git.oblomov.eu Git - linux-2.6/blob - net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/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 static inline int
  54 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  55 {
  56         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  57                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  58                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  59                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  60                 (fl->proto == sel->proto || !sel->proto) &&
  61                 (fl->oif == sel->ifindex || !sel->ifindex);
  62 }
  63
  64 static inline int
  65 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  66 {
  67         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  68                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  69                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  70                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  71                 (fl->proto == sel->proto || !sel->proto) &&
  72                 (fl->oif == sel->ifindex || !sel->ifindex);
  73 }
  74
  75 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  76                     unsigned short family)
  77 {
  78         switch (family) {
  79         case AF_INET:
  80                 return __xfrm4_selector_match(sel, fl);
  81         case AF_INET6:
  82                 return __xfrm6_selector_match(sel, fl);
  83         }
  84         return 0;
  85 }
  86
  87 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
  88 {
  89         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  90         struct xfrm_type **typemap;
  91         int err = 0;
  92
  93         if (unlikely(afinfo == NULL))
  94                 return -EAFNOSUPPORT;
  95         typemap = afinfo->type_map;
  96
  97         if (likely(typemap[type->proto] == NULL))
  98                 typemap[type->proto] = type;
  99         else
 100                 err = -EEXIST;
 101         xfrm_policy_unlock_afinfo(afinfo);
 102         return err;
 103 }
 104 EXPORT_SYMBOL(xfrm_register_type);
 105
 106 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
 107 {
 108         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
 109         struct xfrm_type **typemap;
 110         int err = 0;
 111
 112         if (unlikely(afinfo == NULL))
 113                 return -EAFNOSUPPORT;
 114         typemap = afinfo->type_map;
 115
 116         if (unlikely(typemap[type->proto] != type))
 117                 err = -ENOENT;
 118         else
 119                 typemap[type->proto] = NULL;
 120         xfrm_policy_unlock_afinfo(afinfo);
 121         return err;
 122 }
 123 EXPORT_SYMBOL(xfrm_unregister_type);
 124
 125 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
 126 {
 127         struct xfrm_policy_afinfo *afinfo;
 128         struct xfrm_type **typemap;
 129         struct xfrm_type *type;
 130         int modload_attempted = 0;
 131
 132 retry:
 133         afinfo = xfrm_policy_get_afinfo(family);
 134         if (unlikely(afinfo == NULL))
 135                 return NULL;
 136         typemap = afinfo->type_map;
 137
 138         type = typemap[proto];
 139         if (unlikely(type && !try_module_get(type->owner)))
 140                 type = NULL;
 141         if (!type && !modload_attempted) {
 142                 xfrm_policy_put_afinfo(afinfo);
 143                 request_module("xfrm-type-%d-%d",
 144                                (int) family, (int) proto);
 145                 modload_attempted = 1;
 146                 goto retry;
 147         }
 148
 149         xfrm_policy_put_afinfo(afinfo);
 150         return type;
 151 }
 152
 153 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
 154                     unsigned short family)
 155 {
 156         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 157         int err = 0;
 158
 159         if (unlikely(afinfo == NULL))
 160                 return -EAFNOSUPPORT;
 161
 162         if (likely(afinfo->dst_lookup != NULL))
 163                 err = afinfo->dst_lookup(dst, fl);
 164         else
 165                 err = -EINVAL;
 166         xfrm_policy_put_afinfo(afinfo);
 167         return err;
 168 }
 169 EXPORT_SYMBOL(xfrm_dst_lookup);
 170
 171 void xfrm_put_type(struct xfrm_type *type)
 172 {
 173         module_put(type->owner);
 174 }
 175
 176 int xfrm_register_mode(struct xfrm_mode *mode, int family)
 177 {
 178         struct xfrm_policy_afinfo *afinfo;
 179         struct xfrm_mode **modemap;
 180         int err;
 181
 182         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 183                 return -EINVAL;
 184
 185         afinfo = xfrm_policy_lock_afinfo(family);
 186         if (unlikely(afinfo == NULL))
 187                 return -EAFNOSUPPORT;
 188
 189         err = -EEXIST;
 190         modemap = afinfo->mode_map;
 191         if (likely(modemap[mode->encap] == NULL)) {
 192                 modemap[mode->encap] = mode;
 193                 err = 0;
 194         }
 195
 196         xfrm_policy_unlock_afinfo(afinfo);
 197         return err;
 198 }
 199 EXPORT_SYMBOL(xfrm_register_mode);
 200
 201 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
 202 {
 203         struct xfrm_policy_afinfo *afinfo;
 204         struct xfrm_mode **modemap;
 205         int err;
 206
 207         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 208                 return -EINVAL;
 209
 210         afinfo = xfrm_policy_lock_afinfo(family);
 211         if (unlikely(afinfo == NULL))
 212                 return -EAFNOSUPPORT;
 213
 214         err = -ENOENT;
 215         modemap = afinfo->mode_map;
 216         if (likely(modemap[mode->encap] == mode)) {
 217                 modemap[mode->encap] = NULL;
 218                 err = 0;
 219         }
 220
 221         xfrm_policy_unlock_afinfo(afinfo);
 222         return err;
 223 }
 224 EXPORT_SYMBOL(xfrm_unregister_mode);
 225
 226 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
 227 {
 228         struct xfrm_policy_afinfo *afinfo;
 229         struct xfrm_mode *mode;
 230         int modload_attempted = 0;
 231
 232         if (unlikely(encap >= XFRM_MODE_MAX))
 233                 return NULL;
 234
 235 retry:
 236         afinfo = xfrm_policy_get_afinfo(family);
 237         if (unlikely(afinfo == NULL))
 238                 return NULL;
 239
 240         mode = afinfo->mode_map[encap];
 241         if (unlikely(mode && !try_module_get(mode->owner)))
 242                 mode = NULL;
 243         if (!mode && !modload_attempted) {
 244                 xfrm_policy_put_afinfo(afinfo);
 245                 request_module("xfrm-mode-%d-%d", family, encap);
 246                 modload_attempted = 1;
 247                 goto retry;
 248         }
 249
 250         xfrm_policy_put_afinfo(afinfo);
 251         return mode;
 252 }
 253
 254 void xfrm_put_mode(struct xfrm_mode *mode)
 255 {
 256         module_put(mode->owner);
 257 }
 258
 259 static inline unsigned long make_jiffies(long secs)
 260 {
 261         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 262                 return MAX_SCHEDULE_TIMEOUT-1;
 263         else
 264                 return secs*HZ;
 265 }
 266
 267 static void xfrm_policy_timer(unsigned long data)
 268 {
 269         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 270         unsigned long now = (unsigned long)xtime.tv_sec;
 271         long next = LONG_MAX;
 272         int warn = 0;
 273         int dir;
 274
 275         read_lock(&xp->lock);
 276
 277         if (xp->dead)
 278                 goto out;
 279
 280         dir = xfrm_policy_id2dir(xp->index);
 281
 282         if (xp->lft.hard_add_expires_seconds) {
 283                 long tmo = xp->lft.hard_add_expires_seconds +
 284                         xp->curlft.add_time - now;
 285                 if (tmo <= 0)
 286                         goto expired;
 287                 if (tmo < next)
 288                         next = tmo;
 289         }
 290         if (xp->lft.hard_use_expires_seconds) {
 291                 long tmo = xp->lft.hard_use_expires_seconds +
 292                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 293                 if (tmo <= 0)
 294                         goto expired;
 295                 if (tmo < next)
 296                         next = tmo;
 297         }
 298         if (xp->lft.soft_add_expires_seconds) {
 299                 long tmo = xp->lft.soft_add_expires_seconds +
 300                         xp->curlft.add_time - now;
 301                 if (tmo <= 0) {
 302                         warn = 1;
 303                         tmo = XFRM_KM_TIMEOUT;
 304                 }
 305                 if (tmo < next)
 306                         next = tmo;
 307         }
 308         if (xp->lft.soft_use_expires_seconds) {
 309                 long tmo = xp->lft.soft_use_expires_seconds +
 310                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 311                 if (tmo <= 0) {
 312                         warn = 1;
 313                         tmo = XFRM_KM_TIMEOUT;
 314                 }
 315                 if (tmo < next)
 316                         next = tmo;
 317         }
 318
 319         if (warn)
 320                 km_policy_expired(xp, dir, 0, 0);
 321         if (next != LONG_MAX &&
 322             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 323                 xfrm_pol_hold(xp);
 324
 325 out:
 326         read_unlock(&xp->lock);
 327         xfrm_pol_put(xp);
 328         return;
 329
 330 expired:
 331         read_unlock(&xp->lock);
 332         if (!xfrm_policy_delete(xp, dir))
 333                 km_policy_expired(xp, dir, 1, 0);
 334         xfrm_pol_put(xp);
 335 }
 336
 337
 338 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 339  * SPD calls.
