SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 <net/dst.h>
  28 #include <net/xfrm.h>
  29 #include <net/ip.h>
  30
  31 #include "xfrm_hash.h"
  32
  33 int sysctl_xfrm_larval_drop __read_mostly;
  34
  35 DEFINE_MUTEX(xfrm_cfg_mutex);
  36 EXPORT_SYMBOL(xfrm_cfg_mutex);
  37
  38 static DEFINE_RWLOCK(xfrm_policy_lock);
  39
  40 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  41 EXPORT_SYMBOL(xfrm_policy_count);
  42
  43 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  44 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  45
  46 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  47
  48 static struct work_struct xfrm_policy_gc_work;
  49 static HLIST_HEAD(xfrm_policy_gc_list);
  50 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  51
  52 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  53 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  54 static void xfrm_init_pmtu(struct dst_entry *dst);
  55
  56 static inline int
  57 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  58 {
  59         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  60                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  61                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  62                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  63                 (fl->proto == sel->proto || !sel->proto) &&
  64                 (fl->oif == sel->ifindex || !sel->ifindex);
  65 }
  66
  67 static inline int
  68 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  69 {
  70         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  71                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  72                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  73                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  74                 (fl->proto == sel->proto || !sel->proto) &&
  75                 (fl->oif == sel->ifindex || !sel->ifindex);
  76 }
  77
  78 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  79                     unsigned short family)
  80 {
  81         switch (family) {
  82         case AF_INET:
  83                 return __xfrm4_selector_match(sel, fl);
  84         case AF_INET6:
  85                 return __xfrm6_selector_match(sel, fl);
  86         }
  87         return 0;
  88 }
  89
  90 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
  91                                                 int family)
  92 {
  93         xfrm_address_t *saddr = &x->props.saddr;
  94         xfrm_address_t *daddr = &x->id.daddr;
  95         struct xfrm_policy_afinfo *afinfo;
  96         struct dst_entry *dst;
  97
  98         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR)
  99                 saddr = x->coaddr;
 100         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR)
 101                 daddr = x->coaddr;
 102
 103         afinfo = xfrm_policy_get_afinfo(family);
 104         if (unlikely(afinfo == NULL))
 105                 return ERR_PTR(-EAFNOSUPPORT);
 106
 107         dst = afinfo->dst_lookup(tos, saddr, daddr);
 108         xfrm_policy_put_afinfo(afinfo);
 109         return dst;
 110 }
 111
 112 static inline unsigned long make_jiffies(long secs)
 113 {
 114         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 115                 return MAX_SCHEDULE_TIMEOUT-1;
 116         else
 117                 return secs*HZ;
 118 }
 119
 120 static void xfrm_policy_timer(unsigned long data)
 121 {
 122         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 123         unsigned long now = get_seconds();
 124         long next = LONG_MAX;
 125         int warn = 0;
 126         int dir;
 127
 128         read_lock(&xp->lock);
 129
 130         if (xp->dead)
 131                 goto out;
 132
 133         dir = xfrm_policy_id2dir(xp->index);
 134
 135         if (xp->lft.hard_add_expires_seconds) {
 136                 long tmo = xp->lft.hard_add_expires_seconds +
 137                         xp->curlft.add_time - now;
 138                 if (tmo <= 0)
 139                         goto expired;
 140                 if (tmo < next)
 141                         next = tmo;
 142         }
 143         if (xp->lft.hard_use_expires_seconds) {
 144                 long tmo = xp->lft.hard_use_expires_seconds +
 145                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 146                 if (tmo <= 0)
 147                         goto expired;
 148                 if (tmo < next)
 149                         next = tmo;
 150         }
 151         if (xp->lft.soft_add_expires_seconds) {
 152                 long tmo = xp->lft.soft_add_expires_seconds +
 153                         xp->curlft.add_time - now;
 154                 if (tmo <= 0) {
 155                         warn = 1;
 156                         tmo = XFRM_KM_TIMEOUT;
 157                 }
 158                 if (tmo < next)
 159                         next = tmo;
 160         }
 161         if (xp->lft.soft_use_expires_seconds) {
 162                 long tmo = xp->lft.soft_use_expires_seconds +
 163                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 164                 if (tmo <= 0) {
 165                         warn = 1;
 166                         tmo = XFRM_KM_TIMEOUT;
 167                 }
 168                 if (tmo < next)
 169                         next = tmo;
 170         }
 171
 172         if (warn)
 173                 km_policy_expired(xp, dir, 0, 0);
 174         if (next != LONG_MAX &&
 175             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 176                 xfrm_pol_hold(xp);
 177
 178 out:
 179         read_unlock(&xp->lock);
 180         xfrm_pol_put(xp);
 181         return;
 182
 183 expired:
 184         read_unlock(&xp->lock);
 185         if (!xfrm_policy_delete(xp, dir))
 186                 km_policy_expired(xp, dir, 1, 0);
 187         xfrm_pol_put(xp);
 188 }
 189
 190
 191 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 192  * SPD calls.
 193  */
 194
 195 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 196 {
 197         struct xfrm_policy *policy;
 198
 199         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 200
 201         if (policy) {
 202                 INIT_HLIST_NODE(&policy->bydst);
 203                 INIT_HLIST_NODE(&policy->byidx);
 204                 rwlock_init(&policy->lock);
 205                 atomic_set(&policy->refcnt, 1);
 206                 setup_timer(&policy->timer, xfrm_policy_timer,
 207                                 (unsigned long)policy);
 208         }
 209         return policy;
 210 }
 211 EXPORT_SYMBOL(xfrm_policy_alloc);
 212
 213 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 214
 215 void __xfrm_policy_destroy(struct xfrm_policy *policy)
 216 {
 217         BUG_ON(!policy->dead);
 218
 219         BUG_ON(policy->bundles);
 220
 221         if (del_timer(&policy->timer))
 222                 BUG();
 223
 224         security_xfrm_policy_free(policy);
 225         kfree(policy);
 226 }
 227 EXPORT_SYMBOL(__xfrm_policy_destroy);
 228
 229 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 230 {
 231         struct dst_entry *dst;
 232
 233         while ((dst = policy->bundles) != NULL) {
 234                 policy->bundles = dst->next;
 235                 dst_free(dst);
 236         }
 237
 238         if (del_timer(&policy->timer))
 239                 atomic_dec(&policy->refcnt);
 240
 241         if (atomic_read(&policy->refcnt) > 1)
 242                 flow_cache_flush();
 243
 244         xfrm_pol_put(policy);
 245 }
 246
 247 static void xfrm_policy_gc_task(struct work_struct *work)
 248 {
 249         struct xfrm_policy *policy;
 250         struct hlist_node *entry, *tmp;
 251         struct hlist_head gc_list;
 252
 253         spin_lock_bh(&xfrm_policy_gc_lock);
 254         gc_list.first = xfrm_policy_gc_list.first;
 255         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 256         spin_unlock_bh(&xfrm_policy_gc_lock);
 257
 258         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 259                 xfrm_policy_gc_kill(policy);
 260 }
 261
 262 /* Rule must be locked. Release descentant resources, announce
 263  * entry dead. The rule must be unlinked from lists to the moment.
 264  */
 265
 266 static void xfrm_policy_kill(struct xfrm_policy *policy)
 267 {
 268         int dead;
 269
 270         write_lock_bh(&policy->lock);
 271         dead = policy->dead;
 272         policy->dead = 1;
 273         write_unlock_bh(&policy->lock);
 274
 275         if (unlikely(dead)) {
 276                 WARN_ON(1);
 277                 return;
 278         }
 279
 280         spin_lock(&xfrm_policy_gc_lock);
 281         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 282         spin_unlock(&xfrm_policy_gc_lock);
 283
 284         schedule_work(&xfrm_policy_gc_work);
 285 }
 286
 287 struct xfrm_policy_hash {
 288         struct hlist_head       *table;
 289         unsigned int            hmask;
 290 };
 291
 292 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 293 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 294 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 295 static unsigned int xfrm_idx_hmask __read_mostly;
 296 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 297
 298 static inline unsigned int idx_hash(u32 index)
 299 {
 300         return __idx_hash(index, xfrm_idx_hmask);
 301 }
 302
 303 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 304 {
 305         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 306         unsigned int hash = __sel_hash(sel, family, hmask);
 307
 308         return (hash == hmask + 1 ?
