net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/xfrm.h>
  30 #include <net/ip.h>
  31 #ifdef CONFIG_XFRM_STATISTICS
  32 #include <net/snmp.h>
  33 #endif
  34
  35 #include "xfrm_hash.h"
  36
  37 int sysctl_xfrm_larval_drop __read_mostly = 1;
  38
  39 #ifdef CONFIG_XFRM_STATISTICS
  40 DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
  41 EXPORT_SYMBOL(xfrm_statistics);
  42 #endif
  43
  44 DEFINE_MUTEX(xfrm_cfg_mutex);
  45 EXPORT_SYMBOL(xfrm_cfg_mutex);
  46
  47 static DEFINE_RWLOCK(xfrm_policy_lock);
  48
  49 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  50 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  51
  52 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  53
  54 static HLIST_HEAD(xfrm_policy_gc_list);
  55 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  56
  57 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  58 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  59 static void xfrm_init_pmtu(struct dst_entry *dst);
  60
  61 static inline int
  62 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  63 {
  64         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  65                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  66                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  67                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  68                 (fl->proto == sel->proto || !sel->proto) &&
  69                 (fl->oif == sel->ifindex || !sel->ifindex);
  70 }
  71
  72 static inline int
  73 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  74 {
  75         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  76                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  77                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  78                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  79                 (fl->proto == sel->proto || !sel->proto) &&
  80                 (fl->oif == sel->ifindex || !sel->ifindex);
  81 }
  82
  83 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  84                     unsigned short family)
  85 {
  86         switch (family) {
  87         case AF_INET:
  88                 return __xfrm4_selector_match(sel, fl);
  89         case AF_INET6:
  90                 return __xfrm6_selector_match(sel, fl);
  91         }
  92         return 0;
  93 }
  94
  95 static inline struct dst_entry *__xfrm_dst_lookup(int tos,
  96                                                   xfrm_address_t *saddr,
  97                                                   xfrm_address_t *daddr,
  98                                                   int family)
  99 {
 100         struct xfrm_policy_afinfo *afinfo;
 101         struct dst_entry *dst;
 102
 103         afinfo = xfrm_policy_get_afinfo(family);
 104         if (unlikely(afinfo == NULL))
 105                 return ERR_PTR(-EAFNOSUPPORT);
 106
 107         dst = afinfo->dst_lookup(tos, saddr, daddr);
 108
 109         xfrm_policy_put_afinfo(afinfo);
 110
 111         return dst;
 112 }
 113
 114 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 115                                                 xfrm_address_t *prev_saddr,
 116                                                 xfrm_address_t *prev_daddr,
 117                                                 int family)
 118 {
 119         xfrm_address_t *saddr = &x->props.saddr;
 120         xfrm_address_t *daddr = &x->id.daddr;
 121         struct dst_entry *dst;
 122
 123         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 124                 saddr = x->coaddr;
 125                 daddr = prev_daddr;
 126         }
 127         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 128                 saddr = prev_saddr;
 129                 daddr = x->coaddr;
 130         }
 131
 132         dst = __xfrm_dst_lookup(tos, saddr, daddr, family);
 133
 134         if (!IS_ERR(dst)) {
 135                 if (prev_saddr != saddr)
 136                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 137                 if (prev_daddr != daddr)
 138                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 139         }
 140
 141         return dst;
 142 }
 143
 144 static inline unsigned long make_jiffies(long secs)
 145 {
 146         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 147                 return MAX_SCHEDULE_TIMEOUT-1;
 148         else
 149                 return secs*HZ;
 150 }
 151
 152 static void xfrm_policy_timer(unsigned long data)
 153 {
 154         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 155         unsigned long now = get_seconds();
 156         long next = LONG_MAX;
 157         int warn = 0;
 158         int dir;
 159
 160         read_lock(&xp->lock);
 161
 162         if (xp->walk.dead)
 163                 goto out;
 164
 165         dir = xfrm_policy_id2dir(xp->index);
 166
 167         if (xp->lft.hard_add_expires_seconds) {
 168                 long tmo = xp->lft.hard_add_expires_seconds +
 169                         xp->curlft.add_time - now;
 170                 if (tmo <= 0)
 171                         goto expired;
 172                 if (tmo < next)
 173                         next = tmo;
 174         }
 175         if (xp->lft.hard_use_expires_seconds) {
 176                 long tmo = xp->lft.hard_use_expires_seconds +
 177                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 178                 if (tmo <= 0)
 179                         goto expired;
 180                 if (tmo < next)
 181                         next = tmo;
 182         }
 183         if (xp->lft.soft_add_expires_seconds) {
 184                 long tmo = xp->lft.soft_add_expires_seconds +
 185                         xp->curlft.add_time - now;
 186                 if (tmo <= 0) {
 187                         warn = 1;
 188                         tmo = XFRM_KM_TIMEOUT;
 189                 }
 190                 if (tmo < next)
 191                         next = tmo;
 192         }
 193         if (xp->lft.soft_use_expires_seconds) {
 194                 long tmo = xp->lft.soft_use_expires_seconds +
 195                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 196                 if (tmo <= 0) {
 197                         warn = 1;
 198                         tmo = XFRM_KM_TIMEOUT;
 199                 }
 200                 if (tmo < next)
 201                         next = tmo;
 202         }
 203
 204         if (warn)
 205                 km_policy_expired(xp, dir, 0, 0);
 206         if (next != LONG_MAX &&
 207             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 208                 xfrm_pol_hold(xp);
 209
 210 out:
 211         read_unlock(&xp->lock);
 212         xfrm_pol_put(xp);
 213         return;
 214
 215 expired:
 216         read_unlock(&xp->lock);
 217         if (!xfrm_policy_delete(xp, dir))
 218                 km_policy_expired(xp, dir, 1, 0);
 219         xfrm_pol_put(xp);
 220 }
 221
 222
 223 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 224  * SPD calls.
 225  */
 226
 227 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 228 {
 229         struct xfrm_policy *policy;
 230
 231         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 232
 233         if (policy) {
 234                 write_pnet(&policy->xp_net, net);
 235                 INIT_LIST_HEAD(&policy->walk.all);
 236                 INIT_HLIST_NODE(&policy->bydst);
 237                 INIT_HLIST_NODE(&policy->byidx);
 238                 rwlock_init(&policy->lock);
 239                 atomic_set(&policy->refcnt, 1);
 240                 setup_timer(&policy->timer, xfrm_policy_timer,
 241                                 (unsigned long)policy);
 242         }
 243         return policy;
 244 }
 245 EXPORT_SYMBOL(xfrm_policy_alloc);
 246
 247 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 248
 249 void xfrm_policy_destroy(struct xfrm_policy *policy)
 250 {
 251         BUG_ON(!policy->walk.dead);
 252
 253         BUG_ON(policy->bundles);
 254
 255         if (del_timer(&policy->timer))
 256                 BUG();
 257
 258         security_xfrm_policy_free(policy->security);
 259         kfree(policy);
 260 }
 261 EXPORT_SYMBOL(xfrm_policy_destroy);
 262
 263 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 264 {
 265         struct dst_entry *dst;
 266
 267         while ((dst = policy->bundles) != NULL) {
 268                 policy->bundles = dst->next;
 269                 dst_free(dst);
 270         }
 271
 272         if (del_timer(&policy->timer))
 273                 atomic_dec(&policy->refcnt);
 274
 275         if (atomic_read(&policy->refcnt) > 1)
 276                 flow_cache_flush();
 277
 278         xfrm_pol_put(policy);
 279 }
 280
 281 static void xfrm_policy_gc_task(struct work_struct *work)
 282 {
 283         struct xfrm_policy *policy;
 284         struct hlist_node *entry, *tmp;
 285         struct hlist_head gc_list;
 286
 287         spin_lock_bh(&xfrm_policy_gc_lock);
 288         gc_list.first = xfrm_policy_gc_list.first;
 289         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 290         spin_unlock_bh(&xfrm_policy_gc_lock);
 291
 292         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 293                 xfrm_policy_gc_kill(policy);
 294 }
 295 static DECLARE_WORK(xfrm_policy_gc_work, xfrm_policy_gc_task);
 296
 297 /* Rule must be locked. Release descentant resources, announce
 298  * entry dead. The rule must be unlinked from lists to the moment.
 299  */
 300
 301 static void xfrm_policy_kill(struct xfrm_policy *policy)
 302 {
 303         int dead;
 304
 305         write_lock_bh(&policy->lock);
 306         dead = policy->walk.dead;
 307         policy->walk.dead = 1;
 308         write_unlock_bh(&policy->lock);
 309
 310         if (unlikely(dead)) {
 311                 WARN_ON(1);
 312                 return;
 313         }
 314
 315         spin_lock_bh(&xfrm_policy_gc_lock);
 316         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 317         spin_unlock_bh(&xfrm_policy_gc_lock);
 318
 319         schedule_work(&xfrm_policy_gc_work);
 320 }
 321
 322 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 323
 324 static inline unsigned int idx_hash(struct net *net, u32 index)
 325 {
 326         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 327 }
 328
 329 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 330 {
 331         unsigned int hmask = init_net.xfrm.policy_bydst[dir].hmask;
 332         unsigned int hash = __sel_hash(sel, family, hmask);
 333
 334         return (hash == hmask + 1 ?
