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) || del_timer(&policy->polq.hold_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                 kfree_skb(skb);
 325 }
 326
 327 /* Rule must be locked. Release descentant resources, announce
 328  * entry dead. The rule must be unlinked from lists to the moment.
 329  */
 330
 331 static void xfrm_policy_kill(struct xfrm_policy *policy)
 332 {
 333         policy->walk.dead = 1;
 334
 335         atomic_inc(&policy->genid);
 336
 337         del_timer(&policy->polq.hold_timer);
 338         xfrm_queue_purge(&policy->polq.hold_queue);
 339
 340         if (del_timer(&policy->timer))
 341                 xfrm_pol_put(policy);
 342
 343         xfrm_pol_put(policy);
 344 }
 345
 346 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 347
 348 static inline unsigned int idx_hash(struct net *net, u32 index)
 349 {
 350         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 351 }
 352
 353 static struct hlist_head *policy_hash_bysel(struct net *net,
 354                                             const struct xfrm_selector *sel,
 355                                             unsigned short family, int dir)
 356 {
 357         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 358         unsigned int hash = __sel_hash(sel, family, hmask);
 359
 360         return (hash == hmask + 1 ?
 361                 &net->xfrm.policy_inexact[dir] :
 362                 net->xfrm.policy_bydst[dir].table + hash);
 363 }
 364
 365 static struct hlist_head *policy_hash_direct(struct net *net,
 366                                              const xfrm_address_t *daddr,
 367                                              const xfrm_address_t *saddr,
 368                                              unsigned short family, int dir)
 369 {
 370         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 371         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 372
 373         return net->xfrm.policy_bydst[dir].table + hash;
 374 }
 375
 376 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 377                                    struct hlist_head *ndsttable,
 378                                    unsigned int nhashmask)
 379 {
 380         struct hlist_node *tmp, *entry0 = NULL;
 381         struct xfrm_policy *pol;
 382         unsigned int h0 = 0;
 383
 384 redo:
 385         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 386                 unsigned int h;
 387
 388                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 389                                 pol->family, nhashmask);
 390                 if (!entry0) {
 391                         hlist_del(&pol->bydst);
 392                         hlist_add_head(&pol->bydst, ndsttable+h);
 393                         h0 = h;
 394                 } else {
 395                         if (h != h0)
 396                                 continue;
 397                         hlist_del(&pol->bydst);
 398                         hlist_add_after(entry0, &pol->bydst);
 399                 }
 400                 entry0 = &pol->bydst;
 401         }
 402         if (!hlist_empty(list)) {
 403                 entry0 = NULL;
 404                 goto redo;
 405         }
 406 }
 407
 408 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 409                                    struct hlist_head *nidxtable,
 410                                    unsigned int nhashmask)
 411 {
 412         struct hlist_node *tmp;
 413         struct xfrm_policy *pol;
 414
 415         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 416                 unsigned int h;
 417
 418                 h = __idx_hash(pol->index, nhashmask);
 419                 hlist_add_head(&pol->byidx, nidxtable+h);
 420         }
 421 }
 422
 423 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 424 {
 425         return ((old_hmask + 1) << 1) - 1;
 426 }
 427
 428 static void xfrm_bydst_resize(struct net *net, int dir)
 429 {
 430         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 431         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 432         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 433         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 434         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 435         int i;
 436
 437         if (!ndst)
 438                 return;
 439
 440         write_lock_bh(&xfrm_policy_lock);
 441
 442         for (i = hmask; i >= 0; i--)
 443                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 444
 445         net->xfrm.policy_bydst[dir].table = ndst;
 446         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 447
 448         write_unlock_bh(&xfrm_policy_lock);
 449
 450         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 451 }
 452
 453 static void xfrm_byidx_resize(struct net *net, int total)
 454 {
 455         unsigned int hmask = net->xfrm.policy_idx_hmask;
 456         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 457         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 458         struct hlist_head *oidx = net->xfrm.policy_byidx;
 459         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 460         int i;
 461
 462         if (!nidx)
 463                 return;
 464
 465         write_lock_bh(&xfrm_policy_lock);
 466
 467         for (i = hmask; i >= 0; i--)
 468                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 469
