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

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/xfrm.h>
  30 #include <net/ip.h>
  31 #ifdef CONFIG_XFRM_STATISTICS
  32 #include <net/snmp.h>
  33 #endif
  34
  35 #include "xfrm_hash.h"
  36
  37 int sysctl_xfrm_larval_drop __read_mostly;
  38
  39 #ifdef CONFIG_XFRM_STATISTICS
  40 DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
  41 EXPORT_SYMBOL(xfrm_statistics);
  42 #endif
  43
  44 DEFINE_MUTEX(xfrm_cfg_mutex);
  45 EXPORT_SYMBOL(xfrm_cfg_mutex);
  46
  47 static DEFINE_RWLOCK(xfrm_policy_lock);
  48
  49 static struct list_head xfrm_policy_bytype[XFRM_POLICY_TYPE_MAX];
  50 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  51 EXPORT_SYMBOL(xfrm_policy_count);
  52
  53 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  54 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  55
  56 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  57
  58 static struct work_struct xfrm_policy_gc_work;
  59 static HLIST_HEAD(xfrm_policy_gc_list);
  60 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  61
  62 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  63 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  64 static void xfrm_init_pmtu(struct dst_entry *dst);
  65
  66 static inline int
  67 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  68 {
  69         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  70                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  71                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  72                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  73                 (fl->proto == sel->proto || !sel->proto) &&
  74                 (fl->oif == sel->ifindex || !sel->ifindex);
  75 }
  76
  77 static inline int
  78 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  79 {
  80         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  81                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  82                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  83                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  84                 (fl->proto == sel->proto || !sel->proto) &&
  85                 (fl->oif == sel->ifindex || !sel->ifindex);
  86 }
  87
  88 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  89                     unsigned short family)
  90 {
  91         switch (family) {
  92         case AF_INET:
  93                 return __xfrm4_selector_match(sel, fl);
  94         case AF_INET6:
  95                 return __xfrm6_selector_match(sel, fl);
  96         }
  97         return 0;
  98 }
  99
 100 static inline struct dst_entry *__xfrm_dst_lookup(int tos,
 101                                                   xfrm_address_t *saddr,
 102                                                   xfrm_address_t *daddr,
 103                                                   int family)
 104 {
 105         struct xfrm_policy_afinfo *afinfo;
 106         struct dst_entry *dst;
 107
 108         afinfo = xfrm_policy_get_afinfo(family);
 109         if (unlikely(afinfo == NULL))
 110                 return ERR_PTR(-EAFNOSUPPORT);
 111
 112         dst = afinfo->dst_lookup(tos, saddr, daddr);
 113
 114         xfrm_policy_put_afinfo(afinfo);
 115
 116         return dst;
 117 }
 118
 119 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 120                                                 xfrm_address_t *prev_saddr,
 121                                                 xfrm_address_t *prev_daddr,
 122                                                 int family)
 123 {
 124         xfrm_address_t *saddr = &x->props.saddr;
 125         xfrm_address_t *daddr = &x->id.daddr;
 126         struct dst_entry *dst;
 127
 128         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 129                 saddr = x->coaddr;
 130                 daddr = prev_daddr;
 131         }
 132         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 133                 saddr = prev_saddr;
 134                 daddr = x->coaddr;
 135         }
 136
 137         dst = __xfrm_dst_lookup(tos, saddr, daddr, family);
 138
 139         if (!IS_ERR(dst)) {
 140                 if (prev_saddr != saddr)
 141                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 142                 if (prev_daddr != daddr)
 143                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 144         }
 145
 146         return dst;
 147 }
 148
 149 static inline unsigned long make_jiffies(long secs)
 150 {
 151         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 152                 return MAX_SCHEDULE_TIMEOUT-1;
 153         else
 154                 return secs*HZ;
 155 }
 156
 157 static void xfrm_policy_timer(unsigned long data)
 158 {
 159         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 160         unsigned long now = get_seconds();
 161         long next = LONG_MAX;
 162         int warn = 0;
 163         int dir;
 164
 165         read_lock(&xp->lock);
 166
 167         if (xp->dead)
 168                 goto out;
 169
 170         dir = xfrm_policy_id2dir(xp->index);
 171
 172         if (xp->lft.hard_add_expires_seconds) {
 173                 long tmo = xp->lft.hard_add_expires_seconds +
 174                         xp->curlft.add_time - now;
 175                 if (tmo <= 0)
 176                         goto expired;
 177                 if (tmo < next)
 178                         next = tmo;
 179         }
 180         if (xp->lft.hard_use_expires_seconds) {
 181                 long tmo = xp->lft.hard_use_expires_seconds +
 182                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 183                 if (tmo <= 0)
 184                         goto expired;
 185                 if (tmo < next)
 186                         next = tmo;
 187         }
 188         if (xp->lft.soft_add_expires_seconds) {
 189                 long tmo = xp->lft.soft_add_expires_seconds +
 190                         xp->curlft.add_time - now;
 191                 if (tmo <= 0) {
 192                         warn = 1;
 193                         tmo = XFRM_KM_TIMEOUT;
 194                 }
 195                 if (tmo < next)
 196                         next = tmo;
 197         }
 198         if (xp->lft.soft_use_expires_seconds) {
 199                 long tmo = xp->lft.soft_use_expires_seconds +
 200                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 201                 if (tmo <= 0) {
 202                         warn = 1;
 203                         tmo = XFRM_KM_TIMEOUT;
 204                 }
 205                 if (tmo < next)
 206                         next = tmo;
 207         }
 208
 209         if (warn)
 210                 km_policy_expired(xp, dir, 0, 0);
 211         if (next != LONG_MAX &&
 212             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 213                 xfrm_pol_hold(xp);
 214
 215 out:
 216         read_unlock(&xp->lock);
 217         xfrm_pol_put(xp);
 218         return;
 219
 220 expired:
 221         read_unlock(&xp->lock);
 222         if (!xfrm_policy_delete(xp, dir))
 223                 km_policy_expired(xp, dir, 1, 0);
 224         xfrm_pol_put(xp);
 225 }
 226
 227
 228 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 229  * SPD calls.
 230  */
 231
 232 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 233 {
 234         struct xfrm_policy *policy;
 235
 236         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 237
 238         if (policy) {
 239                 INIT_LIST_HEAD(&policy->bytype);
 240                 INIT_HLIST_NODE(&policy->bydst);
 241                 INIT_HLIST_NODE(&policy->byidx);
 242                 rwlock_init(&policy->lock);
 243                 atomic_set(&policy->refcnt, 1);
 244                 setup_timer(&policy->timer, xfrm_policy_timer,
 245                                 (unsigned long)policy);
 246         }
 247         return policy;
 248 }
 249 EXPORT_SYMBOL(xfrm_policy_alloc);
 250
 251 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 252
 253 void xfrm_policy_destroy(struct xfrm_policy *policy)
 254 {
 255         BUG_ON(!policy->dead);
 256
 257         BUG_ON(policy->bundles);
 258
 259         if (del_timer(&policy->timer))
 260                 BUG();
 261
 262         write_lock_bh(&xfrm_policy_lock);
 263         list_del(&policy->bytype);
 264         write_unlock_bh(&xfrm_policy_lock);
 265
 266         security_xfrm_policy_free(policy->security);
 267         kfree(policy);
 268 }
 269 EXPORT_SYMBOL(xfrm_policy_destroy);
 270
 271 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 272 {
 273         struct dst_entry *dst;
 274
 275         while ((dst = policy->bundles) != NULL) {
 276                 policy->bundles = dst->next;
 277                 dst_free(dst);
 278         }
 279
 280         if (del_timer(&policy->timer))
 281                 atomic_dec(&policy->refcnt);
 282
 283         if (atomic_read(&policy->refcnt) > 1)
 284                 flow_cache_flush();
 285
 286         xfrm_pol_put(policy);
 287 }
 288
 289 static void xfrm_policy_gc_task(struct work_struct *work)
 290 {
 291         struct xfrm_policy *policy;
 292         struct hlist_node *entry, *tmp;
 293         struct hlist_head gc_list;
 294
 295         spin_lock_bh(&xfrm_policy_gc_lock);
 296         gc_list.first = xfrm_policy_gc_list.first;
 297         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 298         spin_unlock_bh(&xfrm_policy_gc_lock);
 299
 300         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 301                 xfrm_policy_gc_kill(policy);
 302 }
 303
 304 /* Rule must be locked. Release descentant resources, announce
 305  * entry dead. The rule must be unlinked from lists to the moment.
 306  */
 307
 308 static void xfrm_policy_kill(struct xfrm_policy *policy)
 309 {
 310         int dead;
 311
 312         write_lock_bh(&policy->lock);
 313         dead = policy->dead;
 314         policy->dead = 1;
 315         write_unlock_bh(&policy->lock);
 316
 317         if (unlikely(dead)) {
 318                 WARN_ON(1);
 319                 return;
 320         }
 321
 322         spin_lock(&xfrm_policy_gc_lock);
 323         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 324         spin_unlock(&xfrm_policy_gc_lock);
 325
 326         schedule_work(&xfrm_policy_gc_work);
 327 }
 328
 329 struct xfrm_policy_hash {
 330         struct hlist_head       *table;
 331         unsigned int            hmask;
 332 };
 333
 334 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 335 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 336 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 337 static unsigned int xfrm_idx_hmask __read_mostly;
 338 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 339
 340 static inline unsigned int idx_hash(u32 index)
 341 {
 342         return __idx_hash(index, xfrm_idx_hmask);
 343 }
 344
 345 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 346 {
 347         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 348         unsigned int hash = __sel_hash(sel, family, hmask);
 349
 350         return (hash == hmask + 1 ?
