err.no Git - linux-2.6/blob - net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/xfrm.h>
  30 #include <net/ip.h>
  31 #ifdef CONFIG_XFRM_STATISTICS
  32 #include <net/snmp.h>
  33 #endif
  34
  35 #include "xfrm_hash.h"
  36
  37 int sysctl_xfrm_larval_drop __read_mostly;
  38
  39 #ifdef CONFIG_XFRM_STATISTICS
  40 DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
  41 EXPORT_SYMBOL(xfrm_statistics);
  42 #endif
  43
  44 DEFINE_MUTEX(xfrm_cfg_mutex);
  45 EXPORT_SYMBOL(xfrm_cfg_mutex);
  46
  47 static DEFINE_RWLOCK(xfrm_policy_lock);
  48
  49 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  50 EXPORT_SYMBOL(xfrm_policy_count);
  51
  52 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  53 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  54
  55 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  56
  57 static struct work_struct xfrm_policy_gc_work;
  58 static HLIST_HEAD(xfrm_policy_gc_list);
  59 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  60
  61 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  62 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  63 static void xfrm_init_pmtu(struct dst_entry *dst);
  64
  65 static inline int
  66 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  67 {
  68         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  69                 addr_match(&fl->fl4_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 static inline int
  77 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  78 {
  79         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  80                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  81                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  82                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  83                 (fl->proto == sel->proto || !sel->proto) &&
  84                 (fl->oif == sel->ifindex || !sel->ifindex);
  85 }
  86
  87 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  88                     unsigned short family)
  89 {
  90         switch (family) {
  91         case AF_INET:
  92                 return __xfrm4_selector_match(sel, fl);
  93         case AF_INET6:
  94                 return __xfrm6_selector_match(sel, fl);
  95         }
  96         return 0;
  97 }
  98
  99 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 100                                                 int family)
 101 {
 102         xfrm_address_t *saddr = &x->props.saddr;
 103         xfrm_address_t *daddr = &x->id.daddr;
 104         struct xfrm_policy_afinfo *afinfo;
 105         struct dst_entry *dst;
 106
 107         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR)
 108                 saddr = x->coaddr;
 109         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR)
 110                 daddr = x->coaddr;
 111
 112         afinfo = xfrm_policy_get_afinfo(family);
 113         if (unlikely(afinfo == NULL))
 114                 return ERR_PTR(-EAFNOSUPPORT);
 115
 116         dst = afinfo->dst_lookup(tos, saddr, daddr);
 117         xfrm_policy_put_afinfo(afinfo);
 118         return dst;
 119 }
 120
 121 static inline unsigned long make_jiffies(long secs)
 122 {
 123         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 124                 return MAX_SCHEDULE_TIMEOUT-1;
 125         else
 126                 return secs*HZ;
 127 }
 128
 129 static void xfrm_policy_timer(unsigned long data)
 130 {
 131         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 132         unsigned long now = get_seconds();
 133         long next = LONG_MAX;
 134         int warn = 0;
 135         int dir;
 136
 137         read_lock(&xp->lock);
 138
 139         if (xp->dead)
 140                 goto out;
 141
 142         dir = xfrm_policy_id2dir(xp->index);
 143
 144         if (xp->lft.hard_add_expires_seconds) {
 145                 long tmo = xp->lft.hard_add_expires_seconds +
 146                         xp->curlft.add_time - now;
 147                 if (tmo <= 0)
 148                         goto expired;
 149                 if (tmo < next)
 150                         next = tmo;
 151         }
 152         if (xp->lft.hard_use_expires_seconds) {
 153                 long tmo = xp->lft.hard_use_expires_seconds +
 154                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 155                 if (tmo <= 0)
 156                         goto expired;
 157                 if (tmo < next)
 158                         next = tmo;
 159         }
 160         if (xp->lft.soft_add_expires_seconds) {
 161                 long tmo = xp->lft.soft_add_expires_seconds +
 162                         xp->curlft.add_time - now;
 163                 if (tmo <= 0) {
 164                         warn = 1;
 165                         tmo = XFRM_KM_TIMEOUT;
 166                 }
 167                 if (tmo < next)
 168                         next = tmo;
 169         }
 170         if (xp->lft.soft_use_expires_seconds) {
 171                 long tmo = xp->lft.soft_use_expires_seconds +
 172                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 173                 if (tmo <= 0) {
 174                         warn = 1;
 175                         tmo = XFRM_KM_TIMEOUT;
 176                 }
 177                 if (tmo < next)
 178                         next = tmo;
 179         }
 180
 181         if (warn)
 182                 km_policy_expired(xp, dir, 0, 0);
 183         if (next != LONG_MAX &&
 184             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 185                 xfrm_pol_hold(xp);
 186
 187 out:
 188         read_unlock(&xp->lock);
 189         xfrm_pol_put(xp);
 190         return;
 191
 192 expired:
 193         read_unlock(&xp->lock);
 194         if (!xfrm_policy_delete(xp, dir))
 195                 km_policy_expired(xp, dir, 1, 0);
 196         xfrm_pol_put(xp);
 197 }
 198
 199
 200 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 201  * SPD calls.
 202  */
 203
 204 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 205 {
 206         struct xfrm_policy *policy;
 207
 208         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 209
 210         if (policy) {
 211                 INIT_HLIST_NODE(&policy->bydst);
 212                 INIT_HLIST_NODE(&policy->byidx);
 213                 rwlock_init(&policy->lock);
 214                 atomic_set(&policy->refcnt, 1);
 215                 setup_timer(&policy->timer, xfrm_policy_timer,
 216                                 (unsigned long)policy);
 217         }
 218         return policy;
 219 }
 220 EXPORT_SYMBOL(xfrm_policy_alloc);
 221
 222 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 223
 224 void xfrm_policy_destroy(struct xfrm_policy *policy)
 225 {
 226         BUG_ON(!policy->dead);
 227
 228         BUG_ON(policy->bundles);
 229
 230         if (del_timer(&policy->timer))
 231                 BUG();
 232
 233         security_xfrm_policy_free(policy);
 234         kfree(policy);
 235 }
 236 EXPORT_SYMBOL(xfrm_policy_destroy);
 237
 238 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 239 {
 240         struct dst_entry *dst;
 241
 242         while ((dst = policy->bundles) != NULL) {
 243                 policy->bundles = dst->next;
 244                 dst_free(dst);
 245         }
 246
 247         if (del_timer(&policy->timer))
 248                 atomic_dec(&policy->refcnt);
 249
 250         if (atomic_read(&policy->refcnt) > 1)
 251                 flow_cache_flush();
 252
 253         xfrm_pol_put(policy);
 254 }
 255
 256 static void xfrm_policy_gc_task(struct work_struct *work)
 257 {
 258         struct xfrm_policy *policy;
 259         struct hlist_node *entry, *tmp;
 260         struct hlist_head gc_list;
 261
 262         spin_lock_bh(&xfrm_policy_gc_lock);
 263         gc_list.first = xfrm_policy_gc_list.first;
 264         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 265         spin_unlock_bh(&xfrm_policy_gc_lock);
 266
 267         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 268                 xfrm_policy_gc_kill(policy);
 269 }
 270
 271 /* Rule must be locked. Release descentant resources, announce
 272  * entry dead. The rule must be unlinked from lists to the moment.
 273  */
 274
 275 static void xfrm_policy_kill(struct xfrm_policy *policy)
 276 {
 277         int dead;
 278
 279         write_lock_bh(&policy->lock);
 280         dead = policy->dead;
 281         policy->dead = 1;
 282         write_unlock_bh(&policy->lock);
 283
 284         if (unlikely(dead)) {
 285                 WARN_ON(1);
 286                 return;
 287         }
 288
 289         spin_lock(&xfrm_policy_gc_lock);
 290         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 291         spin_unlock(&xfrm_policy_gc_lock);
 292
 293         schedule_work(&xfrm_policy_gc_work);
 294 }
 295
 296 struct xfrm_policy_hash {
 297         struct hlist_head       *table;
 298         unsigned int            hmask;
 299 };
 300
 301 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 302 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 303 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 304 static unsigned int xfrm_idx_hmask __read_mostly;
 305 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 306
 307 static inline unsigned int idx_hash(u32 index)
 308 {
 309         return __idx_hash(index, xfrm_idx_hmask);
 310 }
 311
 312 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 313 {
 314         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 315         unsigned int hash = __sel_hash(sel, family, hmask);
 316
 317         return (hash == hmask + 1 ?
