err.no Git - 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 <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/xfrm.h>
  30 #include <net/ip.h>
  31 #ifdef CONFIG_XFRM_STATISTICS
  32 #include <net/snmp.h>
  33 #endif
  34
  35 #include "xfrm_hash.h"
  36
  37 int sysctl_xfrm_larval_drop __read_mostly;
  38
  39 #ifdef CONFIG_XFRM_STATISTICS
  40 DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
  41 EXPORT_SYMBOL(xfrm_statistics);
  42 #endif
  43
  44 DEFINE_MUTEX(xfrm_cfg_mutex);
  45 EXPORT_SYMBOL(xfrm_cfg_mutex);
  46
  47 static DEFINE_RWLOCK(xfrm_policy_lock);
  48
  49 static struct list_head xfrm_policy_bytype[XFRM_POLICY_TYPE_MAX];
  50 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
  51 EXPORT_SYMBOL(xfrm_policy_count);
  52
  53 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  54 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  55
  56 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  57
  58 static struct work_struct xfrm_policy_gc_work;
  59 static HLIST_HEAD(xfrm_policy_gc_list);
  60 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
  61
  62 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  63 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  64 static void xfrm_init_pmtu(struct dst_entry *dst);
  65
  66 static inline int
  67 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  68 {
  69         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  70                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  71                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  72                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  73                 (fl->proto == sel->proto || !sel->proto) &&
  74                 (fl->oif == sel->ifindex || !sel->ifindex);
  75 }
  76
  77 static inline int
  78 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
  79 {
  80         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  81                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  82                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
  83                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
  84                 (fl->proto == sel->proto || !sel->proto) &&
  85                 (fl->oif == sel->ifindex || !sel->ifindex);
  86 }
  87
  88 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
  89                     unsigned short family)
  90 {
  91         switch (family) {
  92         case AF_INET:
  93                 return __xfrm4_selector_match(sel, fl);
  94         case AF_INET6:
  95                 return __xfrm6_selector_match(sel, fl);
  96         }
  97         return 0;
  98 }
  99
 100 static inline struct dst_entry *__xfrm_dst_lookup(int tos,
 101                                                   xfrm_address_t *saddr,
 102                                                   xfrm_address_t *daddr,
 103                                                   int family)
 104 {
 105         struct xfrm_policy_afinfo *afinfo;
 106         struct dst_entry *dst;
 107
 108         afinfo = xfrm_policy_get_afinfo(family);
 109         if (unlikely(afinfo == NULL))
 110                 return ERR_PTR(-EAFNOSUPPORT);
 111
 112         dst = afinfo->dst_lookup(tos, saddr, daddr);
 113
 114         xfrm_policy_put_afinfo(afinfo);
 115
 116         return dst;
 117 }
 118
 119 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 120                                                 xfrm_address_t *prev_saddr,
 121                                                 xfrm_address_t *prev_daddr,
 122                                                 int family)
 123 {
 124         xfrm_address_t *saddr = &x->props.saddr;
 125         xfrm_address_t *daddr = &x->id.daddr;
 126         struct dst_entry *dst;
 127
 128         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 129                 saddr = x->coaddr;
 130                 daddr = prev_daddr;
 131         }
 132         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 133                 saddr = prev_saddr;
 134                 daddr = x->coaddr;
 135         }
 136
 137         dst = __xfrm_dst_lookup(tos, saddr, daddr, family);
 138
 139         if (!IS_ERR(dst)) {
 140                 if (prev_saddr != saddr)
 141                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 142                 if (prev_daddr != daddr)
 143                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 144         }
 145
 146         return dst;
 147 }
 148
 149 static inline unsigned long make_jiffies(long secs)
 150 {
 151         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 152                 return MAX_SCHEDULE_TIMEOUT-1;
 153         else
 154                 return secs*HZ;
 155 }
 156
 157 static void xfrm_policy_timer(unsigned long data)
 158 {
 159         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 160         unsigned long now = get_seconds();
 161         long next = LONG_MAX;
 162         int warn = 0;
 163         int dir;
 164
 165         read_lock(&xp->lock);
 166
 167         if (xp->dead)
 168                 goto out;
 169
 170         dir = xfrm_policy_id2dir(xp->index);
 171
 172         if (xp->lft.hard_add_expires_seconds) {
 173                 long tmo = xp->lft.hard_add_expires_seconds +
 174                         xp->curlft.add_time - now;
 175                 if (tmo <= 0)
 176                         goto expired;
 177                 if (tmo < next)
 178                         next = tmo;
 179         }
 180         if (xp->lft.hard_use_expires_seconds) {
 181                 long tmo = xp->lft.hard_use_expires_seconds +
 182                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 183                 if (tmo <= 0)
 184                         goto expired;
 185                 if (tmo < next)
 186                         next = tmo;
 187         }
 188         if (xp->lft.soft_add_expires_seconds) {
 189                 long tmo = xp->lft.soft_add_expires_seconds +
 190                         xp->curlft.add_time - now;
 191                 if (tmo <= 0) {
 192                         warn = 1;
 193                         tmo = XFRM_KM_TIMEOUT;
 194                 }
 195                 if (tmo < next)
 196                         next = tmo;
 197         }
 198         if (xp->lft.soft_use_expires_seconds) {
 199                 long tmo = xp->lft.soft_use_expires_seconds +
 200                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 201                 if (tmo <= 0) {
 202                         warn = 1;
 203                         tmo = XFRM_KM_TIMEOUT;
 204                 }
 205                 if (tmo < next)
 206                         next = tmo;
 207         }
 208
 209         if (warn)
 210                 km_policy_expired(xp, dir, 0, 0);
 211         if (next != LONG_MAX &&
 212             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 213                 xfrm_pol_hold(xp);
 214
 215 out:
 216         read_unlock(&xp->lock);
 217         xfrm_pol_put(xp);
 218         return;
 219
 220 expired:
 221         read_unlock(&xp->lock);
 222         if (!xfrm_policy_delete(xp, dir))
 223                 km_policy_expired(xp, dir, 1, 0);
 224         xfrm_pol_put(xp);
 225 }
 226
 227
 228 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 229  * SPD calls.
 230  */
 231
 232 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 233 {
 234         struct xfrm_policy *policy;
 235
 236         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 237
 238         if (policy) {
 239                 INIT_LIST_HEAD(&policy->bytype);
 240                 INIT_HLIST_NODE(&policy->bydst);
 241                 INIT_HLIST_NODE(&policy->byidx);
 242                 rwlock_init(&policy->lock);
 243                 atomic_set(&policy->refcnt, 1);
 244                 setup_timer(&policy->timer, xfrm_policy_timer,
 245                                 (unsigned long)policy);
 246         }
 247         return policy;
 248 }
 249 EXPORT_SYMBOL(xfrm_policy_alloc);
 250
 251 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 252
 253 void xfrm_policy_destroy(struct xfrm_policy *policy)
 254 {
 255         BUG_ON(!policy->dead);
 256
 257         BUG_ON(policy->bundles);
 258
 259         if (del_timer(&policy->timer))
 260                 BUG();
 261
 262         write_lock_bh(&xfrm_policy_lock);
 263         list_del(&policy->bytype);
 264         write_unlock_bh(&xfrm_policy_lock);
 265
 266         security_xfrm_policy_free(policy->security);
 267         kfree(policy);
 268 }
 269 EXPORT_SYMBOL(xfrm_policy_destroy);
 270
 271 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
 272 {
 273         struct dst_entry *dst;
 274
 275         while ((dst = policy->bundles) != NULL) {
 276                 policy->bundles = dst->next;
 277                 dst_free(dst);
 278         }
 279
 280         if (del_timer(&policy->timer))
 281                 atomic_dec(&policy->refcnt);
 282
 283         if (atomic_read(&policy->refcnt) > 1)
 284                 flow_cache_flush();
 285
 286         xfrm_pol_put(policy);
 287 }
 288
 289 static void xfrm_policy_gc_task(struct work_struct *work)
 290 {
 291         struct xfrm_policy *policy;
 292         struct hlist_node *entry, *tmp;
 293         struct hlist_head gc_list;
 294
 295         spin_lock_bh(&xfrm_policy_gc_lock);
 296         gc_list.first = xfrm_policy_gc_list.first;
 297         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
 298         spin_unlock_bh(&xfrm_policy_gc_lock);
 299
 300         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
 301                 xfrm_policy_gc_kill(policy);
 302 }
 303
 304 /* Rule must be locked. Release descentant resources, announce
 305  * entry dead. The rule must be unlinked from lists to the moment.
 306  */
 307
 308 static void xfrm_policy_kill(struct xfrm_policy *policy)
 309 {
 310         int dead;
 311
 312         write_lock_bh(&policy->lock);
 313         dead = policy->dead;
 314         policy->dead = 1;
 315         write_unlock_bh(&policy->lock);
 316
 317         if (unlikely(dead)) {
 318                 WARN_ON(1);
 319                 return;
 320         }
 321
 322         spin_lock(&xfrm_policy_gc_lock);
 323         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
 324         spin_unlock(&xfrm_policy_gc_lock);
 325
 326         schedule_work(&xfrm_policy_gc_work);
 327 }
 328
 329 struct xfrm_policy_hash {
 330         struct hlist_head       *table;
 331         unsigned int            hmask;
 332 };
 333
 334 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
 335 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
 336 static struct hlist_head *xfrm_policy_byidx __read_mostly;
 337 static unsigned int xfrm_idx_hmask __read_mostly;
 338 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 339
 340 static inline unsigned int idx_hash(u32 index)
 341 {
 342         return __idx_hash(index, xfrm_idx_hmask);
 343 }
 344
 345 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
 346 {
 347         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 348         unsigned int hash = __sel_hash(sel, family, hmask);
 349
 350         return (hash == hmask + 1 ?