 309                 &xfrm_policy_inexact[dir] :
 310                 xfrm_policy_bydst[dir].table + hash);
 311 }
 312
 313 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 314 {
 315         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 316         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 317
 318         return xfrm_policy_bydst[dir].table + hash;
 319 }
 320
 321 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 322                                    struct hlist_head *ndsttable,
 323                                    unsigned int nhashmask)
 324 {
 325         struct hlist_node *entry, *tmp;
 326         struct xfrm_policy *pol;
 327
 328         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 329                 unsigned int h;
 330
 331                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 332                                 pol->family, nhashmask);
 333                 hlist_add_head(&pol->bydst, ndsttable+h);
 334         }
 335 }
 336
 337 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 338                                    struct hlist_head *nidxtable,
 339                                    unsigned int nhashmask)
 340 {
 341         struct hlist_node *entry, *tmp;
 342         struct xfrm_policy *pol;
 343
 344         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 345                 unsigned int h;
 346
 347                 h = __idx_hash(pol->index, nhashmask);
 348                 hlist_add_head(&pol->byidx, nidxtable+h);
 349         }
 350 }
 351
 352 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 353 {
 354         return ((old_hmask + 1) << 1) - 1;
 355 }
 356
 357 static void xfrm_bydst_resize(int dir)
 358 {
 359         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 360         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 361         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 362         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 363         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 364         int i;
 365
 366         if (!ndst)
 367                 return;
 368
 369         write_lock_bh(&xfrm_policy_lock);
 370
 371         for (i = hmask; i >= 0; i--)
 372                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 373
 374         xfrm_policy_bydst[dir].table = ndst;
 375         xfrm_policy_bydst[dir].hmask = nhashmask;
 376
 377         write_unlock_bh(&xfrm_policy_lock);
 378
 379         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 380 }
 381
 382 static void xfrm_byidx_resize(int total)
 383 {
 384         unsigned int hmask = xfrm_idx_hmask;
 385         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 386         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 387         struct hlist_head *oidx = xfrm_policy_byidx;
 388         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 389         int i;
 390
 391         if (!nidx)
 392                 return;
 393
 394         write_lock_bh(&xfrm_policy_lock);
 395
 396         for (i = hmask; i >= 0; i--)
 397                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 398
 399         xfrm_policy_byidx = nidx;
 400         xfrm_idx_hmask = nhashmask;
 401
 402         write_unlock_bh(&xfrm_policy_lock);
 403
 404         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 405 }
 406
 407 static inline int xfrm_bydst_should_resize(int dir, int *total)
 408 {
 409         unsigned int cnt = xfrm_policy_count[dir];
 410         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 411
 412         if (total)
 413                 *total += cnt;
 414
 415         if ((hmask + 1) < xfrm_policy_hashmax &&
 416             cnt > hmask)
 417                 return 1;
 418
 419         return 0;
 420 }
 421
 422 static inline int xfrm_byidx_should_resize(int total)
 423 {
 424         unsigned int hmask = xfrm_idx_hmask;
 425
 426         if ((hmask + 1) < xfrm_policy_hashmax &&
 427             total > hmask)
 428                 return 1;
 429
 430         return 0;
 431 }
 432
 433 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 434 {
 435         read_lock_bh(&xfrm_policy_lock);
 436         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 437         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 438         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 439         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 440         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 441         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 442         si->spdhcnt = xfrm_idx_hmask;
 443         si->spdhmcnt = xfrm_policy_hashmax;
 444         read_unlock_bh(&xfrm_policy_lock);
 445 }
 446 EXPORT_SYMBOL(xfrm_spd_getinfo);
 447
 448 static DEFINE_MUTEX(hash_resize_mutex);
 449 static void xfrm_hash_resize(struct work_struct *__unused)
 450 {
 451         int dir, total;
 452
 453         mutex_lock(&hash_resize_mutex);
 454
 455         total = 0;
 456         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 457                 if (xfrm_bydst_should_resize(dir, &total))
 458                         xfrm_bydst_resize(dir);
 459         }
 460         if (xfrm_byidx_should_resize(total))
 461                 xfrm_byidx_resize(total);
 462
 463         mutex_unlock(&hash_resize_mutex);
 464 }
 465
 466 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 467
 468 /* Generate new index... KAME seems to generate them ordered by cost
 469  * of an absolute inpredictability of ordering of rules. This will not pass. */
 470 static u32 xfrm_gen_index(u8 type, int dir)
 471 {
 472         static u32 idx_generator;
 473
 474         for (;;) {
 475                 struct hlist_node *entry;
 476                 struct hlist_head *list;
 477                 struct xfrm_policy *p;
 478                 u32 idx;
 479                 int found;
 480
 481                 idx = (idx_generator | dir);
 482                 idx_generator += 8;
 483                 if (idx == 0)
 484                         idx = 8;
 485                 list = xfrm_policy_byidx + idx_hash(idx);
 486                 found = 0;
 487                 hlist_for_each_entry(p, entry, list, byidx) {
 488                         if (p->index == idx) {
 489                                 found = 1;
 490                                 break;
 491                         }
 492                 }
 493                 if (!found)
 494                         return idx;
 495         }
 496 }
 497
 498 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 499 {
 500         u32 *p1 = (u32 *) s1;
 501         u32 *p2 = (u32 *) s2;
 502         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 503         int i;
 504
 505         for (i = 0; i < len; i++) {
 506                 if (p1[i] != p2[i])
 507                         return 1;
 508         }
 509
 510         return 0;
 511 }
 512
 513 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 514 {
 515         struct xfrm_policy *pol;
 516         struct xfrm_policy *delpol;
 517         struct hlist_head *chain;
 518         struct hlist_node *entry, *newpos;
 519         struct dst_entry *gc_list;
 520
 521         write_lock_bh(&xfrm_policy_lock);
 522         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 523         delpol = NULL;
 524         newpos = NULL;
 525         hlist_for_each_entry(pol, entry, chain, bydst) {
 526                 if (pol->type == policy->type &&
 527                     !selector_cmp(&pol->selector, &policy->selector) &&
 528                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 529                     !WARN_ON(delpol)) {
 530                         if (excl) {
 531                                 write_unlock_bh(&xfrm_policy_lock);
 532                                 return -EEXIST;
 533                         }
 534                         delpol = pol;
 535                         if (policy->priority > pol->priority)
 536                                 continue;
 537                 } else if (policy->priority >= pol->priority) {
 538                         newpos = &pol->bydst;
 539                         continue;
 540                 }
 541                 if (delpol)
 542                         break;
 543         }
 544         if (newpos)
 545                 hlist_add_after(newpos, &policy->bydst);
 546         else
 547                 hlist_add_head(&policy->bydst, chain);
 548         xfrm_pol_hold(policy);
 549         xfrm_policy_count[dir]++;
 550         atomic_inc(&flow_cache_genid);
 551         if (delpol) {
 552                 hlist_del(&delpol->bydst);
 553                 hlist_del(&delpol->byidx);
 554                 xfrm_policy_count[dir]--;
 555         }
 556         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 557         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 558         policy->curlft.add_time = get_seconds();
 559         policy->curlft.use_time = 0;
 560         if (!mod_timer(&policy->timer, jiffies + HZ))
 561                 xfrm_pol_hold(policy);
 562         write_unlock_bh(&xfrm_policy_lock);
 563
 564         if (delpol)
 565                 xfrm_policy_kill(delpol);
 566         else if (xfrm_bydst_should_resize(dir, NULL))
 567                 schedule_work(&xfrm_hash_work);
 568
 569         read_lock_bh(&xfrm_policy_lock);
 570         gc_list = NULL;
 571         entry = &policy->bydst;
 572         hlist_for_each_entry_continue(policy, entry, bydst) {
 573                 struct dst_entry *dst;
 574
 575                 write_lock(&policy->lock);
 576                 dst = policy->bundles;
 577                 if (dst) {
 578                         struct dst_entry *tail = dst;
 579                         while (tail->next)
 580                                 tail = tail->next;
 581                         tail->next = gc_list;
 582                         gc_list = dst;
 583
 584                         policy->bundles = NULL;
 585                 }
 586                 write_unlock(&policy->lock);
 587         }
 588         read_unlock_bh(&xfrm_policy_lock);
 589
 590         while (gc_list) {
 591                 struct dst_entry *dst = gc_list;
 592
 593                 gc_list = dst->next;
 594                 dst_free(dst);
 595         }
 596
 597         return 0;
 598 }
 599 EXPORT_SYMBOL(xfrm_policy_insert);
 600
 601 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 602                                           struct xfrm_selector *sel,
 603                                           struct xfrm_sec_ctx *ctx, int delete,
 604                                           int *err)
 605 {
 606         struct xfrm_policy *pol, *ret;
 607         struct hlist_head *chain;
 608         struct hlist_node *entry;
 609
 610         *err = 0;
 611         write_lock_bh(&xfrm_policy_lock);
 612         chain = policy_hash_bysel(sel, sel->family, dir);
 613         ret = NULL;
 614         hlist_for_each_entry(pol, entry, chain, bydst) {
 615                 if (pol->type == type &&
 616                     !selector_cmp(sel, &pol->selector) &&
 617                     xfrm_sec_ctx_match(ctx, pol->security)) {
 618                         xfrm_pol_hold(pol);
 619                         if (delete) {
 620                                 *err = security_xfrm_policy_delete(pol);
 621                                 if (*err) {
 622                                         write_unlock_bh(&xfrm_policy_lock);
 623                                         return pol;
 624                                 }
 625                                 hlist_del(&pol->bydst);
 626                                 hlist_del(&pol->byidx);
 627                                 xfrm_policy_count[dir]--;
 628                         }
 629                         ret = pol;
 630                         break;
 631                 }
 632         }
 633         write_unlock_bh(&xfrm_policy_lock);
 634
 635         if (ret && delete) {