 340  */
 341
 342 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 343 {
 344         struct xfrm_policy *policy;
 345
 346         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 347
 348         if (policy) {
 349                 INIT_HLIST_NODE(&policy->bydst);
 350                 INIT_HLIST_NODE(&policy->byidx);
 351                 rwlock_init(&policy->lock);
 352                 atomic_set(&policy->refcnt, 1);
 353                 init_timer(&policy->timer);
 354                 policy->timer.data = (unsigned long)policy;
 355                 policy->timer.function = xfrm_policy_timer;
 356         }
 357         return policy;
 358 }
 359 EXPORT_SYMBOL(xfrm_policy_alloc);
 360
 361 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 362
 363 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 364 {
 365         BUG_ON(!policy->dead);
 366
 367         BUG_ON(policy->bundles);
 368
 369         if (del_timer(&policy->timer))
 370                 BUG();
 371
 372         security_xfrm_policy_free(policy);
 373         kfree(policy);
 374 }
 375 EXPORT_SYMBOL(__xfrm_policy_destroy);
 376
 377 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 378 {
 379         struct dst_entry *dst;
 380
 381         while ((dst = policy->bundles) != NULL) {
 382                 policy->bundles = dst->next;
 383                 dst_free(dst);
 384         }
 385
 386         if (del_timer(&policy->timer))
 387                 atomic_dec(&policy->refcnt);
 388
 389         if (atomic_read(&policy->refcnt) > 1)
 390                 flow_cache_flush();
 391
 392         xfrm_pol_put(policy);
 393 }
 394
 395 static void xfrm_policy_gc_task(struct work_struct *work)
 396 {
 397         struct xfrm_policy *policy;
 398         struct hlist_node *entry, *tmp;
 399         struct hlist_head gc_list;
 400
 401         spin_lock_bh(&xfrm_policy_gc_lock);
 402         gc_list.first = xfrm_policy_gc_list.first;
 403         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 404         spin_unlock_bh(&xfrm_policy_gc_lock);
 405
 406         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 407                 xfrm_policy_gc_kill(policy);
 408 }
 409
 410 /* Rule must be locked. Release descentant resources, announce
 411  * entry dead. The rule must be unlinked from lists to the moment.
 412  */
 413
 414 static void xfrm_policy_kill(struct xfrm_policy *policy)
 415 {
 416         int dead;
 417
 418         write_lock_bh(&policy->lock);
 419         dead = policy->dead;
 420         policy->dead = 1;
 421         write_unlock_bh(&policy->lock);
 422
 423         if (unlikely(dead)) {
 424                 WARN_ON(1);
 425                 return;
 426         }
 427
 428         spin_lock(&xfrm_policy_gc_lock);
 429         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 430         spin_unlock(&xfrm_policy_gc_lock);
 431
 432         schedule_work(&xfrm_policy_gc_work);
 433 }
 434
 435 struct xfrm_policy_hash {
 436         struct hlist_head       *table;
 437         unsigned int            hmask;
 438 };
 439
 440 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 441 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 442 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 443 static unsigned int xfrm_idx_hmask __read_mostly;
 444 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 445
 446 static inline unsigned int idx_hash(u32 index)
 447 {
 448         return __idx_hash(index, xfrm_idx_hmask);
 449 }
 450
 451 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 452 {
 453         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 454         unsigned int hash = __sel_hash(sel, family, hmask);
 455
 456         return (hash == hmask + 1 ?
 457                 &xfrm_policy_inexact[dir] :
 458                 xfrm_policy_bydst[dir].table + hash);
 459 }
 460
 461 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 462 {
 463         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 464         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 465
 466         return xfrm_policy_bydst[dir].table + hash;
 467 }
 468
 469 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 470                                    struct hlist_head *ndsttable,
 471                                    unsigned int nhashmask)
 472 {
 473         struct hlist_node *entry, *tmp;
 474         struct xfrm_policy *pol;
 475
 476         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 477                 unsigned int h;
 478
 479                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 480                                 pol->family, nhashmask);
 481                 hlist_add_head(&pol->bydst, ndsttable+h);
 482         }
 483 }
 484
 485 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 486                                    struct hlist_head *nidxtable,
 487                                    unsigned int nhashmask)
 488 {
 489         struct hlist_node *entry, *tmp;
 490         struct xfrm_policy *pol;
 491
 492         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 493                 unsigned int h;
 494
 495                 h = __idx_hash(pol->index, nhashmask);
 496                 hlist_add_head(&pol->byidx, nidxtable+h);
 497         }
 498 }
 499
 500 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 501 {
 502         return ((old_hmask + 1) << 1) - 1;
 503 }
 504
 505 static void xfrm_bydst_resize(int dir)
 506 {
 507         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 508         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 509         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 510         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 511         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 512         int i;
 513
 514         if (!ndst)
 515                 return;
 516
 517         write_lock_bh(&xfrm_policy_lock);
 518
 519         for (i = hmask; i >= 0; i--)
 520                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 521
 522         xfrm_policy_bydst[dir].table = ndst;
 523         xfrm_policy_bydst[dir].hmask = nhashmask;
 524
 525         write_unlock_bh(&xfrm_policy_lock);
 526
 527         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 528 }
 529
 530 static void xfrm_byidx_resize(int total)
 531 {
 532         unsigned int hmask = xfrm_idx_hmask;
 533         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 534         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 535         struct hlist_head *oidx = xfrm_policy_byidx;
 536         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 537         int i;
 538
 539         if (!nidx)
 540                 return;
 541
 542         write_lock_bh(&xfrm_policy_lock);
 543
 544         for (i = hmask; i >= 0; i--)
 545                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 546
 547         xfrm_policy_byidx = nidx;
 548         xfrm_idx_hmask = nhashmask;
 549
 550         write_unlock_bh(&xfrm_policy_lock);
 551
 552         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 553 }
 554
 555 static inline int xfrm_bydst_should_resize(int dir, int *total)
 556 {
 557         unsigned int cnt = xfrm_policy_count[dir];
 558         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 559
 560         if (total)
 561                 *total += cnt;
 562
 563         if ((hmask + 1) < xfrm_policy_hashmax &&
 564             cnt > hmask)
 565                 return 1;
 566
 567         return 0;
 568 }
 569
 570 static inline int xfrm_byidx_should_resize(int total)
 571 {
 572         unsigned int hmask = xfrm_idx_hmask;
 573
 574         if ((hmask + 1) < xfrm_policy_hashmax &&
 575             total > hmask)
 576                 return 1;
 577
 578         return 0;
 579 }
 580
 581 static DEFINE_MUTEX(hash_resize_mutex);
 582
 583 static void xfrm_hash_resize(struct work_struct *__unused)
 584 {
 585         int dir, total;
 586
 587         mutex_lock(&hash_resize_mutex);
 588
 589         total = 0;
 590         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 591                 if (xfrm_bydst_should_resize(dir, &total))
 592                         xfrm_bydst_resize(dir);
 593         }
 594         if (xfrm_byidx_should_resize(total))
 595                 xfrm_byidx_resize(total);
 596
 597         mutex_unlock(&hash_resize_mutex);
 598 }
 599
 600 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 601
 602 /* Generate new index... KAME seems to generate them ordered by cost
 603  * of an absolute inpredictability of ordering of rules. This will not pass. */
 604 static u32 xfrm_gen_index(u8 type, int dir)
 605 {
 606         static u32 idx_generator;
 607
 608         for (;;) {
 609                 struct hlist_node *entry;
 610                 struct hlist_head *list;
 611                 struct xfrm_policy *p;
 612                 u32 idx;
 613                 int found;
 614
 615                 idx = (idx_generator | dir);
 616                 idx_generator += 8;
 617                 if (idx == 0)
 618                         idx = 8;
 619                 list = xfrm_policy_byidx + idx_hash(idx);
 620                 found = 0;
 621                 hlist_for_each_entry(p, entry, list, byidx) {
 622                         if (p->index == idx) {
