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 <net/dst.h>
  28 #include <net/xfrm.h>
  29 #include <net/ip.h>
  30 #ifdef CONFIG_XFRM_STATISTICS
  31 #include <net/snmp.h>
  32 #endif
  33
  34 #include "xfrm_hash.h"
  35
  36 int sysctl_xfrm_larval_drop __read_mostly;
  37
  38 #ifdef CONFIG_XFRM_STATISTICS
  39 DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
  40 EXPORT_SYMBOL(xfrm_statistics);
  41 #endif
  42
  43 DEFINE_MUTEX(xfrm_cfg_mutex);
  44 EXPORT_SYMBOL(xfrm_cfg_mutex);
  45
  46 static DEFINE_RWLOCK(xfrm_policy_lock);
  47
  48 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  49 EXPORT_SYMBOL(xfrm_policy_count);
  50
  51 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  52 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  53
  54 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  55
  56 static struct work_struct xfrm_policy_gc_work;
  57 static HLIST_HEAD(xfrm_policy_gc_list);
  58 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  59
  60 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  61 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  62 static void xfrm_init_pmtu(struct dst_entry *dst);
  63
  64 static inline int
  65 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  66 {
  67         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  68                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  69                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  70                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  71                 (fl->proto == sel->proto || !sel->proto) &&
  72                 (fl->oif == sel->ifindex || !sel->ifindex);
  73 }
  74
  75 static inline int
  76 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  77 {
  78         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  79                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  80                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  81                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  82                 (fl->proto == sel->proto || !sel->proto) &&
  83                 (fl->oif == sel->ifindex || !sel->ifindex);
  84 }
  85
  86 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  87                     unsigned short family)
  88 {
  89         switch (family) {
  90         case AF_INET:
  91                 return __xfrm4_selector_match(sel, fl);
  92         case AF_INET6:
  93                 return __xfrm6_selector_match(sel, fl);
  94         }
  95         return 0;
  96 }
  97
  98 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
  99                                                 int family)
 100 {
 101         xfrm_address_t *saddr = &x->props.saddr;
 102         xfrm_address_t *daddr = &x->id.daddr;
 103         struct xfrm_policy_afinfo *afinfo;
 104         struct dst_entry *dst;
 105
 106         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR)
 107                 saddr = x->coaddr;
 108         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR)
 109                 daddr = x->coaddr;
 110
 111         afinfo = xfrm_policy_get_afinfo(family);
 112         if (unlikely(afinfo == NULL))
 113                 return ERR_PTR(-EAFNOSUPPORT);
 114
 115         dst = afinfo->dst_lookup(tos, saddr, daddr);
 116         xfrm_policy_put_afinfo(afinfo);
 117         return dst;
 118 }
 119
 120 static inline unsigned long make_jiffies(long secs)
 121 {
 122         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 123                 return MAX_SCHEDULE_TIMEOUT-1;
 124         else
 125                 return secs*HZ;
 126 }
 127
 128 static void xfrm_policy_timer(unsigned long data)
 129 {
 130         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 131         unsigned long now = get_seconds();
 132         long next = LONG_MAX;
 133         int warn = 0;
 134         int dir;
 135
 136         read_lock(&xp->lock);
 137
 138         if (xp->dead)
 139                 goto out;
 140
 141         dir = xfrm_policy_id2dir(xp->index);
 142
 143         if (xp->lft.hard_add_expires_seconds) {
 144                 long tmo = xp->lft.hard_add_expires_seconds +
 145                         xp->curlft.add_time - now;
 146                 if (tmo <= 0)
 147                         goto expired;
 148                 if (tmo < next)
 149                         next = tmo;
 150         }
 151         if (xp->lft.hard_use_expires_seconds) {
 152                 long tmo = xp->lft.hard_use_expires_seconds +
 153                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 154                 if (tmo <= 0)
 155                         goto expired;
 156                 if (tmo < next)
 157                         next = tmo;
 158         }
 159         if (xp->lft.soft_add_expires_seconds) {
 160                 long tmo = xp->lft.soft_add_expires_seconds +
 161                         xp->curlft.add_time - now;
 162                 if (tmo <= 0) {
 163                         warn = 1;
 164                         tmo = XFRM_KM_TIMEOUT;
 165                 }
 166                 if (tmo < next)
 167                         next = tmo;
 168         }
 169         if (xp->lft.soft_use_expires_seconds) {
 170                 long tmo = xp->lft.soft_use_expires_seconds +
 171                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 172                 if (tmo <= 0) {
 173                         warn = 1;
 174                         tmo = XFRM_KM_TIMEOUT;
 175                 }
 176                 if (tmo < next)
 177                         next = tmo;
 178         }
 179
 180         if (warn)
 181                 km_policy_expired(xp, dir, 0, 0);
 182         if (next != LONG_MAX &&
 183             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 184                 xfrm_pol_hold(xp);
 185
 186 out:
 187         read_unlock(&xp->lock);
 188         xfrm_pol_put(xp);
 189         return;
 190
 191 expired:
 192         read_unlock(&xp->lock);
 193         if (!xfrm_policy_delete(xp, dir))
 194                 km_policy_expired(xp, dir, 1, 0);
 195         xfrm_pol_put(xp);
 196 }
 197
 198
 199 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 200  * SPD calls.
 201  */
 202
 203 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 204 {
 205         struct xfrm_policy *policy;
 206
 207         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 208
 209         if (policy) {
 210                 INIT_HLIST_NODE(&policy->bydst);
 211                 INIT_HLIST_NODE(&policy->byidx);
 212                 rwlock_init(&policy->lock);
 213                 atomic_set(&policy->refcnt, 1);
 214                 setup_timer(&policy->timer, xfrm_policy_timer,
 215                                 (unsigned long)policy);
 216         }
 217         return policy;
 218 }
 219 EXPORT_SYMBOL(xfrm_policy_alloc);
 220
 221 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 222
 223 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 224 {
 225         BUG_ON(!policy->dead);
 226
 227         BUG_ON(policy->bundles);
 228
 229         if (del_timer(&policy->timer))
 230                 BUG();
 231
 232         security_xfrm_policy_free(policy);
 233         kfree(policy);
 234 }
 235 EXPORT_SYMBOL(__xfrm_policy_destroy);
 236
 237 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 238 {
 239         struct dst_entry *dst;
 240
 241         while ((dst = policy->bundles) != NULL) {
 242                 policy->bundles = dst->next;
 243                 dst_free(dst);
 244         }
 245
 246         if (del_timer(&policy->timer))
 247                 atomic_dec(&policy->refcnt);
 248
 249         if (atomic_read(&policy->refcnt) > 1)
 250                 flow_cache_flush();
 251
 252         xfrm_pol_put(policy);
 253 }
 254
 255 static void xfrm_policy_gc_task(struct work_struct *work)
 256 {
 257         struct xfrm_policy *policy;
 258         struct hlist_node *entry, *tmp;
 259         struct hlist_head gc_list;
 260
 261         spin_lock_bh(&xfrm_policy_gc_lock);
 262         gc_list.first = xfrm_policy_gc_list.first;
 263         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 264         spin_unlock_bh(&xfrm_policy_gc_lock);
 265
 266         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 267                 xfrm_policy_gc_kill(policy);
 268 }
 269
 270 /* Rule must be locked. Release descentant resources, announce
 271  * entry dead. The rule must be unlinked from lists to the moment.
 272  */
 273
 274 static void xfrm_policy_kill(struct xfrm_policy *policy)
 275 {
 276         int dead;
 277
 278         write_lock_bh(&policy->lock);
 279         dead = policy->dead;
 280         policy->dead = 1;
 281         write_unlock_bh(&policy->lock);
 282
 283         if (unlikely(dead)) {
 284                 WARN_ON(1);
 285                 return;
 286         }
 287
 288         spin_lock(&xfrm_policy_gc_lock);
 289         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 290         spin_unlock(&xfrm_policy_gc_lock);
 291
 292         schedule_work(&xfrm_policy_gc_work);
 293 }
 294
 295 struct xfrm_policy_hash {
 296         struct hlist_head       *table;
 297         unsigned int            hmask;
 298 };
 299
 300 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 301 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 302 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 303 static unsigned int xfrm_idx_hmask __read_mostly;
 304 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 305
 306 static inline unsigned int idx_hash(u32 index)
 307 {
 308         return __idx_hash(index, xfrm_idx_hmask);
 309 }
 310
 311 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 312 {
 313         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 314         unsigned int hash = __sel_hash(sel, family, hmask);
 315
 316         return (hash == hmask + 1 ?
