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 #include <linux/audit.h>
  29
  30 #include "xfrm_hash.h"
  31
  32 DEFINE_MUTEX(xfrm_cfg_mutex);
  33 EXPORT_SYMBOL(xfrm_cfg_mutex);
  34
  35 static DEFINE_RWLOCK(xfrm_policy_lock);
  36
  37 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  38 EXPORT_SYMBOL(xfrm_policy_count);
  39
  40 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  41 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  42
  43 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  44
  45 static struct work_struct xfrm_policy_gc_work;
  46 static HLIST_HEAD(xfrm_policy_gc_list);
  47 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  48
  49 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  50 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  51 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
  52 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
  53
  54 static inline int
  55 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  56 {
  57         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  58                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  59                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  60                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  61                 (fl->proto == sel->proto || !sel->proto) &&
  62                 (fl->oif == sel->ifindex || !sel->ifindex);
  63 }
  64
  65 static inline int
  66 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  67 {
  68         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  69                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  70                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  71                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  72                 (fl->proto == sel->proto || !sel->proto) &&
  73                 (fl->oif == sel->ifindex || !sel->ifindex);
  74 }
  75
  76 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  77                     unsigned short family)
  78 {
  79         switch (family) {
  80         case AF_INET:
  81                 return __xfrm4_selector_match(sel, fl);
  82         case AF_INET6:
  83                 return __xfrm6_selector_match(sel, fl);
  84         }
  85         return 0;
  86 }
  87
  88 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
  89 {
  90         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
  91         struct xfrm_type **typemap;
  92         int err = 0;
  93
  94         if (unlikely(afinfo == NULL))
  95                 return -EAFNOSUPPORT;
  96         typemap = afinfo->type_map;
  97
  98         if (likely(typemap[type->proto] == NULL))
  99                 typemap[type->proto] = type;
 100         else
 101                 err = -EEXIST;
 102         xfrm_policy_unlock_afinfo(afinfo);
 103         return err;
 104 }
 105 EXPORT_SYMBOL(xfrm_register_type);
 106
 107 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
 108 {
 109         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
 110         struct xfrm_type **typemap;
 111         int err = 0;
 112
 113         if (unlikely(afinfo == NULL))
 114                 return -EAFNOSUPPORT;
 115         typemap = afinfo->type_map;
 116
 117         if (unlikely(typemap[type->proto] != type))
 118                 err = -ENOENT;
 119         else
 120                 typemap[type->proto] = NULL;
 121         xfrm_policy_unlock_afinfo(afinfo);
 122         return err;
 123 }
 124 EXPORT_SYMBOL(xfrm_unregister_type);
 125
 126 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
 127 {
 128         struct xfrm_policy_afinfo *afinfo;
 129         struct xfrm_type **typemap;
 130         struct xfrm_type *type;
 131         int modload_attempted = 0;
 132
 133 retry:
 134         afinfo = xfrm_policy_get_afinfo(family);
 135         if (unlikely(afinfo == NULL))
 136                 return NULL;
 137         typemap = afinfo->type_map;
 138
 139         type = typemap[proto];
 140         if (unlikely(type && !try_module_get(type->owner)))
 141                 type = NULL;
 142         if (!type && !modload_attempted) {
 143                 xfrm_policy_put_afinfo(afinfo);
 144                 request_module("xfrm-type-%d-%d",
 145                                (int) family, (int) proto);
 146                 modload_attempted = 1;
 147                 goto retry;
 148         }
 149
 150         xfrm_policy_put_afinfo(afinfo);
 151         return type;
 152 }
 153
 154 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
 155                     unsigned short family)
 156 {
 157         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
 158         int err = 0;
 159
 160         if (unlikely(afinfo == NULL))
 161                 return -EAFNOSUPPORT;
 162
 163         if (likely(afinfo->dst_lookup != NULL))
 164                 err = afinfo->dst_lookup(dst, fl);
 165         else
 166                 err = -EINVAL;
 167         xfrm_policy_put_afinfo(afinfo);
 168         return err;
 169 }
 170 EXPORT_SYMBOL(xfrm_dst_lookup);
 171
 172 void xfrm_put_type(struct xfrm_type *type)
 173 {
 174         module_put(type->owner);
 175 }
 176
 177 int xfrm_register_mode(struct xfrm_mode *mode, int family)
 178 {
 179         struct xfrm_policy_afinfo *afinfo;
 180         struct xfrm_mode **modemap;
 181         int err;
 182
 183         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 184                 return -EINVAL;
 185
 186         afinfo = xfrm_policy_lock_afinfo(family);
 187         if (unlikely(afinfo == NULL))
 188                 return -EAFNOSUPPORT;
 189
 190         err = -EEXIST;
 191         modemap = afinfo->mode_map;
 192         if (likely(modemap[mode->encap] == NULL)) {
 193                 modemap[mode->encap] = mode;
 194                 err = 0;
 195         }
 196
 197         xfrm_policy_unlock_afinfo(afinfo);
 198         return err;
 199 }
 200 EXPORT_SYMBOL(xfrm_register_mode);
 201
 202 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
 203 {
 204         struct xfrm_policy_afinfo *afinfo;
 205         struct xfrm_mode **modemap;
 206         int err;
 207
 208         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 209                 return -EINVAL;
 210
 211         afinfo = xfrm_policy_lock_afinfo(family);
 212         if (unlikely(afinfo == NULL))
 213                 return -EAFNOSUPPORT;
 214
 215         err = -ENOENT;
 216         modemap = afinfo->mode_map;
 217         if (likely(modemap[mode->encap] == mode)) {
 218                 modemap[mode->encap] = NULL;
 219                 err = 0;
 220         }
 221
 222         xfrm_policy_unlock_afinfo(afinfo);
 223         return err;
 224 }
 225 EXPORT_SYMBOL(xfrm_unregister_mode);
 226
 227 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
 228 {
 229         struct xfrm_policy_afinfo *afinfo;
 230         struct xfrm_mode *mode;
 231         int modload_attempted = 0;
 232
 233         if (unlikely(encap >= XFRM_MODE_MAX))
 234                 return NULL;
 235
 236 retry:
 237         afinfo = xfrm_policy_get_afinfo(family);
 238         if (unlikely(afinfo == NULL))
 239                 return NULL;
 240
 241         mode = afinfo->mode_map[encap];
 242         if (unlikely(mode && !try_module_get(mode->owner)))
 243                 mode = NULL;
 244         if (!mode && !modload_attempted) {
 245                 xfrm_policy_put_afinfo(afinfo);
 246                 request_module("xfrm-mode-%d-%d", family, encap);
 247                 modload_attempted = 1;
 248                 goto retry;
 249         }
 250
 251         xfrm_policy_put_afinfo(afinfo);
 252         return mode;
 253 }
 254
 255 void xfrm_put_mode(struct xfrm_mode *mode)
 256 {
 257         module_put(mode->owner);
 258 }
 259
 260 static inline unsigned long make_jiffies(long secs)
 261 {
 262         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 263                 return MAX_SCHEDULE_TIMEOUT-1;
 264         else
 265                 return secs*HZ;
 266 }
 267
 268 static void xfrm_policy_timer(unsigned long data)
 269 {
 270         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 271         unsigned long now = get_seconds();
 272         long next = LONG_MAX;
 273         int warn = 0;
 274         int dir;
 275
 276         read_lock(&xp->lock);
 277
 278         if (xp->dead)
 279                 goto out;
 280
 281         dir = xfrm_policy_id2dir(xp->index);
 282
 283         if (xp->lft.hard_add_expires_seconds) {
 284                 long tmo = xp->lft.hard_add_expires_seconds +
 285                         xp->curlft.add_time - now;
 286                 if (tmo <= 0)
 287                         goto expired;
 288                 if (tmo < next)
 289                         next = tmo;
 290         }
 291         if (xp->lft.hard_use_expires_seconds) {
 292                 long tmo = xp->lft.hard_use_expires_seconds +
 293                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 294                 if (tmo <= 0)
 295                         goto expired;
 296                 if (tmo < next)
 297                         next = tmo;
 298         }
 299         if (xp->lft.soft_add_expires_seconds) {
 300                 long tmo = xp->lft.soft_add_expires_seconds +
 301                         xp->curlft.add_time - now;
 302                 if (tmo <= 0) {
 303                         warn = 1;
 304                         tmo = XFRM_KM_TIMEOUT;
 305                 }
 306                 if (tmo < next)
 307                         next = tmo;
 308         }
 309         if (xp->lft.soft_use_expires_seconds) {
 310                 long tmo = xp->lft.soft_use_expires_seconds +
 311                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 312                 if (tmo <= 0) {
 313                         warn = 1;
 314                         tmo = XFRM_KM_TIMEOUT;
 315                 }
 316                 if (tmo < next)
 317                         next = tmo;
 318         }
 319
 320         if (warn)
 321                 km_policy_expired(xp, dir, 0, 0);
 322         if (next != LONG_MAX &&
 323             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 324                 xfrm_pol_hold(xp);
 325
 326 out:
 327         read_unlock(&xp->lock);
 328         xfrm_pol_put(xp);
 329         return;
 330
 331 expired:
 332         read_unlock(&xp->lock);
 333         if (!xfrm_policy_delete(xp, dir))
 334                 km_policy_expired(xp, dir, 1, 0);
 335         xfrm_pol_put(xp);
 336 }
 337
 338
 339 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 340  * SPD calls.