 636                 atomic_inc(&flow_cache_genid);
 637                 xfrm_policy_kill(ret);
 638         }
 639         return ret;
 640 }
 641 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 642
 643 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 644                                      int *err)
 645 {
 646         struct xfrm_policy *pol, *ret;
 647         struct hlist_head *chain;
 648         struct hlist_node *entry;
 649
 650         *err = -ENOENT;
 651         if (xfrm_policy_id2dir(id) != dir)
 652                 return NULL;
 653
 654         *err = 0;
 655         write_lock_bh(&xfrm_policy_lock);
 656         chain = xfrm_policy_byidx + idx_hash(id);
 657         ret = NULL;
 658         hlist_for_each_entry(pol, entry, chain, byidx) {
 659                 if (pol->type == type && pol->index == id) {
 660                         xfrm_pol_hold(pol);
 661                         if (delete) {
 662                                 *err = security_xfrm_policy_delete(pol);
 663                                 if (*err) {
 664                                         write_unlock_bh(&xfrm_policy_lock);
 665                                         return pol;
 666                                 }
 667                                 hlist_del(&pol->bydst);
 668                                 hlist_del(&pol->byidx);
 669                                 xfrm_policy_count[dir]--;
 670                         }
 671                         ret = pol;
 672                         break;
 673                 }
 674         }
 675         write_unlock_bh(&xfrm_policy_lock);
 676
 677         if (ret && delete) {
 678                 atomic_inc(&flow_cache_genid);
 679                 xfrm_policy_kill(ret);
 680         }
 681         return ret;
 682 }
 683 EXPORT_SYMBOL(xfrm_policy_byid);
 684
 685 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 686 static inline int
 687 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 688 {
 689         int dir, err = 0;
 690
 691         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 692                 struct xfrm_policy *pol;
 693                 struct hlist_node *entry;
 694                 int i;
 695
 696                 hlist_for_each_entry(pol, entry,
 697                                      &xfrm_policy_inexact[dir], bydst) {
 698                         if (pol->type != type)
 699                                 continue;
 700                         err = security_xfrm_policy_delete(pol);
 701                         if (err) {
 702                                 xfrm_audit_policy_delete(pol, 0,
 703                                                          audit_info->loginuid,
 704                                                          audit_info->secid);
 705                                 return err;
 706                         }
 707                 }
 708                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 709                         hlist_for_each_entry(pol, entry,
 710                                              xfrm_policy_bydst[dir].table + i,
 711                                              bydst) {
 712                                 if (pol->type != type)
 713                                         continue;
 714                                 err = security_xfrm_policy_delete(pol);
 715                                 if (err) {
 716                                         xfrm_audit_policy_delete(pol, 0,
 717                                                         audit_info->loginuid,
 718                                                         audit_info->secid);
 719                                         return err;
 720                                 }
 721                         }
 722                 }
 723         }
 724         return err;
 725 }
 726 #else
 727 static inline int
 728 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 729 {
 730         return 0;
 731 }
 732 #endif
 733
 734 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 735 {
 736         int dir, err = 0;
 737
 738         write_lock_bh(&xfrm_policy_lock);
 739
 740         err = xfrm_policy_flush_secctx_check(type, audit_info);
 741         if (err)
 742                 goto out;
 743
 744         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 745                 struct xfrm_policy *pol;
 746                 struct hlist_node *entry;
 747                 int i, killed;
 748
 749                 killed = 0;
 750         again1:
 751                 hlist_for_each_entry(pol, entry,
 752                                      &xfrm_policy_inexact[dir], bydst) {
 753                         if (pol->type != type)
 754                                 continue;
 755                         hlist_del(&pol->bydst);
 756                         hlist_del(&pol->byidx);
 757                         write_unlock_bh(&xfrm_policy_lock);
 758
 759                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 760                                                  audit_info->secid);
 761
 762                         xfrm_policy_kill(pol);
 763                         killed++;
 764
 765                         write_lock_bh(&xfrm_policy_lock);
 766                         goto again1;
 767                 }
 768
 769                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 770         again2:
 771                         hlist_for_each_entry(pol, entry,
 772                                              xfrm_policy_bydst[dir].table + i,
 773                                              bydst) {
 774                                 if (pol->type != type)
 775                                         continue;
 776                                 hlist_del(&pol->bydst);
 777                                 hlist_del(&pol->byidx);
 778                                 write_unlock_bh(&xfrm_policy_lock);
 779
 780                                 xfrm_audit_policy_delete(pol, 1,
 781                                                          audit_info->loginuid,
 782                                                          audit_info->secid);
 783                                 xfrm_policy_kill(pol);
 784                                 killed++;
 785
 786                                 write_lock_bh(&xfrm_policy_lock);
 787                                 goto again2;
 788                         }
 789                 }
 790
 791                 xfrm_policy_count[dir] -= killed;
 792         }
 793         atomic_inc(&flow_cache_genid);
 794 out:
 795         write_unlock_bh(&xfrm_policy_lock);
 796         return err;
 797 }
 798 EXPORT_SYMBOL(xfrm_policy_flush);
 799
 800 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
 801                      void *data)
 802 {
 803         struct xfrm_policy *pol, *last = NULL;
 804         struct hlist_node *entry;
 805         int dir, last_dir = 0, count, error;
 806
 807         read_lock_bh(&xfrm_policy_lock);
 808         count = 0;
 809
 810         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
 811                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
 812                 int i;
 813
 814                 hlist_for_each_entry(pol, entry,
 815                                      &xfrm_policy_inexact[dir], bydst) {
 816                         if (pol->type != type)
 817                                 continue;
 818                         if (last) {
 819                                 error = func(last, last_dir % XFRM_POLICY_MAX,
 820                                              count, data);
 821                                 if (error)
 822                                         goto out;
 823                         }
 824                         last = pol;
 825                         last_dir = dir;
 826                         count++;
 827                 }
 828                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 829                         hlist_for_each_entry(pol, entry, table + i, bydst) {
 830                                 if (pol->type != type)
 831                                         continue;
 832                                 if (last) {
 833                                         error = func(last, last_dir % XFRM_POLICY_MAX,
 834                                                      count, data);
 835                                         if (error)
 836                                                 goto out;
 837                                 }
 838                                 last = pol;
 839                                 last_dir = dir;
 840                                 count++;
 841                         }
 842                 }
 843         }
 844         if (count == 0) {
 845                 error = -ENOENT;
 846                 goto out;
 847         }
 848         error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
 849 out:
 850         read_unlock_bh(&xfrm_policy_lock);
 851         return error;
 852 }
 853 EXPORT_SYMBOL(xfrm_policy_walk);
 854
 855 /*
 856  * Find policy to apply to this flow.
 857  *
 858  * Returns 0 if policy found, else an -errno.
 859  */
 860 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 861                              u8 type, u16 family, int dir)
 862 {
 863         struct xfrm_selector *sel = &pol->selector;
 864         int match, ret = -ESRCH;
 865
 866         if (pol->family != family ||
 867             pol->type != type)
 868                 return ret;
 869
 870         match = xfrm_selector_match(sel, fl, family);
 871         if (match)
 872                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
 873
 874         return ret;
 875 }
 876
 877 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 878                                                      u16 family, u8 dir)
 879 {
 880         int err;
 881         struct xfrm_policy *pol, *ret;
 882         xfrm_address_t *daddr, *saddr;
 883         struct hlist_node *entry;
 884         struct hlist_head *chain;
 885         u32 priority = ~0U;
 886
 887         daddr = xfrm_flowi_daddr(fl, family);
 888         saddr = xfrm_flowi_saddr(fl, family);
 889         if (unlikely(!daddr || !saddr))
 890                 return NULL;
 891
 892         read_lock_bh(&xfrm_policy_lock);
 893         chain = policy_hash_direct(daddr, saddr, family, dir);
 894         ret = NULL;
 895         hlist_for_each_entry(pol, entry, chain, bydst) {
 896                 err = xfrm_policy_match(pol, fl, type, family, dir);
 897                 if (err) {
 898                         if (err == -ESRCH)
 899                                 continue;
 900                         else {
 901                                 ret = ERR_PTR(err);
 902                                 goto fail;
 903                         }
 904                 } else {
 905                         ret = pol;
 906                         priority = ret->priority;
 907                         break;
 908                 }
 909         }
 910         chain = &xfrm_policy_inexact[dir];
 911         hlist_for_each_entry(pol, entry, chain, bydst) {
 912                 err = xfrm_policy_match(pol, fl, type, family, dir);
 913                 if (err) {
 914                         if (err == -ESRCH)
 915                                 continue;
 916                         else {
 917                                 ret = ERR_PTR(err);
 918                                 goto fail;
 919                         }
 920                 } else if (pol->priority < priority) {
 921                         ret = pol;
 922                         break;
 923                 }
 924         }
 925         if (ret)
 926                 xfrm_pol_hold(ret);
 927 fail:
 928         read_unlock_bh(&xfrm_policy_lock);
 929
 930         return ret;
 931 }
 932
 933 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
 934                                void **objp, atomic_t **obj_refp)
 935 {
 936         struct xfrm_policy *pol;