 623                                 found = 1;
 624                                 break;
 625                         }
 626                 }
 627                 if (!found)
 628                         return idx;
 629         }
 630 }
 631
 632 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 633 {
 634         u32 *p1 = (u32 *) s1;
 635         u32 *p2 = (u32 *) s2;
 636         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 637         int i;
 638
 639         for (i = 0; i < len; i++) {
 640                 if (p1[i] != p2[i])
 641                         return 1;
 642         }
 643
 644         return 0;
 645 }
 646
 647 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 648 {
 649         struct xfrm_policy *pol;
 650         struct xfrm_policy *delpol;
 651         struct hlist_head *chain;
 652         struct hlist_node *entry, *newpos, *last;
 653         struct dst_entry *gc_list;
 654
 655         write_lock_bh(&xfrm_policy_lock);
 656         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 657         delpol = NULL;
 658         newpos = NULL;
 659         last = NULL;
 660         hlist_for_each_entry(pol, entry, chain, bydst) {
 661                 if (!delpol &&
 662                     pol->type == policy->type &&
 663                     !selector_cmp(&pol->selector, &policy->selector) &&
 664                     xfrm_sec_ctx_match(pol->security, policy->security)) {
 665                         if (excl) {
 666                                 write_unlock_bh(&xfrm_policy_lock);
 667                                 return -EEXIST;
 668                         }
 669                         delpol = pol;
 670                         if (policy->priority > pol->priority)
 671                                 continue;
 672                 } else if (policy->priority >= pol->priority) {
 673                         last = &pol->bydst;
 674                         continue;
 675                 }
 676                 if (!newpos)
 677                         newpos = &pol->bydst;
 678                 if (delpol)
 679                         break;
 680                 last = &pol->bydst;
 681         }
 682         if (!newpos)
 683                 newpos = last;
 684         if (newpos)
 685                 hlist_add_after(newpos, &policy->bydst);
 686         else
 687                 hlist_add_head(&policy->bydst, chain);
 688         xfrm_pol_hold(policy);
 689         xfrm_policy_count[dir]++;
 690         atomic_inc(&flow_cache_genid);
 691         if (delpol) {
 692                 hlist_del(&delpol->bydst);
 693                 hlist_del(&delpol->byidx);
 694                 xfrm_policy_count[dir]--;
 695         }
 696         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 697         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 698         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
 699         policy->curlft.use_time = 0;
 700         if (!mod_timer(&policy->timer, jiffies + HZ))
 701                 xfrm_pol_hold(policy);
 702         write_unlock_bh(&xfrm_policy_lock);
 703
 704         if (delpol)
 705                 xfrm_policy_kill(delpol);
 706         else if (xfrm_bydst_should_resize(dir, NULL))
 707                 schedule_work(&xfrm_hash_work);
 708
 709         read_lock_bh(&xfrm_policy_lock);
 710         gc_list = NULL;
 711         entry = &policy->bydst;
 712         hlist_for_each_entry_continue(policy, entry, bydst) {
 713                 struct dst_entry *dst;
 714
 715                 write_lock(&policy->lock);
 716                 dst = policy->bundles;
 717                 if (dst) {
 718                         struct dst_entry *tail = dst;
 719                         while (tail->next)
 720                                 tail = tail->next;
 721                         tail->next = gc_list;
 722                         gc_list = dst;
 723
 724                         policy->bundles = NULL;
 725                 }
 726                 write_unlock(&policy->lock);
 727         }
 728         read_unlock_bh(&xfrm_policy_lock);
 729
 730         while (gc_list) {
 731                 struct dst_entry *dst = gc_list;
 732
 733                 gc_list = dst->next;
 734                 dst_free(dst);
 735         }
 736
 737         return 0;
 738 }
 739 EXPORT_SYMBOL(xfrm_policy_insert);
 740
 741 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 742                                           struct xfrm_selector *sel,
 743                                           struct xfrm_sec_ctx *ctx, int delete)
 744 {
 745         struct xfrm_policy *pol, *ret;
 746         struct hlist_head *chain;
 747         struct hlist_node *entry;
 748
 749         write_lock_bh(&xfrm_policy_lock);
 750         chain = policy_hash_bysel(sel, sel->family, dir);
 751         ret = NULL;
 752         hlist_for_each_entry(pol, entry, chain, bydst) {
 753                 if (pol->type == type &&
 754                     !selector_cmp(sel, &pol->selector) &&
 755                     xfrm_sec_ctx_match(ctx, pol->security)) {
 756                         xfrm_pol_hold(pol);
 757                         if (delete) {
 758                                 hlist_del(&pol->bydst);
 759                                 hlist_del(&pol->byidx);
 760                                 xfrm_policy_count[dir]--;
 761                         }
 762                         ret = pol;
 763                         break;
 764                 }
 765         }
 766         write_unlock_bh(&xfrm_policy_lock);
 767
 768         if (ret && delete) {
 769                 atomic_inc(&flow_cache_genid);
 770                 xfrm_policy_kill(ret);
 771         }
 772         return ret;
 773 }
 774 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 775
 776 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete)
 777 {
 778         struct xfrm_policy *pol, *ret;
 779         struct hlist_head *chain;
 780         struct hlist_node *entry;
 781
 782         write_lock_bh(&xfrm_policy_lock);
 783         chain = xfrm_policy_byidx + idx_hash(id);
 784         ret = NULL;
 785         hlist_for_each_entry(pol, entry, chain, byidx) {
 786                 if (pol->type == type && pol->index == id) {
 787                         xfrm_pol_hold(pol);
 788                         if (delete) {
 789                                 hlist_del(&pol->bydst);
 790                                 hlist_del(&pol->byidx);
 791                                 xfrm_policy_count[dir]--;
 792                         }
 793                         ret = pol;
 794                         break;
 795                 }
 796         }
 797         write_unlock_bh(&xfrm_policy_lock);
 798
 799         if (ret && delete) {
 800                 atomic_inc(&flow_cache_genid);
 801                 xfrm_policy_kill(ret);
 802         }
 803         return ret;
 804 }
 805 EXPORT_SYMBOL(xfrm_policy_byid);
 806
 807 void xfrm_policy_flush(u8 type)
 808 {
 809         int dir;
 810
 811         write_lock_bh(&xfrm_policy_lock);
 812         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 813                 struct xfrm_policy *pol;
 814                 struct hlist_node *entry;
 815                 int i, killed;
 816
 817                 killed = 0;
 818         again1:
 819                 hlist_for_each_entry(pol, entry,
 820                                      &xfrm_policy_inexact[dir], bydst) {
 821                         if (pol->type != type)
 822                                 continue;
 823                         hlist_del(&pol->bydst);
 824                         hlist_del(&pol->byidx);
 825                         write_unlock_bh(&xfrm_policy_lock);
 826
 827                         xfrm_policy_kill(pol);
 828                         killed++;
 829
 830                         write_lock_bh(&xfrm_policy_lock);
 831                         goto again1;
 832                 }
 833
 834                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 835         again2:
 836                         hlist_for_each_entry(pol, entry,
 837                                              xfrm_policy_bydst[dir].table + i,
 838                                              bydst) {
 839                                 if (pol->type != type)
 840                                         continue;
 841                                 hlist_del(&pol->bydst);
 842                                 hlist_del(&pol->byidx);
 843                                 write_unlock_bh(&xfrm_policy_lock);
 844
 845                                 xfrm_policy_kill(pol);
 846                                 killed++;
 847
 848                                 write_lock_bh(&xfrm_policy_lock);
 849                                 goto again2;
 850                         }
 851                 }
 852
 853                 xfrm_policy_count[dir] -= killed;
 854         }
 855         atomic_inc(&flow_cache_genid);
 856         write_unlock_bh(&xfrm_policy_lock);
 857 }
 858 EXPORT_SYMBOL(xfrm_policy_flush);
 859
 860 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 861                      void *data)
 862 {
 863         struct xfrm_policy *pol, *last = NULL;
 864         struct hlist_node *entry;
 865         int dir, last_dir = 0, count, error;
 866
 867         read_lock_bh(&xfrm_policy_lock);