 317                 &xfrm_policy_inexact[dir] :
 318                 xfrm_policy_bydst[dir].table + hash);
 319 }
 320
 321 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 322 {
 323         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 324         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 325
 326         return xfrm_policy_bydst[dir].table + hash;
 327 }
 328
 329 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 330                                    struct hlist_head *ndsttable,
 331                                    unsigned int nhashmask)
 332 {
 333         struct hlist_node *entry, *tmp;
 334         struct xfrm_policy *pol;
 335
 336         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 337                 unsigned int h;
 338
 339                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 340                                 pol->family, nhashmask);
 341                 hlist_add_head(&pol->bydst, ndsttable+h);
 342         }
 343 }
 344
 345 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 346                                    struct hlist_head *nidxtable,
 347                                    unsigned int nhashmask)
 348 {
 349         struct hlist_node *entry, *tmp;
 350         struct xfrm_policy *pol;
 351
 352         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 353                 unsigned int h;
 354
 355                 h = __idx_hash(pol->index, nhashmask);
 356                 hlist_add_head(&pol->byidx, nidxtable+h);
 357         }
 358 }
 359
 360 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 361 {
 362         return ((old_hmask + 1) << 1) - 1;
 363 }
 364
 365 static void xfrm_bydst_resize(int dir)
 366 {
 367         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 368         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 369         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 370         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 371         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 372         int i;
 373
 374         if (!ndst)
 375                 return;
 376
 377         write_lock_bh(&xfrm_policy_lock);
 378
 379         for (i = hmask; i >= 0; i--)
 380                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 381
 382         xfrm_policy_bydst[dir].table = ndst;
 383         xfrm_policy_bydst[dir].hmask = nhashmask;
 384
 385         write_unlock_bh(&xfrm_policy_lock);
 386
 387         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 388 }
 389
 390 static void xfrm_byidx_resize(int total)
 391 {
 392         unsigned int hmask = xfrm_idx_hmask;
 393         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 394         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 395         struct hlist_head *oidx = xfrm_policy_byidx;
 396         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 397         int i;
 398
 399         if (!nidx)
 400                 return;
 401
 402         write_lock_bh(&xfrm_policy_lock);
 403
 404         for (i = hmask; i >= 0; i--)
 405                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 406
 407         xfrm_policy_byidx = nidx;
 408         xfrm_idx_hmask = nhashmask;
 409
 410         write_unlock_bh(&xfrm_policy_lock);
 411
 412         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 413 }
 414
 415 static inline int xfrm_bydst_should_resize(int dir, int *total)
 416 {
 417         unsigned int cnt = xfrm_policy_count[dir];
 418         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 419
 420         if (total)
 421                 *total += cnt;
 422
 423         if ((hmask + 1) < xfrm_policy_hashmax &&
 424             cnt > hmask)
 425                 return 1;
 426
 427         return 0;
 428 }
 429
 430 static inline int xfrm_byidx_should_resize(int total)
 431 {
 432         unsigned int hmask = xfrm_idx_hmask;
 433
 434         if ((hmask + 1) < xfrm_policy_hashmax &&
 435             total > hmask)
 436                 return 1;
 437
 438         return 0;
 439 }
 440
 441 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 442 {
 443         read_lock_bh(&xfrm_policy_lock);
 444         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 445         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 446         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 447         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 448         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 449         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 450         si->spdhcnt = xfrm_idx_hmask;
 451         si->spdhmcnt = xfrm_policy_hashmax;
 452         read_unlock_bh(&xfrm_policy_lock);
 453 }
 454 EXPORT_SYMBOL(xfrm_spd_getinfo);
 455
 456 static DEFINE_MUTEX(hash_resize_mutex);
 457 static void xfrm_hash_resize(struct work_struct *__unused)
 458 {
 459         int dir, total;
 460
 461         mutex_lock(&hash_resize_mutex);
 462
 463         total = 0;
 464         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 465                 if (xfrm_bydst_should_resize(dir, &total))
 466                         xfrm_bydst_resize(dir);
 467         }
 468         if (xfrm_byidx_should_resize(total))
 469                 xfrm_byidx_resize(total);
 470
 471         mutex_unlock(&hash_resize_mutex);
 472 }
 473
 474 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 475
 476 /* Generate new index... KAME seems to generate them ordered by cost
 477  * of an absolute inpredictability of ordering of rules. This will not pass. */
 478 static u32 xfrm_gen_index(u8 type, int dir)
 479 {
 480         static u32 idx_generator;
 481
 482         for (;;) {
 483                 struct hlist_node *entry;
 484                 struct hlist_head *list;
 485                 struct xfrm_policy *p;
 486                 u32 idx;
 487                 int found;
 488
 489                 idx = (idx_generator | dir);
 490                 idx_generator += 8;
 491                 if (idx == 0)
 492                         idx = 8;
 493                 list = xfrm_policy_byidx + idx_hash(idx);
 494                 found = 0;
 495                 hlist_for_each_entry(p, entry, list, byidx) {
 496                         if (p->index == idx) {
 497                                 found = 1;
 498                                 break;
 499                         }
 500                 }
 501                 if (!found)
 502                         return idx;
 503         }
 504 }
 505
 506 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 507 {
 508         u32 *p1 = (u32 *) s1;
 509         u32 *p2 = (u32 *) s2;
 510         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 511         int i;
 512
 513         for (i = 0; i < len; i++) {
 514                 if (p1[i] != p2[i])
 515                         return 1;
 516         }
 517
 518         return 0;
 519 }
 520
 521 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 522 {
 523         struct xfrm_policy *pol;
 524         struct xfrm_policy *delpol;
 525         struct hlist_head *chain;
 526         struct hlist_node *entry, *newpos;
 527         struct dst_entry *gc_list;
 528
 529         write_lock_bh(&xfrm_policy_lock);
 530         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 531         delpol = NULL;
 532         newpos = NULL;
 533         hlist_for_each_entry(pol, entry, chain, bydst) {
 534                 if (pol->type == policy->type &&
 535                     !selector_cmp(&pol->selector, &policy->selector) &&
 536                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 537                     !WARN_ON(delpol)) {
 538                         if (excl) {
 539                                 write_unlock_bh(&xfrm_policy_lock);
 540                                 return -EEXIST;
 541                         }
 542                         delpol = pol;
 543                         if (policy->priority > pol->priority)
 544                                 continue;
 545                 } else if (policy->priority >= pol->priority) {
 546                         newpos = &pol->bydst;
 547                         continue;
 548                 }
 549                 if (delpol)
 550                         break;
 551         }
 552         if (newpos)
 553                 hlist_add_after(newpos, &policy->bydst);
 554         else
 555                 hlist_add_head(&policy->bydst, chain);
 556         xfrm_pol_hold(policy);
 557         xfrm_policy_count[dir]++;
 558         atomic_inc(&flow_cache_genid);
 559         if (delpol) {
 560                 hlist_del(&delpol->bydst);
 561                 hlist_del(&delpol->byidx);
 562                 xfrm_policy_count[dir]--;
 563         }
 564         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 565         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 566         policy->curlft.add_time = get_seconds();
 567         policy->curlft.use_time = 0;
 568         if (!mod_timer(&policy->timer, jiffies + HZ))
 569                 xfrm_pol_hold(policy);
 570         write_unlock_bh(&xfrm_policy_lock);
 571
 572         if (delpol)
 573                 xfrm_policy_kill(delpol);
 574         else if (xfrm_bydst_should_resize(dir, NULL))
 575                 schedule_work(&xfrm_hash_work);
 576
 577         read_lock_bh(&xfrm_policy_lock);
 578         gc_list = NULL;
 579         entry = &policy->bydst;
 580         hlist_for_each_entry_continue(policy, entry, bydst) {
 581                 struct dst_entry *dst;
 582
 583                 write_lock(&policy->lock);
 584                 dst = policy->bundles;
 585                 if (dst) {
 586                         struct dst_entry *tail = dst;
 587                         while (tail->next)
 588                                 tail = tail->next;
 589                         tail->next = gc_list;
 590                         gc_list = dst;
 591
 592                         policy->bundles = NULL;
 593                 }
 594                 write_unlock(&policy->lock);
 595         }
 596         read_unlock_bh(&xfrm_policy_lock);
 597
 598         while (gc_list) {
 599                 struct dst_entry *dst = gc_list;
 600
 601                 gc_list = dst->next;
 602                 dst_free(dst);
 603         }
 604
 605         return 0;
 606 }
 607 EXPORT_SYMBOL(xfrm_policy_insert);
 608
 609 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 610                                           struct xfrm_selector *sel,
 611                                           struct xfrm_sec_ctx *ctx, int delete,
 612                                           int *err)
 613 {
 614         struct xfrm_policy *pol, *ret;
 615         struct hlist_head *chain;
 616         struct hlist_node *entry;
 617
 618         *err = 0;
 619         write_lock_bh(&xfrm_policy_lock);
 620         chain = policy_hash_bysel(sel, sel->family, dir);
 621         ret = NULL;
 622         hlist_for_each_entry(pol, entry, chain, bydst) {
 623                 if (pol->type == type &&
 624                     !selector_cmp(sel, &pol->selector) &&
 625                     xfrm_sec_ctx_match(ctx, pol->security)) {
 626                         xfrm_pol_hold(pol);
 627                         if (delete) {
 628                                 *err = security_xfrm_policy_delete(pol);
 629                                 if (*err) {
 630                                         write_unlock_bh(&xfrm_policy_lock);
 631                                         return pol;
 632                                 }
 633                                 hlist_del(&pol->bydst);
 634                                 hlist_del(&pol->byidx);
 635                                 xfrm_policy_count[dir]--;
 636                         }
 637                         ret = pol;
 638                         break;
 639                 }
 640         }
 641         write_unlock_bh(&xfrm_policy_lock);
 642
 643         if (ret && delete) {
 644                 atomic_inc(&flow_cache_genid);
 645                 xfrm_policy_kill(ret);
 646         }
 647         return ret;
 648 }
