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 static int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *xdst,
  54                           struct flowi *fl, int family, int strict);
  55
  56
  57 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  58                                                 int dir);
  59
  60 static inline int
  61 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  62 {
  63         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  64                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  65                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  66                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  67                 (fl->proto == sel->proto || !sel->proto) &&
  68                 (fl->oif == sel->ifindex || !sel->ifindex);
  69 }
  70
  71 static inline int
  72 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  73 {
  74         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  75                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  76                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  77                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  78                 (fl->proto == sel->proto || !sel->proto) &&
  79                 (fl->oif == sel->ifindex || !sel->ifindex);
  80 }
  81
  82 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  83                     unsigned short family)
  84 {
  85         switch (family) {
  86         case AF_INET:
  87                 return __xfrm4_selector_match(sel, fl);
  88         case AF_INET6:
  89                 return __xfrm6_selector_match(sel, fl);
  90         }
  91         return 0;
  92 }
  93
  94 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
  95                                                   xfrm_address_t *saddr,
  96                                                   xfrm_address_t *daddr,
  97                                                   int family)
  98 {
  99         struct xfrm_policy_afinfo *afinfo;
 100         struct dst_entry *dst;
 101
 102         afinfo = xfrm_policy_get_afinfo(family);
 103         if (unlikely(afinfo == NULL))
 104                 return ERR_PTR(-EAFNOSUPPORT);
 105
 106         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 107
 108         xfrm_policy_put_afinfo(afinfo);
 109
 110         return dst;
 111 }
 112
 113 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 114                                                 xfrm_address_t *prev_saddr,
 115                                                 xfrm_address_t *prev_daddr,
 116                                                 int family)
 117 {
 118         struct net *net = xs_net(x);
 119         xfrm_address_t *saddr = &x->props.saddr;
 120         xfrm_address_t *daddr = &x->id.daddr;
 121         struct dst_entry *dst;
 122
 123         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 124                 saddr = x->coaddr;
 125                 daddr = prev_daddr;
 126         }
 127         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 128                 saddr = prev_saddr;
 129                 daddr = x->coaddr;
 130         }
 131
 132         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 133
 134         if (!IS_ERR(dst)) {
 135                 if (prev_saddr != saddr)
 136                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 137                 if (prev_daddr != daddr)
 138                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 139         }
 140
 141         return dst;
 142 }
 143
 144 static inline unsigned long make_jiffies(long secs)
 145 {
 146         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 147                 return MAX_SCHEDULE_TIMEOUT-1;
 148         else
 149                 return secs*HZ;
 150 }
 151
 152 static void xfrm_policy_timer(unsigned long data)
 153 {
 154         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 155         unsigned long now = get_seconds();
 156         long next = LONG_MAX;
 157         int warn = 0;
 158         int dir;
 159
 160         read_lock(&xp->lock);
 161
 162         if (unlikely(xp->walk.dead))
 163                 goto out;
 164
 165         dir = xfrm_policy_id2dir(xp->index);
 166
 167         if (xp->lft.hard_add_expires_seconds) {
 168                 long tmo = xp->lft.hard_add_expires_seconds +
 169                         xp->curlft.add_time - now;
 170                 if (tmo <= 0)
 171                         goto expired;
 172                 if (tmo < next)
 173                         next = tmo;
 174         }
 175         if (xp->lft.hard_use_expires_seconds) {
 176                 long tmo = xp->lft.hard_use_expires_seconds +
 177                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 178                 if (tmo <= 0)
 179                         goto expired;
 180                 if (tmo < next)
 181                         next = tmo;
 182         }
 183         if (xp->lft.soft_add_expires_seconds) {
 184                 long tmo = xp->lft.soft_add_expires_seconds +
 185                         xp->curlft.add_time - now;
 186                 if (tmo <= 0) {
 187                         warn = 1;
 188                         tmo = XFRM_KM_TIMEOUT;
 189                 }
 190                 if (tmo < next)
 191                         next = tmo;
 192         }
 193         if (xp->lft.soft_use_expires_seconds) {
 194                 long tmo = xp->lft.soft_use_expires_seconds +
 195                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 196                 if (tmo <= 0) {
 197                         warn = 1;
 198                         tmo = XFRM_KM_TIMEOUT;
 199                 }
 200                 if (tmo < next)
 201                         next = tmo;
 202         }
 203
 204         if (warn)
 205                 km_policy_expired(xp, dir, 0, 0);
 206         if (next != LONG_MAX &&
 207             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 208                 xfrm_pol_hold(xp);
 209
 210 out:
 211         read_unlock(&xp->lock);
 212         xfrm_pol_put(xp);
 213         return;
 214
 215 expired:
 216         read_unlock(&xp->lock);
 217         if (!xfrm_policy_delete(xp, dir))
 218                 km_policy_expired(xp, dir, 1, 0);
 219         xfrm_pol_put(xp);
 220 }
 221
 222 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 223 {
 224         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 225
 226         if (unlikely(pol->walk.dead))
 227                 flo = NULL;
 228         else
 229                 xfrm_pol_hold(pol);
 230
 231         return flo;
 232 }
 233
 234 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 235 {
 236         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 237
 238         return !pol->walk.dead;
 239 }
 240
 241 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 242 {
 243         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 244 }
 245
 246 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 247         .get = xfrm_policy_flo_get,
 248         .check = xfrm_policy_flo_check,
 249         .delete = xfrm_policy_flo_delete,
 250 };
 251
 252 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 253  * SPD calls.
 254  */
 255
 256 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 257 {
 258         struct xfrm_policy *policy;
 259
 260         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 261
 262         if (policy) {
 263                 write_pnet(&policy->xp_net, net);
 264                 INIT_LIST_HEAD(&policy->walk.all);
 265                 INIT_HLIST_NODE(&policy->bydst);
 266                 INIT_HLIST_NODE(&policy->byidx);
 267                 rwlock_init(&policy->lock);
 268                 atomic_set(&policy->refcnt, 1);
 269                 setup_timer(&policy->timer, xfrm_policy_timer,
 270                                 (unsigned long)policy);
 271                 policy->flo.ops = &xfrm_policy_fc_ops;
 272         }
 273         return policy;
 274 }
 275 EXPORT_SYMBOL(xfrm_policy_alloc);
 276
 277 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 278
 279 void xfrm_policy_destroy(struct xfrm_policy *policy)
 280 {
 281         BUG_ON(!policy->walk.dead);
 282
 283         if (del_timer(&policy->timer))
 284                 BUG();
 285
 286         security_xfrm_policy_free(policy->security);
 287         kfree(policy);
 288 }
 289 EXPORT_SYMBOL(xfrm_policy_destroy);
 290
 291 /* Rule must be locked. Release descentant resources, announce
 292  * entry dead. The rule must be unlinked from lists to the moment.
 293  */
 294
 295 static void xfrm_policy_kill(struct xfrm_policy *policy)
 296 {
 297         policy->walk.dead = 1;
 298
 299         atomic_inc(&policy->genid);
 300
 301         if (del_timer(&policy->timer))
 302                 xfrm_pol_put(policy);
 303
 304         xfrm_pol_put(policy);
 305 }
 306
 307 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 308
 309 static inline unsigned int idx_hash(struct net *net, u32 index)
 310 {
 311         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 312 }
 313
 314 static struct hlist_head *policy_hash_bysel(struct net *net, struct xfrm_selector *sel, unsigned short family, int dir)
 315 {
 316         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 317         unsigned int hash = __sel_hash(sel, family, hmask);
 318
 319         return (hash == hmask + 1 ?
