net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/xfrm.h>
  30 #include <net/ip.h>
  31 #ifdef CONFIG_XFRM_STATISTICS
  32 #include <net/snmp.h>
  33 #endif
  34
  35 #include "xfrm_hash.h"
  36
  37 DEFINE_MUTEX(xfrm_cfg_mutex);
  38 EXPORT_SYMBOL(xfrm_cfg_mutex);
  39
  40 static DEFINE_RWLOCK(xfrm_policy_lock);
  41
  42 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  43 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  44
  45 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  46
  47 static HLIST_HEAD(xfrm_policy_gc_list);
  48 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  49
  50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  52 static void xfrm_init_pmtu(struct dst_entry *dst);
  53
  54 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  55                                                 int dir);
  56
  57 static inline int
  58 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  59 {
  60         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  61                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  62                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  63                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  64                 (fl->proto == sel->proto || !sel->proto) &&
  65                 (fl->oif == sel->ifindex || !sel->ifindex);
  66 }
  67
  68 static inline int
  69 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  70 {
  71         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  72                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  73                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  74                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  75                 (fl->proto == sel->proto || !sel->proto) &&
  76                 (fl->oif == sel->ifindex || !sel->ifindex);
  77 }
  78
  79 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  80                     unsigned short family)
  81 {
  82         switch (family) {
  83         case AF_INET:
  84                 return __xfrm4_selector_match(sel, fl);
  85         case AF_INET6:
  86                 return __xfrm6_selector_match(sel, fl);
  87         }
  88         return 0;
  89 }
  90
  91 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
  92                                                   xfrm_address_t *saddr,
  93                                                   xfrm_address_t *daddr,
  94                                                   int family)
  95 {
  96         struct xfrm_policy_afinfo *afinfo;
  97         struct dst_entry *dst;
  98
  99         afinfo = xfrm_policy_get_afinfo(family);
 100         if (unlikely(afinfo == NULL))
 101                 return ERR_PTR(-EAFNOSUPPORT);
 102
 103         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 104
 105         xfrm_policy_put_afinfo(afinfo);
 106
 107         return dst;
 108 }
 109
 110 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 111                                                 xfrm_address_t *prev_saddr,
 112                                                 xfrm_address_t *prev_daddr,
 113                                                 int family)
 114 {
 115         struct net *net = xs_net(x);
 116         xfrm_address_t *saddr = &x->props.saddr;
 117         xfrm_address_t *daddr = &x->id.daddr;
 118         struct dst_entry *dst;
 119
 120         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 121                 saddr = x->coaddr;
 122                 daddr = prev_daddr;
 123         }
 124         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 125                 saddr = prev_saddr;
 126                 daddr = x->coaddr;
 127         }
 128
 129         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 130
 131         if (!IS_ERR(dst)) {
 132                 if (prev_saddr != saddr)
 133                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 134                 if (prev_daddr != daddr)
 135                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 136         }
 137
 138         return dst;
 139 }
 140
 141 static inline unsigned long make_jiffies(long secs)
 142 {
 143         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 144                 return MAX_SCHEDULE_TIMEOUT-1;
 145         else
 146                 return secs*HZ;
 147 }
 148
 149 static void xfrm_policy_timer(unsigned long data)
 150 {
 151         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 152         unsigned long now = get_seconds();
 153         long next = LONG_MAX;
 154         int warn = 0;
 155         int dir;
 156
 157         read_lock(&xp->lock);
 158
 159         if (xp->walk.dead)
 160                 goto out;
 161
 162         dir = xfrm_policy_id2dir(xp->index);
 163
 164         if (xp->lft.hard_add_expires_seconds) {
 165                 long tmo = xp->lft.hard_add_expires_seconds +
 166                         xp->curlft.add_time - now;
 167                 if (tmo <= 0)
 168                         goto expired;
 169                 if (tmo < next)
 170                         next = tmo;
 171         }
 172         if (xp->lft.hard_use_expires_seconds) {
 173                 long tmo = xp->lft.hard_use_expires_seconds +
 174                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 175                 if (tmo <= 0)
 176                         goto expired;
 177                 if (tmo < next)
 178                         next = tmo;
 179         }
 180         if (xp->lft.soft_add_expires_seconds) {
 181                 long tmo = xp->lft.soft_add_expires_seconds +
 182                         xp->curlft.add_time - now;
 183                 if (tmo <= 0) {
 184                         warn = 1;
 185                         tmo = XFRM_KM_TIMEOUT;
 186                 }
 187                 if (tmo < next)
 188                         next = tmo;
 189         }
 190         if (xp->lft.soft_use_expires_seconds) {
 191                 long tmo = xp->lft.soft_use_expires_seconds +
 192                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 193                 if (tmo <= 0) {
 194                         warn = 1;
 195                         tmo = XFRM_KM_TIMEOUT;
 196                 }
 197                 if (tmo < next)
 198                         next = tmo;
 199         }
 200
 201         if (warn)
 202                 km_policy_expired(xp, dir, 0, 0);
 203         if (next != LONG_MAX &&
 204             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 205                 xfrm_pol_hold(xp);
 206
 207 out:
 208         read_unlock(&xp->lock);
 209         xfrm_pol_put(xp);
 210         return;
 211
 212 expired:
 213         read_unlock(&xp->lock);
 214         if (!xfrm_policy_delete(xp, dir))
 215                 km_policy_expired(xp, dir, 1, 0);
 216         xfrm_pol_put(xp);
 217 }
 218
 219
 220 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 221  * SPD calls.
 222  */
 223
 224 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 225 {
 226         struct xfrm_policy *policy;
 227
 228         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 229
 230         if (policy) {
 231                 write_pnet(&policy->xp_net, net);
 232                 INIT_LIST_HEAD(&policy->walk.all);
 233                 INIT_HLIST_NODE(&policy->bydst);
 234                 INIT_HLIST_NODE(&policy->byidx);
 235                 rwlock_init(&policy->lock);
 236                 atomic_set(&policy->refcnt, 1);
 237                 setup_timer(&policy->timer, xfrm_policy_timer,
 238                                 (unsigned long)policy);
 239         }
 240         return policy;
 241 }
 242 EXPORT_SYMBOL(xfrm_policy_alloc);
 243
 244 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 245
 246 void xfrm_policy_destroy(struct xfrm_policy *policy)
 247 {
 248         BUG_ON(!policy->walk.dead);
 249
 250         BUG_ON(policy->bundles);
 251
 252         if (del_timer(&policy->timer))
 253                 BUG();
 254
 255         security_xfrm_policy_free(policy->security);
 256         kfree(policy);
 257 }
 258 EXPORT_SYMBOL(xfrm_policy_destroy);
 259
 260 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 261 {
 262         struct dst_entry *dst;
 263
 264         while ((dst = policy->bundles) != NULL) {
 265                 policy->bundles = dst->next;
 266                 dst_free(dst);
 267         }
 268
 269         if (del_timer(&policy->timer))
 270                 atomic_dec(&policy->refcnt);
 271
 272         if (atomic_read(&policy->refcnt) > 1)
 273                 flow_cache_flush();
 274
 275         xfrm_pol_put(policy);
 276 }
 277
 278 static void xfrm_policy_gc_task(struct work_struct *work)
 279 {
 280         struct xfrm_policy *policy;
 281         struct hlist_node *entry, *tmp;
 282         struct hlist_head gc_list;
 283
 284         spin_lock_bh(&xfrm_policy_gc_lock);
 285         gc_list.first = xfrm_policy_gc_list.first;
 286         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 287         spin_unlock_bh(&xfrm_policy_gc_lock);
 288
 289         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 290                 xfrm_policy_gc_kill(policy);
 291 }
 292 static DECLARE_WORK(xfrm_policy_gc_work, xfrm_policy_gc_task);
 293
 294 /* Rule must be locked. Release descentant resources, announce
 295  * entry dead. The rule must be unlinked from lists to the moment.
 296  */
 297
 298 static void xfrm_policy_kill(struct xfrm_policy *policy)
 299 {
 300         int dead;
 301
 302         write_lock_bh(&policy->lock);
 303         dead = policy->walk.dead;
 304         policy->walk.dead = 1;
 305         write_unlock_bh(&policy->lock);
 306
 307         if (unlikely(dead)) {
 308                 WARN_ON(1);
 309                 return;
 310         }
 311
 312         spin_lock_bh(&xfrm_policy_gc_lock);
 313         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 314         spin_unlock_bh(&xfrm_policy_gc_lock);
 315
 316         schedule_work(&xfrm_policy_gc_work);
 317 }
 318
 319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 320
 321 static inline unsigned int idx_hash(struct net *net, u32 index)
 322 {
 323         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 324 }
 325
 326 static struct hlist_head *policy_hash_bysel(struct net *net, struct xfrm_selector *sel, unsigned short family, int dir)
 327 {
 328         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 329         unsigned int hash = __sel_hash(sel, family, hmask);
 330
 331         return (hash == hmask + 1 ?
