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 int sysctl_xfrm_larval_drop __read_mostly;
  32
  33 DEFINE_MUTEX(xfrm_cfg_mutex);
  34 EXPORT_SYMBOL(xfrm_cfg_mutex);
  35
  36 static DEFINE_RWLOCK(xfrm_policy_lock);
  37
  38 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  39 EXPORT_SYMBOL(xfrm_policy_count);
  40
  41 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  42 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  43
  44 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  45
  46 static struct work_struct xfrm_policy_gc_work;
  47 static HLIST_HEAD(xfrm_policy_gc_list);
  48 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  49
  50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  52 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
  53 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
  54
  55 static inline int
  56 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  57 {
  58         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  59                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  60                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  61                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  62                 (fl->proto == sel->proto || !sel->proto) &&
  63                 (fl->oif == sel->ifindex || !sel->ifindex);
  64 }
  65
  66 static inline int
  67 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  68 {
  69         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  70                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  71                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  72                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  73                 (fl->proto == sel->proto || !sel->proto) &&
  74                 (fl->oif == sel->ifindex || !sel->ifindex);
  75 }
  76
  77 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  78                     unsigned short family)
  79 {
  80         switch (family) {
  81         case AF_INET:
  82                 return __xfrm4_selector_match(sel, fl);
  83         case AF_INET6:
  84                 return __xfrm6_selector_match(sel, fl);
  85         }
  86         return 0;
  87 }
  88
  89 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
  90 {
  91         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  92         struct xfrm_type **typemap;
  93         int err = 0;
  94
  95         if (unlikely(afinfo == NULL))
  96                 return -EAFNOSUPPORT;
  97         typemap = afinfo->type_map;
  98
  99         if (likely(typemap[type->proto] == NULL))
 100                 typemap[type->proto] = type;
 101         else
 102                 err = -EEXIST;
 103         xfrm_policy_unlock_afinfo(afinfo);
 104         return err;
 105 }
 106 EXPORT_SYMBOL(xfrm_register_type);
 107
 108 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
 109 {
 110         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
 111         struct xfrm_type **typemap;
 112         int err = 0;
 113
 114         if (unlikely(afinfo == NULL))
 115                 return -EAFNOSUPPORT;
 116         typemap = afinfo->type_map;
 117
 118         if (unlikely(typemap[type->proto] != type))
 119                 err = -ENOENT;
 120         else
 121                 typemap[type->proto] = NULL;
 122         xfrm_policy_unlock_afinfo(afinfo);
 123         return err;
 124 }
 125 EXPORT_SYMBOL(xfrm_unregister_type);
 126
 127 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
 128 {
 129         struct xfrm_policy_afinfo *afinfo;
 130         struct xfrm_type **typemap;
 131         struct xfrm_type *type;
 132         int modload_attempted = 0;
 133
 134 retry:
 135         afinfo = xfrm_policy_get_afinfo(family);
 136         if (unlikely(afinfo == NULL))
 137                 return NULL;
 138         typemap = afinfo->type_map;
 139
 140         type = typemap[proto];
 141         if (unlikely(type && !try_module_get(type->owner)))
 142                 type = NULL;
 143         if (!type && !modload_attempted) {
 144                 xfrm_policy_put_afinfo(afinfo);
 145                 request_module("xfrm-type-%d-%d",
 146                                (int) family, (int) proto);
 147                 modload_attempted = 1;
 148                 goto retry;
 149         }
 150
 151         xfrm_policy_put_afinfo(afinfo);
 152         return type;
 153 }
 154
 155 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
 156                     unsigned short family)
 157 {
 158         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 159         int err = 0;
 160
 161         if (unlikely(afinfo == NULL))
 162                 return -EAFNOSUPPORT;
 163
 164         if (likely(afinfo->dst_lookup != NULL))
 165                 err = afinfo->dst_lookup(dst, fl);
 166         else
 167                 err = -EINVAL;
 168         xfrm_policy_put_afinfo(afinfo);
 169         return err;
 170 }
 171 EXPORT_SYMBOL(xfrm_dst_lookup);
 172
 173 void xfrm_put_type(struct xfrm_type *type)
 174 {
 175         module_put(type->owner);
 176 }
 177
 178 int xfrm_register_mode(struct xfrm_mode *mode, int family)
 179 {
 180         struct xfrm_policy_afinfo *afinfo;
 181         struct xfrm_mode **modemap;
 182         int err;
 183
 184         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 185                 return -EINVAL;
 186
 187         afinfo = xfrm_policy_lock_afinfo(family);
 188         if (unlikely(afinfo == NULL))
 189                 return -EAFNOSUPPORT;
 190
 191         err = -EEXIST;
 192         modemap = afinfo->mode_map;
 193         if (likely(modemap[mode->encap] == NULL)) {
 194                 modemap[mode->encap] = mode;
 195                 err = 0;
 196         }
 197
 198         xfrm_policy_unlock_afinfo(afinfo);
 199         return err;
 200 }
 201 EXPORT_SYMBOL(xfrm_register_mode);
 202
 203 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
 204 {
 205         struct xfrm_policy_afinfo *afinfo;
 206         struct xfrm_mode **modemap;
 207         int err;
 208
 209         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 210                 return -EINVAL;
 211
 212         afinfo = xfrm_policy_lock_afinfo(family);
 213         if (unlikely(afinfo == NULL))
 214                 return -EAFNOSUPPORT;
 215
 216         err = -ENOENT;
 217         modemap = afinfo->mode_map;
 218         if (likely(modemap[mode->encap] == mode)) {
 219                 modemap[mode->encap] = NULL;
 220                 err = 0;
 221         }
 222
 223         xfrm_policy_unlock_afinfo(afinfo);
 224         return err;
 225 }
 226 EXPORT_SYMBOL(xfrm_unregister_mode);
 227
 228 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
 229 {
 230         struct xfrm_policy_afinfo *afinfo;
 231         struct xfrm_mode *mode;
 232         int modload_attempted = 0;
 233
 234         if (unlikely(encap >= XFRM_MODE_MAX))
 235                 return NULL;
 236
 237 retry:
 238         afinfo = xfrm_policy_get_afinfo(family);
 239         if (unlikely(afinfo == NULL))
 240                 return NULL;
 241
 242         mode = afinfo->mode_map[encap];
 243         if (unlikely(mode && !try_module_get(mode->owner)))
 244                 mode = NULL;
 245         if (!mode && !modload_attempted) {
 246                 xfrm_policy_put_afinfo(afinfo);
 247                 request_module("xfrm-mode-%d-%d", family, encap);
 248                 modload_attempted = 1;
 249                 goto retry;
 250         }
 251
 252         xfrm_policy_put_afinfo(afinfo);
 253         return mode;
 254 }
 255
 256 void xfrm_put_mode(struct xfrm_mode *mode)
 257 {
 258         module_put(mode->owner);
 259 }
 260
 261 static inline unsigned long make_jiffies(long secs)
 262 {
 263         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 264                 return MAX_SCHEDULE_TIMEOUT-1;
 265         else
 266                 return secs*HZ;
 267 }
 268
 269 static void xfrm_policy_timer(unsigned long data)
 270 {
 271         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 272         unsigned long now = get_seconds();
 273         long next = LONG_MAX;
 274         int warn = 0;
 275         int dir;
 276
 277         read_lock(&xp->lock);
 278
 279         if (xp->dead)
 280                 goto out;
 281
 282         dir = xfrm_policy_id2dir(xp->index);
 283
 284         if (xp->lft.hard_add_expires_seconds) {
 285                 long tmo = xp->lft.hard_add_expires_seconds +
 286                         xp->curlft.add_time - now;
 287                 if (tmo <= 0)
 288                         goto expired;
 289                 if (tmo < next)
 290                         next = tmo;
 291         }
 292         if (xp->lft.hard_use_expires_seconds) {
 293                 long tmo = xp->lft.hard_use_expires_seconds +
 294                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 295                 if (tmo <= 0)
 296                         goto expired;
 297                 if (tmo < next)
 298                         next = tmo;
 299         }
 300         if (xp->lft.soft_add_expires_seconds) {
 301                 long tmo = xp->lft.soft_add_expires_seconds +
 302                         xp->curlft.add_time - now;
 303                 if (tmo <= 0) {
 304                         warn = 1;
 305                         tmo = XFRM_KM_TIMEOUT;
 306                 }
 307                 if (tmo < next)
 308                         next = tmo;
 309         }
 310         if (xp->lft.soft_use_expires_seconds) {
 311                 long tmo = xp->lft.soft_use_expires_seconds +
 312                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 313                 if (tmo <= 0) {
 314                         warn = 1;
 315                         tmo = XFRM_KM_TIMEOUT;
 316                 }
 317                 if (tmo < next)
 318                         next = tmo;
 319         }
 320
 321         if (warn)
 322                 km_policy_expired(xp, dir, 0, 0);
 323         if (next != LONG_MAX &&
 324             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 325                 xfrm_pol_hold(xp);
 326
 327 out:
 328         read_unlock(&xp->lock);
 329         xfrm_pol_put(xp);
 330         return;
 331
 332 expired:
 333         read_unlock(&xp->lock);
 334         if (!xfrm_policy_delete(xp, dir))
 335                 km_policy_expired(xp, dir, 1, 0);
 336         xfrm_pol_put(xp);
 337 }
 338
 339
 340 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 341  * SPD calls.