 320                 &net->xfrm.policy_inexact[dir] :
 321                 net->xfrm.policy_bydst[dir].table + hash);
 322 }
 323
 324 static struct hlist_head *policy_hash_direct(struct net *net, xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 325 {
 326         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 327         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 328
 329         return net->xfrm.policy_bydst[dir].table + hash;
 330 }
 331
 332 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 333                                    struct hlist_head *ndsttable,
 334                                    unsigned int nhashmask)
 335 {
 336         struct hlist_node *entry, *tmp, *entry0 = NULL;
 337         struct xfrm_policy *pol;
 338         unsigned int h0 = 0;
 339
 340 redo:
 341         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 342                 unsigned int h;
 343
 344                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 345                                 pol->family, nhashmask);
 346                 if (!entry0) {
 347                         hlist_del(entry);
 348                         hlist_add_head(&pol->bydst, ndsttable+h);
 349                         h0 = h;
 350                 } else {
 351                         if (h != h0)
 352                                 continue;
 353                         hlist_del(entry);
 354                         hlist_add_after(entry0, &pol->bydst);
 355                 }
 356                 entry0 = entry;
 357         }
 358         if (!hlist_empty(list)) {
 359                 entry0 = NULL;
 360                 goto redo;
 361         }
 362 }
 363
 364 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 365                                    struct hlist_head *nidxtable,
 366                                    unsigned int nhashmask)
 367 {
 368         struct hlist_node *entry, *tmp;
 369         struct xfrm_policy *pol;
 370
 371         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 372                 unsigned int h;
 373
 374                 h = __idx_hash(pol->index, nhashmask);
 375                 hlist_add_head(&pol->byidx, nidxtable+h);
 376         }
 377 }
 378
 379 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 380 {
 381         return ((old_hmask + 1) << 1) - 1;
 382 }
 383
 384 static void xfrm_bydst_resize(struct net *net, int dir)
 385 {
 386         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 387         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 388         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 389         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 390         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 391         int i;
 392
 393         if (!ndst)
 394                 return;
 395
 396         write_lock_bh(&xfrm_policy_lock);
 397
 398         for (i = hmask; i >= 0; i--)
 399                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 400
 401         net->xfrm.policy_bydst[dir].table = ndst;
 402         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 403
 404         write_unlock_bh(&xfrm_policy_lock);
 405
 406         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 407 }
 408
 409 static void xfrm_byidx_resize(struct net *net, int total)
 410 {
 411         unsigned int hmask = net->xfrm.policy_idx_hmask;
 412         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 413         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 414         struct hlist_head *oidx = net->xfrm.policy_byidx;
 415         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 416         int i;
 417
 418         if (!nidx)
 419                 return;
 420
 421         write_lock_bh(&xfrm_policy_lock);
 422
 423         for (i = hmask; i >= 0; i--)
 424                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 425
 426         net->xfrm.policy_byidx = nidx;
 427         net->xfrm.policy_idx_hmask = nhashmask;
 428
 429         write_unlock_bh(&xfrm_policy_lock);
 430
 431         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 432 }
 433
 434 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 435 {
 436         unsigned int cnt = net->xfrm.policy_count[dir];
 437         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 438
 439         if (total)
 440                 *total += cnt;
 441
 442         if ((hmask + 1) < xfrm_policy_hashmax &&
 443             cnt > hmask)
 444                 return 1;
 445
 446         return 0;
 447 }
 448
 449 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 450 {
 451         unsigned int hmask = net->xfrm.policy_idx_hmask;
 452
 453         if ((hmask + 1) < xfrm_policy_hashmax &&
 454             total > hmask)
 455                 return 1;
 456
 457         return 0;
 458 }
 459
 460 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 461 {
 462         read_lock_bh(&xfrm_policy_lock);
 463         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 464         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 465         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 466         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 467         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 468         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 469         si->spdhcnt = net->xfrm.policy_idx_hmask;
 470         si->spdhmcnt = xfrm_policy_hashmax;
 471         read_unlock_bh(&xfrm_policy_lock);
 472 }
 473 EXPORT_SYMBOL(xfrm_spd_getinfo);
 474
 475 static DEFINE_MUTEX(hash_resize_mutex);
 476 static void xfrm_hash_resize(struct work_struct *work)
 477 {
 478         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 479         int dir, total;
 480
 481         mutex_lock(&hash_resize_mutex);
 482
 483         total = 0;
 484         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 485                 if (xfrm_bydst_should_resize(net, dir, &total))
 486                         xfrm_bydst_resize(net, dir);
 487         }
 488         if (xfrm_byidx_should_resize(net, total))
 489                 xfrm_byidx_resize(net, total);
 490
 491         mutex_unlock(&hash_resize_mutex);
 492 }
 493
 494 /* Generate new index... KAME seems to generate them ordered by cost
 495  * of an absolute inpredictability of ordering of rules. This will not pass. */
 496 static u32 xfrm_gen_index(struct net *net, int dir)
 497 {
 498         static u32 idx_generator;
 499
 500         for (;;) {
 501                 struct hlist_node *entry;
 502                 struct hlist_head *list;
 503                 struct xfrm_policy *p;
 504                 u32 idx;
 505                 int found;
 506
 507                 idx = (idx_generator | dir);
 508                 idx_generator += 8;
 509                 if (idx == 0)
 510                         idx = 8;
 511                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 512                 found = 0;
 513                 hlist_for_each_entry(p, entry, list, byidx) {
 514                         if (p->index == idx) {
 515                                 found = 1;
 516                                 break;
 517                         }
 518                 }
 519                 if (!found)
 520                         return idx;
 521         }
 522 }
 523
 524 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 525 {
 526         u32 *p1 = (u32 *) s1;
 527         u32 *p2 = (u32 *) s2;
 528         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 529         int i;
 530
 531         for (i = 0; i < len; i++) {
 532                 if (p1[i] != p2[i])
 533                         return 1;
 534         }
 535
 536         return 0;
 537 }
 538
 539 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 540 {
 541         struct net *net = xp_net(policy);
 542         struct xfrm_policy *pol;
 543         struct xfrm_policy *delpol;
 544         struct hlist_head *chain;
 545         struct hlist_node *entry, *newpos;
 546         u32 mark = policy->mark.v & policy->mark.m;
 547
 548         write_lock_bh(&xfrm_policy_lock);
 549         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 550         delpol = NULL;
 551         newpos = NULL;
 552         hlist_for_each_entry(pol, entry, chain, bydst) {
 553                 if (pol->type == policy->type &&
 554                     !selector_cmp(&pol->selector, &policy->selector) &&
 555                     (mark & pol->mark.m) == pol->mark.v &&
 556                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 557                     !WARN_ON(delpol)) {
 558                         if (excl) {
 559                                 write_unlock_bh(&xfrm_policy_lock);
 560                                 return -EEXIST;
 561                         }
 562                         delpol = pol;
 563                         if (policy->priority > pol->priority)
 564                                 continue;
 565                 } else if (policy->priority >= pol->priority) {
 566                         newpos = &pol->bydst;
 567                         continue;
 568                 }
 569                 if (delpol)
 570                         break;
 571         }
 572         if (newpos)
 573                 hlist_add_after(newpos, &policy->bydst);
 574         else
 575                 hlist_add_head(&policy->bydst, chain);
 576         xfrm_pol_hold(policy);
 577         net->xfrm.policy_count[dir]++;
 578         atomic_inc(&flow_cache_genid);
 579         if (delpol)
 580                 __xfrm_policy_unlink(delpol, dir);
 581         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 582         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 583         policy->curlft.add_time = get_seconds();
 584         policy->curlft.use_time = 0;
 585         if (!mod_timer(&policy->timer, jiffies + HZ))
 586                 xfrm_pol_hold(policy);
 587         list_add(&policy->walk.all, &net->xfrm.policy_all);
 588         write_unlock_bh(&xfrm_policy_lock);
 589
 590         if (delpol)
 591                 xfrm_policy_kill(delpol);
 592         else if (xfrm_bydst_should_resize(net, dir, NULL))
 593                 schedule_work(&net->xfrm.policy_hash_work);
 594
 595         return 0;
 596 }
 597 EXPORT_SYMBOL(xfrm_policy_insert);
 598
 599 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 600                                           int dir, struct xfrm_selector *sel,
 601                                           struct xfrm_sec_ctx *ctx, int delete,
 602                                           int *err)
 603 {
 604         struct xfrm_policy *pol, *ret;
 605         struct hlist_head *chain;
 606         struct hlist_node *entry;
 607
 608         *err = 0;
 609         write_lock_bh(&xfrm_policy_lock);
 610         chain = policy_hash_bysel(net, sel, sel->family, dir);
 611         ret = NULL;
 612         hlist_for_each_entry(pol, entry, chain, bydst) {
 613                 if (pol->type == type &&
 614                     (mark & pol->mark.m) == pol->mark.v &&
 615                     !selector_cmp(sel, &pol->selector) &&
 616                     xfrm_sec_ctx_match(ctx, pol->security)) {
 617                         xfrm_pol_hold(pol);
 618                         if (delete) {
 619                                 *err = security_xfrm_policy_delete(
 620                                                                 pol->security);
 621                                 if (*err) {
 622                                         write_unlock_bh(&xfrm_policy_lock);
 623                                         return pol;
 624                                 }
 625                                 __xfrm_policy_unlink(pol, dir);
 626                         }
 627                         ret = pol;
 628                         break;
 629                 }
 630         }
 631         write_unlock_bh(&xfrm_policy_lock);
 632
 633         if (ret && delete)
 634                 xfrm_policy_kill(ret);
 635         return ret;
 636 }
 637 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 638
 639 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 640                                      int dir, u32 id, int delete, int *err)
 641 {
 642         struct xfrm_policy *pol, *ret;
 643         struct hlist_head *chain;
 644         struct hlist_node *entry;
 645
 646         *err = -ENOENT;
 647         if (xfrm_policy_id2dir(id) != dir)
 648                 return NULL;
 649
 650         *err = 0;
 651         write_lock_bh(&xfrm_policy_lock);
 652         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 653         ret = NULL;
 654         hlist_for_each_entry(pol, entry, chain, byidx) {
 655                 if (pol->type == type && pol->index == id &&
 656                     (mark & pol->mark.m) == pol->mark.v) {
 657                         xfrm_pol_hold(pol);
 658                         if (delete) {
 659                                 *err = security_xfrm_policy_delete(
 660                                                                 pol->security);
 661                                 if (*err) {
 662                                         write_unlock_bh(&xfrm_policy_lock);
 663                                         return pol;
 664                                 }
 665                                 __xfrm_policy_unlink(pol, dir);
 666                         }
 667                         ret = pol;
 668                         break;
 669                 }
 670         }
 671         write_unlock_bh(&xfrm_policy_lock);
 672
 673         if (ret && delete)
 674                 xfrm_policy_kill(ret);
 675         return ret;
 676 }
 677 EXPORT_SYMBOL(xfrm_policy_byid);
 678
 679 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 680 static inline int
 681 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 682 {
 683         int dir, err = 0;
 684
 685         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 686                 struct xfrm_policy *pol;
 687                 struct hlist_node *entry;
 688                 int i;
 689
 690                 hlist_for_each_entry(pol, entry,
 691                                      &net->xfrm.policy_inexact[dir], bydst) {
 692                         if (pol->type != type)
 693                                 continue;
 694                         err = security_xfrm_policy_delete(pol->security);
 695                         if (err) {
 696                                 xfrm_audit_policy_delete(pol, 0,
 697                                                          audit_info->loginuid,
 698                                                          audit_info->sessionid,
 699                                                          audit_info->secid);
 700                                 return err;
