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_SPINLOCK(xfrm_policy_sk_bundle_lock);
  41 static struct dst_entry *xfrm_policy_sk_bundles;
  42 static DEFINE_RWLOCK(xfrm_policy_lock);
  43
  44 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  45 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  46
  47 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  48
  49 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  50 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  51 static void xfrm_init_pmtu(struct dst_entry *dst);
  52 static int stale_bundle(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 (unlikely(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 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 220 {
 221         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 222
 223         if (unlikely(pol->walk.dead))
 224                 flo = NULL;
 225         else
 226                 xfrm_pol_hold(pol);
 227
 228         return flo;
 229 }
 230
 231 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 232 {
 233         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 234
 235         return !pol->walk.dead;
 236 }
 237
 238 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 239 {
 240         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 241 }
 242
 243 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 244         .get = xfrm_policy_flo_get,
 245         .check = xfrm_policy_flo_check,
 246         .delete = xfrm_policy_flo_delete,
 247 };
 248
 249 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 250  * SPD calls.
 251  */
 252
 253 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 254 {
 255         struct xfrm_policy *policy;
 256
 257         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 258
 259         if (policy) {
 260                 write_pnet(&policy->xp_net, net);
 261                 INIT_LIST_HEAD(&policy->walk.all);
 262                 INIT_HLIST_NODE(&policy->bydst);
 263                 INIT_HLIST_NODE(&policy->byidx);
 264                 rwlock_init(&policy->lock);
 265                 atomic_set(&policy->refcnt, 1);
 266                 setup_timer(&policy->timer, xfrm_policy_timer,
 267                                 (unsigned long)policy);
 268                 policy->flo.ops = &xfrm_policy_fc_ops;
 269         }
 270         return policy;
 271 }
 272 EXPORT_SYMBOL(xfrm_policy_alloc);
 273
 274 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 275
 276 void xfrm_policy_destroy(struct xfrm_policy *policy)
 277 {
 278         BUG_ON(!policy->walk.dead);
 279
 280         if (del_timer(&policy->timer))
 281                 BUG();
 282
 283         security_xfrm_policy_free(policy->security);
 284         kfree(policy);
 285 }
 286 EXPORT_SYMBOL(xfrm_policy_destroy);
 287
 288 /* Rule must be locked. Release descentant resources, announce
 289  * entry dead. The rule must be unlinked from lists to the moment.
 290  */
 291
 292 static void xfrm_policy_kill(struct xfrm_policy *policy)
 293 {
 294         policy->walk.dead = 1;
 295
 296         atomic_inc(&policy->genid);
 297
 298         if (del_timer(&policy->timer))
 299                 xfrm_pol_put(policy);
 300
 301         xfrm_pol_put(policy);
 302 }
 303
 304 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 305
 306 static inline unsigned int idx_hash(struct net *net, u32 index)
 307 {
 308         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 309 }
 310
 311 static struct hlist_head *policy_hash_bysel(struct net *net, struct xfrm_selector *sel, unsigned short family, int dir)
 312 {
 313         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 314         unsigned int hash = __sel_hash(sel, family, hmask);
 315
 316         return (hash == hmask + 1 ?
 317                 &net->xfrm.policy_inexact[dir] :
 318                 net->xfrm.policy_bydst[dir].table + hash);
 319 }
 320
 321 static struct hlist_head *policy_hash_direct(struct net *net, xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 322 {
 323         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 324         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 325
 326         return net->xfrm.policy_bydst[dir].table + hash;
 327 }
 328
 329 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 330                                    struct hlist_head *ndsttable,
 331                                    unsigned int nhashmask)
 332 {
 333         struct hlist_node *entry, *tmp, *entry0 = NULL;
 334         struct xfrm_policy *pol;
 335         unsigned int h0 = 0;
 336
 337 redo:
 338         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 339                 unsigned int h;
 340
 341                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 342                                 pol->family, nhashmask);
 343                 if (!entry0) {
 344                         hlist_del(entry);
 345                         hlist_add_head(&pol->bydst, ndsttable+h);
 346                         h0 = h;
 347                 } else {
 348                         if (h != h0)
 349                                 continue;
 350                         hlist_del(entry);
 351                         hlist_add_after(entry0, &pol->bydst);
 352                 }
 353                 entry0 = entry;
 354         }
 355         if (!hlist_empty(list)) {
 356                 entry0 = NULL;
 357                 goto redo;
 358         }
 359 }
 360
 361 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 362                                    struct hlist_head *nidxtable,
 363                                    unsigned int nhashmask)
 364 {
 365         struct hlist_node *entry, *tmp;
 366         struct xfrm_policy *pol;
 367
 368         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 369                 unsigned int h;
 370
 371                 h = __idx_hash(pol->index, nhashmask);
 372                 hlist_add_head(&pol->byidx, nidxtable+h);
 373         }
 374 }
 375
 376 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 377 {
 378         return ((old_hmask + 1) << 1) - 1;
 379 }
 380
 381 static void xfrm_bydst_resize(struct net *net, int dir)
 382 {
 383         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 384         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 385         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 386         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 387         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 388         int i;
 389
 390         if (!ndst)
 391                 return;
 392
 393         write_lock_bh(&xfrm_policy_lock);
 394
 395         for (i = hmask; i >= 0; i--)
 396                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 397
 398         net->xfrm.policy_bydst[dir].table = ndst;
 399         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 400
 401         write_unlock_bh(&xfrm_policy_lock);
 402
 403         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 404 }
 405
 406 static void xfrm_byidx_resize(struct net *net, int total)
 407 {
 408         unsigned int hmask = net->xfrm.policy_idx_hmask;
 409         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 410         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 411         struct hlist_head *oidx = net->xfrm.policy_byidx;
 412         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 413         int i;
 414
 415         if (!nidx)
 416                 return;
 417
 418         write_lock_bh(&xfrm_policy_lock);
 419
 420         for (i = hmask; i >= 0; i--)
 421                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 422
 423         net->xfrm.policy_byidx = nidx;
 424         net->xfrm.policy_idx_hmask = nhashmask;
 425
 426         write_unlock_bh(&xfrm_policy_lock);
 427
 428         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 429 }
 430
 431 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 432 {
 433         unsigned int cnt = net->xfrm.policy_count[dir];
 434         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 435
 436         if (total)
 437                 *total += cnt;
 438
 439         if ((hmask + 1) < xfrm_policy_hashmax &&
 440             cnt > hmask)
 441                 return 1;
 442
 443         return 0;
 444 }
 445
 446 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 447 {
 448         unsigned int hmask = net->xfrm.policy_idx_hmask;
 449
 450         if ((hmask + 1) < xfrm_policy_hashmax &&
 451             total > hmask)
 452                 return 1;
 453
 454         return 0;
 455 }
 456
 457 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 458 {
 459         read_lock_bh(&xfrm_policy_lock);
 460         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 461         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 462         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 463         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 464         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 465         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 466         si->spdhcnt = net->xfrm.policy_idx_hmask;
 467         si->spdhmcnt = xfrm_policy_hashmax;
 468         read_unlock_bh(&xfrm_policy_lock);
 469 }
 470 EXPORT_SYMBOL(xfrm_spd_getinfo);
 471
 472 static DEFINE_MUTEX(hash_resize_mutex);
 473 static void xfrm_hash_resize(struct work_struct *work)
 474 {
 475         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 476         int dir, total;
 477
 478         mutex_lock(&hash_resize_mutex);
 479
 480         total = 0;
 481         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 482                 if (xfrm_bydst_should_resize(net, dir, &total))
 483                         xfrm_bydst_resize(net, dir);
 484         }
 485         if (xfrm_byidx_should_resize(net, total))
 486                 xfrm_byidx_resize(net, total);
 487
 488         mutex_unlock(&hash_resize_mutex);
 489 }
 490
 491 /* Generate new index... KAME seems to generate them ordered by cost
 492  * of an absolute inpredictability of ordering of rules. This will not pass. */
 493 static u32 xfrm_gen_index(struct net *net, int dir)
 494 {
 495         static u32 idx_generator;
 496
 497         for (;;) {
 498                 struct hlist_node *entry;
 499                 struct hlist_head *list;
 500                 struct xfrm_policy *p;
 501                 u32 idx;
 502                 int found;
 503
 504                 idx = (idx_generator | dir);
 505                 idx_generator += 8;
 506                 if (idx == 0)
 507                         idx = 8;
 508                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 509                 found = 0;
 510                 hlist_for_each_entry(p, entry, list, byidx) {
 511                         if (p->index == idx) {
 512                                 found = 1;
 513                                 break;
 514                         }
 515                 }
 516                 if (!found)
 517                         return idx;
 518         }
 519 }
 520
 521 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 522 {
 523         u32 *p1 = (u32 *) s1;
 524         u32 *p2 = (u32 *) s2;
 525         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 526         int i;
 527
 528         for (i = 0; i < len; i++) {
 529                 if (p1[i] != p2[i])
 530                         return 1;
 531         }
 532
 533         return 0;
 534 }
 535
 536 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 537 {
 538         struct net *net = xp_net(policy);
 539         struct xfrm_policy *pol;
 540         struct xfrm_policy *delpol;
 541         struct hlist_head *chain;
 542         struct hlist_node *entry, *newpos;
 543         u32 mark = policy->mark.v & policy->mark.m;
 544
 545         write_lock_bh(&xfrm_policy_lock);
 546         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 547         delpol = NULL;
 548         newpos = NULL;
 549         hlist_for_each_entry(pol, entry, chain, bydst) {
 550                 if (pol->type == policy->type &&
 551                     !selector_cmp(&pol->selector, &policy->selector) &&
 552                     (mark & pol->mark.m) == pol->mark.v &&
 553                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 554                     !WARN_ON(delpol)) {
 555                         if (excl) {
 556                                 write_unlock_bh(&xfrm_policy_lock);
 557                                 return -EEXIST;