 470         net->xfrm.policy_byidx = nidx;
 471         net->xfrm.policy_idx_hmask = nhashmask;
 472
 473         write_unlock_bh(&xfrm_policy_lock);
 474
 475         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 476 }
 477
 478 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 479 {
 480         unsigned int cnt = net->xfrm.policy_count[dir];
 481         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 482
 483         if (total)
 484                 *total += cnt;
 485
 486         if ((hmask + 1) < xfrm_policy_hashmax &&
 487             cnt > hmask)
 488                 return 1;
 489
 490         return 0;
 491 }
 492
 493 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 494 {
 495         unsigned int hmask = net->xfrm.policy_idx_hmask;
 496
 497         if ((hmask + 1) < xfrm_policy_hashmax &&
 498             total > hmask)
 499                 return 1;
 500
 501         return 0;
 502 }
 503
 504 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 505 {
 506         read_lock_bh(&xfrm_policy_lock);
 507         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 508         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 509         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 510         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 511         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 512         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 513         si->spdhcnt = net->xfrm.policy_idx_hmask;
 514         si->spdhmcnt = xfrm_policy_hashmax;
 515         read_unlock_bh(&xfrm_policy_lock);
 516 }
 517 EXPORT_SYMBOL(xfrm_spd_getinfo);
 518
 519 static DEFINE_MUTEX(hash_resize_mutex);
 520 static void xfrm_hash_resize(struct work_struct *work)
 521 {
 522         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 523         int dir, total;
 524
 525         mutex_lock(&hash_resize_mutex);
 526
 527         total = 0;
 528         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 529                 if (xfrm_bydst_should_resize(net, dir, &total))
 530                         xfrm_bydst_resize(net, dir);
 531         }
 532         if (xfrm_byidx_should_resize(net, total))
 533                 xfrm_byidx_resize(net, total);
 534
 535         mutex_unlock(&hash_resize_mutex);
 536 }
 537
 538 /* Generate new index... KAME seems to generate them ordered by cost
 539  * of an absolute inpredictability of ordering of rules. This will not pass. */
 540 static u32 xfrm_gen_index(struct net *net, int dir)
 541 {
 542         static u32 idx_generator;
 543
 544         for (;;) {
 545                 struct hlist_head *list;
 546                 struct xfrm_policy *p;
 547                 u32 idx;
 548                 int found;
 549
 550                 idx = (idx_generator | dir);
 551                 idx_generator += 8;
 552                 if (idx == 0)
 553                         idx = 8;
 554                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 555                 found = 0;
 556                 hlist_for_each_entry(p, list, byidx) {
 557                         if (p->index == idx) {
 558                                 found = 1;
 559                                 break;
 560                         }
 561                 }
 562                 if (!found)
 563                         return idx;
 564         }
 565 }
 566
 567 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 568 {
 569         u32 *p1 = (u32 *) s1;
 570         u32 *p2 = (u32 *) s2;
 571         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 572         int i;
 573
 574         for (i = 0; i < len; i++) {
 575                 if (p1[i] != p2[i])
 576                         return 1;
 577         }
 578
 579         return 0;
 580 }
 581
 582 static void xfrm_policy_requeue(struct xfrm_policy *old,
 583                                 struct xfrm_policy *new)
 584 {
 585         struct xfrm_policy_queue *pq = &old->polq;
 586         struct sk_buff_head list;
 587
 588         __skb_queue_head_init(&list);
 589
 590         spin_lock_bh(&pq->hold_queue.lock);
 591         skb_queue_splice_init(&pq->hold_queue, &list);
 592         del_timer(&pq->hold_timer);
 593         spin_unlock_bh(&pq->hold_queue.lock);
 594
 595         if (skb_queue_empty(&list))
 596                 return;
 597
 598         pq = &new->polq;
 599
 600         spin_lock_bh(&pq->hold_queue.lock);
 601         skb_queue_splice(&list, &pq->hold_queue);
 602         pq->timeout = XFRM_QUEUE_TMO_MIN;
 603         mod_timer(&pq->hold_timer, jiffies);
 604         spin_unlock_bh(&pq->hold_queue.lock);
 605 }
 606
 607 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 608                                    struct xfrm_policy *pol)
 609 {
 610         u32 mark = policy->mark.v & policy->mark.m;
 611
 612         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 613                 return true;
 614
 615         if ((mark & pol->mark.m) == pol->mark.v &&
 616             policy->priority == pol->priority)
 617                 return true;
 618
 619         return false;
 620 }
 621
 622 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 623 {
 624         struct net *net = xp_net(policy);
 625         struct xfrm_policy *pol;
 626         struct xfrm_policy *delpol;
 627         struct hlist_head *chain;
 628         struct hlist_node *newpos;
 629
 630         write_lock_bh(&xfrm_policy_lock);
 631         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 632         delpol = NULL;
