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);
 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(pol);
 680                                 if (*err) {
 681                                         write_unlock_bh(&xfrm_policy_lock);
 682                                         return pol;
 683                                 }
 684                                 hlist_del(&pol->bydst);
 685                                 hlist_del(&pol->byidx);
 686                                 xfrm_policy_count[dir]--;
 687                         }
 688                         ret = pol;
 689                         break;
 690                 }
 691         }
 692         write_unlock_bh(&xfrm_policy_lock);
 693
 694         if (ret && delete) {
 695                 atomic_inc(&flow_cache_genid);
 696                 xfrm_policy_kill(ret);
 697         }
 698         return ret;
 699 }
 700 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 701
 702 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 703                                      int *err)
 704 {
 705         struct xfrm_policy *pol, *ret;
 706         struct hlist_head *chain;
 707         struct hlist_node *entry;
 708
 709         *err = -ENOENT;
 710         if (xfrm_policy_id2dir(id) != dir)
 711                 return NULL;
 712
 713         *err = 0;
 714         write_lock_bh(&xfrm_policy_lock);
 715         chain = xfrm_policy_byidx + idx_hash(id);
 716         ret = NULL;
 717         hlist_for_each_entry(pol, entry, chain, byidx) {
 718                 if (pol->type == type && pol->index == id) {
 719                         xfrm_pol_hold(pol);
 720                         if (delete) {
 721                                 *err = security_xfrm_policy_delete(pol);
 722                                 if (*err) {
 723                                         write_unlock_bh(&xfrm_policy_lock);
 724                                         return pol;
 725                                 }
 726                                 hlist_del(&pol->bydst);
 727                                 hlist_del(&pol->byidx);
 728                                 xfrm_policy_count[dir]--;
 729                         }
 730                         ret = pol;
 731                         break;
 732                 }
 733         }
 734         write_unlock_bh(&xfrm_policy_lock);
 735
 736         if (ret && delete) {
 737                 atomic_inc(&flow_cache_genid);
 738                 xfrm_policy_kill(ret);
 739         }
 740         return ret;
 741 }
 742 EXPORT_SYMBOL(xfrm_policy_byid);
 743
 744 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 745 static inline int
 746 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 747 {
 748         int dir, err = 0;
 749
 750         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 751                 struct xfrm_policy *pol;
 752                 struct hlist_node *entry;
 753                 int i;
 754
 755                 hlist_for_each_entry(pol, entry,
 756                                      &xfrm_policy_inexact[dir], bydst) {
 757                         if (pol->type != type)
 758                                 continue;
 759                         err = security_xfrm_policy_delete(pol);
 760                         if (err) {
 761                                 xfrm_audit_policy_delete(pol, 0,
 762                                                          audit_info->loginuid,
 763                                                          audit_info->secid);
 764                                 return err;
 765                         }
 766                 }
 767                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 768                         hlist_for_each_entry(pol, entry,
 769                                              xfrm_policy_bydst[dir].table + i,
 770                                              bydst) {
 771                                 if (pol->type != type)
 772                                         continue;
 773                                 err = security_xfrm_policy_delete(pol);
 774                                 if (err) {
 775                                         xfrm_audit_policy_delete(pol, 0,
 776                                                         audit_info->loginuid,
 777                                                         audit_info->secid);
 778                                         return err;
 779                                 }
 780                         }
 781                 }
 782         }
 783         return err;
 784 }
 785 #else
 786 static inline int
 787 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 788 {
 789         return 0;
 790 }
 791 #endif
 792
 793 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 794 {
 795         int dir, err = 0;
 796
 797         write_lock_bh(&xfrm_policy_lock);
 798
 799         err = xfrm_policy_flush_secctx_check(type, audit_info);
 800         if (err)
 801                 goto out;
 802
 803         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 804                 struct xfrm_policy *pol;
 805                 struct hlist_node *entry;
 806                 int i, killed;
 807
 808                 killed = 0;
 809         again1:
 810                 hlist_for_each_entry(pol, entry,
 811                                      &xfrm_policy_inexact[dir], bydst) {
 812                         if (pol->type != type)
 813                                 continue;
 814                         hlist_del(&pol->bydst);
 815                         hlist_del(&pol->byidx);
 816                         write_unlock_bh(&xfrm_policy_lock);
 817
 818                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 819                                                  audit_info->secid);
 820
 821                         xfrm_policy_kill(pol);
 822                         killed++;
 823
 824                         write_lock_bh(&xfrm_policy_lock);
 825                         goto again1;
 826                 }
 827
 828                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 829         again2:
 830                         hlist_for_each_entry(pol, entry,
 831                                              xfrm_policy_bydst[dir].table + i,
 832                                              bydst) {
 833                                 if (pol->type != type)
 834                                         continue;
 835                                 hlist_del(&pol->bydst);
 836                                 hlist_del(&pol->byidx);
 837                                 write_unlock_bh(&xfrm_policy_lock);
 838
 839                                 xfrm_audit_policy_delete(pol, 1,
 840                                                          audit_info->loginuid,
 841                                                          audit_info->secid);
 842                                 xfrm_policy_kill(pol);
 843                                 killed++;
 844
 845                                 write_lock_bh(&xfrm_policy_lock);
 846                                 goto again2;
 847                         }
 848                 }
 849
 850                 xfrm_policy_count[dir] -= killed;
 851         }
 852         atomic_inc(&flow_cache_genid);
 853 out:
 854         write_unlock_bh(&xfrm_policy_lock);
 855         return err;
 856 }
 857 EXPORT_SYMBOL(xfrm_policy_flush);
 858
 859 int xfrm_policy_walk(struct xfrm_policy_walk *walk,
 860                      int (*func)(struct xfrm_policy *, int, int, void*),
 861                      void *data)
 862 {
 863         struct xfrm_policy *old, *pol, *last = NULL;
 864         int error = 0;
 865
 866         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 867             walk->type != XFRM_POLICY_TYPE_ANY)
 868                 return -EINVAL;
 869
 870         if (walk->policy == NULL && walk->count != 0)
 871                 return 0;
 872
 873         old = pol = walk->policy;
 874         walk->policy = NULL;
 875         read_lock_bh(&xfrm_policy_lock);
 876
 877         for (; walk->cur_type < XFRM_POLICY_TYPE_MAX; walk->cur_type++) {
 878                 if (walk->type != walk->cur_type &&
 879                     walk->type != XFRM_POLICY_TYPE_ANY)
 880                         continue;
 881
 882                 if (pol == NULL) {
 883                         pol = list_first_entry(&xfrm_policy_bytype[walk->cur_type],
 884                                                struct xfrm_policy, bytype);
 885                 }
 886                 list_for_each_entry_from(pol, &xfrm_policy_bytype[walk->cur_type], bytype) {
 887                         if (pol->dead)
 888                                 continue;
 889                         if (last) {
 890                                 error = func(last, xfrm_policy_id2dir(last->index),
 891                                              walk->count, data);
 892                                 if (error) {
 893                                         xfrm_pol_hold(last);
 894                                         walk->policy = last;
 895                                         goto out;
 896                                 }
 897                         }
 898                         last = pol;
 899                         walk->count++;
 900                 }
 901                 pol = NULL;
 902         }
 903         if (walk->count == 0) {
 904                 error = -ENOENT;
 905                 goto out;
 906         }
 907         if (last)
 908                 error = func(last, xfrm_policy_id2dir(last->index), 0, data);
 909 out:
 910         read_unlock_bh(&xfrm_policy_lock);
 911         if (old != NULL)
 912                 xfrm_pol_put(old);
 913         return error;
 914 }
 915 EXPORT_SYMBOL(xfrm_policy_walk);
 916
 917 /*
 918  * Find policy to apply to this flow.