 868         count = 0;
 869
 870         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 871                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 872                 int i;
 873
 874                 hlist_for_each_entry(pol, entry,
 875                                      &xfrm_policy_inexact[dir], bydst) {
 876                         if (pol->type != type)
 877                                 continue;
 878                         if (last) {
 879                                 error = func(last, last_dir % XFRM_POLICY_MAX,
 880                                              count, data);
 881                                 if (error)
 882                                         goto out;
 883                         }
 884                         last = pol;
 885                         last_dir = dir;
 886                         count++;
 887                 }
 888                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 889                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 890                                 if (pol->type != type)
 891                                         continue;
 892                                 if (last) {
 893                                         error = func(last, last_dir % XFRM_POLICY_MAX,
 894                                                      count, data);
 895                                         if (error)
 896                                                 goto out;
 897                                 }
 898                                 last = pol;
 899                                 last_dir = dir;
 900                                 count++;
 901                         }
 902                 }
 903         }
 904         if (count == 0) {
 905                 error = -ENOENT;
 906                 goto out;
 907         }
 908         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
 909 out:
 910         read_unlock_bh(&xfrm_policy_lock);
 911         return error;
 912 }
 913 EXPORT_SYMBOL(xfrm_policy_walk);
 914
 915 /*
 916  * Find policy to apply to this flow.
 917  *
 918  * Returns 0 if policy found, else an -errno.
 919  */
 920 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 921                              u8 type, u16 family, int dir)
 922 {
 923         struct xfrm_selector *sel = &pol->selector;
 924         int match, ret = -ESRCH;
 925
 926         if (pol->family != family ||
 927             pol->type != type)
 928                 return ret;
 929
 930         match = xfrm_selector_match(sel, fl, family);
 931         if (match)
 932                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
 933
 934         return ret;
 935 }
 936
 937 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 938                                                      u16 family, u8 dir)
 939 {
 940         int err;
 941         struct xfrm_policy *pol, *ret;
 942         xfrm_address_t *daddr, *saddr;
 943         struct hlist_node *entry;
 944         struct hlist_head *chain;
 945         u32 priority = ~0U;
 946
 947         daddr = xfrm_flowi_daddr(fl, family);
 948         saddr = xfrm_flowi_saddr(fl, family);
 949         if (unlikely(!daddr || !saddr))
 950                 return NULL;
 951
 952         read_lock_bh(&xfrm_policy_lock);
 953         chain = policy_hash_direct(daddr, saddr, family, dir);
 954         ret = NULL;
 955         hlist_for_each_entry(pol, entry, chain, bydst) {
 956                 err = xfrm_policy_match(pol, fl, type, family, dir);
 957                 if (err) {
 958                         if (err == -ESRCH)
 959                                 continue;
 960                         else {
 961                                 ret = ERR_PTR(err);
 962                                 goto fail;
 963                         }
 964                 } else {
 965                         ret = pol;
 966                         priority = ret->priority;
 967                         break;
 968                 }
 969         }
 970         chain = &xfrm_policy_inexact[dir];
 971         hlist_for_each_entry(pol, entry, chain, bydst) {
 972                 err = xfrm_policy_match(pol, fl, type, family, dir);
 973                 if (err) {
 974                         if (err == -ESRCH)
 975                                 continue;
 976                         else {
 977                                 ret = ERR_PTR(err);
 978                                 goto fail;
 979                         }
 980                 } else if (pol->priority < priority) {
 981                         ret = pol;
 982                         break;
 983                 }
 984         }
 985         if (ret)
 986                 xfrm_pol_hold(ret);
 987 fail:
 988         read_unlock_bh(&xfrm_policy_lock);
 989
 990         return ret;
 991 }
 992
 993 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 994                                void **objp, atomic_t **obj_refp)
 995 {
 996         struct xfrm_policy *pol;
 997         int err = 0;
 998
 999 #ifdef CONFIG_XFRM_SUB_POLICY
1000         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1001         if (IS_ERR(pol)) {
1002                 err = PTR_ERR(pol);
1003                 pol = NULL;
1004         }
1005         if (pol || err)
1006                 goto end;
1007 #endif
1008         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1009         if (IS_ERR(pol)) {
1010                 err = PTR_ERR(pol);
1011                 pol = NULL;
1012         }
1013 #ifdef CONFIG_XFRM_SUB_POLICY
1014 end:
1015 #endif
1016         if ((*objp = (void *) pol) != NULL)
1017                 *obj_refp = &pol->refcnt;
1018         return err;
1019 }
1020
1021 static inline int policy_to_flow_dir(int dir)
1022 {
1023         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1024             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1025             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1026                 return dir;
1027         switch (dir) {
1028         default:
1029         case XFRM_POLICY_IN:
1030                 return FLOW_DIR_IN;
1031         case XFRM_POLICY_OUT:
1032                 return FLOW_DIR_OUT;
1033         case XFRM_POLICY_FWD:
1034                 return FLOW_DIR_FWD;
1035         };
1036 }
1037
1038 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1039 {
1040         struct xfrm_policy *pol;
1041
1042         read_lock_bh(&xfrm_policy_lock);
1043         if ((pol = sk->sk_policy[dir]) != NULL) {
1044                 int match = xfrm_selector_match(&pol->selector, fl,
1045                                                 sk->sk_family);
1046                 int err = 0;
1047
1048                 if (match) {
1049                         err = security_xfrm_policy_lookup(pol, fl->secid,
1050                                         policy_to_flow_dir(dir));
1051                         if (!err)
1052                                 xfrm_pol_hold(pol);
1053                         else if (err == -ESRCH)
1054                                 pol = NULL;
1055                         else
1056                                 pol = ERR_PTR(err);
1057                 } else
1058                         pol = NULL;
1059         }
1060         read_unlock_bh(&xfrm_policy_lock);
1061         return pol;
1062 }
1063
1064 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1065 {
1066         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1067                                                      pol->family, dir);
1068
1069         hlist_add_head(&pol->bydst, chain);
1070         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1071         xfrm_policy_count[dir]++;
1072         xfrm_pol_hold(pol);
1073
1074         if (xfrm_bydst_should_resize(dir, NULL))
1075                 schedule_work(&xfrm_hash_work);
1076 }
1077
1078 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1079                                                 int dir)
1080 {
1081         if (hlist_unhashed(&pol->bydst))
1082                 return NULL;
1083
1084         hlist_del(&pol->bydst);
1085         hlist_del(&pol->byidx);
1086         xfrm_policy_count[dir]--;
1087
1088         return pol;
1089 }
1090
1091 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1092 {
1093         write_lock_bh(&xfrm_policy_lock);
1094         pol = __xfrm_policy_unlink(pol, dir);
1095         write_unlock_bh(&xfrm_policy_lock);
1096         if (pol) {
1097                 if (dir < XFRM_POLICY_MAX)
1098                         atomic_inc(&flow_cache_genid);
1099                 xfrm_policy_kill(pol);
1100                 return 0;
1101         }
1102         return -ENOENT;
1103 }
1104 EXPORT_SYMBOL(xfrm_policy_delete);
1105
1106 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1107 {
1108         struct xfrm_policy *old_pol;
1109
1110 #ifdef CONFIG_XFRM_SUB_POLICY
1111         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1112                 return -EINVAL;
1113 #endif
1114
1115         write_lock_bh(&xfrm_policy_lock);
1116         old_pol = sk->sk_policy[dir];
1117         sk->sk_policy[dir] = pol;
1118         if (pol) {
1119                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
