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 DEFINE_MUTEX(xfrm_cfg_mutex);
  38 EXPORT_SYMBOL(xfrm_cfg_mutex);
  39
  40 static DEFINE_RWLOCK(xfrm_policy_lock);
  41
  42 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  43 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  44
  45 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  46
  47 static HLIST_HEAD(xfrm_policy_gc_list);
  48 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  49
  50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  52 static void xfrm_init_pmtu(struct dst_entry *dst);
  53
  54 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  55                                                 int dir);
  56
  57 static inline int
  58 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  59 {
  60         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  61                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  62                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  63                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  64                 (fl->proto == sel->proto || !sel->proto) &&
  65                 (fl->oif == sel->ifindex || !sel->ifindex);
  66 }
  67
  68 static inline int
  69 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  70 {
  71         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  72                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  73                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  74                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  75                 (fl->proto == sel->proto || !sel->proto) &&
  76                 (fl->oif == sel->ifindex || !sel->ifindex);
  77 }
  78
  79 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  80                     unsigned short family)
  81 {
  82         switch (family) {
  83         case AF_INET:
  84                 return __xfrm4_selector_match(sel, fl);
  85         case AF_INET6:
  86                 return __xfrm6_selector_match(sel, fl);
  87         }
  88         return 0;
  89 }
  90
  91 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
  92                                                   xfrm_address_t *saddr,
  93                                                   xfrm_address_t *daddr,
  94                                                   int family)
  95 {
  96         struct xfrm_policy_afinfo *afinfo;
  97         struct dst_entry *dst;
  98
  99         afinfo = xfrm_policy_get_afinfo(family);
 100         if (unlikely(afinfo == NULL))
 101                 return ERR_PTR(-EAFNOSUPPORT);
 102
 103         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 104
 105         xfrm_policy_put_afinfo(afinfo);
 106
 107         return dst;
 108 }
 109
 110 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 111                                                 xfrm_address_t *prev_saddr,
 112                                                 xfrm_address_t *prev_daddr,
 113                                                 int family)
 114 {
 115         struct net *net = xs_net(x);
 116         xfrm_address_t *saddr = &x->props.saddr;
 117         xfrm_address_t *daddr = &x->id.daddr;
 118         struct dst_entry *dst;
 119
 120         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 121                 saddr = x->coaddr;
 122                 daddr = prev_daddr;
 123         }
 124         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 125                 saddr = prev_saddr;
 126                 daddr = x->coaddr;
 127         }
 128
 129         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 130
 131         if (!IS_ERR(dst)) {
 132                 if (prev_saddr != saddr)
 133                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 134                 if (prev_daddr != daddr)
 135                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 136         }
 137
 138         return dst;
 139 }
 140
 141 static inline unsigned long make_jiffies(long secs)
 142 {
 143         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 144                 return MAX_SCHEDULE_TIMEOUT-1;
 145         else
 146                 return secs*HZ;
 147 }
 148
 149 static void xfrm_policy_timer(unsigned long data)
 150 {
 151         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 152         unsigned long now = get_seconds();
 153         long next = LONG_MAX;
 154         int warn = 0;
 155         int dir;
 156
 157         read_lock(&xp->lock);
 158
 159         if (xp->walk.dead)
 160                 goto out;
 161
 162         dir = xfrm_policy_id2dir(xp->index);
 163
 164         if (xp->lft.hard_add_expires_seconds) {
 165                 long tmo = xp->lft.hard_add_expires_seconds +
 166                         xp->curlft.add_time - now;
 167                 if (tmo <= 0)
 168                         goto expired;
 169                 if (tmo < next)
 170                         next = tmo;
 171         }
 172         if (xp->lft.hard_use_expires_seconds) {
 173                 long tmo = xp->lft.hard_use_expires_seconds +
 174                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 175                 if (tmo <= 0)
 176                         goto expired;
 177                 if (tmo < next)
 178                         next = tmo;
 179         }
 180         if (xp->lft.soft_add_expires_seconds) {
 181                 long tmo = xp->lft.soft_add_expires_seconds +
 182                         xp->curlft.add_time - now;
 183                 if (tmo <= 0) {
 184                         warn = 1;
 185                         tmo = XFRM_KM_TIMEOUT;
 186                 }
 187                 if (tmo < next)
 188                         next = tmo;
 189         }
 190         if (xp->lft.soft_use_expires_seconds) {
 191                 long tmo = xp->lft.soft_use_expires_seconds +
 192                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 193                 if (tmo <= 0) {
 194                         warn = 1;
 195                         tmo = XFRM_KM_TIMEOUT;
 196                 }
 197                 if (tmo < next)
 198                         next = tmo;
 199         }
 200
 201         if (warn)
 202                 km_policy_expired(xp, dir, 0, 0);
 203         if (next != LONG_MAX &&
 204             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 205                 xfrm_pol_hold(xp);
 206
 207 out:
 208         read_unlock(&xp->lock);
 209         xfrm_pol_put(xp);
 210         return;
 211
 212 expired:
 213         read_unlock(&xp->lock);
 214         if (!xfrm_policy_delete(xp, dir))
 215                 km_policy_expired(xp, dir, 1, 0);
 216         xfrm_pol_put(xp);
 217 }
 218
 219
 220 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 221  * SPD calls.
 222  */
 223
 224 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 225 {
 226         struct xfrm_policy *policy;
 227
 228         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 229
 230         if (policy) {
 231                 write_pnet(&policy->xp_net, net);
 232                 INIT_LIST_HEAD(&policy->walk.all);
 233                 INIT_HLIST_NODE(&policy->bydst);
 234                 INIT_HLIST_NODE(&policy->byidx);
 235                 rwlock_init(&policy->lock);
 236                 atomic_set(&policy->refcnt, 1);
 237                 setup_timer(&policy->timer, xfrm_policy_timer,
 238                                 (unsigned long)policy);
 239         }
 240         return policy;
 241 }
 242 EXPORT_SYMBOL(xfrm_policy_alloc);
 243
 244 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 245
 246 void xfrm_policy_destroy(struct xfrm_policy *policy)
 247 {
 248         BUG_ON(!policy->walk.dead);
 249
 250         BUG_ON(policy->bundles);
 251
 252         if (del_timer(&policy->timer))
 253                 BUG();
 254
 255         security_xfrm_policy_free(policy->security);
 256         kfree(policy);
 257 }
 258 EXPORT_SYMBOL(xfrm_policy_destroy);
 259
 260 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 261 {
 262         struct dst_entry *dst;
 263
 264         while ((dst = policy->bundles) != NULL) {
 265                 policy->bundles = dst->next;
 266                 dst_free(dst);
 267         }
 268
 269         if (del_timer(&policy->timer))
 270                 atomic_dec(&policy->refcnt);
 271
 272         if (atomic_read(&policy->refcnt) > 1)
 273                 flow_cache_flush();
 274
 275         xfrm_pol_put(policy);
 276 }
 277
 278 static void xfrm_policy_gc_task(struct work_struct *work)
 279 {
 280         struct xfrm_policy *policy;
 281         struct hlist_node *entry, *tmp;
 282         struct hlist_head gc_list;
 283
 284         spin_lock_bh(&xfrm_policy_gc_lock);
 285         gc_list.first = xfrm_policy_gc_list.first;
 286         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 287         spin_unlock_bh(&xfrm_policy_gc_lock);
 288
 289         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 290                 xfrm_policy_gc_kill(policy);
 291 }
 292 static DECLARE_WORK(xfrm_policy_gc_work, xfrm_policy_gc_task);
 293
 294 /* Rule must be locked. Release descentant resources, announce
 295  * entry dead. The rule must be unlinked from lists to the moment.
 296  */
 297
 298 static void xfrm_policy_kill(struct xfrm_policy *policy)
 299 {
 300         int dead;
 301
 302         write_lock_bh(&policy->lock);
 303         dead = policy->walk.dead;
 304         policy->walk.dead = 1;
 305         write_unlock_bh(&policy->lock);
 306
 307         if (unlikely(dead)) {
 308                 WARN_ON(1);
 309                 return;
 310         }
 311
 312         spin_lock_bh(&xfrm_policy_gc_lock);
 313         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 314         spin_unlock_bh(&xfrm_policy_gc_lock);
 315
 316         schedule_work(&xfrm_policy_gc_work);
 317 }
 318
 319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 320
 321 static inline unsigned int idx_hash(struct net *net, u32 index)
 322 {
 323         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 324 }
 325
 326 static struct hlist_head *policy_hash_bysel(struct net *net, struct xfrm_selector *sel, unsigned short family, int dir)
 327 {
 328         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 329         unsigned int hash = __sel_hash(sel, family, hmask);
 330
 331         return (hash == hmask + 1 ?