 558                         }
 559                         delpol = pol;
 560                         if (policy->priority > pol->priority)
 561                                 continue;
 562                 } else if (policy->priority >= pol->priority) {
 563                         newpos = &pol->bydst;
 564                         continue;
 565                 }
 566                 if (delpol)
 567                         break;
 568         }
 569         if (newpos)
 570                 hlist_add_after(newpos, &policy->bydst);
 571         else
 572                 hlist_add_head(&policy->bydst, chain);
 573         xfrm_pol_hold(policy);
 574         net->xfrm.policy_count[dir]++;
 575         atomic_inc(&flow_cache_genid);
 576         if (delpol)
 577                 __xfrm_policy_unlink(delpol, dir);
 578         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 579         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 580         policy->curlft.add_time = get_seconds();
 581         policy->curlft.use_time = 0;
 582         if (!mod_timer(&policy->timer, jiffies + HZ))
 583                 xfrm_pol_hold(policy);
 584         list_add(&policy->walk.all, &net->xfrm.policy_all);
 585         write_unlock_bh(&xfrm_policy_lock);
 586
 587         if (delpol)
 588                 xfrm_policy_kill(delpol);
 589         else if (xfrm_bydst_should_resize(net, dir, NULL))
 590                 schedule_work(&net->xfrm.policy_hash_work);
 591
 592         return 0;
 593 }
 594 EXPORT_SYMBOL(xfrm_policy_insert);
 595
 596 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 597                                           int dir, struct xfrm_selector *sel,
 598                                           struct xfrm_sec_ctx *ctx, int delete,
 599                                           int *err)
 600 {
 601         struct xfrm_policy *pol, *ret;
 602         struct hlist_head *chain;
 603         struct hlist_node *entry;
 604
 605         *err = 0;
 606         write_lock_bh(&xfrm_policy_lock);
 607         chain = policy_hash_bysel(net, sel, sel->family, dir);
 608         ret = NULL;
 609         hlist_for_each_entry(pol, entry, chain, bydst) {
 610                 if (pol->type == type &&
 611                     (mark & pol->mark.m) == pol->mark.v &&
 612                     !selector_cmp(sel, &pol->selector) &&
 613                     xfrm_sec_ctx_match(ctx, pol->security)) {
 614                         xfrm_pol_hold(pol);
 615                         if (delete) {
 616                                 *err = security_xfrm_policy_delete(
 617                                                                 pol->security);
 618                                 if (*err) {
 619                                         write_unlock_bh(&xfrm_policy_lock);
 620                                         return pol;
 621                                 }
 622                                 __xfrm_policy_unlink(pol, dir);
 623                         }
 624                         ret = pol;
 625                         break;
 626                 }
 627         }
 628         write_unlock_bh(&xfrm_policy_lock);
 629
 630         if (ret && delete)
 631                 xfrm_policy_kill(ret);
 632         return ret;
 633 }
 634 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 635
 636 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 637                                      int dir, u32 id, int delete, int *err)
 638 {
 639         struct xfrm_policy *pol, *ret;
 640         struct hlist_head *chain;
 641         struct hlist_node *entry;
 642
 643         *err = -ENOENT;
 644         if (xfrm_policy_id2dir(id) != dir)
 645                 return NULL;
 646
 647         *err = 0;
 648         write_lock_bh(&xfrm_policy_lock);
 649         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 650         ret = NULL;
 651         hlist_for_each_entry(pol, entry, chain, byidx) {
 652                 if (pol->type == type && pol->index == id &&
 653                     (mark & pol->mark.m) == pol->mark.v) {
 654                         xfrm_pol_hold(pol);
 655                         if (delete) {
 656                                 *err = security_xfrm_policy_delete(
 657                                                                 pol->security);
 658                                 if (*err) {
 659                                         write_unlock_bh(&xfrm_policy_lock);
 660                                         return pol;
 661                                 }
 662                                 __xfrm_policy_unlink(pol, dir);
 663                         }
 664                         ret = pol;
 665                         break;
 666                 }
 667         }
 668         write_unlock_bh(&xfrm_policy_lock);
 669
 670         if (ret && delete)
 671                 xfrm_policy_kill(ret);
 672         return ret;
 673 }
 674 EXPORT_SYMBOL(xfrm_policy_byid);
 675
 676 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 677 static inline int
 678 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 679 {
 680         int dir, err = 0;
 681
 682         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 683                 struct xfrm_policy *pol;
 684                 struct hlist_node *entry;
 685                 int i;
 686
 687                 hlist_for_each_entry(pol, entry,
 688                                      &net->xfrm.policy_inexact[dir], bydst) {
 689                         if (pol->type != type)
 690                                 continue;
 691                         err = security_xfrm_policy_delete(pol->security);
 692                         if (err) {
 693                                 xfrm_audit_policy_delete(pol, 0,
 694                                                          audit_info->loginuid,
 695                                                          audit_info->sessionid,
 696                                                          audit_info->secid);
 697                                 return err;
 698                         }
 699                 }
 700                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 701                         hlist_for_each_entry(pol, entry,
 702                                              net->xfrm.policy_bydst[dir].table + i,
 703                                              bydst) {
 704                                 if (pol->type != type)
 705                                         continue;
 706                                 err = security_xfrm_policy_delete(
 707                                                                 pol->security);
 708                                 if (err) {
 709                                         xfrm_audit_policy_delete(pol, 0,
 710                                                         audit_info->loginuid,
 711                                                         audit_info->sessionid,
 712                                                         audit_info->secid);
 713                                         return err;
 714                                 }
 715                         }
 716                 }
 717         }
 718         return err;
 719 }
 720 #else
 721 static inline int
 722 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 723 {
 724         return 0;
 725 }
 726 #endif
 727
 728 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 729 {
 730         int dir, err = 0, cnt = 0;
 731
 732         write_lock_bh(&xfrm_policy_lock);
 733
 734         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 735         if (err)
 736                 goto out;
 737
 738         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 739                 struct xfrm_policy *pol;
 740                 struct hlist_node *entry;
 741                 int i;
 742
 743         again1:
 744                 hlist_for_each_entry(pol, entry,
 745                                      &net->xfrm.policy_inexact[dir], bydst) {
 746                         if (pol->type != type)
 747                                 continue;
 748                         __xfrm_policy_unlink(pol, dir);
 749                         write_unlock_bh(&xfrm_policy_lock);
 750                         cnt++;
 751
 752                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 753                                                  audit_info->sessionid,
 754                                                  audit_info->secid);
 755
 756                         xfrm_policy_kill(pol);
 757
 758                         write_lock_bh(&xfrm_policy_lock);
 759                         goto again1;
 760                 }
 761
 762                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 763         again2:
 764                         hlist_for_each_entry(pol, entry,
 765                                              net->xfrm.policy_bydst[dir].table + i,
 766                                              bydst) {
 767                                 if (pol->type != type)
 768                                         continue;
 769                                 __xfrm_policy_unlink(pol, dir);
 770                                 write_unlock_bh(&xfrm_policy_lock);
 771                                 cnt++;
 772
 773                                 xfrm_audit_policy_delete(pol, 1,
 774                                                          audit_info->loginuid,
 775                                                          audit_info->sessionid,
 776                                                          audit_info->secid);
 777                                 xfrm_policy_kill(pol);
 778
 779                                 write_lock_bh(&xfrm_policy_lock);
 780                                 goto again2;
 781                         }
 782                 }
 783
 784         }
 785         if (!cnt)
 786                 err = -ESRCH;
 787 out:
 788         write_unlock_bh(&xfrm_policy_lock);
 789         return err;
 790 }
 791 EXPORT_SYMBOL(xfrm_policy_flush);
 792
 793 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 794                      int (*func)(struct xfrm_policy *, int, int, void*),
 795                      void *data)
 796 {
 797         struct xfrm_policy *pol;
 798         struct xfrm_policy_walk_entry *x;
 799         int error = 0;
 800
 801         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 802             walk->type != XFRM_POLICY_TYPE_ANY)
 803                 return -EINVAL;
 804
 805         if (list_empty(&walk->walk.all) && walk->seq != 0)
 806                 return 0;
 807
 808         write_lock_bh(&xfrm_policy_lock);
 809         if (list_empty(&walk->walk.all))
 810                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 811         else
 812                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 813         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 814                 if (x->dead)
 815                         continue;
 816                 pol = container_of(x, struct xfrm_policy, walk);
 817                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 818                     walk->type != pol->type)
 819                         continue;
 820                 error = func(pol, xfrm_policy_id2dir(pol->index),
 821                              walk->seq, data);
 822                 if (error) {
 823                         list_move_tail(&walk->walk.all, &x->all);
 824                         goto out;
 825                 }
 826                 walk->seq++;
 827         }
 828         if (walk->seq == 0) {
 829                 error = -ENOENT;
 830                 goto out;
 831         }
 832         list_del_init(&walk->walk.all);
 833 out:
 834         write_unlock_bh(&xfrm_policy_lock);
 835         return error;
 836 }
 837 EXPORT_SYMBOL(xfrm_policy_walk);
 838
 839 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 840 {
 841         INIT_LIST_HEAD(&walk->walk.all);
 842         walk->walk.dead = 1;
 843         walk->type = type;
 844         walk->seq = 0;
 845 }
 846 EXPORT_SYMBOL(xfrm_policy_walk_init);
 847
 848 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 849 {
 850         if (list_empty(&walk->walk.all))
 851                 return;
 852
 853         write_lock_bh(&xfrm_policy_lock);
 854         list_del(&walk->walk.all);
 855         write_unlock_bh(&xfrm_policy_lock);
 856 }
 857 EXPORT_SYMBOL(xfrm_policy_walk_done);
 858
 859 /*
 860  * Find policy to apply to this flow.
 861  *
 862  * Returns 0 if policy found, else an -errno.