1120                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1121                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1122         }
1123         if (old_pol)
1124                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1125         write_unlock_bh(&xfrm_policy_lock);
1126
1127         if (old_pol) {
1128                 xfrm_policy_kill(old_pol);
1129         }
1130         return 0;
1131 }
1132
1133 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1134 {
1135         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1136
1137         if (newp) {
1138                 newp->selector = old->selector;
1139                 if (security_xfrm_policy_clone(old, newp)) {
1140                         kfree(newp);
1141                         return NULL;  /* ENOMEM */
1142                 }
1143                 newp->lft = old->lft;
1144                 newp->curlft = old->curlft;
1145                 newp->action = old->action;
1146                 newp->flags = old->flags;
1147                 newp->xfrm_nr = old->xfrm_nr;
1148                 newp->index = old->index;
1149                 newp->type = old->type;
1150                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1151                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1152                 write_lock_bh(&xfrm_policy_lock);
1153                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1154                 write_unlock_bh(&xfrm_policy_lock);
1155                 xfrm_pol_put(newp);
1156         }
1157         return newp;
1158 }
1159
1160 int __xfrm_sk_clone_policy(struct sock *sk)
1161 {
1162         struct xfrm_policy *p0 = sk->sk_policy[0],
1163                            *p1 = sk->sk_policy[1];
1164
1165         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1166         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1167                 return -ENOMEM;
1168         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1169                 return -ENOMEM;
1170         return 0;
1171 }
1172
1173 static int
1174 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1175                unsigned short family)
1176 {
1177         int err;
1178         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1179
1180         if (unlikely(afinfo == NULL))
1181                 return -EINVAL;
1182         err = afinfo->get_saddr(local, remote);
1183         xfrm_policy_put_afinfo(afinfo);
1184         return err;
1185 }
1186
1187 /* Resolve list of templates for the flow, given policy. */
1188
1189 static int
1190 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1191                       struct xfrm_state **xfrm,
1192                       unsigned short family)
1193 {
1194         int nx;
1195         int i, error;
1196         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1197         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1198         xfrm_address_t tmp;
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                         family = tmpl->encap_family;
1210                         if (xfrm_addr_any(local, family)) {
1211                                 error = xfrm_get_saddr(&tmp, remote, family);
1212                                 if (error)
1213                                         goto fail;
1214                                 local = &tmp;
1215                         }
1216                 }
1217
1218                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1219
1220                 if (x && x->km.state == XFRM_STATE_VALID) {
1221                         xfrm[nx++] = x;
1222                         daddr = remote;
1223                         saddr = local;
1224                         continue;
1225                 }
1226                 if (x) {
1227                         error = (x->km.state == XFRM_STATE_ERROR ?
1228                                  -EINVAL : -EAGAIN);
1229                         xfrm_state_put(x);
1230                 }
1231
1232                 if (!tmpl->optional)
1233                         goto fail;
1234         }
1235         return nx;
1236
1237 fail:
1238         for (nx--; nx>=0; nx--)
1239                 xfrm_state_put(xfrm[nx]);
1240         return error;
1241 }
1242
1243 static int
1244 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1245                   struct xfrm_state **xfrm,
1246                   unsigned short family)
1247 {
1248         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1249         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1250         int cnx = 0;
1251         int error;
1252         int ret;
1253         int i;
1254
1255         for (i = 0; i < npols; i++) {
1256                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1257                         error = -ENOBUFS;
1258                         goto fail;
1259                 }
1260
1261                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1262                 if (ret < 0) {
1263                         error = ret;
1264                         goto fail;
1265                 } else
1266                         cnx += ret;
1267         }
1268
1269         /* found states are sorted for outbound processing */
1270         if (npols > 1)
1271                 xfrm_state_sort(xfrm, tpp, cnx, family);
1272
1273         return cnx;
1274
1275  fail:
1276         for (cnx--; cnx>=0; cnx--)
1277                 xfrm_state_put(tpp[cnx]);
1278         return error;
1279
1280 }
1281
1282 /* Check that the bundle accepts the flow and its components are
1283  * still valid.
1284  */
1285
1286 static struct dst_entry *
1287 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1288 {
1289         struct dst_entry *x;
1290         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1291         if (unlikely(afinfo == NULL))
1292                 return ERR_PTR(-EINVAL);
1293         x = afinfo->find_bundle(fl, policy);
1294         xfrm_policy_put_afinfo(afinfo);
1295         return x;
1296 }
1297
1298 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1299  * all the metrics... Shortly, bundle a bundle.
1300  */
1301
1302 static int
1303 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1304                    struct flowi *fl, struct dst_entry **dst_p,
1305                    unsigned short family)
1306 {
1307         int err;
1308         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1309         if (unlikely(afinfo == NULL))
1310                 return -EINVAL;
1311         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1312         xfrm_policy_put_afinfo(afinfo);
1313         return err;
1314 }
1315
1316
1317 static int stale_bundle(struct dst_entry *dst);
1318
1319 /* Main function: finds/creates a bundle for given flow.
1320  *
1321  * At the moment we eat a raw IP route. Mostly to speed up lookups
1322  * on interfaces with disabled IPsec.
1323  */
1324 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1325                 struct sock *sk, int flags)
1326 {
1327         struct xfrm_policy *policy;
1328         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1329         int npols;
1330         int pol_dead;
1331         int xfrm_nr;
1332         int pi;
1333         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1334         struct dst_entry *dst, *dst_orig = *dst_p;
1335         int nx = 0;
1336         int err;
1337         u32 genid;
1338         u16 family;
1339         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1340
1341 restart:
1342         genid = atomic_read(&flow_cache_genid);
1343         policy = NULL;
1344         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1345                 pols[pi] = NULL;
1346         npols = 0;
1347         pol_dead = 0;
1348         xfrm_nr = 0;
1349
1350         if (sk && sk->sk_policy[1]) {
1351                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1352                 if (IS_ERR(policy))
1353                         return PTR_ERR(policy);
1354         }
1355
1356         if (!policy) {
1357                 /* To accelerate a bit...  */
1358                 if ((dst_orig->flags & DST_NOXFRM) ||
1359                     !xfrm_policy_count[XFRM_POLICY_OUT])
1360                         return 0;
1361
1362                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1363                                            dir, xfrm_policy_lookup);
1364                 if (IS_ERR(policy))
1365                         return PTR_ERR(policy);
1366         }
1367
1368         if (!policy)
1369                 return 0;
1370
1371         family = dst_orig->ops->family;
1372         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
1373         pols[0] = policy;
1374         npols ++;
1375         xfrm_nr += pols[0]->xfrm_nr;
1376
1377         switch (policy->action) {
1378         case XFRM_POLICY_BLOCK:
1379                 /* Prohibit the flow */
1380                 err = -EPERM;
1381                 goto error;
1382
1383         case XFRM_POLICY_ALLOW:
1384 #ifndef CONFIG_XFRM_SUB_POLICY
1385                 if (policy->xfrm_nr == 0) {
1386                         /* Flow passes not transformed. */
1387                         xfrm_pol_put(policy);
1388                         return 0;
1389                 }
1390 #endif
1391
1392                 /* Try to find matching bundle.