 701                         }
 702                 }
 703                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 704                         hlist_for_each_entry(pol, entry,
 705                                              net->xfrm.policy_bydst[dir].table + i,
 706                                              bydst) {
 707                                 if (pol->type != type)
 708                                         continue;
 709                                 err = security_xfrm_policy_delete(
 710                                                                 pol->security);
 711                                 if (err) {
 712                                         xfrm_audit_policy_delete(pol, 0,
 713                                                         audit_info->loginuid,
 714                                                         audit_info->sessionid,
 715                                                         audit_info->secid);
 716                                         return err;
 717                                 }
 718                         }
 719                 }
 720         }
 721         return err;
 722 }
 723 #else
 724 static inline int
 725 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 726 {
 727         return 0;
 728 }
 729 #endif
 730
 731 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 732 {
 733         int dir, err = 0, cnt = 0;
 734
 735         write_lock_bh(&xfrm_policy_lock);
 736
 737         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 738         if (err)
 739                 goto out;
 740
 741         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 742                 struct xfrm_policy *pol;
 743                 struct hlist_node *entry;
 744                 int i;
 745
 746         again1:
 747                 hlist_for_each_entry(pol, entry,
 748                                      &net->xfrm.policy_inexact[dir], bydst) {
 749                         if (pol->type != type)
 750                                 continue;
 751                         __xfrm_policy_unlink(pol, dir);
 752                         write_unlock_bh(&xfrm_policy_lock);
 753                         cnt++;
 754
 755                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 756                                                  audit_info->sessionid,
 757                                                  audit_info->secid);
 758
 759                         xfrm_policy_kill(pol);
 760
 761                         write_lock_bh(&xfrm_policy_lock);
 762                         goto again1;
 763                 }
 764
 765                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 766         again2:
 767                         hlist_for_each_entry(pol, entry,
 768                                              net->xfrm.policy_bydst[dir].table + i,
 769                                              bydst) {
 770                                 if (pol->type != type)
 771                                         continue;
 772                                 __xfrm_policy_unlink(pol, dir);
 773                                 write_unlock_bh(&xfrm_policy_lock);
 774                                 cnt++;
 775
 776                                 xfrm_audit_policy_delete(pol, 1,
 777                                                          audit_info->loginuid,
 778                                                          audit_info->sessionid,
 779                                                          audit_info->secid);
 780                                 xfrm_policy_kill(pol);
 781
 782                                 write_lock_bh(&xfrm_policy_lock);
 783                                 goto again2;
 784                         }
 785                 }
 786
 787         }
 788         if (!cnt)
 789                 err = -ESRCH;
 790 out:
 791         write_unlock_bh(&xfrm_policy_lock);
 792         return err;
 793 }
 794 EXPORT_SYMBOL(xfrm_policy_flush);
 795
 796 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 797                      int (*func)(struct xfrm_policy *, int, int, void*),
 798                      void *data)
 799 {
 800         struct xfrm_policy *pol;
 801         struct xfrm_policy_walk_entry *x;
 802         int error = 0;
 803
 804         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 805             walk->type != XFRM_POLICY_TYPE_ANY)
 806                 return -EINVAL;
 807
 808         if (list_empty(&walk->walk.all) && walk->seq != 0)
 809                 return 0;
 810
 811         write_lock_bh(&xfrm_policy_lock);
 812         if (list_empty(&walk->walk.all))
 813                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 814         else
 815                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 816         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 817                 if (x->dead)
 818                         continue;
 819                 pol = container_of(x, struct xfrm_policy, walk);
 820                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 821                     walk->type != pol->type)
 822                         continue;
 823                 error = func(pol, xfrm_policy_id2dir(pol->index),
 824                              walk->seq, data);
 825                 if (error) {
 826                         list_move_tail(&walk->walk.all, &x->all);
 827                         goto out;
 828                 }
 829                 walk->seq++;
 830         }
 831         if (walk->seq == 0) {
 832                 error = -ENOENT;
 833                 goto out;
 834         }
 835         list_del_init(&walk->walk.all);
 836 out:
 837         write_unlock_bh(&xfrm_policy_lock);
 838         return error;
 839 }
 840 EXPORT_SYMBOL(xfrm_policy_walk);
 841
 842 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 843 {
 844         INIT_LIST_HEAD(&walk->walk.all);
 845         walk->walk.dead = 1;
 846         walk->type = type;
 847         walk->seq = 0;
 848 }
 849 EXPORT_SYMBOL(xfrm_policy_walk_init);
 850
 851 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 852 {
 853         if (list_empty(&walk->walk.all))
 854                 return;
 855
 856         write_lock_bh(&xfrm_policy_lock);
 857         list_del(&walk->walk.all);
 858         write_unlock_bh(&xfrm_policy_lock);
 859 }
 860 EXPORT_SYMBOL(xfrm_policy_walk_done);
 861
 862 /*
 863  * Find policy to apply to this flow.
 864  *
 865  * Returns 0 if policy found, else an -errno.
 866  */
 867 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 868                              u8 type, u16 family, int dir)
 869 {
 870         struct xfrm_selector *sel = &pol->selector;
 871         int match, ret = -ESRCH;
 872
 873         if (pol->family != family ||
 874             (fl->mark & pol->mark.m) != pol->mark.v ||
 875             pol->type != type)
 876                 return ret;
 877
 878         match = xfrm_selector_match(sel, fl, family);
 879         if (match)
 880                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 881                                                   dir);
 882
 883         return ret;
 884 }
 885
 886 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 887                                                      struct flowi *fl,
 888                                                      u16 family, u8 dir)
 889 {
 890         int err;
 891         struct xfrm_policy *pol, *ret;
 892         xfrm_address_t *daddr, *saddr;
 893         struct hlist_node *entry;
 894         struct hlist_head *chain;
 895         u32 priority = ~0U;
 896
 897         daddr = xfrm_flowi_daddr(fl, family);
 898         saddr = xfrm_flowi_saddr(fl, family);
 899         if (unlikely(!daddr || !saddr))
 900                 return NULL;
 901
 902         read_lock_bh(&xfrm_policy_lock);
 903         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 904         ret = NULL;
 905         hlist_for_each_entry(pol, entry, chain, bydst) {
 906                 err = xfrm_policy_match(pol, fl, type, family, dir);
 907                 if (err) {
 908                         if (err == -ESRCH)
 909                                 continue;
 910                         else {
 911                                 ret = ERR_PTR(err);
 912                                 goto fail;
 913                         }
 914                 } else {
 915                         ret = pol;
 916                         priority = ret->priority;
 917                         break;
 918                 }
 919         }
 920         chain = &net->xfrm.policy_inexact[dir];
 921         hlist_for_each_entry(pol, entry, chain, bydst) {
 922                 err = xfrm_policy_match(pol, fl, type, family, dir);
 923                 if (err) {
 924                         if (err == -ESRCH)
 925                                 continue;
 926                         else {
 927                                 ret = ERR_PTR(err);
 928                                 goto fail;
 929                         }
 930                 } else if (pol->priority < priority) {
 931                         ret = pol;
 932                         break;
 933                 }
 934         }
 935         if (ret)
 936                 xfrm_pol_hold(ret);
 937 fail:
 938         read_unlock_bh(&xfrm_policy_lock);
 939
 940         return ret;
 941 }
 942
 943 static struct xfrm_policy *
 944 __xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family, u8 dir)
 945 {
 946 #ifdef CONFIG_XFRM_SUB_POLICY
 947         struct xfrm_policy *pol;
 948
 949         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
 950         if (pol != NULL)
 951                 return pol;
 952 #endif
 953         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 954 }
 955
 956 static struct flow_cache_object *
 957 xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
 958                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
 959 {
 960         struct xfrm_policy *pol;
 961
 962         if (old_obj)
 963                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
 964
 965         pol = __xfrm_policy_lookup(net, fl, family, dir);
 966         if (IS_ERR_OR_NULL(pol))
 967                 return ERR_CAST(pol);
 968
 969         /* Resolver returns two references:
 970          * one for cache and one for caller of flow_cache_lookup() */
 971         xfrm_pol_hold(pol);
 972
 973         return &pol->flo;
 974 }
 975
 976 static inline int policy_to_flow_dir(int dir)
 977 {
 978         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 979             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 980             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 981                 return dir;
 982         switch (dir) {
 983         default:
 984         case XFRM_POLICY_IN:
 985                 return FLOW_DIR_IN;
 986         case XFRM_POLICY_OUT:
 987                 return FLOW_DIR_OUT;
 988         case XFRM_POLICY_FWD:
 989                 return FLOW_DIR_FWD;
 990         }
 991 }
 992
 993 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
 994 {
 995         struct xfrm_policy *pol;
 996
 997         read_lock_bh(&xfrm_policy_lock);
 998         if ((pol = sk->sk_policy[dir]) != NULL) {
 999                 int match = xfrm_selector_match(&pol->selector, fl,
1000                                                 sk->sk_family);
1001                 int err = 0;
1002
1003                 if (match) {
1004                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1005                                 pol = NULL;
1006                                 goto out;
1007                         }
1008                         err = security_xfrm_policy_lookup(pol->security,
1009                                                       fl->secid,
1010                                                       policy_to_flow_dir(dir));
1011                         if (!err)
1012                                 xfrm_pol_hold(pol);
1013                         else if (err == -ESRCH)
1014                                 pol = NULL;
1015                         else
1016                                 pol = ERR_PTR(err);
1017                 } else
1018                         pol = NULL;
1019         }
1020 out:
1021         read_unlock_bh(&xfrm_policy_lock);
1022         return pol;
1023 }
1024
1025 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1026 {
1027         struct net *net = xp_net(pol);
1028         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1029                                                      pol->family, dir);
1030
1031         list_add(&pol->walk.all, &net->xfrm.policy_all);
1032         hlist_add_head(&pol->bydst, chain);
1033         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1034         net->xfrm.policy_count[dir]++;
1035         xfrm_pol_hold(pol);
1036
1037         if (xfrm_bydst_should_resize(net, dir, NULL))
1038                 schedule_work(&net->xfrm.policy_hash_work);
1039 }
1040
1041 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1042                                                 int dir)
1043 {
1044         struct net *net = xp_net(pol);
1045
1046         if (hlist_unhashed(&pol->bydst))
1047                 return NULL;
1048
1049         hlist_del(&pol->bydst);
1050         hlist_del(&pol->byidx);
1051         list_del(&pol->walk.all);
1052         net->xfrm.policy_count[dir]--;
1053
1054         return pol;
1055 }
1056
1057 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1058 {
1059         write_lock_bh(&xfrm_policy_lock);
1060         pol = __xfrm_policy_unlink(pol, dir);
1061         write_unlock_bh(&xfrm_policy_lock);
1062         if (pol) {
1063                 xfrm_policy_kill(pol);
1064                 return 0;
1065         }
1066         return -ENOENT;
1067 }
1068 EXPORT_SYMBOL(xfrm_policy_delete);
1069
1070 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1071 {
1072         struct net *net = xp_net(pol);
1073         struct xfrm_policy *old_pol;
1074
1075 #ifdef CONFIG_XFRM_SUB_POLICY
1076         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1077                 return -EINVAL;
1078 #endif
1079
1080         write_lock_bh(&xfrm_policy_lock);
1081         old_pol = sk->sk_policy[dir];
1082         sk->sk_policy[dir] = pol;
1083         if (pol) {
1084                 pol->curlft.add_time = get_seconds();
1085                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1086                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1087         }
1088         if (old_pol)
1089                 /* Unlinking succeeds always. This is the only function
1090                  * allowed to delete or replace socket policy.
