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(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(&init_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                                 init_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(u8 type, int dir, u32 id, int delete,
 690                                      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 = init_net.xfrm.policy_byidx + idx_hash(&init_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                                 init_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(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                                      &init_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 = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 758                         hlist_for_each_entry(pol, entry,
 759                                              init_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(u8 type, struct xfrm_audit *audit_info)
 780 {
 781         return 0;
 782 }
 783 #endif
 784
 785 int xfrm_policy_flush(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(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                                      &init_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 = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 822         again2:
 823                         hlist_for_each_entry(pol, entry,
 824                                              init_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                 init_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 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(&init_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, &init_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(u8 type, struct flowi *fl,
 944                                                      u16 family, u8 dir)
 945 {
 946         int err;
 947         struct xfrm_policy *pol, *ret;
 948         xfrm_address_t *daddr, *saddr;
 949         struct hlist_node *entry;
 950         struct hlist_head *chain;
 951         u32 priority = ~0U;
 952
 953         daddr = xfrm_flowi_daddr(fl, family);
 954         saddr = xfrm_flowi_saddr(fl, family);
 955         if (unlikely(!daddr || !saddr))
 956                 return NULL;
 957
 958         read_lock_bh(&xfrm_policy_lock);
 959         chain = policy_hash_direct(&init_net, daddr, saddr, family, dir);
 960         ret = NULL;
 961         hlist_for_each_entry(pol, entry, chain, bydst) {
 962                 err = xfrm_policy_match(pol, fl, type, family, dir);
 963                 if (err) {
 964                         if (err == -ESRCH)
 965                                 continue;
 966                         else {
 967                                 ret = ERR_PTR(err);
 968                                 goto fail;
 969                         }
 970                 } else {
 971                         ret = pol;
 972                         priority = ret->priority;
 973                         break;
 974                 }
 975         }
 976         chain = &init_net.xfrm.policy_inexact[dir];
 977         hlist_for_each_entry(pol, entry, chain, bydst) {
 978                 err = xfrm_policy_match(pol, fl, type, family, dir);
 979                 if (err) {
 980                         if (err == -ESRCH)
 981                                 continue;
 982                         else {
 983                                 ret = ERR_PTR(err);
 984                                 goto fail;
 985                         }
 986                 } else if (pol->priority < priority) {
 987                         ret = pol;
 988                         break;
 989                 }
 990         }
 991         if (ret)
 992                 xfrm_pol_hold(ret);
 993 fail:
 994         read_unlock_bh(&xfrm_policy_lock);
 995
 996         return ret;
 997 }
 998
 999 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1000                                void **objp, atomic_t **obj_refp)
1001 {
1002         struct xfrm_policy *pol;
1003         int err = 0;
1004
1005 #ifdef CONFIG_XFRM_SUB_POLICY
1006         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1007         if (IS_ERR(pol)) {
1008                 err = PTR_ERR(pol);
1009                 pol = NULL;
1010         }
1011         if (pol || err)
1012                 goto end;
1013 #endif
1014         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1015         if (IS_ERR(pol)) {
1016                 err = PTR_ERR(pol);
1017                 pol = NULL;
1018         }
1019 #ifdef CONFIG_XFRM_SUB_POLICY
1020 end:
1021 #endif
1022         if ((*objp = (void *) pol) != NULL)
1023                 *obj_refp = &pol->refcnt;
1024         return err;
1025 }
1026
1027 static inline int policy_to_flow_dir(int dir)
1028 {
1029         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1030             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1031             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1032                 return dir;
1033         switch (dir) {
1034         default:
1035         case XFRM_POLICY_IN:
1036                 return FLOW_DIR_IN;
1037         case XFRM_POLICY_OUT:
1038                 return FLOW_DIR_OUT;
1039         case XFRM_POLICY_FWD:
1040                 return FLOW_DIR_FWD;
1041         }
1042 }
1043
1044 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1045 {
1046         struct xfrm_policy *pol;
1047
1048         read_lock_bh(&xfrm_policy_lock);
1049         if ((pol = sk->sk_policy[dir]) != NULL) {
1050                 int match = xfrm_selector_match(&pol->selector, fl,
1051                                                 sk->sk_family);
1052                 int err = 0;
1053
1054                 if (match) {
1055                         err = security_xfrm_policy_lookup(pol->security,
1056                                                       fl->secid,
1057                                                       policy_to_flow_dir(dir));
1058                         if (!err)
1059                                 xfrm_pol_hold(pol);
1060                         else if (err == -ESRCH)
1061                                 pol = NULL;
1062                         else
1063                                 pol = ERR_PTR(err);
1064                 } else
1065                         pol = NULL;
1066         }
1067         read_unlock_bh(&xfrm_policy_lock);
1068         return pol;
1069 }
1070
1071 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1072 {
1073         struct net *net = xp_net(pol);
1074         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1075                                                      pol->family, dir);
1076
1077         list_add(&pol->walk.all, &net->xfrm.policy_all);
1078         hlist_add_head(&pol->bydst, chain);
1079         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1080         net->xfrm.policy_count[dir]++;
1081         xfrm_pol_hold(pol);
1082
1083         if (xfrm_bydst_should_resize(net, dir, NULL))
1084                 schedule_work(&net->xfrm.policy_hash_work);
1085 }
1086
1087 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1088                                                 int dir)
1089 {
1090         struct net *net = xp_net(pol);
1091
1092         if (hlist_unhashed(&pol->bydst))
1093                 return NULL;
1094
1095         hlist_del(&pol->bydst);
1096         hlist_del(&pol->byidx);
1097         list_del(&pol->walk.all);
1098         net->xfrm.policy_count[dir]--;
1099
1100         return pol;
1101 }
1102
1103 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1104 {
1105         write_lock_bh(&xfrm_policy_lock);
1106         pol = __xfrm_policy_unlink(pol, dir);
1107         write_unlock_bh(&xfrm_policy_lock);
1108         if (pol) {
1109                 if (dir < XFRM_POLICY_MAX)
1110                         atomic_inc(&flow_cache_genid);
1111                 xfrm_policy_kill(pol);
1112                 return 0;
1113         }
1114         return -ENOENT;
1115 }
1116 EXPORT_SYMBOL(xfrm_policy_delete);
1117
1118 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1119 {
1120         struct net *net = xp_net(pol);
1121         struct xfrm_policy *old_pol;
1122
1123 #ifdef CONFIG_XFRM_SUB_POLICY
1124         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1125                 return -EINVAL;
1126 #endif
1127
1128         write_lock_bh(&xfrm_policy_lock);
1129         old_pol = sk->sk_policy[dir];
1130         sk->sk_policy[dir] = pol;
