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/flow.h>
  30 #include <net/xfrm.h>
  31 #include <net/ip.h>
  32 #ifdef CONFIG_XFRM_STATISTICS
  33 #include <net/snmp.h>
  34 #endif
  35
  36 #include "xfrm_hash.h"
  37
  38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  40 #define XFRM_MAX_QUEUE_LEN      100
  41
  42 DEFINE_MUTEX(xfrm_cfg_mutex);
  43 EXPORT_SYMBOL(xfrm_cfg_mutex);
  44
  45 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
  46 static struct dst_entry *xfrm_policy_sk_bundles;
  47 static DEFINE_RWLOCK(xfrm_policy_lock);
  48
  49 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  50 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
  51                                                 __read_mostly;
  52
  53 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  54
  55 static void xfrm_init_pmtu(struct dst_entry *dst);
  56 static int stale_bundle(struct dst_entry *dst);
  57 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  58 static void xfrm_policy_queue_process(unsigned long arg);
  59
  60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  61                                                 int dir);
  62
  63 static inline bool
  64 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  65 {
  66         const struct flowi4 *fl4 = &fl->u.ip4;
  67
  68         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  69                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  70                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  71                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  72                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  73                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  74 }
  75
  76 static inline bool
  77 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  78 {
  79         const struct flowi6 *fl6 = &fl->u.ip6;
  80
  81         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  82                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  83                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  84                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  85                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  86                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  87 }
  88
  89 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  90                          unsigned short family)
  91 {
  92         switch (family) {
  93         case AF_INET:
  94                 return __xfrm4_selector_match(sel, fl);
  95         case AF_INET6:
  96                 return __xfrm6_selector_match(sel, fl);
  97         }
  98         return false;
  99 }
 100
 101 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 102 {
 103         struct xfrm_policy_afinfo *afinfo;
 104
 105         if (unlikely(family >= NPROTO))
 106                 return NULL;
 107         rcu_read_lock();
 108         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 109         if (unlikely(!afinfo))
 110                 rcu_read_unlock();
 111         return afinfo;
 112 }
 113
 114 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
 115 {
 116         rcu_read_unlock();
 117 }
 118
 119 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
 120                                                   const xfrm_address_t *saddr,
 121                                                   const xfrm_address_t *daddr,
 122                                                   int family)
 123 {
 124         struct xfrm_policy_afinfo *afinfo;
 125         struct dst_entry *dst;
 126
 127         afinfo = xfrm_policy_get_afinfo(family);
 128         if (unlikely(afinfo == NULL))
 129                 return ERR_PTR(-EAFNOSUPPORT);
 130
 131         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 132
 133         xfrm_policy_put_afinfo(afinfo);
 134
 135         return dst;
 136 }
 137
 138 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 139                                                 xfrm_address_t *prev_saddr,
 140                                                 xfrm_address_t *prev_daddr,
 141                                                 int family)
 142 {
 143         struct net *net = xs_net(x);
 144         xfrm_address_t *saddr = &x->props.saddr;
 145         xfrm_address_t *daddr = &x->id.daddr;
 146         struct dst_entry *dst;
 147
 148         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 149                 saddr = x->coaddr;
 150                 daddr = prev_daddr;
 151         }
 152         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 153                 saddr = prev_saddr;
 154                 daddr = x->coaddr;
 155         }
 156
 157         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 158
 159         if (!IS_ERR(dst)) {
 160                 if (prev_saddr != saddr)
 161                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 162                 if (prev_daddr != daddr)
 163                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 164         }
 165
 166         return dst;
 167 }
 168
 169 static inline unsigned long make_jiffies(long secs)
 170 {
 171         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 172                 return MAX_SCHEDULE_TIMEOUT-1;
 173         else
 174                 return secs*HZ;
 175 }
 176
 177 static void xfrm_policy_timer(unsigned long data)
 178 {
 179         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 180         unsigned long now = get_seconds();
 181         long next = LONG_MAX;
 182         int warn = 0;
 183         int dir;
 184
 185         read_lock(&xp->lock);
 186
 187         if (unlikely(xp->walk.dead))
 188                 goto out;
 189
 190         dir = xfrm_policy_id2dir(xp->index);
 191
 192         if (xp->lft.hard_add_expires_seconds) {
 193                 long tmo = xp->lft.hard_add_expires_seconds +
 194                         xp->curlft.add_time - now;
 195                 if (tmo <= 0)
 196                         goto expired;
 197                 if (tmo < next)
 198                         next = tmo;
 199         }
 200         if (xp->lft.hard_use_expires_seconds) {
 201                 long tmo = xp->lft.hard_use_expires_seconds +
 202                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 203                 if (tmo <= 0)
 204                         goto expired;
 205                 if (tmo < next)
 206                         next = tmo;
 207         }
 208         if (xp->lft.soft_add_expires_seconds) {
 209                 long tmo = xp->lft.soft_add_expires_seconds +
 210                         xp->curlft.add_time - now;
 211                 if (tmo <= 0) {
 212                         warn = 1;
 213                         tmo = XFRM_KM_TIMEOUT;
 214                 }
 215                 if (tmo < next)
 216                         next = tmo;
 217         }
 218         if (xp->lft.soft_use_expires_seconds) {
 219                 long tmo = xp->lft.soft_use_expires_seconds +
 220                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 221                 if (tmo <= 0) {
 222                         warn = 1;
 223                         tmo = XFRM_KM_TIMEOUT;
 224                 }
 225                 if (tmo < next)
 226                         next = tmo;
 227         }
 228
 229         if (warn)
 230                 km_policy_expired(xp, dir, 0, 0);
 231         if (next != LONG_MAX &&
 232             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 233                 xfrm_pol_hold(xp);
 234
 235 out:
 236         read_unlock(&xp->lock);
 237         xfrm_pol_put(xp);
 238         return;
 239
 240 expired:
 241         read_unlock(&xp->lock);
 242         if (!xfrm_policy_delete(xp, dir))
 243                 km_policy_expired(xp, dir, 1, 0);
 244         xfrm_pol_put(xp);
 245 }
 246
 247 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 248 {
 249         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 250
 251         if (unlikely(pol->walk.dead))
 252                 flo = NULL;
 253         else
 254                 xfrm_pol_hold(pol);
 255
 256         return flo;
 257 }
 258
 259 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 260 {
 261         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 262
 263         return !pol->walk.dead;
 264 }
 265
 266 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 267 {
 268         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 269 }
 270
 271 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 272         .get = xfrm_policy_flo_get,
 273         .check = xfrm_policy_flo_check,
 274         .delete = xfrm_policy_flo_delete,
 275 };
 276
 277 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 278  * SPD calls.
 279  */
 280
 281 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 282 {
 283         struct xfrm_policy *policy;
 284
 285         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 286
 287         if (policy) {
 288                 write_pnet(&policy->xp_net, net);
 289                 INIT_LIST_HEAD(&policy->walk.all);
 290                 INIT_HLIST_NODE(&policy->bydst);
 291                 INIT_HLIST_NODE(&policy->byidx);
 292                 rwlock_init(&policy->lock);
 293                 atomic_set(&policy->refcnt, 1);
 294                 skb_queue_head_init(&policy->polq.hold_queue);
 295                 setup_timer(&policy->timer, xfrm_policy_timer,
 296                                 (unsigned long)policy);
 297                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 298                             (unsigned long)policy);
 299                 policy->flo.ops = &xfrm_policy_fc_ops;
 300         }
 301         return policy;
 302 }
 303 EXPORT_SYMBOL(xfrm_policy_alloc);
 304
 305 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 306
 307 void xfrm_policy_destroy(struct xfrm_policy *policy)
 308 {
 309         BUG_ON(!policy->walk.dead);
 310
 311         if (del_timer(&policy->timer))
 312                 BUG();
 313
 314         security_xfrm_policy_free(policy->security);
 315         kfree(policy);
 316 }
 317 EXPORT_SYMBOL(xfrm_policy_destroy);
 318
 319 static void xfrm_queue_purge(struct sk_buff_head *list)
 320 {
 321         struct sk_buff *skb;
 322
 323         while ((skb = skb_dequeue(list)) != NULL) {
 324                 dev_put(skb->dev);
 325                 kfree_skb(skb);
 326         }
 327 }
 328
 329 /* Rule must be locked. Release descentant resources, announce
 330  * entry dead. The rule must be unlinked from lists to the moment.
 331  */
 332
 333 static void xfrm_policy_kill(struct xfrm_policy *policy)
 334 {
 335         policy->walk.dead = 1;
 336
 337         atomic_inc(&policy->genid);
 338
 339         del_timer(&policy->polq.hold_timer);
 340         xfrm_queue_purge(&policy->polq.hold_queue);
 341
 342         if (del_timer(&policy->timer))
 343                 xfrm_pol_put(policy);
 344
 345         xfrm_pol_put(policy);
 346 }
 347
 348 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 349
 350 static inline unsigned int idx_hash(struct net *net, u32 index)
 351 {
 352         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 353 }
 354
 355 static struct hlist_head *policy_hash_bysel(struct net *net,
 356                                             const struct xfrm_selector *sel,
 357                                             unsigned short family, int dir)
 358 {
 359         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 360         unsigned int hash = __sel_hash(sel, family, hmask);
 361
 362         return (hash == hmask + 1 ?