 937         int err = 0;
 938
 939 #ifdef CONFIG_XFRM_SUB_POLICY
 940         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
 941         if (IS_ERR(pol)) {
 942                 err = PTR_ERR(pol);
 943                 pol = NULL;
 944         }
 945         if (pol || err)
 946                 goto end;
 947 #endif
 948         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 949         if (IS_ERR(pol)) {
 950                 err = PTR_ERR(pol);
 951                 pol = NULL;
 952         }
 953 #ifdef CONFIG_XFRM_SUB_POLICY
 954 end:
 955 #endif
 956         if ((*objp = (void *) pol) != NULL)
 957                 *obj_refp = &pol->refcnt;
 958         return err;
 959 }
 960
 961 static inline int policy_to_flow_dir(int dir)
 962 {
 963         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 964             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 965             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 966                 return dir;
 967         switch (dir) {
 968         default:
 969         case XFRM_POLICY_IN:
 970                 return FLOW_DIR_IN;
 971         case XFRM_POLICY_OUT:
 972                 return FLOW_DIR_OUT;
 973         case XFRM_POLICY_FWD:
 974                 return FLOW_DIR_FWD;
 975         }
 976 }
 977
 978 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
 979 {
 980         struct xfrm_policy *pol;
 981
 982         read_lock_bh(&xfrm_policy_lock);
 983         if ((pol = sk->sk_policy[dir]) != NULL) {
 984                 int match = xfrm_selector_match(&pol->selector, fl,
 985                                                 sk->sk_family);
 986                 int err = 0;
 987
 988                 if (match) {
 989                         err = security_xfrm_policy_lookup(pol, fl->secid,
 990                                         policy_to_flow_dir(dir));
 991                         if (!err)
 992                                 xfrm_pol_hold(pol);
 993                         else if (err == -ESRCH)
 994                                 pol = NULL;
 995                         else
 996                                 pol = ERR_PTR(err);
 997                 } else
 998                         pol = NULL;
 999         }
1000         read_unlock_bh(&xfrm_policy_lock);
1001         return pol;
1002 }
1003
1004 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1005 {
1006         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1007                                                      pol->family, dir);
1008
1009         hlist_add_head(&pol->bydst, chain);
1010         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1011         xfrm_policy_count[dir]++;
1012         xfrm_pol_hold(pol);
1013
1014         if (xfrm_bydst_should_resize(dir, NULL))
1015                 schedule_work(&xfrm_hash_work);
1016 }
1017
1018 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1019                                                 int dir)
1020 {
1021         if (hlist_unhashed(&pol->bydst))
1022                 return NULL;
1023
1024         hlist_del(&pol->bydst);
1025         hlist_del(&pol->byidx);
1026         xfrm_policy_count[dir]--;
1027
1028         return pol;
1029 }
1030
1031 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1032 {
1033         write_lock_bh(&xfrm_policy_lock);
1034         pol = __xfrm_policy_unlink(pol, dir);
1035         write_unlock_bh(&xfrm_policy_lock);
1036         if (pol) {
1037                 if (dir < XFRM_POLICY_MAX)
1038                         atomic_inc(&flow_cache_genid);
1039                 xfrm_policy_kill(pol);
1040                 return 0;
1041         }
1042         return -ENOENT;
1043 }
1044 EXPORT_SYMBOL(xfrm_policy_delete);
1045
1046 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1047 {
1048         struct xfrm_policy *old_pol;
1049
1050 #ifdef CONFIG_XFRM_SUB_POLICY
1051         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1052                 return -EINVAL;
1053 #endif
1054
1055         write_lock_bh(&xfrm_policy_lock);
1056         old_pol = sk->sk_policy[dir];
1057         sk->sk_policy[dir] = pol;
1058         if (pol) {
1059                 pol->curlft.add_time = get_seconds();
1060                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1061                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1062         }
1063         if (old_pol)
1064                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1065         write_unlock_bh(&xfrm_policy_lock);
1066
1067         if (old_pol) {
1068                 xfrm_policy_kill(old_pol);
1069         }
1070         return 0;
1071 }
1072
1073 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1074 {
1075         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1076
1077         if (newp) {
1078                 newp->selector = old->selector;
1079                 if (security_xfrm_policy_clone(old, newp)) {
1080                         kfree(newp);
1081                         return NULL;  /* ENOMEM */
1082                 }
1083                 newp->lft = old->lft;
1084                 newp->curlft = old->curlft;
1085                 newp->action = old->action;
1086                 newp->flags = old->flags;
1087                 newp->xfrm_nr = old->xfrm_nr;
1088                 newp->index = old->index;
1089                 newp->type = old->type;
1090                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1091                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1092                 write_lock_bh(&xfrm_policy_lock);
1093                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1094                 write_unlock_bh(&xfrm_policy_lock);
1095                 xfrm_pol_put(newp);
1096         }
1097         return newp;
1098 }
1099
1100 int __xfrm_sk_clone_policy(struct sock *sk)
1101 {
1102         struct xfrm_policy *p0 = sk->sk_policy[0],
1103                            *p1 = sk->sk_policy[1];
1104
1105         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1106         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1107                 return -ENOMEM;
1108         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1109                 return -ENOMEM;
1110         return 0;
1111 }
1112
1113 static int
1114 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1115                unsigned short family)
1116 {
1117         int err;
1118         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1119
1120         if (unlikely(afinfo == NULL))
1121                 return -EINVAL;
1122         err = afinfo->get_saddr(local, remote);
1123         xfrm_policy_put_afinfo(afinfo);
1124         return err;
1125 }
1126
1127 /* Resolve list of templates for the flow, given policy. */
1128
1129 static int
1130 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1131                       struct xfrm_state **xfrm,
1132                       unsigned short family)
1133 {
1134         int nx;
1135         int i, error;
1136         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1137         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1138         xfrm_address_t tmp;
1139
1140         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1141                 struct xfrm_state *x;
1142                 xfrm_address_t *remote = daddr;
1143                 xfrm_address_t *local  = saddr;
1144                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1145
1146                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1147                     tmpl->mode == XFRM_MODE_BEET) {
1148                         remote = &tmpl->id.daddr;
1149                         local = &tmpl->saddr;
1150                         family = tmpl->encap_family;
1151                         if (xfrm_addr_any(local, family)) {
1152                                 error = xfrm_get_saddr(&tmp, remote, family);
1153                                 if (error)
1154                                         goto fail;
1155                                 local = &tmp;
1156                         }
1157                 }
1158
1159                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1160
1161                 if (x && x->km.state == XFRM_STATE_VALID) {
1162                         xfrm[nx++] = x;
1163                         daddr = remote;
1164                         saddr = local;
1165                         continue;
1166                 }
1167                 if (x) {
1168                         error = (x->km.state == XFRM_STATE_ERROR ?
1169                                  -EINVAL : -EAGAIN);
1170                         xfrm_state_put(x);
1171                 }
1172
1173                 if (!tmpl->optional)
1174                         goto fail;
1175         }
1176         return nx;
1177
1178 fail:
1179         for (nx--; nx>=0; nx--)
1180                 xfrm_state_put(xfrm[nx]);
1181         return error;
1182 }
1183
1184 static int
1185 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1186                   struct xfrm_state **xfrm,
1187                   unsigned short family)
1188 {
1189         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1190         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1191         int cnx = 0;
1192         int error;
1193         int ret;
1194         int i;
1195
1196         for (i = 0; i < npols; i++) {
1197                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1198                         error = -ENOBUFS;
1199                         goto fail;
1200                 }
1201
1202                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1203                 if (ret < 0) {
1204                         error = ret;
1205                         goto fail;
1206                 } else
1207                         cnx += ret;
1208         }
1209
1210         /* found states are sorted for outbound processing */
1211         if (npols > 1)
1212                 xfrm_state_sort(xfrm, tpp, cnx, family);
1213
1214         return cnx;
1215
1216  fail:
1217         for (cnx--; cnx>=0; cnx--)
1218                 xfrm_state_put(tpp[cnx]);
1219         return error;
1220
1221 }
1222
1223 /* Check that the bundle accepts the flow and its components are
1224  * still valid.
1225  */
1226
1227 static struct dst_entry *
1228 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1229 {
1230         struct dst_entry *x;
1231         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1232         if (unlikely(afinfo == NULL))
1233                 return ERR_PTR(-EINVAL);
1234         x = afinfo->find_bundle(fl, policy);
1235         xfrm_policy_put_afinfo(afinfo);
1236         return x;
1237 }
1238
1239 static inline int xfrm_get_tos(struct flowi *fl, int family)
1240 {
1241         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1242         int tos;
1243
1244         if (!afinfo)
1245                 return -EINVAL;
1246
1247         tos = afinfo->get_tos(fl);
1248
1249         xfrm_policy_put_afinfo(afinfo);
1250
1251         return tos;
1252 }
1253
1254 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1255 {
1256         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1257         struct xfrm_dst *xdst;
1258
1259         if (!afinfo)
1260                 return ERR_PTR(-EINVAL);
1261
1262         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1263
1264         xfrm_policy_put_afinfo(afinfo);
1265
1266         return xdst;
1267 }
1268
1269 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1270 {
1271         struct xfrm_policy_afinfo *afinfo =
1272                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1273         int err;
1274
1275         if (!afinfo)
1276                 return -EINVAL;
1277
1278         err = afinfo->fill_dst(xdst, dev);
1279
1280         xfrm_policy_put_afinfo(afinfo);
1281
1282         return err;
1283 }
1284
1285 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1286  * all the metrics... Shortly, bundle a bundle.