 335                 &init_net.xfrm.policy_inexact[dir] :
 336                 init_net.xfrm.policy_bydst[dir].table + hash);
 337 }
 338
 339 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 340 {
 341         unsigned int hmask = init_net.xfrm.policy_bydst[dir].hmask;
 342         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 343
 344         return init_net.xfrm.policy_bydst[dir].table + hash;
 345 }
 346
 347 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 348                                    struct hlist_head *ndsttable,
 349                                    unsigned int nhashmask)
 350 {
 351         struct hlist_node *entry, *tmp, *entry0 = NULL;
 352         struct xfrm_policy *pol;
 353         unsigned int h0 = 0;
 354
 355 redo:
 356         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 357                 unsigned int h;
 358
 359                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 360                                 pol->family, nhashmask);
 361                 if (!entry0) {
 362                         hlist_del(entry);
 363                         hlist_add_head(&pol->bydst, ndsttable+h);
 364                         h0 = h;
 365                 } else {
 366                         if (h != h0)
 367                                 continue;
 368                         hlist_del(entry);
 369                         hlist_add_after(entry0, &pol->bydst);
 370                 }
 371                 entry0 = entry;
 372         }
 373         if (!hlist_empty(list)) {
 374                 entry0 = NULL;
 375                 goto redo;
 376         }
 377 }
 378
 379 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 380                                    struct hlist_head *nidxtable,
 381                                    unsigned int nhashmask)
 382 {
 383         struct hlist_node *entry, *tmp;
 384         struct xfrm_policy *pol;
 385
 386         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 387                 unsigned int h;
 388
 389                 h = __idx_hash(pol->index, nhashmask);
 390                 hlist_add_head(&pol->byidx, nidxtable+h);
 391         }
 392 }
 393
 394 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 395 {
 396         return ((old_hmask + 1) << 1) - 1;
 397 }
 398
 399 static void xfrm_bydst_resize(struct net *net, int dir)
 400 {
 401         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 402         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 403         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 404         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 405         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 406         int i;
 407
 408         if (!ndst)
 409                 return;
 410
 411         write_lock_bh(&xfrm_policy_lock);
 412
 413         for (i = hmask; i >= 0; i--)
 414                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 415
 416         net->xfrm.policy_bydst[dir].table = ndst;
 417         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 418
 419         write_unlock_bh(&xfrm_policy_lock);
 420
 421         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 422 }
 423
 424 static void xfrm_byidx_resize(struct net *net, int total)
 425 {
 426         unsigned int hmask = net->xfrm.policy_idx_hmask;
 427         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 428         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 429         struct hlist_head *oidx = net->xfrm.policy_byidx;
 430         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 431         int i;
 432
 433         if (!nidx)
 434                 return;
 435
 436         write_lock_bh(&xfrm_policy_lock);
 437
 438         for (i = hmask; i >= 0; i--)
 439                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 440
 441         net->xfrm.policy_byidx = nidx;
 442         net->xfrm.policy_idx_hmask = nhashmask;
 443
 444         write_unlock_bh(&xfrm_policy_lock);
 445
 446         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 447 }
 448
 449 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 450 {
 451         unsigned int cnt = net->xfrm.policy_count[dir];
 452         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 453
 454         if (total)
 455                 *total += cnt;
 456
 457         if ((hmask + 1) < xfrm_policy_hashmax &&
 458             cnt > hmask)
 459                 return 1;
 460
 461         return 0;
 462 }
 463
 464 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 465 {
 466         unsigned int hmask = net->xfrm.policy_idx_hmask;
 467
 468         if ((hmask + 1) < xfrm_policy_hashmax &&
 469             total > hmask)
 470                 return 1;
 471
 472         return 0;
 473 }
 474
 475 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 476 {
 477         read_lock_bh(&xfrm_policy_lock);
 478         si->incnt = init_net.xfrm.policy_count[XFRM_POLICY_IN];
 479         si->outcnt = init_net.xfrm.policy_count[XFRM_POLICY_OUT];
 480         si->fwdcnt = init_net.xfrm.policy_count[XFRM_POLICY_FWD];
 481         si->inscnt = init_net.xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 482         si->outscnt = init_net.xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 483         si->fwdscnt = init_net.xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 484         si->spdhcnt = init_net.xfrm.policy_idx_hmask;
 485         si->spdhmcnt = xfrm_policy_hashmax;
 486         read_unlock_bh(&xfrm_policy_lock);
 487 }
 488 EXPORT_SYMBOL(xfrm_spd_getinfo);
 489
 490 static DEFINE_MUTEX(hash_resize_mutex);
 491 static void xfrm_hash_resize(struct work_struct *work)
 492 {
 493         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 494         int dir, total;
 495
 496         mutex_lock(&hash_resize_mutex);
 497
 498         total = 0;
 499         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 500                 if (xfrm_bydst_should_resize(net, dir, &total))
 501                         xfrm_bydst_resize(net, dir);
 502         }
 503         if (xfrm_byidx_should_resize(net, total))
 504                 xfrm_byidx_resize(net, total);
 505
 506         mutex_unlock(&hash_resize_mutex);
 507 }
 508
 509 /* Generate new index... KAME seems to generate them ordered by cost
 510  * of an absolute inpredictability of ordering of rules. This will not pass. */
 511 static u32 xfrm_gen_index(int dir)
 512 {
 513         static u32 idx_generator;
 514
 515         for (;;) {
 516                 struct hlist_node *entry;
 517                 struct hlist_head *list;
 518                 struct xfrm_policy *p;
 519                 u32 idx;
 520                 int found;
 521
 522                 idx = (idx_generator | dir);
 523                 idx_generator += 8;
 524                 if (idx == 0)
 525                         idx = 8;
 526                 list = init_net.xfrm.policy_byidx + idx_hash(&init_net, idx);
 527                 found = 0;
 528                 hlist_for_each_entry(p, entry, list, byidx) {
 529                         if (p->index == idx) {
 530                                 found = 1;
 531                                 break;
 532                         }
 533                 }
 534                 if (!found)
 535                         return idx;
 536         }
 537 }
 538
 539 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 540 {
 541         u32 *p1 = (u32 *) s1;
 542         u32 *p2 = (u32 *) s2;
 543         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 544         int i;
 545
 546         for (i = 0; i < len; i++) {
 547                 if (p1[i] != p2[i])
 548                         return 1;
 549         }
 550
 551         return 0;
 552 }
 553
 554 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 555 {
 556         struct xfrm_policy *pol;
 557         struct xfrm_policy *delpol;
 558         struct hlist_head *chain;
 559         struct hlist_node *entry, *newpos;
 560         struct dst_entry *gc_list;
 561
 562         write_lock_bh(&xfrm_policy_lock);
 563         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 564         delpol = NULL;
 565         newpos = NULL;
 566         hlist_for_each_entry(pol, entry, chain, bydst) {
 567                 if (pol->type == policy->type &&
 568                     !selector_cmp(&pol->selector, &policy->selector) &&
 569                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 570                     !WARN_ON(delpol)) {
 571                         if (excl) {
 572                                 write_unlock_bh(&xfrm_policy_lock);
 573                                 return -EEXIST;
 574                         }
 575                         delpol = pol;
 576                         if (policy->priority > pol->priority)
 577                                 continue;
 578                 } else if (policy->priority >= pol->priority) {
 579                         newpos = &pol->bydst;
 580                         continue;
 581                 }
 582                 if (delpol)
 583                         break;
 584         }
 585         if (newpos)
 586                 hlist_add_after(newpos, &policy->bydst);
 587         else
 588                 hlist_add_head(&policy->bydst, chain);
 589         xfrm_pol_hold(policy);
 590         init_net.xfrm.policy_count[dir]++;
 591         atomic_inc(&flow_cache_genid);
 592         if (delpol) {
 593                 hlist_del(&delpol->bydst);
 594                 hlist_del(&delpol->byidx);
 595                 list_del(&delpol->walk.all);
 596                 init_net.xfrm.policy_count[dir]--;
 597         }
 598         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
 599         hlist_add_head(&policy->byidx, init_net.xfrm.policy_byidx+idx_hash(&init_net, policy->index));
 600         policy->curlft.add_time = get_seconds();
 601         policy->curlft.use_time = 0;
 602         if (!mod_timer(&policy->timer, jiffies + HZ))
 603                 xfrm_pol_hold(policy);
 604         list_add(&policy->walk.all, &init_net.xfrm.policy_all);
 605         write_unlock_bh(&xfrm_policy_lock);
 606
 607         if (delpol)
 608                 xfrm_policy_kill(delpol);
 609         else if (xfrm_bydst_should_resize(&init_net, dir, NULL))
 610                 schedule_work(&init_net.xfrm.policy_hash_work);
 611
 612         read_lock_bh(&xfrm_policy_lock);
 613         gc_list = NULL;
 614         entry = &policy->bydst;
 615         hlist_for_each_entry_continue(policy, entry, bydst) {
 616                 struct dst_entry *dst;
 617
 618                 write_lock(&policy->lock);
 619                 dst = policy->bundles;
 620                 if (dst) {
 621                         struct dst_entry *tail = dst;
 622                         while (tail->next)
 623                                 tail = tail->next;
 624                         tail->next = gc_list;
 625                         gc_list = dst;
 626
 627                         policy->bundles = NULL;
 628                 }
 629                 write_unlock(&policy->lock);
 630         }
 631         read_unlock_bh(&xfrm_policy_lock);
 632
 633         while (gc_list) {
 634                 struct dst_entry *dst = gc_list;
 635
 636                 gc_list = dst->next;
 637                 dst_free(dst);
 638         }
 639
 640         return 0;
 641 }
 642 EXPORT_SYMBOL(xfrm_policy_insert);
 643
 644 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 645                                           struct xfrm_selector *sel,