 919  *
 920  * Returns 0 if policy found, else an -errno.
 921  */
 922 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 923                              u8 type, u16 family, int dir)
 924 {
 925         struct xfrm_selector *sel = &pol->selector;
 926         int match, ret = -ESRCH;
 927
 928         if (pol->family != family ||
 929             pol->type != type)
 930                 return ret;
 931
 932         match = xfrm_selector_match(sel, fl, family);
 933         if (match)
 934                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
 935
 936         return ret;
 937 }
 938
 939 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 940                                                      u16 family, u8 dir)
 941 {
 942         int err;
 943         struct xfrm_policy *pol, *ret;
 944         xfrm_address_t *daddr, *saddr;
 945         struct hlist_node *entry;
 946         struct hlist_head *chain;
 947         u32 priority = ~0U;
 948
 949         daddr = xfrm_flowi_daddr(fl, family);
 950         saddr = xfrm_flowi_saddr(fl, family);
 951         if (unlikely(!daddr || !saddr))
 952                 return NULL;
 953
 954         read_lock_bh(&xfrm_policy_lock);
 955         chain = policy_hash_direct(daddr, saddr, family, dir);
 956         ret = NULL;
 957         hlist_for_each_entry(pol, entry, chain, bydst) {
 958                 err = xfrm_policy_match(pol, fl, type, family, dir);
 959                 if (err) {
 960                         if (err == -ESRCH)
 961                                 continue;
 962                         else {
 963                                 ret = ERR_PTR(err);
 964                                 goto fail;
 965                         }
 966                 } else {
 967                         ret = pol;
 968                         priority = ret->priority;
 969                         break;
 970                 }
 971         }
 972         chain = &xfrm_policy_inexact[dir];
 973         hlist_for_each_entry(pol, entry, chain, bydst) {
 974                 err = xfrm_policy_match(pol, fl, type, family, dir);
 975                 if (err) {
 976                         if (err == -ESRCH)
 977                                 continue;
 978                         else {
 979                                 ret = ERR_PTR(err);
 980                                 goto fail;
 981                         }
 982                 } else if (pol->priority < priority) {
 983                         ret = pol;
 984                         break;
 985                 }
 986         }
 987         if (ret)
 988                 xfrm_pol_hold(ret);
 989 fail:
 990         read_unlock_bh(&xfrm_policy_lock);
 991
 992         return ret;
 993 }
 994
 995 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 996                                void **objp, atomic_t **obj_refp)
 997 {
 998         struct xfrm_policy *pol;
 999         int err = 0;
1000
1001 #ifdef CONFIG_XFRM_SUB_POLICY
1002         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1003         if (IS_ERR(pol)) {
1004                 err = PTR_ERR(pol);
1005                 pol = NULL;
1006         }
1007         if (pol || err)
1008                 goto end;
1009 #endif
1010         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1011         if (IS_ERR(pol)) {
1012                 err = PTR_ERR(pol);
1013                 pol = NULL;
1014         }
1015 #ifdef CONFIG_XFRM_SUB_POLICY
1016 end:
1017 #endif
1018         if ((*objp = (void *) pol) != NULL)
1019                 *obj_refp = &pol->refcnt;
1020         return err;
1021 }
1022
1023 static inline int policy_to_flow_dir(int dir)
1024 {
1025         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1026             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1027             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1028                 return dir;
1029         switch (dir) {
1030         default:
1031         case XFRM_POLICY_IN:
1032                 return FLOW_DIR_IN;
1033         case XFRM_POLICY_OUT:
1034                 return FLOW_DIR_OUT;
1035         case XFRM_POLICY_FWD:
1036                 return FLOW_DIR_FWD;
1037         }
1038 }
1039
1040 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1041 {
1042         struct xfrm_policy *pol;
1043
1044         read_lock_bh(&xfrm_policy_lock);
1045         if ((pol = sk->sk_policy[dir]) != NULL) {
1046                 int match = xfrm_selector_match(&pol->selector, fl,
1047                                                 sk->sk_family);
1048                 int err = 0;
1049
1050                 if (match) {
1051                         err = security_xfrm_policy_lookup(pol, fl->secid,
1052                                         policy_to_flow_dir(dir));
1053                         if (!err)
1054                                 xfrm_pol_hold(pol);
1055                         else if (err == -ESRCH)
1056                                 pol = NULL;
1057                         else
1058                                 pol = ERR_PTR(err);
1059                 } else
1060                         pol = NULL;
1061         }
1062         read_unlock_bh(&xfrm_policy_lock);
1063         return pol;
1064 }
1065
1066 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1067 {
1068         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1069                                                      pol->family, dir);
1070
1071         hlist_add_head(&pol->bydst, chain);
1072         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1073         xfrm_policy_count[dir]++;
1074         xfrm_pol_hold(pol);
1075
1076         if (xfrm_bydst_should_resize(dir, NULL))
1077                 schedule_work(&xfrm_hash_work);
1078 }
1079
1080 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1081                                                 int dir)
1082 {
1083         if (hlist_unhashed(&pol->bydst))
1084                 return NULL;
1085
1086         hlist_del(&pol->bydst);
1087         hlist_del(&pol->byidx);
1088         xfrm_policy_count[dir]--;
1089
1090         return pol;
1091 }
1092
1093 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1094 {
1095         write_lock_bh(&xfrm_policy_lock);
1096         pol = __xfrm_policy_unlink(pol, dir);
1097         write_unlock_bh(&xfrm_policy_lock);
1098         if (pol) {
1099                 if (dir < XFRM_POLICY_MAX)
1100                         atomic_inc(&flow_cache_genid);
1101                 xfrm_policy_kill(pol);
1102                 return 0;
1103         }
1104         return -ENOENT;
1105 }
1106 EXPORT_SYMBOL(xfrm_policy_delete);
1107
1108 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1109 {
1110         struct xfrm_policy *old_pol;
1111
1112 #ifdef CONFIG_XFRM_SUB_POLICY
1113         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1114                 return -EINVAL;
1115 #endif
1116
1117         write_lock_bh(&xfrm_policy_lock);
1118         old_pol = sk->sk_policy[dir];
1119         sk->sk_policy[dir] = pol;
1120         if (pol) {
1121                 pol->curlft.add_time = get_seconds();
1122                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1123                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1124         }
1125         if (old_pol)
1126                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1127         write_unlock_bh(&xfrm_policy_lock);
1128
1129         if (old_pol) {
1130                 xfrm_policy_kill(old_pol);
1131         }
1132         return 0;
1133 }
1134
1135 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1136 {
1137         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1138
1139         if (newp) {
1140                 newp->selector = old->selector;
1141                 if (security_xfrm_policy_clone(old, newp)) {
1142                         kfree(newp);
1143                         return NULL;  /* ENOMEM */
1144                 }
1145                 newp->lft = old->lft;
1146                 newp->curlft = old->curlft;
1147                 newp->action = old->action;
1148                 newp->flags = old->flags;
1149                 newp->xfrm_nr = old->xfrm_nr;
1150                 newp->index = old->index;
1151                 newp->type = old->type;
1152                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1153                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1154                 write_lock_bh(&xfrm_policy_lock);
1155                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1156                 write_unlock_bh(&xfrm_policy_lock);
1157                 xfrm_pol_put(newp);
1158         }
1159         return newp;
1160 }
1161
1162 int __xfrm_sk_clone_policy(struct sock *sk)
1163 {
1164         struct xfrm_policy *p0 = sk->sk_policy[0],
1165                            *p1 = sk->sk_policy[1];
1166
1167         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1168         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1169                 return -ENOMEM;
1170         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1171                 return -ENOMEM;
1172         return 0;
1173 }
1174
1175 static int
1176 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1177                unsigned short family)
1178 {
1179         int err;
1180         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1181
1182         if (unlikely(afinfo == NULL))
1183                 return -EINVAL;
1184         err = afinfo->get_saddr(local, remote);
1185         xfrm_policy_put_afinfo(afinfo);
1186         return err;
1187 }
1188
1189 /* Resolve list of templates for the flow, given policy. */
1190
1191 static int
1192 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1193                       struct xfrm_state **xfrm,
1194                       unsigned short family)
1195 {
1196         int nx;
1197         int i, error;
1198         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1199         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1200         xfrm_address_t tmp;
1201
1202         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1203                 struct xfrm_state *x;
1204                 xfrm_address_t *remote = daddr;
1205                 xfrm_address_t *local  = saddr;
1206                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1207
1208                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1209                     tmpl->mode == XFRM_MODE_BEET) {
1210                         remote = &tmpl->id.daddr;
1211                         local = &tmpl->saddr;
1212                         family = tmpl->encap_family;
1213                         if (xfrm_addr_any(local, family)) {
1214                                 error = xfrm_get_saddr(&tmp, remote, family);
1215                                 if (error)
1216                                         goto fail;
1217                                 local = &tmp;
1218                         }
1219                 }
1220
1221                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1222
1223                 if (x && x->km.state == XFRM_STATE_VALID) {
1224                         xfrm[nx++] = x;
1225                         daddr = remote;
1226                         saddr = local;
1227                         continue;
1228                 }
1229                 if (x) {
1230                         error = (x->km.state == XFRM_STATE_ERROR ?