1393                  *
1394                  * LATER: help from flow cache. It is optional, this
1395                  * is required only for output policy.
1396                  */
1397                 dst = xfrm_find_bundle(fl, policy, family);
1398                 if (IS_ERR(dst)) {
1399                         err = PTR_ERR(dst);
1400                         goto error;
1401                 }
1402
1403                 if (dst)
1404                         break;
1405
1406 #ifdef CONFIG_XFRM_SUB_POLICY
1407                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1408                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1409                                                             fl, family,
1410                                                             XFRM_POLICY_OUT);
1411                         if (pols[1]) {
1412                                 if (IS_ERR(pols[1])) {
1413                                         err = PTR_ERR(pols[1]);
1414                                         goto error;
1415                                 }
1416                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1417                                         err = -EPERM;
1418                                         goto error;
1419                                 }
1420                                 npols ++;
1421                                 xfrm_nr += pols[1]->xfrm_nr;
1422                         }
1423                 }
1424
1425                 /*
1426                  * Because neither flowi nor bundle information knows about
1427                  * transformation template size. On more than one policy usage
1428                  * we can realize whether all of them is bypass or not after
1429                  * they are searched. See above not-transformed bypass
1430                  * is surrounded by non-sub policy configuration, too.
1431                  */
1432                 if (xfrm_nr == 0) {
1433                         /* Flow passes not transformed. */
1434                         xfrm_pols_put(pols, npols);
1435                         return 0;
1436                 }
1437
1438 #endif
1439                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1440
1441                 if (unlikely(nx<0)) {
1442                         err = nx;
1443                         if (err == -EAGAIN && flags) {
1444                                 DECLARE_WAITQUEUE(wait, current);
1445
1446                                 add_wait_queue(&km_waitq, &wait);
1447                                 set_current_state(TASK_INTERRUPTIBLE);
1448                                 schedule();
1449                                 set_current_state(TASK_RUNNING);
1450                                 remove_wait_queue(&km_waitq, &wait);
1451
1452                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1453
1454                                 if (nx == -EAGAIN && signal_pending(current)) {
1455                                         err = -ERESTART;
1456                                         goto error;
1457                                 }
1458                                 if (nx == -EAGAIN ||
1459                                     genid != atomic_read(&flow_cache_genid)) {
1460                                         xfrm_pols_put(pols, npols);
1461                                         goto restart;
1462                                 }
1463                                 err = nx;
1464                         }
1465                         if (err < 0)
1466                                 goto error;
1467                 }
1468                 if (nx == 0) {
1469                         /* Flow passes not transformed. */
1470                         xfrm_pols_put(pols, npols);
1471                         return 0;
1472                 }
1473
1474                 dst = dst_orig;
1475                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1476
1477                 if (unlikely(err)) {
1478                         int i;
1479                         for (i=0; i<nx; i++)
1480                                 xfrm_state_put(xfrm[i]);
1481                         goto error;
1482                 }
1483
1484                 for (pi = 0; pi < npols; pi++) {
1485                         read_lock_bh(&pols[pi]->lock);
1486                         pol_dead |= pols[pi]->dead;
1487                         read_unlock_bh(&pols[pi]->lock);
1488                 }
1489
1490                 write_lock_bh(&policy->lock);
1491                 if (unlikely(pol_dead || stale_bundle(dst))) {
1492                         /* Wow! While we worked on resolving, this
1493                          * policy has gone. Retry. It is not paranoia,
1494                          * we just cannot enlist new bundle to dead object.
1495                          * We can't enlist stable bundles either.
1496                          */
1497                         write_unlock_bh(&policy->lock);
1498                         if (dst)
1499                                 dst_free(dst);
1500
1501                         err = -EHOSTUNREACH;
1502                         goto error;
1503                 }
1504                 dst->next = policy->bundles;
1505                 policy->bundles = dst;
1506                 dst_hold(dst);
1507                 write_unlock_bh(&policy->lock);
1508         }
1509         *dst_p = dst;
1510         dst_release(dst_orig);
1511         xfrm_pols_put(pols, npols);
1512         return 0;
1513
1514 error:
1515         dst_release(dst_orig);
1516         xfrm_pols_put(pols, npols);
1517         *dst_p = NULL;
1518         return err;
1519 }
1520 EXPORT_SYMBOL(xfrm_lookup);
1521
1522 static inline int
1523 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1524 {
1525         struct xfrm_state *x;
1526         int err;
1527
1528         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1529                 return 0;
1530         x = skb->sp->xvec[idx];
1531         if (!x->type->reject)
1532                 return 0;
1533         xfrm_state_hold(x);
1534         err = x->type->reject(x, skb, fl);
1535         xfrm_state_put(x);
1536         return err;
1537 }
1538
1539 /* When skb is transformed back to its "native" form, we have to
1540  * check policy restrictions. At the moment we make this in maximally
1541  * stupid way. Shame on me. :-) Of course, connected sockets must
1542  * have policy cached at them.
1543  */
1544
1545 static inline int
1546 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1547               unsigned short family)
1548 {
1549         if (xfrm_state_kern(x))
1550                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1551         return  x->id.proto == tmpl->id.proto &&
1552                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1553                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1554                 x->props.mode == tmpl->mode &&
1555                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1556                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1557                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1558                   xfrm_state_addr_cmp(tmpl, x, family));
1559 }
1560
1561 /*
1562  * 0 or more than 0 is returned when validation is succeeded (either bypass
1563  * because of optional transport mode, or next index of the mathced secpath
1564  * state with the template.
1565  * -1 is returned when no matching template is found.
1566  * Otherwise "-2 - errored_index" is returned.