 863  */
 864 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 865                              u8 type, u16 family, int dir)
 866 {
 867         struct xfrm_selector *sel = &pol->selector;
 868         int match, ret = -ESRCH;
 869
 870         if (pol->family != family ||
 871             (fl->mark & pol->mark.m) != pol->mark.v ||
 872             pol->type != type)
 873                 return ret;
 874
 875         match = xfrm_selector_match(sel, fl, family);
 876         if (match)
 877                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 878                                                   dir);
 879
 880         return ret;
 881 }
 882
 883 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 884                                                      struct flowi *fl,
 885                                                      u16 family, u8 dir)
 886 {
 887         int err;
 888         struct xfrm_policy *pol, *ret;
 889         xfrm_address_t *daddr, *saddr;
 890         struct hlist_node *entry;
 891         struct hlist_head *chain;
 892         u32 priority = ~0U;
 893
 894         daddr = xfrm_flowi_daddr(fl, family);
 895         saddr = xfrm_flowi_saddr(fl, family);
 896         if (unlikely(!daddr || !saddr))
 897                 return NULL;
 898
 899         read_lock_bh(&xfrm_policy_lock);
 900         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 901         ret = NULL;
 902         hlist_for_each_entry(pol, entry, chain, bydst) {
 903                 err = xfrm_policy_match(pol, fl, type, family, dir);
 904                 if (err) {
 905                         if (err == -ESRCH)
 906                                 continue;
 907                         else {
 908                                 ret = ERR_PTR(err);
 909                                 goto fail;
 910                         }
 911                 } else {
 912                         ret = pol;
 913                         priority = ret->priority;
 914                         break;
 915                 }
 916         }
 917         chain = &net->xfrm.policy_inexact[dir];
 918         hlist_for_each_entry(pol, entry, chain, bydst) {
 919                 err = xfrm_policy_match(pol, fl, type, family, dir);
 920                 if (err) {
 921                         if (err == -ESRCH)
 922                                 continue;
 923                         else {
 924                                 ret = ERR_PTR(err);
 925                                 goto fail;
 926                         }
 927                 } else if (pol->priority < priority) {
 928                         ret = pol;
 929                         break;
 930                 }
 931         }
 932         if (ret)
 933                 xfrm_pol_hold(ret);
 934 fail:
 935         read_unlock_bh(&xfrm_policy_lock);
 936
 937         return ret;
 938 }
 939
 940 static struct xfrm_policy *
 941 __xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family, u8 dir)
 942 {
 943 #ifdef CONFIG_XFRM_SUB_POLICY
 944         struct xfrm_policy *pol;
 945
 946         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
 947         if (pol != NULL)
 948                 return pol;
 949 #endif
 950         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 951 }
 952
 953 static struct flow_cache_object *
 954 xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
 955                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
 956 {
 957         struct xfrm_policy *pol;
 958
 959         if (old_obj)
 960                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
 961
 962         pol = __xfrm_policy_lookup(net, fl, family, dir);
 963         if (IS_ERR_OR_NULL(pol))
 964                 return ERR_CAST(pol);
 965
 966         /* Resolver returns two references:
 967          * one for cache and one for caller of flow_cache_lookup() */
 968         xfrm_pol_hold(pol);
 969
 970         return &pol->flo;
 971 }
 972
 973 static inline int policy_to_flow_dir(int dir)
 974 {
 975         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 976             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 977             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 978                 return dir;
 979         switch (dir) {
 980         default:
 981         case XFRM_POLICY_IN:
 982                 return FLOW_DIR_IN;
 983         case XFRM_POLICY_OUT:
 984                 return FLOW_DIR_OUT;
 985         case XFRM_POLICY_FWD:
 986                 return FLOW_DIR_FWD;
 987         }
 988 }
 989
 990 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
 991 {
 992         struct xfrm_policy *pol;
 993
 994         read_lock_bh(&xfrm_policy_lock);
 995         if ((pol = sk->sk_policy[dir]) != NULL) {
 996                 int match = xfrm_selector_match(&pol->selector, fl,
 997                                                 sk->sk_family);
 998                 int err = 0;
 999
1000                 if (match) {
1001                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1002                                 pol = NULL;
1003                                 goto out;
1004                         }
1005                         err = security_xfrm_policy_lookup(pol->security,
1006                                                       fl->secid,
1007                                                       policy_to_flow_dir(dir));
1008                         if (!err)
1009                                 xfrm_pol_hold(pol);
1010                         else if (err == -ESRCH)
1011                                 pol = NULL;
1012                         else
1013                                 pol = ERR_PTR(err);
1014                 } else
1015                         pol = NULL;
1016         }
1017 out:
1018         read_unlock_bh(&xfrm_policy_lock);
1019         return pol;
1020 }
1021
1022 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1023 {
1024         struct net *net = xp_net(pol);
1025         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1026                                                      pol->family, dir);
1027
1028         list_add(&pol->walk.all, &net->xfrm.policy_all);
1029         hlist_add_head(&pol->bydst, chain);
1030         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1031         net->xfrm.policy_count[dir]++;
1032         xfrm_pol_hold(pol);
1033
1034         if (xfrm_bydst_should_resize(net, dir, NULL))
1035                 schedule_work(&net->xfrm.policy_hash_work);
1036 }
1037
1038 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1039                                                 int dir)
1040 {
1041         struct net *net = xp_net(pol);
1042
1043         if (hlist_unhashed(&pol->bydst))
1044                 return NULL;
1045
1046         hlist_del(&pol->bydst);
1047         hlist_del(&pol->byidx);
1048         list_del(&pol->walk.all);
1049         net->xfrm.policy_count[dir]--;
1050
1051         return pol;
1052 }
1053
1054 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1055 {
1056         write_lock_bh(&xfrm_policy_lock);
1057         pol = __xfrm_policy_unlink(pol, dir);
1058         write_unlock_bh(&xfrm_policy_lock);
1059         if (pol) {
1060                 xfrm_policy_kill(pol);
1061                 return 0;
1062         }
1063         return -ENOENT;
1064 }
1065 EXPORT_SYMBOL(xfrm_policy_delete);
1066
1067 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1068 {
1069         struct net *net = xp_net(pol);
1070         struct xfrm_policy *old_pol;
1071
1072 #ifdef CONFIG_XFRM_SUB_POLICY
1073         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1074                 return -EINVAL;
1075 #endif
1076
1077         write_lock_bh(&xfrm_policy_lock);
1078         old_pol = sk->sk_policy[dir];
1079         sk->sk_policy[dir] = pol;
1080         if (pol) {
1081                 pol->curlft.add_time = get_seconds();
1082                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1083                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1084         }
1085         if (old_pol)
1086                 /* Unlinking succeeds always. This is the only function
1087                  * allowed to delete or replace socket policy.
1088                  */
1089                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1090         write_unlock_bh(&xfrm_policy_lock);
1091
1092         if (old_pol) {
1093                 xfrm_policy_kill(old_pol);
1094         }
1095         return 0;
1096 }
1097
1098 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1099 {
1100         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1101
1102         if (newp) {
1103                 newp->selector = old->selector;
1104                 if (security_xfrm_policy_clone(old->security,
1105                                                &newp->security)) {
1106                         kfree(newp);
1107                         return NULL;  /* ENOMEM */
1108                 }
1109                 newp->lft = old->lft;
1110                 newp->curlft = old->curlft;
1111                 newp->mark = old->mark;
1112                 newp->action = old->action;
1113                 newp->flags = old->flags;
1114                 newp->xfrm_nr = old->xfrm_nr;
1115                 newp->index = old->index;
1116                 newp->type = old->type;
1117                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1118                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1119                 write_lock_bh(&xfrm_policy_lock);
1120                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1121                 write_unlock_bh(&xfrm_policy_lock);
1122                 xfrm_pol_put(newp);
1123         }
1124         return newp;
1125 }
1126
1127 int __xfrm_sk_clone_policy(struct sock *sk)
1128 {
1129         struct xfrm_policy *p0 = sk->sk_policy[0],
1130                            *p1 = sk->sk_policy[1];
1131
1132         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1133         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1134                 return -ENOMEM;
1135         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1136                 return -ENOMEM;
1137         return 0;
1138 }
1139
1140 static int
1141 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1142                unsigned short family)
1143 {
1144         int err;
1145         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1146
1147         if (unlikely(afinfo == NULL))
1148                 return -EINVAL;
1149         err = afinfo->get_saddr(net, local, remote);
1150         xfrm_policy_put_afinfo(afinfo);
1151         return err;
1152 }
1153
1154 /* Resolve list of templates for the flow, given policy. */
1155
1156 static int
1157 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1158                       struct xfrm_state **xfrm,
1159                       unsigned short family)
1160 {
1161         struct net *net = xp_net(policy);
1162         int nx;
1163         int i, error;
1164         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1165         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1166         xfrm_address_t tmp;
1167
1168         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1169                 struct xfrm_state *x;
1170                 xfrm_address_t *remote = daddr;
1171                 xfrm_address_t *local  = saddr;
1172                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1173
1174                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1175                     tmpl->mode == XFRM_MODE_BEET) {
1176                         remote = &tmpl->id.daddr;
1177                         local = &tmpl->saddr;
1178                         family = tmpl->encap_family;
1179                         if (xfrm_addr_any(local, family)) {
1180                                 error = xfrm_get_saddr(net, &tmp, remote, family);
1181                                 if (error)
1182                                         goto fail;
1183                                 local = &tmp;
1184                         }
1185                 }
1186
1187                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1188
1189                 if (x && x->km.state == XFRM_STATE_VALID) {
1190                         xfrm[nx++] = x;
1191                         daddr = remote;
1192                         saddr = local;
1193                         continue;
1194                 }
1195                 if (x) {
1196                         error = (x->km.state == XFRM_STATE_ERROR ?
