SAFE public projects git trees. - safe/jmp/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 = 1;
  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 DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  50 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  51
  52 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  53
  54 static HLIST_HEAD(xfrm_policy_gc_list);
  55 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  56
  57 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  58 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  59 static void xfrm_init_pmtu(struct dst_entry *dst);
  60
  61 static inline int
  62 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  63 {
  64         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  65                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  66                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  67                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  68                 (fl->proto == sel->proto || !sel->proto) &&
  69                 (fl->oif == sel->ifindex || !sel->ifindex);
  70 }
  71
  72 static inline int
  73 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  74 {
  75         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  76                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  77                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  78                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  79                 (fl->proto == sel->proto || !sel->proto) &&
  80                 (fl->oif == sel->ifindex || !sel->ifindex);
  81 }
  82
  83 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  84                     unsigned short family)
  85 {
  86         switch (family) {
  87         case AF_INET:
  88                 return __xfrm4_selector_match(sel, fl);
  89         case AF_INET6:
  90                 return __xfrm6_selector_match(sel, fl);
  91         }
  92         return 0;
  93 }
  94
  95 static inline struct dst_entry *__xfrm_dst_lookup(int tos,
  96                                                   xfrm_address_t *saddr,
  97                                                   xfrm_address_t *daddr,
  98                                                   int family)
  99 {
 100         struct xfrm_policy_afinfo *afinfo;
 101         struct dst_entry *dst;
 102
 103         afinfo = xfrm_policy_get_afinfo(family);
 104         if (unlikely(afinfo == NULL))
 105                 return ERR_PTR(-EAFNOSUPPORT);
 106
 107         dst = afinfo->dst_lookup(tos, saddr, daddr);
 108
 109         xfrm_policy_put_afinfo(afinfo);
 110
 111         return dst;
 112 }
 113
 114 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 115                                                 xfrm_address_t *prev_saddr,
 116                                                 xfrm_address_t *prev_daddr,
 117                                                 int family)
 118 {
 119         xfrm_address_t *saddr = &x->props.saddr;
 120         xfrm_address_t *daddr = &x->id.daddr;
 121         struct dst_entry *dst;
 122
 123         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 124                 saddr = x->coaddr;
 125                 daddr = prev_daddr;
 126         }
 127         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 128                 saddr = prev_saddr;
 129                 daddr = x->coaddr;
 130         }
 131
 132         dst = __xfrm_dst_lookup(tos, saddr, daddr, family);
 133
 134         if (!IS_ERR(dst)) {
 135                 if (prev_saddr != saddr)
 136                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 137                 if (prev_daddr != daddr)
 138                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 139         }
 140
 141         return dst;
 142 }
 143
 144 static inline unsigned long make_jiffies(long secs)
 145 {
 146         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 147                 return MAX_SCHEDULE_TIMEOUT-1;
 148         else
 149                 return secs*HZ;
 150 }
 151
 152 static void xfrm_policy_timer(unsigned long data)
 153 {
 154         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 155         unsigned long now = get_seconds();
 156         long next = LONG_MAX;
 157         int warn = 0;
 158         int dir;
 159
 160         read_lock(&xp->lock);
 161
 162         if (xp->walk.dead)
 163                 goto out;
 164
 165         dir = xfrm_policy_id2dir(xp->index);
 166
 167         if (xp->lft.hard_add_expires_seconds) {
 168                 long tmo = xp->lft.hard_add_expires_seconds +
 169                         xp->curlft.add_time - now;
 170                 if (tmo <= 0)
 171                         goto expired;
 172                 if (tmo < next)
 173                         next = tmo;
 174         }
 175         if (xp->lft.hard_use_expires_seconds) {
 176                 long tmo = xp->lft.hard_use_expires_seconds +
 177                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 178                 if (tmo <= 0)
 179                         goto expired;
 180                 if (tmo < next)
 181                         next = tmo;
 182         }
 183         if (xp->lft.soft_add_expires_seconds) {
 184                 long tmo = xp->lft.soft_add_expires_seconds +
 185                         xp->curlft.add_time - now;
 186                 if (tmo <= 0) {
 187                         warn = 1;
 188                         tmo = XFRM_KM_TIMEOUT;
 189                 }
 190                 if (tmo < next)
 191                         next = tmo;
 192         }
 193         if (xp->lft.soft_use_expires_seconds) {
 194                 long tmo = xp->lft.soft_use_expires_seconds +
 195                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 196                 if (tmo <= 0) {
 197                         warn = 1;
 198                         tmo = XFRM_KM_TIMEOUT;
 199                 }
 200                 if (tmo < next)
 201                         next = tmo;
 202         }
 203
 204         if (warn)
 205                 km_policy_expired(xp, dir, 0, 0);
 206         if (next != LONG_MAX &&
 207             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 208                 xfrm_pol_hold(xp);
 209
 210 out:
 211         read_unlock(&xp->lock);
 212         xfrm_pol_put(xp);
 213         return;
 214
 215 expired:
 216         read_unlock(&xp->lock);
 217         if (!xfrm_policy_delete(xp, dir))
 218                 km_policy_expired(xp, dir, 1, 0);
 219         xfrm_pol_put(xp);
 220 }
 221
 222
 223 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 224  * SPD calls.
 225  */
 226
 227 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 228 {
 229         struct xfrm_policy *policy;
 230
 231         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 232
 233         if (policy) {
 234                 write_pnet(&policy->xp_net, net);
 235                 INIT_LIST_HEAD(&policy->walk.all);
 236                 INIT_HLIST_NODE(&policy->bydst);
 237                 INIT_HLIST_NODE(&policy->byidx);
 238                 rwlock_init(&policy->lock);
 239                 atomic_set(&policy->refcnt, 1);
 240                 setup_timer(&policy->timer, xfrm_policy_timer,
 241                                 (unsigned long)policy);
 242         }
 243         return policy;
 244 }
 245 EXPORT_SYMBOL(xfrm_policy_alloc);
 246
 247 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 248
 249 void xfrm_policy_destroy(struct xfrm_policy *policy)
 250 {
 251         BUG_ON(!policy->walk.dead);
 252
 253         BUG_ON(policy->bundles);
 254
 255         if (del_timer(&policy->timer))
 256                 BUG();
 257
 258         security_xfrm_policy_free(policy->security);
 259         kfree(policy);
 260 }
 261 EXPORT_SYMBOL(xfrm_policy_destroy);
 262
 263 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 264 {
 265         struct dst_entry *dst;
 266
 267         while ((dst = policy->bundles) != NULL) {
 268                 policy->bundles = dst->next;
 269                 dst_free(dst);
 270         }
 271
 272         if (del_timer(&policy->timer))
 273                 atomic_dec(&policy->refcnt);
 274
 275         if (atomic_read(&policy->refcnt) > 1)
 276                 flow_cache_flush();
 277
 278         xfrm_pol_put(policy);
 279 }
 280
 281 static void xfrm_policy_gc_task(struct work_struct *work)
 282 {
 283         struct xfrm_policy *policy;
 284         struct hlist_node *entry, *tmp;
 285         struct hlist_head gc_list;
 286
 287         spin_lock_bh(&xfrm_policy_gc_lock);
 288         gc_list.first = xfrm_policy_gc_list.first;
 289         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 290         spin_unlock_bh(&xfrm_policy_gc_lock);
 291
 292         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 293                 xfrm_policy_gc_kill(policy);
 294 }
 295 static DECLARE_WORK(xfrm_policy_gc_work, xfrm_policy_gc_task);
 296
 297 /* Rule must be locked. Release descentant resources, announce
 298  * entry dead. The rule must be unlinked from lists to the moment.
 299  */
 300
 301 static void xfrm_policy_kill(struct xfrm_policy *policy)
 302 {
 303         int dead;
 304
 305         write_lock_bh(&policy->lock);
 306         dead = policy->walk.dead;
 307         policy->walk.dead = 1;
 308         write_unlock_bh(&policy->lock);
 309
 310         if (unlikely(dead)) {
 311                 WARN_ON(1);
 312                 return;
 313         }
 314
 315         spin_lock_bh(&xfrm_policy_gc_lock);
 316         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 317         spin_unlock_bh(&xfrm_policy_gc_lock);
 318
 319         schedule_work(&xfrm_policy_gc_work);
 320 }
 321
 322 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 323
 324 static inline unsigned int idx_hash(struct net *net, u32 index)
 325 {
 326         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 327 }
 328
 329 static struct hlist_head *policy_hash_bysel(struct net *net, struct xfrm_selector *sel, unsigned short family, int dir)
 330 {
 331         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 332         unsigned int hash = __sel_hash(sel, family, hmask);
 333
 334         return (hash == hmask + 1 ?