 332                 &net->xfrm.policy_inexact[dir] :
 333                 net->xfrm.policy_bydst[dir].table + hash);
 334 }
 335
 336 static struct hlist_head *policy_hash_direct(struct net *net, xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 337 {
 338         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 339         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 340
 341         return net->xfrm.policy_bydst[dir].table + hash;
 342 }
 343
 344 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 345                                    struct hlist_head *ndsttable,
 346                                    unsigned int nhashmask)
 347 {
 348         struct hlist_node *entry, *tmp, *entry0 = NULL;
 349         struct xfrm_policy *pol;
 350         unsigned int h0 = 0;
 351
 352 redo:
 353         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 354                 unsigned int h;
 355
 356                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 357                                 pol->family, nhashmask);
 358                 if (!entry0) {
 359                         hlist_del(entry);
 360                         hlist_add_head(&pol->bydst, ndsttable+h);
 361                         h0 = h;
 362                 } else {
 363                         if (h != h0)
 364                                 continue;
 365                         hlist_del(entry);
 366                         hlist_add_after(entry0, &pol->bydst);
 367                 }
 368                 entry0 = entry;
 369         }
 370         if (!hlist_empty(list)) {
 371                 entry0 = NULL;
 372                 goto redo;
 373         }
 374 }
 375
 376 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 377                                    struct hlist_head *nidxtable,
 378                                    unsigned int nhashmask)
 379 {
 380         struct hlist_node *entry, *tmp;
 381         struct xfrm_policy *pol;
 382
 383         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 384                 unsigned int h;
 385
 386                 h = __idx_hash(pol->index, nhashmask);
 387                 hlist_add_head(&pol->byidx, nidxtable+h);
 388         }
 389 }
 390
 391 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 392 {
 393         return ((old_hmask + 1) << 1) - 1;
 394 }
 395
 396 static void xfrm_bydst_resize(struct net *net, int dir)
 397 {
 398         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 399         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 400         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 401         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 402         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 403         int i;
 404
 405         if (!ndst)
 406                 return;
 407
 408         write_lock_bh(&xfrm_policy_lock);
 409
 410         for (i = hmask; i >= 0; i--)
 411                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 412
 413         net->xfrm.policy_bydst[dir].table = ndst;
 414         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 415
 416         write_unlock_bh(&xfrm_policy_lock);
 417
 418         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 419 }
 420
 421 static void xfrm_byidx_resize(struct net *net, int total)
 422 {
 423         unsigned int hmask = net->xfrm.policy_idx_hmask;
 424         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 425         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 426         struct hlist_head *oidx = net->xfrm.policy_byidx;
 427         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 428         int i;
 429
 430         if (!nidx)
 431                 return;
 432
 433         write_lock_bh(&xfrm_policy_lock);
 434
 435         for (i = hmask; i >= 0; i--)
 436                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 437
 438         net->xfrm.policy_byidx = nidx;
 439         net->xfrm.policy_idx_hmask = nhashmask;
 440
 441         write_unlock_bh(&xfrm_policy_lock);
 442
 443         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 444 }
 445
 446 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 447 {
 448         unsigned int cnt = net->xfrm.policy_count[dir];
 449         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 450
 451         if (total)
 452                 *total += cnt;
 453
 454         if ((hmask + 1) < xfrm_policy_hashmax &&
 455             cnt > hmask)
 456                 return 1;
 457
 458         return 0;
 459 }
 460
 461 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 462 {
 463         unsigned int hmask = net->xfrm.policy_idx_hmask;
 464
 465         if ((hmask + 1) < xfrm_policy_hashmax &&
 466             total > hmask)
 467                 return 1;
 468
 469         return 0;
 470 }
 471
 472 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 473 {
 474         read_lock_bh(&xfrm_policy_lock);
 475         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 476         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 477         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 478         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 479         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 480         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 481         si->spdhcnt = net->xfrm.policy_idx_hmask;
 482         si->spdhmcnt = xfrm_policy_hashmax;
 483         read_unlock_bh(&xfrm_policy_lock);
 484 }
 485 EXPORT_SYMBOL(xfrm_spd_getinfo);
 486
 487 static DEFINE_MUTEX(hash_resize_mutex);
 488 static void xfrm_hash_resize(struct work_struct *work)
 489 {
 490         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 491         int dir, total;
 492
 493         mutex_lock(&hash_resize_mutex);
 494
 495         total = 0;
 496         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 497                 if (xfrm_bydst_should_resize(net, dir, &total))
 498                         xfrm_bydst_resize(net, dir);
 499         }
 500         if (xfrm_byidx_should_resize(net, total))
 501                 xfrm_byidx_resize(net, total);
 502
 503         mutex_unlock(&hash_resize_mutex);
 504 }
 505
 506 /* Generate new index... KAME seems to generate them ordered by cost
 507  * of an absolute inpredictability of ordering of rules. This will not pass. */
 508 static u32 xfrm_gen_index(struct net *net, int dir)
 509 {
 510         static u32 idx_generator;
 511
 512         for (;;) {
 513                 struct hlist_node *entry;
 514                 struct hlist_head *list;
 515                 struct xfrm_policy *p;
 516                 u32 idx;
 517                 int found;
 518
 519                 idx = (idx_generator | dir);
 520                 idx_generator += 8;
 521                 if (idx == 0)
 522                         idx = 8;
 523                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 524                 found = 0;
 525                 hlist_for_each_entry(p, entry, list, byidx) {
 526                         if (p->index == idx) {
 527                                 found = 1;
 528                                 break;
 529                         }
 530                 }
 531                 if (!found)
 532                         return idx;
 533         }
 534 }
 535
 536 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 537 {
 538         u32 *p1 = (u32 *) s1;
 539         u32 *p2 = (u32 *) s2;
 540         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 541         int i;
 542
 543         for (i = 0; i < len; i++) {
 544                 if (p1[i] != p2[i])
 545                         return 1;
 546         }
 547
 548         return 0;
 549 }
 550
 551 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 552 {
 553         struct net *net = xp_net(policy);
 554         struct xfrm_policy *pol;
 555         struct xfrm_policy *delpol;
 556         struct hlist_head *chain;
 557         struct hlist_node *entry, *newpos;
 558         struct dst_entry *gc_list;
 559         u32 mark = policy->mark.v & policy->mark.m;
 560
 561         write_lock_bh(&xfrm_policy_lock);
 562         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 563         delpol = NULL;
 564         newpos = NULL;
 565         hlist_for_each_entry(pol, entry, chain, bydst) {
 566                 if (pol->type == policy->type &&
 567                     !selector_cmp(&pol->selector, &policy->selector) &&
 568                     (mark & pol->mark.m) == pol->mark.v &&
 569                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 570                     !WARN_ON(delpol)) {
 571                         if (excl) {
 572                                 write_unlock_bh(&xfrm_policy_lock);
 573                                 return -EEXIST;
 574                         }
 575                         delpol = pol;
 576                         if (policy->priority > pol->priority)
 577                                 continue;
 578                 } else if (policy->priority >= pol->priority) {
 579                         newpos = &pol->bydst;
 580                         continue;
 581                 }
 582                 if (delpol)
 583                         break;
 584         }
 585         if (newpos)
 586                 hlist_add_after(newpos, &policy->bydst);
 587         else
 588                 hlist_add_head(&policy->bydst, chain);
 589         xfrm_pol_hold(policy);
 590         net->xfrm.policy_count[dir]++;
 591         atomic_inc(&flow_cache_genid);
 592         if (delpol)
 593                 __xfrm_policy_unlink(delpol, dir);
 594         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 595         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 596         policy->curlft.add_time = get_seconds();
 597         policy->curlft.use_time = 0;
 598         if (!mod_timer(&policy->timer, jiffies + HZ))
 599                 xfrm_pol_hold(policy);
 600         list_add(&policy->walk.all, &net->xfrm.policy_all);
 601         write_unlock_bh(&xfrm_policy_lock);
 602
 603         if (delpol)
 604                 xfrm_policy_kill(delpol);
 605         else if (xfrm_bydst_should_resize(net, dir, NULL))
 606                 schedule_work(&net->xfrm.policy_hash_work);
 607
 608         read_lock_bh(&xfrm_policy_lock);
 609         gc_list = NULL;
 610         entry = &policy->bydst;
 611         hlist_for_each_entry_continue(policy, entry, bydst) {
 612                 struct dst_entry *dst;
 613
 614                 write_lock(&policy->lock);
 615                 dst = policy->bundles;
 616                 if (dst) {
 617                         struct dst_entry *tail = dst;
 618                         while (tail->next)
 619                                 tail = tail->next;
 620                         tail->next = gc_list;
 621                         gc_list = dst;
 622
 623                         policy->bundles = NULL;
 624                 }
 625                 write_unlock(&policy->lock);
 626         }
 627         read_unlock_bh(&xfrm_policy_lock);
 628
 629         while (gc_list) {
 630                 struct dst_entry *dst = gc_list;
 631
 632                 gc_list = dst->next;
 633                 dst_free(dst);
 634         }
 635
 636         return 0;
 637 }
 638 EXPORT_SYMBOL(xfrm_policy_insert);
 639
 640 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 641                                           int dir, struct xfrm_selector *sel,
 642                                           struct xfrm_sec_ctx *ctx, int delete,
 643                                           int *err)
 644 {
 645         struct xfrm_policy *pol, *ret;
 646         struct hlist_head *chain;
 647         struct hlist_node *entry;
 648
 649         *err = 0;
 650         write_lock_bh(&xfrm_policy_lock);
 651         chain = policy_hash_bysel(net, sel, sel->family, dir);
 652         ret = NULL;
 653         hlist_for_each_entry(pol, entry, chain, bydst) {
 654                 if (pol->type == type &&
 655                     (mark & pol->mark.m) == pol->mark.v &&
 656                     !selector_cmp(sel, &pol->selector) &&
 657                     xfrm_sec_ctx_match(ctx, pol->security)) {
 658                         xfrm_pol_hold(pol);
 659                         if (delete) {
 660                                 *err = security_xfrm_policy_delete(
 661                                                                 pol->security);
 662                                 if (*err) {
 663                                         write_unlock_bh(&xfrm_policy_lock);
 664                                         return pol;
 665                                 }
 666                                 __xfrm_policy_unlink(pol, dir);
 667                         }
 668                         ret = pol;
 669                         break;
 670                 }
 671         }
 672         write_unlock_bh(&xfrm_policy_lock);
 673
 674         if (ret && delete) {
 675                 atomic_inc(&flow_cache_genid);
 676                 xfrm_policy_kill(ret);
 677         }
 678         return ret;
 679 }
 680 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 681
 682 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 683                                      int dir, u32 id, int delete, int *err)
 684 {
 685         struct xfrm_policy *pol, *ret;
 686         struct hlist_head *chain;
 687         struct hlist_node *entry;
 688
 689         *err = -ENOENT;
 690         if (xfrm_policy_id2dir(id) != dir)
 691                 return NULL;
 692
 693         *err = 0;