 341  */
 342
 343 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 344 {
 345         struct xfrm_policy *policy;
 346
 347         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 348
 349         if (policy) {
 350                 INIT_HLIST_NODE(&policy->bydst);
 351                 INIT_HLIST_NODE(&policy->byidx);
 352                 rwlock_init(&policy->lock);
 353                 atomic_set(&policy->refcnt, 1);
 354                 init_timer(&policy->timer);
 355                 policy->timer.data = (unsigned long)policy;
 356                 policy->timer.function = xfrm_policy_timer;
 357         }
 358         return policy;
 359 }
 360 EXPORT_SYMBOL(xfrm_policy_alloc);
 361
 362 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 363
 364 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 365 {
 366         BUG_ON(!policy->dead);
 367
 368         BUG_ON(policy->bundles);
 369
 370         if (del_timer(&policy->timer))
 371                 BUG();
 372
 373         security_xfrm_policy_free(policy);
 374         kfree(policy);
 375 }
 376 EXPORT_SYMBOL(__xfrm_policy_destroy);
 377
 378 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 379 {
 380         struct dst_entry *dst;
 381
 382         while ((dst = policy->bundles) != NULL) {
 383                 policy->bundles = dst->next;
 384                 dst_free(dst);
 385         }
 386
 387         if (del_timer(&policy->timer))
 388                 atomic_dec(&policy->refcnt);
 389
 390         if (atomic_read(&policy->refcnt) > 1)
 391                 flow_cache_flush();
 392
 393         xfrm_pol_put(policy);
 394 }
 395
 396 static void xfrm_policy_gc_task(struct work_struct *work)
 397 {
 398         struct xfrm_policy *policy;
 399         struct hlist_node *entry, *tmp;
 400         struct hlist_head gc_list;
 401
 402         spin_lock_bh(&xfrm_policy_gc_lock);
 403         gc_list.first = xfrm_policy_gc_list.first;
 404         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 405         spin_unlock_bh(&xfrm_policy_gc_lock);
 406
 407         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 408                 xfrm_policy_gc_kill(policy);
 409 }
 410
 411 /* Rule must be locked. Release descentant resources, announce
 412  * entry dead. The rule must be unlinked from lists to the moment.
 413  */
 414
 415 static void xfrm_policy_kill(struct xfrm_policy *policy)
 416 {
 417         int dead;
 418
 419         write_lock_bh(&policy->lock);
 420         dead = policy->dead;
 421         policy->dead = 1;
 422         write_unlock_bh(&policy->lock);
 423
 424         if (unlikely(dead)) {
 425                 WARN_ON(1);
 426                 return;
 427         }
 428
 429         spin_lock(&xfrm_policy_gc_lock);
 430         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 431         spin_unlock(&xfrm_policy_gc_lock);
 432
 433         schedule_work(&xfrm_policy_gc_work);
 434 }
 435
 436 struct xfrm_policy_hash {
 437         struct hlist_head       *table;
 438         unsigned int            hmask;
 439 };
 440
 441 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 442 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 443 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 444 static unsigned int xfrm_idx_hmask __read_mostly;
 445 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 446
 447 static inline unsigned int idx_hash(u32 index)
 448 {
 449         return __idx_hash(index, xfrm_idx_hmask);
 450 }
 451
 452 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 453 {
 454         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 455         unsigned int hash = __sel_hash(sel, family, hmask);
 456
 457         return (hash == hmask + 1 ?
 458                 &xfrm_policy_inexact[dir] :
 459                 xfrm_policy_bydst[dir].table + hash);
 460 }
 461
 462 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 463 {
 464         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 465         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 466
 467         return xfrm_policy_bydst[dir].table + hash;
 468 }
 469
 470 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 471                                    struct hlist_head *ndsttable,
 472                                    unsigned int nhashmask)
 473 {
 474         struct hlist_node *entry, *tmp;
 475         struct xfrm_policy *pol;
 476
 477         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 478                 unsigned int h;
 479
 480                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 481                                 pol->family, nhashmask);
 482                 hlist_add_head(&pol->bydst, ndsttable+h);
 483         }
 484 }
 485
 486 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 487                                    struct hlist_head *nidxtable,
 488                                    unsigned int nhashmask)
 489 {
 490         struct hlist_node *entry, *tmp;
 491         struct xfrm_policy *pol;
 492
 493         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 494                 unsigned int h;
 495
 496                 h = __idx_hash(pol->index, nhashmask);
 497                 hlist_add_head(&pol->byidx, nidxtable+h);
 498         }
 499 }
 500
 501 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 502 {
 503         return ((old_hmask + 1) << 1) - 1;
 504 }
 505
 506 static void xfrm_bydst_resize(int dir)
 507 {
 508         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 509         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 510         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 511         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 512         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 513         int i;
 514
 515         if (!ndst)
 516                 return;
 517
 518         write_lock_bh(&xfrm_policy_lock);
 519
 520         for (i = hmask; i >= 0; i--)
 521                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 522
 523         xfrm_policy_bydst[dir].table = ndst;
 524         xfrm_policy_bydst[dir].hmask = nhashmask;
 525
 526         write_unlock_bh(&xfrm_policy_lock);
 527
 528         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 529 }
 530
 531 static void xfrm_byidx_resize(int total)
 532 {
 533         unsigned int hmask = xfrm_idx_hmask;
 534         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 535         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 536         struct hlist_head *oidx = xfrm_policy_byidx;
 537         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 538         int i;
 539
 540         if (!nidx)
 541                 return;
 542
 543         write_lock_bh(&xfrm_policy_lock);
 544
 545         for (i = hmask; i >= 0; i--)
 546                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 547
 548         xfrm_policy_byidx = nidx;
 549         xfrm_idx_hmask = nhashmask;
 550
 551         write_unlock_bh(&xfrm_policy_lock);
 552
 553         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 554 }
 555
 556 static inline int xfrm_bydst_should_resize(int dir, int *total)
 557 {
 558         unsigned int cnt = xfrm_policy_count[dir];
 559         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 560
 561         if (total)
 562                 *total += cnt;
 563
 564         if ((hmask + 1) < xfrm_policy_hashmax &&
 565             cnt > hmask)
 566                 return 1;
 567
 568         return 0;
 569 }
 570
 571 static inline int xfrm_byidx_should_resize(int total)
 572 {
 573         unsigned int hmask = xfrm_idx_hmask;
 574
 575         if ((hmask + 1) < xfrm_policy_hashmax &&
 576             total > hmask)
 577                 return 1;
 578
 579         return 0;
 580 }
 581
 582 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 583 {
 584         read_lock_bh(&xfrm_policy_lock);
 585         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 586         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 587         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 588         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 589         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 590         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 591         si->spdhcnt = xfrm_idx_hmask;
 592         si->spdhmcnt = xfrm_policy_hashmax;
 593         read_unlock_bh(&xfrm_policy_lock);
 594 }
 595 EXPORT_SYMBOL(xfrm_spd_getinfo);
 596
 597 static DEFINE_MUTEX(hash_resize_mutex);
 598 static void xfrm_hash_resize(struct work_struct *__unused)
 599 {
 600         int dir, total;
 601
 602         mutex_lock(&hash_resize_mutex);
 603
 604         total = 0;
 605         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 606                 if (xfrm_bydst_should_resize(dir, &total))
 607                         xfrm_bydst_resize(dir);
 608         }
 609         if (xfrm_byidx_should_resize(total))
 610                 xfrm_byidx_resize(total);
 611
 612         mutex_unlock(&hash_resize_mutex);
 613 }
 614
 615 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 616
 617 /* Generate new index... KAME seems to generate them ordered by cost
 618  * of an absolute inpredictability of ordering of rules. This will not pass. */
 619 static u32 xfrm_gen_index(u8 type, int dir)
 620 {
 621         static u32 idx_generator;
 622
 623         for (;;) {
 624                 struct hlist_node *entry;
 625                 struct hlist_head *list;
 626                 struct xfrm_policy *p;
 627                 u32 idx;
 628                 int found;
 629
 630                 idx = (idx_generator | dir);
 631                 idx_generator += 8;
 632                 if (idx == 0)
 633                         idx = 8;
 634                 list = xfrm_policy_byidx + idx_hash(idx);
 635                 found = 0;
 636                 hlist_for_each_entry(p, entry, list, byidx) {
 637                         if (p->index == idx) {
 638                                 found = 1;
 639                                 break;
 640                         }
 641                 }
 642                 if (!found)
 643                         return idx;
 644         }
 645 }
 646
 647 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 648 {
 649         u32 *p1 = (u32 *) s1;
 650         u32 *p2 = (u32 *) s2;
 651         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 652         int i;
 653
 654         for (i = 0; i < len; i++) {
 655                 if (p1[i] != p2[i])
 656                         return 1;
 657         }
 658
 659         return 0;
 660 }
 661
 662 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 663 {
 664         struct xfrm_policy *pol;
 665         struct xfrm_policy *delpol;
 666         struct hlist_head *chain;
 667         struct hlist_node *entry, *newpos;
 668         struct dst_entry *gc_list;
 669
 670         write_lock_bh(&xfrm_policy_lock);
 671         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 672         delpol = NULL;
 673         newpos = NULL;
 674         hlist_for_each_entry(pol, entry, chain, bydst) {
 675                 if (pol->type == policy->type &&
 676                     !selector_cmp(&pol->selector, &policy->selector) &&