 318                 &xfrm_policy_inexact[dir] :
 319                 xfrm_policy_bydst[dir].table + hash);
 320 }
 321
 322 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 323 {
 324         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 325         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 326
 327         return xfrm_policy_bydst[dir].table + hash;
 328 }
 329
 330 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 331                                    struct hlist_head *ndsttable,
 332                                    unsigned int nhashmask)
 333 {
 334         struct hlist_node *entry, *tmp, *entry0 = NULL;
 335         struct xfrm_policy *pol;
 336         unsigned int h0 = 0;
 337
 338 redo:
 339         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 340                 unsigned int h;
 341
 342                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 343                                 pol->family, nhashmask);
 344                 if (!entry0) {
 345                         hlist_del(entry);
 346                         hlist_add_head(&pol->bydst, ndsttable+h);
 347                         h0 = h;
 348                 } else {
 349                         if (h != h0)
 350                                 continue;
 351                         hlist_del(entry);
 352                         hlist_add_after(entry0, &pol->bydst);
 353                 }
 354                 entry0 = entry;
 355         }
 356         if (!hlist_empty(list)) {
 357                 entry0 = NULL;
 358                 goto redo;
 359         }
 360 }
 361
 362 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 363                                    struct hlist_head *nidxtable,
 364                                    unsigned int nhashmask)
 365 {
 366         struct hlist_node *entry, *tmp;
 367         struct xfrm_policy *pol;
 368
 369         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 370                 unsigned int h;
 371
 372                 h = __idx_hash(pol->index, nhashmask);
 373                 hlist_add_head(&pol->byidx, nidxtable+h);
 374         }
 375 }
 376
 377 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 378 {
 379         return ((old_hmask + 1) << 1) - 1;
 380 }
 381
 382 static void xfrm_bydst_resize(int dir)
 383 {
 384         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 385         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 386         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 387         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 388         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 389         int i;
 390
 391         if (!ndst)
 392                 return;
 393
 394         write_lock_bh(&xfrm_policy_lock);
 395
 396         for (i = hmask; i >= 0; i--)
 397                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 398
 399         xfrm_policy_bydst[dir].table = ndst;
 400         xfrm_policy_bydst[dir].hmask = nhashmask;
 401
 402         write_unlock_bh(&xfrm_policy_lock);
 403
 404         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 405 }
 406
 407 static void xfrm_byidx_resize(int total)
 408 {
 409         unsigned int hmask = xfrm_idx_hmask;
 410         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 411         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 412         struct hlist_head *oidx = xfrm_policy_byidx;
 413         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 414         int i;
 415
 416         if (!nidx)
 417                 return;
 418
 419         write_lock_bh(&xfrm_policy_lock);
 420
 421         for (i = hmask; i >= 0; i--)
 422                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 423
 424         xfrm_policy_byidx = nidx;
 425         xfrm_idx_hmask = nhashmask;
 426
 427         write_unlock_bh(&xfrm_policy_lock);
 428
 429         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 430 }
 431
 432 static inline int xfrm_bydst_should_resize(int dir, int *total)
 433 {
 434         unsigned int cnt = xfrm_policy_count[dir];
 435         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 436
 437         if (total)
 438                 *total += cnt;
 439
 440         if ((hmask + 1) < xfrm_policy_hashmax &&
 441             cnt > hmask)
 442                 return 1;
 443
 444         return 0;
 445 }
 446
 447 static inline int xfrm_byidx_should_resize(int total)
 448 {
 449         unsigned int hmask = xfrm_idx_hmask;
 450
 451         if ((hmask + 1) < xfrm_policy_hashmax &&
 452             total > hmask)
 453                 return 1;
 454
 455         return 0;
 456 }
 457
 458 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 459 {
 460         read_lock_bh(&xfrm_policy_lock);
 461         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 462         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 463         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 464         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 465         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 466         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 467         si->spdhcnt = xfrm_idx_hmask;
 468         si->spdhmcnt = xfrm_policy_hashmax;
 469         read_unlock_bh(&xfrm_policy_lock);
 470 }
 471 EXPORT_SYMBOL(xfrm_spd_getinfo);
 472
 473 static DEFINE_MUTEX(hash_resize_mutex);
 474 static void xfrm_hash_resize(struct work_struct *__unused)
 475 {
 476         int dir, total;
 477
 478         mutex_lock(&hash_resize_mutex);
 479
 480         total = 0;
 481         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 482                 if (xfrm_bydst_should_resize(dir, &total))
 483                         xfrm_bydst_resize(dir);
 484         }
 485         if (xfrm_byidx_should_resize(total))
 486                 xfrm_byidx_resize(total);
 487
 488         mutex_unlock(&hash_resize_mutex);
 489 }
 490
 491 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 492
 493 /* Generate new index... KAME seems to generate them ordered by cost
 494  * of an absolute inpredictability of ordering of rules. This will not pass. */
 495 static u32 xfrm_gen_index(u8 type, int dir)
 496 {
 497         static u32 idx_generator;
 498
 499         for (;;) {
 500                 struct hlist_node *entry;
 501                 struct hlist_head *list;
 502                 struct xfrm_policy *p;
 503                 u32 idx;
 504                 int found;
 505
 506                 idx = (idx_generator | dir);
 507                 idx_generator += 8;
 508                 if (idx == 0)
 509                         idx = 8;
 510                 list = xfrm_policy_byidx + idx_hash(idx);
 511                 found = 0;
 512                 hlist_for_each_entry(p, entry, list, byidx) {
 513                         if (p->index == idx) {
 514                                 found = 1;
 515                                 break;
 516                         }
 517                 }
 518                 if (!found)
 519                         return idx;
 520         }
 521 }
 522
 523 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 524 {
 525         u32 *p1 = (u32 *) s1;
 526         u32 *p2 = (u32 *) s2;
 527         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 528         int i;
 529
 530         for (i = 0; i < len; i++) {
 531                 if (p1[i] != p2[i])
 532                         return 1;
 533         }
 534
 535         return 0;
 536 }
 537
 538 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 539 {
 540         struct xfrm_policy *pol;
 541         struct xfrm_policy *delpol;
 542         struct hlist_head *chain;
 543         struct hlist_node *entry, *newpos;
 544         struct dst_entry *gc_list;
 545
 546         write_lock_bh(&xfrm_policy_lock);
 547         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 548         delpol = NULL;
 549         newpos = NULL;
 550         hlist_for_each_entry(pol, entry, chain, bydst) {
 551                 if (pol->type == policy->type &&
 552                     !selector_cmp(&pol->selector, &policy->selector) &&
 553                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 554                     !WARN_ON(delpol)) {
 555                         if (excl) {
 556                                 write_unlock_bh(&xfrm_policy_lock);
 557                                 return -EEXIST;
 558                         }
 559                         delpol = pol;
 560                         if (policy->priority > pol->priority)
 561                                 continue;
 562                 } else if (policy->priority >= pol->priority) {
 563                         newpos = &pol->bydst;
 564                         continue;
 565                 }
 566                 if (delpol)
 567                         break;
 568         }
 569         if (newpos)
 570                 hlist_add_after(newpos, &policy->bydst);
 571         else
 572                 hlist_add_head(&policy->bydst, chain);
 573         xfrm_pol_hold(policy);
 574         xfrm_policy_count[dir]++;
 575         atomic_inc(&flow_cache_genid);
 576         if (delpol) {
 577                 hlist_del(&delpol->bydst);
 578                 hlist_del(&delpol->byidx);
 579                 xfrm_policy_count[dir]--;
 580         }
 581         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 582         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 583         policy->curlft.add_time = get_seconds();
 584         policy->curlft.use_time = 0;
 585         if (!mod_timer(&policy->timer, jiffies + HZ))
 586                 xfrm_pol_hold(policy);
 587         write_unlock_bh(&xfrm_policy_lock);
 588
 589         if (delpol)
 590                 xfrm_policy_kill(delpol);
 591         else if (xfrm_bydst_should_resize(dir, NULL))
 592                 schedule_work(&xfrm_hash_work);
 593
 594         read_lock_bh(&xfrm_policy_lock);
 595         gc_list = NULL;
 596         entry = &policy->bydst;
 597         hlist_for_each_entry_continue(policy, entry, bydst) {
 598                 struct dst_entry *dst;
 599
 600                 write_lock(&policy->lock);
 601                 dst = policy->bundles;
 602                 if (dst) {
 603                         struct dst_entry *tail = dst;
 604                         while (tail->next)
 605                                 tail = tail->next;
 606                         tail->next = gc_list;
 607                         gc_list = dst;
 608
 609                         policy->bundles = NULL;
 610                 }
 611                 write_unlock(&policy->lock);
 612         }
 613         read_unlock_bh(&xfrm_policy_lock);
 614
 615         while (gc_list) {
 616                 struct dst_entry *dst = gc_list;
 617
 618                 gc_list = dst->next;
 619                 dst_free(dst);
 620         }
 621
 622         return 0;
 623 }
 624 EXPORT_SYMBOL(xfrm_policy_insert);
 625
 626 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 627                                           struct xfrm_selector *sel,
 628                                           struct xfrm_sec_ctx *ctx, int delete,
 629                                           int *err)
 630 {
 631         struct xfrm_policy *pol, *ret;
 632         struct hlist_head *chain;
 633         struct hlist_node *entry;
 634
 635         *err = 0;
 636         write_lock_bh(&xfrm_policy_lock);
 637         chain = policy_hash_bysel(sel, sel->family, dir);
 638         ret = NULL;
 639         hlist_for_each_entry(pol, entry, chain, bydst) {
 640                 if (pol->type == type &&
 641                     !selector_cmp(sel, &pol->selector) &&
 642                     xfrm_sec_ctx_match(ctx, pol->security)) {
 643                         xfrm_pol_hold(pol);
 644                         if (delete) {
 645                                 *err = security_xfrm_policy_delete(pol);
 646                                 if (*err) {
 647                                         write_unlock_bh(&xfrm_policy_lock);
 648                                         return pol;
 649                                 }
 650                                 hlist_del(&pol->bydst);
 651                                 hlist_del(&pol->byidx);
 652                                 xfrm_policy_count[dir]--;