 351                 &xfrm_policy_inexact[dir] :
 352                 xfrm_policy_bydst[dir].table + hash);
 353 }
 354
 355 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
 356 {
 357         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 358         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 359
 360         return xfrm_policy_bydst[dir].table + hash;
 361 }
 362
 363 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 364                                    struct hlist_head *ndsttable,
 365                                    unsigned int nhashmask)
 366 {
 367         struct hlist_node *entry, *tmp, *entry0 = NULL;
 368         struct xfrm_policy *pol;
 369         unsigned int h0 = 0;
 370
 371 redo:
 372         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 373                 unsigned int h;
 374
 375                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 376                                 pol->family, nhashmask);
 377                 if (!entry0) {
 378                         hlist_del(entry);
 379                         hlist_add_head(&pol->bydst, ndsttable+h);
 380                         h0 = h;
 381                 } else {
 382                         if (h != h0)
 383                                 continue;
 384                         hlist_del(entry);
 385                         hlist_add_after(entry0, &pol->bydst);
 386                 }
 387                 entry0 = entry;
 388         }
 389         if (!hlist_empty(list)) {
 390                 entry0 = NULL;
 391                 goto redo;
 392         }
 393 }
 394
 395 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 396                                    struct hlist_head *nidxtable,
 397                                    unsigned int nhashmask)
 398 {
 399         struct hlist_node *entry, *tmp;
 400         struct xfrm_policy *pol;
 401
 402         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 403                 unsigned int h;
 404
 405                 h = __idx_hash(pol->index, nhashmask);
 406                 hlist_add_head(&pol->byidx, nidxtable+h);
 407         }
 408 }
 409
 410 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 411 {
 412         return ((old_hmask + 1) << 1) - 1;
 413 }
 414
 415 static void xfrm_bydst_resize(int dir)
 416 {
 417         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 418         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 419         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 420         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
 421         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 422         int i;
 423
 424         if (!ndst)
 425                 return;
 426
 427         write_lock_bh(&xfrm_policy_lock);
 428
 429         for (i = hmask; i >= 0; i--)
 430                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 431
 432         xfrm_policy_bydst[dir].table = ndst;
 433         xfrm_policy_bydst[dir].hmask = nhashmask;
 434
 435         write_unlock_bh(&xfrm_policy_lock);
 436
 437         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 438 }
 439
 440 static void xfrm_byidx_resize(int total)
 441 {
 442         unsigned int hmask = xfrm_idx_hmask;
 443         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 444         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 445         struct hlist_head *oidx = xfrm_policy_byidx;
 446         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 447         int i;
 448
 449         if (!nidx)
 450                 return;
 451
 452         write_lock_bh(&xfrm_policy_lock);
 453
 454         for (i = hmask; i >= 0; i--)
 455                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 456
 457         xfrm_policy_byidx = nidx;
 458         xfrm_idx_hmask = nhashmask;
 459
 460         write_unlock_bh(&xfrm_policy_lock);
 461
 462         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 463 }
 464
 465 static inline int xfrm_bydst_should_resize(int dir, int *total)
 466 {
 467         unsigned int cnt = xfrm_policy_count[dir];
 468         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
 469
 470         if (total)
 471                 *total += cnt;
 472
 473         if ((hmask + 1) < xfrm_policy_hashmax &&
 474             cnt > hmask)
 475                 return 1;
 476
 477         return 0;
 478 }
 479
 480 static inline int xfrm_byidx_should_resize(int total)
 481 {
 482         unsigned int hmask = xfrm_idx_hmask;
 483
 484         if ((hmask + 1) < xfrm_policy_hashmax &&
 485             total > hmask)
 486                 return 1;
 487
 488         return 0;
 489 }
 490
 491 void xfrm_spd_getinfo(struct xfrmk_spdinfo *si)
 492 {
 493         read_lock_bh(&xfrm_policy_lock);
 494         si->incnt = xfrm_policy_count[XFRM_POLICY_IN];
 495         si->outcnt = xfrm_policy_count[XFRM_POLICY_OUT];
 496         si->fwdcnt = xfrm_policy_count[XFRM_POLICY_FWD];
 497         si->inscnt = xfrm_policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 498         si->outscnt = xfrm_policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 499         si->fwdscnt = xfrm_policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 500         si->spdhcnt = xfrm_idx_hmask;
 501         si->spdhmcnt = xfrm_policy_hashmax;
 502         read_unlock_bh(&xfrm_policy_lock);
 503 }
 504 EXPORT_SYMBOL(xfrm_spd_getinfo);
 505
 506 static DEFINE_MUTEX(hash_resize_mutex);
 507 static void xfrm_hash_resize(struct work_struct *__unused)
 508 {
 509         int dir, total;
 510
 511         mutex_lock(&hash_resize_mutex);
 512
 513         total = 0;
 514         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 515                 if (xfrm_bydst_should_resize(dir, &total))
 516                         xfrm_bydst_resize(dir);
 517         }
 518         if (xfrm_byidx_should_resize(total))
 519                 xfrm_byidx_resize(total);
 520
 521         mutex_unlock(&hash_resize_mutex);
 522 }
 523
 524 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
 525
 526 /* Generate new index... KAME seems to generate them ordered by cost
 527  * of an absolute inpredictability of ordering of rules. This will not pass. */
 528 static u32 xfrm_gen_index(u8 type, int dir)
 529 {
 530         static u32 idx_generator;
 531
 532         for (;;) {
 533                 struct hlist_node *entry;
 534                 struct hlist_head *list;
 535                 struct xfrm_policy *p;
 536                 u32 idx;
 537                 int found;
 538
 539                 idx = (idx_generator | dir);
 540                 idx_generator += 8;
 541                 if (idx == 0)
 542                         idx = 8;
 543                 list = xfrm_policy_byidx + idx_hash(idx);
 544                 found = 0;
 545                 hlist_for_each_entry(p, entry, list, byidx) {
 546                         if (p->index == idx) {
 547                                 found = 1;
 548                                 break;
 549                         }
 550                 }
 551                 if (!found)
 552                         return idx;
 553         }
 554 }
 555
 556 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 557 {
 558         u32 *p1 = (u32 *) s1;
 559         u32 *p2 = (u32 *) s2;
 560         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 561         int i;
 562
 563         for (i = 0; i < len; i++) {
 564                 if (p1[i] != p2[i])
 565                         return 1;
 566         }
 567
 568         return 0;
 569 }
 570
 571 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 572 {
 573         struct xfrm_policy *pol;
 574         struct xfrm_policy *delpol;
 575         struct hlist_head *chain;
 576         struct hlist_node *entry, *newpos;
 577         struct dst_entry *gc_list;
 578
 579         write_lock_bh(&xfrm_policy_lock);
 580         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
 581         delpol = NULL;
 582         newpos = NULL;
 583         hlist_for_each_entry(pol, entry, chain, bydst) {
 584                 if (pol->type == policy->type &&
 585                     !selector_cmp(&pol->selector, &policy->selector) &&
 586                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 587                     !WARN_ON(delpol)) {
 588                         if (excl) {
 589                                 write_unlock_bh(&xfrm_policy_lock);
 590                                 return -EEXIST;
 591                         }
 592                         delpol = pol;
 593                         if (policy->priority > pol->priority)
 594                                 continue;
 595                 } else if (policy->priority >= pol->priority) {
 596                         newpos = &pol->bydst;
 597                         continue;
 598                 }
 599                 if (delpol)
 600                         break;
 601         }
 602         if (newpos)
 603                 hlist_add_after(newpos, &policy->bydst);
 604         else
 605                 hlist_add_head(&policy->bydst, chain);
 606         xfrm_pol_hold(policy);
 607         xfrm_policy_count[dir]++;
 608         atomic_inc(&flow_cache_genid);
 609         if (delpol) {
 610                 hlist_del(&delpol->bydst);
 611                 hlist_del(&delpol->byidx);
 612                 xfrm_policy_count[dir]--;
 613         }
 614         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
 615         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
 616         policy->curlft.add_time = get_seconds();
 617         policy->curlft.use_time = 0;
 618         if (!mod_timer(&policy->timer, jiffies + HZ))
 619                 xfrm_pol_hold(policy);
 620         list_add_tail(&policy->bytype, &xfrm_policy_bytype[policy->type]);
 621         write_unlock_bh(&xfrm_policy_lock);
 622
 623         if (delpol)
 624                 xfrm_policy_kill(delpol);
 625         else if (xfrm_bydst_should_resize(dir, NULL))
 626                 schedule_work(&xfrm_hash_work);
 627
 628         read_lock_bh(&xfrm_policy_lock);
 629         gc_list = NULL;
 630         entry = &policy->bydst;
 631         hlist_for_each_entry_continue(policy, entry, bydst) {
 632                 struct dst_entry *dst;
 633
 634                 write_lock(&policy->lock);
 635                 dst = policy->bundles;
 636                 if (dst) {
 637                         struct dst_entry *tail = dst;
 638                         while (tail->next)
 639                                 tail = tail->next;
 640                         tail->next = gc_list;
 641                         gc_list = dst;
 642
 643                         policy->bundles = NULL;
 644                 }
 645                 write_unlock(&policy->lock);
 646         }
 647         read_unlock_bh(&xfrm_policy_lock);
 648
 649         while (gc_list) {
 650                 struct dst_entry *dst = gc_list;
 651
 652                 gc_list = dst->next;
 653                 dst_free(dst);
 654         }
 655
 656         return 0;
 657 }
 658 EXPORT_SYMBOL(xfrm_policy_insert);