 649 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 650
 651 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 652                                      int *err)
 653 {
 654         struct xfrm_policy *pol, *ret;
 655         struct hlist_head *chain;
 656         struct hlist_node *entry;
 657
 658         *err = -ENOENT;
 659         if (xfrm_policy_id2dir(id) != dir)
 660                 return NULL;
 661
 662         *err = 0;
 663         write_lock_bh(&xfrm_policy_lock);
 664         chain = xfrm_policy_byidx + idx_hash(id);
 665         ret = NULL;
 666         hlist_for_each_entry(pol, entry, chain, byidx) {
 667                 if (pol->type == type && pol->index == id) {
 668                         xfrm_pol_hold(pol);
 669                         if (delete) {
 670                                 *err = security_xfrm_policy_delete(pol);
 671                                 if (*err) {
 672                                         write_unlock_bh(&xfrm_policy_lock);
 673                                         return pol;
 674                                 }
 675                                 hlist_del(&pol->bydst);
 676                                 hlist_del(&pol->byidx);
 677                                 xfrm_policy_count[dir]--;
 678                         }
 679                         ret = pol;
 680                         break;
 681                 }
 682         }
 683         write_unlock_bh(&xfrm_policy_lock);
 684
 685         if (ret && delete) {
 686                 atomic_inc(&flow_cache_genid);
 687                 xfrm_policy_kill(ret);
 688         }
 689         return ret;
 690 }
 691 EXPORT_SYMBOL(xfrm_policy_byid);
 692
 693 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 694 static inline int
 695 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 696 {
 697         int dir, err = 0;
 698
 699         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 700                 struct xfrm_policy *pol;
 701                 struct hlist_node *entry;
 702                 int i;
 703
 704                 hlist_for_each_entry(pol, entry,
 705                                      &xfrm_policy_inexact[dir], bydst) {
 706                         if (pol->type != type)
 707                                 continue;
 708                         err = security_xfrm_policy_delete(pol);
 709                         if (err) {
 710                                 xfrm_audit_policy_delete(pol, 0,
 711                                                          audit_info->loginuid,
 712                                                          audit_info->secid);
 713                                 return err;
 714                         }
 715                 }
 716                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 717                         hlist_for_each_entry(pol, entry,
 718                                              xfrm_policy_bydst[dir].table + i,
 719                                              bydst) {
 720                                 if (pol->type != type)
 721                                         continue;
 722                                 err = security_xfrm_policy_delete(pol);
 723                                 if (err) {
 724                                         xfrm_audit_policy_delete(pol, 0,
 725                                                         audit_info->loginuid,
 726                                                         audit_info->secid);
 727                                         return err;
 728                                 }
 729                         }
 730                 }
 731         }
 732         return err;
 733 }
 734 #else
 735 static inline int
 736 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 737 {
 738         return 0;
 739 }
 740 #endif
 741
 742 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 743 {
 744         int dir, err = 0;
 745
 746         write_lock_bh(&xfrm_policy_lock);
 747
 748         err = xfrm_policy_flush_secctx_check(type, audit_info);
 749         if (err)
 750                 goto out;
 751
 752         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 753                 struct xfrm_policy *pol;
 754                 struct hlist_node *entry;
 755                 int i, killed;
 756
 757                 killed = 0;
 758         again1:
 759                 hlist_for_each_entry(pol, entry,
 760                                      &xfrm_policy_inexact[dir], bydst) {
 761                         if (pol->type != type)
 762                                 continue;
 763                         hlist_del(&pol->bydst);
 764                         hlist_del(&pol->byidx);
 765                         write_unlock_bh(&xfrm_policy_lock);
 766
 767                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 768                                                  audit_info->secid);
 769
 770                         xfrm_policy_kill(pol);
 771                         killed++;
 772
 773                         write_lock_bh(&xfrm_policy_lock);
 774                         goto again1;
 775                 }
 776
 777                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 778         again2:
 779                         hlist_for_each_entry(pol, entry,
 780                                              xfrm_policy_bydst[dir].table + i,
 781                                              bydst) {
 782                                 if (pol->type != type)
 783                                         continue;
 784                                 hlist_del(&pol->bydst);
 785                                 hlist_del(&pol->byidx);
 786                                 write_unlock_bh(&xfrm_policy_lock);
 787
 788                                 xfrm_audit_policy_delete(pol, 1,
 789                                                          audit_info->loginuid,
 790                                                          audit_info->secid);
 791                                 xfrm_policy_kill(pol);
 792                                 killed++;
 793
 794                                 write_lock_bh(&xfrm_policy_lock);
 795                                 goto again2;
 796                         }
 797                 }
 798
 799                 xfrm_policy_count[dir] -= killed;
 800         }
 801         atomic_inc(&flow_cache_genid);
 802 out:
 803         write_unlock_bh(&xfrm_policy_lock);
 804         return err;
 805 }
 806 EXPORT_SYMBOL(xfrm_policy_flush);
 807
 808 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 809                      void *data)
 810 {
 811         struct xfrm_policy *pol, *last = NULL;
 812         struct hlist_node *entry;
 813         int dir, last_dir = 0, count, error;
 814
 815         read_lock_bh(&xfrm_policy_lock);
 816         count = 0;
 817
 818         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 819                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 820                 int i;
 821
 822                 hlist_for_each_entry(pol, entry,
 823                                      &xfrm_policy_inexact[dir], bydst) {
 824                         if (pol->type != type)
 825                                 continue;
 826                         if (last) {
 827                                 error = func(last, last_dir % XFRM_POLICY_MAX,
 828                                              count, data);
 829                                 if (error)
 830                                         goto out;
 831                         }
 832                         last = pol;
 833                         last_dir = dir;
 834                         count++;
 835                 }
 836                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 837                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 838                                 if (pol->type != type)
 839                                         continue;
 840                                 if (last) {
 841                                         error = func(last, last_dir % XFRM_POLICY_MAX,
 842                                                      count, data);
 843                                         if (error)
 844                                                 goto out;
 845                                 }
 846                                 last = pol;
 847                                 last_dir = dir;
 848                                 count++;
 849                         }
 850                 }
 851         }
 852         if (count == 0) {
 853                 error = -ENOENT;
 854                 goto out;
 855         }
 856         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
 857 out:
 858         read_unlock_bh(&xfrm_policy_lock);
 859         return error;
 860 }
 861 EXPORT_SYMBOL(xfrm_policy_walk);
 862
 863 /*
 864  * Find policy to apply to this flow.
 865  *
 866  * Returns 0 if policy found, else an -errno.
 867  */
 868 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 869                              u8 type, u16 family, int dir)
 870 {
 871         struct xfrm_selector *sel = &pol->selector;
 872         int match, ret = -ESRCH;
 873
 874         if (pol->family != family ||
 875             pol->type != type)
 876                 return ret;
 877
 878         match = xfrm_selector_match(sel, fl, family);
 879         if (match)
 880                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
 881
 882         return ret;
 883 }
 884
 885 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 886                                                      u16 family, u8 dir)
 887 {
 888         int err;
 889         struct xfrm_policy *pol, *ret;
 890         xfrm_address_t *daddr, *saddr;
 891         struct hlist_node *entry;
 892         struct hlist_head *chain;
 893         u32 priority = ~0U;
 894
 895         daddr = xfrm_flowi_daddr(fl, family);
 896         saddr = xfrm_flowi_saddr(fl, family);
 897         if (unlikely(!daddr || !saddr))
 898                 return NULL;
 899
 900         read_lock_bh(&xfrm_policy_lock);
 901         chain = policy_hash_direct(daddr, saddr, family, dir);
 902         ret = NULL;
 903         hlist_for_each_entry(pol, entry, chain, bydst) {
 904                 err = xfrm_policy_match(pol, fl, type, family, dir);
 905                 if (err) {
 906                         if (err == -ESRCH)
 907                                 continue;
 908                         else {
 909                                 ret = ERR_PTR(err);
 910                                 goto fail;
 911                         }
 912                 } else {
 913                         ret = pol;
 914                         priority = ret->priority;
 915                         break;
 916                 }
 917         }
 918         chain = &xfrm_policy_inexact[dir];
 919         hlist_for_each_entry(pol, entry, chain, bydst) {
 920                 err = xfrm_policy_match(pol, fl, type, family, dir);
 921                 if (err) {
 922                         if (err == -ESRCH)
 923                                 continue;
 924                         else {
 925                                 ret = ERR_PTR(err);
 926                                 goto fail;
 927                         }
 928                 } else if (pol->priority < priority) {
 929                         ret = pol;
 930                         break;
 931                 }
 932         }
 933         if (ret)
 934                 xfrm_pol_hold(ret);
 935 fail:
 936         read_unlock_bh(&xfrm_policy_lock);
 937
 938         return ret;
 939 }
 940
 941 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 942                                void **objp, atomic_t **obj_refp)
 943 {
 944         struct xfrm_policy *pol;
 945         int err = 0;
 946
 947 #ifdef CONFIG_XFRM_SUB_POLICY
 948         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
 949         if (IS_ERR(pol)) {
 950                 err = PTR_ERR(pol);
 951                 pol = NULL;
 952         }
 953         if (pol || err)
 954                 goto end;
 955 #endif