 677                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 678                     !WARN_ON(delpol)) {
 679                         if (excl) {
 680                                 write_unlock_bh(&xfrm_policy_lock);
 681                                 return -EEXIST;
 682                         }
 683                         delpol = pol;
 684                         if (policy->priority > pol->priority)
 685                                 continue;
 686                 } else if (policy->priority >= pol->priority) {
 687                         newpos = &pol->bydst;
 688                         continue;
 689                 }
 690                 if (delpol)
 691                         break;
 692         }
 693         if (newpos)
 694                 hlist_add_after(newpos, &policy->bydst);
 695         else
 696                 hlist_add_head(&policy->bydst, chain);
 697         xfrm_pol_hold(policy);
 698         xfrm_policy_count[dir]++;
 699         atomic_inc(&flow_cache_genid);
 700         if (delpol) {
 701                 hlist_del(&delpol->bydst);
 702                 hlist_del(&delpol->byidx);
 703                 xfrm_policy_count[dir]--;
 704         }
 705         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 706         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 707         policy->curlft.add_time = get_seconds();
 708         policy->curlft.use_time = 0;
 709         if (!mod_timer(&policy->timer, jiffies + HZ))
 710                 xfrm_pol_hold(policy);
 711         write_unlock_bh(&xfrm_policy_lock);
 712
 713         if (delpol)
 714                 xfrm_policy_kill(delpol);
 715         else if (xfrm_bydst_should_resize(dir, NULL))
 716                 schedule_work(&xfrm_hash_work);
 717
 718         read_lock_bh(&xfrm_policy_lock);
 719         gc_list = NULL;
 720         entry = &policy->bydst;
 721         hlist_for_each_entry_continue(policy, entry, bydst) {
 722                 struct dst_entry *dst;
 723
 724                 write_lock(&policy->lock);
 725                 dst = policy->bundles;
 726                 if (dst) {
 727                         struct dst_entry *tail = dst;
 728                         while (tail->next)
 729                                 tail = tail->next;
 730                         tail->next = gc_list;
 731                         gc_list = dst;
 732
 733                         policy->bundles = NULL;
 734                 }
 735                 write_unlock(&policy->lock);
 736         }
 737         read_unlock_bh(&xfrm_policy_lock);
 738
 739         while (gc_list) {
 740                 struct dst_entry *dst = gc_list;
 741
 742                 gc_list = dst->next;
 743                 dst_free(dst);
 744         }
 745
 746         return 0;
 747 }
 748 EXPORT_SYMBOL(xfrm_policy_insert);
 749
 750 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 751                                           struct xfrm_selector *sel,
 752                                           struct xfrm_sec_ctx *ctx, int delete,
 753                                           int *err)
 754 {
 755         struct xfrm_policy *pol, *ret;
 756         struct hlist_head *chain;
 757         struct hlist_node *entry;
 758
 759         *err = 0;
 760         write_lock_bh(&xfrm_policy_lock);
 761         chain = policy_hash_bysel(sel, sel->family, dir);
 762         ret = NULL;
 763         hlist_for_each_entry(pol, entry, chain, bydst) {
 764                 if (pol->type == type &&
 765                     !selector_cmp(sel, &pol->selector) &&
 766                     xfrm_sec_ctx_match(ctx, pol->security)) {
 767                         xfrm_pol_hold(pol);
 768                         if (delete) {
 769                                 *err = security_xfrm_policy_delete(pol);
 770                                 if (*err) {
 771                                         write_unlock_bh(&xfrm_policy_lock);
 772                                         return pol;
 773                                 }
 774                                 hlist_del(&pol->bydst);
 775                                 hlist_del(&pol->byidx);
 776                                 xfrm_policy_count[dir]--;
 777                         }
 778                         ret = pol;
 779                         break;
 780                 }
 781         }
 782         write_unlock_bh(&xfrm_policy_lock);
 783
 784         if (ret && delete) {
 785                 atomic_inc(&flow_cache_genid);
 786                 xfrm_policy_kill(ret);
 787         }
 788         return ret;
 789 }
 790 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 791
 792 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 793                                      int *err)
 794 {
 795         struct xfrm_policy *pol, *ret;
 796         struct hlist_head *chain;
 797         struct hlist_node *entry;
 798
 799         *err = -ENOENT;
 800         if (xfrm_policy_id2dir(id) != dir)
 801                 return NULL;
 802
 803         *err = 0;
 804         write_lock_bh(&xfrm_policy_lock);
 805         chain = xfrm_policy_byidx + idx_hash(id);
 806         ret = NULL;
 807         hlist_for_each_entry(pol, entry, chain, byidx) {
 808                 if (pol->type == type && pol->index == id) {
 809                         xfrm_pol_hold(pol);
 810                         if (delete) {
 811                                 *err = security_xfrm_policy_delete(pol);
 812                                 if (*err) {
 813                                         write_unlock_bh(&xfrm_policy_lock);
 814                                         return pol;
 815                                 }
 816                                 hlist_del(&pol->bydst);
 817                                 hlist_del(&pol->byidx);
 818                                 xfrm_policy_count[dir]--;
 819                         }
 820                         ret = pol;
 821                         break;
 822                 }
 823         }
 824         write_unlock_bh(&xfrm_policy_lock);
 825
 826         if (ret && delete) {
 827                 atomic_inc(&flow_cache_genid);
 828                 xfrm_policy_kill(ret);
 829         }
 830         return ret;
 831 }
 832 EXPORT_SYMBOL(xfrm_policy_byid);
 833
 834 void xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 835 {
 836         int dir;
 837
 838         write_lock_bh(&xfrm_policy_lock);
 839         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 840                 struct xfrm_policy *pol;
 841                 struct hlist_node *entry;
 842                 int i, killed;
 843
 844                 killed = 0;
 845         again1:
 846                 hlist_for_each_entry(pol, entry,
 847                                      &xfrm_policy_inexact[dir], bydst) {
 848                         if (pol->type != type)
 849                                 continue;
 850                         hlist_del(&pol->bydst);
 851                         hlist_del(&pol->byidx);
 852                         write_unlock_bh(&xfrm_policy_lock);
 853
 854                         xfrm_audit_log(audit_info->loginuid, audit_info->secid,
 855                                        AUDIT_MAC_IPSEC_DELSPD, 1, pol, NULL);
 856
 857                         xfrm_policy_kill(pol);
 858                         killed++;
 859
 860                         write_lock_bh(&xfrm_policy_lock);
 861                         goto again1;
 862                 }
 863
 864                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 865         again2:
 866                         hlist_for_each_entry(pol, entry,
 867                                              xfrm_policy_bydst[dir].table + i,
 868                                              bydst) {
 869                                 if (pol->type != type)
 870                                         continue;
 871                                 hlist_del(&pol->bydst);
 872                                 hlist_del(&pol->byidx);
 873                                 write_unlock_bh(&xfrm_policy_lock);
 874
 875                                 xfrm_audit_log(audit_info->loginuid,
 876                                                audit_info->secid,
 877                                                AUDIT_MAC_IPSEC_DELSPD, 1,
 878                                                pol, NULL);
 879
 880                                 xfrm_policy_kill(pol);
 881                                 killed++;
 882
 883                                 write_lock_bh(&xfrm_policy_lock);
 884                                 goto again2;
 885                         }
 886                 }
 887
 888                 xfrm_policy_count[dir] -= killed;
 889         }
 890         atomic_inc(&flow_cache_genid);
 891         write_unlock_bh(&xfrm_policy_lock);
 892 }
 893 EXPORT_SYMBOL(xfrm_policy_flush);
 894
 895 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 896                      void *data)
 897 {
 898         struct xfrm_policy *pol, *last = NULL;
 899         struct hlist_node *entry;
 900         int dir, last_dir = 0, count, error;
 901
 902         read_lock_bh(&xfrm_policy_lock);
 903         count = 0;
 904
 905         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 906                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 907                 int i;
 908
 909                 hlist_for_each_entry(pol, entry,
 910                                      &xfrm_policy_inexact[dir], bydst) {
 911                         if (pol->type != type)
 912                                 continue;
 913                         if (last) {
 914                                 error = func(last, last_dir % XFRM_POLICY_MAX,
 915                                              count, data);
 916                                 if (error)
 917                                         goto out;
 918                         }
 919                         last = pol;
 920                         last_dir = dir;
 921                         count++;
 922                 }
 923                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 924                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 925                                 if (pol->type != type)
 926                                         continue;
 927                                 if (last) {
 928                                         error = func(last, last_dir % XFRM_POLICY_MAX,
 929                                                      count, data);
 930                                         if (error)
 931                                                 goto out;
 932                                 }
 933                                 last = pol;
 934                                 last_dir = dir;
 935                                 count++;
 936                         }
 937                 }
 938         }
 939         if (count == 0) {
 940                 error = -ENOENT;
 941                 goto out;
 942         }
 943         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
 944 out:
 945         read_unlock_bh(&xfrm_policy_lock);
 946         return error;
 947 }
 948 EXPORT_SYMBOL(xfrm_policy_walk);
 949
 950 /*
 951  * Find policy to apply to this flow.
 952  *
 953  * Returns 0 if policy found, else an -errno.
