net/xfrm/xfrm_policy.c

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