 332                 &net->xfrm.policy_inexact[dir] :
 333                 net->xfrm.policy_bydst[dir].table + hash);
 334 }
 335
 336 static struct hlist_head *policy_hash_direct(struct net *net, xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 337 {
 338         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 339         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 340
 341         return net->xfrm.policy_bydst[dir].table + hash;
 342 }
 343
 344 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 345                                    struct hlist_head *ndsttable,
 346                                    unsigned int nhashmask)
 347 {
 348         struct hlist_node *entry, *tmp, *entry0 = NULL;
 349         struct xfrm_policy *pol;
 350         unsigned int h0 = 0;
 351
 352 redo:
 353         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 354                 unsigned int h;
 355
 356                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 357                                 pol->family, nhashmask);
 358                 if (!entry0) {
 359                         hlist_del(entry);
 360                         hlist_add_head(&pol->bydst, ndsttable+h);
 361                         h0 = h;
 362                 } else {
 363                         if (h != h0)
 364                                 continue;
 365                         hlist_del(entry);
 366                         hlist_add_after(entry0, &pol->bydst);
 367                 }
 368                 entry0 = entry;
 369         }
 370         if (!hlist_empty(list)) {
 371                 entry0 = NULL;
 372                 goto redo;
 373         }
 374 }
 375
 376 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 377                                    struct hlist_head *nidxtable,
 378                                    unsigned int nhashmask)
 379 {
 380         struct hlist_node *entry, *tmp;
 381         struct xfrm_policy *pol;
 382
 383         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 384                 unsigned int h;
 385
 386                 h = __idx_hash(pol->index, nhashmask);
 387                 hlist_add_head(&pol->byidx, nidxtable+h);
 388         }
 389 }
 390
 391 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 392 {
 393         return ((old_hmask + 1) << 1) - 1;
 394 }
 395
 396 static void xfrm_bydst_resize(struct net *net, int dir)
 397 {
 398         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 399         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 400         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 401         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 402         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 403         int i;
 404
 405         if (!ndst)
 406                 return;
 407
 408         write_lock_bh(&xfrm_policy_lock);
 409
 410         for (i = hmask; i >= 0; i--)
 411                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 412
 413         net->xfrm.policy_bydst[dir].table = ndst;
 414         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 415
 416         write_unlock_bh(&xfrm_policy_lock);
 417
 418         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 419 }
 420
 421 static void xfrm_byidx_resize(struct net *net, int total)
 422 {
 423         unsigned int hmask = net->xfrm.policy_idx_hmask;
 424         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 425         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 426         struct hlist_head *oidx = net->xfrm.policy_byidx;
 427         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 428         int i;
 429
 430         if (!nidx)
 431                 return;
 432
 433         write_lock_bh(&xfrm_policy_lock);
 434
 435         for (i = hmask; i >= 0; i--)
 436                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 437
 438         net->xfrm.policy_byidx = nidx;
 439         net->xfrm.policy_idx_hmask = nhashmask;
 440
 441         write_unlock_bh(&xfrm_policy_lock);
 442
 443         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 444 }
 445
 446 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 447 {
 448         unsigned int cnt = net->xfrm.policy_count[dir];
 449         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 450
 451         if (total)
 452                 *total += cnt;
 453
 454         if ((hmask + 1) < xfrm_policy_hashmax &&
 455             cnt > hmask)
 456                 return 1;
 457
 458         return 0;
 459 }
 460
 461 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 462 {
 463         unsigned int hmask = net->xfrm.policy_idx_hmask;
 464
 465         if ((hmask + 1) < xfrm_policy_hashmax &&
 466             total > hmask)
 467                 return 1;
 468
 469         return 0;
 470 }
 471
 472 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 473 {
 474         read_lock_bh(&xfrm_policy_lock);
 475         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 476         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 477         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 478         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 479         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 480         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 481         si->spdhcnt = net->xfrm.policy_idx_hmask;
 482         si->spdhmcnt = xfrm_policy_hashmax;
 483         read_unlock_bh(&xfrm_policy_lock);
 484 }
 485 EXPORT_SYMBOL(xfrm_spd_getinfo);
 486
 487 static DEFINE_MUTEX(hash_resize_mutex);
 488 static void xfrm_hash_resize(struct work_struct *work)
 489 {
 490         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 491         int dir, total;
 492
 493         mutex_lock(&hash_resize_mutex);
 494
 495         total = 0;
 496         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 497                 if (xfrm_bydst_should_resize(net, dir, &total))
 498                         xfrm_bydst_resize(net, dir);
 499         }
 500         if (xfrm_byidx_should_resize(net, total))
 501                 xfrm_byidx_resize(net, total);
 502
 503         mutex_unlock(&hash_resize_mutex);
 504 }
 505
 506 /* Generate new index... KAME seems to generate them ordered by cost
 507  * of an absolute inpredictability of ordering of rules. This will not pass. */
 508 static u32 xfrm_gen_index(struct net *net, int dir)
 509 {
 510         static u32 idx_generator;
 511
 512         for (;;) {
 513                 struct hlist_node *entry;
 514                 struct hlist_head *list;
 515                 struct xfrm_policy *p;
 516                 u32 idx;
 517                 int found;
 518
 519                 idx = (idx_generator | dir);
 520                 idx_generator += 8;
 521                 if (idx == 0)
 522                         idx = 8;
 523                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 524                 found = 0;
 525                 hlist_for_each_entry(p, entry, list, byidx) {
 526                         if (p->index == idx) {
 527                                 found = 1;
 528                                 break;
 529                         }
 530                 }
 531                 if (!found)
 532                         return idx;
 533         }
 534 }
 535
 536 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 537 {
 538         u32 *p1 = (u32 *) s1;
 539         u32 *p2 = (u32 *) s2;
 540         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 541         int i;
 542
 543         for (i = 0; i < len; i++) {
 544                 if (p1[i] != p2[i])
 545                         return 1;
 546         }
 547
 548         return 0;
 549 }
 550
 551 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 552 {
 553         struct net *net = xp_net(policy);
 554         struct xfrm_policy *pol;
 555         struct xfrm_policy *delpol;
 556         struct hlist_head *chain;
 557         struct hlist_node *entry, *newpos;
 558         struct dst_entry *gc_list;
 559         u32 mark = policy->mark.v & policy->mark.m;
 560
 561         write_lock_bh(&xfrm_policy_lock);
 562         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 563         delpol = NULL;
 564         newpos = NULL;
 565         hlist_for_each_entry(pol, entry, chain, bydst) {
 566                 if (pol->type == policy->type &&
 567                     !selector_cmp(&pol->selector, &policy->selector) &&
 568                     (mark & pol->mark.m) == pol->mark.v &&
 569                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 570                     !WARN_ON(delpol)) {
 571                         if (excl) {
 572                                 write_unlock_bh(&xfrm_policy_lock);
 573                                 return -EEXIST;
 574                         }
 575                         delpol = pol;
 576                         if (policy->priority > pol->priority)
 577                                 continue;
 578                 } else if (policy->priority >= pol->priority) {
 579                         newpos = &pol->bydst;
 580                         continue;
 581                 }
 582                 if (delpol)
 583                         break;
 584         }
 585         if (newpos)
 586                 hlist_add_after(newpos, &policy->bydst);
 587         else
 588                 hlist_add_head(&policy->bydst, chain);
 589         xfrm_pol_hold(policy);
 590         net->xfrm.policy_count[dir]++;
 591         atomic_inc(&flow_cache_genid);
 592         if (delpol)
 593                 __xfrm_policy_unlink(delpol, dir);
 594         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 595         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 596         policy->curlft.add_time = get_seconds();
 597         policy->curlft.use_time = 0;
 598         if (!mod_timer(&policy->timer, jiffies + HZ))
 599                 xfrm_pol_hold(policy);
 600         list_add(&policy->walk.all, &net->xfrm.policy_all);
 601         write_unlock_bh(&xfrm_policy_lock);
 602
 603         if (delpol)
 604                 xfrm_policy_kill(delpol);
 605         else if (xfrm_bydst_should_resize(net, dir, NULL))
 606                 schedule_work(&net->xfrm.policy_hash_work);
 607
 608         read_lock_bh(&xfrm_policy_lock);
 609         gc_list = NULL;
 610         entry = &policy->bydst;
 611         hlist_for_each_entry_continue(policy, entry, bydst) {
 612                 struct dst_entry *dst;
 613
 614                 write_lock(&policy->lock);
 615                 dst = policy->bundles;
 616                 if (dst) {
 617                         struct dst_entry *tail = dst;
 618                         while (tail->next)
 619                                 tail = tail->next;
 620                         tail->next = gc_list;
 621                         gc_list = dst;
 622
 623                         policy->bundles = NULL;
 624                 }
 625                 write_unlock(&policy->lock);
 626         }
 627         read_unlock_bh(&xfrm_policy_lock);
 628
 629         while (gc_list) {
 630                 struct dst_entry *dst = gc_list;
 631
 632                 gc_list = dst->next;
 633                 dst_free(dst);
 634         }
 635
 636         return 0;
 637 }
 638 EXPORT_SYMBOL(xfrm_policy_insert);
 639
 640 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 641                                           int dir, struct xfrm_selector *sel,
 642                                           struct xfrm_sec_ctx *ctx, int delete,
 643                                           int *err)
 644 {
 645         struct xfrm_policy *pol, *ret;
 646         struct hlist_head *chain;
 647         struct hlist_node *entry;
 648
 649         *err = 0;
 650         write_lock_bh(&xfrm_policy_lock);
 651         chain = policy_hash_bysel(net, sel, sel->family, dir);
 652         ret = NULL;
 653         hlist_for_each_entry(pol, entry, chain, bydst) {
 654                 if (pol->type == type &&
 655                     (mark & pol->mark.m) == pol->mark.v &&
 656                     !selector_cmp(sel, &pol->selector) &&
 657                     xfrm_sec_ctx_match(ctx, pol->security)) {
 658                         xfrm_pol_hold(pol);
 659                         if (delete) {
 660                                 *err = security_xfrm_policy_delete(
 661                                                                 pol->security);
 662                                 if (*err) {
 663                                         write_unlock_bh(&xfrm_policy_lock);
 664                                         return pol;
 665                                 }
 666                                 __xfrm_policy_unlink(pol, dir);
 667                         }
 668                         ret = pol;
 669                         break;
 670                 }
 671         }
 672         write_unlock_bh(&xfrm_policy_lock);
 673
 674         if (ret && delete) {
 675                 atomic_inc(&flow_cache_genid);
 676                 xfrm_policy_kill(ret);
 677         }
 678         return ret;
 679 }