 954  */
 955 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 956                              u8 type, u16 family, int dir)
 957 {
 958         struct xfrm_selector *sel = &pol->selector;
 959         int match, ret = -ESRCH;
 960
 961         if (pol->family != family ||
 962             pol->type != type)
 963                 return ret;
 964
 965         match = xfrm_selector_match(sel, fl, family);
 966         if (match)
 967                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
 968
 969         return ret;
 970 }
 971
 972 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 973                                                      u16 family, u8 dir)
 974 {
 975         int err;
 976         struct xfrm_policy *pol, *ret;
 977         xfrm_address_t *daddr, *saddr;
 978         struct hlist_node *entry;
 979         struct hlist_head *chain;
 980         u32 priority = ~0U;
 981
 982         daddr = xfrm_flowi_daddr(fl, family);
 983         saddr = xfrm_flowi_saddr(fl, family);
 984         if (unlikely(!daddr || !saddr))
 985                 return NULL;
 986
 987         read_lock_bh(&xfrm_policy_lock);
 988         chain = policy_hash_direct(daddr, saddr, family, dir);
 989         ret = NULL;
 990         hlist_for_each_entry(pol, entry, chain, bydst) {
 991                 err = xfrm_policy_match(pol, fl, type, family, dir);
 992                 if (err) {
 993                         if (err == -ESRCH)
 994                                 continue;
 995                         else {
 996                                 ret = ERR_PTR(err);
 997                                 goto fail;
 998                         }
 999                 } else {
1000                         ret = pol;
1001                         priority = ret->priority;
1002                         break;
1003                 }
1004         }
1005         chain = &xfrm_policy_inexact[dir];
1006         hlist_for_each_entry(pol, entry, chain, bydst) {
1007                 err = xfrm_policy_match(pol, fl, type, family, dir);
1008                 if (err) {
1009                         if (err == -ESRCH)
1010                                 continue;
1011                         else {
1012                                 ret = ERR_PTR(err);
1013                                 goto fail;
1014                         }
1015                 } else if (pol->priority < priority) {
1016                         ret = pol;
1017                         break;
1018                 }
1019         }
1020         if (ret)
1021                 xfrm_pol_hold(ret);
1022 fail:
1023         read_unlock_bh(&xfrm_policy_lock);
1024
1025         return ret;
1026 }
1027
1028 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1029                                void **objp, atomic_t **obj_refp)
1030 {
1031         struct xfrm_policy *pol;
1032         int err = 0;
1033
1034 #ifdef CONFIG_XFRM_SUB_POLICY
1035         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1036         if (IS_ERR(pol)) {
1037                 err = PTR_ERR(pol);
1038                 pol = NULL;
1039         }
1040         if (pol || err)
1041                 goto end;
1042 #endif
1043         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1044         if (IS_ERR(pol)) {
1045                 err = PTR_ERR(pol);
1046                 pol = NULL;
1047         }
1048 #ifdef CONFIG_XFRM_SUB_POLICY
1049 end:
1050 #endif
1051         if ((*objp = (void *) pol) != NULL)
1052                 *obj_refp = &pol->refcnt;
1053         return err;
1054 }
1055
1056 static inline int policy_to_flow_dir(int dir)
1057 {
1058         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1059             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1060             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1061                 return dir;
1062         switch (dir) {
1063         default:
1064         case XFRM_POLICY_IN:
1065                 return FLOW_DIR_IN;
1066         case XFRM_POLICY_OUT:
1067                 return FLOW_DIR_OUT;
1068         case XFRM_POLICY_FWD:
1069                 return FLOW_DIR_FWD;
1070         }
1071 }
1072
1073 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1074 {
1075         struct xfrm_policy *pol;
1076
1077         read_lock_bh(&xfrm_policy_lock);
1078         if ((pol = sk->sk_policy[dir]) != NULL) {
1079                 int match = xfrm_selector_match(&pol->selector, fl,
1080                                                 sk->sk_family);
1081                 int err = 0;
1082
1083                 if (match) {
1084                         err = security_xfrm_policy_lookup(pol, fl->secid,
1085                                         policy_to_flow_dir(dir));
1086                         if (!err)
1087                                 xfrm_pol_hold(pol);
1088                         else if (err == -ESRCH)
1089                                 pol = NULL;
1090                         else
1091                                 pol = ERR_PTR(err);
1092                 } else
1093                         pol = NULL;
1094         }
1095         read_unlock_bh(&xfrm_policy_lock);
1096         return pol;
1097 }
1098
1099 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1100 {
1101         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1102                                                      pol->family, dir);
1103
1104         hlist_add_head(&pol->bydst, chain);
1105         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1106         xfrm_policy_count[dir]++;
1107         xfrm_pol_hold(pol);
1108
1109         if (xfrm_bydst_should_resize(dir, NULL))
1110                 schedule_work(&xfrm_hash_work);
1111 }
1112
1113 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1114                                                 int dir)
1115 {
1116         if (hlist_unhashed(&pol->bydst))
1117                 return NULL;
1118
1119         hlist_del(&pol->bydst);
1120         hlist_del(&pol->byidx);
1121         xfrm_policy_count[dir]--;
1122
1123         return pol;
1124 }
1125
1126 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1127 {
1128         write_lock_bh(&xfrm_policy_lock);
1129         pol = __xfrm_policy_unlink(pol, dir);
1130         write_unlock_bh(&xfrm_policy_lock);
1131         if (pol) {
1132                 if (dir < XFRM_POLICY_MAX)
1133                         atomic_inc(&flow_cache_genid);
1134                 xfrm_policy_kill(pol);
1135                 return 0;
1136         }
1137         return -ENOENT;
1138 }
1139 EXPORT_SYMBOL(xfrm_policy_delete);
1140
1141 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1142 {
1143         struct xfrm_policy *old_pol;
1144
1145 #ifdef CONFIG_XFRM_SUB_POLICY
1146         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1147                 return -EINVAL;
1148 #endif
1149
1150         write_lock_bh(&xfrm_policy_lock);
1151         old_pol = sk->sk_policy[dir];
1152         sk->sk_policy[dir] = pol;
1153         if (pol) {
1154                 pol->curlft.add_time = get_seconds();
1155                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1156                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1157         }
1158         if (old_pol)
1159                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1160         write_unlock_bh(&xfrm_policy_lock);
1161
1162         if (old_pol) {
1163                 xfrm_policy_kill(old_pol);
1164         }
1165         return 0;
1166 }
1167
1168 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1169 {
1170         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1171
1172         if (newp) {
1173                 newp->selector = old->selector;
1174                 if (security_xfrm_policy_clone(old, newp)) {
1175                         kfree(newp);
1176                         return NULL;  /* ENOMEM */
1177                 }
1178                 newp->lft = old->lft;
1179                 newp->curlft = old->curlft;
1180                 newp->action = old->action;
1181                 newp->flags = old->flags;
1182                 newp->xfrm_nr = old->xfrm_nr;
1183                 newp->index = old->index;
1184                 newp->type = old->type;
1185                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1186                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1187                 write_lock_bh(&xfrm_policy_lock);
1188                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1189                 write_unlock_bh(&xfrm_policy_lock);
1190                 xfrm_pol_put(newp);
1191         }
1192         return newp;
1193 }
1194
1195 int __xfrm_sk_clone_policy(struct sock *sk)
1196 {
1197         struct xfrm_policy *p0 = sk->sk_policy[0],
1198                            *p1 = sk->sk_policy[1];
1199
1200         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1201         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1202                 return -ENOMEM;
1203         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1204                 return -ENOMEM;
1205         return 0;
1206 }
1207
1208 static int
1209 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1210                unsigned short family)
1211 {
1212         int err;
1213         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1214
1215         if (unlikely(afinfo == NULL))
1216                 return -EINVAL;
1217         err = afinfo->get_saddr(local, remote);
1218         xfrm_policy_put_afinfo(afinfo);
1219         return err;
1220 }
1221
1222 /* Resolve list of templates for the flow, given policy. */
1223
1224 static int
1225 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1226                       struct xfrm_state **xfrm,
1227                       unsigned short family)
1228 {
1229         int nx;
1230         int i, error;
1231         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1232         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1233         xfrm_address_t tmp;
1234
1235         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1236                 struct xfrm_state *x;
1237                 xfrm_address_t *remote = daddr;
1238                 xfrm_address_t *local  = saddr;
1239                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1240
1241                 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1242                         remote = &tmpl->id.daddr;
1243                         local = &tmpl->saddr;
1244                         family = tmpl->encap_family;
1245                         if (xfrm_addr_any(local, family)) {
1246                                 error = xfrm_get_saddr(&tmp, remote, family);
1247                                 if (error)
1248                                         goto fail;
1249                                 local = &tmp;
1250                         }
1251                 }
1252
1253                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1254
1255                 if (x && x->km.state == XFRM_STATE_VALID) {
1256                         xfrm[nx++] = x;
1257                         daddr = remote;
1258                         saddr = local;
1259                         continue;
1260                 }
1261                 if (x) {
1262                         error = (x->km.state == XFRM_STATE_ERROR ?
1263                                  -EINVAL : -EAGAIN);
1264                         xfrm_state_put(x);
1265                 }
1266
1267                 if (!tmpl->optional)
1268                         goto fail;
1269         }
1270         return nx;
1271
1272 fail:
1273         for (nx--; nx>=0; nx--)
1274                 xfrm_state_put(xfrm[nx]);
1275         return error;
1276 }
1277
1278 static int
1279 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1280                   struct xfrm_state **xfrm,
1281                   unsigned short family)
1282 {
1283         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1284         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1285         int cnx = 0;
1286         int error;
1287         int ret;
1288         int i;
1289
1290         for (i = 0; i < npols; i++) {
1291                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1292                         error = -ENOBUFS;
1293                         goto fail;
1294                 }
1295
1296                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1297                 if (ret < 0) {
1298                         error = ret;
1299                         goto fail;
1300                 } else
1301                         cnx += ret;
1302         }
1303
1304         /* found states are sorted for outbound processing */
1305         if (npols > 1)
1306                 xfrm_state_sort(xfrm, tpp, cnx, family);
1307
1308         return cnx;
1309
1310  fail:
1311         for (cnx--; cnx>=0; cnx--)
1312                 xfrm_state_put(tpp[cnx]);
1313         return error;
1314
1315 }
1316
1317 /* Check that the bundle accepts the flow and its components are
1318  * still valid.
