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(u32 index)
 325 {
 326         return __idx_hash(index, init_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(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(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(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 hlist_head *chain = policy_hash_bysel(&pol->selector,
1073                                                      pol->family, dir);
1074
1075         list_add(&pol->walk.all, &init_net.xfrm.policy_all);
1076         hlist_add_head(&pol->bydst, chain);
1077         hlist_add_head(&pol->byidx, init_net.xfrm.policy_byidx+idx_hash(pol->index));
1078         init_net.xfrm.policy_count[dir]++;
1079         xfrm_pol_hold(pol);
1080
1081         if (xfrm_bydst_should_resize(&init_net, dir, NULL))
1082                 schedule_work(&init_net.xfrm.policy_hash_work);
1083 }
1084
1085 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1086                                                 int dir)
1087 {
1088         if (hlist_unhashed(&pol->bydst))
1089                 return NULL;
1090
1091         hlist_del(&pol->bydst);
1092         hlist_del(&pol->byidx);
1093         list_del(&pol->walk.all);
1094         init_net.xfrm.policy_count[dir]--;
1095
1096         return pol;
1097 }
1098
1099 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1100 {
1101         write_lock_bh(&xfrm_policy_lock);
1102         pol = __xfrm_policy_unlink(pol, dir);
1103         write_unlock_bh(&xfrm_policy_lock);
1104         if (pol) {
1105                 if (dir < XFRM_POLICY_MAX)
1106                         atomic_inc(&flow_cache_genid);
1107                 xfrm_policy_kill(pol);
1108                 return 0;
1109         }
1110         return -ENOENT;
1111 }
1112 EXPORT_SYMBOL(xfrm_policy_delete);
1113
1114 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1115 {
1116         struct xfrm_policy *old_pol;
1117
1118 #ifdef CONFIG_XFRM_SUB_POLICY
1119         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1120                 return -EINVAL;
1121 #endif
1122
1123         write_lock_bh(&xfrm_policy_lock);
1124         old_pol = sk->sk_policy[dir];
1125         sk->sk_policy[dir] = pol;
1126         if (pol) {
1127                 pol->curlft.add_time = get_seconds();
1128                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
1129                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1130         }
1131         if (old_pol)
1132                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1133         write_unlock_bh(&xfrm_policy_lock);
1134
1135         if (old_pol) {
1136                 xfrm_policy_kill(old_pol);
1137         }
1138         return 0;
1139 }
1140
1141 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1142 {
1143         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1144
1145         if (newp) {
1146                 newp->selector = old->selector;
1147                 if (security_xfrm_policy_clone(old->security,
1148                                                &newp->security)) {
1149                         kfree(newp);
1150                         return NULL;  /* ENOMEM */
1151                 }
1152                 newp->lft = old->lft;
1153                 newp->curlft = old->curlft;
1154                 newp->action = old->action;
1155                 newp->flags = old->flags;
1156                 newp->xfrm_nr = old->xfrm_nr;
1157                 newp->index = old->index;
1158                 newp->type = old->type;
1159                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1160                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1161                 write_lock_bh(&xfrm_policy_lock);
1162                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1163                 write_unlock_bh(&xfrm_policy_lock);
1164                 xfrm_pol_put(newp);
1165         }
1166         return newp;
1167 }
1168
1169 int __xfrm_sk_clone_policy(struct sock *sk)
1170 {
1171         struct xfrm_policy *p0 = sk->sk_policy[0],
1172                            *p1 = sk->sk_policy[1];
1173
1174         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1175         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1176                 return -ENOMEM;
1177         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1178                 return -ENOMEM;
1179         return 0;
1180 }
1181
1182 static int
1183 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1184                unsigned short family)
1185 {
1186         int err;
1187         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1188
1189         if (unlikely(afinfo == NULL))
1190                 return -EINVAL;
1191         err = afinfo->get_saddr(local, remote);
1192         xfrm_policy_put_afinfo(afinfo);
1193         return err;
1194 }
1195
1196 /* Resolve list of templates for the flow, given policy. */
1197
1198 static int
1199 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1200                       struct xfrm_state **xfrm,
1201                       unsigned short family)
1202 {
1203         int nx;
1204         int i, error;
1205         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1206         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1207         xfrm_address_t tmp;
1208
1209         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1210                 struct xfrm_state *x;
1211                 xfrm_address_t *remote = daddr;
1212                 xfrm_address_t *local  = saddr;
1213                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1214
1215                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1216                     tmpl->mode == XFRM_MODE_BEET) {
1217                         remote = &tmpl->id.daddr;
1218                         local = &tmpl->saddr;
1219                         family = tmpl->encap_family;
1220                         if (xfrm_addr_any(local, family)) {
1221                                 error = xfrm_get_saddr(&tmp, remote, family);
1222                                 if (error)
1223                                         goto fail;
1224                                 local = &tmp;
1225                         }
1226                 }
1227
1228                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1229
1230                 if (x && x->km.state == XFRM_STATE_VALID) {
1231                         xfrm[nx++] = x;
1232                         daddr = remote;
1233                         saddr = local;
1234                         continue;
1235                 }
1236                 if (x) {
1237                         error = (x->km.state == XFRM_STATE_ERROR ?
1238                                  -EINVAL : -EAGAIN);
1239                         xfrm_state_put(x);
1240                 }
1241                 else if (error == -ESRCH)
1242                         error = -EAGAIN;
1243
1244                 if (!tmpl->optional)
1245                         goto fail;
1246         }
1247         return nx;
1248
1249 fail:
1250         for (nx--; nx>=0; nx--)
1251                 xfrm_state_put(xfrm[nx]);
1252         return error;
1253 }
1254
1255 static int
1256 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1257                   struct xfrm_state **xfrm,
1258                   unsigned short family)
1259 {
1260         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1261         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1262         int cnx = 0;
1263         int error;
1264         int ret;
1265         int i;
1266
1267         for (i = 0; i < npols; i++) {
1268                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1269                         error = -ENOBUFS;
1270                         goto fail;
1271                 }
1272
1273                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1274                 if (ret < 0) {
1275                         error = ret;
1276                         goto fail;
1277                 } else
1278                         cnx += ret;
1279         }
1280
1281         /* found states are sorted for outbound processing */
1282         if (npols > 1)
1283                 xfrm_state_sort(xfrm, tpp, cnx, family);
1284
1285         return cnx;
1286
1287  fail:
1288         for (cnx--; cnx>=0; cnx--)
1289                 xfrm_state_put(tpp[cnx]);
1290         return error;
1291
1292 }
1293
1294 /* Check that the bundle accepts the flow and its components are
1295  * still valid.