 680 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 681
 682 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 683                                      int dir, u32 id, int delete, int *err)
 684 {
 685         struct xfrm_policy *pol, *ret;
 686         struct hlist_head *chain;
 687         struct hlist_node *entry;
 688
 689         *err = -ENOENT;
 690         if (xfrm_policy_id2dir(id) != dir)
 691                 return NULL;
 692
 693         *err = 0;
 694         write_lock_bh(&xfrm_policy_lock);
 695         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 696         ret = NULL;
 697         hlist_for_each_entry(pol, entry, chain, byidx) {
 698                 if (pol->type == type && pol->index == id &&
 699                     (mark & pol->mark.m) == pol->mark.v) {
 700                         xfrm_pol_hold(pol);
 701                         if (delete) {
 702                                 *err = security_xfrm_policy_delete(
 703                                                                 pol->security);
 704                                 if (*err) {
 705                                         write_unlock_bh(&xfrm_policy_lock);
 706                                         return pol;
 707                                 }
 708                                 __xfrm_policy_unlink(pol, dir);
 709                         }
 710                         ret = pol;
 711                         break;
 712                 }
 713         }
 714         write_unlock_bh(&xfrm_policy_lock);
 715
 716         if (ret && delete) {
 717                 atomic_inc(&flow_cache_genid);
 718                 xfrm_policy_kill(ret);
 719         }
 720         return ret;
 721 }
 722 EXPORT_SYMBOL(xfrm_policy_byid);
 723
 724 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 725 static inline int
 726 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 727 {
 728         int dir, err = 0;
 729
 730         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 731                 struct xfrm_policy *pol;
 732                 struct hlist_node *entry;
 733                 int i;
 734
 735                 hlist_for_each_entry(pol, entry,
 736                                      &net->xfrm.policy_inexact[dir], bydst) {
 737                         if (pol->type != type)
 738                                 continue;
 739                         err = security_xfrm_policy_delete(pol->security);
 740                         if (err) {
 741                                 xfrm_audit_policy_delete(pol, 0,
 742                                                          audit_info->loginuid,
 743                                                          audit_info->sessionid,
 744                                                          audit_info->secid);
 745                                 return err;
 746                         }
 747                 }
 748                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 749                         hlist_for_each_entry(pol, entry,
 750                                              net->xfrm.policy_bydst[dir].table + i,
 751                                              bydst) {
 752                                 if (pol->type != type)
 753                                         continue;
 754                                 err = security_xfrm_policy_delete(
 755                                                                 pol->security);
 756                                 if (err) {
 757                                         xfrm_audit_policy_delete(pol, 0,
 758                                                         audit_info->loginuid,
 759                                                         audit_info->sessionid,
 760                                                         audit_info->secid);
 761                                         return err;
 762                                 }
 763                         }
 764                 }
 765         }
 766         return err;
 767 }
 768 #else
 769 static inline int
 770 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 771 {
 772         return 0;
 773 }
 774 #endif
 775
 776 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 777 {
 778         int dir, err = 0, cnt = 0;
 779         struct xfrm_policy *dp;
 780
 781         write_lock_bh(&xfrm_policy_lock);
 782
 783         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 784         if (err)
 785                 goto out;
 786
 787         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 788                 struct xfrm_policy *pol;
 789                 struct hlist_node *entry;
 790                 int i;
 791
 792         again1:
 793                 hlist_for_each_entry(pol, entry,
 794                                      &net->xfrm.policy_inexact[dir], bydst) {
 795                         if (pol->type != type)
 796                                 continue;
 797                         dp = __xfrm_policy_unlink(pol, dir);
 798                         write_unlock_bh(&xfrm_policy_lock);
 799                         if (dp)
 800                                 cnt++;
 801
 802                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 803                                                  audit_info->sessionid,
 804                                                  audit_info->secid);
 805
 806                         xfrm_policy_kill(pol);
 807
 808                         write_lock_bh(&xfrm_policy_lock);
 809                         goto again1;
 810                 }
 811
 812                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 813         again2:
 814                         hlist_for_each_entry(pol, entry,
 815                                              net->xfrm.policy_bydst[dir].table + i,
 816                                              bydst) {
 817                                 if (pol->type != type)
 818                                         continue;
 819                                 dp = __xfrm_policy_unlink(pol, dir);
 820                                 write_unlock_bh(&xfrm_policy_lock);
 821                                 if (dp)
 822                                         cnt++;
 823
 824                                 xfrm_audit_policy_delete(pol, 1,
 825                                                          audit_info->loginuid,
 826                                                          audit_info->sessionid,
 827                                                          audit_info->secid);
 828                                 xfrm_policy_kill(pol);
 829
 830                                 write_lock_bh(&xfrm_policy_lock);
 831                                 goto again2;
 832                         }
 833                 }
 834
 835         }
 836         if (!cnt)
 837                 err = -ESRCH;
 838         atomic_inc(&flow_cache_genid);
 839 out:
 840         write_unlock_bh(&xfrm_policy_lock);
 841         return err;
 842 }
 843 EXPORT_SYMBOL(xfrm_policy_flush);
 844
 845 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 846                      int (*func)(struct xfrm_policy *, int, int, void*),
 847                      void *data)
 848 {
 849         struct xfrm_policy *pol;
 850         struct xfrm_policy_walk_entry *x;
 851         int error = 0;
 852
 853         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 854             walk->type != XFRM_POLICY_TYPE_ANY)
 855                 return -EINVAL;
 856
 857         if (list_empty(&walk->walk.all) && walk->seq != 0)
 858                 return 0;
 859
 860         write_lock_bh(&xfrm_policy_lock);
 861         if (list_empty(&walk->walk.all))
 862                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 863         else
 864                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 865         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 866                 if (x->dead)
 867                         continue;
 868                 pol = container_of(x, struct xfrm_policy, walk);
 869                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 870                     walk->type != pol->type)
 871                         continue;
 872                 error = func(pol, xfrm_policy_id2dir(pol->index),
 873                              walk->seq, data);
 874                 if (error) {
 875                         list_move_tail(&walk->walk.all, &x->all);
 876                         goto out;
 877                 }
 878                 walk->seq++;
 879         }
 880         if (walk->seq == 0) {
 881                 error = -ENOENT;
 882                 goto out;
 883         }
 884         list_del_init(&walk->walk.all);
 885 out:
 886         write_unlock_bh(&xfrm_policy_lock);
 887         return error;
 888 }
 889 EXPORT_SYMBOL(xfrm_policy_walk);
 890
 891 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 892 {
 893         INIT_LIST_HEAD(&walk->walk.all);
 894         walk->walk.dead = 1;
 895         walk->type = type;
 896         walk->seq = 0;
 897 }
 898 EXPORT_SYMBOL(xfrm_policy_walk_init);
 899
 900 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 901 {
 902         if (list_empty(&walk->walk.all))
 903                 return;
 904
 905         write_lock_bh(&xfrm_policy_lock);
 906         list_del(&walk->walk.all);
 907         write_unlock_bh(&xfrm_policy_lock);
 908 }
 909 EXPORT_SYMBOL(xfrm_policy_walk_done);
 910
 911 /*
 912  * Find policy to apply to this flow.
 913  *
 914  * Returns 0 if policy found, else an -errno.
 915  */
 916 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 917                              u8 type, u16 family, int dir)
 918 {
 919         struct xfrm_selector *sel = &pol->selector;
 920         int match, ret = -ESRCH;
 921
 922         if (pol->family != family ||
 923             (fl->mark & pol->mark.m) != pol->mark.v ||
 924             pol->type != type)
 925                 return ret;
 926
 927         match = xfrm_selector_match(sel, fl, family);
 928         if (match)
 929                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 930                                                   dir);
 931
 932         return ret;
 933 }
 934
 935 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 936                                                      struct flowi *fl,
 937                                                      u16 family, u8 dir)
 938 {
 939         int err;
 940         struct xfrm_policy *pol, *ret;
 941         xfrm_address_t *daddr, *saddr;
 942         struct hlist_node *entry;
 943         struct hlist_head *chain;
 944         u32 priority = ~0U;
 945
 946         daddr = xfrm_flowi_daddr(fl, family);
 947         saddr = xfrm_flowi_saddr(fl, family);
 948         if (unlikely(!daddr || !saddr))
 949                 return NULL;
 950
 951         read_lock_bh(&xfrm_policy_lock);
 952         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 953         ret = NULL;
 954         hlist_for_each_entry(pol, entry, chain, bydst) {
 955                 err = xfrm_policy_match(pol, fl, type, family, dir);
 956                 if (err) {
 957                         if (err == -ESRCH)
 958                                 continue;
 959                         else {
 960                                 ret = ERR_PTR(err);
 961                                 goto fail;
 962                         }
 963                 } else {
 964                         ret = pol;
 965                         priority = ret->priority;
 966                         break;
 967                 }
 968         }
 969         chain = &net->xfrm.policy_inexact[dir];
 970         hlist_for_each_entry(pol, entry, chain, bydst) {
 971                 err = xfrm_policy_match(pol, fl, type, family, dir);
 972                 if (err) {
 973                         if (err == -ESRCH)
 974                                 continue;
 975                         else {
 976                                 ret = ERR_PTR(err);
 977                                 goto fail;
 978                         }
 979                 } else if (pol->priority < priority) {
 980                         ret = pol;
 981                         break;
 982                 }
 983         }
 984         if (ret)
 985                 xfrm_pol_hold(ret);
 986 fail:
 987         read_unlock_bh(&xfrm_policy_lock);
 988
 989         return ret;
 990 }
 991
 992 static int xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
 993                               u8 dir, void **objp, atomic_t **obj_refp)
 994 {
 995         struct xfrm_policy *pol;
 996         int err = 0;
 997
 998 #ifdef CONFIG_XFRM_SUB_POLICY
 999         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1000         if (IS_ERR(pol)) {
1001                 err = PTR_ERR(pol);
1002                 pol = NULL;
1003         }
1004         if (pol || err)
1005                 goto end;
1006 #endif
1007         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1008         if (IS_ERR(pol)) {
1009                 err = PTR_ERR(pol);
1010                 pol = NULL;
1011         }
1012 #ifdef CONFIG_XFRM_SUB_POLICY