 335                 &net->xfrm.policy_inexact[dir] :
 336                 net->xfrm.policy_bydst[dir].table + hash);
 337 }
 338
 339 static struct hlist_head *policy_hash_direct(struct net *net, xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 340 {
 341         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 342         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 343
 344         return net->xfrm.policy_bydst[dir].table + hash;
 345 }
 346
 347 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 348                                    struct hlist_head *ndsttable,
 349                                    unsigned int nhashmask)
 350 {
 351         struct hlist_node *entry, *tmp, *entry0 = NULL;
 352         struct xfrm_policy *pol;
 353         unsigned int h0 = 0;
 354
 355 redo:
 356         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 357                 unsigned int h;
 358
 359                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 360                                 pol->family, nhashmask);
 361                 if (!entry0) {
 362                         hlist_del(entry);
 363                         hlist_add_head(&pol->bydst, ndsttable+h);
 364                         h0 = h;
 365                 } else {
 366                         if (h != h0)
 367                                 continue;
 368                         hlist_del(entry);
 369                         hlist_add_after(entry0, &pol->bydst);
 370                 }
 371                 entry0 = entry;
 372         }
 373         if (!hlist_empty(list)) {
 374                 entry0 = NULL;
 375                 goto redo;
 376         }
 377 }
 378
 379 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 380                                    struct hlist_head *nidxtable,
 381                                    unsigned int nhashmask)
 382 {
 383         struct hlist_node *entry, *tmp;
 384         struct xfrm_policy *pol;
 385
 386         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 387                 unsigned int h;
 388
 389                 h = __idx_hash(pol->index, nhashmask);
 390                 hlist_add_head(&pol->byidx, nidxtable+h);
 391         }
 392 }
 393
 394 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 395 {
 396         return ((old_hmask + 1) << 1) - 1;
 397 }
 398
 399 static void xfrm_bydst_resize(struct net *net, int dir)
 400 {
 401         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 402         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 403         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 404         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 405         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 406         int i;
 407
 408         if (!ndst)
 409                 return;
 410
 411         write_lock_bh(&xfrm_policy_lock);
 412
 413         for (i = hmask; i >= 0; i--)
 414                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 415
 416         net->xfrm.policy_bydst[dir].table = ndst;
 417         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 418
 419         write_unlock_bh(&xfrm_policy_lock);
 420
 421         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 422 }
 423
 424 static void xfrm_byidx_resize(struct net *net, int total)
 425 {
 426         unsigned int hmask = net->xfrm.policy_idx_hmask;
 427         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 428         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 429         struct hlist_head *oidx = net->xfrm.policy_byidx;
 430         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 431         int i;
 432
 433         if (!nidx)
 434                 return;
 435
 436         write_lock_bh(&xfrm_policy_lock);
 437
 438         for (i = hmask; i >= 0; i--)
 439                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 440
 441         net->xfrm.policy_byidx = nidx;
 442         net->xfrm.policy_idx_hmask = nhashmask;
 443
 444         write_unlock_bh(&xfrm_policy_lock);
 445
 446         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 447 }
 448
 449 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 450 {
 451         unsigned int cnt = net->xfrm.policy_count[dir];
 452         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 453
 454         if (total)
 455                 *total += cnt;
 456
 457         if ((hmask + 1) < xfrm_policy_hashmax &&
 458             cnt > hmask)
 459                 return 1;
 460
 461         return 0;
 462 }
 463
 464 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 465 {
 466         unsigned int hmask = net->xfrm.policy_idx_hmask;
 467
 468         if ((hmask + 1) < xfrm_policy_hashmax &&
 469             total > hmask)
 470                 return 1;
 471
 472         return 0;
 473 }
 474
 475 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 476 {
 477         read_lock_bh(&xfrm_policy_lock);
 478         si->incnt = init_net.xfrm.policy_count[XFRM_POLICY_IN];
 479         si->outcnt = init_net.xfrm.policy_count[XFRM_POLICY_OUT];
 480         si->fwdcnt = init_net.xfrm.policy_count[XFRM_POLICY_FWD];
 481         si->inscnt = init_net.xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 482         si->outscnt = init_net.xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 483         si->fwdscnt = init_net.xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 484         si->spdhcnt = init_net.xfrm.policy_idx_hmask;
 485         si->spdhmcnt = xfrm_policy_hashmax;
 486         read_unlock_bh(&xfrm_policy_lock);
 487 }
 488 EXPORT_SYMBOL(xfrm_spd_getinfo);
 489
 490 static DEFINE_MUTEX(hash_resize_mutex);
 491 static void xfrm_hash_resize(struct work_struct *work)
 492 {
 493         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 494         int dir, total;
 495
 496         mutex_lock(&hash_resize_mutex);
 497
 498         total = 0;
 499         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 500                 if (xfrm_bydst_should_resize(net, dir, &total))
 501                         xfrm_bydst_resize(net, dir);
 502         }
 503         if (xfrm_byidx_should_resize(net, total))
 504                 xfrm_byidx_resize(net, total);
 505
 506         mutex_unlock(&hash_resize_mutex);
 507 }
 508
 509 /* Generate new index... KAME seems to generate them ordered by cost
 510  * of an absolute inpredictability of ordering of rules. This will not pass. */
 511 static u32 xfrm_gen_index(struct net *net, int dir)
 512 {
 513         static u32 idx_generator;
 514
 515         for (;;) {
 516                 struct hlist_node *entry;
 517                 struct hlist_head *list;
 518                 struct xfrm_policy *p;
 519                 u32 idx;
 520                 int found;
 521
 522                 idx = (idx_generator | dir);
 523                 idx_generator += 8;
 524                 if (idx == 0)
 525                         idx = 8;
 526                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 527                 found = 0;
 528                 hlist_for_each_entry(p, entry, list, byidx) {
 529                         if (p->index == idx) {
 530                                 found = 1;
 531                                 break;
 532                         }
 533                 }
 534                 if (!found)
 535                         return idx;
 536         }
 537 }
 538
 539 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 540 {
 541         u32 *p1 = (u32 *) s1;
 542         u32 *p2 = (u32 *) s2;
 543         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 544         int i;
 545
 546         for (i = 0; i < len; i++) {
 547                 if (p1[i] != p2[i])
 548                         return 1;
 549         }
 550
 551         return 0;
 552 }
 553
 554 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 555 {
 556         struct net *net = xp_net(policy);
 557         struct xfrm_policy *pol;
 558         struct xfrm_policy *delpol;
 559         struct hlist_head *chain;
 560         struct hlist_node *entry, *newpos;
 561         struct dst_entry *gc_list;
 562
 563         write_lock_bh(&xfrm_policy_lock);
 564         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 565         delpol = NULL;
 566         newpos = NULL;
 567         hlist_for_each_entry(pol, entry, chain, bydst) {
 568                 if (pol->type == policy->type &&
 569                     !selector_cmp(&pol->selector, &policy->selector) &&
 570                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 571                     !WARN_ON(delpol)) {
 572                         if (excl) {
 573                                 write_unlock_bh(&xfrm_policy_lock);
 574                                 return -EEXIST;
 575                         }
 576                         delpol = pol;
 577                         if (policy->priority > pol->priority)
 578                                 continue;
 579                 } else if (policy->priority >= pol->priority) {
 580                         newpos = &pol->bydst;
 581                         continue;
 582                 }
 583                 if (delpol)
 584                         break;
 585         }
 586         if (newpos)
 587                 hlist_add_after(newpos, &policy->bydst);
 588         else
 589                 hlist_add_head(&policy->bydst, chain);
 590         xfrm_pol_hold(policy);
 591         net->xfrm.policy_count[dir]++;
 592         atomic_inc(&flow_cache_genid);
 593         if (delpol) {
 594                 hlist_del(&delpol->bydst);
 595                 hlist_del(&delpol->byidx);
 596                 list_del(&delpol->walk.all);
 597                 net->xfrm.policy_count[dir]--;
 598         }
 599         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 600         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 601         policy->curlft.add_time = get_seconds();
 602         policy->curlft.use_time = 0;
 603         if (!mod_timer(&policy->timer, jiffies + HZ))
 604                 xfrm_pol_hold(policy);
 605         list_add(&policy->walk.all, &net->xfrm.policy_all);
 606         write_unlock_bh(&xfrm_policy_lock);
 607
 608         if (delpol)
 609                 xfrm_policy_kill(delpol);
 610         else if (xfrm_bydst_should_resize(net, dir, NULL))
 611                 schedule_work(&net->xfrm.policy_hash_work);
 612
 613         read_lock_bh(&xfrm_policy_lock);
 614         gc_list = NULL;
 615         entry = &policy->bydst;
 616         hlist_for_each_entry_continue(policy, entry, bydst) {
 617                 struct dst_entry *dst;
 618
 619                 write_lock(&policy->lock);
 620                 dst = policy->bundles;
 621                 if (dst) {
 622                         struct dst_entry *tail = dst;
 623                         while (tail->next)
 624                                 tail = tail->next;
 625                         tail->next = gc_list;
 626                         gc_list = dst;
 627
 628                         policy->bundles = NULL;
 629                 }
 630                 write_unlock(&policy->lock);
 631         }
 632         read_unlock_bh(&xfrm_policy_lock);
 633
 634         while (gc_list) {
 635                 struct dst_entry *dst = gc_list;
 636
 637                 gc_list = dst->next;
 638                 dst_free(dst);
 639         }
 640
 641         return 0;
 642 }
 643 EXPORT_SYMBOL(xfrm_policy_insert);
 644
 645 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u8 type, int dir,
 646                                           struct xfrm_selector *sel,
 647                                           struct xfrm_sec_ctx *ctx, int delete,
 648                                           int *err)
 649 {
 650         struct xfrm_policy *pol, *ret;
 651         struct hlist_head *chain;
 652         struct hlist_node *entry;
 653
 654         *err = 0;
 655         write_lock_bh(&xfrm_policy_lock);
 656         chain = policy_hash_bysel(net, sel, sel->family, dir);
 657         ret = NULL;
 658         hlist_for_each_entry(pol, entry, chain, bydst) {
 659                 if (pol->type == type &&
 660                     !selector_cmp(sel, &pol->selector) &&
 661                     xfrm_sec_ctx_match(ctx, pol->security)) {
 662                         xfrm_pol_hold(pol);
 663                         if (delete) {
 664                                 *err = security_xfrm_policy_delete(
 665                                                                 pol->security);
 666                                 if (*err) {
 667                                         write_unlock_bh(&xfrm_policy_lock);
 668                                         return pol;
 669                                 }
 670                                 hlist_del(&pol->bydst);
 671                                 hlist_del(&pol->byidx);
 672                                 list_del(&pol->walk.all);
 673                                 net->xfrm.policy_count[dir]--;
 674                         }