 956         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 957         if (IS_ERR(pol)) {
 958                 err = PTR_ERR(pol);
 959                 pol = NULL;
 960         }
 961 #ifdef CONFIG_XFRM_SUB_POLICY
 962 end:
 963 #endif
 964         if ((*objp = (void *) pol) != NULL)
 965                 *obj_refp = &pol->refcnt;
 966         return err;
 967 }
 968
 969 static inline int policy_to_flow_dir(int dir)
 970 {
 971         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 972             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 973             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 974                 return dir;
 975         switch (dir) {
 976         default:
 977         case XFRM_POLICY_IN:
 978                 return FLOW_DIR_IN;
 979         case XFRM_POLICY_OUT:
 980                 return FLOW_DIR_OUT;
 981         case XFRM_POLICY_FWD:
 982                 return FLOW_DIR_FWD;
 983         }
 984 }
 985
 986 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
 987 {
 988         struct xfrm_policy *pol;
 989
 990         read_lock_bh(&xfrm_policy_lock);
 991         if ((pol = sk->sk_policy[dir]) != NULL) {
 992                 int match = xfrm_selector_match(&pol->selector, fl,
 993                                                 sk->sk_family);
 994                 int err = 0;
 995
 996                 if (match) {
 997                         err = security_xfrm_policy_lookup(pol, fl->secid,
 998                                         policy_to_flow_dir(dir));
 999                         if (!err)
1000                                 xfrm_pol_hold(pol);
1001                         else if (err == -ESRCH)
1002                                 pol = NULL;
1003                         else
1004                                 pol = ERR_PTR(err);
1005                 } else
1006                         pol = NULL;
1007         }
1008         read_unlock_bh(&xfrm_policy_lock);
1009         return pol;
1010 }
1011
1012 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1013 {
1014         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1015                                                      pol->family, dir);
1016
1017         hlist_add_head(&pol->bydst, chain);
1018         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1019         xfrm_policy_count[dir]++;
1020         xfrm_pol_hold(pol);
1021
1022         if (xfrm_bydst_should_resize(dir, NULL))
1023                 schedule_work(&xfrm_hash_work);
1024 }
1025
1026 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1027                                                 int dir)
1028 {
1029         if (hlist_unhashed(&pol->bydst))
1030                 return NULL;
1031
1032         hlist_del(&pol->bydst);
1033         hlist_del(&pol->byidx);
1034         xfrm_policy_count[dir]--;
1035
1036         return pol;
1037 }
1038
1039 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1040 {
1041         write_lock_bh(&xfrm_policy_lock);
1042         pol = __xfrm_policy_unlink(pol, dir);
1043         write_unlock_bh(&xfrm_policy_lock);
1044         if (pol) {
1045                 if (dir < XFRM_POLICY_MAX)
1046                         atomic_inc(&flow_cache_genid);
1047                 xfrm_policy_kill(pol);
1048                 return 0;
1049         }
1050         return -ENOENT;
1051 }
1052 EXPORT_SYMBOL(xfrm_policy_delete);
1053
1054 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1055 {
1056         struct xfrm_policy *old_pol;
1057
1058 #ifdef CONFIG_XFRM_SUB_POLICY
1059         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1060                 return -EINVAL;
1061 #endif
1062
1063         write_lock_bh(&xfrm_policy_lock);
1064         old_pol = sk->sk_policy[dir];
1065         sk->sk_policy[dir] = pol;
1066         if (pol) {
1067                 pol->curlft.add_time = get_seconds();
1068                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1069                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1070         }
1071         if (old_pol)
1072                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1073         write_unlock_bh(&xfrm_policy_lock);
1074
1075         if (old_pol) {
1076                 xfrm_policy_kill(old_pol);
1077         }
1078         return 0;
1079 }
1080
1081 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1082 {
1083         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1084
1085         if (newp) {
1086                 newp->selector = old->selector;
1087                 if (security_xfrm_policy_clone(old, newp)) {
1088                         kfree(newp);
1089                         return NULL;  /* ENOMEM */
1090                 }
1091                 newp->lft = old->lft;
1092                 newp->curlft = old->curlft;
1093                 newp->action = old->action;
1094                 newp->flags = old->flags;
1095                 newp->xfrm_nr = old->xfrm_nr;
1096                 newp->index = old->index;
1097                 newp->type = old->type;
1098                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1099                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1100                 write_lock_bh(&xfrm_policy_lock);
1101                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1102                 write_unlock_bh(&xfrm_policy_lock);
1103                 xfrm_pol_put(newp);
1104         }
1105         return newp;
1106 }
1107
1108 int __xfrm_sk_clone_policy(struct sock *sk)
1109 {
1110         struct xfrm_policy *p0 = sk->sk_policy[0],
1111                            *p1 = sk->sk_policy[1];
1112
1113         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1114         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1115                 return -ENOMEM;
1116         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1117                 return -ENOMEM;
1118         return 0;
1119 }
1120
1121 static int
1122 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1123                unsigned short family)
1124 {
1125         int err;
1126         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1127
1128         if (unlikely(afinfo == NULL))
1129                 return -EINVAL;
1130         err = afinfo->get_saddr(local, remote);
1131         xfrm_policy_put_afinfo(afinfo);
1132         return err;
1133 }
1134
1135 /* Resolve list of templates for the flow, given policy. */
1136
1137 static int
1138 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1139                       struct xfrm_state **xfrm,
1140                       unsigned short family)
1141 {
1142         int nx;
1143         int i, error;
1144         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1145         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1146         xfrm_address_t tmp;
1147
1148         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1149                 struct xfrm_state *x;
1150                 xfrm_address_t *remote = daddr;
1151                 xfrm_address_t *local  = saddr;
1152                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1153
1154                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1155                     tmpl->mode == XFRM_MODE_BEET) {
1156                         remote = &tmpl->id.daddr;
1157                         local = &tmpl->saddr;
1158                         family = tmpl->encap_family;
1159                         if (xfrm_addr_any(local, family)) {
1160                                 error = xfrm_get_saddr(&tmp, remote, family);
1161                                 if (error)
1162                                         goto fail;
1163                                 local = &tmp;
1164                         }
1165                 }
1166
1167                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1168
1169                 if (x && x->km.state == XFRM_STATE_VALID) {
1170                         xfrm[nx++] = x;
1171                         daddr = remote;
1172                         saddr = local;
1173                         continue;
1174                 }
1175                 if (x) {
1176                         error = (x->km.state == XFRM_STATE_ERROR ?
1177                                  -EINVAL : -EAGAIN);
1178                         xfrm_state_put(x);
1179                 }
1180
1181                 if (!tmpl->optional)
1182                         goto fail;
1183         }
1184         return nx;
1185
1186 fail:
1187         for (nx--; nx>=0; nx--)
1188                 xfrm_state_put(xfrm[nx]);
1189         return error;
1190 }
1191
1192 static int
1193 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1194                   struct xfrm_state **xfrm,
1195                   unsigned short family)
1196 {
1197         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1198         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1199         int cnx = 0;
1200         int error;
1201         int ret;
1202         int i;
1203
1204         for (i = 0; i < npols; i++) {
1205                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1206                         error = -ENOBUFS;
1207                         goto fail;
1208                 }
1209
1210                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1211                 if (ret < 0) {
1212                         error = ret;
1213                         goto fail;
1214                 } else
1215                         cnx += ret;
1216         }
1217
1218         /* found states are sorted for outbound processing */
1219         if (npols > 1)
1220                 xfrm_state_sort(xfrm, tpp, cnx, family);
1221
1222         return cnx;
1223
1224  fail:
1225         for (cnx--; cnx>=0; cnx--)
1226                 xfrm_state_put(tpp[cnx]);
1227         return error;
1228
1229 }
1230
1231 /* Check that the bundle accepts the flow and its components are
1232  * still valid.
1233  */
1234
1235 static struct dst_entry *
1236 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1237 {
1238         struct dst_entry *x;
1239         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1240         if (unlikely(afinfo == NULL))
1241                 return ERR_PTR(-EINVAL);
1242         x = afinfo->find_bundle(fl, policy);
1243         xfrm_policy_put_afinfo(afinfo);
1244         return x;
1245 }
1246
1247 static inline int xfrm_get_tos(struct flowi *fl, int family)
1248 {
1249         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1250         int tos;
1251
1252         if (!afinfo)
1253                 return -EINVAL;
1254
1255         tos = afinfo->get_tos(fl);
1256
1257         xfrm_policy_put_afinfo(afinfo);
1258
1259         return tos;
1260 }
1261
1262 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1263 {
1264         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1265         struct xfrm_dst *xdst;
1266
1267         if (!afinfo)
1268                 return ERR_PTR(-EINVAL);
1269
1270         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1271
1272         xfrm_policy_put_afinfo(afinfo);
1273
1274         return xdst;
1275 }
1276
1277 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1278                                  int nfheader_len)
1279 {
1280         struct xfrm_policy_afinfo *afinfo =
1281                 xfrm_policy_get_afinfo(dst->ops->family);
1282         int err;
1283
1284         if (!afinfo)
1285                 return -EINVAL;
1286
1287         err = afinfo->init_path(path, dst, nfheader_len);
1288
1289         xfrm_policy_put_afinfo(afinfo);
1290
1291         return err;
1292 }
1293
1294 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1295 {
1296         struct xfrm_policy_afinfo *afinfo =
1297                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1298         int err;
1299
1300         if (!afinfo)
1301                 return -EINVAL;
1302
1303         err = afinfo->fill_dst(xdst, dev);
1304
1305         xfrm_policy_put_afinfo(afinfo);
1306
1307         return err;
1308 }
1309
1310 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1311  * all the metrics... Shortly, bundle a bundle.