 694         write_lock_bh(&xfrm_policy_lock);
 695         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 696         ret = NULL;
 697         hlist_for_each_entry(pol, entry, chain, byidx) {
 698                 if (pol->type == type && pol->index == id &&
 699                     (mark & pol->mark.m) == pol->mark.v) {
 700                         xfrm_pol_hold(pol);
 701                         if (delete) {
 702                                 *err = security_xfrm_policy_delete(
 703                                                                 pol->security);
 704                                 if (*err) {
 705                                         write_unlock_bh(&xfrm_policy_lock);
 706                                         return pol;
 707                                 }
 708                                 __xfrm_policy_unlink(pol, dir);
 709                         }
 710                         ret = pol;
 711                         break;
 712                 }
 713         }
 714         write_unlock_bh(&xfrm_policy_lock);
 715
 716         if (ret && delete) {
 717                 atomic_inc(&flow_cache_genid);
 718                 xfrm_policy_kill(ret);
 719         }
 720         return ret;
 721 }
 722 EXPORT_SYMBOL(xfrm_policy_byid);
 723
 724 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 725 static inline int
 726 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 727 {
 728         int dir, err = 0;
 729
 730         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 731                 struct xfrm_policy *pol;
 732                 struct hlist_node *entry;
 733                 int i;
 734
 735                 hlist_for_each_entry(pol, entry,
 736                                      &net->xfrm.policy_inexact[dir], bydst) {
 737                         if (pol->type != type)
 738                                 continue;
 739                         err = security_xfrm_policy_delete(pol->security);
 740                         if (err) {
 741                                 xfrm_audit_policy_delete(pol, 0,
 742                                                          audit_info->loginuid,
 743                                                          audit_info->sessionid,
 744                                                          audit_info->secid);
 745                                 return err;
 746                         }
 747                 }
 748                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 749                         hlist_for_each_entry(pol, entry,
 750                                              net->xfrm.policy_bydst[dir].table + i,
 751                                              bydst) {
 752                                 if (pol->type != type)
 753                                         continue;
 754                                 err = security_xfrm_policy_delete(
 755                                                                 pol->security);
 756                                 if (err) {
 757                                         xfrm_audit_policy_delete(pol, 0,
 758                                                         audit_info->loginuid,
 759                                                         audit_info->sessionid,
 760                                                         audit_info->secid);
 761                                         return err;
 762                                 }
 763                         }
 764                 }
 765         }
 766         return err;
 767 }
 768 #else
 769 static inline int
 770 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 771 {
 772         return 0;
 773 }
 774 #endif
 775
 776 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 777 {
 778         int dir, err = 0, cnt = 0;
 779         struct xfrm_policy *dp;
 780
 781         write_lock_bh(&xfrm_policy_lock);
 782
 783         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 784         if (err)
 785                 goto out;
 786
 787         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 788                 struct xfrm_policy *pol;
 789                 struct hlist_node *entry;
 790                 int i;
 791
 792         again1:
 793                 hlist_for_each_entry(pol, entry,
 794                                      &net->xfrm.policy_inexact[dir], bydst) {
 795                         if (pol->type != type)
 796                                 continue;
 797                         dp = __xfrm_policy_unlink(pol, dir);
 798                         write_unlock_bh(&xfrm_policy_lock);
 799                         if (dp)
 800                                 cnt++;
 801
 802                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 803                                                  audit_info->sessionid,
 804                                                  audit_info->secid);
 805
 806                         xfrm_policy_kill(pol);
 807
 808                         write_lock_bh(&xfrm_policy_lock);
 809                         goto again1;
 810                 }
 811
 812                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 813         again2:
 814                         hlist_for_each_entry(pol, entry,
 815                                              net->xfrm.policy_bydst[dir].table + i,
 816                                              bydst) {
 817                                 if (pol->type != type)
 818                                         continue;
 819                                 dp = __xfrm_policy_unlink(pol, dir);
 820                                 write_unlock_bh(&xfrm_policy_lock);
 821                                 if (dp)
 822                                         cnt++;
 823
 824                                 xfrm_audit_policy_delete(pol, 1,
 825                                                          audit_info->loginuid,
 826                                                          audit_info->sessionid,
 827                                                          audit_info->secid);
 828                                 xfrm_policy_kill(pol);
 829
 830                                 write_lock_bh(&xfrm_policy_lock);
 831                                 goto again2;
 832                         }
 833                 }
 834
 835         }
 836         if (!cnt)
 837                 err = -ESRCH;
 838         atomic_inc(&flow_cache_genid);
 839 out:
 840         write_unlock_bh(&xfrm_policy_lock);
 841         return err;
 842 }
 843 EXPORT_SYMBOL(xfrm_policy_flush);
 844
 845 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 846                      int (*func)(struct xfrm_policy *, int, int, void*),
 847                      void *data)
 848 {
 849         struct xfrm_policy *pol;
 850         struct xfrm_policy_walk_entry *x;
 851         int error = 0;
 852
 853         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 854             walk->type != XFRM_POLICY_TYPE_ANY)
 855                 return -EINVAL;
 856
 857         if (list_empty(&walk->walk.all) && walk->seq != 0)
 858                 return 0;
 859
 860         write_lock_bh(&xfrm_policy_lock);
 861         if (list_empty(&walk->walk.all))
 862                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 863         else
 864                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 865         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 866                 if (x->dead)
 867                         continue;
 868                 pol = container_of(x, struct xfrm_policy, walk);
 869                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 870                     walk->type != pol->type)
 871                         continue;
 872                 error = func(pol, xfrm_policy_id2dir(pol->index),
 873                              walk->seq, data);
 874                 if (error) {
 875                         list_move_tail(&walk->walk.all, &x->all);
 876                         goto out;
 877                 }
 878                 walk->seq++;
 879         }
 880         if (walk->seq == 0) {
 881                 error = -ENOENT;
 882                 goto out;
 883         }
 884         list_del_init(&walk->walk.all);
 885 out:
 886         write_unlock_bh(&xfrm_policy_lock);
 887         return error;
 888 }
 889 EXPORT_SYMBOL(xfrm_policy_walk);
 890
 891 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 892 {
 893         INIT_LIST_HEAD(&walk->walk.all);
 894         walk->walk.dead = 1;
 895         walk->type = type;
 896         walk->seq = 0;
 897 }
 898 EXPORT_SYMBOL(xfrm_policy_walk_init);
 899
 900 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 901 {
 902         if (list_empty(&walk->walk.all))
 903                 return;
 904
 905         write_lock_bh(&xfrm_policy_lock);
 906         list_del(&walk->walk.all);
 907         write_unlock_bh(&xfrm_policy_lock);
 908 }
 909 EXPORT_SYMBOL(xfrm_policy_walk_done);
 910
 911 /*
 912  * Find policy to apply to this flow.
 913  *
 914  * Returns 0 if policy found, else an -errno.
 915  */
 916 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 917                              u8 type, u16 family, int dir)
 918 {
 919         struct xfrm_selector *sel = &pol->selector;
 920         int match, ret = -ESRCH;
 921
 922         if (pol->family != family ||
 923             (fl->mark & pol->mark.m) != pol->mark.v ||
 924             pol->type != type)
 925                 return ret;
 926
 927         match = xfrm_selector_match(sel, fl, family);
 928         if (match)
 929                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 930                                                   dir);
 931
 932         return ret;
 933 }
 934
 935 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 936                                                      struct flowi *fl,
 937                                                      u16 family, u8 dir)
 938 {
 939         int err;
 940         struct xfrm_policy *pol, *ret;
 941         xfrm_address_t *daddr, *saddr;
 942         struct hlist_node *entry;
 943         struct hlist_head *chain;
 944         u32 priority = ~0U;
 945
 946         daddr = xfrm_flowi_daddr(fl, family);
 947         saddr = xfrm_flowi_saddr(fl, family);
 948         if (unlikely(!daddr || !saddr))
 949                 return NULL;
 950
 951         read_lock_bh(&xfrm_policy_lock);
 952         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 953         ret = NULL;
 954         hlist_for_each_entry(pol, entry, chain, bydst) {
 955                 err = xfrm_policy_match(pol, fl, type, family, dir);
 956                 if (err) {
 957                         if (err == -ESRCH)
 958                                 continue;
 959                         else {
 960                                 ret = ERR_PTR(err);
 961                                 goto fail;
 962                         }
 963                 } else {
 964                         ret = pol;
 965                         priority = ret->priority;
 966                         break;
 967                 }
 968         }
 969         chain = &net->xfrm.policy_inexact[dir];
 970         hlist_for_each_entry(pol, entry, chain, bydst) {
 971                 err = xfrm_policy_match(pol, fl, type, family, dir);
 972                 if (err) {
 973                         if (err == -ESRCH)
 974                                 continue;
 975                         else {
 976                                 ret = ERR_PTR(err);
 977                                 goto fail;
 978                         }
 979                 } else if (pol->priority < priority) {
 980                         ret = pol;
 981                         break;
 982                 }
 983         }
 984         if (ret)
 985                 xfrm_pol_hold(ret);
 986 fail:
 987         read_unlock_bh(&xfrm_policy_lock);
 988
 989         return ret;
 990 }
 991
 992 static int xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
 993                               u8 dir, void **objp, atomic_t **obj_refp)
 994 {
 995         struct xfrm_policy *pol;
 996         int err = 0;
 997
 998 #ifdef CONFIG_XFRM_SUB_POLICY
 999         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1000         if (IS_ERR(pol)) {
1001                 err = PTR_ERR(pol);
1002                 pol = NULL;
1003         }
1004         if (pol || err)
1005                 goto end;
1006 #endif
1007         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1008         if (IS_ERR(pol)) {
1009                 err = PTR_ERR(pol);
1010                 pol = NULL;
1011         }
1012 #ifdef CONFIG_XFRM_SUB_POLICY
1013 end:
1014 #endif
1015         if ((*objp = (void *) pol) != NULL)
1016                 *obj_refp = &pol->refcnt;
1017         return err;
1018 }
1019
1020 static inline int policy_to_flow_dir(int dir)
1021 {
1022         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1023             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1024             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1025                 return dir;
1026         switch (dir) {
1027         default:
1028         case XFRM_POLICY_IN:
1029                 return FLOW_DIR_IN;
1030         case XFRM_POLICY_OUT:
1031                 return FLOW_DIR_OUT;
1032         case XFRM_POLICY_FWD:
1033                 return FLOW_DIR_FWD;
1034         }
1035 }
1036
1037 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1038 {
1039         struct xfrm_policy *pol;
1040
1041         read_lock_bh(&xfrm_policy_lock);