1197                                  -EINVAL : -EAGAIN);
1198                         xfrm_state_put(x);
1199                 }
1200                 else if (error == -ESRCH)
1201                         error = -EAGAIN;
1202
1203                 if (!tmpl->optional)
1204                         goto fail;
1205         }
1206         return nx;
1207
1208 fail:
1209         for (nx--; nx>=0; nx--)
1210                 xfrm_state_put(xfrm[nx]);
1211         return error;
1212 }
1213
1214 static int
1215 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1216                   struct xfrm_state **xfrm,
1217                   unsigned short family)
1218 {
1219         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1220         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1221         int cnx = 0;
1222         int error;
1223         int ret;
1224         int i;
1225
1226         for (i = 0; i < npols; i++) {
1227                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1228                         error = -ENOBUFS;
1229                         goto fail;
1230                 }
1231
1232                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1233                 if (ret < 0) {
1234                         error = ret;
1235                         goto fail;
1236                 } else
1237                         cnx += ret;
1238         }
1239
1240         /* found states are sorted for outbound processing */
1241         if (npols > 1)
1242                 xfrm_state_sort(xfrm, tpp, cnx, family);
1243
1244         return cnx;
1245
1246  fail:
1247         for (cnx--; cnx>=0; cnx--)
1248                 xfrm_state_put(tpp[cnx]);
1249         return error;
1250
1251 }
1252
1253 /* Check that the bundle accepts the flow and its components are
1254  * still valid.
1255  */
1256
1257 static inline int xfrm_get_tos(struct flowi *fl, int family)
1258 {
1259         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1260         int tos;
1261
1262         if (!afinfo)
1263                 return -EINVAL;
1264
1265         tos = afinfo->get_tos(fl);
1266
1267         xfrm_policy_put_afinfo(afinfo);
1268
1269         return tos;
1270 }
1271
1272 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1273 {
1274         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1275         struct dst_entry *dst = &xdst->u.dst;
1276
1277         if (xdst->route == NULL) {
1278                 /* Dummy bundle - if it has xfrms we were not
1279                  * able to build bundle as template resolution failed.
1280                  * It means we need to try again resolving. */
1281                 if (xdst->num_xfrms > 0)
1282                         return NULL;
1283         } else {
1284                 /* Real bundle */
1285                 if (stale_bundle(dst))
1286                         return NULL;
1287         }
1288
1289         dst_hold(dst);
1290         return flo;
1291 }
1292
1293 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1294 {
1295         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1296         struct dst_entry *dst = &xdst->u.dst;
1297
1298         if (!xdst->route)
1299                 return 0;
1300         if (stale_bundle(dst))
1301                 return 0;
1302
1303         return 1;
1304 }
1305
1306 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1307 {
1308         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1309         struct dst_entry *dst = &xdst->u.dst;
1310
1311         dst_free(dst);
1312 }
1313
1314 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1315         .get = xfrm_bundle_flo_get,
1316         .check = xfrm_bundle_flo_check,
1317         .delete = xfrm_bundle_flo_delete,
1318 };
1319
1320 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1321 {
1322         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1323         struct dst_ops *dst_ops;
1324         struct xfrm_dst *xdst;
1325
1326         if (!afinfo)
1327                 return ERR_PTR(-EINVAL);
1328
1329         switch (family) {
1330         case AF_INET:
1331                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1332                 break;
1333 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1334         case AF_INET6:
1335                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1336                 break;
1337 #endif
1338         default:
1339                 BUG();
1340         }
1341         xdst = dst_alloc(dst_ops) ?: ERR_PTR(-ENOBUFS);
1342         xfrm_policy_put_afinfo(afinfo);
1343
1344         xdst->flo.ops = &xfrm_bundle_fc_ops;
1345
1346         return xdst;
1347 }
1348
1349 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1350                                  int nfheader_len)
1351 {
1352         struct xfrm_policy_afinfo *afinfo =
1353                 xfrm_policy_get_afinfo(dst->ops->family);
1354         int err;
1355
1356         if (!afinfo)
1357                 return -EINVAL;
1358
1359         err = afinfo->init_path(path, dst, nfheader_len);
1360
1361         xfrm_policy_put_afinfo(afinfo);
1362
1363         return err;
1364 }
1365
1366 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1367                                 struct flowi *fl)
1368 {
1369         struct xfrm_policy_afinfo *afinfo =
1370                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1371         int err;
1372
1373         if (!afinfo)
1374                 return -EINVAL;
1375
1376         err = afinfo->fill_dst(xdst, dev, fl);
1377
1378         xfrm_policy_put_afinfo(afinfo);
1379
1380         return err;
1381 }
1382
1383
1384 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1385  * all the metrics... Shortly, bundle a bundle.
1386  */
1387
1388 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1389                                             struct xfrm_state **xfrm, int nx,
1390                                             struct flowi *fl,
1391                                             struct dst_entry *dst)
1392 {
1393         struct net *net = xp_net(policy);
1394         unsigned long now = jiffies;
1395         struct net_device *dev;
1396         struct dst_entry *dst_prev = NULL;
1397         struct dst_entry *dst0 = NULL;
1398         int i = 0;
1399         int err;
1400         int header_len = 0;
1401         int nfheader_len = 0;
1402         int trailer_len = 0;
1403         int tos;
1404         int family = policy->selector.family;
1405         xfrm_address_t saddr, daddr;
1406
1407         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1408
1409         tos = xfrm_get_tos(fl, family);
1410         err = tos;
1411         if (tos < 0)
1412                 goto put_states;
1413
1414         dst_hold(dst);
1415
1416         for (; i < nx; i++) {
1417                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1418                 struct dst_entry *dst1 = &xdst->u.dst;
1419
1420                 err = PTR_ERR(xdst);
1421                 if (IS_ERR(xdst)) {
1422                         dst_release(dst);
1423                         goto put_states;
1424                 }
1425
1426                 if (!dst_prev)
1427                         dst0 = dst1;
1428                 else {
1429                         dst_prev->child = dst_clone(dst1);
1430                         dst1->flags |= DST_NOHASH;
1431                 }
1432
1433                 xdst->route = dst;
1434                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1435
1436                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1437                         family = xfrm[i]->props.family;
1438                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1439                                               family);
1440                         err = PTR_ERR(dst);
1441                         if (IS_ERR(dst))
1442                                 goto put_states;
1443                 } else
1444                         dst_hold(dst);
1445
1446                 dst1->xfrm = xfrm[i];
1447                 xdst->xfrm_genid = xfrm[i]->genid;
1448
1449                 dst1->obsolete = -1;
1450                 dst1->flags |= DST_HOST;
1451                 dst1->lastuse = now;
1452
1453                 dst1->input = dst_discard;
1454                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1455
1456                 dst1->next = dst_prev;
1457                 dst_prev = dst1;
1458
1459                 header_len += xfrm[i]->props.header_len;
1460                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1461                         nfheader_len += xfrm[i]->props.header_len;
1462                 trailer_len += xfrm[i]->props.trailer_len;
1463         }
1464
1465         dst_prev->child = dst;
1466         dst0->path = dst;
1467
1468         err = -ENODEV;
1469         dev = dst->dev;
1470         if (!dev)
1471                 goto free_dst;
1472
1473         /* Copy neighbour for reachability confirmation */
1474         dst0->neighbour = neigh_clone(dst->neighbour);
1475
1476         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1477         xfrm_init_pmtu(dst_prev);
1478
1479         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1480                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1481
1482                 err = xfrm_fill_dst(xdst, dev, fl);
1483                 if (err)
1484                         goto free_dst;
1485
1486                 dst_prev->header_len = header_len;
1487                 dst_prev->trailer_len = trailer_len;
1488                 header_len -= xdst->u.dst.xfrm->props.header_len;
1489                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1490         }
1491
1492 out:
1493         return dst0;
1494
1495 put_states:
1496         for (; i < nx; i++)
1497                 xfrm_state_put(xfrm[i]);
1498 free_dst:
1499         if (dst0)
1500                 dst_free(dst0);
1501         dst0 = ERR_PTR(err);
1502         goto out;
1503 }
1504
1505 static int inline
1506 xfrm_dst_alloc_copy(void **target, void *src, int size)
1507 {
1508         if (!*target) {
1509                 *target = kmalloc(size, GFP_ATOMIC);
1510                 if (!*target)
1511                         return -ENOMEM;
1512         }
1513         memcpy(*target, src, size);
1514         return 0;
1515 }
1516
1517 static int inline
1518 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1519 {
1520 #ifdef CONFIG_XFRM_SUB_POLICY
1521         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1522         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1523                                    sel, sizeof(*sel));
1524 #else
1525         return 0;
1526 #endif
1527 }
1528
1529 static int inline
1530 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1531 {
1532 #ifdef CONFIG_XFRM_SUB_POLICY
1533         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1534         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1535 #else
1536         return 0;
1537 #endif
1538 }
1539
1540 static int xfrm_expand_policies(struct flowi *fl, u16 family,
1541                                 struct xfrm_policy **pols,
1542                                 int *num_pols, int *num_xfrms)
1543 {
1544         int i;
1545
1546         if (*num_pols == 0 || !pols[0]) {
1547                 *num_pols = 0;
1548                 *num_xfrms = 0;
1549                 return 0;
1550         }
1551         if (IS_ERR(pols[0]))
1552                 return PTR_ERR(pols[0]);
1553
1554         *num_xfrms = pols[0]->xfrm_nr;
1555
1556 #ifdef CONFIG_XFRM_SUB_POLICY
1557         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1558             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1559                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1560                                                     XFRM_POLICY_TYPE_MAIN,
1561                                                     fl, family,
1562                                                     XFRM_POLICY_OUT);
1563                 if (pols[1]) {
1564                         if (IS_ERR(pols[1])) {
1565                                 xfrm_pols_put(pols, *num_pols);
1566                                 return PTR_ERR(pols[1]);
1567                         }
1568                         (*num_pols) ++;
1569                         (*num_xfrms) += pols[1]->xfrm_nr;
1570                 }
1571         }
1572 #endif
1573         for (i = 0; i < *num_pols; i++) {
1574                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1575                         *num_xfrms = -1;
1576                         break;
1577                 }
1578         }
1579
1580         return 0;
1581
1582 }
1583
1584 static struct xfrm_dst *
1585 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1586                                struct flowi *fl, u16 family,
1587                                struct dst_entry *dst_orig)
1588 {
1589         struct net *net = xp_net(pols[0]);
1590         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1591         struct dst_entry *dst;