1287  */
1288
1289 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1290                                             struct xfrm_state **xfrm, int nx,
1291                                             struct flowi *fl,
1292                                             struct dst_entry *dst)
1293 {
1294         unsigned long now = jiffies;
1295         struct net_device *dev;
1296         struct dst_entry *dst_prev = NULL;
1297         struct dst_entry *dst0 = NULL;
1298         int i = 0;
1299         int err;
1300         int header_len = 0;
1301         int trailer_len = 0;
1302         int tos;
1303         int family = policy->selector.family;
1304
1305         tos = xfrm_get_tos(fl, family);
1306         err = tos;
1307         if (tos < 0)
1308                 goto put_states;
1309
1310         dst_hold(dst);
1311
1312         for (; i < nx; i++) {
1313                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1314                 struct dst_entry *dst1 = &xdst->u.dst;
1315
1316                 err = PTR_ERR(xdst);
1317                 if (IS_ERR(xdst)) {
1318                         dst_release(dst);
1319                         goto put_states;
1320                 }
1321
1322                 if (!dst_prev)
1323                         dst0 = dst1;
1324                 else {
1325                         dst_prev->child = dst_clone(dst1);
1326                         dst1->flags |= DST_NOHASH;
1327                 }
1328
1329                 xdst->route = dst;
1330                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1331
1332                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1333                         family = xfrm[i]->props.family;
1334                         dst = xfrm_dst_lookup(xfrm[i], tos, family);
1335                         err = PTR_ERR(dst);
1336                         if (IS_ERR(dst))
1337                                 goto put_states;
1338                 } else
1339                         dst_hold(dst);
1340
1341                 dst1->xfrm = xfrm[i];
1342                 xdst->genid = xfrm[i]->genid;
1343
1344                 dst1->obsolete = -1;
1345                 dst1->flags |= DST_HOST;
1346                 dst1->lastuse = now;
1347
1348                 dst1->input = dst_discard;
1349                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1350
1351                 dst1->next = dst_prev;
1352                 dst_prev = dst1;
1353
1354                 header_len += xfrm[i]->props.header_len;
1355                 trailer_len += xfrm[i]->props.trailer_len;
1356         }
1357
1358         dst_prev->child = dst;
1359         dst0->path = dst;
1360
1361         err = -ENODEV;
1362         dev = dst->dev;
1363         if (!dev)
1364                 goto free_dst;
1365
1366         /* Copy neighbout for reachability confirmation */
1367         dst0->neighbour = neigh_clone(dst->neighbour);
1368
1369         xfrm_init_pmtu(dst_prev);
1370
1371         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1372                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1373
1374                 err = xfrm_fill_dst(xdst, dev);
1375                 if (err)
1376                         goto free_dst;
1377
1378                 dst_prev->header_len = header_len;
1379                 dst_prev->trailer_len = trailer_len;
1380                 header_len -= xdst->u.dst.xfrm->props.header_len;
1381                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1382         }
1383
1384 out:
1385         return dst0;
1386
1387 put_states:
1388         for (; i < nx; i++)
1389                 xfrm_state_put(xfrm[i]);
1390 free_dst:
1391         if (dst0)
1392                 dst_free(dst0);
1393         dst0 = ERR_PTR(err);
1394         goto out;
1395 }
1396
1397 static int inline
1398 xfrm_dst_alloc_copy(void **target, void *src, int size)
1399 {
1400         if (!*target) {
1401                 *target = kmalloc(size, GFP_ATOMIC);
1402                 if (!*target)
1403                         return -ENOMEM;
1404         }
1405         memcpy(*target, src, size);
1406         return 0;
1407 }
1408
1409 static int inline
1410 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1411 {
1412 #ifdef CONFIG_XFRM_SUB_POLICY
1413         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1414         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1415                                    sel, sizeof(*sel));
1416 #else
1417         return 0;
1418 #endif
1419 }
1420
1421 static int inline
1422 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1423 {
1424 #ifdef CONFIG_XFRM_SUB_POLICY
1425         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1426         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1427 #else
1428         return 0;
1429 #endif
1430 }
1431
1432 static int stale_bundle(struct dst_entry *dst);
1433
1434 /* Main function: finds/creates a bundle for given flow.
1435  *
1436  * At the moment we eat a raw IP route. Mostly to speed up lookups
1437  * on interfaces with disabled IPsec.
1438  */
1439 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1440                   struct sock *sk, int flags)
1441 {
1442         struct xfrm_policy *policy;
1443         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1444         int npols;
1445         int pol_dead;
1446         int xfrm_nr;
1447         int pi;
1448         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1449         struct dst_entry *dst, *dst_orig = *dst_p;
1450         int nx = 0;
1451         int err;
1452         u32 genid;
1453         u16 family;
1454         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1455
1456 restart:
1457         genid = atomic_read(&flow_cache_genid);
1458         policy = NULL;
1459         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1460                 pols[pi] = NULL;
1461         npols = 0;
1462         pol_dead = 0;
1463         xfrm_nr = 0;
1464
1465         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1466                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1467                 err = PTR_ERR(policy);
1468                 if (IS_ERR(policy))
1469                         goto dropdst;
1470         }
1471
1472         err = -ENOENT;
1473
1474         if (!policy) {
1475                 /* To accelerate a bit...  */
1476                 if ((dst_orig->flags & DST_NOXFRM) ||
1477                     !xfrm_policy_count[XFRM_POLICY_OUT])
1478                         goto nopol;
1479
1480                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1481                                            dir, xfrm_policy_lookup);
1482                 err = PTR_ERR(policy);
1483                 if (IS_ERR(policy))
1484                         goto dropdst;
1485         }
1486
1487         if (!policy)
1488                 goto nopol;
1489
1490         family = dst_orig->ops->family;
1491         pols[0] = policy;
1492         npols ++;
1493         xfrm_nr += pols[0]->xfrm_nr;
1494
1495         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1496                 goto error;
1497
1498         policy->curlft.use_time = get_seconds();
1499
1500         switch (policy->action) {
1501         default:
1502         case XFRM_POLICY_BLOCK:
1503                 /* Prohibit the flow */
1504                 err = -EPERM;
1505                 goto error;
1506
1507         case XFRM_POLICY_ALLOW:
1508 #ifndef CONFIG_XFRM_SUB_POLICY
1509                 if (policy->xfrm_nr == 0) {
1510                         /* Flow passes not transformed. */
1511                         xfrm_pol_put(policy);
1512                         return 0;
1513                 }
1514 #endif
1515
1516                 /* Try to find matching bundle.
1517                  *
1518                  * LATER: help from flow cache. It is optional, this
1519                  * is required only for output policy.
1520                  */
1521                 dst = xfrm_find_bundle(fl, policy, family);
1522                 if (IS_ERR(dst)) {
1523                         err = PTR_ERR(dst);
1524                         goto error;
1525                 }
1526
1527                 if (dst)
1528                         break;
1529
1530 #ifdef CONFIG_XFRM_SUB_POLICY
1531                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1532                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1533                                                             fl, family,
1534                                                             XFRM_POLICY_OUT);
1535                         if (pols[1]) {
1536                                 if (IS_ERR(pols[1])) {
1537                                         err = PTR_ERR(pols[1]);
1538                                         goto error;
1539                                 }
1540                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1541                                         err = -EPERM;
1542                                         goto error;
1543                                 }
1544                                 npols ++;
1545                                 xfrm_nr += pols[1]->xfrm_nr;
1546                         }
1547                 }
1548
1549                 /*
1550                  * Because neither flowi nor bundle information knows about
1551                  * transformation template size. On more than one policy usage
1552                  * we can realize whether all of them is bypass or not after
1553                  * they are searched. See above not-transformed bypass
1554                  * is surrounded by non-sub policy configuration, too.
1555                  */
1556                 if (xfrm_nr == 0) {
1557                         /* Flow passes not transformed. */
1558                         xfrm_pols_put(pols, npols);
1559                         return 0;
1560                 }
1561
1562 #endif
1563                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1564
1565                 if (unlikely(nx<0)) {
1566                         err = nx;
1567                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1568                                 /* EREMOTE tells the caller to generate
1569                                  * a one-shot blackhole route.
1570                                  */
1571                                 xfrm_pol_put(policy);
1572                                 return -EREMOTE;
1573                         }
1574                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1575                                 DECLARE_WAITQUEUE(wait, current);
1576
1577                                 add_wait_queue(&km_waitq, &wait);
1578                                 set_current_state(TASK_INTERRUPTIBLE);
1579                                 schedule();
1580                                 set_current_state(TASK_RUNNING);
1581                                 remove_wait_queue(&km_waitq, &wait);
1582
1583                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1584
1585                                 if (nx == -EAGAIN && signal_pending(current)) {
1586                                         err = -ERESTART;
1587                                         goto error;
1588                                 }
1589                                 if (nx == -EAGAIN ||
1590                                     genid != atomic_read(&flow_cache_genid)) {
1591                                         xfrm_pols_put(pols, npols);
1592                                         goto restart;
1593                                 }
1594                                 err = nx;
1595                         }
1596                         if (err < 0)
1597                                 goto error;
1598                 }
1599                 if (nx == 0) {
1600                         /* Flow passes not transformed. */
1601                         xfrm_pols_put(pols, npols);
1602                         return 0;
1603                 }
1604
1605                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1606                 err = PTR_ERR(dst);
1607                 if (IS_ERR(dst))
1608                         goto error;
1609
1610                 for (pi = 0; pi < npols; pi++) {
1611                         read_lock_bh(&pols[pi]->lock);
1612                         pol_dead |= pols[pi]->dead;
1613                         read_unlock_bh(&pols[pi]->lock);
1614                 }
1615
1616                 write_lock_bh(&policy->lock);
1617                 if (unlikely(pol_dead || stale_bundle(dst))) {
1618                         /* Wow! While we worked on resolving, this
1619                          * policy has gone. Retry. It is not paranoia,
1620                          * we just cannot enlist new bundle to dead object.
1621                          * We can't enlist stable bundles either.
