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 = (unsigned long)xtime.tv_sec;
 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 static DEFINE_MUTEX(hash_resize_mutex);
 583
 584 static void xfrm_hash_resize(struct work_struct *__unused)
 585 {
 586         int dir, total;
 587
 588         mutex_lock(&hash_resize_mutex);
 589
 590         total = 0;
 591         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 592                 if (xfrm_bydst_should_resize(dir, &total))
 593                         xfrm_bydst_resize(dir);
 594         }
 595         if (xfrm_byidx_should_resize(total))
 596                 xfrm_byidx_resize(total);
 597
 598         mutex_unlock(&hash_resize_mutex);
 599 }
 600
 601 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 602
 603 /* Generate new index... KAME seems to generate them ordered by cost
 604  * of an absolute inpredictability of ordering of rules. This will not pass. */
 605 static u32 xfrm_gen_index(u8 type, int dir)
 606 {
 607         static u32 idx_generator;
 608
 609         for (;;) {
 610                 struct hlist_node *entry;
 611                 struct hlist_head *list;
 612                 struct xfrm_policy *p;
 613                 u32 idx;
 614                 int found;
 615
 616                 idx = (idx_generator | dir);
 617                 idx_generator += 8;
 618                 if (idx == 0)
 619                         idx = 8;
 620                 list = xfrm_policy_byidx + idx_hash(idx);
 621                 found = 0;
 622                 hlist_for_each_entry(p, entry, list, byidx) {
 623                         if (p->index == idx) {
 624                                 found = 1;
 625                                 break;
 626                         }
 627                 }
 628                 if (!found)
 629                         return idx;
 630         }
 631 }
 632
 633 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 634 {
 635         u32 *p1 = (u32 *) s1;
 636         u32 *p2 = (u32 *) s2;
 637         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 638         int i;
 639
 640         for (i = 0; i < len; i++) {
 641                 if (p1[i] != p2[i])
 642                         return 1;
 643         }
 644
 645         return 0;
 646 }
 647
 648 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 649 {
 650         struct xfrm_policy *pol;
 651         struct xfrm_policy *delpol;
 652         struct hlist_head *chain;
 653         struct hlist_node *entry, *newpos;
 654         struct dst_entry *gc_list;
 655
 656         write_lock_bh(&xfrm_policy_lock);
 657         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 658         delpol = NULL;
 659         newpos = NULL;
 660         hlist_for_each_entry(pol, entry, chain, bydst) {
 661                 if (pol->type == policy->type &&
 662                     !selector_cmp(&pol->selector, &policy->selector) &&
 663                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 664                     !WARN_ON(delpol)) {
 665                         if (excl) {
 666                                 write_unlock_bh(&xfrm_policy_lock);
 667                                 return -EEXIST;
 668                         }
 669                         delpol = pol;
 670                         if (policy->priority > pol->priority)
 671                                 continue;
 672                 } else if (policy->priority >= pol->priority) {
 673                         newpos = &pol->bydst;
 674                         continue;
 675                 }
 676                 if (delpol)
 677                         break;
 678         }
 679         if (newpos)
 680                 hlist_add_after(newpos, &policy->bydst);
 681         else
 682                 hlist_add_head(&policy->bydst, chain);
 683         xfrm_pol_hold(policy);
 684         xfrm_policy_count[dir]++;
 685         atomic_inc(&flow_cache_genid);
 686         if (delpol) {
 687                 hlist_del(&delpol->bydst);
 688                 hlist_del(&delpol->byidx);
 689                 xfrm_policy_count[dir]--;
 690         }
 691         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 692         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 693         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
 694         policy->curlft.use_time = 0;
 695         if (!mod_timer(&policy->timer, jiffies + HZ))
 696                 xfrm_pol_hold(policy);
 697         write_unlock_bh(&xfrm_policy_lock);
 698
 699         if (delpol)
 700                 xfrm_policy_kill(delpol);
 701         else if (xfrm_bydst_should_resize(dir, NULL))
 702                 schedule_work(&xfrm_hash_work);
 703
 704         read_lock_bh(&xfrm_policy_lock);
 705         gc_list = NULL;
 706         entry = &policy->bydst;
 707         hlist_for_each_entry_continue(policy, entry, bydst) {
 708                 struct dst_entry *dst;
 709
 710                 write_lock(&policy->lock);
 711                 dst = policy->bundles;
 712                 if (dst) {
 713                         struct dst_entry *tail = dst;
 714                         while (tail->next)
 715                                 tail = tail->next;
 716                         tail->next = gc_list;
 717                         gc_list = dst;
 718
 719                         policy->bundles = NULL;
 720                 }
 721                 write_unlock(&policy->lock);
 722         }
 723         read_unlock_bh(&xfrm_policy_lock);
 724
 725         while (gc_list) {
 726                 struct dst_entry *dst = gc_list;
 727
 728                 gc_list = dst->next;
 729                 dst_free(dst);
 730         }
 731
 732         return 0;
 733 }
 734 EXPORT_SYMBOL(xfrm_policy_insert);
 735
 736 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 737                                           struct xfrm_selector *sel,
 738                                           struct xfrm_sec_ctx *ctx, int delete)
 739 {
 740         struct xfrm_policy *pol, *ret;
 741         struct hlist_head *chain;
 742         struct hlist_node *entry;
 743
 744         write_lock_bh(&xfrm_policy_lock);
 745         chain = policy_hash_bysel(sel, sel->family, dir);
 746         ret = NULL;
 747         hlist_for_each_entry(pol, entry, chain, bydst) {
 748                 if (pol->type == type &&
 749                     !selector_cmp(sel, &pol->selector) &&
 750                     xfrm_sec_ctx_match(ctx, pol->security)) {
 751                         xfrm_pol_hold(pol);
 752                         if (delete) {
 753                                 hlist_del(&pol->bydst);
 754                                 hlist_del(&pol->byidx);
 755                                 xfrm_policy_count[dir]--;
 756                         }
 757                         ret = pol;
 758                         break;
 759                 }
 760         }
 761         write_unlock_bh(&xfrm_policy_lock);
 762
 763         if (ret && delete) {
 764                 atomic_inc(&flow_cache_genid);
 765                 xfrm_policy_kill(ret);
 766         }
 767         return ret;
 768 }
 769 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 770
 771 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete)
 772 {
 773         struct xfrm_policy *pol, *ret;
 774         struct hlist_head *chain;
 775         struct hlist_node *entry;
 776
 777         write_lock_bh(&xfrm_policy_lock);
 778         chain = xfrm_policy_byidx + idx_hash(id);
 779         ret = NULL;
 780         hlist_for_each_entry(pol, entry, chain, byidx) {
 781                 if (pol->type == type && pol->index == id) {
 782                         xfrm_pol_hold(pol);
 783                         if (delete) {
 784                                 hlist_del(&pol->bydst);
 785                                 hlist_del(&pol->byidx);
 786                                 xfrm_policy_count[dir]--;
 787                         }
 788                         ret = pol;
 789                         break;
 790                 }
 791         }
 792         write_unlock_bh(&xfrm_policy_lock);
 793
 794         if (ret && delete) {
 795                 atomic_inc(&flow_cache_genid);
 796                 xfrm_policy_kill(ret);
 797         }
 798         return ret;
 799 }
 800 EXPORT_SYMBOL(xfrm_policy_byid);
 801
 802 void xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 803 {
 804         int dir;
 805
 806         write_lock_bh(&xfrm_policy_lock);
 807         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 808                 struct xfrm_policy *pol;
 809                 struct hlist_node *entry;
 810                 int i, killed;
 811
 812                 killed = 0;
 813         again1:
 814                 hlist_for_each_entry(pol, entry,