 342  */
 343
 344 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 345 {
 346         struct xfrm_policy *policy;
 347
 348         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 349
 350         if (policy) {
 351                 INIT_HLIST_NODE(&policy->bydst);
 352                 INIT_HLIST_NODE(&policy->byidx);
 353                 rwlock_init(&policy->lock);
 354                 atomic_set(&policy->refcnt, 1);
 355                 init_timer(&policy->timer);
 356                 policy->timer.data = (unsigned long)policy;
 357                 policy->timer.function = xfrm_policy_timer;
 358         }
 359         return policy;
 360 }
 361 EXPORT_SYMBOL(xfrm_policy_alloc);
 362
 363 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 364
 365 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 366 {
 367         BUG_ON(!policy->dead);
 368
 369         BUG_ON(policy->bundles);
 370
 371         if (del_timer(&policy->timer))
 372                 BUG();
 373
 374         security_xfrm_policy_free(policy);
 375         kfree(policy);
 376 }
 377 EXPORT_SYMBOL(__xfrm_policy_destroy);
 378
 379 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 380 {
 381         struct dst_entry *dst;
 382
 383         while ((dst = policy->bundles) != NULL) {
 384                 policy->bundles = dst->next;
 385                 dst_free(dst);
 386         }
 387
 388         if (del_timer(&policy->timer))
 389                 atomic_dec(&policy->refcnt);
 390
 391         if (atomic_read(&policy->refcnt) > 1)
 392                 flow_cache_flush();
 393
 394         xfrm_pol_put(policy);
 395 }
 396
 397 static void xfrm_policy_gc_task(struct work_struct *work)
 398 {
 399         struct xfrm_policy *policy;
 400         struct hlist_node *entry, *tmp;
 401         struct hlist_head gc_list;
 402
 403         spin_lock_bh(&xfrm_policy_gc_lock);
 404         gc_list.first = xfrm_policy_gc_list.first;
 405         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 406         spin_unlock_bh(&xfrm_policy_gc_lock);
 407
 408         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 409                 xfrm_policy_gc_kill(policy);
 410 }
 411
 412 /* Rule must be locked. Release descentant resources, announce
 413  * entry dead. The rule must be unlinked from lists to the moment.
 414  */
 415
 416 static void xfrm_policy_kill(struct xfrm_policy *policy)
 417 {
 418         int dead;
 419
 420         write_lock_bh(&policy->lock);
 421         dead = policy->dead;
 422         policy->dead = 1;
 423         write_unlock_bh(&policy->lock);
 424
 425         if (unlikely(dead)) {
 426                 WARN_ON(1);
 427                 return;
 428         }
 429
 430         spin_lock(&xfrm_policy_gc_lock);
 431         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 432         spin_unlock(&xfrm_policy_gc_lock);
 433
 434         schedule_work(&xfrm_policy_gc_work);
 435 }
 436
 437 struct xfrm_policy_hash {
 438         struct hlist_head       *table;
 439         unsigned int            hmask;
 440 };
 441
 442 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 443 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 444 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 445 static unsigned int xfrm_idx_hmask __read_mostly;
 446 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 447
 448 static inline unsigned int idx_hash(u32 index)
 449 {
 450         return __idx_hash(index, xfrm_idx_hmask);
 451 }
 452
 453 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 454 {
 455         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 456         unsigned int hash = __sel_hash(sel, family, hmask);
 457
 458         return (hash == hmask + 1 ?
 459                 &xfrm_policy_inexact[dir] :
 460                 xfrm_policy_bydst[dir].table + hash);
 461 }
 462
 463 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 464 {
 465         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 466         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 467
 468         return xfrm_policy_bydst[dir].table + hash;
 469 }
 470
 471 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 472                                    struct hlist_head *ndsttable,
 473                                    unsigned int nhashmask)
 474 {
 475         struct hlist_node *entry, *tmp;
 476         struct xfrm_policy *pol;
 477
 478         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 479                 unsigned int h;
 480
 481                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 482                                 pol->family, nhashmask);
 483                 hlist_add_head(&pol->bydst, ndsttable+h);
 484         }
 485 }
 486
 487 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 488                                    struct hlist_head *nidxtable,
 489                                    unsigned int nhashmask)
 490 {
 491         struct hlist_node *entry, *tmp;
 492         struct xfrm_policy *pol;
 493
 494         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 495                 unsigned int h;
 496
 497                 h = __idx_hash(pol->index, nhashmask);
 498                 hlist_add_head(&pol->byidx, nidxtable+h);
 499         }
 500 }
 501
 502 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 503 {
 504         return ((old_hmask + 1) << 1) - 1;
 505 }
 506
 507 static void xfrm_bydst_resize(int dir)
 508 {
 509         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 510         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 511         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 512         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 513         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 514         int i;
 515
 516         if (!ndst)
 517                 return;
 518
 519         write_lock_bh(&xfrm_policy_lock);
 520
 521         for (i = hmask; i >= 0; i--)
 522                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 523
 524         xfrm_policy_bydst[dir].table = ndst;
 525         xfrm_policy_bydst[dir].hmask = nhashmask;
 526
 527         write_unlock_bh(&xfrm_policy_lock);
 528
 529         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 530 }
 531
 532 static void xfrm_byidx_resize(int total)
 533 {
 534         unsigned int hmask = xfrm_idx_hmask;
 535         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 536         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 537         struct hlist_head *oidx = xfrm_policy_byidx;
 538         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 539         int i;
 540
 541         if (!nidx)
 542                 return;
 543
 544         write_lock_bh(&xfrm_policy_lock);
 545
 546         for (i = hmask; i >= 0; i--)
 547                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 548
 549         xfrm_policy_byidx = nidx;
 550         xfrm_idx_hmask = nhashmask;
 551
 552         write_unlock_bh(&xfrm_policy_lock);
 553
 554         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 555 }
 556
 557 static inline int xfrm_bydst_should_resize(int dir, int *total)
 558 {
 559         unsigned int cnt = xfrm_policy_count[dir];
 560         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 561
 562         if (total)
 563                 *total += cnt;
 564
 565         if ((hmask + 1) < xfrm_policy_hashmax &&
 566             cnt > hmask)
 567                 return 1;
 568
 569         return 0;
 570 }
 571
 572 static inline int xfrm_byidx_should_resize(int total)
 573 {
 574         unsigned int hmask = xfrm_idx_hmask;
 575
 576         if ((hmask + 1) < xfrm_policy_hashmax &&
 577             total > hmask)
 578                 return 1;
 579
 580         return 0;
 581 }
 582
 583 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 584 {
 585         read_lock_bh(&xfrm_policy_lock);
 586         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 587         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 588         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 589         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 590         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 591         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 592         si->spdhcnt = xfrm_idx_hmask;
 593         si->spdhmcnt = xfrm_policy_hashmax;
 594         read_unlock_bh(&xfrm_policy_lock);
 595 }
 596 EXPORT_SYMBOL(xfrm_spd_getinfo);
 597
 598 static DEFINE_MUTEX(hash_resize_mutex);
 599 static void xfrm_hash_resize(struct work_struct *__unused)
 600 {
 601         int dir, total;
 602
 603         mutex_lock(&hash_resize_mutex);
 604
 605         total = 0;
 606         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 607                 if (xfrm_bydst_should_resize(dir, &total))
 608                         xfrm_bydst_resize(dir);
 609         }
 610         if (xfrm_byidx_should_resize(total))
 611                 xfrm_byidx_resize(total);
 612
 613         mutex_unlock(&hash_resize_mutex);
 614 }
 615
 616 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 617
 618 /* Generate new index... KAME seems to generate them ordered by cost
 619  * of an absolute inpredictability of ordering of rules. This will not pass. */
 620 static u32 xfrm_gen_index(u8 type, int dir)
 621 {
 622         static u32 idx_generator;
 623
 624         for (;;) {
 625                 struct hlist_node *entry;
 626                 struct hlist_head *list;
 627                 struct xfrm_policy *p;
 628                 u32 idx;
 629                 int found;
 630
 631                 idx = (idx_generator | dir);
 632                 idx_generator += 8;
 633                 if (idx == 0)
 634                         idx = 8;
 635                 list = xfrm_policy_byidx + idx_hash(idx);
 636                 found = 0;
 637                 hlist_for_each_entry(p, entry, list, byidx) {
 638                         if (p->index == idx) {
 639                                 found = 1;
 640                                 break;
 641                         }
 642                 }
 643                 if (!found)
 644                         return idx;
 645         }
 646 }
 647
 648 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 649 {
 650         u32 *p1 = (u32 *) s1;
 651         u32 *p2 = (u32 *) s2;
 652         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 653         int i;
 654
 655         for (i = 0; i < len; i++) {
 656                 if (p1[i] != p2[i])
 657                         return 1;
 658         }
 659
 660         return 0;
 661 }
 662
 663 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 664 {
 665         struct xfrm_policy *pol;
 666         struct xfrm_policy *delpol;
 667         struct hlist_head *chain;
 668         struct hlist_node *entry, *newpos;
 669         struct dst_entry *gc_list;
 670
 671         write_lock_bh(&xfrm_policy_lock);
 672         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 673         delpol = NULL;
 674         newpos = NULL;
 675         hlist_for_each_entry(pol, entry, chain, bydst) {
 676                 if (pol->type == policy->type &&
 677                     !selector_cmp(&pol->selector, &policy->selector) &&
 678                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 679                     !WARN_ON(delpol)) {
 680                         if (excl) {
 681                                 write_unlock_bh(&xfrm_policy_lock);
 682                                 return -EEXIST;
 683                         }
 684                         delpol = pol;
 685                         if (policy->priority > pol->priority)
 686                                 continue;
 687                 } else if (policy->priority >= pol->priority) {
 688                         newpos = &pol->bydst;
 689                         continue;
 690                 }
 691                 if (delpol)
 692                         break;
 693         }
 694         if (newpos)
 695                 hlist_add_after(newpos, &policy->bydst);
 696         else
 697                 hlist_add_head(&policy->bydst, chain);
 698         xfrm_pol_hold(policy);
 699         xfrm_policy_count[dir]++;
 700         atomic_inc(&flow_cache_genid);
 701         if (delpol) {
 702                 hlist_del(&delpol->bydst);
 703                 hlist_del(&delpol->byidx);