 646                                           struct xfrm_sec_ctx *ctx, int delete,
 647                                           int *err)
 648 {
 649         struct xfrm_policy *pol, *ret;
 650         struct hlist_head *chain;
 651         struct hlist_node *entry;
 652
 653         *err = 0;
 654         write_lock_bh(&xfrm_policy_lock);
 655         chain = policy_hash_bysel(sel, sel->family, dir);
 656         ret = NULL;
 657         hlist_for_each_entry(pol, entry, chain, bydst) {
 658                 if (pol->type == type &&
 659                     !selector_cmp(sel, &pol->selector) &&
 660                     xfrm_sec_ctx_match(ctx, pol->security)) {
 661                         xfrm_pol_hold(pol);
 662                         if (delete) {
 663                                 *err = security_xfrm_policy_delete(
 664                                                                 pol->security);
 665                                 if (*err) {
 666                                         write_unlock_bh(&xfrm_policy_lock);
 667                                         return pol;
 668                                 }
 669                                 hlist_del(&pol->bydst);
 670                                 hlist_del(&pol->byidx);
 671                                 list_del(&pol->walk.all);
 672                                 init_net.xfrm.policy_count[dir]--;
 673                         }
 674                         ret = pol;
 675                         break;
 676                 }
 677         }
 678         write_unlock_bh(&xfrm_policy_lock);
 679
 680         if (ret && delete) {
 681                 atomic_inc(&flow_cache_genid);
 682                 xfrm_policy_kill(ret);
 683         }
 684         return ret;
 685 }
 686 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 687
 688 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 689                                      int *err)
 690 {
 691         struct xfrm_policy *pol, *ret;
 692         struct hlist_head *chain;
 693         struct hlist_node *entry;
 694
 695         *err = -ENOENT;
 696         if (xfrm_policy_id2dir(id) != dir)
 697                 return NULL;
 698
 699         *err = 0;
 700         write_lock_bh(&xfrm_policy_lock);
 701         chain = init_net.xfrm.policy_byidx + idx_hash(&init_net, id);
 702         ret = NULL;
 703         hlist_for_each_entry(pol, entry, chain, byidx) {
 704                 if (pol->type == type && pol->index == id) {
 705                         xfrm_pol_hold(pol);
 706                         if (delete) {
 707                                 *err = security_xfrm_policy_delete(
 708                                                                 pol->security);
 709                                 if (*err) {
 710                                         write_unlock_bh(&xfrm_policy_lock);
 711                                         return pol;
 712                                 }
 713                                 hlist_del(&pol->bydst);
 714                                 hlist_del(&pol->byidx);
 715                                 list_del(&pol->walk.all);
 716                                 init_net.xfrm.policy_count[dir]--;
 717                         }
 718                         ret = pol;
 719                         break;
 720                 }
 721         }
 722         write_unlock_bh(&xfrm_policy_lock);
 723
 724         if (ret && delete) {
 725                 atomic_inc(&flow_cache_genid);
 726                 xfrm_policy_kill(ret);
 727         }
 728         return ret;
 729 }
 730 EXPORT_SYMBOL(xfrm_policy_byid);
 731
 732 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 733 static inline int
 734 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 735 {
 736         int dir, err = 0;
 737
 738         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 739                 struct xfrm_policy *pol;
 740                 struct hlist_node *entry;
 741                 int i;
 742
 743                 hlist_for_each_entry(pol, entry,
 744                                      &init_net.xfrm.policy_inexact[dir], bydst) {
 745                         if (pol->type != type)
 746                                 continue;
 747                         err = security_xfrm_policy_delete(pol->security);
 748                         if (err) {
 749                                 xfrm_audit_policy_delete(pol, 0,
 750                                                          audit_info->loginuid,
 751                                                          audit_info->sessionid,
 752                                                          audit_info->secid);
 753                                 return err;
 754                         }
 755                 }
 756                 for (i = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 757                         hlist_for_each_entry(pol, entry,
 758                                              init_net.xfrm.policy_bydst[dir].table + i,
 759                                              bydst) {
 760                                 if (pol->type != type)
 761                                         continue;
 762                                 err = security_xfrm_policy_delete(
 763                                                                 pol->security);
 764                                 if (err) {
 765                                         xfrm_audit_policy_delete(pol, 0,
 766                                                         audit_info->loginuid,
 767                                                         audit_info->sessionid,
 768                                                         audit_info->secid);
 769                                         return err;
 770                                 }
 771                         }
 772                 }
 773         }
 774         return err;
 775 }
 776 #else
 777 static inline int
 778 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 779 {
 780         return 0;
 781 }
 782 #endif
 783
 784 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 785 {
 786         int dir, err = 0;
 787
 788         write_lock_bh(&xfrm_policy_lock);
 789
 790         err = xfrm_policy_flush_secctx_check(type, audit_info);
 791         if (err)
 792                 goto out;
 793
 794         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 795                 struct xfrm_policy *pol;
 796                 struct hlist_node *entry;
 797                 int i, killed;
 798
 799                 killed = 0;
 800         again1:
 801                 hlist_for_each_entry(pol, entry,
 802                                      &init_net.xfrm.policy_inexact[dir], bydst) {
 803                         if (pol->type != type)
 804                                 continue;
 805                         hlist_del(&pol->bydst);
 806                         hlist_del(&pol->byidx);
 807                         write_unlock_bh(&xfrm_policy_lock);
 808
 809                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 810                                                  audit_info->sessionid,
 811                                                  audit_info->secid);
 812
 813                         xfrm_policy_kill(pol);
 814                         killed++;
 815
 816                         write_lock_bh(&xfrm_policy_lock);
 817                         goto again1;
 818                 }
 819
 820                 for (i = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 821         again2:
 822                         hlist_for_each_entry(pol, entry,
 823                                              init_net.xfrm.policy_bydst[dir].table + i,
 824                                              bydst) {
 825                                 if (pol->type != type)
 826                                         continue;
 827                                 hlist_del(&pol->bydst);
 828                                 hlist_del(&pol->byidx);
 829                                 list_del(&pol->walk.all);
 830                                 write_unlock_bh(&xfrm_policy_lock);
 831
 832                                 xfrm_audit_policy_delete(pol, 1,
 833                                                          audit_info->loginuid,
 834                                                          audit_info->sessionid,
 835                                                          audit_info->secid);
 836                                 xfrm_policy_kill(pol);
 837                                 killed++;
 838
 839                                 write_lock_bh(&xfrm_policy_lock);
 840                                 goto again2;
 841                         }
 842                 }
 843
 844                 init_net.xfrm.policy_count[dir] -= killed;
 845         }
 846         atomic_inc(&flow_cache_genid);
 847 out:
 848         write_unlock_bh(&xfrm_policy_lock);
 849         return err;
 850 }
 851 EXPORT_SYMBOL(xfrm_policy_flush);
 852
 853 int xfrm_policy_walk(struct xfrm_policy_walk *walk,
 854                      int (*func)(struct xfrm_policy *, int, int, void*),
 855                      void *data)
 856 {
 857         struct xfrm_policy *pol;
 858         struct xfrm_policy_walk_entry *x;
 859         int error = 0;
 860
 861         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 862             walk->type != XFRM_POLICY_TYPE_ANY)
 863                 return -EINVAL;
 864
 865         if (list_empty(&walk->walk.all) && walk->seq != 0)
 866                 return 0;
 867
 868         write_lock_bh(&xfrm_policy_lock);
 869         if (list_empty(&walk->walk.all))
 870                 x = list_first_entry(&init_net.xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 871         else
 872                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 873         list_for_each_entry_from(x, &init_net.xfrm.policy_all, all) {
 874                 if (x->dead)
 875                         continue;
 876                 pol = container_of(x, struct xfrm_policy, walk);
 877                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 878                     walk->type != pol->type)
 879                         continue;
 880                 error = func(pol, xfrm_policy_id2dir(pol->index),
 881                              walk->seq, data);
 882                 if (error) {
 883                         list_move_tail(&walk->walk.all, &x->all);
 884                         goto out;
 885                 }
 886                 walk->seq++;
 887         }
 888         if (walk->seq == 0) {
 889                 error = -ENOENT;
 890                 goto out;
 891         }
 892         list_del_init(&walk->walk.all);
 893 out:
 894         write_unlock_bh(&xfrm_policy_lock);
 895         return error;
 896 }
 897 EXPORT_SYMBOL(xfrm_policy_walk);
 898
 899 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 900 {
 901         INIT_LIST_HEAD(&walk->walk.all);
 902         walk->walk.dead = 1;
 903         walk->type = type;
 904         walk->seq = 0;
 905 }
 906 EXPORT_SYMBOL(xfrm_policy_walk_init);
 907
 908 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 909 {
 910         if (list_empty(&walk->walk.all))
 911                 return;
 912
 913         write_lock_bh(&xfrm_policy_lock);
 914         list_del(&walk->walk.all);
 915         write_unlock_bh(&xfrm_policy_lock);
 916 }
 917 EXPORT_SYMBOL(xfrm_policy_walk_done);
 918
 919 /*
 920  * Find policy to apply to this flow.
 921  *
 922  * Returns 0 if policy found, else an -errno.