1131         if (pol) {
1132                 pol->curlft.add_time = get_seconds();
1133                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1134                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1135         }
1136         if (old_pol)
1137                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1138         write_unlock_bh(&xfrm_policy_lock);
1139
1140         if (old_pol) {
1141                 xfrm_policy_kill(old_pol);
1142         }
1143         return 0;
1144 }
1145
1146 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1147 {
1148         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1149
1150         if (newp) {
1151                 newp->selector = old->selector;
1152                 if (security_xfrm_policy_clone(old->security,
1153                                                &newp->security)) {
1154                         kfree(newp);
1155                         return NULL;  /* ENOMEM */
1156                 }
1157                 newp->lft = old->lft;
1158                 newp->curlft = old->curlft;
1159                 newp->action = old->action;
1160                 newp->flags = old->flags;
1161                 newp->xfrm_nr = old->xfrm_nr;
1162                 newp->index = old->index;
1163                 newp->type = old->type;
1164                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1165                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1166                 write_lock_bh(&xfrm_policy_lock);
1167                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1168                 write_unlock_bh(&xfrm_policy_lock);
1169                 xfrm_pol_put(newp);
1170         }
1171         return newp;
1172 }
1173
1174 int __xfrm_sk_clone_policy(struct sock *sk)
1175 {
1176         struct xfrm_policy *p0 = sk->sk_policy[0],
1177                            *p1 = sk->sk_policy[1];
1178
1179         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1180         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1181                 return -ENOMEM;
1182         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1183                 return -ENOMEM;
1184         return 0;
1185 }
1186
1187 static int
1188 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1189                unsigned short family)
1190 {
1191         int err;
1192         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1193
1194         if (unlikely(afinfo == NULL))
1195                 return -EINVAL;
1196         err = afinfo->get_saddr(local, remote);
1197         xfrm_policy_put_afinfo(afinfo);
1198         return err;
1199 }
1200
1201 /* Resolve list of templates for the flow, given policy. */
1202
1203 static int
1204 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1205                       struct xfrm_state **xfrm,
1206                       unsigned short family)
1207 {
1208         int nx;
1209         int i, error;
1210         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1211         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1212         xfrm_address_t tmp;
1213
1214         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1215                 struct xfrm_state *x;
1216                 xfrm_address_t *remote = daddr;
1217                 xfrm_address_t *local  = saddr;
1218                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1219
1220                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1221                     tmpl->mode == XFRM_MODE_BEET) {
1222                         remote = &tmpl->id.daddr;
1223                         local = &tmpl->saddr;
1224                         family = tmpl->encap_family;
1225                         if (xfrm_addr_any(local, family)) {
1226                                 error = xfrm_get_saddr(&tmp, remote, family);
1227                                 if (error)
1228                                         goto fail;
1229                                 local = &tmp;
1230                         }
1231                 }
1232
1233                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1234
1235                 if (x && x->km.state == XFRM_STATE_VALID) {
1236                         xfrm[nx++] = x;
1237                         daddr = remote;
1238                         saddr = local;
1239                         continue;
1240                 }
1241                 if (x) {
1242                         error = (x->km.state == XFRM_STATE_ERROR ?
1243                                  -EINVAL : -EAGAIN);
1244                         xfrm_state_put(x);
1245                 }
1246                 else if (error == -ESRCH)
1247                         error = -EAGAIN;
1248
1249                 if (!tmpl->optional)
1250                         goto fail;
1251         }
1252         return nx;
1253
1254 fail:
1255         for (nx--; nx>=0; nx--)
1256                 xfrm_state_put(xfrm[nx]);
1257         return error;
1258 }
1259
1260 static int
1261 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1262                   struct xfrm_state **xfrm,
1263                   unsigned short family)
1264 {
1265         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1266         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1267         int cnx = 0;
1268         int error;
1269         int ret;
1270         int i;
1271
1272         for (i = 0; i < npols; i++) {
1273                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1274                         error = -ENOBUFS;
1275                         goto fail;
1276                 }
1277
1278                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1279                 if (ret < 0) {
1280                         error = ret;
1281                         goto fail;
1282                 } else
1283                         cnx += ret;
1284         }
1285
1286         /* found states are sorted for outbound processing */
1287         if (npols > 1)
1288                 xfrm_state_sort(xfrm, tpp, cnx, family);
1289
1290         return cnx;
1291
1292  fail:
1293         for (cnx--; cnx>=0; cnx--)
1294                 xfrm_state_put(tpp[cnx]);
1295         return error;
1296
1297 }
1298
1299 /* Check that the bundle accepts the flow and its components are
1300  * still valid.
1301  */
1302
1303 static struct dst_entry *
1304 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1305 {
1306         struct dst_entry *x;
1307         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1308         if (unlikely(afinfo == NULL))
1309                 return ERR_PTR(-EINVAL);
1310         x = afinfo->find_bundle(fl, policy);
1311         xfrm_policy_put_afinfo(afinfo);
1312         return x;
1313 }
1314
1315 static inline int xfrm_get_tos(struct flowi *fl, int family)
1316 {
1317         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1318         int tos;
1319
1320         if (!afinfo)
1321                 return -EINVAL;
1322
1323         tos = afinfo->get_tos(fl);
1324
1325         xfrm_policy_put_afinfo(afinfo);
1326
1327         return tos;
1328 }
1329
1330 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1331 {
1332         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1333         struct xfrm_dst *xdst;
1334
1335         if (!afinfo)
1336                 return ERR_PTR(-EINVAL);
1337
1338         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1339
1340         xfrm_policy_put_afinfo(afinfo);
1341
1342         return xdst;
1343 }
1344
1345 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1346                                  int nfheader_len)
1347 {
1348         struct xfrm_policy_afinfo *afinfo =
1349                 xfrm_policy_get_afinfo(dst->ops->family);
1350         int err;
1351
1352         if (!afinfo)
1353                 return -EINVAL;
1354
1355         err = afinfo->init_path(path, dst, nfheader_len);
1356
1357         xfrm_policy_put_afinfo(afinfo);
1358
1359         return err;
1360 }
1361
1362 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1363 {
1364         struct xfrm_policy_afinfo *afinfo =
1365                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1366         int err;
1367
1368         if (!afinfo)
1369                 return -EINVAL;
1370
1371         err = afinfo->fill_dst(xdst, dev);
1372
1373         xfrm_policy_put_afinfo(afinfo);
1374
1375         return err;
1376 }
1377
1378 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1379  * all the metrics... Shortly, bundle a bundle.