1567  */
1568 static inline int
1569 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1570                unsigned short family)
1571 {
1572         int idx = start;
1573
1574         if (tmpl->optional) {
1575                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1576                         return start;
1577         } else
1578                 start = -1;
1579         for (; idx < sp->len; idx++) {
1580                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1581                         return ++idx;
1582                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1583                         if (start == -1)
1584                                 start = -2-idx;
1585                         break;
1586                 }
1587         }
1588         return start;
1589 }
1590
1591 int
1592 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1593 {
1594         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1595         int err;
1596
1597         if (unlikely(afinfo == NULL))
1598                 return -EAFNOSUPPORT;
1599
1600         afinfo->decode_session(skb, fl);
1601         err = security_xfrm_decode_session(skb, &fl->secid);
1602         xfrm_policy_put_afinfo(afinfo);
1603         return err;
1604 }
1605 EXPORT_SYMBOL(xfrm_decode_session);
1606
1607 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1608 {
1609         for (; k < sp->len; k++) {
1610                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1611                         *idxp = k;
1612                         return 1;
1613                 }
1614         }
1615
1616         return 0;
1617 }
1618
1619 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1620                         unsigned short family)
1621 {
1622         struct xfrm_policy *pol;
1623         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1624         int npols = 0;
1625         int xfrm_nr;
1626         int pi;
1627         struct flowi fl;
1628         u8 fl_dir = policy_to_flow_dir(dir);
1629         int xerr_idx = -1;
1630
1631         if (xfrm_decode_session(skb, &fl, family) < 0)
1632                 return 0;
1633         nf_nat_decode_session(skb, &fl, family);
1634
1635         /* First, check used SA against their selectors. */
1636         if (skb->sp) {
1637                 int i;
1638
1639                 for (i=skb->sp->len-1; i>=0; i--) {
1640                         struct xfrm_state *x = skb->sp->xvec[i];
1641                         if (!xfrm_selector_match(&x->sel, &fl, family))
1642                                 return 0;
1643                 }
1644         }
1645
1646         pol = NULL;
1647         if (sk && sk->sk_policy[dir]) {
1648                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1649                 if (IS_ERR(pol))
1650                         return 0;
1651         }
1652
1653         if (!pol)
1654                 pol = flow_cache_lookup(&fl, family, fl_dir,
1655                                         xfrm_policy_lookup);
1656
1657         if (IS_ERR(pol))
1658                 return 0;
1659
1660         if (!pol) {
1661                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1662                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1663                         return 0;
1664                 }
1665                 return 1;
1666         }
1667
1668         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1669
1670         pols[0] = pol;
1671         npols ++;
1672 #ifdef CONFIG_XFRM_SUB_POLICY
1673         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1674                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1675                                                     &fl, family,
1676                                                     XFRM_POLICY_IN);
1677                 if (pols[1]) {
1678                         if (IS_ERR(pols[1]))
1679                                 return 0;
1680                         pols[1]->curlft.use_time = (unsigned long)xtime.tv_sec;
1681                         npols ++;
1682                 }
1683         }
1684 #endif
1685
1686         if (pol->action == XFRM_POLICY_ALLOW) {
1687                 struct sec_path *sp;
1688                 static struct sec_path dummy;
1689                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1690                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1691                 struct xfrm_tmpl **tpp = tp;
1692                 int ti = 0;
1693                 int i, k;
1694
1695                 if ((sp = skb->sp) == NULL)
1696                         sp = &dummy;
1697
1698                 for (pi = 0; pi < npols; pi++) {
1699                         if (pols[pi] != pol &&
1700                             pols[pi]->action != XFRM_POLICY_ALLOW)
1701                                 goto reject;
1702                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1703                                 goto reject_error;
1704                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1705                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1706                 }
1707                 xfrm_nr = ti;
1708                 if (npols > 1) {
1709                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1710                         tpp = stp;
1711                 }
1712
1713                 /* For each tunnel xfrm, find the first matching tmpl.
1714                  * For each tmpl before that, find corresponding xfrm.
1715                  * Order is _important_. Later we will implement
1716                  * some barriers, but at the moment barriers
1717                  * are implied between each two transformations.
1718                  */
1719                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1720                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1721                         if (k < 0) {
1722                                 if (k < -1)
1723                                         /* "-2 - errored_index" returned */
1724                                         xerr_idx = -(2+k);
1725                                 goto reject;
1726                         }
1727                 }
1728
1729                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1730                         goto reject;
1731
1732                 xfrm_pols_put(pols, npols);
1733                 return 1;
1734         }
1735
1736 reject:
1737         xfrm_secpath_reject(xerr_idx, skb, &fl);
1738 reject_error:
1739         xfrm_pols_put(pols, npols);
1740         return 0;
1741 }
1742 EXPORT_SYMBOL(__xfrm_policy_check);
1743
1744 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1745 {
1746         struct flowi fl;
1747
1748         if (xfrm_decode_session(skb, &fl, family) < 0)
1749                 return 0;
1750
1751         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1752 }
1753 EXPORT_SYMBOL(__xfrm_route_forward);
1754
1755 /* Optimize later using cookies and generation ids. */
1756
1757 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1758 {
1759         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1760          * to "-1" to force all XFRM destinations to get validated by
1761          * dst_ops->check on every use.  We do this because when a
1762          * normal route referenced by an XFRM dst is obsoleted we do
1763          * not go looking around for all parent referencing XFRM dsts
1764          * so that we can invalidate them.  It is just too much work.
1765          * Instead we make the checks here on every use.  For example:
1766          *
1767          *      XFRM dst A --> IPv4 dst X
1768          *
1769          * X is the "xdst->route" of A (X is also the "dst->path" of A
1770          * in this example).  If X is marked obsolete, "A" will not
1771          * notice.  That's what we are validating here via the
1772          * stale_bundle() check.
1773          *
1774          * When a policy's bundle is pruned, we dst_free() the XFRM
1775          * dst which causes it's ->obsolete field to be set to a
1776          * positive non-zero integer.  If an XFRM dst has been pruned
1777          * like this, we want to force a new route lookup.
1778          */
1779         if (dst->obsolete < 0 && !stale_bundle(dst))
1780                 return dst;
1781
1782         return NULL;
1783 }
1784
1785 static int stale_bundle(struct dst_entry *dst)
1786 {
1787         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1788 }
1789
1790 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1791 {
1792         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1793                 dst->dev = &loopback_dev;
1794                 dev_hold(&loopback_dev);
1795                 dev_put(dev);
1796         }
1797 }
1798 EXPORT_SYMBOL(xfrm_dst_ifdown);
1799
1800 static void xfrm_link_failure(struct sk_buff *skb)
1801 {
1802         /* Impossible. Such dst must be popped before reaches point of failure. */
1803         return;
1804 }
1805
1806 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1807 {
1808         if (dst) {
1809                 if (dst->obsolete) {
1810                         dst_release(dst);
1811                         dst = NULL;
1812                 }
1813         }
1814         return dst;
1815 }
1816
1817 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1818 {
1819         struct dst_entry *dst, **dstp;
1820
1821         write_lock(&pol->lock);
1822         dstp = &pol->bundles;
1823         while ((dst=*dstp) != NULL) {
1824                 if (func(dst)) {
1825                         *dstp = dst->next;
1826                         dst->next = *gc_list_p;
1827                         *gc_list_p = dst;
1828                 } else {
1829                         dstp = &dst->next;
1830                 }
1831         }
1832         write_unlock(&pol->lock);
1833 }
1834
1835 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1836 {
1837         struct dst_entry *gc_list = NULL;
1838         int dir;
1839
1840         read_lock_bh(&xfrm_policy_lock);
1841         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1842                 struct xfrm_policy *pol;
1843                 struct hlist_node *entry;
1844                 struct hlist_head *table;
1845                 int i;
1846
1847                 hlist_for_each_entry(pol, entry,
1848                                      &xfrm_policy_inexact[dir], bydst)
1849                         prune_one_bundle(pol, func, &gc_list);
1850
1851                 table = xfrm_policy_bydst[dir].table;
1852                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1853                         hlist_for_each_entry(pol, entry, table + i, bydst)
1854                                 prune_one_bundle(pol, func, &gc_list);
1855                 }
1856         }
1857         read_unlock_bh(&xfrm_policy_lock);
1858
1859         while (gc_list) {
1860                 struct dst_entry *dst = gc_list;
1861                 gc_list = dst->next;
1862                 dst_free(dst);
1863         }
1864 }
1865
1866 static int unused_bundle(struct dst_entry *dst)
1867 {
1868         return !atomic_read(&dst->__refcnt);
1869 }
1870
1871 static void __xfrm_garbage_collect(void)
1872 {
1873         xfrm_prune_bundles(unused_bundle);
1874 }
1875
1876 static int xfrm_flush_bundles(void)
1877 {
1878         xfrm_prune_bundles(stale_bundle);
1879         return 0;
1880 }
1881
1882 void xfrm_init_pmtu(struct dst_entry *dst)
1883 {
1884         do {
1885                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1886                 u32 pmtu, route_mtu_cached;
1887
1888                 pmtu = dst_mtu(dst->child);
1889                 xdst->child_mtu_cached = pmtu;
1890
1891                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1892
1893                 route_mtu_cached = dst_mtu(xdst->route);
1894                 xdst->route_mtu_cached = route_mtu_cached;
1895
1896                 if (pmtu > route_mtu_cached)
1897                         pmtu = route_mtu_cached;
1898
1899                 dst->metrics[RTAX_MTU-1] = pmtu;
1900         } while ((dst = dst->next));
1901 }
1902
1903 EXPORT_SYMBOL(xfrm_init_pmtu);
1904
1905 /* Check that the bundle accepts the flow and its components are
1906  * still valid.