1592         struct xfrm_dst *xdst;
1593         int err;
1594
1595         /* Try to instantiate a bundle */
1596         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1597         if (err <= 0) {
1598                 if (err != 0 && err != -EAGAIN)
1599                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1600                 return ERR_PTR(err);
1601         }
1602
1603         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1604         if (IS_ERR(dst)) {
1605                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1606                 return ERR_CAST(dst);
1607         }
1608
1609         xdst = (struct xfrm_dst *)dst;
1610         xdst->num_xfrms = err;
1611         if (num_pols > 1)
1612                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1613         else
1614                 err = xfrm_dst_update_origin(dst, fl);
1615         if (unlikely(err)) {
1616                 dst_free(dst);
1617                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1618                 return ERR_PTR(err);
1619         }
1620
1621         xdst->num_pols = num_pols;
1622         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1623         xdst->policy_genid = atomic_read(&pols[0]->genid);
1624
1625         return xdst;
1626 }
1627
1628 static struct flow_cache_object *
1629 xfrm_bundle_lookup(struct net *net, struct flowi *fl, u16 family, u8 dir,
1630                    struct flow_cache_object *oldflo, void *ctx)
1631 {
1632         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1633         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1634         struct xfrm_dst *xdst, *new_xdst;
1635         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1636
1637         /* Check if the policies from old bundle are usable */
1638         xdst = NULL;
1639         if (oldflo) {
1640                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1641                 num_pols = xdst->num_pols;
1642                 num_xfrms = xdst->num_xfrms;
1643                 pol_dead = 0;
1644                 for (i = 0; i < num_pols; i++) {
1645                         pols[i] = xdst->pols[i];
1646                         pol_dead |= pols[i]->walk.dead;
1647                 }
1648                 if (pol_dead) {
1649                         dst_free(&xdst->u.dst);
1650                         xdst = NULL;
1651                         num_pols = 0;
1652                         num_xfrms = 0;
1653                         oldflo = NULL;
1654                 }
1655         }
1656
1657         /* Resolve policies to use if we couldn't get them from
1658          * previous cache entry */
1659         if (xdst == NULL) {
1660                 num_pols = 1;
1661                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1662                 err = xfrm_expand_policies(fl, family, pols,
1663                                            &num_pols, &num_xfrms);
1664                 if (err < 0)
1665                         goto inc_error;
1666                 if (num_pols == 0)
1667                         return NULL;
1668                 if (num_xfrms <= 0)
1669                         goto make_dummy_bundle;
1670         }
1671
1672         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1673         if (IS_ERR(new_xdst)) {
1674                 err = PTR_ERR(new_xdst);
1675                 if (err != -EAGAIN)
1676                         goto error;
1677                 if (oldflo == NULL)
1678                         goto make_dummy_bundle;
1679                 dst_hold(&xdst->u.dst);
1680                 return oldflo;
1681         } else if (new_xdst == NULL) {
1682                 num_xfrms = 0;
1683                 if (oldflo == NULL)
1684                         goto make_dummy_bundle;
1685                 xdst->num_xfrms = 0;
1686                 dst_hold(&xdst->u.dst);
1687                 return oldflo;
1688         }
1689
1690         /* Kill the previous bundle */
1691         if (xdst) {
1692                 /* The policies were stolen for newly generated bundle */
1693                 xdst->num_pols = 0;
1694                 dst_free(&xdst->u.dst);
1695         }
1696
1697         /* Flow cache does not have reference, it dst_free()'s,
1698          * but we do need to return one reference for original caller */
1699         dst_hold(&new_xdst->u.dst);
1700         return &new_xdst->flo;
1701
1702 make_dummy_bundle:
1703         /* We found policies, but there's no bundles to instantiate:
1704          * either because the policy blocks, has no transformations or
1705          * we could not build template (no xfrm_states).*/
1706         xdst = xfrm_alloc_dst(net, family);
1707         if (IS_ERR(xdst)) {
1708                 xfrm_pols_put(pols, num_pols);
1709                 return ERR_CAST(xdst);
1710         }
1711         xdst->num_pols = num_pols;
1712         xdst->num_xfrms = num_xfrms;
1713         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1714
1715         dst_hold(&xdst->u.dst);
1716         return &xdst->flo;
1717
1718 inc_error:
1719         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1720 error:
1721         if (xdst != NULL)
1722                 dst_free(&xdst->u.dst);
1723         else
1724                 xfrm_pols_put(pols, num_pols);
1725         return ERR_PTR(err);
1726 }
1727
1728 /* Main function: finds/creates a bundle for given flow.
1729  *
1730  * At the moment we eat a raw IP route. Mostly to speed up lookups
1731  * on interfaces with disabled IPsec.
1732  */
1733 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1734                   struct sock *sk, int flags)
1735 {
1736         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1737         struct flow_cache_object *flo;
1738         struct xfrm_dst *xdst;
1739         struct dst_entry *dst, *dst_orig = *dst_p, *route;
1740         u16 family = dst_orig->ops->family;
1741         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1742         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1743
1744 restart:
1745         dst = NULL;
1746         xdst = NULL;
1747         route = NULL;
1748
1749         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1750                 num_pols = 1;
1751                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1752                 err = xfrm_expand_policies(fl, family, pols,
1753                                            &num_pols, &num_xfrms);
1754                 if (err < 0)
1755                         goto dropdst;
1756
1757                 if (num_pols) {
1758                         if (num_xfrms <= 0) {
1759                                 drop_pols = num_pols;
1760                                 goto no_transform;
1761                         }
1762
1763                         xdst = xfrm_resolve_and_create_bundle(
1764                                         pols, num_pols, fl,
1765                                         family, dst_orig);
1766                         if (IS_ERR(xdst)) {
1767                                 xfrm_pols_put(pols, num_pols);
1768                                 err = PTR_ERR(xdst);
1769                                 goto dropdst;
1770                         } else if (xdst == NULL) {
1771                                 num_xfrms = 0;
1772                                 drop_pols = num_pols;
1773                                 goto no_transform;
1774                         }
1775
1776                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1777                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1778                         xfrm_policy_sk_bundles = &xdst->u.dst;
1779                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1780
1781                         route = xdst->route;
1782                 }
1783         }
1784
1785         if (xdst == NULL) {
1786                 /* To accelerate a bit...  */
1787                 if ((dst_orig->flags & DST_NOXFRM) ||
1788                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1789                         goto nopol;
1790
1791                 flo = flow_cache_lookup(net, fl, family, dir,
1792                                         xfrm_bundle_lookup, dst_orig);
1793                 if (flo == NULL)
1794                         goto nopol;
1795                 if (IS_ERR(flo)) {
1796                         err = PTR_ERR(flo);
1797                         goto dropdst;
1798                 }
1799                 xdst = container_of(flo, struct xfrm_dst, flo);
1800
1801                 num_pols = xdst->num_pols;
1802                 num_xfrms = xdst->num_xfrms;
1803                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1804                 route = xdst->route;
1805         }
1806
1807         dst = &xdst->u.dst;
1808         if (route == NULL && num_xfrms > 0) {
1809                 /* The only case when xfrm_bundle_lookup() returns a
1810                  * bundle with null route, is when the template could
1811                  * not be resolved. It means policies are there, but
1812                  * bundle could not be created, since we don't yet
1813                  * have the xfrm_state's. We need to wait for KM to
1814                  * negotiate new SA's or bail out with error.*/
1815                 if (net->xfrm.sysctl_larval_drop) {
1816                         /* EREMOTE tells the caller to generate
1817                          * a one-shot blackhole route. */
1818                         dst_release(dst);
1819                         xfrm_pols_put(pols, drop_pols);
1820                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1821                         return -EREMOTE;
1822                 }
1823                 if (flags & XFRM_LOOKUP_WAIT) {
1824                         DECLARE_WAITQUEUE(wait, current);
1825
1826                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1827                         set_current_state(TASK_INTERRUPTIBLE);
1828                         schedule();
1829                         set_current_state(TASK_RUNNING);
1830                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1831
1832                         if (!signal_pending(current)) {
1833                                 dst_release(dst);
1834                                 goto restart;
1835                         }
1836
1837                         err = -ERESTART;
1838                 } else
1839                         err = -EAGAIN;
1840
1841                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1842                 goto error;
1843         }
1844
1845 no_transform:
1846         if (num_pols == 0)
1847                 goto nopol;
1848
1849         if ((flags & XFRM_LOOKUP_ICMP) &&
1850             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1851                 err = -ENOENT;
1852                 goto error;
1853         }
1854
1855         for (i = 0; i < num_pols; i++)
1856                 pols[i]->curlft.use_time = get_seconds();
1857
1858         if (num_xfrms < 0) {
1859                 /* Prohibit the flow */
1860                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1861                 err = -EPERM;
1862                 goto error;
1863         } else if (num_xfrms > 0) {
1864                 /* Flow transformed */
1865                 *dst_p = dst;
1866                 dst_release(dst_orig);
1867         } else {
1868                 /* Flow passes untransformed */
1869                 dst_release(dst);
1870         }
1871 ok:
1872         xfrm_pols_put(pols, drop_pols);
1873         return 0;
1874
1875 nopol:
1876         if (!(flags & XFRM_LOOKUP_ICMP))
1877                 goto ok;
1878         err = -ENOENT;
1879 error:
1880         dst_release(dst);
1881 dropdst:
1882         dst_release(dst_orig);
1883         *dst_p = NULL;
1884         xfrm_pols_put(pols, drop_pols);
1885         return err;
1886 }
1887 EXPORT_SYMBOL(__xfrm_lookup);
1888
1889 int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1890                 struct sock *sk, int flags)
1891 {
1892         int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1893
1894         if (err == -EREMOTE) {
1895                 dst_release(*dst_p);
1896                 *dst_p = NULL;
1897                 err = -EAGAIN;
1898         }
1899
1900         return err;
1901 }
1902 EXPORT_SYMBOL(xfrm_lookup);
1903
1904 static inline int
1905 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1906 {
1907         struct xfrm_state *x;
1908
1909         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1910                 return 0;
1911         x = skb->sp->xvec[idx];
1912         if (!x->type->reject)
1913                 return 0;
1914         return x->type->reject(x, skb, fl);
1915 }
1916
1917 /* When skb is transformed back to its "native" form, we have to
1918  * check policy restrictions. At the moment we make this in maximally
1919  * stupid way. Shame on me. :-) Of course, connected sockets must
1920  * have policy cached at them.