1380  */
1381
1382 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1383                                             struct xfrm_state **xfrm, int nx,
1384                                             struct flowi *fl,
1385                                             struct dst_entry *dst)
1386 {
1387         unsigned long now = jiffies;
1388         struct net_device *dev;
1389         struct dst_entry *dst_prev = NULL;
1390         struct dst_entry *dst0 = NULL;
1391         int i = 0;
1392         int err;
1393         int header_len = 0;
1394         int nfheader_len = 0;
1395         int trailer_len = 0;
1396         int tos;
1397         int family = policy->selector.family;
1398         xfrm_address_t saddr, daddr;
1399
1400         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1401
1402         tos = xfrm_get_tos(fl, family);
1403         err = tos;
1404         if (tos < 0)
1405                 goto put_states;
1406
1407         dst_hold(dst);
1408
1409         for (; i < nx; i++) {
1410                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1411                 struct dst_entry *dst1 = &xdst->u.dst;
1412
1413                 err = PTR_ERR(xdst);
1414                 if (IS_ERR(xdst)) {
1415                         dst_release(dst);
1416                         goto put_states;
1417                 }
1418
1419                 if (!dst_prev)
1420                         dst0 = dst1;
1421                 else {
1422                         dst_prev->child = dst_clone(dst1);
1423                         dst1->flags |= DST_NOHASH;
1424                 }
1425
1426                 xdst->route = dst;
1427                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1428
1429                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1430                         family = xfrm[i]->props.family;
1431                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1432                                               family);
1433                         err = PTR_ERR(dst);
1434                         if (IS_ERR(dst))
1435                                 goto put_states;
1436                 } else
1437                         dst_hold(dst);
1438
1439                 dst1->xfrm = xfrm[i];
1440                 xdst->genid = xfrm[i]->genid;
1441
1442                 dst1->obsolete = -1;
1443                 dst1->flags |= DST_HOST;
1444                 dst1->lastuse = now;
1445
1446                 dst1->input = dst_discard;
1447                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1448
1449                 dst1->next = dst_prev;
1450                 dst_prev = dst1;
1451
1452                 header_len += xfrm[i]->props.header_len;
1453                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1454                         nfheader_len += xfrm[i]->props.header_len;
1455                 trailer_len += xfrm[i]->props.trailer_len;
1456         }
1457
1458         dst_prev->child = dst;
1459         dst0->path = dst;
1460
1461         err = -ENODEV;
1462         dev = dst->dev;
1463         if (!dev)
1464                 goto free_dst;
1465
1466         /* Copy neighbout for reachability confirmation */
1467         dst0->neighbour = neigh_clone(dst->neighbour);
1468
1469         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1470         xfrm_init_pmtu(dst_prev);
1471
1472         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1473                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1474
1475                 err = xfrm_fill_dst(xdst, dev);
1476                 if (err)
1477                         goto free_dst;
1478
1479                 dst_prev->header_len = header_len;
1480                 dst_prev->trailer_len = trailer_len;
1481                 header_len -= xdst->u.dst.xfrm->props.header_len;
1482                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1483         }
1484
1485 out:
1486         return dst0;
1487
1488 put_states:
1489         for (; i < nx; i++)
1490                 xfrm_state_put(xfrm[i]);
1491 free_dst:
1492         if (dst0)
1493                 dst_free(dst0);
1494         dst0 = ERR_PTR(err);
1495         goto out;
1496 }
1497
1498 static int inline
1499 xfrm_dst_alloc_copy(void **target, void *src, int size)
1500 {
1501         if (!*target) {
1502                 *target = kmalloc(size, GFP_ATOMIC);
1503                 if (!*target)
1504                         return -ENOMEM;
1505         }
1506         memcpy(*target, src, size);
1507         return 0;
1508 }
1509
1510 static int inline
1511 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1512 {
1513 #ifdef CONFIG_XFRM_SUB_POLICY
1514         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1515         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1516                                    sel, sizeof(*sel));
1517 #else
1518         return 0;
1519 #endif
1520 }
1521
1522 static int inline
1523 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1524 {
1525 #ifdef CONFIG_XFRM_SUB_POLICY
1526         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1527         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1528 #else
1529         return 0;
1530 #endif
1531 }
1532
1533 static int stale_bundle(struct dst_entry *dst);
1534
1535 /* Main function: finds/creates a bundle for given flow.
1536  *
1537  * At the moment we eat a raw IP route. Mostly to speed up lookups
1538  * on interfaces with disabled IPsec.
1539  */
1540 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1541                   struct sock *sk, int flags)
1542 {
1543         struct xfrm_policy *policy;
1544         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1545         int npols;
1546         int pol_dead;
1547         int xfrm_nr;
1548         int pi;
1549         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1550         struct dst_entry *dst, *dst_orig = *dst_p;
1551         int nx = 0;
1552         int err;
1553         u32 genid;
1554         u16 family;
1555         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1556
1557 restart:
1558         genid = atomic_read(&flow_cache_genid);
1559         policy = NULL;
1560         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1561                 pols[pi] = NULL;
1562         npols = 0;
1563         pol_dead = 0;
1564         xfrm_nr = 0;
1565
1566         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1567                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1568                 err = PTR_ERR(policy);
1569                 if (IS_ERR(policy)) {
1570                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1571                         goto dropdst;
1572                 }
1573         }
1574
1575         if (!policy) {
1576                 /* To accelerate a bit...  */
1577                 if ((dst_orig->flags & DST_NOXFRM) ||
1578                     !init_net.xfrm.policy_count[XFRM_POLICY_OUT])
1579                         goto nopol;
1580
1581                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1582                                            dir, xfrm_policy_lookup);
1583                 err = PTR_ERR(policy);
1584                 if (IS_ERR(policy)) {
1585                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1586                         goto dropdst;
1587                 }
1588         }
1589
1590         if (!policy)
1591                 goto nopol;
1592
1593         family = dst_orig->ops->family;
1594         pols[0] = policy;
1595         npols ++;
1596         xfrm_nr += pols[0]->xfrm_nr;
1597
1598         err = -ENOENT;
1599         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1600                 goto error;
1601
1602         policy->curlft.use_time = get_seconds();
1603
1604         switch (policy->action) {
1605         default:
1606         case XFRM_POLICY_BLOCK:
1607                 /* Prohibit the flow */
1608                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1609                 err = -EPERM;
1610                 goto error;
1611
1612         case XFRM_POLICY_ALLOW:
1613 #ifndef CONFIG_XFRM_SUB_POLICY
1614                 if (policy->xfrm_nr == 0) {
1615                         /* Flow passes not transformed. */
1616                         xfrm_pol_put(policy);
1617                         return 0;
1618                 }
1619 #endif
1620
1621                 /* Try to find matching bundle.
1622                  *
1623                  * LATER: help from flow cache. It is optional, this
1624                  * is required only for output policy.