 815                                      &xfrm_policy_inexact[dir], bydst) {
 816                         if (pol->type != type)
 817                                 continue;
 818                         hlist_del(&pol->bydst);
 819                         hlist_del(&pol->byidx);
 820                         write_unlock_bh(&xfrm_policy_lock);
 821
 822                         xfrm_audit_log(audit_info->loginuid, audit_info->secid,
 823                                        AUDIT_MAC_IPSEC_DELSPD, 1, pol, NULL);
 824
 825                         xfrm_policy_kill(pol);
 826                         killed++;
 827
 828                         write_lock_bh(&xfrm_policy_lock);
 829                         goto again1;
 830                 }
 831
 832                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 833         again2:
 834                         hlist_for_each_entry(pol, entry,
 835                                              xfrm_policy_bydst[dir].table + i,
 836                                              bydst) {
 837                                 if (pol->type != type)
 838                                         continue;
 839                                 hlist_del(&pol->bydst);
 840                                 hlist_del(&pol->byidx);
 841                                 write_unlock_bh(&xfrm_policy_lock);
 842
 843                                 xfrm_audit_log(audit_info->loginuid,
 844                                                audit_info->secid,
 845                                                AUDIT_MAC_IPSEC_DELSPD, 1,
 846                                                pol, NULL);
 847
 848                                 xfrm_policy_kill(pol);
 849                                 killed++;
 850
 851                                 write_lock_bh(&xfrm_policy_lock);
 852                                 goto again2;
 853                         }
 854                 }
 855
 856                 xfrm_policy_count[dir] -= killed;
 857         }
 858         atomic_inc(&flow_cache_genid);
 859         write_unlock_bh(&xfrm_policy_lock);
 860 }
 861 EXPORT_SYMBOL(xfrm_policy_flush);
 862
 863 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 864                      void *data)
 865 {
 866         struct xfrm_policy *pol, *last = NULL;
 867         struct hlist_node *entry;
 868         int dir, last_dir = 0, count, error;
 869
 870         read_lock_bh(&xfrm_policy_lock);
 871         count = 0;
 872
 873         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 874                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 875                 int i;
 876
 877                 hlist_for_each_entry(pol, entry,
 878                                      &xfrm_policy_inexact[dir], bydst) {
 879                         if (pol->type != type)
 880                                 continue;
 881                         if (last) {
 882                                 error = func(last, last_dir % XFRM_POLICY_MAX,
 883                                              count, data);
 884                                 if (error)
 885                                         goto out;
 886                         }
 887                         last = pol;
 888                         last_dir = dir;
 889                         count++;
 890                 }
 891                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 892                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 893                                 if (pol->type != type)
 894                                         continue;
 895                                 if (last) {
 896                                         error = func(last, last_dir % XFRM_POLICY_MAX,
 897                                                      count, data);
 898                                         if (error)
 899                                                 goto out;
 900                                 }
 901                                 last = pol;
 902                                 last_dir = dir;
 903                                 count++;
 904                         }
 905                 }
 906         }
 907         if (count == 0) {
 908                 error = -ENOENT;
 909                 goto out;
 910         }
 911         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
 912 out:
 913         read_unlock_bh(&xfrm_policy_lock);
 914         return error;
 915 }
 916 EXPORT_SYMBOL(xfrm_policy_walk);
 917
 918 /*
 919  * Find policy to apply to this flow.
 920  *
 921  * Returns 0 if policy found, else an -errno.
 922  */
 923 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 924                              u8 type, u16 family, int dir)
 925 {
 926         struct xfrm_selector *sel = &pol->selector;
 927         int match, ret = -ESRCH;
 928
 929         if (pol->family != family ||
 930             pol->type != type)
 931                 return ret;
 932
 933         match = xfrm_selector_match(sel, fl, family);
 934         if (match)
 935                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
 936
 937         return ret;
 938 }
 939
 940 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 941                                                      u16 family, u8 dir)
 942 {
 943         int err;
 944         struct xfrm_policy *pol, *ret;
 945         xfrm_address_t *daddr, *saddr;
 946         struct hlist_node *entry;
 947         struct hlist_head *chain;
 948         u32 priority = ~0U;
 949
 950         daddr = xfrm_flowi_daddr(fl, family);
 951         saddr = xfrm_flowi_saddr(fl, family);
 952         if (unlikely(!daddr || !saddr))
 953                 return NULL;
 954
 955         read_lock_bh(&xfrm_policy_lock);
 956         chain = policy_hash_direct(daddr, saddr, family, dir);
 957         ret = NULL;
 958         hlist_for_each_entry(pol, entry, chain, bydst) {
 959                 err = xfrm_policy_match(pol, fl, type, family, dir);
 960                 if (err) {
 961                         if (err == -ESRCH)
 962                                 continue;
 963                         else {
 964                                 ret = ERR_PTR(err);
 965                                 goto fail;
 966                         }
 967                 } else {
 968                         ret = pol;
 969                         priority = ret->priority;
 970                         break;
 971                 }
 972         }
 973         chain = &xfrm_policy_inexact[dir];
 974         hlist_for_each_entry(pol, entry, chain, bydst) {
 975                 err = xfrm_policy_match(pol, fl, type, family, dir);
 976                 if (err) {
 977                         if (err == -ESRCH)
 978                                 continue;
 979                         else {
 980                                 ret = ERR_PTR(err);
 981                                 goto fail;
 982                         }
 983                 } else if (pol->priority < priority) {
 984                         ret = pol;
 985                         break;
 986                 }
 987         }
 988         if (ret)
 989                 xfrm_pol_hold(ret);
 990 fail:
 991         read_unlock_bh(&xfrm_policy_lock);
 992
 993         return ret;
 994 }
 995
 996 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 997                                void **objp, atomic_t **obj_refp)
 998 {
 999         struct xfrm_policy *pol;
1000         int err = 0;
1001
1002 #ifdef CONFIG_XFRM_SUB_POLICY
1003         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1004         if (IS_ERR(pol)) {
1005                 err = PTR_ERR(pol);
1006                 pol = NULL;
1007         }
1008         if (pol || err)
1009                 goto end;
1010 #endif
1011         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1012         if (IS_ERR(pol)) {
1013                 err = PTR_ERR(pol);
1014                 pol = NULL;
1015         }
1016 #ifdef CONFIG_XFRM_SUB_POLICY
1017 end:
1018 #endif
1019         if ((*objp = (void *) pol) != NULL)
1020                 *obj_refp = &pol->refcnt;
1021         return err;
1022 }
1023
1024 static inline int policy_to_flow_dir(int dir)
1025 {
1026         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1027             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1028             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1029                 return dir;
1030         switch (dir) {
1031         default:
1032         case XFRM_POLICY_IN:
1033                 return FLOW_DIR_IN;
1034         case XFRM_POLICY_OUT:
1035                 return FLOW_DIR_OUT;
1036         case XFRM_POLICY_FWD:
1037                 return FLOW_DIR_FWD;
1038         };
1039 }
1040
1041 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1042 {
1043         struct xfrm_policy *pol;