 704                 xfrm_policy_count[dir]--;
 705         }
 706         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 707         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 708         policy->curlft.add_time = get_seconds();
 709         policy->curlft.use_time = 0;
 710         if (!mod_timer(&policy->timer, jiffies + HZ))
 711                 xfrm_pol_hold(policy);
 712         write_unlock_bh(&xfrm_policy_lock);
 713
 714         if (delpol)
 715                 xfrm_policy_kill(delpol);
 716         else if (xfrm_bydst_should_resize(dir, NULL))
 717                 schedule_work(&xfrm_hash_work);
 718
 719         read_lock_bh(&xfrm_policy_lock);
 720         gc_list = NULL;
 721         entry = &policy->bydst;
 722         hlist_for_each_entry_continue(policy, entry, bydst) {
 723                 struct dst_entry *dst;
 724
 725                 write_lock(&policy->lock);
 726                 dst = policy->bundles;
 727                 if (dst) {
 728                         struct dst_entry *tail = dst;
 729                         while (tail->next)
 730                                 tail = tail->next;
 731                         tail->next = gc_list;
 732                         gc_list = dst;
 733
 734                         policy->bundles = NULL;
 735                 }
 736                 write_unlock(&policy->lock);
 737         }
 738         read_unlock_bh(&xfrm_policy_lock);
 739
 740         while (gc_list) {
 741                 struct dst_entry *dst = gc_list;
 742
 743                 gc_list = dst->next;
 744                 dst_free(dst);
 745         }
 746
 747         return 0;
 748 }
 749 EXPORT_SYMBOL(xfrm_policy_insert);
 750
 751 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 752                                           struct xfrm_selector *sel,
 753                                           struct xfrm_sec_ctx *ctx, int delete,
 754                                           int *err)
 755 {
 756         struct xfrm_policy *pol, *ret;
 757         struct hlist_head *chain;
 758         struct hlist_node *entry;
 759
 760         *err = 0;
 761         write_lock_bh(&xfrm_policy_lock);
 762         chain = policy_hash_bysel(sel, sel->family, dir);
 763         ret = NULL;
 764         hlist_for_each_entry(pol, entry, chain, bydst) {
 765                 if (pol->type == type &&
 766                     !selector_cmp(sel, &pol->selector) &&
 767                     xfrm_sec_ctx_match(ctx, pol->security)) {
 768                         xfrm_pol_hold(pol);
 769                         if (delete) {
 770                                 *err = security_xfrm_policy_delete(pol);
 771                                 if (*err) {
 772                                         write_unlock_bh(&xfrm_policy_lock);
 773                                         return pol;
 774                                 }
 775                                 hlist_del(&pol->bydst);
 776                                 hlist_del(&pol->byidx);
 777                                 xfrm_policy_count[dir]--;
 778                         }
 779                         ret = pol;
 780                         break;
 781                 }
 782         }
 783         write_unlock_bh(&xfrm_policy_lock);
 784
 785         if (ret && delete) {
 786                 atomic_inc(&flow_cache_genid);
 787                 xfrm_policy_kill(ret);
 788         }
 789         return ret;
 790 }
 791 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 792
 793 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 794                                      int *err)
 795 {
 796         struct xfrm_policy *pol, *ret;
 797         struct hlist_head *chain;
 798         struct hlist_node *entry;
 799
 800         *err = -ENOENT;
 801         if (xfrm_policy_id2dir(id) != dir)
 802                 return NULL;
 803
 804         *err = 0;
 805         write_lock_bh(&xfrm_policy_lock);
 806         chain = xfrm_policy_byidx + idx_hash(id);
 807         ret = NULL;
 808         hlist_for_each_entry(pol, entry, chain, byidx) {
 809                 if (pol->type == type && pol->index == id) {
 810                         xfrm_pol_hold(pol);
 811                         if (delete) {
 812                                 *err = security_xfrm_policy_delete(pol);
 813                                 if (*err) {
 814                                         write_unlock_bh(&xfrm_policy_lock);
 815                                         return pol;
 816                                 }
 817                                 hlist_del(&pol->bydst);
 818                                 hlist_del(&pol->byidx);
 819                                 xfrm_policy_count[dir]--;
 820                         }
 821                         ret = pol;
 822                         break;
 823                 }
 824         }
 825         write_unlock_bh(&xfrm_policy_lock);
 826
 827         if (ret && delete) {
 828                 atomic_inc(&flow_cache_genid);
 829                 xfrm_policy_kill(ret);
 830         }
 831         return ret;
 832 }
 833 EXPORT_SYMBOL(xfrm_policy_byid);
 834
 835 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 836 static inline int
 837 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 838 {
 839         int dir, err = 0;
 840
 841         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 842                 struct xfrm_policy *pol;
 843                 struct hlist_node *entry;
 844                 int i;
 845
 846                 hlist_for_each_entry(pol, entry,
 847                                      &xfrm_policy_inexact[dir], bydst) {
 848                         if (pol->type != type)
 849                                 continue;
 850                         err = security_xfrm_policy_delete(pol);
 851                         if (err) {
 852                                 xfrm_audit_policy_delete(pol, 0,
 853                                                          audit_info->loginuid,
 854                                                          audit_info->secid);
 855                                 return err;
 856                         }
 857                 }
 858                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 859                         hlist_for_each_entry(pol, entry,
 860                                              xfrm_policy_bydst[dir].table + i,
 861                                              bydst) {
 862                                 if (pol->type != type)
 863                                         continue;
 864                                 err = security_xfrm_policy_delete(pol);
 865                                 if (err) {
 866                                         xfrm_audit_policy_delete(pol, 0,
 867                                                         audit_info->loginuid,
 868                                                         audit_info->secid);
 869                                         return err;
 870                                 }
 871                         }
 872                 }
 873         }
 874         return err;
 875 }
 876 #else
 877 static inline int
 878 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 879 {
 880         return 0;
 881 }
 882 #endif
 883
 884 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 885 {
 886         int dir, err = 0;
 887
 888         write_lock_bh(&xfrm_policy_lock);
 889
 890         err = xfrm_policy_flush_secctx_check(type, audit_info);
 891         if (err)
 892                 goto out;
 893
 894         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 895                 struct xfrm_policy *pol;
 896                 struct hlist_node *entry;
 897                 int i, killed;
 898
 899                 killed = 0;
 900         again1:
 901                 hlist_for_each_entry(pol, entry,
 902                                      &xfrm_policy_inexact[dir], bydst) {
 903                         if (pol->type != type)
 904                                 continue;
 905                         hlist_del(&pol->bydst);
 906                         hlist_del(&pol->byidx);
 907                         write_unlock_bh(&xfrm_policy_lock);
 908
 909                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 910                                                  audit_info->secid);
 911
 912                         xfrm_policy_kill(pol);
 913                         killed++;
 914
 915                         write_lock_bh(&xfrm_policy_lock);
 916                         goto again1;
 917                 }
 918
 919                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 920         again2:
 921                         hlist_for_each_entry(pol, entry,
 922                                              xfrm_policy_bydst[dir].table + i,
 923                                              bydst) {
 924                                 if (pol->type != type)
 925                                         continue;
 926                                 hlist_del(&pol->bydst);
 927                                 hlist_del(&pol->byidx);
 928                                 write_unlock_bh(&xfrm_policy_lock);
 929
 930                                 xfrm_audit_policy_delete(pol, 1,
 931                                                          audit_info->loginuid,
 932                                                          audit_info->secid);
 933                                 xfrm_policy_kill(pol);
 934                                 killed++;
 935
 936                                 write_lock_bh(&xfrm_policy_lock);
 937                                 goto again2;
 938                         }
 939                 }
 940
 941                 xfrm_policy_count[dir] -= killed;
 942         }
 943         atomic_inc(&flow_cache_genid);
 944 out:
 945         write_unlock_bh(&xfrm_policy_lock);
 946         return err;
 947 }
 948 EXPORT_SYMBOL(xfrm_policy_flush);
 949
 950 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 951                      void *data)
 952 {
 953         struct xfrm_policy *pol, *last = NULL;
 954         struct hlist_node *entry;
 955         int dir, last_dir = 0, count, error;
 956
 957         read_lock_bh(&xfrm_policy_lock);
 958         count = 0;
 959
 960         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 961                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 962                 int i;
 963
 964                 hlist_for_each_entry(pol, entry,
 965                                      &xfrm_policy_inexact[dir], bydst) {
 966                         if (pol->type != type)
 967                                 continue;
 968                         if (last) {
 969                                 error = func(last, last_dir % XFRM_POLICY_MAX,
 970                                              count, data);
 971                                 if (error)
 972                                         goto out;
 973                         }
 974                         last = pol;
 975                         last_dir = dir;
 976                         count++;
 977                 }
 978                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 979                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 980                                 if (pol->type != type)
 981                                         continue;
 982                                 if (last) {
 983                                         error = func(last, last_dir % XFRM_POLICY_MAX,
 984                                                      count, data);
 985                                         if (error)
 986                                                 goto out;
 987                                 }
 988                                 last = pol;
 989                                 last_dir = dir;
 990                                 count++;
 991                         }
 992                 }
 993         }
 994         if (count == 0) {
 995                 error = -ENOENT;
 996                 goto out;
 997         }
 998         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
 999 out:
1000         read_unlock_bh(&xfrm_policy_lock);
1001         return error;
1002 }
1003 EXPORT_SYMBOL(xfrm_policy_walk);
1004
1005 /*
1006  * Find policy to apply to this flow.