1921  */
1922
1923 static inline int
1924 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1925               unsigned short family)
1926 {
1927         if (xfrm_state_kern(x))
1928                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1929         return  x->id.proto == tmpl->id.proto &&
1930                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1931                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1932                 x->props.mode == tmpl->mode &&
1933                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1934                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1935                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1936                   xfrm_state_addr_cmp(tmpl, x, family));
1937 }
1938
1939 /*
1940  * 0 or more than 0 is returned when validation is succeeded (either bypass
1941  * because of optional transport mode, or next index of the mathced secpath
1942  * state with the template.
1943  * -1 is returned when no matching template is found.
1944  * Otherwise "-2 - errored_index" is returned.
1945  */
1946 static inline int
1947 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1948                unsigned short family)
1949 {
1950         int idx = start;
1951
1952         if (tmpl->optional) {
1953                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1954                         return start;
1955         } else
1956                 start = -1;
1957         for (; idx < sp->len; idx++) {
1958                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1959                         return ++idx;
1960                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1961                         if (start == -1)
1962                                 start = -2-idx;
1963                         break;
1964                 }
1965         }
1966         return start;
1967 }
1968
1969 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1970                           unsigned int family, int reverse)
1971 {
1972         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1973         int err;
1974
1975         if (unlikely(afinfo == NULL))
1976                 return -EAFNOSUPPORT;
1977
1978         afinfo->decode_session(skb, fl, reverse);
1979         err = security_xfrm_decode_session(skb, &fl->secid);
1980         xfrm_policy_put_afinfo(afinfo);
1981         return err;
1982 }
1983 EXPORT_SYMBOL(__xfrm_decode_session);
1984
1985 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1986 {
1987         for (; k < sp->len; k++) {
1988                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1989                         *idxp = k;
1990                         return 1;
1991                 }
1992         }
1993
1994         return 0;
1995 }
1996
1997 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1998                         unsigned short family)
1999 {
2000         struct net *net = dev_net(skb->dev);
2001         struct xfrm_policy *pol;
2002         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2003         int npols = 0;
2004         int xfrm_nr;
2005         int pi;
2006         int reverse;
2007         struct flowi fl;
2008         u8 fl_dir;
2009         int xerr_idx = -1;
2010
2011         reverse = dir & ~XFRM_POLICY_MASK;
2012         dir &= XFRM_POLICY_MASK;
2013         fl_dir = policy_to_flow_dir(dir);
2014
2015         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2016                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2017                 return 0;
2018         }
2019
2020         nf_nat_decode_session(skb, &fl, family);
2021
2022         /* First, check used SA against their selectors. */
2023         if (skb->sp) {
2024                 int i;
2025
2026                 for (i=skb->sp->len-1; i>=0; i--) {
2027                         struct xfrm_state *x = skb->sp->xvec[i];
2028                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2029                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2030                                 return 0;
2031                         }
2032                 }
2033         }
2034
2035         pol = NULL;
2036         if (sk && sk->sk_policy[dir]) {
2037                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2038                 if (IS_ERR(pol)) {
2039                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2040                         return 0;
2041                 }
2042         }
2043
2044         if (!pol) {
2045                 struct flow_cache_object *flo;
2046
2047                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2048                                         xfrm_policy_lookup, NULL);
2049                 if (IS_ERR_OR_NULL(flo))
2050                         pol = ERR_CAST(flo);
2051                 else
2052                         pol = container_of(flo, struct xfrm_policy, flo);
2053         }
2054
2055         if (IS_ERR(pol)) {
2056                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2057                 return 0;
2058         }
2059
2060         if (!pol) {
2061                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2062                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2063                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2064                         return 0;
2065                 }
2066                 return 1;
2067         }
2068
2069         pol->curlft.use_time = get_seconds();
2070
2071         pols[0] = pol;
2072         npols ++;
2073 #ifdef CONFIG_XFRM_SUB_POLICY
2074         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2075                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2076                                                     &fl, family,
2077                                                     XFRM_POLICY_IN);
2078                 if (pols[1]) {
2079                         if (IS_ERR(pols[1])) {
2080                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2081                                 return 0;
2082                         }
2083                         pols[1]->curlft.use_time = get_seconds();
2084                         npols ++;
2085                 }
2086         }
2087 #endif
2088
2089         if (pol->action == XFRM_POLICY_ALLOW) {
2090                 struct sec_path *sp;
2091                 static struct sec_path dummy;
2092                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2093                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2094                 struct xfrm_tmpl **tpp = tp;
2095                 int ti = 0;
2096                 int i, k;
2097
2098                 if ((sp = skb->sp) == NULL)
2099                         sp = &dummy;
2100
2101                 for (pi = 0; pi < npols; pi++) {
2102                         if (pols[pi] != pol &&
2103                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2104                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2105                                 goto reject;
2106                         }
2107                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2108                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2109                                 goto reject_error;
2110                         }
2111                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2112                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2113                 }
2114                 xfrm_nr = ti;
2115                 if (npols > 1) {
2116                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2117                         tpp = stp;
2118                 }
2119
2120                 /* For each tunnel xfrm, find the first matching tmpl.
2121                  * For each tmpl before that, find corresponding xfrm.
2122                  * Order is _important_. Later we will implement
2123                  * some barriers, but at the moment barriers
2124                  * are implied between each two transformations.
2125                  */
2126                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2127                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2128                         if (k < 0) {
2129                                 if (k < -1)
2130                                         /* "-2 - errored_index" returned */
2131                                         xerr_idx = -(2+k);
2132                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2133                                 goto reject;
2134                         }
2135                 }
2136
2137                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2138                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2139                         goto reject;
2140                 }
2141
2142                 xfrm_pols_put(pols, npols);
2143                 return 1;
2144         }
2145         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2146
2147 reject:
2148         xfrm_secpath_reject(xerr_idx, skb, &fl);
2149 reject_error:
2150         xfrm_pols_put(pols, npols);
2151         return 0;
2152 }
2153 EXPORT_SYMBOL(__xfrm_policy_check);
2154
2155 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2156 {
2157         struct net *net = dev_net(skb->dev);
2158         struct flowi fl;
2159         struct dst_entry *dst;
2160         int res;
2161
2162         if (xfrm_decode_session(skb, &fl, family) < 0) {
2163                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2164                 return 0;
2165         }
2166
2167         skb_dst_force(skb);
2168         dst = skb_dst(skb);
2169
2170         res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
2171         skb_dst_set(skb, dst);
2172         return res;
2173 }
2174 EXPORT_SYMBOL(__xfrm_route_forward);
2175
2176 /* Optimize later using cookies and generation ids. */
2177
2178 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2179 {
2180         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2181          * to "-1" to force all XFRM destinations to get validated by
2182          * dst_ops->check on every use.  We do this because when a
2183          * normal route referenced by an XFRM dst is obsoleted we do
2184          * not go looking around for all parent referencing XFRM dsts
2185          * so that we can invalidate them.  It is just too much work.
2186          * Instead we make the checks here on every use.  For example:
2187          *
2188          *      XFRM dst A --> IPv4 dst X
2189          *
2190          * X is the "xdst->route" of A (X is also the "dst->path" of A
2191          * in this example).  If X is marked obsolete, "A" will not
2192          * notice.  That's what we are validating here via the
2193          * stale_bundle() check.
2194          *
2195          * When a policy's bundle is pruned, we dst_free() the XFRM
2196          * dst which causes it's ->obsolete field to be set to a
2197          * positive non-zero integer.  If an XFRM dst has been pruned
2198          * like this, we want to force a new route lookup.
2199          */
2200         if (dst->obsolete < 0 && !stale_bundle(dst))
2201                 return dst;
2202
2203         return NULL;
2204 }
2205
2206 static int stale_bundle(struct dst_entry *dst)
2207 {
2208         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2209 }
2210
2211 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2212 {
2213         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2214                 dst->dev = dev_net(dev)->loopback_dev;
2215                 dev_hold(dst->dev);
2216                 dev_put(dev);
2217         }
2218 }
2219 EXPORT_SYMBOL(xfrm_dst_ifdown);
2220
2221 static void xfrm_link_failure(struct sk_buff *skb)
2222 {
2223         /* Impossible. Such dst must be popped before reaches point of failure. */
2224 }
2225
2226 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2227 {
2228         if (dst) {
2229                 if (dst->obsolete) {
2230                         dst_release(dst);
2231                         dst = NULL;
2232                 }
2233         }
2234         return dst;
2235 }
2236
2237 static void __xfrm_garbage_collect(struct net *net)
2238 {
2239         struct dst_entry *head, *next;
2240
2241         flow_cache_flush();
2242
2243         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2244         head = xfrm_policy_sk_bundles;
2245         xfrm_policy_sk_bundles = NULL;
2246         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2247
2248         while (head) {
2249                 next = head->next;
2250                 dst_free(head);
2251                 head = next;
2252         }
2253 }
2254
2255 static void xfrm_init_pmtu(struct dst_entry *dst)
2256 {
2257         do {
2258                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2259                 u32 pmtu, route_mtu_cached;
2260
2261                 pmtu = dst_mtu(dst->child);
2262                 xdst->child_mtu_cached = pmtu;
2263
2264                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2265
2266                 route_mtu_cached = dst_mtu(xdst->route);
2267                 xdst->route_mtu_cached = route_mtu_cached;
2268
2269                 if (pmtu > route_mtu_cached)
2270                         pmtu = route_mtu_cached;
2271
2272                 dst->metrics[RTAX_MTU-1] = pmtu;
2273         } while ((dst = dst->next));
2274 }
2275
2276 /* Check that the bundle accepts the flow and its components are
2277  * still valid.