1296  */
1297
1298 static struct dst_entry *
1299 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1300 {
1301         struct dst_entry *x;
1302         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1303         if (unlikely(afinfo == NULL))
1304                 return ERR_PTR(-EINVAL);
1305         x = afinfo->find_bundle(fl, policy);
1306         xfrm_policy_put_afinfo(afinfo);
1307         return x;
1308 }
1309
1310 static inline int xfrm_get_tos(struct flowi *fl, int family)
1311 {
1312         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1313         int tos;
1314
1315         if (!afinfo)
1316                 return -EINVAL;
1317
1318         tos = afinfo->get_tos(fl);
1319
1320         xfrm_policy_put_afinfo(afinfo);
1321
1322         return tos;
1323 }
1324
1325 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1326 {
1327         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1328         struct xfrm_dst *xdst;
1329
1330         if (!afinfo)
1331                 return ERR_PTR(-EINVAL);
1332
1333         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1334
1335         xfrm_policy_put_afinfo(afinfo);
1336
1337         return xdst;
1338 }
1339
1340 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1341                                  int nfheader_len)
1342 {
1343         struct xfrm_policy_afinfo *afinfo =
1344                 xfrm_policy_get_afinfo(dst->ops->family);
1345         int err;
1346
1347         if (!afinfo)
1348                 return -EINVAL;
1349
1350         err = afinfo->init_path(path, dst, nfheader_len);
1351
1352         xfrm_policy_put_afinfo(afinfo);
1353
1354         return err;
1355 }
1356
1357 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1358 {
1359         struct xfrm_policy_afinfo *afinfo =
1360                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1361         int err;
1362
1363         if (!afinfo)
1364                 return -EINVAL;
1365
1366         err = afinfo->fill_dst(xdst, dev);
1367
1368         xfrm_policy_put_afinfo(afinfo);
1369
1370         return err;
1371 }
1372
1373 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1374  * all the metrics... Shortly, bundle a bundle.
1375  */
1376
1377 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1378                                             struct xfrm_state **xfrm, int nx,
1379                                             struct flowi *fl,
1380                                             struct dst_entry *dst)
1381 {
1382         unsigned long now = jiffies;
1383         struct net_device *dev;
1384         struct dst_entry *dst_prev = NULL;
1385         struct dst_entry *dst0 = NULL;
1386         int i = 0;
1387         int err;
1388         int header_len = 0;
1389         int nfheader_len = 0;
1390         int trailer_len = 0;
1391         int tos;
1392         int family = policy->selector.family;
1393         xfrm_address_t saddr, daddr;
1394
1395         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1396
1397         tos = xfrm_get_tos(fl, family);
1398         err = tos;
1399         if (tos < 0)
1400                 goto put_states;
1401
1402         dst_hold(dst);
1403
1404         for (; i < nx; i++) {
1405                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1406                 struct dst_entry *dst1 = &xdst->u.dst;
1407
1408                 err = PTR_ERR(xdst);
1409                 if (IS_ERR(xdst)) {
1410                         dst_release(dst);
1411                         goto put_states;
1412                 }
1413
1414                 if (!dst_prev)
1415                         dst0 = dst1;
1416                 else {
1417                         dst_prev->child = dst_clone(dst1);
1418                         dst1->flags |= DST_NOHASH;
1419                 }
1420
1421                 xdst->route = dst;
1422                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1423
1424                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1425                         family = xfrm[i]->props.family;
1426                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1427                                               family);
1428                         err = PTR_ERR(dst);
1429                         if (IS_ERR(dst))
1430                                 goto put_states;
1431                 } else
1432                         dst_hold(dst);
1433
1434                 dst1->xfrm = xfrm[i];
1435                 xdst->genid = xfrm[i]->genid;
1436
1437                 dst1->obsolete = -1;
1438                 dst1->flags |= DST_HOST;
1439                 dst1->lastuse = now;
1440
1441                 dst1->input = dst_discard;
1442                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1443
1444                 dst1->next = dst_prev;
1445                 dst_prev = dst1;
1446
1447                 header_len += xfrm[i]->props.header_len;
1448                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1449                         nfheader_len += xfrm[i]->props.header_len;
1450                 trailer_len += xfrm[i]->props.trailer_len;
1451         }
1452
1453         dst_prev->child = dst;
1454         dst0->path = dst;
1455
1456         err = -ENODEV;
1457         dev = dst->dev;
1458         if (!dev)
1459                 goto free_dst;
1460
1461         /* Copy neighbout for reachability confirmation */
1462         dst0->neighbour = neigh_clone(dst->neighbour);
1463
1464         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1465         xfrm_init_pmtu(dst_prev);
1466
1467         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1468                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1469
1470                 err = xfrm_fill_dst(xdst, dev);
1471                 if (err)
1472                         goto free_dst;
1473
1474                 dst_prev->header_len = header_len;
1475                 dst_prev->trailer_len = trailer_len;
1476                 header_len -= xdst->u.dst.xfrm->props.header_len;
1477                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1478         }
1479
1480 out:
1481         return dst0;
1482
1483 put_states:
1484         for (; i < nx; i++)
1485                 xfrm_state_put(xfrm[i]);
1486 free_dst:
1487         if (dst0)
1488                 dst_free(dst0);
1489         dst0 = ERR_PTR(err);
1490         goto out;
1491 }
1492
1493 static int inline
1494 xfrm_dst_alloc_copy(void **target, void *src, int size)
1495 {
1496         if (!*target) {
1497                 *target = kmalloc(size, GFP_ATOMIC);
1498                 if (!*target)
1499                         return -ENOMEM;
1500         }
1501         memcpy(*target, src, size);
1502         return 0;
1503 }
1504
1505 static int inline
1506 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1507 {
1508 #ifdef CONFIG_XFRM_SUB_POLICY
1509         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1510         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1511                                    sel, sizeof(*sel));
1512 #else
1513         return 0;
1514 #endif
1515 }
1516
1517 static int inline
1518 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1519 {
1520 #ifdef CONFIG_XFRM_SUB_POLICY
1521         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1522         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1523 #else
1524         return 0;
1525 #endif
1526 }
1527
1528 static int stale_bundle(struct dst_entry *dst);
1529
1530 /* Main function: finds/creates a bundle for given flow.