1007  *
1008  * Returns 0 if policy found, else an -errno.
1009  */
1010 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
1011                              u8 type, u16 family, int dir)
1012 {
1013         struct xfrm_selector *sel = &pol->selector;
1014         int match, ret = -ESRCH;
1015
1016         if (pol->family != family ||
1017             pol->type != type)
1018                 return ret;
1019
1020         match = xfrm_selector_match(sel, fl, family);
1021         if (match)
1022                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
1023
1024         return ret;
1025 }
1026
1027 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
1028                                                      u16 family, u8 dir)
1029 {
1030         int err;
1031         struct xfrm_policy *pol, *ret;
1032         xfrm_address_t *daddr, *saddr;
1033         struct hlist_node *entry;
1034         struct hlist_head *chain;
1035         u32 priority = ~0U;
1036
1037         daddr = xfrm_flowi_daddr(fl, family);
1038         saddr = xfrm_flowi_saddr(fl, family);
1039         if (unlikely(!daddr || !saddr))
1040                 return NULL;
1041
1042         read_lock_bh(&xfrm_policy_lock);
1043         chain = policy_hash_direct(daddr, saddr, family, dir);
1044         ret = NULL;
1045         hlist_for_each_entry(pol, entry, chain, bydst) {
1046                 err = xfrm_policy_match(pol, fl, type, family, dir);
1047                 if (err) {
1048                         if (err == -ESRCH)
1049                                 continue;
1050                         else {
1051                                 ret = ERR_PTR(err);
1052                                 goto fail;
1053                         }
1054                 } else {
1055                         ret = pol;
1056                         priority = ret->priority;
1057                         break;
1058                 }
1059         }
1060         chain = &xfrm_policy_inexact[dir];
1061         hlist_for_each_entry(pol, entry, chain, bydst) {
1062                 err = xfrm_policy_match(pol, fl, type, family, dir);
1063                 if (err) {
1064                         if (err == -ESRCH)
1065                                 continue;
1066                         else {
1067                                 ret = ERR_PTR(err);
1068                                 goto fail;
1069                         }
1070                 } else if (pol->priority < priority) {
1071                         ret = pol;
1072                         break;
1073                 }
1074         }
1075         if (ret)
1076                 xfrm_pol_hold(ret);
1077 fail:
1078         read_unlock_bh(&xfrm_policy_lock);
1079
1080         return ret;
1081 }
1082
1083 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1084                                void **objp, atomic_t **obj_refp)
1085 {
1086         struct xfrm_policy *pol;
1087         int err = 0;
1088
1089 #ifdef CONFIG_XFRM_SUB_POLICY
1090         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1091         if (IS_ERR(pol)) {
1092                 err = PTR_ERR(pol);
1093                 pol = NULL;
1094         }
1095         if (pol || err)
1096                 goto end;
1097 #endif
1098         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1099         if (IS_ERR(pol)) {
1100                 err = PTR_ERR(pol);
1101                 pol = NULL;
1102         }
1103 #ifdef CONFIG_XFRM_SUB_POLICY
1104 end:
1105 #endif
1106         if ((*objp = (void *) pol) != NULL)
1107                 *obj_refp = &pol->refcnt;
1108         return err;
1109 }
1110
1111 static inline int policy_to_flow_dir(int dir)
1112 {
1113         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1114             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1115             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1116                 return dir;
1117         switch (dir) {
1118         default:
1119         case XFRM_POLICY_IN:
1120                 return FLOW_DIR_IN;
1121         case XFRM_POLICY_OUT:
1122                 return FLOW_DIR_OUT;
1123         case XFRM_POLICY_FWD:
1124                 return FLOW_DIR_FWD;
1125         }
1126 }
1127
1128 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1129 {
1130         struct xfrm_policy *pol;
1131
1132         read_lock_bh(&xfrm_policy_lock);
1133         if ((pol = sk->sk_policy[dir]) != NULL) {
1134                 int match = xfrm_selector_match(&pol->selector, fl,
1135                                                 sk->sk_family);
1136                 int err = 0;
1137
1138                 if (match) {
1139                         err = security_xfrm_policy_lookup(pol, fl->secid,
1140                                         policy_to_flow_dir(dir));
1141                         if (!err)
1142                                 xfrm_pol_hold(pol);
1143                         else if (err == -ESRCH)
1144                                 pol = NULL;
1145                         else
1146                                 pol = ERR_PTR(err);
1147                 } else
1148                         pol = NULL;
1149         }
1150         read_unlock_bh(&xfrm_policy_lock);
1151         return pol;
1152 }
1153
1154 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1155 {
1156         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1157                                                      pol->family, dir);
1158
1159         hlist_add_head(&pol->bydst, chain);
1160         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1161         xfrm_policy_count[dir]++;
1162         xfrm_pol_hold(pol);
1163
1164         if (xfrm_bydst_should_resize(dir, NULL))
1165                 schedule_work(&xfrm_hash_work);
1166 }
1167
1168 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1169                                                 int dir)
1170 {
1171         if (hlist_unhashed(&pol->bydst))
1172                 return NULL;
1173
1174         hlist_del(&pol->bydst);
1175         hlist_del(&pol->byidx);
1176         xfrm_policy_count[dir]--;
1177
1178         return pol;
1179 }
1180
1181 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1182 {
1183         write_lock_bh(&xfrm_policy_lock);
1184         pol = __xfrm_policy_unlink(pol, dir);
1185         write_unlock_bh(&xfrm_policy_lock);
1186         if (pol) {
1187                 if (dir < XFRM_POLICY_MAX)
1188                         atomic_inc(&flow_cache_genid);
1189                 xfrm_policy_kill(pol);
1190                 return 0;
1191         }
1192         return -ENOENT;
1193 }
1194 EXPORT_SYMBOL(xfrm_policy_delete);
1195
1196 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1197 {
1198         struct xfrm_policy *old_pol;
1199
1200 #ifdef CONFIG_XFRM_SUB_POLICY
1201         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1202                 return -EINVAL;
1203 #endif
1204
1205         write_lock_bh(&xfrm_policy_lock);
1206         old_pol = sk->sk_policy[dir];
1207         sk->sk_policy[dir] = pol;
1208         if (pol) {
1209                 pol->curlft.add_time = get_seconds();
1210                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1211                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1212         }
1213         if (old_pol)
1214                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1215         write_unlock_bh(&xfrm_policy_lock);
1216
1217         if (old_pol) {
1218                 xfrm_policy_kill(old_pol);
1219         }
1220         return 0;
1221 }
1222
1223 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1224 {
1225         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1226
1227         if (newp) {
1228                 newp->selector = old->selector;
1229                 if (security_xfrm_policy_clone(old, newp)) {
1230                         kfree(newp);
1231                         return NULL;  /* ENOMEM */
1232                 }
1233                 newp->lft = old->lft;
1234                 newp->curlft = old->curlft;
1235                 newp->action = old->action;
1236                 newp->flags = old->flags;
1237                 newp->xfrm_nr = old->xfrm_nr;
1238                 newp->index = old->index;
1239                 newp->type = old->type;
1240                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1241                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1242                 write_lock_bh(&xfrm_policy_lock);
1243                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1244                 write_unlock_bh(&xfrm_policy_lock);
1245                 xfrm_pol_put(newp);
1246         }
1247         return newp;
1248 }
1249
1250 int __xfrm_sk_clone_policy(struct sock *sk)
1251 {
1252         struct xfrm_policy *p0 = sk->sk_policy[0],
1253                            *p1 = sk->sk_policy[1];
1254
1255         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1256         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1257                 return -ENOMEM;
1258         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1259                 return -ENOMEM;
1260         return 0;
1261 }
1262
1263 static int
1264 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1265                unsigned short family)
1266 {
1267         int err;
1268         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1269
1270         if (unlikely(afinfo == NULL))
1271                 return -EINVAL;
1272         err = afinfo->get_saddr(local, remote);
1273         xfrm_policy_put_afinfo(afinfo);
1274         return err;
1275 }
1276
1277 /* Resolve list of templates for the flow, given policy. */
1278
1279 static int
1280 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1281                       struct xfrm_state **xfrm,
1282                       unsigned short family)
1283 {
1284         int nx;
1285         int i, error;
1286         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1287         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1288         xfrm_address_t tmp;
1289
1290         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1291                 struct xfrm_state *x;
1292                 xfrm_address_t *remote = daddr;
1293                 xfrm_address_t *local  = saddr;
1294                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1295
1296                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1297                     tmpl->mode == XFRM_MODE_BEET) {
1298                         remote = &tmpl->id.daddr;
1299                         local = &tmpl->saddr;
1300                         family = tmpl->encap_family;
1301                         if (xfrm_addr_any(local, family)) {
1302                                 error = xfrm_get_saddr(&tmp, remote, family);
1303                                 if (error)
1304                                         goto fail;
1305                                 local = &tmp;
1306                         }
1307                 }
1308
1309                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1310
1311                 if (x && x->km.state == XFRM_STATE_VALID) {
1312                         xfrm[nx++] = x;
1313                         daddr = remote;
1314                         saddr = local;
1315                         continue;
1316                 }
1317                 if (x) {
1318                         error = (x->km.state == XFRM_STATE_ERROR ?