1319  */
1320
1321 static struct dst_entry *
1322 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1323 {
1324         struct dst_entry *x;
1325         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1326         if (unlikely(afinfo == NULL))
1327                 return ERR_PTR(-EINVAL);
1328         x = afinfo->find_bundle(fl, policy);
1329         xfrm_policy_put_afinfo(afinfo);
1330         return x;
1331 }
1332
1333 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1334  * all the metrics... Shortly, bundle a bundle.
1335  */
1336
1337 static int
1338 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1339                    struct flowi *fl, struct dst_entry **dst_p,
1340                    unsigned short family)
1341 {
1342         int err;
1343         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1344         if (unlikely(afinfo == NULL))
1345                 return -EINVAL;
1346         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1347         xfrm_policy_put_afinfo(afinfo);
1348         return err;
1349 }
1350
1351 static int inline
1352 xfrm_dst_alloc_copy(void **target, void *src, int size)
1353 {
1354         if (!*target) {
1355                 *target = kmalloc(size, GFP_ATOMIC);
1356                 if (!*target)
1357                         return -ENOMEM;
1358         }
1359         memcpy(*target, src, size);
1360         return 0;
1361 }
1362
1363 static int inline
1364 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1365 {
1366 #ifdef CONFIG_XFRM_SUB_POLICY
1367         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1368         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1369                                    sel, sizeof(*sel));
1370 #else
1371         return 0;
1372 #endif
1373 }
1374
1375 static int inline
1376 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1377 {
1378 #ifdef CONFIG_XFRM_SUB_POLICY
1379         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1380         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1381 #else
1382         return 0;
1383 #endif
1384 }
1385
1386 static int stale_bundle(struct dst_entry *dst);
1387
1388 /* Main function: finds/creates a bundle for given flow.
1389  *
1390  * At the moment we eat a raw IP route. Mostly to speed up lookups
1391  * on interfaces with disabled IPsec.
1392  */
1393 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1394                 struct sock *sk, int flags)
1395 {
1396         struct xfrm_policy *policy;
1397         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1398         int npols;
1399         int pol_dead;
1400         int xfrm_nr;
1401         int pi;
1402         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1403         struct dst_entry *dst, *dst_orig = *dst_p;
1404         int nx = 0;
1405         int err;
1406         u32 genid;
1407         u16 family;
1408         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1409
1410 restart:
1411         genid = atomic_read(&flow_cache_genid);
1412         policy = NULL;
1413         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1414                 pols[pi] = NULL;
1415         npols = 0;
1416         pol_dead = 0;
1417         xfrm_nr = 0;
1418
1419         if (sk && sk->sk_policy[1]) {
1420                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1421                 if (IS_ERR(policy))
1422                         return PTR_ERR(policy);
1423         }
1424
1425         if (!policy) {
1426                 /* To accelerate a bit...  */
1427                 if ((dst_orig->flags & DST_NOXFRM) ||
1428                     !xfrm_policy_count[XFRM_POLICY_OUT])
1429                         return 0;
1430
1431                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1432                                            dir, xfrm_policy_lookup);
1433                 if (IS_ERR(policy))
1434                         return PTR_ERR(policy);
1435         }
1436
1437         if (!policy)
1438                 return 0;
1439
1440         family = dst_orig->ops->family;
1441         policy->curlft.use_time = get_seconds();
1442         pols[0] = policy;
1443         npols ++;
1444         xfrm_nr += pols[0]->xfrm_nr;
1445
1446         switch (policy->action) {
1447         case XFRM_POLICY_BLOCK:
1448                 /* Prohibit the flow */
1449                 err = -EPERM;
1450                 goto error;
1451
1452         case XFRM_POLICY_ALLOW:
1453 #ifndef CONFIG_XFRM_SUB_POLICY
1454                 if (policy->xfrm_nr == 0) {
1455                         /* Flow passes not transformed. */
1456                         xfrm_pol_put(policy);
1457                         return 0;
1458                 }
1459 #endif
1460
1461                 /* Try to find matching bundle.
1462                  *
1463                  * LATER: help from flow cache. It is optional, this
1464                  * is required only for output policy.
1465                  */
1466                 dst = xfrm_find_bundle(fl, policy, family);
1467                 if (IS_ERR(dst)) {
1468                         err = PTR_ERR(dst);
1469                         goto error;
1470                 }
1471
1472                 if (dst)
1473                         break;
1474
1475 #ifdef CONFIG_XFRM_SUB_POLICY
1476                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1477                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1478                                                             fl, family,
1479                                                             XFRM_POLICY_OUT);
1480                         if (pols[1]) {
1481                                 if (IS_ERR(pols[1])) {
1482                                         err = PTR_ERR(pols[1]);
1483                                         goto error;
1484                                 }
1485                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1486                                         err = -EPERM;
1487                                         goto error;
1488                                 }
1489                                 npols ++;
1490                                 xfrm_nr += pols[1]->xfrm_nr;
1491                         }
1492                 }
1493
1494                 /*
1495                  * Because neither flowi nor bundle information knows about
1496                  * transformation template size. On more than one policy usage
1497                  * we can realize whether all of them is bypass or not after
1498                  * they are searched. See above not-transformed bypass
1499                  * is surrounded by non-sub policy configuration, too.
1500                  */
1501                 if (xfrm_nr == 0) {
1502                         /* Flow passes not transformed. */
1503                         xfrm_pols_put(pols, npols);
1504                         return 0;
1505                 }
1506
1507 #endif
1508                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1509
1510                 if (unlikely(nx<0)) {
1511                         err = nx;
1512                         if (err == -EAGAIN && flags) {
1513                                 DECLARE_WAITQUEUE(wait, current);
1514
1515                                 add_wait_queue(&km_waitq, &wait);
1516                                 set_current_state(TASK_INTERRUPTIBLE);
1517                                 schedule();
1518                                 set_current_state(TASK_RUNNING);
1519                                 remove_wait_queue(&km_waitq, &wait);
1520
1521                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1522
1523                                 if (nx == -EAGAIN && signal_pending(current)) {
1524                                         err = -ERESTART;
1525                                         goto error;
1526                                 }
1527                                 if (nx == -EAGAIN ||
1528                                     genid != atomic_read(&flow_cache_genid)) {
1529                                         xfrm_pols_put(pols, npols);
1530                                         goto restart;
1531                                 }
1532                                 err = nx;
1533                         }
1534                         if (err < 0)
1535                                 goto error;
1536                 }
1537                 if (nx == 0) {
1538                         /* Flow passes not transformed. */
1539                         xfrm_pols_put(pols, npols);
1540                         return 0;
1541                 }
1542
1543                 dst = dst_orig;
1544                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1545
1546                 if (unlikely(err)) {
1547                         int i;
1548                         for (i=0; i<nx; i++)
1549                                 xfrm_state_put(xfrm[i]);
1550                         goto error;
1551                 }
1552
1553                 for (pi = 0; pi < npols; pi++) {
1554                         read_lock_bh(&pols[pi]->lock);
1555                         pol_dead |= pols[pi]->dead;
1556                         read_unlock_bh(&pols[pi]->lock);
1557                 }
1558
1559                 write_lock_bh(&policy->lock);
1560                 if (unlikely(pol_dead || stale_bundle(dst))) {
1561                         /* Wow! While we worked on resolving, this
1562                          * policy has gone. Retry. It is not paranoia,
1563                          * we just cannot enlist new bundle to dead object.
1564                          * We can't enlist stable bundles either.
1565                          */
1566                         write_unlock_bh(&policy->lock);
1567                         if (dst)
1568                                 dst_free(dst);
1569
1570                         err = -EHOSTUNREACH;
1571                         goto error;
1572                 }
1573
1574                 if (npols > 1)
1575                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1576                 else
1577                         err = xfrm_dst_update_origin(dst, fl);
1578                 if (unlikely(err)) {
1579                         write_unlock_bh(&policy->lock);
1580                         if (dst)
1581                                 dst_free(dst);
1582                         goto error;
1583                 }
1584
1585                 dst->next = policy->bundles;
1586                 policy->bundles = dst;
1587                 dst_hold(dst);
1588                 write_unlock_bh(&policy->lock);
1589         }
1590         *dst_p = dst;
1591         dst_release(dst_orig);
1592         xfrm_pols_put(pols, npols);
1593         return 0;
1594
1595 error:
1596         dst_release(dst_orig);
1597         xfrm_pols_put(pols, npols);
1598         *dst_p = NULL;
1599         return err;
1600 }
1601 EXPORT_SYMBOL(xfrm_lookup);
1602
1603 static inline int
1604 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1605 {
1606         struct xfrm_state *x;
1607         int err;
1608
1609         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1610                 return 0;
1611         x = skb->sp->xvec[idx];
1612         if (!x->type->reject)
1613                 return 0;
1614         xfrm_state_hold(x);
1615         err = x->type->reject(x, skb, fl);
1616         xfrm_state_put(x);
1617         return err;
1618 }
1619
1620 /* When skb is transformed back to its "native" form, we have to
1621  * check policy restrictions. At the moment we make this in maximally
1622  * stupid way. Shame on me. :-) Of course, connected sockets must
1623  * have policy cached at them.