 653                         }
 654                         ret = pol;
 655                         break;
 656                 }
 657         }
 658         write_unlock_bh(&xfrm_policy_lock);
 659
 660         if (ret && delete) {
 661                 atomic_inc(&flow_cache_genid);
 662                 xfrm_policy_kill(ret);
 663         }
 664         return ret;
 665 }
 666 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 667
 668 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 669                                      int *err)
 670 {
 671         struct xfrm_policy *pol, *ret;
 672         struct hlist_head *chain;
 673         struct hlist_node *entry;
 674
 675         *err = -ENOENT;
 676         if (xfrm_policy_id2dir(id) != dir)
 677                 return NULL;
 678
 679         *err = 0;
 680         write_lock_bh(&xfrm_policy_lock);
 681         chain = xfrm_policy_byidx + idx_hash(id);
 682         ret = NULL;
 683         hlist_for_each_entry(pol, entry, chain, byidx) {
 684                 if (pol->type == type && pol->index == id) {
 685                         xfrm_pol_hold(pol);
 686                         if (delete) {
 687                                 *err = security_xfrm_policy_delete(pol);
 688                                 if (*err) {
 689                                         write_unlock_bh(&xfrm_policy_lock);
 690                                         return pol;
 691                                 }
 692                                 hlist_del(&pol->bydst);
 693                                 hlist_del(&pol->byidx);
 694                                 xfrm_policy_count[dir]--;
 695                         }
 696                         ret = pol;
 697                         break;
 698                 }
 699         }
 700         write_unlock_bh(&xfrm_policy_lock);
 701
 702         if (ret && delete) {
 703                 atomic_inc(&flow_cache_genid);
 704                 xfrm_policy_kill(ret);
 705         }
 706         return ret;
 707 }
 708 EXPORT_SYMBOL(xfrm_policy_byid);
 709
 710 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 711 static inline int
 712 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 713 {
 714         int dir, err = 0;
 715
 716         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 717                 struct xfrm_policy *pol;
 718                 struct hlist_node *entry;
 719                 int i;
 720
 721                 hlist_for_each_entry(pol, entry,
 722                                      &xfrm_policy_inexact[dir], bydst) {
 723                         if (pol->type != type)
 724                                 continue;
 725                         err = security_xfrm_policy_delete(pol);
 726                         if (err) {
 727                                 xfrm_audit_policy_delete(pol, 0,
 728                                                          audit_info->loginuid,
 729                                                          audit_info->secid);
 730                                 return err;
 731                         }
 732                 }
 733                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 734                         hlist_for_each_entry(pol, entry,
 735                                              xfrm_policy_bydst[dir].table + i,
 736                                              bydst) {
 737                                 if (pol->type != type)
 738                                         continue;
 739                                 err = security_xfrm_policy_delete(pol);
 740                                 if (err) {
 741                                         xfrm_audit_policy_delete(pol, 0,
 742                                                         audit_info->loginuid,
 743                                                         audit_info->secid);
 744                                         return err;
 745                                 }
 746                         }
 747                 }
 748         }
 749         return err;
 750 }
 751 #else
 752 static inline int
 753 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 754 {
 755         return 0;
 756 }
 757 #endif
 758
 759 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 760 {
 761         int dir, err = 0;
 762
 763         write_lock_bh(&xfrm_policy_lock);
 764
 765         err = xfrm_policy_flush_secctx_check(type, audit_info);
 766         if (err)
 767                 goto out;
 768
 769         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 770                 struct xfrm_policy *pol;
 771                 struct hlist_node *entry;
 772                 int i, killed;
 773
 774                 killed = 0;
 775         again1:
 776                 hlist_for_each_entry(pol, entry,
 777                                      &xfrm_policy_inexact[dir], bydst) {
 778                         if (pol->type != type)
 779                                 continue;
 780                         hlist_del(&pol->bydst);
 781                         hlist_del(&pol->byidx);
 782                         write_unlock_bh(&xfrm_policy_lock);
 783
 784                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 785                                                  audit_info->secid);
 786
 787                         xfrm_policy_kill(pol);
 788                         killed++;
 789
 790                         write_lock_bh(&xfrm_policy_lock);
 791                         goto again1;
 792                 }
 793
 794                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 795         again2:
 796                         hlist_for_each_entry(pol, entry,
 797                                              xfrm_policy_bydst[dir].table + i,
 798                                              bydst) {
 799                                 if (pol->type != type)
 800                                         continue;
 801                                 hlist_del(&pol->bydst);
 802                                 hlist_del(&pol->byidx);
 803                                 write_unlock_bh(&xfrm_policy_lock);
 804
 805                                 xfrm_audit_policy_delete(pol, 1,
 806                                                          audit_info->loginuid,
 807                                                          audit_info->secid);
 808                                 xfrm_policy_kill(pol);
 809                                 killed++;
 810
 811                                 write_lock_bh(&xfrm_policy_lock);
 812                                 goto again2;
 813                         }
 814                 }
 815
 816                 xfrm_policy_count[dir] -= killed;
 817         }
 818         atomic_inc(&flow_cache_genid);
 819 out:
 820         write_unlock_bh(&xfrm_policy_lock);
 821         return err;
 822 }
 823 EXPORT_SYMBOL(xfrm_policy_flush);
 824
 825 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 826                      void *data)
 827 {
 828         struct xfrm_policy *pol, *last = NULL;
 829         struct hlist_node *entry;
 830         int dir, last_dir = 0, count, error;
 831
 832         read_lock_bh(&xfrm_policy_lock);
 833         count = 0;
 834
 835         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 836                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 837                 int i;
 838
 839                 hlist_for_each_entry(pol, entry,
 840                                      &xfrm_policy_inexact[dir], bydst) {
 841                         if (pol->type != type)
 842                                 continue;
 843                         if (last) {
 844                                 error = func(last, last_dir % XFRM_POLICY_MAX,
 845                                              count, data);
 846                                 if (error)
 847                                         goto out;
 848                         }
 849                         last = pol;
 850                         last_dir = dir;
 851                         count++;
 852                 }
 853                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 854                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 855                                 if (pol->type != type)
 856                                         continue;
 857                                 if (last) {
 858                                         error = func(last, last_dir % XFRM_POLICY_MAX,
 859                                                      count, data);
 860                                         if (error)
 861                                                 goto out;
 862                                 }
 863                                 last = pol;
 864                                 last_dir = dir;
 865                                 count++;
 866                         }
 867                 }
 868         }
 869         if (count == 0) {
 870                 error = -ENOENT;
 871                 goto out;
 872         }
 873         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
 874 out:
 875         read_unlock_bh(&xfrm_policy_lock);
 876         return error;
 877 }
 878 EXPORT_SYMBOL(xfrm_policy_walk);
 879
 880 /*
 881  * Find policy to apply to this flow.
 882  *
 883  * Returns 0 if policy found, else an -errno.
 884  */
 885 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 886                              u8 type, u16 family, int dir)
 887 {
 888         struct xfrm_selector *sel = &pol->selector;
 889         int match, ret = -ESRCH;
 890
 891         if (pol->family != family ||
 892             pol->type != type)
 893                 return ret;
 894
 895         match = xfrm_selector_match(sel, fl, family);
 896         if (match)
 897                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
 898
 899         return ret;
 900 }
 901
 902 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 903                                                      u16 family, u8 dir)
 904 {
 905         int err;
 906         struct xfrm_policy *pol, *ret;
 907         xfrm_address_t *daddr, *saddr;
 908         struct hlist_node *entry;
 909         struct hlist_head *chain;
 910         u32 priority = ~0U;
 911
 912         daddr = xfrm_flowi_daddr(fl, family);
 913         saddr = xfrm_flowi_saddr(fl, family);
 914         if (unlikely(!daddr || !saddr))
 915                 return NULL;
 916
 917         read_lock_bh(&xfrm_policy_lock);
 918         chain = policy_hash_direct(daddr, saddr, family, dir);
 919         ret = NULL;
 920         hlist_for_each_entry(pol, entry, chain, bydst) {
 921                 err = xfrm_policy_match(pol, fl, type, family, dir);
 922                 if (err) {
 923                         if (err == -ESRCH)
 924                                 continue;
 925                         else {
 926                                 ret = ERR_PTR(err);
 927                                 goto fail;
 928                         }
 929                 } else {
 930                         ret = pol;
 931                         priority = ret->priority;
 932                         break;
 933                 }
 934         }
 935         chain = &xfrm_policy_inexact[dir];
 936         hlist_for_each_entry(pol, entry, chain, bydst) {
 937                 err = xfrm_policy_match(pol, fl, type, family, dir);
 938                 if (err) {
 939                         if (err == -ESRCH)
 940                                 continue;
 941                         else {
 942                                 ret = ERR_PTR(err);
 943                                 goto fail;
 944                         }
 945                 } else if (pol->priority < priority) {
 946                         ret = pol;
 947                         break;
 948                 }
 949         }
 950         if (ret)
 951                 xfrm_pol_hold(ret);
 952 fail:
 953         read_unlock_bh(&xfrm_policy_lock);
 954
 955         return ret;
 956 }
 957
 958 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 959                                void **objp, atomic_t **obj_refp)
 960 {
 961         struct xfrm_policy *pol;
 962         int err = 0;
 963
 964 #ifdef CONFIG_XFRM_SUB_POLICY
 965         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