1622                          */
1623                         write_unlock_bh(&policy->lock);
1624                         if (dst)
1625                                 dst_free(dst);
1626
1627                         err = -EHOSTUNREACH;
1628                         goto error;
1629                 }
1630
1631                 if (npols > 1)
1632                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1633                 else
1634                         err = xfrm_dst_update_origin(dst, fl);
1635                 if (unlikely(err)) {
1636                         write_unlock_bh(&policy->lock);
1637                         if (dst)
1638                                 dst_free(dst);
1639                         goto error;
1640                 }
1641
1642                 dst->next = policy->bundles;
1643                 policy->bundles = dst;
1644                 dst_hold(dst);
1645                 write_unlock_bh(&policy->lock);
1646         }
1647         *dst_p = dst;
1648         dst_release(dst_orig);
1649         xfrm_pols_put(pols, npols);
1650         return 0;
1651
1652 error:
1653         xfrm_pols_put(pols, npols);
1654 dropdst:
1655         dst_release(dst_orig);
1656         *dst_p = NULL;
1657         return err;
1658
1659 nopol:
1660         if (flags & XFRM_LOOKUP_ICMP)
1661                 goto dropdst;
1662         return 0;
1663 }
1664 EXPORT_SYMBOL(__xfrm_lookup);
1665
1666 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1667                 struct sock *sk, int flags)
1668 {
1669         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1670
1671         if (err == -EREMOTE) {
1672                 dst_release(*dst_p);
1673                 *dst_p = NULL;
1674                 err = -EAGAIN;
1675         }
1676
1677         return err;
1678 }
1679 EXPORT_SYMBOL(xfrm_lookup);
1680
1681 static inline int
1682 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1683 {
1684         struct xfrm_state *x;
1685
1686         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1687                 return 0;
1688         x = skb->sp->xvec[idx];
1689         if (!x->type->reject)
1690                 return 0;
1691         return x->type->reject(x, skb, fl);
1692 }
1693
1694 /* When skb is transformed back to its "native" form, we have to
1695  * check policy restrictions. At the moment we make this in maximally
1696  * stupid way. Shame on me. :-) Of course, connected sockets must
1697  * have policy cached at them.
1698  */
1699
1700 static inline int
1701 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1702               unsigned short family)
1703 {
1704         if (xfrm_state_kern(x))
1705                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1706         return  x->id.proto == tmpl->id.proto &&
1707                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1708                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1709                 x->props.mode == tmpl->mode &&
1710                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1711                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1712                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1713                   xfrm_state_addr_cmp(tmpl, x, family));
1714 }
1715
1716 /*
1717  * 0 or more than 0 is returned when validation is succeeded (either bypass
1718  * because of optional transport mode, or next index of the mathced secpath
1719  * state with the template.
1720  * -1 is returned when no matching template is found.
1721  * Otherwise "-2 - errored_index" is returned.
1722  */
1723 static inline int
1724 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1725                unsigned short family)
1726 {
1727         int idx = start;
1728
1729         if (tmpl->optional) {
1730                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1731                         return start;
1732         } else
1733                 start = -1;
1734         for (; idx < sp->len; idx++) {
1735                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1736                         return ++idx;
1737                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1738                         if (start == -1)
1739                                 start = -2-idx;
1740                         break;
1741                 }
1742         }
1743         return start;
1744 }
1745
1746 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1747                           unsigned int family, int reverse)
1748 {
1749         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1750         int err;
1751
1752         if (unlikely(afinfo == NULL))
1753                 return -EAFNOSUPPORT;
1754
1755         afinfo->decode_session(skb, fl, reverse);
1756         err = security_xfrm_decode_session(skb, &fl->secid);
1757         xfrm_policy_put_afinfo(afinfo);
1758         return err;
1759 }
1760 EXPORT_SYMBOL(__xfrm_decode_session);
1761
1762 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1763 {
1764         for (; k < sp->len; k++) {
1765                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1766                         *idxp = k;
1767                         return 1;
1768                 }
1769         }
1770
1771         return 0;
1772 }
1773
1774 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1775                         unsigned short family)
1776 {
1777         struct xfrm_policy *pol;
1778         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1779         int npols = 0;
1780         int xfrm_nr;
1781         int pi;
1782         int reverse;
1783         struct flowi fl;
1784         u8 fl_dir;
1785         int xerr_idx = -1;
1786
1787         reverse = dir & ~XFRM_POLICY_MASK;
1788         dir &= XFRM_POLICY_MASK;
1789         fl_dir = policy_to_flow_dir(dir);
1790
1791         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0)
1792                 return 0;
1793         nf_nat_decode_session(skb, &fl, family);
1794
1795         /* First, check used SA against their selectors. */
1796         if (skb->sp) {
1797                 int i;
1798
1799                 for (i=skb->sp->len-1; i>=0; i--) {
1800                         struct xfrm_state *x = skb->sp->xvec[i];
1801                         if (!xfrm_selector_match(&x->sel, &fl, family))
1802                                 return 0;
1803                 }
1804         }
1805
1806         pol = NULL;
1807         if (sk && sk->sk_policy[dir]) {
1808                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1809                 if (IS_ERR(pol))
1810                         return 0;
1811         }
1812
1813         if (!pol)
1814                 pol = flow_cache_lookup(&fl, family, fl_dir,
1815                                         xfrm_policy_lookup);
1816
1817         if (IS_ERR(pol))
1818                 return 0;
1819
1820         if (!pol) {
1821                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1822                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1823                         return 0;
1824                 }
1825                 return 1;
1826         }
1827
1828         pol->curlft.use_time = get_seconds();
1829
1830         pols[0] = pol;
1831         npols ++;
1832 #ifdef CONFIG_XFRM_SUB_POLICY
1833         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1834                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1835                                                     &fl, family,
1836                                                     XFRM_POLICY_IN);
1837                 if (pols[1]) {
1838                         if (IS_ERR(pols[1]))
1839                                 return 0;
1840                         pols[1]->curlft.use_time = get_seconds();
1841                         npols ++;
1842                 }
1843         }
1844 #endif
1845
1846         if (pol->action == XFRM_POLICY_ALLOW) {
1847                 struct sec_path *sp;
1848                 static struct sec_path dummy;
1849                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1850                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1851                 struct xfrm_tmpl **tpp = tp;
1852                 int ti = 0;
1853                 int i, k;
1854
1855                 if ((sp = skb->sp) == NULL)
1856                         sp = &dummy;
1857
1858                 for (pi = 0; pi < npols; pi++) {
1859                         if (pols[pi] != pol &&
1860                             pols[pi]->action != XFRM_POLICY_ALLOW)
1861                                 goto reject;
1862                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1863                                 goto reject_error;
1864                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1865                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1866                 }
1867                 xfrm_nr = ti;
1868                 if (npols > 1) {
1869                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1870                         tpp = stp;
1871                 }
1872
1873                 /* For each tunnel xfrm, find the first matching tmpl.
1874                  * For each tmpl before that, find corresponding xfrm.
1875                  * Order is _important_. Later we will implement
1876                  * some barriers, but at the moment barriers
1877                  * are implied between each two transformations.
1878                  */
1879                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1880                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1881                         if (k < 0) {
1882                                 if (k < -1)
1883                                         /* "-2 - errored_index" returned */
1884                                         xerr_idx = -(2+k);
1885                                 goto reject;
1886                         }
1887                 }
1888
1889                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1890                         goto reject;
1891
1892                 xfrm_pols_put(pols, npols);
1893                 return 1;
1894         }
1895
1896 reject:
1897         xfrm_secpath_reject(xerr_idx, skb, &fl);
1898 reject_error:
1899         xfrm_pols_put(pols, npols);
1900         return 0;
1901 }
1902 EXPORT_SYMBOL(__xfrm_policy_check);
1903
1904 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1905 {
1906         struct flowi fl;
1907
1908         if (xfrm_decode_session(skb, &fl, family) < 0)
1909                 return 0;
1910
1911         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1912 }
1913 EXPORT_SYMBOL(__xfrm_route_forward);
1914
1915 /* Optimize later using cookies and generation ids. */
1916
1917 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1918 {
1919         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1920          * to "-1" to force all XFRM destinations to get validated by
1921          * dst_ops->check on every use.  We do this because when a
1922          * normal route referenced by an XFRM dst is obsoleted we do
1923          * not go looking around for all parent referencing XFRM dsts
1924          * so that we can invalidate them.  It is just too much work.
1925          * Instead we make the checks here on every use.  For example:
1926          *
1927          *      XFRM dst A --> IPv4 dst X
1928          *
1929          * X is the "xdst->route" of A (X is also the "dst->path" of A
1930          * in this example).  If X is marked obsolete, "A" will not
1931          * notice.  That's what we are validating here via the
1932          * stale_bundle() check.
1933          *
1934          * When a policy's bundle is pruned, we dst_free() the XFRM
1935          * dst which causes it's ->obsolete field to be set to a
1936          * positive non-zero integer.  If an XFRM dst has been pruned
1937          * like this, we want to force a new route lookup.