 363                 &net->xfrm.policy_inexact[dir] :
 364                 net->xfrm.policy_bydst[dir].table + hash);
 365 }
 366
 367 static struct hlist_head *policy_hash_direct(struct net *net,
 368                                              const xfrm_address_t *daddr,
 369                                              const xfrm_address_t *saddr,
 370                                              unsigned short family, int dir)
 371 {
 372         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 373         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 374
 375         return net->xfrm.policy_bydst[dir].table + hash;
 376 }
 377
 378 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 379                                    struct hlist_head *ndsttable,
 380                                    unsigned int nhashmask)
 381 {
 382         struct hlist_node *entry, *tmp, *entry0 = NULL;
 383         struct xfrm_policy *pol;
 384         unsigned int h0 = 0;
 385
 386 redo:
 387         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 388                 unsigned int h;
 389
 390                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 391                                 pol->family, nhashmask);
 392                 if (!entry0) {
 393                         hlist_del(entry);
 394                         hlist_add_head(&pol->bydst, ndsttable+h);
 395                         h0 = h;
 396                 } else {
 397                         if (h != h0)
 398                                 continue;
 399                         hlist_del(entry);
 400                         hlist_add_after(entry0, &pol->bydst);
 401                 }
 402                 entry0 = entry;
 403         }
 404         if (!hlist_empty(list)) {
 405                 entry0 = NULL;
 406                 goto redo;
 407         }
 408 }
 409
 410 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 411                                    struct hlist_head *nidxtable,
 412                                    unsigned int nhashmask)
 413 {
 414         struct hlist_node *entry, *tmp;
 415         struct xfrm_policy *pol;
 416
 417         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 418                 unsigned int h;
 419
 420                 h = __idx_hash(pol->index, nhashmask);
 421                 hlist_add_head(&pol->byidx, nidxtable+h);
 422         }
 423 }
 424
 425 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 426 {
 427         return ((old_hmask + 1) << 1) - 1;
 428 }
 429
 430 static void xfrm_bydst_resize(struct net *net, int dir)
 431 {
 432         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 433         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 434         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 435         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 436         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 437         int i;
 438
 439         if (!ndst)
 440                 return;
 441
 442         write_lock_bh(&xfrm_policy_lock);
 443
 444         for (i = hmask; i >= 0; i--)
 445                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 446
 447         net->xfrm.policy_bydst[dir].table = ndst;
 448         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 449
 450         write_unlock_bh(&xfrm_policy_lock);
 451
 452         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 453 }
 454
 455 static void xfrm_byidx_resize(struct net *net, int total)
 456 {
 457         unsigned int hmask = net->xfrm.policy_idx_hmask;
 458         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 459         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 460         struct hlist_head *oidx = net->xfrm.policy_byidx;
 461         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 462         int i;
 463
 464         if (!nidx)
 465                 return;
 466
 467         write_lock_bh(&xfrm_policy_lock);
 468
 469         for (i = hmask; i >= 0; i--)
 470                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 471
 472         net->xfrm.policy_byidx = nidx;
 473         net->xfrm.policy_idx_hmask = nhashmask;
 474
 475         write_unlock_bh(&xfrm_policy_lock);
 476
 477         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 478 }
 479
 480 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 481 {
 482         unsigned int cnt = net->xfrm.policy_count[dir];
 483         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 484
 485         if (total)
 486                 *total += cnt;
 487
 488         if ((hmask + 1) < xfrm_policy_hashmax &&
 489             cnt > hmask)
 490                 return 1;
 491
 492         return 0;
 493 }
 494
 495 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 496 {
 497         unsigned int hmask = net->xfrm.policy_idx_hmask;
 498
 499         if ((hmask + 1) < xfrm_policy_hashmax &&
 500             total > hmask)
 501                 return 1;
 502
 503         return 0;
 504 }
 505
 506 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 507 {
 508         read_lock_bh(&xfrm_policy_lock);
 509         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 510         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 511         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 512         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 513         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 514         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 515         si->spdhcnt = net->xfrm.policy_idx_hmask;
 516         si->spdhmcnt = xfrm_policy_hashmax;
 517         read_unlock_bh(&xfrm_policy_lock);
 518 }
 519 EXPORT_SYMBOL(xfrm_spd_getinfo);
 520
 521 static DEFINE_MUTEX(hash_resize_mutex);
 522 static void xfrm_hash_resize(struct work_struct *work)
 523 {
 524         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 525         int dir, total;
 526
 527         mutex_lock(&hash_resize_mutex);
 528
 529         total = 0;
 530         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 531                 if (xfrm_bydst_should_resize(net, dir, &total))
 532                         xfrm_bydst_resize(net, dir);
 533         }
 534         if (xfrm_byidx_should_resize(net, total))
 535                 xfrm_byidx_resize(net, total);
 536
 537         mutex_unlock(&hash_resize_mutex);
 538 }
 539
 540 /* Generate new index... KAME seems to generate them ordered by cost
 541  * of an absolute inpredictability of ordering of rules. This will not pass. */
 542 static u32 xfrm_gen_index(struct net *net, int dir)
 543 {
 544         static u32 idx_generator;
 545
 546         for (;;) {
 547                 struct hlist_node *entry;
 548                 struct hlist_head *list;
 549                 struct xfrm_policy *p;
 550                 u32 idx;
 551                 int found;
 552
 553                 idx = (idx_generator | dir);
 554                 idx_generator += 8;
 555                 if (idx == 0)
 556                         idx = 8;
 557                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 558                 found = 0;
 559                 hlist_for_each_entry(p, entry, list, byidx) {
 560                         if (p->index == idx) {
 561                                 found = 1;
 562                                 break;
 563                         }
 564                 }
 565                 if (!found)
 566                         return idx;
 567         }
 568 }
 569
 570 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 571 {
 572         u32 *p1 = (u32 *) s1;
 573         u32 *p2 = (u32 *) s2;
 574         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 575         int i;
 576
 577         for (i = 0; i < len; i++) {
 578                 if (p1[i] != p2[i])
 579                         return 1;
 580         }
 581
 582         return 0;
 583 }
 584
 585 static void xfrm_policy_requeue(struct xfrm_policy *old,
 586                                 struct xfrm_policy *new)
 587 {
 588         struct xfrm_policy_queue *pq = &old->polq;
 589         struct sk_buff_head list;
 590
 591         __skb_queue_head_init(&list);
 592
 593         spin_lock_bh(&pq->hold_queue.lock);
 594         skb_queue_splice_init(&pq->hold_queue, &list);
 595         del_timer(&pq->hold_timer);
 596         spin_unlock_bh(&pq->hold_queue.lock);
 597
 598         if (skb_queue_empty(&list))
 599                 return;
 600
 601         pq = &new->polq;
 602
 603         spin_lock_bh(&pq->hold_queue.lock);
 604         skb_queue_splice(&list, &pq->hold_queue);
 605         pq->timeout = XFRM_QUEUE_TMO_MIN;
 606         mod_timer(&pq->hold_timer, jiffies);
 607         spin_unlock_bh(&pq->hold_queue.lock);
 608 }
 609
 610 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 611                                    struct xfrm_policy *pol)
 612 {
 613         u32 mark = policy->mark.v & policy->mark.m;
 614
 615         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 616                 return true;
 617
 618         if ((mark & pol->mark.m) == pol->mark.v &&
 619             policy->priority == pol->priority)
 620                 return true;
 621
 622         return false;
 623 }
 624
 625 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 626 {
 627         struct net *net = xp_net(policy);
 628         struct xfrm_policy *pol;
 629         struct xfrm_policy *delpol;
 630         struct hlist_head *chain;
 631         struct hlist_node *entry, *newpos;
 632
 633         write_lock_bh(&xfrm_policy_lock);
 634         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 635         delpol = NULL;
 636         newpos = NULL;
 637         hlist_for_each_entry(pol, entry, chain, bydst) {
 638                 if (pol->type == policy->type &&
 639                     !selector_cmp(&pol->selector, &policy->selector) &&
 640                     xfrm_policy_mark_match(policy, pol) &&
 641                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 642                     !WARN_ON(delpol)) {
 643                         if (excl) {
 644                                 write_unlock_bh(&xfrm_policy_lock);
 645                                 return -EEXIST;
 646                         }
 647                         delpol = pol;
 648                         if (policy->priority > pol->priority)
 649                                 continue;
 650                 } else if (policy->priority >= pol->priority) {
 651                         newpos = &pol->bydst;
 652                         continue;
 653                 }
 654                 if (delpol)
 655                         break;
 656         }
 657         if (newpos)
 658                 hlist_add_after(newpos, &policy->bydst);
 659         else
 660                 hlist_add_head(&policy->bydst, chain);
 661         xfrm_pol_hold(policy);
 662         net->xfrm.policy_count[dir]++;
 663         atomic_inc(&flow_cache_genid);
 664         rt_genid_bump(net);
 665         if (delpol) {
 666                 xfrm_policy_requeue(delpol, policy);
 667                 __xfrm_policy_unlink(delpol, dir);
 668         }
 669         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 670         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 671         policy->curlft.add_time = get_seconds();
 672         policy->curlft.use_time = 0;
 673         if (!mod_timer(&policy->timer, jiffies + HZ))
 674                 xfrm_pol_hold(policy);
 675         list_add(&policy->walk.all, &net->xfrm.policy_all);
 676         write_unlock_bh(&xfrm_policy_lock);
 677
 678         if (delpol)
 679                 xfrm_policy_kill(delpol);
 680         else if (xfrm_bydst_should_resize(net, dir, NULL))
 681                 schedule_work(&net->xfrm.policy_hash_work);
 682
 683         return 0;
 684 }
 685 EXPORT_SYMBOL(xfrm_policy_insert);
 686
 687 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 688                                           int dir, struct xfrm_selector *sel,
 689                                           struct xfrm_sec_ctx *ctx, int delete,
 690                                           int *err)
 691 {
 692         struct xfrm_policy *pol, *ret;
 693         struct hlist_head *chain;
 694         struct hlist_node *entry;
 695
 696         *err = 0;
 697         write_lock_bh(&xfrm_policy_lock);
 698         chain = policy_hash_bysel(net, sel, sel->family, dir);
 699         ret = NULL;
 700         hlist_for_each_entry(pol, entry, chain, bydst) {
 701                 if (pol->type == type &&
 702                     (mark & pol->mark.m) == pol->mark.v &&
 703                     !selector_cmp(sel, &pol->selector) &&
 704                     xfrm_sec_ctx_match(ctx, pol->security)) {
 705                         xfrm_pol_hold(pol);
 706                         if (delete) {
 707                                 *err = security_xfrm_policy_delete(
 708                                                                 pol->security);
 709                                 if (*err) {
 710                                         write_unlock_bh(&xfrm_policy_lock);
 711                                         return pol;
 712                                 }
 713                                 __xfrm_policy_unlink(pol, dir);
 714                         }
 715                         ret = pol;
 716                         break;
 717                 }
 718         }
 719         write_unlock_bh(&xfrm_policy_lock);
 720
 721         if (ret && delete)
 722                 xfrm_policy_kill(ret);
 723         return ret;
 724 }
 725 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 726
 727 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 728                                      int dir, u32 id, int delete, int *err)
 729 {
 730         struct xfrm_policy *pol, *ret;
 731         struct hlist_head *chain;
 732         struct hlist_node *entry;
 733
 734         *err = -ENOENT;
 735         if (xfrm_policy_id2dir(id) != dir)
 736                 return NULL;
 737
 738         *err = 0;
 739         write_lock_bh(&xfrm_policy_lock);
 740         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 741         ret = NULL;
 742         hlist_for_each_entry(pol, entry, chain, byidx) {
 743                 if (pol->type == type && pol->index == id &&
 744                     (mark & pol->mark.m) == pol->mark.v) {
 745                         xfrm_pol_hold(pol);
 746                         if (delete) {
 747                                 *err = security_xfrm_policy_delete(
 748                                                                 pol->security);
 749                                 if (*err) {
 750                                         write_unlock_bh(&xfrm_policy_lock);
 751                                         return pol;
 752                                 }
 753                                 __xfrm_policy_unlink(pol, dir);
 754                         }
 755                         ret = pol;
 756                         break;
 757                 }
 758         }
 759         write_unlock_bh(&xfrm_policy_lock);
 760
 761         if (ret && delete)
 762                 xfrm_policy_kill(ret);
 763         return ret;
 764 }
 765 EXPORT_SYMBOL(xfrm_policy_byid);
 766
 767 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 768 static inline int
 769 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 770 {
 771         int dir, err = 0;
 772
 773         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 774                 struct xfrm_policy *pol;
 775                 struct hlist_node *entry;
 776                 int i;
 777
 778                 hlist_for_each_entry(pol, entry,