1013 end:
1014 #endif
1015         if ((*objp = (void *) pol) != NULL)
1016                 *obj_refp = &pol->refcnt;
1017         return err;
1018 }
1019
1020 static inline int policy_to_flow_dir(int dir)
1021 {
1022         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1023             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1024             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1025                 return dir;
1026         switch (dir) {
1027         default:
1028         case XFRM_POLICY_IN:
1029                 return FLOW_DIR_IN;
1030         case XFRM_POLICY_OUT:
1031                 return FLOW_DIR_OUT;
1032         case XFRM_POLICY_FWD:
1033                 return FLOW_DIR_FWD;
1034         }
1035 }
1036
1037 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1038 {
1039         struct xfrm_policy *pol;
1040
1041         read_lock_bh(&xfrm_policy_lock);
1042         if ((pol = sk->sk_policy[dir]) != NULL) {
1043                 int match = xfrm_selector_match(&pol->selector, fl,
1044                                                 sk->sk_family);
1045                 int err = 0;
1046
1047                 if (match) {
1048                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1049                                 pol = NULL;
1050                                 goto out;
1051                         }
1052                         err = security_xfrm_policy_lookup(pol->security,
1053                                                       fl->secid,
1054                                                       policy_to_flow_dir(dir));
1055                         if (!err)
1056                                 xfrm_pol_hold(pol);
1057                         else if (err == -ESRCH)
1058                                 pol = NULL;
1059                         else
1060                                 pol = ERR_PTR(err);
1061                 } else
1062                         pol = NULL;
1063         }
1064 out:
1065         read_unlock_bh(&xfrm_policy_lock);
1066         return pol;
1067 }
1068
1069 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1070 {
1071         struct net *net = xp_net(pol);
1072         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1073                                                      pol->family, dir);
1074
1075         list_add(&pol->walk.all, &net->xfrm.policy_all);
1076         hlist_add_head(&pol->bydst, chain);
1077         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1078         net->xfrm.policy_count[dir]++;
1079         xfrm_pol_hold(pol);
1080
1081         if (xfrm_bydst_should_resize(net, dir, NULL))
1082                 schedule_work(&net->xfrm.policy_hash_work);
1083 }
1084
1085 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1086                                                 int dir)
1087 {
1088         struct net *net = xp_net(pol);
1089
1090         if (hlist_unhashed(&pol->bydst))
1091                 return NULL;
1092
1093         hlist_del(&pol->bydst);
1094         hlist_del(&pol->byidx);
1095         list_del(&pol->walk.all);
1096         net->xfrm.policy_count[dir]--;
1097
1098         return pol;
1099 }
1100
1101 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1102 {
1103         write_lock_bh(&xfrm_policy_lock);
1104         pol = __xfrm_policy_unlink(pol, dir);
1105         write_unlock_bh(&xfrm_policy_lock);
1106         if (pol) {
1107                 if (dir < XFRM_POLICY_MAX)
1108                         atomic_inc(&flow_cache_genid);
1109                 xfrm_policy_kill(pol);
1110                 return 0;
1111         }
1112         return -ENOENT;
1113 }
1114 EXPORT_SYMBOL(xfrm_policy_delete);
1115
1116 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1117 {
1118         struct net *net = xp_net(pol);
1119         struct xfrm_policy *old_pol;
1120
1121 #ifdef CONFIG_XFRM_SUB_POLICY
1122         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1123                 return -EINVAL;
1124 #endif
1125
1126         write_lock_bh(&xfrm_policy_lock);
1127         old_pol = sk->sk_policy[dir];
1128         sk->sk_policy[dir] = pol;
1129         if (pol) {
1130                 pol->curlft.add_time = get_seconds();
1131                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1132                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1133         }
1134         if (old_pol)
1135                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1136         write_unlock_bh(&xfrm_policy_lock);
1137
1138         if (old_pol) {
1139                 xfrm_policy_kill(old_pol);
1140         }
1141         return 0;
1142 }
1143
1144 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1145 {
1146         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1147
1148         if (newp) {
1149                 newp->selector = old->selector;
1150                 if (security_xfrm_policy_clone(old->security,
1151                                                &newp->security)) {
1152                         kfree(newp);
1153                         return NULL;  /* ENOMEM */
1154                 }
1155                 newp->lft = old->lft;
1156                 newp->curlft = old->curlft;
1157                 newp->action = old->action;
1158                 newp->flags = old->flags;
1159                 newp->xfrm_nr = old->xfrm_nr;
1160                 newp->index = old->index;
1161                 newp->type = old->type;
1162                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1163                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1164                 write_lock_bh(&xfrm_policy_lock);
1165                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1166                 write_unlock_bh(&xfrm_policy_lock);
1167                 xfrm_pol_put(newp);
1168         }
1169         return newp;
1170 }
1171
1172 int __xfrm_sk_clone_policy(struct sock *sk)
1173 {
1174         struct xfrm_policy *p0 = sk->sk_policy[0],
1175                            *p1 = sk->sk_policy[1];
1176
1177         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1178         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1179                 return -ENOMEM;
1180         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1181                 return -ENOMEM;
1182         return 0;
1183 }
1184
1185 static int
1186 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1187                unsigned short family)
1188 {
1189         int err;
1190         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1191
1192         if (unlikely(afinfo == NULL))
1193                 return -EINVAL;
1194         err = afinfo->get_saddr(net, local, remote);
1195         xfrm_policy_put_afinfo(afinfo);
1196         return err;
1197 }
1198
1199 /* Resolve list of templates for the flow, given policy. */
1200
1201 static int
1202 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1203                       struct xfrm_state **xfrm,
1204                       unsigned short family)
1205 {
1206         struct net *net = xp_net(policy);
1207         int nx;
1208         int i, error;
1209         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1210         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1211         xfrm_address_t tmp;
1212
1213         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1214                 struct xfrm_state *x;
1215                 xfrm_address_t *remote = daddr;
1216                 xfrm_address_t *local  = saddr;
1217                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1218
1219                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1220                     tmpl->mode == XFRM_MODE_BEET) {
1221                         remote = &tmpl->id.daddr;
1222                         local = &tmpl->saddr;
1223                         family = tmpl->encap_family;
1224                         if (xfrm_addr_any(local, family)) {
1225                                 error = xfrm_get_saddr(net, &tmp, remote, family);
1226                                 if (error)
1227                                         goto fail;
1228                                 local = &tmp;
1229                         }
1230                 }
1231
1232                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1233
1234                 if (x && x->km.state == XFRM_STATE_VALID) {
1235                         xfrm[nx++] = x;
1236                         daddr = remote;
1237                         saddr = local;
1238                         continue;
1239                 }
1240                 if (x) {
1241                         error = (x->km.state == XFRM_STATE_ERROR ?
1242                                  -EINVAL : -EAGAIN);
1243                         xfrm_state_put(x);
1244                 }
1245                 else if (error == -ESRCH)
1246                         error = -EAGAIN;
1247
1248                 if (!tmpl->optional)
1249                         goto fail;
1250         }
1251         return nx;
1252
1253 fail:
1254         for (nx--; nx>=0; nx--)
1255                 xfrm_state_put(xfrm[nx]);
1256         return error;
1257 }
1258
1259 static int
1260 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1261                   struct xfrm_state **xfrm,
1262                   unsigned short family)
1263 {
1264         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1265         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1266         int cnx = 0;
1267         int error;
1268         int ret;
1269         int i;
1270
1271         for (i = 0; i < npols; i++) {
1272                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1273                         error = -ENOBUFS;
1274                         goto fail;
1275                 }
1276
1277                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1278                 if (ret < 0) {
1279                         error = ret;
1280                         goto fail;
1281                 } else
1282                         cnx += ret;
1283         }
1284
1285         /* found states are sorted for outbound processing */
1286         if (npols > 1)
1287                 xfrm_state_sort(xfrm, tpp, cnx, family);
1288
1289         return cnx;
1290
1291  fail:
1292         for (cnx--; cnx>=0; cnx--)
1293                 xfrm_state_put(tpp[cnx]);
1294         return error;
1295
1296 }
1297
1298 /* Check that the bundle accepts the flow and its components are
1299  * still valid.
1300  */
1301
1302 static struct dst_entry *
1303 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1304 {
1305         struct dst_entry *x;
1306         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1307         if (unlikely(afinfo == NULL))
1308                 return ERR_PTR(-EINVAL);
1309         x = afinfo->find_bundle(fl, policy);
1310         xfrm_policy_put_afinfo(afinfo);
1311         return x;
1312 }
1313
1314 static inline int xfrm_get_tos(struct flowi *fl, int family)
1315 {
1316         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1317         int tos;
1318
1319         if (!afinfo)
1320                 return -EINVAL;
1321
1322         tos = afinfo->get_tos(fl);
1323
1324         xfrm_policy_put_afinfo(afinfo);
1325
1326         return tos;
1327 }
1328
1329 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1330 {
1331         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1332         struct dst_ops *dst_ops;
1333         struct xfrm_dst *xdst;
1334
1335         if (!afinfo)
1336                 return ERR_PTR(-EINVAL);
1337
1338         switch (family) {
1339         case AF_INET:
1340                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1341                 break;
1342 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1343         case AF_INET6:
1344                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1345                 break;
1346 #endif
1347         default:
1348                 BUG();
1349         }
1350         xdst = dst_alloc(dst_ops) ?: ERR_PTR(-ENOBUFS);
1351
1352         xfrm_policy_put_afinfo(afinfo);
1353
1354         return xdst;
1355 }
1356
1357 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1358                                  int nfheader_len)
1359 {
1360         struct xfrm_policy_afinfo *afinfo =
1361                 xfrm_policy_get_afinfo(dst->ops->family);
1362         int err;
1363
1364         if (!afinfo)
1365                 return -EINVAL;
1366
1367         err = afinfo->init_path(path, dst, nfheader_len);
1368
1369         xfrm_policy_put_afinfo(afinfo);
1370
1371         return err;
1372 }
1373
1374 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1375 {
1376         struct xfrm_policy_afinfo *afinfo =
1377                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1378         int err;
1379
1380         if (!afinfo)
1381                 return -EINVAL;
1382
1383         err = afinfo->fill_dst(xdst, dev);
1384
1385         xfrm_policy_put_afinfo(afinfo);
1386
1387         return err;
1388 }
1389
1390 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1391  * all the metrics... Shortly, bundle a bundle.