1231                                  -EINVAL : -EAGAIN);
1232                         xfrm_state_put(x);
1233                 }
1234
1235                 if (!tmpl->optional)
1236                         goto fail;
1237         }
1238         return nx;
1239
1240 fail:
1241         for (nx--; nx>=0; nx--)
1242                 xfrm_state_put(xfrm[nx]);
1243         return error;
1244 }
1245
1246 static int
1247 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1248                   struct xfrm_state **xfrm,
1249                   unsigned short family)
1250 {
1251         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1252         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1253         int cnx = 0;
1254         int error;
1255         int ret;
1256         int i;
1257
1258         for (i = 0; i < npols; i++) {
1259                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1260                         error = -ENOBUFS;
1261                         goto fail;
1262                 }
1263
1264                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1265                 if (ret < 0) {
1266                         error = ret;
1267                         goto fail;
1268                 } else
1269                         cnx += ret;
1270         }
1271
1272         /* found states are sorted for outbound processing */
1273         if (npols > 1)
1274                 xfrm_state_sort(xfrm, tpp, cnx, family);
1275
1276         return cnx;
1277
1278  fail:
1279         for (cnx--; cnx>=0; cnx--)
1280                 xfrm_state_put(tpp[cnx]);
1281         return error;
1282
1283 }
1284
1285 /* Check that the bundle accepts the flow and its components are
1286  * still valid.
1287  */
1288
1289 static struct dst_entry *
1290 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1291 {
1292         struct dst_entry *x;
1293         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1294         if (unlikely(afinfo == NULL))
1295                 return ERR_PTR(-EINVAL);
1296         x = afinfo->find_bundle(fl, policy);
1297         xfrm_policy_put_afinfo(afinfo);
1298         return x;
1299 }
1300
1301 static inline int xfrm_get_tos(struct flowi *fl, int family)
1302 {
1303         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1304         int tos;
1305
1306         if (!afinfo)
1307                 return -EINVAL;
1308
1309         tos = afinfo->get_tos(fl);
1310
1311         xfrm_policy_put_afinfo(afinfo);
1312
1313         return tos;
1314 }
1315
1316 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1317 {
1318         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1319         struct xfrm_dst *xdst;
1320
1321         if (!afinfo)
1322                 return ERR_PTR(-EINVAL);
1323
1324         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1325
1326         xfrm_policy_put_afinfo(afinfo);
1327
1328         return xdst;
1329 }
1330
1331 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1332                                  int nfheader_len)
1333 {
1334         struct xfrm_policy_afinfo *afinfo =
1335                 xfrm_policy_get_afinfo(dst->ops->family);
1336         int err;
1337
1338         if (!afinfo)
1339                 return -EINVAL;
1340
1341         err = afinfo->init_path(path, dst, nfheader_len);
1342
1343         xfrm_policy_put_afinfo(afinfo);
1344
1345         return err;
1346 }
1347
1348 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1349 {
1350         struct xfrm_policy_afinfo *afinfo =
1351                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1352         int err;
1353
1354         if (!afinfo)
1355                 return -EINVAL;
1356
1357         err = afinfo->fill_dst(xdst, dev);
1358
1359         xfrm_policy_put_afinfo(afinfo);
1360
1361         return err;
1362 }
1363
1364 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1365  * all the metrics... Shortly, bundle a bundle.
1366  */
1367
1368 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1369                                             struct xfrm_state **xfrm, int nx,
1370                                             struct flowi *fl,
1371                                             struct dst_entry *dst)
1372 {
1373         unsigned long now = jiffies;
1374         struct net_device *dev;
1375         struct dst_entry *dst_prev = NULL;
1376         struct dst_entry *dst0 = NULL;
1377         int i = 0;
1378         int err;
1379         int header_len = 0;
1380         int nfheader_len = 0;
1381         int trailer_len = 0;
1382         int tos;
1383         int family = policy->selector.family;
1384         xfrm_address_t saddr, daddr;
1385
1386         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1387
1388         tos = xfrm_get_tos(fl, family);
1389         err = tos;
1390         if (tos < 0)
1391                 goto put_states;
1392
1393         dst_hold(dst);
1394
1395         for (; i < nx; i++) {
1396                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1397                 struct dst_entry *dst1 = &xdst->u.dst;
1398
1399                 err = PTR_ERR(xdst);
1400                 if (IS_ERR(xdst)) {
1401                         dst_release(dst);
1402                         goto put_states;
1403                 }
1404
1405                 if (!dst_prev)
1406                         dst0 = dst1;
1407                 else {
1408                         dst_prev->child = dst_clone(dst1);
1409                         dst1->flags |= DST_NOHASH;
1410                 }
1411
1412                 xdst->route = dst;
1413                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1414
1415                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1416                         family = xfrm[i]->props.family;
1417                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1418                                               family);
1419                         err = PTR_ERR(dst);
1420                         if (IS_ERR(dst))
1421                                 goto put_states;
1422                 } else
1423                         dst_hold(dst);
1424
1425                 dst1->xfrm = xfrm[i];
1426                 xdst->genid = xfrm[i]->genid;
1427
1428                 dst1->obsolete = -1;
1429                 dst1->flags |= DST_HOST;
1430                 dst1->lastuse = now;
1431
1432                 dst1->input = dst_discard;
1433                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1434
1435                 dst1->next = dst_prev;
1436                 dst_prev = dst1;
1437
1438                 header_len += xfrm[i]->props.header_len;
1439                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1440                         nfheader_len += xfrm[i]->props.header_len;
1441                 trailer_len += xfrm[i]->props.trailer_len;
1442         }
1443
1444         dst_prev->child = dst;
1445         dst0->path = dst;
1446
1447         err = -ENODEV;
1448         dev = dst->dev;
1449         if (!dev)
1450                 goto free_dst;
1451
1452         /* Copy neighbout for reachability confirmation */
1453         dst0->neighbour = neigh_clone(dst->neighbour);
1454
1455         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1456         xfrm_init_pmtu(dst_prev);
1457
1458         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1459                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1460
1461                 err = xfrm_fill_dst(xdst, dev);
1462                 if (err)
1463                         goto free_dst;
1464
1465                 dst_prev->header_len = header_len;
1466                 dst_prev->trailer_len = trailer_len;
1467                 header_len -= xdst->u.dst.xfrm->props.header_len;
1468                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1469         }
1470
1471 out:
1472         return dst0;
1473
1474 put_states:
1475         for (; i < nx; i++)
1476                 xfrm_state_put(xfrm[i]);
1477 free_dst:
1478         if (dst0)
1479                 dst_free(dst0);
1480         dst0 = ERR_PTR(err);
1481         goto out;
1482 }
1483
1484 static int inline
1485 xfrm_dst_alloc_copy(void **target, void *src, int size)
1486 {
1487         if (!*target) {
1488                 *target = kmalloc(size, GFP_ATOMIC);
1489                 if (!*target)
1490                         return -ENOMEM;
1491         }
1492         memcpy(*target, src, size);
1493         return 0;
1494 }
1495
1496 static int inline
1497 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1498 {
1499 #ifdef CONFIG_XFRM_SUB_POLICY
1500         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1501         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1502                                    sel, sizeof(*sel));
1503 #else
1504         return 0;
1505 #endif
1506 }
1507
1508 static int inline
1509 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1510 {
1511 #ifdef CONFIG_XFRM_SUB_POLICY
1512         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1513         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1514 #else
1515         return 0;
1516 #endif
1517 }
1518
1519 static int stale_bundle(struct dst_entry *dst);
1520
1521 /* Main function: finds/creates a bundle for given flow.