 659
 660 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
 661                                           struct xfrm_selector *sel,
 662                                           struct xfrm_sec_ctx *ctx, int delete,
 663                                           int *err)
 664 {
 665         struct xfrm_policy *pol, *ret;
 666         struct hlist_head *chain;
 667         struct hlist_node *entry;
 668
 669         *err = 0;
 670         write_lock_bh(&xfrm_policy_lock);
 671         chain = policy_hash_bysel(sel, sel->family, dir);
 672         ret = NULL;
 673         hlist_for_each_entry(pol, entry, chain, bydst) {
 674                 if (pol->type == type &&
 675                     !selector_cmp(sel, &pol->selector) &&
 676                     xfrm_sec_ctx_match(ctx, pol->security)) {
 677                         xfrm_pol_hold(pol);
 678                         if (delete) {
 679                                 *err = security_xfrm_policy_delete(
 680                                                                 pol->security);
 681                                 if (*err) {
 682                                         write_unlock_bh(&xfrm_policy_lock);
 683                                         return pol;
 684                                 }
 685                                 hlist_del(&pol->bydst);
 686                                 hlist_del(&pol->byidx);
 687                                 xfrm_policy_count[dir]--;
 688                         }
 689                         ret = pol;
 690                         break;
 691                 }
 692         }
 693         write_unlock_bh(&xfrm_policy_lock);
 694
 695         if (ret && delete) {
 696                 atomic_inc(&flow_cache_genid);
 697                 xfrm_policy_kill(ret);
 698         }
 699         return ret;
 700 }
 701 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 702
 703 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete,
 704                                      int *err)
 705 {
 706         struct xfrm_policy *pol, *ret;
 707         struct hlist_head *chain;
 708         struct hlist_node *entry;
 709
 710         *err = -ENOENT;
 711         if (xfrm_policy_id2dir(id) != dir)
 712                 return NULL;
 713
 714         *err = 0;
 715         write_lock_bh(&xfrm_policy_lock);
 716         chain = xfrm_policy_byidx + idx_hash(id);
 717         ret = NULL;
 718         hlist_for_each_entry(pol, entry, chain, byidx) {
 719                 if (pol->type == type && pol->index == id) {
 720                         xfrm_pol_hold(pol);
 721                         if (delete) {
 722                                 *err = security_xfrm_policy_delete(
 723                                                                 pol->security);
 724                                 if (*err) {
 725                                         write_unlock_bh(&xfrm_policy_lock);
 726                                         return pol;
 727                                 }
 728                                 hlist_del(&pol->bydst);
 729                                 hlist_del(&pol->byidx);
 730                                 xfrm_policy_count[dir]--;
 731                         }
 732                         ret = pol;
 733                         break;
 734                 }
 735         }
 736         write_unlock_bh(&xfrm_policy_lock);
 737
 738         if (ret && delete) {
 739                 atomic_inc(&flow_cache_genid);
 740                 xfrm_policy_kill(ret);
 741         }
 742         return ret;
 743 }
 744 EXPORT_SYMBOL(xfrm_policy_byid);
 745
 746 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 747 static inline int
 748 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 749 {
 750         int dir, err = 0;
 751
 752         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 753                 struct xfrm_policy *pol;
 754                 struct hlist_node *entry;
 755                 int i;
 756
 757                 hlist_for_each_entry(pol, entry,
 758                                      &xfrm_policy_inexact[dir], bydst) {
 759                         if (pol->type != type)
 760                                 continue;
 761                         err = security_xfrm_policy_delete(pol->security);
 762                         if (err) {
 763                                 xfrm_audit_policy_delete(pol, 0,
 764                                                          audit_info->loginuid,
 765                                                          audit_info->secid);
 766                                 return err;
 767                         }
 768                 }
 769                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 770                         hlist_for_each_entry(pol, entry,
 771                                              xfrm_policy_bydst[dir].table + i,
 772                                              bydst) {
 773                                 if (pol->type != type)
 774                                         continue;
 775                                 err = security_xfrm_policy_delete(
 776                                                                 pol->security);
 777                                 if (err) {
 778                                         xfrm_audit_policy_delete(pol, 0,
 779                                                         audit_info->loginuid,
 780                                                         audit_info->secid);
 781                                         return err;
 782                                 }
 783                         }
 784                 }
 785         }
 786         return err;
 787 }
 788 #else
 789 static inline int
 790 xfrm_policy_flush_secctx_check(u8 type, struct xfrm_audit *audit_info)
 791 {
 792         return 0;
 793 }
 794 #endif
 795
 796 int xfrm_policy_flush(u8 type, struct xfrm_audit *audit_info)
 797 {
 798         int dir, err = 0;
 799
 800         write_lock_bh(&xfrm_policy_lock);
 801
 802         err = xfrm_policy_flush_secctx_check(type, audit_info);
 803         if (err)
 804                 goto out;
 805
 806         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 807                 struct xfrm_policy *pol;
 808                 struct hlist_node *entry;
 809                 int i, killed;
 810
 811                 killed = 0;
 812         again1:
 813                 hlist_for_each_entry(pol, entry,
 814                                      &xfrm_policy_inexact[dir], bydst) {
 815                         if (pol->type != type)
 816                                 continue;
 817                         hlist_del(&pol->bydst);
 818                         hlist_del(&pol->byidx);
 819                         write_unlock_bh(&xfrm_policy_lock);
 820
 821                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 822                                                  audit_info->secid);
 823
 824                         xfrm_policy_kill(pol);
 825                         killed++;
 826
 827                         write_lock_bh(&xfrm_policy_lock);
 828                         goto again1;
 829                 }
 830
 831                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
 832         again2:
 833                         hlist_for_each_entry(pol, entry,
 834                                              xfrm_policy_bydst[dir].table + i,
 835                                              bydst) {
 836                                 if (pol->type != type)
 837                                         continue;
 838                                 hlist_del(&pol->bydst);
 839                                 hlist_del(&pol->byidx);
 840                                 write_unlock_bh(&xfrm_policy_lock);
 841
 842                                 xfrm_audit_policy_delete(pol, 1,
 843                                                          audit_info->loginuid,
 844                                                          audit_info->secid);
 845                                 xfrm_policy_kill(pol);
 846                                 killed++;
 847
 848                                 write_lock_bh(&xfrm_policy_lock);
 849                                 goto again2;
 850                         }
 851                 }
 852
 853                 xfrm_policy_count[dir] -= killed;
 854         }
 855         atomic_inc(&flow_cache_genid);
 856 out:
 857         write_unlock_bh(&xfrm_policy_lock);
 858         return err;
 859 }
 860 EXPORT_SYMBOL(xfrm_policy_flush);
 861
 862 int xfrm_policy_walk(struct xfrm_policy_walk *walk,
 863                      int (*func)(struct xfrm_policy *, int, int, void*),
 864                      void *data)
 865 {
 866         struct xfrm_policy *old, *pol, *last = NULL;
 867         int error = 0;
 868
 869         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 870             walk->type != XFRM_POLICY_TYPE_ANY)
 871                 return -EINVAL;
 872
 873         if (walk->policy == NULL && walk->count != 0)
 874                 return 0;
 875
 876         old = pol = walk->policy;
 877         walk->policy = NULL;
 878         read_lock_bh(&xfrm_policy_lock);
 879
 880         for (; walk->cur_type < XFRM_POLICY_TYPE_MAX; walk->cur_type++) {
 881                 if (walk->type != walk->cur_type &&
 882                     walk->type != XFRM_POLICY_TYPE_ANY)
 883                         continue;
 884
 885                 if (pol == NULL) {
 886                         pol = list_first_entry(&xfrm_policy_bytype[walk->cur_type],
 887                                                struct xfrm_policy, bytype);
 888                 }
 889                 list_for_each_entry_from(pol, &xfrm_policy_bytype[walk->cur_type], bytype) {
 890                         if (pol->dead)
 891                                 continue;
 892                         if (last) {
 893                                 error = func(last, xfrm_policy_id2dir(last->index),
 894                                              walk->count, data);
 895                                 if (error) {
 896                                         xfrm_pol_hold(last);
 897                                         walk->policy = last;
 898                                         goto out;
 899                                 }
 900                         }
 901                         last = pol;
 902                         walk->count++;
 903                 }
 904                 pol = NULL;
 905         }
 906         if (walk->count == 0) {
 907                 error = -ENOENT;
 908                 goto out;
 909         }
 910         if (last)
 911                 error = func(last, xfrm_policy_id2dir(last->index), 0, data);
 912 out:
 913         read_unlock_bh(&xfrm_policy_lock);
 914         if (old != NULL)
 915                 xfrm_pol_put(old);
 916         return error;
 917 }
 918 EXPORT_SYMBOL(xfrm_policy_walk);
 919
 920 /*
 921  * Find policy to apply to this flow.
 922  *
 923  * Returns 0 if policy found, else an -errno.