1392  */
1393
1394 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1395                                             struct xfrm_state **xfrm, int nx,
1396                                             struct flowi *fl,
1397                                             struct dst_entry *dst)
1398 {
1399         struct net *net = xp_net(policy);
1400         unsigned long now = jiffies;
1401         struct net_device *dev;
1402         struct dst_entry *dst_prev = NULL;
1403         struct dst_entry *dst0 = NULL;
1404         int i = 0;
1405         int err;
1406         int header_len = 0;
1407         int nfheader_len = 0;
1408         int trailer_len = 0;
1409         int tos;
1410         int family = policy->selector.family;
1411         xfrm_address_t saddr, daddr;
1412
1413         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1414
1415         tos = xfrm_get_tos(fl, family);
1416         err = tos;
1417         if (tos < 0)
1418                 goto put_states;
1419
1420         dst_hold(dst);
1421
1422         for (; i < nx; i++) {
1423                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1424                 struct dst_entry *dst1 = &xdst->u.dst;
1425
1426                 err = PTR_ERR(xdst);
1427                 if (IS_ERR(xdst)) {
1428                         dst_release(dst);
1429                         goto put_states;
1430                 }
1431
1432                 if (!dst_prev)
1433                         dst0 = dst1;
1434                 else {
1435                         dst_prev->child = dst_clone(dst1);
1436                         dst1->flags |= DST_NOHASH;
1437                 }
1438
1439                 xdst->route = dst;
1440                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1441
1442                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1443                         family = xfrm[i]->props.family;
1444                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1445                                               family);
1446                         err = PTR_ERR(dst);
1447                         if (IS_ERR(dst))
1448                                 goto put_states;
1449                 } else
1450                         dst_hold(dst);
1451
1452                 dst1->xfrm = xfrm[i];
1453                 xdst->genid = xfrm[i]->genid;
1454
1455                 dst1->obsolete = -1;
1456                 dst1->flags |= DST_HOST;
1457                 dst1->lastuse = now;
1458
1459                 dst1->input = dst_discard;
1460                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1461
1462                 dst1->next = dst_prev;
1463                 dst_prev = dst1;
1464
1465                 header_len += xfrm[i]->props.header_len;
1466                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1467                         nfheader_len += xfrm[i]->props.header_len;
1468                 trailer_len += xfrm[i]->props.trailer_len;
1469         }
1470
1471         dst_prev->child = dst;
1472         dst0->path = dst;
1473
1474         err = -ENODEV;
1475         dev = dst->dev;
1476         if (!dev)
1477                 goto free_dst;
1478
1479         /* Copy neighbour for reachability confirmation */
1480         dst0->neighbour = neigh_clone(dst->neighbour);
1481
1482         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1483         xfrm_init_pmtu(dst_prev);
1484
1485         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1486                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1487
1488                 err = xfrm_fill_dst(xdst, dev);
1489                 if (err)
1490                         goto free_dst;
1491
1492                 dst_prev->header_len = header_len;
1493                 dst_prev->trailer_len = trailer_len;
1494                 header_len -= xdst->u.dst.xfrm->props.header_len;
1495                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1496         }
1497
1498 out:
1499         return dst0;
1500
1501 put_states:
1502         for (; i < nx; i++)
1503                 xfrm_state_put(xfrm[i]);
1504 free_dst:
1505         if (dst0)
1506                 dst_free(dst0);
1507         dst0 = ERR_PTR(err);
1508         goto out;
1509 }
1510
1511 static int inline
1512 xfrm_dst_alloc_copy(void **target, void *src, int size)
1513 {
1514         if (!*target) {
1515                 *target = kmalloc(size, GFP_ATOMIC);
1516                 if (!*target)
1517                         return -ENOMEM;
1518         }
1519         memcpy(*target, src, size);
1520         return 0;
1521 }
1522
1523 static int inline
1524 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1525 {
1526 #ifdef CONFIG_XFRM_SUB_POLICY
1527         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1528         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1529                                    sel, sizeof(*sel));
1530 #else
1531         return 0;
1532 #endif
1533 }
1534
1535 static int inline
1536 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1537 {
1538 #ifdef CONFIG_XFRM_SUB_POLICY
1539         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1540         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1541 #else
1542         return 0;
1543 #endif
1544 }
1545
1546 static int stale_bundle(struct dst_entry *dst);
1547
1548 /* Main function: finds/creates a bundle for given flow.
1549  *
1550  * At the moment we eat a raw IP route. Mostly to speed up lookups
1551  * on interfaces with disabled IPsec.
1552  */
1553 int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1554                   struct sock *sk, int flags)
1555 {
1556         struct xfrm_policy *policy;
1557         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1558         int npols;
1559         int pol_dead;
1560         int xfrm_nr;
1561         int pi;
1562         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1563         struct dst_entry *dst, *dst_orig = *dst_p;
1564         int nx = 0;
1565         int err;
1566         u32 genid;
1567         u16 family;
1568         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1569
1570 restart:
1571         genid = atomic_read(&flow_cache_genid);
1572         policy = NULL;
1573         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1574                 pols[pi] = NULL;
1575         npols = 0;
1576         pol_dead = 0;
1577         xfrm_nr = 0;
1578
1579         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1580                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1581                 err = PTR_ERR(policy);
1582                 if (IS_ERR(policy)) {
1583                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1584                         goto dropdst;
1585                 }
1586         }
1587
1588         if (!policy) {
1589                 /* To accelerate a bit...  */
1590                 if ((dst_orig->flags & DST_NOXFRM) ||
1591                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1592                         goto nopol;
1593
1594                 policy = flow_cache_lookup(net, fl, dst_orig->ops->family,
1595                                            dir, xfrm_policy_lookup);
1596                 err = PTR_ERR(policy);
1597                 if (IS_ERR(policy)) {
1598                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1599                         goto dropdst;
1600                 }
1601         }
1602
1603         if (!policy)
1604                 goto nopol;
1605
1606         family = dst_orig->ops->family;
1607         pols[0] = policy;
1608         npols ++;
1609         xfrm_nr += pols[0]->xfrm_nr;
1610
1611         err = -ENOENT;
1612         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1613                 goto error;
1614
1615         policy->curlft.use_time = get_seconds();
1616
1617         switch (policy->action) {
1618         default:
1619         case XFRM_POLICY_BLOCK:
1620                 /* Prohibit the flow */
1621                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1622                 err = -EPERM;
1623                 goto error;
1624
1625         case XFRM_POLICY_ALLOW:
1626 #ifndef CONFIG_XFRM_SUB_POLICY
1627                 if (policy->xfrm_nr == 0) {
1628                         /* Flow passes not transformed. */
1629                         xfrm_pol_put(policy);
1630                         return 0;
1631                 }
1632 #endif
1633
1634                 /* Try to find matching bundle.
1635                  *
1636                  * LATER: help from flow cache. It is optional, this
1637                  * is required only for output policy.
1638                  */
1639                 dst = xfrm_find_bundle(fl, policy, family);
1640                 if (IS_ERR(dst)) {
1641                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1642                         err = PTR_ERR(dst);
1643                         goto error;
1644                 }
1645
1646                 if (dst)
1647                         break;
1648
1649 #ifdef CONFIG_XFRM_SUB_POLICY
1650                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1651                         pols[1] = xfrm_policy_lookup_bytype(net,
1652                                                             XFRM_POLICY_TYPE_MAIN,
1653                                                             fl, family,
1654                                                             XFRM_POLICY_OUT);
1655                         if (pols[1]) {
1656                                 if (IS_ERR(pols[1])) {
1657                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1658                                         err = PTR_ERR(pols[1]);
1659                                         goto error;
1660                                 }
1661                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1662                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1663                                         err = -EPERM;
1664                                         goto error;
1665                                 }
1666                                 npols ++;
1667                                 xfrm_nr += pols[1]->xfrm_nr;
1668                         }
1669                 }
1670
1671                 /*
1672                  * Because neither flowi nor bundle information knows about
1673                  * transformation template size. On more than one policy usage
1674                  * we can realize whether all of them is bypass or not after
1675                  * they are searched. See above not-transformed bypass
1676                  * is surrounded by non-sub policy configuration, too.
1677                  */
1678                 if (xfrm_nr == 0) {
1679                         /* Flow passes not transformed. */
1680                         xfrm_pols_put(pols, npols);
1681                         return 0;
1682                 }
1683
1684 #endif
1685                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1686
1687                 if (unlikely(nx<0)) {
1688                         err = nx;
1689                         if (err == -EAGAIN && net->xfrm.sysctl_larval_drop) {
1690                                 /* EREMOTE tells the caller to generate
1691                                  * a one-shot blackhole route.
1692                                  */
1693                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1694                                 xfrm_pol_put(policy);
1695                                 return -EREMOTE;
1696                         }
1697                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1698                                 DECLARE_WAITQUEUE(wait, current);
1699
1700                                 add_wait_queue(&net->xfrm.km_waitq, &wait);
1701                                 set_current_state(TASK_INTERRUPTIBLE);
1702                                 schedule();
1703                                 set_current_state(TASK_RUNNING);
1704                                 remove_wait_queue(&net->xfrm.km_waitq, &wait);
1705
1706                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1707
1708                                 if (nx == -EAGAIN && signal_pending(current)) {
1709                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1710                                         err = -ERESTART;
1711                                         goto error;
1712                                 }
1713                                 if (nx == -EAGAIN ||
1714                                     genid != atomic_read(&flow_cache_genid)) {
1715                                         xfrm_pols_put(pols, npols);
1716                                         goto restart;
1717                                 }
1718                                 err = nx;
1719                         }
1720                         if (err < 0) {
1721                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1722                                 goto error;
1723                         }
1724                 }
1725                 if (nx == 0) {
1726                         /* Flow passes not transformed. */
1727                         xfrm_pols_put(pols, npols);
1728                         return 0;
1729                 }
1730
1731                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1732                 err = PTR_ERR(dst);
1733                 if (IS_ERR(dst)) {
1734                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1735                         goto error;
1736                 }
1737
1738                 for (pi = 0; pi < npols; pi++) {
1739                         read_lock_bh(&pols[pi]->lock);
1740                         pol_dead |= pols[pi]->walk.dead;
1741                         read_unlock_bh(&pols[pi]->lock);
1742                 }
1743
1744                 write_lock_bh(&policy->lock);
1745                 if (unlikely(pol_dead || stale_bundle(dst))) {
1746                         /* Wow! While we worked on resolving, this
1747                          * policy has gone. Retry. It is not paranoia,
1748                          * we just cannot enlist new bundle to dead object.
1749                          * We can't enlist stable bundles either.
1750                          */
1751                         write_unlock_bh(&policy->lock);
1752                         dst_free(dst);
1753
1754                         if (pol_dead)
1755                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLDEAD);
1756                         else
1757                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1758                         err = -EHOSTUNREACH;
1759                         goto error;
1760                 }
1761
1762                 if (npols > 1)
1763                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1764                 else
1765                         err = xfrm_dst_update_origin(dst, fl);
1766                 if (unlikely(err)) {
1767                         write_unlock_bh(&policy->lock);
1768                         dst_free(dst);
1769                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1770                         goto error;
1771                 }
1772
1773                 dst->next = policy->bundles;
1774                 policy->bundles = dst;
1775                 dst_hold(dst);
1776                 write_unlock_bh(&policy->lock);
1777         }
1778         *dst_p = dst;
1779         dst_release(dst_orig);
1780         xfrm_pols_put(pols, npols);
1781         return 0;
1782
1783 error:
1784         xfrm_pols_put(pols, npols);
1785 dropdst:
1786         dst_release(dst_orig);
1787         *dst_p = NULL;
1788         return err;
1789
1790 nopol:
1791         err = -ENOENT;
1792         if (flags & XFRM_LOOKUP_ICMP)
1793                 goto dropdst;
1794         return 0;
1795 }
1796 EXPORT_SYMBOL(__xfrm_lookup);
1797
1798 int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
1799                 struct sock *sk, int flags)
1800 {
1801         int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
1802
1803         if (err == -EREMOTE) {
1804                 dst_release(*dst_p);
1805                 *dst_p = NULL;
1806                 err = -EAGAIN;
1807         }
1808
1809         return err;
1810 }
1811 EXPORT_SYMBOL(xfrm_lookup);
1812
1813 static inline int
1814 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1815 {
1816         struct xfrm_state *x;
1817
1818         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1819                 return 0;
1820         x = skb->sp->xvec[idx];
1821         if (!x->type->reject)
1822                 return 0;
1823         return x->type->reject(x, skb, fl);
1824 }
1825
1826 /* When skb is transformed back to its "native" form, we have to
1827  * check policy restrictions. At the moment we make this in maximally
1828  * stupid way. Shame on me. :-) Of course, connected sockets must
1829  * have policy cached at them.
