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