 924  */
 925 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
 926                              u8 type, u16 family, int dir)
 927 {
 928         struct xfrm_selector *sel = &pol->selector;
 929         int match, ret = -ESRCH;
 930
 931         if (pol->family != family ||
 932             pol->type != type)
 933                 return ret;
 934
 935         match = xfrm_selector_match(sel, fl, family);
 936         if (match)
 937                 ret = security_xfrm_policy_lookup(pol->security, fl->secid,
 938                                                   dir);
 939
 940         return ret;
 941 }
 942
 943 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
 944                                                      u16 family, u8 dir)
 945 {
 946         int err;
 947         struct xfrm_policy *pol, *ret;
 948         xfrm_address_t *daddr, *saddr;
 949         struct hlist_node *entry;
 950         struct hlist_head *chain;
 951         u32 priority = ~0U;
 952
 953         daddr = xfrm_flowi_daddr(fl, family);
 954         saddr = xfrm_flowi_saddr(fl, family);
 955         if (unlikely(!daddr || !saddr))
 956                 return NULL;
 957
 958         read_lock_bh(&xfrm_policy_lock);
 959         chain = policy_hash_direct(daddr, saddr, family, dir);
 960         ret = NULL;
 961         hlist_for_each_entry(pol, entry, chain, bydst) {
 962                 err = xfrm_policy_match(pol, fl, type, family, dir);
 963                 if (err) {
 964                         if (err == -ESRCH)
 965                                 continue;
 966                         else {
 967                                 ret = ERR_PTR(err);
 968                                 goto fail;
 969                         }
 970                 } else {
 971                         ret = pol;
 972                         priority = ret->priority;
 973                         break;
 974                 }
 975         }
 976         chain = &xfrm_policy_inexact[dir];
 977         hlist_for_each_entry(pol, entry, chain, bydst) {
 978                 err = xfrm_policy_match(pol, fl, type, family, dir);
 979                 if (err) {
 980                         if (err == -ESRCH)
 981                                 continue;
 982                         else {
 983                                 ret = ERR_PTR(err);
 984                                 goto fail;
 985                         }
 986                 } else if (pol->priority < priority) {
 987                         ret = pol;
 988                         break;
 989                 }
 990         }
 991         if (ret)
 992                 xfrm_pol_hold(ret);
 993 fail:
 994         read_unlock_bh(&xfrm_policy_lock);
 995
 996         return ret;
 997 }
 998
 999 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
1000                                void **objp, atomic_t **obj_refp)
1001 {
1002         struct xfrm_policy *pol;
1003         int err = 0;
1004
1005 #ifdef CONFIG_XFRM_SUB_POLICY
1006         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1007         if (IS_ERR(pol)) {
1008                 err = PTR_ERR(pol);
1009                 pol = NULL;
1010         }
1011         if (pol || err)
1012                 goto end;
1013 #endif
1014         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1015         if (IS_ERR(pol)) {
1016                 err = PTR_ERR(pol);
1017                 pol = NULL;
1018         }
1019 #ifdef CONFIG_XFRM_SUB_POLICY
1020 end:
1021 #endif
1022         if ((*objp = (void *) pol) != NULL)
1023                 *obj_refp = &pol->refcnt;
1024         return err;
1025 }
1026
1027 static inline int policy_to_flow_dir(int dir)
1028 {
1029         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1030             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1031             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1032                 return dir;
1033         switch (dir) {
1034         default:
1035         case XFRM_POLICY_IN:
1036                 return FLOW_DIR_IN;
1037         case XFRM_POLICY_OUT:
1038                 return FLOW_DIR_OUT;
1039         case XFRM_POLICY_FWD:
1040                 return FLOW_DIR_FWD;
1041         }
1042 }
1043
1044 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1045 {
1046         struct xfrm_policy *pol;
1047
1048         read_lock_bh(&xfrm_policy_lock);
1049         if ((pol = sk->sk_policy[dir]) != NULL) {
1050                 int match = xfrm_selector_match(&pol->selector, fl,
1051                                                 sk->sk_family);
1052                 int err = 0;
1053
1054                 if (match) {
1055                         err = security_xfrm_policy_lookup(pol->security,
1056                                                       fl->secid,
1057                                                       policy_to_flow_dir(dir));
1058                         if (!err)
1059                                 xfrm_pol_hold(pol);
1060                         else if (err == -ESRCH)
1061                                 pol = NULL;
1062                         else
1063                                 pol = ERR_PTR(err);
1064                 } else
1065                         pol = NULL;
1066         }
1067         read_unlock_bh(&xfrm_policy_lock);
1068         return pol;
1069 }
1070
1071 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1072 {
1073         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1074                                                      pol->family, dir);
1075
1076         hlist_add_head(&pol->bydst, chain);
1077         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1078         xfrm_policy_count[dir]++;
1079         xfrm_pol_hold(pol);
1080
1081         if (xfrm_bydst_should_resize(dir, NULL))
1082                 schedule_work(&xfrm_hash_work);
1083 }
1084
1085 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1086                                                 int dir)
1087 {
1088         if (hlist_unhashed(&pol->bydst))
1089                 return NULL;
1090
1091         hlist_del(&pol->bydst);
1092         hlist_del(&pol->byidx);
1093         xfrm_policy_count[dir]--;
1094
1095         return pol;
1096 }
1097
1098 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1099 {
1100         write_lock_bh(&xfrm_policy_lock);
1101         pol = __xfrm_policy_unlink(pol, dir);
1102         write_unlock_bh(&xfrm_policy_lock);
1103         if (pol) {
1104                 if (dir < XFRM_POLICY_MAX)
1105                         atomic_inc(&flow_cache_genid);
1106                 xfrm_policy_kill(pol);
1107                 return 0;
1108         }
1109         return -ENOENT;
1110 }
1111 EXPORT_SYMBOL(xfrm_policy_delete);
1112
1113 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1114 {
1115         struct xfrm_policy *old_pol;
1116
1117 #ifdef CONFIG_XFRM_SUB_POLICY
1118         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1119                 return -EINVAL;
1120 #endif
1121
1122         write_lock_bh(&xfrm_policy_lock);
1123         old_pol = sk->sk_policy[dir];
1124         sk->sk_policy[dir] = pol;
1125         if (pol) {
1126                 pol->curlft.add_time = get_seconds();
1127                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1128                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1129         }
1130         if (old_pol)
1131                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1132         write_unlock_bh(&xfrm_policy_lock);
1133
1134         if (old_pol) {
1135                 xfrm_policy_kill(old_pol);
1136         }
1137         return 0;
1138 }
1139
1140 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1141 {
1142         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1143
1144         if (newp) {
1145                 newp->selector = old->selector;
1146                 if (security_xfrm_policy_clone(old->security,
1147                                                &newp->security)) {
1148                         kfree(newp);
1149                         return NULL;  /* ENOMEM */
1150                 }
1151                 newp->lft = old->lft;
1152                 newp->curlft = old->curlft;
1153                 newp->action = old->action;
1154                 newp->flags = old->flags;
1155                 newp->xfrm_nr = old->xfrm_nr;
1156                 newp->index = old->index;
1157                 newp->type = old->type;
1158                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1159                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1160                 write_lock_bh(&xfrm_policy_lock);
1161                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1162                 write_unlock_bh(&xfrm_policy_lock);
1163                 xfrm_pol_put(newp);
1164         }
1165         return newp;
1166 }
1167
1168 int __xfrm_sk_clone_policy(struct sock *sk)
1169 {
1170         struct xfrm_policy *p0 = sk->sk_policy[0],
1171                            *p1 = sk->sk_policy[1];
1172
1173         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1174         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1175                 return -ENOMEM;
1176         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1177                 return -ENOMEM;
1178         return 0;
1179 }
1180
1181 static int
1182 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1183                unsigned short family)
1184 {
1185         int err;
1186         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1187
1188         if (unlikely(afinfo == NULL))
1189                 return -EINVAL;
1190         err = afinfo->get_saddr(local, remote);
1191         xfrm_policy_put_afinfo(afinfo);
1192         return err;
1193 }
1194
1195 /* Resolve list of templates for the flow, given policy. */
1196
1197 static int
1198 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1199                       struct xfrm_state **xfrm,
1200                       unsigned short family)
1201 {
1202         int nx;
1203         int i, error;
1204         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1205         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1206         xfrm_address_t tmp;
1207
1208         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1209                 struct xfrm_state *x;
1210                 xfrm_address_t *remote = daddr;
1211                 xfrm_address_t *local  = saddr;
1212                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1213
1214                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1215                     tmpl->mode == XFRM_MODE_BEET) {
1216                         remote = &tmpl->id.daddr;
1217                         local = &tmpl->saddr;
1218                         family = tmpl->encap_family;
1219                         if (xfrm_addr_any(local, family)) {
1220                                 error = xfrm_get_saddr(&tmp, remote, family);
1221                                 if (error)
1222                                         goto fail;
1223                                 local = &tmp;
1224                         }
1225                 }
1226
1227                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1228
1229                 if (x && x->km.state == XFRM_STATE_VALID) {
1230                         xfrm[nx++] = x;
1231                         daddr = remote;
1232                         saddr = local;
1233                         continue;
1234                 }
1235                 if (x) {
1236                         error = (x->km.state == XFRM_STATE_ERROR ?
1237                                  -EINVAL : -EAGAIN);
1238                         xfrm_state_put(x);
1239                 }
1240
1241                 if (!tmpl->optional)
1242                         goto fail;
1243         }
1244         return nx;
1245
1246 fail:
1247         for (nx--; nx>=0; nx--)
1248                 xfrm_state_put(xfrm[nx]);
1249         return error;
1250 }
1251
1252 static int
1253 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1254                   struct xfrm_state **xfrm,
1255                   unsigned short family)
1256 {
1257         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1258         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1259         int cnx = 0;
1260         int error;
1261         int ret;
1262         int i;
1263
1264         for (i = 0; i < npols; i++) {
1265                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1266                         error = -ENOBUFS;
1267                         goto fail;
1268                 }
1269
1270                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1271                 if (ret < 0) {
1272                         error = ret;
1273                         goto fail;
1274                 } else
1275                         cnx += ret;
1276         }
1277
1278         /* found states are sorted for outbound processing */
1279         if (npols > 1)
1280                 xfrm_state_sort(xfrm, tpp, cnx, family);
1281
1282         return cnx;
1283
1284  fail:
1285         for (cnx--; cnx>=0; cnx--)
1286                 xfrm_state_put(tpp[cnx]);
1287         return error;
1288
1289 }
1290
1291 /* Check that the bundle accepts the flow and its components are
1292  * still valid.
1293  */
1294
1295 static struct dst_entry *
1296 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1297 {
1298         struct dst_entry *x;
1299         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1300         if (unlikely(afinfo == NULL))
1301                 return ERR_PTR(-EINVAL);
1302         x = afinfo->find_bundle(fl, policy);
1303         xfrm_policy_put_afinfo(afinfo);
1304         return x;
1305 }
1306
1307 static inline int xfrm_get_tos(struct flowi *fl, int family)
1308 {
1309         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1310         int tos;
1311
1312         if (!afinfo)
1313                 return -EINVAL;
1314
1315         tos = afinfo->get_tos(fl);
1316
1317         xfrm_policy_put_afinfo(afinfo);
1318
1319         return tos;
1320 }
1321
1322 static inline struct xfrm_dst *xfrm_alloc_dst(int family)
1323 {
1324         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1325         struct xfrm_dst *xdst;
1326
1327         if (!afinfo)
1328                 return ERR_PTR(-EINVAL);
1329
1330         xdst = dst_alloc(afinfo->dst_ops) ?: ERR_PTR(-ENOBUFS);
1331
1332         xfrm_policy_put_afinfo(afinfo);
1333
1334         return xdst;
1335 }
1336
1337 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1338                                  int nfheader_len)
1339 {
1340         struct xfrm_policy_afinfo *afinfo =
1341                 xfrm_policy_get_afinfo(dst->ops->family);
1342         int err;
1343
1344         if (!afinfo)
1345                 return -EINVAL;
1346
1347         err = afinfo->init_path(path, dst, nfheader_len);
1348
1349         xfrm_policy_put_afinfo(afinfo);
1350
1351         return err;
1352 }
1353
1354 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev)
1355 {
1356         struct xfrm_policy_afinfo *afinfo =
1357                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1358         int err;
1359
1360         if (!afinfo)
1361                 return -EINVAL;
1362
1363         err = afinfo->fill_dst(xdst, dev);
1364
1365         xfrm_policy_put_afinfo(afinfo);
1366
1367         return err;
1368 }
1369
1370 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1371  * all the metrics... Shortly, bundle a bundle.
