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