 923  */
 924 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 925                              u8 type, u16 family, int dir)
 926 {
 927         struct xfrm_selector *sel = &pol->selector;
 928         int match, ret = -ESRCH;
 929
 930         if (pol->family != family ||
 931             pol->type != type)
 932                 return ret;
 933
 934         match = xfrm_selector_match(sel, fl, family);
 935         if (match)
 936                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 937                                                   dir);
 938
 939         return ret;
 940 }
 941
 942 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 943                                                      u16 family, u8 dir)
 944 {
 945         int err;
 946         struct xfrm_policy *pol, *ret;
 947         xfrm_address_t *daddr, *saddr;
 948         struct hlist_node *entry;
 949         struct hlist_head *chain;
 950         u32 priority = ~0U;
 951
 952         daddr = xfrm_flowi_daddr(fl, family);
 953         saddr = xfrm_flowi_saddr(fl, family);
 954         if (unlikely(!daddr || !saddr))
 955                 return NULL;
 956
 957         read_lock_bh(&xfrm_policy_lock);
 958         chain = policy_hash_direct(daddr, saddr, family, dir);
 959         ret = NULL;
 960         hlist_for_each_entry(pol, entry, chain, bydst) {
 961                 err = xfrm_policy_match(pol, fl, type, family, dir);
 962                 if (err) {
 963                         if (err == -ESRCH)
 964                                 continue;
 965                         else {
 966                                 ret = ERR_PTR(err);
 967                                 goto fail;
 968                         }
 969                 } else {
 970                         ret = pol;
 971                         priority = ret->priority;
 972                         break;
 973                 }
 974         }
 975         chain = &init_net.xfrm.policy_inexact[dir];
 976         hlist_for_each_entry(pol, entry, chain, bydst) {
 977                 err = xfrm_policy_match(pol, fl, type, family, dir);
 978                 if (err) {
 979                         if (err == -ESRCH)
 980                                 continue;
 981                         else {
 982                                 ret = ERR_PTR(err);
 983                                 goto fail;
 984                         }
 985                 } else if (pol->priority < priority) {
 986                         ret = pol;
 987                         break;
 988                 }
 989         }
 990         if (ret)
 991                 xfrm_pol_hold(ret);
 992 fail:
 993         read_unlock_bh(&xfrm_policy_lock);
 994
 995         return ret;
 996 }
 997
 998 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 999                                void **objp, atomic_t **obj_refp)
1000 {
1001         struct xfrm_policy *pol;
1002         int err = 0;
1003
1004 #ifdef CONFIG_XFRM_SUB_POLICY
1005         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1006         if (IS_ERR(pol)) {
1007                 err = PTR_ERR(pol);
1008                 pol = NULL;
1009         }
1010         if (pol || err)
1011                 goto end;
1012 #endif
1013         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1014         if (IS_ERR(pol)) {
1015                 err = PTR_ERR(pol);
1016                 pol = NULL;
1017         }
1018 #ifdef CONFIG_XFRM_SUB_POLICY
1019 end:
1020 #endif
1021         if ((*objp = (void *) pol) != NULL)
1022                 *obj_refp = &pol->refcnt;
1023         return err;
1024 }
1025
1026 static inline int policy_to_flow_dir(int dir)
1027 {
1028         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1029             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1030             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1031                 return dir;
1032         switch (dir) {
1033         default:
1034         case XFRM_POLICY_IN:
1035                 return FLOW_DIR_IN;
1036         case XFRM_POLICY_OUT:
1037                 return FLOW_DIR_OUT;
1038         case XFRM_POLICY_FWD:
1039                 return FLOW_DIR_FWD;
1040         }
1041 }
1042
1043 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1044 {
1045         struct xfrm_policy *pol;
1046
1047         read_lock_bh(&xfrm_policy_lock);
1048         if ((pol = sk->sk_policy[dir]) != NULL) {
1049                 int match = xfrm_selector_match(&pol->selector, fl,
1050                                                 sk->sk_family);
1051                 int err = 0;
1052
1053                 if (match) {
1054                         err = security_xfrm_policy_lookup(pol->security,
1055                                                       fl->secid,
1056                                                       policy_to_flow_dir(dir));
1057                         if (!err)
1058                                 xfrm_pol_hold(pol);
1059                         else if (err == -ESRCH)
1060                                 pol = NULL;
1061                         else
1062                                 pol = ERR_PTR(err);
1063                 } else
1064                         pol = NULL;
1065         }
1066         read_unlock_bh(&xfrm_policy_lock);
1067         return pol;
1068 }
1069
1070 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1071 {
1072         struct net *net = xp_net(pol);
1073         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1074                                                      pol->family, dir);
1075
1076         list_add(&pol->walk.all, &net->xfrm.policy_all);
1077         hlist_add_head(&pol->bydst, chain);
1078         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1079         net->xfrm.policy_count[dir]++;
1080         xfrm_pol_hold(pol);
1081
1082         if (xfrm_bydst_should_resize(net, dir, NULL))
1083                 schedule_work(&net->xfrm.policy_hash_work);
1084 }
1085
1086 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1087                                                 int dir)
1088 {
1089         struct net *net = xp_net(pol);
1090
1091         if (hlist_unhashed(&pol->bydst))
1092                 return NULL;
1093
1094         hlist_del(&pol->bydst);
1095         hlist_del(&pol->byidx);
1096         list_del(&pol->walk.all);
1097         net->xfrm.policy_count[dir]--;
1098
1099         return pol;
1100 }
1101
1102 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1103 {
1104         write_lock_bh(&xfrm_policy_lock);
1105         pol = __xfrm_policy_unlink(pol, dir);
1106         write_unlock_bh(&xfrm_policy_lock);
1107         if (pol) {
1108                 if (dir < XFRM_POLICY_MAX)
1109                         atomic_inc(&flow_cache_genid);
1110                 xfrm_policy_kill(pol);
1111                 return 0;
1112         }
1113         return -ENOENT;
1114 }
1115 EXPORT_SYMBOL(xfrm_policy_delete);
1116
1117 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1118 {
1119         struct xfrm_policy *old_pol;
1120
1121 #ifdef CONFIG_XFRM_SUB_POLICY
1122         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1123                 return -EINVAL;
1124 #endif
1125
1126         write_lock_bh(&xfrm_policy_lock);
1127         old_pol = sk->sk_policy[dir];
1128         sk->sk_policy[dir] = pol;
1129         if (pol) {
1130                 pol->curlft.add_time = get_seconds();
1131                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
1132                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1133         }
1134         if (old_pol)
1135                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1136         write_unlock_bh(&xfrm_policy_lock);
1137
1138         if (old_pol) {
1139                 xfrm_policy_kill(old_pol);
1140         }
1141         return 0;
1142 }
1143
1144 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1145 {
1146         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1147
1148         if (newp) {
1149                 newp->selector = old->selector;
1150                 if (security_xfrm_policy_clone(old->security,
1151                                                &newp->security)) {
1152                         kfree(newp);
1153                         return NULL;  /* ENOMEM */
1154                 }
1155                 newp->lft = old->lft;
1156                 newp->curlft = old->curlft;
1157                 newp->action = old->action;
1158                 newp->flags = old->flags;
1159                 newp->xfrm_nr = old->xfrm_nr;
1160                 newp->index = old->index;
1161                 newp->type = old->type;
1162                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1163                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1164                 write_lock_bh(&xfrm_policy_lock);
1165                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1166                 write_unlock_bh(&xfrm_policy_lock);
1167                 xfrm_pol_put(newp);
1168         }
1169         return newp;
1170 }
1171
1172 int __xfrm_sk_clone_policy(struct sock *sk)
1173 {
1174         struct xfrm_policy *p0 = sk->sk_policy[0],
1175                            *p1 = sk->sk_policy[1];
1176
1177         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1178         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1179                 return -ENOMEM;
1180         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1181                 return -ENOMEM;
1182         return 0;
1183 }
1184
1185 static int
1186 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1187                unsigned short family)
1188 {
1189         int err;
1190         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1191
1192         if (unlikely(afinfo == NULL))
1193                 return -EINVAL;
1194         err = afinfo->get_saddr(local, remote);
1195         xfrm_policy_put_afinfo(afinfo);
1196         return err;
1197 }
1198
1199 /* Resolve list of templates for the flow, given policy. */
1200
1201 static int
1202 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1203                       struct xfrm_state **xfrm,
1204                       unsigned short family)
1205 {
1206         int nx;
1207         int i, error;
1208         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1209         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1210         xfrm_address_t tmp;
1211
1212         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1213                 struct xfrm_state *x;
1214                 xfrm_address_t *remote = daddr;
1215                 xfrm_address_t *local  = saddr;
1216                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1217
1218                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1219                     tmpl->mode == XFRM_MODE_BEET) {
1220                         remote = &tmpl->id.daddr;
1221                         local = &tmpl->saddr;
1222                         family = tmpl->encap_family;
1223                         if (xfrm_addr_any(local, family)) {
1224                                 error = xfrm_get_saddr(&tmp, remote, family);
1225                                 if (error)
1226                                         goto fail;
1227                                 local = &tmp;
1228                         }
1229                 }
1230
1231                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1232
1233                 if (x && x->km.state == XFRM_STATE_VALID) {
1234                         xfrm[nx++] = x;
1235                         daddr = remote;
1236                         saddr = local;
1237                         continue;
1238                 }
1239                 if (x) {
1240                         error = (x->km.state == XFRM_STATE_ERROR ?
1241                                  -EINVAL : -EAGAIN);
1242                         xfrm_state_put(x);
1243                 }
1244                 else if (error == -ESRCH)
1245                         error = -EAGAIN;
1246
1247                 if (!tmpl->optional)
1248                         goto fail;
1249         }
1250         return nx;
1251
1252 fail:
1253         for (nx--; nx>=0; nx--)
1254                 xfrm_state_put(xfrm[nx]);
1255         return error;
1256 }
1257
1258 static int
1259 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1260                   struct xfrm_state **xfrm,
1261                   unsigned short family)
1262 {
1263         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1264         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1265         int cnx = 0;
1266         int error;
1267         int ret;
1268         int i;
1269
1270         for (i = 0; i < npols; i++) {
1271                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1272                         error = -ENOBUFS;
1273                         goto fail;
1274                 }
1275
1276                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1277                 if (ret < 0) {
1278                         error = ret;
1279                         goto fail;
1280                 } else
1281                         cnx += ret;
1282         }
1283
1284         /* found states are sorted for outbound processing */
1285         if (npols > 1)
1286                 xfrm_state_sort(xfrm, tpp, cnx, family);
1287
1288         return cnx;
1289
1290  fail:
1291         for (cnx--; cnx>=0; cnx--)
1292                 xfrm_state_put(tpp[cnx]);
1293         return error;
1294
1295 }
1296
1297 /* Check that the bundle accepts the flow and its components are
1298  * still valid.
1299  */
1300
1301 static struct dst_entry *
1302 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1303 {
1304         struct dst_entry *x;
1305         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1306         if (unlikely(afinfo == NULL))
1307                 return ERR_PTR(-EINVAL);
1308         x = afinfo->find_bundle(fl, policy);
1309         xfrm_policy_put_afinfo(afinfo);
1310         return x;
1311 }
1312
1313 static inline int xfrm_get_tos(struct flowi *fl, int family)
1314 {
1315         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1316         int tos;
1317
1318         if (!afinfo)
1319                 return -EINVAL;
1320
1321         tos = afinfo->get_tos(fl);
1322
1323         xfrm_policy_put_afinfo(afinfo);
1324
1325         return tos;
1326 }
1327
1328 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1329 {
1330         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1331         struct xfrm_dst *xdst;
1332
1333         if (!afinfo)
1334                 return ERR_PTR(-EINVAL);
1335
1336         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1337
1338         xfrm_policy_put_afinfo(afinfo);
1339
1340         return xdst;
1341 }
1342
1343 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1344                                  int nfheader_len)
1345 {
1346         struct xfrm_policy_afinfo *afinfo =
1347                 xfrm_policy_get_afinfo(dst->ops->family);
1348         int err;
1349
1350         if (!afinfo)
1351                 return -EINVAL;
1352
1353         err = afinfo->init_path(path, dst, nfheader_len);
1354
1355         xfrm_policy_put_afinfo(afinfo);
1356
1357         return err;
1358 }
1359
1360 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1361 {
1362         struct xfrm_policy_afinfo *afinfo =
1363                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1364         int err;
1365
1366         if (!afinfo)
1367                 return -EINVAL;
1368
1369         err = afinfo->fill_dst(xdst, dev);
1370
1371         xfrm_policy_put_afinfo(afinfo);
1372
1373         return err;
1374 }
1375
1376 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1377  * all the metrics... Shortly, bundle a bundle.