2278  */
2279
2280 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2281                 struct flowi *fl, int family, int strict)
2282 {
2283         struct dst_entry *dst = &first->u.dst;
2284         struct xfrm_dst *last;
2285         u32 mtu;
2286
2287         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2288             (dst->dev && !netif_running(dst->dev)))
2289                 return 0;
2290 #ifdef CONFIG_XFRM_SUB_POLICY
2291         if (fl) {
2292                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2293                         return 0;
2294                 if (first->partner &&
2295                     !xfrm_selector_match(first->partner, fl, family))
2296                         return 0;
2297         }
2298 #endif
2299
2300         last = NULL;
2301
2302         do {
2303                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2304
2305                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2306                         return 0;
2307                 if (fl && pol &&
2308                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2309                         return 0;
2310                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2311                         return 0;
2312                 if (xdst->xfrm_genid != dst->xfrm->genid)
2313                         return 0;
2314                 if (xdst->num_pols > 0 &&
2315                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2316                         return 0;
2317
2318                 if (strict && fl &&
2319                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2320                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2321                         return 0;
2322
2323                 mtu = dst_mtu(dst->child);
2324                 if (xdst->child_mtu_cached != mtu) {
2325                         last = xdst;
2326                         xdst->child_mtu_cached = mtu;
2327                 }
2328
2329                 if (!dst_check(xdst->route, xdst->route_cookie))
2330                         return 0;
2331                 mtu = dst_mtu(xdst->route);
2332                 if (xdst->route_mtu_cached != mtu) {
2333                         last = xdst;
2334                         xdst->route_mtu_cached = mtu;
2335                 }
2336
2337                 dst = dst->child;
2338         } while (dst->xfrm);
2339
2340         if (likely(!last))
2341                 return 1;
2342
2343         mtu = last->child_mtu_cached;
2344         for (;;) {
2345                 dst = &last->u.dst;
2346
2347                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2348                 if (mtu > last->route_mtu_cached)
2349                         mtu = last->route_mtu_cached;
2350                 dst->metrics[RTAX_MTU-1] = mtu;
2351
2352                 if (last == first)
2353                         break;
2354
2355                 last = (struct xfrm_dst *)last->u.dst.next;
2356                 last->child_mtu_cached = mtu;
2357         }
2358
2359         return 1;
2360 }
2361
2362 EXPORT_SYMBOL(xfrm_bundle_ok);
2363
2364 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2365 {
2366         struct net *net;
2367         int err = 0;
2368         if (unlikely(afinfo == NULL))
2369                 return -EINVAL;
2370         if (unlikely(afinfo->family >= NPROTO))
2371                 return -EAFNOSUPPORT;
2372         write_lock_bh(&xfrm_policy_afinfo_lock);
2373         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2374                 err = -ENOBUFS;
2375         else {
2376                 struct dst_ops *dst_ops = afinfo->dst_ops;
2377                 if (likely(dst_ops->kmem_cachep == NULL))
2378                         dst_ops->kmem_cachep = xfrm_dst_cache;
2379                 if (likely(dst_ops->check == NULL))
2380                         dst_ops->check = xfrm_dst_check;
2381                 if (likely(dst_ops->negative_advice == NULL))
2382                         dst_ops->negative_advice = xfrm_negative_advice;
2383                 if (likely(dst_ops->link_failure == NULL))
2384                         dst_ops->link_failure = xfrm_link_failure;
2385                 if (likely(afinfo->garbage_collect == NULL))
2386                         afinfo->garbage_collect = __xfrm_garbage_collect;
2387                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2388         }
2389         write_unlock_bh(&xfrm_policy_afinfo_lock);
2390
2391         rtnl_lock();
2392         for_each_net(net) {
2393                 struct dst_ops *xfrm_dst_ops;
2394
2395                 switch (afinfo->family) {
2396                 case AF_INET:
2397                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2398                         break;
2399 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2400                 case AF_INET6:
2401                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2402                         break;
2403 #endif
2404                 default:
2405                         BUG();
2406                 }
2407                 *xfrm_dst_ops = *afinfo->dst_ops;
2408         }
2409         rtnl_unlock();
2410
2411         return err;
2412 }
2413 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2414
2415 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2416 {
2417         int err = 0;
2418         if (unlikely(afinfo == NULL))
2419                 return -EINVAL;
2420         if (unlikely(afinfo->family >= NPROTO))
2421                 return -EAFNOSUPPORT;
2422         write_lock_bh(&xfrm_policy_afinfo_lock);
2423         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2424                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2425                         err = -EINVAL;
2426                 else {
2427                         struct dst_ops *dst_ops = afinfo->dst_ops;
2428                         xfrm_policy_afinfo[afinfo->family] = NULL;
2429                         dst_ops->kmem_cachep = NULL;
2430                         dst_ops->check = NULL;
2431                         dst_ops->negative_advice = NULL;
2432                         dst_ops->link_failure = NULL;
2433                         afinfo->garbage_collect = NULL;
2434                 }
2435         }
2436         write_unlock_bh(&xfrm_policy_afinfo_lock);
2437         return err;
2438 }
2439 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2440
2441 static void __net_init xfrm_dst_ops_init(struct net *net)
2442 {
2443         struct xfrm_policy_afinfo *afinfo;
2444
2445         read_lock_bh(&xfrm_policy_afinfo_lock);
2446         afinfo = xfrm_policy_afinfo[AF_INET];
2447         if (afinfo)
2448                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2449 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2450         afinfo = xfrm_policy_afinfo[AF_INET6];
2451         if (afinfo)
2452                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2453 #endif
2454         read_unlock_bh(&xfrm_policy_afinfo_lock);
2455 }
2456
2457 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2458 {
2459         struct xfrm_policy_afinfo *afinfo;
2460         if (unlikely(family >= NPROTO))
2461                 return NULL;
2462         read_lock(&xfrm_policy_afinfo_lock);
2463         afinfo = xfrm_policy_afinfo[family];
2464         if (unlikely(!afinfo))
2465                 read_unlock(&xfrm_policy_afinfo_lock);
2466         return afinfo;
2467 }
2468
2469 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2470 {
2471         read_unlock(&xfrm_policy_afinfo_lock);
2472 }
2473
2474 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2475 {
2476         struct net_device *dev = ptr;
2477
2478         switch (event) {
2479         case NETDEV_DOWN:
2480                 __xfrm_garbage_collect(dev_net(dev));
2481         }
2482         return NOTIFY_DONE;
2483 }
2484
2485 static struct notifier_block xfrm_dev_notifier = {
2486         .notifier_call  = xfrm_dev_event,
2487 };
2488
2489 #ifdef CONFIG_XFRM_STATISTICS
2490 static int __net_init xfrm_statistics_init(struct net *net)
2491 {
2492         int rv;
2493
2494         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2495                           sizeof(struct linux_xfrm_mib)) < 0)
2496                 return -ENOMEM;
2497         rv = xfrm_proc_init(net);
2498         if (rv < 0)
2499                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2500         return rv;
2501 }
2502
2503 static void xfrm_statistics_fini(struct net *net)
2504 {
2505         xfrm_proc_fini(net);
2506         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2507 }
2508 #else
2509 static int __net_init xfrm_statistics_init(struct net *net)
2510 {
2511         return 0;
2512 }
2513
2514 static void xfrm_statistics_fini(struct net *net)
2515 {
2516 }
2517 #endif
2518
2519 static int __net_init xfrm_policy_init(struct net *net)
2520 {
2521         unsigned int hmask, sz;
2522         int dir;
2523
2524         if (net_eq(net, &init_net))
2525                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2526                                            sizeof(struct xfrm_dst),
2527                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2528                                            NULL);
2529
2530         hmask = 8 - 1;
2531         sz = (hmask+1) * sizeof(struct hlist_head);
2532
2533         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2534         if (!net->xfrm.policy_byidx)
2535                 goto out_byidx;
2536         net->xfrm.policy_idx_hmask = hmask;
2537
2538         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2539                 struct xfrm_policy_hash *htab;
2540
2541                 net->xfrm.policy_count[dir] = 0;
2542                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2543
2544                 htab = &net->xfrm.policy_bydst[dir];
2545                 htab->table = xfrm_hash_alloc(sz);
2546                 if (!htab->table)
2547                         goto out_bydst;
2548                 htab->hmask = hmask;
2549         }
2550
2551         INIT_LIST_HEAD(&net->xfrm.policy_all);
2552         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2553         if (net_eq(net, &init_net))
2554                 register_netdevice_notifier(&xfrm_dev_notifier);
2555         return 0;
2556
2557 out_bydst:
2558         for (dir--; dir >= 0; dir--) {
2559                 struct xfrm_policy_hash *htab;
2560
2561                 htab = &net->xfrm.policy_bydst[dir];
2562                 xfrm_hash_free(htab->table, sz);
2563         }
2564         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2565 out_byidx:
2566         return -ENOMEM;
2567 }
2568
2569 static void xfrm_policy_fini(struct net *net)
2570 {
2571         struct xfrm_audit audit_info;
2572         unsigned int sz;
2573         int dir;
2574
2575         flush_work(&net->xfrm.policy_hash_work);
2576 #ifdef CONFIG_XFRM_SUB_POLICY
2577         audit_info.loginuid = -1;
2578         audit_info.sessionid = -1;
2579         audit_info.secid = 0;
2580         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2581 #endif
2582         audit_info.loginuid = -1;
2583         audit_info.sessionid = -1;
2584         audit_info.secid = 0;
2585         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2586
2587         WARN_ON(!list_empty(&net->xfrm.policy_all));
2588
2589         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2590                 struct xfrm_policy_hash *htab;
2591
2592                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2593
2594                 htab = &net->xfrm.policy_bydst[dir];
2595                 sz = (htab->hmask + 1);
2596                 WARN_ON(!hlist_empty(htab->table));
2597                 xfrm_hash_free(htab->table, sz);
2598         }
2599
2600         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2601         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2602         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2603 }
2604
2605 static int __net_init xfrm_net_init(struct net *net)
2606 {
2607         int rv;
2608
2609         rv = xfrm_statistics_init(net);
2610         if (rv < 0)
2611                 goto out_statistics;
2612         rv = xfrm_state_init(net);
2613         if (rv < 0)
2614                 goto out_state;
2615         rv = xfrm_policy_init(net);
2616         if (rv < 0)
2617                 goto out_policy;
2618         xfrm_dst_ops_init(net);
2619         rv = xfrm_sysctl_init(net);
2620         if (rv < 0)
2621                 goto out_sysctl;
2622         return 0;
2623
2624 out_sysctl:
2625         xfrm_policy_fini(net);
