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