1907  */
1908
1909 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
1910                 struct flowi *fl, int family, int strict)
1911 {
1912         struct dst_entry *dst = &first->u.dst;
1913         struct xfrm_dst *last;
1914         u32 mtu;
1915
1916         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1917             (dst->dev && !netif_running(dst->dev)))
1918                 return 0;
1919
1920         last = NULL;
1921
1922         do {
1923                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1924
1925                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1926                         return 0;
1927                 if (fl && pol &&
1928                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
1929                         return 0;
1930                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1931                         return 0;
1932                 if (xdst->genid != dst->xfrm->genid)
1933                         return 0;
1934
1935                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
1936                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1937                         return 0;
1938
1939                 mtu = dst_mtu(dst->child);
1940                 if (xdst->child_mtu_cached != mtu) {
1941                         last = xdst;
1942                         xdst->child_mtu_cached = mtu;
1943                 }
1944
1945                 if (!dst_check(xdst->route, xdst->route_cookie))
1946                         return 0;
1947                 mtu = dst_mtu(xdst->route);
1948                 if (xdst->route_mtu_cached != mtu) {
1949                         last = xdst;
1950                         xdst->route_mtu_cached = mtu;
1951                 }
1952
1953                 dst = dst->child;
1954         } while (dst->xfrm);
1955
1956         if (likely(!last))
1957                 return 1;
1958
1959         mtu = last->child_mtu_cached;
1960         for (;;) {
1961                 dst = &last->u.dst;
1962
1963                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1964                 if (mtu > last->route_mtu_cached)
1965                         mtu = last->route_mtu_cached;
1966                 dst->metrics[RTAX_MTU-1] = mtu;
1967
1968                 if (last == first)
1969                         break;
1970
1971                 last = last->u.next;
1972                 last->child_mtu_cached = mtu;
1973         }
1974
1975         return 1;
1976 }
1977
1978 EXPORT_SYMBOL(xfrm_bundle_ok);
1979
1980 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1981 {
1982         int err = 0;
1983         if (unlikely(afinfo == NULL))
1984                 return -EINVAL;
1985         if (unlikely(afinfo->family >= NPROTO))
1986                 return -EAFNOSUPPORT;
1987         write_lock_bh(&xfrm_policy_afinfo_lock);
1988         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1989                 err = -ENOBUFS;
1990         else {
1991                 struct dst_ops *dst_ops = afinfo->dst_ops;
1992                 if (likely(dst_ops->kmem_cachep == NULL))
1993                         dst_ops->kmem_cachep = xfrm_dst_cache;
1994                 if (likely(dst_ops->check == NULL))
1995                         dst_ops->check = xfrm_dst_check;
1996                 if (likely(dst_ops->negative_advice == NULL))
1997                         dst_ops->negative_advice = xfrm_negative_advice;
1998                 if (likely(dst_ops->link_failure == NULL))
1999                         dst_ops->link_failure = xfrm_link_failure;
2000                 if (likely(afinfo->garbage_collect == NULL))
2001                         afinfo->garbage_collect = __xfrm_garbage_collect;
2002                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2003         }
2004         write_unlock_bh(&xfrm_policy_afinfo_lock);
2005         return err;
2006 }
2007 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2008
2009 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2010 {
2011         int err = 0;
2012         if (unlikely(afinfo == NULL))
2013                 return -EINVAL;
2014         if (unlikely(afinfo->family >= NPROTO))
2015                 return -EAFNOSUPPORT;
2016         write_lock_bh(&xfrm_policy_afinfo_lock);
2017         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2018                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2019                         err = -EINVAL;
2020                 else {
2021                         struct dst_ops *dst_ops = afinfo->dst_ops;
2022                         xfrm_policy_afinfo[afinfo->family] = NULL;
2023                         dst_ops->kmem_cachep = NULL;
2024                         dst_ops->check = NULL;
2025                         dst_ops->negative_advice = NULL;
2026                         dst_ops->link_failure = NULL;
2027                         afinfo->garbage_collect = NULL;
2028                 }
2029         }
2030         write_unlock_bh(&xfrm_policy_afinfo_lock);
2031         return err;
2032 }
2033 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2034
2035 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2036 {
2037         struct xfrm_policy_afinfo *afinfo;
2038         if (unlikely(family >= NPROTO))
2039                 return NULL;
2040         read_lock(&xfrm_policy_afinfo_lock);
2041         afinfo = xfrm_policy_afinfo[family];
2042         if (unlikely(!afinfo))
2043                 read_unlock(&xfrm_policy_afinfo_lock);
2044         return afinfo;
2045 }
2046
2047 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2048 {
2049         read_unlock(&xfrm_policy_afinfo_lock);
2050 }
2051
2052 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2053 {
2054         struct xfrm_policy_afinfo *afinfo;
2055         if (unlikely(family >= NPROTO))
2056                 return NULL;
2057         write_lock_bh(&xfrm_policy_afinfo_lock);
2058         afinfo = xfrm_policy_afinfo[family];
2059         if (unlikely(!afinfo))
2060                 write_unlock_bh(&xfrm_policy_afinfo_lock);
2061         return afinfo;
2062 }
2063
2064 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2065 {
2066         write_unlock_bh(&xfrm_policy_afinfo_lock);
2067 }
2068
2069 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2070 {
2071         switch (event) {
2072         case NETDEV_DOWN:
2073                 xfrm_flush_bundles();
2074         }
2075         return NOTIFY_DONE;
2076 }
2077
2078 static struct notifier_block xfrm_dev_notifier = {
2079         xfrm_dev_event,
2080         NULL,
2081         0
2082 };
2083
2084 static void __init xfrm_policy_init(void)
2085 {
2086         unsigned int hmask, sz;
2087         int dir;
2088
2089         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2090                                            sizeof(struct xfrm_dst),
2091                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2092                                            NULL, NULL);
2093
2094         hmask = 8 - 1;
2095         sz = (hmask+1) * sizeof(struct hlist_head);
2096
2097         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2098         xfrm_idx_hmask = hmask;
2099         if (!xfrm_policy_byidx)
2100                 panic("XFRM: failed to allocate byidx hash\n");
2101
2102         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2103                 struct xfrm_policy_hash *htab;
2104
2105                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2106
2107                 htab = &xfrm_policy_bydst[dir];
2108                 htab->table = xfrm_hash_alloc(sz);
2109                 htab->hmask = hmask;
2110                 if (!htab->table)
2111                         panic("XFRM: failed to allocate bydst hash\n");
2112         }
2113
2114         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2115         register_netdevice_notifier(&xfrm_dev_notifier);
2116 }
2117
2118 void __init xfrm_init(void)
2119 {
2120         xfrm_state_init();
2121         xfrm_policy_init();
2122         xfrm_input_init();
2123 }
2124