1091                  */
1092                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1093         write_unlock_bh(&xfrm_policy_lock);
1094
1095         if (old_pol) {
1096                 xfrm_policy_kill(old_pol);
1097         }
1098         return 0;
1099 }
1100
1101 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1102 {
1103         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1104
1105         if (newp) {
1106                 newp->selector = old->selector;
1107                 if (security_xfrm_policy_clone(old->security,
1108                                                &newp->security)) {
1109                         kfree(newp);
1110                         return NULL;  /* ENOMEM */
1111                 }
1112                 newp->lft = old->lft;
1113                 newp->curlft = old->curlft;
1114                 newp->mark = old->mark;
1115                 newp->action = old->action;
1116                 newp->flags = old->flags;
1117                 newp->xfrm_nr = old->xfrm_nr;
1118                 newp->index = old->index;
1119                 newp->type = old->type;
1120                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1121                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1122                 write_lock_bh(&xfrm_policy_lock);
1123                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1124                 write_unlock_bh(&xfrm_policy_lock);
1125                 xfrm_pol_put(newp);
1126         }
1127         return newp;
1128 }
1129
1130 int __xfrm_sk_clone_policy(struct sock *sk)
1131 {
1132         struct xfrm_policy *p0 = sk->sk_policy[0],
1133                            *p1 = sk->sk_policy[1];
1134
1135         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1136         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1137                 return -ENOMEM;
1138         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1139                 return -ENOMEM;
1140         return 0;
1141 }
1142
1143 static int
1144 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1145                unsigned short family)
1146 {
1147         int err;
1148         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1149
1150         if (unlikely(afinfo == NULL))
1151                 return -EINVAL;
1152         err = afinfo->get_saddr(net, local, remote);
1153         xfrm_policy_put_afinfo(afinfo);
1154         return err;
1155 }
1156
1157 /* Resolve list of templates for the flow, given policy. */
1158
1159 static int
1160 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1161                       struct xfrm_state **xfrm,
1162                       unsigned short family)
1163 {
1164         struct net *net = xp_net(policy);
1165         int nx;
1166         int i, error;
1167         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1168         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1169         xfrm_address_t tmp;
1170
1171         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1172                 struct xfrm_state *x;
1173                 xfrm_address_t *remote = daddr;
1174                 xfrm_address_t *local  = saddr;
1175                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1176
1177                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1178                     tmpl->mode == XFRM_MODE_BEET) {
1179                         remote = &tmpl->id.daddr;
1180                         local = &tmpl->saddr;
1181                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1182                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1183                                 if (error)
1184                                         goto fail;
1185                                 local = &tmp;
1186                         }
1187                 }
1188
1189                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1190
1191                 if (x && x->km.state == XFRM_STATE_VALID) {
1192                         xfrm[nx++] = x;
1193                         daddr = remote;
1194                         saddr = local;
1195                         continue;
1196                 }
1197                 if (x) {
1198                         error = (x->km.state == XFRM_STATE_ERROR ?
1199                                  -EINVAL : -EAGAIN);
1200                         xfrm_state_put(x);
1201                 }
1202                 else if (error == -ESRCH)
1203                         error = -EAGAIN;
1204
1205                 if (!tmpl->optional)
1206                         goto fail;
1207         }
1208         return nx;
1209
1210 fail:
1211         for (nx--; nx>=0; nx--)
1212                 xfrm_state_put(xfrm[nx]);
1213         return error;
1214 }
1215
1216 static int
1217 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1218                   struct xfrm_state **xfrm,
1219                   unsigned short family)
1220 {
1221         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1222         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1223         int cnx = 0;
1224         int error;
1225         int ret;
1226         int i;
1227
1228         for (i = 0; i < npols; i++) {
1229                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1230                         error = -ENOBUFS;
1231                         goto fail;
1232                 }
1233
1234                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1235                 if (ret < 0) {
1236                         error = ret;
1237                         goto fail;
1238                 } else
1239                         cnx += ret;
1240         }
1241
1242         /* found states are sorted for outbound processing */
1243         if (npols > 1)
1244                 xfrm_state_sort(xfrm, tpp, cnx, family);
1245
1246         return cnx;
1247
1248  fail:
1249         for (cnx--; cnx>=0; cnx--)
1250                 xfrm_state_put(tpp[cnx]);
1251         return error;
1252
1253 }
1254
1255 /* Check that the bundle accepts the flow and its components are
1256  * still valid.
1257  */
1258
1259 static inline int xfrm_get_tos(struct flowi *fl, int family)
1260 {
1261         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1262         int tos;
1263
1264         if (!afinfo)
1265                 return -EINVAL;
1266
1267         tos = afinfo->get_tos(fl);
1268
1269         xfrm_policy_put_afinfo(afinfo);
1270
1271         return tos;
1272 }
1273
1274 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1275 {
1276         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1277         struct dst_entry *dst = &xdst->u.dst;
1278
1279         if (xdst->route == NULL) {
1280                 /* Dummy bundle - if it has xfrms we were not
1281                  * able to build bundle as template resolution failed.
1282                  * It means we need to try again resolving. */
1283                 if (xdst->num_xfrms > 0)
1284                         return NULL;
1285         } else {
1286                 /* Real bundle */
1287                 if (stale_bundle(dst))
1288                         return NULL;
1289         }
1290
1291         dst_hold(dst);
1292         return flo;
1293 }
1294
1295 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1296 {
1297         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1298         struct dst_entry *dst = &xdst->u.dst;
1299
1300         if (!xdst->route)
1301                 return 0;
1302         if (stale_bundle(dst))
1303                 return 0;
1304
1305         return 1;
1306 }
1307
1308 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1309 {
1310         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1311         struct dst_entry *dst = &xdst->u.dst;
1312
1313         dst_free(dst);
1314 }
1315
1316 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1317         .get = xfrm_bundle_flo_get,
1318         .check = xfrm_bundle_flo_check,
1319         .delete = xfrm_bundle_flo_delete,
1320 };
1321
1322 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1323 {
1324         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1325         struct dst_ops *dst_ops;
1326         struct xfrm_dst *xdst;
1327
1328         if (!afinfo)
1329                 return ERR_PTR(-EINVAL);
1330
1331         switch (family) {
1332         case AF_INET:
1333                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1334                 break;
1335 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1336         case AF_INET6:
1337                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1338                 break;
1339 #endif
1340         default:
1341                 BUG();
1342         }
1343         xdst = dst_alloc(dst_ops) ?: ERR_PTR(-ENOBUFS);
1344         xfrm_policy_put_afinfo(afinfo);
1345
1346         xdst->flo.ops = &xfrm_bundle_fc_ops;
1347
1348         return xdst;
1349 }
1350
1351 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1352                                  int nfheader_len)
1353 {
1354         struct xfrm_policy_afinfo *afinfo =
1355                 xfrm_policy_get_afinfo(dst->ops->family);
1356         int err;
1357
1358         if (!afinfo)
1359                 return -EINVAL;
1360
1361         err = afinfo->init_path(path, dst, nfheader_len);
1362
1363         xfrm_policy_put_afinfo(afinfo);
1364
1365         return err;
1366 }
1367
1368 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1369                                 struct flowi *fl)
1370 {
1371         struct xfrm_policy_afinfo *afinfo =
1372                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1373         int err;
1374
1375         if (!afinfo)
1376                 return -EINVAL;
1377
1378         err = afinfo->fill_dst(xdst, dev, fl);
1379
1380         xfrm_policy_put_afinfo(afinfo);
1381
1382         return err;
1383 }
1384
1385
1386 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1387  * all the metrics... Shortly, bundle a bundle.