1522  *
1523  * At the moment we eat a raw IP route. Mostly to speed up lookups
1524  * on interfaces with disabled IPsec.
1525  */
1526 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1527                   struct sock *sk, int flags)
1528 {
1529         struct xfrm_policy *policy;
1530         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1531         int npols;
1532         int pol_dead;
1533         int xfrm_nr;
1534         int pi;
1535         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1536         struct dst_entry *dst, *dst_orig = *dst_p;
1537         int nx = 0;
1538         int err;
1539         u32 genid;
1540         u16 family;
1541         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1542
1543 restart:
1544         genid = atomic_read(&flow_cache_genid);
1545         policy = NULL;
1546         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1547                 pols[pi] = NULL;
1548         npols = 0;
1549         pol_dead = 0;
1550         xfrm_nr = 0;
1551
1552         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1553                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1554                 err = PTR_ERR(policy);
1555                 if (IS_ERR(policy)) {
1556                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1557                         goto dropdst;
1558                 }
1559         }
1560
1561         if (!policy) {
1562                 /* To accelerate a bit...  */
1563                 if ((dst_orig->flags & DST_NOXFRM) ||
1564                     !xfrm_policy_count[XFRM_POLICY_OUT])
1565                         goto nopol;
1566
1567                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1568                                            dir, xfrm_policy_lookup);
1569                 err = PTR_ERR(policy);
1570                 if (IS_ERR(policy)) {
1571                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1572                         goto dropdst;
1573                 }
1574         }
1575
1576         if (!policy)
1577                 goto nopol;
1578
1579         family = dst_orig->ops->family;
1580         pols[0] = policy;
1581         npols ++;
1582         xfrm_nr += pols[0]->xfrm_nr;
1583
1584         err = -ENOENT;
1585         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1586                 goto error;
1587
1588         policy->curlft.use_time = get_seconds();
1589
1590         switch (policy->action) {
1591         default:
1592         case XFRM_POLICY_BLOCK:
1593                 /* Prohibit the flow */
1594                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1595                 err = -EPERM;
1596                 goto error;
1597
1598         case XFRM_POLICY_ALLOW:
1599 #ifndef CONFIG_XFRM_SUB_POLICY
1600                 if (policy->xfrm_nr == 0) {
1601                         /* Flow passes not transformed. */
1602                         xfrm_pol_put(policy);
1603                         return 0;
1604                 }
1605 #endif
1606
1607                 /* Try to find matching bundle.
1608                  *
1609                  * LATER: help from flow cache. It is optional, this
1610                  * is required only for output policy.
1611                  */
1612                 dst = xfrm_find_bundle(fl, policy, family);
1613                 if (IS_ERR(dst)) {
1614                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1615                         err = PTR_ERR(dst);
1616                         goto error;
1617                 }
1618
1619                 if (dst)
1620                         break;
1621
1622 #ifdef CONFIG_XFRM_SUB_POLICY
1623                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1624                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1625                                                             fl, family,
1626                                                             XFRM_POLICY_OUT);
1627                         if (pols[1]) {
1628                                 if (IS_ERR(pols[1])) {
1629                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1630                                         err = PTR_ERR(pols[1]);
1631                                         goto error;
1632                                 }
1633                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1634                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1635                                         err = -EPERM;
1636                                         goto error;
1637                                 }
1638                                 npols ++;
1639                                 xfrm_nr += pols[1]->xfrm_nr;
1640                         }
1641                 }
1642
1643                 /*
1644                  * Because neither flowi nor bundle information knows about
1645                  * transformation template size. On more than one policy usage
1646                  * we can realize whether all of them is bypass or not after
1647                  * they are searched. See above not-transformed bypass
1648                  * is surrounded by non-sub policy configuration, too.
1649                  */
1650                 if (xfrm_nr == 0) {
1651                         /* Flow passes not transformed. */
1652                         xfrm_pols_put(pols, npols);
1653                         return 0;
1654                 }
1655
1656 #endif
1657                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1658
1659                 if (unlikely(nx<0)) {
1660                         err = nx;
1661                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1662                                 /* EREMOTE tells the caller to generate
1663                                  * a one-shot blackhole route.
1664                                  */
1665                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1666                                 xfrm_pol_put(policy);
1667                                 return -EREMOTE;
1668                         }
1669                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1670                                 DECLARE_WAITQUEUE(wait, current);
1671
1672                                 add_wait_queue(&km_waitq, &wait);
1673                                 set_current_state(TASK_INTERRUPTIBLE);
1674                                 schedule();
1675                                 set_current_state(TASK_RUNNING);
1676                                 remove_wait_queue(&km_waitq, &wait);
1677
1678                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1679
1680                                 if (nx == -EAGAIN && signal_pending(current)) {
1681                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1682                                         err = -ERESTART;
1683                                         goto error;
1684                                 }
1685                                 if (nx == -EAGAIN ||
1686                                     genid != atomic_read(&flow_cache_genid)) {
1687                                         xfrm_pols_put(pols, npols);
1688                                         goto restart;
1689                                 }
1690                                 err = nx;
1691                         }
1692                         if (err < 0) {
1693                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1694                                 goto error;
1695                         }
1696                 }
1697                 if (nx == 0) {
1698                         /* Flow passes not transformed. */
1699                         xfrm_pols_put(pols, npols);
1700                         return 0;
1701                 }
1702
1703                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1704                 err = PTR_ERR(dst);
1705                 if (IS_ERR(dst)) {
1706                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1707                         goto error;
1708                 }
1709
1710                 for (pi = 0; pi < npols; pi++) {
1711                         read_lock_bh(&pols[pi]->lock);
1712                         pol_dead |= pols[pi]->dead;
1713                         read_unlock_bh(&pols[pi]->lock);
1714                 }
1715
1716                 write_lock_bh(&policy->lock);
1717                 if (unlikely(pol_dead || stale_bundle(dst))) {
1718                         /* Wow! While we worked on resolving, this
1719                          * policy has gone. Retry. It is not paranoia,
1720                          * we just cannot enlist new bundle to dead object.