1625                  */
1626                 dst = xfrm_find_bundle(fl, policy, family);
1627                 if (IS_ERR(dst)) {
1628                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1629                         err = PTR_ERR(dst);
1630                         goto error;
1631                 }
1632
1633                 if (dst)
1634                         break;
1635
1636 #ifdef CONFIG_XFRM_SUB_POLICY
1637                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1638                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1639                                                             fl, family,
1640                                                             XFRM_POLICY_OUT);
1641                         if (pols[1]) {
1642                                 if (IS_ERR(pols[1])) {
1643                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1644                                         err = PTR_ERR(pols[1]);
1645                                         goto error;
1646                                 }
1647                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1648                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1649                                         err = -EPERM;
1650                                         goto error;
1651                                 }
1652                                 npols ++;
1653                                 xfrm_nr += pols[1]->xfrm_nr;
1654                         }
1655                 }
1656
1657                 /*
1658                  * Because neither flowi nor bundle information knows about
1659                  * transformation template size. On more than one policy usage
1660                  * we can realize whether all of them is bypass or not after
1661                  * they are searched. See above not-transformed bypass
1662                  * is surrounded by non-sub policy configuration, too.
1663                  */
1664                 if (xfrm_nr == 0) {
1665                         /* Flow passes not transformed. */
1666                         xfrm_pols_put(pols, npols);
1667                         return 0;
1668                 }
1669
1670 #endif
1671                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1672
1673                 if (unlikely(nx<0)) {
1674                         err = nx;
1675                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1676                                 /* EREMOTE tells the caller to generate
1677                                  * a one-shot blackhole route.
1678                                  */
1679                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1680                                 xfrm_pol_put(policy);
1681                                 return -EREMOTE;
1682                         }
1683                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1684                                 DECLARE_WAITQUEUE(wait, current);
1685
1686                                 add_wait_queue(&init_net.xfrm.km_waitq, &wait);
1687                                 set_current_state(TASK_INTERRUPTIBLE);
1688                                 schedule();
1689                                 set_current_state(TASK_RUNNING);
1690                                 remove_wait_queue(&init_net.xfrm.km_waitq, &wait);
1691
1692                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1693
1694                                 if (nx == -EAGAIN && signal_pending(current)) {
1695                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1696                                         err = -ERESTART;
1697                                         goto error;
1698                                 }
1699                                 if (nx == -EAGAIN ||
1700                                     genid != atomic_read(&flow_cache_genid)) {
1701                                         xfrm_pols_put(pols, npols);
1702                                         goto restart;
1703                                 }
1704                                 err = nx;
1705                         }
1706                         if (err < 0) {
1707                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1708                                 goto error;
1709                         }
1710                 }
1711                 if (nx == 0) {
1712                         /* Flow passes not transformed. */
1713                         xfrm_pols_put(pols, npols);
1714                         return 0;
1715                 }
1716
1717                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1718                 err = PTR_ERR(dst);
1719                 if (IS_ERR(dst)) {
1720                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1721                         goto error;
1722                 }
1723
1724                 for (pi = 0; pi < npols; pi++) {
1725                         read_lock_bh(&pols[pi]->lock);
1726                         pol_dead |= pols[pi]->walk.dead;
1727                         read_unlock_bh(&pols[pi]->lock);
1728                 }
1729
1730                 write_lock_bh(&policy->lock);
1731                 if (unlikely(pol_dead || stale_bundle(dst))) {
1732                         /* Wow! While we worked on resolving, this
1733                          * policy has gone. Retry. It is not paranoia,
1734                          * we just cannot enlist new bundle to dead object.
1735                          * We can't enlist stable bundles either.
1736                          */
1737                         write_unlock_bh(&policy->lock);
1738                         dst_free(dst);
1739
1740                         if (pol_dead)
1741                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1742                         else
1743                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1744                         err = -EHOSTUNREACH;
1745                         goto error;
1746                 }
1747
1748                 if (npols > 1)
1749                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1750                 else
1751                         err = xfrm_dst_update_origin(dst, fl);
1752                 if (unlikely(err)) {
1753                         write_unlock_bh(&policy->lock);
1754                         dst_free(dst);
1755                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1756                         goto error;
1757                 }
1758
1759                 dst->next = policy->bundles;
1760                 policy->bundles = dst;
1761                 dst_hold(dst);
1762                 write_unlock_bh(&policy->lock);
1763         }
1764         *dst_p = dst;
1765         dst_release(dst_orig);
1766         xfrm_pols_put(pols, npols);
1767         return 0;
1768
1769 error:
1770         xfrm_pols_put(pols, npols);
1771 dropdst:
1772         dst_release(dst_orig);
1773         *dst_p = NULL;
1774         return err;
1775
1776 nopol:
1777         err = -ENOENT;
1778         if (flags & XFRM_LOOKUP_ICMP)
1779                 goto dropdst;
1780         return 0;
1781 }
1782 EXPORT_SYMBOL(__xfrm_lookup);
1783
1784 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1785                 struct sock *sk, int flags)
1786 {
1787         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1788
1789         if (err == -EREMOTE) {
1790                 dst_release(*dst_p);
1791                 *dst_p = NULL;
1792                 err = -EAGAIN;
1793         }
1794
1795         return err;
1796 }
1797 EXPORT_SYMBOL(xfrm_lookup);
1798
1799 static inline int
1800 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1801 {
1802         struct xfrm_state *x;
1803
1804         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1805                 return 0;
1806         x = skb->sp->xvec[idx];
1807         if (!x->type->reject)
1808                 return 0;
1809         return x->type->reject(x, skb, fl);
1810 }
1811
1812 /* When skb is transformed back to its "native" form, we have to
1813  * check policy restrictions. At the moment we make this in maximally
1814  * stupid way. Shame on me. :-) Of course, connected sockets must
1815  * have policy cached at them.
1816  */
1817
1818 static inline int
1819 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1820               unsigned short family)
1821 {
1822         if (xfrm_state_kern(x))
1823                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1824         return  x->id.proto == tmpl->id.proto &&
1825                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1826                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1827                 x->props.mode == tmpl->mode &&
1828                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1829                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1830                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1831                   xfrm_state_addr_cmp(tmpl, x, family));
1832 }
1833
1834 /*
1835  * 0 or more than 0 is returned when validation is succeeded (either bypass
1836  * because of optional transport mode, or next index of the mathced secpath
1837  * state with the template.
1838  * -1 is returned when no matching template is found.
1839  * Otherwise "-2 - errored_index" is returned.