1042         if ((pol = sk->sk_policy[dir]) != NULL) {
1043                 int match = xfrm_selector_match(&pol->selector, fl,
1044                                                 sk->sk_family);
1045                 int err = 0;
1046
1047                 if (match) {
1048                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1049                                 pol = NULL;
1050                                 goto out;
1051                         }
1052                         err = security_xfrm_policy_lookup(pol->security,
1053                                                       fl->secid,
1054                                                       policy_to_flow_dir(dir));
1055                         if (!err)
1056                                 xfrm_pol_hold(pol);
1057                         else if (err == -ESRCH)
1058                                 pol = NULL;
1059                         else
1060                                 pol = ERR_PTR(err);
1061                 } else
1062                         pol = NULL;
1063         }
1064 out:
1065         read_unlock_bh(&xfrm_policy_lock);
1066         return pol;
1067 }
1068
1069 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1070 {
1071         struct net *net = xp_net(pol);
1072         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1073                                                      pol->family, dir);
1074
1075         list_add(&pol->walk.all, &net->xfrm.policy_all);
1076         hlist_add_head(&pol->bydst, chain);
1077         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1078         net->xfrm.policy_count[dir]++;
1079         xfrm_pol_hold(pol);
1080
1081         if (xfrm_bydst_should_resize(net, dir, NULL))
1082                 schedule_work(&net->xfrm.policy_hash_work);
1083 }
1084
1085 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1086                                                 int dir)
1087 {
1088         struct net *net = xp_net(pol);
1089
1090         if (hlist_unhashed(&pol->bydst))
1091                 return NULL;
1092
1093         hlist_del(&pol->bydst);
1094         hlist_del(&pol->byidx);
1095         list_del(&pol->walk.all);
1096         net->xfrm.policy_count[dir]--;
1097
1098         return pol;
1099 }
1100
1101 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1102 {
1103         write_lock_bh(&xfrm_policy_lock);
1104         pol = __xfrm_policy_unlink(pol, dir);
1105         write_unlock_bh(&xfrm_policy_lock);
1106         if (pol) {
1107                 if (dir < XFRM_POLICY_MAX)
1108                         atomic_inc(&flow_cache_genid);
1109                 xfrm_policy_kill(pol);
1110                 return 0;
1111         }
1112         return -ENOENT;
1113 }
1114 EXPORT_SYMBOL(xfrm_policy_delete);
1115
1116 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1117 {
1118         struct net *net = xp_net(pol);
1119         struct xfrm_policy *old_pol;
1120
1121 #ifdef CONFIG_XFRM_SUB_POLICY
1122         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1123                 return -EINVAL;
1124 #endif
1125
1126         write_lock_bh(&xfrm_policy_lock);
1127         old_pol = sk->sk_policy[dir];
1128         sk->sk_policy[dir] = pol;
1129         if (pol) {
1130                 pol->curlft.add_time = get_seconds();
1131                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1132                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1133         }
1134         if (old_pol)
1135                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1136         write_unlock_bh(&xfrm_policy_lock);
1137
1138         if (old_pol) {
1139                 xfrm_policy_kill(old_pol);
1140         }
1141         return 0;
1142 }
1143
1144 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1145 {
1146         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1147
1148         if (newp) {
1149                 newp->selector = old->selector;
1150                 if (security_xfrm_policy_clone(old->security,
1151                                                &newp->security)) {
1152                         kfree(newp);
1153                         return NULL;  /* ENOMEM */
1154                 }
1155                 newp->lft = old->lft;
1156                 newp->curlft = old->curlft;
1157                 newp->mark = old->mark;
1158                 newp->action = old->action;
1159                 newp->flags = old->flags;
1160                 newp->xfrm_nr = old->xfrm_nr;
1161                 newp->index = old->index;
1162                 newp->type = old->type;
1163                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1164                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1165                 write_lock_bh(&xfrm_policy_lock);
1166                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1167                 write_unlock_bh(&xfrm_policy_lock);
1168                 xfrm_pol_put(newp);
1169         }
1170         return newp;
1171 }
1172
1173 int __xfrm_sk_clone_policy(struct sock *sk)
1174 {
1175         struct xfrm_policy *p0 = sk->sk_policy[0],
1176                            *p1 = sk->sk_policy[1];
1177
1178         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1179         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1180                 return -ENOMEM;
1181         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1182                 return -ENOMEM;
1183         return 0;
1184 }
1185
1186 static int
1187 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1188                unsigned short family)
1189 {
1190         int err;
1191         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1192
1193         if (unlikely(afinfo == NULL))
1194                 return -EINVAL;
1195         err = afinfo->get_saddr(net, local, remote);
1196         xfrm_policy_put_afinfo(afinfo);
1197         return err;
1198 }
1199
1200 /* Resolve list of templates for the flow, given policy. */
1201
1202 static int
1203 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1204                       struct xfrm_state **xfrm,
1205                       unsigned short family)
1206 {
1207         struct net *net = xp_net(policy);
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(net, &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(struct net *net, int family)
1331 {
1332         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1333         struct dst_ops *dst_ops;
1334         struct xfrm_dst *xdst;
1335
1336         if (!afinfo)
1337                 return ERR_PTR(-EINVAL);
1338
1339         switch (family) {
1340         case AF_INET:
1341                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1342                 break;
1343 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1344         case AF_INET6:
1345                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1346                 break;
1347 #endif
1348         default:
1349                 BUG();
1350         }
1351         xdst = dst_alloc(dst_ops) ?: ERR_PTR(-ENOBUFS);
1352
1353         xfrm_policy_put_afinfo(afinfo);
1354
1355         return xdst;
1356 }
1357
1358 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1359                                  int nfheader_len)
1360 {
1361         struct xfrm_policy_afinfo *afinfo =
1362                 xfrm_policy_get_afinfo(dst->ops->family);
1363         int err;
1364
1365         if (!afinfo)
1366                 return -EINVAL;
1367
1368         err = afinfo->init_path(path, dst, nfheader_len);
1369
1370         xfrm_policy_put_afinfo(afinfo);
1371
1372         return err;
1373 }
1374
1375 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1376                                 struct flowi *fl)
1377 {
1378         struct xfrm_policy_afinfo *afinfo =
1379                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1380         int err;
1381
1382         if (!afinfo)
1383                 return -EINVAL;
1384
1385         err = afinfo->fill_dst(xdst, dev, fl);
1386
1387         xfrm_policy_put_afinfo(afinfo);
1388
1389         return err;
1390 }
1391
1392 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1393  * all the metrics... Shortly, bundle a bundle.
1394  */
1395
1396 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1397                                             struct xfrm_state **xfrm, int nx,
1398                                             struct flowi *fl,
1399                                             struct dst_entry *dst)
1400 {
1401         struct net *net = xp_net(policy);
1402         unsigned long now = jiffies;
1403         struct net_device *dev;
1404         struct dst_entry *dst_prev = NULL;
1405         struct dst_entry *dst0 = NULL;
1406         int i = 0;
1407         int err;
1408         int header_len = 0;
1409         int nfheader_len = 0;
1410         int trailer_len = 0;
1411         int tos;
1412         int family = policy->selector.family;
1413         xfrm_address_t saddr, daddr;
1414
1415         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1416
1417         tos = xfrm_get_tos(fl, family);
1418         err = tos;
1419         if (tos < 0)
1420                 goto put_states;
1421
1422         dst_hold(dst);
1423
1424         for (; i < nx; i++) {
1425                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1426                 struct dst_entry *dst1 = &xdst->u.dst;
1427
1428                 err = PTR_ERR(xdst);
1429                 if (IS_ERR(xdst)) {
1430                         dst_release(dst);
1431                         goto put_states;
1432                 }
1433
1434                 if (!dst_prev)
1435                         dst0 = dst1;
1436                 else {
1437                         dst_prev->child = dst_clone(dst1);
1438                         dst1->flags |= DST_NOHASH;
1439                 }
1440
1441                 xdst->route = dst;
1442                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1443
1444                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1445                         family = xfrm[i]->props.family;
1446                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1447                                               family);
1448                         err = PTR_ERR(dst);
1449                         if (IS_ERR(dst))
1450                                 goto put_states;
1451                 } else
1452                         dst_hold(dst);
1453
1454                 dst1->xfrm = xfrm[i];
1455                 xdst->genid = xfrm[i]->genid;
1456
1457                 dst1->obsolete = -1;
1458                 dst1->flags |= DST_HOST;
1459                 dst1->lastuse = now;
1460
1461                 dst1->input = dst_discard;
1462                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1463
1464                 dst1->next = dst_prev;
1465                 dst_prev = dst1;
1466
1467                 header_len += xfrm[i]->props.header_len;
1468                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1469                         nfheader_len += xfrm[i]->props.header_len;
1470                 trailer_len += xfrm[i]->props.trailer_len;
1471         }
1472
1473         dst_prev->child = dst;
1474         dst0->path = dst;
1475
1476         err = -ENODEV;
1477         dev = dst->dev;
1478         if (!dev)
1479                 goto free_dst;
1480
1481         /* Copy neighbour for reachability confirmation */
1482         dst0->neighbour = neigh_clone(dst->neighbour);
1483
1484         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1485         xfrm_init_pmtu(dst_prev);
1486
1487         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1488                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1489
1490                 err = xfrm_fill_dst(xdst, dev, fl);
1491                 if (err)
1492                         goto free_dst;
1493
1494                 dst_prev->header_len = header_len;
1495                 dst_prev->trailer_len = trailer_len;
1496                 header_len -= xdst->u.dst.xfrm->props.header_len;
1497                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1498         }
1499
1500 out:
1501         return dst0;
1502
1503 put_states:
1504         for (; i < nx; i++)
1505                 xfrm_state_put(xfrm[i]);
1506 free_dst:
1507         if (dst0)
1508                 dst_free(dst0);
1509         dst0 = ERR_PTR(err);
1510         goto out;
1511 }
1512
1513 static int inline
1514 xfrm_dst_alloc_copy(void **target, void *src, int size)
1515 {
1516         if (!*target) {
1517                 *target = kmalloc(size, GFP_ATOMIC);
1518                 if (!*target)
1519                         return -ENOMEM;
1520         }
1521         memcpy(*target, src, size);
1522         return 0;
1523 }
1524
1525 static int inline
1526 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1527 {
1528 #ifdef CONFIG_XFRM_SUB_POLICY
1529         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1530         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1531                                    sel, sizeof(*sel));
1532 #else
1533         return 0;
1534 #endif
1535 }
1536
1537 static int inline
1538 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1539 {
1540 #ifdef CONFIG_XFRM_SUB_POLICY
1541         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1542         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1543 #else
1544         return 0;
1545 #endif
1546 }
1547
1548 static int stale_bundle(struct dst_entry *dst);
1549
1550 /* Main function: finds/creates a bundle for given flow.