 675                         ret = pol;
 676                         break;
 677                 }
 678         }
 679         write_unlock_bh(&xfrm_policy_lock);
 680
 681         if (ret && delete) {
 682                 atomic_inc(&flow_cache_genid);
 683                 xfrm_policy_kill(ret);
 684         }
 685         return ret;
 686 }
 687 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 688
 689 struct xfrm_policy *xfrm_policy_byid(struct net *net, u8 type, int dir, u32 id,
 690                                      int delete, int *err)
 691 {
 692         struct xfrm_policy *pol, *ret;
 693         struct hlist_head *chain;
 694         struct hlist_node *entry;
 695
 696         *err = -ENOENT;
 697         if (xfrm_policy_id2dir(id) != dir)
 698                 return NULL;
 699
 700         *err = 0;
 701         write_lock_bh(&xfrm_policy_lock);
 702         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 703         ret = NULL;
 704         hlist_for_each_entry(pol, entry, chain, byidx) {
 705                 if (pol->type == type && pol->index == id) {
 706                         xfrm_pol_hold(pol);
 707                         if (delete) {
 708                                 *err = security_xfrm_policy_delete(
 709                                                                 pol->security);
 710                                 if (*err) {
 711                                         write_unlock_bh(&xfrm_policy_lock);
 712                                         return pol;
 713                                 }
 714                                 hlist_del(&pol->bydst);
 715                                 hlist_del(&pol->byidx);
 716                                 list_del(&pol->walk.all);
 717                                 net->xfrm.policy_count[dir]--;
 718                         }
 719                         ret = pol;
 720                         break;
 721                 }
 722         }
 723         write_unlock_bh(&xfrm_policy_lock);
 724
 725         if (ret && delete) {
 726                 atomic_inc(&flow_cache_genid);
 727                 xfrm_policy_kill(ret);
 728         }
 729         return ret;
 730 }
 731 EXPORT_SYMBOL(xfrm_policy_byid);
 732
 733 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 734 static inline int
 735 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 736 {
 737         int dir, err = 0;
 738
 739         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 740                 struct xfrm_policy *pol;
 741                 struct hlist_node *entry;
 742                 int i;
 743
 744                 hlist_for_each_entry(pol, entry,
 745                                      &net->xfrm.policy_inexact[dir], bydst) {
 746                         if (pol->type != type)
 747                                 continue;
 748                         err = security_xfrm_policy_delete(pol->security);
 749                         if (err) {
 750                                 xfrm_audit_policy_delete(pol, 0,
 751                                                          audit_info->loginuid,
 752                                                          audit_info->sessionid,
 753                                                          audit_info->secid);
 754                                 return err;
 755                         }
 756                 }
 757                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 758                         hlist_for_each_entry(pol, entry,
 759                                              net->xfrm.policy_bydst[dir].table + i,
 760                                              bydst) {
 761                                 if (pol->type != type)
 762                                         continue;
 763                                 err = security_xfrm_policy_delete(
 764                                                                 pol->security);
 765                                 if (err) {
 766                                         xfrm_audit_policy_delete(pol, 0,
 767                                                         audit_info->loginuid,
 768                                                         audit_info->sessionid,
 769                                                         audit_info->secid);
 770                                         return err;
 771                                 }
 772                         }
 773                 }
 774         }
 775         return err;
 776 }
 777 #else
 778 static inline int
 779 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 780 {
 781         return 0;
 782 }
 783 #endif
 784
 785 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 786 {
 787         int dir, err = 0;
 788
 789         write_lock_bh(&xfrm_policy_lock);
 790
 791         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 792         if (err)
 793                 goto out;
 794
 795         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 796                 struct xfrm_policy *pol;
 797                 struct hlist_node *entry;
 798                 int i, killed;
 799
 800                 killed = 0;
 801         again1:
 802                 hlist_for_each_entry(pol, entry,
 803                                      &net->xfrm.policy_inexact[dir], bydst) {
 804                         if (pol->type != type)
 805                                 continue;
 806                         hlist_del(&pol->bydst);
 807                         hlist_del(&pol->byidx);
 808                         write_unlock_bh(&xfrm_policy_lock);
 809
 810                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 811                                                  audit_info->sessionid,
 812                                                  audit_info->secid);
 813
 814                         xfrm_policy_kill(pol);
 815                         killed++;
 816
 817                         write_lock_bh(&xfrm_policy_lock);
 818                         goto again1;
 819                 }
 820
 821                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 822         again2:
 823                         hlist_for_each_entry(pol, entry,
 824                                              net->xfrm.policy_bydst[dir].table + i,
 825                                              bydst) {
 826                                 if (pol->type != type)
 827                                         continue;
 828                                 hlist_del(&pol->bydst);
 829                                 hlist_del(&pol->byidx);
 830                                 list_del(&pol->walk.all);
 831                                 write_unlock_bh(&xfrm_policy_lock);
 832
 833                                 xfrm_audit_policy_delete(pol, 1,
 834                                                          audit_info->loginuid,
 835                                                          audit_info->sessionid,
 836                                                          audit_info->secid);
 837                                 xfrm_policy_kill(pol);
 838                                 killed++;
 839
 840                                 write_lock_bh(&xfrm_policy_lock);
 841                                 goto again2;
 842                         }
 843                 }
 844
 845                 net->xfrm.policy_count[dir] -= killed;
 846         }
 847         atomic_inc(&flow_cache_genid);
 848 out:
 849         write_unlock_bh(&xfrm_policy_lock);
 850         return err;
 851 }
 852 EXPORT_SYMBOL(xfrm_policy_flush);
 853
 854 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 855                      int (*func)(struct xfrm_policy *, int, int, void*),
 856                      void *data)
 857 {
 858         struct xfrm_policy *pol;
 859         struct xfrm_policy_walk_entry *x;
 860         int error = 0;
 861
 862         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 863             walk->type != XFRM_POLICY_TYPE_ANY)
 864                 return -EINVAL;
 865
 866         if (list_empty(&walk->walk.all) && walk->seq != 0)
 867                 return 0;
 868
 869         write_lock_bh(&xfrm_policy_lock);
 870         if (list_empty(&walk->walk.all))
 871                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 872         else
 873                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 874         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 875                 if (x->dead)
 876                         continue;
 877                 pol = container_of(x, struct xfrm_policy, walk);
 878                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 879                     walk->type != pol->type)
 880                         continue;
 881                 error = func(pol, xfrm_policy_id2dir(pol->index),
 882                              walk->seq, data);
 883                 if (error) {
 884                         list_move_tail(&walk->walk.all, &x->all);
 885                         goto out;
 886                 }
 887                 walk->seq++;
 888         }
 889         if (walk->seq == 0) {
 890                 error = -ENOENT;
 891                 goto out;
 892         }
 893         list_del_init(&walk->walk.all);
 894 out:
 895         write_unlock_bh(&xfrm_policy_lock);
 896         return error;
 897 }
 898 EXPORT_SYMBOL(xfrm_policy_walk);
 899
 900 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 901 {
 902         INIT_LIST_HEAD(&walk->walk.all);
 903         walk->walk.dead = 1;
 904         walk->type = type;
 905         walk->seq = 0;
 906 }
 907 EXPORT_SYMBOL(xfrm_policy_walk_init);
 908
 909 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 910 {
 911         if (list_empty(&walk->walk.all))
 912                 return;
 913
 914         write_lock_bh(&xfrm_policy_lock);
 915         list_del(&walk->walk.all);
 916         write_unlock_bh(&xfrm_policy_lock);
 917 }
 918 EXPORT_SYMBOL(xfrm_policy_walk_done);
 919
 920 /*
 921  * Find policy to apply to this flow.
 922  *
 923  * Returns 0 if policy found, else an -errno.
 924  */
 925 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 926                              u8 type, u16 family, int dir)
 927 {
 928         struct xfrm_selector *sel = &pol->selector;
 929         int match, ret = -ESRCH;
 930
 931         if (pol->family != family ||
 932             pol->type != type)
 933                 return ret;
 934
 935         match = xfrm_selector_match(sel, fl, family);
 936         if (match)
 937                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 938                                                   dir);
 939
 940         return ret;
 941 }
 942
 943 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 944                                                      struct flowi *fl,
 945                                                      u16 family, u8 dir)
 946 {
 947         int err;
 948         struct xfrm_policy *pol, *ret;
 949         xfrm_address_t *daddr, *saddr;
 950         struct hlist_node *entry;
 951         struct hlist_head *chain;
 952         u32 priority = ~0U;
 953
 954         daddr = xfrm_flowi_daddr(fl, family);
 955         saddr = xfrm_flowi_saddr(fl, family);
 956         if (unlikely(!daddr || !saddr))
 957                 return NULL;
 958
 959         read_lock_bh(&xfrm_policy_lock);
 960         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 961         ret = NULL;
 962         hlist_for_each_entry(pol, entry, chain, bydst) {
 963                 err = xfrm_policy_match(pol, fl, type, family, dir);
 964                 if (err) {
 965                         if (err == -ESRCH)
 966                                 continue;
 967                         else {
 968                                 ret = ERR_PTR(err);
 969                                 goto fail;
 970                         }
 971                 } else {
 972                         ret = pol;
 973                         priority = ret->priority;
 974                         break;
 975                 }
 976         }
 977         chain = &net->xfrm.policy_inexact[dir];
 978         hlist_for_each_entry(pol, entry, chain, bydst) {
 979                 err = xfrm_policy_match(pol, fl, type, family, dir);
 980                 if (err) {
 981                         if (err == -ESRCH)
 982                                 continue;
 983                         else {
 984                                 ret = ERR_PTR(err);
 985                                 goto fail;
 986                         }
 987                 } else if (pol->priority < priority) {
 988                         ret = pol;
 989                         break;
 990                 }
 991         }
 992         if (ret)
 993                 xfrm_pol_hold(ret);
 994 fail:
 995         read_unlock_bh(&xfrm_policy_lock);
 996
 997         return ret;
 998 }
 999
1000 static int xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
1001                               u8 dir, void **objp, atomic_t **obj_refp)