1319                                  -EINVAL : -EAGAIN);
1320                         xfrm_state_put(x);
1321                 }
1322
1323                 if (!tmpl->optional)
1324                         goto fail;
1325         }
1326         return nx;
1327
1328 fail:
1329         for (nx--; nx>=0; nx--)
1330                 xfrm_state_put(xfrm[nx]);
1331         return error;
1332 }
1333
1334 static int
1335 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1336                   struct xfrm_state **xfrm,
1337                   unsigned short family)
1338 {
1339         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1340         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1341         int cnx = 0;
1342         int error;
1343         int ret;
1344         int i;
1345
1346         for (i = 0; i < npols; i++) {
1347                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1348                         error = -ENOBUFS;
1349                         goto fail;
1350                 }
1351
1352                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1353                 if (ret < 0) {
1354                         error = ret;
1355                         goto fail;
1356                 } else
1357                         cnx += ret;
1358         }
1359
1360         /* found states are sorted for outbound processing */
1361         if (npols > 1)
1362                 xfrm_state_sort(xfrm, tpp, cnx, family);
1363
1364         return cnx;
1365
1366  fail:
1367         for (cnx--; cnx>=0; cnx--)
1368                 xfrm_state_put(tpp[cnx]);
1369         return error;
1370
1371 }
1372
1373 /* Check that the bundle accepts the flow and its components are
1374  * still valid.
1375  */
1376
1377 static struct dst_entry *
1378 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1379 {
1380         struct dst_entry *x;
1381         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1382         if (unlikely(afinfo == NULL))
1383                 return ERR_PTR(-EINVAL);
1384         x = afinfo->find_bundle(fl, policy);
1385         xfrm_policy_put_afinfo(afinfo);
1386         return x;
1387 }
1388
1389 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1390  * all the metrics... Shortly, bundle a bundle.
1391  */
1392
1393 static int
1394 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1395                    struct flowi *fl, struct dst_entry **dst_p,
1396                    unsigned short family)
1397 {
1398         int err;
1399         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1400         if (unlikely(afinfo == NULL))
1401                 return -EINVAL;
1402         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1403         xfrm_policy_put_afinfo(afinfo);
1404         return err;
1405 }
1406
1407 static int inline
1408 xfrm_dst_alloc_copy(void **target, void *src, int size)
1409 {
1410         if (!*target) {
1411                 *target = kmalloc(size, GFP_ATOMIC);
1412                 if (!*target)
1413                         return -ENOMEM;
1414         }
1415         memcpy(*target, src, size);
1416         return 0;
1417 }
1418
1419 static int inline
1420 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1421 {
1422 #ifdef CONFIG_XFRM_SUB_POLICY
1423         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1424         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1425                                    sel, sizeof(*sel));
1426 #else
1427         return 0;
1428 #endif
1429 }
1430
1431 static int inline
1432 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1433 {
1434 #ifdef CONFIG_XFRM_SUB_POLICY
1435         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1436         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1437 #else
1438         return 0;
1439 #endif
1440 }
1441
1442 static int stale_bundle(struct dst_entry *dst);
1443
1444 /* Main function: finds/creates a bundle for given flow.
1445  *
1446  * At the moment we eat a raw IP route. Mostly to speed up lookups
1447  * on interfaces with disabled IPsec.
1448  */
1449 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1450                   struct sock *sk, int flags)
1451 {
1452         struct xfrm_policy *policy;
1453         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1454         int npols;
1455         int pol_dead;
1456         int xfrm_nr;
1457         int pi;
1458         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1459         struct dst_entry *dst, *dst_orig = *dst_p;
1460         int nx = 0;
1461         int err;
1462         u32 genid;
1463         u16 family;
1464         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1465
1466 restart:
1467         genid = atomic_read(&flow_cache_genid);
1468         policy = NULL;
1469         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1470                 pols[pi] = NULL;
1471         npols = 0;
1472         pol_dead = 0;
1473         xfrm_nr = 0;
1474
1475         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1476                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1477                 if (IS_ERR(policy))
1478                         return PTR_ERR(policy);
1479         }
1480
1481         if (!policy) {
1482                 /* To accelerate a bit...  */
1483                 if ((dst_orig->flags & DST_NOXFRM) ||
1484                     !xfrm_policy_count[XFRM_POLICY_OUT])
1485                         return 0;
1486
1487                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1488                                            dir, xfrm_policy_lookup);
1489                 if (IS_ERR(policy))
1490                         return PTR_ERR(policy);
1491         }
1492
1493         if (!policy)
1494                 return 0;
1495
1496         family = dst_orig->ops->family;
1497         policy->curlft.use_time = get_seconds();
1498         pols[0] = policy;
1499         npols ++;
1500         xfrm_nr += pols[0]->xfrm_nr;
1501
1502         switch (policy->action) {
1503         case XFRM_POLICY_BLOCK:
1504                 /* Prohibit the flow */
1505                 err = -EPERM;
1506                 goto error;
1507
1508         case XFRM_POLICY_ALLOW:
1509 #ifndef CONFIG_XFRM_SUB_POLICY
1510                 if (policy->xfrm_nr == 0) {
1511                         /* Flow passes not transformed. */
1512                         xfrm_pol_put(policy);
1513                         return 0;
1514                 }
1515 #endif
1516
1517                 /* Try to find matching bundle.
1518                  *
1519                  * LATER: help from flow cache. It is optional, this
1520                  * is required only for output policy.
1521                  */
1522                 dst = xfrm_find_bundle(fl, policy, family);
1523                 if (IS_ERR(dst)) {
1524                         err = PTR_ERR(dst);
1525                         goto error;
1526                 }
1527
1528                 if (dst)
1529                         break;
1530
1531 #ifdef CONFIG_XFRM_SUB_POLICY
1532                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1533                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1534                                                             fl, family,
1535                                                             XFRM_POLICY_OUT);
1536                         if (pols[1]) {
1537                                 if (IS_ERR(pols[1])) {
1538                                         err = PTR_ERR(pols[1]);
1539                                         goto error;
1540                                 }
1541                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1542                                         err = -EPERM;
1543                                         goto error;
1544                                 }
1545                                 npols ++;
1546                                 xfrm_nr += pols[1]->xfrm_nr;
1547                         }
1548                 }
1549
1550                 /*
1551                  * Because neither flowi nor bundle information knows about
1552                  * transformation template size. On more than one policy usage
1553                  * we can realize whether all of them is bypass or not after
1554                  * they are searched. See above not-transformed bypass
1555                  * is surrounded by non-sub policy configuration, too.
1556                  */
1557                 if (xfrm_nr == 0) {
1558                         /* Flow passes not transformed. */
1559                         xfrm_pols_put(pols, npols);
1560                         return 0;
1561                 }
1562
1563 #endif
1564                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1565
1566                 if (unlikely(nx<0)) {
1567                         err = nx;
1568                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1569                                 /* EREMOTE tells the caller to generate
1570                                  * a one-shot blackhole route.
1571                                  */
1572                                 xfrm_pol_put(policy);
1573                                 return -EREMOTE;
1574                         }
1575                         if (err == -EAGAIN && flags) {
1576                                 DECLARE_WAITQUEUE(wait, current);
1577
1578                                 add_wait_queue(&km_waitq, &wait);
1579                                 set_current_state(TASK_INTERRUPTIBLE);
1580                                 schedule();
1581                                 set_current_state(TASK_RUNNING);
1582                                 remove_wait_queue(&km_waitq, &wait);
1583
1584                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1585
1586                                 if (nx == -EAGAIN && signal_pending(current)) {
1587                                         err = -ERESTART;
1588                                         goto error;
1589                                 }
1590                                 if (nx == -EAGAIN ||
1591                                     genid != atomic_read(&flow_cache_genid)) {
1592                                         xfrm_pols_put(pols, npols);
1593                                         goto restart;
1594                                 }
1595                                 err = nx;
1596                         }
1597                         if (err < 0)
1598                                 goto error;
1599                 }
1600                 if (nx == 0) {
1601                         /* Flow passes not transformed. */
1602                         xfrm_pols_put(pols, npols);
1603                         return 0;
1604                 }
1605
1606                 dst = dst_orig;
1607                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1608
1609                 if (unlikely(err)) {
1610                         int i;
1611                         for (i=0; i<nx; i++)
1612                                 xfrm_state_put(xfrm[i]);
1613                         goto error;
1614                 }
1615
1616                 for (pi = 0; pi < npols; pi++) {
1617                         read_lock_bh(&pols[pi]->lock);
1618                         pol_dead |= pols[pi]->dead;
1619                         read_unlock_bh(&pols[pi]->lock);
1620                 }
1621
1622                 write_lock_bh(&policy->lock);
1623                 if (unlikely(pol_dead || stale_bundle(dst))) {
1624                         /* Wow! While we worked on resolving, this
1625                          * policy has gone. Retry. It is not paranoia,
1626                          * we just cannot enlist new bundle to dead object.