1624  */
1625
1626 static inline int
1627 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1628               unsigned short family)
1629 {
1630         if (xfrm_state_kern(x))
1631                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1632         return  x->id.proto == tmpl->id.proto &&
1633                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1634                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1635                 x->props.mode == tmpl->mode &&
1636                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1637                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1638                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1639                   xfrm_state_addr_cmp(tmpl, x, family));
1640 }
1641
1642 /*
1643  * 0 or more than 0 is returned when validation is succeeded (either bypass
1644  * because of optional transport mode, or next index of the mathced secpath
1645  * state with the template.
1646  * -1 is returned when no matching template is found.
1647  * Otherwise "-2 - errored_index" is returned.
1648  */
1649 static inline int
1650 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1651                unsigned short family)
1652 {
1653         int idx = start;
1654
1655         if (tmpl->optional) {
1656                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1657                         return start;
1658         } else
1659                 start = -1;
1660         for (; idx < sp->len; idx++) {
1661                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1662                         return ++idx;
1663                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1664                         if (start == -1)
1665                                 start = -2-idx;
1666                         break;
1667                 }
1668         }
1669         return start;
1670 }
1671
1672 int
1673 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1674 {
1675         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1676         int err;
1677
1678         if (unlikely(afinfo == NULL))
1679                 return -EAFNOSUPPORT;
1680
1681         afinfo->decode_session(skb, fl);
1682         err = security_xfrm_decode_session(skb, &fl->secid);
1683         xfrm_policy_put_afinfo(afinfo);
1684         return err;
1685 }
1686 EXPORT_SYMBOL(xfrm_decode_session);
1687
1688 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1689 {
1690         for (; k < sp->len; k++) {
1691                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1692                         *idxp = k;
1693                         return 1;
1694                 }
1695         }
1696
1697         return 0;
1698 }
1699
1700 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1701                         unsigned short family)
1702 {
1703         struct xfrm_policy *pol;
1704         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1705         int npols = 0;
1706         int xfrm_nr;
1707         int pi;
1708         struct flowi fl;
1709         u8 fl_dir = policy_to_flow_dir(dir);
1710         int xerr_idx = -1;
1711
1712         if (xfrm_decode_session(skb, &fl, family) < 0)
1713                 return 0;
1714         nf_nat_decode_session(skb, &fl, family);
1715
1716         /* First, check used SA against their selectors. */
1717         if (skb->sp) {
1718                 int i;
1719
1720                 for (i=skb->sp->len-1; i>=0; i--) {
1721                         struct xfrm_state *x = skb->sp->xvec[i];
1722                         if (!xfrm_selector_match(&x->sel, &fl, family))
1723                                 return 0;
1724                 }
1725         }
1726
1727         pol = NULL;
1728         if (sk && sk->sk_policy[dir]) {
1729                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1730                 if (IS_ERR(pol))
1731                         return 0;
1732         }
1733
1734         if (!pol)
1735                 pol = flow_cache_lookup(&fl, family, fl_dir,
1736                                         xfrm_policy_lookup);
1737
1738         if (IS_ERR(pol))
1739                 return 0;
1740
1741         if (!pol) {
1742                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1743                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1744                         return 0;
1745                 }
1746                 return 1;
1747         }
1748
1749         pol->curlft.use_time = get_seconds();
1750
1751         pols[0] = pol;
1752         npols ++;
1753 #ifdef CONFIG_XFRM_SUB_POLICY
1754         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1755                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1756                                                     &fl, family,
1757                                                     XFRM_POLICY_IN);
1758                 if (pols[1]) {
1759                         if (IS_ERR(pols[1]))
1760                                 return 0;
1761                         pols[1]->curlft.use_time = get_seconds();
1762                         npols ++;
1763                 }
1764         }
1765 #endif
1766
1767         if (pol->action == XFRM_POLICY_ALLOW) {
1768                 struct sec_path *sp;
1769                 static struct sec_path dummy;
1770                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1771                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1772                 struct xfrm_tmpl **tpp = tp;
1773                 int ti = 0;
1774                 int i, k;
1775
1776                 if ((sp = skb->sp) == NULL)
1777                         sp = &dummy;
1778
1779                 for (pi = 0; pi < npols; pi++) {
1780                         if (pols[pi] != pol &&
1781                             pols[pi]->action != XFRM_POLICY_ALLOW)
1782                                 goto reject;
1783                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1784                                 goto reject_error;
1785                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1786                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1787                 }
1788                 xfrm_nr = ti;
1789                 if (npols > 1) {
1790                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1791                         tpp = stp;
1792                 }
1793
1794                 /* For each tunnel xfrm, find the first matching tmpl.
1795                  * For each tmpl before that, find corresponding xfrm.
1796                  * Order is _important_. Later we will implement
1797                  * some barriers, but at the moment barriers
1798                  * are implied between each two transformations.
1799                  */
1800                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1801                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1802                         if (k < 0) {
1803                                 if (k < -1)
1804                                         /* "-2 - errored_index" returned */
1805                                         xerr_idx = -(2+k);
1806                                 goto reject;
1807                         }
1808                 }
1809
1810                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1811                         goto reject;
1812
1813                 xfrm_pols_put(pols, npols);
1814                 return 1;
1815         }
1816
1817 reject:
1818         xfrm_secpath_reject(xerr_idx, skb, &fl);
1819 reject_error:
1820         xfrm_pols_put(pols, npols);
1821         return 0;
1822 }
1823 EXPORT_SYMBOL(__xfrm_policy_check);
1824
1825 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1826 {
1827         struct flowi fl;
1828
1829         if (xfrm_decode_session(skb, &fl, family) < 0)
1830                 return 0;
1831
1832         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1833 }
1834 EXPORT_SYMBOL(__xfrm_route_forward);
1835
1836 /* Optimize later using cookies and generation ids. */
1837
1838 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1839 {
1840         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1841          * to "-1" to force all XFRM destinations to get validated by
1842          * dst_ops->check on every use.  We do this because when a
1843          * normal route referenced by an XFRM dst is obsoleted we do
1844          * not go looking around for all parent referencing XFRM dsts
1845          * so that we can invalidate them.  It is just too much work.
1846          * Instead we make the checks here on every use.  For example:
1847          *
1848          *      XFRM dst A --> IPv4 dst X
1849          *
1850          * X is the "xdst->route" of A (X is also the "dst->path" of A
1851          * in this example).  If X is marked obsolete, "A" will not
1852          * notice.  That's what we are validating here via the
1853          * stale_bundle() check.
1854          *
1855          * When a policy's bundle is pruned, we dst_free() the XFRM
1856          * dst which causes it's ->obsolete field to be set to a
1857          * positive non-zero integer.  If an XFRM dst has been pruned
1858          * like this, we want to force a new route lookup.
1859          */
1860         if (dst->obsolete < 0 && !stale_bundle(dst))
1861                 return dst;
1862
1863         return NULL;
1864 }
1865
1866 static int stale_bundle(struct dst_entry *dst)
1867 {
1868         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1869 }
1870
1871 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1872 {
1873         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1874                 dst->dev = &loopback_dev;
1875                 dev_hold(&loopback_dev);
1876                 dev_put(dev);
1877         }
1878 }
1879 EXPORT_SYMBOL(xfrm_dst_ifdown);
1880
1881 static void xfrm_link_failure(struct sk_buff *skb)
1882 {
1883         /* Impossible. Such dst must be popped before reaches point of failure. */
1884         return;
1885 }
1886
1887 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1888 {
1889         if (dst) {
1890                 if (dst->obsolete) {
1891                         dst_release(dst);
1892                         dst = NULL;
1893                 }
1894         }
1895         return dst;
1896 }
1897
1898 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1899 {
1900         struct dst_entry *dst, **dstp;
1901
1902         write_lock(&pol->lock);
1903         dstp = &pol->bundles;
1904         while ((dst=*dstp) != NULL) {
1905                 if (func(dst)) {
1906                         *dstp = dst->next;
1907                         dst->next = *gc_list_p;
1908                         *gc_list_p = dst;
1909                 } else {
1910                         dstp = &dst->next;
1911                 }
1912         }
1913         write_unlock(&pol->lock);
1914 }
1915
1916 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1917 {
1918         struct dst_entry *gc_list = NULL;
1919         int dir;
1920
1921         read_lock_bh(&xfrm_policy_lock);
1922         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1923                 struct xfrm_policy *pol;
1924                 struct hlist_node *entry;
1925                 struct hlist_head *table;
1926                 int i;
1927
1928                 hlist_for_each_entry(pol, entry,
1929                                      &xfrm_policy_inexact[dir], bydst)
1930                         prune_one_bundle(pol, func, &gc_list);
1931
1932                 table = xfrm_policy_bydst[dir].table;
1933                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1934                         hlist_for_each_entry(pol, entry, table + i, bydst)
1935                                 prune_one_bundle(pol, func, &gc_list);
1936                 }
1937         }
1938         read_unlock_bh(&xfrm_policy_lock);
1939
1940         while (gc_list) {
1941                 struct dst_entry *dst = gc_list;
1942                 gc_list = dst->next;
1943                 dst_free(dst);
1944         }
1945 }
1946
1947 static int unused_bundle(struct dst_entry *dst)
1948 {
1949         return !atomic_read(&dst->__refcnt);
1950 }
1951
1952 static void __xfrm_garbage_collect(void)
1953 {
1954         xfrm_prune_bundles(unused_bundle);
1955 }
1956
1957 static int xfrm_flush_bundles(void)
1958 {
1959         xfrm_prune_bundles(stale_bundle);
1960         return 0;
1961 }
1962
1963 void xfrm_init_pmtu(struct dst_entry *dst)
1964 {
1965         do {
1966                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1967                 u32 pmtu, route_mtu_cached;
1968
1969                 pmtu = dst_mtu(dst->child);
1970                 xdst->child_mtu_cached = pmtu;
1971
1972                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1973
1974                 route_mtu_cached = dst_mtu(xdst->route);
1975                 xdst->route_mtu_cached = route_mtu_cached;
1976
1977                 if (pmtu > route_mtu_cached)
1978                         pmtu = route_mtu_cached;
1979
1980                 dst->metrics[RTAX_MTU-1] = pmtu;
1981         } while ((dst = dst->next));
1982 }
1983
1984 EXPORT_SYMBOL(xfrm_init_pmtu);
1985
1986 /* Check that the bundle accepts the flow and its components are
1987  * still valid.