1002 {
1003         struct xfrm_policy *pol;
1004         int err = 0;
1005
1006 #ifdef CONFIG_XFRM_SUB_POLICY
1007         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1008         if (IS_ERR(pol)) {
1009                 err = PTR_ERR(pol);
1010                 pol = NULL;
1011         }
1012         if (pol || err)
1013                 goto end;
1014 #endif
1015         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1016         if (IS_ERR(pol)) {
1017                 err = PTR_ERR(pol);
1018                 pol = NULL;
1019         }
1020 #ifdef CONFIG_XFRM_SUB_POLICY
1021 end:
1022 #endif
1023         if ((*objp = (void *) pol) != NULL)
1024                 *obj_refp = &pol->refcnt;
1025         return err;
1026 }
1027
1028 static inline int policy_to_flow_dir(int dir)
1029 {
1030         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1031             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1032             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1033                 return dir;
1034         switch (dir) {
1035         default:
1036         case XFRM_POLICY_IN:
1037                 return FLOW_DIR_IN;
1038         case XFRM_POLICY_OUT:
1039                 return FLOW_DIR_OUT;
1040         case XFRM_POLICY_FWD:
1041                 return FLOW_DIR_FWD;
1042         }
1043 }
1044
1045 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1046 {
1047         struct xfrm_policy *pol;
1048
1049         read_lock_bh(&xfrm_policy_lock);
1050         if ((pol = sk->sk_policy[dir]) != NULL) {
1051                 int match = xfrm_selector_match(&pol->selector, fl,
1052                                                 sk->sk_family);
1053                 int err = 0;
1054
1055                 if (match) {
1056                         err = security_xfrm_policy_lookup(pol->security,
1057                                                       fl->secid,
1058                                                       policy_to_flow_dir(dir));
1059                         if (!err)
1060                                 xfrm_pol_hold(pol);
1061                         else if (err == -ESRCH)
1062                                 pol = NULL;
1063                         else
1064                                 pol = ERR_PTR(err);
1065                 } else
1066                         pol = NULL;
1067         }
1068         read_unlock_bh(&xfrm_policy_lock);
1069         return pol;
1070 }
1071
1072 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1073 {
1074         struct net *net = xp_net(pol);
1075         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1076                                                      pol->family, dir);
1077
1078         list_add(&pol->walk.all, &net->xfrm.policy_all);
1079         hlist_add_head(&pol->bydst, chain);
1080         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1081         net->xfrm.policy_count[dir]++;
1082         xfrm_pol_hold(pol);
1083
1084         if (xfrm_bydst_should_resize(net, dir, NULL))
1085                 schedule_work(&net->xfrm.policy_hash_work);
1086 }
1087
1088 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1089                                                 int dir)
1090 {
1091         struct net *net = xp_net(pol);
1092
1093         if (hlist_unhashed(&pol->bydst))
1094                 return NULL;
1095
1096         hlist_del(&pol->bydst);
1097         hlist_del(&pol->byidx);
1098         list_del(&pol->walk.all);
1099         net->xfrm.policy_count[dir]--;
1100
1101         return pol;
1102 }
1103
1104 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1105 {
1106         write_lock_bh(&xfrm_policy_lock);
1107         pol = __xfrm_policy_unlink(pol, dir);
1108         write_unlock_bh(&xfrm_policy_lock);
1109         if (pol) {
1110                 if (dir < XFRM_POLICY_MAX)
1111                         atomic_inc(&flow_cache_genid);
1112                 xfrm_policy_kill(pol);
1113                 return 0;
1114         }
1115         return -ENOENT;
1116 }
1117 EXPORT_SYMBOL(xfrm_policy_delete);
1118
1119 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1120 {
1121         struct net *net = xp_net(pol);
1122         struct xfrm_policy *old_pol;
1123
1124 #ifdef CONFIG_XFRM_SUB_POLICY
1125         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1126                 return -EINVAL;
1127 #endif
1128
1129         write_lock_bh(&xfrm_policy_lock);
1130         old_pol = sk->sk_policy[dir];
1131         sk->sk_policy[dir] = pol;
1132         if (pol) {
1133                 pol->curlft.add_time = get_seconds();
1134                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1135                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1136         }
1137         if (old_pol)
1138                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1139         write_unlock_bh(&xfrm_policy_lock);
1140
1141         if (old_pol) {
1142                 xfrm_policy_kill(old_pol);
1143         }
1144         return 0;
1145 }
1146
1147 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1148 {
1149         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1150
1151         if (newp) {
1152                 newp->selector = old->selector;
1153                 if (security_xfrm_policy_clone(old->security,
1154                                                &newp->security)) {
1155                         kfree(newp);
1156                         return NULL;  /* ENOMEM */
1157                 }
1158                 newp->lft = old->lft;
1159                 newp->curlft = old->curlft;
1160                 newp->action = old->action;
1161                 newp->flags = old->flags;
1162                 newp->xfrm_nr = old->xfrm_nr;
1163                 newp->index = old->index;
1164                 newp->type = old->type;
1165                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1166                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1167                 write_lock_bh(&xfrm_policy_lock);
1168                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1169                 write_unlock_bh(&xfrm_policy_lock);
1170                 xfrm_pol_put(newp);
1171         }
1172         return newp;
1173 }
1174
1175 int __xfrm_sk_clone_policy(struct sock *sk)
1176 {
1177         struct xfrm_policy *p0 = sk->sk_policy[0],
1178                            *p1 = sk->sk_policy[1];
1179
1180         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1181         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1182                 return -ENOMEM;
1183         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1184                 return -ENOMEM;
1185         return 0;
1186 }
1187
1188 static int
1189 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1190                unsigned short family)
1191 {
1192         int err;
1193         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1194
1195         if (unlikely(afinfo == NULL))
1196                 return -EINVAL;
1197         err = afinfo->get_saddr(local, remote);
1198         xfrm_policy_put_afinfo(afinfo);
1199         return err;
1200 }
1201
1202 /* Resolve list of templates for the flow, given policy. */
1203
1204 static int
1205 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1206                       struct xfrm_state **xfrm,
1207                       unsigned short family)
1208 {
1209         int nx;
1210         int i, error;
1211         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1212         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1213         xfrm_address_t tmp;
1214
1215         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1216                 struct xfrm_state *x;
1217                 xfrm_address_t *remote = daddr;
1218                 xfrm_address_t *local  = saddr;
1219                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1220
1221                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1222                     tmpl->mode == XFRM_MODE_BEET) {
1223                         remote = &tmpl->id.daddr;
1224                         local = &tmpl->saddr;
1225                         family = tmpl->encap_family;
1226                         if (xfrm_addr_any(local, family)) {
1227                                 error = xfrm_get_saddr(&tmp, remote, family);
1228                                 if (error)
1229                                         goto fail;
1230                                 local = &tmp;
1231                         }
1232                 }
1233
1234                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1235
1236                 if (x && x->km.state == XFRM_STATE_VALID) {
1237                         xfrm[nx++] = x;
1238                         daddr = remote;
1239                         saddr = local;
1240                         continue;
1241                 }
1242                 if (x) {
1243                         error = (x->km.state == XFRM_STATE_ERROR ?
1244                                  -EINVAL : -EAGAIN);
1245                         xfrm_state_put(x);
1246                 }
1247                 else if (error == -ESRCH)
1248                         error = -EAGAIN;
1249
1250                 if (!tmpl->optional)
1251                         goto fail;
1252         }
1253         return nx;
1254
1255 fail:
1256         for (nx--; nx>=0; nx--)
1257                 xfrm_state_put(xfrm[nx]);
1258         return error;
1259 }
1260
1261 static int
1262 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1263                   struct xfrm_state **xfrm,
1264                   unsigned short family)
1265 {
1266         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1267         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1268         int cnx = 0;
1269         int error;
1270         int ret;
1271         int i;
1272
1273         for (i = 0; i < npols; i++) {
1274                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1275                         error = -ENOBUFS;
1276                         goto fail;
1277                 }
1278
1279                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1280                 if (ret < 0) {
1281                         error = ret;
1282                         goto fail;
1283                 } else
1284                         cnx += ret;
1285         }
1286
1287         /* found states are sorted for outbound processing */
1288         if (npols > 1)
1289                 xfrm_state_sort(xfrm, tpp, cnx, family);
1290
1291         return cnx;
1292
1293  fail:
1294         for (cnx--; cnx>=0; cnx--)
1295                 xfrm_state_put(tpp[cnx]);
1296         return error;
1297
1298 }
1299
1300 /* Check that the bundle accepts the flow and its components are
1301  * still valid.
1302  */
1303
1304 static struct dst_entry *
1305 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1306 {
1307         struct dst_entry *x;
1308         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1309         if (unlikely(afinfo == NULL))
1310                 return ERR_PTR(-EINVAL);
1311         x = afinfo->find_bundle(fl, policy);
1312         xfrm_policy_put_afinfo(afinfo);
1313         return x;
1314 }
1315
1316 static inline int xfrm_get_tos(struct flowi *fl, int family)
1317 {
1318         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1319         int tos;
1320
1321         if (!afinfo)
1322                 return -EINVAL;
1323
1324         tos = afinfo->get_tos(fl);
1325
1326         xfrm_policy_put_afinfo(afinfo);
1327
1328         return tos;
1329 }
1330
1331 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1332 {
1333         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1334         struct xfrm_dst *xdst;
1335
1336         if (!afinfo)
1337                 return ERR_PTR(-EINVAL);
1338
1339         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1340
1341         xfrm_policy_put_afinfo(afinfo);
1342
1343         return xdst;
1344 }
1345
1346 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1347                                  int nfheader_len)
1348 {
1349         struct xfrm_policy_afinfo *afinfo =
1350                 xfrm_policy_get_afinfo(dst->ops->family);
1351         int err;
1352
1353         if (!afinfo)
1354                 return -EINVAL;
1355
1356         err = afinfo->init_path(path, dst, nfheader_len);
1357
1358         xfrm_policy_put_afinfo(afinfo);
1359
1360         return err;
1361 }
1362
1363 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1364 {
1365         struct xfrm_policy_afinfo *afinfo =
1366                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1367         int err;
1368
1369         if (!afinfo)
1370                 return -EINVAL;
1371
1372         err = afinfo->fill_dst(xdst, dev);
1373
1374         xfrm_policy_put_afinfo(afinfo);
1375
1376         return err;
1377 }
1378
1379 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1380  * all the metrics... Shortly, bundle a bundle.