 351                 &xfrm_policy_inexact[dir] :
 352                 xfrm_policy_bydst[dir].table + hash);
 353 }
 354
 355 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 356 {
 357         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 358         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 359
 360         return xfrm_policy_bydst[dir].table + hash;
 361 }
 362
 363 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 364                                    struct hlist_head *ndsttable,
 365                                    unsigned int nhashmask)
 366 {
 367         struct hlist_node *entry, *tmp, *entry0 = NULL;
 368         struct xfrm_policy *pol;
 369         unsigned int h0 = 0;
 370
 371 redo:
 372         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 373                 unsigned int h;
 374
 375                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 376                                 pol->family, nhashmask);
 377                 if (!entry0) {
 378                         hlist_del(entry);
 379                         hlist_add_head(&pol->bydst, ndsttable+h);
 380                         h0 = h;
 381                 } else {
 382                         if (h != h0)
 383                                 continue;
 384                         hlist_del(entry);
 385                         hlist_add_after(entry0, &pol->bydst);
 386                 }
 387                 entry0 = entry;
 388         }
 389         if (!hlist_empty(list)) {
 390                 entry0 = NULL;
 391                 goto redo;
 392         }
 393 }
 394
 395 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 396                                    struct hlist_head *nidxtable,
 397                                    unsigned int nhashmask)
 398 {
 399         struct hlist_node *entry, *tmp;
 400         struct xfrm_policy *pol;
 401
 402         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 403                 unsigned int h;
 404
 405                 h = __idx_hash(pol->index, nhashmask);
 406                 hlist_add_head(&pol->byidx, nidxtable+h);
 407         }
 408 }
 409
 410 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 411 {
 412         return ((old_hmask + 1) << 1) - 1;
 413 }
 414
 415 static void xfrm_bydst_resize(int dir)
 416 {
 417         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 418         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 419         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 420         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 421         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 422         int i;
 423
 424         if (!ndst)
 425                 return;
 426
 427         write_lock_bh(&xfrm_policy_lock);
 428
 429         for (i = hmask; i >= 0; i--)
 430                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 431
 432         xfrm_policy_bydst[dir].table = ndst;
 433         xfrm_policy_bydst[dir].hmask = nhashmask;
 434
 435         write_unlock_bh(&xfrm_policy_lock);
 436
 437         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 438 }
 439
 440 static void xfrm_byidx_resize(int total)
 441 {
 442         unsigned int hmask = xfrm_idx_hmask;
 443         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 444         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 445         struct hlist_head *oidx = xfrm_policy_byidx;
 446         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 447         int i;
 448
 449         if (!nidx)
 450                 return;
 451
 452         write_lock_bh(&xfrm_policy_lock);
 453
 454         for (i = hmask; i >= 0; i--)
 455                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 456
 457         xfrm_policy_byidx = nidx;
 458         xfrm_idx_hmask = nhashmask;
 459
 460         write_unlock_bh(&xfrm_policy_lock);
 461
 462         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 463 }
 464
 465 static inline int xfrm_bydst_should_resize(int dir, int *total)
 466 {
 467         unsigned int cnt = xfrm_policy_count[dir];
 468         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 469
 470         if (total)
 471                 *total += cnt;
 472
 473         if ((hmask + 1) < xfrm_policy_hashmax &&
 474             cnt > hmask)
 475                 return 1;
 476
 477         return 0;
 478 }
 479
 480 static inline int xfrm_byidx_should_resize(int total)
 481 {
 482         unsigned int hmask = xfrm_idx_hmask;
 483
 484         if ((hmask + 1) < xfrm_policy_hashmax &&
 485             total > hmask)
 486                 return 1;
 487
 488         return 0;
 489 }
 490
 491 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 492 {
 493         read_lock_bh(&xfrm_policy_lock);
 494         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 495         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 496         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 497         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 498         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 499         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 500         si->spdhcnt = xfrm_idx_hmask;
 501         si->spdhmcnt = xfrm_policy_hashmax;
 502         read_unlock_bh(&xfrm_policy_lock);
 503 }
 504 EXPORT_SYMBOL(xfrm_spd_getinfo);
 505
 506 static DEFINE_MUTEX(hash_resize_mutex);
 507 static void xfrm_hash_resize(struct work_struct *__unused)
 508 {
 509         int dir, total;
 510
 511         mutex_lock(&hash_resize_mutex);
 512
 513         total = 0;
 514         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 515                 if (xfrm_bydst_should_resize(dir, &total))
 516                         xfrm_bydst_resize(dir);
 517         }
 518         if (xfrm_byidx_should_resize(total))
 519                 xfrm_byidx_resize(total);
 520
 521         mutex_unlock(&hash_resize_mutex);
 522 }
 523
 524 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 525
 526 /* Generate new index... KAME seems to generate them ordered by cost
 527  * of an absolute inpredictability of ordering of rules. This will not pass. */
 528 static u32 xfrm_gen_index(u8 type, int dir)
 529 {
 530         static u32 idx_generator;
 531
 532         for (;;) {
 533                 struct hlist_node *entry;
 534                 struct hlist_head *list;
 535                 struct xfrm_policy *p;
 536                 u32 idx;
 537                 int found;
 538
 539                 idx = (idx_generator | dir);
 540                 idx_generator += 8;
 541                 if (idx == 0)
 542                         idx = 8;
 543                 list = xfrm_policy_byidx + idx_hash(idx);
 544                 found = 0;
 545                 hlist_for_each_entry(p, entry, list, byidx) {
 546                         if (p->index == idx) {
 547                                 found = 1;
 548                                 break;
 549                         }
 550                 }
 551                 if (!found)
 552                         return idx;
 553         }
 554 }
 555
 556 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 557 {
 558         u32 *p1 = (u32 *) s1;
 559         u32 *p2 = (u32 *) s2;
 560         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 561         int i;
 562
 563         for (i = 0; i < len; i++) {
 564                 if (p1[i] != p2[i])
 565                         return 1;
 566         }
 567
 568         return 0;
 569 }
 570
 571 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 572 {
 573         struct xfrm_policy *pol;
 574         struct xfrm_policy *delpol;
 575         struct hlist_head *chain;
 576         struct hlist_node *entry, *newpos;
 577         struct dst_entry *gc_list;
 578
 579         write_lock_bh(&xfrm_policy_lock);
 580         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 581         delpol = NULL;
 582         newpos = NULL;
 583         hlist_for_each_entry(pol, entry, chain, bydst) {
 584                 if (pol->type == policy->type &&
 585                     !selector_cmp(&pol->selector, &policy->selector) &&
 586                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 587                     !WARN_ON(delpol)) {
 588                         if (excl) {
 589                                 write_unlock_bh(&xfrm_policy_lock);
 590                                 return -EEXIST;
 591                         }
 592                         delpol = pol;
 593                         if (policy->priority > pol->priority)
 594                                 continue;
 595                 } else if (policy->priority >= pol->priority) {
 596                         newpos = &pol->bydst;
 597                         continue;
 598                 }
 599                 if (delpol)
 600                         break;
 601         }
 602         if (newpos)
 603                 hlist_add_after(newpos, &policy->bydst);
 604         else
 605                 hlist_add_head(&policy->bydst, chain);
 606         xfrm_pol_hold(policy);
 607         xfrm_policy_count[dir]++;
 608         atomic_inc(&flow_cache_genid);
 609         if (delpol) {
 610                 hlist_del(&delpol->bydst);
 611                 hlist_del(&delpol->byidx);
 612                 xfrm_policy_count[dir]--;
 613         }
 614         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 615         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 616         policy->curlft.add_time = get_seconds();
 617         policy->curlft.use_time = 0;
 618         if (!mod_timer(&policy->timer, jiffies + HZ))
 619                 xfrm_pol_hold(policy);
 620         list_add_tail(&policy->bytype, &xfrm_policy_bytype[policy->type]);
 621         write_unlock_bh(&xfrm_policy_lock);
 622
 623         if (delpol)
 624                 xfrm_policy_kill(delpol);
 625         else if (xfrm_bydst_should_resize(dir, NULL))
 626                 schedule_work(&xfrm_hash_work);
 627
 628         read_lock_bh(&xfrm_policy_lock);
 629         gc_list = NULL;
 630         entry = &policy->bydst;
 631         hlist_for_each_entry_continue(policy, entry, bydst) {
 632                 struct dst_entry *dst;
 633
 634                 write_lock(&policy->lock);
 635                 dst = policy->bundles;
 636                 if (dst) {
 637                         struct dst_entry *tail = dst;
 638                         while (tail->next)
 639                                 tail = tail->next;
 640                         tail->next = gc_list;
 641                         gc_list = dst;
 642
 643                         policy->bundles = NULL;
 644                 }
 645                 write_unlock(&policy->lock);
 646         }
 647         read_unlock_bh(&xfrm_policy_lock);
 648
 649         while (gc_list) {
 650                 struct dst_entry *dst = gc_list;
 651
 652                 gc_list = dst->next;
 653                 dst_free(dst);
 654         }
 655
 656         return 0;
 657 }
 658 EXPORT_SYMBOL(xfrm_policy_insert);
 659
 660 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 661                                           struct xfrm_selector *sel,