1988  */
1989
1990 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
1991                 struct flowi *fl, int family, int strict)
1992 {
1993         struct dst_entry *dst = &first->u.dst;
1994         struct xfrm_dst *last;
1995         u32 mtu;
1996
1997         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1998             (dst->dev && !netif_running(dst->dev)))
1999                 return 0;
2000 #ifdef CONFIG_XFRM_SUB_POLICY
2001         if (fl) {
2002                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2003                         return 0;
2004                 if (first->partner &&
2005                     !xfrm_selector_match(first->partner, fl, family))
2006                         return 0;
2007         }
2008 #endif
2009
2010         last = NULL;
2011
2012         do {
2013                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2014
2015                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2016                         return 0;
2017                 if (fl && pol &&
2018                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2019                         return 0;
2020                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2021                         return 0;
2022                 if (xdst->genid != dst->xfrm->genid)
2023                         return 0;
2024
2025                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
2026                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2027                         return 0;
2028
2029                 mtu = dst_mtu(dst->child);
2030                 if (xdst->child_mtu_cached != mtu) {
2031                         last = xdst;
2032                         xdst->child_mtu_cached = mtu;
2033                 }
2034
2035                 if (!dst_check(xdst->route, xdst->route_cookie))
2036                         return 0;
2037                 mtu = dst_mtu(xdst->route);
2038                 if (xdst->route_mtu_cached != mtu) {
2039                         last = xdst;
2040                         xdst->route_mtu_cached = mtu;
2041                 }
2042
2043                 dst = dst->child;
2044         } while (dst->xfrm);
2045
2046         if (likely(!last))
2047                 return 1;
2048
2049         mtu = last->child_mtu_cached;
2050         for (;;) {
2051                 dst = &last->u.dst;
2052
2053                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2054                 if (mtu > last->route_mtu_cached)
2055                         mtu = last->route_mtu_cached;
2056                 dst->metrics[RTAX_MTU-1] = mtu;
2057
2058                 if (last == first)
2059                         break;
2060
2061                 last = last->u.next;
2062                 last->child_mtu_cached = mtu;
2063         }
2064
2065         return 1;
2066 }
2067
2068 EXPORT_SYMBOL(xfrm_bundle_ok);
2069
2070 #ifdef CONFIG_AUDITSYSCALL
2071 /* Audit addition and deletion of SAs and ipsec policy */
2072
2073 void xfrm_audit_log(uid_t auid, u32 sid, int type, int result,
2074                     struct xfrm_policy *xp, struct xfrm_state *x)
2075 {
2076
2077         char *secctx;
2078         u32 secctx_len;
2079         struct xfrm_sec_ctx *sctx = NULL;
2080         struct audit_buffer *audit_buf;
2081         int family;
2082         extern int audit_enabled;
2083
2084         if (audit_enabled == 0)
2085                 return;
2086
2087         BUG_ON((type == AUDIT_MAC_IPSEC_ADDSA ||
2088                 type == AUDIT_MAC_IPSEC_DELSA) && !x);
2089         BUG_ON((type == AUDIT_MAC_IPSEC_ADDSPD ||
2090                 type == AUDIT_MAC_IPSEC_DELSPD) && !xp);
2091
2092         audit_buf = audit_log_start(current->audit_context, GFP_ATOMIC, type);
2093         if (audit_buf == NULL)
2094                 return;
2095
2096         switch(type) {
2097         case AUDIT_MAC_IPSEC_ADDSA:
2098                 audit_log_format(audit_buf, "SAD add: auid=%u", auid);
2099                 break;
2100         case AUDIT_MAC_IPSEC_DELSA:
2101                 audit_log_format(audit_buf, "SAD delete: auid=%u", auid);
2102                 break;
2103         case AUDIT_MAC_IPSEC_ADDSPD:
2104                 audit_log_format(audit_buf, "SPD add: auid=%u", auid);
2105                 break;
2106         case AUDIT_MAC_IPSEC_DELSPD:
2107                 audit_log_format(audit_buf, "SPD delete: auid=%u", auid);
2108                 break;
2109         default:
2110                 return;
2111         }
2112
2113         if (sid != 0 &&
2114                 security_secid_to_secctx(sid, &secctx, &secctx_len) == 0)
2115                 audit_log_format(audit_buf, " subj=%s", secctx);
2116         else
2117                 audit_log_task_context(audit_buf);
2118
2119         if (xp) {
2120                 family = xp->selector.family;
2121                 if (xp->security)
2122                         sctx = xp->security;
2123         } else {
2124                 family = x->props.family;
2125                 if (x->security)
2126                         sctx = x->security;
2127         }
2128
2129         if (sctx)
2130                 audit_log_format(audit_buf,
2131                                 " sec_alg=%u sec_doi=%u sec_obj=%s",
2132                                 sctx->ctx_alg, sctx->ctx_doi, sctx->ctx_str);
2133
2134         switch(family) {
2135         case AF_INET:
2136                 {
2137                         struct in_addr saddr, daddr;
2138                         if (xp) {
2139                                 saddr.s_addr = xp->selector.saddr.a4;
2140                                 daddr.s_addr = xp->selector.daddr.a4;
2141                         } else {
2142                                 saddr.s_addr = x->props.saddr.a4;
2143                                 daddr.s_addr = x->id.daddr.a4;
2144                         }
2145                         audit_log_format(audit_buf,
2146                                          " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2147                                          NIPQUAD(saddr), NIPQUAD(daddr));
2148                 }
2149                         break;
2150         case AF_INET6:
2151                 {
2152                         struct in6_addr saddr6, daddr6;
2153                         if (xp) {
2154                                 memcpy(&saddr6, xp->selector.saddr.a6,
2155                                         sizeof(struct in6_addr));
2156                                 memcpy(&daddr6, xp->selector.daddr.a6,
2157                                         sizeof(struct in6_addr));
2158                         } else {
2159                                 memcpy(&saddr6, x->props.saddr.a6,
2160                                         sizeof(struct in6_addr));
2161                                 memcpy(&daddr6, x->id.daddr.a6,
2162                                         sizeof(struct in6_addr));
2163                         }
2164                         audit_log_format(audit_buf,
2165                                          " src=" NIP6_FMT " dst=" NIP6_FMT,
2166                                          NIP6(saddr6), NIP6(daddr6));
2167                 }
2168                 break;
2169         }
2170
2171         if (x)
2172                 audit_log_format(audit_buf, " spi=%lu(0x%lx) protocol=%s",
2173                                 (unsigned long)ntohl(x->id.spi),
2174                                 (unsigned long)ntohl(x->id.spi),
2175                                 x->id.proto == IPPROTO_AH ? "AH" :
2176                                 (x->id.proto == IPPROTO_ESP ?