1381  */
1382
1383 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1384                                             struct xfrm_state **xfrm, int nx,
1385                                             struct flowi *fl,
1386                                             struct dst_entry *dst)
1387 {
1388         unsigned long now = jiffies;
1389         struct net_device *dev;
1390         struct dst_entry *dst_prev = NULL;
1391         struct dst_entry *dst0 = NULL;
1392         int i = 0;
1393         int err;
1394         int header_len = 0;
1395         int nfheader_len = 0;
1396         int trailer_len = 0;
1397         int tos;
1398         int family = policy->selector.family;
1399         xfrm_address_t saddr, daddr;
1400
1401         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1402
1403         tos = xfrm_get_tos(fl, family);
1404         err = tos;
1405         if (tos < 0)
1406                 goto put_states;
1407
1408         dst_hold(dst);
1409
1410         for (; i < nx; i++) {
1411                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1412                 struct dst_entry *dst1 = &xdst->u.dst;
1413
1414                 err = PTR_ERR(xdst);
1415                 if (IS_ERR(xdst)) {
1416                         dst_release(dst);
1417                         goto put_states;
1418                 }
1419
1420                 if (!dst_prev)
1421                         dst0 = dst1;
1422                 else {
1423                         dst_prev->child = dst_clone(dst1);
1424                         dst1->flags |= DST_NOHASH;
1425                 }
1426
1427                 xdst->route = dst;
1428                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1429
1430                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1431                         family = xfrm[i]->props.family;
1432                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1433                                               family);
1434                         err = PTR_ERR(dst);
1435                         if (IS_ERR(dst))
1436                                 goto put_states;
1437                 } else
1438                         dst_hold(dst);
1439
1440                 dst1->xfrm = xfrm[i];
1441                 xdst->genid = xfrm[i]->genid;
1442
1443                 dst1->obsolete = -1;
1444                 dst1->flags |= DST_HOST;
1445                 dst1->lastuse = now;
1446
1447                 dst1->input = dst_discard;
1448                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1449
1450                 dst1->next = dst_prev;
1451                 dst_prev = dst1;
1452
1453                 header_len += xfrm[i]->props.header_len;
1454                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1455                         nfheader_len += xfrm[i]->props.header_len;
1456                 trailer_len += xfrm[i]->props.trailer_len;
1457         }
1458
1459         dst_prev->child = dst;
1460         dst0->path = dst;
1461
1462         err = -ENODEV;
1463         dev = dst->dev;
1464         if (!dev)
1465                 goto free_dst;
1466
1467         /* Copy neighbout for reachability confirmation */
1468         dst0->neighbour = neigh_clone(dst->neighbour);
1469
1470         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1471         xfrm_init_pmtu(dst_prev);
1472
1473         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1474                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1475
1476                 err = xfrm_fill_dst(xdst, dev);
1477                 if (err)
1478                         goto free_dst;
1479
1480                 dst_prev->header_len = header_len;
1481                 dst_prev->trailer_len = trailer_len;
1482                 header_len -= xdst->u.dst.xfrm->props.header_len;
1483                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1484         }
1485
1486 out:
1487         return dst0;
1488
1489 put_states:
1490         for (; i < nx; i++)
1491                 xfrm_state_put(xfrm[i]);
1492 free_dst:
1493         if (dst0)
1494                 dst_free(dst0);
1495         dst0 = ERR_PTR(err);
1496         goto out;
1497 }
1498
1499 static int inline
1500 xfrm_dst_alloc_copy(void **target, void *src, int size)
1501 {
1502         if (!*target) {
1503                 *target = kmalloc(size, GFP_ATOMIC);
1504                 if (!*target)
1505                         return -ENOMEM;
1506         }
1507         memcpy(*target, src, size);
1508         return 0;
1509 }
1510
1511 static int inline
1512 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1513 {
1514 #ifdef CONFIG_XFRM_SUB_POLICY
1515         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1516         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1517                                    sel, sizeof(*sel));
1518 #else
1519         return 0;
1520 #endif
1521 }
1522
1523 static int inline
1524 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1525 {
1526 #ifdef CONFIG_XFRM_SUB_POLICY
1527         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1528         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1529 #else
1530         return 0;
1531 #endif
1532 }
1533
1534 static int stale_bundle(struct dst_entry *dst);
1535
1536 /* Main function: finds/creates a bundle for given flow.
1537  *
1538  * At the moment we eat a raw IP route. Mostly to speed up lookups
1539  * on interfaces with disabled IPsec.
1540  */
1541 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1542                   struct sock *sk, int flags)
1543 {
1544         struct xfrm_policy *policy;
1545         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1546         int npols;
1547         int pol_dead;
1548         int xfrm_nr;
1549         int pi;
1550         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1551         struct dst_entry *dst, *dst_orig = *dst_p;
1552         int nx = 0;
1553         int err;
1554         u32 genid;
1555         u16 family;
1556         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1557
1558 restart:
1559         genid = atomic_read(&flow_cache_genid);
1560         policy = NULL;
1561         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1562                 pols[pi] = NULL;
1563         npols = 0;
1564         pol_dead = 0;
1565         xfrm_nr = 0;
1566
1567         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1568                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
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                 /* To accelerate a bit...  */
1578                 if ((dst_orig->flags & DST_NOXFRM) ||
1579                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1580                         goto nopol;
1581
1582                 policy = flow_cache_lookup(net, fl, dst_orig->ops->family,
1583                                            dir, xfrm_policy_lookup);
1584                 err = PTR_ERR(policy);
1585                 if (IS_ERR(policy)) {
1586                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1587                         goto dropdst;
1588                 }
1589         }
1590
1591         if (!policy)
1592                 goto nopol;
1593
1594         family = dst_orig->ops->family;
1595         pols[0] = policy;
1596         npols ++;
1597         xfrm_nr += pols[0]->xfrm_nr;
1598
1599         err = -ENOENT;
1600         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1601                 goto error;
1602
1603         policy->curlft.use_time = get_seconds();
1604
1605         switch (policy->action) {
1606         default:
1607         case XFRM_POLICY_BLOCK:
1608                 /* Prohibit the flow */
1609                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1610                 err = -EPERM;
1611                 goto error;
1612
1613         case XFRM_POLICY_ALLOW:
1614 #ifndef CONFIG_XFRM_SUB_POLICY
1615                 if (policy->xfrm_nr == 0) {
1616                         /* Flow passes not transformed. */
1617                         xfrm_pol_put(policy);
1618                         return 0;
1619                 }
1620 #endif
1621
1622                 /* Try to find matching bundle.
1623                  *
1624                  * LATER: help from flow cache. It is optional, this
1625                  * is required only for output policy.
1626                  */
1627                 dst = xfrm_find_bundle(fl, policy, family);
1628                 if (IS_ERR(dst)) {
1629                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1630                         err = PTR_ERR(dst);
1631                         goto error;
1632                 }
1633
1634                 if (dst)
1635                         break;
1636
1637 #ifdef CONFIG_XFRM_SUB_POLICY
1638                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1639                         pols[1] = xfrm_policy_lookup_bytype(net,
1640                                                             XFRM_POLICY_TYPE_MAIN,
1641                                                             fl, family,
1642                                                             XFRM_POLICY_OUT);
1643                         if (pols[1]) {
1644                                 if (IS_ERR(pols[1])) {
1645                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1646                                         err = PTR_ERR(pols[1]);
1647                                         goto error;
1648                                 }
1649                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1650                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1651                                         err = -EPERM;
1652                                         goto error;
1653                                 }
1654                                 npols ++;
1655                                 xfrm_nr += pols[1]->xfrm_nr;
1656                         }
1657                 }
1658
1659                 /*
1660                  * Because neither flowi nor bundle information knows about
1661                  * transformation template size. On more than one policy usage
1662                  * we can realize whether all of them is bypass or not after
1663                  * they are searched. See above not-transformed bypass
1664                  * is surrounded by non-sub policy configuration, too.
1665                  */
1666                 if (xfrm_nr == 0) {
1667                         /* Flow passes not transformed. */
1668                         xfrm_pols_put(pols, npols);
1669                         return 0;
1670                 }
1671
1672 #endif
1673                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1674
1675                 if (unlikely(nx<0)) {
1676                         err = nx;
1677                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1678                                 /* EREMOTE tells the caller to generate
1679                                  * a one-shot blackhole route.
1680                                  */
1681                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1682                                 xfrm_pol_put(policy);
1683                                 return -EREMOTE;
1684                         }
1685                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1686                                 DECLARE_WAITQUEUE(wait, current);
1687
1688                                 add_wait_queue(&net->xfrm.km_waitq, &wait);
1689                                 set_current_state(TASK_INTERRUPTIBLE);
1690                                 schedule();
1691                                 set_current_state(TASK_RUNNING);
1692                                 remove_wait_queue(&net->xfrm.km_waitq, &wait);
1693
1694                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1695
1696                                 if (nx == -EAGAIN && signal_pending(current)) {
1697                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1698                                         err = -ERESTART;
1699                                         goto error;
1700                                 }
1701                                 if (nx == -EAGAIN ||
1702                                     genid != atomic_read(&flow_cache_genid)) {
1703                                         xfrm_pols_put(pols, npols);
1704                                         goto restart;
1705                                 }
1706                                 err = nx;
1707                         }
1708                         if (err < 0) {
1709                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1710                                 goto error;
1711                         }
1712                 }
1713                 if (nx == 0) {
1714                         /* Flow passes not transformed. */
1715                         xfrm_pols_put(pols, npols);
1716                         return 0;
1717                 }
1718
1719                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1720                 err = PTR_ERR(dst);
1721                 if (IS_ERR(dst)) {
1722                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1723                         goto error;
1724                 }
1725
1726                 for (pi = 0; pi < npols; pi++) {
1727                         read_lock_bh(&pols[pi]->lock);
1728                         pol_dead |= pols[pi]->walk.dead;
1729                         read_unlock_bh(&pols[pi]->lock);
1730                 }
1731
1732                 write_lock_bh(&policy->lock);
1733                 if (unlikely(pol_dead || stale_bundle(dst))) {
1734                         /* Wow! While we worked on resolving, this
1735                          * policy has gone. Retry. It is not paranoia,
1736                          * we just cannot enlist new bundle to dead object.
1737                          * We can't enlist stable bundles either.
