when validation is succeeded (either bypass
1547  * because of optional transport mode, or next index of the mathced secpath
1548  * state with the template.
1549  * -1 is returned when no matching template is found.
1550  * Otherwise "-2 - errored_index" is returned.
1551  */
1552 static inline int
1553 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1554                unsigned short family)
1555 {
1556         int idx = start;
1557
1558         if (tmpl->optional) {
1559                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1560                         return start;
1561         } else
1562                 start = -1;
1563         for (; idx < sp->len; idx++) {
1564                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1565                         return ++idx;
1566                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1567                         if (start == -1)
1568                                 start = -2-idx;
1569                         break;
1570                 }
1571         }
1572         return start;
1573 }
1574
1575 int
1576 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1577 {
1578         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1579         int err;
1580
1581         if (unlikely(afinfo == NULL))
1582                 return -EAFNOSUPPORT;
1583
1584         afinfo->decode_session(skb, fl);
1585         err = security_xfrm_decode_session(skb, &fl->secid);
1586         xfrm_policy_put_afinfo(afinfo);
1587         return err;
1588 }
1589 EXPORT_SYMBOL(xfrm_decode_session);
1590
1591 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1592 {
1593         for (; k < sp->len; k++) {
1594                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1595                         if (idxp)
1596                                 *idxp = k;
1597                         return 1;
1598                 }
1599         }
1600
1601         return 0;
1602 }
1603
1604 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1605                         unsigned short family)
1606 {
1607         struct xfrm_policy *pol;
1608         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1609         int npols = 0;
1610         int xfrm_nr;
1611         int pi;
1612         struct flowi fl;
1613         u8 fl_dir = policy_to_flow_dir(dir);
1614         int xerr_idx = -1;
1615         int *xerr_idxp = &xerr_idx;
1616
1617         if (xfrm_decode_session(skb, &fl, family) < 0)
1618                 return 0;
1619         nf_nat_decode_session(skb, &fl, family);
1620
1621         /* First, check used SA against their selectors. */
1622         if (skb->sp) {
1623                 int i;
1624
1625                 for (i=skb->sp->len-1; i>=0; i--) {
1626                         struct xfrm_state *x = skb->sp->xvec[i];
1627                         if (!xfrm_selector_match(&x->sel, &fl, family))
1628                                 return 0;
1629                 }
1630         }
1631
1632         pol = NULL;
1633         if (sk && sk->sk_policy[dir])
1634                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1635
1636         if (!pol)
1637                 pol = flow_cache_lookup(&fl, family, fl_dir,
1638                                         xfrm_policy_lookup);
1639
1640         if (!pol) {
1641                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, xerr_idxp)) {
1642                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1643                         return 0;
1644                 }
1645                 return 1;
1646         }
1647
1648         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1649
1650         pols[0] = pol;
1651         npols ++;
1652 #ifdef CONFIG_XFRM_SUB_POLICY
1653         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1654                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1655                                                     &fl, family,
1656                                                     XFRM_POLICY_IN);
1657                 if (pols[1]) {
1658                         pols[1]->curlft.use_time = (unsigned long)xtime.tv_sec;
1659                         npols ++;
1660                 }
1661         }
1662 #endif
1663
1664         if (pol->action == XFRM_POLICY_ALLOW) {
1665                 struct sec_path *sp;
1666                 static struct sec_path dummy;
1667                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1668                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1669                 struct xfrm_tmpl **tpp = tp;
1670                 int ti = 0;
1671                 int i, k;
1672
1673                 if ((sp = skb->sp) == NULL)
1674                         sp = &dummy;
1675
1676                 for (pi = 0; pi < npols; pi++) {
1677                         if (pols[pi] != pol &&
1678                             pols[pi]->action != XFRM_POLICY_ALLOW)
1679                                 goto reject;
1680                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1681                                 goto reject_error;
1682                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1683                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1684                 }
1685                 xfrm_nr = ti;
1686                 if (npols > 1) {
1687                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1688                         tpp = stp;
1689                 }
1690
1691                 /* For each tunnel xfrm, find the first matching tmpl.
1692                  * For each tmpl before that, find corresponding xfrm.
1693                  * Order is _important_. Later we will implement
1694                  * some barriers, but at the moment barriers
1695                  * are implied between each two transformations.