1044
1045         read_lock_bh(&xfrm_policy_lock);
1046         if ((pol = sk->sk_policy[dir]) != NULL) {
1047                 int match = xfrm_selector_match(&pol->selector, fl,
1048                                                 sk->sk_family);
1049                 int err = 0;
1050
1051                 if (match) {
1052                         err = security_xfrm_policy_lookup(pol, fl->secid,
1053                                         policy_to_flow_dir(dir));
1054                         if (!err)
1055                                 xfrm_pol_hold(pol);
1056                         else if (err == -ESRCH)
1057                                 pol = NULL;
1058                         else
1059                                 pol = ERR_PTR(err);
1060                 } else
1061                         pol = NULL;
1062         }
1063         read_unlock_bh(&xfrm_policy_lock);
1064         return pol;
1065 }
1066
1067 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1068 {
1069         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1070                                                      pol->family, dir);
1071
1072         hlist_add_head(&pol->bydst, chain);
1073         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1074         xfrm_policy_count[dir]++;
1075         xfrm_pol_hold(pol);
1076
1077         if (xfrm_bydst_should_resize(dir, NULL))
1078                 schedule_work(&xfrm_hash_work);
1079 }
1080
1081 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1082                                                 int dir)
1083 {
1084         if (hlist_unhashed(&pol->bydst))
1085                 return NULL;
1086
1087         hlist_del(&pol->bydst);
1088         hlist_del(&pol->byidx);
1089         xfrm_policy_count[dir]--;
1090
1091         return pol;
1092 }
1093
1094 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1095 {
1096         write_lock_bh(&xfrm_policy_lock);
1097         pol = __xfrm_policy_unlink(pol, dir);
1098         write_unlock_bh(&xfrm_policy_lock);
1099         if (pol) {
1100                 if (dir < XFRM_POLICY_MAX)
1101                         atomic_inc(&flow_cache_genid);
1102                 xfrm_policy_kill(pol);
1103                 return 0;
1104         }
1105         return -ENOENT;
1106 }
1107 EXPORT_SYMBOL(xfrm_policy_delete);
1108
1109 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1110 {
1111         struct xfrm_policy *old_pol;
1112
1113 #ifdef CONFIG_XFRM_SUB_POLICY
1114         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1115                 return -EINVAL;
1116 #endif
1117
1118         write_lock_bh(&xfrm_policy_lock);
1119         old_pol = sk->sk_policy[dir];
1120         sk->sk_policy[dir] = pol;
1121         if (pol) {
1122                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
1123                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1124                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1125         }
1126         if (old_pol)
1127                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1128         write_unlock_bh(&xfrm_policy_lock);
1129
1130         if (old_pol) {
1131                 xfrm_policy_kill(old_pol);
1132         }
1133         return 0;
1134 }
1135
1136 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1137 {
1138         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1139
1140         if (newp) {
1141                 newp->selector = old->selector;
1142                 if (security_xfrm_policy_clone(old, newp)) {
1143                         kfree(newp);
1144                         return NULL;  /* ENOMEM */
1145                 }
1146                 newp->lft = old->lft;
1147                 newp->curlft = old->curlft;
1148                 newp->action = old->action;
1149                 newp->flags = old->flags;
1150                 newp->xfrm_nr = old->xfrm_nr;
1151                 newp->index = old->index;
1152                 newp->type = old->type;
1153                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1154                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1155                 write_lock_bh(&xfrm_policy_lock);
1156                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1157                 write_unlock_bh(&xfrm_policy_lock);
1158                 xfrm_pol_put(newp);
1159         }
1160         return newp;
1161 }
1162
1163 int __xfrm_sk_clone_policy(struct sock *sk)
1164 {
1165         struct xfrm_policy *p0 = sk->sk_policy[0],
1166                            *p1 = sk->sk_policy[1];
1167
1168         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1169         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1170                 return -ENOMEM;
1171         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1172                 return -ENOMEM;
1173         return 0;
1174 }
1175
1176 static int
1177 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1178                unsigned short family)
1179 {
1180         int err;
1181         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1182
1183         if (unlikely(afinfo == NULL))
1184                 return -EINVAL;
1185         err = afinfo->get_saddr(local, remote);
1186         xfrm_policy_put_afinfo(afinfo);
1187         return err;
1188 }
1189
1190 /* Resolve list of templates for the flow, given policy. */
1191
1192 static int
1193 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1194                       struct xfrm_state **xfrm,
1195                       unsigned short family)
1196 {
1197         int nx;
1198         int i, error;
1199         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1200         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1201         xfrm_address_t tmp;
1202
1203         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1204                 struct xfrm_state *x;
1205                 xfrm_address_t *remote = daddr;
1206                 xfrm_address_t *local  = saddr;
1207                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1208
1209                 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1210                         remote = &tmpl->id.daddr;
1211                         local = &tmpl->saddr;
1212                         family = tmpl->encap_family;
1213                         if (xfrm_addr_any(local, family)) {
1214                                 error = xfrm_get_saddr(&tmp, remote, family);
1215                                 if (error)
1216                                         goto fail;
1217                                 local = &tmp;
1218                         }
1219                 }
1220
1221                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1222
1223                 if (x && x->km.state == XFRM_STATE_VALID) {
1224                         xfrm[nx++] = x;
1225                         daddr = remote;
1226                         saddr = local;
1227                         continue;
1228                 }
1229                 if (x) {
1230                         error = (x->km.state == XFRM_STATE_ERROR ?
1231                                  -EINVAL : -EAGAIN);
1232                         xfrm_state_put(x);
1233                 }
1234
1235                 if (!tmpl->optional)
1236                         goto fail;
1237         }
1238         return nx;
1239
1240 fail:
1241         for (nx--; nx>=0; nx--)
1242                 xfrm_state_put(xfrm[nx]);
1243         return error;
1244 }
1245
1246 static int
1247 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1248                   struct xfrm_state **xfrm,
1249                   unsigned short family)
1250 {
1251         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1252         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1253         int cnx = 0;
1254         int error;
1255         int ret;
1256         int i;
1257
1258         for (i = 0; i < npols; i++) {
1259                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1260                         error = -ENOBUFS;
1261                         goto fail;
1262                 }
1263
1264                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1265                 if (ret < 0) {
1266                         error = ret;
1267                         goto fail;
1268                 } else
1269                         cnx += ret;
1270         }
1271
1272         /* found states are sorted for outbound processing */
1273         if (npols > 1)
1274                 xfrm_state_sort(xfrm, tpp, cnx, family);
1275
1276         return cnx;
1277
1278  fail:
1279         for (cnx--; cnx>=0; cnx--)
1280                 xfrm_state_put(tpp[cnx]);
1281         return error;
1282
1283 }
1284
1285 /* Check that the bundle accepts the flow and its components are
1286  * still valid.