1627                          * We can't enlist stable bundles either.
1628                          */
1629                         write_unlock_bh(&policy->lock);
1630                         if (dst)
1631                                 dst_free(dst);
1632
1633                         err = -EHOSTUNREACH;
1634                         goto error;
1635                 }
1636
1637                 if (npols > 1)
1638                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1639                 else
1640                         err = xfrm_dst_update_origin(dst, fl);
1641                 if (unlikely(err)) {
1642                         write_unlock_bh(&policy->lock);
1643                         if (dst)
1644                                 dst_free(dst);
1645                         goto error;
1646                 }
1647
1648                 dst->next = policy->bundles;
1649                 policy->bundles = dst;
1650                 dst_hold(dst);
1651                 write_unlock_bh(&policy->lock);
1652         }
1653         *dst_p = dst;
1654         dst_release(dst_orig);
1655         xfrm_pols_put(pols, npols);
1656         return 0;
1657
1658 error:
1659         dst_release(dst_orig);
1660         xfrm_pols_put(pols, npols);
1661         *dst_p = NULL;
1662         return err;
1663 }
1664 EXPORT_SYMBOL(__xfrm_lookup);
1665
1666 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1667                 struct sock *sk, int flags)
1668 {
1669         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1670
1671         if (err == -EREMOTE) {
1672                 dst_release(*dst_p);
1673                 *dst_p = NULL;
1674                 err = -EAGAIN;
1675         }
1676
1677         return err;
1678 }
1679 EXPORT_SYMBOL(xfrm_lookup);
1680
1681 static inline int
1682 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1683 {
1684         struct xfrm_state *x;
1685
1686         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1687                 return 0;
1688         x = skb->sp->xvec[idx];
1689         if (!x->type->reject)
1690                 return 0;
1691         return x->type->reject(x, skb, fl);
1692 }
1693
1694 /* When skb is transformed back to its "native" form, we have to
1695  * check policy restrictions. At the moment we make this in maximally
1696  * stupid way. Shame on me. :-) Of course, connected sockets must
1697  * have policy cached at them.
1698  */
1699
1700 static inline int
1701 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1702               unsigned short family)
1703 {
1704         if (xfrm_state_kern(x))
1705                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1706         return  x->id.proto == tmpl->id.proto &&
1707                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1708                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1709                 x->props.mode == tmpl->mode &&
1710                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1711                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1712                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1713                   xfrm_state_addr_cmp(tmpl, x, family));
1714 }
1715
1716 /*
1717  * 0 or more than 0 is returned when validation is succeeded (either bypass
1718  * because of optional transport mode, or next index of the mathced secpath
1719  * state with the template.
1720  * -1 is returned when no matching template is found.
1721  * Otherwise "-2 - errored_index" is returned.
1722  */
1723 static inline int
1724 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1725                unsigned short family)
1726 {
1727         int idx = start;
1728
1729         if (tmpl->optional) {
1730                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1731                         return start;
1732         } else
1733                 start = -1;
1734         for (; idx < sp->len; idx++) {
1735                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1736                         return ++idx;
1737                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1738                         if (start == -1)
1739                                 start = -2-idx;
1740                         break;
1741                 }
1742         }
1743         return start;
1744 }
1745
1746 int
1747 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1748 {
1749         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1750         int err;
1751
1752         if (unlikely(afinfo == NULL))
1753                 return -EAFNOSUPPORT;
1754
1755         afinfo->decode_session(skb, fl);
1756         err = security_xfrm_decode_session(skb, &fl->secid);
1757         xfrm_policy_put_afinfo(afinfo);
1758         return err;
1759 }
1760 EXPORT_SYMBOL(xfrm_decode_session);
1761
1762 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1763 {
1764         for (; k < sp->len; k++) {
1765                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1766                         *idxp = k;
1767                         return 1;
1768                 }
1769         }
1770
1771         return 0;
1772 }
1773
1774 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1775                         unsigned short family)
1776 {
1777         struct xfrm_policy *pol;
1778         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1779         int npols = 0;
1780         int xfrm_nr;
1781         int pi;
1782         struct flowi fl;
1783         u8 fl_dir = policy_to_flow_dir(dir);
1784         int xerr_idx = -1;
1785
1786         if (xfrm_decode_session(skb, &fl, family) < 0)
1787                 return 0;
1788         nf_nat_decode_session(skb, &fl, family);
1789
1790         /* First, check used SA against their selectors. */
1791         if (skb->sp) {
1792                 int i;
1793
1794                 for (i=skb->sp->len-1; i>=0; i--) {
1795                         struct xfrm_state *x = skb->sp->xvec[i];
1796                         if (!xfrm_selector_match(&x->sel, &fl, family))
1797                                 return 0;
1798                 }
1799         }
1800
1801         pol = NULL;
1802         if (sk && sk->sk_policy[dir]) {
1803                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1804                 if (IS_ERR(pol))
1805                         return 0;
1806         }
1807
1808         if (!pol)
1809                 pol = flow_cache_lookup(&fl, family, fl_dir,
1810                                         xfrm_policy_lookup);
1811
1812         if (IS_ERR(pol))
1813                 return 0;
1814
1815         if (!pol) {
1816                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1817                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1818                         return 0;
1819                 }
1820                 return 1;
1821         }
1822
1823         pol->curlft.use_time = get_seconds();
1824
1825         pols[0] = pol;
1826         npols ++;
1827 #ifdef CONFIG_XFRM_SUB_POLICY
1828         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1829                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1830                                                     &fl, family,
1831                                                     XFRM_POLICY_IN);
1832                 if (pols[1]) {
1833                         if (IS_ERR(pols[1]))
1834                                 return 0;
1835                         pols[1]->curlft.use_time = get_seconds();
1836                         npols ++;
1837                 }
1838         }
1839 #endif
1840
1841         if (pol->action == XFRM_POLICY_ALLOW) {
1842                 struct sec_path *sp;
1843                 static struct sec_path dummy;
1844                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1845                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1846                 struct xfrm_tmpl **tpp = tp;
1847                 int ti = 0;
1848                 int i, k;
1849
1850                 if ((sp = skb->sp) == NULL)
1851                         sp = &dummy;
1852
1853                 for (pi = 0; pi < npols; pi++) {
1854                         if (pols[pi] != pol &&
1855                             pols[pi]->action != XFRM_POLICY_ALLOW)
1856                                 goto reject;
1857                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1858                                 goto reject_error;
1859                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1860                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1861                 }
1862                 xfrm_nr = ti;
1863                 if (npols > 1) {
1864                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1865                         tpp = stp;
1866                 }
1867
1868                 /* For each tunnel xfrm, find the first matching tmpl.
1869                  * For each tmpl before that, find corresponding xfrm.
1870                  * Order is _important_. Later we will implement
1871                  * some barriers, but at the moment barriers
1872                  * are implied between each two transformations.
1873                  */
1874                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1875                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1876                         if (k < 0) {
1877                                 if (k < -1)
1878                                         /* "-2 - errored_index" returned */
1879                                         xerr_idx = -(2+k);
1880                                 goto reject;
1881                         }
1882                 }
1883
1884                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1885                         goto reject;
1886
1887                 xfrm_pols_put(pols, npols);
1888                 return 1;
1889         }
1890
1891 reject:
1892         xfrm_secpath_reject(xerr_idx, skb, &fl);
1893 reject_error:
1894         xfrm_pols_put(pols, npols);
1895         return 0;
1896 }
1897 EXPORT_SYMBOL(__xfrm_policy_check);
1898
1899 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1900 {
1901         struct flowi fl;
1902
1903         if (xfrm_decode_session(skb, &fl, family) < 0)
1904                 return 0;
1905
1906         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1907 }
1908 EXPORT_SYMBOL(__xfrm_route_forward);
1909
1910 /* Optimize later using cookies and generation ids. */
1911
1912 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1913 {
1914         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1915          * to "-1" to force all XFRM destinations to get validated by
1916          * dst_ops->check on every use.  We do this because when a
1917          * normal route referenced by an XFRM dst is obsoleted we do
1918          * not go looking around for all parent referencing XFRM dsts
1919          * so that we can invalidate them.  It is just too much work.
1920          * Instead we make the checks here on every use.  For example:
1921          *
1922          *      XFRM dst A --> IPv4 dst X
1923          *
1924          * X is the "xdst->route" of A (X is also the "dst->path" of A
1925          * in this example).  If X is marked obsolete, "A" will not
1926          * notice.  That's what we are validating here via the
1927          * stale_bundle() check.
1928          *
1929          * When a policy's bundle is pruned, we dst_free() the XFRM
1930          * dst which causes it's ->obsolete field to be set to a
1931          * positive non-zero integer.  If an XFRM dst has been pruned
1932          * like this, we want to force a new route lookup.