 779                                      &net->xfrm.policy_inexact[dir], bydst) {
 780                         if (pol->type != type)
 781                                 continue;
 782                         err = security_xfrm_policy_delete(pol->security);
 783                         if (err) {
 784                                 xfrm_audit_policy_delete(pol, 0,
 785                                                          audit_info->loginuid,
 786                                                          audit_info->sessionid,
 787                                                          audit_info->secid);
 788                                 return err;
 789                         }
 790                 }
 791                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 792                         hlist_for_each_entry(pol, entry,
 793                                              net->xfrm.policy_bydst[dir].table + i,
 794                                              bydst) {
 795                                 if (pol->type != type)
 796                                         continue;
 797                                 err = security_xfrm_policy_delete(
 798                                                                 pol->security);
 799                                 if (err) {
 800                                         xfrm_audit_policy_delete(pol, 0,
 801                                                         audit_info->loginuid,
 802                                                         audit_info->sessionid,
 803                                                         audit_info->secid);
 804                                         return err;
 805                                 }
 806                         }
 807                 }
 808         }
 809         return err;
 810 }
 811 #else
 812 static inline int
 813 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 814 {
 815         return 0;
 816 }
 817 #endif
 818
 819 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 820 {
 821         int dir, err = 0, cnt = 0;
 822
 823         write_lock_bh(&xfrm_policy_lock);
 824
 825         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 826         if (err)
 827                 goto out;
 828
 829         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 830                 struct xfrm_policy *pol;
 831                 struct hlist_node *entry;
 832                 int i;
 833
 834         again1:
 835                 hlist_for_each_entry(pol, entry,
 836                                      &net->xfrm.policy_inexact[dir], bydst) {
 837                         if (pol->type != type)
 838                                 continue;
 839                         __xfrm_policy_unlink(pol, dir);
 840                         write_unlock_bh(&xfrm_policy_lock);
 841                         cnt++;
 842
 843                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 844                                                  audit_info->sessionid,
 845                                                  audit_info->secid);
 846
 847                         xfrm_policy_kill(pol);
 848
 849                         write_lock_bh(&xfrm_policy_lock);
 850                         goto again1;
 851                 }
 852
 853                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 854         again2:
 855                         hlist_for_each_entry(pol, entry,
 856                                              net->xfrm.policy_bydst[dir].table + i,
 857                                              bydst) {
 858                                 if (pol->type != type)
 859                                         continue;
 860                                 __xfrm_policy_unlink(pol, dir);
 861                                 write_unlock_bh(&xfrm_policy_lock);
 862                                 cnt++;
 863
 864                                 xfrm_audit_policy_delete(pol, 1,
 865                                                          audit_info->loginuid,
 866                                                          audit_info->sessionid,
 867                                                          audit_info->secid);
 868                                 xfrm_policy_kill(pol);
 869
 870                                 write_lock_bh(&xfrm_policy_lock);
 871                                 goto again2;
 872                         }
 873                 }
 874
 875         }
 876         if (!cnt)
 877                 err = -ESRCH;
 878 out:
 879         write_unlock_bh(&xfrm_policy_lock);
 880         return err;
 881 }
 882 EXPORT_SYMBOL(xfrm_policy_flush);
 883
 884 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 885                      int (*func)(struct xfrm_policy *, int, int, void*),
 886                      void *data)
 887 {
 888         struct xfrm_policy *pol;
 889         struct xfrm_policy_walk_entry *x;
 890         int error = 0;
 891
 892         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 893             walk->type != XFRM_POLICY_TYPE_ANY)
 894                 return -EINVAL;
 895
 896         if (list_empty(&walk->walk.all) && walk->seq != 0)
 897                 return 0;
 898
 899         write_lock_bh(&xfrm_policy_lock);
 900         if (list_empty(&walk->walk.all))
 901                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 902         else
 903                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 904         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 905                 if (x->dead)
 906                         continue;
 907                 pol = container_of(x, struct xfrm_policy, walk);
 908                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 909                     walk->type != pol->type)
 910                         continue;
 911                 error = func(pol, xfrm_policy_id2dir(pol->index),
 912                              walk->seq, data);
 913                 if (error) {
 914                         list_move_tail(&walk->walk.all, &x->all);
 915                         goto out;
 916                 }
 917                 walk->seq++;
 918         }
 919         if (walk->seq == 0) {
 920                 error = -ENOENT;
 921                 goto out;
 922         }
 923         list_del_init(&walk->walk.all);
 924 out:
 925         write_unlock_bh(&xfrm_policy_lock);
 926         return error;
 927 }
 928 EXPORT_SYMBOL(xfrm_policy_walk);
 929
 930 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 931 {
 932         INIT_LIST_HEAD(&walk->walk.all);
 933         walk->walk.dead = 1;
 934         walk->type = type;
 935         walk->seq = 0;
 936 }
 937 EXPORT_SYMBOL(xfrm_policy_walk_init);
 938
 939 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 940 {
 941         if (list_empty(&walk->walk.all))
 942                 return;
 943
 944         write_lock_bh(&xfrm_policy_lock);
 945         list_del(&walk->walk.all);
 946         write_unlock_bh(&xfrm_policy_lock);
 947 }
 948 EXPORT_SYMBOL(xfrm_policy_walk_done);
 949
 950 /*
 951  * Find policy to apply to this flow.
 952  *
 953  * Returns 0 if policy found, else an -errno.
 954  */
 955 static int xfrm_policy_match(const struct xfrm_policy *pol,
 956                              const struct flowi *fl,
 957                              u8 type, u16 family, int dir)
 958 {
 959         const struct xfrm_selector *sel = &pol->selector;
 960         int ret = -ESRCH;
 961         bool match;
 962
 963         if (pol->family != family ||
 964             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
 965             pol->type != type)
 966                 return ret;
 967
 968         match = xfrm_selector_match(sel, fl, family);
 969         if (match)
 970                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
 971                                                   dir);
 972
 973         return ret;
 974 }
 975
 976 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 977                                                      const struct flowi *fl,
 978                                                      u16 family, u8 dir)
 979 {
 980         int err;
 981         struct xfrm_policy *pol, *ret;
 982         const xfrm_address_t *daddr, *saddr;
 983         struct hlist_node *entry;
 984         struct hlist_head *chain;
 985         u32 priority = ~0U;
 986
 987         daddr = xfrm_flowi_daddr(fl, family);
 988         saddr = xfrm_flowi_saddr(fl, family);
 989         if (unlikely(!daddr || !saddr))
 990                 return NULL;
 991
 992         read_lock_bh(&xfrm_policy_lock);
 993         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 994         ret = NULL;
 995         hlist_for_each_entry(pol, entry, chain, bydst) {
 996                 err = xfrm_policy_match(pol, fl, type, family, dir);
 997                 if (err) {
 998                         if (err == -ESRCH)
 999                                 continue;
1000                         else {
1001                                 ret = ERR_PTR(err);
1002                                 goto fail;
1003                         }
1004                 } else {
1005                         ret = pol;
1006                         priority = ret->priority;
1007                         break;
1008                 }
1009         }
1010         chain = &net->xfrm.policy_inexact[dir];
1011         hlist_for_each_entry(pol, entry, chain, bydst) {
1012                 err = xfrm_policy_match(pol, fl, type, family, dir);
1013                 if (err) {
1014                         if (err == -ESRCH)
1015                                 continue;
1016                         else {
1017                                 ret = ERR_PTR(err);
1018                                 goto fail;
1019                         }
1020                 } else if (pol->priority < priority) {
1021                         ret = pol;
1022                         break;
1023                 }
1024         }
1025         if (ret)
1026                 xfrm_pol_hold(ret);
1027 fail:
1028         read_unlock_bh(&xfrm_policy_lock);
1029
1030         return ret;
1031 }
1032
1033 static struct xfrm_policy *
1034 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1035 {
1036 #ifdef CONFIG_XFRM_SUB_POLICY
1037         struct xfrm_policy *pol;
1038
1039         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1040         if (pol != NULL)
1041                 return pol;
1042 #endif
1043         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1044 }
1045
1046 static struct flow_cache_object *
1047 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1048                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1049 {
1050         struct xfrm_policy *pol;
1051
1052         if (old_obj)
1053                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1054
1055         pol = __xfrm_policy_lookup(net, fl, family, dir);
1056         if (IS_ERR_OR_NULL(pol))
1057                 return ERR_CAST(pol);
1058
1059         /* Resolver returns two references:
1060          * one for cache and one for caller of flow_cache_lookup() */
1061         xfrm_pol_hold(pol);
1062
1063         return &pol->flo;
1064 }
1065
1066 static inline int policy_to_flow_dir(int dir)
1067 {
1068         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1069             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1070             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1071                 return dir;
1072         switch (dir) {
1073         default:
1074         case XFRM_POLICY_IN:
1075                 return FLOW_DIR_IN;
1076         case XFRM_POLICY_OUT:
1077                 return FLOW_DIR_OUT;
1078         case XFRM_POLICY_FWD:
1079                 return FLOW_DIR_FWD;
1080         }
1081 }
1082
1083 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1084                                                  const struct flowi *fl)
1085 {
1086         struct xfrm_policy *pol;
1087
1088         read_lock_bh(&xfrm_policy_lock);
1089         if ((pol = sk->sk_policy[dir]) != NULL) {
1090                 bool match = xfrm_selector_match(&pol->selector, fl,
1091                                                  sk->sk_family);
1092                 int err = 0;
1093
1094                 if (match) {
1095                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1096                                 pol = NULL;
1097                                 goto out;
1098                         }
1099                         err = security_xfrm_policy_lookup(pol->security,
1100                                                       fl->flowi_secid,
1101                                                       policy_to_flow_dir(dir));
1102                         if (!err)
1103                                 xfrm_pol_hold(pol);
1104                         else if (err == -ESRCH)
1105                                 pol = NULL;
1106                         else
1107                                 pol = ERR_PTR(err);
1108                 } else
1109                         pol = NULL;
1110         }
1111 out:
1112         read_unlock_bh(&xfrm_policy_lock);
1113         return pol;
1114 }
1115
1116 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1117 {
1118         struct net *net = xp_net(pol);
1119         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1120                                                      pol->family, dir);
1121
1122         list_add(&pol->walk.all, &net->xfrm.policy_all);
1123         hlist_add_head(&pol->bydst, chain);
1124         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1125         net->xfrm.policy_count[dir]++;
1126         xfrm_pol_hold(pol);
1127
1128         if (xfrm_bydst_should_resize(net, dir, NULL))
1129                 schedule_work(&net->xfrm.policy_hash_work);
1130 }
1131
1132 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1133                                                 int dir)
1134 {
1135         struct net *net = xp_net(pol);
1136
1137         if (hlist_unhashed(&pol->bydst))
1138                 return NULL;
1139
1140         hlist_del(&pol->bydst);
1141         hlist_del(&pol->byidx);
1142         list_del(&pol->walk.all);
1143         net->xfrm.policy_count[dir]--;
1144
1145         return pol;
1146 }
1147
1148 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1149 {
1150         write_lock_bh(&xfrm_policy_lock);
1151         pol = __xfrm_policy_unlink(pol, dir);
1152         write_unlock_bh(&xfrm_policy_lock);
1153         if (pol) {
1154                 xfrm_policy_kill(pol);
1155                 return 0;
1156         }
1157         return -ENOENT;
1158 }
1159 EXPORT_SYMBOL(xfrm_policy_delete);
1160
1161 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1162 {
1163         struct net *net = xp_net(pol);
1164         struct xfrm_policy *old_pol;
1165
1166 #ifdef CONFIG_XFRM_SUB_POLICY
1167         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1168                 return -EINVAL;
1169 #endif
1170
1171         write_lock_bh(&xfrm_policy_lock);
1172         old_pol = sk->sk_policy[dir];
1173         sk->sk_policy[dir] = pol;
1174         if (pol) {
1175                 pol->curlft.add_time = get_seconds();
1176                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1177                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1178         }
1179         if (old_pol) {
1180                 if (pol)
1181                         xfrm_policy_requeue(old_pol, pol);
1182
1183                 /* Unlinking succeeds always. This is the only function
1184                  * allowed to delete or replace socket policy.