1372  */
1373
1374 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1375                                             struct xfrm_state **xfrm, int nx,
1376                                             struct flowi *fl,
1377                                             struct dst_entry *dst)
1378 {
1379         unsigned long now = jiffies;
1380         struct net_device *dev;
1381         struct dst_entry *dst_prev = NULL;
1382         struct dst_entry *dst0 = NULL;
1383         int i = 0;
1384         int err;
1385         int header_len = 0;
1386         int nfheader_len = 0;
1387         int trailer_len = 0;
1388         int tos;
1389         int family = policy->selector.family;
1390         xfrm_address_t saddr, daddr;
1391
1392         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1393
1394         tos = xfrm_get_tos(fl, family);
1395         err = tos;
1396         if (tos < 0)
1397                 goto put_states;
1398
1399         dst_hold(dst);
1400
1401         for (; i < nx; i++) {
1402                 struct xfrm_dst *xdst = xfrm_alloc_dst(family);
1403                 struct dst_entry *dst1 = &xdst->u.dst;
1404
1405                 err = PTR_ERR(xdst);
1406                 if (IS_ERR(xdst)) {
1407                         dst_release(dst);
1408                         goto put_states;
1409                 }
1410
1411                 if (!dst_prev)
1412                         dst0 = dst1;
1413                 else {
1414                         dst_prev->child = dst_clone(dst1);
1415                         dst1->flags |= DST_NOHASH;
1416                 }
1417
1418                 xdst->route = dst;
1419                 memcpy(&dst1->metrics, &dst->metrics, sizeof(dst->metrics));
1420
1421                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1422                         family = xfrm[i]->props.family;
1423                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1424                                               family);
1425                         err = PTR_ERR(dst);
1426                         if (IS_ERR(dst))
1427                                 goto put_states;
1428                 } else
1429                         dst_hold(dst);
1430
1431                 dst1->xfrm = xfrm[i];
1432                 xdst->genid = xfrm[i]->genid;
1433
1434                 dst1->obsolete = -1;
1435                 dst1->flags |= DST_HOST;
1436                 dst1->lastuse = now;
1437
1438                 dst1->input = dst_discard;
1439                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1440
1441                 dst1->next = dst_prev;
1442                 dst_prev = dst1;
1443
1444                 header_len += xfrm[i]->props.header_len;
1445                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1446                         nfheader_len += xfrm[i]->props.header_len;
1447                 trailer_len += xfrm[i]->props.trailer_len;
1448         }
1449
1450         dst_prev->child = dst;
1451         dst0->path = dst;
1452
1453         err = -ENODEV;
1454         dev = dst->dev;
1455         if (!dev)
1456                 goto free_dst;
1457
1458         /* Copy neighbout for reachability confirmation */
1459         dst0->neighbour = neigh_clone(dst->neighbour);
1460
1461         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1462         xfrm_init_pmtu(dst_prev);
1463
1464         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1465                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1466
1467                 err = xfrm_fill_dst(xdst, dev);
1468                 if (err)
1469                         goto free_dst;
1470
1471                 dst_prev->header_len = header_len;
1472                 dst_prev->trailer_len = trailer_len;
1473                 header_len -= xdst->u.dst.xfrm->props.header_len;
1474                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1475         }
1476
1477 out:
1478         return dst0;
1479
1480 put_states:
1481         for (; i < nx; i++)
1482                 xfrm_state_put(xfrm[i]);
1483 free_dst:
1484         if (dst0)
1485                 dst_free(dst0);
1486         dst0 = ERR_PTR(err);
1487         goto out;
1488 }
1489
1490 static int inline
1491 xfrm_dst_alloc_copy(void **target, void *src, int size)
1492 {
1493         if (!*target) {
1494                 *target = kmalloc(size, GFP_ATOMIC);
1495                 if (!*target)
1496                         return -ENOMEM;
1497         }
1498         memcpy(*target, src, size);
1499         return 0;
1500 }
1501
1502 static int inline
1503 xfrm_dst_update_parent(struct dst_entry *dst, struct xfrm_selector *sel)
1504 {
1505 #ifdef CONFIG_XFRM_SUB_POLICY
1506         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1507         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1508                                    sel, sizeof(*sel));
1509 #else
1510         return 0;
1511 #endif
1512 }
1513
1514 static int inline
1515 xfrm_dst_update_origin(struct dst_entry *dst, struct flowi *fl)
1516 {
1517 #ifdef CONFIG_XFRM_SUB_POLICY
1518         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1519         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1520 #else
1521         return 0;
1522 #endif
1523 }
1524
1525 static int stale_bundle(struct dst_entry *dst);
1526
1527 /* Main function: finds/creates a bundle for given flow.
1528  *
1529  * At the moment we eat a raw IP route. Mostly to speed up lookups
1530  * on interfaces with disabled IPsec.
1531  */
1532 int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1533                   struct sock *sk, int flags)
1534 {
1535         struct xfrm_policy *policy;
1536         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1537         int npols;
1538         int pol_dead;
1539         int xfrm_nr;
1540         int pi;
1541         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1542         struct dst_entry *dst, *dst_orig = *dst_p;
1543         int nx = 0;
1544         int err;
1545         u32 genid;
1546         u16 family;
1547         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1548
1549 restart:
1550         genid = atomic_read(&flow_cache_genid);
1551         policy = NULL;
1552         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1553                 pols[pi] = NULL;
1554         npols = 0;
1555         pol_dead = 0;
1556         xfrm_nr = 0;
1557
1558         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1559                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1560                 err = PTR_ERR(policy);
1561                 if (IS_ERR(policy)) {
1562                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1563                         goto dropdst;
1564                 }
1565         }
1566
1567         if (!policy) {
1568                 /* To accelerate a bit...  */
1569                 if ((dst_orig->flags & DST_NOXFRM) ||
1570                     !xfrm_policy_count[XFRM_POLICY_OUT])
1571                         goto nopol;
1572
1573                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1574                                            dir, xfrm_policy_lookup);
1575                 err = PTR_ERR(policy);
1576                 if (IS_ERR(policy)) {
1577                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1578                         goto dropdst;
1579                 }
1580         }
1581
1582         if (!policy)
1583                 goto nopol;
1584
1585         family = dst_orig->ops->family;
1586         pols[0] = policy;
1587         npols ++;
1588         xfrm_nr += pols[0]->xfrm_nr;
1589
1590         err = -ENOENT;
1591         if ((flags & XFRM_LOOKUP_ICMP) && !(policy->flags & XFRM_POLICY_ICMP))
1592                 goto error;
1593
1594         policy->curlft.use_time = get_seconds();
1595
1596         switch (policy->action) {
1597         default:
1598         case XFRM_POLICY_BLOCK:
1599                 /* Prohibit the flow */
1600                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1601                 err = -EPERM;
1602                 goto error;
1603
1604         case XFRM_POLICY_ALLOW:
1605 #ifndef CONFIG_XFRM_SUB_POLICY
1606                 if (policy->xfrm_nr == 0) {
1607                         /* Flow passes not transformed. */
1608                         xfrm_pol_put(policy);
1609                         return 0;
1610                 }
1611 #endif
1612
1613                 /* Try to find matching bundle.
1614                  *
1615                  * LATER: help from flow cache. It is optional, this
1616                  * is required only for output policy.
1617                  */
1618                 dst = xfrm_find_bundle(fl, policy, family);
1619                 if (IS_ERR(dst)) {
1620                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1621                         err = PTR_ERR(dst);
1622                         goto error;
1623                 }
1624
1625                 if (dst)
1626                         break;
1627
1628 #ifdef CONFIG_XFRM_SUB_POLICY
1629                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1630                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1631                                                             fl, family,
1632                                                             XFRM_POLICY_OUT);
1633                         if (pols[1]) {
1634                                 if (IS_ERR(pols[1])) {
1635                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
1636                                         err = PTR_ERR(pols[1]);
1637                                         goto error;
1638                                 }
1639                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1640                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
1641                                         err = -EPERM;
1642                                         goto error;
1643                                 }
1644                                 npols ++;
1645                                 xfrm_nr += pols[1]->xfrm_nr;
1646                         }
1647                 }
1648
1649                 /*
1650                  * Because neither flowi nor bundle information knows about
1651                  * transformation template size. On more than one policy usage
1652                  * we can realize whether all of them is bypass or not after
1653                  * they are searched. See above not-transformed bypass
1654                  * is surrounded by non-sub policy configuration, too.
1655                  */
1656                 if (xfrm_nr == 0) {
1657                         /* Flow passes not transformed. */
1658                         xfrm_pols_put(pols, npols);
1659                         return 0;
1660                 }
1661
1662 #endif
1663                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1664
1665                 if (unlikely(nx<0)) {
1666                         err = nx;
1667                         if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
1668                                 /* EREMOTE tells the caller to generate
1669                                  * a one-shot blackhole route.
1670                                  */
1671                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1672                                 xfrm_pol_put(policy);
1673                                 return -EREMOTE;
1674                         }
1675                         if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
1676                                 DECLARE_WAITQUEUE(wait, current);
1677
1678                                 add_wait_queue(&km_waitq, &wait);
1679                                 set_current_state(TASK_INTERRUPTIBLE);
1680                                 schedule();
1681                                 set_current_state(TASK_RUNNING);
1682                                 remove_wait_queue(&km_waitq, &wait);
1683
1684                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1685
1686                                 if (nx == -EAGAIN && signal_pending(current)) {
1687                                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1688                                         err = -ERESTART;
1689                                         goto error;
1690                                 }
1691                                 if (nx == -EAGAIN ||
1692                                     genid != atomic_read(&flow_cache_genid)) {
1693                                         xfrm_pols_put(pols, npols);
1694                                         goto restart;
1695                                 }
1696                                 err = nx;
1697                         }
1698                         if (err < 0) {
1699                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
1700                                 goto error;
1701                         }
1702                 }
1703                 if (nx == 0) {
1704                         /* Flow passes not transformed. */
1705                         xfrm_pols_put(pols, npols);
1706                         return 0;
1707                 }
1708
1709                 dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
1710                 err = PTR_ERR(dst);
1711                 if (IS_ERR(dst)) {
1712                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1713                         goto error;
1714                 }
1715
1716                 for (pi = 0; pi < npols; pi++) {
1717                         read_lock_bh(&pols[pi]->lock);
1718                         pol_dead |= pols[pi]->dead;
1719                         read_unlock_bh(&pols[pi]->lock);
1720                 }
1721
1722                 write_lock_bh(&policy->lock);
1723                 if (unlikely(pol_dead || stale_bundle(dst))) {
1724                         /* Wow! While we worked on resolving, this
1725                          * policy has gone. Retry. It is not paranoia,
1726                          * we just cannot enlist new bundle to dead object.