1938          */
1939         if (dst->obsolete < 0 && !stale_bundle(dst))
1940                 return dst;
1941
1942         return NULL;
1943 }
1944
1945 static int stale_bundle(struct dst_entry *dst)
1946 {
1947         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1948 }
1949
1950 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1951 {
1952         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1953                 dst->dev = dev->nd_net->loopback_dev;
1954                 dev_hold(dst->dev);
1955                 dev_put(dev);
1956         }
1957 }
1958 EXPORT_SYMBOL(xfrm_dst_ifdown);
1959
1960 static void xfrm_link_failure(struct sk_buff *skb)
1961 {
1962         /* Impossible. Such dst must be popped before reaches point of failure. */
1963         return;
1964 }
1965
1966 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1967 {
1968         if (dst) {
1969                 if (dst->obsolete) {
1970                         dst_release(dst);
1971                         dst = NULL;
1972                 }
1973         }
1974         return dst;
1975 }
1976
1977 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1978 {
1979         struct dst_entry *dst, **dstp;
1980
1981         write_lock(&pol->lock);
1982         dstp = &pol->bundles;
1983         while ((dst=*dstp) != NULL) {
1984                 if (func(dst)) {
1985                         *dstp = dst->next;
1986                         dst->next = *gc_list_p;
1987                         *gc_list_p = dst;
1988                 } else {
1989                         dstp = &dst->next;
1990                 }
1991         }
1992         write_unlock(&pol->lock);
1993 }
1994
1995 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1996 {
1997         struct dst_entry *gc_list = NULL;
1998         int dir;
1999
2000         read_lock_bh(&xfrm_policy_lock);
2001         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2002                 struct xfrm_policy *pol;
2003                 struct hlist_node *entry;
2004                 struct hlist_head *table;
2005                 int i;
2006
2007                 hlist_for_each_entry(pol, entry,
2008                                      &xfrm_policy_inexact[dir], bydst)
2009                         prune_one_bundle(pol, func, &gc_list);
2010
2011                 table = xfrm_policy_bydst[dir].table;
2012                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2013                         hlist_for_each_entry(pol, entry, table + i, bydst)
2014                                 prune_one_bundle(pol, func, &gc_list);
2015                 }
2016         }
2017         read_unlock_bh(&xfrm_policy_lock);
2018
2019         while (gc_list) {
2020                 struct dst_entry *dst = gc_list;
2021                 gc_list = dst->next;
2022                 dst_free(dst);
2023         }
2024 }
2025
2026 static int unused_bundle(struct dst_entry *dst)
2027 {
2028         return !atomic_read(&dst->__refcnt);
2029 }
2030
2031 static void __xfrm_garbage_collect(void)
2032 {
2033         xfrm_prune_bundles(unused_bundle);
2034 }
2035
2036 static int xfrm_flush_bundles(void)
2037 {
2038         xfrm_prune_bundles(stale_bundle);
2039         return 0;
2040 }
2041
2042 static void xfrm_init_pmtu(struct dst_entry *dst)
2043 {
2044         do {
2045                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2046                 u32 pmtu, route_mtu_cached;
2047
2048                 pmtu = dst_mtu(dst->child);
2049                 xdst->child_mtu_cached = pmtu;
2050
2051                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2052
2053                 route_mtu_cached = dst_mtu(xdst->route);
2054                 xdst->route_mtu_cached = route_mtu_cached;
2055
2056                 if (pmtu > route_mtu_cached)
2057                         pmtu = route_mtu_cached;
2058
2059                 dst->metrics[RTAX_MTU-1] = pmtu;
2060         } while ((dst = dst->next));
2061 }
2062
2063 /* Check that the bundle accepts the flow and its components are
2064  * still valid.
2065  */
2066
2067 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2068                 struct flowi *fl, int family, int strict)
2069 {
2070         struct dst_entry *dst = &first->u.dst;
2071         struct xfrm_dst *last;
2072         u32 mtu;
2073
2074         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2075             (dst->dev && !netif_running(dst->dev)))
2076                 return 0;
2077 #ifdef CONFIG_XFRM_SUB_POLICY
2078         if (fl) {
2079                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2080                         return 0;
2081                 if (first->partner &&
2082                     !xfrm_selector_match(first->partner, fl, family))
2083                         return 0;
2084         }
2085 #endif
2086
2087         last = NULL;
2088
2089         do {
2090                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2091
2092                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2093                         return 0;
2094                 if (fl && pol &&
2095                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2096                         return 0;
2097                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2098                         return 0;
2099                 if (xdst->genid != dst->xfrm->genid)
2100                         return 0;
2101
2102                 if (strict && fl &&
2103                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2104                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2105                         return 0;
2106
2107                 mtu = dst_mtu(dst->child);
2108                 if (xdst->child_mtu_cached != mtu) {
2109                         last = xdst;
2110                         xdst->child_mtu_cached = mtu;
2111                 }
2112
2113                 if (!dst_check(xdst->route, xdst->route_cookie))
2114                         return 0;
2115                 mtu = dst_mtu(xdst->route);
2116                 if (xdst->route_mtu_cached != mtu) {
2117                         last = xdst;
2118                         xdst->route_mtu_cached = mtu;
2119                 }
2120
2121                 dst = dst->child;
2122         } while (dst->xfrm);
2123
2124         if (likely(!last))
2125                 return 1;
2126
2127         mtu = last->child_mtu_cached;
2128         for (;;) {
2129                 dst = &last->u.dst;
2130
2131                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2132                 if (mtu > last->route_mtu_cached)
2133                         mtu = last->route_mtu_cached;
2134                 dst->metrics[RTAX_MTU-1] = mtu;
2135
2136                 if (last == first)
2137                         break;
2138
2139                 last = (struct xfrm_dst *)last->u.dst.next;
2140                 last->child_mtu_cached = mtu;
2141         }
2142
2143         return 1;
2144 }
2145
2146 EXPORT_SYMBOL(xfrm_bundle_ok);
2147
2148 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2149 {
2150         int err = 0;
2151         if (unlikely(afinfo == NULL))
2152                 return -EINVAL;
2153         if (unlikely(afinfo->family >= NPROTO))
2154                 return -EAFNOSUPPORT;
2155         write_lock_bh(&xfrm_policy_afinfo_lock);
2156         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2157                 err = -ENOBUFS;
2158         else {
2159                 struct dst_ops *dst_ops = afinfo->dst_ops;
2160                 if (likely(dst_ops->kmem_cachep == NULL))
2161                         dst_ops->kmem_cachep = xfrm_dst_cache;
2162                 if (likely(dst_ops->check == NULL))
2163                         dst_ops->check = xfrm_dst_check;
2164                 if (likely(dst_ops->negative_advice == NULL))
2165                         dst_ops->negative_advice = xfrm_negative_advice;
2166                 if (likely(dst_ops->link_failure == NULL))
2167                         dst_ops->link_failure = xfrm_link_failure;
2168                 if (likely(afinfo->garbage_collect == NULL))
2169                         afinfo->garbage_collect = __xfrm_garbage_collect;
2170                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2171         }
2172         write_unlock_bh(&xfrm_policy_afinfo_lock);
2173         return err;
2174 }
2175 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2176
2177 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2178 {
2179         int err = 0;
2180         if (unlikely(afinfo == NULL))
2181                 return -EINVAL;
2182         if (unlikely(afinfo->family >= NPROTO))
2183                 return -EAFNOSUPPORT;
2184         write_lock_bh(&xfrm_policy_afinfo_lock);
2185         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2186                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2187                         err = -EINVAL;
2188                 else {
2189                         struct dst_ops *dst_ops = afinfo->dst_ops;
2190                         xfrm_policy_afinfo[afinfo->family] = NULL;
2191                         dst_ops->kmem_cachep = NULL;
2192                         dst_ops->check = NULL;
2193                         dst_ops->negative_advice = NULL;
2194                         dst_ops->link_failure = NULL;
2195                         afinfo->garbage_collect = NULL;
2196                 }
2197         }
2198         write_unlock_bh(&xfrm_policy_afinfo_lock);
2199         return err;
2200 }
2201 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2202
2203 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2204 {
2205         struct xfrm_policy_afinfo *afinfo;
2206         if (unlikely(family >= NPROTO))
2207                 return NULL;
2208         read_lock(&xfrm_policy_afinfo_lock);
2209         afinfo = xfrm_policy_afinfo[family];
2210         if (unlikely(!afinfo))
2211                 read_unlock(&xfrm_policy_afinfo_lock);
2212         return afinfo;
2213 }
2214
2215 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2216 {
2217         read_unlock(&xfrm_policy_afinfo_lock);
2218 }
2219
2220 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2221 {
2222         struct net_device *dev = ptr;
2223
2224         if (dev->nd_net != &init_net)
2225                 return NOTIFY_DONE;
2226
2227         switch (event) {
2228         case NETDEV_DOWN:
2229                 xfrm_flush_bundles();
2230         }
2231         return NOTIFY_DONE;
2232 }
2233
2234 static struct notifier_block xfrm_dev_notifier = {
2235         xfrm_dev_event,
2236         NULL,
2237         0
2238 };
2239
2240 static void __init xfrm_policy_init(void)
2241 {
2242         unsigned int hmask, sz;
2243         int dir;
2244
2245         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2246                                            sizeof(struct xfrm_dst),
2247                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2248                                            NULL);
2249
2250         hmask = 8 - 1;
2251         sz = (hmask+1) * sizeof(struct hlist_head);
2252
2253         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2254         xfrm_idx_hmask = hmask;
2255         if (!xfrm_policy_byidx)
2256                 panic("XFRM: failed to allocate byidx hash\n");
2257
2258         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2259                 struct xfrm_policy_hash *htab;
2260
2261                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2262
2263                 htab = &xfrm_policy_bydst[dir];
2264                 htab->table = xfrm_hash_alloc(sz);
2265                 htab->hmask = hmask;
2266                 if (!htab->table)
2267                         panic("XFRM: failed to allocate bydst hash\n");
2268         }
2269
2270         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2271         register_netdevice_notifier(&xfrm_dev_notifier);
2272 }
2273
2274 void __init xfrm_init(void)
2275 {
2276         xfrm_state_init();
2277         xfrm_policy_init();
2278         xfrm_input_init();
2279 }
2280
2281 #ifdef CONFIG_AUDITSYSCALL
2282 static inline void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2283                                                 struct audit_buffer *audit_buf)