1531  *
1532  * At the moment we eat a raw IP route. Mostly to speed up lookups
1533  * on interfaces with disabled IPsec.
1534  */
1535 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1536                   struct sock *sk, int flags)
1537 {
1538         struct xfrm_policy *policy;
1539         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1540         int npols;
1541         int pol_dead;
1542         int xfrm_nr;
1543         int pi;
1544         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1545         struct dst_entry *dst, *dst_orig = *dst_p;
1546         int nx = 0;
1547         int err;
1548         u32 genid;
1549         u16 family;
1550         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1551
1552 restart:
1553         genid = atomic_read(&flow_cache_genid);
1554         policy = NULL;
1555         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1556                 pols[pi] = NULL;
1557         npols = 0;
1558         pol_dead = 0;
1559         xfrm_nr = 0;
1560
1561         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1562                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1563                 err = PTR_ERR(policy);
1564                 if (IS_ERR(policy)) {
1565                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1566                         goto dropdst;
1567                 }
1568         }
1569
1570         if (!policy) {
1571                 /* To accelerate a bit...  */
1572                 if ((dst_orig->flags & DST_NOXFRM) ||
1573                     !init_net.xfrm.policy_count[XFRM_POLICY_OUT])
1574                         goto nopol;
1575
1576                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1577                                            dir, xfrm_policy_lookup);
1578                 err = PTR_ERR(policy);
1579                 if (IS_ERR(policy)) {
1580                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1581                         goto dropdst;
1582                 }
1583         }
1584
1585         if (!policy)
1586                 goto nopol;
1587
1588         family = dst_orig->ops->family;
1589         pols[0] = policy;
1590         npols ++;
1591         xfrm_nr += pols[0]->xfrm_nr;
1592
1593         err = -ENOENT;
1594         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1595                 goto error;
1596
1597         policy->curlft.use_time = get_seconds();
1598
1599         switch (policy->action) {
1600         default:
1601         case XFRM_POLICY_BLOCK:
1602                 /* Prohibit the flow */
1603                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1604                 err = -EPERM;
1605                 goto error;
1606
1607         case XFRM_POLICY_ALLOW:
1608 #ifndef CONFIG_XFRM_SUB_POLICY
1609                 if (policy->xfrm_nr == 0) {
1610                         /* Flow passes not transformed. */
1611                         xfrm_pol_put(policy);
1612                         return 0;
1613                 }
1614 #endif
1615
1616                 /* Try to find matching bundle.
1617                  *
1618                  * LATER: help from flow cache. It is optional, this
1619                  * is required only for output policy.
1620                  */
1621                 dst = xfrm_find_bundle(fl, policy, family);
1622                 if (IS_ERR(dst)) {
1623                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1624                         err = PTR_ERR(dst);
1625                         goto error;
1626                 }
1627
1628                 if (dst)
1629                         break;
1630
1631 #ifdef CONFIG_XFRM_SUB_POLICY
1632                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1633                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1634                                                             fl, family,
1635                                                             XFRM_POLICY_OUT);
1636                         if (pols[1]) {
1637                                 if (IS_ERR(pols[1])) {
1638                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1639                                         err = PTR_ERR(pols[1]);
1640                                         goto error;
1641                                 }
1642                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1643                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1644                                         err = -EPERM;
1645                                         goto error;
1646                                 }
1647                                 npols ++;
1648                                 xfrm_nr += pols[1]->xfrm_nr;
1649                         }
1650                 }
1651
1652                 /*
1653                  * Because neither flowi nor bundle information knows about
1654                  * transformation template size. On more than one policy usage
1655                  * we can realize whether all of them is bypass or not after
1656                  * they are searched. See above not-transformed bypass
1657                  * is surrounded by non-sub policy configuration, too.
1658                  */
1659                 if (xfrm_nr == 0) {
1660                         /* Flow passes not transformed. */
1661                         xfrm_pols_put(pols, npols);
1662                         return 0;
1663                 }
1664
1665 #endif
1666                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1667
1668                 if (unlikely(nx<0)) {
1669                         err = nx;
1670                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1671                                 /* EREMOTE tells the caller to generate
1672                                  * a one-shot blackhole route.
1673                                  */
1674                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1675                                 xfrm_pol_put(policy);
1676                                 return -EREMOTE;
1677                         }
1678                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1679                                 DECLARE_WAITQUEUE(wait, current);
1680
1681                                 add_wait_queue(&init_net.xfrm.km_waitq, &wait);
1682                                 set_current_state(TASK_INTERRUPTIBLE);
1683                                 schedule();
1684                                 set_current_state(TASK_RUNNING);
1685                                 remove_wait_queue(&init_net.xfrm.km_waitq, &wait);
1686
1687                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1688
1689                                 if (nx == -EAGAIN && signal_pending(current)) {
1690                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1691                                         err = -ERESTART;
1692                                         goto error;
1693                                 }
1694                                 if (nx == -EAGAIN ||
1695                                     genid != atomic_read(&flow_cache_genid)) {
1696                                         xfrm_pols_put(pols, npols);
1697                                         goto restart;
1698                                 }
1699                                 err = nx;
1700                         }
1701                         if (err < 0) {
1702                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1703                                 goto error;
1704                         }
1705                 }
1706                 if (nx == 0) {
1707                         /* Flow passes not transformed. */
1708                         xfrm_pols_put(pols, npols);
1709                         return 0;
1710                 }
1711
1712                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1713                 err = PTR_ERR(dst);
1714                 if (IS_ERR(dst)) {
1715                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1716                         goto error;
1717                 }
1718
1719                 for (pi = 0; pi < npols; pi++) {
1720                         read_lock_bh(&pols[pi]->lock);
1721                         pol_dead |= pols[pi]->walk.dead;
1722                         read_unlock_bh(&pols[pi]->lock);
1723                 }
1724
1725                 write_lock_bh(&policy->lock);
1726                 if (unlikely(pol_dead || stale_bundle(dst))) {
1727                         /* Wow! While we worked on resolving, this
1728                          * policy has gone. Retry. It is not paranoia,
1729                          * we just cannot enlist new bundle to dead object.