1185                  */
1186                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1187         }
1188         write_unlock_bh(&xfrm_policy_lock);
1189
1190         if (old_pol) {
1191                 xfrm_policy_kill(old_pol);
1192         }
1193         return 0;
1194 }
1195
1196 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1197 {
1198         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1199
1200         if (newp) {
1201                 newp->selector = old->selector;
1202                 if (security_xfrm_policy_clone(old->security,
1203                                                &newp->security)) {
1204                         kfree(newp);
1205                         return NULL;  /* ENOMEM */
1206                 }
1207                 newp->lft = old->lft;
1208                 newp->curlft = old->curlft;
1209                 newp->mark = old->mark;
1210                 newp->action = old->action;
1211                 newp->flags = old->flags;
1212                 newp->xfrm_nr = old->xfrm_nr;
1213                 newp->index = old->index;
1214                 newp->type = old->type;
1215                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1216                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1217                 write_lock_bh(&xfrm_policy_lock);
1218                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1219                 write_unlock_bh(&xfrm_policy_lock);
1220                 xfrm_pol_put(newp);
1221         }
1222         return newp;
1223 }
1224
1225 int __xfrm_sk_clone_policy(struct sock *sk)
1226 {
1227         struct xfrm_policy *p0 = sk->sk_policy[0],
1228                            *p1 = sk->sk_policy[1];
1229
1230         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1231         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1232                 return -ENOMEM;
1233         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1234                 return -ENOMEM;
1235         return 0;
1236 }
1237
1238 static int
1239 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1240                unsigned short family)
1241 {
1242         int err;
1243         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1244
1245         if (unlikely(afinfo == NULL))
1246                 return -EINVAL;
1247         err = afinfo->get_saddr(net, local, remote);
1248         xfrm_policy_put_afinfo(afinfo);
1249         return err;
1250 }
1251
1252 /* Resolve list of templates for the flow, given policy. */
1253
1254 static int
1255 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1256                       struct xfrm_state **xfrm, unsigned short family)
1257 {
1258         struct net *net = xp_net(policy);
1259         int nx;
1260         int i, error;
1261         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1262         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1263         xfrm_address_t tmp;
1264
1265         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1266                 struct xfrm_state *x;
1267                 xfrm_address_t *remote = daddr;
1268                 xfrm_address_t *local  = saddr;
1269                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1270
1271                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1272                     tmpl->mode == XFRM_MODE_BEET) {
1273                         remote = &tmpl->id.daddr;
1274                         local = &tmpl->saddr;
1275                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1276                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1277                                 if (error)
1278                                         goto fail;
1279                                 local = &tmp;
1280                         }
1281                 }
1282
1283                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1284
1285                 if (x && x->km.state == XFRM_STATE_VALID) {
1286                         xfrm[nx++] = x;
1287                         daddr = remote;
1288                         saddr = local;
1289                         continue;
1290                 }
1291                 if (x) {
1292                         error = (x->km.state == XFRM_STATE_ERROR ?
1293                                  -EINVAL : -EAGAIN);
1294                         xfrm_state_put(x);
1295                 }
1296                 else if (error == -ESRCH)
1297                         error = -EAGAIN;
1298
1299                 if (!tmpl->optional)
1300                         goto fail;
1301         }
1302         return nx;
1303
1304 fail:
1305         for (nx--; nx>=0; nx--)
1306                 xfrm_state_put(xfrm[nx]);
1307         return error;
1308 }
1309
1310 static int
1311 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1312                   struct xfrm_state **xfrm, unsigned short family)
1313 {
1314         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1315         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1316         int cnx = 0;
1317         int error;
1318         int ret;
1319         int i;
1320
1321         for (i = 0; i < npols; i++) {
1322                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1323                         error = -ENOBUFS;
1324                         goto fail;
1325                 }
1326
1327                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1328                 if (ret < 0) {
1329                         error = ret;
1330                         goto fail;
1331                 } else
1332                         cnx += ret;
1333         }
1334
1335         /* found states are sorted for outbound processing */
1336         if (npols > 1)
1337                 xfrm_state_sort(xfrm, tpp, cnx, family);
1338
1339         return cnx;
1340
1341  fail:
1342         for (cnx--; cnx>=0; cnx--)
1343                 xfrm_state_put(tpp[cnx]);
1344         return error;
1345
1346 }
1347
1348 /* Check that the bundle accepts the flow and its components are
1349  * still valid.
1350  */
1351
1352 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1353 {
1354         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1355         int tos;
1356
1357         if (!afinfo)
1358                 return -EINVAL;
1359
1360         tos = afinfo->get_tos(fl);
1361
1362         xfrm_policy_put_afinfo(afinfo);
1363
1364         return tos;
1365 }
1366
1367 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1368 {
1369         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1370         struct dst_entry *dst = &xdst->u.dst;
1371
1372         if (xdst->route == NULL) {
1373                 /* Dummy bundle - if it has xfrms we were not
1374                  * able to build bundle as template resolution failed.
1375                  * It means we need to try again resolving. */
1376                 if (xdst->num_xfrms > 0)
1377                         return NULL;
1378         } else if (dst->flags & DST_XFRM_QUEUE) {
1379                 return NULL;
1380         } else {
1381                 /* Real bundle */
1382                 if (stale_bundle(dst))
1383                         return NULL;
1384         }
1385
1386         dst_hold(dst);
1387         return flo;
1388 }
1389
1390 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1391 {
1392         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1393         struct dst_entry *dst = &xdst->u.dst;
1394
1395         if (!xdst->route)
1396                 return 0;
1397         if (stale_bundle(dst))
1398                 return 0;
1399
1400         return 1;
1401 }
1402
1403 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1404 {
1405         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1406         struct dst_entry *dst = &xdst->u.dst;
1407
1408         dst_free(dst);
1409 }
1410
1411 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1412         .get = xfrm_bundle_flo_get,
1413         .check = xfrm_bundle_flo_check,
1414         .delete = xfrm_bundle_flo_delete,
1415 };
1416
1417 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1418 {
1419         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1420         struct dst_ops *dst_ops;
1421         struct xfrm_dst *xdst;
1422
1423         if (!afinfo)
1424                 return ERR_PTR(-EINVAL);
1425
1426         switch (family) {
1427         case AF_INET:
1428                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1429                 break;
1430 #if IS_ENABLED(CONFIG_IPV6)
1431         case AF_INET6:
1432                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1433                 break;
1434 #endif
1435         default:
1436                 BUG();
1437         }
1438         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1439
1440         if (likely(xdst)) {
1441                 struct dst_entry *dst = &xdst->u.dst;
1442
1443                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1444                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1445                 if (afinfo->init_dst)
1446                         afinfo->init_dst(net, xdst);
1447         } else
1448                 xdst = ERR_PTR(-ENOBUFS);
1449
1450         xfrm_policy_put_afinfo(afinfo);
1451
1452         return xdst;
1453 }
1454
1455 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1456                                  int nfheader_len)
1457 {
1458         struct xfrm_policy_afinfo *afinfo =
1459                 xfrm_policy_get_afinfo(dst->ops->family);
1460         int err;
1461
1462         if (!afinfo)
1463                 return -EINVAL;
1464
1465         err = afinfo->init_path(path, dst, nfheader_len);
1466
1467         xfrm_policy_put_afinfo(afinfo);
1468
1469         return err;
1470 }
1471
1472 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1473                                 const struct flowi *fl)
1474 {
1475         struct xfrm_policy_afinfo *afinfo =
1476                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1477         int err;
1478
1479         if (!afinfo)
1480                 return -EINVAL;
1481
1482         err = afinfo->fill_dst(xdst, dev, fl);
1483
1484         xfrm_policy_put_afinfo(afinfo);
1485
1486         return err;
1487 }
1488
1489
1490 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1491  * all the metrics... Shortly, bundle a bundle.
1492  */
1493
1494 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1495                                             struct xfrm_state **xfrm, int nx,
1496                                             const struct flowi *fl,
1497                                             struct dst_entry *dst)
1498 {
1499         struct net *net = xp_net(policy);
1500         unsigned long now = jiffies;
1501         struct net_device *dev;
1502         struct xfrm_mode *inner_mode;
1503         struct dst_entry *dst_prev = NULL;
1504         struct dst_entry *dst0 = NULL;
1505         int i = 0;
1506         int err;
1507         int header_len = 0;
1508         int nfheader_len = 0;
1509         int trailer_len = 0;
1510         int tos;
1511         int family = policy->selector.family;
1512         xfrm_address_t saddr, daddr;
1513
1514         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1515
1516         tos = xfrm_get_tos(fl, family);
1517         err = tos;
1518         if (tos < 0)
1519                 goto put_states;
1520
1521         dst_hold(dst);
1522
1523         for (; i < nx; i++) {
1524                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1525                 struct dst_entry *dst1 = &xdst->u.dst;
1526
1527                 err = PTR_ERR(xdst);
1528                 if (IS_ERR(xdst)) {
1529                         dst_release(dst);
1530                         goto put_states;
1531                 }
1532
1533                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1534                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1535                                                         xfrm_af2proto(family));
1536                         if (!inner_mode) {
1537                                 err = -EAFNOSUPPORT;
1538                                 dst_release(dst);
1539                                 goto put_states;
1540                         }
1541                 } else
1542                         inner_mode = xfrm[i]->inner_mode;
1543
1544                 if (!dst_prev)
1545                         dst0 = dst1;
1546                 else {
1547                         dst_prev->child = dst_clone(dst1);
1548                         dst1->flags |= DST_NOHASH;
1549                 }
1550
1551                 xdst->route = dst;
1552                 dst_copy_metrics(dst1, dst);
1553
1554                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1555                         family = xfrm[i]->props.family;
1556                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1557                                               family);
1558                         err = PTR_ERR(dst);
1559                         if (IS_ERR(dst))
1560                                 goto put_states;
1561                 } else
1562                         dst_hold(dst);
1563
1564                 dst1->xfrm = xfrm[i];
1565                 xdst->xfrm_genid = xfrm[i]->genid;
1566
1567                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1568                 dst1->flags |= DST_HOST;
1569                 dst1->lastuse = now;
1570
1571                 dst1->input = dst_discard;
1572                 dst1->output = inner_mode->afinfo->output;
1573
1574                 dst1->next = dst_prev;
1575                 dst_prev = dst1;
1576
1577                 header_len += xfrm[i]->props.header_len;
1578                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1579                         nfheader_len += xfrm[i]->props.header_len;
1580                 trailer_len += xfrm[i]->props.trailer_len;
1581         }
1582
1583         dst_prev->child = dst;
1584         dst0->path = dst;
1585
1586         err = -ENODEV;
1587         dev = dst->dev;
1588         if (!dev)
1589                 goto free_dst;
1590
1591         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1592         xfrm_init_pmtu(dst_prev);
1593
1594         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1595                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1596
1597                 err = xfrm_fill_dst(xdst, dev, fl);
1598                 if (err)
1599                         goto free_dst;
1600
1601                 dst_prev->header_len = header_len;
1602                 dst_prev->trailer_len = trailer_len;
1603                 header_len -= xdst->u.dst.xfrm->props.header_len;
1604                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1605         }