 966         if (IS_ERR(pol)) {
 967                 err = PTR_ERR(pol);
 968                 pol = NULL;
 969         }
 970         if (pol || err)
 971                 goto end;
 972 #endif
 973         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 974         if (IS_ERR(pol)) {
 975                 err = PTR_ERR(pol);
 976                 pol = NULL;
 977         }
 978 #ifdef CONFIG_XFRM_SUB_POLICY
 979 end:
 980 #endif
 981         if ((*objp = (void *) pol) != NULL)
 982                 *obj_refp = &pol->refcnt;
 983         return err;
 984 }
 985
 986 static inline int policy_to_flow_dir(int dir)
 987 {
 988         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 989             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 990             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 991                 return dir;
 992         switch (dir) {
 993         default:
 994         case XFRM_POLICY_IN:
 995                 return FLOW_DIR_IN;
 996         case XFRM_POLICY_OUT:
 997                 return FLOW_DIR_OUT;
 998         case XFRM_POLICY_FWD:
 999                 return FLOW_DIR_FWD;
1000         }
1001 }
1002
1003 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1004 {
1005         struct xfrm_policy *pol;
1006
1007         read_lock_bh(&xfrm_policy_lock);
1008         if ((pol = sk->sk_policy[dir]) != NULL) {
1009                 int match = xfrm_selector_match(&pol->selector, fl,
1010                                                 sk->sk_family);
1011                 int err = 0;
1012
1013                 if (match) {
1014                         err = security_xfrm_policy_lookup(pol, fl->secid,
1015                                         policy_to_flow_dir(dir));
1016                         if (!err)
1017                                 xfrm_pol_hold(pol);
1018                         else if (err == -ESRCH)
1019                                 pol = NULL;
1020                         else
1021                                 pol = ERR_PTR(err);
1022                 } else
1023                         pol = NULL;
1024         }
1025         read_unlock_bh(&xfrm_policy_lock);
1026         return pol;
1027 }
1028
1029 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1030 {
1031         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1032                                                      pol->family, dir);
1033
1034         hlist_add_head(&pol->bydst, chain);
1035         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1036         xfrm_policy_count[dir]++;
1037         xfrm_pol_hold(pol);
1038
1039         if (xfrm_bydst_should_resize(dir, NULL))
1040                 schedule_work(&xfrm_hash_work);
1041 }
1042
1043 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1044                                                 int dir)
1045 {
1046         if (hlist_unhashed(&pol->bydst))
1047                 return NULL;
1048
1049         hlist_del(&pol->bydst);
1050         hlist_del(&pol->byidx);
1051         xfrm_policy_count[dir]--;
1052
1053         return pol;
1054 }
1055
1056 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1057 {
1058         write_lock_bh(&xfrm_policy_lock);
1059         pol = __xfrm_policy_unlink(pol, dir);
1060         write_unlock_bh(&xfrm_policy_lock);
1061         if (pol) {
1062                 if (dir < XFRM_POLICY_MAX)
1063                         atomic_inc(&flow_cache_genid);
1064                 xfrm_policy_kill(pol);
1065                 return 0;
1066         }
1067         return -ENOENT;
1068 }
1069 EXPORT_SYMBOL(xfrm_policy_delete);
1070
1071 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1072 {
1073         struct xfrm_policy *old_pol;
1074
1075 #ifdef CONFIG_XFRM_SUB_POLICY
1076         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1077                 return -EINVAL;
1078 #endif
1079
1080         write_lock_bh(&xfrm_policy_lock);
1081         old_pol = sk->sk_policy[dir];
1082         sk->sk_policy[dir] = pol;
1083         if (pol) {
1084                 pol->curlft.add_time = get_seconds();
1085                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1086                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1087         }
1088         if (old_pol)
1089                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1090         write_unlock_bh(&xfrm_policy_lock);
1091
1092         if (old_pol) {
1093                 xfrm_policy_kill(old_pol);
1094         }
1095         return 0;
1096 }
1097
1098 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1099 {
1100         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1101
1102         if (newp) {
1103                 newp->selector = old->selector;
1104                 if (security_xfrm_policy_clone(old, newp)) {
1105                         kfree(newp);
1106                         return NULL;  /* ENOMEM */
1107                 }
1108                 newp->lft = old->lft;
1109                 newp->curlft = old->curlft;
1110                 newp->action = old->action;
1111                 newp->flags = old->flags;
1112                 newp->xfrm_nr = old->xfrm_nr;
1113                 newp->index = old->index;
1114                 newp->type = old->type;
1115                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1116                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1117                 write_lock_bh(&xfrm_policy_lock);
1118                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1119                 write_unlock_bh(&xfrm_policy_lock);
1120                 xfrm_pol_put(newp);
1121         }
1122         return newp;
1123 }
1124
1125 int __xfrm_sk_clone_policy(struct sock *sk)
1126 {
1127         struct xfrm_policy *p0 = sk->sk_policy[0],
1128                            *p1 = sk->sk_policy[1];
1129
1130         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1131         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1132                 return -ENOMEM;
1133         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1134                 return -ENOMEM;
1135         return 0;
1136 }
1137
1138 static int
1139 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1140                unsigned short family)
1141 {
1142         int err;
1143         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1144
1145         if (unlikely(afinfo == NULL))
1146                 return -EINVAL;
1147         err = afinfo->get_saddr(local, remote);
1148         xfrm_policy_put_afinfo(afinfo);
1149         return err;
1150 }
1151
1152 /* Resolve list of templates for the flow, given policy. */
1153
1154 static int
1155 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1156                       struct xfrm_state **xfrm,
1157                       unsigned short family)
1158 {
1159         int nx;
1160         int i, error;
1161         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1162         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1163         xfrm_address_t tmp;
1164
1165         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1166                 struct xfrm_state *x;
1167                 xfrm_address_t *remote = daddr;
1168                 xfrm_address_t *local  = saddr;
1169                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1170
1171                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1172                     tmpl->mode == XFRM_MODE_BEET) {
1173                         remote = &tmpl->id.daddr;
1174                         local = &tmpl->saddr;
1175                         family = tmpl->encap_family;
1176                         if (xfrm_addr_any(local, family)) {
1177                                 error = xfrm_get_saddr(&tmp, remote, family);
1178                                 if (error)
1179                                         goto fail;
1180                                 local = &tmp;
1181                         }
1182                 }
1183
1184                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1185
1186                 if (x && x->km.state == XFRM_STATE_VALID) {
1187                         xfrm[nx++] = x;
1188                         daddr = remote;
1189                         saddr = local;
1190                         continue;
1191                 }
1192                 if (x) {
1193                         error = (x->km.state == XFRM_STATE_ERROR ?
1194                                  -EINVAL : -EAGAIN);
1195                         xfrm_state_put(x);
1196                 }
1197
1198                 if (!tmpl->optional)
1199                         goto fail;
1200         }
1201         return nx;
1202
1203 fail:
1204         for (nx--; nx>=0; nx--)
1205                 xfrm_state_put(xfrm[nx]);
1206         return error;
1207 }
1208
1209 static int
1210 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1211                   struct xfrm_state **xfrm,
1212                   unsigned short family)
1213 {
1214         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1215         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1216         int cnx = 0;
1217         int error;
1218         int ret;
1219         int i;
1220
1221         for (i = 0; i < npols; i++) {
1222                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1223                         error = -ENOBUFS;
1224                         goto fail;
1225                 }
1226
1227                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1228                 if (ret < 0) {
1229                         error = ret;
1230                         goto fail;
1231                 } else
1232                         cnx += ret;
1233         }
1234
1235         /* found states are sorted for outbound processing */
1236         if (npols > 1)
1237                 xfrm_state_sort(xfrm, tpp, cnx, family);
1238
1239         return cnx;
1240
1241  fail:
1242         for (cnx--; cnx>=0; cnx--)
1243                 xfrm_state_put(tpp[cnx]);
1244         return error;
1245
1246 }
1247
1248 /* Check that the bundle accepts the flow and its components are
1249  * still valid.
1250  */
1251
1252 static struct dst_entry *
1253 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1254 {
1255         struct dst_entry *x;
1256         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1257         if (unlikely(afinfo == NULL))
1258                 return ERR_PTR(-EINVAL);
1259         x = afinfo->find_bundle(fl, policy);
1260         xfrm_policy_put_afinfo(afinfo);
1261         return x;
1262 }
1263
1264 static inline int xfrm_get_tos(struct flowi *fl, int family)
1265 {
1266         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1267         int tos;
1268
1269         if (!afinfo)
1270                 return -EINVAL;
1271
1272         tos = afinfo->get_tos(fl);
1273
1274         xfrm_policy_put_afinfo(afinfo);
1275
1276         return tos;
1277 }
1278
1279 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1280 {
1281         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1282         struct xfrm_dst *xdst;
1283
1284         if (!afinfo)
1285                 return ERR_PTR(-EINVAL);
1286
1287         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1288
1289         xfrm_policy_put_afinfo(afinfo);
1290
1291         return xdst;
1292 }
1293
1294 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1295                                  int nfheader_len)
1296 {
1297         struct xfrm_policy_afinfo *afinfo =
1298                 xfrm_policy_get_afinfo(dst->ops->family);
1299         int err;
1300
1301         if (!afinfo)
1302                 return -EINVAL;
1303
1304         err = afinfo->init_path(path, dst, nfheader_len);
1305
1306         xfrm_policy_put_afinfo(afinfo);
1307
1308         return err;
1309 }
1310
1311 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1312 {
1313         struct xfrm_policy_afinfo *afinfo =
1314                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1315         int err;
1316
1317         if (!afinfo)
1318                 return -EINVAL;
1319
1320         err = afinfo->fill_dst(xdst, dev);
1321
1322         xfrm_policy_put_afinfo(afinfo);
1323
1324         return err;
1325 }
1326
1327 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1328  * all the metrics... Shortly, bundle a bundle.