1378  */
1379
1380 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1381                                             struct xfrm_state **xfrm, int nx,
1382                                             struct flowi *fl,
1383                                             struct dst_entry *dst)
1384 {
1385         unsigned long now = jiffies;
1386         struct net_device *dev;
1387         struct dst_entry *dst_prev = NULL;
1388         struct dst_entry *dst0 = NULL;
1389         int i = 0;
1390         int err;
1391         int header_len = 0;
1392         int nfheader_len = 0;
1393         int trailer_len = 0;
1394         int tos;
1395         int family = policy->selector.family;
1396         xfrm_address_t saddr, daddr;
1397
1398         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1399
1400         tos = xfrm_get_tos(fl, family);
1401         err = tos;
1402         if (tos < 0)
1403                 goto put_states;
1404
1405         dst_hold(dst);
1406
1407         for (; i < nx; i++) {
1408                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1409                 struct dst_entry *dst1 = &xdst->u.dst;
1410
1411                 err = PTR_ERR(xdst);
1412                 if (IS_ERR(xdst)) {
1413                         dst_release(dst);
1414                         goto put_states;
1415                 }
1416
1417                 if (!dst_prev)
1418                         dst0 = dst1;
1419                 else {
1420                         dst_prev->child = dst_clone(dst1);
1421                         dst1->flags |= DST_NOHASH;
1422                 }
1423
1424                 xdst->route = dst;
1425                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1426
1427                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1428                         family = xfrm[i]->props.family;
1429                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1430                                               family);
1431                         err = PTR_ERR(dst);
1432                         if (IS_ERR(dst))
1433                                 goto put_states;
1434                 } else
1435                         dst_hold(dst);
1436
1437                 dst1->xfrm = xfrm[i];
1438                 xdst->genid = xfrm[i]->genid;
1439
1440                 dst1->obsolete = -1;
1441                 dst1->flags |= DST_HOST;
1442                 dst1->lastuse = now;
1443
1444                 dst1->input = dst_discard;
1445                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1446
1447                 dst1->next = dst_prev;
1448                 dst_prev = dst1;
1449
1450                 header_len += xfrm[i]->props.header_len;
1451                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1452                         nfheader_len += xfrm[i]->props.header_len;
1453                 trailer_len += xfrm[i]->props.trailer_len;
1454         }
1455
1456         dst_prev->child = dst;
1457         dst0->path = dst;
1458
1459         err = -ENODEV;
1460         dev = dst->dev;
1461         if (!dev)
1462                 goto free_dst;
1463
1464         /* Copy neighbout for reachability confirmation */
1465         dst0->neighbour = neigh_clone(dst->neighbour);
1466
1467         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1468         xfrm_init_pmtu(dst_prev);
1469
1470         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1471                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1472
1473                 err = xfrm_fill_dst(xdst, dev);
1474                 if (err)
1475                         goto free_dst;
1476
1477                 dst_prev->header_len = header_len;
1478                 dst_prev->trailer_len = trailer_len;
1479                 header_len -= xdst->u.dst.xfrm->props.header_len;
1480                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1481         }
1482
1483 out:
1484         return dst0;
1485
1486 put_states:
1487         for (; i < nx; i++)
1488                 xfrm_state_put(xfrm[i]);
1489 free_dst:
1490         if (dst0)
1491                 dst_free(dst0);
1492         dst0 = ERR_PTR(err);
1493         goto out;
1494 }
1495
1496 static int inline
1497 xfrm_dst_alloc_copy(void **target, void *src, int size)
1498 {
1499         if (!*target) {
1500                 *target = kmalloc(size, GFP_ATOMIC);
1501                 if (!*target)
1502                         return -ENOMEM;
1503         }
1504         memcpy(*target, src, size);
1505         return 0;
1506 }
1507
1508 static int inline
1509 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1510 {
1511 #ifdef CONFIG_XFRM_SUB_POLICY
1512         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1513         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1514                                    sel, sizeof(*sel));
1515 #else
1516         return 0;
1517 #endif
1518 }
1519
1520 static int inline
1521 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1522 {
1523 #ifdef CONFIG_XFRM_SUB_POLICY
1524         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1525         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1526 #else
1527         return 0;
1528 #endif
1529 }
1530
1531 static int stale_bundle(struct dst_entry *dst);
1532
1533 /* Main function: finds/creates a bundle for given flow.
1534  *
1535  * At the moment we eat a raw IP route. Mostly to speed up lookups
1536  * on interfaces with disabled IPsec.
1537  */
1538 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1539                   struct sock *sk, int flags)
1540 {
1541         struct xfrm_policy *policy;
1542         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1543         int npols;
1544         int pol_dead;
1545         int xfrm_nr;
1546         int pi;
1547         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1548         struct dst_entry *dst, *dst_orig = *dst_p;
1549         int nx = 0;
1550         int err;
1551         u32 genid;
1552         u16 family;
1553         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1554
1555 restart:
1556         genid = atomic_read(&flow_cache_genid);
1557         policy = NULL;
1558         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1559                 pols[pi] = NULL;
1560         npols = 0;
1561         pol_dead = 0;
1562         xfrm_nr = 0;
1563
1564         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1565                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1566                 err = PTR_ERR(policy);
1567                 if (IS_ERR(policy)) {
1568                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1569                         goto dropdst;
1570                 }
1571         }
1572
1573         if (!policy) {
1574                 /* To accelerate a bit...  */
1575                 if ((dst_orig->flags & DST_NOXFRM) ||
1576                     !init_net.xfrm.policy_count[XFRM_POLICY_OUT])
1577                         goto nopol;
1578
1579                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1580                                            dir, xfrm_policy_lookup);
1581                 err = PTR_ERR(policy);
1582                 if (IS_ERR(policy)) {
1583                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1584                         goto dropdst;
1585                 }
1586         }
1587
1588         if (!policy)
1589                 goto nopol;
1590
1591         family = dst_orig->ops->family;
1592         pols[0] = policy;
1593         npols ++;
1594         xfrm_nr += pols[0]->xfrm_nr;
1595
1596         err = -ENOENT;
1597         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1598                 goto error;
1599
1600         policy->curlft.use_time = get_seconds();
1601
1602         switch (policy->action) {
1603         default:
1604         case XFRM_POLICY_BLOCK:
1605                 /* Prohibit the flow */
1606                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1607                 err = -EPERM;
1608                 goto error;
1609
1610         case XFRM_POLICY_ALLOW:
1611 #ifndef CONFIG_XFRM_SUB_POLICY
1612                 if (policy->xfrm_nr == 0) {
1613                         /* Flow passes not transformed. */
1614                         xfrm_pol_put(policy);
1615                         return 0;
1616                 }
1617 #endif
1618
1619                 /* Try to find matching bundle.
1620                  *
1621                  * LATER: help from flow cache. It is optional, this
1622                  * is required only for output policy.
1623                  */
1624                 dst = xfrm_find_bundle(fl, policy, family);
1625                 if (IS_ERR(dst)) {
1626                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1627                         err = PTR_ERR(dst);
1628                         goto error;
1629                 }
1630
1631                 if (dst)
1632                         break;
1633
1634 #ifdef CONFIG_XFRM_SUB_POLICY
1635                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1636                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1637                                                             fl, family,
1638                                                             XFRM_POLICY_OUT);
1639                         if (pols[1]) {
1640                                 if (IS_ERR(pols[1])) {
1641                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1642                                         err = PTR_ERR(pols[1]);
1643                                         goto error;
1644                                 }
1645                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1646                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1647                                         err = -EPERM;
1648                                         goto error;
1649                                 }
1650                                 npols ++;
1651                                 xfrm_nr += pols[1]->xfrm_nr;
1652                         }
1653                 }
1654
1655                 /*
1656                  * Because neither flowi nor bundle information knows about
1657                  * transformation template size. On more than one policy usage
1658                  * we can realize whether all of them is bypass or not after
1659                  * they are searched. See above not-transformed bypass
1660                  * is surrounded by non-sub policy configuration, too.
1661                  */
1662                 if (xfrm_nr == 0) {
1663                         /* Flow passes not transformed. */
1664                         xfrm_pols_put(pols, npols);
1665                         return 0;
1666                 }
1667
1668 #endif
1669                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1670
1671                 if (unlikely(nx<0)) {
1672                         err = nx;
1673                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1674                                 /* EREMOTE tells the caller to generate
1675                                  * a one-shot blackhole route.
1676                                  */
1677                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1678                                 xfrm_pol_put(policy);
1679                                 return -EREMOTE;
1680                         }
1681                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1682                                 DECLARE_WAITQUEUE(wait, current);
1683
1684                                 add_wait_queue(&init_net.xfrm.km_waitq, &wait);
1685                                 set_current_state(TASK_INTERRUPTIBLE);
1686                                 schedule();
1687                                 set_current_state(TASK_RUNNING);
1688                                 remove_wait_queue(&init_net.xfrm.km_waitq, &wait);
1689
1690                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1691
1692                                 if (nx == -EAGAIN && signal_pending(current)) {
1693                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1694                                         err = -ERESTART;
1695                                         goto error;
1696                                 }
1697                                 if (nx == -EAGAIN ||
1698                                     genid != atomic_read(&flow_cache_genid)) {
1699                                         xfrm_pols_put(pols, npols);
1700                                         goto restart;
1701                                 }
1702                                 err = nx;
1703                         }
1704                         if (err < 0) {
1705                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1706                                 goto error;
1707                         }
1708                 }
1709                 if (nx == 0) {
1710                         /* Flow passes not transformed. */
1711                         xfrm_pols_put(pols, npols);
1712                         return 0;
1713                 }
1714
1715                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1716                 err = PTR_ERR(dst);
1717                 if (IS_ERR(dst)) {
1718                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1719                         goto error;
1720                 }
1721
1722                 for (pi = 0; pi < npols; pi++) {
1723                         read_lock_bh(&pols[pi]->lock);
1724                         pol_dead |= pols[pi]->walk.dead;
1725                         read_unlock_bh(&pols[pi]->lock);
1726                 }
1727
1728                 write_lock_bh(&policy->lock);
1729                 if (unlikely(pol_dead || stale_bundle(dst))) {
1730                         /* Wow! While we worked on resolving, this
1731                          * policy has gone. Retry. It is not paranoia,
1732                          * we just cannot enlist new bundle to dead object.