1721                          * We can't enlist stable bundles either.
1722                          */
1723                         write_unlock_bh(&policy->lock);
1724                         if (dst)
1725                                 dst_free(dst);
1726
1727                         if (pol_dead)
1728                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1729                         else
1730                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1731                         err = -EHOSTUNREACH;
1732                         goto error;
1733                 }
1734
1735                 if (npols > 1)
1736                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1737                 else
1738                         err = xfrm_dst_update_origin(dst, fl);
1739                 if (unlikely(err)) {
1740                         write_unlock_bh(&policy->lock);
1741                         if (dst)
1742                                 dst_free(dst);
1743                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1744                         goto error;
1745                 }
1746
1747                 dst->next = policy->bundles;
1748                 policy->bundles = dst;
1749                 dst_hold(dst);
1750                 write_unlock_bh(&policy->lock);
1751         }
1752         *dst_p = dst;
1753         dst_release(dst_orig);
1754         xfrm_pols_put(pols, npols);
1755         return 0;
1756
1757 error:
1758         xfrm_pols_put(pols, npols);
1759 dropdst:
1760         dst_release(dst_orig);
1761         *dst_p = NULL;
1762         return err;
1763
1764 nopol:
1765         err = -ENOENT;
1766         if (flags & XFRM_LOOKUP_ICMP)
1767                 goto dropdst;
1768         return 0;
1769 }
1770 EXPORT_SYMBOL(__xfrm_lookup);
1771
1772 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1773                 struct sock *sk, int flags)
1774 {
1775         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1776
1777         if (err == -EREMOTE) {
1778                 dst_release(*dst_p);
1779                 *dst_p = NULL;
1780                 err = -EAGAIN;
1781         }
1782
1783         return err;
1784 }
1785 EXPORT_SYMBOL(xfrm_lookup);
1786
1787 static inline int
1788 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1789 {
1790         struct xfrm_state *x;
1791
1792         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1793                 return 0;
1794         x = skb->sp->xvec[idx];
1795         if (!x->type->reject)
1796                 return 0;
1797         return x->type->reject(x, skb, fl);
1798 }
1799
1800 /* When skb is transformed back to its "native" form, we have to
1801  * check policy restrictions. At the moment we make this in maximally
1802  * stupid way. Shame on me. :-) Of course, connected sockets must
1803  * have policy cached at them.
1804  */
1805
1806 static inline int
1807 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1808               unsigned short family)
1809 {
1810         if (xfrm_state_kern(x))
1811                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1812         return  x->id.proto == tmpl->id.proto &&
1813                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1814                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1815                 x->props.mode == tmpl->mode &&
1816                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1817                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1818                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1819                   xfrm_state_addr_cmp(tmpl, x, family));
1820 }
1821
1822 /*
1823  * 0 or more than 0 is returned when validation is succeeded (either bypass
1824  * because of optional transport mode, or next index of the mathced secpath
1825  * state with the template.
1826  * -1 is returned when no matching template is found.
1827  * Otherwise "-2 - errored_index" is returned.
1828  */
1829 static inline int
1830 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1831                unsigned short family)
1832 {
1833         int idx = start;
1834
1835         if (tmpl->optional) {
1836                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1837                         return start;
1838         } else
1839                 start = -1;
1840         for (; idx < sp->len; idx++) {
1841                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1842                         return ++idx;
1843                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1844                         if (start == -1)
1845                                 start = -2-idx;
1846                         break;
1847                 }
1848         }
1849         return start;
1850 }
1851
1852 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1853                           unsigned int family, int reverse)
1854 {
1855         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1856         int err;
1857
1858         if (unlikely(afinfo == NULL))
1859                 return -EAFNOSUPPORT;
1860
1861         afinfo->decode_session(skb, fl, reverse);
1862         err = security_xfrm_decode_session(skb, &fl->secid);
1863         xfrm_policy_put_afinfo(afinfo);
1864         return err;
1865 }
1866 EXPORT_SYMBOL(__xfrm_decode_session);
1867
1868 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1869 {
1870         for (; k < sp->len; k++) {
1871                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1872                         *idxp = k;
1873                         return 1;
1874                 }
1875         }
1876
1877         return 0;
1878 }
1879
1880 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1881                         unsigned short family)
1882 {
1883         struct xfrm_policy *pol;
1884         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1885         int npols = 0;
1886         int xfrm_nr;
1887         int pi;
1888         int reverse;
1889         struct flowi fl;
1890         u8 fl_dir;
1891         int xerr_idx = -1;
1892
1893         reverse = dir & ~XFRM_POLICY_MASK;
1894         dir &= XFRM_POLICY_MASK;
1895         fl_dir = policy_to_flow_dir(dir);
1896
1897         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1898                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1899                 return 0;
1900         }
1901
1902         nf_nat_decode_session(skb, &fl, family);
1903
1904         /* First, check used SA against their selectors. */
1905         if (skb->sp) {
1906                 int i;
1907
1908                 for (i=skb->sp->len-1; i>=0; i--) {
1909                         struct xfrm_state *x = skb->sp->xvec[i];
1910                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1911                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1912                                 return 0;
1913                         }
1914                 }
1915         }
1916
1917         pol = NULL;
1918         if (sk && sk->sk_policy[dir]) {
1919                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1920                 if (IS_ERR(pol)) {
1921                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1922                         return 0;
1923                 }
1924         }
1925
1926         if (!pol)
1927                 pol = flow_cache_lookup(&fl, family, fl_dir,
1928                                         xfrm_policy_lookup);
1929
1930         if (IS_ERR(pol)) {
1931                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1932                 return 0;
1933         }
1934
1935         if (!pol) {
1936                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1937                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1938                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1939                         return 0;
1940                 }
1941                 return 1;
1942         }
1943
1944         pol->curlft.use_time = get_seconds();
1945
1946         pols[0] = pol;
1947         npols ++;
1948 #ifdef CONFIG_XFRM_SUB_POLICY
1949         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1950                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1951                                                     &fl, family,
1952                                                     XFRM_POLICY_IN);
1953                 if (pols[1]) {
1954                         if (IS_ERR(pols[1])) {
1955                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1956                                 return 0;
1957                         }
1958                         pols[1]->curlft.use_time = get_seconds();
1959                         npols ++;
1960                 }
1961         }
1962 #endif
1963
1964         if (pol->action == XFRM_POLICY_ALLOW) {
1965                 struct sec_path *sp;
1966                 static struct sec_path dummy;
1967                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1968                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1969                 struct xfrm_tmpl **tpp = tp;
1970                 int ti = 0;
1971                 int i, k;
1972
1973                 if ((sp = skb->sp) == NULL)
1974                         sp = &dummy;
1975
1976                 for (pi = 0; pi < npols; pi++) {
1977                         if (pols[pi] != pol &&
1978                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1979                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1980                                 goto reject;
1981                         }
1982                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1983                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1984                                 goto reject_error;
1985                         }
1986                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1987                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1988                 }
1989                 xfrm_nr = ti;
1990                 if (npols > 1) {
1991                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1992                         tpp = stp;
1993                 }
1994
1995                 /* For each tunnel xfrm, find the first matching tmpl.