1606
1607 out:
1608         return dst0;
1609
1610 put_states:
1611         for (; i < nx; i++)
1612                 xfrm_state_put(xfrm[i]);
1613 free_dst:
1614         if (dst0)
1615                 dst_free(dst0);
1616         dst0 = ERR_PTR(err);
1617         goto out;
1618 }
1619
1620 static int inline
1621 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1622 {
1623         if (!*target) {
1624                 *target = kmalloc(size, GFP_ATOMIC);
1625                 if (!*target)
1626                         return -ENOMEM;
1627         }
1628         memcpy(*target, src, size);
1629         return 0;
1630 }
1631
1632 static int inline
1633 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1634 {
1635 #ifdef CONFIG_XFRM_SUB_POLICY
1636         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1637         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1638                                    sel, sizeof(*sel));
1639 #else
1640         return 0;
1641 #endif
1642 }
1643
1644 static int inline
1645 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1646 {
1647 #ifdef CONFIG_XFRM_SUB_POLICY
1648         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1649         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1650 #else
1651         return 0;
1652 #endif
1653 }
1654
1655 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1656                                 struct xfrm_policy **pols,
1657                                 int *num_pols, int *num_xfrms)
1658 {
1659         int i;
1660
1661         if (*num_pols == 0 || !pols[0]) {
1662                 *num_pols = 0;
1663                 *num_xfrms = 0;
1664                 return 0;
1665         }
1666         if (IS_ERR(pols[0]))
1667                 return PTR_ERR(pols[0]);
1668
1669         *num_xfrms = pols[0]->xfrm_nr;
1670
1671 #ifdef CONFIG_XFRM_SUB_POLICY
1672         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1673             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1674                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1675                                                     XFRM_POLICY_TYPE_MAIN,
1676                                                     fl, family,
1677                                                     XFRM_POLICY_OUT);
1678                 if (pols[1]) {
1679                         if (IS_ERR(pols[1])) {
1680                                 xfrm_pols_put(pols, *num_pols);
1681                                 return PTR_ERR(pols[1]);
1682                         }
1683                         (*num_pols) ++;
1684                         (*num_xfrms) += pols[1]->xfrm_nr;
1685                 }
1686         }
1687 #endif
1688         for (i = 0; i < *num_pols; i++) {
1689                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1690                         *num_xfrms = -1;
1691                         break;
1692                 }
1693         }
1694
1695         return 0;
1696
1697 }
1698
1699 static struct xfrm_dst *
1700 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1701                                const struct flowi *fl, u16 family,
1702                                struct dst_entry *dst_orig)
1703 {
1704         struct net *net = xp_net(pols[0]);
1705         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1706         struct dst_entry *dst;
1707         struct xfrm_dst *xdst;
1708         int err;
1709
1710         /* Try to instantiate a bundle */
1711         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1712         if (err <= 0) {
1713                 if (err != 0 && err != -EAGAIN)
1714                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1715                 return ERR_PTR(err);
1716         }
1717
1718         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1719         if (IS_ERR(dst)) {
1720                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1721                 return ERR_CAST(dst);
1722         }
1723
1724         xdst = (struct xfrm_dst *)dst;
1725         xdst->num_xfrms = err;
1726         if (num_pols > 1)
1727                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1728         else
1729                 err = xfrm_dst_update_origin(dst, fl);
1730         if (unlikely(err)) {
1731                 dst_free(dst);
1732                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1733                 return ERR_PTR(err);
1734         }
1735
1736         xdst->num_pols = num_pols;
1737         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1738         xdst->policy_genid = atomic_read(&pols[0]->genid);
1739
1740         return xdst;
1741 }
1742
1743 static void xfrm_policy_queue_process(unsigned long arg)
1744 {
1745         int err = 0;
1746         struct sk_buff *skb;
1747         struct sock *sk;
1748         struct dst_entry *dst;
1749         struct net_device *dev;
1750         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1751         struct xfrm_policy_queue *pq = &pol->polq;
1752         struct flowi fl;
1753         struct sk_buff_head list;
1754
1755         spin_lock(&pq->hold_queue.lock);
1756         skb = skb_peek(&pq->hold_queue);
1757         dst = skb_dst(skb);
1758         sk = skb->sk;
1759         xfrm_decode_session(skb, &fl, dst->ops->family);
1760         spin_unlock(&pq->hold_queue.lock);
1761
1762         dst_hold(dst->path);
1763         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1764                           sk, 0);
1765         if (IS_ERR(dst))
1766                 goto purge_queue;
1767
1768         if (dst->flags & DST_XFRM_QUEUE) {
1769                 dst_release(dst);
1770
1771                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1772                         goto purge_queue;
1773
1774                 pq->timeout = pq->timeout << 1;
1775                 mod_timer(&pq->hold_timer, jiffies + pq->timeout);
1776                 return;
1777         }
1778
1779         dst_release(dst);
1780
1781         __skb_queue_head_init(&list);
1782
1783         spin_lock(&pq->hold_queue.lock);
1784         pq->timeout = 0;
1785         skb_queue_splice_init(&pq->hold_queue, &list);
1786         spin_unlock(&pq->hold_queue.lock);
1787
1788         while (!skb_queue_empty(&list)) {
1789                 skb = __skb_dequeue(&list);
1790
1791                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1792                 dst_hold(skb_dst(skb)->path);
1793                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1794                                   &fl, skb->sk, 0);
1795                 if (IS_ERR(dst)) {
1796                         dev_put(skb->dev);
1797                         kfree_skb(skb);
1798                         continue;
1799                 }
1800
1801                 nf_reset(skb);
1802                 skb_dst_drop(skb);
1803                 skb_dst_set(skb, dst);
1804
1805                 dev = skb->dev;
1806                 err = dst_output(skb);
1807                 dev_put(dev);
1808         }
1809
1810         return;
1811
1812 purge_queue:
1813         pq->timeout = 0;
1814         xfrm_queue_purge(&pq->hold_queue);
1815 }
1816
1817 static int xdst_queue_output(struct sk_buff *skb)
1818 {
1819         unsigned long sched_next;
1820         struct dst_entry *dst = skb_dst(skb);
1821         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1822         struct xfrm_policy_queue *pq = &xdst->pols[0]->polq;
1823
1824         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1825                 kfree_skb(skb);
1826                 return -EAGAIN;
1827         }
1828
1829         skb_dst_force(skb);
1830         dev_hold(skb->dev);
1831
1832         spin_lock_bh(&pq->hold_queue.lock);
1833
1834         if (!pq->timeout)
1835                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1836
1837         sched_next = jiffies + pq->timeout;
1838
1839         if (del_timer(&pq->hold_timer)) {
1840                 if (time_before(pq->hold_timer.expires, sched_next))
1841                         sched_next = pq->hold_timer.expires;
1842         }
1843
1844         __skb_queue_tail(&pq->hold_queue, skb);
1845         mod_timer(&pq->hold_timer, sched_next);
1846
1847         spin_unlock_bh(&pq->hold_queue.lock);
1848
1849         return 0;
1850 }
1851
1852 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1853                                                  struct dst_entry *dst,
1854                                                  const struct flowi *fl,
1855                                                  int num_xfrms,
1856                                                  u16 family)
1857 {
1858         int err;
1859         struct net_device *dev;
1860         struct dst_entry *dst1;
1861         struct xfrm_dst *xdst;
1862
1863         xdst = xfrm_alloc_dst(net, family);
1864         if (IS_ERR(xdst))
1865                 return xdst;
1866
1867         if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0 ||
1868             (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP))
1869                 return xdst;
1870
1871         dst1 = &xdst->u.dst;
1872         dst_hold(dst);
1873         xdst->route = dst;
1874
1875         dst_copy_metrics(dst1, dst);
1876
1877         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1878         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1879         dst1->lastuse = jiffies;
1880
1881         dst1->input = dst_discard;
1882         dst1->output = xdst_queue_output;
1883
1884         dst_hold(dst);
1885         dst1->child = dst;
1886         dst1->path = dst;
1887
1888         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1889
1890         err = -ENODEV;
1891         dev = dst->dev;
1892         if (!dev)
1893                 goto free_dst;
1894
1895         err = xfrm_fill_dst(xdst, dev, fl);
1896         if (err)
1897                 goto free_dst;
1898
1899 out:
1900         return xdst;
1901
1902 free_dst:
1903         dst_release(dst1);
1904         xdst = ERR_PTR(err);
1905         goto out;
1906 }
1907
1908 static struct flow_cache_object *
1909 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1910                    struct flow_cache_object *oldflo, void *ctx)
1911 {
1912         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1913         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1914         struct xfrm_dst *xdst, *new_xdst;
1915         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1916
1917         /* Check if the policies from old bundle are usable */
1918         xdst = NULL;
1919         if (oldflo) {
1920                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1921                 num_pols = xdst->num_pols;
1922                 num_xfrms = xdst->num_xfrms;
1923                 pol_dead = 0;
1924                 for (i = 0; i < num_pols; i++) {
1925                         pols[i] = xdst->pols[i];
1926                         pol_dead |= pols[i]->walk.dead;
1927                 }
1928                 if (pol_dead) {
1929                         dst_free(&xdst->u.dst);
1930                         xdst = NULL;
1931                         num_pols = 0;
1932                         num_xfrms = 0;
1933                         oldflo = NULL;
1934                 }
1935         }
1936
1937         /* Resolve policies to use if we couldn't get them from
1938          * previous cache entry */
1939         if (xdst == NULL) {
1940                 num_pols = 1;
1941                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1942                 err = xfrm_expand_policies(fl, family, pols,
1943                                            &num_pols, &num_xfrms);
1944                 if (err < 0)
1945                         goto inc_error;
1946                 if (num_pols == 0)
1947                         return NULL;
1948                 if (num_xfrms <= 0)
1949                         goto make_dummy_bundle;
1950         }
1951
1952         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1953         if (IS_ERR(new_xdst)) {
1954                 err = PTR_ERR(new_xdst);
1955                 if (err != -EAGAIN)
1956                         goto error;
1957                 if (oldflo == NULL)
1958                         goto make_dummy_bundle;
1959                 dst_hold(&xdst->u.dst);
1960                 return oldflo;
1961         } else if (new_xdst == NULL) {
1962                 num_xfrms = 0;
1963                 if (oldflo == NULL)
1964                         goto make_dummy_bundle;
1965                 xdst->num_xfrms = 0;
1966                 dst_hold(&xdst->u.dst);
1967                 return oldflo;
1968         }
1969
1970         /* Kill the previous bundle */
1971         if (xdst) {
1972                 /* The policies were stolen for newly generated bundle */
1973                 xdst->num_pols = 0;
1974                 dst_free(&xdst->u.dst);
1975         }
1976
1977         /* Flow cache does not have reference, it dst_free()'s,
1978          * but we do need to return one reference for original caller */
1979         dst_hold(&new_xdst->u.dst);
1980         return &new_xdst->flo;
1981
1982 make_dummy_bundle:
1983         /* We found policies, but there's no bundles to instantiate:
1984          * either because the policy blocks, has no transformations or
1985          * we could not build template (no xfrm_states).*/
1986         xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
1987         if (IS_ERR(xdst)) {
1988                 xfrm_pols_put(pols, num_pols);
1989                 return ERR_CAST(xdst);
1990         }
1991         xdst->num_pols = num_pols;
1992         xdst->num_xfrms = num_xfrms;
1993         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1994
1995         dst_hold(&xdst->u.dst);
1996         return &xdst->flo;
1997
1998 inc_error:
1999         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2000 error:
2001         if (xdst != NULL)
2002                 dst_free(&xdst->u.dst);
2003         else
2004                 xfrm_pols_put(pols, num_pols);
2005         return ERR_PTR(err);
2006 }
2007
2008 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2009                                         struct dst_entry *dst_orig)
2010 {
2011         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2012         struct dst_entry *ret;
2013
2014         if (!afinfo) {
2015                 dst_release(dst_orig);
2016                 return ERR_PTR(-EINVAL);
2017         } else {
2018                 ret = afinfo->blackhole_route(net, dst_orig);
2019         }
2020         xfrm_policy_put_afinfo(afinfo);
2021
2022         return ret;
2023 }
2024
2025 /* Main function: finds/creates a bundle for given flow.
2026  *
2027  * At the moment we eat a raw IP route. Mostly to speed up lookups
2028  * on interfaces with disabled IPsec.