 662                                           struct xfrm_sec_ctx *ctx, int delete,
 663                                           int *err)
 664 {
 665         struct xfrm_policy *pol, *ret;
 666         struct hlist_head *chain;
 667         struct hlist_node *entry;
 668
 669         *err = 0;
 670         write_lock_bh(&xfrm_policy_lock);
 671         chain = policy_hash_bysel(sel, sel->family, dir);
 672         ret = NULL;
 673         hlist_for_each_entry(pol, entry, chain, bydst) {
 674                 if (pol->type == type &&
 675                     !selector_cmp(sel, &pol->selector) &&
 676                     xfrm_sec_ctx_match(ctx, pol->security)) {
 677                         xfrm_pol_hold(pol);
 678                         if (delete) {
 679                                 *err = security_xfrm_policy_delete(
 680                                                                 pol->security);
 681                                 if (*err) {
 682                                         write_unlock_bh(&xfrm_policy_lock);
 683                                         return pol;
 684                                 }
 685                                 hlist_del(&pol->bydst);
 686                                 hlist_del(&pol->byidx);
 687                                 xfrm_policy_count[dir]--;
 688                         }
 689                         ret = pol;
 690                         break;
 691                 }
 692         }
 693         write_unlock_bh(&xfrm_policy_lock);
 694
 695         if (ret && delete) {
 696                 atomic_inc(&flow_cache_genid);
 697                 xfrm_policy_kill(ret);
 698         }
 699         return ret;
 700 }
 701 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 702
 703 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 704                                      int *err)
 705 {
 706         struct xfrm_policy *pol, *ret;
 707         struct hlist_head *chain;
 708         struct hlist_node *entry;
 709
 710         *err = -ENOENT;
 711         if (xfrm_policy_id2dir(id) != dir)
 712                 return NULL;
 713
 714         *err = 0;
 715         write_lock_bh(&xfrm_policy_lock);
 716         chain = xfrm_policy_byidx + idx_hash(id);
 717         ret = NULL;
 718         hlist_for_each_entry(pol, entry, chain, byidx) {
 719                 if (pol->type == type && pol->index == id) {
 720                         xfrm_pol_hold(pol);
 721                         if (delete) {
 722                                 *err = security_xfrm_policy_delete(
 723                                                                 pol->security);
 724                                 if (*err) {
 725                                         write_unlock_bh(&xfrm_policy_lock);
 726                                         return pol;
 727                                 }
 728                                 hlist_del(&pol->bydst);
 729                                 hlist_del(&pol->byidx);
 730                                 xfrm_policy_count[dir]--;
 731                         }
 732                         ret = pol;
 733                         break;
 734                 }
 735         }
 736         write_unlock_bh(&xfrm_policy_lock);
 737
 738         if (ret && delete) {
 739                 atomic_inc(&flow_cache_genid);
 740                 xfrm_policy_kill(ret);
 741         }
 742         return ret;
 743 }
 744 EXPORT_SYMBOL(xfrm_policy_byid);
 745
 746 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 747 static inline int
 748 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 749 {
 750         int dir, err = 0;
 751
 752         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 753                 struct xfrm_policy *pol;
 754                 struct hlist_node *entry;
 755                 int i;
 756
 757                 hlist_for_each_entry(pol, entry,
 758                                      &xfrm_policy_inexact[dir], bydst) {
 759                         if (pol->type != type)
 760                                 continue;
 761                         err = security_xfrm_policy_delete(pol->security);
 762                         if (err) {
 763                                 xfrm_audit_policy_delete(pol, 0,
 764                                                          audit_info->loginuid,
 765                                                          audit_info->sessionid,
 766                                                          audit_info->secid);
 767                                 return err;
 768                         }
 769                 }
 770                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 771                         hlist_for_each_entry(pol, entry,
 772                                              xfrm_policy_bydst[dir].table + i,
 773                                              bydst) {
 774                                 if (pol->type != type)
 775                                         continue;
 776                                 err = security_xfrm_policy_delete(
 777                                                                 pol->security);
 778                                 if (err) {
 779                                         xfrm_audit_policy_delete(pol, 0,
 780                                                         audit_info->loginuid,
 781                                                         audit_info->sessionid,
 782                                                         audit_info->secid);
 783                                         return err;
 784                                 }
 785                         }
 786                 }
 787         }
 788         return err;
 789 }
 790 #else
 791 static inline int
 792 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 793 {
 794         return 0;
 795 }
 796 #endif
 797
 798 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 799 {
 800         int dir, err = 0;
 801
 802         write_lock_bh(&xfrm_policy_lock);
 803
 804         err = xfrm_policy_flush_secctx_check(type, audit_info);
 805         if (err)
 806                 goto out;
 807
 808         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 809                 struct xfrm_policy *pol;
 810                 struct hlist_node *entry;
 811                 int i, killed;
 812
 813                 killed = 0;
 814         again1:
 815                 hlist_for_each_entry(pol, entry,
 816                                      &xfrm_policy_inexact[dir], bydst) {
 817                         if (pol->type != type)
 818                                 continue;
 819                         hlist_del(&pol->bydst);
 820                         hlist_del(&pol->byidx);
 821                         write_unlock_bh(&xfrm_policy_lock);
 822
 823                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 824                                                  audit_info->sessionid,
 825                                                  audit_info->secid);
 826
 827                         xfrm_policy_kill(pol);
 828                         killed++;
 829
 830                         write_lock_bh(&xfrm_policy_lock);
 831                         goto again1;
 832                 }
 833
 834                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 835         again2:
 836                         hlist_for_each_entry(pol, entry,
 837                                              xfrm_policy_bydst[dir].table + i,
 838                                              bydst) {
 839                                 if (pol->type != type)
 840                                         continue;
 841                                 hlist_del(&pol->bydst);
 842                                 hlist_del(&pol->byidx);
 843                                 write_unlock_bh(&xfrm_policy_lock);
 844
 845                                 xfrm_audit_policy_delete(pol, 1,
 846                                                          audit_info->loginuid,
 847                                                          audit_info->sessionid,
 848                                                          audit_info->secid);
 849                                 xfrm_policy_kill(pol);
 850                                 killed++;
 851
 852                                 write_lock_bh(&xfrm_policy_lock);
 853                                 goto again2;
 854                         }
 855                 }
 856
 857                 xfrm_policy_count[dir] -= killed;
 858         }
 859         atomic_inc(&flow_cache_genid);
 860 out:
 861         write_unlock_bh(&xfrm_policy_lock);
 862         return err;
 863 }
 864 EXPORT_SYMBOL(xfrm_policy_flush);
 865
 866 int xfrm_policy_walk(struct xfrm_policy_walk *walk,
 867                      int (*func)(struct xfrm_policy *, int, int, void*),
 868                      void *data)
 869 {
 870         struct xfrm_policy *old, *pol, *last = NULL;
 871         int error = 0;
 872
 873         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 874             walk->type != XFRM_POLICY_TYPE_ANY)
 875                 return -EINVAL;
 876
 877         if (walk->policy == NULL && walk->count != 0)
 878                 return 0;
 879
 880         old = pol = walk->policy;
 881         walk->policy = NULL;
 882         read_lock_bh(&xfrm_policy_lock);
 883
 884         for (; walk->cur_type < XFRM_POLICY_TYPE_MAX; walk->cur_type++) {
 885                 if (walk->type != walk->cur_type &&
 886                     walk->type != XFRM_POLICY_TYPE_ANY)
 887                         continue;
 888
 889                 if (pol == NULL) {
 890                         pol = list_first_entry(&xfrm_policy_bytype[walk->cur_type],
 891                                                struct xfrm_policy, bytype);
 892                 }
 893                 list_for_each_entry_from(pol, &xfrm_policy_bytype[walk->cur_type], bytype) {
 894                         if (pol->dead)
 895                                 continue;
 896                         if (last) {
 897                                 error = func(last, xfrm_policy_id2dir(last->index),
 898                                              walk->count, data);
 899                                 if (error) {
 900                                         xfrm_pol_hold(last);
 901                                         walk->policy = last;
 902                                         goto out;
 903                                 }
 904                         }
 905                         last = pol;
 906                         walk->count++;
 907                 }
 908                 pol = NULL;
 909         }
 910         if (walk->count == 0) {
 911                 error = -ENOENT;
 912                 goto out;
 913         }
 914         if (last)
 915                 error = func(last, xfrm_policy_id2dir(last->index), 0, data);
 916 out:
 917         read_unlock_bh(&xfrm_policy_lock);
 918         if (old != NULL)
 919                 xfrm_pol_put(old);
 920         return error;
 921 }
 922 EXPORT_SYMBOL(xfrm_policy_walk);
 923
 924 /*
 925  * Find policy to apply to this flow.
 926  *
 927  * Returns 0 if policy found, else an -errno.