1996                  * For each tmpl before that, find corresponding xfrm.
1997                  * Order is _important_. Later we will implement
1998                  * some barriers, but at the moment barriers
1999                  * are implied between each two transformations.
2000                  */
2001                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2002                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2003                         if (k < 0) {
2004                                 if (k < -1)
2005                                         /* "-2 - errored_index" returned */
2006                                         xerr_idx = -(2+k);
2007                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2008                                 goto reject;
2009                         }
2010                 }
2011
2012                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2013                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2014                         goto reject;
2015                 }
2016
2017                 xfrm_pols_put(pols, npols);
2018                 return 1;
2019         }
2020         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2021
2022 reject:
2023         xfrm_secpath_reject(xerr_idx, skb, &fl);
2024 reject_error:
2025         xfrm_pols_put(pols, npols);
2026         return 0;
2027 }
2028 EXPORT_SYMBOL(__xfrm_policy_check);
2029
2030 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2031 {
2032         struct flowi fl;
2033
2034         if (xfrm_decode_session(skb, &fl, family) < 0) {
2035                 /* XXX: we should have something like FWDHDRERROR here. */
2036                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2037                 return 0;
2038         }
2039
2040         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2041 }
2042 EXPORT_SYMBOL(__xfrm_route_forward);
2043
2044 /* Optimize later using cookies and generation ids. */
2045
2046 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2047 {
2048         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2049          * to "-1" to force all XFRM destinations to get validated by
2050          * dst_ops->check on every use.  We do this because when a
2051          * normal route referenced by an XFRM dst is obsoleted we do
2052          * not go looking around for all parent referencing XFRM dsts
2053          * so that we can invalidate them.  It is just too much work.
2054          * Instead we make the checks here on every use.  For example:
2055          *
2056          *      XFRM dst A --> IPv4 dst X
2057          *
2058          * X is the "xdst->route" of A (X is also the "dst->path" of A
2059          * in this example).  If X is marked obsolete, "A" will not
2060          * notice.  That's what we are validating here via the
2061          * stale_bundle() check.
2062          *
2063          * When a policy's bundle is pruned, we dst_free() the XFRM
2064          * dst which causes it's ->obsolete field to be set to a
2065          * positive non-zero integer.  If an XFRM dst has been pruned
2066          * like this, we want to force a new route lookup.
2067          */
2068         if (dst->obsolete < 0 && !stale_bundle(dst))
2069                 return dst;
2070
2071         return NULL;
2072 }
2073
2074 static int stale_bundle(struct dst_entry *dst)
2075 {
2076         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2077 }
2078
2079 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2080 {
2081         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2082                 dst->dev = dev_net(dev)->loopback_dev;
2083                 dev_hold(dst->dev);
2084                 dev_put(dev);
2085         }
2086 }
2087 EXPORT_SYMBOL(xfrm_dst_ifdown);
2088
2089 static void xfrm_link_failure(struct sk_buff *skb)
2090 {
2091         /* Impossible. Such dst must be popped before reaches point of failure. */
2092         return;
2093 }
2094
2095 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2096 {
2097         if (dst) {
2098                 if (dst->obsolete) {
2099                         dst_release(dst);
2100                         dst = NULL;
2101                 }
2102         }
2103         return dst;
2104 }
2105
2106 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2107 {
2108         struct dst_entry *dst, **dstp;
2109
2110         write_lock(&pol->lock);
2111         dstp = &pol->bundles;
2112         while ((dst=*dstp) != NULL) {
2113                 if (func(dst)) {
2114                         *dstp = dst->next;
2115                         dst->next = *gc_list_p;
2116                         *gc_list_p = dst;
2117                 } else {
2118                         dstp = &dst->next;
2119                 }
2120         }
2121         write_unlock(&pol->lock);
2122 }
2123
2124 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2125 {
2126         struct dst_entry *gc_list = NULL;
2127         int dir;
2128
2129         read_lock_bh(&xfrm_policy_lock);
2130         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2131                 struct xfrm_policy *pol;
2132                 struct hlist_node *entry;
2133                 struct hlist_head *table;
2134                 int i;
2135
2136                 hlist_for_each_entry(pol, entry,
2137                                      &xfrm_policy_inexact[dir], bydst)
2138                         prune_one_bundle(pol, func, &gc_list);
2139
2140                 table = xfrm_policy_bydst[dir].table;
2141                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2142                         hlist_for_each_entry(pol, entry, table + i, bydst)
2143                                 prune_one_bundle(pol, func, &gc_list);
2144                 }
2145         }
2146         read_unlock_bh(&xfrm_policy_lock);
2147
2148         while (gc_list) {
2149                 struct dst_entry *dst = gc_list;
2150                 gc_list = dst->next;
2151                 dst_free(dst);
2152         }
2153 }
2154
2155 static int unused_bundle(struct dst_entry *dst)
2156 {
2157         return !atomic_read(&dst->__refcnt);
2158 }
2159
2160 static void __xfrm_garbage_collect(void)
2161 {
2162         xfrm_prune_bundles(unused_bundle);
2163 }
2164
2165 static int xfrm_flush_bundles(void)
2166 {
2167         xfrm_prune_bundles(stale_bundle);
2168         return 0;
2169 }
2170
2171 static void xfrm_init_pmtu(struct dst_entry *dst)
2172 {
2173         do {
2174                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2175                 u32 pmtu, route_mtu_cached;
2176
2177                 pmtu = dst_mtu(dst->child);
2178                 xdst->child_mtu_cached = pmtu;
2179
2180                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2181
2182                 route_mtu_cached = dst_mtu(xdst->route);
2183                 xdst->route_mtu_cached = route_mtu_cached;
2184
2185                 if (pmtu > route_mtu_cached)
2186                         pmtu = route_mtu_cached;
2187
2188                 dst->metrics[RTAX_MTU-1] = pmtu;
2189         } while ((dst = dst->next));
2190 }
2191
2192 /* Check that the bundle accepts the flow and its components are
2193  * still valid.