1730                          * We can't enlist stable bundles either.
1731                          */
1732                         write_unlock_bh(&policy->lock);
1733                         dst_free(dst);
1734
1735                         if (pol_dead)
1736                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1737                         else
1738                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1739                         err = -EHOSTUNREACH;
1740                         goto error;
1741                 }
1742
1743                 if (npols > 1)
1744                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1745                 else
1746                         err = xfrm_dst_update_origin(dst, fl);
1747                 if (unlikely(err)) {
1748                         write_unlock_bh(&policy->lock);
1749                         dst_free(dst);
1750                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1751                         goto error;
1752                 }
1753
1754                 dst->next = policy->bundles;
1755                 policy->bundles = dst;
1756                 dst_hold(dst);
1757                 write_unlock_bh(&policy->lock);
1758         }
1759         *dst_p = dst;
1760         dst_release(dst_orig);
1761         xfrm_pols_put(pols, npols);
1762         return 0;
1763
1764 error:
1765         xfrm_pols_put(pols, npols);
1766 dropdst:
1767         dst_release(dst_orig);
1768         *dst_p = NULL;
1769         return err;
1770
1771 nopol:
1772         err = -ENOENT;
1773         if (flags & XFRM_LOOKUP_ICMP)
1774                 goto dropdst;
1775         return 0;
1776 }
1777 EXPORT_SYMBOL(__xfrm_lookup);
1778
1779 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1780                 struct sock *sk, int flags)
1781 {
1782         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1783
1784         if (err == -EREMOTE) {
1785                 dst_release(*dst_p);
1786                 *dst_p = NULL;
1787                 err = -EAGAIN;
1788         }
1789
1790         return err;
1791 }
1792 EXPORT_SYMBOL(xfrm_lookup);
1793
1794 static inline int
1795 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1796 {
1797         struct xfrm_state *x;
1798
1799         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1800                 return 0;
1801         x = skb->sp->xvec[idx];
1802         if (!x->type->reject)
1803                 return 0;
1804         return x->type->reject(x, skb, fl);
1805 }
1806
1807 /* When skb is transformed back to its "native" form, we have to
1808  * check policy restrictions. At the moment we make this in maximally
1809  * stupid way. Shame on me. :-) Of course, connected sockets must
1810  * have policy cached at them.
1811  */
1812
1813 static inline int
1814 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1815               unsigned short family)
1816 {
1817         if (xfrm_state_kern(x))
1818                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1819         return  x->id.proto == tmpl->id.proto &&
1820                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1821                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1822                 x->props.mode == tmpl->mode &&
1823                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1824                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1825                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1826                   xfrm_state_addr_cmp(tmpl, x, family));
1827 }
1828
1829 /*
1830  * 0 or more than 0 is returned when validation is succeeded (either bypass
1831  * because of optional transport mode, or next index of the mathced secpath
1832  * state with the template.
1833  * -1 is returned when no matching template is found.
1834  * Otherwise "-2 - errored_index" is returned.
1835  */
1836 static inline int
1837 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1838                unsigned short family)
1839 {
1840         int idx = start;
1841
1842         if (tmpl->optional) {
1843                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1844                         return start;
1845         } else
1846                 start = -1;
1847         for (; idx < sp->len; idx++) {
1848                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1849                         return ++idx;
1850                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1851                         if (start == -1)
1852                                 start = -2-idx;
1853                         break;
1854                 }
1855         }
1856         return start;
1857 }
1858
1859 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1860                           unsigned int family, int reverse)
1861 {
1862         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1863         int err;
1864
1865         if (unlikely(afinfo == NULL))
1866                 return -EAFNOSUPPORT;
1867
1868         afinfo->decode_session(skb, fl, reverse);
1869         err = security_xfrm_decode_session(skb, &fl->secid);
1870         xfrm_policy_put_afinfo(afinfo);
1871         return err;
1872 }
1873 EXPORT_SYMBOL(__xfrm_decode_session);
1874
1875 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1876 {
1877         for (; k < sp->len; k++) {
1878                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1879                         *idxp = k;
1880                         return 1;
1881                 }
1882         }
1883
1884         return 0;
1885 }
1886
1887 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1888                         unsigned short family)
1889 {
1890         struct xfrm_policy *pol;
1891         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1892         int npols = 0;
1893         int xfrm_nr;
1894         int pi;
1895         int reverse;
1896         struct flowi fl;
1897         u8 fl_dir;
1898         int xerr_idx = -1;
1899
1900         reverse = dir & ~XFRM_POLICY_MASK;
1901         dir &= XFRM_POLICY_MASK;
1902         fl_dir = policy_to_flow_dir(dir);
1903
1904         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1905                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1906                 return 0;
1907         }
1908
1909         nf_nat_decode_session(skb, &fl, family);
1910
1911         /* First, check used SA against their selectors. */
1912         if (skb->sp) {
1913                 int i;
1914
1915                 for (i=skb->sp->len-1; i>=0; i--) {
1916                         struct xfrm_state *x = skb->sp->xvec[i];
1917                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1918                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1919                                 return 0;
1920                         }
1921                 }
1922         }
1923
1924         pol = NULL;
1925         if (sk && sk->sk_policy[dir]) {
1926                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1927                 if (IS_ERR(pol)) {
1928                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1929                         return 0;
1930                 }
1931         }
1932
1933         if (!pol)
1934                 pol = flow_cache_lookup(&fl, family, fl_dir,
1935                                         xfrm_policy_lookup);
1936
1937         if (IS_ERR(pol)) {
1938                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1939                 return 0;
1940         }
1941
1942         if (!pol) {
1943                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1944                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1945                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1946                         return 0;
1947                 }
1948                 return 1;
1949         }
1950
1951         pol->curlft.use_time = get_seconds();
1952
1953         pols[0] = pol;
1954         npols ++;
1955 #ifdef CONFIG_XFRM_SUB_POLICY
1956         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1957                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1958                                                     &fl, family,
1959                                                     XFRM_POLICY_IN);
1960                 if (pols[1]) {
1961                         if (IS_ERR(pols[1])) {
1962                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1963                                 return 0;
1964                         }
1965                         pols[1]->curlft.use_time = get_seconds();
1966                         npols ++;
1967                 }
1968         }
1969 #endif
1970
1971         if (pol->action == XFRM_POLICY_ALLOW) {
1972                 struct sec_path *sp;
1973                 static struct sec_path dummy;
1974                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1975                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1976                 struct xfrm_tmpl **tpp = tp;
1977                 int ti = 0;
1978                 int i, k;
1979
1980                 if ((sp = skb->sp) == NULL)
1981                         sp = &dummy;
1982
1983                 for (pi = 0; pi < npols; pi++) {
1984                         if (pols[pi] != pol &&
1985                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1986                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1987                                 goto reject;
1988                         }
1989                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1990                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1991                                 goto reject_error;
1992                         }
1993                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1994                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1995                 }
1996                 xfrm_nr = ti;
1997                 if (npols > 1) {
1998                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1999                         tpp = stp;
2000                 }
2001
2002                 /* For each tunnel xfrm, find the first matching tmpl.