 633         newpos = NULL;
 634         hlist_for_each_entry(pol, chain, bydst) {
 635                 if (pol->type == policy->type &&
 636                     !selector_cmp(&pol->selector, &policy->selector) &&
 637                     xfrm_policy_mark_match(policy, pol) &&
 638                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 639                     !WARN_ON(delpol)) {
 640                         if (excl) {
 641                                 write_unlock_bh(&xfrm_policy_lock);
 642                                 return -EEXIST;
 643                         }
 644                         delpol = pol;
 645                         if (policy->priority > pol->priority)
 646                                 continue;
 647                 } else if (policy->priority >= pol->priority) {
 648                         newpos = &pol->bydst;
 649                         continue;
 650                 }
 651                 if (delpol)
 652                         break;
 653         }
 654         if (newpos)
 655                 hlist_add_after(newpos, &policy->bydst);
 656         else
 657                 hlist_add_head(&policy->bydst, chain);
 658         xfrm_pol_hold(policy);
 659         net->xfrm.policy_count[dir]++;
 660         atomic_inc(&flow_cache_genid);
 661
 662         /* After previous checking, family can either be AF_INET or AF_INET6 */
 663         if (policy->family == AF_INET)
 664                 rt_genid_bump_ipv4(net);
 665         else
 666                 rt_genid_bump_ipv6(net);
 667
 668         if (delpol) {
 669                 xfrm_policy_requeue(delpol, policy);
 670                 __xfrm_policy_unlink(delpol, dir);
 671         }
 672         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 673         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 674         policy->curlft.add_time = get_seconds();
 675         policy->curlft.use_time = 0;
 676         if (!mod_timer(&policy->timer, jiffies + HZ))
 677                 xfrm_pol_hold(policy);
 678         list_add(&policy->walk.all, &net->xfrm.policy_all);
 679         write_unlock_bh(&xfrm_policy_lock);
 680
 681         if (delpol)
 682                 xfrm_policy_kill(delpol);
 683         else if (xfrm_bydst_should_resize(net, dir, NULL))
 684                 schedule_work(&net->xfrm.policy_hash_work);
 685
 686         return 0;
 687 }
 688 EXPORT_SYMBOL(xfrm_policy_insert);
 689
 690 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 691                                           int dir, struct xfrm_selector *sel,
 692                                           struct xfrm_sec_ctx *ctx, int delete,
 693                                           int *err)
 694 {
 695         struct xfrm_policy *pol, *ret;
 696         struct hlist_head *chain;
 697
 698         *err = 0;
 699         write_lock_bh(&xfrm_policy_lock);
 700         chain = policy_hash_bysel(net, sel, sel->family, dir);
 701         ret = NULL;
 702         hlist_for_each_entry(pol, chain, bydst) {
 703                 if (pol->type == type &&
 704                     (mark & pol->mark.m) == pol->mark.v &&
 705                     !selector_cmp(sel, &pol->selector) &&
 706                     xfrm_sec_ctx_match(ctx, pol->security)) {
 707                         xfrm_pol_hold(pol);
 708                         if (delete) {
 709                                 *err = security_xfrm_policy_delete(
 710                                                                 pol->security);
 711                                 if (*err) {
 712                                         write_unlock_bh(&xfrm_policy_lock);
 713                                         return pol;
 714                                 }
 715                                 __xfrm_policy_unlink(pol, dir);
 716                         }
 717                         ret = pol;
 718                         break;
 719                 }
 720         }
 721         write_unlock_bh(&xfrm_policy_lock);
 722
 723         if (ret && delete)
 724                 xfrm_policy_kill(ret);
 725         return ret;
 726 }
 727 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 728
 729 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 730                                      int dir, u32 id, int delete, int *err)
 731 {
 732         struct xfrm_policy *pol, *ret;
 733         struct hlist_head *chain;
 734
 735         *err = -ENOENT;
 736         if (xfrm_policy_id2dir(id) != dir)
 737                 return NULL;
 738
 739         *err = 0;
 740         write_lock_bh(&xfrm_policy_lock);
 741         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 742         ret = NULL;
 743         hlist_for_each_entry(pol, chain, byidx) {
 744                 if (pol->type == type && pol->index == id &&
 745                     (mark & pol->mark.m) == pol->mark.v) {
 746                         xfrm_pol_hold(pol);
 747                         if (delete) {
 748                                 *err = security_xfrm_policy_delete(
 749                                                                 pol->security);
 750                                 if (*err) {
 751                                         write_unlock_bh(&xfrm_policy_lock);
 752                                         return pol;
 753                                 }
 754                                 __xfrm_policy_unlink(pol, dir);
 755                         }