1738                          */
1739                         write_unlock_bh(&policy->lock);
1740                         dst_free(dst);
1741
1742                         if (pol_dead)
1743                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1744                         else
1745                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1746                         err = -EHOSTUNREACH;
1747                         goto error;
1748                 }
1749
1750                 if (npols > 1)
1751                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1752                 else
1753                         err = xfrm_dst_update_origin(dst, fl);
1754                 if (unlikely(err)) {
1755                         write_unlock_bh(&policy->lock);
1756                         dst_free(dst);
1757                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1758                         goto error;
1759                 }
1760
1761                 dst->next = policy->bundles;
1762                 policy->bundles = dst;
1763                 dst_hold(dst);
1764                 write_unlock_bh(&policy->lock);
1765         }
1766         *dst_p = dst;
1767         dst_release(dst_orig);
1768         xfrm_pols_put(pols, npols);
1769         return 0;
1770
1771 error:
1772         xfrm_pols_put(pols, npols);
1773 dropdst:
1774         dst_release(dst_orig);
1775         *dst_p = NULL;
1776         return err;
1777
1778 nopol:
1779         err = -ENOENT;
1780         if (flags & XFRM_LOOKUP_ICMP)
1781                 goto dropdst;
1782         return 0;
1783 }
1784 EXPORT_SYMBOL(__xfrm_lookup);
1785
1786 int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1787                 struct sock *sk, int flags)
1788 {
1789         int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1790
1791         if (err == -EREMOTE) {
1792                 dst_release(*dst_p);
1793                 *dst_p = NULL;
1794                 err = -EAGAIN;
1795         }
1796
1797         return err;
1798 }
1799 EXPORT_SYMBOL(xfrm_lookup);
1800
1801 static inline int
1802 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1803 {
1804         struct xfrm_state *x;
1805
1806         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1807                 return 0;
1808         x = skb->sp->xvec[idx];
1809         if (!x->type->reject)
1810                 return 0;
1811         return x->type->reject(x, skb, fl);
1812 }
1813
1814 /* When skb is transformed back to its "native" form, we have to
1815  * check policy restrictions. At the moment we make this in maximally
1816  * stupid way. Shame on me. :-) Of course, connected sockets must
1817  * have policy cached at them.
1818  */
1819
1820 static inline int
1821 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1822               unsigned short family)
1823 {
1824         if (xfrm_state_kern(x))
1825                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1826         return  x->id.proto == tmpl->id.proto &&
1827                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1828                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1829                 x->props.mode == tmpl->mode &&
1830                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1831                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1832                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1833                   xfrm_state_addr_cmp(tmpl, x, family));
1834 }
1835
1836 /*
1837  * 0 or more than 0 is returned when validation is succeeded (either bypass
1838  * because of optional transport mode, or next index of the mathced secpath
1839  * state with the template.
1840  * -1 is returned when no matching template is found.
1841  * Otherwise "-2 - errored_index" is returned.
1842  */
1843 static inline int
1844 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1845                unsigned short family)
1846 {
1847         int idx = start;
1848
1849         if (tmpl->optional) {
1850                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1851                         return start;
1852         } else
1853                 start = -1;
1854         for (; idx < sp->len; idx++) {
1855                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1856                         return ++idx;
1857                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1858                         if (start == -1)
1859                                 start = -2-idx;
1860                         break;
1861                 }
1862         }
1863         return start;
1864 }
1865
1866 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1867                           unsigned int family, int reverse)
1868 {
1869         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1870         int err;
1871
1872         if (unlikely(afinfo == NULL))
1873                 return -EAFNOSUPPORT;
1874
1875         afinfo->decode_session(skb, fl, reverse);
1876         err = security_xfrm_decode_session(skb, &fl->secid);
1877         xfrm_policy_put_afinfo(afinfo);
1878         return err;
1879 }
1880 EXPORT_SYMBOL(__xfrm_decode_session);
1881
1882 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1883 {
1884         for (; k < sp->len; k++) {
1885                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1886                         *idxp = k;
1887                         return 1;
1888                 }
1889         }
1890
1891         return 0;
1892 }
1893
1894 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1895                         unsigned short family)
1896 {
1897         struct net *net = dev_net(skb->dev);
1898         struct xfrm_policy *pol;
1899         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1900         int npols = 0;
1901         int xfrm_nr;
1902         int pi;
1903         int reverse;
1904         struct flowi fl;
1905         u8 fl_dir;
1906         int xerr_idx = -1;
1907
1908         reverse = dir & ~XFRM_POLICY_MASK;
1909         dir &= XFRM_POLICY_MASK;
1910         fl_dir = policy_to_flow_dir(dir);
1911
1912         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1913                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1914                 return 0;
1915         }
1916
1917         nf_nat_decode_session(skb, &fl, family);
1918
1919         /* First, check used SA against their selectors. */
1920         if (skb->sp) {
1921                 int i;
1922
1923                 for (i=skb->sp->len-1; i>=0; i--) {
1924                         struct xfrm_state *x = skb->sp->xvec[i];
1925                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1926                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1927                                 return 0;
1928                         }
1929                 }
1930         }
1931
1932         pol = NULL;
1933         if (sk && sk->sk_policy[dir]) {
1934                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1935                 if (IS_ERR(pol)) {
1936                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1937                         return 0;
1938                 }
1939         }
1940
1941         if (!pol)
1942                 pol = flow_cache_lookup(net, &fl, family, fl_dir,
1943                                         xfrm_policy_lookup);
1944
1945         if (IS_ERR(pol)) {
1946                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1947                 return 0;
1948         }
1949
1950         if (!pol) {
1951                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1952                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1953                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1954                         return 0;
1955                 }
1956                 return 1;
1957         }
1958
1959         pol->curlft.use_time = get_seconds();
1960
1961         pols[0] = pol;
1962         npols ++;
1963 #ifdef CONFIG_XFRM_SUB_POLICY
1964         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1965                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
1966                                                     &fl, family,
1967                                                     XFRM_POLICY_IN);
1968                 if (pols[1]) {
1969                         if (IS_ERR(pols[1])) {
1970                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1971                                 return 0;
1972                         }
1973                         pols[1]->curlft.use_time = get_seconds();
1974                         npols ++;
1975                 }
1976         }
1977 #endif
1978
1979         if (pol->action == XFRM_POLICY_ALLOW) {
1980                 struct sec_path *sp;
1981                 static struct sec_path dummy;
1982                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1983                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1984                 struct xfrm_tmpl **tpp = tp;
1985                 int ti = 0;
1986                 int i, k;
1987
1988                 if ((sp = skb->sp) == NULL)
1989                         sp = &dummy;
1990
1991                 for (pi = 0; pi < npols; pi++) {
1992                         if (pols[pi] != pol &&
1993                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1994                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1995                                 goto reject;
1996                         }
1997                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1998                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1999                                 goto reject_error;
2000                         }
2001                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2002                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2003                 }
2004                 xfrm_nr = ti;
2005                 if (npols > 1) {
2006                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2007                         tpp = stp;
2008                 }
2009
2010                 /* For each tunnel xfrm, find the first matching tmpl.
2011                  * For each tmpl before that, find corresponding xfrm.
2012                  * Order is _important_. Later we will implement
2013                  * some barriers, but at the moment barriers
2014                  * are implied between each two transformations.
2015                  */
2016                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2017                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2018                         if (k < 0) {
2019                                 if (k < -1)
2020                                         /* "-2 - errored_index" returned */
2021                                         xerr_idx = -(2+k);
2022                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2023                                 goto reject;
2024                         }
2025                 }
2026
2027                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2028                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2029                         goto reject;
2030                 }
2031
2032                 xfrm_pols_put(pols, npols);
2033                 return 1;
2034         }
2035         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2036
2037 reject:
2038         xfrm_secpath_reject(xerr_idx, skb, &fl);
2039 reject_error:
2040         xfrm_pols_put(pols, npols);
2041         return 0;
2042 }
2043 EXPORT_SYMBOL(__xfrm_policy_check);
2044
2045 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2046 {
2047         struct flowi fl;
2048
2049         if (xfrm_decode_session(skb, &fl, family) < 0) {
2050                 /* XXX: we should have something like FWDHDRERROR here. */
2051                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2052                 return 0;
2053         }
2054
2055         return xfrm_lookup(&init_net, &skb->dst, &fl, NULL, 0) == 0;
2056 }
2057 EXPORT_SYMBOL(__xfrm_route_forward);
2058
2059 /* Optimize later using cookies and generation ids. */
2060
2061 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2062 {
2063         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2064          * to "-1" to force all XFRM destinations to get validated by
2065          * dst_ops->check on every use.  We do this because when a
2066          * normal route referenced by an XFRM dst is obsoleted we do
2067          * not go looking around for all parent referencing XFRM dsts
2068          * so that we can invalidate them.  It is just too much work.
2069          * Instead we make the checks here on every use.  For example:
2070          *
2071          *      XFRM dst A --> IPv4 dst X
2072          *
2073          * X is the "xdst->route" of A (X is also the "dst->path" of A
2074          * in this example).  If X is marked obsolete, "A" will not
2075          * notice.  That's what we are validating here via the
2076          * stale_bundle() check.
2077          *
2078          * When a policy's bundle is pruned, we dst_free() the XFRM
2079          * dst which causes it's ->obsolete field to be set to a
2080          * positive non-zero integer.  If an XFRM dst has been pruned
2081          * like this, we want to force a new route lookup.