1388  */
1389
1390 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1391                                             struct xfrm_state **xfrm, int nx,
1392                                             struct flowi *fl,
1393                                             struct dst_entry *dst)
1394 {
1395         struct net *net = xp_net(policy);
1396         unsigned long now = jiffies;
1397         struct net_device *dev;
1398         struct dst_entry *dst_prev = NULL;
1399         struct dst_entry *dst0 = NULL;
1400         int i = 0;
1401         int err;
1402         int header_len = 0;
1403         int nfheader_len = 0;
1404         int trailer_len = 0;
1405         int tos;
1406         int family = policy->selector.family;
1407         xfrm_address_t saddr, daddr;
1408
1409         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1410
1411         tos = xfrm_get_tos(fl, family);
1412         err = tos;
1413         if (tos < 0)
1414                 goto put_states;
1415
1416         dst_hold(dst);
1417
1418         for (; i < nx; i++) {
1419                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1420                 struct dst_entry *dst1 = &xdst->u.dst;
1421
1422                 err = PTR_ERR(xdst);
1423                 if (IS_ERR(xdst)) {
1424                         dst_release(dst);
1425                         goto put_states;
1426                 }
1427
1428                 if (!dst_prev)
1429                         dst0 = dst1;
1430                 else {
1431                         dst_prev->child = dst_clone(dst1);
1432                         dst1->flags |= DST_NOHASH;
1433                 }
1434
1435                 xdst->route = dst;
1436                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1437
1438                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1439                         family = xfrm[i]->props.family;
1440                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1441                                               family);
1442                         err = PTR_ERR(dst);
1443                         if (IS_ERR(dst))
1444                                 goto put_states;
1445                 } else
1446                         dst_hold(dst);
1447
1448                 dst1->xfrm = xfrm[i];
1449                 xdst->xfrm_genid = xfrm[i]->genid;
1450
1451                 dst1->obsolete = -1;
1452                 dst1->flags |= DST_HOST;
1453                 dst1->lastuse = now;
1454
1455                 dst1->input = dst_discard;
1456                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1457
1458                 dst1->next = dst_prev;
1459                 dst_prev = dst1;
1460
1461                 header_len += xfrm[i]->props.header_len;
1462                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1463                         nfheader_len += xfrm[i]->props.header_len;
1464                 trailer_len += xfrm[i]->props.trailer_len;
1465         }
1466
1467         dst_prev->child = dst;
1468         dst0->path = dst;
1469
1470         err = -ENODEV;
1471         dev = dst->dev;
1472         if (!dev)
1473                 goto free_dst;
1474
1475         /* Copy neighbour for reachability confirmation */
1476         dst0->neighbour = neigh_clone(dst->neighbour);
1477
1478         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1479         xfrm_init_pmtu(dst_prev);
1480
1481         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1482                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1483
1484                 err = xfrm_fill_dst(xdst, dev, fl);
1485                 if (err)
1486                         goto free_dst;
1487
1488                 dst_prev->header_len = header_len;
1489                 dst_prev->trailer_len = trailer_len;
1490                 header_len -= xdst->u.dst.xfrm->props.header_len;
1491                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1492         }
1493
1494 out:
1495         return dst0;
1496
1497 put_states:
1498         for (; i < nx; i++)
1499                 xfrm_state_put(xfrm[i]);
1500 free_dst:
1501         if (dst0)
1502                 dst_free(dst0);
1503         dst0 = ERR_PTR(err);
1504         goto out;
1505 }
1506
1507 static int inline
1508 xfrm_dst_alloc_copy(void **target, void *src, int size)
1509 {
1510         if (!*target) {
1511                 *target = kmalloc(size, GFP_ATOMIC);
1512                 if (!*target)
1513                         return -ENOMEM;
1514         }
1515         memcpy(*target, src, size);
1516         return 0;
1517 }
1518
1519 static int inline
1520 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1521 {
1522 #ifdef CONFIG_XFRM_SUB_POLICY
1523         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1524         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1525                                    sel, sizeof(*sel));
1526 #else
1527         return 0;
1528 #endif
1529 }
1530
1531 static int inline
1532 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1533 {
1534 #ifdef CONFIG_XFRM_SUB_POLICY
1535         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1536         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1537 #else
1538         return 0;
1539 #endif
1540 }
1541
1542 static int xfrm_expand_policies(struct flowi *fl, u16 family,
1543                                 struct xfrm_policy **pols,
1544                                 int *num_pols, int *num_xfrms)
1545 {
1546         int i;
1547
1548         if (*num_pols == 0 || !pols[0]) {
1549                 *num_pols = 0;
1550                 *num_xfrms = 0;
1551                 return 0;
1552         }
1553         if (IS_ERR(pols[0]))
1554                 return PTR_ERR(pols[0]);
1555
1556         *num_xfrms = pols[0]->xfrm_nr;
1557
1558 #ifdef CONFIG_XFRM_SUB_POLICY
1559         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1560             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1561                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1562                                                     XFRM_POLICY_TYPE_MAIN,
1563                                                     fl, family,
1564                                                     XFRM_POLICY_OUT);
1565                 if (pols[1]) {
1566                         if (IS_ERR(pols[1])) {
1567                                 xfrm_pols_put(pols, *num_pols);
1568                                 return PTR_ERR(pols[1]);
1569                         }
1570                         (*num_pols) ++;
1571                         (*num_xfrms) += pols[1]->xfrm_nr;
1572                 }
1573         }
1574 #endif
1575         for (i = 0; i < *num_pols; i++) {
1576                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1577                         *num_xfrms = -1;
1578                         break;
1579                 }
1580         }
1581
1582         return 0;
1583
1584 }
1585
1586 static struct xfrm_dst *
1587 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1588                                struct flowi *fl, u16 family,
1589                                struct dst_entry *dst_orig)
1590 {
1591         struct net *net = xp_net(pols[0]);
1592         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1593         struct dst_entry *dst;
1594         struct xfrm_dst *xdst;
1595         int err;
1596
1597         /* Try to instantiate a bundle */
1598         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1599         if (err <= 0) {
1600                 if (err != 0 && err != -EAGAIN)
1601                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1602                 return ERR_PTR(err);
1603         }
1604
1605         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1606         if (IS_ERR(dst)) {
1607                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1608                 return ERR_CAST(dst);
1609         }
1610
1611         xdst = (struct xfrm_dst *)dst;
1612         xdst->num_xfrms = err;
1613         if (num_pols > 1)
1614                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1615         else
1616                 err = xfrm_dst_update_origin(dst, fl);
1617         if (unlikely(err)) {
1618                 dst_free(dst);
1619                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1620                 return ERR_PTR(err);
1621         }
1622
1623         xdst->num_pols = num_pols;
1624         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1625         xdst->policy_genid = atomic_read(&pols[0]->genid);
1626
1627         return xdst;
1628 }
1629
1630 static struct flow_cache_object *
1631 xfrm_bundle_lookup(struct net *net, struct flowi *fl, u16 family, u8 dir,
1632                    struct flow_cache_object *oldflo, void *ctx)
1633 {
1634         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1635         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1636         struct xfrm_dst *xdst, *new_xdst;
1637         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1638
1639         /* Check if the policies from old bundle are usable */
1640         xdst = NULL;
1641         if (oldflo) {
1642                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1643                 num_pols = xdst->num_pols;
1644                 num_xfrms = xdst->num_xfrms;
1645                 pol_dead = 0;
1646                 for (i = 0; i < num_pols; i++) {
1647                         pols[i] = xdst->pols[i];
1648                         pol_dead |= pols[i]->walk.dead;
1649                 }
1650                 if (pol_dead) {
1651                         dst_free(&xdst->u.dst);
1652                         xdst = NULL;
1653                         num_pols = 0;
1654                         num_xfrms = 0;
1655                         oldflo = NULL;
1656                 }
1657         }
1658
1659         /* Resolve policies to use if we couldn't get them from
1660          * previous cache entry */
1661         if (xdst == NULL) {
1662                 num_pols = 1;
1663                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1664                 err = xfrm_expand_policies(fl, family, pols,
1665                                            &num_pols, &num_xfrms);
1666                 if (err < 0)
1667                         goto inc_error;
1668                 if (num_pols == 0)
1669                         return NULL;
1670                 if (num_xfrms <= 0)
1671                         goto make_dummy_bundle;
1672         }
1673
1674         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1675         if (IS_ERR(new_xdst)) {
1676                 err = PTR_ERR(new_xdst);
1677                 if (err != -EAGAIN)
1678                         goto error;
1679                 if (oldflo == NULL)
1680                         goto make_dummy_bundle;
1681                 dst_hold(&xdst->u.dst);
1682                 return oldflo;
1683         } else if (new_xdst == NULL) {
1684                 num_xfrms = 0;
1685                 if (oldflo == NULL)
1686                         goto make_dummy_bundle;
1687                 xdst->num_xfrms = 0;
1688                 dst_hold(&xdst->u.dst);
1689                 return oldflo;
1690         }
1691
1692         /* Kill the previous bundle */
1693         if (xdst) {
1694                 /* The policies were stolen for newly generated bundle */
1695                 xdst->num_pols = 0;
1696                 dst_free(&xdst->u.dst);
1697         }
1698
1699         /* Flow cache does not have reference, it dst_free()'s,
1700          * but we do need to return one reference for original caller */
1701         dst_hold(&new_xdst->u.dst);
1702         return &new_xdst->flo;
1703
1704 make_dummy_bundle:
1705         /* We found policies, but there's no bundles to instantiate:
1706          * either because the policy blocks, has no transformations or
1707          * we could not build template (no xfrm_states).*/
1708         xdst = xfrm_alloc_dst(net, family);
1709         if (IS_ERR(xdst)) {
1710                 xfrm_pols_put(pols, num_pols);
1711                 return ERR_CAST(xdst);
1712         }
1713         xdst->num_pols = num_pols;
1714         xdst->num_xfrms = num_xfrms;
1715         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1716
1717         dst_hold(&xdst->u.dst);
1718         return &xdst->flo;
1719
1720 inc_error:
1721         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1722 error:
1723         if (xdst != NULL)
1724                 dst_free(&xdst->u.dst);
1725         else
1726                 xfrm_pols_put(pols, num_pols);
1727         return ERR_PTR(err);
1728 }
1729
1730 /* Main function: finds/creates a bundle for given flow.
1731  *
1732  * At the moment we eat a raw IP route. Mostly to speed up lookups
1733  * on interfaces with disabled IPsec.