1830  */
1831
1832 static inline int
1833 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1834               unsigned short family)
1835 {
1836         if (xfrm_state_kern(x))
1837                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1838         return  x->id.proto == tmpl->id.proto &&
1839                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1840                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1841                 x->props.mode == tmpl->mode &&
1842                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1843                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1844                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1845                   xfrm_state_addr_cmp(tmpl, x, family));
1846 }
1847
1848 /*
1849  * 0 or more than 0 is returned when validation is succeeded (either bypass
1850  * because of optional transport mode, or next index of the mathced secpath
1851  * state with the template.
1852  * -1 is returned when no matching template is found.
1853  * Otherwise "-2 - errored_index" is returned.
1854  */
1855 static inline int
1856 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1857                unsigned short family)
1858 {
1859         int idx = start;
1860
1861         if (tmpl->optional) {
1862                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1863                         return start;
1864         } else
1865                 start = -1;
1866         for (; idx < sp->len; idx++) {
1867                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1868                         return ++idx;
1869                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1870                         if (start == -1)
1871                                 start = -2-idx;
1872                         break;
1873                 }
1874         }
1875         return start;
1876 }
1877
1878 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1879                           unsigned int family, int reverse)
1880 {
1881         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1882         int err;
1883
1884         if (unlikely(afinfo == NULL))
1885                 return -EAFNOSUPPORT;
1886
1887         afinfo->decode_session(skb, fl, reverse);
1888         err = security_xfrm_decode_session(skb, &fl->secid);
1889         xfrm_policy_put_afinfo(afinfo);
1890         return err;
1891 }
1892 EXPORT_SYMBOL(__xfrm_decode_session);
1893
1894 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1895 {
1896         for (; k < sp->len; k++) {
1897                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1898                         *idxp = k;
1899                         return 1;
1900                 }
1901         }
1902
1903         return 0;
1904 }
1905
1906 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1907                         unsigned short family)
1908 {
1909         struct net *net = dev_net(skb->dev);
1910         struct xfrm_policy *pol;
1911         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1912         int npols = 0;
1913         int xfrm_nr;
1914         int pi;
1915         int reverse;
1916         struct flowi fl;
1917         u8 fl_dir;
1918         int xerr_idx = -1;
1919
1920         reverse = dir & ~XFRM_POLICY_MASK;
1921         dir &= XFRM_POLICY_MASK;
1922         fl_dir = policy_to_flow_dir(dir);
1923
1924         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1925                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
1926                 return 0;
1927         }
1928
1929         nf_nat_decode_session(skb, &fl, family);
1930
1931         /* First, check used SA against their selectors. */
1932         if (skb->sp) {
1933                 int i;
1934
1935                 for (i=skb->sp->len-1; i>=0; i--) {
1936                         struct xfrm_state *x = skb->sp->xvec[i];
1937                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1938                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
1939                                 return 0;
1940                         }
1941                 }
1942         }
1943
1944         pol = NULL;
1945         if (sk && sk->sk_policy[dir]) {
1946                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1947                 if (IS_ERR(pol)) {
1948                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1949                         return 0;
1950                 }
1951         }
1952
1953         if (!pol)
1954                 pol = flow_cache_lookup(net, &fl, family, fl_dir,
1955                                         xfrm_policy_lookup);
1956
1957         if (IS_ERR(pol)) {
1958                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1959                 return 0;
1960         }
1961
1962         if (!pol) {
1963                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1964                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1965                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
1966                         return 0;
1967                 }
1968                 return 1;
1969         }
1970
1971         pol->curlft.use_time = get_seconds();
1972
1973         pols[0] = pol;
1974         npols ++;
1975 #ifdef CONFIG_XFRM_SUB_POLICY
1976         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1977                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
1978                                                     &fl, family,
1979                                                     XFRM_POLICY_IN);
1980                 if (pols[1]) {
1981                         if (IS_ERR(pols[1])) {
1982                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
1983                                 return 0;
1984                         }
1985                         pols[1]->curlft.use_time = get_seconds();
1986                         npols ++;
1987                 }
1988         }
1989 #endif
1990
1991         if (pol->action == XFRM_POLICY_ALLOW) {
1992                 struct sec_path *sp;
1993                 static struct sec_path dummy;
1994                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1995                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1996                 struct xfrm_tmpl **tpp = tp;
1997                 int ti = 0;
1998                 int i, k;
1999
2000                 if ((sp = skb->sp) == NULL)
2001                         sp = &dummy;
2002
2003                 for (pi = 0; pi < npols; pi++) {
2004                         if (pols[pi] != pol &&
2005                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2006                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2007                                 goto reject;
2008                         }
2009                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2010                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2011                                 goto reject_error;
2012                         }
2013                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2014                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2015                 }
2016                 xfrm_nr = ti;
2017                 if (npols > 1) {
2018                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2019                         tpp = stp;
2020                 }
2021
2022                 /* For each tunnel xfrm, find the first matching tmpl.
2023                  * For each tmpl before that, find corresponding xfrm.
2024                  * Order is _important_. Later we will implement
2025                  * some barriers, but at the moment barriers
2026                  * are implied between each two transformations.
2027                  */
2028                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2029                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2030                         if (k < 0) {
2031                                 if (k < -1)
2032                                         /* "-2 - errored_index" returned */
2033                                         xerr_idx = -(2+k);
2034                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2035                                 goto reject;
2036                         }
2037                 }
2038
2039                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2040                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2041                         goto reject;
2042                 }
2043
2044                 xfrm_pols_put(pols, npols);
2045                 return 1;
2046         }
2047         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2048
2049 reject:
2050         xfrm_secpath_reject(xerr_idx, skb, &fl);
2051 reject_error:
2052         xfrm_pols_put(pols, npols);
2053         return 0;
2054 }
2055 EXPORT_SYMBOL(__xfrm_policy_check);
2056
2057 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2058 {
2059         struct net *net = dev_net(skb->dev);
2060         struct flowi fl;
2061         struct dst_entry *dst;
2062         int res;
2063
2064         if (xfrm_decode_session(skb, &fl, family) < 0) {
2065                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2066                 return 0;
2067         }
2068
2069         dst = skb_dst(skb);
2070
2071         res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
2072         skb_dst_set(skb, dst);
2073         return res;
2074 }
2075 EXPORT_SYMBOL(__xfrm_route_forward);
2076
2077 /* Optimize later using cookies and generation ids. */
2078
2079 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2080 {
2081         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2082          * to "-1" to force all XFRM destinations to get validated by
2083          * dst_ops->check on every use.  We do this because when a
2084          * normal route referenced by an XFRM dst is obsoleted we do
2085          * not go looking around for all parent referencing XFRM dsts
2086          * so that we can invalidate them.  It is just too much work.
2087          * Instead we make the checks here on every use.  For example:
2088          *
2089          *      XFRM dst A --> IPv4 dst X
2090          *
2091          * X is the "xdst->route" of A (X is also the "dst->path" of A
2092          * in this example).  If X is marked obsolete, "A" will not
2093          * notice.  That's what we are validating here via the
2094          * stale_bundle() check.
2095          *
2096          * When a policy's bundle is pruned, we dst_free() the XFRM
2097          * dst which causes it's ->obsolete field to be set to a
2098          * positive non-zero integer.  If an XFRM dst has been pruned
2099          * like this, we want to force a new route lookup.
2100          */
2101         if (dst->obsolete < 0 && !stale_bundle(dst))
2102                 return dst;
2103
2104         return NULL;
2105 }
2106
2107 static int stale_bundle(struct dst_entry *dst)
2108 {
2109         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2110 }
2111
2112 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2113 {
2114         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2115                 dst->dev = dev_net(dev)->loopback_dev;
2116                 dev_hold(dst->dev);
2117                 dev_put(dev);
2118         }
2119 }
2120 EXPORT_SYMBOL(xfrm_dst_ifdown);
2121
2122 static void xfrm_link_failure(struct sk_buff *skb)
2123 {
2124         /* Impossible. Such dst must be popped before reaches point of failure. */
2125         return;
2126 }
2127
2128 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2129 {
2130         if (dst) {
2131                 if (dst->obsolete) {
2132                         dst_release(dst);
2133                         dst = NULL;
2134                 }
2135         }
2136         return dst;
2137 }
2138
2139 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2140 {
2141         struct dst_entry *dst, **dstp;
2142
2143         write_lock(&pol->lock);
2144         dstp = &pol->bundles;
2145         while ((dst=*dstp) != NULL) {
2146                 if (func(dst)) {
2147                         *dstp = dst->next;
2148                         dst->next = *gc_list_p;
2149                         *gc_list_p = dst;
2150                 } else {
2151                         dstp = &dst->next;
2152                 }
2153         }
2154         write_unlock(&pol->lock);
2155 }
2156
2157 static void xfrm_prune_bundles(struct net *net, int (*func)(struct dst_entry *))
2158 {
2159         struct dst_entry *gc_list = NULL;
2160         int dir;
2161
2162         read_lock_bh(&xfrm_policy_lock);
2163         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2164                 struct xfrm_policy *pol;
2165                 struct hlist_node *entry;
2166                 struct hlist_head *table;
2167                 int i;
2168
2169                 hlist_for_each_entry(pol, entry,
2170                                      &net->xfrm.policy_inexact[dir], bydst)
2171                         prune_one_bundle(pol, func, &gc_list);
2172
2173                 table = net->xfrm.policy_bydst[dir].table;
2174                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
2175                         hlist_for_each_entry(pol, entry, table + i, bydst)
2176                                 prune_one_bundle(pol, func, &gc_list);
2177                 }
2178         }
2179         read_unlock_bh(&xfrm_policy_lock);
2180
2181         while (gc_list) {
2182                 struct dst_entry *dst = gc_list;
2183                 gc_list = dst->next;
2184                 dst_free(dst);
2185         }
2186 }
2187
2188 static int unused_bundle(struct dst_entry *dst)
2189 {
2190         return !atomic_read(&dst->__refcnt);
2191 }
2192
2193 static void __xfrm_garbage_collect(struct net *net)
2194 {
2195         xfrm_prune_bundles(net, unused_bundle);
2196 }
2197
2198 static int xfrm_flush_bundles(struct net *net)
2199 {
2200         xfrm_prune_bundles(net, stale_bundle);
2201         return 0;
2202 }
2203
2204 static void xfrm_init_pmtu(struct dst_entry *dst)
2205 {
2206         do {
2207                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2208                 u32 pmtu, route_mtu_cached;
2209
2210                 pmtu = dst_mtu(dst->child);
2211                 xdst->child_mtu_cached = pmtu;
2212
2213                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2214
2215                 route_mtu_cached = dst_mtu(xdst->route);
2216                 xdst->route_mtu_cached = route_mtu_cached;
2217
2218                 if (pmtu > route_mtu_cached)
2219                         pmtu = route_mtu_cached;
2220
2221                 dst->metrics[RTAX_MTU-1] = pmtu;
2222         } while ((dst = dst->next));
2223 }
2224
2225 /* Check that the bundle accepts the flow and its components are
2226  * still valid.