2082          */
2083         if (dst->obsolete < 0 && !stale_bundle(dst))
2084                 return dst;
2085
2086         return NULL;
2087 }
2088
2089 static int stale_bundle(struct dst_entry *dst)
2090 {
2091         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2092 }
2093
2094 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2095 {
2096         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2097                 dst->dev = dev_net(dev)->loopback_dev;
2098                 dev_hold(dst->dev);
2099                 dev_put(dev);
2100         }
2101 }
2102 EXPORT_SYMBOL(xfrm_dst_ifdown);
2103
2104 static void xfrm_link_failure(struct sk_buff *skb)
2105 {
2106         /* Impossible. Such dst must be popped before reaches point of failure. */
2107         return;
2108 }
2109
2110 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2111 {
2112         if (dst) {
2113                 if (dst->obsolete) {
2114                         dst_release(dst);
2115                         dst = NULL;
2116                 }
2117         }
2118         return dst;
2119 }
2120
2121 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2122 {
2123         struct dst_entry *dst, **dstp;
2124
2125         write_lock(&pol->lock);
2126         dstp = &pol->bundles;
2127         while ((dst=*dstp) != NULL) {
2128                 if (func(dst)) {
2129                         *dstp = dst->next;
2130                         dst->next = *gc_list_p;
2131                         *gc_list_p = dst;
2132                 } else {
2133                         dstp = &dst->next;
2134                 }
2135         }
2136         write_unlock(&pol->lock);
2137 }
2138
2139 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2140 {
2141         struct dst_entry *gc_list = NULL;
2142         int dir;
2143
2144         read_lock_bh(&xfrm_policy_lock);
2145         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2146                 struct xfrm_policy *pol;
2147                 struct hlist_node *entry;
2148                 struct hlist_head *table;
2149                 int i;
2150
2151                 hlist_for_each_entry(pol, entry,
2152                                      &init_net.xfrm.policy_inexact[dir], bydst)
2153                         prune_one_bundle(pol, func, &gc_list);
2154
2155                 table = init_net.xfrm.policy_bydst[dir].table;
2156                 for (i = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2157                         hlist_for_each_entry(pol, entry, table + i, bydst)
2158                                 prune_one_bundle(pol, func, &gc_list);
2159                 }
2160         }
2161         read_unlock_bh(&xfrm_policy_lock);
2162
2163         while (gc_list) {
2164                 struct dst_entry *dst = gc_list;
2165                 gc_list = dst->next;
2166                 dst_free(dst);
2167         }
2168 }
2169
2170 static int unused_bundle(struct dst_entry *dst)
2171 {
2172         return !atomic_read(&dst->__refcnt);
2173 }
2174
2175 static void __xfrm_garbage_collect(void)
2176 {
2177         xfrm_prune_bundles(unused_bundle);
2178 }
2179
2180 static int xfrm_flush_bundles(void)
2181 {
2182         xfrm_prune_bundles(stale_bundle);
2183         return 0;
2184 }
2185
2186 static void xfrm_init_pmtu(struct dst_entry *dst)
2187 {
2188         do {
2189                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2190                 u32 pmtu, route_mtu_cached;
2191
2192                 pmtu = dst_mtu(dst->child);
2193                 xdst->child_mtu_cached = pmtu;
2194
2195                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2196
2197                 route_mtu_cached = dst_mtu(xdst->route);
2198                 xdst->route_mtu_cached = route_mtu_cached;
2199
2200                 if (pmtu > route_mtu_cached)
2201                         pmtu = route_mtu_cached;
2202
2203                 dst->metrics[RTAX_MTU-1] = pmtu;
2204         } while ((dst = dst->next));
2205 }
2206
2207 /* Check that the bundle accepts the flow and its components are
2208  * still valid.
2209  */
2210
2211 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2212                 struct flowi *fl, int family, int strict)
2213 {
2214         struct dst_entry *dst = &first->u.dst;
2215         struct xfrm_dst *last;
2216         u32 mtu;
2217
2218         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2219             (dst->dev && !netif_running(dst->dev)))
2220                 return 0;
2221 #ifdef CONFIG_XFRM_SUB_POLICY
2222         if (fl) {
2223                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2224                         return 0;
2225                 if (first->partner &&
2226                     !xfrm_selector_match(first->partner, fl, family))
2227                         return 0;
2228         }
2229 #endif
2230
2231         last = NULL;
2232
2233         do {
2234                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2235
2236                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2237                         return 0;
2238                 if (fl && pol &&
2239                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2240                         return 0;
2241                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2242                         return 0;
2243                 if (xdst->genid != dst->xfrm->genid)
2244                         return 0;
2245
2246                 if (strict && fl &&
2247                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2248                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2249                         return 0;
2250
2251                 mtu = dst_mtu(dst->child);
2252                 if (xdst->child_mtu_cached != mtu) {
2253                         last = xdst;
2254                         xdst->child_mtu_cached = mtu;
2255                 }
2256
2257                 if (!dst_check(xdst->route, xdst->route_cookie))
2258                         return 0;
2259                 mtu = dst_mtu(xdst->route);
2260                 if (xdst->route_mtu_cached != mtu) {
2261                         last = xdst;
2262                         xdst->route_mtu_cached = mtu;
2263                 }
2264
2265                 dst = dst->child;
2266         } while (dst->xfrm);
2267
2268         if (likely(!last))
2269                 return 1;
2270
2271         mtu = last->child_mtu_cached;
2272         for (;;) {
2273                 dst = &last->u.dst;
2274
2275                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2276                 if (mtu > last->route_mtu_cached)
2277                         mtu = last->route_mtu_cached;
2278                 dst->metrics[RTAX_MTU-1] = mtu;
2279
2280                 if (last == first)
2281                         break;
2282
2283                 last = (struct xfrm_dst *)last->u.dst.next;
2284                 last->child_mtu_cached = mtu;
2285         }
2286
2287         return 1;
2288 }
2289
2290 EXPORT_SYMBOL(xfrm_bundle_ok);
2291
2292 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2293 {
2294         int err = 0;
2295         if (unlikely(afinfo == NULL))
2296                 return -EINVAL;
2297         if (unlikely(afinfo->family >= NPROTO))
2298                 return -EAFNOSUPPORT;
2299         write_lock_bh(&xfrm_policy_afinfo_lock);
2300         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2301                 err = -ENOBUFS;
2302         else {
2303                 struct dst_ops *dst_ops = afinfo->dst_ops;
2304                 if (likely(dst_ops->kmem_cachep == NULL))
2305                         dst_ops->kmem_cachep = xfrm_dst_cache;
2306                 if (likely(dst_ops->check == NULL))
2307                         dst_ops->check = xfrm_dst_check;
2308                 if (likely(dst_ops->negative_advice == NULL))
2309                         dst_ops->negative_advice = xfrm_negative_advice;
2310                 if (likely(dst_ops->link_failure == NULL))
2311                         dst_ops->link_failure = xfrm_link_failure;
2312                 if (likely(afinfo->garbage_collect == NULL))
2313                         afinfo->garbage_collect = __xfrm_garbage_collect;
2314                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2315         }
2316         write_unlock_bh(&xfrm_policy_afinfo_lock);
2317         return err;
2318 }
2319 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2320
2321 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2322 {
2323         int err = 0;
2324         if (unlikely(afinfo == NULL))
2325                 return -EINVAL;
2326         if (unlikely(afinfo->family >= NPROTO))
2327                 return -EAFNOSUPPORT;
2328         write_lock_bh(&xfrm_policy_afinfo_lock);
2329         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2330                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2331                         err = -EINVAL;
2332                 else {
2333                         struct dst_ops *dst_ops = afinfo->dst_ops;
2334                         xfrm_policy_afinfo[afinfo->family] = NULL;
2335                         dst_ops->kmem_cachep = NULL;
2336                         dst_ops->check = NULL;
2337                         dst_ops->negative_advice = NULL;
2338                         dst_ops->link_failure = NULL;
2339                         afinfo->garbage_collect = NULL;
2340                 }
2341         }
2342         write_unlock_bh(&xfrm_policy_afinfo_lock);
2343         return err;
2344 }
2345 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2346
2347 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2348 {
2349         struct xfrm_policy_afinfo *afinfo;
2350         if (unlikely(family >= NPROTO))
2351                 return NULL;
2352         read_lock(&xfrm_policy_afinfo_lock);
2353         afinfo = xfrm_policy_afinfo[family];
2354         if (unlikely(!afinfo))
2355                 read_unlock(&xfrm_policy_afinfo_lock);
2356         return afinfo;
2357 }
2358
2359 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2360 {
2361         read_unlock(&xfrm_policy_afinfo_lock);
2362 }
2363
2364 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2365 {
2366         struct net_device *dev = ptr;
2367
2368         if (!net_eq(dev_net(dev), &init_net))
2369                 return NOTIFY_DONE;
2370
2371         switch (event) {
2372         case NETDEV_DOWN:
2373                 xfrm_flush_bundles();
2374         }
2375         return NOTIFY_DONE;
2376 }
2377
2378 static struct notifier_block xfrm_dev_notifier = {
2379         .notifier_call  = xfrm_dev_event,
2380 };
2381
2382 #ifdef CONFIG_XFRM_STATISTICS
2383 static int __init xfrm_statistics_init(void)
2384 {
2385         if (snmp_mib_init((void **)xfrm_statistics,
2386                           sizeof(struct linux_xfrm_mib)) < 0)
2387                 return -ENOMEM;
2388         return 0;
2389 }
2390 #endif
2391
2392 static int __net_init xfrm_policy_init(struct net *net)
2393 {
2394         unsigned int hmask, sz;
2395         int dir;
2396
2397         if (net_eq(net, &init_net))
2398                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2399                                            sizeof(struct xfrm_dst),
2400                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2401                                            NULL);
2402
2403         hmask = 8 - 1;
2404         sz = (hmask+1) * sizeof(struct hlist_head);
2405
2406         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2407         if (!net->xfrm.policy_byidx)
2408                 goto out_byidx;
2409         net->xfrm.policy_idx_hmask = hmask;
2410
2411         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2412                 struct xfrm_policy_hash *htab;
2413
2414                 net->xfrm.policy_count[dir] = 0;
2415                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2416
2417                 htab = &net->xfrm.policy_bydst[dir];
2418                 htab->table = xfrm_hash_alloc(sz);
2419                 if (!htab->table)
2420                         goto out_bydst;
2421                 htab->hmask = hmask;
2422         }
2423
2424         INIT_LIST_HEAD(&net->xfrm.policy_all);
2425         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2426         if (net_eq(net, &init_net))
2427                 register_netdevice_notifier(&xfrm_dev_notifier);
2428         return 0;
2429
2430 out_bydst:
2431         for (dir--; dir >= 0; dir--) {
2432                 struct xfrm_policy_hash *htab;
2433
2434                 htab = &net->xfrm.policy_bydst[dir];
2435                 xfrm_hash_free(htab->table, sz);
2436         }
2437         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2438 out_byidx:
2439         return -ENOMEM;
2440 }
2441
2442 static void xfrm_policy_fini(struct net *net)
2443 {
2444         unsigned int sz;
2445         int dir;
2446
2447         WARN_ON(!list_empty(&net->xfrm.policy_all));
2448
2449         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2450                 struct xfrm_policy_hash *htab;
2451
2452                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2453
2454                 htab = &net->xfrm.policy_bydst[dir];
2455                 sz = (htab->hmask + 1);
2456                 WARN_ON(!hlist_empty(htab->table));
2457                 xfrm_hash_free(htab->table, sz);
2458         }