2194  */
2195
2196 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2197                 struct flowi *fl, int family, int strict)
2198 {
2199         struct dst_entry *dst = &first->u.dst;
2200         struct xfrm_dst *last;
2201         u32 mtu;
2202
2203         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2204             (dst->dev && !netif_running(dst->dev)))
2205                 return 0;
2206 #ifdef CONFIG_XFRM_SUB_POLICY
2207         if (fl) {
2208                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2209                         return 0;
2210                 if (first->partner &&
2211                     !xfrm_selector_match(first->partner, fl, family))
2212                         return 0;
2213         }
2214 #endif
2215
2216         last = NULL;
2217
2218         do {
2219                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2220
2221                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2222                         return 0;
2223                 if (fl && pol &&
2224                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2225                         return 0;
2226                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2227                         return 0;
2228                 if (xdst->genid != dst->xfrm->genid)
2229                         return 0;
2230
2231                 if (strict && fl &&
2232                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2233                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2234                         return 0;
2235
2236                 mtu = dst_mtu(dst->child);
2237                 if (xdst->child_mtu_cached != mtu) {
2238                         last = xdst;
2239                         xdst->child_mtu_cached = mtu;
2240                 }
2241
2242                 if (!dst_check(xdst->route, xdst->route_cookie))
2243                         return 0;
2244                 mtu = dst_mtu(xdst->route);
2245                 if (xdst->route_mtu_cached != mtu) {
2246                         last = xdst;
2247                         xdst->route_mtu_cached = mtu;
2248                 }
2249
2250                 dst = dst->child;
2251         } while (dst->xfrm);
2252
2253         if (likely(!last))
2254                 return 1;
2255
2256         mtu = last->child_mtu_cached;
2257         for (;;) {
2258                 dst = &last->u.dst;
2259
2260                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2261                 if (mtu > last->route_mtu_cached)
2262                         mtu = last->route_mtu_cached;
2263                 dst->metrics[RTAX_MTU-1] = mtu;
2264
2265                 if (last == first)
2266                         break;
2267
2268                 last = (struct xfrm_dst *)last->u.dst.next;
2269                 last->child_mtu_cached = mtu;
2270         }
2271
2272         return 1;
2273 }
2274
2275 EXPORT_SYMBOL(xfrm_bundle_ok);
2276
2277 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2278 {
2279         int err = 0;
2280         if (unlikely(afinfo == NULL))
2281                 return -EINVAL;
2282         if (unlikely(afinfo->family >= NPROTO))
2283                 return -EAFNOSUPPORT;
2284         write_lock_bh(&xfrm_policy_afinfo_lock);
2285         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2286                 err = -ENOBUFS;
2287         else {
2288                 struct dst_ops *dst_ops = afinfo->dst_ops;
2289                 if (likely(dst_ops->kmem_cachep == NULL))
2290                         dst_ops->kmem_cachep = xfrm_dst_cache;
2291                 if (likely(dst_ops->check == NULL))
2292                         dst_ops->check = xfrm_dst_check;
2293                 if (likely(dst_ops->negative_advice == NULL))
2294                         dst_ops->negative_advice = xfrm_negative_advice;
2295                 if (likely(dst_ops->link_failure == NULL))
2296                         dst_ops->link_failure = xfrm_link_failure;
2297                 if (likely(afinfo->garbage_collect == NULL))
2298                         afinfo->garbage_collect = __xfrm_garbage_collect;
2299                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2300         }
2301         write_unlock_bh(&xfrm_policy_afinfo_lock);
2302         return err;
2303 }
2304 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2305
2306 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2307 {
2308         int err = 0;
2309         if (unlikely(afinfo == NULL))
2310                 return -EINVAL;
2311         if (unlikely(afinfo->family >= NPROTO))
2312                 return -EAFNOSUPPORT;
2313         write_lock_bh(&xfrm_policy_afinfo_lock);
2314         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2315                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2316                         err = -EINVAL;
2317                 else {
2318                         struct dst_ops *dst_ops = afinfo->dst_ops;
2319                         xfrm_policy_afinfo[afinfo->family] = NULL;
2320                         dst_ops->kmem_cachep = NULL;
2321                         dst_ops->check = NULL;
2322                         dst_ops->negative_advice = NULL;
2323                         dst_ops->link_failure = NULL;
2324                         afinfo->garbage_collect = NULL;
2325                 }
2326         }
2327         write_unlock_bh(&xfrm_policy_afinfo_lock);
2328         return err;
2329 }
2330 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2331
2332 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2333 {
2334         struct xfrm_policy_afinfo *afinfo;
2335         if (unlikely(family >= NPROTO))
2336                 return NULL;
2337         read_lock(&xfrm_policy_afinfo_lock);
2338         afinfo = xfrm_policy_afinfo[family];
2339         if (unlikely(!afinfo))
2340                 read_unlock(&xfrm_policy_afinfo_lock);
2341         return afinfo;
2342 }
2343
2344 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2345 {
2346         read_unlock(&xfrm_policy_afinfo_lock);
2347 }
2348
2349 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2350 {
2351         struct net_device *dev = ptr;
2352
2353         if (dev_net(dev) != &init_net)
2354                 return NOTIFY_DONE;
2355
2356         switch (event) {
2357         case NETDEV_DOWN:
2358                 xfrm_flush_bundles();
2359         }
2360         return NOTIFY_DONE;
2361 }
2362
2363 static struct notifier_block xfrm_dev_notifier = {
2364         xfrm_dev_event,
2365         NULL,
2366         0
2367 };
2368
2369 #ifdef CONFIG_XFRM_STATISTICS
2370 static int __init xfrm_statistics_init(void)
2371 {
2372         if (snmp_mib_init((void **)xfrm_statistics,
2373                           sizeof(struct linux_xfrm_mib)) < 0)
2374                 return -ENOMEM;
2375         return 0;
2376 }
2377 #endif
2378
2379 static void __init xfrm_policy_init(void)
2380 {
2381         unsigned int hmask, sz;
2382         int dir;
2383
2384         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2385                                            sizeof(struct xfrm_dst),
2386                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2387                                            NULL);
2388
2389         hmask = 8 - 1;
2390         sz = (hmask+1) * sizeof(struct hlist_head);
2391
2392         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2393         xfrm_idx_hmask = hmask;
2394         if (!xfrm_policy_byidx)
2395                 panic("XFRM: failed to allocate byidx hash\n");
2396
2397         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2398                 struct xfrm_policy_hash *htab;
2399
2400                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2401
2402                 htab = &xfrm_policy_bydst[dir];
2403                 htab->table = xfrm_hash_alloc(sz);
2404                 htab->hmask = hmask;
2405                 if (!htab->table)
2406                         panic("XFRM: failed to allocate bydst hash\n");
2407         }
2408
2409         for (dir = 0; dir < XFRM_POLICY_TYPE_MAX; dir++)
2410                 INIT_LIST_HEAD(&xfrm_policy_bytype[dir]);
2411
2412         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2413         register_netdevice_notifier(&xfrm_dev_notifier);
2414 }
2415
2416 void __init xfrm_init(void)
2417 {
2418 #ifdef CONFIG_XFRM_STATISTICS
2419         xfrm_statistics_init();
2420 #endif
2421         xfrm_state_init();
2422         xfrm_policy_init();
2423         xfrm_input_init();
2424 #ifdef CONFIG_XFRM_STATISTICS
2425         xfrm_proc_init();
2426 #endif
2427 }
2428
2429 #ifdef CONFIG_AUDITSYSCALL
2430 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2431                                          struct audit_buffer *audit_buf)
2432 {
2433         struct xfrm_sec_ctx *ctx = xp->security;
2434         struct xfrm_selector *sel = &xp->selector;
2435
2436         if (ctx)
2437                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2438                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2439
2440         switch(sel->family) {
2441         case AF_INET:
2442                 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2443                                  NIPQUAD(sel->saddr.a4));
2444                 if (sel->prefixlen_s != 32)
2445                         audit_log_format(audit_buf, " src_prefixlen=%d",
2446                                          sel->prefixlen_s);
2447                 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2448                                  NIPQUAD(sel->daddr.a4));
2449                 if (sel->prefixlen_d != 32)
2450                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2451                                          sel->prefixlen_d);
2452                 break;
2453         case AF_INET6:
2454                 audit_log_format(audit_buf, " src=" NIP6_FMT,
2455                                  NIP6(*(struct in6_addr *)sel->saddr.a6));
2456                 if (sel->prefixlen_s != 128)
2457                         audit_log_format(audit_buf, " src_prefixlen=%d",
2458                                          sel->prefixlen_s);
2459                 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2460                                  NIP6(*(struct in6_addr *)sel->daddr.a6));
2461                 if (sel->prefixlen_d != 128)
2462                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2463                                          sel->prefixlen_d);