1551  *
1552  * At the moment we eat a raw IP route. Mostly to speed up lookups
1553  * on interfaces with disabled IPsec.
1554  */
1555 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1556                   struct sock *sk, int flags)
1557 {
1558         struct xfrm_policy *policy;
1559         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1560         int npols;
1561         int pol_dead;
1562         int xfrm_nr;
1563         int pi;
1564         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1565         struct dst_entry *dst, *dst_orig = *dst_p;
1566         int nx = 0;
1567         int err;
1568         u32 genid;
1569         u16 family;
1570         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1571
1572 restart:
1573         genid = atomic_read(&flow_cache_genid);
1574         policy = NULL;
1575         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1576                 pols[pi] = NULL;
1577         npols = 0;
1578         pol_dead = 0;
1579         xfrm_nr = 0;
1580
1581         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1582                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1583                 err = PTR_ERR(policy);
1584                 if (IS_ERR(policy)) {
1585                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1586                         goto dropdst;
1587                 }
1588         }
1589
1590         if (!policy) {
1591                 /* To accelerate a bit...  */
1592                 if ((dst_orig->flags & DST_NOXFRM) ||
1593                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1594                         goto nopol;
1595
1596                 policy = flow_cache_lookup(net, fl, dst_orig->ops->family,
1597                                            dir, xfrm_policy_lookup);
1598                 err = PTR_ERR(policy);
1599                 if (IS_ERR(policy)) {
1600                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1601                         goto dropdst;
1602                 }
1603         }
1604
1605         if (!policy)
1606                 goto nopol;
1607
1608         family = dst_orig->ops->family;
1609         pols[0] = policy;
1610         npols ++;
1611         xfrm_nr += pols[0]->xfrm_nr;
1612
1613         err = -ENOENT;
1614         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1615                 goto error;
1616
1617         policy->curlft.use_time = get_seconds();
1618
1619         switch (policy->action) {
1620         default:
1621         case XFRM_POLICY_BLOCK:
1622                 /* Prohibit the flow */
1623                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1624                 err = -EPERM;
1625                 goto error;
1626
1627         case XFRM_POLICY_ALLOW:
1628 #ifndef CONFIG_XFRM_SUB_POLICY
1629                 if (policy->xfrm_nr == 0) {
1630                         /* Flow passes not transformed. */
1631                         xfrm_pol_put(policy);
1632                         return 0;
1633                 }
1634 #endif
1635
1636                 /* Try to find matching bundle.
1637                  *
1638                  * LATER: help from flow cache. It is optional, this
1639                  * is required only for output policy.
1640                  */
1641                 dst = xfrm_find_bundle(fl, policy, family);
1642                 if (IS_ERR(dst)) {
1643                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1644                         err = PTR_ERR(dst);
1645                         goto error;
1646                 }
1647
1648                 if (dst)
1649                         break;
1650
1651 #ifdef CONFIG_XFRM_SUB_POLICY
1652                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1653                         pols[1] = xfrm_policy_lookup_bytype(net,
1654                                                             XFRM_POLICY_TYPE_MAIN,
1655                                                             fl, family,
1656                                                             XFRM_POLICY_OUT);
1657                         if (pols[1]) {
1658                                 if (IS_ERR(pols[1])) {
1659                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1660                                         err = PTR_ERR(pols[1]);
1661                                         goto error;
1662                                 }
1663                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1664                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1665                                         err = -EPERM;
1666                                         goto error;
1667                                 }
1668                                 npols ++;
1669                                 xfrm_nr += pols[1]->xfrm_nr;
1670                         }
1671                 }
1672
1673                 /*
1674                  * Because neither flowi nor bundle information knows about
1675                  * transformation template size. On more than one policy usage
1676                  * we can realize whether all of them is bypass or not after
1677                  * they are searched. See above not-transformed bypass
1678                  * is surrounded by non-sub policy configuration, too.
1679                  */
1680                 if (xfrm_nr == 0) {
1681                         /* Flow passes not transformed. */
1682                         xfrm_pols_put(pols, npols);
1683                         return 0;
1684                 }
1685
1686 #endif
1687                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1688
1689                 if (unlikely(nx<0)) {
1690                         err = nx;
1691                         if (err == -EAGAIN && net->xfrm.sysctl_larval_drop) {
1692                                 /* EREMOTE tells the caller to generate
1693                                  * a one-shot blackhole route.
1694                                  */
1695                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1696                                 xfrm_pol_put(policy);
1697                                 return -EREMOTE;
1698                         }
1699                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1700                                 DECLARE_WAITQUEUE(wait, current);
1701
1702                                 add_wait_queue(&net->xfrm.km_waitq, &wait);
1703                                 set_current_state(TASK_INTERRUPTIBLE);
1704                                 schedule();
1705                                 set_current_state(TASK_RUNNING);
1706                                 remove_wait_queue(&net->xfrm.km_waitq, &wait);
1707
1708                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1709
1710                                 if (nx == -EAGAIN && signal_pending(current)) {
1711                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1712                                         err = -ERESTART;
1713                                         goto error;
1714                                 }
1715                                 if (nx == -EAGAIN ||
1716                                     genid != atomic_read(&flow_cache_genid)) {
1717                                         xfrm_pols_put(pols, npols);
1718                                         goto restart;
1719                                 }
1720                                 err = nx;
1721                         }
1722                         if (err < 0) {
1723                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1724                                 goto error;
1725                         }
1726                 }
1727                 if (nx == 0) {
1728                         /* Flow passes not transformed. */
1729                         xfrm_pols_put(pols, npols);
1730                         return 0;
1731                 }
1732
1733                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1734                 err = PTR_ERR(dst);
1735                 if (IS_ERR(dst)) {
1736                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1737                         goto error;
1738                 }
1739
1740                 for (pi = 0; pi < npols; pi++) {
1741                         read_lock_bh(&pols[pi]->lock);
1742                         pol_dead |= pols[pi]->walk.dead;
1743                         read_unlock_bh(&pols[pi]->lock);
1744                 }
1745
1746                 write_lock_bh(&policy->lock);
1747                 if (unlikely(pol_dead || stale_bundle(dst))) {
1748                         /* Wow! While we worked on resolving, this
1749                          * policy has gone. Retry. It is not paranoia,
1750                          * we just cannot enlist new bundle to dead object.
1751                          * We can't enlist stable bundles either.
1752                          */
1753                         write_unlock_bh(&policy->lock);
1754                         dst_free(dst);
1755
1756                         if (pol_dead)
1757                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLDEAD);
1758                         else
1759                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1760                         err = -EHOSTUNREACH;
1761                         goto error;
1762                 }
1763
1764                 if (npols > 1)
1765                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1766                 else
1767                         err = xfrm_dst_update_origin(dst, fl);
1768                 if (unlikely(err)) {
1769                         write_unlock_bh(&policy->lock);
1770                         dst_free(dst);
1771                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1772                         goto error;
1773                 }
1774
1775                 dst->next = policy->bundles;
1776                 policy->bundles = dst;
1777                 dst_hold(dst);
1778                 write_unlock_bh(&policy->lock);
1779         }
1780         *dst_p = dst;
1781         dst_release(dst_orig);
1782         xfrm_pols_put(pols, npols);
1783         return 0;
1784
1785 error:
1786         xfrm_pols_put(pols, npols);
1787 dropdst:
1788         dst_release(dst_orig);
1789         *dst_p = NULL;
1790         return err;
1791
1792 nopol:
1793         err = -ENOENT;
1794         if (flags & XFRM_LOOKUP_ICMP)
1795                 goto dropdst;
1796         return 0;
1797 }
1798 EXPORT_SYMBOL(__xfrm_lookup);
1799
1800 int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1801                 struct sock *sk, int flags)
1802 {
1803         int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1804
1805         if (err == -EREMOTE) {
1806                 dst_release(*dst_p);
1807                 *dst_p = NULL;
1808                 err = -EAGAIN;
1809         }
1810
1811         return err;
1812 }
1813 EXPORT_SYMBOL(xfrm_lookup);
1814
1815 static inline int
1816 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1817 {
1818         struct xfrm_state *x;
1819
1820         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1821                 return 0;
1822         x = skb->sp->xvec[idx];
1823         if (!x->type->reject)
1824                 return 0;
1825         return x->type->reject(x, skb, fl);
1826 }
1827
1828 /* When skb is transformed back to its "native" form, we have to
1829  * check policy restrictions. At the moment we make this in maximally
1830  * stupid way. Shame on me. :-) Of course, connected sockets must
1831  * have policy cached at them.
1832  */
1833
1834 static inline int
1835 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1836               unsigned short family)
1837 {
1838         if (xfrm_state_kern(x))
1839                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1840         return  x->id.proto == tmpl->id.proto &&
1841                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1842                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1843                 x->props.mode == tmpl->mode &&
1844                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1845                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1846                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1847                   xfrm_state_addr_cmp(tmpl, x, family));
1848 }
1849
1850 /*
1851  * 0 or more than 0 is returned when validation is succeeded (either bypass
1852  * because of optional transport mode, or next index of the mathced secpath
1853  * state with the template.
1854  * -1 is returned when no matching template is found.
1855  * Otherwise "-2 - errored_index" is returned.