1840  */
1841 static inline int
1842 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1843                unsigned short family)
1844 {
1845         int idx = start;
1846
1847         if (tmpl->optional) {
1848                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1849                         return start;
1850         } else
1851                 start = -1;
1852         for (; idx < sp->len; idx++) {
1853                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1854                         return ++idx;
1855                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1856                         if (start == -1)
1857                                 start = -2-idx;
1858                         break;
1859                 }
1860         }
1861         return start;
1862 }
1863
1864 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1865                           unsigned int family, int reverse)
1866 {
1867         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1868         int err;
1869
1870         if (unlikely(afinfo == NULL))
1871                 return -EAFNOSUPPORT;
1872
1873         afinfo->decode_session(skb, fl, reverse);
1874         err = security_xfrm_decode_session(skb, &fl->secid);
1875         xfrm_policy_put_afinfo(afinfo);
1876         return err;
1877 }
1878 EXPORT_SYMBOL(__xfrm_decode_session);
1879
1880 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1881 {
1882         for (; k < sp->len; k++) {
1883                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1884                         *idxp = k;
1885                         return 1;
1886                 }
1887         }
1888
1889         return 0;
1890 }
1891
1892 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1893                         unsigned short family)
1894 {
1895         struct xfrm_policy *pol;
1896         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1897         int npols = 0;
1898         int xfrm_nr;
1899         int pi;
1900         int reverse;
1901         struct flowi fl;
1902         u8 fl_dir;
1903         int xerr_idx = -1;
1904
1905         reverse = dir & ~XFRM_POLICY_MASK;
1906         dir &= XFRM_POLICY_MASK;
1907         fl_dir = policy_to_flow_dir(dir);
1908
1909         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1910                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1911                 return 0;
1912         }
1913
1914         nf_nat_decode_session(skb, &fl, family);
1915
1916         /* First, check used SA against their selectors. */
1917         if (skb->sp) {
1918                 int i;
1919
1920                 for (i=skb->sp->len-1; i>=0; i--) {
1921                         struct xfrm_state *x = skb->sp->xvec[i];
1922                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1923                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1924                                 return 0;
1925                         }
1926                 }
1927         }
1928
1929         pol = NULL;
1930         if (sk && sk->sk_policy[dir]) {
1931                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1932                 if (IS_ERR(pol)) {
1933                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1934                         return 0;
1935                 }
1936         }
1937
1938         if (!pol)
1939                 pol = flow_cache_lookup(&fl, family, fl_dir,
1940                                         xfrm_policy_lookup);
1941
1942         if (IS_ERR(pol)) {
1943                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1944                 return 0;
1945         }
1946
1947         if (!pol) {
1948                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1949                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1950                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1951                         return 0;
1952                 }
1953                 return 1;
1954         }
1955
1956         pol->curlft.use_time = get_seconds();
1957
1958         pols[0] = pol;
1959         npols ++;
1960 #ifdef CONFIG_XFRM_SUB_POLICY
1961         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1962                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1963                                                     &fl, family,
1964                                                     XFRM_POLICY_IN);
1965                 if (pols[1]) {
1966                         if (IS_ERR(pols[1])) {
1967                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1968                                 return 0;
1969                         }
1970                         pols[1]->curlft.use_time = get_seconds();
1971                         npols ++;
1972                 }
1973         }
1974 #endif
1975
1976         if (pol->action == XFRM_POLICY_ALLOW) {
1977                 struct sec_path *sp;
1978                 static struct sec_path dummy;
1979                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1980                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1981                 struct xfrm_tmpl **tpp = tp;
1982                 int ti = 0;
1983                 int i, k;
1984
1985                 if ((sp = skb->sp) == NULL)
1986                         sp = &dummy;
1987
1988                 for (pi = 0; pi < npols; pi++) {
1989                         if (pols[pi] != pol &&
1990                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1991                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1992                                 goto reject;
1993                         }
1994                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1995                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1996                                 goto reject_error;
1997                         }
1998                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1999                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2000                 }
2001                 xfrm_nr = ti;
2002                 if (npols > 1) {
2003                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2004                         tpp = stp;
2005                 }
2006
2007                 /* For each tunnel xfrm, find the first matching tmpl.
2008                  * For each tmpl before that, find corresponding xfrm.
2009                  * Order is _important_. Later we will implement
2010                  * some barriers, but at the moment barriers
2011                  * are implied between each two transformations.
2012                  */
2013                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2014                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2015                         if (k < 0) {
2016                                 if (k < -1)
2017                                         /* "-2 - errored_index" returned */
2018                                         xerr_idx = -(2+k);
2019                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2020                                 goto reject;
2021                         }
2022                 }
2023
2024                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2025                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2026                         goto reject;
2027                 }
2028
2029                 xfrm_pols_put(pols, npols);
2030                 return 1;
2031         }
2032         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2033
2034 reject:
2035         xfrm_secpath_reject(xerr_idx, skb, &fl);
2036 reject_error:
2037         xfrm_pols_put(pols, npols);
2038         return 0;
2039 }
2040 EXPORT_SYMBOL(__xfrm_policy_check);
2041
2042 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2043 {
2044         struct flowi fl;
2045
2046         if (xfrm_decode_session(skb, &fl, family) < 0) {
2047                 /* XXX: we should have something like FWDHDRERROR here. */
2048                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2049                 return 0;
2050         }
2051
2052         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2053 }
2054 EXPORT_SYMBOL(__xfrm_route_forward);
2055
2056 /* Optimize later using cookies and generation ids. */
2057
2058 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2059 {
2060         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2061          * to "-1" to force all XFRM destinations to get validated by
2062          * dst_ops->check on every use.  We do this because when a
2063          * normal route referenced by an XFRM dst is obsoleted we do
2064          * not go looking around for all parent referencing XFRM dsts
2065          * so that we can invalidate them.  It is just too much work.
2066          * Instead we make the checks here on every use.  For example:
2067          *
2068          *      XFRM dst A --> IPv4 dst X
2069          *
2070          * X is the "xdst->route" of A (X is also the "dst->path" of A
2071          * in this example).  If X is marked obsolete, "A" will not
2072          * notice.  That's what we are validating here via the
2073          * stale_bundle() check.
2074          *
2075          * When a policy's bundle is pruned, we dst_free() the XFRM
2076          * dst which causes it's ->obsolete field to be set to a
2077          * positive non-zero integer.  If an XFRM dst has been pruned
2078          * like this, we want to force a new route lookup.