1329  */
1330
1331 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1332                                             struct xfrm_state **xfrm, int nx,
1333                                             struct flowi *fl,
1334                                             struct dst_entry *dst)
1335 {
1336         unsigned long now = jiffies;
1337         struct net_device *dev;
1338         struct dst_entry *dst_prev = NULL;
1339         struct dst_entry *dst0 = NULL;
1340         int i = 0;
1341         int err;
1342         int header_len = 0;
1343         int nfheader_len = 0;
1344         int trailer_len = 0;
1345         int tos;
1346         int family = policy->selector.family;
1347
1348         tos = xfrm_get_tos(fl, family);
1349         err = tos;
1350         if (tos < 0)
1351                 goto put_states;
1352
1353         dst_hold(dst);
1354
1355         for (; i < nx; i++) {
1356                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1357                 struct dst_entry *dst1 = &xdst->u.dst;
1358
1359                 err = PTR_ERR(xdst);
1360                 if (IS_ERR(xdst)) {
1361                         dst_release(dst);
1362                         goto put_states;
1363                 }
1364
1365                 if (!dst_prev)
1366                         dst0 = dst1;
1367                 else {
1368                         dst_prev->child = dst_clone(dst1);
1369                         dst1->flags |= DST_NOHASH;
1370                 }
1371
1372                 xdst->route = dst;
1373                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1374
1375                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1376                         family = xfrm[i]->props.family;
1377                         dst = xfrm_dst_lookup(xfrm[i], tos, family);
1378                         err = PTR_ERR(dst);
1379                         if (IS_ERR(dst))
1380                                 goto put_states;
1381                 } else
1382                         dst_hold(dst);
1383
1384                 dst1->xfrm = xfrm[i];
1385                 xdst->genid = xfrm[i]->genid;
1386
1387                 dst1->obsolete = -1;
1388                 dst1->flags |= DST_HOST;
1389                 dst1->lastuse = now;
1390
1391                 dst1->input = dst_discard;
1392                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1393
1394                 dst1->next = dst_prev;
1395                 dst_prev = dst1;
1396
1397                 header_len += xfrm[i]->props.header_len;
1398                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1399                         nfheader_len += xfrm[i]->props.header_len;
1400                 trailer_len += xfrm[i]->props.trailer_len;
1401         }
1402
1403         dst_prev->child = dst;
1404         dst0->path = dst;
1405
1406         err = -ENODEV;
1407         dev = dst->dev;
1408         if (!dev)
1409                 goto free_dst;
1410
1411         /* Copy neighbout for reachability confirmation */
1412         dst0->neighbour = neigh_clone(dst->neighbour);
1413
1414         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1415         xfrm_init_pmtu(dst_prev);
1416
1417         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1418                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1419
1420                 err = xfrm_fill_dst(xdst, dev);
1421                 if (err)
1422                         goto free_dst;
1423
1424                 dst_prev->header_len = header_len;
1425                 dst_prev->trailer_len = trailer_len;
1426                 header_len -= xdst->u.dst.xfrm->props.header_len;
1427                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1428         }
1429
1430 out:
1431         return dst0;
1432
1433 put_states:
1434         for (; i < nx; i++)
1435                 xfrm_state_put(xfrm[i]);
1436 free_dst:
1437         if (dst0)
1438                 dst_free(dst0);
1439         dst0 = ERR_PTR(err);
1440         goto out;
1441 }
1442
1443 static int inline
1444 xfrm_dst_alloc_copy(void **target, void *src, int size)
1445 {
1446         if (!*target) {
1447                 *target = kmalloc(size, GFP_ATOMIC);
1448                 if (!*target)
1449                         return -ENOMEM;
1450         }
1451         memcpy(*target, src, size);
1452         return 0;
1453 }
1454
1455 static int inline
1456 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1457 {
1458 #ifdef CONFIG_XFRM_SUB_POLICY
1459         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1460         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1461                                    sel, sizeof(*sel));
1462 #else
1463         return 0;
1464 #endif
1465 }
1466
1467 static int inline
1468 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1469 {
1470 #ifdef CONFIG_XFRM_SUB_POLICY
1471         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1472         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1473 #else
1474         return 0;
1475 #endif
1476 }
1477
1478 static int stale_bundle(struct dst_entry *dst);
1479
1480 /* Main function: finds/creates a bundle for given flow.
1481  *
1482  * At the moment we eat a raw IP route. Mostly to speed up lookups
1483  * on interfaces with disabled IPsec.
1484  */
1485 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1486                   struct sock *sk, int flags)
1487 {
1488         struct xfrm_policy *policy;
1489         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1490         int npols;
1491         int pol_dead;
1492         int xfrm_nr;
1493         int pi;
1494         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1495         struct dst_entry *dst, *dst_orig = *dst_p;
1496         int nx = 0;
1497         int err;
1498         u32 genid;
1499         u16 family;
1500         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1501
1502 restart:
1503         genid = atomic_read(&flow_cache_genid);
1504         policy = NULL;
1505         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1506                 pols[pi] = NULL;
1507         npols = 0;
1508         pol_dead = 0;
1509         xfrm_nr = 0;
1510
1511         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1512                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1513                 err = PTR_ERR(policy);
1514                 if (IS_ERR(policy)) {
1515                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1516                         goto dropdst;
1517                 }
1518         }
1519
1520         if (!policy) {
1521                 /* To accelerate a bit...  */
1522                 if ((dst_orig->flags & DST_NOXFRM) ||
1523                     !xfrm_policy_count[XFRM_POLICY_OUT])
1524                         goto nopol;
1525
1526                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1527                                            dir, xfrm_policy_lookup);
1528                 err = PTR_ERR(policy);
1529                 if (IS_ERR(policy)) {
1530                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1531                         goto dropdst;
1532                 }
1533         }
1534
1535         if (!policy)
1536                 goto nopol;
1537
1538         family = dst_orig->ops->family;
1539         pols[0] = policy;
1540         npols ++;
1541         xfrm_nr += pols[0]->xfrm_nr;
1542
1543         err = -ENOENT;
1544         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1545                 goto error;
1546
1547         policy->curlft.use_time = get_seconds();
1548
1549         switch (policy->action) {
1550         default:
1551         case XFRM_POLICY_BLOCK:
1552                 /* Prohibit the flow */
1553                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1554                 err = -EPERM;
1555                 goto error;
1556
1557         case XFRM_POLICY_ALLOW:
1558 #ifndef CONFIG_XFRM_SUB_POLICY
1559                 if (policy->xfrm_nr == 0) {
1560                         /* Flow passes not transformed. */
1561                         xfrm_pol_put(policy);
1562                         return 0;
1563                 }
1564 #endif
1565
1566                 /* Try to find matching bundle.
1567                  *
1568                  * LATER: help from flow cache. It is optional, this
1569                  * is required only for output policy.
1570                  */
1571                 dst = xfrm_find_bundle(fl, policy, family);
1572                 if (IS_ERR(dst)) {
1573                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1574                         err = PTR_ERR(dst);
1575                         goto error;
1576                 }
1577
1578                 if (dst)
1579                         break;
1580
1581 #ifdef CONFIG_XFRM_SUB_POLICY
1582                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1583                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1584                                                             fl, family,
1585                                                             XFRM_POLICY_OUT);
1586                         if (pols[1]) {
1587                                 if (IS_ERR(pols[1])) {
1588                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1589                                         err = PTR_ERR(pols[1]);
1590                                         goto error;
1591                                 }
1592                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1593                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1594                                         err = -EPERM;
1595                                         goto error;
1596                                 }
1597                                 npols ++;
1598                                 xfrm_nr += pols[1]->xfrm_nr;
1599                         }
1600                 }
1601
1602                 /*
1603                  * Because neither flowi nor bundle information knows about
1604                  * transformation template size. On more than one policy usage
1605                  * we can realize whether all of them is bypass or not after
1606                  * they are searched. See above not-transformed bypass
1607                  * is surrounded by non-sub policy configuration, too.
1608                  */
1609                 if (xfrm_nr == 0) {
1610                         /* Flow passes not transformed. */
1611                         xfrm_pols_put(pols, npols);
1612                         return 0;
1613                 }
1614
1615 #endif
1616                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1617
1618                 if (unlikely(nx<0)) {
1619                         err = nx;
1620                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1621                                 /* EREMOTE tells the caller to generate
1622                                  * a one-shot blackhole route.
1623                                  */
1624                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1625                                 xfrm_pol_put(policy);
1626                                 return -EREMOTE;
1627                         }
1628                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1629                                 DECLARE_WAITQUEUE(wait, current);
1630
1631                                 add_wait_queue(&km_waitq, &wait);
1632                                 set_current_state(TASK_INTERRUPTIBLE);
1633                                 schedule();
1634                                 set_current_state(TASK_RUNNING);
1635                                 remove_wait_queue(&km_waitq, &wait);
1636
1637                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1638
1639                                 if (nx == -EAGAIN && signal_pending(current)) {
1640                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1641                                         err = -ERESTART;
1642                                         goto error;
1643                                 }
1644                                 if (nx == -EAGAIN ||
1645                                     genid != atomic_read(&flow_cache_genid)) {
1646                                         xfrm_pols_put(pols, npols);
1647                                         goto restart;
1648                                 }
1649                                 err = nx;
1650                         }
1651                         if (err < 0) {
1652                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1653                                 goto error;
1654                         }
1655                 }
1656                 if (nx == 0) {
1657                         /* Flow passes not transformed. */
1658                         xfrm_pols_put(pols, npols);
1659                         return 0;
1660                 }
1661
1662                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1663                 err = PTR_ERR(dst);
1664                 if (IS_ERR(dst)) {
1665                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1666                         goto error;
1667                 }
1668
1669                 for (pi = 0; pi < npols; pi++) {
1670                         read_lock_bh(&pols[pi]->lock);
1671                         pol_dead |= pols[pi]->dead;
1672                         read_unlock_bh(&pols[pi]->lock);
1673                 }
1674
1675                 write_lock_bh(&policy->lock);
1676                 if (unlikely(pol_dead || stale_bundle(dst))) {
1677                         /* Wow! While we worked on resolving, this
1678                          * policy has gone. Retry. It is not paranoia,
1679                          * we just cannot enlist new bundle to dead object.
1680                          * We can't enlist stable bundles either.