1312  */
1313
1314 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1315                                             struct xfrm_state **xfrm, int nx,
1316                                             struct flowi *fl,
1317                                             struct dst_entry *dst)
1318 {
1319         unsigned long now = jiffies;
1320         struct net_device *dev;
1321         struct dst_entry *dst_prev = NULL;
1322         struct dst_entry *dst0 = NULL;
1323         int i = 0;
1324         int err;
1325         int header_len = 0;
1326         int nfheader_len = 0;
1327         int trailer_len = 0;
1328         int tos;
1329         int family = policy->selector.family;
1330
1331         tos = xfrm_get_tos(fl, family);
1332         err = tos;
1333         if (tos < 0)
1334                 goto put_states;
1335
1336         dst_hold(dst);
1337
1338         for (; i < nx; i++) {
1339                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1340                 struct dst_entry *dst1 = &xdst->u.dst;
1341
1342                 err = PTR_ERR(xdst);
1343                 if (IS_ERR(xdst)) {
1344                         dst_release(dst);
1345                         goto put_states;
1346                 }
1347
1348                 if (!dst_prev)
1349                         dst0 = dst1;
1350                 else {
1351                         dst_prev->child = dst_clone(dst1);
1352                         dst1->flags |= DST_NOHASH;
1353                 }
1354
1355                 xdst->route = dst;
1356                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1357
1358                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1359                         family = xfrm[i]->props.family;
1360                         dst = xfrm_dst_lookup(xfrm[i], tos, family);
1361                         err = PTR_ERR(dst);
1362                         if (IS_ERR(dst))
1363                                 goto put_states;
1364                 } else
1365                         dst_hold(dst);
1366
1367                 dst1->xfrm = xfrm[i];
1368                 xdst->genid = xfrm[i]->genid;
1369
1370                 dst1->obsolete = -1;
1371                 dst1->flags |= DST_HOST;
1372                 dst1->lastuse = now;
1373
1374                 dst1->input = dst_discard;
1375                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1376
1377                 dst1->next = dst_prev;
1378                 dst_prev = dst1;
1379
1380                 header_len += xfrm[i]->props.header_len;
1381                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1382                         nfheader_len += xfrm[i]->props.header_len;
1383                 trailer_len += xfrm[i]->props.trailer_len;
1384         }
1385
1386         dst_prev->child = dst;
1387         dst0->path = dst;
1388
1389         err = -ENODEV;
1390         dev = dst->dev;
1391         if (!dev)
1392                 goto free_dst;
1393
1394         /* Copy neighbout for reachability confirmation */
1395         dst0->neighbour = neigh_clone(dst->neighbour);
1396
1397         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1398         xfrm_init_pmtu(dst_prev);
1399
1400         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1401                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1402
1403                 err = xfrm_fill_dst(xdst, dev);
1404                 if (err)
1405                         goto free_dst;
1406
1407                 dst_prev->header_len = header_len;
1408                 dst_prev->trailer_len = trailer_len;
1409                 header_len -= xdst->u.dst.xfrm->props.header_len;
1410                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1411         }
1412
1413 out:
1414         return dst0;
1415
1416 put_states:
1417         for (; i < nx; i++)
1418                 xfrm_state_put(xfrm[i]);
1419 free_dst:
1420         if (dst0)
1421                 dst_free(dst0);
1422         dst0 = ERR_PTR(err);
1423         goto out;
1424 }
1425
1426 static int inline
1427 xfrm_dst_alloc_copy(void **target, void *src, int size)
1428 {
1429         if (!*target) {
1430                 *target = kmalloc(size, GFP_ATOMIC);
1431                 if (!*target)
1432                         return -ENOMEM;
1433         }
1434         memcpy(*target, src, size);
1435         return 0;
1436 }
1437
1438 static int inline
1439 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1440 {
1441 #ifdef CONFIG_XFRM_SUB_POLICY
1442         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1443         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1444                                    sel, sizeof(*sel));
1445 #else
1446         return 0;
1447 #endif
1448 }
1449
1450 static int inline
1451 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1452 {
1453 #ifdef CONFIG_XFRM_SUB_POLICY
1454         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1455         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1456 #else
1457         return 0;
1458 #endif
1459 }
1460
1461 static int stale_bundle(struct dst_entry *dst);
1462
1463 /* Main function: finds/creates a bundle for given flow.
1464  *
1465  * At the moment we eat a raw IP route. Mostly to speed up lookups
1466  * on interfaces with disabled IPsec.
1467  */
1468 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1469                   struct sock *sk, int flags)
1470 {
1471         struct xfrm_policy *policy;
1472         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1473         int npols;
1474         int pol_dead;
1475         int xfrm_nr;
1476         int pi;
1477         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1478         struct dst_entry *dst, *dst_orig = *dst_p;
1479         int nx = 0;
1480         int err;
1481         u32 genid;
1482         u16 family;
1483         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1484
1485 restart:
1486         genid = atomic_read(&flow_cache_genid);
1487         policy = NULL;
1488         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1489                 pols[pi] = NULL;
1490         npols = 0;
1491         pol_dead = 0;
1492         xfrm_nr = 0;
1493
1494         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1495                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1496                 err = PTR_ERR(policy);
1497                 if (IS_ERR(policy))
1498                         goto dropdst;
1499         }
1500
1501         if (!policy) {
1502                 /* To accelerate a bit...  */
1503                 if ((dst_orig->flags & DST_NOXFRM) ||
1504                     !xfrm_policy_count[XFRM_POLICY_OUT])
1505                         goto nopol;
1506
1507                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1508                                            dir, xfrm_policy_lookup);
1509                 err = PTR_ERR(policy);
1510                 if (IS_ERR(policy))
1511                         goto dropdst;
1512         }
1513
1514         if (!policy)
1515                 goto nopol;
1516
1517         family = dst_orig->ops->family;
1518         pols[0] = policy;
1519         npols ++;
1520         xfrm_nr += pols[0]->xfrm_nr;
1521
1522         err = -ENOENT;
1523         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1524                 goto error;
1525
1526         policy->curlft.use_time = get_seconds();
1527
1528         switch (policy->action) {
1529         default:
1530         case XFRM_POLICY_BLOCK:
1531                 /* Prohibit the flow */
1532                 err = -EPERM;
1533                 goto error;
1534
1535         case XFRM_POLICY_ALLOW:
1536 #ifndef CONFIG_XFRM_SUB_POLICY
1537                 if (policy->xfrm_nr == 0) {
1538                         /* Flow passes not transformed. */
1539                         xfrm_pol_put(policy);
1540                         return 0;
1541                 }
1542 #endif
1543
1544                 /* Try to find matching bundle.
1545                  *
1546                  * LATER: help from flow cache. It is optional, this
1547                  * is required only for output policy.
1548                  */
1549                 dst = xfrm_find_bundle(fl, policy, family);
1550                 if (IS_ERR(dst)) {
1551                         err = PTR_ERR(dst);
1552                         goto error;
1553                 }
1554
1555                 if (dst)
1556                         break;
1557
1558 #ifdef CONFIG_XFRM_SUB_POLICY
1559                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1560                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1561                                                             fl, family,
1562                                                             XFRM_POLICY_OUT);
1563                         if (pols[1]) {
1564                                 if (IS_ERR(pols[1])) {
1565                                         err = PTR_ERR(pols[1]);
1566                                         goto error;
1567                                 }
1568                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1569                                         err = -EPERM;
1570                                         goto error;
1571                                 }
1572                                 npols ++;
1573                                 xfrm_nr += pols[1]->xfrm_nr;
1574                         }
1575                 }
1576
1577                 /*
1578                  * Because neither flowi nor bundle information knows about
1579                  * transformation template size. On more than one policy usage
1580                  * we can realize whether all of them is bypass or not after
1581                  * they are searched. See above not-transformed bypass
1582                  * is surrounded by non-sub policy configuration, too.
1583                  */
1584                 if (xfrm_nr == 0) {
1585                         /* Flow passes not transformed. */
1586                         xfrm_pols_put(pols, npols);
1587                         return 0;
1588                 }
1589
1590 #endif
1591                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1592
1593                 if (unlikely(nx<0)) {
1594                         err = nx;
1595                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1596                                 /* EREMOTE tells the caller to generate
1597                                  * a one-shot blackhole route.
1598                                  */
1599                                 xfrm_pol_put(policy);
1600                                 return -EREMOTE;
1601                         }
1602                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1603                                 DECLARE_WAITQUEUE(wait, current);
1604
1605                                 add_wait_queue(&km_waitq, &wait);
1606                                 set_current_state(TASK_INTERRUPTIBLE);
1607                                 schedule();
1608                                 set_current_state(TASK_RUNNING);
1609                                 remove_wait_queue(&km_waitq, &wait);
1610
1611                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1612
1613                                 if (nx == -EAGAIN && signal_pending(current)) {
1614                                         err = -ERESTART;
1615                                         goto error;
1616                                 }
1617                                 if (nx == -EAGAIN ||
1618                                     genid != atomic_read(&flow_cache_genid)) {
1619                                         xfrm_pols_put(pols, npols);
1620                                         goto restart;
1621                                 }
1622                                 err = nx;
1623                         }
1624                         if (err < 0)
1625                                 goto error;
1626                 }
1627                 if (nx == 0) {
1628                         /* Flow passes not transformed. */
1629                         xfrm_pols_put(pols, npols);
1630                         return 0;
1631                 }
1632
1633                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1634                 err = PTR_ERR(dst);
1635                 if (IS_ERR(dst))
1636                         goto error;
1637
1638                 for (pi = 0; pi < npols; pi++) {
1639                         read_lock_bh(&pols[pi]->lock);
1640                         pol_dead |= pols[pi]->dead;
1641                         read_unlock_bh(&pols[pi]->lock);
1642                 }
1643
1644                 write_lock_bh(&policy->lock);
1645                 if (unlikely(pol_dead || stale_bundle(dst))) {
1646                         /* Wow! While we worked on resolving, this
1647                          * policy has gone. Retry. It is not paranoia,
1648                          * we just cannot enlist new bundle to dead object.
1649                          * We can't enlist stable bundles either.