 928  */
 929 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 930                              u8 type, u16 family, int dir)
 931 {
 932         struct xfrm_selector *sel = &pol->selector;
 933         int match, ret = -ESRCH;
 934
 935         if (pol->family != family ||
 936             pol->type != type)
 937                 return ret;
 938
 939         match = xfrm_selector_match(sel, fl, family);
 940         if (match)
 941                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 942                                                   dir);
 943
 944         return ret;
 945 }
 946
 947 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 948                                                      u16 family, u8 dir)
 949 {
 950         int err;
 951         struct xfrm_policy *pol, *ret;
 952         xfrm_address_t *daddr, *saddr;
 953         struct hlist_node *entry;
 954         struct hlist_head *chain;
 955         u32 priority = ~0U;
 956
 957         daddr = xfrm_flowi_daddr(fl, family);
 958         saddr = xfrm_flowi_saddr(fl, family);
 959         if (unlikely(!daddr || !saddr))
 960                 return NULL;
 961
 962         read_lock_bh(&xfrm_policy_lock);
 963         chain = policy_hash_direct(daddr, saddr, family, dir);
 964         ret = NULL;
 965         hlist_for_each_entry(pol, entry, chain, bydst) {
 966                 err = xfrm_policy_match(pol, fl, type, family, dir);
 967                 if (err) {
 968                         if (err == -ESRCH)
 969                                 continue;
 970                         else {
 971                                 ret = ERR_PTR(err);
 972                                 goto fail;
 973                         }
 974                 } else {
 975                         ret = pol;
 976                         priority = ret->priority;
 977                         break;
 978                 }
 979         }
 980         chain = &xfrm_policy_inexact[dir];
 981         hlist_for_each_entry(pol, entry, chain, bydst) {
 982                 err = xfrm_policy_match(pol, fl, type, family, dir);
 983                 if (err) {
 984                         if (err == -ESRCH)
 985                                 continue;
 986                         else {
 987                                 ret = ERR_PTR(err);
 988                                 goto fail;
 989                         }
 990                 } else if (pol->priority < priority) {
 991                         ret = pol;
 992                         break;
 993                 }
 994         }
 995         if (ret)
 996                 xfrm_pol_hold(ret);
 997 fail:
 998         read_unlock_bh(&xfrm_policy_lock);
 999
1000         return ret;
1001 }
1002
1003 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1004                                void **objp, atomic_t **obj_refp)
1005 {
1006         struct xfrm_policy *pol;
1007         int err = 0;
1008
1009 #ifdef CONFIG_XFRM_SUB_POLICY
1010         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1011         if (IS_ERR(pol)) {
1012                 err = PTR_ERR(pol);
1013                 pol = NULL;
1014         }
1015         if (pol || err)
1016                 goto end;
1017 #endif
1018         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1019         if (IS_ERR(pol)) {
1020                 err = PTR_ERR(pol);
1021                 pol = NULL;
1022         }
1023 #ifdef CONFIG_XFRM_SUB_POLICY
1024 end:
1025 #endif
1026         if ((*objp = (void *) pol) != NULL)
1027                 *obj_refp = &pol->refcnt;
1028         return err;
1029 }
1030
1031 static inline int policy_to_flow_dir(int dir)
1032 {
1033         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1034             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1035             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1036                 return dir;
1037         switch (dir) {
1038         default:
1039         case XFRM_POLICY_IN:
1040                 return FLOW_DIR_IN;
1041         case XFRM_POLICY_OUT:
1042                 return FLOW_DIR_OUT;
1043         case XFRM_POLICY_FWD:
1044                 return FLOW_DIR_FWD;
1045         }
1046 }
1047
1048 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1049 {
1050         struct xfrm_policy *pol;
1051
1052         read_lock_bh(&xfrm_policy_lock);
1053         if ((pol = sk->sk_policy[dir]) != NULL) {
1054                 int match = xfrm_selector_match(&pol->selector, fl,
1055                                                 sk->sk_family);
1056                 int err = 0;
1057
1058                 if (match) {
1059                         err = security_xfrm_policy_lookup(pol->security,
1060                                                       fl->secid,
1061                                                       policy_to_flow_dir(dir));
1062                         if (!err)
1063                                 xfrm_pol_hold(pol);
1064                         else if (err == -ESRCH)
1065                                 pol = NULL;
1066                         else
1067                                 pol = ERR_PTR(err);
1068                 } else
1069                         pol = NULL;
1070         }
1071         read_unlock_bh(&xfrm_policy_lock);
1072         return pol;
1073 }
1074
1075 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1076 {
1077         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1078                                                      pol->family, dir);
1079
1080         hlist_add_head(&pol->bydst, chain);
1081         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1082         xfrm_policy_count[dir]++;
1083         xfrm_pol_hold(pol);
1084
1085         if (xfrm_bydst_should_resize(dir, NULL))
1086                 schedule_work(&xfrm_hash_work);
1087 }
1088
1089 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1090                                                 int dir)
1091 {
1092         if (hlist_unhashed(&pol->bydst))
1093                 return NULL;
1094
1095         hlist_del(&pol->bydst);
1096         hlist_del(&pol->byidx);
1097         xfrm_policy_count[dir]--;
1098
1099         return pol;
1100 }
1101
1102 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1103 {
1104         write_lock_bh(&xfrm_policy_lock);
1105         pol = __xfrm_policy_unlink(pol, dir);
1106         write_unlock_bh(&xfrm_policy_lock);
1107         if (pol) {
1108                 if (dir < XFRM_POLICY_MAX)
1109                         atomic_inc(&flow_cache_genid);
1110                 xfrm_policy_kill(pol);
1111                 return 0;
1112         }
1113         return -ENOENT;
1114 }
1115 EXPORT_SYMBOL(xfrm_policy_delete);
1116
1117 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1118 {
1119         struct xfrm_policy *old_pol;
1120
1121 #ifdef CONFIG_XFRM_SUB_POLICY
1122         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1123                 return -EINVAL;
1124 #endif
1125
1126         write_lock_bh(&xfrm_policy_lock);
1127         old_pol = sk->sk_policy[dir];
1128         sk->sk_policy[dir] = pol;
1129         if (pol) {
1130                 pol->curlft.add_time = get_seconds();
1131                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1132                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1133         }
1134         if (old_pol)
1135                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1136         write_unlock_bh(&xfrm_policy_lock);
1137
1138         if (old_pol) {
1139                 xfrm_policy_kill(old_pol);
1140         }
1141         return 0;
1142 }
1143
1144 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1145 {
1146         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1147
1148         if (newp) {
1149                 newp->selector = old->selector;
1150                 if (security_xfrm_policy_clone(old->security,
1151                                                &newp->security)) {
1152                         kfree(newp);
1153                         return NULL;  /* ENOMEM */
1154                 }
1155                 newp->lft = old->lft;
1156                 newp->curlft = old->curlft;
1157                 newp->action = old->action;
1158                 newp->flags = old->flags;
1159                 newp->xfrm_nr = old->xfrm_nr;
1160                 newp->index = old->index;
1161                 newp->type = old->type;
1162                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1163                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1164                 write_lock_bh(&xfrm_policy_lock);
1165                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1166                 write_unlock_bh(&xfrm_policy_lock);
1167                 xfrm_pol_put(newp);
1168         }
1169         return newp;
1170 }
1171
1172 int __xfrm_sk_clone_policy(struct sock *sk)
1173 {
1174         struct xfrm_policy *p0 = sk->sk_policy[0],
1175                            *p1 = sk->sk_policy[1];
1176
1177         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1178         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1179                 return -ENOMEM;
1180         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1181                 return -ENOMEM;
1182         return 0;
1183 }
1184
1185 static int
1186 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1187                unsigned short family)
1188 {
1189         int err;
1190         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1191
1192         if (unlikely(afinfo == NULL))
1193                 return -EINVAL;
1194         err = afinfo->get_saddr(local, remote);
1195         xfrm_policy_put_afinfo(afinfo);
1196         return err;
1197 }
1198
1199 /* Resolve list of templates for the flow, given policy. */
1200
1201 static int
1202 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1203                       struct xfrm_state **xfrm,
1204                       unsigned short family)
1205 {
1206         int nx;
1207         int i, error;
1208         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1209         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1210         xfrm_address_t tmp;
1211
1212         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1213                 struct xfrm_state *x;
1214                 xfrm_address_t *remote = daddr;
1215                 xfrm_address_t *local  = saddr;
1216                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1217
1218                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1219                     tmpl->mode == XFRM_MODE_BEET) {
1220                         remote = &tmpl->id.daddr;
1221                         local = &tmpl->saddr;
1222                         family = tmpl->encap_family;
1223                         if (xfrm_addr_any(local, family)) {
1224                                 error = xfrm_get_saddr(&tmp, remote, family);
1225                                 if (error)
1226                                         goto fail;
1227                                 local = &tmp;
1228                         }
1229                 }
1230
1231                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1232
1233                 if (x && x->km.state == XFRM_STATE_VALID) {
1234                         xfrm[nx++] = x;
1235                         daddr = remote;
1236                         saddr = local;
1237                         continue;
1238                 }
1239                 if (x) {
1240                         error = (x->km.state == XFRM_STATE_ERROR ?
1241                                  -EINVAL : -EAGAIN);
1242                         xfrm_state_put(x);
1243                 }
1244
1245                 if (!tmpl->optional)
1246                         goto fail;
1247         }
1248         return nx;
1249
1250 fail:
1251         for (nx--; nx>=0; nx--)
1252                 xfrm_state_put(xfrm[nx]);
1253         return error;
1254 }
1255
1256 static int
1257 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1258                   struct xfrm_state **xfrm,
1259                   unsigned short family)
1260 {
1261         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1262         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1263         int cnx = 0;
1264         int error;
1265         int ret;
1266         int i;
1267
1268         for (i = 0; i < npols; i++) {
1269                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1270                         error = -ENOBUFS;
1271                         goto fail;
1272                 }
1273
1274                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1275                 if (ret < 0) {
1276                         error = ret;
1277                         goto fail;
1278                 } else
1279                         cnx += ret;
1280         }
1281
1282         /* found states are sorted for outbound processing */
1283         if (npols > 1)
1284                 xfrm_state_sort(xfrm, tpp, cnx, family);
1285
1286         return cnx;
1287
1288  fail:
1289         for (cnx--; cnx>=0; cnx--)
1290                 xfrm_state_put(tpp[cnx]);
1291         return error;
1292
1293 }
1294
1295 /* Check that the bundle accepts the flow and its components are
1296  * still valid.
1297  */
1298
1299 static struct dst_entry *
1300 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1301 {
1302         struct dst_entry *x;
1303         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1304         if (unlikely(afinfo == NULL))
1305                 return ERR_PTR(-EINVAL);
1306         x = afinfo->find_bundle(fl, policy);
1307         xfrm_policy_put_afinfo(afinfo);
1308         return x;
1309 }
1310
1311 static inline int xfrm_get_tos(struct flowi *fl, int family)
1312 {
1313         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1314         int tos;
1315
1316         if (!afinfo)
1317                 return -EINVAL;
1318
1319         tos = afinfo->get_tos(fl);
1320
1321         xfrm_policy_put_afinfo(afinfo);
1322
1323         return tos;
1324 }
1325
1326 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1327 {
1328         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1329         struct xfrm_dst *xdst;
1330
1331         if (!afinfo)
1332                 return ERR_PTR(-EINVAL);
1333
1334         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1335
1336         xfrm_policy_put_afinfo(afinfo);
1337
1338         return xdst;
1339 }
1340
1341 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1342                                  int nfheader_len)
1343 {
1344         struct xfrm_policy_afinfo *afinfo =
1345                 xfrm_policy_get_afinfo(dst->ops->family);
1346         int err;
1347
1348         if (!afinfo)
1349                 return -EINVAL;
1350
1351         err = afinfo->init_path(path, dst, nfheader_len);
1352
1353         xfrm_policy_put_afinfo(afinfo);
1354
1355         return err;
1356 }
1357
1358 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1359 {
1360         struct xfrm_policy_afinfo *afinfo =
1361                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1362         int err;
1363
1364         if (!afinfo)
1365                 return -EINVAL;
1366
1367         err = afinfo->fill_dst(xdst, dev);
1368
1369         xfrm_policy_put_afinfo(afinfo);
1370
1371         return err;
1372 }
1373
1374 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1375  * all the metrics... Shortly, bundle a bundle.