1681                          */
1682                         write_unlock_bh(&policy->lock);
1683                         if (dst)
1684                                 dst_free(dst);
1685
1686                         if (pol_dead)
1687                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1688                         else
1689                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1690                         err = -EHOSTUNREACH;
1691                         goto error;
1692                 }
1693
1694                 if (npols > 1)
1695                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1696                 else
1697                         err = xfrm_dst_update_origin(dst, fl);
1698                 if (unlikely(err)) {
1699                         write_unlock_bh(&policy->lock);
1700                         if (dst)
1701                                 dst_free(dst);
1702                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1703                         goto error;
1704                 }
1705
1706                 dst->next = policy->bundles;
1707                 policy->bundles = dst;
1708                 dst_hold(dst);
1709                 write_unlock_bh(&policy->lock);
1710         }
1711         *dst_p = dst;
1712         dst_release(dst_orig);
1713         xfrm_pols_put(pols, npols);
1714         return 0;
1715
1716 error:
1717         xfrm_pols_put(pols, npols);
1718 dropdst:
1719         dst_release(dst_orig);
1720         *dst_p = NULL;
1721         return err;
1722
1723 nopol:
1724         err = -ENOENT;
1725         if (flags & XFRM_LOOKUP_ICMP)
1726                 goto dropdst;
1727         return 0;
1728 }
1729 EXPORT_SYMBOL(__xfrm_lookup);
1730
1731 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1732                 struct sock *sk, int flags)
1733 {
1734         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1735
1736         if (err == -EREMOTE) {
1737                 dst_release(*dst_p);
1738                 *dst_p = NULL;
1739                 err = -EAGAIN;
1740         }
1741
1742         return err;
1743 }
1744 EXPORT_SYMBOL(xfrm_lookup);
1745
1746 static inline int
1747 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1748 {
1749         struct xfrm_state *x;
1750
1751         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1752                 return 0;
1753         x = skb->sp->xvec[idx];
1754         if (!x->type->reject)
1755                 return 0;
1756         return x->type->reject(x, skb, fl);
1757 }
1758
1759 /* When skb is transformed back to its "native" form, we have to
1760  * check policy restrictions. At the moment we make this in maximally
1761  * stupid way. Shame on me. :-) Of course, connected sockets must
1762  * have policy cached at them.
1763  */
1764
1765 static inline int
1766 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1767               unsigned short family)
1768 {
1769         if (xfrm_state_kern(x))
1770                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1771         return  x->id.proto == tmpl->id.proto &&
1772                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1773                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1774                 x->props.mode == tmpl->mode &&
1775                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1776                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1777                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1778                   xfrm_state_addr_cmp(tmpl, x, family));
1779 }
1780
1781 /*
1782  * 0 or more than 0 is returned when validation is succeeded (either bypass
1783  * because of optional transport mode, or next index of the mathced secpath
1784  * state with the template.
1785  * -1 is returned when no matching template is found.
1786  * Otherwise "-2 - errored_index" is returned.
1787  */
1788 static inline int
1789 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1790                unsigned short family)
1791 {
1792         int idx = start;
1793
1794         if (tmpl->optional) {
1795                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1796                         return start;
1797         } else
1798                 start = -1;
1799         for (; idx < sp->len; idx++) {
1800                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1801                         return ++idx;
1802                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1803                         if (start == -1)
1804                                 start = -2-idx;
1805                         break;
1806                 }
1807         }
1808         return start;
1809 }
1810
1811 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1812                           unsigned int family, int reverse)
1813 {
1814         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1815         int err;
1816
1817         if (unlikely(afinfo == NULL))
1818                 return -EAFNOSUPPORT;
1819
1820         afinfo->decode_session(skb, fl, reverse);
1821         err = security_xfrm_decode_session(skb, &fl->secid);
1822         xfrm_policy_put_afinfo(afinfo);
1823         return err;
1824 }
1825 EXPORT_SYMBOL(__xfrm_decode_session);
1826
1827 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1828 {
1829         for (; k < sp->len; k++) {
1830                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1831                         *idxp = k;
1832                         return 1;
1833                 }
1834         }
1835
1836         return 0;
1837 }
1838
1839 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1840                         unsigned short family)
1841 {
1842         struct xfrm_policy *pol;
1843         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1844         int npols = 0;
1845         int xfrm_nr;
1846         int pi;
1847         int reverse;
1848         struct flowi fl;
1849         u8 fl_dir;
1850         int xerr_idx = -1;
1851
1852         reverse = dir & ~XFRM_POLICY_MASK;
1853         dir &= XFRM_POLICY_MASK;
1854         fl_dir = policy_to_flow_dir(dir);
1855
1856         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1857                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1858                 return 0;
1859         }
1860
1861         nf_nat_decode_session(skb, &fl, family);
1862
1863         /* First, check used SA against their selectors. */
1864         if (skb->sp) {
1865                 int i;
1866
1867                 for (i=skb->sp->len-1; i>=0; i--) {
1868                         struct xfrm_state *x = skb->sp->xvec[i];
1869                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1870                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1871                                 return 0;
1872                         }
1873                 }
1874         }
1875
1876         pol = NULL;
1877         if (sk && sk->sk_policy[dir]) {
1878                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1879                 if (IS_ERR(pol)) {
1880                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1881                         return 0;
1882                 }
1883         }
1884
1885         if (!pol)
1886                 pol = flow_cache_lookup(&fl, family, fl_dir,
1887                                         xfrm_policy_lookup);
1888
1889         if (IS_ERR(pol)) {
1890                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1891                 return 0;
1892         }
1893
1894         if (!pol) {
1895                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1896                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1897                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1898                         return 0;
1899                 }
1900                 return 1;
1901         }
1902
1903         pol->curlft.use_time = get_seconds();
1904
1905         pols[0] = pol;
1906         npols ++;
1907 #ifdef CONFIG_XFRM_SUB_POLICY
1908         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1909                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1910                                                     &fl, family,
1911                                                     XFRM_POLICY_IN);
1912                 if (pols[1]) {
1913                         if (IS_ERR(pols[1])) {
1914                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1915                                 return 0;
1916                         }
1917                         pols[1]->curlft.use_time = get_seconds();
1918                         npols ++;
1919                 }
1920         }
1921 #endif
1922
1923         if (pol->action == XFRM_POLICY_ALLOW) {
1924                 struct sec_path *sp;
1925                 static struct sec_path dummy;
1926                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1927                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1928                 struct xfrm_tmpl **tpp = tp;
1929                 int ti = 0;
1930                 int i, k;
1931
1932                 if ((sp = skb->sp) == NULL)
1933                         sp = &dummy;
1934
1935                 for (pi = 0; pi < npols; pi++) {
1936                         if (pols[pi] != pol &&
1937                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1938                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1939                                 goto reject;
1940                         }
1941                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1942                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1943                                 goto reject_error;
1944                         }
1945                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1946                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1947                 }
1948                 xfrm_nr = ti;
1949                 if (npols > 1) {
1950                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1951                         tpp = stp;
1952                 }
1953
1954                 /* For each tunnel xfrm, find the first matching tmpl.
1955                  * For each tmpl before that, find corresponding xfrm.
1956                  * Order is _important_. Later we will implement
1957                  * some barriers, but at the moment barriers
1958                  * are implied between each two transformations.
1959                  */
1960                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1961                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1962                         if (k < 0) {
1963                                 if (k < -1)
1964                                         /* "-2 - errored_index" returned */
1965                                         xerr_idx = -(2+k);
1966                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
1967                                 goto reject;
1968                         }
1969                 }
1970
1971                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
1972                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
1973                         goto reject;
1974                 }
1975
1976                 xfrm_pols_put(pols, npols);
1977                 return 1;
1978         }
1979         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1980
1981 reject:
1982         xfrm_secpath_reject(xerr_idx, skb, &fl);
1983 reject_error:
1984         xfrm_pols_put(pols, npols);
1985         return 0;
1986 }
1987 EXPORT_SYMBOL(__xfrm_policy_check);
1988
1989 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1990 {
1991         struct flowi fl;
1992
1993         if (xfrm_decode_session(skb, &fl, family) < 0) {
1994                 /* XXX: we should have something like FWDHDRERROR here. */
1995                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1996                 return 0;
1997         }
1998
1999         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2000 }
2001 EXPORT_SYMBOL(__xfrm_route_forward);
2002
2003 /* Optimize later using cookies and generation ids. */
2004
2005 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2006 {
2007         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2008          * to "-1" to force all XFRM destinations to get validated by
2009          * dst_ops->check on every use.  We do this because when a
2010          * normal route referenced by an XFRM dst is obsoleted we do
2011          * not go looking around for all parent referencing XFRM dsts
2012          * so that we can invalidate them.  It is just too much work.
2013          * Instead we make the checks here on every use.  For example:
2014          *
2015          *      XFRM dst A --> IPv4 dst X
2016          *
2017          * X is the "xdst->route" of A (X is also the "dst->path" of A
2018          * in this example).  If X is marked obsolete, "A" will not
2019          * notice.  That's what we are validating here via the
2020          * stale_bundle() check.
2021          *
2022          * When a policy's bundle is pruned, we dst_free() the XFRM
2023          * dst which causes it's ->obsolete field to be set to a
2024          * positive non-zero integer.  If an XFRM dst has been pruned
2025          * like this, we want to force a new route lookup.