2003                  * For each tmpl before that, find corresponding xfrm.
2004                  * Order is _important_. Later we will implement
2005                  * some barriers, but at the moment barriers
2006                  * are implied between each two transformations.
2007                  */
2008                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2009                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2010                         if (k < 0) {
2011                                 if (k < -1)
2012                                         /* "-2 - errored_index" returned */
2013                                         xerr_idx = -(2+k);
2014                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2015                                 goto reject;
2016                         }
2017                 }
2018
2019                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2020                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2021                         goto reject;
2022                 }
2023
2024                 xfrm_pols_put(pols, npols);
2025                 return 1;
2026         }
2027         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2028
2029 reject:
2030         xfrm_secpath_reject(xerr_idx, skb, &fl);
2031 reject_error:
2032         xfrm_pols_put(pols, npols);
2033         return 0;
2034 }
2035 EXPORT_SYMBOL(__xfrm_policy_check);
2036
2037 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2038 {
2039         struct flowi fl;
2040
2041         if (xfrm_decode_session(skb, &fl, family) < 0) {
2042                 /* XXX: we should have something like FWDHDRERROR here. */
2043                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2044                 return 0;
2045         }
2046
2047         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2048 }
2049 EXPORT_SYMBOL(__xfrm_route_forward);
2050
2051 /* Optimize later using cookies and generation ids. */
2052
2053 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2054 {
2055         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2056          * to "-1" to force all XFRM destinations to get validated by
2057          * dst_ops->check on every use.  We do this because when a
2058          * normal route referenced by an XFRM dst is obsoleted we do
2059          * not go looking around for all parent referencing XFRM dsts
2060          * so that we can invalidate them.  It is just too much work.
2061          * Instead we make the checks here on every use.  For example:
2062          *
2063          *      XFRM dst A --> IPv4 dst X
2064          *
2065          * X is the "xdst->route" of A (X is also the "dst->path" of A
2066          * in this example).  If X is marked obsolete, "A" will not
2067          * notice.  That's what we are validating here via the
2068          * stale_bundle() check.
2069          *
2070          * When a policy's bundle is pruned, we dst_free() the XFRM
2071          * dst which causes it's ->obsolete field to be set to a
2072          * positive non-zero integer.  If an XFRM dst has been pruned
2073          * like this, we want to force a new route lookup.
2074          */
2075         if (dst->obsolete < 0 && !stale_bundle(dst))
2076                 return dst;
2077
2078         return NULL;
2079 }
2080
2081 static int stale_bundle(struct dst_entry *dst)
2082 {
2083         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2084 }
2085
2086 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2087 {
2088         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2089                 dst->dev = dev_net(dev)->loopback_dev;
2090                 dev_hold(dst->dev);
2091                 dev_put(dev);
2092         }
2093 }
2094 EXPORT_SYMBOL(xfrm_dst_ifdown);
2095
2096 static void xfrm_link_failure(struct sk_buff *skb)
2097 {
2098         /* Impossible. Such dst must be popped before reaches point of failure. */
2099         return;
2100 }
2101
2102 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2103 {
2104         if (dst) {
2105                 if (dst->obsolete) {
2106                         dst_release(dst);
2107                         dst = NULL;
2108                 }
2109         }
2110         return dst;
2111 }
2112
2113 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2114 {
2115         struct dst_entry *dst, **dstp;
2116
2117         write_lock(&pol->lock);
2118         dstp = &pol->bundles;
2119         while ((dst=*dstp) != NULL) {
2120                 if (func(dst)) {
2121                         *dstp = dst->next;
2122                         dst->next = *gc_list_p;
2123                         *gc_list_p = dst;
2124                 } else {
2125                         dstp = &dst->next;
2126                 }
2127         }
2128         write_unlock(&pol->lock);
2129 }
2130
2131 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2132 {
2133         struct dst_entry *gc_list = NULL;
2134         int dir;
2135
2136         read_lock_bh(&xfrm_policy_lock);
2137         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2138                 struct xfrm_policy *pol;
2139                 struct hlist_node *entry;
2140                 struct hlist_head *table;
2141                 int i;
2142
2143                 hlist_for_each_entry(pol, entry,
2144                                      &init_net.xfrm.policy_inexact[dir], bydst)
2145                         prune_one_bundle(pol, func, &gc_list);
2146
2147                 table = init_net.xfrm.policy_bydst[dir].table;
2148                 for (i = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2149                         hlist_for_each_entry(pol, entry, table + i, bydst)
2150                                 prune_one_bundle(pol, func, &gc_list);
2151                 }
2152         }
2153         read_unlock_bh(&xfrm_policy_lock);
2154
2155         while (gc_list) {
2156                 struct dst_entry *dst = gc_list;
2157                 gc_list = dst->next;
2158                 dst_free(dst);
2159         }
2160 }
2161
2162 static int unused_bundle(struct dst_entry *dst)
2163 {
2164         return !atomic_read(&dst->__refcnt);
2165 }
2166
2167 static void __xfrm_garbage_collect(void)
2168 {
2169         xfrm_prune_bundles(unused_bundle);
2170 }
2171
2172 static int xfrm_flush_bundles(void)
2173 {
2174         xfrm_prune_bundles(stale_bundle);
2175         return 0;
2176 }
2177
2178 static void xfrm_init_pmtu(struct dst_entry *dst)
2179 {
2180         do {
2181                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2182                 u32 pmtu, route_mtu_cached;
2183
2184                 pmtu = dst_mtu(dst->child);
2185                 xdst->child_mtu_cached = pmtu;
2186
2187                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2188
2189                 route_mtu_cached = dst_mtu(xdst->route);
2190                 xdst->route_mtu_cached = route_mtu_cached;
2191
2192                 if (pmtu > route_mtu_cached)
2193                         pmtu = route_mtu_cached;
2194
2195                 dst->metrics[RTAX_MTU-1] = pmtu;
2196         } while ((dst = dst->next));
2197 }
2198
2199 /* Check that the bundle accepts the flow and its components are
2200  * still valid.