1733                          * We can't enlist stable bundles either.
1734                          */
1735                         write_unlock_bh(&policy->lock);
1736                         dst_free(dst);
1737
1738                         if (pol_dead)
1739                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1740                         else
1741                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1742                         err = -EHOSTUNREACH;
1743                         goto error;
1744                 }
1745
1746                 if (npols > 1)
1747                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1748                 else
1749                         err = xfrm_dst_update_origin(dst, fl);
1750                 if (unlikely(err)) {
1751                         write_unlock_bh(&policy->lock);
1752                         dst_free(dst);
1753                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1754                         goto error;
1755                 }
1756
1757                 dst->next = policy->bundles;
1758                 policy->bundles = dst;
1759                 dst_hold(dst);
1760                 write_unlock_bh(&policy->lock);
1761         }
1762         *dst_p = dst;
1763         dst_release(dst_orig);
1764         xfrm_pols_put(pols, npols);
1765         return 0;
1766
1767 error:
1768         xfrm_pols_put(pols, npols);
1769 dropdst:
1770         dst_release(dst_orig);
1771         *dst_p = NULL;
1772         return err;
1773
1774 nopol:
1775         err = -ENOENT;
1776         if (flags & XFRM_LOOKUP_ICMP)
1777                 goto dropdst;
1778         return 0;
1779 }
1780 EXPORT_SYMBOL(__xfrm_lookup);
1781
1782 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1783                 struct sock *sk, int flags)
1784 {
1785         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1786
1787         if (err == -EREMOTE) {
1788                 dst_release(*dst_p);
1789                 *dst_p = NULL;
1790                 err = -EAGAIN;
1791         }
1792
1793         return err;
1794 }
1795 EXPORT_SYMBOL(xfrm_lookup);
1796
1797 static inline int
1798 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1799 {
1800         struct xfrm_state *x;
1801
1802         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1803                 return 0;
1804         x = skb->sp->xvec[idx];
1805         if (!x->type->reject)
1806                 return 0;
1807         return x->type->reject(x, skb, fl);
1808 }
1809
1810 /* When skb is transformed back to its "native" form, we have to
1811  * check policy restrictions. At the moment we make this in maximally
1812  * stupid way. Shame on me. :-) Of course, connected sockets must
1813  * have policy cached at them.
1814  */
1815
1816 static inline int
1817 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1818               unsigned short family)
1819 {
1820         if (xfrm_state_kern(x))
1821                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1822         return  x->id.proto == tmpl->id.proto &&
1823                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1824                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1825                 x->props.mode == tmpl->mode &&
1826                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1827                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1828                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1829                   xfrm_state_addr_cmp(tmpl, x, family));
1830 }
1831
1832 /*
1833  * 0 or more than 0 is returned when validation is succeeded (either bypass
1834  * because of optional transport mode, or next index of the mathced secpath
1835  * state with the template.
1836  * -1 is returned when no matching template is found.
1837  * Otherwise "-2 - errored_index" is returned.
1838  */
1839 static inline int
1840 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1841                unsigned short family)
1842 {
1843         int idx = start;
1844
1845         if (tmpl->optional) {
1846                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1847                         return start;
1848         } else
1849                 start = -1;
1850         for (; idx < sp->len; idx++) {
1851                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1852                         return ++idx;
1853                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1854                         if (start == -1)
1855                                 start = -2-idx;
1856                         break;
1857                 }
1858         }
1859         return start;
1860 }
1861
1862 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1863                           unsigned int family, int reverse)
1864 {
1865         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1866         int err;
1867
1868         if (unlikely(afinfo == NULL))
1869                 return -EAFNOSUPPORT;
1870
1871         afinfo->decode_session(skb, fl, reverse);
1872         err = security_xfrm_decode_session(skb, &fl->secid);
1873         xfrm_policy_put_afinfo(afinfo);
1874         return err;
1875 }
1876 EXPORT_SYMBOL(__xfrm_decode_session);
1877
1878 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1879 {
1880         for (; k < sp->len; k++) {
1881                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1882                         *idxp = k;
1883                         return 1;
1884                 }
1885         }
1886
1887         return 0;
1888 }
1889
1890 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1891                         unsigned short family)
1892 {
1893         struct xfrm_policy *pol;
1894         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1895         int npols = 0;
1896         int xfrm_nr;
1897         int pi;
1898         int reverse;
1899         struct flowi fl;
1900         u8 fl_dir;
1901         int xerr_idx = -1;
1902
1903         reverse = dir & ~XFRM_POLICY_MASK;
1904         dir &= XFRM_POLICY_MASK;
1905         fl_dir = policy_to_flow_dir(dir);
1906
1907         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1908                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1909                 return 0;
1910         }
1911
1912         nf_nat_decode_session(skb, &fl, family);
1913
1914         /* First, check used SA against their selectors. */
1915         if (skb->sp) {
1916                 int i;
1917
1918                 for (i=skb->sp->len-1; i>=0; i--) {
1919                         struct xfrm_state *x = skb->sp->xvec[i];
1920                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1921                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1922                                 return 0;
1923                         }
1924                 }
1925         }
1926
1927         pol = NULL;
1928         if (sk && sk->sk_policy[dir]) {
1929                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1930                 if (IS_ERR(pol)) {
1931                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1932                         return 0;
1933                 }
1934         }
1935
1936         if (!pol)
1937                 pol = flow_cache_lookup(&fl, family, fl_dir,
1938                                         xfrm_policy_lookup);
1939
1940         if (IS_ERR(pol)) {
1941                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1942                 return 0;
1943         }
1944
1945         if (!pol) {
1946                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1947                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1948                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1949                         return 0;
1950                 }
1951                 return 1;
1952         }
1953
1954         pol->curlft.use_time = get_seconds();
1955
1956         pols[0] = pol;
1957         npols ++;
1958 #ifdef CONFIG_XFRM_SUB_POLICY
1959         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1960                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1961                                                     &fl, family,
1962                                                     XFRM_POLICY_IN);
1963                 if (pols[1]) {
1964                         if (IS_ERR(pols[1])) {
1965                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1966                                 return 0;
1967                         }
1968                         pols[1]->curlft.use_time = get_seconds();
1969                         npols ++;
1970                 }
1971         }
1972 #endif
1973
1974         if (pol->action == XFRM_POLICY_ALLOW) {
1975                 struct sec_path *sp;
1976                 static struct sec_path dummy;
1977                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1978                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1979                 struct xfrm_tmpl **tpp = tp;
1980                 int ti = 0;
1981                 int i, k;
1982
1983                 if ((sp = skb->sp) == NULL)
1984                         sp = &dummy;
1985
1986                 for (pi = 0; pi < npols; pi++) {
1987                         if (pols[pi] != pol &&
1988                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1989                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1990                                 goto reject;
1991                         }
1992                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1993                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1994                                 goto reject_error;
1995                         }
1996                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1997                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1998                 }
1999                 xfrm_nr = ti;
2000                 if (npols > 1) {
2001                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2002                         tpp = stp;
2003                 }
2004
2005                 /* For each tunnel xfrm, find the first matching tmpl.
2006                  * For each tmpl before that, find corresponding xfrm.
2007                  * Order is _important_. Later we will implement
2008                  * some barriers, but at the moment barriers
2009                  * are implied between each two transformations.
2010                  */
2011                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2012                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2013                         if (k < 0) {
2014                                 if (k < -1)
2015                                         /* "-2 - errored_index" returned */
2016                                         xerr_idx = -(2+k);
2017                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2018                                 goto reject;
2019                         }
2020                 }
2021
2022                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2023                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2024                         goto reject;
2025                 }
2026
2027                 xfrm_pols_put(pols, npols);
2028                 return 1;
2029         }
2030         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2031
2032 reject:
2033         xfrm_secpath_reject(xerr_idx, skb, &fl);
2034 reject_error:
2035         xfrm_pols_put(pols, npols);
2036         return 0;
2037 }
2038 EXPORT_SYMBOL(__xfrm_policy_check);
2039
2040 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2041 {
2042         struct flowi fl;
2043
2044         if (xfrm_decode_session(skb, &fl, family) < 0) {
2045                 /* XXX: we should have something like FWDHDRERROR here. */
2046                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2047                 return 0;
2048         }
2049
2050         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2051 }
2052 EXPORT_SYMBOL(__xfrm_route_forward);
2053
2054 /* Optimize later using cookies and generation ids. */
2055
2056 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2057 {
2058         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2059          * to "-1" to force all XFRM destinations to get validated by
2060          * dst_ops->check on every use.  We do this because when a
2061          * normal route referenced by an XFRM dst is obsoleted we do
2062          * not go looking around for all parent referencing XFRM dsts
2063          * so that we can invalidate them.  It is just too much work.
2064          * Instead we make the checks here on every use.  For example:
2065          *
2066          *      XFRM dst A --> IPv4 dst X
2067          *
2068          * X is the "xdst->route" of A (X is also the "dst->path" of A
2069          * in this example).  If X is marked obsolete, "A" will not
2070          * notice.  That's what we are validating here via the
2071          * stale_bundle() check.
2072          *
2073          * When a policy's bundle is pruned, we dst_free() the XFRM
2074          * dst which causes it's ->obsolete field to be set to a
2075          * positive non-zero integer.  If an XFRM dst has been pruned
2076          * like this, we want to force a new route lookup.