1650                          */
1651                         write_unlock_bh(&policy->lock);
1652                         if (dst)
1653                                 dst_free(dst);
1654
1655                         err = -EHOSTUNREACH;
1656                         goto error;
1657                 }
1658
1659                 if (npols > 1)
1660                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1661                 else
1662                         err = xfrm_dst_update_origin(dst, fl);
1663                 if (unlikely(err)) {
1664                         write_unlock_bh(&policy->lock);
1665                         if (dst)
1666                                 dst_free(dst);
1667                         goto error;
1668                 }
1669
1670                 dst->next = policy->bundles;
1671                 policy->bundles = dst;
1672                 dst_hold(dst);
1673                 write_unlock_bh(&policy->lock);
1674         }
1675         *dst_p = dst;
1676         dst_release(dst_orig);
1677         xfrm_pols_put(pols, npols);
1678         return 0;
1679
1680 error:
1681         xfrm_pols_put(pols, npols);
1682 dropdst:
1683         dst_release(dst_orig);
1684         *dst_p = NULL;
1685         return err;
1686
1687 nopol:
1688         err = -ENOENT;
1689         if (flags & XFRM_LOOKUP_ICMP)
1690                 goto dropdst;
1691         return 0;
1692 }
1693 EXPORT_SYMBOL(__xfrm_lookup);
1694
1695 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1696                 struct sock *sk, int flags)
1697 {
1698         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1699
1700         if (err == -EREMOTE) {
1701                 dst_release(*dst_p);
1702                 *dst_p = NULL;
1703                 err = -EAGAIN;
1704         }
1705
1706         return err;
1707 }
1708 EXPORT_SYMBOL(xfrm_lookup);
1709
1710 static inline int
1711 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1712 {
1713         struct xfrm_state *x;
1714
1715         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1716                 return 0;
1717         x = skb->sp->xvec[idx];
1718         if (!x->type->reject)
1719                 return 0;
1720         return x->type->reject(x, skb, fl);
1721 }
1722
1723 /* When skb is transformed back to its "native" form, we have to
1724  * check policy restrictions. At the moment we make this in maximally
1725  * stupid way. Shame on me. :-) Of course, connected sockets must
1726  * have policy cached at them.
1727  */
1728
1729 static inline int
1730 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1731               unsigned short family)
1732 {
1733         if (xfrm_state_kern(x))
1734                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1735         return  x->id.proto == tmpl->id.proto &&
1736                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1737                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1738                 x->props.mode == tmpl->mode &&
1739                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1740                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1741                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1742                   xfrm_state_addr_cmp(tmpl, x, family));
1743 }
1744
1745 /*
1746  * 0 or more than 0 is returned when validation is succeeded (either bypass
1747  * because of optional transport mode, or next index of the mathced secpath
1748  * state with the template.
1749  * -1 is returned when no matching template is found.
1750  * Otherwise "-2 - errored_index" is returned.
1751  */
1752 static inline int
1753 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1754                unsigned short family)
1755 {
1756         int idx = start;
1757
1758         if (tmpl->optional) {
1759                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1760                         return start;
1761         } else
1762                 start = -1;
1763         for (; idx < sp->len; idx++) {
1764                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1765                         return ++idx;
1766                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1767                         if (start == -1)
1768                                 start = -2-idx;
1769                         break;
1770                 }
1771         }
1772         return start;
1773 }
1774
1775 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1776                           unsigned int family, int reverse)
1777 {
1778         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1779         int err;
1780
1781         if (unlikely(afinfo == NULL))
1782                 return -EAFNOSUPPORT;
1783
1784         afinfo->decode_session(skb, fl, reverse);
1785         err = security_xfrm_decode_session(skb, &fl->secid);
1786         xfrm_policy_put_afinfo(afinfo);
1787         return err;
1788 }
1789 EXPORT_SYMBOL(__xfrm_decode_session);
1790
1791 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1792 {
1793         for (; k < sp->len; k++) {
1794                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1795                         *idxp = k;
1796                         return 1;
1797                 }
1798         }
1799
1800         return 0;
1801 }
1802
1803 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1804                         unsigned short family)
1805 {
1806         struct xfrm_policy *pol;
1807         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1808         int npols = 0;
1809         int xfrm_nr;
1810         int pi;
1811         int reverse;
1812         struct flowi fl;
1813         u8 fl_dir;
1814         int xerr_idx = -1;
1815
1816         reverse = dir & ~XFRM_POLICY_MASK;
1817         dir &= XFRM_POLICY_MASK;
1818         fl_dir = policy_to_flow_dir(dir);
1819
1820         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0)
1821                 return 0;
1822         nf_nat_decode_session(skb, &fl, family);
1823
1824         /* First, check used SA against their selectors. */
1825         if (skb->sp) {
1826                 int i;
1827
1828                 for (i=skb->sp->len-1; i>=0; i--) {
1829                         struct xfrm_state *x = skb->sp->xvec[i];
1830                         if (!xfrm_selector_match(&x->sel, &fl, family))
1831                                 return 0;
1832                 }
1833         }
1834
1835         pol = NULL;
1836         if (sk && sk->sk_policy[dir]) {
1837                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1838                 if (IS_ERR(pol))
1839                         return 0;
1840         }
1841
1842         if (!pol)
1843                 pol = flow_cache_lookup(&fl, family, fl_dir,
1844                                         xfrm_policy_lookup);
1845
1846         if (IS_ERR(pol))
1847                 return 0;
1848
1849         if (!pol) {
1850                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1851                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1852                         return 0;
1853                 }
1854                 return 1;
1855         }
1856
1857         pol->curlft.use_time = get_seconds();
1858
1859         pols[0] = pol;
1860         npols ++;
1861 #ifdef CONFIG_XFRM_SUB_POLICY
1862         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1863                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1864                                                     &fl, family,
1865                                                     XFRM_POLICY_IN);
1866                 if (pols[1]) {
1867                         if (IS_ERR(pols[1]))
1868                                 return 0;
1869                         pols[1]->curlft.use_time = get_seconds();
1870                         npols ++;
1871                 }
1872         }
1873 #endif
1874
1875         if (pol->action == XFRM_POLICY_ALLOW) {
1876                 struct sec_path *sp;
1877                 static struct sec_path dummy;
1878                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1879                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1880                 struct xfrm_tmpl **tpp = tp;
1881                 int ti = 0;
1882                 int i, k;
1883
1884                 if ((sp = skb->sp) == NULL)
1885                         sp = &dummy;
1886
1887                 for (pi = 0; pi < npols; pi++) {
1888                         if (pols[pi] != pol &&
1889                             pols[pi]->action != XFRM_POLICY_ALLOW)
1890                                 goto reject;
1891                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1892                                 goto reject_error;
1893                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1894                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1895                 }
1896                 xfrm_nr = ti;
1897                 if (npols > 1) {
1898                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1899                         tpp = stp;
1900                 }
1901
1902                 /* For each tunnel xfrm, find the first matching tmpl.
1903                  * For each tmpl before that, find corresponding xfrm.
1904                  * Order is _important_. Later we will implement
1905                  * some barriers, but at the moment barriers
1906                  * are implied between each two transformations.
1907                  */
1908                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1909                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1910                         if (k < 0) {
1911                                 if (k < -1)
1912                                         /* "-2 - errored_index" returned */
1913                                         xerr_idx = -(2+k);
1914                                 goto reject;
1915                         }
1916                 }
1917
1918                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1919                         goto reject;
1920
1921                 xfrm_pols_put(pols, npols);
1922                 return 1;
1923         }
1924
1925 reject:
1926         xfrm_secpath_reject(xerr_idx, skb, &fl);
1927 reject_error:
1928         xfrm_pols_put(pols, npols);
1929         return 0;
1930 }
1931 EXPORT_SYMBOL(__xfrm_policy_check);
1932
1933 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1934 {
1935         struct flowi fl;
1936
1937         if (xfrm_decode_session(skb, &fl, family) < 0)
1938                 return 0;
1939
1940         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1941 }
1942 EXPORT_SYMBOL(__xfrm_route_forward);
1943
1944 /* Optimize later using cookies and generation ids. */
1945
1946 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1947 {
1948         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1949          * to "-1" to force all XFRM destinations to get validated by
1950          * dst_ops->check on every use.  We do this because when a
1951          * normal route referenced by an XFRM dst is obsoleted we do
1952          * not go looking around for all parent referencing XFRM dsts
1953          * so that we can invalidate them.  It is just too much work.
1954          * Instead we make the checks here on every use.  For example:
1955          *
1956          *      XFRM dst A --> IPv4 dst X
1957          *
1958          * X is the "xdst->route" of A (X is also the "dst->path" of A
1959          * in this example).  If X is marked obsolete, "A" will not
1960          * notice.  That's what we are validating here via the
1961          * stale_bundle() check.
1962          *
1963          * When a policy's bundle is pruned, we dst_free() the XFRM
1964          * dst which causes it's ->obsolete field to be set to a
1965          * positive non-zero integer.  If an XFRM dst has been pruned
1966          * like this, we want to force a new route lookup.