1376  */
1377
1378 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1379                                             struct xfrm_state **xfrm, int nx,
1380                                             struct flowi *fl,
1381                                             struct dst_entry *dst)
1382 {
1383         unsigned long now = jiffies;
1384         struct net_device *dev;
1385         struct dst_entry *dst_prev = NULL;
1386         struct dst_entry *dst0 = NULL;
1387         int i = 0;
1388         int err;
1389         int header_len = 0;
1390         int nfheader_len = 0;
1391         int trailer_len = 0;
1392         int tos;
1393         int family = policy->selector.family;
1394         xfrm_address_t saddr, daddr;
1395
1396         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1397
1398         tos = xfrm_get_tos(fl, family);
1399         err = tos;
1400         if (tos < 0)
1401                 goto put_states;
1402
1403         dst_hold(dst);
1404
1405         for (; i < nx; i++) {
1406                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1407                 struct dst_entry *dst1 = &xdst->u.dst;
1408
1409                 err = PTR_ERR(xdst);
1410                 if (IS_ERR(xdst)) {
1411                         dst_release(dst);
1412                         goto put_states;
1413                 }
1414
1415                 if (!dst_prev)
1416                         dst0 = dst1;
1417                 else {
1418                         dst_prev->child = dst_clone(dst1);
1419                         dst1->flags |= DST_NOHASH;
1420                 }
1421
1422                 xdst->route = dst;
1423                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1424
1425                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1426                         family = xfrm[i]->props.family;
1427                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1428                                               family);
1429                         err = PTR_ERR(dst);
1430                         if (IS_ERR(dst))
1431                                 goto put_states;
1432                 } else
1433                         dst_hold(dst);
1434
1435                 dst1->xfrm = xfrm[i];
1436                 xdst->genid = xfrm[i]->genid;
1437
1438                 dst1->obsolete = -1;
1439                 dst1->flags |= DST_HOST;
1440                 dst1->lastuse = now;
1441
1442                 dst1->input = dst_discard;
1443                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1444
1445                 dst1->next = dst_prev;
1446                 dst_prev = dst1;
1447
1448                 header_len += xfrm[i]->props.header_len;
1449                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1450                         nfheader_len += xfrm[i]->props.header_len;
1451                 trailer_len += xfrm[i]->props.trailer_len;
1452         }
1453
1454         dst_prev->child = dst;
1455         dst0->path = dst;
1456
1457         err = -ENODEV;
1458         dev = dst->dev;
1459         if (!dev)
1460                 goto free_dst;
1461
1462         /* Copy neighbout for reachability confirmation */
1463         dst0->neighbour = neigh_clone(dst->neighbour);
1464
1465         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1466         xfrm_init_pmtu(dst_prev);
1467
1468         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1469                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1470
1471                 err = xfrm_fill_dst(xdst, dev);
1472                 if (err)
1473                         goto free_dst;
1474
1475                 dst_prev->header_len = header_len;
1476                 dst_prev->trailer_len = trailer_len;
1477                 header_len -= xdst->u.dst.xfrm->props.header_len;
1478                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1479         }
1480
1481 out:
1482         return dst0;
1483
1484 put_states:
1485         for (; i < nx; i++)
1486                 xfrm_state_put(xfrm[i]);
1487 free_dst:
1488         if (dst0)
1489                 dst_free(dst0);
1490         dst0 = ERR_PTR(err);
1491         goto out;
1492 }
1493
1494 static int inline
1495 xfrm_dst_alloc_copy(void **target, void *src, int size)
1496 {
1497         if (!*target) {
1498                 *target = kmalloc(size, GFP_ATOMIC);
1499                 if (!*target)
1500                         return -ENOMEM;
1501         }
1502         memcpy(*target, src, size);
1503         return 0;
1504 }
1505
1506 static int inline
1507 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1508 {
1509 #ifdef CONFIG_XFRM_SUB_POLICY
1510         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1511         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1512                                    sel, sizeof(*sel));
1513 #else
1514         return 0;
1515 #endif
1516 }
1517
1518 static int inline
1519 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1520 {
1521 #ifdef CONFIG_XFRM_SUB_POLICY
1522         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1523         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1524 #else
1525         return 0;
1526 #endif
1527 }
1528
1529 static int stale_bundle(struct dst_entry *dst);
1530
1531 /* Main function: finds/creates a bundle for given flow.
1532  *
1533  * At the moment we eat a raw IP route. Mostly to speed up lookups
1534  * on interfaces with disabled IPsec.
1535  */
1536 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1537                   struct sock *sk, int flags)
1538 {
1539         struct xfrm_policy *policy;
1540         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1541         int npols;
1542         int pol_dead;
1543         int xfrm_nr;
1544         int pi;
1545         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1546         struct dst_entry *dst, *dst_orig = *dst_p;
1547         int nx = 0;
1548         int err;
1549         u32 genid;
1550         u16 family;
1551         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1552
1553 restart:
1554         genid = atomic_read(&flow_cache_genid);
1555         policy = NULL;
1556         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1557                 pols[pi] = NULL;
1558         npols = 0;
1559         pol_dead = 0;
1560         xfrm_nr = 0;
1561
1562         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1563                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1564                 err = PTR_ERR(policy);
1565                 if (IS_ERR(policy)) {
1566                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1567                         goto dropdst;
1568                 }
1569         }
1570
1571         if (!policy) {
1572                 /* To accelerate a bit...  */
1573                 if ((dst_orig->flags & DST_NOXFRM) ||
1574                     !xfrm_policy_count[XFRM_POLICY_OUT])
1575                         goto nopol;
1576
1577                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1578                                            dir, xfrm_policy_lookup);
1579                 err = PTR_ERR(policy);
1580                 if (IS_ERR(policy)) {
1581                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1582                         goto dropdst;
1583                 }
1584         }
1585
1586         if (!policy)
1587                 goto nopol;
1588
1589         family = dst_orig->ops->family;
1590         pols[0] = policy;
1591         npols ++;
1592         xfrm_nr += pols[0]->xfrm_nr;
1593
1594         err = -ENOENT;
1595         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1596                 goto error;
1597
1598         policy->curlft.use_time = get_seconds();
1599
1600         switch (policy->action) {
1601         default:
1602         case XFRM_POLICY_BLOCK:
1603                 /* Prohibit the flow */
1604                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1605                 err = -EPERM;
1606                 goto error;
1607
1608         case XFRM_POLICY_ALLOW:
1609 #ifndef CONFIG_XFRM_SUB_POLICY
1610                 if (policy->xfrm_nr == 0) {
1611                         /* Flow passes not transformed. */
1612                         xfrm_pol_put(policy);
1613                         return 0;
1614                 }
1615 #endif
1616
1617                 /* Try to find matching bundle.
1618                  *
1619                  * LATER: help from flow cache. It is optional, this
1620                  * is required only for output policy.
1621                  */
1622                 dst = xfrm_find_bundle(fl, policy, family);
1623                 if (IS_ERR(dst)) {
1624                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1625                         err = PTR_ERR(dst);
1626                         goto error;
1627                 }
1628
1629                 if (dst)
1630                         break;
1631
1632 #ifdef CONFIG_XFRM_SUB_POLICY
1633                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1634                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1635                                                             fl, family,
1636                                                             XFRM_POLICY_OUT);
1637                         if (pols[1]) {
1638                                 if (IS_ERR(pols[1])) {
1639                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1640                                         err = PTR_ERR(pols[1]);
1641                                         goto error;
1642                                 }
1643                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1644                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1645                                         err = -EPERM;
1646                                         goto error;
1647                                 }
1648                                 npols ++;
1649                                 xfrm_nr += pols[1]->xfrm_nr;
1650                         }
1651                 }
1652
1653                 /*
1654                  * Because neither flowi nor bundle information knows about
1655                  * transformation template size. On more than one policy usage
1656                  * we can realize whether all of them is bypass or not after
1657                  * they are searched. See above not-transformed bypass
1658                  * is surrounded by non-sub policy configuration, too.
1659                  */
1660                 if (xfrm_nr == 0) {
1661                         /* Flow passes not transformed. */
1662                         xfrm_pols_put(pols, npols);
1663                         return 0;
1664                 }
1665
1666 #endif
1667                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1668
1669                 if (unlikely(nx<0)) {
1670                         err = nx;
1671                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1672                                 /* EREMOTE tells the caller to generate
1673                                  * a one-shot blackhole route.
1674                                  */
1675                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1676                                 xfrm_pol_put(policy);
1677                                 return -EREMOTE;
1678                         }
1679                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1680                                 DECLARE_WAITQUEUE(wait, current);
1681
1682                                 add_wait_queue(&km_waitq, &wait);
1683                                 set_current_state(TASK_INTERRUPTIBLE);
1684                                 schedule();
1685                                 set_current_state(TASK_RUNNING);
1686                                 remove_wait_queue(&km_waitq, &wait);
1687
1688                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1689
1690                                 if (nx == -EAGAIN && signal_pending(current)) {
1691                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1692                                         err = -ERESTART;
1693                                         goto error;
1694                                 }
1695                                 if (nx == -EAGAIN ||
1696                                     genid != atomic_read(&flow_cache_genid)) {
1697                                         xfrm_pols_put(pols, npols);
1698                                         goto restart;
1699                                 }
1700                                 err = nx;
1701                         }
1702                         if (err < 0) {
1703                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1704                                 goto error;
1705                         }
1706                 }
1707                 if (nx == 0) {
1708                         /* Flow passes not transformed. */
1709                         xfrm_pols_put(pols, npols);
1710                         return 0;
1711                 }
1712
1713                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1714                 err = PTR_ERR(dst);
1715                 if (IS_ERR(dst)) {
1716                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1717                         goto error;
1718                 }
1719
1720                 for (pi = 0; pi < npols; pi++) {
1721                         read_lock_bh(&pols[pi]->lock);
1722                         pol_dead |= pols[pi]->dead;
1723                         read_unlock_bh(&pols[pi]->lock);
1724                 }
1725
1726                 write_lock_bh(&policy->lock);
1727                 if (unlikely(pol_dead || stale_bundle(dst))) {
1728                         /* Wow! While we worked on resolving, this
1729                          * policy has gone. Retry. It is not paranoia,
1730                          * we just cannot enlist new bundle to dead object.