1727                          * We can't enlist stable bundles either.
1728                          */
1729                         write_unlock_bh(&policy->lock);
1730                         if (dst)
1731                                 dst_free(dst);
1732
1733                         if (pol_dead)
1734                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
1735                         else
1736                                 XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1737                         err = -EHOSTUNREACH;
1738                         goto error;
1739                 }
1740
1741                 if (npols > 1)
1742                         err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1743                 else
1744                         err = xfrm_dst_update_origin(dst, fl);
1745                 if (unlikely(err)) {
1746                         write_unlock_bh(&policy->lock);
1747                         if (dst)
1748                                 dst_free(dst);
1749                         XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1750                         goto error;
1751                 }
1752
1753                 dst->next = policy->bundles;
1754                 policy->bundles = dst;
1755                 dst_hold(dst);
1756                 write_unlock_bh(&policy->lock);
1757         }
1758         *dst_p = dst;
1759         dst_release(dst_orig);
1760         xfrm_pols_put(pols, npols);
1761         return 0;
1762
1763 error:
1764         xfrm_pols_put(pols, npols);
1765 dropdst:
1766         dst_release(dst_orig);
1767         *dst_p = NULL;
1768         return err;
1769
1770 nopol:
1771         err = -ENOENT;
1772         if (flags & XFRM_LOOKUP_ICMP)
1773                 goto dropdst;
1774         return 0;
1775 }
1776 EXPORT_SYMBOL(__xfrm_lookup);
1777
1778 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1779                 struct sock *sk, int flags)
1780 {
1781         int err = __xfrm_lookup(dst_p, fl, sk, flags);
1782
1783         if (err == -EREMOTE) {
1784                 dst_release(*dst_p);
1785                 *dst_p = NULL;
1786                 err = -EAGAIN;
1787         }
1788
1789         return err;
1790 }
1791 EXPORT_SYMBOL(xfrm_lookup);
1792
1793 static inline int
1794 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1795 {
1796         struct xfrm_state *x;
1797
1798         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1799                 return 0;
1800         x = skb->sp->xvec[idx];
1801         if (!x->type->reject)
1802                 return 0;
1803         return x->type->reject(x, skb, fl);
1804 }
1805
1806 /* When skb is transformed back to its "native" form, we have to
1807  * check policy restrictions. At the moment we make this in maximally
1808  * stupid way. Shame on me. :-) Of course, connected sockets must
1809  * have policy cached at them.
1810  */
1811
1812 static inline int
1813 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
1814               unsigned short family)
1815 {
1816         if (xfrm_state_kern(x))
1817                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1818         return  x->id.proto == tmpl->id.proto &&
1819                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1820                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1821                 x->props.mode == tmpl->mode &&
1822                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1823                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1824                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1825                   xfrm_state_addr_cmp(tmpl, x, family));
1826 }
1827
1828 /*
1829  * 0 or more than 0 is returned when validation is succeeded (either bypass
1830  * because of optional transport mode, or next index of the mathced secpath
1831  * state with the template.
1832  * -1 is returned when no matching template is found.
1833  * Otherwise "-2 - errored_index" is returned.
1834  */
1835 static inline int
1836 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1837                unsigned short family)
1838 {
1839         int idx = start;
1840
1841         if (tmpl->optional) {
1842                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1843                         return start;
1844         } else
1845                 start = -1;
1846         for (; idx < sp->len; idx++) {
1847                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1848                         return ++idx;
1849                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1850                         if (start == -1)
1851                                 start = -2-idx;
1852                         break;
1853                 }
1854         }
1855         return start;
1856 }
1857
1858 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1859                           unsigned int family, int reverse)
1860 {
1861         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1862         int err;
1863
1864         if (unlikely(afinfo == NULL))
1865                 return -EAFNOSUPPORT;
1866
1867         afinfo->decode_session(skb, fl, reverse);
1868         err = security_xfrm_decode_session(skb, &fl->secid);
1869         xfrm_policy_put_afinfo(afinfo);
1870         return err;
1871 }
1872 EXPORT_SYMBOL(__xfrm_decode_session);
1873
1874 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1875 {
1876         for (; k < sp->len; k++) {
1877                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1878                         *idxp = k;
1879                         return 1;
1880                 }
1881         }
1882
1883         return 0;
1884 }
1885
1886 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1887                         unsigned short family)
1888 {
1889         struct xfrm_policy *pol;
1890         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1891         int npols = 0;
1892         int xfrm_nr;
1893         int pi;
1894         int reverse;
1895         struct flowi fl;
1896         u8 fl_dir;
1897         int xerr_idx = -1;
1898
1899         reverse = dir & ~XFRM_POLICY_MASK;
1900         dir &= XFRM_POLICY_MASK;
1901         fl_dir = policy_to_flow_dir(dir);
1902
1903         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
1904                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
1905                 return 0;
1906         }
1907
1908         nf_nat_decode_session(skb, &fl, family);
1909
1910         /* First, check used SA against their selectors. */
1911         if (skb->sp) {
1912                 int i;
1913
1914                 for (i=skb->sp->len-1; i>=0; i--) {
1915                         struct xfrm_state *x = skb->sp->xvec[i];
1916                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
1917                                 XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
1918                                 return 0;
1919                         }
1920                 }
1921         }
1922
1923         pol = NULL;
1924         if (sk && sk->sk_policy[dir]) {
1925                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1926                 if (IS_ERR(pol)) {
1927                         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1928                         return 0;
1929                 }
1930         }
1931
1932         if (!pol)
1933                 pol = flow_cache_lookup(&fl, family, fl_dir,
1934                                         xfrm_policy_lookup);
1935
1936         if (IS_ERR(pol)) {
1937                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1938                 return 0;
1939         }
1940
1941         if (!pol) {
1942                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1943                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1944                         XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
1945                         return 0;
1946                 }
1947                 return 1;
1948         }
1949
1950         pol->curlft.use_time = get_seconds();
1951
1952         pols[0] = pol;
1953         npols ++;
1954 #ifdef CONFIG_XFRM_SUB_POLICY
1955         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1956                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1957                                                     &fl, family,
1958                                                     XFRM_POLICY_IN);
1959                 if (pols[1]) {
1960                         if (IS_ERR(pols[1])) {
1961                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
1962                                 return 0;
1963                         }
1964                         pols[1]->curlft.use_time = get_seconds();
1965                         npols ++;
1966                 }
1967         }
1968 #endif
1969
1970         if (pol->action == XFRM_POLICY_ALLOW) {
1971                 struct sec_path *sp;
1972                 static struct sec_path dummy;
1973                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1974                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1975                 struct xfrm_tmpl **tpp = tp;
1976                 int ti = 0;
1977                 int i, k;
1978
1979                 if ((sp = skb->sp) == NULL)
1980                         sp = &dummy;
1981
1982                 for (pi = 0; pi < npols; pi++) {
1983                         if (pols[pi] != pol &&
1984                             pols[pi]->action != XFRM_POLICY_ALLOW) {
1985                                 XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
1986                                 goto reject;
1987                         }
1988                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
1989                                 XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
1990                                 goto reject_error;
1991                         }
1992                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1993                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1994                 }
1995                 xfrm_nr = ti;
1996                 if (npols > 1) {
1997                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1998                         tpp = stp;
1999                 }
2000
2001                 /* For each tunnel xfrm, find the first matching tmpl.
2002                  * For each tmpl before that, find corresponding xfrm.
2003                  * Order is _important_. Later we will implement
2004                  * some barriers, but at the moment barriers
2005                  * are implied between each two transformations.
2006                  */
2007                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2008                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2009                         if (k < 0) {
2010                                 if (k < -1)
2011                                         /* "-2 - errored_index" returned */
2012                                         xerr_idx = -(2+k);
2013                                 XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2014                                 goto reject;
2015                         }
2016                 }
2017
2018                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2019                         XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
2020                         goto reject;
2021                 }
2022
2023                 xfrm_pols_put(pols, npols);
2024                 return 1;
2025         }
2026         XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
2027
2028 reject:
2029         xfrm_secpath_reject(xerr_idx, skb, &fl);
2030 reject_error:
2031         xfrm_pols_put(pols, npols);
2032         return 0;
2033 }
2034 EXPORT_SYMBOL(__xfrm_policy_check);
2035
2036 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2037 {
2038         struct flowi fl;
2039
2040         if (xfrm_decode_session(skb, &fl, family) < 0) {
2041                 /* XXX: we should have something like FWDHDRERROR here. */
2042                 XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
2043                 return 0;
2044         }
2045
2046         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
2047 }
2048 EXPORT_SYMBOL(__xfrm_route_forward);
2049
2050 /* Optimize later using cookies and generation ids. */
2051
2052 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2053 {
2054         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2055          * to "-1" to force all XFRM destinations to get validated by
2056          * dst_ops->check on every use.  We do this because when a
2057          * normal route referenced by an XFRM dst is obsoleted we do
2058          * not go looking around for all parent referencing XFRM dsts
2059          * so that we can invalidate them.  It is just too much work.
2060          * Instead we make the checks here on every use.  For example:
2061          *
2062          *      XFRM dst A --> IPv4 dst X
2063          *
2064          * X is the "xdst->route" of A (X is also the "dst->path" of A
2065          * in this example).  If X is marked obsolete, "A" will not
2066          * notice.  That's what we are validating here via the
2067          * stale_bundle() check.
2068          *
2069          * When a policy's bundle is pruned, we dst_free() the XFRM
2070          * dst which causes it's ->obsolete field to be set to a
2071          * positive non-zero integer.  If an XFRM dst has been pruned
2072          * like this, we want to force a new route lookup.