2077          */
2078         if (dst->obsolete < 0 && !stale_bundle(dst))
2079                 return dst;
2080
2081         return NULL;
2082 }
2083
2084 static int stale_bundle(struct dst_entry *dst)
2085 {
2086         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2087 }
2088
2089 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2090 {
2091         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2092                 dst->dev = dev_net(dev)->loopback_dev;
2093                 dev_hold(dst->dev);
2094                 dev_put(dev);
2095         }
2096 }
2097 EXPORT_SYMBOL(xfrm_dst_ifdown);
2098
2099 static void xfrm_link_failure(struct sk_buff *skb)
2100 {
2101         /* Impossible. Such dst must be popped before reaches point of failure. */
2102         return;
2103 }
2104
2105 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2106 {
2107         if (dst) {
2108                 if (dst->obsolete) {
2109                         dst_release(dst);
2110                         dst = NULL;
2111                 }
2112         }
2113         return dst;
2114 }
2115
2116 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2117 {
2118         struct dst_entry *dst, **dstp;
2119
2120         write_lock(&pol->lock);
2121         dstp = &pol->bundles;
2122         while ((dst=*dstp) != NULL) {
2123                 if (func(dst)) {
2124                         *dstp = dst->next;
2125                         dst->next = *gc_list_p;
2126                         *gc_list_p = dst;
2127                 } else {
2128                         dstp = &dst->next;
2129                 }
2130         }
2131         write_unlock(&pol->lock);
2132 }
2133
2134 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2135 {
2136         struct dst_entry *gc_list = NULL;
2137         int dir;
2138
2139         read_lock_bh(&xfrm_policy_lock);
2140         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2141                 struct xfrm_policy *pol;
2142                 struct hlist_node *entry;
2143                 struct hlist_head *table;
2144                 int i;
2145
2146                 hlist_for_each_entry(pol, entry,
2147                                      &init_net.xfrm.policy_inexact[dir], bydst)
2148                         prune_one_bundle(pol, func, &gc_list);
2149
2150                 table = init_net.xfrm.policy_bydst[dir].table;
2151                 for (i = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2152                         hlist_for_each_entry(pol, entry, table + i, bydst)
2153                                 prune_one_bundle(pol, func, &gc_list);
2154                 }
2155         }
2156         read_unlock_bh(&xfrm_policy_lock);
2157
2158         while (gc_list) {
2159                 struct dst_entry *dst = gc_list;
2160                 gc_list = dst->next;
2161                 dst_free(dst);
2162         }
2163 }
2164
2165 static int unused_bundle(struct dst_entry *dst)
2166 {
2167         return !atomic_read(&dst->__refcnt);
2168 }
2169
2170 static void __xfrm_garbage_collect(void)
2171 {
2172         xfrm_prune_bundles(unused_bundle);
2173 }
2174
2175 static int xfrm_flush_bundles(void)
2176 {
2177         xfrm_prune_bundles(stale_bundle);
2178         return 0;
2179 }
2180
2181 static void xfrm_init_pmtu(struct dst_entry *dst)
2182 {
2183         do {
2184                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2185                 u32 pmtu, route_mtu_cached;
2186
2187                 pmtu = dst_mtu(dst->child);
2188                 xdst->child_mtu_cached = pmtu;
2189
2190                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2191
2192                 route_mtu_cached = dst_mtu(xdst->route);
2193                 xdst->route_mtu_cached = route_mtu_cached;
2194
2195                 if (pmtu > route_mtu_cached)
2196                         pmtu = route_mtu_cached;
2197
2198                 dst->metrics[RTAX_MTU-1] = pmtu;
2199         } while ((dst = dst->next));
2200 }
2201
2202 /* Check that the bundle accepts the flow and its components are
2203  * still valid.
2204  */
2205
2206 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2207                 struct flowi *fl, int family, int strict)
2208 {
2209         struct dst_entry *dst = &first->u.dst;
2210         struct xfrm_dst *last;
2211         u32 mtu;
2212
2213         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2214             (dst->dev && !netif_running(dst->dev)))
2215                 return 0;
2216 #ifdef CONFIG_XFRM_SUB_POLICY
2217         if (fl) {
2218                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2219                         return 0;
2220                 if (first->partner &&
2221                     !xfrm_selector_match(first->partner, fl, family))
2222                         return 0;
2223         }
2224 #endif
2225
2226         last = NULL;
2227
2228         do {
2229                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2230
2231                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2232                         return 0;
2233                 if (fl && pol &&
2234                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2235                         return 0;
2236                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2237                         return 0;
2238                 if (xdst->genid != dst->xfrm->genid)
2239                         return 0;
2240
2241                 if (strict && fl &&
2242                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2243                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2244                         return 0;
2245
2246                 mtu = dst_mtu(dst->child);
2247                 if (xdst->child_mtu_cached != mtu) {
2248                         last = xdst;
2249                         xdst->child_mtu_cached = mtu;
2250                 }
2251
2252                 if (!dst_check(xdst->route, xdst->route_cookie))
2253                         return 0;
2254                 mtu = dst_mtu(xdst->route);
2255                 if (xdst->route_mtu_cached != mtu) {
2256                         last = xdst;
2257                         xdst->route_mtu_cached = mtu;
2258                 }
2259
2260                 dst = dst->child;
2261         } while (dst->xfrm);
2262
2263         if (likely(!last))
2264                 return 1;
2265
2266         mtu = last->child_mtu_cached;
2267         for (;;) {
2268                 dst = &last->u.dst;
2269
2270                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2271                 if (mtu > last->route_mtu_cached)
2272                         mtu = last->route_mtu_cached;
2273                 dst->metrics[RTAX_MTU-1] = mtu;
2274
2275                 if (last == first)
2276                         break;
2277
2278                 last = (struct xfrm_dst *)last->u.dst.next;
2279                 last->child_mtu_cached = mtu;
2280         }
2281
2282         return 1;
2283 }
2284
2285 EXPORT_SYMBOL(xfrm_bundle_ok);
2286
2287 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2288 {
2289         int err = 0;
2290         if (unlikely(afinfo == NULL))
2291                 return -EINVAL;
2292         if (unlikely(afinfo->family >= NPROTO))
2293                 return -EAFNOSUPPORT;
2294         write_lock_bh(&xfrm_policy_afinfo_lock);
2295         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2296                 err = -ENOBUFS;
2297         else {
2298                 struct dst_ops *dst_ops = afinfo->dst_ops;
2299                 if (likely(dst_ops->kmem_cachep == NULL))
2300                         dst_ops->kmem_cachep = xfrm_dst_cache;
2301                 if (likely(dst_ops->check == NULL))
2302                         dst_ops->check = xfrm_dst_check;
2303                 if (likely(dst_ops->negative_advice == NULL))
2304                         dst_ops->negative_advice = xfrm_negative_advice;
2305                 if (likely(dst_ops->link_failure == NULL))
2306                         dst_ops->link_failure = xfrm_link_failure;
2307                 if (likely(afinfo->garbage_collect == NULL))
2308                         afinfo->garbage_collect = __xfrm_garbage_collect;
2309                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2310         }
2311         write_unlock_bh(&xfrm_policy_afinfo_lock);
2312         return err;
2313 }
2314 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2315
2316 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2317 {
2318         int err = 0;
2319         if (unlikely(afinfo == NULL))
2320                 return -EINVAL;
2321         if (unlikely(afinfo->family >= NPROTO))
2322                 return -EAFNOSUPPORT;
2323         write_lock_bh(&xfrm_policy_afinfo_lock);
2324         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2325                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2326                         err = -EINVAL;
2327                 else {
2328                         struct dst_ops *dst_ops = afinfo->dst_ops;
2329                         xfrm_policy_afinfo[afinfo->family] = NULL;
2330                         dst_ops->kmem_cachep = NULL;
2331                         dst_ops->check = NULL;
2332                         dst_ops->negative_advice = NULL;
2333                         dst_ops->link_failure = NULL;
2334                         afinfo->garbage_collect = NULL;
2335                 }
2336         }
2337         write_unlock_bh(&xfrm_policy_afinfo_lock);
2338         return err;
2339 }
2340 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2341
2342 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2343 {
2344         struct xfrm_policy_afinfo *afinfo;
2345         if (unlikely(family >= NPROTO))
2346                 return NULL;
2347         read_lock(&xfrm_policy_afinfo_lock);
2348         afinfo = xfrm_policy_afinfo[family];
2349         if (unlikely(!afinfo))
2350                 read_unlock(&xfrm_policy_afinfo_lock);
2351         return afinfo;
2352 }
2353
2354 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2355 {
2356         read_unlock(&xfrm_policy_afinfo_lock);
2357 }
2358
2359 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2360 {
2361         struct net_device *dev = ptr;
2362
2363         if (!net_eq(dev_net(dev), &init_net))
2364                 return NOTIFY_DONE;
2365
2366         switch (event) {
2367         case NETDEV_DOWN:
2368                 xfrm_flush_bundles();
2369         }
2370         return NOTIFY_DONE;
2371 }
2372
2373 static struct notifier_block xfrm_dev_notifier = {
2374         .notifier_call  = xfrm_dev_event,
2375 };
2376
2377 #ifdef CONFIG_XFRM_STATISTICS
2378 static int __init xfrm_statistics_init(void)
2379 {
2380         if (snmp_mib_init((void **)xfrm_statistics,
2381                           sizeof(struct linux_xfrm_mib)) < 0)
2382                 return -ENOMEM;
2383         return 0;
2384 }
2385 #endif
2386
2387 static int __net_init xfrm_policy_init(struct net *net)
2388 {
2389         unsigned int hmask, sz;
2390         int dir;
2391
2392         if (net_eq(net, &init_net))
2393                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2394                                            sizeof(struct xfrm_dst),
2395                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2396                                            NULL);
2397
2398         hmask = 8 - 1;
2399         sz = (hmask+1) * sizeof(struct hlist_head);
2400
2401         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2402         if (!net->xfrm.policy_byidx)
2403                 goto out_byidx;
2404         net->xfrm.policy_idx_hmask = hmask;
2405
2406         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2407                 struct xfrm_policy_hash *htab;
2408
2409                 net->xfrm.policy_count[dir] = 0;
2410                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2411
2412                 htab = &net->xfrm.policy_bydst[dir];
2413                 htab->table = xfrm_hash_alloc(sz);
2414                 if (!htab->table)
2415                         goto out_bydst;
2416                 htab->hmask = hmask;
2417         }
2418
2419         INIT_LIST_HEAD(&net->xfrm.policy_all);
2420         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2421         if (net_eq(net, &init_net))
2422                 register_netdevice_notifier(&xfrm_dev_notifier);
2423         return 0;
2424
2425 out_bydst:
2426         for (dir--; dir >= 0; dir--) {
2427                 struct xfrm_policy_hash *htab;
2428
2429                 htab = &net->xfrm.policy_bydst[dir];
2430                 xfrm_hash_free(htab->table, sz);
2431         }
2432         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2433 out_byidx:
2434         return -ENOMEM;
2435 }
2436
2437 static void xfrm_policy_fini(struct net *net)
2438 {
2439         unsigned int sz;
2440         int dir;
2441
2442         WARN_ON(!list_empty(&net->xfrm.policy_all));
2443
2444         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2445                 struct xfrm_policy_hash *htab;
2446
2447                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2448
2449                 htab = &net->xfrm.policy_bydst[dir];
2450                 sz = (htab->hmask + 1);
2451                 WARN_ON(!hlist_empty(htab->table));
2452                 xfrm_hash_free(htab->table, sz);
2453         }