1734  */
1735 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1736                   struct sock *sk, int flags)
1737 {
1738         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1739         struct flow_cache_object *flo;
1740         struct xfrm_dst *xdst;
1741         struct dst_entry *dst, *dst_orig = *dst_p, *route;
1742         u16 family = dst_orig->ops->family;
1743         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1744         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1745
1746 restart:
1747         dst = NULL;
1748         xdst = NULL;
1749         route = NULL;
1750
1751         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1752                 num_pols = 1;
1753                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1754                 err = xfrm_expand_policies(fl, family, pols,
1755                                            &num_pols, &num_xfrms);
1756                 if (err < 0)
1757                         goto dropdst;
1758
1759                 if (num_pols) {
1760                         if (num_xfrms <= 0) {
1761                                 drop_pols = num_pols;
1762                                 goto no_transform;
1763                         }
1764
1765                         xdst = xfrm_resolve_and_create_bundle(
1766                                         pols, num_pols, fl,
1767                                         family, dst_orig);
1768                         if (IS_ERR(xdst)) {
1769                                 xfrm_pols_put(pols, num_pols);
1770                                 err = PTR_ERR(xdst);
1771                                 goto dropdst;
1772                         } else if (xdst == NULL) {
1773                                 num_xfrms = 0;
1774                                 drop_pols = num_pols;
1775                                 goto no_transform;
1776                         }
1777
1778                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1779                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1780                         xfrm_policy_sk_bundles = &xdst->u.dst;
1781                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1782
1783                         route = xdst->route;
1784                 }
1785         }
1786
1787         if (xdst == NULL) {
1788                 /* To accelerate a bit...  */
1789                 if ((dst_orig->flags & DST_NOXFRM) ||
1790                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1791                         goto nopol;
1792
1793                 flo = flow_cache_lookup(net, fl, family, dir,
1794                                         xfrm_bundle_lookup, dst_orig);
1795                 if (flo == NULL)
1796                         goto nopol;
1797                 if (IS_ERR(flo)) {
1798                         err = PTR_ERR(flo);
1799                         goto dropdst;
1800                 }
1801                 xdst = container_of(flo, struct xfrm_dst, flo);
1802
1803                 num_pols = xdst->num_pols;
1804                 num_xfrms = xdst->num_xfrms;
1805                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1806                 route = xdst->route;
1807         }
1808
1809         dst = &xdst->u.dst;
1810         if (route == NULL && num_xfrms > 0) {
1811                 /* The only case when xfrm_bundle_lookup() returns a
1812                  * bundle with null route, is when the template could
1813                  * not be resolved. It means policies are there, but
1814                  * bundle could not be created, since we don't yet
1815                  * have the xfrm_state's. We need to wait for KM to
1816                  * negotiate new SA's or bail out with error.*/
1817                 if (net->xfrm.sysctl_larval_drop) {
1818                         /* EREMOTE tells the caller to generate
1819                          * a one-shot blackhole route. */
1820                         dst_release(dst);
1821                         xfrm_pols_put(pols, drop_pols);
1822                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1823                         return -EREMOTE;
1824                 }
1825                 if (flags & XFRM_LOOKUP_WAIT) {
1826                         DECLARE_WAITQUEUE(wait, current);
1827
1828                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1829                         set_current_state(TASK_INTERRUPTIBLE);
1830                         schedule();
1831                         set_current_state(TASK_RUNNING);
1832                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1833
1834                         if (!signal_pending(current)) {
1835                                 dst_release(dst);
1836                                 goto restart;
1837                         }
1838
1839                         err = -ERESTART;
1840                 } else
1841                         err = -EAGAIN;
1842
1843                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1844                 goto error;
1845         }
1846
1847 no_transform:
1848         if (num_pols == 0)
1849                 goto nopol;
1850
1851         if ((flags & XFRM_LOOKUP_ICMP) &&
1852             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1853                 err = -ENOENT;
1854                 goto error;
1855         }
1856
1857         for (i = 0; i < num_pols; i++)
1858                 pols[i]->curlft.use_time = get_seconds();
1859
1860         if (num_xfrms < 0) {
1861                 /* Prohibit the flow */
1862                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1863                 err = -EPERM;
1864                 goto error;
1865         } else if (num_xfrms > 0) {
1866                 /* Flow transformed */
1867                 *dst_p = dst;
1868                 dst_release(dst_orig);
1869         } else {
1870                 /* Flow passes untransformed */
1871                 dst_release(dst);
1872         }
1873 ok:
1874         xfrm_pols_put(pols, drop_pols);
1875         return 0;
1876
1877 nopol:
1878         if (!(flags & XFRM_LOOKUP_ICMP))
1879                 goto ok;
1880         err = -ENOENT;
1881 error:
1882         dst_release(dst);
1883 dropdst:
1884         dst_release(dst_orig);
1885         *dst_p = NULL;
1886         xfrm_pols_put(pols, drop_pols);
1887         return err;
1888 }
1889 EXPORT_SYMBOL(__xfrm_lookup);
1890
1891 int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1892                 struct sock *sk, int flags)
1893 {
1894         int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1895
1896         if (err == -EREMOTE) {
1897                 dst_release(*dst_p);
1898                 *dst_p = NULL;
1899                 err = -EAGAIN;
1900         }
1901
1902         return err;
1903 }
1904 EXPORT_SYMBOL(xfrm_lookup);
1905
1906 static inline int
1907 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1908 {
1909         struct xfrm_state *x;
1910
1911         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1912                 return 0;
1913         x = skb->sp->xvec[idx];
1914         if (!x->type->reject)
1915                 return 0;
1916         return x->type->reject(x, skb, fl);
1917 }
1918
1919 /* When skb is transformed back to its "native" form, we have to
1920  * check policy restrictions. At the moment we make this in maximally
1921  * stupid way. Shame on me. :-) Of course, connected sockets must
1922  * have policy cached at them.
1923  */
1924
1925 static inline int
1926 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1927               unsigned short family)
1928 {
1929         if (xfrm_state_kern(x))
1930                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1931         return  x->id.proto == tmpl->id.proto &&
1932                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1933                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1934                 x->props.mode == tmpl->mode &&
1935                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1936                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1937                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1938                   xfrm_state_addr_cmp(tmpl, x, family));
1939 }
1940
1941 /*
1942  * 0 or more than 0 is returned when validation is succeeded (either bypass
1943  * because of optional transport mode, or next index of the mathced secpath
1944  * state with the template.
1945  * -1 is returned when no matching template is found.
1946  * Otherwise "-2 - errored_index" is returned.
1947  */
1948 static inline int
1949 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1950                unsigned short family)
1951 {
1952         int idx = start;
1953
1954         if (tmpl->optional) {
1955                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1956                         return start;
1957         } else
1958                 start = -1;
1959         for (; idx < sp->len; idx++) {
1960                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1961                         return ++idx;
1962                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1963                         if (start == -1)
1964                                 start = -2-idx;
1965                         break;
1966                 }
1967         }
1968         return start;
1969 }
1970
1971 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1972                           unsigned int family, int reverse)
1973 {
1974         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1975         int err;
1976
1977         if (unlikely(afinfo == NULL))
1978                 return -EAFNOSUPPORT;
1979
1980         afinfo->decode_session(skb, fl, reverse);
1981         err = security_xfrm_decode_session(skb, &fl->secid);
1982         xfrm_policy_put_afinfo(afinfo);
1983         return err;
1984 }
1985 EXPORT_SYMBOL(__xfrm_decode_session);
1986
1987 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1988 {
1989         for (; k < sp->len; k++) {
1990                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1991                         *idxp = k;
1992                         return 1;
1993                 }
1994         }
1995
1996         return 0;
1997 }
1998
1999 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2000                         unsigned short family)
2001 {
2002         struct net *net = dev_net(skb->dev);
2003         struct xfrm_policy *pol;
2004         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2005         int npols = 0;
2006         int xfrm_nr;
2007         int pi;
2008         int reverse;
2009         struct flowi fl;
2010         u8 fl_dir;
2011         int xerr_idx = -1;
2012
2013         reverse = dir & ~XFRM_POLICY_MASK;
2014         dir &= XFRM_POLICY_MASK;
2015         fl_dir = policy_to_flow_dir(dir);
2016
2017         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2018                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2019                 return 0;
2020         }
2021
2022         nf_nat_decode_session(skb, &fl, family);
2023
2024         /* First, check used SA against their selectors. */
2025         if (skb->sp) {
2026                 int i;
2027
2028                 for (i=skb->sp->len-1; i>=0; i--) {
2029                         struct xfrm_state *x = skb->sp->xvec[i];
2030                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2031                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2032                                 return 0;
2033                         }
2034                 }
2035         }
2036
2037         pol = NULL;
2038         if (sk && sk->sk_policy[dir]) {
2039                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2040                 if (IS_ERR(pol)) {
2041                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2042                         return 0;
2043                 }
2044         }
2045
2046         if (!pol) {
2047                 struct flow_cache_object *flo;
2048
2049                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2050                                         xfrm_policy_lookup, NULL);
2051                 if (IS_ERR_OR_NULL(flo))
2052                         pol = ERR_CAST(flo);
2053                 else
2054                         pol = container_of(flo, struct xfrm_policy, flo);
2055         }
2056
2057         if (IS_ERR(pol)) {
2058                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2059                 return 0;
2060         }
2061
2062         if (!pol) {
2063                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2064                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2065                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2066                         return 0;
2067                 }
2068                 return 1;
2069         }
2070
2071         pol->curlft.use_time = get_seconds();
2072
2073         pols[0] = pol;
2074         npols ++;
2075 #ifdef CONFIG_XFRM_SUB_POLICY
2076         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2077                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2078                                                     &fl, family,
2079                                                     XFRM_POLICY_IN);
2080                 if (pols[1]) {
2081                         if (IS_ERR(pols[1])) {
2082                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2083                                 return 0;
2084                         }
2085                         pols[1]->curlft.use_time = get_seconds();
2086                         npols ++;
2087                 }
2088         }
2089 #endif
2090
2091         if (pol->action == XFRM_POLICY_ALLOW) {
2092                 struct sec_path *sp;
2093                 static struct sec_path dummy;
2094                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2095                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2096                 struct xfrm_tmpl **tpp = tp;
2097                 int ti = 0;
2098                 int i, k;
2099
2100                 if ((sp = skb->sp) == NULL)
2101                         sp = &dummy;
2102
2103                 for (pi = 0; pi < npols; pi++) {
2104                         if (pols[pi] != pol &&
2105                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2106                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2107                                 goto reject;
2108                         }
2109                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2110                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2111                                 goto reject_error;
2112                         }
2113                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2114                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2115                 }
2116                 xfrm_nr = ti;
2117                 if (npols > 1) {
2118                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2119                         tpp = stp;
2120                 }
2121
2122                 /* For each tunnel xfrm, find the first matching tmpl.
2123                  * For each tmpl before that, find corresponding xfrm.