2201  */
2202
2203 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2204                 struct flowi *fl, int family, int strict)
2205 {
2206         struct dst_entry *dst = &first->u.dst;
2207         struct xfrm_dst *last;
2208         u32 mtu;
2209
2210         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2211             (dst->dev && !netif_running(dst->dev)))
2212                 return 0;
2213 #ifdef CONFIG_XFRM_SUB_POLICY
2214         if (fl) {
2215                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2216                         return 0;
2217                 if (first->partner &&
2218                     !xfrm_selector_match(first->partner, fl, family))
2219                         return 0;
2220         }
2221 #endif
2222
2223         last = NULL;
2224
2225         do {
2226                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2227
2228                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2229                         return 0;
2230                 if (fl && pol &&
2231                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2232                         return 0;
2233                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2234                         return 0;
2235                 if (xdst->genid != dst->xfrm->genid)
2236                         return 0;
2237
2238                 if (strict && fl &&
2239                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2240                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2241                         return 0;
2242
2243                 mtu = dst_mtu(dst->child);
2244                 if (xdst->child_mtu_cached != mtu) {
2245                         last = xdst;
2246                         xdst->child_mtu_cached = mtu;
2247                 }
2248
2249                 if (!dst_check(xdst->route, xdst->route_cookie))
2250                         return 0;
2251                 mtu = dst_mtu(xdst->route);
2252                 if (xdst->route_mtu_cached != mtu) {
2253                         last = xdst;
2254                         xdst->route_mtu_cached = mtu;
2255                 }
2256
2257                 dst = dst->child;
2258         } while (dst->xfrm);
2259
2260         if (likely(!last))
2261                 return 1;
2262
2263         mtu = last->child_mtu_cached;
2264         for (;;) {
2265                 dst = &last->u.dst;
2266
2267                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2268                 if (mtu > last->route_mtu_cached)
2269                         mtu = last->route_mtu_cached;
2270                 dst->metrics[RTAX_MTU-1] = mtu;
2271
2272                 if (last == first)
2273                         break;
2274
2275                 last = (struct xfrm_dst *)last->u.dst.next;
2276                 last->child_mtu_cached = mtu;
2277         }
2278
2279         return 1;
2280 }
2281
2282 EXPORT_SYMBOL(xfrm_bundle_ok);
2283
2284 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2285 {
2286         int err = 0;
2287         if (unlikely(afinfo == NULL))
2288                 return -EINVAL;
2289         if (unlikely(afinfo->family >= NPROTO))
2290                 return -EAFNOSUPPORT;
2291         write_lock_bh(&xfrm_policy_afinfo_lock);
2292         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2293                 err = -ENOBUFS;
2294         else {
2295                 struct dst_ops *dst_ops = afinfo->dst_ops;
2296                 if (likely(dst_ops->kmem_cachep == NULL))
2297                         dst_ops->kmem_cachep = xfrm_dst_cache;
2298                 if (likely(dst_ops->check == NULL))
2299                         dst_ops->check = xfrm_dst_check;
2300                 if (likely(dst_ops->negative_advice == NULL))
2301                         dst_ops->negative_advice = xfrm_negative_advice;
2302                 if (likely(dst_ops->link_failure == NULL))
2303                         dst_ops->link_failure = xfrm_link_failure;
2304                 if (likely(afinfo->garbage_collect == NULL))
2305                         afinfo->garbage_collect = __xfrm_garbage_collect;
2306                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2307         }
2308         write_unlock_bh(&xfrm_policy_afinfo_lock);
2309         return err;
2310 }
2311 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2312
2313 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2314 {
2315         int err = 0;
2316         if (unlikely(afinfo == NULL))
2317                 return -EINVAL;
2318         if (unlikely(afinfo->family >= NPROTO))
2319                 return -EAFNOSUPPORT;
2320         write_lock_bh(&xfrm_policy_afinfo_lock);
2321         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2322                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2323                         err = -EINVAL;
2324                 else {
2325                         struct dst_ops *dst_ops = afinfo->dst_ops;
2326                         xfrm_policy_afinfo[afinfo->family] = NULL;
2327                         dst_ops->kmem_cachep = NULL;
2328                         dst_ops->check = NULL;
2329                         dst_ops->negative_advice = NULL;
2330                         dst_ops->link_failure = NULL;
2331                         afinfo->garbage_collect = NULL;
2332                 }
2333         }
2334         write_unlock_bh(&xfrm_policy_afinfo_lock);
2335         return err;
2336 }
2337 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2338
2339 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2340 {
2341         struct xfrm_policy_afinfo *afinfo;
2342         if (unlikely(family >= NPROTO))
2343                 return NULL;
2344         read_lock(&xfrm_policy_afinfo_lock);
2345         afinfo = xfrm_policy_afinfo[family];
2346         if (unlikely(!afinfo))
2347                 read_unlock(&xfrm_policy_afinfo_lock);
2348         return afinfo;
2349 }
2350
2351 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2352 {
2353         read_unlock(&xfrm_policy_afinfo_lock);
2354 }
2355
2356 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2357 {
2358         struct net_device *dev = ptr;
2359
2360         if (!net_eq(dev_net(dev), &init_net))
2361                 return NOTIFY_DONE;
2362
2363         switch (event) {
2364         case NETDEV_DOWN:
2365                 xfrm_flush_bundles();
2366         }
2367         return NOTIFY_DONE;
2368 }
2369
2370 static struct notifier_block xfrm_dev_notifier = {
2371         .notifier_call  = xfrm_dev_event,
2372 };
2373
2374 #ifdef CONFIG_XFRM_STATISTICS
2375 static int __init xfrm_statistics_init(void)
2376 {
2377         if (snmp_mib_init((void **)xfrm_statistics,
2378                           sizeof(struct linux_xfrm_mib)) < 0)
2379                 return -ENOMEM;
2380         return 0;
2381 }
2382 #endif
2383
2384 static int __net_init xfrm_policy_init(struct net *net)
2385 {
2386         unsigned int hmask, sz;
2387         int dir;
2388
2389         if (net_eq(net, &init_net))
2390                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2391                                            sizeof(struct xfrm_dst),
2392                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2393                                            NULL);
2394
2395         hmask = 8 - 1;
2396         sz = (hmask+1) * sizeof(struct hlist_head);
2397
2398         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2399         if (!net->xfrm.policy_byidx)
2400                 goto out_byidx;
2401         net->xfrm.policy_idx_hmask = hmask;