2177                                 "ESP" : "IPCOMP"));
2178
2179         audit_log_format(audit_buf, " res=%u", result);
2180         audit_log_end(audit_buf);
2181 }
2182
2183 EXPORT_SYMBOL(xfrm_audit_log);
2184 #endif /* CONFIG_AUDITSYSCALL */
2185
2186 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2187 {
2188         int err = 0;
2189         if (unlikely(afinfo == NULL))
2190                 return -EINVAL;
2191         if (unlikely(afinfo->family >= NPROTO))
2192                 return -EAFNOSUPPORT;
2193         write_lock_bh(&xfrm_policy_afinfo_lock);
2194         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2195                 err = -ENOBUFS;
2196         else {
2197                 struct dst_ops *dst_ops = afinfo->dst_ops;
2198                 if (likely(dst_ops->kmem_cachep == NULL))
2199                         dst_ops->kmem_cachep = xfrm_dst_cache;
2200                 if (likely(dst_ops->check == NULL))
2201                         dst_ops->check = xfrm_dst_check;
2202                 if (likely(dst_ops->negative_advice == NULL))
2203                         dst_ops->negative_advice = xfrm_negative_advice;
2204                 if (likely(dst_ops->link_failure == NULL))
2205                         dst_ops->link_failure = xfrm_link_failure;
2206                 if (likely(afinfo->garbage_collect == NULL))
2207                         afinfo->garbage_collect = __xfrm_garbage_collect;
2208                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2209         }
2210         write_unlock_bh(&xfrm_policy_afinfo_lock);
2211         return err;
2212 }
2213 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2214
2215 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2216 {
2217         int err = 0;
2218         if (unlikely(afinfo == NULL))
2219                 return -EINVAL;
2220         if (unlikely(afinfo->family >= NPROTO))
2221                 return -EAFNOSUPPORT;
2222         write_lock_bh(&xfrm_policy_afinfo_lock);
2223         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2224                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2225                         err = -EINVAL;
2226                 else {
2227                         struct dst_ops *dst_ops = afinfo->dst_ops;
2228                         xfrm_policy_afinfo[afinfo->family] = NULL;
2229                         dst_ops->kmem_cachep = NULL;
2230                         dst_ops->check = NULL;
2231                         dst_ops->negative_advice = NULL;
2232                         dst_ops->link_failure = NULL;
2233                         afinfo->garbage_collect = NULL;
2234                 }
2235         }
2236         write_unlock_bh(&xfrm_policy_afinfo_lock);
2237         return err;
2238 }
2239 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2240
2241 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2242 {
2243         struct xfrm_policy_afinfo *afinfo;
2244         if (unlikely(family >= NPROTO))
2245                 return NULL;
2246         read_lock(&xfrm_policy_afinfo_lock);
2247         afinfo = xfrm_policy_afinfo[family];
2248         if (unlikely(!afinfo))
2249                 read_unlock(&xfrm_policy_afinfo_lock);
2250         return afinfo;
2251 }
2252
2253 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2254 {
2255         read_unlock(&xfrm_policy_afinfo_lock);
2256 }
2257
2258 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2259 {
2260         struct xfrm_policy_afinfo *afinfo;
2261         if (unlikely(family >= NPROTO))
2262                 return NULL;
2263         write_lock_bh(&xfrm_policy_afinfo_lock);
2264         afinfo = xfrm_policy_afinfo[family];
2265         if (unlikely(!afinfo))
2266                 write_unlock_bh(&xfrm_policy_afinfo_lock);
2267         return afinfo;
2268 }
2269
2270 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2271 {
2272         write_unlock_bh(&xfrm_policy_afinfo_lock);
2273 }
2274
2275 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2276 {
2277         switch (event) {
2278         case NETDEV_DOWN:
2279                 xfrm_flush_bundles();
2280         }
2281         return NOTIFY_DONE;
2282 }
2283
2284 static struct notifier_block xfrm_dev_notifier = {
2285         xfrm_dev_event,
2286         NULL,
2287         0
2288 };
2289
2290 static void __init xfrm_policy_init(void)
2291 {
2292         unsigned int hmask, sz;
2293         int dir;
2294
2295         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2296                                            sizeof(struct xfrm_dst),
2297                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2298                                            NULL, NULL);
2299
2300         hmask = 8 - 1;
2301         sz = (hmask+1) * sizeof(struct hlist_head);
2302
2303         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2304         xfrm_idx_hmask = hmask;
2305         if (!xfrm_policy_byidx)
2306                 panic("XFRM: failed to allocate byidx hash\n");
2307
2308         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2309                 struct xfrm_policy_hash *htab;
2310
2311                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2312
2313                 htab = &xfrm_policy_bydst[dir];
2314                 htab->table = xfrm_hash_alloc(sz);
2315                 htab->hmask = hmask;
2316                 if (!htab->table)
2317                         panic("XFRM: failed to allocate bydst hash\n");
2318         }
2319
2320         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2321         register_netdevice_notifier(&xfrm_dev_notifier);
2322 }
2323
2324 void __init xfrm_init(void)
2325 {
2326         xfrm_state_init();
2327         xfrm_policy_init();
2328         xfrm_input_init();
2329 }
2330
2331 #ifdef CONFIG_XFRM_MIGRATE
2332 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2333                                        struct xfrm_selector *sel_tgt)
2334 {
2335         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2336                 if (sel_tgt->family == sel_cmp->family &&
2337                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2338                                   sel_cmp->family) == 0 &&
2339                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2340                                   sel_cmp->family) == 0 &&
2341                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2342                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2343                         return 1;
2344                 }
2345         } else {
2346                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2347                         return 1;
2348                 }
2349         }
2350         return 0;
2351 }
2352
2353 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2354                                                      u8 dir, u8 type)
2355 {
2356         struct xfrm_policy *pol, *ret = NULL;
2357         struct hlist_node *entry;
2358         struct hlist_head *chain;
2359         u32 priority = ~0U;
2360
2361         read_lock_bh(&xfrm_policy_lock);
2362         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2363         hlist_for_each_entry(pol, entry, chain, bydst) {
2364                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2365                     pol->type == type) {
2366                         ret = pol;
2367                         priority = ret->priority;
2368                         break;
2369                 }
2370         }
2371         chain = &xfrm_policy_inexact[dir];
2372         hlist_for_each_entry(pol, entry, chain, bydst) {
2373                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2374                     pol->type == type &&
2375                     pol->priority < priority) {
2376                         ret = pol;
2377                         break;
2378                 }
2379         }
2380
2381         if (ret)
2382                 xfrm_pol_hold(ret);
2383
2384         read_unlock_bh(&xfrm_policy_lock);
2385
2386         return ret;
2387 }
2388
2389 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2390 {
2391         int match = 0;
2392
2393         if (t->mode == m->mode && t->id.proto == m->proto &&
2394             (m->reqid == 0 || t->reqid == m->reqid)) {
2395                 switch (t->mode) {
2396                 case XFRM_MODE_TUNNEL:
2397                 case XFRM_MODE_BEET:
2398                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2399                                           m->old_family) == 0 &&
2400                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2401                                           m->old_family) == 0) {
2402                                 match = 1;
2403                         }
2404                         break;
2405                 case XFRM_MODE_TRANSPORT:
2406                         /* in case of transport mode, template does not store
2407                            any IP addresses, hence we just compare mode and
2408                            protocol */
2409                         match = 1;
2410                         break;
2411                 default:
2412                         break;
2413                 }
2414         }
2415         return match;
2416 }
2417
2418 /* update endpoint address(es) of template(s) */
2419 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2420                                struct xfrm_migrate *m, int num_migrate)
2421 {
2422         struct xfrm_migrate *mp;
2423         struct dst_entry *dst;
2424         int i, j, n = 0;
2425
2426         write_lock_bh(&pol->lock);
2427         if (unlikely(pol->dead)) {
2428                 /* target policy has been deleted */
2429                 write_unlock_bh(&pol->lock);
2430                 return -ENOENT;
2431         }
2432
2433         for (i = 0; i < pol->xfrm_nr; i++) {
2434                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2435                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2436                                 continue;
2437                         n++;
2438                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL)
2439                                 continue;
2440                         /* update endpoints */
2441                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2442                                sizeof(pol->xfrm_vec[i].id.daddr));
2443                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2444                                sizeof(pol->xfrm_vec[i].saddr));
2445                         pol->xfrm_vec[i].encap_family = mp->new_family;
2446                         /* flush bundles */
2447                         while ((dst = pol->bundles) != NULL) {
2448                                 pol->bundles = dst->next;
2449                                 dst_free(dst);
2450                         }
2451                 }
2452         }
2453
2454         write_unlock_bh(&pol->lock);
2455
2456         if (!n)
2457                 return -ENODATA;
2458
2459         return 0;
2460 }
2461
2462 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2463 {
2464         int i, j;
2465
2466         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2467                 return -EINVAL;
2468
2469         for (i = 0; i < num_migrate; i++) {
2470                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2471                                    m[i].old_family) == 0) &&
2472                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2473                                    m[i].old_family) == 0))
2474                         return -EINVAL;
2475                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2476                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2477                         return -EINVAL;
2478
2479                 /* check if there is any duplicated entry */
2480                 for (j = i + 1; j < num_migrate; j++) {
2481                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2482                                     sizeof(m[i].old_daddr)) &&
2483                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2484                                     sizeof(m[i].old_saddr)) &&
2485                             m[i].proto == m[j].proto &&
2486                             m[i].mode == m[j].mode &&
2487                             m[i].reqid == m[j].reqid &&
2488                             m[i].old_family == m[j].old_family)
2489                                 return -EINVAL;
2490                 }
2491         }
2492
2493         return 0;
2494 }
2495
2496 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2497                  struct xfrm_migrate *m, int num_migrate)
2498 {
2499         int i, err, nx_cur = 0, nx_new = 0;
2500         struct xfrm_policy *pol = NULL;
2501         struct xfrm_state *x, *xc;
2502         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2503         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2504         struct xfrm_migrate *mp;
2505
2506         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2507                 goto out;
2508
2509         /* Stage 1 - find policy */
2510         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2511                 err = -ENOENT;
2512                 goto out;
2513         }
2514
2515         /* Stage 2 - find and update state(s) */
2516         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2517                 if ((x = xfrm_migrate_state_find(mp))) {
2518                         x_cur[nx_cur] = x;
2519                         nx_cur++;
2520                         if ((xc = xfrm_state_migrate(x, mp))) {
2521                                 x_new[nx_new] = xc;
2522                                 nx_new++;
2523                         } else {
2524                                 err = -ENODATA;
2525                                 goto restore_state;
2526                         }
2527                 }
2528         }
2529
2530         /* Stage 3 - update policy */
2531         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2532                 goto restore_state;
2533
2534         /* Stage 4 - delete old state(s) */
2535         if (nx_cur) {
2536                 xfrm_states_put(x_cur, nx_cur);
2537                 xfrm_states_delete(x_cur, nx_cur);
2538         }
2539
2540         /* Stage 5 - announce */
2541         km_migrate(sel, dir, type, m, num_migrate);
2542
2543         xfrm_pol_put(pol);
2544
2545         return 0;
2546 out:
2547         return err;
2548
2549 restore_state:
2550         if (pol)
2551                 xfrm_pol_put(pol);
2552         if (nx_cur)
2553                 xfrm_states_put(x_cur, nx_cur);
2554         if (nx_new)
2555                 xfrm_states_delete(x_new, nx_new);
2556
2557         return err;
2558 }
2559 EXPORT_SYMBOL(xfrm_migrate);
2560 #endif
2561