2073          */
2074         if (dst->obsolete < 0 && !stale_bundle(dst))
2075                 return dst;
2076
2077         return NULL;
2078 }
2079
2080 static int stale_bundle(struct dst_entry *dst)
2081 {
2082         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
2083 }
2084
2085 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2086 {
2087         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2088                 dst->dev = dev_net(dev)->loopback_dev;
2089                 dev_hold(dst->dev);
2090                 dev_put(dev);
2091         }
2092 }
2093 EXPORT_SYMBOL(xfrm_dst_ifdown);
2094
2095 static void xfrm_link_failure(struct sk_buff *skb)
2096 {
2097         /* Impossible. Such dst must be popped before reaches point of failure. */
2098         return;
2099 }
2100
2101 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2102 {
2103         if (dst) {
2104                 if (dst->obsolete) {
2105                         dst_release(dst);
2106                         dst = NULL;
2107                 }
2108         }
2109         return dst;
2110 }
2111
2112 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
2113 {
2114         struct dst_entry *dst, **dstp;
2115
2116         write_lock(&pol->lock);
2117         dstp = &pol->bundles;
2118         while ((dst=*dstp) != NULL) {
2119                 if (func(dst)) {
2120                         *dstp = dst->next;
2121                         dst->next = *gc_list_p;
2122                         *gc_list_p = dst;
2123                 } else {
2124                         dstp = &dst->next;
2125                 }
2126         }
2127         write_unlock(&pol->lock);
2128 }
2129
2130 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
2131 {
2132         struct dst_entry *gc_list = NULL;
2133         int dir;
2134
2135         read_lock_bh(&xfrm_policy_lock);
2136         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2137                 struct xfrm_policy *pol;
2138                 struct hlist_node *entry;
2139                 struct hlist_head *table;
2140                 int i;
2141
2142                 hlist_for_each_entry(pol, entry,
2143                                      &xfrm_policy_inexact[dir], bydst)
2144                         prune_one_bundle(pol, func, &gc_list);
2145
2146                 table = xfrm_policy_bydst[dir].table;
2147                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
2148                         hlist_for_each_entry(pol, entry, table + i, bydst)
2149                                 prune_one_bundle(pol, func, &gc_list);
2150                 }
2151         }
2152         read_unlock_bh(&xfrm_policy_lock);
2153
2154         while (gc_list) {
2155                 struct dst_entry *dst = gc_list;
2156                 gc_list = dst->next;
2157                 dst_free(dst);
2158         }
2159 }
2160
2161 static int unused_bundle(struct dst_entry *dst)
2162 {
2163         return !atomic_read(&dst->__refcnt);
2164 }
2165
2166 static void __xfrm_garbage_collect(void)
2167 {
2168         xfrm_prune_bundles(unused_bundle);
2169 }
2170
2171 static int xfrm_flush_bundles(void)
2172 {
2173         xfrm_prune_bundles(stale_bundle);
2174         return 0;
2175 }
2176
2177 static void xfrm_init_pmtu(struct dst_entry *dst)
2178 {
2179         do {
2180                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2181                 u32 pmtu, route_mtu_cached;
2182
2183                 pmtu = dst_mtu(dst->child);
2184                 xdst->child_mtu_cached = pmtu;
2185
2186                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2187
2188                 route_mtu_cached = dst_mtu(xdst->route);
2189                 xdst->route_mtu_cached = route_mtu_cached;
2190
2191                 if (pmtu > route_mtu_cached)
2192                         pmtu = route_mtu_cached;
2193
2194                 dst->metrics[RTAX_MTU-1] = pmtu;
2195         } while ((dst = dst->next));
2196 }
2197
2198 /* Check that the bundle accepts the flow and its components are
2199  * still valid.
2200  */
2201
2202 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
2203                 struct flowi *fl, int family, int strict)
2204 {
2205         struct dst_entry *dst = &first->u.dst;
2206         struct xfrm_dst *last;
2207         u32 mtu;
2208
2209         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2210             (dst->dev && !netif_running(dst->dev)))
2211                 return 0;
2212 #ifdef CONFIG_XFRM_SUB_POLICY
2213         if (fl) {
2214                 if (first->origin && !flow_cache_uli_match(first->origin, fl))
2215                         return 0;
2216                 if (first->partner &&
2217                     !xfrm_selector_match(first->partner, fl, family))
2218                         return 0;
2219         }
2220 #endif
2221
2222         last = NULL;
2223
2224         do {
2225                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2226
2227                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
2228                         return 0;
2229                 if (fl && pol &&
2230                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
2231                         return 0;
2232                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2233                         return 0;
2234                 if (xdst->genid != dst->xfrm->genid)
2235                         return 0;
2236
2237                 if (strict && fl &&
2238                     !(dst->xfrm->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2239                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
2240                         return 0;
2241
2242                 mtu = dst_mtu(dst->child);
2243                 if (xdst->child_mtu_cached != mtu) {
2244                         last = xdst;
2245                         xdst->child_mtu_cached = mtu;
2246                 }
2247
2248                 if (!dst_check(xdst->route, xdst->route_cookie))
2249                         return 0;
2250                 mtu = dst_mtu(xdst->route);
2251                 if (xdst->route_mtu_cached != mtu) {
2252                         last = xdst;
2253                         xdst->route_mtu_cached = mtu;
2254                 }
2255
2256                 dst = dst->child;
2257         } while (dst->xfrm);
2258
2259         if (likely(!last))
2260                 return 1;
2261
2262         mtu = last->child_mtu_cached;
2263         for (;;) {
2264                 dst = &last->u.dst;
2265
2266                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2267                 if (mtu > last->route_mtu_cached)
2268                         mtu = last->route_mtu_cached;
2269                 dst->metrics[RTAX_MTU-1] = mtu;
2270
2271                 if (last == first)
2272                         break;
2273
2274                 last = (struct xfrm_dst *)last->u.dst.next;
2275                 last->child_mtu_cached = mtu;
2276         }
2277
2278         return 1;
2279 }
2280
2281 EXPORT_SYMBOL(xfrm_bundle_ok);
2282
2283 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2284 {
2285         int err = 0;
2286         if (unlikely(afinfo == NULL))
2287                 return -EINVAL;
2288         if (unlikely(afinfo->family >= NPROTO))
2289                 return -EAFNOSUPPORT;
2290         write_lock_bh(&xfrm_policy_afinfo_lock);
2291         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2292                 err = -ENOBUFS;
2293         else {
2294                 struct dst_ops *dst_ops = afinfo->dst_ops;
2295                 if (likely(dst_ops->kmem_cachep == NULL))
2296                         dst_ops->kmem_cachep = xfrm_dst_cache;
2297                 if (likely(dst_ops->check == NULL))
2298                         dst_ops->check = xfrm_dst_check;
2299                 if (likely(dst_ops->negative_advice == NULL))
2300                         dst_ops->negative_advice = xfrm_negative_advice;
2301                 if (likely(dst_ops->link_failure == NULL))
2302                         dst_ops->link_failure = xfrm_link_failure;
2303                 if (likely(afinfo->garbage_collect == NULL))
2304                         afinfo->garbage_collect = __xfrm_garbage_collect;
2305                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2306         }
2307         write_unlock_bh(&xfrm_policy_afinfo_lock);
2308         return err;
2309 }
2310 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2311
2312 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2313 {
2314         int err = 0;
2315         if (unlikely(afinfo == NULL))
2316                 return -EINVAL;
2317         if (unlikely(afinfo->family >= NPROTO))
2318                 return -EAFNOSUPPORT;
2319         write_lock_bh(&xfrm_policy_afinfo_lock);
2320         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2321                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2322                         err = -EINVAL;
2323                 else {
2324                         struct dst_ops *dst_ops = afinfo->dst_ops;
2325                         xfrm_policy_afinfo[afinfo->family] = NULL;
2326                         dst_ops->kmem_cachep = NULL;
2327                         dst_ops->check = NULL;
2328                         dst_ops->negative_advice = NULL;
2329                         dst_ops->link_failure = NULL;
2330                         afinfo->garbage_collect = NULL;
2331                 }
2332         }
2333         write_unlock_bh(&xfrm_policy_afinfo_lock);
2334         return err;
2335 }
2336 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2337
2338 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2339 {
2340         struct xfrm_policy_afinfo *afinfo;
2341         if (unlikely(family >= NPROTO))
2342                 return NULL;
2343         read_lock(&xfrm_policy_afinfo_lock);
2344         afinfo = xfrm_policy_afinfo[family];
2345         if (unlikely(!afinfo))
2346                 read_unlock(&xfrm_policy_afinfo_lock);
2347         return afinfo;
2348 }
2349
2350 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2351 {
2352         read_unlock(&xfrm_policy_afinfo_lock);
2353 }
2354
2355 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2356 {
2357         struct net_device *dev = ptr;
2358
2359         if (dev_net(dev) != &init_net)
2360                 return NOTIFY_DONE;
2361
2362         switch (event) {
2363         case NETDEV_DOWN:
2364                 xfrm_flush_bundles();
2365         }
2366         return NOTIFY_DONE;
2367 }
2368
2369 static struct notifier_block xfrm_dev_notifier = {
2370         xfrm_dev_event,
2371         NULL,
2372         0
2373 };
2374
2375 #ifdef CONFIG_XFRM_STATISTICS
2376 static int __init xfrm_statistics_init(void)
2377 {
2378         if (snmp_mib_init((void **)xfrm_statistics,
2379                           sizeof(struct linux_xfrm_mib)) < 0)
2380                 return -ENOMEM;
2381         return 0;
2382 }
2383 #endif
2384
2385 static void __init xfrm_policy_init(void)
2386 {
2387         unsigned int hmask, sz;
2388         int dir;
2389
2390         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2391                                            sizeof(struct xfrm_dst),
2392                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2393                                            NULL);
2394
2395         hmask = 8 - 1;
2396         sz = (hmask+1) * sizeof(struct hlist_head);
2397
2398         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2399         xfrm_idx_hmask = hmask;
2400         if (!xfrm_policy_byidx)
2401                 panic("XFRM: failed to allocate byidx hash\n");
2402
2403         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2404                 struct xfrm_policy_hash *htab;
2405
2406                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2407
2408                 htab = &xfrm_policy_bydst[dir];
2409                 htab->table = xfrm_hash_alloc(sz);
2410                 htab->hmask = hmask;
2411                 if (!htab->table)
2412                         panic("XFRM: failed to allocate bydst hash\n");
2413         }
2414
2415         for (dir = 0; dir < XFRM_POLICY_TYPE_MAX; dir++)
2416                 INIT_LIST_HEAD(&xfrm_policy_bytype[dir]);
2417
2418         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2419         register_netdevice_notifier(&xfrm_dev_notifier);
2420 }
2421
2422 void __init xfrm_init(void)
2423 {
2424 #ifdef CONFIG_XFRM_STATISTICS
2425         xfrm_statistics_init();
2426 #endif
2427         xfrm_state_init();
2428         xfrm_policy_init();