1967          */
1968         if (dst->obsolete < 0 && !stale_bundle(dst))
1969                 return dst;
1970
1971         return NULL;
1972 }
1973
1974 static int stale_bundle(struct dst_entry *dst)
1975 {
1976         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1977 }
1978
1979 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1980 {
1981         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1982                 dst->dev = dev->nd_net->loopback_dev;
1983                 dev_hold(dst->dev);
1984                 dev_put(dev);
1985         }
1986 }
1987 EXPORT_SYMBOL(xfrm_dst_ifdown);
1988
1989 static void xfrm_link_failure(struct sk_buff *skb)
1990 {
1991         /* Impossible. Such dst must be popped before reaches point of failure. */
1992         return;
1993 }
1994
1995 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1996 {
1997         if (dst) {
1998                 if (dst->obsolete) {
1999                         dst_release(dst);
2000                         dst = NULL;
2001                 }
2002         }
2003         return dst;
2004 }
2005
2006 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2007 {
2008         struct dst_entry *dst, **dstp;
2009
2010         write_lock(&pol->lock);
2011         dstp = &pol->bundles;
2012         while ((dst=*dstp) != NULL) {
2013                 if (func(dst)) {
2014                         *dstp = dst->next;
2015                         dst->next = *gc_list_p;
2016                         *gc_list_p = dst;
2017                 } else {
2018                         dstp = &dst->next;
2019                 }
2020         }
2021         write_unlock(&pol->lock);
2022 }
2023
2024 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2025 {
2026         struct dst_entry *gc_list = NULL;
2027         int dir;
2028
2029         read_lock_bh(&xfrm_policy_lock);
2030         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2031                 struct xfrm_policy *pol;
2032                 struct hlist_node *entry;
2033                 struct hlist_head *table;
2034                 int i;
2035
2036                 hlist_for_each_entry(pol, entry,
2037                                      &xfrm_policy_inexact[dir], bydst)
2038                         prune_one_bundle(pol, func, &gc_list);
2039
2040                 table = xfrm_policy_bydst[dir].table;
2041                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2042                         hlist_for_each_entry(pol, entry, table + i, bydst)
2043                                 prune_one_bundle(pol, func, &gc_list);
2044                 }
2045         }
2046         read_unlock_bh(&xfrm_policy_lock);
2047
2048         while (gc_list) {
2049                 struct dst_entry *dst = gc_list;
2050                 gc_list = dst->next;
2051                 dst_free(dst);
2052         }
2053 }
2054
2055 static int unused_bundle(struct dst_entry *dst)
2056 {
2057         return !atomic_read(&dst->__refcnt);
2058 }
2059
2060 static void __xfrm_garbage_collect(void)
2061 {
2062         xfrm_prune_bundles(unused_bundle);
2063 }
2064
2065 static int xfrm_flush_bundles(void)
2066 {
2067         xfrm_prune_bundles(stale_bundle);
2068         return 0;
2069 }
2070
2071 static void xfrm_init_pmtu(struct dst_entry *dst)
2072 {
2073         do {
2074                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2075                 u32 pmtu, route_mtu_cached;
2076
2077                 pmtu = dst_mtu(dst->child);
2078                 xdst->child_mtu_cached = pmtu;
2079
2080                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2081
2082                 route_mtu_cached = dst_mtu(xdst->route);
2083                 xdst->route_mtu_cached = route_mtu_cached;
2084
2085                 if (pmtu > route_mtu_cached)
2086                         pmtu = route_mtu_cached;
2087
2088                 dst->metrics[RTAX_MTU-1] = pmtu;
2089         } while ((dst = dst->next));
2090 }
2091
2092 /* Check that the bundle accepts the flow and its components are
2093  * still valid.
2094  */
2095
2096 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2097                 struct flowi *fl, int family, int strict)
2098 {
2099         struct dst_entry *dst = &first->u.dst;
2100         struct xfrm_dst *last;
2101         u32 mtu;
2102
2103         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2104             (dst->dev && !netif_running(dst->dev)))
2105                 return 0;
2106 #ifdef CONFIG_XFRM_SUB_POLICY
2107         if (fl) {
2108                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2109                         return 0;
2110                 if (first->partner &&
2111                     !xfrm_selector_match(first->partner, fl, family))
2112                         return 0;
2113         }
2114 #endif
2115
2116         last = NULL;
2117
2118         do {
2119                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2120
2121                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2122                         return 0;
2123                 if (fl && pol &&
2124                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2125                         return 0;
2126                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2127                         return 0;
2128                 if (xdst->genid != dst->xfrm->genid)
2129                         return 0;
2130
2131                 if (strict && fl &&
2132                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2133                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2134                         return 0;
2135
2136                 mtu = dst_mtu(dst->child);
2137                 if (xdst->child_mtu_cached != mtu) {
2138                         last = xdst;
2139                         xdst->child_mtu_cached = mtu;
2140                 }
2141
2142                 if (!dst_check(xdst->route, xdst->route_cookie))
2143                         return 0;
2144                 mtu = dst_mtu(xdst->route);
2145                 if (xdst->route_mtu_cached != mtu) {
2146                         last = xdst;
2147                         xdst->route_mtu_cached = mtu;
2148                 }
2149
2150                 dst = dst->child;
2151         } while (dst->xfrm);
2152
2153         if (likely(!last))
2154                 return 1;
2155
2156         mtu = last->child_mtu_cached;
2157         for (;;) {
2158                 dst = &last->u.dst;
2159
2160                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2161                 if (mtu > last->route_mtu_cached)
2162                         mtu = last->route_mtu_cached;
2163                 dst->metrics[RTAX_MTU-1] = mtu;
2164
2165                 if (last == first)
2166                         break;
2167
2168                 last = (struct xfrm_dst *)last->u.dst.next;
2169                 last->child_mtu_cached = mtu;
2170         }
2171
2172         return 1;
2173 }
2174
2175 EXPORT_SYMBOL(xfrm_bundle_ok);
2176
2177 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2178 {
2179         int err = 0;
2180         if (unlikely(afinfo == NULL))
2181                 return -EINVAL;
2182         if (unlikely(afinfo->family >= NPROTO))
2183                 return -EAFNOSUPPORT;
2184         write_lock_bh(&xfrm_policy_afinfo_lock);
2185         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2186                 err = -ENOBUFS;
2187         else {
2188                 struct dst_ops *dst_ops = afinfo->dst_ops;
2189                 if (likely(dst_ops->kmem_cachep == NULL))
2190                         dst_ops->kmem_cachep = xfrm_dst_cache;
2191                 if (likely(dst_ops->check == NULL))
2192                         dst_ops->check = xfrm_dst_check;
2193                 if (likely(dst_ops->negative_advice == NULL))
2194                         dst_ops->negative_advice = xfrm_negative_advice;
2195                 if (likely(dst_ops->link_failure == NULL))
2196                         dst_ops->link_failure = xfrm_link_failure;
2197                 if (likely(afinfo->garbage_collect == NULL))
2198                         afinfo->garbage_collect = __xfrm_garbage_collect;
2199                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2200         }
2201         write_unlock_bh(&xfrm_policy_afinfo_lock);
2202         return err;
2203 }
2204 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2205
2206 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2207 {
2208         int err = 0;
2209         if (unlikely(afinfo == NULL))
2210                 return -EINVAL;
2211         if (unlikely(afinfo->family >= NPROTO))
2212                 return -EAFNOSUPPORT;
2213         write_lock_bh(&xfrm_policy_afinfo_lock);
2214         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2215                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2216                         err = -EINVAL;
2217                 else {
2218                         struct dst_ops *dst_ops = afinfo->dst_ops;
2219                         xfrm_policy_afinfo[afinfo->family] = NULL;
2220                         dst_ops->kmem_cachep = NULL;
2221                         dst_ops->check = NULL;
2222                         dst_ops->negative_advice = NULL;
2223                         dst_ops->link_failure = NULL;
2224                         afinfo->garbage_collect = NULL;
2225                 }
2226         }
2227         write_unlock_bh(&xfrm_policy_afinfo_lock);
2228         return err;
2229 }
2230 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2231
2232 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2233 {
2234         struct xfrm_policy_afinfo *afinfo;
2235         if (unlikely(family >= NPROTO))
2236                 return NULL;
2237         read_lock(&xfrm_policy_afinfo_lock);
2238         afinfo = xfrm_policy_afinfo[family];
2239         if (unlikely(!afinfo))
2240                 read_unlock(&xfrm_policy_afinfo_lock);
2241         return afinfo;
2242 }
2243
2244 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2245 {
2246         read_unlock(&xfrm_policy_afinfo_lock);
2247 }
2248
2249 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2250 {
2251         struct net_device *dev = ptr;
2252
2253         if (dev->nd_net != &init_net)
2254                 return NOTIFY_DONE;
2255
2256         switch (event) {
2257         case NETDEV_DOWN:
2258                 xfrm_flush_bundles();
2259         }
2260         return NOTIFY_DONE;
2261 }
2262
2263 static struct notifier_block xfrm_dev_notifier = {
2264         xfrm_dev_event,
2265         NULL,
2266         0
2267 };
2268
2269 #ifdef CONFIG_XFRM_STATISTICS
2270 static int __init xfrm_statistics_init(void)
2271 {
2272         if (snmp_mib_init((void **)xfrm_statistics,
2273                           sizeof(struct linux_xfrm_mib)) < 0)
2274                 return -ENOMEM;
2275         return 0;
2276 }
2277 #endif
2278
2279 static void __init xfrm_policy_init(void)
2280 {
2281         unsigned int hmask, sz;
2282         int dir;
2283
2284         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2285                                            sizeof(struct xfrm_dst),
2286                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2287                                            NULL);
2288
2289         hmask = 8 - 1;
2290         sz = (hmask+1) * sizeof(struct hlist_head);
2291
2292         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2293         xfrm_idx_hmask = hmask;
2294         if (!xfrm_policy_byidx)
2295                 panic("XFRM: failed to allocate byidx hash\n");
2296
2297         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2298                 struct xfrm_policy_hash *htab;
2299
2300                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2301
2302                 htab = &xfrm_policy_bydst[dir];
2303                 htab->table = xfrm_hash_alloc(sz);
2304                 htab->hmask = hmask;
2305                 if (!htab->table)
2306                         panic("XFRM: failed to allocate bydst hash\n");
2307         }
2308
2309         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2310         register_netdevice_notifier(&xfrm_dev_notifier);
2311 }
2312
2313 void __init xfrm_init(void)
2314 {
2315 #ifdef CONFIG_XFRM_STATISTICS
2316         xfrm_statistics_init();
2317 #endif
2318         xfrm_state_init();
2319         xfrm_policy_init();
2320         xfrm_input_init();
2321 #ifdef CONFIG_XFRM_STATISTICS