1933          */
1934         if (dst->obsolete < 0 && !stale_bundle(dst))
1935                 return dst;
1936
1937         return NULL;
1938 }
1939
1940 static int stale_bundle(struct dst_entry *dst)
1941 {
1942         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1943 }
1944
1945 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1946 {
1947         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1948                 dst->dev = init_net.loopback_dev;
1949                 dev_hold(dst->dev);
1950                 dev_put(dev);
1951         }
1952 }
1953 EXPORT_SYMBOL(xfrm_dst_ifdown);
1954
1955 static void xfrm_link_failure(struct sk_buff *skb)
1956 {
1957         /* Impossible. Such dst must be popped before reaches point of failure. */
1958         return;
1959 }
1960
1961 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1962 {
1963         if (dst) {
1964                 if (dst->obsolete) {
1965                         dst_release(dst);
1966                         dst = NULL;
1967                 }
1968         }
1969         return dst;
1970 }
1971
1972 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1973 {
1974         struct dst_entry *dst, **dstp;
1975
1976         write_lock(&pol->lock);
1977         dstp = &pol->bundles;
1978         while ((dst=*dstp) != NULL) {
1979                 if (func(dst)) {
1980                         *dstp = dst->next;
1981                         dst->next = *gc_list_p;
1982                         *gc_list_p = dst;
1983                 } else {
1984                         dstp = &dst->next;
1985                 }
1986         }
1987         write_unlock(&pol->lock);
1988 }
1989
1990 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1991 {
1992         struct dst_entry *gc_list = NULL;
1993         int dir;
1994
1995         read_lock_bh(&xfrm_policy_lock);
1996         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1997                 struct xfrm_policy *pol;
1998                 struct hlist_node *entry;
1999                 struct hlist_head *table;
2000                 int i;
2001
2002                 hlist_for_each_entry(pol, entry,
2003                                      &xfrm_policy_inexact[dir], bydst)
2004                         prune_one_bundle(pol, func, &gc_list);
2005
2006                 table = xfrm_policy_bydst[dir].table;
2007                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2008                         hlist_for_each_entry(pol, entry, table + i, bydst)
2009                                 prune_one_bundle(pol, func, &gc_list);
2010                 }
2011         }
2012         read_unlock_bh(&xfrm_policy_lock);
2013
2014         while (gc_list) {
2015                 struct dst_entry *dst = gc_list;
2016                 gc_list = dst->next;
2017                 dst_free(dst);
2018         }
2019 }
2020
2021 static int unused_bundle(struct dst_entry *dst)
2022 {
2023         return !atomic_read(&dst->__refcnt);
2024 }
2025
2026 static void __xfrm_garbage_collect(void)
2027 {
2028         xfrm_prune_bundles(unused_bundle);
2029 }
2030
2031 static int xfrm_flush_bundles(void)
2032 {
2033         xfrm_prune_bundles(stale_bundle);
2034         return 0;
2035 }
2036
2037 void xfrm_init_pmtu(struct dst_entry *dst)
2038 {
2039         do {
2040                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2041                 u32 pmtu, route_mtu_cached;
2042
2043                 pmtu = dst_mtu(dst->child);
2044                 xdst->child_mtu_cached = pmtu;
2045
2046                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2047
2048                 route_mtu_cached = dst_mtu(xdst->route);
2049                 xdst->route_mtu_cached = route_mtu_cached;
2050
2051                 if (pmtu > route_mtu_cached)
2052                         pmtu = route_mtu_cached;
2053
2054                 dst->metrics[RTAX_MTU-1] = pmtu;
2055         } while ((dst = dst->next));
2056 }
2057
2058 EXPORT_SYMBOL(xfrm_init_pmtu);
2059
2060 /* Check that the bundle accepts the flow and its components are
2061  * still valid.
2062  */
2063
2064 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2065                 struct flowi *fl, int family, int strict)
2066 {
2067         struct dst_entry *dst = &first->u.dst;
2068         struct xfrm_dst *last;
2069         u32 mtu;
2070
2071         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2072             (dst->dev && !netif_running(dst->dev)))
2073                 return 0;
2074 #ifdef CONFIG_XFRM_SUB_POLICY
2075         if (fl) {
2076                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2077                         return 0;
2078                 if (first->partner &&
2079                     !xfrm_selector_match(first->partner, fl, family))
2080                         return 0;
2081         }
2082 #endif
2083
2084         last = NULL;
2085
2086         do {
2087                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2088
2089                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2090                         return 0;
2091                 if (fl && pol &&
2092                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2093                         return 0;
2094                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2095                         return 0;
2096                 if (xdst->genid != dst->xfrm->genid)
2097                         return 0;
2098
2099                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
2100                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2101                         return 0;
2102
2103                 mtu = dst_mtu(dst->child);
2104                 if (xdst->child_mtu_cached != mtu) {
2105                         last = xdst;
2106                         xdst->child_mtu_cached = mtu;
2107                 }
2108
2109                 if (!dst_check(xdst->route, xdst->route_cookie))
2110                         return 0;
2111                 mtu = dst_mtu(xdst->route);
2112                 if (xdst->route_mtu_cached != mtu) {
2113                         last = xdst;
2114                         xdst->route_mtu_cached = mtu;
2115                 }
2116
2117                 dst = dst->child;
2118         } while (dst->xfrm);
2119
2120         if (likely(!last))
2121                 return 1;
2122
2123         mtu = last->child_mtu_cached;
2124         for (;;) {
2125                 dst = &last->u.dst;
2126
2127                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2128                 if (mtu > last->route_mtu_cached)
2129                         mtu = last->route_mtu_cached;
2130                 dst->metrics[RTAX_MTU-1] = mtu;
2131
2132                 if (last == first)
2133                         break;
2134
2135                 last = (struct xfrm_dst *)last->u.dst.next;
2136                 last->child_mtu_cached = mtu;
2137         }
2138
2139         return 1;
2140 }
2141
2142 EXPORT_SYMBOL(xfrm_bundle_ok);
2143
2144 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2145 {
2146         int err = 0;
2147         if (unlikely(afinfo == NULL))
2148                 return -EINVAL;
2149         if (unlikely(afinfo->family >= NPROTO))
2150                 return -EAFNOSUPPORT;
2151         write_lock_bh(&xfrm_policy_afinfo_lock);
2152         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2153                 err = -ENOBUFS;
2154         else {
2155                 struct dst_ops *dst_ops = afinfo->dst_ops;
2156                 if (likely(dst_ops->kmem_cachep == NULL))
2157                         dst_ops->kmem_cachep = xfrm_dst_cache;
2158                 if (likely(dst_ops->check == NULL))
2159                         dst_ops->check = xfrm_dst_check;
2160                 if (likely(dst_ops->negative_advice == NULL))
2161                         dst_ops->negative_advice = xfrm_negative_advice;
2162                 if (likely(dst_ops->link_failure == NULL))
2163                         dst_ops->link_failure = xfrm_link_failure;
2164                 if (likely(afinfo->garbage_collect == NULL))
2165                         afinfo->garbage_collect = __xfrm_garbage_collect;
2166                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2167         }
2168         write_unlock_bh(&xfrm_policy_afinfo_lock);
2169         return err;
2170 }
2171 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2172
2173 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2174 {
2175         int err = 0;
2176         if (unlikely(afinfo == NULL))
2177                 return -EINVAL;
2178         if (unlikely(afinfo->family >= NPROTO))
2179                 return -EAFNOSUPPORT;
2180         write_lock_bh(&xfrm_policy_afinfo_lock);
2181         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2182                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2183                         err = -EINVAL;
2184                 else {
2185                         struct dst_ops *dst_ops = afinfo->dst_ops;
2186                         xfrm_policy_afinfo[afinfo->family] = NULL;
2187                         dst_ops->kmem_cachep = NULL;
2188                         dst_ops->check = NULL;
2189                         dst_ops->negative_advice = NULL;
2190                         dst_ops->link_failure = NULL;
2191                         afinfo->garbage_collect = NULL;
2192                 }
2193         }
2194         write_unlock_bh(&xfrm_policy_afinfo_lock);
2195         return err;
2196 }
2197 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2198
2199 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2200 {
2201         struct xfrm_policy_afinfo *afinfo;
2202         if (unlikely(family >= NPROTO))
2203                 return NULL;
2204         read_lock(&xfrm_policy_afinfo_lock);
2205         afinfo = xfrm_policy_afinfo[family];
2206         if (unlikely(!afinfo))
2207                 read_unlock(&xfrm_policy_afinfo_lock);
2208         return afinfo;
2209 }
2210
2211 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2212 {
2213         read_unlock(&xfrm_policy_afinfo_lock);
2214 }
2215
2216 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2217 {
2218         struct xfrm_policy_afinfo *afinfo;
2219         if (unlikely(family >= NPROTO))
2220                 return NULL;
2221         write_lock_bh(&xfrm_policy_afinfo_lock);
2222         afinfo = xfrm_policy_afinfo[family];
2223         if (unlikely(!afinfo))
2224                 write_unlock_bh(&xfrm_policy_afinfo_lock);
2225         return afinfo;
2226 }
2227
2228 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2229 {
2230         write_unlock_bh(&xfrm_policy_afinfo_lock);
2231 }
2232
2233 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2234 {
2235         struct net_device *dev = ptr;
2236
2237         if (dev->nd_net != &init_net)
2238                 return NOTIFY_DONE;
2239
2240         switch (event) {
2241         case NETDEV_DOWN:
2242                 xfrm_flush_bundles();
2243         }
2244         return NOTIFY_DONE;
2245 }
2246
2247 static struct notifier_block xfrm_dev_notifier = {
2248         xfrm_dev_event,
2249         NULL,
2250         0
2251 };
2252
2253 static void __init xfrm_policy_init(void)
2254 {
2255         unsigned int hmask, sz;
2256         int dir;
2257
2258         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2259                                            sizeof(struct xfrm_dst),
2260                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2261                                            NULL);
2262
2263         hmask = 8 - 1;
2264         sz = (hmask+1) * sizeof(struct hlist_head);
2265
2266         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2267         xfrm_idx_hmask = hmask;
2268         if (!xfrm_policy_byidx)
2269                 panic("XFRM: failed to allocate byidx hash\n");
2270
2271         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2272                 struct xfrm_policy_hash *htab;
2273
2274                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2275
2276                 htab = &xfrm_policy_bydst[dir];
2277                 htab->table = xfrm_hash_alloc(sz);
2278                 htab->hmask = hmask;