2029  */
2030 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2031                               const struct flowi *fl,
2032                               struct sock *sk, int flags)
2033 {
2034         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2035         struct flow_cache_object *flo;
2036         struct xfrm_dst *xdst;
2037         struct dst_entry *dst, *route;
2038         u16 family = dst_orig->ops->family;
2039         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2040         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2041
2042 restart:
2043         dst = NULL;
2044         xdst = NULL;
2045         route = NULL;
2046
2047         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2048                 num_pols = 1;
2049                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2050                 err = xfrm_expand_policies(fl, family, pols,
2051                                            &num_pols, &num_xfrms);
2052                 if (err < 0)
2053                         goto dropdst;
2054
2055                 if (num_pols) {
2056                         if (num_xfrms <= 0) {
2057                                 drop_pols = num_pols;
2058                                 goto no_transform;
2059                         }
2060
2061                         xdst = xfrm_resolve_and_create_bundle(
2062                                         pols, num_pols, fl,
2063                                         family, dst_orig);
2064                         if (IS_ERR(xdst)) {
2065                                 xfrm_pols_put(pols, num_pols);
2066                                 err = PTR_ERR(xdst);
2067                                 goto dropdst;
2068                         } else if (xdst == NULL) {
2069                                 num_xfrms = 0;
2070                                 drop_pols = num_pols;
2071                                 goto no_transform;
2072                         }
2073
2074                         dst_hold(&xdst->u.dst);
2075
2076                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2077                         xdst->u.dst.next = xfrm_policy_sk_bundles;
2078                         xfrm_policy_sk_bundles = &xdst->u.dst;
2079                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2080
2081                         route = xdst->route;
2082                 }
2083         }
2084
2085         if (xdst == NULL) {
2086                 /* To accelerate a bit...  */
2087                 if ((dst_orig->flags & DST_NOXFRM) ||
2088                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2089                         goto nopol;
2090
2091                 flo = flow_cache_lookup(net, fl, family, dir,
2092                                         xfrm_bundle_lookup, dst_orig);
2093                 if (flo == NULL)
2094                         goto nopol;
2095                 if (IS_ERR(flo)) {
2096                         err = PTR_ERR(flo);
2097                         goto dropdst;
2098                 }
2099                 xdst = container_of(flo, struct xfrm_dst, flo);
2100
2101                 num_pols = xdst->num_pols;
2102                 num_xfrms = xdst->num_xfrms;
2103                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
2104                 route = xdst->route;
2105         }
2106
2107         dst = &xdst->u.dst;
2108         if (route == NULL && num_xfrms > 0) {
2109                 /* The only case when xfrm_bundle_lookup() returns a
2110                  * bundle with null route, is when the template could
2111                  * not be resolved. It means policies are there, but
2112                  * bundle could not be created, since we don't yet
2113                  * have the xfrm_state's. We need to wait for KM to
2114                  * negotiate new SA's or bail out with error.*/
2115                 if (net->xfrm.sysctl_larval_drop) {
2116                         /* EREMOTE tells the caller to generate
2117                          * a one-shot blackhole route. */
2118                         dst_release(dst);
2119                         xfrm_pols_put(pols, drop_pols);
2120                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2121
2122                         return make_blackhole(net, family, dst_orig);
2123                 }
2124                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
2125                         DECLARE_WAITQUEUE(wait, current);
2126
2127                         add_wait_queue(&net->xfrm.km_waitq, &wait);
2128                         set_current_state(TASK_INTERRUPTIBLE);
2129                         schedule();
2130                         set_current_state(TASK_RUNNING);
2131                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
2132
2133                         if (!signal_pending(current)) {
2134                                 dst_release(dst);
2135                                 goto restart;
2136                         }
2137
2138                         err = -ERESTART;
2139                 } else
2140                         err = -EAGAIN;
2141
2142                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2143                 goto error;
2144         }
2145
2146 no_transform:
2147         if (num_pols == 0)
2148                 goto nopol;
2149
2150         if ((flags & XFRM_LOOKUP_ICMP) &&
2151             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2152                 err = -ENOENT;
2153                 goto error;
2154         }
2155
2156         for (i = 0; i < num_pols; i++)
2157                 pols[i]->curlft.use_time = get_seconds();
2158
2159         if (num_xfrms < 0) {
2160                 /* Prohibit the flow */
2161                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2162                 err = -EPERM;
2163                 goto error;
2164         } else if (num_xfrms > 0) {
2165                 /* Flow transformed */
2166                 dst_release(dst_orig);
2167         } else {
2168                 /* Flow passes untransformed */
2169                 dst_release(dst);
2170                 dst = dst_orig;
2171         }
2172 ok:
2173         xfrm_pols_put(pols, drop_pols);
2174         if (dst && dst->xfrm &&
2175             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2176                 dst->flags |= DST_XFRM_TUNNEL;
2177         return dst;
2178
2179 nopol:
2180         if (!(flags & XFRM_LOOKUP_ICMP)) {
2181                 dst = dst_orig;
2182                 goto ok;
2183         }
2184         err = -ENOENT;
2185 error:
2186         dst_release(dst);
2187 dropdst:
2188         dst_release(dst_orig);
2189         xfrm_pols_put(pols, drop_pols);
2190         return ERR_PTR(err);
2191 }
2192 EXPORT_SYMBOL(xfrm_lookup);
2193
2194 static inline int
2195 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2196 {
2197         struct xfrm_state *x;
2198
2199         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2200                 return 0;
2201         x = skb->sp->xvec[idx];
2202         if (!x->type->reject)
2203                 return 0;
2204         return x->type->reject(x, skb, fl);
2205 }
2206
2207 /* When skb is transformed back to its "native" form, we have to
2208  * check policy restrictions. At the moment we make this in maximally
2209  * stupid way. Shame on me. :-) Of course, connected sockets must
2210  * have policy cached at them.
2211  */
2212
2213 static inline int
2214 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2215               unsigned short family)
2216 {
2217         if (xfrm_state_kern(x))
2218                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2219         return  x->id.proto == tmpl->id.proto &&
2220                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2221                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2222                 x->props.mode == tmpl->mode &&
2223                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2224                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2225                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2226                   xfrm_state_addr_cmp(tmpl, x, family));
2227 }
2228
2229 /*
2230  * 0 or more than 0 is returned when validation is succeeded (either bypass
2231  * because of optional transport mode, or next index of the mathced secpath
2232  * state with the template.
2233  * -1 is returned when no matching template is found.
2234  * Otherwise "-2 - errored_index" is returned.
2235  */
2236 static inline int
2237 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2238                unsigned short family)
2239 {
2240         int idx = start;
2241
2242         if (tmpl->optional) {
2243                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2244                         return start;
2245         } else
2246                 start = -1;
2247         for (; idx < sp->len; idx++) {
2248                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2249                         return ++idx;
2250                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2251                         if (start == -1)
2252                                 start = -2-idx;
2253                         break;
2254                 }
2255         }
2256         return start;
2257 }
2258
2259 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2260                           unsigned int family, int reverse)
2261 {
2262         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2263         int err;
2264
2265         if (unlikely(afinfo == NULL))
2266                 return -EAFNOSUPPORT;
2267
2268         afinfo->decode_session(skb, fl, reverse);
2269         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2270         xfrm_policy_put_afinfo(afinfo);
2271         return err;
2272 }
2273 EXPORT_SYMBOL(__xfrm_decode_session);
2274
2275 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2276 {
2277         for (; k < sp->len; k++) {
2278                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2279                         *idxp = k;
2280                         return 1;
2281                 }
2282         }
2283
2284         return 0;
2285 }
2286
2287 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2288                         unsigned short family)
2289 {
2290         struct net *net = dev_net(skb->dev);
2291         struct xfrm_policy *pol;
2292         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2293         int npols = 0;
2294         int xfrm_nr;
2295         int pi;
2296         int reverse;
2297         struct flowi fl;
2298         u8 fl_dir;
2299         int xerr_idx = -1;
2300
2301         reverse = dir & ~XFRM_POLICY_MASK;
2302         dir &= XFRM_POLICY_MASK;
2303         fl_dir = policy_to_flow_dir(dir);
2304
2305         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2306                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2307                 return 0;
2308         }
2309
2310         nf_nat_decode_session(skb, &fl, family);
2311
2312         /* First, check used SA against their selectors. */
2313         if (skb->sp) {
2314                 int i;
2315
2316                 for (i=skb->sp->len-1; i>=0; i--) {
2317                         struct xfrm_state *x = skb->sp->xvec[i];
2318                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2319                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2320                                 return 0;
2321                         }
2322                 }
2323         }
2324
2325         pol = NULL;
2326         if (sk && sk->sk_policy[dir]) {
2327                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2328                 if (IS_ERR(pol)) {
2329                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2330                         return 0;
2331                 }
2332         }
2333
2334         if (!pol) {
2335                 struct flow_cache_object *flo;
2336
2337                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2338                                         xfrm_policy_lookup, NULL);
2339                 if (IS_ERR_OR_NULL(flo))
2340                         pol = ERR_CAST(flo);
2341                 else
2342                         pol = container_of(flo, struct xfrm_policy, flo);
2343         }
2344
2345         if (IS_ERR(pol)) {
2346                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2347                 return 0;
2348         }
2349
2350         if (!pol) {
2351                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2352                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2353                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2354                         return 0;
2355                 }
2356                 return 1;
2357         }
2358
2359         pol->curlft.use_time = get_seconds();
2360
2361         pols[0] = pol;
2362         npols ++;
2363 #ifdef CONFIG_XFRM_SUB_POLICY
2364         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2365                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2366                                                     &fl, family,
2367                                                     XFRM_POLICY_IN);
2368                 if (pols[1]) {
2369                         if (IS_ERR(pols[1])) {
2370                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2371                                 return 0;
2372                         }
2373                         pols[1]->curlft.use_time = get_seconds();
2374                         npols ++;
2375                 }
2376         }
2377 #endif
2378
2379         if (pol->action == XFRM_POLICY_ALLOW) {
2380                 struct sec_path *sp;
2381                 static struct sec_path dummy;
2382                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2383                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2384                 struct xfrm_tmpl **tpp = tp;
2385                 int ti = 0;
2386                 int i, k;
2387
2388                 if ((sp = skb->sp) == NULL)
2389                         sp = &dummy;
2390
2391                 for (pi = 0; pi < npols; pi++) {
2392                         if (pols[pi] != pol &&
2393                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2394                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2395                                 goto reject;
2396                         }
2397                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2398                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2399                                 goto reject_error;
2400                         }
2401                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2402                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2403                 }
2404                 xfrm_nr = ti;
2405                 if (npols > 1) {
2406                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2407                         tpp = stp;
2408                 }
2409
2410                 /* For each tunnel xfrm, find the first matching tmpl.
2411                  * For each tmpl before that, find corresponding xfrm.
2412                  * Order is _important_. Later we will implement
2413                  * some barriers, but at the moment barriers
2414                  * are implied between each two transformations.