2079          */
2080         if (dst->obsolete < 0 && !stale_bundle(dst))
2081                 return dst;
2082
2083         return NULL;
2084 }
2085
2086 static int stale_bundle(struct dst_entry *dst)
2087 {
2088         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2089 }
2090
2091 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2092 {
2093         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2094                 dst->dev = dev_net(dev)->loopback_dev;
2095                 dev_hold(dst->dev);
2096                 dev_put(dev);
2097         }
2098 }
2099 EXPORT_SYMBOL(xfrm_dst_ifdown);
2100
2101 static void xfrm_link_failure(struct sk_buff *skb)
2102 {
2103         /* Impossible. Such dst must be popped before reaches point of failure. */
2104         return;
2105 }
2106
2107 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2108 {
2109         if (dst) {
2110                 if (dst->obsolete) {
2111                         dst_release(dst);
2112                         dst = NULL;
2113                 }
2114         }
2115         return dst;
2116 }
2117
2118 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2119 {
2120         struct dst_entry *dst, **dstp;
2121
2122         write_lock(&pol->lock);
2123         dstp = &pol->bundles;
2124         while ((dst=*dstp) != NULL) {
2125                 if (func(dst)) {
2126                         *dstp = dst->next;
2127                         dst->next = *gc_list_p;
2128                         *gc_list_p = dst;
2129                 } else {
2130                         dstp = &dst->next;
2131                 }
2132         }
2133         write_unlock(&pol->lock);
2134 }
2135
2136 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2137 {
2138         struct dst_entry *gc_list = NULL;
2139         int dir;
2140
2141         read_lock_bh(&xfrm_policy_lock);
2142         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2143                 struct xfrm_policy *pol;
2144                 struct hlist_node *entry;
2145                 struct hlist_head *table;
2146                 int i;
2147
2148                 hlist_for_each_entry(pol, entry,
2149                                      &init_net.xfrm.policy_inexact[dir], bydst)
2150                         prune_one_bundle(pol, func, &gc_list);
2151
2152                 table = init_net.xfrm.policy_bydst[dir].table;
2153                 for (i = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2154                         hlist_for_each_entry(pol, entry, table + i, bydst)
2155                                 prune_one_bundle(pol, func, &gc_list);
2156                 }
2157         }
2158         read_unlock_bh(&xfrm_policy_lock);
2159
2160         while (gc_list) {
2161                 struct dst_entry *dst = gc_list;
2162                 gc_list = dst->next;
2163                 dst_free(dst);
2164         }
2165 }
2166
2167 static int unused_bundle(struct dst_entry *dst)
2168 {
2169         return !atomic_read(&dst->__refcnt);
2170 }
2171
2172 static void __xfrm_garbage_collect(void)
2173 {
2174         xfrm_prune_bundles(unused_bundle);
2175 }
2176
2177 static int xfrm_flush_bundles(void)
2178 {
2179         xfrm_prune_bundles(stale_bundle);
2180         return 0;
2181 }
2182
2183 static void xfrm_init_pmtu(struct dst_entry *dst)
2184 {
2185         do {
2186                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2187                 u32 pmtu, route_mtu_cached;
2188
2189                 pmtu = dst_mtu(dst->child);
2190                 xdst->child_mtu_cached = pmtu;
2191
2192                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2193
2194                 route_mtu_cached = dst_mtu(xdst->route);
2195                 xdst->route_mtu_cached = route_mtu_cached;
2196
2197                 if (pmtu > route_mtu_cached)
2198                         pmtu = route_mtu_cached;
2199
2200                 dst->metrics[RTAX_MTU-1] = pmtu;
2201         } while ((dst = dst->next));
2202 }
2203
2204 /* Check that the bundle accepts the flow and its components are
2205  * still valid.
2206  */
2207
2208 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2209                 struct flowi *fl, int family, int strict)
2210 {
2211         struct dst_entry *dst = &first->u.dst;
2212         struct xfrm_dst *last;
2213         u32 mtu;
2214
2215         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2216             (dst->dev && !netif_running(dst->dev)))
2217                 return 0;
2218 #ifdef CONFIG_XFRM_SUB_POLICY
2219         if (fl) {
2220                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2221                         return 0;
2222                 if (first->partner &&
2223                     !xfrm_selector_match(first->partner, fl, family))
2224                         return 0;
2225         }
2226 #endif
2227
2228         last = NULL;
2229
2230         do {
2231                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2232
2233                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2234                         return 0;
2235                 if (fl && pol &&
2236                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2237                         return 0;
2238                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2239                         return 0;
2240                 if (xdst->genid != dst->xfrm->genid)
2241                         return 0;
2242
2243                 if (strict && fl &&
2244                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2245                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2246                         return 0;
2247
2248                 mtu = dst_mtu(dst->child);
2249                 if (xdst->child_mtu_cached != mtu) {
2250                         last = xdst;
2251                         xdst->child_mtu_cached = mtu;
2252                 }
2253
2254                 if (!dst_check(xdst->route, xdst->route_cookie))
2255                         return 0;
2256                 mtu = dst_mtu(xdst->route);
2257                 if (xdst->route_mtu_cached != mtu) {
2258                         last = xdst;
2259                         xdst->route_mtu_cached = mtu;
2260                 }
2261
2262                 dst = dst->child;
2263         } while (dst->xfrm);
2264
2265         if (likely(!last))
2266                 return 1;
2267
2268         mtu = last->child_mtu_cached;
2269         for (;;) {
2270                 dst = &last->u.dst;
2271
2272                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2273                 if (mtu > last->route_mtu_cached)
2274                         mtu = last->route_mtu_cached;
2275                 dst->metrics[RTAX_MTU-1] = mtu;
2276
2277                 if (last == first)
2278                         break;
2279
2280                 last = (struct xfrm_dst *)last->u.dst.next;
2281                 last->child_mtu_cached = mtu;
2282         }
2283
2284         return 1;
2285 }
2286
2287 EXPORT_SYMBOL(xfrm_bundle_ok);
2288
2289 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2290 {
2291         int err = 0;
2292         if (unlikely(afinfo == NULL))
2293                 return -EINVAL;
2294         if (unlikely(afinfo->family >= NPROTO))
2295                 return -EAFNOSUPPORT;
2296         write_lock_bh(&xfrm_policy_afinfo_lock);
2297         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2298                 err = -ENOBUFS;
2299         else {
2300                 struct dst_ops *dst_ops = afinfo->dst_ops;
2301                 if (likely(dst_ops->kmem_cachep == NULL))
2302                         dst_ops->kmem_cachep = xfrm_dst_cache;
2303                 if (likely(dst_ops->check == NULL))
2304                         dst_ops->check = xfrm_dst_check;
2305                 if (likely(dst_ops->negative_advice == NULL))
2306                         dst_ops->negative_advice = xfrm_negative_advice;
2307                 if (likely(dst_ops->link_failure == NULL))
2308                         dst_ops->link_failure = xfrm_link_failure;
2309                 if (likely(afinfo->garbage_collect == NULL))
2310                         afinfo->garbage_collect = __xfrm_garbage_collect;
2311                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2312         }
2313         write_unlock_bh(&xfrm_policy_afinfo_lock);
2314         return err;
2315 }
2316 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2317
2318 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2319 {
2320         int err = 0;
2321         if (unlikely(afinfo == NULL))
2322                 return -EINVAL;
2323         if (unlikely(afinfo->family >= NPROTO))
2324                 return -EAFNOSUPPORT;
2325         write_lock_bh(&xfrm_policy_afinfo_lock);
2326         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2327                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2328                         err = -EINVAL;