2459
2460         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2461         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2462         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2463 }
2464
2465 static int __net_init xfrm_net_init(struct net *net)
2466 {
2467         int rv;
2468
2469         rv = xfrm_state_init(net);
2470         if (rv < 0)
2471                 goto out_state;
2472         rv = xfrm_policy_init(net);
2473         if (rv < 0)
2474                 goto out_policy;
2475         return 0;
2476
2477 out_policy:
2478         xfrm_state_fini(net);
2479 out_state:
2480         return rv;
2481 }
2482
2483 static void __net_exit xfrm_net_exit(struct net *net)
2484 {
2485         xfrm_policy_fini(net);
2486         xfrm_state_fini(net);
2487 }
2488
2489 static struct pernet_operations __net_initdata xfrm_net_ops = {
2490         .init = xfrm_net_init,
2491         .exit = xfrm_net_exit,
2492 };
2493
2494 void __init xfrm_init(void)
2495 {
2496         register_pernet_subsys(&xfrm_net_ops);
2497 #ifdef CONFIG_XFRM_STATISTICS
2498         xfrm_statistics_init();
2499 #endif
2500         xfrm_input_init();
2501 #ifdef CONFIG_XFRM_STATISTICS
2502         xfrm_proc_init();
2503 #endif
2504 }
2505
2506 #ifdef CONFIG_AUDITSYSCALL
2507 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2508                                          struct audit_buffer *audit_buf)
2509 {
2510         struct xfrm_sec_ctx *ctx = xp->security;
2511         struct xfrm_selector *sel = &xp->selector;
2512
2513         if (ctx)
2514                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2515                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2516
2517         switch(sel->family) {
2518         case AF_INET:
2519                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2520                 if (sel->prefixlen_s != 32)
2521                         audit_log_format(audit_buf, " src_prefixlen=%d",
2522                                          sel->prefixlen_s);
2523                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2524                 if (sel->prefixlen_d != 32)
2525                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2526                                          sel->prefixlen_d);
2527                 break;
2528         case AF_INET6:
2529                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2530                 if (sel->prefixlen_s != 128)
2531                         audit_log_format(audit_buf, " src_prefixlen=%d",
2532                                          sel->prefixlen_s);
2533                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2534                 if (sel->prefixlen_d != 128)
2535                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2536                                          sel->prefixlen_d);
2537                 break;
2538         }
2539 }
2540
2541 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2542                            uid_t auid, u32 sessionid, u32 secid)
2543 {
2544         struct audit_buffer *audit_buf;
2545
2546         audit_buf = xfrm_audit_start("SPD-add");
2547         if (audit_buf == NULL)
2548                 return;
2549         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2550         audit_log_format(audit_buf, " res=%u", result);
2551         xfrm_audit_common_policyinfo(xp, audit_buf);
2552         audit_log_end(audit_buf);
2553 }
2554 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2555
2556 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2557                               uid_t auid, u32 sessionid, u32 secid)
2558 {
2559         struct audit_buffer *audit_buf;
2560
2561         audit_buf = xfrm_audit_start("SPD-delete");
2562         if (audit_buf == NULL)
2563                 return;
2564         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2565         audit_log_format(audit_buf, " res=%u", result);
2566         xfrm_audit_common_policyinfo(xp, audit_buf);
2567         audit_log_end(audit_buf);
2568 }
2569 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2570 #endif
2571
2572 #ifdef CONFIG_XFRM_MIGRATE
2573 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2574                                        struct xfrm_selector *sel_tgt)
2575 {
2576         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2577                 if (sel_tgt->family == sel_cmp->family &&
2578                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2579                                   sel_cmp->family) == 0 &&
2580                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2581                                   sel_cmp->family) == 0 &&
2582                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2583                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2584                         return 1;
2585                 }
2586         } else {
2587                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2588                         return 1;
2589                 }
2590         }
2591         return 0;
2592 }
2593
2594 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2595                                                      u8 dir, u8 type)
2596 {
2597         struct xfrm_policy *pol, *ret = NULL;
2598         struct hlist_node *entry;
2599         struct hlist_head *chain;
2600         u32 priority = ~0U;
2601
2602         read_lock_bh(&xfrm_policy_lock);
2603         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2604         hlist_for_each_entry(pol, entry, chain, bydst) {
2605                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2606                     pol->type == type) {
2607                         ret = pol;
2608                         priority = ret->priority;
2609                         break;
2610                 }
2611         }
2612         chain = &init_net.xfrm.policy_inexact[dir];
2613         hlist_for_each_entry(pol, entry, chain, bydst) {
2614                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2615                     pol->type == type &&
2616                     pol->priority < priority) {
2617                         ret = pol;
2618                         break;
2619                 }
2620         }
2621
2622         if (ret)
2623                 xfrm_pol_hold(ret);
2624
2625         read_unlock_bh(&xfrm_policy_lock);
2626
2627         return ret;
2628 }
2629
2630 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2631 {
2632         int match = 0;
2633
2634         if (t->mode == m->mode && t->id.proto == m->proto &&
2635             (m->reqid == 0 || t->reqid == m->reqid)) {
2636                 switch (t->mode) {
2637                 case XFRM_MODE_TUNNEL:
2638                 case XFRM_MODE_BEET:
2639                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2640                                           m->old_family) == 0 &&
2641                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2642                                           m->old_family) == 0) {
2643                                 match = 1;
2644                         }
2645                         break;
2646                 case XFRM_MODE_TRANSPORT:
2647                         /* in case of transport mode, template does not store
2648                            any IP addresses, hence we just compare mode and
2649                            protocol */
2650                         match = 1;
2651                         break;
2652                 default:
2653                         break;
2654                 }
2655         }
2656         return match;
2657 }
2658
2659 /* update endpoint address(es) of template(s) */
2660 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2661                                struct xfrm_migrate *m, int num_migrate)
2662 {
2663         struct xfrm_migrate *mp;
2664         struct dst_entry *dst;
2665         int i, j, n = 0;
2666
2667         write_lock_bh(&pol->lock);
2668         if (unlikely(pol->walk.dead)) {
2669                 /* target policy has been deleted */
2670                 write_unlock_bh(&pol->lock);
2671                 return -ENOENT;
2672         }
2673
2674         for (i = 0; i < pol->xfrm_nr; i++) {
2675                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2676                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2677                                 continue;
2678                         n++;
2679                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2680                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2681                                 continue;
2682                         /* update endpoints */
2683                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2684                                sizeof(pol->xfrm_vec[i].id.daddr));
2685                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2686                                sizeof(pol->xfrm_vec[i].saddr));
2687                         pol->xfrm_vec[i].encap_family = mp->new_family;
2688                         /* flush bundles */
2689                         while ((dst = pol->bundles) != NULL) {
2690                                 pol->bundles = dst->next;
2691                                 dst_free(dst);
2692                         }
2693                 }
2694         }
2695
2696         write_unlock_bh(&pol->lock);
2697
2698         if (!n)
2699                 return -ENODATA;
2700
2701         return 0;
2702 }
2703
2704 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2705 {
2706         int i, j;
2707
2708         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2709                 return -EINVAL;
2710
2711         for (i = 0; i < num_migrate; i++) {
2712                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2713                                    m[i].old_family) == 0) &&
2714                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2715                                    m[i].old_family) == 0))
2716                         return -EINVAL;
2717                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2718                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2719                         return -EINVAL;
2720
2721                 /* check if there is any duplicated entry */
2722                 for (j = i + 1; j < num_migrate; j++) {
2723                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2724                                     sizeof(m[i].old_daddr)) &&
2725                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2726                                     sizeof(m[i].old_saddr)) &&
2727                             m[i].proto == m[j].proto &&
2728                             m[i].mode == m[j].mode &&
2729                             m[i].reqid == m[j].reqid &&
2730                             m[i].old_family == m[j].old_family)
2731                                 return -EINVAL;
2732                 }
2733         }
2734
2735         return 0;
2736 }
2737
2738 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2739                  struct xfrm_migrate *m, int num_migrate,
2740                  struct xfrm_kmaddress *k)
2741 {
2742         int i, err, nx_cur = 0, nx_new = 0;
2743         struct xfrm_policy *pol = NULL;
2744         struct xfrm_state *x, *xc;
2745         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2746         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2747         struct xfrm_migrate *mp;
2748
2749         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2750                 goto out;
2751
2752         /* Stage 1 - find policy */
2753         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2754                 err = -ENOENT;
2755                 goto out;
2756         }
2757
2758         /* Stage 2 - find and update state(s) */
2759         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2760                 if ((x = xfrm_migrate_state_find(mp))) {
2761                         x_cur[nx_cur] = x;
2762                         nx_cur++;
2763                         if ((xc = xfrm_state_migrate(x, mp))) {
2764                                 x_new[nx_new] = xc;
2765                                 nx_new++;
2766                         } else {
2767                                 err = -ENODATA;
2768                                 goto restore_state;
2769                         }
2770                 }
2771         }
2772
2773         /* Stage 3 - update policy */
2774         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2775                 goto restore_state;
2776
2777         /* Stage 4 - delete old state(s) */
2778         if (nx_cur) {
2779                 xfrm_states_put(x_cur, nx_cur);
2780                 xfrm_states_delete(x_cur, nx_cur);
2781         }
2782
2783         /* Stage 5 - announce */
2784         km_migrate(sel, dir, type, m, num_migrate, k);
2785
2786         xfrm_pol_put(pol);
2787
2788         return 0;
2789 out:
2790         return err;
2791
2792 restore_state:
2793         if (pol)
2794                 xfrm_pol_put(pol);
2795         if (nx_cur)
2796                 xfrm_states_put(x_cur, nx_cur);
2797         if (nx_new)
2798                 xfrm_states_delete(x_new, nx_new);
2799
2800         return err;
2801 }
2802 EXPORT_SYMBOL(xfrm_migrate);
2803 #endif