1731                          * We can't enlist stable bundles either.
1732                          */
1733                         write_unlock_bh(&policy->lock);
1734                         if (dst)
1735                                 dst_free(dst);
1736
1737                         if (pol_dead)
1738                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1739                         else
1740                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1741                         err = -EHOSTUNREACH;
1742                         goto error;
1743                 }
1744
1745                 if (npols > 1)
1746                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1747                 else
1748                         err = xfrm_dst_update_origin(dst, fl);
1749                 if (unlikely(err)) {
1750                         write_unlock_bh(&policy->lock);
1751                         if (dst)
1752                                 dst_free(dst);
1753                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1754                         goto error;
1755                 }
1756
1757                 dst->next = policy->bundles;
1758                 policy->bundles = dst;
1759                 dst_hold(dst);
1760                 write_unlock_bh(&policy->lock);
1761         }
1762         *dst_p = dst;
1763         dst_release(dst_orig);
1764         xfrm_pols_put(pols, npols);
1765         return 0;
1766
1767 error:
1768         xfrm_pols_put(pols, npols);
1769 dropdst:
1770         dst_release(dst_orig);
1771         *dst_p = NULL;
1772         return err;
1773
1774 nopol:
1775         err = -ENOENT;
1776         if (flags & XFRM_LOOKUP_ICMP)
1777                 goto dropdst;
1778         return 0;
1779 }
1780 EXPORT_SYMBOL(__xfrm_lookup);
1781
1782 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1783                 struct sock *sk, int flags)
1784 {
1785         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1786
1787         if (err == -EREMOTE) {
1788                 dst_release(*dst_p);
1789                 *dst_p = NULL;
1790                 err = -EAGAIN;
1791         }
1792
1793         return err;
1794 }
1795 EXPORT_SYMBOL(xfrm_lookup);
1796
1797 static inline int
1798 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1799 {
1800         struct xfrm_state *x;
1801
1802         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1803                 return 0;
1804         x = skb->sp->xvec[idx];
1805         if (!x->type->reject)
1806                 return 0;
1807         return x->type->reject(x, skb, fl);
1808 }
1809
1810 /* When skb is transformed back to its "native" form, we have to
1811  * check policy restrictions. At the moment we make this in maximally
1812  * stupid way. Shame on me. :-) Of course, connected sockets must
1813  * have policy cached at them.
1814  */
1815
1816 static inline int
1817 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1818               unsigned short family)
1819 {
1820         if (xfrm_state_kern(x))
1821                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1822         return  x->id.proto == tmpl->id.proto &&
1823                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1824                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1825                 x->props.mode == tmpl->mode &&
1826                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1827                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1828                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1829                   xfrm_state_addr_cmp(tmpl, x, family));
1830 }
1831
1832 /*
1833  * 0 or more than 0 is returned when validation is succeeded (either bypass
1834  * because of optional transport mode, or next index of the mathced secpath
1835  * state with the template.
1836  * -1 is returned when no matching template is found.
1837  * Otherwise "-2 - errored_index" is returned.
1838  */
1839 static inline int
1840 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1841                unsigned short family)
1842 {
1843         int idx = start;
1844
1845         if (tmpl->optional) {
1846                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1847                         return start;
1848         } else
1849                 start = -1;
1850         for (; idx < sp->len; idx++) {
1851                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1852                         return ++idx;
1853                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1854                         if (start == -1)
1855                                 start = -2-idx;
1856                         break;
1857                 }
1858         }
1859         return start;
1860 }
1861
1862 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1863                           unsigned int family, int reverse)
1864 {
1865         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1866         int err;
1867
1868         if (unlikely(afinfo == NULL))
1869                 return -EAFNOSUPPORT;
1870
1871         afinfo->decode_session(skb, fl, reverse);
1872         err = security_xfrm_decode_session(skb, &fl->secid);
1873         xfrm_policy_put_afinfo(afinfo);
1874         return err;
1875 }
1876 EXPORT_SYMBOL(__xfrm_decode_session);
1877
1878 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1879 {
1880         for (; k < sp->len; k++) {
1881                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1882                         *idxp = k;
1883                         return 1;
1884                 }
1885         }
1886
1887         return 0;
1888 }
1889
1890 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1891                         unsigned short family)
1892 {
1893         struct xfrm_policy *pol;
1894         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1895         int npols = 0;
1896         int xfrm_nr;
1897         int pi;
1898         int reverse;
1899         struct flowi fl;
1900         u8 fl_dir;
1901         int xerr_idx = -1;
1902
1903         reverse = dir & ~XFRM_POLICY_MASK;
1904         dir &= XFRM_POLICY_MASK;
1905         fl_dir = policy_to_flow_dir(dir);
1906
1907         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1908                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1909                 return 0;
1910         }
1911
1912         nf_nat_decode_session(skb, &fl, family);
1913
1914         /* First, check used SA against their selectors. */
1915         if (skb->sp) {
1916                 int i;
1917
1918                 for (i=skb->sp->len-1; i>=0; i--) {
1919                         struct xfrm_state *x = skb->sp->xvec[i];
1920                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1921                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1922                                 return 0;
1923                         }
1924                 }
1925         }
1926
1927         pol = NULL;
1928         if (sk && sk->sk_policy[dir]) {
1929                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1930                 if (IS_ERR(pol)) {
1931                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1932                         return 0;
1933                 }
1934         }
1935
1936         if (!pol)
1937                 pol = flow_cache_lookup(&fl, family, fl_dir,
1938                                         xfrm_policy_lookup);
1939
1940         if (IS_ERR(pol)) {
1941                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1942                 return 0;
1943         }
1944
1945         if (!pol) {
1946                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1947                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1948                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1949                         return 0;
1950                 }
1951                 return 1;
1952         }
1953
1954         pol->curlft.use_time = get_seconds();
1955
1956         pols[0] = pol;
1957         npols ++;
1958 #ifdef CONFIG_XFRM_SUB_POLICY
1959         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1960                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1961                                                     &fl, family,
1962                                                     XFRM_POLICY_IN);
1963                 if (pols[1]) {
1964                         if (IS_ERR(pols[1])) {
1965                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1966                                 return 0;
1967                         }
1968                         pols[1]->curlft.use_time = get_seconds();
1969                         npols ++;
1970                 }
1971         }
1972 #endif
1973
1974         if (pol->action == XFRM_POLICY_ALLOW) {
1975                 struct sec_path *sp;
1976                 static struct sec_path dummy;
1977                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1978                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1979                 struct xfrm_tmpl **tpp = tp;
1980                 int ti = 0;
1981                 int i, k;
1982
1983                 if ((sp = skb->sp) == NULL)
1984                         sp = &dummy;
1985
1986                 for (pi = 0; pi < npols; pi++) {
1987                         if (pols[pi] != pol &&
1988                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1989                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1990                                 goto reject;
1991                         }
1992                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1993                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1994                                 goto reject_error;
1995                         }
1996                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1997                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1998                 }
1999                 xfrm_nr = ti;
2000                 if (npols > 1) {
2001                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2002                         tpp = stp;
2003                 }
2004
2005                 /* For each tunnel xfrm, find the first matching tmpl.
2006                  * For each tmpl before that, find corresponding xfrm.
2007                  * Order is _important_. Later we will implement
2008                  * some barriers, but at the moment barriers
2009                  * are implied between each two transformations.
2010                  */
2011                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2012                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2013                         if (k < 0) {
2014                                 if (k < -1)
2015                                         /* "-2 - errored_index" returned */
2016                                         xerr_idx = -(2+k);
2017                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2018                                 goto reject;
2019                         }
2020                 }
2021
2022                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2023                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2024                         goto reject;
2025                 }
2026
2027                 xfrm_pols_put(pols, npols);
2028                 return 1;
2029         }
2030         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2031
2032 reject:
2033         xfrm_secpath_reject(xerr_idx, skb, &fl);
2034 reject_error:
2035         xfrm_pols_put(pols, npols);
2036         return 0;
2037 }
2038 EXPORT_SYMBOL(__xfrm_policy_check);
2039
2040 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2041 {
2042         struct flowi fl;
2043
2044         if (xfrm_decode_session(skb, &fl, family) < 0) {
2045                 /* XXX: we should have something like FWDHDRERROR here. */
2046                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2047                 return 0;
2048         }
2049
2050         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2051 }
2052 EXPORT_SYMBOL(__xfrm_route_forward);
2053
2054 /* Optimize later using cookies and generation ids. */
2055
2056 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2057 {
2058         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2059          * to "-1" to force all XFRM destinations to get validated by
2060          * dst_ops->check on every use.  We do this because when a
2061          * normal route referenced by an XFRM dst is obsoleted we do
2062          * not go looking around for all parent referencing XFRM dsts
2063          * so that we can invalidate them.  It is just too much work.
2064          * Instead we make the checks here on every use.  For example:
2065          *
2066          *      XFRM dst A --> IPv4 dst X
2067          *
2068          * X is the "xdst->route" of A (X is also the "dst->path" of A
2069          * in this example).  If X is marked obsolete, "A" will not
2070          * notice.  That's what we are validating here via the
2071          * stale_bundle() check.
2072          *
2073          * When a policy's bundle is pruned, we dst_free() the XFRM
2074          * dst which causes it's ->obsolete field to be set to a
2075          * positive non-zero integer.  If an XFRM dst has been pruned
2076          * like this, we want to force a new route lookup.