1287  */
1288
1289 static struct dst_entry *
1290 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1291 {
1292         struct dst_entry *x;
1293         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1294         if (unlikely(afinfo == NULL))
1295                 return ERR_PTR(-EINVAL);
1296         x = afinfo->find_bundle(fl, policy);
1297         xfrm_policy_put_afinfo(afinfo);
1298         return x;
1299 }
1300
1301 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1302  * all the metrics... Shortly, bundle a bundle.
1303  */
1304
1305 static int
1306 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1307                    struct flowi *fl, struct dst_entry **dst_p,
1308                    unsigned short family)
1309 {
1310         int err;
1311         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1312         if (unlikely(afinfo == NULL))
1313                 return -EINVAL;
1314         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1315         xfrm_policy_put_afinfo(afinfo);
1316         return err;
1317 }
1318
1319
1320 static int stale_bundle(struct dst_entry *dst);
1321
1322 /* Main function: finds/creates a bundle for given flow.
1323  *
1324  * At the moment we eat a raw IP route. Mostly to speed up lookups
1325  * on interfaces with disabled IPsec.
1326  */
1327 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1328                 struct sock *sk, int flags)
1329 {
1330         struct xfrm_policy *policy;
1331         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1332         int npols;
1333         int pol_dead;
1334         int xfrm_nr;
1335         int pi;
1336         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1337         struct dst_entry *dst, *dst_orig = *dst_p;
1338         int nx = 0;
1339         int err;
1340         u32 genid;
1341         u16 family;
1342         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1343
1344 restart:
1345         genid = atomic_read(&flow_cache_genid);
1346         policy = NULL;
1347         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1348                 pols[pi] = NULL;
1349         npols = 0;
1350         pol_dead = 0;
1351         xfrm_nr = 0;
1352
1353         if (sk && sk->sk_policy[1]) {
1354                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1355                 if (IS_ERR(policy))
1356                         return PTR_ERR(policy);
1357         }
1358
1359         if (!policy) {
1360                 /* To accelerate a bit...  */
1361                 if ((dst_orig->flags & DST_NOXFRM) ||
1362                     !xfrm_policy_count[XFRM_POLICY_OUT])
1363                         return 0;
1364
1365                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1366                                            dir, xfrm_policy_lookup);
1367                 if (IS_ERR(policy))
1368                         return PTR_ERR(policy);
1369         }
1370
1371         if (!policy)
1372                 return 0;
1373
1374         family = dst_orig->ops->family;
1375         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
1376         pols[0] = policy;
1377         npols ++;
1378         xfrm_nr += pols[0]->xfrm_nr;
1379
1380         switch (policy->action) {
1381         case XFRM_POLICY_BLOCK:
1382                 /* Prohibit the flow */
1383                 err = -EPERM;
1384                 goto error;
1385
1386         case XFRM_POLICY_ALLOW:
1387 #ifndef CONFIG_XFRM_SUB_POLICY
1388                 if (policy->xfrm_nr == 0) {
1389                         /* Flow passes not transformed. */
1390                         xfrm_pol_put(policy);
1391                         return 0;
1392                 }
1393 #endif
1394
1395                 /* Try to find matching bundle.
1396                  *
1397                  * LATER: help from flow cache. It is optional, this
1398                  * is required only for output policy.
1399                  */
1400                 dst = xfrm_find_bundle(fl, policy, family);
1401                 if (IS_ERR(dst)) {
1402                         err = PTR_ERR(dst);
1403                         goto error;
1404                 }
1405
1406                 if (dst)
1407                         break;
1408
1409 #ifdef CONFIG_XFRM_SUB_POLICY
1410                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1411                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1412                                                             fl, family,
1413                                                             XFRM_POLICY_OUT);
1414                         if (pols[1]) {
1415                                 if (IS_ERR(pols[1])) {
1416                                         err = PTR_ERR(pols[1]);
1417                                         goto error;
1418                                 }
1419                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1420                                         err = -EPERM;
1421                                         goto error;
1422                                 }
1423                                 npols ++;
1424                                 xfrm_nr += pols[1]->xfrm_nr;
1425                         }
1426                 }
1427
1428                 /*
1429                  * Because neither flowi nor bundle information knows about
1430                  * transformation template size. On more than one policy usage
1431                  * we can realize whether all of them is bypass or not after
1432                  * they are searched. See above not-transformed bypass
1433                  * is surrounded by non-sub policy configuration, too.
1434                  */
1435                 if (xfrm_nr == 0) {
1436                         /* Flow passes not transformed. */
1437                         xfrm_pols_put(pols, npols);
1438                         return 0;
1439                 }
1440
1441 #endif
1442                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1443
1444                 if (unlikely(nx<0)) {
1445                         err = nx;
1446                         if (err == -EAGAIN && flags) {
1447                                 DECLARE_WAITQUEUE(wait, current);
1448
1449                                 add_wait_queue(&km_waitq, &wait);
1450                                 set_current_state(TASK_INTERRUPTIBLE);
1451                                 schedule();
1452                                 set_current_state(TASK_RUNNING);
1453                                 remove_wait_queue(&km_waitq, &wait);
1454
1455                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1456
1457                                 if (nx == -EAGAIN && signal_pending(current)) {
1458                                         err = -ERESTART;
1459                                         goto error;
1460                                 }
1461                                 if (nx == -EAGAIN ||
1462                                     genid != atomic_read(&flow_cache_genid)) {
1463                                         xfrm_pols_put(pols, npols);
1464                                         goto restart;
1465                                 }
1466                                 err = nx;
1467                         }
1468                         if (err < 0)
1469                                 goto error;
1470                 }
1471                 if (nx == 0) {
1472                         /* Flow passes not transformed. */
1473                         xfrm_pols_put(pols, npols);
1474                         return 0;
1475                 }
1476
1477                 dst = dst_orig;
1478                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1479
1480                 if (unlikely(err)) {
1481                         int i;
1482                         for (i=0; i<nx; i++)
1483                                 xfrm_state_put(xfrm[i]);
1484                         goto error;
1485                 }
1486
1487                 for (pi = 0; pi < npols; pi++) {
1488                         read_lock_bh(&pols[pi]->lock);
1489                         pol_dead |= pols[pi]->dead;
1490                         read_unlock_bh(&pols[pi]->lock);
1491                 }
1492
1493                 write_lock_bh(&policy->lock);
1494                 if (unlikely(pol_dead || stale_bundle(dst))) {
1495                         /* Wow! While we worked on resolving, this
1496                          * policy has gone. Retry. It is not paranoia,
1497                          * we just cannot enlist new bundle to dead object.
1498                          * We can't enlist stable bundles either.
1499                          */
1500                         write_unlock_bh(&policy->lock);
1501                         if (dst)
1502                                 dst_free(dst);
1503
1504                         err = -EHOSTUNREACH;
1505                         goto error;
1506                 }
1507                 dst->next = policy->bundles;
1508                 policy->bundles = dst;
1509                 dst_hold(dst);
1510                 write_unlock_bh(&policy->lock);
1511         }
1512         *dst_p = dst;
1513         dst_release(dst_orig);
1514         xfrm_pols_put(pols, npols);
1515         return 0;
1516
1517 error:
1518         dst_release(dst_orig);
1519         xfrm_pols_put(pols, npols);
1520         *dst_p = NULL;
1521         return err;
1522 }
1523 EXPORT_SYMBOL(xfrm_lookup);
1524
1525 static inline int
1526 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1527 {
1528         struct xfrm_state *x;
1529         int err;
1530
1531         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1532                 return 0;
1533         x = skb->sp->xvec[idx];
1534         if (!x->type->reject)
1535                 return 0;
1536         xfrm_state_hold(x);
1537         err = x->type->reject(x, skb, fl);
1538         xfrm_state_put(x);
1539         return err;
1540 }
1541
1542 /* When skb is transformed back to its "native" form, we have to
1543  * check policy restrictions. At the moment we make this in maximally
1544  * stupid way. Shame on me. :-) Of course, connected sockets must
1545  * have policy cached at them.