2415                  */
2416                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2417                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2418                         if (k < 0) {
2419                                 if (k < -1)
2420                                         /* "-2 - errored_index" returned */
2421                                         xerr_idx = -(2+k);
2422                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2423                                 goto reject;
2424                         }
2425                 }
2426
2427                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2428                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2429                         goto reject;
2430                 }
2431
2432                 xfrm_pols_put(pols, npols);
2433                 return 1;
2434         }
2435         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2436
2437 reject:
2438         xfrm_secpath_reject(xerr_idx, skb, &fl);
2439 reject_error:
2440         xfrm_pols_put(pols, npols);
2441         return 0;
2442 }
2443 EXPORT_SYMBOL(__xfrm_policy_check);
2444
2445 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2446 {
2447         struct net *net = dev_net(skb->dev);
2448         struct flowi fl;
2449         struct dst_entry *dst;
2450         int res = 1;
2451
2452         if (xfrm_decode_session(skb, &fl, family) < 0) {
2453                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2454                 return 0;
2455         }
2456
2457         skb_dst_force(skb);
2458
2459         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2460         if (IS_ERR(dst)) {
2461                 res = 0;
2462                 dst = NULL;
2463         }
2464         skb_dst_set(skb, dst);
2465         return res;
2466 }
2467 EXPORT_SYMBOL(__xfrm_route_forward);
2468
2469 /* Optimize later using cookies and generation ids. */
2470
2471 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2472 {
2473         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2474          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2475          * get validated by dst_ops->check on every use.  We do this
2476          * because when a normal route referenced by an XFRM dst is
2477          * obsoleted we do not go looking around for all parent
2478          * referencing XFRM dsts so that we can invalidate them.  It
2479          * is just too much work.  Instead we make the checks here on
2480          * every use.  For example:
2481          *
2482          *      XFRM dst A --> IPv4 dst X
2483          *
2484          * X is the "xdst->route" of A (X is also the "dst->path" of A
2485          * in this example).  If X is marked obsolete, "A" will not
2486          * notice.  That's what we are validating here via the
2487          * stale_bundle() check.
2488          *
2489          * When a policy's bundle is pruned, we dst_free() the XFRM
2490          * dst which causes it's ->obsolete field to be set to
2491          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2492          * this, we want to force a new route lookup.
2493          */
2494         if (dst->obsolete < 0 && !stale_bundle(dst))
2495                 return dst;
2496
2497         return NULL;
2498 }
2499
2500 static int stale_bundle(struct dst_entry *dst)
2501 {
2502         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2503 }
2504
2505 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2506 {
2507         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2508                 dst->dev = dev_net(dev)->loopback_dev;
2509                 dev_hold(dst->dev);
2510                 dev_put(dev);
2511         }
2512 }
2513 EXPORT_SYMBOL(xfrm_dst_ifdown);
2514
2515 static void xfrm_link_failure(struct sk_buff *skb)
2516 {
2517         /* Impossible. Such dst must be popped before reaches point of failure. */
2518 }
2519
2520 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2521 {
2522         if (dst) {
2523                 if (dst->obsolete) {
2524                         dst_release(dst);
2525                         dst = NULL;
2526                 }
2527         }
2528         return dst;
2529 }
2530
2531 static void __xfrm_garbage_collect(struct net *net)
2532 {
2533         struct dst_entry *head, *next;
2534
2535         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2536         head = xfrm_policy_sk_bundles;
2537         xfrm_policy_sk_bundles = NULL;
2538         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2539
2540         while (head) {
2541                 next = head->next;
2542                 dst_free(head);
2543                 head = next;
2544         }
2545 }
2546
2547 static void xfrm_garbage_collect(struct net *net)
2548 {
2549         flow_cache_flush();
2550         __xfrm_garbage_collect(net);
2551 }
2552
2553 static void xfrm_garbage_collect_deferred(struct net *net)
2554 {
2555         flow_cache_flush_deferred();
2556         __xfrm_garbage_collect(net);
2557 }
2558
2559 static void xfrm_init_pmtu(struct dst_entry *dst)
2560 {
2561         do {
2562                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2563                 u32 pmtu, route_mtu_cached;
2564
2565                 pmtu = dst_mtu(dst->child);
2566                 xdst->child_mtu_cached = pmtu;
2567
2568                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2569
2570                 route_mtu_cached = dst_mtu(xdst->route);
2571                 xdst->route_mtu_cached = route_mtu_cached;
2572
2573                 if (pmtu > route_mtu_cached)
2574                         pmtu = route_mtu_cached;
2575
2576                 dst_metric_set(dst, RTAX_MTU, pmtu);
2577         } while ((dst = dst->next));
2578 }
2579
2580 /* Check that the bundle accepts the flow and its components are
2581  * still valid.
2582  */
2583
2584 static int xfrm_bundle_ok(struct xfrm_dst *first)
2585 {
2586         struct dst_entry *dst = &first->u.dst;
2587         struct xfrm_dst *last;
2588         u32 mtu;
2589
2590         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2591             (dst->dev && !netif_running(dst->dev)))
2592                 return 0;
2593
2594         if (dst->flags & DST_XFRM_QUEUE)
2595                 return 1;
2596
2597         last = NULL;
2598
2599         do {
2600                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2601
2602                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2603                         return 0;
2604                 if (xdst->xfrm_genid != dst->xfrm->genid)
2605                         return 0;
2606                 if (xdst->num_pols > 0 &&
2607                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2608                         return 0;
2609
2610                 mtu = dst_mtu(dst->child);
2611                 if (xdst->child_mtu_cached != mtu) {
2612                         last = xdst;
2613                         xdst->child_mtu_cached = mtu;
2614                 }
2615
2616                 if (!dst_check(xdst->route, xdst->route_cookie))
2617                         return 0;
2618                 mtu = dst_mtu(xdst->route);
2619                 if (xdst->route_mtu_cached != mtu) {
2620                         last = xdst;
2621                         xdst->route_mtu_cached = mtu;
2622                 }
2623
2624                 dst = dst->child;
2625         } while (dst->xfrm);
2626
2627         if (likely(!last))
2628                 return 1;
2629
2630         mtu = last->child_mtu_cached;
2631         for (;;) {
2632                 dst = &last->u.dst;
2633
2634                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2635                 if (mtu > last->route_mtu_cached)
2636                         mtu = last->route_mtu_cached;
2637                 dst_metric_set(dst, RTAX_MTU, mtu);
2638
2639                 if (last == first)
2640                         break;
2641
2642                 last = (struct xfrm_dst *)last->u.dst.next;
2643                 last->child_mtu_cached = mtu;
2644         }
2645
2646         return 1;
2647 }
2648
2649 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2650 {
2651         return dst_metric_advmss(dst->path);
2652 }
2653
2654 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2655 {
2656         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2657
2658         return mtu ? : dst_mtu(dst->path);
2659 }
2660
2661 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2662                                            struct sk_buff *skb,
2663                                            const void *daddr)
2664 {
2665         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2666 }
2667
2668 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2669 {
2670         struct net *net;
2671         int err = 0;
2672         if (unlikely(afinfo == NULL))
2673                 return -EINVAL;
2674         if (unlikely(afinfo->family >= NPROTO))
2675                 return -EAFNOSUPPORT;
2676         spin_lock(&xfrm_policy_afinfo_lock);
2677         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2678                 err = -ENOBUFS;
2679         else {
2680                 struct dst_ops *dst_ops = afinfo->dst_ops;
2681                 if (likely(dst_ops->kmem_cachep == NULL))
2682                         dst_ops->kmem_cachep = xfrm_dst_cache;
2683                 if (likely(dst_ops->check == NULL))
2684                         dst_ops->check = xfrm_dst_check;
2685                 if (likely(dst_ops->default_advmss == NULL))
2686                         dst_ops->default_advmss = xfrm_default_advmss;
2687                 if (likely(dst_ops->mtu == NULL))
2688                         dst_ops->mtu = xfrm_mtu;
2689                 if (likely(dst_ops->negative_advice == NULL))
2690                         dst_ops->negative_advice = xfrm_negative_advice;
2691                 if (likely(dst_ops->link_failure == NULL))
2692                         dst_ops->link_failure = xfrm_link_failure;
2693                 if (likely(dst_ops->neigh_lookup == NULL))
2694                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2695                 if (likely(afinfo->garbage_collect == NULL))
2696                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2697                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2698         }
2699         spin_unlock(&xfrm_policy_afinfo_lock);
2700
2701         rtnl_lock();
2702         for_each_net(net) {
2703                 struct dst_ops *xfrm_dst_ops;
2704
2705                 switch (afinfo->family) {
2706                 case AF_INET:
2707                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2708                         break;
2709 #if IS_ENABLED(CONFIG_IPV6)
2710                 case AF_INET6:
2711                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2712                         break;
2713 #endif
2714                 default:
2715                         BUG();
2716                 }
2717                 *xfrm_dst_ops = *afinfo->dst_ops;
2718         }
2719         rtnl_unlock();
2720
2721         return err;
2722 }
2723 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2724
2725 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2726 {
2727         int err = 0;
2728         if (unlikely(afinfo == NULL))
2729                 return -EINVAL;
2730         if (unlikely(afinfo->family >= NPROTO))
2731                 return -EAFNOSUPPORT;
2732         spin_lock(&xfrm_policy_afinfo_lock);
2733         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2734                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2735                         err = -EINVAL;
2736                 else
2737                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2738                                          NULL);
2739         }
2740         spin_unlock(&xfrm_policy_afinfo_lock);
2741         if (!err) {
2742                 struct dst_ops *dst_ops = afinfo->dst_ops;
2743
2744                 synchronize_rcu();
2745
2746                 dst_ops->kmem_cachep = NULL;
2747                 dst_ops->check = NULL;
2748                 dst_ops->negative_advice = NULL;
2749                 dst_ops->link_failure = NULL;
2750                 afinfo->garbage_collect = NULL;
2751         }
2752         return err;
2753 }
2754 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2755
2756 static void __net_init xfrm_dst_ops_init(struct net *net)
2757 {
2758         struct xfrm_policy_afinfo *afinfo;
2759
2760         rcu_read_lock();
2761         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2762         if (afinfo)
2763                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2764 #if IS_ENABLED(CONFIG_IPV6)
2765         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2766         if (afinfo)
2767                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2768 #endif
2769         rcu_read_unlock();
2770 }
2771
2772 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2773 {
2774         struct net_device *dev = ptr;
2775
2776         switch (event) {
2777         case NETDEV_DOWN:
2778                 xfrm_garbage_collect(dev_net(dev));
2779         }
2780         return NOTIFY_DONE;
2781 }
2782
2783 static struct notifier_block xfrm_dev_notifier = {
2784         .notifier_call  = xfrm_dev_event,
2785 };
2786
2787 #ifdef CONFIG_XFRM_STATISTICS
2788 static int __net_init xfrm_statistics_init(struct net *net)
2789 {
2790         int rv;
2791
2792         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2793                           sizeof(struct linux_xfrm_mib),
2794                           __alignof__(struct linux_xfrm_mib)) < 0)
2795                 return -ENOMEM;
2796         rv = xfrm_proc_init(net);
2797         if (rv < 0)
2798                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2799         return rv;
2800 }
2801
2802 static void xfrm_statistics_fini(struct net *net)
2803 {
2804         xfrm_proc_fini(net);
2805         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2806 }
2807 #else
2808 static int __net_init xfrm_statistics_init(struct net *net)
2809 {
2810         return 0;
2811 }
2812
2813 static void xfrm_statistics_fini(struct net *net)
2814 {
2815 }
2816 #endif
2817
2818 static int __net_init xfrm_policy_init(struct net *net)
2819 {
2820         unsigned int hmask, sz;
2821         int dir;
2822
2823         if (net_eq(net, &init_net))
2824                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2825                                            sizeof(struct xfrm_dst),
2826                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2827                                            NULL);
2828
2829         hmask = 8 - 1;
2830         sz = (hmask+1) * sizeof(struct hlist_head);
2831
2832         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2833         if (!net->xfrm.policy_byidx)
2834                 goto out_byidx;
2835         net->xfrm.policy_idx_hmask = hmask;
2836
2837         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2838                 struct xfrm_policy_hash *htab;
2839
2840                 net->xfrm.policy_count[dir] = 0;
2841                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2842
2843                 htab = &net->xfrm.policy_bydst[dir];
2844                 htab->table = xfrm_hash_alloc(sz);