2454
2455         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2456         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2457         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2458 }
2459
2460 static int __net_init xfrm_net_init(struct net *net)
2461 {
2462         int rv;
2463
2464         rv = xfrm_state_init(net);
2465         if (rv < 0)
2466                 goto out_state;
2467         rv = xfrm_policy_init(net);
2468         if (rv < 0)
2469                 goto out_policy;
2470         return 0;
2471
2472 out_policy:
2473         xfrm_state_fini(net);
2474 out_state:
2475         return rv;
2476 }
2477
2478 static void __net_exit xfrm_net_exit(struct net *net)
2479 {
2480         xfrm_policy_fini(net);
2481         xfrm_state_fini(net);
2482 }
2483
2484 static struct pernet_operations __net_initdata xfrm_net_ops = {
2485         .init = xfrm_net_init,
2486         .exit = xfrm_net_exit,
2487 };
2488
2489 void __init xfrm_init(void)
2490 {
2491         register_pernet_subsys(&xfrm_net_ops);
2492 #ifdef CONFIG_XFRM_STATISTICS
2493         xfrm_statistics_init();
2494 #endif
2495         xfrm_input_init();
2496 #ifdef CONFIG_XFRM_STATISTICS
2497         xfrm_proc_init();
2498 #endif
2499 }
2500
2501 #ifdef CONFIG_AUDITSYSCALL
2502 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2503                                          struct audit_buffer *audit_buf)
2504 {
2505         struct xfrm_sec_ctx *ctx = xp->security;
2506         struct xfrm_selector *sel = &xp->selector;
2507
2508         if (ctx)
2509                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2510                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2511
2512         switch(sel->family) {
2513         case AF_INET:
2514                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2515                 if (sel->prefixlen_s != 32)
2516                         audit_log_format(audit_buf, " src_prefixlen=%d",
2517                                          sel->prefixlen_s);
2518                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2519                 if (sel->prefixlen_d != 32)
2520                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2521                                          sel->prefixlen_d);
2522                 break;
2523         case AF_INET6:
2524                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2525                 if (sel->prefixlen_s != 128)
2526                         audit_log_format(audit_buf, " src_prefixlen=%d",
2527                                          sel->prefixlen_s);
2528                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2529                 if (sel->prefixlen_d != 128)
2530                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2531                                          sel->prefixlen_d);
2532                 break;
2533         }
2534 }
2535
2536 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2537                            uid_t auid, u32 sessionid, u32 secid)
2538 {
2539         struct audit_buffer *audit_buf;
2540
2541         audit_buf = xfrm_audit_start("SPD-add");
2542         if (audit_buf == NULL)
2543                 return;
2544         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2545         audit_log_format(audit_buf, " res=%u", result);
2546         xfrm_audit_common_policyinfo(xp, audit_buf);
2547         audit_log_end(audit_buf);
2548 }
2549 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2550
2551 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2552                               uid_t auid, u32 sessionid, u32 secid)
2553 {
2554         struct audit_buffer *audit_buf;
2555
2556         audit_buf = xfrm_audit_start("SPD-delete");
2557         if (audit_buf == NULL)
2558                 return;
2559         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2560         audit_log_format(audit_buf, " res=%u", result);
2561         xfrm_audit_common_policyinfo(xp, audit_buf);
2562         audit_log_end(audit_buf);
2563 }
2564 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2565 #endif
2566
2567 #ifdef CONFIG_XFRM_MIGRATE
2568 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2569                                        struct xfrm_selector *sel_tgt)
2570 {
2571         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2572                 if (sel_tgt->family == sel_cmp->family &&
2573                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2574                                   sel_cmp->family) == 0 &&
2575                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2576                                   sel_cmp->family) == 0 &&
2577                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2578                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2579                         return 1;
2580                 }
2581         } else {
2582                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2583                         return 1;
2584                 }
2585         }
2586         return 0;
2587 }
2588
2589 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2590                                                      u8 dir, u8 type)
2591 {
2592         struct xfrm_policy *pol, *ret = NULL;
2593         struct hlist_node *entry;
2594         struct hlist_head *chain;
2595         u32 priority = ~0U;
2596
2597         read_lock_bh(&xfrm_policy_lock);
2598         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2599         hlist_for_each_entry(pol, entry, chain, bydst) {
2600                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2601                     pol->type == type) {
2602                         ret = pol;
2603                         priority = ret->priority;
2604                         break;
2605                 }
2606         }
2607         chain = &init_net.xfrm.policy_inexact[dir];
2608         hlist_for_each_entry(pol, entry, chain, bydst) {
2609                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2610                     pol->type == type &&
2611                     pol->priority < priority) {
2612                         ret = pol;
2613                         break;
2614                 }
2615         }
2616
2617         if (ret)
2618                 xfrm_pol_hold(ret);
2619
2620         read_unlock_bh(&xfrm_policy_lock);
2621
2622         return ret;
2623 }
2624
2625 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2626 {
2627         int match = 0;
2628
2629         if (t->mode == m->mode && t->id.proto == m->proto &&
2630             (m->reqid == 0 || t->reqid == m->reqid)) {
2631                 switch (t->mode) {
2632                 case XFRM_MODE_TUNNEL:
2633                 case XFRM_MODE_BEET:
2634                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2635                                           m->old_family) == 0 &&
2636                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2637                                           m->old_family) == 0) {
2638                                 match = 1;
2639                         }
2640                         break;
2641                 case XFRM_MODE_TRANSPORT:
2642                         /* in case of transport mode, template does not store
2643                            any IP addresses, hence we just compare mode and
2644                            protocol */
2645                         match = 1;
2646                         break;
2647                 default:
2648                         break;
2649                 }
2650         }
2651         return match;
2652 }
2653
2654 /* update endpoint address(es) of template(s) */
2655 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2656                                struct xfrm_migrate *m, int num_migrate)
2657 {
2658         struct xfrm_migrate *mp;
2659         struct dst_entry *dst;
2660         int i, j, n = 0;
2661
2662         write_lock_bh(&pol->lock);
2663         if (unlikely(pol->walk.dead)) {
2664                 /* target policy has been deleted */
2665                 write_unlock_bh(&pol->lock);
2666                 return -ENOENT;
2667         }
2668
2669         for (i = 0; i < pol->xfrm_nr; i++) {
2670                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2671                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2672                                 continue;
2673                         n++;
2674                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2675                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2676                                 continue;
2677                         /* update endpoints */
2678                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2679                                sizeof(pol->xfrm_vec[i].id.daddr));
2680                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2681                                sizeof(pol->xfrm_vec[i].saddr));
2682                         pol->xfrm_vec[i].encap_family = mp->new_family;
2683                         /* flush bundles */
2684                         while ((dst = pol->bundles) != NULL) {
2685                                 pol->bundles = dst->next;
2686                                 dst_free(dst);
2687                         }
2688                 }
2689         }
2690
2691         write_unlock_bh(&pol->lock);
2692
2693         if (!n)
2694                 return -ENODATA;
2695
2696         return 0;
2697 }
2698
2699 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2700 {
2701         int i, j;
2702
2703         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2704                 return -EINVAL;
2705
2706         for (i = 0; i < num_migrate; i++) {
2707                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2708                                    m[i].old_family) == 0) &&
2709                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2710                                    m[i].old_family) == 0))
2711                         return -EINVAL;
2712                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2713                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2714                         return -EINVAL;
2715
2716                 /* check if there is any duplicated entry */
2717                 for (j = i + 1; j < num_migrate; j++) {
2718                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2719                                     sizeof(m[i].old_daddr)) &&
2720                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2721                                     sizeof(m[i].old_saddr)) &&
2722                             m[i].proto == m[j].proto &&
2723                             m[i].mode == m[j].mode &&
2724                             m[i].reqid == m[j].reqid &&
2725                             m[i].old_family == m[j].old_family)
2726                                 return -EINVAL;
2727                 }
2728         }
2729
2730         return 0;
2731 }
2732
2733 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2734                  struct xfrm_migrate *m, int num_migrate,
2735                  struct xfrm_kmaddress *k)
2736 {
2737         int i, err, nx_cur = 0, nx_new = 0;
2738         struct xfrm_policy *pol = NULL;
2739         struct xfrm_state *x, *xc;
2740         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2741         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2742         struct xfrm_migrate *mp;
2743
2744         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2745                 goto out;
2746
2747         /* Stage 1 - find policy */
2748         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2749                 err = -ENOENT;
2750                 goto out;
2751         }
2752
2753         /* Stage 2 - find and update state(s) */
2754         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2755                 if ((x = xfrm_migrate_state_find(mp))) {
2756                         x_cur[nx_cur] = x;
2757                         nx_cur++;
2758                         if ((xc = xfrm_state_migrate(x, mp))) {
2759                                 x_new[nx_new] = xc;
2760                                 nx_new++;
2761                         } else {
2762                                 err = -ENODATA;
2763                                 goto restore_state;
2764                         }
2765                 }
2766         }
2767
2768         /* Stage 3 - update policy */
2769         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2770                 goto restore_state;
2771
2772         /* Stage 4 - delete old state(s) */
2773         if (nx_cur) {
2774                 xfrm_states_put(x_cur, nx_cur);
2775                 xfrm_states_delete(x_cur, nx_cur);
2776         }
2777
2778         /* Stage 5 - announce */
2779         km_migrate(sel, dir, type, m, num_migrate, k);
2780
2781         xfrm_pol_put(pol);
2782
2783         return 0;
2784 out:
2785         return err;
2786
2787 restore_state:
2788         if (pol)
2789                 xfrm_pol_put(pol);
2790         if (nx_cur)
2791                 xfrm_states_put(x_cur, nx_cur);
2792         if (nx_new)
2793                 xfrm_states_delete(x_new, nx_new);
2794
2795         return err;
2796 }
2797 EXPORT_SYMBOL(xfrm_migrate);
2798 #endif