2402
2403         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2404                 struct xfrm_policy_hash *htab;
2405
2406                 net->xfrm.policy_count[dir] = 0;
2407                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2408
2409                 htab = &net->xfrm.policy_bydst[dir];
2410                 htab->table = xfrm_hash_alloc(sz);
2411                 if (!htab->table)
2412                         goto out_bydst;
2413                 htab->hmask = hmask;
2414         }
2415
2416         INIT_LIST_HEAD(&net->xfrm.policy_all);
2417         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2418         if (net_eq(net, &init_net))
2419                 register_netdevice_notifier(&xfrm_dev_notifier);
2420         return 0;
2421
2422 out_bydst:
2423         for (dir--; dir >= 0; dir--) {
2424                 struct xfrm_policy_hash *htab;
2425
2426                 htab = &net->xfrm.policy_bydst[dir];
2427                 xfrm_hash_free(htab->table, sz);
2428         }
2429         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2430 out_byidx:
2431         return -ENOMEM;
2432 }
2433
2434 static void xfrm_policy_fini(struct net *net)
2435 {
2436         unsigned int sz;
2437         int dir;
2438
2439         WARN_ON(!list_empty(&net->xfrm.policy_all));
2440
2441         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2442                 struct xfrm_policy_hash *htab;
2443
2444                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2445
2446                 htab = &net->xfrm.policy_bydst[dir];
2447                 sz = (htab->hmask + 1);
2448                 WARN_ON(!hlist_empty(htab->table));
2449                 xfrm_hash_free(htab->table, sz);
2450         }
2451
2452         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2453         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2454         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2455 }
2456
2457 static int __net_init xfrm_net_init(struct net *net)
2458 {
2459         int rv;
2460
2461         rv = xfrm_state_init(net);
2462         if (rv < 0)
2463                 goto out_state;
2464         rv = xfrm_policy_init(net);
2465         if (rv < 0)
2466                 goto out_policy;
2467         return 0;
2468
2469 out_policy:
2470         xfrm_state_fini(net);
2471 out_state:
2472         return rv;
2473 }
2474
2475 static void __net_exit xfrm_net_exit(struct net *net)
2476 {
2477         xfrm_policy_fini(net);
2478         xfrm_state_fini(net);
2479 }
2480
2481 static struct pernet_operations __net_initdata xfrm_net_ops = {
2482         .init = xfrm_net_init,
2483         .exit = xfrm_net_exit,
2484 };
2485
2486 void __init xfrm_init(void)
2487 {
2488         register_pernet_subsys(&xfrm_net_ops);
2489 #ifdef CONFIG_XFRM_STATISTICS
2490         xfrm_statistics_init();
2491 #endif
2492         xfrm_input_init();
2493 #ifdef CONFIG_XFRM_STATISTICS
2494         xfrm_proc_init();
2495 #endif
2496 }
2497
2498 #ifdef CONFIG_AUDITSYSCALL
2499 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2500                                          struct audit_buffer *audit_buf)
2501 {
2502         struct xfrm_sec_ctx *ctx = xp->security;
2503         struct xfrm_selector *sel = &xp->selector;
2504
2505         if (ctx)
2506                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2507                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2508
2509         switch(sel->family) {
2510         case AF_INET:
2511                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2512                 if (sel->prefixlen_s != 32)
2513                         audit_log_format(audit_buf, " src_prefixlen=%d",
2514                                          sel->prefixlen_s);
2515                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2516                 if (sel->prefixlen_d != 32)
2517                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2518                                          sel->prefixlen_d);
2519                 break;
2520         case AF_INET6:
2521                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2522                 if (sel->prefixlen_s != 128)
2523                         audit_log_format(audit_buf, " src_prefixlen=%d",
2524                                          sel->prefixlen_s);
2525                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2526                 if (sel->prefixlen_d != 128)
2527                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2528                                          sel->prefixlen_d);
2529                 break;
2530         }
2531 }
2532
2533 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2534                            uid_t auid, u32 sessionid, u32 secid)
2535 {
2536         struct audit_buffer *audit_buf;
2537
2538         audit_buf = xfrm_audit_start("SPD-add");
2539         if (audit_buf == NULL)
2540                 return;
2541         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2542         audit_log_format(audit_buf, " res=%u", result);
2543         xfrm_audit_common_policyinfo(xp, audit_buf);
2544         audit_log_end(audit_buf);
2545 }
2546 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2547
2548 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2549                               uid_t auid, u32 sessionid, u32 secid)
2550 {
2551         struct audit_buffer *audit_buf;
2552
2553         audit_buf = xfrm_audit_start("SPD-delete");
2554         if (audit_buf == NULL)
2555                 return;
2556         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2557         audit_log_format(audit_buf, " res=%u", result);
2558         xfrm_audit_common_policyinfo(xp, audit_buf);
2559         audit_log_end(audit_buf);
2560 }
2561 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2562 #endif
2563
2564 #ifdef CONFIG_XFRM_MIGRATE
2565 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2566                                        struct xfrm_selector *sel_tgt)
2567 {
2568         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2569                 if (sel_tgt->family == sel_cmp->family &&
2570                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2571                                   sel_cmp->family) == 0 &&
2572                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2573                                   sel_cmp->family) == 0 &&
2574                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2575                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2576                         return 1;
2577                 }
2578         } else {
2579                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2580                         return 1;
2581                 }
2582         }
2583         return 0;
2584 }
2585
2586 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2587                                                      u8 dir, u8 type)
2588 {
2589         struct xfrm_policy *pol, *ret = NULL;
2590         struct hlist_node *entry;
2591         struct hlist_head *chain;
2592         u32 priority = ~0U;
2593
2594         read_lock_bh(&xfrm_policy_lock);
2595         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2596         hlist_for_each_entry(pol, entry, chain, bydst) {
2597                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2598                     pol->type == type) {
2599                         ret = pol;
2600                         priority = ret->priority;
2601                         break;