2077          */
2078         if (dst->obsolete < 0 && !stale_bundle(dst))
2079                 return dst;
2080
2081         return NULL;
2082 }
2083
2084 static int stale_bundle(struct dst_entry *dst)
2085 {
2086         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2087 }
2088
2089 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2090 {
2091         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2092                 dst->dev = dev_net(dev)->loopback_dev;
2093                 dev_hold(dst->dev);
2094                 dev_put(dev);
2095         }
2096 }
2097 EXPORT_SYMBOL(xfrm_dst_ifdown);
2098
2099 static void xfrm_link_failure(struct sk_buff *skb)
2100 {
2101         /* Impossible. Such dst must be popped before reaches point of failure. */
2102         return;
2103 }
2104
2105 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2106 {
2107         if (dst) {
2108                 if (dst->obsolete) {
2109                         dst_release(dst);
2110                         dst = NULL;
2111                 }
2112         }
2113         return dst;
2114 }
2115
2116 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2117 {
2118         struct dst_entry *dst, **dstp;
2119
2120         write_lock(&pol->lock);
2121         dstp = &pol->bundles;
2122         while ((dst=*dstp) != NULL) {
2123                 if (func(dst)) {
2124                         *dstp = dst->next;
2125                         dst->next = *gc_list_p;
2126                         *gc_list_p = dst;
2127                 } else {
2128                         dstp = &dst->next;
2129                 }
2130         }
2131         write_unlock(&pol->lock);
2132 }
2133
2134 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2135 {
2136         struct dst_entry *gc_list = NULL;
2137         int dir;
2138
2139         read_lock_bh(&xfrm_policy_lock);
2140         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2141                 struct xfrm_policy *pol;
2142                 struct hlist_node *entry;
2143                 struct hlist_head *table;
2144                 int i;
2145
2146                 hlist_for_each_entry(pol, entry,
2147                                      &xfrm_policy_inexact[dir], bydst)
2148                         prune_one_bundle(pol, func, &gc_list);
2149
2150                 table = xfrm_policy_bydst[dir].table;
2151                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2152                         hlist_for_each_entry(pol, entry, table + i, bydst)
2153                                 prune_one_bundle(pol, func, &gc_list);
2154                 }
2155         }
2156         read_unlock_bh(&xfrm_policy_lock);
2157
2158         while (gc_list) {
2159                 struct dst_entry *dst = gc_list;
2160                 gc_list = dst->next;
2161                 dst_free(dst);
2162         }
2163 }
2164
2165 static int unused_bundle(struct dst_entry *dst)
2166 {
2167         return !atomic_read(&dst->__refcnt);
2168 }
2169
2170 static void __xfrm_garbage_collect(void)
2171 {
2172         xfrm_prune_bundles(unused_bundle);
2173 }
2174
2175 static int xfrm_flush_bundles(void)
2176 {
2177         xfrm_prune_bundles(stale_bundle);
2178         return 0;
2179 }
2180
2181 static void xfrm_init_pmtu(struct dst_entry *dst)
2182 {
2183         do {
2184                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2185                 u32 pmtu, route_mtu_cached;
2186
2187                 pmtu = dst_mtu(dst->child);
2188                 xdst->child_mtu_cached = pmtu;
2189
2190                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2191
2192                 route_mtu_cached = dst_mtu(xdst->route);
2193                 xdst->route_mtu_cached = route_mtu_cached;
2194
2195                 if (pmtu > route_mtu_cached)
2196                         pmtu = route_mtu_cached;
2197
2198                 dst->metrics[RTAX_MTU-1] = pmtu;
2199         } while ((dst = dst->next));
2200 }
2201
2202 /* Check that the bundle accepts the flow and its components are
2203  * still valid.
2204  */
2205
2206 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2207                 struct flowi *fl, int family, int strict)
2208 {
2209         struct dst_entry *dst = &first->u.dst;
2210         struct xfrm_dst *last;
2211         u32 mtu;
2212
2213         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2214             (dst->dev && !netif_running(dst->dev)))
2215                 return 0;
2216 #ifdef CONFIG_XFRM_SUB_POLICY
2217         if (fl) {
2218                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2219                         return 0;
2220                 if (first->partner &&
2221                     !xfrm_selector_match(first->partner, fl, family))
2222                         return 0;
2223         }
2224 #endif
2225
2226         last = NULL;
2227
2228         do {
2229                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2230
2231                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2232                         return 0;
2233                 if (fl && pol &&
2234                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2235                         return 0;
2236                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2237                         return 0;
2238                 if (xdst->genid != dst->xfrm->genid)
2239                         return 0;
2240
2241                 if (strict && fl &&
2242                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2243                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2244                         return 0;
2245
2246                 mtu = dst_mtu(dst->child);
2247                 if (xdst->child_mtu_cached != mtu) {
2248                         last = xdst;
2249                         xdst->child_mtu_cached = mtu;
2250                 }
2251
2252                 if (!dst_check(xdst->route, xdst->route_cookie))
2253                         return 0;
2254                 mtu = dst_mtu(xdst->route);
2255                 if (xdst->route_mtu_cached != mtu) {
2256                         last = xdst;
2257                         xdst->route_mtu_cached = mtu;
2258                 }
2259
2260                 dst = dst->child;
2261         } while (dst->xfrm);
2262
2263         if (likely(!last))
2264                 return 1;
2265
2266         mtu = last->child_mtu_cached;
2267         for (;;) {
2268                 dst = &last->u.dst;
2269
2270                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2271                 if (mtu > last->route_mtu_cached)
2272                         mtu = last->route_mtu_cached;
2273                 dst->metrics[RTAX_MTU-1] = mtu;
2274
2275                 if (last == first)
2276                         break;
2277
2278                 last = (struct xfrm_dst *)last->u.dst.next;
2279                 last->child_mtu_cached = mtu;
2280         }
2281
2282         return 1;
2283 }
2284
2285 EXPORT_SYMBOL(xfrm_bundle_ok);
2286
2287 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2288 {
2289         int err = 0;
2290         if (unlikely(afinfo == NULL))
2291                 return -EINVAL;
2292         if (unlikely(afinfo->family >= NPROTO))
2293                 return -EAFNOSUPPORT;
2294         write_lock_bh(&xfrm_policy_afinfo_lock);
2295         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2296                 err = -ENOBUFS;
2297         else {
2298                 struct dst_ops *dst_ops = afinfo->dst_ops;
2299                 if (likely(dst_ops->kmem_cachep == NULL))
2300                         dst_ops->kmem_cachep = xfrm_dst_cache;
2301                 if (likely(dst_ops->check == NULL))
2302                         dst_ops->check = xfrm_dst_check;
2303                 if (likely(dst_ops->negative_advice == NULL))
2304                         dst_ops->negative_advice = xfrm_negative_advice;
2305                 if (likely(dst_ops->link_failure == NULL))
2306                         dst_ops->link_failure = xfrm_link_failure;
2307                 if (likely(afinfo->garbage_collect == NULL))
2308                         afinfo->garbage_collect = __xfrm_garbage_collect;
2309                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2310         }
2311         write_unlock_bh(&xfrm_policy_afinfo_lock);
2312         return err;
2313 }
2314 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2315
2316 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2317 {
2318         int err = 0;
2319         if (unlikely(afinfo == NULL))
2320                 return -EINVAL;
2321         if (unlikely(afinfo->family >= NPROTO))
2322                 return -EAFNOSUPPORT;
2323         write_lock_bh(&xfrm_policy_afinfo_lock);
2324         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2325                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2326                         err = -EINVAL;
2327                 else {
2328                         struct dst_ops *dst_ops = afinfo->dst_ops;
2329                         xfrm_policy_afinfo[afinfo->family] = NULL;
2330                         dst_ops->kmem_cachep = NULL;
2331                         dst_ops->check = NULL;
2332                         dst_ops->negative_advice = NULL;
2333                         dst_ops->link_failure = NULL;
2334                         afinfo->garbage_collect = NULL;
2335                 }
2336         }
2337         write_unlock_bh(&xfrm_policy_afinfo_lock);
2338         return err;
2339 }
2340 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2341
2342 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2343 {
2344         struct xfrm_policy_afinfo *afinfo;
2345         if (unlikely(family >= NPROTO))
2346                 return NULL;
2347         read_lock(&xfrm_policy_afinfo_lock);
2348         afinfo = xfrm_policy_afinfo[family];
2349         if (unlikely(!afinfo))
2350                 read_unlock(&xfrm_policy_afinfo_lock);
2351         return afinfo;
2352 }
2353
2354 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2355 {
2356         read_unlock(&xfrm_policy_afinfo_lock);
2357 }
2358
2359 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2360 {
2361         struct net_device *dev = ptr;
2362
2363         if (dev_net(dev) != &init_net)
2364                 return NOTIFY_DONE;
2365
2366         switch (event) {
2367         case NETDEV_DOWN:
2368                 xfrm_flush_bundles();
2369         }
2370         return NOTIFY_DONE;
2371 }
2372
2373 static struct notifier_block xfrm_dev_notifier = {
2374         xfrm_dev_event,
2375         NULL,
2376         0
2377 };
2378
2379 #ifdef CONFIG_XFRM_STATISTICS
2380 static int __init xfrm_statistics_init(void)
2381 {
2382         if (snmp_mib_init((void **)xfrm_statistics,
2383                           sizeof(struct linux_xfrm_mib)) < 0)
2384                 return -ENOMEM;
2385         return 0;
2386 }
2387 #endif
2388
2389 static void __init xfrm_policy_init(void)
2390 {
2391         unsigned int hmask, sz;
2392         int dir;
2393
2394         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2395                                            sizeof(struct xfrm_dst),
2396                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2397                                            NULL);
2398
2399         hmask = 8 - 1;
2400         sz = (hmask+1) * sizeof(struct hlist_head);
2401
2402         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2403         xfrm_idx_hmask = hmask;
2404         if (!xfrm_policy_byidx)
2405                 panic("XFRM: failed to allocate byidx hash\n");
2406
2407         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2408                 struct xfrm_policy_hash *htab;
2409
2410                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2411
2412                 htab = &xfrm_policy_bydst[dir];
2413                 htab->table = xfrm_hash_alloc(sz);
2414                 htab->hmask = hmask;
2415                 if (!htab->table)
2416                         panic("XFRM: failed to allocate bydst hash\n");
2417         }
2418
2419         for (dir = 0; dir < XFRM_POLICY_TYPE_MAX; dir++)
2420                 INIT_LIST_HEAD(&xfrm_policy_bytype[dir]);
2421
2422         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2423         register_netdevice_notifier(&xfrm_dev_notifier);
2424 }
2425
2426 void __init xfrm_init(void)
2427 {
2428 #ifdef CONFIG_XFRM_STATISTICS
2429         xfrm_statistics_init();
2430 #endif
2431         xfrm_state_init();