2227  */
2228
2229 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2230                 struct flowi *fl, int family, int strict)
2231 {
2232         struct dst_entry *dst = &first->u.dst;
2233         struct xfrm_dst *last;
2234         u32 mtu;
2235
2236         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2237             (dst->dev && !netif_running(dst->dev)))
2238                 return 0;
2239 #ifdef CONFIG_XFRM_SUB_POLICY
2240         if (fl) {
2241                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2242                         return 0;
2243                 if (first->partner &&
2244                     !xfrm_selector_match(first->partner, fl, family))
2245                         return 0;
2246         }
2247 #endif
2248
2249         last = NULL;
2250
2251         do {
2252                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2253
2254                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2255                         return 0;
2256                 if (fl && pol &&
2257                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2258                         return 0;
2259                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2260                         return 0;
2261                 if (xdst->genid != dst->xfrm->genid)
2262                         return 0;
2263
2264                 if (strict && fl &&
2265                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2266                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2267                         return 0;
2268
2269                 mtu = dst_mtu(dst->child);
2270                 if (xdst->child_mtu_cached != mtu) {
2271                         last = xdst;
2272                         xdst->child_mtu_cached = mtu;
2273                 }
2274
2275                 if (!dst_check(xdst->route, xdst->route_cookie))
2276                         return 0;
2277                 mtu = dst_mtu(xdst->route);
2278                 if (xdst->route_mtu_cached != mtu) {
2279                         last = xdst;
2280                         xdst->route_mtu_cached = mtu;
2281                 }
2282
2283                 dst = dst->child;
2284         } while (dst->xfrm);
2285
2286         if (likely(!last))
2287                 return 1;
2288
2289         mtu = last->child_mtu_cached;
2290         for (;;) {
2291                 dst = &last->u.dst;
2292
2293                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2294                 if (mtu > last->route_mtu_cached)
2295                         mtu = last->route_mtu_cached;
2296                 dst->metrics[RTAX_MTU-1] = mtu;
2297
2298                 if (last == first)
2299                         break;
2300
2301                 last = (struct xfrm_dst *)last->u.dst.next;
2302                 last->child_mtu_cached = mtu;
2303         }
2304
2305         return 1;
2306 }
2307
2308 EXPORT_SYMBOL(xfrm_bundle_ok);
2309
2310 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2311 {
2312         struct net *net;
2313         int err = 0;
2314         if (unlikely(afinfo == NULL))
2315                 return -EINVAL;
2316         if (unlikely(afinfo->family >= NPROTO))
2317                 return -EAFNOSUPPORT;
2318         write_lock_bh(&xfrm_policy_afinfo_lock);
2319         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2320                 err = -ENOBUFS;
2321         else {
2322                 struct dst_ops *dst_ops = afinfo->dst_ops;
2323                 if (likely(dst_ops->kmem_cachep == NULL))
2324                         dst_ops->kmem_cachep = xfrm_dst_cache;
2325                 if (likely(dst_ops->check == NULL))
2326                         dst_ops->check = xfrm_dst_check;
2327                 if (likely(dst_ops->negative_advice == NULL))
2328                         dst_ops->negative_advice = xfrm_negative_advice;
2329                 if (likely(dst_ops->link_failure == NULL))
2330                         dst_ops->link_failure = xfrm_link_failure;
2331                 if (likely(afinfo->garbage_collect == NULL))
2332                         afinfo->garbage_collect = __xfrm_garbage_collect;
2333                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2334         }
2335         write_unlock_bh(&xfrm_policy_afinfo_lock);
2336
2337         rtnl_lock();
2338         for_each_net(net) {
2339                 struct dst_ops *xfrm_dst_ops;
2340
2341                 switch (afinfo->family) {
2342                 case AF_INET:
2343                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2344                         break;
2345 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2346                 case AF_INET6:
2347                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2348                         break;
2349 #endif
2350                 default:
2351                         BUG();
2352                 }
2353                 *xfrm_dst_ops = *afinfo->dst_ops;
2354         }
2355         rtnl_unlock();
2356
2357         return err;
2358 }
2359 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2360
2361 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2362 {
2363         int err = 0;
2364         if (unlikely(afinfo == NULL))
2365                 return -EINVAL;
2366         if (unlikely(afinfo->family >= NPROTO))
2367                 return -EAFNOSUPPORT;
2368         write_lock_bh(&xfrm_policy_afinfo_lock);
2369         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2370                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2371                         err = -EINVAL;
2372                 else {
2373                         struct dst_ops *dst_ops = afinfo->dst_ops;
2374                         xfrm_policy_afinfo[afinfo->family] = NULL;
2375                         dst_ops->kmem_cachep = NULL;
2376                         dst_ops->check = NULL;
2377                         dst_ops->negative_advice = NULL;
2378                         dst_ops->link_failure = NULL;
2379                         afinfo->garbage_collect = NULL;
2380                 }
2381         }
2382         write_unlock_bh(&xfrm_policy_afinfo_lock);
2383         return err;
2384 }
2385 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2386
2387 static void __net_init xfrm_dst_ops_init(struct net *net)
2388 {
2389         struct xfrm_policy_afinfo *afinfo;
2390
2391         read_lock_bh(&xfrm_policy_afinfo_lock);
2392         afinfo = xfrm_policy_afinfo[AF_INET];
2393         if (afinfo)
2394                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2395 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2396         afinfo = xfrm_policy_afinfo[AF_INET6];
2397         if (afinfo)
2398                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2399 #endif
2400         read_unlock_bh(&xfrm_policy_afinfo_lock);
2401 }
2402
2403 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2404 {
2405         struct xfrm_policy_afinfo *afinfo;
2406         if (unlikely(family >= NPROTO))
2407                 return NULL;
2408         read_lock(&xfrm_policy_afinfo_lock);
2409         afinfo = xfrm_policy_afinfo[family];
2410         if (unlikely(!afinfo))
2411                 read_unlock(&xfrm_policy_afinfo_lock);
2412         return afinfo;
2413 }
2414
2415 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2416 {
2417         read_unlock(&xfrm_policy_afinfo_lock);
2418 }
2419
2420 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2421 {
2422         struct net_device *dev = ptr;
2423
2424         switch (event) {
2425         case NETDEV_DOWN:
2426                 xfrm_flush_bundles(dev_net(dev));
2427         }
2428         return NOTIFY_DONE;
2429 }
2430
2431 static struct notifier_block xfrm_dev_notifier = {
2432         .notifier_call  = xfrm_dev_event,
2433 };
2434
2435 #ifdef CONFIG_XFRM_STATISTICS
2436 static int __net_init xfrm_statistics_init(struct net *net)
2437 {
2438         int rv;
2439
2440         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2441                           sizeof(struct linux_xfrm_mib)) < 0)
2442                 return -ENOMEM;
2443         rv = xfrm_proc_init(net);
2444         if (rv < 0)
2445                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2446         return rv;
2447 }
2448
2449 static void xfrm_statistics_fini(struct net *net)
2450 {
2451         xfrm_proc_fini(net);
2452         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2453 }
2454 #else
2455 static int __net_init xfrm_statistics_init(struct net *net)
2456 {
2457         return 0;
2458 }
2459
2460 static void xfrm_statistics_fini(struct net *net)
2461 {
2462 }
2463 #endif
2464
2465 static int __net_init xfrm_policy_init(struct net *net)
2466 {
2467         unsigned int hmask, sz;
2468         int dir;
2469
2470         if (net_eq(net, &init_net))
2471                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2472                                            sizeof(struct xfrm_dst),
2473                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2474                                            NULL);
2475
2476         hmask = 8 - 1;
2477         sz = (hmask+1) * sizeof(struct hlist_head);
2478
2479         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2480         if (!net->xfrm.policy_byidx)
2481                 goto out_byidx;
2482         net->xfrm.policy_idx_hmask = hmask;
2483
2484         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2485                 struct xfrm_policy_hash *htab;
2486
2487                 net->xfrm.policy_count[dir] = 0;
2488                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2489
2490                 htab = &net->xfrm.policy_bydst[dir];
2491                 htab->table = xfrm_hash_alloc(sz);
2492                 if (!htab->table)
2493                         goto out_bydst;
2494                 htab->hmask = hmask;
2495         }
2496
2497         INIT_LIST_HEAD(&net->xfrm.policy_all);
2498         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2499         if (net_eq(net, &init_net))
2500                 register_netdevice_notifier(&xfrm_dev_notifier);
2501         return 0;
2502
2503 out_bydst:
2504         for (dir--; dir >= 0; dir--) {
2505                 struct xfrm_policy_hash *htab;
2506
2507                 htab = &net->xfrm.policy_bydst[dir];
2508                 xfrm_hash_free(htab->table, sz);
2509         }
2510         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2511 out_byidx:
2512         return -ENOMEM;
2513 }
2514
2515 static void xfrm_policy_fini(struct net *net)
2516 {
2517         struct xfrm_audit audit_info;
2518         unsigned int sz;
2519         int dir;
2520
2521         flush_work(&net->xfrm.policy_hash_work);
2522 #ifdef CONFIG_XFRM_SUB_POLICY
2523         audit_info.loginuid = -1;
2524         audit_info.sessionid = -1;
2525         audit_info.secid = 0;
2526         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2527 #endif
2528         audit_info.loginuid = -1;
2529         audit_info.sessionid = -1;
2530         audit_info.secid = 0;
2531         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2532         flush_work(&xfrm_policy_gc_work);
2533
2534         WARN_ON(!list_empty(&net->xfrm.policy_all));
2535
2536         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2537                 struct xfrm_policy_hash *htab;
2538
2539                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2540
2541                 htab = &net->xfrm.policy_bydst[dir];
2542                 sz = (htab->hmask + 1);
2543                 WARN_ON(!hlist_empty(htab->table));
2544                 xfrm_hash_free(htab->table, sz);
2545         }
2546
2547         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2548         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2549         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2550 }
2551
2552 static int __net_init xfrm_net_init(struct net *net)
2553 {
2554         int rv;
2555
2556         rv = xfrm_statistics_init(net);
2557         if (rv < 0)
2558                 goto out_statistics;
2559         rv = xfrm_state_init(net);
2560         if (rv < 0)
2561                 goto out_state;
2562         rv = xfrm_policy_init(net);
2563         if (rv < 0)
2564                 goto out_policy;
2565         xfrm_dst_ops_init(net);
2566         rv = xfrm_sysctl_init(net);
2567         if (rv < 0)
2568                 goto out_sysctl;
2569         return 0;
2570
2571 out_sysctl:
2572         xfrm_policy_fini(net);
2573 out_policy:
2574         xfrm_state_fini(net);
2575 out_state:
2576         xfrm_statistics_fini(net);
2577 out_statistics:
2578         return rv;