1696                  */
1697                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1698                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1699                         if (k < 0) {
1700                                 if (k < -1 && xerr_idxp)
1701                                         *xerr_idxp = -(2+k);
1702                                 goto reject;
1703                         }
1704                 }
1705
1706                 if (secpath_has_nontransport(sp, k, xerr_idxp))
1707                         goto reject;
1708
1709                 xfrm_pols_put(pols, npols);
1710                 return 1;
1711         }
1712
1713 reject:
1714         xfrm_secpath_reject(xerr_idx, skb, &fl);
1715 reject_error:
1716         xfrm_pols_put(pols, npols);
1717         return 0;
1718 }
1719 EXPORT_SYMBOL(__xfrm_policy_check);
1720
1721 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1722 {
1723         struct flowi fl;
1724
1725         if (xfrm_decode_session(skb, &fl, family) < 0)
1726                 return 0;
1727
1728         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1729 }
1730 EXPORT_SYMBOL(__xfrm_route_forward);
1731
1732 /* Optimize later using cookies and generation ids. */
1733
1734 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1735 {
1736         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1737          * to "-1" to force all XFRM destinations to get validated by
1738          * dst_ops->check on every use.  We do this because when a
1739          * normal route referenced by an XFRM dst is obsoleted we do
1740          * not go looking around for all parent referencing XFRM dsts
1741          * so that we can invalidate them.  It is just too much work.
1742          * Instead we make the checks here on every use.  For example:
1743          *
1744          *      XFRM dst A --> IPv4 dst X
1745          *
1746          * X is the "xdst->route" of A (X is also the "dst->path" of A
1747          * in this example).  If X is marked obsolete, "A" will not
1748          * notice.  That's what we are validating here via the
1749          * stale_bundle() check.
1750          *
1751          * When a policy's bundle is pruned, we dst_free() the XFRM
1752          * dst which causes it's ->obsolete field to be set to a
1753          * positive non-zero integer.  If an XFRM dst has been pruned
1754          * like this, we want to force a new route lookup.
1755          */
1756         if (dst->obsolete < 0 && !stale_bundle(dst))
1757                 return dst;
1758
1759         return NULL;
1760 }
1761
1762 static int stale_bundle(struct dst_entry *dst)
1763 {
1764         return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1765 }
1766
1767 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1768 {
1769         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1770                 dst->dev = &loopback_dev;
1771                 dev_hold(&loopback_dev);
1772                 dev_put(dev);
1773         }
1774 }
1775 EXPORT_SYMBOL(xfrm_dst_ifdown);
1776
1777 static void xfrm_link_failure(struct sk_buff *skb)
1778 {
1779         /* Impossible. Such dst must be popped before reaches point of failure. */
1780         return;
1781 }
1782
1783 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1784 {
1785         if (dst) {
1786                 if (dst->obsolete) {
1787                         dst_release(dst);
1788                         dst = NULL;
1789                 }
1790         }
1791         return dst;
1792 }
1793
1794 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1795 {
1796         struct dst_entry *dst, **dstp;
1797
1798         write_lock(&pol->lock);
1799         dstp = &pol->bundles;
1800         while ((dst=*dstp) != NULL) {
1801                 if (func(dst)) {
1802                         *dstp = dst->next;
1803                         dst->next = *gc_list_p;
1804                         *gc_list_p = dst;
1805                 } else {
1806                         dstp = &dst->next;
1807                 }
1808         }
1809         write_unlock(&pol->lock);
1810 }
1811
1812 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1813 {
1814         struct dst_entry *gc_list = NULL;
1815         int dir;
1816
1817         read_lock_bh(&xfrm_policy_lock);
1818         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1819                 struct xfrm_policy *pol;
1820                 struct hlist_node *entry;
1821                 struct hlist_head *table;
1822                 int i;
1823
1824                 hlist_for_each_entry(pol, entry,
1825                                      &xfrm_policy_inexact[dir], bydst)
1826                         prune_one_bundle(pol, func, &gc_list);
1827
1828                 table = xfrm_policy_bydst[dir].table;
1829                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1830                         hlist_for_each_entry(pol, entry, table + i, bydst)