1546  */
1547
1548 static inline int
1549 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1550               unsigned short family)
1551 {
1552         if (xfrm_state_kern(x))
1553                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1554         return  x->id.proto == tmpl->id.proto &&
1555                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1556                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1557                 x->props.mode == tmpl->mode &&
1558                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1559                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1560                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1561                   xfrm_state_addr_cmp(tmpl, x, family));
1562 }
1563
1564 /*
1565  * 0 or more than 0 is returned when validation is succeeded (either bypass
1566  * because of optional transport mode, or next index of the mathced secpath
1567  * state with the template.
1568  * -1 is returned when no matching template is found.
1569  * Otherwise "-2 - errored_index" is returned.
1570  */
1571 static inline int
1572 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1573                unsigned short family)
1574 {
1575         int idx = start;
1576
1577         if (tmpl->optional) {
1578                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1579                         return start;
1580         } else
1581                 start = -1;
1582         for (; idx < sp->len; idx++) {
1583                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1584                         return ++idx;
1585                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1586                         if (start == -1)
1587                                 start = -2-idx;
1588                         break;
1589                 }
1590         }
1591         return start;
1592 }
1593
1594 int
1595 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1596 {
1597         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1598         int err;
1599
1600         if (unlikely(afinfo == NULL))
1601                 return -EAFNOSUPPORT;
1602
1603         afinfo->decode_session(skb, fl);
1604         err = security_xfrm_decode_session(skb, &fl->secid);
1605         xfrm_policy_put_afinfo(afinfo);
1606         return err;
1607 }
1608 EXPORT_SYMBOL(xfrm_decode_session);
1609
1610 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1611 {
1612         for (; k < sp->len; k++) {
1613                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1614                         *idxp = k;
1615                         return 1;
1616                 }
1617         }
1618
1619         return 0;
1620 }
1621
1622 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1623                         unsigned short family)
1624 {
1625         struct xfrm_policy *pol;
1626         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1627         int npols = 0;
1628         int xfrm_nr;
1629         int pi;
1630         struct flowi fl;
1631         u8 fl_dir = policy_to_flow_dir(dir);
1632         int xerr_idx = -1;
1633
1634         if (xfrm_decode_session(skb, &fl, family) < 0)
1635                 return 0;
1636         nf_nat_decode_session(skb, &fl, family);
1637
1638         /* First, check used SA against their selectors. */
1639         if (skb->sp) {
1640                 int i;
1641
1642                 for (i=skb->sp->len-1; i>=0; i--) {
1643                         struct xfrm_state *x = skb->sp->xvec[i];
1644                         if (!xfrm_selector_match(&x->sel, &fl, family))
1645                                 return 0;
1646                 }
1647         }
1648
1649         pol = NULL;
1650         if (sk && sk->sk_policy[dir]) {
1651                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1652                 if (IS_ERR(pol))
1653                         return 0;
1654         }
1655
1656         if (!pol)
1657                 pol = flow_cache_lookup(&fl, family, fl_dir,
1658                                         xfrm_policy_lookup);
1659
1660         if (IS_ERR(pol))
1661                 return 0;
1662
1663         if (!pol) {
1664                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1665                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1666                         return 0;
1667                 }
1668                 return 1;
1669         }
1670
1671         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1672
1673         pols[0] = pol;
1674         npols ++;
1675 #ifdef CONFIG_XFRM_SUB_POLICY
1676         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1677                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1678                                                     &fl, family,
1679                                                     XFRM_POLICY_IN);
1680                 if (pols[1]) {
1681                         if (IS_ERR(pols[1]))
1682                                 return 0;
1683                         pols[1]->curlft.use_time = (unsigned long)xtime.tv_sec;
1684                         npols ++;
1685                 }
1686         }
1687 #endif
1688
1689         if (pol->action == XFRM_POLICY_ALLOW) {
1690                 struct sec_path *sp;
1691                 static struct sec_path dummy;
1692                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1693                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1694                 struct xfrm_tmpl **tpp = tp;
1695                 int ti = 0;
1696                 int i, k;
1697
1698                 if ((sp = skb->sp) == NULL)
1699                         sp = &dummy;
1700
1701                 for (pi = 0; pi < npols; pi++) {
1702                         if (pols[pi] != pol &&
1703                             pols[pi]->action != XFRM_POLICY_ALLOW)
1704                                 goto reject;
1705                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1706                                 goto reject_error;
1707                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1708                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1709                 }
1710                 xfrm_nr = ti;
1711                 if (npols > 1) {
1712                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1713                         tpp = stp;
1714                 }
1715
1716                 /* For each tunnel xfrm, find the first matching tmpl.
1717                  * For each tmpl before that, find corresponding xfrm.
1718                  * Order is _important_. Later we will implement
1719                  * some barriers, but at the moment barriers
1720                  * are implied between each two transformations.
1721                  */
1722                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1723                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1724                         if (k < 0) {
1725                                 if (k < -1)
1726                                         /* "-2 - errored_index" returned */
1727                                         xerr_idx = -(2+k);
1728                                 goto reject;
1729                         }
1730                 }
1731
1732                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1733                         goto reject;
1734
1735                 xfrm_pols_put(pols, npols);
1736                 return 1;
1737         }
1738
1739 reject:
1740         xfrm_secpath_reject(xerr_idx, skb, &fl);
1741 reject_error:
1742         xfrm_pols_put(pols, npols);
1743         return 0;
1744 }
1745 EXPORT_SYMBOL(__xfrm_policy_check);
1746
1747 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1748 {
1749         struct flowi fl;
1750
1751         if (xfrm_decode_session(skb, &fl, family) < 0)
1752                 return 0;
1753
1754         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1755 }
1756 EXPORT_SYMBOL(__xfrm_route_forward);
1757
1758 /* Optimize later using cookies and generation ids. */
1759
1760 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1761 {
1762         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1763          * to "-1" to force all XFRM destinations to get validated by
1764          * dst_ops->check on every use.  We do this because when a
1765          * normal route referenced by an XFRM dst is obsoleted we do
1766          * not go looking around for all parent referencing XFRM dsts
1767          * so that we can invalidate them.  It is just too much work.
1768          * Instead we make the checks here on every use.  For example:
1769          *
1770          *      XFRM dst A --> IPv4 dst X
1771          *
1772          * X is the "xdst->route" of A (X is also the "dst->path" of A
1773          * in this example).  If X is marked obsolete, "A" will not
1774          * notice.  That's what we are validating here via the
1775          * stale_bundle() check.
