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