2124                  * Order is _important_. Later we will implement
2125                  * some barriers, but at the moment barriers
2126                  * are implied between each two transformations.
2127                  */
2128                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2129                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2130                         if (k < 0) {
2131                                 if (k < -1)
2132                                         /* "-2 - errored_index" returned */
2133                                         xerr_idx = -(2+k);
2134                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2135                                 goto reject;
2136                         }
2137                 }
2138
2139                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2140                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2141                         goto reject;
2142                 }
2143
2144                 xfrm_pols_put(pols, npols);
2145                 return 1;
2146         }
2147         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2148
2149 reject:
2150         xfrm_secpath_reject(xerr_idx, skb, &fl);
2151 reject_error:
2152         xfrm_pols_put(pols, npols);
2153         return 0;
2154 }
2155 EXPORT_SYMBOL(__xfrm_policy_check);
2156
2157 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2158 {
2159         struct net *net = dev_net(skb->dev);
2160         struct flowi fl;
2161         struct dst_entry *dst;
2162         int res;
2163
2164         if (xfrm_decode_session(skb, &fl, family) < 0) {
2165                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2166                 return 0;
2167         }
2168
2169         skb_dst_force(skb);
2170         dst = skb_dst(skb);
2171
2172         res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
2173         skb_dst_set(skb, dst);
2174         return res;
2175 }
2176 EXPORT_SYMBOL(__xfrm_route_forward);
2177
2178 /* Optimize later using cookies and generation ids. */
2179
2180 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2181 {
2182         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2183          * to "-1" to force all XFRM destinations to get validated by
2184          * dst_ops->check on every use.  We do this because when a
2185          * normal route referenced by an XFRM dst is obsoleted we do
2186          * not go looking around for all parent referencing XFRM dsts
2187          * so that we can invalidate them.  It is just too much work.
2188          * Instead we make the checks here on every use.  For example:
2189          *
2190          *      XFRM dst A --> IPv4 dst X
2191          *
2192          * X is the "xdst->route" of A (X is also the "dst->path" of A
2193          * in this example).  If X is marked obsolete, "A" will not
2194          * notice.  That's what we are validating here via the
2195          * stale_bundle() check.
2196          *
2197          * When a policy's bundle is pruned, we dst_free() the XFRM
2198          * dst which causes it's ->obsolete field to be set to a
2199          * positive non-zero integer.  If an XFRM dst has been pruned
2200          * like this, we want to force a new route lookup.
2201          */
2202         if (dst->obsolete < 0 && !stale_bundle(dst))
2203                 return dst;
2204
2205         return NULL;
2206 }
2207
2208 static int stale_bundle(struct dst_entry *dst)
2209 {
2210         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2211 }
2212
2213 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2214 {
2215         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2216                 dst->dev = dev_net(dev)->loopback_dev;
2217                 dev_hold(dst->dev);
2218                 dev_put(dev);
2219         }
2220 }
2221 EXPORT_SYMBOL(xfrm_dst_ifdown);
2222
2223 static void xfrm_link_failure(struct sk_buff *skb)
2224 {
2225         /* Impossible. Such dst must be popped before reaches point of failure. */
2226 }
2227
2228 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2229 {
2230         if (dst) {
2231                 if (dst->obsolete) {
2232                         dst_release(dst);
2233                         dst = NULL;
2234                 }
2235         }
2236         return dst;
2237 }
2238
2239 static void __xfrm_garbage_collect(struct net *net)
2240 {
2241         struct dst_entry *head, *next;
2242
2243         flow_cache_flush();
2244
2245         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2246         head = xfrm_policy_sk_bundles;
2247         xfrm_policy_sk_bundles = NULL;
2248         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2249
2250         while (head) {
2251                 next = head->next;
2252                 dst_free(head);
2253                 head = next;
2254         }
2255 }
2256
2257 static void xfrm_init_pmtu(struct dst_entry *dst)
2258 {
2259         do {
2260                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2261                 u32 pmtu, route_mtu_cached;
2262
2263                 pmtu = dst_mtu(dst->child);
2264                 xdst->child_mtu_cached = pmtu;
2265
2266                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2267
2268                 route_mtu_cached = dst_mtu(xdst->route);
2269                 xdst->route_mtu_cached = route_mtu_cached;
2270
2271                 if (pmtu > route_mtu_cached)
2272                         pmtu = route_mtu_cached;
2273
2274                 dst->metrics[RTAX_MTU-1] = pmtu;
2275         } while ((dst = dst->next));
2276 }
2277
2278 /* Check that the bundle accepts the flow and its components are
2279  * still valid.
2280  */
2281
2282 static int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2283                 struct flowi *fl, int family, int strict)
2284 {
2285         struct dst_entry *dst = &first->u.dst;
2286         struct xfrm_dst *last;
2287         u32 mtu;
2288
2289         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2290             (dst->dev && !netif_running(dst->dev)))
2291                 return 0;
2292 #ifdef CONFIG_XFRM_SUB_POLICY
2293         if (fl) {
2294                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2295                         return 0;
2296                 if (first->partner &&
2297                     !xfrm_selector_match(first->partner, fl, family))
2298                         return 0;
2299         }
2300 #endif
2301
2302         last = NULL;
2303
2304         do {
2305                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2306
2307                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2308                         return 0;
2309                 if (fl && pol &&
2310                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2311                         return 0;
2312                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2313                         return 0;
2314                 if (xdst->xfrm_genid != dst->xfrm->genid)
2315                         return 0;
2316                 if (xdst->num_pols > 0 &&
2317                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2318                         return 0;
2319
2320                 if (strict && fl &&
2321                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2322                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2323                         return 0;
2324
2325                 mtu = dst_mtu(dst->child);
2326                 if (xdst->child_mtu_cached != mtu) {
2327                         last = xdst;
2328                         xdst->child_mtu_cached = mtu;
2329                 }
2330
2331                 if (!dst_check(xdst->route, xdst->route_cookie))
2332                         return 0;
2333                 mtu = dst_mtu(xdst->route);
2334                 if (xdst->route_mtu_cached != mtu) {
2335                         last = xdst;
2336                         xdst->route_mtu_cached = mtu;
2337                 }
2338
2339                 dst = dst->child;
2340         } while (dst->xfrm);
2341
2342         if (likely(!last))
2343                 return 1;
2344
2345         mtu = last->child_mtu_cached;
2346         for (;;) {
2347                 dst = &last->u.dst;
2348
2349                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2350                 if (mtu > last->route_mtu_cached)
2351                         mtu = last->route_mtu_cached;
2352                 dst->metrics[RTAX_MTU-1] = mtu;
2353
2354                 if (last == first)
2355                         break;
2356
2357                 last = (struct xfrm_dst *)last->u.dst.next;
2358                 last->child_mtu_cached = mtu;
2359         }
2360
2361         return 1;
2362 }
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),
2496                           __alignof__(struct linux_xfrm_mib)) < 0)
2497                 return -ENOMEM;
2498         rv = xfrm_proc_init(net);
2499         if (rv < 0)
2500                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2501         return rv;
2502 }
2503
2504 static void xfrm_statistics_fini(struct net *net)
2505 {
2506         xfrm_proc_fini(net);
2507         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2508 }
2509 #else
2510 static int __net_init xfrm_statistics_init(struct net *net)
2511 {
2512         return 0;
2513 }
2514
2515 static void xfrm_statistics_fini(struct net *net)
2516 {
2517 }
2518 #endif
2519
2520 static int __net_init xfrm_policy_init(struct net *net)
2521 {
2522         unsigned int hmask, sz;
2523         int dir;
2524
2525         if (net_eq(net, &init_net))
2526                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2527                                            sizeof(struct xfrm_dst),
2528                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2529                                            NULL);
2530
2531         hmask = 8 - 1;
2532         sz = (hmask+1) * sizeof(struct hlist_head);
2533
2534         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2535         if (!net->xfrm.policy_byidx)
2536                 goto out_byidx;
2537         net->xfrm.policy_idx_hmask = hmask;
2538
2539         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2540                 struct xfrm_policy_hash *htab;
2541
2542                 net->xfrm.policy_count[dir] = 0;
2543                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2544
2545                 htab = &net->xfrm.policy_bydst[dir];
2546                 htab->table = xfrm_hash_alloc(sz);
2547                 if (!htab->table)
2548                         goto out_bydst;
2549                 htab->hmask = hmask;
2550         }
2551
2552         INIT_LIST_HEAD(&net->xfrm.policy_all);
2553         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2554         if (net_eq(net, &init_net))
2555                 register_netdevice_notifier(&xfrm_dev_notifier);
2556         return 0;
2557
2558 out_bydst:
2559         for (dir--; dir >= 0; dir--) {
2560                 struct xfrm_policy_hash *htab;
2561
2562                 htab = &net->xfrm.policy_bydst[dir];
2563                 xfrm_hash_free(htab->table, sz);
2564         }
2565         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2566 out_byidx:
2567         return -ENOMEM;
2568 }
2569
2570 static void xfrm_policy_fini(struct net *net)
2571 {
2572         struct xfrm_audit audit_info;
2573         unsigned int sz;
2574         int dir;
2575
2576         flush_work(&net->xfrm.policy_hash_work);
2577 #ifdef CONFIG_XFRM_SUB_POLICY
2578         audit_info.loginuid = -1;
2579         audit_info.sessionid = -1;
2580         audit_info.secid = 0;
2581         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2582 #endif
2583         audit_info.loginuid = -1;
2584         audit_info.sessionid = -1;
2585         audit_info.secid = 0;
2586         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2587
2588         WARN_ON(!list_empty(&net->xfrm.policy_all));
2589
2590         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2591                 struct xfrm_policy_hash *htab;
2592
2593                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2594
2595                 htab = &net->xfrm.policy_bydst[dir];
2596                 sz = (htab->hmask + 1);
2597                 WARN_ON(!hlist_empty(htab->table));
2598                 xfrm_hash_free(htab->table, sz);
2599         }
2600
2601         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2602         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2603         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2604 }
2605
2606 static int __net_init xfrm_net_init(struct net *net)
2607 {
2608         int rv;
2609
2610         rv = xfrm_statistics_init(net);
2611         if (rv < 0)
2612                 goto out_statistics;
2613         rv = xfrm_state_init(net);
2614         if (rv < 0)
