SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/xfrm/xfrm_policy.c

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