1856  */
1857 static inline int
1858 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1859                unsigned short family)
1860 {
1861         int idx = start;
1862
1863         if (tmpl->optional) {
1864                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1865                         return start;
1866         } else
1867                 start = -1;
1868         for (; idx < sp->len; idx++) {
1869                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1870                         return ++idx;
1871                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1872                         if (start == -1)
1873                                 start = -2-idx;
1874                         break;
1875                 }
1876         }
1877         return start;
1878 }
1879
1880 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1881                           unsigned int family, int reverse)
1882 {
1883         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1884         int err;
1885
1886         if (unlikely(afinfo == NULL))
1887                 return -EAFNOSUPPORT;
1888
1889         afinfo->decode_session(skb, fl, reverse);
1890         err = security_xfrm_decode_session(skb, &fl->secid);
1891         xfrm_policy_put_afinfo(afinfo);
1892         return err;
1893 }
1894 EXPORT_SYMBOL(__xfrm_decode_session);
1895
1896 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1897 {
1898         for (; k < sp->len; k++) {
1899                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1900                         *idxp = k;
1901                         return 1;
1902                 }
1903         }
1904
1905         return 0;
1906 }
1907
1908 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1909                         unsigned short family)
1910 {
1911         struct net *net = dev_net(skb->dev);
1912         struct xfrm_policy *pol;
1913         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1914         int npols = 0;
1915         int xfrm_nr;
1916         int pi;
1917         int reverse;
1918         struct flowi fl;
1919         u8 fl_dir;
1920         int xerr_idx = -1;
1921
1922         reverse = dir & ~XFRM_POLICY_MASK;
1923         dir &= XFRM_POLICY_MASK;
1924         fl_dir = policy_to_flow_dir(dir);
1925
1926         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1927                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
1928                 return 0;
1929         }
1930
1931         nf_nat_decode_session(skb, &fl, family);
1932
1933         /* First, check used SA against their selectors. */
1934         if (skb->sp) {
1935                 int i;
1936
1937                 for (i=skb->sp->len-1; i>=0; i--) {
1938                         struct xfrm_state *x = skb->sp->xvec[i];
1939                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1940                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
1941                                 return 0;
1942                         }
1943                 }
1944         }
1945
1946         pol = NULL;
1947         if (sk && sk->sk_policy[dir]) {
1948                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1949                 if (IS_ERR(pol)) {
1950                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1951                         return 0;
1952                 }
1953         }
1954
1955         if (!pol)
1956                 pol = flow_cache_lookup(net, &fl, family, fl_dir,
1957                                         xfrm_policy_lookup);
1958
1959         if (IS_ERR(pol)) {
1960                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1961                 return 0;
1962         }
1963
1964         if (!pol) {
1965                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1966                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1967                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
1968                         return 0;
1969                 }
1970                 return 1;
1971         }
1972
1973         pol->curlft.use_time = get_seconds();
1974
1975         pols[0] = pol;
1976         npols ++;
1977 #ifdef CONFIG_XFRM_SUB_POLICY
1978         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1979                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
1980                                                     &fl, family,
1981                                                     XFRM_POLICY_IN);
1982                 if (pols[1]) {
1983                         if (IS_ERR(pols[1])) {
1984                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1985                                 return 0;
1986                         }
1987                         pols[1]->curlft.use_time = get_seconds();
1988                         npols ++;
1989                 }
1990         }
1991 #endif
1992
1993         if (pol->action == XFRM_POLICY_ALLOW) {
1994                 struct sec_path *sp;
1995                 static struct sec_path dummy;
1996                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1997                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1998                 struct xfrm_tmpl **tpp = tp;
1999                 int ti = 0;
2000                 int i, k;
2001
2002                 if ((sp = skb->sp) == NULL)
2003                         sp = &dummy;
2004
2005                 for (pi = 0; pi < npols; pi++) {
2006                         if (pols[pi] != pol &&
2007                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2008                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2009                                 goto reject;
2010                         }
2011                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2012                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2013                                 goto reject_error;
2014                         }
2015                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2016                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2017                 }
2018                 xfrm_nr = ti;
2019                 if (npols > 1) {
2020                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2021                         tpp = stp;
2022                 }
2023
2024                 /* For each tunnel xfrm, find the first matching tmpl.
2025                  * For each tmpl before that, find corresponding xfrm.
2026                  * Order is _important_. Later we will implement
2027                  * some barriers, but at the moment barriers
2028                  * are implied between each two transformations.
2029                  */
2030                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2031                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2032                         if (k < 0) {
2033                                 if (k < -1)
2034                                         /* "-2 - errored_index" returned */
2035                                         xerr_idx = -(2+k);
2036                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2037                                 goto reject;
2038                         }
2039                 }
2040
2041                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2042                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2043                         goto reject;
2044                 }
2045
2046                 xfrm_pols_put(pols, npols);
2047                 return 1;
2048         }
2049         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2050
2051 reject:
2052         xfrm_secpath_reject(xerr_idx, skb, &fl);
2053 reject_error:
2054         xfrm_pols_put(pols, npols);
2055         return 0;
2056 }
2057 EXPORT_SYMBOL(__xfrm_policy_check);
2058
2059 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2060 {
2061         struct net *net = dev_net(skb->dev);
2062         struct flowi fl;
2063         struct dst_entry *dst;
2064         int res;
2065
2066         if (xfrm_decode_session(skb, &fl, family) < 0) {
2067                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2068                 return 0;
2069         }
2070
2071         dst = skb_dst(skb);
2072
2073         res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
2074         skb_dst_set(skb, dst);
2075         return res;
2076 }
2077 EXPORT_SYMBOL(__xfrm_route_forward);
2078
2079 /* Optimize later using cookies and generation ids. */
2080
2081 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2082 {
2083         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2084          * to "-1" to force all XFRM destinations to get validated by
2085          * dst_ops->check on every use.  We do this because when a
2086          * normal route referenced by an XFRM dst is obsoleted we do
2087          * not go looking around for all parent referencing XFRM dsts
2088          * so that we can invalidate them.  It is just too much work.
2089          * Instead we make the checks here on every use.  For example:
2090          *
2091          *      XFRM dst A --> IPv4 dst X
2092          *
2093          * X is the "xdst->route" of A (X is also the "dst->path" of A
2094          * in this example).  If X is marked obsolete, "A" will not
2095          * notice.  That's what we are validating here via the
2096          * stale_bundle() check.
2097          *
2098          * When a policy's bundle is pruned, we dst_free() the XFRM
2099          * dst which causes it's ->obsolete field to be set to a
2100          * positive non-zero integer.  If an XFRM dst has been pruned
2101          * like this, we want to force a new route lookup.
2102          */
2103         if (dst->obsolete < 0 && !stale_bundle(dst))
2104                 return dst;
2105
2106         return NULL;
2107 }
2108
2109 static int stale_bundle(struct dst_entry *dst)
2110 {
2111         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2112 }
2113
2114 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2115 {
2116         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2117                 dst->dev = dev_net(dev)->loopback_dev;
2118                 dev_hold(dst->dev);
2119                 dev_put(dev);
2120         }
2121 }
2122 EXPORT_SYMBOL(xfrm_dst_ifdown);
2123
2124 static void xfrm_link_failure(struct sk_buff *skb)
2125 {
2126         /* Impossible. Such dst must be popped before reaches point of failure. */
2127         return;
2128 }
2129
2130 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2131 {
2132         if (dst) {
2133                 if (dst->obsolete) {
2134                         dst_release(dst);
2135                         dst = NULL;
2136                 }
2137         }
2138         return dst;
2139 }
2140
2141 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2142 {
2143         struct dst_entry *dst, **dstp;
2144
2145         write_lock(&pol->lock);
2146         dstp = &pol->bundles;
2147         while ((dst=*dstp) != NULL) {
2148                 if (func(dst)) {
2149                         *dstp = dst->next;
2150                         dst->next = *gc_list_p;
2151                         *gc_list_p = dst;
2152                 } else {
2153                         dstp = &dst->next;
2154                 }
2155         }
2156         write_unlock(&pol->lock);
2157 }
2158
2159 static void xfrm_prune_bundles(struct net *net, int (*func)(struct dst_entry *))
2160 {
2161         struct dst_entry *gc_list = NULL;
2162         int dir;
2163
2164         read_lock_bh(&xfrm_policy_lock);
2165         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2166                 struct xfrm_policy *pol;
2167                 struct hlist_node *entry;
2168                 struct hlist_head *table;
2169                 int i;
2170
2171                 hlist_for_each_entry(pol, entry,
2172                                      &net->xfrm.policy_inexact[dir], bydst)
2173                         prune_one_bundle(pol, func, &gc_list);
2174
2175                 table = net->xfrm.policy_bydst[dir].table;
2176                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2177                         hlist_for_each_entry(pol, entry, table + i, bydst)
2178                                 prune_one_bundle(pol, func, &gc_list);
2179                 }
2180         }
2181         read_unlock_bh(&xfrm_policy_lock);
2182
2183         while (gc_list) {
2184                 struct dst_entry *dst = gc_list;
2185                 gc_list = dst->next;
2186                 dst_free(dst);
2187         }
2188 }
2189
2190 static int unused_bundle(struct dst_entry *dst)
2191 {
2192         return !atomic_read(&dst->__refcnt);
2193 }
2194
2195 static void __xfrm_garbage_collect(struct net *net)
2196 {
2197         xfrm_prune_bundles(net, unused_bundle);
2198 }
2199
2200 static int xfrm_flush_bundles(struct net *net)
2201 {
2202         xfrm_prune_bundles(net, stale_bundle);
2203         return 0;
2204 }
2205
2206 static void xfrm_init_pmtu(struct dst_entry *dst)
2207 {
2208         do {
2209                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2210                 u32 pmtu, route_mtu_cached;
2211
2212                 pmtu = dst_mtu(dst->child);
2213                 xdst->child_mtu_cached = pmtu;
2214
2215                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2216
2217                 route_mtu_cached = dst_mtu(xdst->route);
2218                 xdst->route_mtu_cached = route_mtu_cached;
2219
2220                 if (pmtu > route_mtu_cached)
2221                         pmtu = route_mtu_cached;
2222
2223                 dst->metrics[RTAX_MTU-1] = pmtu;
2224         } while ((dst = dst->next));
2225 }
2226
2227 /* Check that the bundle accepts the flow and its components are
2228  * still valid.