2464                 break;
2465         }
2466 }
2467
2468 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2469                            u32 auid, u32 secid)
2470 {
2471         struct audit_buffer *audit_buf;
2472
2473         audit_buf = xfrm_audit_start("SPD-add");
2474         if (audit_buf == NULL)
2475                 return;
2476         xfrm_audit_helper_usrinfo(auid, secid, audit_buf);
2477         audit_log_format(audit_buf, " res=%u", result);
2478         xfrm_audit_common_policyinfo(xp, audit_buf);
2479         audit_log_end(audit_buf);
2480 }
2481 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2482
2483 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2484                               u32 auid, u32 secid)
2485 {
2486         struct audit_buffer *audit_buf;
2487
2488         audit_buf = xfrm_audit_start("SPD-delete");
2489         if (audit_buf == NULL)
2490                 return;
2491         xfrm_audit_helper_usrinfo(auid, secid, audit_buf);
2492         audit_log_format(audit_buf, " res=%u", result);
2493         xfrm_audit_common_policyinfo(xp, audit_buf);
2494         audit_log_end(audit_buf);
2495 }
2496 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2497 #endif
2498
2499 #ifdef CONFIG_XFRM_MIGRATE
2500 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2501                                        struct xfrm_selector *sel_tgt)
2502 {
2503         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2504                 if (sel_tgt->family == sel_cmp->family &&
2505                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2506                                   sel_cmp->family) == 0 &&
2507                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2508                                   sel_cmp->family) == 0 &&
2509                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2510                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2511                         return 1;
2512                 }
2513         } else {
2514                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2515                         return 1;
2516                 }
2517         }
2518         return 0;
2519 }
2520
2521 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2522                                                      u8 dir, u8 type)
2523 {
2524         struct xfrm_policy *pol, *ret = NULL;
2525         struct hlist_node *entry;
2526         struct hlist_head *chain;
2527         u32 priority = ~0U;
2528
2529         read_lock_bh(&xfrm_policy_lock);
2530         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2531         hlist_for_each_entry(pol, entry, chain, bydst) {
2532                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2533                     pol->type == type) {
2534                         ret = pol;
2535                         priority = ret->priority;
2536                         break;
2537                 }
2538         }
2539         chain = &xfrm_policy_inexact[dir];
2540         hlist_for_each_entry(pol, entry, chain, bydst) {
2541                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2542                     pol->type == type &&
2543                     pol->priority < priority) {
2544                         ret = pol;
2545                         break;
2546                 }
2547         }
2548
2549         if (ret)
2550                 xfrm_pol_hold(ret);
2551
2552         read_unlock_bh(&xfrm_policy_lock);
2553
2554         return ret;
2555 }
2556
2557 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2558 {
2559         int match = 0;
2560
2561         if (t->mode == m->mode && t->id.proto == m->proto &&
2562             (m->reqid == 0 || t->reqid == m->reqid)) {
2563                 switch (t->mode) {
2564                 case XFRM_MODE_TUNNEL:
2565                 case XFRM_MODE_BEET:
2566                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2567                                           m->old_family) == 0 &&
2568                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2569                                           m->old_family) == 0) {
2570                                 match = 1;
2571                         }
2572                         break;
2573                 case XFRM_MODE_TRANSPORT:
2574                         /* in case of transport mode, template does not store
2575                            any IP addresses, hence we just compare mode and
2576                            protocol */
2577                         match = 1;
2578                         break;
2579                 default:
2580                         break;
2581                 }
2582         }
2583         return match;
2584 }
2585
2586 /* update endpoint address(es) of template(s) */
2587 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2588                                struct xfrm_migrate *m, int num_migrate)
2589 {
2590         struct xfrm_migrate *mp;
2591         struct dst_entry *dst;
2592         int i, j, n = 0;
2593
2594         write_lock_bh(&pol->lock);
2595         if (unlikely(pol->dead)) {
2596                 /* target policy has been deleted */
2597                 write_unlock_bh(&pol->lock);
2598                 return -ENOENT;
2599         }
2600
2601         for (i = 0; i < pol->xfrm_nr; i++) {
2602                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2603                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2604                                 continue;
2605                         n++;
2606                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2607                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2608                                 continue;
2609                         /* update endpoints */
2610                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2611                                sizeof(pol->xfrm_vec[i].id.daddr));
2612                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2613                                sizeof(pol->xfrm_vec[i].saddr));
2614                         pol->xfrm_vec[i].encap_family = mp->new_family;
2615                         /* flush bundles */
2616                         while ((dst = pol->bundles) != NULL) {
2617                                 pol->bundles = dst->next;
2618                                 dst_free(dst);
2619                         }
2620                 }
2621         }
2622
2623         write_unlock_bh(&pol->lock);
2624
2625         if (!n)
2626                 return -ENODATA;
2627
2628         return 0;
2629 }
2630
2631 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2632 {
2633         int i, j;
2634
2635         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2636                 return -EINVAL;
2637
2638         for (i = 0; i < num_migrate; i++) {
2639                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2640                                    m[i].old_family) == 0) &&
2641                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2642                                    m[i].old_family) == 0))
2643                         return -EINVAL;
2644                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2645                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2646                         return -EINVAL;
2647
2648                 /* check if there is any duplicated entry */
2649                 for (j = i + 1; j < num_migrate; j++) {
2650                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2651                                     sizeof(m[i].old_daddr)) &&
2652                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2653                                     sizeof(m[i].old_saddr)) &&
2654                             m[i].proto == m[j].proto &&
2655                             m[i].mode == m[j].mode &&
2656                             m[i].reqid == m[j].reqid &&
2657                             m[i].old_family == m[j].old_family)
2658                                 return -EINVAL;
2659                 }
2660         }
2661
2662         return 0;
2663 }
2664
2665 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2666                  struct xfrm_migrate *m, int num_migrate)
2667 {
2668         int i, err, nx_cur = 0, nx_new = 0;
2669         struct xfrm_policy *pol = NULL;
2670         struct xfrm_state *x, *xc;
2671         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2672         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2673         struct xfrm_migrate *mp;
2674
2675         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2676                 goto out;
2677
2678         /* Stage 1 - find policy */
2679         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2680                 err = -ENOENT;
2681                 goto out;
2682         }
2683
2684         /* Stage 2 - find and update state(s) */
2685         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2686                 if ((x = xfrm_migrate_state_find(mp))) {
2687                         x_cur[nx_cur] = x;
2688                         nx_cur++;
2689                         if ((xc = xfrm_state_migrate(x, mp))) {
2690                                 x_new[nx_new] = xc;
2691                                 nx_new++;
2692                         } else {
2693                                 err = -ENODATA;
2694                                 goto restore_state;
2695                         }
2696                 }
2697         }
2698
2699         /* Stage 3 - update policy */
2700         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2701                 goto restore_state;
2702
2703         /* Stage 4 - delete old state(s) */
2704         if (nx_cur) {
2705                 xfrm_states_put(x_cur, nx_cur);
2706                 xfrm_states_delete(x_cur, nx_cur);
2707         }
2708
2709         /* Stage 5 - announce */
2710         km_migrate(sel, dir, type, m, num_migrate);
2711
2712         xfrm_pol_put(pol);
2713
2714         return 0;
2715 out:
2716         return err;
2717
2718 restore_state:
2719         if (pol)
2720                 xfrm_pol_put(pol);
2721         if (nx_cur)
2722                 xfrm_states_put(x_cur, nx_cur);
2723         if (nx_new)
2724                 xfrm_states_delete(x_new, nx_new);
2725
2726         return err;
2727 }
2728 EXPORT_SYMBOL(xfrm_migrate);
2729 #endif