1776          *
1777          * When a policy's bundle is pruned, we dst_free() the XFRM
1778          * dst which causes it's ->obsolete field to be set to a
1779          * positive non-zero integer.  If an XFRM dst has been pruned
1780          * like this, we want to force a new route lookup.
1781          */
1782         if (dst->obsolete < 0 && !stale_bundle(dst))
1783                 return dst;
1784
1785         return NULL;
1786 }
1787
1788 static int stale_bundle(struct dst_entry *dst)
1789 {
1790         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1791 }
1792
1793 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1794 {
1795         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1796                 dst->dev = &loopback_dev;
1797                 dev_hold(&loopback_dev);
1798                 dev_put(dev);
1799         }
1800 }
1801 EXPORT_SYMBOL(xfrm_dst_ifdown);
1802
1803 static void xfrm_link_failure(struct sk_buff *skb)
1804 {
1805         /* Impossible. Such dst must be popped before reaches point of failure. */
1806         return;
1807 }
1808
1809 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1810 {
1811         if (dst) {
1812                 if (dst->obsolete) {
1813                         dst_release(dst);
1814                         dst = NULL;
1815                 }
1816         }
1817         return dst;
1818 }
1819
1820 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1821 {
1822         struct dst_entry *dst, **dstp;
1823
1824         write_lock(&pol->lock);
1825         dstp = &pol->bundles;
1826         while ((dst=*dstp) != NULL) {
1827                 if (func(dst)) {
1828                         *dstp = dst->next;
1829                         dst->next = *gc_list_p;
1830                         *gc_list_p = dst;
1831                 } else {
1832                         dstp = &dst->next;
1833                 }
1834         }
1835         write_unlock(&pol->lock);
1836 }
1837
1838 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1839 {
1840         struct dst_entry *gc_list = NULL;
1841         int dir;
1842
1843         read_lock_bh(&xfrm_policy_lock);
1844         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1845                 struct xfrm_policy *pol;
1846                 struct hlist_node *entry;
1847                 struct hlist_head *table;
1848                 int i;
1849
1850                 hlist_for_each_entry(pol, entry,
1851                                      &xfrm_policy_inexact[dir], bydst)
1852                         prune_one_bundle(pol, func, &gc_list);
1853
1854                 table = xfrm_policy_bydst[dir].table;
1855                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1856                         hlist_for_each_entry(pol, entry, table + i, bydst)
1857                                 prune_one_bundle(pol, func, &gc_list);
1858                 }
1859         }
1860         read_unlock_bh(&xfrm_policy_lock);
1861
1862         while (gc_list) {
1863                 struct dst_entry *dst = gc_list;
1864                 gc_list = dst->next;
1865                 dst_free(dst);
1866         }
1867 }
1868
1869 static int unused_bundle(struct dst_entry *dst)
1870 {
1871         return !atomic_read(&dst->__refcnt);
1872 }
1873
1874 static void __xfrm_garbage_collect(void)
1875 {
1876         xfrm_prune_bundles(unused_bundle);
1877 }
1878
1879 static int xfrm_flush_bundles(void)
1880 {
1881         xfrm_prune_bundles(stale_bundle);
1882         return 0;
1883 }
1884
1885 void xfrm_init_pmtu(struct dst_entry *dst)
1886 {
1887         do {
1888                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1889                 u32 pmtu, route_mtu_cached;
1890
1891                 pmtu = dst_mtu(dst->child);
1892                 xdst->child_mtu_cached = pmtu;
1893
1894                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1895
1896                 route_mtu_cached = dst_mtu(xdst->route);
1897                 xdst->route_mtu_cached = route_mtu_cached;
1898
1899                 if (pmtu > route_mtu_cached)
1900                         pmtu = route_mtu_cached;
1901
1902                 dst->metrics[RTAX_MTU-1] = pmtu;
1903         } while ((dst = dst->next));
1904 }
1905
1906 EXPORT_SYMBOL(xfrm_init_pmtu);
1907
1908 /* Check that the bundle accepts the flow and its components are
1909  * still valid.
1910  */
1911
1912 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
1913                 struct flowi *fl, int family, int strict)
1914 {
1915         struct dst_entry *dst = &first->u.dst;
1916         struct xfrm_dst *last;
1917         u32 mtu;
1918
1919         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1920             (dst->dev && !netif_running(dst->dev)))
1921                 return 0;
1922
1923         last = NULL;
1924
1925         do {
1926                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1927
1928                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1929                         return 0;
1930                 if (fl && pol &&
1931                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
1932                         return 0;
1933                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1934                         return 0;
1935                 if (xdst->genid != dst->xfrm->genid)
1936                         return 0;
1937
1938                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
1939                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1940                         return 0;
1941
1942                 mtu = dst_mtu(dst->child);
1943                 if (xdst->child_mtu_cached != mtu) {
1944                         last = xdst;
1945                         xdst->child_mtu_cached = mtu;
1946                 }
1947
1948                 if (!dst_check(xdst->route, xdst->route_cookie))
1949                         return 0;
1950                 mtu = dst_mtu(xdst->route);
1951                 if (xdst->route_mtu_cached != mtu) {
1952                         last = xdst;
1953                         xdst->route_mtu_cached = mtu;
1954                 }
1955
1956                 dst = dst->child;
1957         } while (dst->xfrm);
1958
1959         if (likely(!last))
1960                 return 1;
1961
1962         mtu = last->child_mtu_cached;
1963         for (;;) {
1964                 dst = &last->u.dst;
1965
1966                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1967                 if (mtu > last->route_mtu_cached)
1968                         mtu = last->route_mtu_cached;
1969                 dst->metrics[RTAX_MTU-1] = mtu;
1970
1971                 if (last == first)
1972                         break;
1973
1974                 last = last->u.next;
1975                 last->child_mtu_cached = mtu;
1976         }
1977
1978         return 1;
1979 }
1980
1981 EXPORT_SYMBOL(xfrm_bundle_ok);
1982
1983 #ifdef CONFIG_AUDITSYSCALL
1984 /* Audit addition and deletion of SAs and ipsec policy */
1985
1986 void xfrm_audit_log(uid_t auid, u32 sid, int type, int result,
1987                     struct xfrm_policy *xp, struct xfrm_state *x)
1988 {
1989
1990         char *secctx;
1991         u32 secctx_len;
1992         struct xfrm_sec_ctx *sctx = NULL;
1993         struct audit_buffer *audit_buf;
1994         int family;
1995         extern int audit_enabled;
1996
1997         if (audit_enabled == 0)
1998                 return;
1999
2000         audit_buf = audit_log_start(current->audit_context, GFP_ATOMIC, type);
2001         if (audit_buf == NULL)
2002         return;
2003
2004         switch(type) {
2005         case AUDIT_MAC_IPSEC_ADDSA:
2006                 audit_log_format(audit_buf, "SAD add: auid=%u", auid);
2007                 break;
2008         case AUDIT_MAC_IPSEC_DELSA:
2009                 audit_log_format(audit_buf, "SAD delete: auid=%u", auid);
2010                 break;
2011         case AUDIT_MAC_IPSEC_ADDSPD:
2012                 audit_log_format(audit_buf, "SPD add: auid=%u", auid);
2013                 break;
2014         case AUDIT_MAC_IPSEC_DELSPD:
2015                 audit_log_format(audit_buf, "SPD delete: auid=%u", auid);
2016                 break;
2017         default:
2018                 return;
2019         }
2020
2021         if (sid != 0 &&
2022                 security_secid_to_secctx(sid, &secctx, &secctx_len) == 0)
2023                 audit_log_format(audit_buf, " subj=%s", secctx);
2024         else
2025                 audit_log_task_context(audit_buf);
2026
2027         if (xp) {
2028                 family = xp->selector.family;
2029                 if (xp->security)
2030                         sctx = xp->security;
2031         } else {
2032                 family = x->props.family;