2579 }
2580
2581 static void __net_exit xfrm_net_exit(struct net *net)
2582 {
2583         xfrm_sysctl_fini(net);
2584         xfrm_policy_fini(net);
2585         xfrm_state_fini(net);
2586         xfrm_statistics_fini(net);
2587 }
2588
2589 static struct pernet_operations __net_initdata xfrm_net_ops = {
2590         .init = xfrm_net_init,
2591         .exit = xfrm_net_exit,
2592 };
2593
2594 void __init xfrm_init(void)
2595 {
2596         register_pernet_subsys(&xfrm_net_ops);
2597         xfrm_input_init();
2598 }
2599
2600 #ifdef CONFIG_AUDITSYSCALL
2601 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2602                                          struct audit_buffer *audit_buf)
2603 {
2604         struct xfrm_sec_ctx *ctx = xp->security;
2605         struct xfrm_selector *sel = &xp->selector;
2606
2607         if (ctx)
2608                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2609                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2610
2611         switch(sel->family) {
2612         case AF_INET:
2613                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2614                 if (sel->prefixlen_s != 32)
2615                         audit_log_format(audit_buf, " src_prefixlen=%d",
2616                                          sel->prefixlen_s);
2617                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2618                 if (sel->prefixlen_d != 32)
2619                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2620                                          sel->prefixlen_d);
2621                 break;
2622         case AF_INET6:
2623                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2624                 if (sel->prefixlen_s != 128)
2625                         audit_log_format(audit_buf, " src_prefixlen=%d",
2626                                          sel->prefixlen_s);
2627                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2628                 if (sel->prefixlen_d != 128)
2629                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2630                                          sel->prefixlen_d);
2631                 break;
2632         }
2633 }
2634
2635 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2636                            uid_t auid, u32 sessionid, u32 secid)
2637 {
2638         struct audit_buffer *audit_buf;
2639
2640         audit_buf = xfrm_audit_start("SPD-add");
2641         if (audit_buf == NULL)
2642                 return;
2643         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2644         audit_log_format(audit_buf, " res=%u", result);
2645         xfrm_audit_common_policyinfo(xp, audit_buf);
2646         audit_log_end(audit_buf);
2647 }
2648 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2649
2650 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2651                               uid_t auid, u32 sessionid, u32 secid)
2652 {
2653         struct audit_buffer *audit_buf;
2654
2655         audit_buf = xfrm_audit_start("SPD-delete");
2656         if (audit_buf == NULL)
2657                 return;
2658         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2659         audit_log_format(audit_buf, " res=%u", result);
2660         xfrm_audit_common_policyinfo(xp, audit_buf);
2661         audit_log_end(audit_buf);
2662 }
2663 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2664 #endif
2665
2666 #ifdef CONFIG_XFRM_MIGRATE
2667 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2668                                        struct xfrm_selector *sel_tgt)
2669 {
2670         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2671                 if (sel_tgt->family == sel_cmp->family &&
2672                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2673                                   sel_cmp->family) == 0 &&
2674                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2675                                   sel_cmp->family) == 0 &&
2676                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2677                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2678                         return 1;
2679                 }
2680         } else {
2681                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2682                         return 1;
2683                 }
2684         }
2685         return 0;
2686 }
2687
2688 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2689                                                      u8 dir, u8 type)
2690 {
2691         struct xfrm_policy *pol, *ret = NULL;
2692         struct hlist_node *entry;
2693         struct hlist_head *chain;
2694         u32 priority = ~0U;
2695
2696         read_lock_bh(&xfrm_policy_lock);
2697         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2698         hlist_for_each_entry(pol, entry, chain, bydst) {
2699                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2700                     pol->type == type) {
2701                         ret = pol;
2702                         priority = ret->priority;
2703                         break;
2704                 }
2705         }
2706         chain = &init_net.xfrm.policy_inexact[dir];
2707         hlist_for_each_entry(pol, entry, chain, bydst) {
2708                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2709                     pol->type == type &&
2710                     pol->priority < priority) {
2711                         ret = pol;
2712                         break;
2713                 }
2714         }
2715
2716         if (ret)
2717                 xfrm_pol_hold(ret);
2718
2719         read_unlock_bh(&xfrm_policy_lock);
2720
2721         return ret;
2722 }
2723
2724 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2725 {
2726         int match = 0;
2727
2728         if (t->mode == m->mode && t->id.proto == m->proto &&
2729             (m->reqid == 0 || t->reqid == m->reqid)) {
2730                 switch (t->mode) {
2731                 case XFRM_MODE_TUNNEL:
2732                 case XFRM_MODE_BEET:
2733                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2734                                           m->old_family) == 0 &&
2735                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2736                                           m->old_family) == 0) {
2737                                 match = 1;
2738                         }
2739                         break;
2740                 case XFRM_MODE_TRANSPORT:
2741                         /* in case of transport mode, template does not store
2742                            any IP addresses, hence we just compare mode and
2743                            protocol */
2744                         match = 1;
2745                         break;
2746                 default:
2747                         break;
2748                 }
2749         }
2750         return match;
2751 }
2752
2753 /* update endpoint address(es) of template(s) */
2754 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2755                                struct xfrm_migrate *m, int num_migrate)
2756 {
2757         struct xfrm_migrate *mp;
2758         struct dst_entry *dst;
2759         int i, j, n = 0;
2760
2761         write_lock_bh(&pol->lock);
2762         if (unlikely(pol->walk.dead)) {
2763                 /* target policy has been deleted */
2764                 write_unlock_bh(&pol->lock);
2765                 return -ENOENT;
2766         }
2767
2768         for (i = 0; i < pol->xfrm_nr; i++) {
2769                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2770                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2771                                 continue;
2772                         n++;
2773                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2774                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2775                                 continue;
2776                         /* update endpoints */
2777                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2778                                sizeof(pol->xfrm_vec[i].id.daddr));
2779                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2780                                sizeof(pol->xfrm_vec[i].saddr));
2781                         pol->xfrm_vec[i].encap_family = mp->new_family;
2782                         /* flush bundles */
2783                         while ((dst = pol->bundles) != NULL) {
2784                                 pol->bundles = dst->next;
2785                                 dst_free(dst);
2786                         }
2787                 }
2788         }
2789
2790         write_unlock_bh(&pol->lock);
2791
2792         if (!n)
2793                 return -ENODATA;
2794
2795         return 0;
2796 }
2797
2798 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2799 {
2800         int i, j;
2801
2802         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2803                 return -EINVAL;
2804
2805         for (i = 0; i < num_migrate; i++) {
2806                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2807                                    m[i].old_family) == 0) &&
2808                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2809                                    m[i].old_family) == 0))
2810                         return -EINVAL;
2811                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2812                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2813                         return -EINVAL;
2814
2815                 /* check if there is any duplicated entry */
2816                 for (j = i + 1; j < num_migrate; j++) {
2817                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2818                                     sizeof(m[i].old_daddr)) &&
2819                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2820                                     sizeof(m[i].old_saddr)) &&
2821                             m[i].proto == m[j].proto &&
2822                             m[i].mode == m[j].mode &&
2823                             m[i].reqid == m[j].reqid &&
2824                             m[i].old_family == m[j].old_family)
2825                                 return -EINVAL;
2826                 }
2827         }
2828
2829         return 0;
2830 }
2831
2832 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2833                  struct xfrm_migrate *m, int num_migrate,
2834                  struct xfrm_kmaddress *k)
2835 {
2836         int i, err, nx_cur = 0, nx_new = 0;
2837         struct xfrm_policy *pol = NULL;
2838         struct xfrm_state *x, *xc;
2839         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2840         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2841         struct xfrm_migrate *mp;
2842
2843         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2844                 goto out;
2845
2846         /* Stage 1 - find policy */
2847         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2848                 err = -ENOENT;
2849                 goto out;
2850         }
2851
2852         /* Stage 2 - find and update state(s) */
2853         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2854                 if ((x = xfrm_migrate_state_find(mp))) {
2855                         x_cur[nx_cur] = x;
2856                         nx_cur++;
2857                         if ((xc = xfrm_state_migrate(x, mp))) {
2858                                 x_new[nx_new] = xc;
2859                                 nx_new++;
2860                         } else {
2861                                 err = -ENODATA;
2862                                 goto restore_state;
2863                         }
2864                 }
2865         }
2866
2867         /* Stage 3 - update policy */
2868         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2869                 goto restore_state;
2870
2871         /* Stage 4 - delete old state(s) */
2872         if (nx_cur) {
2873                 xfrm_states_put(x_cur, nx_cur);
2874                 xfrm_states_delete(x_cur, nx_cur);
2875         }
2876
2877         /* Stage 5 - announce */
2878         km_migrate(sel, dir, type, m, num_migrate, k);
2879
2880         xfrm_pol_put(pol);
2881
2882         return 0;
2883 out:
2884         return err;
2885
2886 restore_state:
2887         if (pol)
2888                 xfrm_pol_put(pol);
2889         if (nx_cur)
2890                 xfrm_states_put(x_cur, nx_cur);
2891         if (nx_new)
2892                 xfrm_states_delete(x_new, nx_new);
2893
2894         return err;
2895 }
2896 EXPORT_SYMBOL(xfrm_migrate);
2897 #endif