2229  */
2230
2231 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2232                 struct flowi *fl, int family, int strict)
2233 {
2234         struct dst_entry *dst = &first->u.dst;
2235         struct xfrm_dst *last;
2236         u32 mtu;
2237
2238         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2239             (dst->dev && !netif_running(dst->dev)))
2240                 return 0;
2241 #ifdef CONFIG_XFRM_SUB_POLICY
2242         if (fl) {
2243                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2244                         return 0;
2245                 if (first->partner &&
2246                     !xfrm_selector_match(first->partner, fl, family))
2247                         return 0;
2248         }
2249 #endif
2250
2251         last = NULL;
2252
2253         do {
2254                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2255
2256                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2257                         return 0;
2258                 if (fl && pol &&
2259                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2260                         return 0;
2261                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2262                         return 0;
2263                 if (xdst->genid != dst->xfrm->genid)
2264                         return 0;
2265
2266                 if (strict && fl &&
2267                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2268                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2269                         return 0;
2270
2271                 mtu = dst_mtu(dst->child);
2272                 if (xdst->child_mtu_cached != mtu) {
2273                         last = xdst;
2274                         xdst->child_mtu_cached = mtu;
2275                 }
2276
2277                 if (!dst_check(xdst->route, xdst->route_cookie))
2278                         return 0;
2279                 mtu = dst_mtu(xdst->route);
2280                 if (xdst->route_mtu_cached != mtu) {
2281                         last = xdst;
2282                         xdst->route_mtu_cached = mtu;
2283                 }
2284
2285                 dst = dst->child;
2286         } while (dst->xfrm);
2287
2288         if (likely(!last))
2289                 return 1;
2290
2291         mtu = last->child_mtu_cached;
2292         for (;;) {
2293                 dst = &last->u.dst;
2294
2295                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2296                 if (mtu > last->route_mtu_cached)
2297                         mtu = last->route_mtu_cached;
2298                 dst->metrics[RTAX_MTU-1] = mtu;
2299
2300                 if (last == first)
2301                         break;
2302
2303                 last = (struct xfrm_dst *)last->u.dst.next;
2304                 last->child_mtu_cached = mtu;
2305         }
2306
2307         return 1;
2308 }
2309
2310 EXPORT_SYMBOL(xfrm_bundle_ok);
2311
2312 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2313 {
2314         struct net *net;
2315         int err = 0;
2316         if (unlikely(afinfo == NULL))
2317                 return -EINVAL;
2318         if (unlikely(afinfo->family >= NPROTO))
2319                 return -EAFNOSUPPORT;
2320         write_lock_bh(&xfrm_policy_afinfo_lock);
2321         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2322                 err = -ENOBUFS;
2323         else {
2324                 struct dst_ops *dst_ops = afinfo->dst_ops;
2325                 if (likely(dst_ops->kmem_cachep == NULL))
2326                         dst_ops->kmem_cachep = xfrm_dst_cache;
2327                 if (likely(dst_ops->check == NULL))
2328                         dst_ops->check = xfrm_dst_check;
2329                 if (likely(dst_ops->negative_advice == NULL))
2330                         dst_ops->negative_advice = xfrm_negative_advice;
2331                 if (likely(dst_ops->link_failure == NULL))
2332                         dst_ops->link_failure = xfrm_link_failure;
2333                 if (likely(afinfo->garbage_collect == NULL))
2334                         afinfo->garbage_collect = __xfrm_garbage_collect;
2335                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2336         }
2337         write_unlock_bh(&xfrm_policy_afinfo_lock);
2338
2339         rtnl_lock();
2340         for_each_net(net) {
2341                 struct dst_ops *xfrm_dst_ops;
2342
2343                 switch (afinfo->family) {
2344                 case AF_INET:
2345                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2346                         break;
2347 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2348                 case AF_INET6:
2349                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2350                         break;
2351 #endif
2352                 default:
2353                         BUG();
2354                 }
2355                 *xfrm_dst_ops = *afinfo->dst_ops;
2356         }
2357         rtnl_unlock();
2358
2359         return err;
2360 }
2361 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2362
2363 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2364 {
2365         int err = 0;
2366         if (unlikely(afinfo == NULL))
2367                 return -EINVAL;
2368         if (unlikely(afinfo->family >= NPROTO))
2369                 return -EAFNOSUPPORT;
2370         write_lock_bh(&xfrm_policy_afinfo_lock);
2371         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2372                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2373                         err = -EINVAL;
2374                 else {
2375                         struct dst_ops *dst_ops = afinfo->dst_ops;
2376                         xfrm_policy_afinfo[afinfo->family] = NULL;
2377                         dst_ops->kmem_cachep = NULL;
2378                         dst_ops->check = NULL;
2379                         dst_ops->negative_advice = NULL;
2380                         dst_ops->link_failure = NULL;
2381                         afinfo->garbage_collect = NULL;
2382                 }
2383         }
2384         write_unlock_bh(&xfrm_policy_afinfo_lock);
2385         return err;
2386 }
2387 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2388
2389 static void __net_init xfrm_dst_ops_init(struct net *net)
2390 {
2391         struct xfrm_policy_afinfo *afinfo;
2392
2393         read_lock_bh(&xfrm_policy_afinfo_lock);
2394         afinfo = xfrm_policy_afinfo[AF_INET];
2395         if (afinfo)
2396                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2397 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2398         afinfo = xfrm_policy_afinfo[AF_INET6];
2399         if (afinfo)
2400                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2401 #endif
2402         read_unlock_bh(&xfrm_policy_afinfo_lock);
2403 }
2404
2405 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2406 {
2407         struct xfrm_policy_afinfo *afinfo;
2408         if (unlikely(family >= NPROTO))
2409                 return NULL;
2410         read_lock(&xfrm_policy_afinfo_lock);
2411         afinfo = xfrm_policy_afinfo[family];
2412         if (unlikely(!afinfo))
2413                 read_unlock(&xfrm_policy_afinfo_lock);
2414         return afinfo;
2415 }
2416
2417 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2418 {
2419         read_unlock(&xfrm_policy_afinfo_lock);
2420 }
2421
2422 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2423 {
2424         struct net_device *dev = ptr;
2425
2426         switch (event) {
2427         case NETDEV_DOWN:
2428                 xfrm_flush_bundles(dev_net(dev));
2429         }
2430         return NOTIFY_DONE;
2431 }
2432
2433 static struct notifier_block xfrm_dev_notifier = {
2434         .notifier_call  = xfrm_dev_event,
2435 };
2436
2437 #ifdef CONFIG_XFRM_STATISTICS
2438 static int __net_init xfrm_statistics_init(struct net *net)
2439 {
2440         int rv;
2441
2442         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2443                           sizeof(struct linux_xfrm_mib)) < 0)
2444                 return -ENOMEM;
2445         rv = xfrm_proc_init(net);
2446         if (rv < 0)
2447                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2448         return rv;
2449 }
2450
2451 static void xfrm_statistics_fini(struct net *net)
2452 {
2453         xfrm_proc_fini(net);
2454         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2455 }
2456 #else
2457 static int __net_init xfrm_statistics_init(struct net *net)
2458 {
2459         return 0;
2460 }
2461
2462 static void xfrm_statistics_fini(struct net *net)
2463 {
2464 }
2465 #endif
2466
2467 static int __net_init xfrm_policy_init(struct net *net)
2468 {
2469         unsigned int hmask, sz;
2470         int dir;
2471
2472         if (net_eq(net, &init_net))
2473                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2474                                            sizeof(struct xfrm_dst),
2475                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2476                                            NULL);
2477
2478         hmask = 8 - 1;
2479         sz = (hmask+1) * sizeof(struct hlist_head);
2480
2481         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2482         if (!net->xfrm.policy_byidx)
2483                 goto out_byidx;
2484         net->xfrm.policy_idx_hmask = hmask;
2485
2486         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2487                 struct xfrm_policy_hash *htab;
2488
2489                 net->xfrm.policy_count[dir] = 0;
2490                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2491
2492                 htab = &net->xfrm.policy_bydst[dir];
2493                 htab->table = xfrm_hash_alloc(sz);
2494                 if (!htab->table)
2495                         goto out_bydst;
2496                 htab->hmask = hmask;
2497         }
2498
2499         INIT_LIST_HEAD(&net->xfrm.policy_all);
2500         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2501         if (net_eq(net, &init_net))
2502                 register_netdevice_notifier(&xfrm_dev_notifier);
2503         return 0;
2504
2505 out_bydst:
2506         for (dir--; dir >= 0; dir--) {
2507                 struct xfrm_policy_hash *htab;
2508
2509                 htab = &net->xfrm.policy_bydst[dir];
2510                 xfrm_hash_free(htab->table, sz);
2511         }
2512         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2513 out_byidx:
2514         return -ENOMEM;
2515 }
2516
2517 static void xfrm_policy_fini(struct net *net)
2518 {
2519         struct xfrm_audit audit_info;
2520         unsigned int sz;
2521         int dir;
2522
2523         flush_work(&net->xfrm.policy_hash_work);
2524 #ifdef CONFIG_XFRM_SUB_POLICY
2525         audit_info.loginuid = -1;
2526         audit_info.sessionid = -1;
2527         audit_info.secid = 0;
2528         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2529 #endif
2530         audit_info.loginuid = -1;
2531         audit_info.sessionid = -1;
2532         audit_info.secid = 0;
2533         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2534         flush_work(&xfrm_policy_gc_work);
2535
2536         WARN_ON(!list_empty(&net->xfrm.policy_all));
2537
2538         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2539                 struct xfrm_policy_hash *htab;
2540
2541                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2542
2543                 htab = &net->xfrm.policy_bydst[dir];
2544                 sz = (htab->hmask + 1);
2545                 WARN_ON(!hlist_empty(htab->table));
2546                 xfrm_hash_free(htab->table, sz);
2547         }
2548
2549         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2550         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2551         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2552 }
2553
2554 static int __net_init xfrm_net_init(struct net *net)
2555 {
2556         int rv;
2557
2558         rv = xfrm_statistics_init(net);
2559         if (rv < 0)
2560                 goto out_statistics;
2561         rv = xfrm_state_init(net);
2562         if (rv < 0)
2563                 goto out_state;
2564         rv = xfrm_policy_init(net);
2565         if (rv < 0)
2566                 goto out_policy;
2567         xfrm_dst_ops_init(net);
2568         rv = xfrm_sysctl_init(net);
2569         if (rv < 0)
2570                 goto out_sysctl;
2571         return 0;
2572
2573 out_sysctl:
2574         xfrm_policy_fini(net);
2575 out_policy:
2576         xfrm_state_fini(net);
2577 out_state:
2578         xfrm_statistics_fini(net);
2579 out_statistics:
2580         return rv;