2329                 else {
2330                         struct dst_ops *dst_ops = afinfo->dst_ops;
2331                         xfrm_policy_afinfo[afinfo->family] = NULL;
2332                         dst_ops->kmem_cachep = NULL;
2333                         dst_ops->check = NULL;
2334                         dst_ops->negative_advice = NULL;
2335                         dst_ops->link_failure = NULL;
2336                         afinfo->garbage_collect = NULL;
2337                 }
2338         }
2339         write_unlock_bh(&xfrm_policy_afinfo_lock);
2340         return err;
2341 }
2342 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2343
2344 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2345 {
2346         struct xfrm_policy_afinfo *afinfo;
2347         if (unlikely(family >= NPROTO))
2348                 return NULL;
2349         read_lock(&xfrm_policy_afinfo_lock);
2350         afinfo = xfrm_policy_afinfo[family];
2351         if (unlikely(!afinfo))
2352                 read_unlock(&xfrm_policy_afinfo_lock);
2353         return afinfo;
2354 }
2355
2356 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2357 {
2358         read_unlock(&xfrm_policy_afinfo_lock);
2359 }
2360
2361 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2362 {
2363         struct net_device *dev = ptr;
2364
2365         if (!net_eq(dev_net(dev), &init_net))
2366                 return NOTIFY_DONE;
2367
2368         switch (event) {
2369         case NETDEV_DOWN:
2370                 xfrm_flush_bundles();
2371         }
2372         return NOTIFY_DONE;
2373 }
2374
2375 static struct notifier_block xfrm_dev_notifier = {
2376         .notifier_call  = xfrm_dev_event,
2377 };
2378
2379 #ifdef CONFIG_XFRM_STATISTICS
2380 static int __init xfrm_statistics_init(void)
2381 {
2382         if (snmp_mib_init((void **)xfrm_statistics,
2383                           sizeof(struct linux_xfrm_mib)) < 0)
2384                 return -ENOMEM;
2385         return 0;
2386 }
2387 #endif
2388
2389 static int __net_init xfrm_policy_init(struct net *net)
2390 {
2391         unsigned int hmask, sz;
2392         int dir;
2393
2394         if (net_eq(net, &init_net))
2395                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2396                                            sizeof(struct xfrm_dst),
2397                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2398                                            NULL);
2399
2400         hmask = 8 - 1;
2401         sz = (hmask+1) * sizeof(struct hlist_head);
2402
2403         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2404         if (!net->xfrm.policy_byidx)
2405                 goto out_byidx;
2406         net->xfrm.policy_idx_hmask = hmask;
2407
2408         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2409                 struct xfrm_policy_hash *htab;
2410
2411                 net->xfrm.policy_count[dir] = 0;
2412                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2413
2414                 htab = &net->xfrm.policy_bydst[dir];
2415                 htab->table = xfrm_hash_alloc(sz);
2416                 if (!htab->table)
2417                         goto out_bydst;
2418                 htab->hmask = hmask;
2419         }
2420
2421         INIT_LIST_HEAD(&net->xfrm.policy_all);
2422         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2423         if (net_eq(net, &init_net))
2424                 register_netdevice_notifier(&xfrm_dev_notifier);
2425         return 0;
2426
2427 out_bydst:
2428         for (dir--; dir >= 0; dir--) {
2429                 struct xfrm_policy_hash *htab;
2430
2431                 htab = &net->xfrm.policy_bydst[dir];
2432                 xfrm_hash_free(htab->table, sz);
2433         }
2434         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2435 out_byidx:
2436         return -ENOMEM;
2437 }
2438
2439 static void xfrm_policy_fini(struct net *net)
2440 {
2441         unsigned int sz;
2442         int dir;
2443
2444         WARN_ON(!list_empty(&net->xfrm.policy_all));
2445
2446         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2447                 struct xfrm_policy_hash *htab;
2448
2449                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2450
2451                 htab = &net->xfrm.policy_bydst[dir];
2452                 sz = (htab->hmask + 1);
2453                 WARN_ON(!hlist_empty(htab->table));
2454                 xfrm_hash_free(htab->table, sz);
2455         }
2456
2457         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2458         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2459         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2460 }
2461
2462 static int __net_init xfrm_net_init(struct net *net)
2463 {
2464         int rv;
2465
2466         rv = xfrm_state_init(net);
2467         if (rv < 0)
2468                 goto out_state;
2469         rv = xfrm_policy_init(net);
2470         if (rv < 0)
2471                 goto out_policy;
2472         return 0;
2473
2474 out_policy:
2475         xfrm_state_fini(net);
2476 out_state:
2477         return rv;
2478 }
2479
2480 static void __net_exit xfrm_net_exit(struct net *net)
2481 {
2482         xfrm_policy_fini(net);
2483         xfrm_state_fini(net);
2484 }
2485
2486 static struct pernet_operations __net_initdata xfrm_net_ops = {
2487         .init = xfrm_net_init,
2488         .exit = xfrm_net_exit,
2489 };
2490
2491 void __init xfrm_init(void)
2492 {
2493         register_pernet_subsys(&xfrm_net_ops);
2494 #ifdef CONFIG_XFRM_STATISTICS
2495         xfrm_statistics_init();
2496 #endif
2497         xfrm_input_init();
2498 #ifdef CONFIG_XFRM_STATISTICS
2499         xfrm_proc_init();
2500 #endif
2501 }
2502
2503 #ifdef CONFIG_AUDITSYSCALL
2504 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2505                                          struct audit_buffer *audit_buf)
2506 {
2507         struct xfrm_sec_ctx *ctx = xp->security;
2508         struct xfrm_selector *sel = &xp->selector;
2509
2510         if (ctx)
2511                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2512                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2513
2514         switch(sel->family) {
2515         case AF_INET:
2516                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2517                 if (sel->prefixlen_s != 32)
2518                         audit_log_format(audit_buf, " src_prefixlen=%d",
2519                                          sel->prefixlen_s);
2520                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2521                 if (sel->prefixlen_d != 32)
2522                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2523                                          sel->prefixlen_d);
2524                 break;
2525         case AF_INET6:
2526                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2527                 if (sel->prefixlen_s != 128)
2528                         audit_log_format(audit_buf, " src_prefixlen=%d",
2529                                          sel->prefixlen_s);
2530                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2531                 if (sel->prefixlen_d != 128)
2532                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2533                                          sel->prefixlen_d);
2534                 break;
2535         }
2536 }
2537
2538 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2539                            uid_t auid, u32 sessionid, u32 secid)
2540 {
2541         struct audit_buffer *audit_buf;
2542
2543         audit_buf = xfrm_audit_start("SPD-add");
2544         if (audit_buf == NULL)
2545                 return;
2546         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2547         audit_log_format(audit_buf, " res=%u", result);
2548         xfrm_audit_common_policyinfo(xp, audit_buf);
2549         audit_log_end(audit_buf);
2550 }
2551 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2552
2553 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2554                               uid_t auid, u32 sessionid, u32 secid)
2555 {
2556         struct audit_buffer *audit_buf;
2557
2558         audit_buf = xfrm_audit_start("SPD-delete");
2559         if (audit_buf == NULL)
2560                 return;
2561         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2562         audit_log_format(audit_buf, " res=%u", result);
2563         xfrm_audit_common_policyinfo(xp, audit_buf);
2564         audit_log_end(audit_buf);
2565 }
2566 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2567 #endif
2568
2569 #ifdef CONFIG_XFRM_MIGRATE
2570 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2571                                        struct xfrm_selector *sel_tgt)
2572 {
2573         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2574                 if (sel_tgt->family == sel_cmp->family &&