2432         xfrm_policy_init();
2433         xfrm_input_init();
2434 #ifdef CONFIG_XFRM_STATISTICS
2435         xfrm_proc_init();
2436 #endif
2437 }
2438
2439 #ifdef CONFIG_AUDITSYSCALL
2440 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2441                                          struct audit_buffer *audit_buf)
2442 {
2443         struct xfrm_sec_ctx *ctx = xp->security;
2444         struct xfrm_selector *sel = &xp->selector;
2445
2446         if (ctx)
2447                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2448                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2449
2450         switch(sel->family) {
2451         case AF_INET:
2452                 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2453                                  NIPQUAD(sel->saddr.a4));
2454                 if (sel->prefixlen_s != 32)
2455                         audit_log_format(audit_buf, " src_prefixlen=%d",
2456                                          sel->prefixlen_s);
2457                 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2458                                  NIPQUAD(sel->daddr.a4));
2459                 if (sel->prefixlen_d != 32)
2460                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2461                                          sel->prefixlen_d);
2462                 break;
2463         case AF_INET6:
2464                 audit_log_format(audit_buf, " src=" NIP6_FMT,
2465                                  NIP6(*(struct in6_addr *)sel->saddr.a6));
2466                 if (sel->prefixlen_s != 128)
2467                         audit_log_format(audit_buf, " src_prefixlen=%d",
2468                                          sel->prefixlen_s);
2469                 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2470                                  NIP6(*(struct in6_addr *)sel->daddr.a6));
2471                 if (sel->prefixlen_d != 128)
2472                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2473                                          sel->prefixlen_d);
2474                 break;
2475         }
2476 }
2477
2478 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2479                            uid_t auid, u32 sessionid, u32 secid)
2480 {
2481         struct audit_buffer *audit_buf;
2482
2483         audit_buf = xfrm_audit_start("SPD-add");
2484         if (audit_buf == NULL)
2485                 return;
2486         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2487         audit_log_format(audit_buf, " res=%u", result);
2488         xfrm_audit_common_policyinfo(xp, audit_buf);
2489         audit_log_end(audit_buf);
2490 }
2491 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2492
2493 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2494                               uid_t auid, u32 sessionid, u32 secid)
2495 {
2496         struct audit_buffer *audit_buf;
2497
2498         audit_buf = xfrm_audit_start("SPD-delete");
2499         if (audit_buf == NULL)
2500                 return;
2501         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2502         audit_log_format(audit_buf, " res=%u", result);
2503         xfrm_audit_common_policyinfo(xp, audit_buf);
2504         audit_log_end(audit_buf);
2505 }
2506 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2507 #endif
2508
2509 #ifdef CONFIG_XFRM_MIGRATE
2510 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2511                                        struct xfrm_selector *sel_tgt)
2512 {
2513         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2514                 if (sel_tgt->family == sel_cmp->family &&
2515                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2516                                   sel_cmp->family) == 0 &&
2517                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2518                                   sel_cmp->family) == 0 &&
2519                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2520                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2521                         return 1;
2522                 }
2523         } else {
2524                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2525                         return 1;
2526                 }
2527         }
2528         return 0;
2529 }
2530
2531 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2532                                                      u8 dir, u8 type)
2533 {
2534         struct xfrm_policy *pol, *ret = NULL;
2535         struct hlist_node *entry;
2536         struct hlist_head *chain;
2537         u32 priority = ~0U;
2538
2539         read_lock_bh(&xfrm_policy_lock);
2540         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2541         hlist_for_each_entry(pol, entry, chain, bydst) {
2542                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2543                     pol->type == type) {
2544                         ret = pol;
2545                         priority = ret->priority;
2546                         break;
2547                 }
2548         }
2549         chain = &xfrm_policy_inexact[dir];
2550         hlist_for_each_entry(pol, entry, chain, bydst) {
2551                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2552                     pol->type == type &&
2553                     pol->priority < priority) {
2554                         ret = pol;
2555                         break;
2556                 }
2557         }
2558
2559         if (ret)
2560                 xfrm_pol_hold(ret);
2561
2562         read_unlock_bh(&xfrm_policy_lock);
2563
2564         return ret;
2565 }
2566
2567 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2568 {
2569         int match = 0;
2570
2571         if (t->mode == m->mode && t->id.proto == m->proto &&
2572             (m->reqid == 0 || t->reqid == m->reqid)) {
2573                 switch (t->mode) {
2574                 case XFRM_MODE_TUNNEL:
2575                 case XFRM_MODE_BEET:
2576                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2577                                           m->old_family) == 0 &&
2578                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2579                                           m->old_family) == 0) {
2580                                 match = 1;
2581                         }
2582                         break;
2583                 case XFRM_MODE_TRANSPORT:
2584                         /* in case of transport mode, template does not store
2585                            any IP addresses, hence we just compare mode and
2586                            protocol */
2587                         match = 1;
2588                         break;
2589                 default:
2590                         break;
2591                 }
2592         }
2593         return match;
2594 }
2595
2596 /* update endpoint address(es) of template(s) */
2597 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2598                                struct xfrm_migrate *m, int num_migrate)
2599 {
2600         struct xfrm_migrate *mp;
2601         struct dst_entry *dst;
2602         int i, j, n = 0;
2603
2604         write_lock_bh(&pol->lock);
2605         if (unlikely(pol->dead)) {
2606                 /* target policy has been deleted */
2607                 write_unlock_bh(&pol->lock);
2608                 return -ENOENT;
2609         }
2610
2611         for (i = 0; i < pol->xfrm_nr; i++) {
2612                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2613                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2614                                 continue;
2615                         n++;
2616                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2617                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2618                                 continue;
2619                         /* update endpoints */
2620                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2621                                sizeof(pol->xfrm_vec[i].id.daddr));
2622                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2623                                sizeof(pol->xfrm_vec[i].saddr));
2624                         pol->xfrm_vec[i].encap_family = mp->new_family;
2625                         /* flush bundles */
2626                         while ((dst = pol->bundles) != NULL) {
2627                                 pol->bundles = dst->next;
2628                                 dst_free(dst);
2629                         }
2630                 }
2631         }
2632
2633         write_unlock_bh(&pol->lock);
2634
2635         if (!n)
2636                 return -ENODATA;
2637
2638         return 0;
2639 }
2640
2641 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2642 {
2643         int i, j;
2644
2645         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2646                 return -EINVAL;
2647
2648         for (i = 0; i < num_migrate; i++) {
2649                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2650                                    m[i].old_family) == 0) &&
2651                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2652                                    m[i].old_family) == 0))
2653                         return -EINVAL;
2654                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2655                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2656                         return -EINVAL;
2657
2658                 /* check if there is any duplicated entry */
2659                 for (j = i + 1; j < num_migrate; j++) {
2660                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2661                                     sizeof(m[i].old_daddr)) &&
2662                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2663                                     sizeof(m[i].old_saddr)) &&
2664                             m[i].proto == m[j].proto &&
2665                             m[i].mode == m[j].mode &&
2666                             m[i].reqid == m[j].reqid &&
2667                             m[i].old_family == m[j].old_family)
2668                                 return -EINVAL;
2669                 }
2670         }
2671
2672         return 0;
2673 }
2674
2675 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2676                  struct xfrm_migrate *m, int num_migrate)
2677 {
2678         int i, err, nx_cur = 0, nx_new = 0;
2679         struct xfrm_policy *pol = NULL;
2680         struct xfrm_state *x, *xc;
2681         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2682         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2683         struct xfrm_migrate *mp;
2684
2685         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2686                 goto out;
2687
2688         /* Stage 1 - find policy */
2689         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2690                 err = -ENOENT;
2691                 goto out;
2692         }
2693
2694         /* Stage 2 - find and update state(s) */
2695         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2696                 if ((x = xfrm_migrate_state_find(mp))) {
2697                         x_cur[nx_cur] = x;
2698                         nx_cur++;
2699                         if ((xc = xfrm_state_migrate(x, mp))) {
2700                                 x_new[nx_new] = xc;
2701                                 nx_new++;
2702                         } else {
2703                                 err = -ENODATA;
2704                                 goto restore_state;
2705                         }
2706                 }
2707         }
2708
2709         /* Stage 3 - update policy */
2710         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2711                 goto restore_state;
2712
2713         /* Stage 4 - delete old state(s) */
2714         if (nx_cur) {
2715                 xfrm_states_put(x_cur, nx_cur);
2716                 xfrm_states_delete(x_cur, nx_cur);
2717         }
2718
2719         /* Stage 5 - announce */
2720         km_migrate(sel, dir, type, m, num_migrate);
2721
2722         xfrm_pol_put(pol);
2723
2724         return 0;
2725 out:
2726         return err;
2727
2728 restore_state:
2729         if (pol)
2730                 xfrm_pol_put(pol);
2731         if (nx_cur)
2732                 xfrm_states_put(x_cur, nx_cur);
2733         if (nx_new)
2734                 xfrm_states_delete(x_new, nx_new);
2735
2736         return err;
2737 }
2738 EXPORT_SYMBOL(xfrm_migrate);
2739 #endif