2033                 if (x->security)
2034                         sctx = x->security;
2035         }
2036
2037         if (sctx)
2038                 audit_log_format(audit_buf,
2039                                 " sec_alg=%u sec_doi=%u sec_obj=%s",
2040                                 sctx->ctx_alg, sctx->ctx_doi, sctx->ctx_str);
2041
2042         switch(family) {
2043         case AF_INET:
2044                 {
2045                         struct in_addr saddr, daddr;
2046                         if (xp) {
2047                                 saddr.s_addr = xp->selector.saddr.a4;
2048                                 daddr.s_addr = xp->selector.daddr.a4;
2049                         } else {
2050                                 saddr.s_addr = x->props.saddr.a4;
2051                                 daddr.s_addr = x->id.daddr.a4;
2052                         }
2053                         audit_log_format(audit_buf,
2054                                          " src=%u.%u.%u.%u dst=%u.%u.%u.%u",
2055                                          NIPQUAD(saddr), NIPQUAD(daddr));
2056                 }
2057                         break;
2058         case AF_INET6:
2059                 {
2060                         struct in6_addr saddr6, daddr6;
2061                         if (xp) {
2062                                 memcpy(&saddr6, xp->selector.saddr.a6,
2063                                         sizeof(struct in6_addr));
2064                                 memcpy(&daddr6, xp->selector.daddr.a6,
2065                                         sizeof(struct in6_addr));
2066                         } else {
2067                                 memcpy(&saddr6, x->props.saddr.a6,
2068                                         sizeof(struct in6_addr));
2069                                 memcpy(&daddr6, x->id.daddr.a6,
2070                                         sizeof(struct in6_addr));
2071                         }
2072                         audit_log_format(audit_buf,
2073                                          " src=" NIP6_FMT "dst=" NIP6_FMT,
2074                                          NIP6(saddr6), NIP6(daddr6));
2075                 }
2076                 break;
2077         }
2078
2079         if (x)
2080                 audit_log_format(audit_buf, " spi=%lu(0x%lx) protocol=%s",
2081                                 (unsigned long)ntohl(x->id.spi),
2082                                 (unsigned long)ntohl(x->id.spi),
2083                                 x->id.proto == IPPROTO_AH ? "AH" :
2084                                 (x->id.proto == IPPROTO_ESP ?
2085                                 "ESP" : "IPCOMP"));
2086
2087         audit_log_format(audit_buf, " res=%u", result);
2088         audit_log_end(audit_buf);
2089 }
2090
2091 EXPORT_SYMBOL(xfrm_audit_log);
2092 #endif /* CONFIG_AUDITSYSCALL */
2093
2094 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2095 {
2096         int err = 0;
2097         if (unlikely(afinfo == NULL))
2098                 return -EINVAL;
2099         if (unlikely(afinfo->family >= NPROTO))
2100                 return -EAFNOSUPPORT;
2101         write_lock_bh(&xfrm_policy_afinfo_lock);
2102         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2103                 err = -ENOBUFS;
2104         else {
2105                 struct dst_ops *dst_ops = afinfo->dst_ops;
2106                 if (likely(dst_ops->kmem_cachep == NULL))
2107                         dst_ops->kmem_cachep = xfrm_dst_cache;
2108                 if (likely(dst_ops->check == NULL))
2109                         dst_ops->check = xfrm_dst_check;
2110                 if (likely(dst_ops->negative_advice == NULL))
2111                         dst_ops->negative_advice = xfrm_negative_advice;
2112                 if (likely(dst_ops->link_failure == NULL))
2113                         dst_ops->link_failure = xfrm_link_failure;
2114                 if (likely(afinfo->garbage_collect == NULL))
2115                         afinfo->garbage_collect = __xfrm_garbage_collect;
2116                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2117         }
2118         write_unlock_bh(&xfrm_policy_afinfo_lock);
2119         return err;
2120 }
2121 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2122
2123 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2124 {
2125         int err = 0;
2126         if (unlikely(afinfo == NULL))
2127                 return -EINVAL;
2128         if (unlikely(afinfo->family >= NPROTO))
2129                 return -EAFNOSUPPORT;
2130         write_lock_bh(&xfrm_policy_afinfo_lock);
2131         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2132                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2133                         err = -EINVAL;
2134                 else {
2135                         struct dst_ops *dst_ops = afinfo->dst_ops;
2136                         xfrm_policy_afinfo[afinfo->family] = NULL;
2137                         dst_ops->kmem_cachep = NULL;
2138                         dst_ops->check = NULL;
2139                         dst_ops->negative_advice = NULL;
2140                         dst_ops->link_failure = NULL;
2141                         afinfo->garbage_collect = NULL;
2142                 }
2143         }
2144         write_unlock_bh(&xfrm_policy_afinfo_lock);
2145         return err;
2146 }
2147 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2148
2149 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2150 {
2151         struct xfrm_policy_afinfo *afinfo;
2152         if (unlikely(family >= NPROTO))
2153                 return NULL;
2154         read_lock(&xfrm_policy_afinfo_lock);
2155         afinfo = xfrm_policy_afinfo[family];
2156         if (unlikely(!afinfo))
2157                 read_unlock(&xfrm_policy_afinfo_lock);
2158         return afinfo;
2159 }
2160
2161 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2162 {
2163         read_unlock(&xfrm_policy_afinfo_lock);
2164 }
2165
2166 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2167 {
2168         struct xfrm_policy_afinfo *afinfo;
2169         if (unlikely(family >= NPROTO))
2170                 return NULL;
2171         write_lock_bh(&xfrm_policy_afinfo_lock);
2172         afinfo = xfrm_policy_afinfo[family];
2173         if (unlikely(!afinfo))
2174                 write_unlock_bh(&xfrm_policy_afinfo_lock);
2175         return afinfo;
2176 }
2177
2178 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2179 {
2180         write_unlock_bh(&xfrm_policy_afinfo_lock);
2181 }
2182
2183 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2184 {
2185         switch (event) {
2186         case NETDEV_DOWN:
2187                 xfrm_flush_bundles();
2188         }
2189         return NOTIFY_DONE;
2190 }
2191
2192 static struct notifier_block xfrm_dev_notifier = {
2193         xfrm_dev_event,
2194         NULL,
2195         0
2196 };
2197
2198 static void __init xfrm_policy_init(void)
2199 {
2200         unsigned int hmask, sz;
2201         int dir;
2202
2203         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2204                                            sizeof(struct xfrm_dst),
2205                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2206                                            NULL, NULL);
2207
2208         hmask = 8 - 1;
2209         sz = (hmask+1) * sizeof(struct hlist_head);
2210
2211         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2212         xfrm_idx_hmask = hmask;
2213         if (!xfrm_policy_byidx)
2214                 panic("XFRM: failed to allocate byidx hash\n");
2215
2216         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2217                 struct xfrm_policy_hash *htab;
2218
2219                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2220
2221                 htab = &xfrm_policy_bydst[dir];
2222                 htab->table = xfrm_hash_alloc(sz);
2223                 htab->hmask = hmask;
2224                 if (!htab->table)
2225                         panic("XFRM: failed to allocate bydst hash\n");
2226         }
2227
2228         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2229         register_netdevice_notifier(&xfrm_dev_notifier);
2230 }
2231
2232 void __init xfrm_init(void)
2233 {
2234         xfrm_state_init();
2235         xfrm_policy_init();
2236         xfrm_input_init();
2237 }
2238
2239 #ifdef CONFIG_XFRM_MIGRATE
2240 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2241                                        struct xfrm_selector *sel_tgt)
2242 {
2243         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2244                 if (sel_tgt->family == sel_cmp->family &&
2245                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2246                                   sel_cmp->family) == 0 &&