2575                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2576                                   sel_cmp->family) == 0 &&
2577                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2578                                   sel_cmp->family) == 0 &&
2579                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2580                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2581                         return 1;
2582                 }
2583         } else {
2584                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2585                         return 1;
2586                 }
2587         }
2588         return 0;
2589 }
2590
2591 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2592                                                      u8 dir, u8 type)
2593 {
2594         struct xfrm_policy *pol, *ret = NULL;
2595         struct hlist_node *entry;
2596         struct hlist_head *chain;
2597         u32 priority = ~0U;
2598
2599         read_lock_bh(&xfrm_policy_lock);
2600         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2601         hlist_for_each_entry(pol, entry, chain, bydst) {
2602                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2603                     pol->type == type) {
2604                         ret = pol;
2605                         priority = ret->priority;
2606                         break;
2607                 }
2608         }
2609         chain = &init_net.xfrm.policy_inexact[dir];
2610         hlist_for_each_entry(pol, entry, chain, bydst) {
2611                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2612                     pol->type == type &&
2613                     pol->priority < priority) {
2614                         ret = pol;
2615                         break;
2616                 }
2617         }
2618
2619         if (ret)
2620                 xfrm_pol_hold(ret);
2621
2622         read_unlock_bh(&xfrm_policy_lock);
2623
2624         return ret;
2625 }
2626
2627 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2628 {
2629         int match = 0;
2630
2631         if (t->mode == m->mode && t->id.proto == m->proto &&
2632             (m->reqid == 0 || t->reqid == m->reqid)) {
2633                 switch (t->mode) {
2634                 case XFRM_MODE_TUNNEL:
2635                 case XFRM_MODE_BEET:
2636                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2637                                           m->old_family) == 0 &&
2638                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2639                                           m->old_family) == 0) {
2640                                 match = 1;
2641                         }
2642                         break;
2643                 case XFRM_MODE_TRANSPORT:
2644                         /* in case of transport mode, template does not store
2645                            any IP addresses, hence we just compare mode and
2646                            protocol */
2647                         match = 1;
2648                         break;
2649                 default:
2650                         break;
2651                 }
2652         }
2653         return match;
2654 }
2655
2656 /* update endpoint address(es) of template(s) */
2657 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2658                                struct xfrm_migrate *m, int num_migrate)
2659 {
2660         struct xfrm_migrate *mp;
2661         struct dst_entry *dst;
2662         int i, j, n = 0;
2663
2664         write_lock_bh(&pol->lock);
2665         if (unlikely(pol->walk.dead)) {
2666                 /* target policy has been deleted */
2667                 write_unlock_bh(&pol->lock);
2668                 return -ENOENT;
2669         }
2670
2671         for (i = 0; i < pol->xfrm_nr; i++) {
2672                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2673                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2674                                 continue;
2675                         n++;
2676                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2677                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2678                                 continue;
2679                         /* update endpoints */
2680                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2681                                sizeof(pol->xfrm_vec[i].id.daddr));
2682                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2683                                sizeof(pol->xfrm_vec[i].saddr));
2684                         pol->xfrm_vec[i].encap_family = mp->new_family;
2685                         /* flush bundles */
2686                         while ((dst = pol->bundles) != NULL) {
2687                                 pol->bundles = dst->next;
2688                                 dst_free(dst);
2689                         }
2690                 }
2691         }
2692
2693         write_unlock_bh(&pol->lock);
2694
2695         if (!n)
2696                 return -ENODATA;
2697
2698         return 0;
2699 }
2700
2701 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2702 {
2703         int i, j;
2704
2705         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2706                 return -EINVAL;
2707
2708         for (i = 0; i < num_migrate; i++) {
2709                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2710                                    m[i].old_family) == 0) &&
2711                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2712                                    m[i].old_family) == 0))
2713                         return -EINVAL;
2714                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2715                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2716                         return -EINVAL;
2717
2718                 /* check if there is any duplicated entry */
2719                 for (j = i + 1; j < num_migrate; j++) {
2720                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2721                                     sizeof(m[i].old_daddr)) &&
2722                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2723                                     sizeof(m[i].old_saddr)) &&
2724                             m[i].proto == m[j].proto &&
2725                             m[i].mode == m[j].mode &&
2726                             m[i].reqid == m[j].reqid &&
2727                             m[i].old_family == m[j].old_family)
2728                                 return -EINVAL;
2729                 }
2730         }
2731
2732         return 0;
2733 }
2734
2735 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2736                  struct xfrm_migrate *m, int num_migrate,
2737                  struct xfrm_kmaddress *k)
2738 {
2739         int i, err, nx_cur = 0, nx_new = 0;
2740         struct xfrm_policy *pol = NULL;
2741         struct xfrm_state *x, *xc;
2742         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2743         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2744         struct xfrm_migrate *mp;
2745
2746         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2747                 goto out;
2748
2749         /* Stage 1 - find policy */
2750         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2751                 err = -ENOENT;
2752                 goto out;
2753         }
2754
2755         /* Stage 2 - find and update state(s) */
2756         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2757                 if ((x = xfrm_migrate_state_find(mp))) {
2758                         x_cur[nx_cur] = x;
2759                         nx_cur++;
2760                         if ((xc = xfrm_state_migrate(x, mp))) {
2761                                 x_new[nx_new] = xc;
2762                                 nx_new++;
2763                         } else {
2764                                 err = -ENODATA;
2765                                 goto restore_state;
2766                         }
2767                 }
2768         }
2769
2770         /* Stage 3 - update policy */
2771         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2772                 goto restore_state;
2773
2774         /* Stage 4 - delete old state(s) */
2775         if (nx_cur) {
2776                 xfrm_states_put(x_cur, nx_cur);
2777                 xfrm_states_delete(x_cur, nx_cur);
2778         }
2779
2780         /* Stage 5 - announce */
2781         km_migrate(sel, dir, type, m, num_migrate, k);
2782
2783         xfrm_pol_put(pol);
2784
2785         return 0;
2786 out:
2787         return err;
2788
2789 restore_state:
2790         if (pol)
2791                 xfrm_pol_put(pol);
2792         if (nx_cur)
2793                 xfrm_states_put(x_cur, nx_cur);
2794         if (nx_new)
2795                 xfrm_states_delete(x_new, nx_new);
2796
2797         return err;
2798 }
2799 EXPORT_SYMBOL(xfrm_migrate);
2800 #endif