1831                                 prune_one_bundle(pol, func, &gc_list);
1832                 }
1833         }
1834         read_unlock_bh(&xfrm_policy_lock);
1835
1836         while (gc_list) {
1837                 struct dst_entry *dst = gc_list;
1838                 gc_list = dst->next;
1839                 dst_free(dst);
1840         }
1841 }
1842
1843 static int unused_bundle(struct dst_entry *dst)
1844 {
1845         return !atomic_read(&dst->__refcnt);
1846 }
1847
1848 static void __xfrm_garbage_collect(void)
1849 {
1850         xfrm_prune_bundles(unused_bundle);
1851 }
1852
1853 static int xfrm_flush_bundles(void)
1854 {
1855         xfrm_prune_bundles(stale_bundle);
1856         return 0;
1857 }
1858
1859 void xfrm_init_pmtu(struct dst_entry *dst)
1860 {
1861         do {
1862                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1863                 u32 pmtu, route_mtu_cached;
1864
1865                 pmtu = dst_mtu(dst->child);
1866                 xdst->child_mtu_cached = pmtu;
1867
1868                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1869
1870                 route_mtu_cached = dst_mtu(xdst->route);
1871                 xdst->route_mtu_cached = route_mtu_cached;
1872
1873                 if (pmtu > route_mtu_cached)
1874                         pmtu = route_mtu_cached;
1875
1876                 dst->metrics[RTAX_MTU-1] = pmtu;
1877         } while ((dst = dst->next));
1878 }
1879
1880 EXPORT_SYMBOL(xfrm_init_pmtu);
1881
1882 /* Check that the bundle accepts the flow and its components are
1883  * still valid.
1884  */
1885
1886 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family, int strict)
1887 {
1888         struct dst_entry *dst = &first->u.dst;
1889         struct xfrm_dst *last;
1890         u32 mtu;
1891
1892         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1893             (dst->dev && !netif_running(dst->dev)))
1894                 return 0;
1895
1896         last = NULL;
1897
1898         do {
1899                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1900
1901                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1902                         return 0;
1903                 if (fl && !security_xfrm_flow_state_match(fl, dst->xfrm))
1904                         return 0;
1905                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1906                         return 0;
1907                 if (xdst->genid != dst->xfrm->genid)
1908                         return 0;
1909
1910                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
1911                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1912                         return 0;
1913
1914                 mtu = dst_mtu(dst->child);
1915                 if (xdst->child_mtu_cached != mtu) {
1916                         last = xdst;
1917                         xdst->child_mtu_cached = mtu;
1918                 }
1919
1920                 if (!dst_check(xdst->route, xdst->route_cookie))
1921                         return 0;
1922                 mtu = dst_mtu(xdst->route);
1923                 if (xdst->route_mtu_cached != mtu) {
1924                         last = xdst;
1925                         xdst->route_mtu_cached = mtu;
1926                 }
1927
1928                 dst = dst->child;
1929         } while (dst->xfrm);
1930
1931         if (likely(!last))
1932                 return 1;
1933
1934         mtu = last->child_mtu_cached;
1935         for (;;) {
1936                 dst = &last->u.dst;
1937
1938                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1939                 if (mtu > last->route_mtu_cached)
1940                         mtu = last->route_mtu_cached;
1941                 dst->metrics[RTAX_MTU-1] = mtu;
1942
1943                 if (last == first)
1944                         break;
1945
1946                 last = last->u.next;
1947                 last->child_mtu_cached = mtu;
1948         }
1949
1950         return 1;
1951 }
1952
1953 EXPORT_SYMBOL(xfrm_bundle_ok);
1954
1955 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1956 {
1957         int err = 0;
1958         if (unlikely(afinfo == NULL))
1959                 return -EINVAL;
1960         if (unlikely(afinfo->family >= NPROTO))
1961                 return -EAFNOSUPPORT;
1962         write_lock_bh(&xfrm_policy_afinfo_lock);
1963         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1964                 err = -ENOBUFS;
1965         else {
1966                 struct dst_ops *dst_ops = afinfo->dst_ops;
1967                 if (likely(dst_ops->kmem_cachep == NULL))
1968                         dst_ops->kmem_cachep = xfrm_dst_cache;
1969                 if (likely(dst_ops->check == NULL))
1970                         dst_ops->check = xfrm_dst_check;