2845                 if (!htab->table)
2846                         goto out_bydst;
2847                 htab->hmask = hmask;
2848         }
2849
2850         INIT_LIST_HEAD(&net->xfrm.policy_all);
2851         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2852         if (net_eq(net, &init_net))
2853                 register_netdevice_notifier(&xfrm_dev_notifier);
2854         return 0;
2855
2856 out_bydst:
2857         for (dir--; dir >= 0; dir--) {
2858                 struct xfrm_policy_hash *htab;
2859
2860                 htab = &net->xfrm.policy_bydst[dir];
2861                 xfrm_hash_free(htab->table, sz);
2862         }
2863         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2864 out_byidx:
2865         return -ENOMEM;
2866 }
2867
2868 static void xfrm_policy_fini(struct net *net)
2869 {
2870         struct xfrm_audit audit_info;
2871         unsigned int sz;
2872         int dir;
2873
2874         flush_work(&net->xfrm.policy_hash_work);
2875 #ifdef CONFIG_XFRM_SUB_POLICY
2876         audit_info.loginuid = INVALID_UID;
2877         audit_info.sessionid = -1;
2878         audit_info.secid = 0;
2879         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2880 #endif
2881         audit_info.loginuid = INVALID_UID;
2882         audit_info.sessionid = -1;
2883         audit_info.secid = 0;
2884         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2885
2886         WARN_ON(!list_empty(&net->xfrm.policy_all));
2887
2888         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2889                 struct xfrm_policy_hash *htab;
2890
2891                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2892
2893                 htab = &net->xfrm.policy_bydst[dir];
2894                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2895                 WARN_ON(!hlist_empty(htab->table));
2896                 xfrm_hash_free(htab->table, sz);
2897         }
2898
2899         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2900         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2901         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2902 }
2903
2904 static int __net_init xfrm_net_init(struct net *net)
2905 {
2906         int rv;
2907
2908         rv = xfrm_statistics_init(net);
2909         if (rv < 0)
2910                 goto out_statistics;
2911         rv = xfrm_state_init(net);
2912         if (rv < 0)
2913                 goto out_state;
2914         rv = xfrm_policy_init(net);
2915         if (rv < 0)
2916                 goto out_policy;
2917         xfrm_dst_ops_init(net);
2918         rv = xfrm_sysctl_init(net);
2919         if (rv < 0)
2920                 goto out_sysctl;
2921         return 0;
2922
2923 out_sysctl:
2924         xfrm_policy_fini(net);
2925 out_policy:
2926         xfrm_state_fini(net);
2927 out_state:
2928         xfrm_statistics_fini(net);
2929 out_statistics:
2930         return rv;
2931 }
2932
2933 static void __net_exit xfrm_net_exit(struct net *net)
2934 {
2935         xfrm_sysctl_fini(net);
2936         xfrm_policy_fini(net);
2937         xfrm_state_fini(net);
2938         xfrm_statistics_fini(net);
2939 }
2940
2941 static struct pernet_operations __net_initdata xfrm_net_ops = {
2942         .init = xfrm_net_init,
2943         .exit = xfrm_net_exit,
2944 };
2945
2946 void __init xfrm_init(void)
2947 {
2948         register_pernet_subsys(&xfrm_net_ops);
2949         xfrm_input_init();
2950 }
2951
2952 #ifdef CONFIG_AUDITSYSCALL
2953 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2954                                          struct audit_buffer *audit_buf)
2955 {
2956         struct xfrm_sec_ctx *ctx = xp->security;
2957         struct xfrm_selector *sel = &xp->selector;
2958
2959         if (ctx)
2960                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2961                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2962
2963         switch(sel->family) {
2964         case AF_INET:
2965                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2966                 if (sel->prefixlen_s != 32)
2967                         audit_log_format(audit_buf, " src_prefixlen=%d",
2968                                          sel->prefixlen_s);
2969                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2970                 if (sel->prefixlen_d != 32)
2971                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2972                                          sel->prefixlen_d);
2973                 break;
2974         case AF_INET6:
2975                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2976                 if (sel->prefixlen_s != 128)
2977                         audit_log_format(audit_buf, " src_prefixlen=%d",
2978                                          sel->prefixlen_s);
2979                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2980                 if (sel->prefixlen_d != 128)
2981                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2982                                          sel->prefixlen_d);
2983                 break;
2984         }
2985 }
2986
2987 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2988                            kuid_t auid, u32 sessionid, u32 secid)
2989 {
2990         struct audit_buffer *audit_buf;
2991
2992         audit_buf = xfrm_audit_start("SPD-add");
2993         if (audit_buf == NULL)
2994                 return;
2995         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2996         audit_log_format(audit_buf, " res=%u", result);
2997         xfrm_audit_common_policyinfo(xp, audit_buf);
2998         audit_log_end(audit_buf);
2999 }
3000 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3001
3002 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3003                               kuid_t auid, u32 sessionid, u32 secid)
3004 {
3005         struct audit_buffer *audit_buf;
3006
3007         audit_buf = xfrm_audit_start("SPD-delete");
3008         if (audit_buf == NULL)
3009                 return;
3010         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3011         audit_log_format(audit_buf, " res=%u", result);
3012         xfrm_audit_common_policyinfo(xp, audit_buf);
3013         audit_log_end(audit_buf);
3014 }
3015 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3016 #endif
3017
3018 #ifdef CONFIG_XFRM_MIGRATE
3019 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3020                                         const struct xfrm_selector *sel_tgt)
3021 {
3022         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3023                 if (sel_tgt->family == sel_cmp->family &&
3024                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3025                                     sel_cmp->family) &&
3026                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3027                                     sel_cmp->family) &&
3028                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3029                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3030                         return true;
3031                 }
3032         } else {
3033                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3034                         return true;
3035                 }
3036         }
3037         return false;
3038 }
3039
3040 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3041                                                      u8 dir, u8 type)
3042 {
3043         struct xfrm_policy *pol, *ret = NULL;
3044         struct hlist_node *entry;
3045         struct hlist_head *chain;
3046         u32 priority = ~0U;
3047
3048         read_lock_bh(&xfrm_policy_lock);
3049         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
3050         hlist_for_each_entry(pol, entry, chain, bydst) {
3051                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3052                     pol->type == type) {
3053                         ret = pol;
3054                         priority = ret->priority;
3055                         break;
3056                 }
3057         }
3058         chain = &init_net.xfrm.policy_inexact[dir];
3059         hlist_for_each_entry(pol, entry, chain, bydst) {
3060                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3061                     pol->type == type &&
3062                     pol->priority < priority) {
3063                         ret = pol;
3064                         break;
3065                 }
3066         }
3067
3068         if (ret)
3069                 xfrm_pol_hold(ret);
3070
3071         read_unlock_bh(&xfrm_policy_lock);
3072
3073         return ret;
3074 }
3075
3076 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3077 {
3078         int match = 0;
3079
3080         if (t->mode == m->mode && t->id.proto == m->proto &&
3081             (m->reqid == 0 || t->reqid == m->reqid)) {
3082                 switch (t->mode) {
3083                 case XFRM_MODE_TUNNEL:
3084                 case XFRM_MODE_BEET:
3085                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3086                                             m->old_family) &&
3087                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3088                                             m->old_family)) {
3089                                 match = 1;
3090                         }
3091                         break;
3092                 case XFRM_MODE_TRANSPORT:
3093                         /* in case of transport mode, template does not store
3094                            any IP addresses, hence we just compare mode and
3095                            protocol */
3096                         match = 1;
3097                         break;
3098                 default:
3099                         break;
3100                 }
3101         }
3102         return match;
3103 }
3104
3105 /* update endpoint address(es) of template(s) */
3106 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3107                                struct xfrm_migrate *m, int num_migrate)
3108 {
3109         struct xfrm_migrate *mp;
3110         int i, j, n = 0;
3111
3112         write_lock_bh(&pol->lock);
3113         if (unlikely(pol->walk.dead)) {
3114                 /* target policy has been deleted */
3115                 write_unlock_bh(&pol->lock);
3116                 return -ENOENT;
3117         }
3118
3119         for (i = 0; i < pol->xfrm_nr; i++) {
3120                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3121                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3122                                 continue;
3123                         n++;
3124                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3125                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3126                                 continue;
3127                         /* update endpoints */
3128                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3129                                sizeof(pol->xfrm_vec[i].id.daddr));
3130                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3131                                sizeof(pol->xfrm_vec[i].saddr));
3132                         pol->xfrm_vec[i].encap_family = mp->new_family;
3133                         /* flush bundles */
3134                         atomic_inc(&pol->genid);
3135                 }
3136         }
3137
3138         write_unlock_bh(&pol->lock);
3139
3140         if (!n)
3141                 return -ENODATA;
3142
3143         return 0;
3144 }
3145
3146 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3147 {
3148         int i, j;
3149
3150         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3151                 return -EINVAL;
3152
3153         for (i = 0; i < num_migrate; i++) {
3154                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3155                                     m[i].old_family) &&
3156                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3157                                     m[i].old_family))
3158                         return -EINVAL;
3159                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3160                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3161                         return -EINVAL;
3162
3163                 /* check if there is any duplicated entry */
3164                 for (j = i + 1; j < num_migrate; j++) {
3165                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3166                                     sizeof(m[i].old_daddr)) &&
3167                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3168                                     sizeof(m[i].old_saddr)) &&
3169                             m[i].proto == m[j].proto &&
3170                             m[i].mode == m[j].mode &&
3171                             m[i].reqid == m[j].reqid &&
3172                             m[i].old_family == m[j].old_family)
3173                                 return -EINVAL;
3174                 }
3175         }
3176
3177         return 0;
3178 }
3179
3180 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3181                  struct xfrm_migrate *m, int num_migrate,
3182                  struct xfrm_kmaddress *k)
3183 {
3184         int i, err, nx_cur = 0, nx_new = 0;
3185         struct xfrm_policy *pol = NULL;
3186         struct xfrm_state *x, *xc;
3187         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3188         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3189         struct xfrm_migrate *mp;
3190
3191         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3192                 goto out;
3193
3194         /* Stage 1 - find policy */
3195         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
3196                 err = -ENOENT;
3197                 goto out;
3198         }
3199
3200         /* Stage 2 - find and update state(s) */
3201         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3202                 if ((x = xfrm_migrate_state_find(mp))) {
3203                         x_cur[nx_cur] = x;
3204                         nx_cur++;
3205                         if ((xc = xfrm_state_migrate(x, mp))) {
3206                                 x_new[nx_new] = xc;
3207                                 nx_new++;
3208                         } else {
3209                                 err = -ENODATA;
3210                                 goto restore_state;
3211                         }
3212                 }
3213         }
3214
3215         /* Stage 3 - update policy */
3216         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3217                 goto restore_state;
3218
3219         /* Stage 4 - delete old state(s) */
3220         if (nx_cur) {
3221                 xfrm_states_put(x_cur, nx_cur);
3222                 xfrm_states_delete(x_cur, nx_cur);
3223         }
3224
3225         /* Stage 5 - announce */
3226         km_migrate(sel, dir, type, m, num_migrate, k);
3227
3228         xfrm_pol_put(pol);
3229
3230         return 0;
3231 out:
3232         return err;
3233
3234 restore_state:
3235         if (pol)
3236                 xfrm_pol_put(pol);
3237         if (nx_cur)
3238                 xfrm_states_put(x_cur, nx_cur);
3239         if (nx_new)
3240                 xfrm_states_delete(x_new, nx_new);
3241
3242         return err;
3243 }
3244 EXPORT_SYMBOL(xfrm_migrate);
3245 #endif