2279                 if (!htab->table)
2280                         panic("XFRM: failed to allocate bydst hash\n");
2281         }
2282
2283         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2284         register_netdevice_notifier(&xfrm_dev_notifier);
2285 }
2286
2287 void __init xfrm_init(void)
2288 {
2289         xfrm_state_init();
2290         xfrm_policy_init();
2291         xfrm_input_init();
2292 }
2293
2294 #ifdef CONFIG_AUDITSYSCALL
2295 static inline void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2296                                                 struct audit_buffer *audit_buf)
2297 {
2298         if (xp->security)
2299                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2300                                  xp->security->ctx_alg, xp->security->ctx_doi,
2301                                  xp->security->ctx_str);
2302
2303         switch(xp->selector.family) {
2304         case AF_INET:
2305                 audit_log_format(audit_buf, " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2306                                  NIPQUAD(xp->selector.saddr.a4),
2307                                  NIPQUAD(xp->selector.daddr.a4));
2308                 break;
2309         case AF_INET6:
2310                 {
2311                         struct in6_addr saddr6, daddr6;
2312
2313                         memcpy(&saddr6, xp->selector.saddr.a6,
2314                                 sizeof(struct in6_addr));
2315                         memcpy(&daddr6, xp->selector.daddr.a6,
2316                                 sizeof(struct in6_addr));
2317                         audit_log_format(audit_buf,
2318                                 " src=" NIP6_FMT " dst=" NIP6_FMT,
2319                                 NIP6(saddr6), NIP6(daddr6));
2320                 }
2321                 break;
2322         }
2323 }
2324
2325 void
2326 xfrm_audit_policy_add(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2327 {
2328         struct audit_buffer *audit_buf;
2329         extern int audit_enabled;
2330
2331         if (audit_enabled == 0)
2332                 return;
2333         audit_buf = xfrm_audit_start(sid, auid);
2334         if (audit_buf == NULL)
2335                 return;
2336         audit_log_format(audit_buf, " op=SPD-add res=%u", result);
2337         xfrm_audit_common_policyinfo(xp, audit_buf);
2338         audit_log_end(audit_buf);
2339 }
2340 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2341
2342 void
2343 xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2344 {
2345         struct audit_buffer *audit_buf;
2346         extern int audit_enabled;
2347
2348         if (audit_enabled == 0)
2349                 return;
2350         audit_buf = xfrm_audit_start(sid, auid);
2351         if (audit_buf == NULL)
2352                 return;
2353         audit_log_format(audit_buf, " op=SPD-delete res=%u", result);
2354         xfrm_audit_common_policyinfo(xp, audit_buf);
2355         audit_log_end(audit_buf);
2356 }
2357 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2358 #endif
2359
2360 #ifdef CONFIG_XFRM_MIGRATE
2361 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2362                                        struct xfrm_selector *sel_tgt)
2363 {
2364         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2365                 if (sel_tgt->family == sel_cmp->family &&
2366                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2367                                   sel_cmp->family) == 0 &&
2368                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2369                                   sel_cmp->family) == 0 &&
2370                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2371                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2372                         return 1;
2373                 }
2374         } else {
2375                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2376                         return 1;
2377                 }
2378         }
2379         return 0;
2380 }
2381
2382 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2383                                                      u8 dir, u8 type)
2384 {
2385         struct xfrm_policy *pol, *ret = NULL;
2386         struct hlist_node *entry;
2387         struct hlist_head *chain;
2388         u32 priority = ~0U;
2389
2390         read_lock_bh(&xfrm_policy_lock);
2391         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2392         hlist_for_each_entry(pol, entry, chain, bydst) {
2393                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2394                     pol->type == type) {
2395                         ret = pol;
2396                         priority = ret->priority;
2397                         break;
2398                 }
2399         }
2400         chain = &xfrm_policy_inexact[dir];
2401         hlist_for_each_entry(pol, entry, chain, bydst) {
2402                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2403                     pol->type == type &&
2404                     pol->priority < priority) {
2405                         ret = pol;
2406                         break;
2407                 }
2408         }
2409
2410         if (ret)
2411                 xfrm_pol_hold(ret);
2412
2413         read_unlock_bh(&xfrm_policy_lock);
2414
2415         return ret;
2416 }
2417
2418 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2419 {
2420         int match = 0;
2421
2422         if (t->mode == m->mode && t->id.proto == m->proto &&
2423             (m->reqid == 0 || t->reqid == m->reqid)) {
2424                 switch (t->mode) {
2425                 case XFRM_MODE_TUNNEL:
2426                 case XFRM_MODE_BEET:
2427                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2428                                           m->old_family) == 0 &&
2429                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2430                                           m->old_family) == 0) {
2431                                 match = 1;
2432                         }
2433                         break;
2434                 case XFRM_MODE_TRANSPORT:
2435                         /* in case of transport mode, template does not store
2436                            any IP addresses, hence we just compare mode and
2437                            protocol */
2438                         match = 1;
2439                         break;
2440                 default:
2441                         break;
2442                 }
2443         }
2444         return match;
2445 }
2446
2447 /* update endpoint address(es) of template(s) */
2448 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2449                                struct xfrm_migrate *m, int num_migrate)
2450 {
2451         struct xfrm_migrate *mp;
2452         struct dst_entry *dst;
2453         int i, j, n = 0;
2454
2455         write_lock_bh(&pol->lock);
2456         if (unlikely(pol->dead)) {
2457                 /* target policy has been deleted */
2458                 write_unlock_bh(&pol->lock);
2459                 return -ENOENT;
2460         }
2461
2462         for (i = 0; i < pol->xfrm_nr; i++) {
2463                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2464                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2465                                 continue;
2466                         n++;
2467                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL)
2468                                 continue;
2469                         /* update endpoints */
2470                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2471                                sizeof(pol->xfrm_vec[i].id.daddr));
2472                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2473                                sizeof(pol->xfrm_vec[i].saddr));
2474                         pol->xfrm_vec[i].encap_family = mp->new_family;
2475                         /* flush bundles */
2476                         while ((dst = pol->bundles) != NULL) {
2477                                 pol->bundles = dst->next;
2478                                 dst_free(dst);
2479                         }
2480                 }
2481         }
2482
2483         write_unlock_bh(&pol->lock);
2484
2485         if (!n)
2486                 return -ENODATA;
2487
2488         return 0;
2489 }
2490
2491 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2492 {
2493         int i, j;
2494
2495         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2496                 return -EINVAL;
2497
2498         for (i = 0; i < num_migrate; i++) {
2499                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2500                                    m[i].old_family) == 0) &&
2501                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2502                                    m[i].old_family) == 0))
2503                         return -EINVAL;
2504                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2505                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2506                         return -EINVAL;
2507
2508                 /* check if there is any duplicated entry */
2509                 for (j = i + 1; j < num_migrate; j++) {
2510                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2511                                     sizeof(m[i].old_daddr)) &&
2512                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2513                                     sizeof(m[i].old_saddr)) &&
2514                             m[i].proto == m[j].proto &&
2515                             m[i].mode == m[j].mode &&
2516                             m[i].reqid == m[j].reqid &&
2517                             m[i].old_family == m[j].old_family)
2518                                 return -EINVAL;
2519                 }
2520         }
2521
2522         return 0;
2523 }
2524
2525 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2526                  struct xfrm_migrate *m, int num_migrate)
2527 {
2528         int i, err, nx_cur = 0, nx_new = 0;
2529         struct xfrm_policy *pol = NULL;
2530         struct xfrm_state *x, *xc;
2531         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2532         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2533         struct xfrm_migrate *mp;
2534
2535         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2536                 goto out;
2537
2538         /* Stage 1 - find policy */
2539         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2540                 err = -ENOENT;
2541                 goto out;
2542         }
2543
2544         /* Stage 2 - find and update state(s) */
2545         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2546                 if ((x = xfrm_migrate_state_find(mp))) {
2547                         x_cur[nx_cur] = x;
2548                         nx_cur++;
2549                         if ((xc = xfrm_state_migrate(x, mp))) {
2550                                 x_new[nx_new] = xc;
2551                                 nx_new++;
2552                         } else {
2553                                 err = -ENODATA;
2554                                 goto restore_state;
2555                         }
2556                 }
2557         }
2558
2559         /* Stage 3 - update policy */
2560         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2561                 goto restore_state;
2562
2563         /* Stage 4 - delete old state(s) */
2564         if (nx_cur) {
2565                 xfrm_states_put(x_cur, nx_cur);
2566                 xfrm_states_delete(x_cur, nx_cur);
2567         }
2568
2569         /* Stage 5 - announce */
2570         km_migrate(sel, dir, type, m, num_migrate);
2571
2572         xfrm_pol_put(pol);
2573
2574         return 0;
2575 out:
2576         return err;
2577
2578 restore_state:
2579         if (pol)
2580                 xfrm_pol_put(pol);
2581         if (nx_cur)
2582                 xfrm_states_put(x_cur, nx_cur);
2583         if (nx_new)
2584                 xfrm_states_delete(x_new, nx_new);
2585
2586         return err;
2587 }
2588 EXPORT_SYMBOL(xfrm_migrate);
2589 #endif