2429         xfrm_input_init();
2430 #ifdef CONFIG_XFRM_STATISTICS
2431         xfrm_proc_init();
2432 #endif
2433 }
2434
2435 #ifdef CONFIG_AUDITSYSCALL
2436 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2437                                          struct audit_buffer *audit_buf)
2438 {
2439         struct xfrm_sec_ctx *ctx = xp->security;
2440         struct xfrm_selector *sel = &xp->selector;
2441
2442         if (ctx)
2443                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2444                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2445
2446         switch(sel->family) {
2447         case AF_INET:
2448                 audit_log_format(audit_buf, " src=" NIPQUAD_FMT,
2449                                  NIPQUAD(sel->saddr.a4));
2450                 if (sel->prefixlen_s != 32)
2451                         audit_log_format(audit_buf, " src_prefixlen=%d",
2452                                          sel->prefixlen_s);
2453                 audit_log_format(audit_buf, " dst=" NIPQUAD_FMT,
2454                                  NIPQUAD(sel->daddr.a4));
2455                 if (sel->prefixlen_d != 32)
2456                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2457                                          sel->prefixlen_d);
2458                 break;
2459         case AF_INET6:
2460                 audit_log_format(audit_buf, " src=" NIP6_FMT,
2461                                  NIP6(*(struct in6_addr *)sel->saddr.a6));
2462                 if (sel->prefixlen_s != 128)
2463                         audit_log_format(audit_buf, " src_prefixlen=%d",
2464                                          sel->prefixlen_s);
2465                 audit_log_format(audit_buf, " dst=" NIP6_FMT,
2466                                  NIP6(*(struct in6_addr *)sel->daddr.a6));
2467                 if (sel->prefixlen_d != 128)
2468                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2469                                          sel->prefixlen_d);
2470                 break;
2471         }
2472 }
2473
2474 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2475                            u32 auid, u32 secid)
2476 {
2477         struct audit_buffer *audit_buf;
2478
2479         audit_buf = xfrm_audit_start("SPD-add");
2480         if (audit_buf == NULL)
2481                 return;
2482         xfrm_audit_helper_usrinfo(auid, secid, audit_buf);
2483         audit_log_format(audit_buf, " res=%u", result);
2484         xfrm_audit_common_policyinfo(xp, audit_buf);
2485         audit_log_end(audit_buf);
2486 }
2487 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2488
2489 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2490                               u32 auid, u32 secid)
2491 {
2492         struct audit_buffer *audit_buf;
2493
2494         audit_buf = xfrm_audit_start("SPD-delete");
2495         if (audit_buf == NULL)
2496                 return;
2497         xfrm_audit_helper_usrinfo(auid, secid, audit_buf);
2498         audit_log_format(audit_buf, " res=%u", result);
2499         xfrm_audit_common_policyinfo(xp, audit_buf);
2500         audit_log_end(audit_buf);
2501 }
2502 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2503 #endif
2504
2505 #ifdef CONFIG_XFRM_MIGRATE
2506 static int xfrm_migrate_selector_match(struct xfrm_selector *sel_cmp,
2507                                        struct xfrm_selector *sel_tgt)
2508 {
2509         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2510                 if (sel_tgt->family == sel_cmp->family &&
2511                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2512                                   sel_cmp->family) == 0 &&
2513                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2514                                   sel_cmp->family) == 0 &&
2515                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2516                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2517                         return 1;
2518                 }
2519         } else {
2520                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2521                         return 1;
2522                 }
2523         }
2524         return 0;
2525 }
2526
2527 static struct xfrm_policy * xfrm_migrate_policy_find(struct xfrm_selector *sel,
2528                                                      u8 dir, u8 type)
2529 {
2530         struct xfrm_policy *pol, *ret = NULL;
2531         struct hlist_node *entry;
2532         struct hlist_head *chain;
2533         u32 priority = ~0U;
2534
2535         read_lock_bh(&xfrm_policy_lock);
2536         chain = policy_hash_direct(&sel->daddr, &sel->saddr, sel->family, dir);
2537         hlist_for_each_entry(pol, entry, chain, bydst) {
2538                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2539                     pol->type == type) {
2540                         ret = pol;
2541                         priority = ret->priority;
2542                         break;
2543                 }
2544         }
2545         chain = &xfrm_policy_inexact[dir];
2546         hlist_for_each_entry(pol, entry, chain, bydst) {
2547                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2548                     pol->type == type &&
2549                     pol->priority < priority) {
2550                         ret = pol;
2551                         break;
2552                 }
2553         }
2554
2555         if (ret)
2556                 xfrm_pol_hold(ret);
2557
2558         read_unlock_bh(&xfrm_policy_lock);
2559
2560         return ret;
2561 }
2562
2563 static int migrate_tmpl_match(struct xfrm_migrate *m, struct xfrm_tmpl *t)
2564 {
2565         int match = 0;
2566
2567         if (t->mode == m->mode && t->id.proto == m->proto &&
2568             (m->reqid == 0 || t->reqid == m->reqid)) {
2569                 switch (t->mode) {
2570                 case XFRM_MODE_TUNNEL:
2571                 case XFRM_MODE_BEET:
2572                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2573                                           m->old_family) == 0 &&
2574                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2575                                           m->old_family) == 0) {
2576                                 match = 1;
2577                         }
2578                         break;
2579                 case XFRM_MODE_TRANSPORT:
2580                         /* in case of transport mode, template does not store
2581                            any IP addresses, hence we just compare mode and
2582                            protocol */
2583                         match = 1;
2584                         break;
2585                 default:
2586                         break;
2587                 }
2588         }
2589         return match;
2590 }
2591
2592 /* update endpoint address(es) of template(s) */
2593 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2594                                struct xfrm_migrate *m, int num_migrate)
2595 {
2596         struct xfrm_migrate *mp;
2597         struct dst_entry *dst;
2598         int i, j, n = 0;
2599
2600         write_lock_bh(&pol->lock);
2601         if (unlikely(pol->dead)) {
2602                 /* target policy has been deleted */
2603                 write_unlock_bh(&pol->lock);
2604                 return -ENOENT;
2605         }
2606
2607         for (i = 0; i < pol->xfrm_nr; i++) {
2608                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2609                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2610                                 continue;
2611                         n++;
2612                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2613                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2614                                 continue;
2615                         /* update endpoints */
2616                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2617                                sizeof(pol->xfrm_vec[i].id.daddr));
2618                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2619                                sizeof(pol->xfrm_vec[i].saddr));
2620                         pol->xfrm_vec[i].encap_family = mp->new_family;
2621                         /* flush bundles */
2622                         while ((dst = pol->bundles) != NULL) {
2623                                 pol->bundles = dst->next;
2624                                 dst_free(dst);
2625                         }
2626                 }
2627         }
2628
2629         write_unlock_bh(&pol->lock);
2630
2631         if (!n)
2632                 return -ENODATA;
2633
2634         return 0;
2635 }
2636
2637 static int xfrm_migrate_check(struct xfrm_migrate *m, int num_migrate)
2638 {
2639         int i, j;
2640
2641         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2642                 return -EINVAL;
2643
2644         for (i = 0; i < num_migrate; i++) {
2645                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2646                                    m[i].old_family) == 0) &&
2647                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2648                                    m[i].old_family) == 0))
2649                         return -EINVAL;
2650                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2651                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2652                         return -EINVAL;
2653
2654                 /* check if there is any duplicated entry */
2655                 for (j = i + 1; j < num_migrate; j++) {
2656                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2657                                     sizeof(m[i].old_daddr)) &&
2658                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2659                                     sizeof(m[i].old_saddr)) &&
2660                             m[i].proto == m[j].proto &&
2661                             m[i].mode == m[j].mode &&
2662                             m[i].reqid == m[j].reqid &&
2663                             m[i].old_family == m[j].old_family)
2664                                 return -EINVAL;
2665                 }
2666         }
2667
2668         return 0;
2669 }
2670
2671 int xfrm_migrate(struct xfrm_selector *sel, u8 dir, u8 type,
2672                  struct xfrm_migrate *m, int num_migrate)
2673 {
2674         int i, err, nx_cur = 0, nx_new = 0;
2675         struct xfrm_policy *pol = NULL;
2676         struct xfrm_state *x, *xc;
2677         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2678         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2679         struct xfrm_migrate *mp;
2680
2681         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2682                 goto out;
2683
2684         /* Stage 1 - find policy */
2685         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2686                 err = -ENOENT;
2687                 goto out;
2688         }
2689
2690         /* Stage 2 - find and update state(s) */
2691         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2692                 if ((x = xfrm_migrate_state_find(mp))) {
2693                         x_cur[nx_cur] = x;
2694                         nx_cur++;
2695                         if ((xc = xfrm_state_migrate(x, mp))) {
2696                                 x_new[nx_new] = xc;
2697                                 nx_new++;
2698                         } else {
2699                                 err = -ENODATA;
2700                                 goto restore_state;
2701                         }
2702                 }
2703         }
2704
2705         /* Stage 3 - update policy */
2706         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2707                 goto restore_state;
2708
2709         /* Stage 4 - delete old state(s) */
2710         if (nx_cur) {
2711                 xfrm_states_put(x_cur, nx_cur);
2712                 xfrm_states_delete(x_cur, nx_cur);
2713         }
2714
2715         /* Stage 5 - announce */
2716         km_migrate(sel, dir, type, m, num_migrate);
2717
2718         xfrm_pol_put(pol);
2719
2720         return 0;
2721 out:
2722         return err;
2723
2724 restore_state:
2725         if (pol)
2726                 xfrm_pol_put(pol);
2727         if (nx_cur)
2728                 xfrm_states_put(x_cur, nx_cur);
2729         if (nx_new)
2730                 xfrm_states_delete(x_new, nx_new);
2731
2732         return err;
2733 }
2734 EXPORT_SYMBOL(xfrm_migrate);
2735 #endif