2602                 }
2603         }
2604         chain = &init_net.xfrm.policy_inexact[dir];
2605         hlist_for_each_entry(pol, entry, chain, bydst) {
2606                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2607                     pol->type == type &&
2608                     pol->priority < priority) {
2609                         ret = pol;
2610                         break;
2611                 }
2612         }
2613
2614         if (ret)
2615                 xfrm_pol_hold(ret);
2616
2617         read_unlock_bh(&xfrm_policy_lock);
2618
2619         return ret;
2620 }
2621
2622 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2623 {
2624         int match = 0;
2625
2626         if (t->mode == m->mode && t->id.proto == m->proto &&
2627             (m->reqid == 0 || t->reqid == m->reqid)) {
2628                 switch (t->mode) {
2629                 case XFRM_MODE_TUNNEL:
2630                 case XFRM_MODE_BEET:
2631                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2632                                           m->old_family) == 0 &&
2633                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2634                                           m->old_family) == 0) {
2635                                 match = 1;
2636                         }
2637                         break;
2638                 case XFRM_MODE_TRANSPORT:
2639                         /* in case of transport mode, template does not store
2640                            any IP addresses, hence we just compare mode and
2641                            protocol */
2642                         match = 1;
2643                         break;
2644                 default:
2645                         break;
2646                 }
2647         }
2648         return match;
2649 }
2650
2651 /* update endpoint address(es) of template(s) */
2652 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2653                                struct xfrm_migrate *m, int num_migrate)
2654 {
2655         struct xfrm_migrate *mp;
2656         struct dst_entry *dst;
2657         int i, j, n = 0;
2658
2659         write_lock_bh(&pol->lock);
2660         if (unlikely(pol->walk.dead)) {
2661                 /* target policy has been deleted */
2662                 write_unlock_bh(&pol->lock);
2663                 return -ENOENT;
2664         }
2665
2666         for (i = 0; i < pol->xfrm_nr; i++) {
2667                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2668                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2669                                 continue;
2670                         n++;
2671                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2672                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2673                                 continue;
2674                         /* update endpoints */
2675                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2676                                sizeof(pol->xfrm_vec[i].id.daddr));
2677                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2678                                sizeof(pol->xfrm_vec[i].saddr));
2679                         pol->xfrm_vec[i].encap_family = mp->new_family;
2680                         /* flush bundles */
2681                         while ((dst = pol->bundles) != NULL) {
2682                                 pol->bundles = dst->next;
2683                                 dst_free(dst);
2684                         }
2685                 }
2686         }
2687
2688         write_unlock_bh(&pol->lock);
2689
2690         if (!n)
2691                 return -ENODATA;
2692
2693         return 0;
2694 }
2695
2696 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2697 {
2698         int i, j;
2699
2700         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2701                 return -EINVAL;
2702
2703         for (i = 0; i < num_migrate; i++) {
2704                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2705                                    m[i].old_family) == 0) &&
2706                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2707                                    m[i].old_family) == 0))
2708                         return -EINVAL;
2709                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2710                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2711                         return -EINVAL;
2712
2713                 /* check if there is any duplicated entry */
2714                 for (j = i + 1; j < num_migrate; j++) {
2715                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2716                                     sizeof(m[i].old_daddr)) &&
2717                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2718                                     sizeof(m[i].old_saddr)) &&
2719                             m[i].proto == m[j].proto &&
2720                             m[i].mode == m[j].mode &&
2721                             m[i].reqid == m[j].reqid &&
2722                             m[i].old_family == m[j].old_family)
2723                                 return -EINVAL;
2724                 }
2725         }
2726
2727         return 0;
2728 }
2729
2730 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2731                  struct xfrm_migrate *m, int num_migrate,
2732                  struct xfrm_kmaddress *k)
2733 {
2734         int i, err, nx_cur = 0, nx_new = 0;
2735         struct xfrm_policy *pol = NULL;
2736         struct xfrm_state *x, *xc;
2737         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2738         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2739         struct xfrm_migrate *mp;
2740
2741         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2742                 goto out;
2743
2744         /* Stage 1 - find policy */
2745         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2746                 err = -ENOENT;
2747                 goto out;
2748         }
2749
2750         /* Stage 2 - find and update state(s) */
2751         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2752                 if ((x = xfrm_migrate_state_find(mp))) {
2753                         x_cur[nx_cur] = x;
2754                         nx_cur++;
2755                         if ((xc = xfrm_state_migrate(x, mp))) {
2756                                 x_new[nx_new] = xc;
2757                                 nx_new++;
2758                         } else {
2759                                 err = -ENODATA;
2760                                 goto restore_state;
2761                         }
2762                 }
2763         }
2764
2765         /* Stage 3 - update policy */
2766         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2767                 goto restore_state;
2768
2769         /* Stage 4 - delete old state(s) */
2770         if (nx_cur) {
2771                 xfrm_states_put(x_cur, nx_cur);
2772                 xfrm_states_delete(x_cur, nx_cur);
2773         }
2774
2775         /* Stage 5 - announce */
2776         km_migrate(sel, dir, type, m, num_migrate, k);
2777
2778         xfrm_pol_put(pol);
2779
2780         return 0;
2781 out:
2782         return err;
2783
2784 restore_state:
2785         if (pol)
2786                 xfrm_pol_put(pol);
2787         if (nx_cur)
2788                 xfrm_states_put(x_cur, nx_cur);
2789         if (nx_new)
2790                 xfrm_states_delete(x_new, nx_new);
2791
2792         return err;
2793 }
2794 EXPORT_SYMBOL(xfrm_migrate);
2795 #endif