2026          */
2027         if (dst->obsolete < 0 && !stale_bundle(dst))
2028                 return dst;
2029
2030         return NULL;
2031 }
2032
2033 static int stale_bundle(struct dst_entry *dst)
2034 {
2035         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2036 }
2037
2038 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2039 {
2040         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2041                 dst->dev = dev->nd_net->loopback_dev;
2042                 dev_hold(dst->dev);
2043                 dev_put(dev);
2044         }
2045 }
2046 EXPORT_SYMBOL(xfrm_dst_ifdown);
2047
2048 static void xfrm_link_failure(struct sk_buff *skb)
2049 {
2050         /* Impossible. Such dst must be popped before reaches point of failure. */
2051         return;
2052 }
2053
2054 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2055 {
2056         if (dst) {
2057                 if (dst->obsolete) {
2058                         dst_release(dst);
2059                         dst = NULL;
2060                 }
2061         }
2062         return dst;
2063 }
2064
2065 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2066 {
2067         struct dst_entry *dst, **dstp;
2068
2069         write_lock(&pol->lock);
2070         dstp = &pol->bundles;
2071         while ((dst=*dstp) != NULL) {
2072                 if (func(dst)) {
2073                         *dstp = dst->next;
2074                         dst->next = *gc_list_p;
2075                         *gc_list_p = dst;
2076                 } else {
2077                         dstp = &dst->next;
2078                 }
2079         }
2080         write_unlock(&pol->lock);
2081 }
2082
2083 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2084 {
2085         struct dst_entry *gc_list = NULL;
2086         int dir;
2087
2088         read_lock_bh(&xfrm_policy_lock);
2089         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2090                 struct xfrm_policy *pol;
2091                 struct hlist_node *entry;
2092                 struct hlist_head *table;
2093                 int i;
2094
2095                 hlist_for_each_entry(pol, entry,
2096                                      &xfrm_policy_inexact[dir], bydst)
2097                         prune_one_bundle(pol, func, &gc_list);
2098
2099                 table = xfrm_policy_bydst[dir].table;
2100                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2101                         hlist_for_each_entry(pol, entry, table + i, bydst)
2102                                 prune_one_bundle(pol, func, &gc_list);
2103                 }
2104         }
2105         read_unlock_bh(&xfrm_policy_lock);
2106
2107         while (gc_list) {
2108                 struct dst_entry *dst = gc_list;
2109                 gc_list = dst->next;
2110                 dst_free(dst);
2111         }
2112 }
2113
2114 static int unused_bundle(struct dst_entry *dst)
2115 {
2116         return !atomic_read(&dst->__refcnt);
2117 }
2118
2119 static void __xfrm_garbage_collect(void)
2120 {
2121         xfrm_prune_bundles(unused_bundle);
2122 }
2123
2124 static int xfrm_flush_bundles(void)
2125 {
2126         xfrm_prune_bundles(stale_bundle);
2127         return 0;
2128 }
2129
2130 static void xfrm_init_pmtu(struct dst_entry *dst)
2131 {
2132         do {
2133                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2134                 u32 pmtu, route_mtu_cached;
2135
2136                 pmtu = dst_mtu(dst->child);
2137                 xdst->child_mtu_cached = pmtu;
2138
2139                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2140
2141                 route_mtu_cached = dst_mtu(xdst->route);
2142                 xdst->route_mtu_cached = route_mtu_cached;
2143
2144                 if (pmtu > route_mtu_cached)
2145                         pmtu = route_mtu_cached;
2146
2147                 dst->metrics[RTAX_MTU-1] = pmtu;
2148         } while ((dst = dst->next));
2149 }
2150
2151 /* Check that the bundle accepts the flow and its components are
2152  * still valid.
2153  */
2154
2155 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2156                 struct flowi *fl, int family, int strict)
2157 {
2158         struct dst_entry *dst = &first->u.dst;
2159         struct xfrm_dst *last;
2160         u32 mtu;
2161
2162         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2163             (dst->dev && !netif_running(dst->dev)))
2164                 return 0;
2165 #ifdef CONFIG_XFRM_SUB_POLICY
2166         if (fl) {
2167                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2168                         return 0;
2169                 if (first->partner &&
2170                     !xfrm_selector_match(first->partner, fl, family))
2171                         return 0;
2172         }
2173 #endif
2174
2175         last = NULL;
2176
2177         do {
2178                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2179
2180                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2181                         return 0;
2182                 if (fl && pol &&
2183                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2184                         return 0;
2185                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2186                         return 0;
2187                 if (xdst->genid != dst->xfrm->genid)
2188                         return 0;
2189
2190                 if (strict && fl &&
2191                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2192                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2193                         return 0;
2194
2195                 mtu = dst_mtu(dst->child);
2196                 if (xdst->child_mtu_cached != mtu) {
2197                         last = xdst;
2198                         xdst->child_mtu_cached = mtu;
2199                 }
2200
2201                 if (!dst_check(xdst->route, xdst->route_cookie))
2202                         return 0;
2203                 mtu = dst_mtu(xdst->route);
2204                 if (xdst->route_mtu_cached != mtu) {
2205                         last = xdst;
2206                         xdst->route_mtu_cached = mtu;
2207                 }
2208
2209                 dst = dst->child;
2210         } while (dst->xfrm);
2211
2212         if (likely(!last))
2213                 return 1;
2214
2215         mtu = last->child_mtu_cached;
2216         for (;;) {
2217                 dst = &last->u.dst;
2218
2219                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2220                 if (mtu > last->route_mtu_cached)
2221                         mtu = last->route_mtu_cached;
2222                 dst->metrics[RTAX_MTU-1] = mtu;
2223
2224                 if (last == first)
2225                         break;
2226
2227                 last = (struct xfrm_dst *)last->u.dst.next;
2228                 last->child_mtu_cached = mtu;
2229         }
2230
2231         return 1;
2232 }
2233
2234 EXPORT_SYMBOL(xfrm_bundle_ok);
2235
2236 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2237 {
2238         int err = 0;
2239         if (unlikely(afinfo == NULL))
2240                 return -EINVAL;
2241         if (unlikely(afinfo->family >= NPROTO))
2242                 return -EAFNOSUPPORT;
2243         write_lock_bh(&xfrm_policy_afinfo_lock);
2244         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2245                 err = -ENOBUFS;
2246         else {
2247                 struct dst_ops *dst_ops = afinfo->dst_ops;
2248                 if (likely(dst_ops->kmem_cachep == NULL))
2249                         dst_ops->kmem_cachep = xfrm_dst_cache;
2250                 if (likely(dst_ops->check == NULL))
2251                         dst_ops->check = xfrm_dst_check;
2252                 if (likely(dst_ops->negative_advice == NULL))
2253                         dst_ops->negative_advice = xfrm_negative_advice;
2254                 if (likely(dst_ops->link_failure == NULL))
2255                         dst_ops->link_failure = xfrm_link_failure;
2256                 if (likely(afinfo->garbage_collect == NULL))
2257                         afinfo->garbage_collect = __xfrm_garbage_collect;
2258                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2259         }
2260         write_unlock_bh(&xfrm_policy_afinfo_lock);
2261         return err;
2262 }
2263 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2264
2265 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2266 {
2267         int err = 0;
2268         if (unlikely(afinfo == NULL))
2269                 return -EINVAL;
2270         if (unlikely(afinfo->family >= NPROTO))
2271                 return -EAFNOSUPPORT;
2272         write_lock_bh(&xfrm_policy_afinfo_lock);
2273         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2274                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2275                         err = -EINVAL;
2276                 else {
2277                         struct dst_ops *dst_ops = afinfo->dst_ops;
2278                         xfrm_policy_afinfo[afinfo->family] = NULL;
2279                         dst_ops->kmem_cachep = NULL;
2280                         dst_ops->check = NULL;
2281                         dst_ops->negative_advice = NULL;
2282                         dst_ops->link_failure = NULL;
2283                         afinfo->garbage_collect = NULL;
2284                 }
2285         }
2286         write_unlock_bh(&xfrm_policy_afinfo_lock);
2287         return err;
2288 }
2289 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2290
2291 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2292 {
2293         struct xfrm_policy_afinfo *afinfo;
2294         if (unlikely(family >= NPROTO))
2295                 return NULL;
2296         read_lock(&xfrm_policy_afinfo_lock);
2297         afinfo = xfrm_policy_afinfo[family];
2298         if (unlikely(!afinfo))
2299                 read_unlock(&xfrm_policy_afinfo_lock);
2300         return afinfo;
2301 }
2302
2303 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2304 {
2305         read_unlock(&xfrm_policy_afinfo_lock);
2306 }
2307
2308 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2309 {
2310         struct net_device *dev = ptr;
2311
2312         if (dev->nd_net != &init_net)
2313                 return NOTIFY_DONE;
2314
2315         switch (event) {
2316         case NETDEV_DOWN:
2317                 xfrm_flush_bundles();
2318         }
2319         return NOTIFY_DONE;
2320 }
2321
2322 static struct notifier_block xfrm_dev_notifier = {
2323         xfrm_dev_event,
2324         NULL,
2325         0
2326 };
2327
2328 #ifdef CONFIG_XFRM_STATISTICS
2329 static int __init xfrm_statistics_init(void)
2330 {
2331         if (snmp_mib_init((void **)xfrm_statistics,
2332                           sizeof(struct linux_xfrm_mib)) < 0)
2333                 return -ENOMEM;
2334         return 0;
2335 }
2336 #endif
2337
2338 static void __init xfrm_policy_init(void)
2339 {
2340         unsigned int hmask, sz;
2341         int dir;
2342
2343         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2344                                            sizeof(struct xfrm_dst),
2345                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2346                                            NULL);
2347
2348         hmask = 8 - 1;
2349         sz = (hmask+1) * sizeof(struct hlist_head);
2350
2351         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2352         xfrm_idx_hmask = hmask;
2353         if (!xfrm_policy_byidx)
2354                 panic("XFRM: failed to allocate byidx hash\n");
2355
2356         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2357                 struct xfrm_policy_hash *htab;
2358
2359                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2360
2361                 htab = &xfrm_policy_bydst[dir];
2362                 htab->table = xfrm_hash_alloc(sz);
2363                 htab->hmask = hmask;
2364                 if (!htab->table)
2365                         panic("XFRM: failed to allocate bydst hash\n");
2366         }
2367
2368         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2369         register_netdevice_notifier(&xfrm_dev_notifier);
2370 }
2371
2372 void __init xfrm_init(void)
2373 {
2374 #ifdef CONFIG_XFRM_STATISTICS
2375         xfrm_statistics_init();
2376 #endif
2377         xfrm_state_init();
2378         xfrm_policy_init();
2379         xfrm_input_init();