2581 }
2582
2583 static void __net_exit xfrm_net_exit(struct net *net)
2584 {
2585         xfrm_sysctl_fini(net);
2586         xfrm_policy_fini(net);
2587         xfrm_state_fini(net);
2588         xfrm_statistics_fini(net);
2589 }
2590
2591 static struct pernet_operations __net_initdata xfrm_net_ops = {
2592         .init = xfrm_net_init,
2593         .exit = xfrm_net_exit,
2594 };
2595
2596 void __init xfrm_init(void)
2597 {
2598         register_pernet_subsys(&xfrm_net_ops);
2599         xfrm_input_init();
2600 }
2601
2602 #ifdef CONFIG_AUDITSYSCALL
2603 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2604                                          struct audit_buffer *audit_buf)
2605 {
2606         struct xfrm_sec_ctx *ctx = xp->security;
2607         struct xfrm_selector *sel = &xp->selector;
2608
2609         if (ctx)
2610                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2611                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2612
2613         switch(sel->family) {
2614         case AF_INET:
2615                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2616                 if (sel->prefixlen_s != 32)
2617                         audit_log_format(audit_buf, " src_prefixlen=%d",
2618                                          sel->prefixlen_s);
2619                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2620                 if (sel->prefixlen_d != 32)
2621                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2622                                          sel->prefixlen_d);
2623                 break;
2624         case AF_INET6:
2625                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2626                 if (sel->prefixlen_s != 128)
2627                         audit_log_format(audit_buf, " src_prefixlen=%d",
2628                                          sel->prefixlen_s);
2629                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2630                 if (sel->prefixlen_d != 128)
2631                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2632                                          sel->prefixlen_d);
2633                 break;
2634         }
2635 }
2636
2637 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2638                            uid_t auid, u32 sessionid, u32 secid)
2639 {
2640         struct audit_buffer *audit_buf;
2641
2642         audit_buf = xfrm_audit_start("SPD-add");
2643         if (audit_buf == NULL)
2644                 return;
2645         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2646         audit_log_format(audit_buf, " res=%u", result);
2647         xfrm_audit_common_policyinfo(xp, audit_buf);
2648         audit_log_end(audit_buf);
2649 }
2650 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2651
2652 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2653                               uid_t auid, u32 sessionid, u32 secid)
2654 {
2655         struct audit_buffer *audit_buf;
2656
2657         audit_buf = xfrm_audit_start("SPD-delete");
2658         if (audit_buf == NULL)
2659                 return;
2660         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2661         audit_log_format(audit_buf, " res=%u", result);
2662         xfrm_audit_common_policyinfo(xp, audit_buf);
2663         audit_log_end(audit_buf);
2664 }
2665 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2666 #endif
2667
2668 #ifdef CONFIG_XFRM_MIGRATE
2669 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2670                                        struct xfrm_selector *sel_tgt)
2671 {
2672         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2673                 if (sel_tgt->family == sel_cmp->family &&
2674                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2675                                   sel_cmp->family) == 0 &&
2676                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2677                                   sel_cmp->family) == 0 &&
2678                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2679                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2680                         return 1;
2681                 }
2682         } else {
2683                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2684                         return 1;
2685                 }
2686         }
2687         return 0;
2688 }
2689
2690 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2691                                                      u8 dir, u8 type)
2692 {
2693         struct xfrm_policy *pol, *ret = NULL;
2694         struct hlist_node *entry;
2695         struct hlist_head *chain;
2696         u32 priority = ~0U;
2697
2698         read_lock_bh(&xfrm_policy_lock);
2699         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2700         hlist_for_each_entry(pol, entry, chain, bydst) {
2701                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2702                     pol->type == type) {
2703                         ret = pol;
2704                         priority = ret->priority;
2705                         break;
2706                 }
2707         }
2708         chain = &init_net.xfrm.policy_inexact[dir];
2709         hlist_for_each_entry(pol, entry, chain, bydst) {
2710                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2711                     pol->type == type &&
2712                     pol->priority < priority) {
2713                         ret = pol;
2714                         break;
2715                 }
2716         }
2717
2718         if (ret)
2719                 xfrm_pol_hold(ret);
2720
2721         read_unlock_bh(&xfrm_policy_lock);
2722
2723         return ret;
2724 }
2725
2726 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2727 {
2728         int match = 0;
2729
2730         if (t->mode == m->mode && t->id.proto == m->proto &&
2731             (m->reqid == 0 || t->reqid == m->reqid)) {
2732                 switch (t->mode) {
2733                 case XFRM_MODE_TUNNEL:
2734                 case XFRM_MODE_BEET:
2735                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2736                                           m->old_family) == 0 &&
2737                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2738                                           m->old_family) == 0) {
2739                                 match = 1;
2740                         }
2741                         break;
2742                 case XFRM_MODE_TRANSPORT:
2743                         /* in case of transport mode, template does not store
2744                            any IP addresses, hence we just compare mode and
2745                            protocol */
2746                         match = 1;
2747                         break;
2748                 default:
2749                         break;
2750                 }
2751         }
2752         return match;
2753 }
2754
2755 /* update endpoint address(es) of template(s) */
2756 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2757                                struct xfrm_migrate *m, int num_migrate)
2758 {
2759         struct xfrm_migrate *mp;
2760         struct dst_entry *dst;
2761         int i, j, n = 0;
2762
2763         write_lock_bh(&pol->lock);
2764         if (unlikely(pol->walk.dead)) {
2765                 /* target policy has been deleted */
2766                 write_unlock_bh(&pol->lock);
2767                 return -ENOENT;
2768         }
2769
2770         for (i = 0; i < pol->xfrm_nr; i++) {
2771                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2772                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2773                                 continue;
2774                         n++;
2775                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2776                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2777                                 continue;
2778                         /* update endpoints */
2779                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2780                                sizeof(pol->xfrm_vec[i].id.daddr));
2781                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2782                                sizeof(pol->xfrm_vec[i].saddr));
2783                         pol->xfrm_vec[i].encap_family = mp->new_family;
2784                         /* flush bundles */
2785                         while ((dst = pol->bundles) != NULL) {
2786                                 pol->bundles = dst->next;
2787                                 dst_free(dst);
2788                         }
2789                 }
2790         }
2791
2792         write_unlock_bh(&pol->lock);
2793
2794         if (!n)
2795                 return -ENODATA;
2796
2797         return 0;
2798 }
2799
2800 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2801 {
2802         int i, j;
2803
2804         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2805                 return -EINVAL;
2806
2807         for (i = 0; i < num_migrate; i++) {
2808                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2809                                    m[i].old_family) == 0) &&
2810                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2811                                    m[i].old_family) == 0))
2812                         return -EINVAL;
2813                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2814                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2815                         return -EINVAL;
2816
2817                 /* check if there is any duplicated entry */
2818                 for (j = i + 1; j < num_migrate; j++) {
2819                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2820                                     sizeof(m[i].old_daddr)) &&
2821                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2822                                     sizeof(m[i].old_saddr)) &&
2823                             m[i].proto == m[j].proto &&
2824                             m[i].mode == m[j].mode &&
2825                             m[i].reqid == m[j].reqid &&
2826                             m[i].old_family == m[j].old_family)
2827                                 return -EINVAL;
2828                 }
2829         }
2830
2831         return 0;
2832 }
2833
2834 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2835                  struct xfrm_migrate *m, int num_migrate,
2836                  struct xfrm_kmaddress *k)
2837 {
2838         int i, err, nx_cur = 0, nx_new = 0;
2839         struct xfrm_policy *pol = NULL;
2840         struct xfrm_state *x, *xc;
2841         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2842         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2843         struct xfrm_migrate *mp;
2844
2845         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2846                 goto out;
2847
2848         /* Stage 1 - find policy */
2849         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2850                 err = -ENOENT;
2851                 goto out;
2852         }
2853
2854         /* Stage 2 - find and update state(s) */
2855         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2856                 if ((x = xfrm_migrate_state_find(mp))) {
2857                         x_cur[nx_cur] = x;
2858                         nx_cur++;
2859                         if ((xc = xfrm_state_migrate(x, mp))) {
2860                                 x_new[nx_new] = xc;
2861                                 nx_new++;
2862                         } else {
2863                                 err = -ENODATA;
2864                                 goto restore_state;
2865                         }
2866                 }
2867         }
2868
2869         /* Stage 3 - update policy */
2870         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2871                 goto restore_state;
2872
2873         /* Stage 4 - delete old state(s) */
2874         if (nx_cur) {
2875                 xfrm_states_put(x_cur, nx_cur);
2876                 xfrm_states_delete(x_cur, nx_cur);
2877         }
2878
2879         /* Stage 5 - announce */
2880         km_migrate(sel, dir, type, m, num_migrate, k);
2881
2882         xfrm_pol_put(pol);
2883
2884         return 0;
2885 out:
2886         return err;
2887
2888 restore_state:
2889         if (pol)
2890                 xfrm_pol_put(pol);
2891         if (nx_cur)
2892                 xfrm_states_put(x_cur, nx_cur);
2893         if (nx_new)
2894                 xfrm_states_delete(x_new, nx_new);
2895
2896         return err;
2897 }
2898 EXPORT_SYMBOL(xfrm_migrate);
2899 #endif