2626 out_policy:
2627         xfrm_state_fini(net);
2628 out_state:
2629         xfrm_statistics_fini(net);
2630 out_statistics:
2631         return rv;
2632 }
2633
2634 static void __net_exit xfrm_net_exit(struct net *net)
2635 {
2636         xfrm_sysctl_fini(net);
2637         xfrm_policy_fini(net);
2638         xfrm_state_fini(net);
2639         xfrm_statistics_fini(net);
2640 }
2641
2642 static struct pernet_operations __net_initdata xfrm_net_ops = {
2643         .init = xfrm_net_init,
2644         .exit = xfrm_net_exit,
2645 };
2646
2647 void __init xfrm_init(void)
2648 {
2649         register_pernet_subsys(&xfrm_net_ops);
2650         xfrm_input_init();
2651 }
2652
2653 #ifdef CONFIG_AUDITSYSCALL
2654 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2655                                          struct audit_buffer *audit_buf)
2656 {
2657         struct xfrm_sec_ctx *ctx = xp->security;
2658         struct xfrm_selector *sel = &xp->selector;
2659
2660         if (ctx)
2661                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2662                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2663
2664         switch(sel->family) {
2665         case AF_INET:
2666                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2667                 if (sel->prefixlen_s != 32)
2668                         audit_log_format(audit_buf, " src_prefixlen=%d",
2669                                          sel->prefixlen_s);
2670                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2671                 if (sel->prefixlen_d != 32)
2672                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2673                                          sel->prefixlen_d);
2674                 break;
2675         case AF_INET6:
2676                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2677                 if (sel->prefixlen_s != 128)
2678                         audit_log_format(audit_buf, " src_prefixlen=%d",
2679                                          sel->prefixlen_s);
2680                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2681                 if (sel->prefixlen_d != 128)
2682                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2683                                          sel->prefixlen_d);
2684                 break;
2685         }
2686 }
2687
2688 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2689                            uid_t auid, u32 sessionid, u32 secid)
2690 {
2691         struct audit_buffer *audit_buf;
2692
2693         audit_buf = xfrm_audit_start("SPD-add");
2694         if (audit_buf == NULL)
2695                 return;
2696         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2697         audit_log_format(audit_buf, " res=%u", result);
2698         xfrm_audit_common_policyinfo(xp, audit_buf);
2699         audit_log_end(audit_buf);
2700 }
2701 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2702
2703 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2704                               uid_t auid, u32 sessionid, u32 secid)
2705 {
2706         struct audit_buffer *audit_buf;
2707
2708         audit_buf = xfrm_audit_start("SPD-delete");
2709         if (audit_buf == NULL)
2710                 return;
2711         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2712         audit_log_format(audit_buf, " res=%u", result);
2713         xfrm_audit_common_policyinfo(xp, audit_buf);
2714         audit_log_end(audit_buf);
2715 }
2716 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2717 #endif
2718
2719 #ifdef CONFIG_XFRM_MIGRATE
2720 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2721                                        struct xfrm_selector *sel_tgt)
2722 {
2723         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2724                 if (sel_tgt->family == sel_cmp->family &&
2725                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2726                                   sel_cmp->family) == 0 &&
2727                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2728                                   sel_cmp->family) == 0 &&
2729                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2730                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2731                         return 1;
2732                 }
2733         } else {
2734                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2735                         return 1;
2736                 }
2737         }
2738         return 0;
2739 }
2740
2741 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2742                                                      u8 dir, u8 type)
2743 {
2744         struct xfrm_policy *pol, *ret = NULL;
2745         struct hlist_node *entry;
2746         struct hlist_head *chain;
2747         u32 priority = ~0U;
2748
2749         read_lock_bh(&xfrm_policy_lock);
2750         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2751         hlist_for_each_entry(pol, entry, chain, bydst) {
2752                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2753                     pol->type == type) {
2754                         ret = pol;
2755                         priority = ret->priority;
2756                         break;
2757                 }
2758         }
2759         chain = &init_net.xfrm.policy_inexact[dir];
2760         hlist_for_each_entry(pol, entry, chain, bydst) {
2761                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2762                     pol->type == type &&
2763                     pol->priority < priority) {
2764                         ret = pol;
2765                         break;
2766                 }
2767         }
2768
2769         if (ret)
2770                 xfrm_pol_hold(ret);
2771
2772         read_unlock_bh(&xfrm_policy_lock);
2773
2774         return ret;
2775 }
2776
2777 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2778 {
2779         int match = 0;
2780
2781         if (t->mode == m->mode && t->id.proto == m->proto &&
2782             (m->reqid == 0 || t->reqid == m->reqid)) {
2783                 switch (t->mode) {
2784                 case XFRM_MODE_TUNNEL:
2785                 case XFRM_MODE_BEET:
2786                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2787                                           m->old_family) == 0 &&
2788                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2789                                           m->old_family) == 0) {
2790                                 match = 1;
2791                         }
2792                         break;
2793                 case XFRM_MODE_TRANSPORT:
2794                         /* in case of transport mode, template does not store
2795                            any IP addresses, hence we just compare mode and
2796                            protocol */
2797                         match = 1;
2798                         break;
2799                 default:
2800                         break;
2801                 }
2802         }
2803         return match;
2804 }
2805
2806 /* update endpoint address(es) of template(s) */
2807 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2808                                struct xfrm_migrate *m, int num_migrate)
2809 {
2810         struct xfrm_migrate *mp;
2811         int i, j, n = 0;
2812
2813         write_lock_bh(&pol->lock);
2814         if (unlikely(pol->walk.dead)) {
2815                 /* target policy has been deleted */
2816                 write_unlock_bh(&pol->lock);
2817                 return -ENOENT;
2818         }
2819
2820         for (i = 0; i < pol->xfrm_nr; i++) {
2821                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2822                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2823                                 continue;
2824                         n++;
2825                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2826                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2827                                 continue;
2828                         /* update endpoints */
2829                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2830                                sizeof(pol->xfrm_vec[i].id.daddr));
2831                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2832                                sizeof(pol->xfrm_vec[i].saddr));
2833                         pol->xfrm_vec[i].encap_family = mp->new_family;
2834                         /* flush bundles */
2835                         atomic_inc(&pol->genid);
2836                 }
2837         }
2838
2839         write_unlock_bh(&pol->lock);
2840
2841         if (!n)
2842                 return -ENODATA;
2843
2844         return 0;
2845 }
2846
2847 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2848 {
2849         int i, j;
2850
2851         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2852                 return -EINVAL;
2853
2854         for (i = 0; i < num_migrate; i++) {
2855                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2856                                    m[i].old_family) == 0) &&
2857                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2858                                    m[i].old_family) == 0))
2859                         return -EINVAL;
2860                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2861                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2862                         return -EINVAL;
2863
2864                 /* check if there is any duplicated entry */
2865                 for (j = i + 1; j < num_migrate; j++) {
2866                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2867                                     sizeof(m[i].old_daddr)) &&
2868                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2869                                     sizeof(m[i].old_saddr)) &&
2870                             m[i].proto == m[j].proto &&
2871                             m[i].mode == m[j].mode &&
2872                             m[i].reqid == m[j].reqid &&
2873                             m[i].old_family == m[j].old_family)
2874                                 return -EINVAL;
2875                 }
2876         }
2877
2878         return 0;
2879 }
2880
2881 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2882                  struct xfrm_migrate *m, int num_migrate,
2883                  struct xfrm_kmaddress *k)
2884 {
2885         int i, err, nx_cur = 0, nx_new = 0;
2886         struct xfrm_policy *pol = NULL;
2887         struct xfrm_state *x, *xc;
2888         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2889         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2890         struct xfrm_migrate *mp;
2891
2892         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2893                 goto out;
2894
2895         /* Stage 1 - find policy */
2896         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2897                 err = -ENOENT;
2898                 goto out;
2899         }
2900
2901         /* Stage 2 - find and update state(s) */
2902         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2903                 if ((x = xfrm_migrate_state_find(mp))) {
2904                         x_cur[nx_cur] = x;
2905                         nx_cur++;
2906                         if ((xc = xfrm_state_migrate(x, mp))) {
2907                                 x_new[nx_new] = xc;
2908                                 nx_new++;
2909                         } else {
2910                                 err = -ENODATA;
2911                                 goto restore_state;
2912                         }
2913                 }
2914         }
2915
2916         /* Stage 3 - update policy */
2917         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2918                 goto restore_state;
2919
2920         /* Stage 4 - delete old state(s) */
2921         if (nx_cur) {
2922                 xfrm_states_put(x_cur, nx_cur);
2923                 xfrm_states_delete(x_cur, nx_cur);
2924         }
2925
2926         /* Stage 5 - announce */
2927         km_migrate(sel, dir, type, m, num_migrate, k);
2928
2929         xfrm_pol_put(pol);
2930
2931         return 0;
2932 out:
2933         return err;
2934
2935 restore_state:
2936         if (pol)
2937                 xfrm_pol_put(pol);
2938         if (nx_cur)
2939                 xfrm_states_put(x_cur, nx_cur);
2940         if (nx_new)
2941                 xfrm_states_delete(x_new, nx_new);
2942
2943         return err;
2944 }
2945 EXPORT_SYMBOL(xfrm_migrate);
2946 #endif