2380 #ifdef CONFIG_XFRM_STATISTICS
2381         xfrm_proc_init();
2382 #endif
2383 }
2384
2385 #ifdef CONFIG_AUDITSYSCALL
2386 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2387                                          struct audit_buffer *audit_buf)
2388 {
2389         struct xfrm_sec_ctx *ctx = xp->security;
2390         struct xfrm_selector *sel = &xp->selector;
2391
2392         if (ctx)
2393                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2394                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2395
2396         switch(sel->family) {
2397         case AF_INET:
2398                 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2399                                  NIPQUAD(sel->saddr.a4));
2400                 if (sel->prefixlen_s != 32)
2401                         audit_log_format(audit_buf, " src_prefixlen=%d",
2402                                          sel->prefixlen_s);
2403                 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2404                                  NIPQUAD(sel->daddr.a4));
2405                 if (sel->prefixlen_d != 32)
2406                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2407                                          sel->prefixlen_d);
2408                 break;
2409         case AF_INET6:
2410                 audit_log_format(audit_buf, " src=" NIP6_FMT,
2411                                  NIP6(*(struct in6_addr *)sel->saddr.a6));
2412                 if (sel->prefixlen_s != 128)
2413                         audit_log_format(audit_buf, " src_prefixlen=%d",
2414                                          sel->prefixlen_s);
2415                 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2416                                  NIP6(*(struct in6_addr *)sel->daddr.a6));
2417                 if (sel->prefixlen_d != 128)
2418                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2419                                          sel->prefixlen_d);
2420                 break;
2421         }
2422 }
2423
2424 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2425                            u32 auid, u32 secid)
2426 {
2427         struct audit_buffer *audit_buf;
2428
2429         audit_buf = xfrm_audit_start("SPD-add");
2430         if (audit_buf == NULL)
2431                 return;
2432         xfrm_audit_helper_usrinfo(auid, secid, audit_buf);
2433         audit_log_format(audit_buf, " res=%u", result);
2434         xfrm_audit_common_policyinfo(xp, audit_buf);
2435         audit_log_end(audit_buf);
2436 }
2437 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2438
2439 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2440                               u32 auid, u32 secid)
2441 {
2442         struct audit_buffer *audit_buf;
2443
2444         audit_buf = xfrm_audit_start("SPD-delete");
2445         if (audit_buf == NULL)
2446                 return;
2447         xfrm_audit_helper_usrinfo(auid, secid, audit_buf);
2448         audit_log_format(audit_buf, " res=%u", result);
2449         xfrm_audit_common_policyinfo(xp, audit_buf);
2450         audit_log_end(audit_buf);
2451 }
2452 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2453 #endif
2454
2455 #ifdef CONFIG_XFRM_MIGRATE
2456 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2457                                        struct xfrm_selector *sel_tgt)
2458 {
2459         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2460                 if (sel_tgt->family == sel_cmp->family &&
2461                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2462                                   sel_cmp->family) == 0 &&
2463                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2464                                   sel_cmp->family) == 0 &&
2465                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2466                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2467                         return 1;
2468                 }
2469         } else {
2470                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2471                         return 1;
2472                 }
2473         }
2474         return 0;
2475 }
2476
2477 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2478                                                      u8 dir, u8 type)
2479 {
2480         struct xfrm_policy *pol, *ret = NULL;
2481         struct hlist_node *entry;
2482         struct hlist_head *chain;
2483         u32 priority = ~0U;
2484
2485         read_lock_bh(&xfrm_policy_lock);
2486         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2487         hlist_for_each_entry(pol, entry, chain, bydst) {
2488                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2489                     pol->type == type) {
2490                         ret = pol;
2491                         priority = ret->priority;
2492                         break;
2493                 }
2494         }
2495         chain = &xfrm_policy_inexact[dir];
2496         hlist_for_each_entry(pol, entry, chain, bydst) {
2497                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2498                     pol->type == type &&
2499                     pol->priority < priority) {
2500                         ret = pol;
2501                         break;
2502                 }
2503         }
2504
2505         if (ret)
2506                 xfrm_pol_hold(ret);
2507
2508         read_unlock_bh(&xfrm_policy_lock);
2509
2510         return ret;
2511 }
2512
2513 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2514 {
2515         int match = 0;
2516
2517         if (t->mode == m->mode && t->id.proto == m->proto &&
2518             (m->reqid == 0 || t->reqid == m->reqid)) {
2519                 switch (t->mode) {
2520                 case XFRM_MODE_TUNNEL:
2521                 case XFRM_MODE_BEET:
2522                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2523                                           m->old_family) == 0 &&
2524                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2525                                           m->old_family) == 0) {
2526                                 match = 1;
2527                         }
2528                         break;
2529                 case XFRM_MODE_TRANSPORT:
2530                         /* in case of transport mode, template does not store
2531                            any IP addresses, hence we just compare mode and
2532                            protocol */
2533                         match = 1;
2534                         break;
2535                 default:
2536                         break;
2537                 }
2538         }
2539         return match;
2540 }
2541
2542 /* update endpoint address(es) of template(s) */
2543 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2544                                struct xfrm_migrate *m, int num_migrate)
2545 {
2546         struct xfrm_migrate *mp;
2547         struct dst_entry *dst;
2548         int i, j, n = 0;
2549
2550         write_lock_bh(&pol->lock);
2551         if (unlikely(pol->dead)) {
2552                 /* target policy has been deleted */
2553                 write_unlock_bh(&pol->lock);
2554                 return -ENOENT;
2555         }
2556
2557         for (i = 0; i < pol->xfrm_nr; i++) {
2558                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2559                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2560                                 continue;
2561                         n++;
2562                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2563                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2564                                 continue;
2565                         /* update endpoints */
2566                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2567                                sizeof(pol->xfrm_vec[i].id.daddr));
2568                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2569                                sizeof(pol->xfrm_vec[i].saddr));
2570                         pol->xfrm_vec[i].encap_family = mp->new_family;
2571                         /* flush bundles */
2572                         while ((dst = pol->bundles) != NULL) {
2573                                 pol->bundles = dst->next;
2574                                 dst_free(dst);
2575                         }
2576                 }
2577         }
2578
2579         write_unlock_bh(&pol->lock);
2580
2581         if (!n)
2582                 return -ENODATA;
2583
2584         return 0;
2585 }
2586
2587 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2588 {
2589         int i, j;
2590
2591         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2592                 return -EINVAL;
2593
2594         for (i = 0; i < num_migrate; i++) {
2595                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2596                                    m[i].old_family) == 0) &&
2597                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2598                                    m[i].old_family) == 0))
2599                         return -EINVAL;
2600                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2601                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2602                         return -EINVAL;
2603
2604                 /* check if there is any duplicated entry */
2605                 for (j = i + 1; j < num_migrate; j++) {
2606                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2607                                     sizeof(m[i].old_daddr)) &&
2608                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2609                                     sizeof(m[i].old_saddr)) &&
2610                             m[i].proto == m[j].proto &&
2611                             m[i].mode == m[j].mode &&
2612                             m[i].reqid == m[j].reqid &&
2613                             m[i].old_family == m[j].old_family)
2614                                 return -EINVAL;
2615                 }
2616         }
2617
2618         return 0;
2619 }
2620
2621 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2622                  struct xfrm_migrate *m, int num_migrate)
2623 {
2624         int i, err, nx_cur = 0, nx_new = 0;
2625         struct xfrm_policy *pol = NULL;
2626         struct xfrm_state *x, *xc;
2627         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2628         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2629         struct xfrm_migrate *mp;
2630
2631         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2632                 goto out;
2633
2634         /* Stage 1 - find policy */
2635         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2636                 err = -ENOENT;
2637                 goto out;
2638         }
2639
2640         /* Stage 2 - find and update state(s) */
2641         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2642                 if ((x = xfrm_migrate_state_find(mp))) {
2643                         x_cur[nx_cur] = x;
2644                         nx_cur++;
2645                         if ((xc = xfrm_state_migrate(x, mp))) {
2646                                 x_new[nx_new] = xc;
2647                                 nx_new++;
2648                         } else {
2649                                 err = -ENODATA;
2650                                 goto restore_state;
2651                         }
2652                 }
2653         }
2654
2655         /* Stage 3 - update policy */
2656         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2657                 goto restore_state;
2658
2659         /* Stage 4 - delete old state(s) */
2660         if (nx_cur) {
2661                 xfrm_states_put(x_cur, nx_cur);
2662                 xfrm_states_delete(x_cur, nx_cur);
2663         }
2664
2665         /* Stage 5 - announce */
2666         km_migrate(sel, dir, type, m, num_migrate);
2667
2668         xfrm_pol_put(pol);
2669
2670         return 0;
2671 out:
2672         return err;
2673
2674 restore_state:
2675         if (pol)
2676                 xfrm_pol_put(pol);
2677         if (nx_cur)
2678                 xfrm_states_put(x_cur, nx_cur);
2679         if (nx_new)
2680                 xfrm_states_delete(x_new, nx_new);
2681
2682         return err;
2683 }
2684 EXPORT_SYMBOL(xfrm_migrate);
2685 #endif