2322         xfrm_proc_init();
2323 #endif
2324 }
2325
2326 #ifdef CONFIG_AUDITSYSCALL
2327 static inline void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2328                                                 struct audit_buffer *audit_buf)
2329 {
2330         struct xfrm_sec_ctx *ctx = xp->security;
2331         struct xfrm_selector *sel = &xp->selector;
2332
2333         if (ctx)
2334                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2335                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2336
2337         switch(sel->family) {
2338         case AF_INET:
2339                 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2340                                  NIPQUAD(sel->saddr.a4));
2341                 if (sel->prefixlen_s != 32)
2342                         audit_log_format(audit_buf, " src_prefixlen=%d",
2343                                          sel->prefixlen_s);
2344                 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2345                                  NIPQUAD(sel->daddr.a4));
2346                 if (sel->prefixlen_d != 32)
2347                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2348                                          sel->prefixlen_d);
2349                 break;
2350         case AF_INET6:
2351                 audit_log_format(audit_buf, " src=" NIP6_FMT,
2352                                  NIP6(*(struct in6_addr *)sel->saddr.a6));
2353                 if (sel->prefixlen_s != 128)
2354                         audit_log_format(audit_buf, " src_prefixlen=%d",
2355                                          sel->prefixlen_s);
2356                 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2357                                  NIP6(*(struct in6_addr *)sel->daddr.a6));
2358                 if (sel->prefixlen_d != 128)
2359                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2360                                          sel->prefixlen_d);
2361                 break;
2362         }
2363 }
2364
2365 void
2366 xfrm_audit_policy_add(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2367 {
2368         struct audit_buffer *audit_buf;
2369         extern int audit_enabled;
2370
2371         if (audit_enabled == 0)
2372                 return;
2373         audit_buf = xfrm_audit_start(auid, sid);
2374         if (audit_buf == NULL)
2375                 return;
2376         audit_log_format(audit_buf, " op=SPD-add res=%u", result);
2377         xfrm_audit_common_policyinfo(xp, audit_buf);
2378         audit_log_end(audit_buf);
2379 }
2380 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2381
2382 void
2383 xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2384 {
2385         struct audit_buffer *audit_buf;
2386         extern int audit_enabled;
2387
2388         if (audit_enabled == 0)
2389                 return;
2390         audit_buf = xfrm_audit_start(auid, sid);
2391         if (audit_buf == NULL)
2392                 return;
2393         audit_log_format(audit_buf, " op=SPD-delete res=%u", result);
2394         xfrm_audit_common_policyinfo(xp, audit_buf);
2395         audit_log_end(audit_buf);
2396 }
2397 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2398 #endif
2399
2400 #ifdef CONFIG_XFRM_MIGRATE
2401 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2402                                        struct xfrm_selector *sel_tgt)
2403 {
2404         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2405                 if (sel_tgt->family == sel_cmp->family &&
2406                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2407                                   sel_cmp->family) == 0 &&
2408                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2409                                   sel_cmp->family) == 0 &&
2410                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2411                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2412                         return 1;
2413                 }
2414         } else {
2415                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2416                         return 1;
2417                 }
2418         }
2419         return 0;
2420 }
2421
2422 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2423                                                      u8 dir, u8 type)
2424 {
2425         struct xfrm_policy *pol, *ret = NULL;
2426         struct hlist_node *entry;
2427         struct hlist_head *chain;
2428         u32 priority = ~0U;
2429
2430         read_lock_bh(&xfrm_policy_lock);
2431         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2432         hlist_for_each_entry(pol, entry, chain, bydst) {
2433                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2434                     pol->type == type) {
2435                         ret = pol;
2436                         priority = ret->priority;
2437                         break;
2438                 }
2439         }
2440         chain = &xfrm_policy_inexact[dir];
2441         hlist_for_each_entry(pol, entry, chain, bydst) {
2442                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2443                     pol->type == type &&
2444                     pol->priority < priority) {
2445                         ret = pol;
2446                         break;
2447                 }
2448         }
2449
2450         if (ret)
2451                 xfrm_pol_hold(ret);
2452
2453         read_unlock_bh(&xfrm_policy_lock);
2454
2455         return ret;
2456 }
2457
2458 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2459 {
2460         int match = 0;
2461
2462         if (t->mode == m->mode && t->id.proto == m->proto &&
2463             (m->reqid == 0 || t->reqid == m->reqid)) {
2464                 switch (t->mode) {
2465                 case XFRM_MODE_TUNNEL:
2466                 case XFRM_MODE_BEET:
2467                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2468                                           m->old_family) == 0 &&
2469                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2470                                           m->old_family) == 0) {
2471                                 match = 1;
2472                         }
2473                         break;
2474                 case XFRM_MODE_TRANSPORT:
2475                         /* in case of transport mode, template does not store
2476                            any IP addresses, hence we just compare mode and
2477                            protocol */
2478                         match = 1;
2479                         break;
2480                 default:
2481                         break;
2482                 }
2483         }
2484         return match;
2485 }
2486
2487 /* update endpoint address(es) of template(s) */
2488 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2489                                struct xfrm_migrate *m, int num_migrate)
2490 {
2491         struct xfrm_migrate *mp;
2492         struct dst_entry *dst;
2493         int i, j, n = 0;
2494
2495         write_lock_bh(&pol->lock);
2496         if (unlikely(pol->dead)) {
2497                 /* target policy has been deleted */
2498                 write_unlock_bh(&pol->lock);
2499                 return -ENOENT;
2500         }
2501
2502         for (i = 0; i < pol->xfrm_nr; i++) {
2503                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2504                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2505                                 continue;
2506                         n++;
2507                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2508                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2509                                 continue;
2510                         /* update endpoints */
2511                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2512                                sizeof(pol->xfrm_vec[i].id.daddr));
2513                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2514                                sizeof(pol->xfrm_vec[i].saddr));
2515                         pol->xfrm_vec[i].encap_family = mp->new_family;
2516                         /* flush bundles */
2517                         while ((dst = pol->bundles) != NULL) {
2518                                 pol->bundles = dst->next;
2519                                 dst_free(dst);
2520                         }
2521                 }
2522         }
2523
2524         write_unlock_bh(&pol->lock);
2525
2526         if (!n)
2527                 return -ENODATA;
2528
2529         return 0;
2530 }
2531
2532 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2533 {
2534         int i, j;
2535
2536         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2537                 return -EINVAL;
2538
2539         for (i = 0; i < num_migrate; i++) {
2540                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2541                                    m[i].old_family) == 0) &&
2542                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2543                                    m[i].old_family) == 0))
2544                         return -EINVAL;
2545                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2546                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2547                         return -EINVAL;
2548
2549                 /* check if there is any duplicated entry */
2550                 for (j = i + 1; j < num_migrate; j++) {
2551                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2552                                     sizeof(m[i].old_daddr)) &&
2553                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2554                                     sizeof(m[i].old_saddr)) &&
2555                             m[i].proto == m[j].proto &&
2556                             m[i].mode == m[j].mode &&
2557                             m[i].reqid == m[j].reqid &&
2558                             m[i].old_family == m[j].old_family)
2559                                 return -EINVAL;
2560                 }
2561         }
2562
2563         return 0;
2564 }
2565
2566 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2567                  struct xfrm_migrate *m, int num_migrate)
2568 {
2569         int i, err, nx_cur = 0, nx_new = 0;
2570         struct xfrm_policy *pol = NULL;
2571         struct xfrm_state *x, *xc;
2572         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2573         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2574         struct xfrm_migrate *mp;
2575
2576         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2577                 goto out;
2578
2579         /* Stage 1 - find policy */
2580         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2581                 err = -ENOENT;
2582                 goto out;
2583         }
2584
2585         /* Stage 2 - find and update state(s) */
2586         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2587                 if ((x = xfrm_migrate_state_find(mp))) {
2588                         x_cur[nx_cur] = x;
2589                         nx_cur++;
2590                         if ((xc = xfrm_state_migrate(x, mp))) {
2591                                 x_new[nx_new] = xc;
2592                                 nx_new++;
2593                         } else {
2594                                 err = -ENODATA;
2595                                 goto restore_state;
2596                         }
2597                 }
2598         }
2599
2600         /* Stage 3 - update policy */
2601         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2602                 goto restore_state;
2603
2604         /* Stage 4 - delete old state(s) */
2605         if (nx_cur) {
2606                 xfrm_states_put(x_cur, nx_cur);
2607                 xfrm_states_delete(x_cur, nx_cur);
2608         }
2609
2610         /* Stage 5 - announce */
2611         km_migrate(sel, dir, type, m, num_migrate);
2612
2613         xfrm_pol_put(pol);
2614
2615         return 0;
2616 out:
2617         return err;
2618
2619 restore_state:
2620         if (pol)
2621                 xfrm_pol_put(pol);
2622         if (nx_cur)
2623                 xfrm_states_put(x_cur, nx_cur);
2624         if (nx_new)
2625                 xfrm_states_delete(x_new, nx_new);
2626
2627         return err;
2628 }
2629 EXPORT_SYMBOL(xfrm_migrate);
2630 #endif