2284 {
2285         struct xfrm_sec_ctx *ctx = xp->security;
2286         struct xfrm_selector *sel = &xp->selector;
2287
2288         if (ctx)
2289                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2290                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2291
2292         switch(sel->family) {
2293         case AF_INET:
2294                 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2295                                  NIPQUAD(sel->saddr.a4));
2296                 if (sel->prefixlen_s != 32)
2297                         audit_log_format(audit_buf, " src_prefixlen=%d",
2298                                          sel->prefixlen_s);
2299                 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2300                                  NIPQUAD(sel->daddr.a4));
2301                 if (sel->prefixlen_d != 32)
2302                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2303                                          sel->prefixlen_d);
2304                 break;
2305         case AF_INET6:
2306                 audit_log_format(audit_buf, " src=" NIP6_FMT,
2307                                  NIP6(*(struct in6_addr *)sel->saddr.a6));
2308                 if (sel->prefixlen_s != 128)
2309                         audit_log_format(audit_buf, " src_prefixlen=%d",
2310                                          sel->prefixlen_s);
2311                 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2312                                  NIP6(*(struct in6_addr *)sel->daddr.a6));
2313                 if (sel->prefixlen_d != 128)
2314                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2315                                          sel->prefixlen_d);
2316                 break;
2317         }
2318 }
2319
2320 void
2321 xfrm_audit_policy_add(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2322 {
2323         struct audit_buffer *audit_buf;
2324         extern int audit_enabled;
2325
2326         if (audit_enabled == 0)
2327                 return;
2328         audit_buf = xfrm_audit_start(auid, sid);
2329         if (audit_buf == NULL)
2330                 return;
2331         audit_log_format(audit_buf, " op=SPD-add res=%u", result);
2332         xfrm_audit_common_policyinfo(xp, audit_buf);
2333         audit_log_end(audit_buf);
2334 }
2335 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2336
2337 void
2338 xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, u32 auid, u32 sid)
2339 {
2340         struct audit_buffer *audit_buf;
2341         extern int audit_enabled;
2342
2343         if (audit_enabled == 0)
2344                 return;
2345         audit_buf = xfrm_audit_start(auid, sid);
2346         if (audit_buf == NULL)
2347                 return;
2348         audit_log_format(audit_buf, " op=SPD-delete res=%u", result);
2349         xfrm_audit_common_policyinfo(xp, audit_buf);
2350         audit_log_end(audit_buf);
2351 }
2352 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2353 #endif
2354
2355 #ifdef CONFIG_XFRM_MIGRATE
2356 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2357                                        struct xfrm_selector *sel_tgt)
2358 {
2359         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2360                 if (sel_tgt->family == sel_cmp->family &&
2361                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2362                                   sel_cmp->family) == 0 &&
2363                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2364                                   sel_cmp->family) == 0 &&
2365                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2366                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2367                         return 1;
2368                 }
2369         } else {
2370                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2371                         return 1;
2372                 }
2373         }
2374         return 0;
2375 }
2376
2377 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2378                                                      u8 dir, u8 type)
2379 {
2380         struct xfrm_policy *pol, *ret = NULL;
2381         struct hlist_node *entry;
2382         struct hlist_head *chain;
2383         u32 priority = ~0U;
2384
2385         read_lock_bh(&xfrm_policy_lock);
2386         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2387         hlist_for_each_entry(pol, entry, chain, bydst) {
2388                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2389                     pol->type == type) {
2390                         ret = pol;
2391                         priority = ret->priority;
2392                         break;
2393                 }
2394         }
2395         chain = &xfrm_policy_inexact[dir];
2396         hlist_for_each_entry(pol, entry, chain, bydst) {
2397                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2398                     pol->type == type &&
2399                     pol->priority < priority) {
2400                         ret = pol;
2401                         break;
2402                 }
2403         }
2404
2405         if (ret)
2406                 xfrm_pol_hold(ret);
2407
2408         read_unlock_bh(&xfrm_policy_lock);
2409
2410         return ret;
2411 }
2412
2413 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2414 {
2415         int match = 0;
2416
2417         if (t->mode == m->mode && t->id.proto == m->proto &&
2418             (m->reqid == 0 || t->reqid == m->reqid)) {
2419                 switch (t->mode) {
2420                 case XFRM_MODE_TUNNEL:
2421                 case XFRM_MODE_BEET:
2422                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2423                                           m->old_family) == 0 &&
2424                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2425                                           m->old_family) == 0) {
2426                                 match = 1;
2427                         }
2428                         break;
2429                 case XFRM_MODE_TRANSPORT:
2430                         /* in case of transport mode, template does not store
2431                            any IP addresses, hence we just compare mode and
2432                            protocol */
2433                         match = 1;
2434                         break;
2435                 default:
2436                         break;
2437                 }
2438         }
2439         return match;
2440 }
2441
2442 /* update endpoint address(es) of template(s) */
2443 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2444                                struct xfrm_migrate *m, int num_migrate)
2445 {
2446         struct xfrm_migrate *mp;
2447         struct dst_entry *dst;
2448         int i, j, n = 0;
2449
2450         write_lock_bh(&pol->lock);
2451         if (unlikely(pol->dead)) {
2452                 /* target policy has been deleted */
2453                 write_unlock_bh(&pol->lock);
2454                 return -ENOENT;
2455         }
2456
2457         for (i = 0; i < pol->xfrm_nr; i++) {
2458                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2459                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2460                                 continue;
2461                         n++;
2462                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2463                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2464                                 continue;
2465                         /* update endpoints */
2466                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2467                                sizeof(pol->xfrm_vec[i].id.daddr));
2468                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2469                                sizeof(pol->xfrm_vec[i].saddr));
2470                         pol->xfrm_vec[i].encap_family = mp->new_family;
2471                         /* flush bundles */
2472                         while ((dst = pol->bundles) != NULL) {
2473                                 pol->bundles = dst->next;
2474                                 dst_free(dst);
2475                         }
2476                 }
2477         }
2478
2479         write_unlock_bh(&pol->lock);
2480
2481         if (!n)
2482                 return -ENODATA;
2483
2484         return 0;
2485 }
2486
2487 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2488 {
2489         int i, j;
2490
2491         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2492                 return -EINVAL;
2493
2494         for (i = 0; i < num_migrate; i++) {
2495                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2496                                    m[i].old_family) == 0) &&
2497                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2498                                    m[i].old_family) == 0))
2499                         return -EINVAL;
2500                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2501                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2502                         return -EINVAL;
2503
2504                 /* check if there is any duplicated entry */
2505                 for (j = i + 1; j < num_migrate; j++) {
2506                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2507                                     sizeof(m[i].old_daddr)) &&
2508                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2509                                     sizeof(m[i].old_saddr)) &&
2510                             m[i].proto == m[j].proto &&
2511                             m[i].mode == m[j].mode &&
2512                             m[i].reqid == m[j].reqid &&
2513                             m[i].old_family == m[j].old_family)
2514                                 return -EINVAL;
2515                 }
2516         }
2517
2518         return 0;
2519 }
2520
2521 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2522                  struct xfrm_migrate *m, int num_migrate)
2523 {
2524         int i, err, nx_cur = 0, nx_new = 0;
2525         struct xfrm_policy *pol = NULL;
2526         struct xfrm_state *x, *xc;
2527         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2528         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2529         struct xfrm_migrate *mp;
2530
2531         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2532                 goto out;
2533
2534         /* Stage 1 - find policy */
2535         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2536                 err = -ENOENT;
2537                 goto out;
2538         }
2539
2540         /* Stage 2 - find and update state(s) */
2541         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2542                 if ((x = xfrm_migrate_state_find(mp))) {
2543                         x_cur[nx_cur] = x;
2544                         nx_cur++;
2545                         if ((xc = xfrm_state_migrate(x, mp))) {
2546                                 x_new[nx_new] = xc;
2547                                 nx_new++;
2548                         } else {
2549                                 err = -ENODATA;
2550                                 goto restore_state;
2551                         }
2552                 }
2553         }
2554
2555         /* Stage 3 - update policy */
2556         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2557                 goto restore_state;
2558
2559         /* Stage 4 - delete old state(s) */
2560         if (nx_cur) {
2561                 xfrm_states_put(x_cur, nx_cur);
2562                 xfrm_states_delete(x_cur, nx_cur);
2563         }
2564
2565         /* Stage 5 - announce */
2566         km_migrate(sel, dir, type, m, num_migrate);
2567
2568         xfrm_pol_put(pol);
2569
2570         return 0;
2571 out:
2572         return err;
2573
2574 restore_state:
2575         if (pol)
2576                 xfrm_pol_put(pol);
2577         if (nx_cur)
2578                 xfrm_states_put(x_cur, nx_cur);
2579         if (nx_new)
2580                 xfrm_states_delete(x_new, nx_new);
2581
2582         return err;
2583 }
2584 EXPORT_SYMBOL(xfrm_migrate);
2585 #endif