2615                 goto out_state;
2616         rv = xfrm_policy_init(net);
2617         if (rv < 0)
2618                 goto out_policy;
2619         xfrm_dst_ops_init(net);
2620         rv = xfrm_sysctl_init(net);
2621         if (rv < 0)
2622                 goto out_sysctl;
2623         return 0;
2624
2625 out_sysctl:
2626         xfrm_policy_fini(net);
2627 out_policy:
2628         xfrm_state_fini(net);
2629 out_state:
2630         xfrm_statistics_fini(net);
2631 out_statistics:
2632         return rv;
2633 }
2634
2635 static void __net_exit xfrm_net_exit(struct net *net)
2636 {
2637         xfrm_sysctl_fini(net);
2638         xfrm_policy_fini(net);
2639         xfrm_state_fini(net);
2640         xfrm_statistics_fini(net);
2641 }
2642
2643 static struct pernet_operations __net_initdata xfrm_net_ops = {
2644         .init = xfrm_net_init,
2645         .exit = xfrm_net_exit,
2646 };
2647
2648 void __init xfrm_init(void)
2649 {
2650         register_pernet_subsys(&xfrm_net_ops);
2651         xfrm_input_init();
2652 }
2653
2654 #ifdef CONFIG_AUDITSYSCALL
2655 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2656                                          struct audit_buffer *audit_buf)
2657 {
2658         struct xfrm_sec_ctx *ctx = xp->security;
2659         struct xfrm_selector *sel = &xp->selector;
2660
2661         if (ctx)
2662                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2663                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2664
2665         switch(sel->family) {
2666         case AF_INET:
2667                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2668                 if (sel->prefixlen_s != 32)
2669                         audit_log_format(audit_buf, " src_prefixlen=%d",
2670                                          sel->prefixlen_s);
2671                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2672                 if (sel->prefixlen_d != 32)
2673                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2674                                          sel->prefixlen_d);
2675                 break;
2676         case AF_INET6:
2677                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2678                 if (sel->prefixlen_s != 128)
2679                         audit_log_format(audit_buf, " src_prefixlen=%d",
2680                                          sel->prefixlen_s);
2681                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2682                 if (sel->prefixlen_d != 128)
2683                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2684                                          sel->prefixlen_d);
2685                 break;
2686         }
2687 }
2688
2689 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2690                            uid_t auid, u32 sessionid, u32 secid)
2691 {
2692         struct audit_buffer *audit_buf;
2693
2694         audit_buf = xfrm_audit_start("SPD-add");
2695         if (audit_buf == NULL)
2696                 return;
2697         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2698         audit_log_format(audit_buf, " res=%u", result);
2699         xfrm_audit_common_policyinfo(xp, audit_buf);
2700         audit_log_end(audit_buf);
2701 }
2702 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2703
2704 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2705                               uid_t auid, u32 sessionid, u32 secid)
2706 {
2707         struct audit_buffer *audit_buf;
2708
2709         audit_buf = xfrm_audit_start("SPD-delete");
2710         if (audit_buf == NULL)
2711                 return;
2712         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2713         audit_log_format(audit_buf, " res=%u", result);
2714         xfrm_audit_common_policyinfo(xp, audit_buf);
2715         audit_log_end(audit_buf);
2716 }
2717 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2718 #endif
2719
2720 #ifdef CONFIG_XFRM_MIGRATE
2721 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2722                                        struct xfrm_selector *sel_tgt)
2723 {
2724         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2725                 if (sel_tgt->family == sel_cmp->family &&
2726                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2727                                   sel_cmp->family) == 0 &&
2728                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2729                                   sel_cmp->family) == 0 &&
2730                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2731                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2732                         return 1;
2733                 }
2734         } else {
2735                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2736                         return 1;
2737                 }
2738         }
2739         return 0;
2740 }
2741
2742 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2743                                                      u8 dir, u8 type)
2744 {
2745         struct xfrm_policy *pol, *ret = NULL;
2746         struct hlist_node *entry;
2747         struct hlist_head *chain;
2748         u32 priority = ~0U;
2749
2750         read_lock_bh(&xfrm_policy_lock);
2751         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2752         hlist_for_each_entry(pol, entry, chain, bydst) {
2753                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2754                     pol->type == type) {
2755                         ret = pol;
2756                         priority = ret->priority;
2757                         break;
2758                 }
2759         }
2760         chain = &init_net.xfrm.policy_inexact[dir];
2761         hlist_for_each_entry(pol, entry, chain, bydst) {
2762                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2763                     pol->type == type &&
2764                     pol->priority < priority) {
2765                         ret = pol;
2766                         break;
2767                 }
2768         }
2769
2770         if (ret)
2771                 xfrm_pol_hold(ret);
2772
2773         read_unlock_bh(&xfrm_policy_lock);
2774
2775         return ret;
2776 }
2777
2778 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2779 {
2780         int match = 0;
2781
2782         if (t->mode == m->mode && t->id.proto == m->proto &&
2783             (m->reqid == 0 || t->reqid == m->reqid)) {
2784                 switch (t->mode) {
2785                 case XFRM_MODE_TUNNEL:
2786                 case XFRM_MODE_BEET:
2787                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2788                                           m->old_family) == 0 &&
2789                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2790                                           m->old_family) == 0) {
2791                                 match = 1;
2792                         }
2793                         break;
2794                 case XFRM_MODE_TRANSPORT:
2795                         /* in case of transport mode, template does not store
2796                            any IP addresses, hence we just compare mode and
2797                            protocol */
2798                         match = 1;
2799                         break;
2800                 default:
2801                         break;
2802                 }
2803         }
2804         return match;
2805 }
2806
2807 /* update endpoint address(es) of template(s) */
2808 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2809                                struct xfrm_migrate *m, int num_migrate)
2810 {
2811         struct xfrm_migrate *mp;
2812         int i, j, n = 0;
2813
2814         write_lock_bh(&pol->lock);
2815         if (unlikely(pol->walk.dead)) {
2816                 /* target policy has been deleted */
2817                 write_unlock_bh(&pol->lock);
2818                 return -ENOENT;
2819         }
2820
2821         for (i = 0; i < pol->xfrm_nr; i++) {
2822                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2823                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2824                                 continue;
2825                         n++;
2826                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2827                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2828                                 continue;
2829                         /* update endpoints */
2830                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2831                                sizeof(pol->xfrm_vec[i].id.daddr));
2832                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2833                                sizeof(pol->xfrm_vec[i].saddr));
2834                         pol->xfrm_vec[i].encap_family = mp->new_family;
2835                         /* flush bundles */
2836                         atomic_inc(&pol->genid);
2837                 }
2838         }
2839
2840         write_unlock_bh(&pol->lock);
2841
2842         if (!n)
2843                 return -ENODATA;
2844
2845         return 0;
2846 }
2847
2848 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2849 {
2850         int i, j;
2851
2852         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2853                 return -EINVAL;
2854
2855         for (i = 0; i < num_migrate; i++) {
2856                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2857                                    m[i].old_family) == 0) &&
2858                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2859                                    m[i].old_family) == 0))
2860                         return -EINVAL;
2861                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2862                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2863                         return -EINVAL;
2864
2865                 /* check if there is any duplicated entry */
2866                 for (j = i + 1; j < num_migrate; j++) {
2867                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2868                                     sizeof(m[i].old_daddr)) &&
2869                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2870                                     sizeof(m[i].old_saddr)) &&
2871                             m[i].proto == m[j].proto &&
2872                             m[i].mode == m[j].mode &&
2873                             m[i].reqid == m[j].reqid &&
2874                             m[i].old_family == m[j].old_family)
2875                                 return -EINVAL;
2876                 }
2877         }
2878
2879         return 0;
2880 }
2881
2882 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2883                  struct xfrm_migrate *m, int num_migrate,
2884                  struct xfrm_kmaddress *k)
2885 {
2886         int i, err, nx_cur = 0, nx_new = 0;
2887         struct xfrm_policy *pol = NULL;
2888         struct xfrm_state *x, *xc;
2889         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2890         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2891         struct xfrm_migrate *mp;
2892
2893         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2894                 goto out;
2895
2896         /* Stage 1 - find policy */
2897         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2898                 err = -ENOENT;
2899                 goto out;
2900         }
2901
2902         /* Stage 2 - find and update state(s) */
2903         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2904                 if ((x = xfrm_migrate_state_find(mp))) {
2905                         x_cur[nx_cur] = x;
2906                         nx_cur++;
2907                         if ((xc = xfrm_state_migrate(x, mp))) {
2908                                 x_new[nx_new] = xc;
2909                                 nx_new++;
2910                         } else {
2911                                 err = -ENODATA;
2912                                 goto restore_state;
2913                         }
2914                 }
2915         }
2916
2917         /* Stage 3 - update policy */
2918         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2919                 goto restore_state;
2920
2921         /* Stage 4 - delete old state(s) */
2922         if (nx_cur) {
2923                 xfrm_states_put(x_cur, nx_cur);
2924                 xfrm_states_delete(x_cur, nx_cur);
2925         }
2926
2927         /* Stage 5 - announce */
2928         km_migrate(sel, dir, type, m, num_migrate, k);
2929
2930         xfrm_pol_put(pol);
2931
2932         return 0;
2933 out:
2934         return err;
2935
2936 restore_state:
2937         if (pol)
2938                 xfrm_pol_put(pol);
2939         if (nx_cur)
2940                 xfrm_states_put(x_cur, nx_cur);
2941         if (nx_new)
2942                 xfrm_states_delete(x_new, nx_new);
2943
2944         return err;
2945 }
2946 EXPORT_SYMBOL(xfrm_migrate);
2947 #endif