1971                 if (likely(dst_ops->negative_advice == NULL))
1972                         dst_ops->negative_advice = xfrm_negative_advice;
1973                 if (likely(dst_ops->link_failure == NULL))
1974                         dst_ops->link_failure = xfrm_link_failure;
1975                 if (likely(afinfo->garbage_collect == NULL))
1976                         afinfo->garbage_collect = __xfrm_garbage_collect;
1977                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1978         }
1979         write_unlock_bh(&xfrm_policy_afinfo_lock);
1980         return err;
1981 }
1982 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1983
1984 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1985 {
1986         int err = 0;
1987         if (unlikely(afinfo == NULL))
1988                 return -EINVAL;
1989         if (unlikely(afinfo->family >= NPROTO))
1990                 return -EAFNOSUPPORT;
1991         write_lock_bh(&xfrm_policy_afinfo_lock);
1992         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1993                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1994                         err = -EINVAL;
1995                 else {
1996                         struct dst_ops *dst_ops = afinfo->dst_ops;
1997                         xfrm_policy_afinfo[afinfo->family] = NULL;
1998                         dst_ops->kmem_cachep = NULL;
1999                         dst_ops->check = NULL;
2000                         dst_ops->negative_advice = NULL;
2001                         dst_ops->link_failure = NULL;
2002                         afinfo->garbage_collect = NULL;
2003                 }
2004         }
2005         write_unlock_bh(&xfrm_policy_afinfo_lock);
2006         return err;
2007 }
2008 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2009
2010 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2011 {
2012         struct xfrm_policy_afinfo *afinfo;
2013         if (unlikely(family >= NPROTO))
2014                 return NULL;
2015         read_lock(&xfrm_policy_afinfo_lock);
2016         afinfo = xfrm_policy_afinfo[family];
2017         if (unlikely(!afinfo))
2018                 read_unlock(&xfrm_policy_afinfo_lock);
2019         return afinfo;
2020 }
2021
2022 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2023 {
2024         read_unlock(&xfrm_policy_afinfo_lock);
2025 }
2026
2027 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2028 {
2029         struct xfrm_policy_afinfo *afinfo;
2030         if (unlikely(family >= NPROTO))
2031                 return NULL;
2032         write_lock_bh(&xfrm_policy_afinfo_lock);
2033         afinfo = xfrm_policy_afinfo[family];
2034         if (unlikely(!afinfo))
2035                 write_unlock_bh(&xfrm_policy_afinfo_lock);
2036         return afinfo;
2037 }
2038
2039 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2040 {
2041         write_unlock_bh(&xfrm_policy_afinfo_lock);
2042 }
2043
2044 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2045 {
2046         switch (event) {
2047         case NETDEV_DOWN:
2048                 xfrm_flush_bundles();
2049         }
2050         return NOTIFY_DONE;
2051 }
2052
2053 static struct notifier_block xfrm_dev_notifier = {
2054         xfrm_dev_event,
2055         NULL,
2056         0
2057 };
2058
2059 static void __init xfrm_policy_init(void)
2060 {
2061         unsigned int hmask, sz;
2062         int dir;
2063
2064         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2065                                            sizeof(struct xfrm_dst),
2066                                            0, SLAB_HWCACHE_ALIGN,
2067                                            NULL, NULL);
2068         if (!xfrm_dst_cache)
2069                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
2070
2071         hmask = 8 - 1;
2072         sz = (hmask+1) * sizeof(struct hlist_head);
2073
2074         xfrm_policy_byidx = xfrm_policy_hash_alloc(sz);
2075         xfrm_idx_hmask = hmask;
2076         if (!xfrm_policy_byidx)
2077                 panic("XFRM: failed to allocate byidx hash\n");
2078
2079         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2080                 struct xfrm_policy_hash *htab;
2081
2082                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2083
2084                 htab = &xfrm_policy_bydst[dir];
2085                 htab->table = xfrm_policy_hash_alloc(sz);
2086                 htab->hmask = hmask;
2087                 if (!htab->table)
2088                         panic("XFRM: failed to allocate bydst hash\n");
2089         }
2090
2091         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
2092         register_netdevice_notifier(&xfrm_dev_notifier);
2093 }
2094
2095 void __init xfrm_init(void)
2096 {
2097         xfrm_state_init();
2098         xfrm_policy_init();
2099         xfrm_input_init();
2100 }
2101