 756                         ret = pol;
 757                         break;
 758                 }
 759         }
 760         write_unlock_bh(&xfrm_policy_lock);
 761
 762         if (ret && delete)
 763                 xfrm_policy_kill(ret);
 764         return ret;
 765 }
 766 EXPORT_SYMBOL(xfrm_policy_byid);
 767
 768 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 769 static inline int
 770 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 771 {
 772         int dir, err = 0;
 773
 774         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 775                 struct xfrm_policy *pol;
 776                 int i;
 777
 778                 hlist_for_each_entry(pol,
 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,
 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                 int i;
 832
 833         again1:
 834                 hlist_for_each_entry(pol,
 835                                      &net->xfrm.policy_inexact[dir], bydst) {
 836                         if (pol->type != type)
 837                                 continue;
 838                         __xfrm_policy_unlink(pol, dir);
 839                         write_unlock_bh(&xfrm_policy_lock);
 840                         cnt++;
 841
 842                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 843                                                  audit_info->sessionid,
 844                                                  audit_info->secid);
 845
 846                         xfrm_policy_kill(pol);
 847
 848                         write_lock_bh(&xfrm_policy_lock);
 849                         goto again1;
 850                 }
 851
 852                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 853         again2:
 854                         hlist_for_each_entry(pol,
 855                                              net->xfrm.policy_bydst[dir].table + i,
 856                                              bydst) {
 857                                 if (pol->type != type)
 858                                         continue;
 859                                 __xfrm_policy_unlink(pol, dir);
 860                                 write_unlock_bh(&xfrm_policy_lock);
 861                                 cnt++;
 862
 863                                 xfrm_audit_policy_delete(pol, 1,
 864                                                          audit_info->loginuid,
 865                                                          audit_info->sessionid,
 866                                                          audit_info->secid);
 867                                 xfrm_policy_kill(pol);
 868
 869                                 write_lock_bh(&xfrm_policy_lock);
 870                                 goto again2;
 871                         }
 872                 }
 873
 874         }
 875         if (!cnt)
 876                 err = -ESRCH;
 877 out:
 878         write_unlock_bh(&xfrm_policy_lock);
 879         return err;
 880 }
 881 EXPORT_SYMBOL(xfrm_policy_flush);
 882
 883 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 884                      int (*func)(struct xfrm_policy *, int, int, void*),
 885                      void *data)
 886 {
 887         struct xfrm_policy *pol;
 888         struct xfrm_policy_walk_entry *x;
 889         int error = 0;
 890
 891         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 892             walk->type != XFRM_POLICY_TYPE_ANY)
 893                 return -EINVAL;
 894
 895         if (list_empty(&walk->walk.all) && walk->seq != 0)
 896                 return 0;
 897
 898         write_lock_bh(&xfrm_policy_lock);
 899         if (list_empty(&walk->walk.all))
 900                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 901         else
 902                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 903         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 904                 if (x->dead)
 905                         continue;
 906                 pol = container_of(x, struct xfrm_policy, walk);
 907                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 908                     walk->type != pol->type)
 909                         continue;
 910                 error = func(pol, xfrm_policy_id2dir(pol->index),
 911                              walk->seq, data);
 912                 if (error) {
 913                         list_move_tail(&walk->walk.all, &x->all);
 914                         goto out;
 915                 }
 916                 walk->seq++;
 917         }
 918         if (walk->seq == 0) {
 919                 error = -ENOENT;
 920                 goto out;
 921         }
 922         list_del_init(&walk->walk.all);
 923 out:
 924         write_unlock_bh(&xfrm_policy_lock);
 925         return error;
 926 }
 927 EXPORT_SYMBOL(xfrm_policy_walk);
 928
 929 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 930 {
 931         INIT_LIST_HEAD(&walk->walk.all);
 932         walk->walk.dead = 1;
 933         walk->type = type;
 934         walk->seq = 0;
 935 }
 936 EXPORT_SYMBOL(xfrm_policy_walk_init);
 937
 938 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 939 {
 940         if (list_empty(&walk->walk.all))
 941                 return;
 942
 943         write_lock_bh(&xfrm_policy_lock);
 944         list_del(&walk->walk.all);
 945         write_unlock_bh(&xfrm_policy_lock);
 946 }
 947 EXPORT_SYMBOL(xfrm_policy_walk_done);
 948
 949 /*
 950  * Find policy to apply to this flow.