2247                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2248                                   sel_cmp->family) == 0 &&
2249                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2250                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2251                         return 1;
2252                 }
2253         } else {
2254                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2255                         return 1;
2256                 }
2257         }
2258         return 0;
2259 }
2260
2261 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2262                                                      u8 dir, u8 type)
2263 {
2264         struct xfrm_policy *pol, *ret = NULL;
2265         struct hlist_node *entry;
2266         struct hlist_head *chain;
2267         u32 priority = ~0U;
2268
2269         read_lock_bh(&xfrm_policy_lock);
2270         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2271         hlist_for_each_entry(pol, entry, chain, bydst) {
2272                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2273                     pol->type == type) {
2274                         ret = pol;
2275                         priority = ret->priority;
2276                         break;
2277                 }
2278         }
2279         chain = &xfrm_policy_inexact[dir];
2280         hlist_for_each_entry(pol, entry, chain, bydst) {
2281                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2282                     pol->type == type &&
2283                     pol->priority < priority) {
2284                         ret = pol;
2285                         break;
2286                 }
2287         }
2288
2289         if (ret)
2290                 xfrm_pol_hold(ret);
2291
2292         read_unlock_bh(&xfrm_policy_lock);
2293
2294         return ret;
2295 }
2296
2297 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2298 {
2299         int match = 0;
2300
2301         if (t->mode == m->mode && t->id.proto == m->proto &&
2302             (m->reqid == 0 || t->reqid == m->reqid)) {
2303                 switch (t->mode) {
2304                 case XFRM_MODE_TUNNEL:
2305                 case XFRM_MODE_BEET:
2306                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2307                                           m->old_family) == 0 &&
2308                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2309                                           m->old_family) == 0) {
2310                                 match = 1;
2311                         }
2312                         break;
2313                 case XFRM_MODE_TRANSPORT:
2314                         /* in case of transport mode, template does not store
2315                            any IP addresses, hence we just compare mode and
2316                            protocol */
2317                         match = 1;
2318                         break;
2319                 default:
2320                         break;
2321                 }
2322         }
2323         return match;
2324 }
2325
2326 /* update endpoint address(es) of template(s) */
2327 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2328                                struct xfrm_migrate *m, int num_migrate)
2329 {
2330         struct xfrm_migrate *mp;
2331         struct dst_entry *dst;
2332         int i, j, n = 0;
2333
2334         write_lock_bh(&pol->lock);
2335         if (unlikely(pol->dead)) {
2336                 /* target policy has been deleted */
2337                 write_unlock_bh(&pol->lock);
2338                 return -ENOENT;
2339         }
2340
2341         for (i = 0; i < pol->xfrm_nr; i++) {
2342                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2343                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2344                                 continue;
2345                         n++;
2346                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL)
2347                                 continue;
2348                         /* update endpoints */
2349                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2350                                sizeof(pol->xfrm_vec[i].id.daddr));
2351                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2352                                sizeof(pol->xfrm_vec[i].saddr));
2353                         pol->xfrm_vec[i].encap_family = mp->new_family;
2354                         /* flush bundles */
2355                         while ((dst = pol->bundles) != NULL) {
2356                                 pol->bundles = dst->next;
2357                                 dst_free(dst);
2358                         }
2359                 }
2360         }
2361
2362         write_unlock_bh(&pol->lock);
2363
2364         if (!n)
2365                 return -ENODATA;
2366
2367         return 0;
2368 }
2369
2370 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2371 {
2372         int i, j;
2373
2374         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2375                 return -EINVAL;
2376
2377         for (i = 0; i < num_migrate; i++) {
2378                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2379                                    m[i].old_family) == 0) &&
2380                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2381                                    m[i].old_family) == 0))
2382                         return -EINVAL;
2383                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2384                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2385                         return -EINVAL;
2386
2387                 /* check if there is any duplicated entry */
2388                 for (j = i + 1; j < num_migrate; j++) {
2389                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2390                                     sizeof(m[i].old_daddr)) &&
2391                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2392                                     sizeof(m[i].old_saddr)) &&
2393                             m[i].proto == m[j].proto &&
2394                             m[i].mode == m[j].mode &&
2395                             m[i].reqid == m[j].reqid &&
2396                             m[i].old_family == m[j].old_family)
2397                                 return -EINVAL;
2398                 }
2399         }
2400
2401         return 0;
2402 }
2403
2404 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2405                  struct xfrm_migrate *m, int num_migrate)
2406 {
2407         int i, err, nx_cur = 0, nx_new = 0;
2408         struct xfrm_policy *pol = NULL;
2409         struct xfrm_state *x, *xc;
2410         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2411         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2412         struct xfrm_migrate *mp;
2413
2414         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2415                 goto out;
2416
2417         /* Stage 1 - find policy */
2418         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2419                 err = -ENOENT;
2420                 goto out;
2421         }
2422
2423         /* Stage 2 - find and update state(s) */
2424         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2425                 if ((x = xfrm_migrate_state_find(mp))) {
2426                         x_cur[nx_cur] = x;
2427                         nx_cur++;
2428                         if ((xc = xfrm_state_migrate(x, mp))) {
2429                                 x_new[nx_new] = xc;
2430                                 nx_new++;
2431                         } else {
2432                                 err = -ENODATA;
2433                                 goto restore_state;
2434                         }
2435                 }
2436         }
2437
2438         /* Stage 3 - update policy */
2439         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2440                 goto restore_state;
2441
2442         /* Stage 4 - delete old state(s) */
2443         if (nx_cur) {
2444                 xfrm_states_put(x_cur, nx_cur);
2445                 xfrm_states_delete(x_cur, nx_cur);
2446         }
2447
2448         /* Stage 5 - announce */
2449         km_migrate(sel, dir, type, m, num_migrate);
2450
2451         xfrm_pol_put(pol);
2452
2453         return 0;
2454 out:
2455         return err;
2456
2457 restore_state:
2458         if (pol)
2459                 xfrm_pol_put(pol);
2460         if (nx_cur)
2461                 xfrm_states_put(x_cur, nx_cur);
2462         if (nx_new)
2463                 xfrm_states_delete(x_new, nx_new);
2464
2465         return err;
2466 }
2467 EXPORT_SYMBOL(xfrm_migrate);
2468 #endif
2469