 951  *
 952  * Returns 0 if policy found, else an -errno.
 953  */
 954 static int xfrm_policy_match(const struct xfrm_policy *pol,
 955                              const struct flowi *fl,
 956                              u8 type, u16 family, int dir)
 957 {
 958         const struct xfrm_selector *sel = &pol->selector;
 959         int ret = -ESRCH;
 960         bool match;
 961
 962         if (pol->family != family ||
 963             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
 964             pol->type != type)
 965                 return ret;
 966
 967         match = xfrm_selector_match(sel, fl, family);
 968         if (match)
 969                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
 970                                                   dir);
 971
 972         return ret;
 973 }
 974
 975 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 976                                                      const struct flowi *fl,
 977                                                      u16 family, u8 dir)
 978 {
 979         int err;
 980         struct xfrm_policy *pol, *ret;
 981         const xfrm_address_t *daddr, *saddr;
 982         struct hlist_head *chain;
 983         u32 priority = ~0U;
 984
 985         daddr = xfrm_flowi_daddr(fl, family);
 986         saddr = xfrm_flowi_saddr(fl, family);
 987         if (unlikely(!daddr || !saddr))
 988                 return NULL;
 989
 990         read_lock_bh(&xfrm_policy_lock);
 991         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 992         ret = NULL;
 993         hlist_for_each_entry(pol, chain, bydst) {
 994                 err = xfrm_policy_match(pol, fl, type, family, dir);
 995                 if (err) {
 996                         if (err == -ESRCH)
 997                                 continue;
 998                         else {
 999                                 ret = ERR_PTR(err);
1000                                 goto fail;
1001                         }
1002                 } else {
1003                         ret = pol;
1004                         priority = ret->priority;
1005                         break;
1006                 }
1007         }
1008         chain = &net->xfrm.policy_inexact[dir];
1009         hlist_for_each_entry(pol, chain, bydst) {
1010                 err = xfrm_policy_match(pol, fl, type, family, dir);
1011                 if (err) {
1012                         if (err == -ESRCH)
1013                                 continue;
1014                         else {
1015                                 ret = ERR_PTR(err);
1016                                 goto fail;
1017                         }
1018                 } else if (pol->priority < priority) {
1019                         ret = pol;
1020                         break;
1021                 }
1022         }
1023         if (ret)
1024                 xfrm_pol_hold(ret);
1025 fail:
1026         read_unlock_bh(&xfrm_policy_lock);
1027
1028         return ret;
1029 }
1030
1031 static struct xfrm_policy *
1032 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1033 {
1034 #ifdef CONFIG_XFRM_SUB_POLICY
1035         struct xfrm_policy *pol;
1036
1037         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1038         if (pol != NULL)
1039                 return pol;
1040 #endif
1041         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1042 }
1043
1044 static int flow_to_policy_dir(int dir)
1045 {
1046         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1047             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1048             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1049                 return dir;
1050
1051         switch (dir) {
1052         default:
1053         case FLOW_DIR_IN:
1054                 return XFRM_POLICY_IN;
1055         case FLOW_DIR_OUT:
1056                 return XFRM_POLICY_OUT;
1057         case FLOW_DIR_FWD:
1058                 return XFRM_POLICY_FWD;
1059         }
1060 }
1061
1062 static struct flow_cache_object *
1063 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1064                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1065 {
1066         struct xfrm_policy *pol;
1067
1068         if (old_obj)
1069                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1070
1071         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1072         if (IS_ERR_OR_NULL(pol))
1073                 return ERR_CAST(pol);
1074
1075         /* Resolver returns two references:
1076          * one for cache and one for caller of flow_cache_lookup() */
1077         xfrm_pol_hold(pol);
1078
1079         return &pol->flo;
1080 }
1081
1082 static inline int policy_to_flow_dir(int dir)
1083 {
1084         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1085             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1086             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1087                 return dir;
1088         switch (dir) {
1089         default:
1090         case XFRM_POLICY_IN:
1091                 return FLOW_DIR_IN;
1092         case XFRM_POLICY_OUT:
1093                 return FLOW_DIR_OUT;
1094         case XFRM_POLICY_FWD:
1095                 return FLOW_DIR_FWD;
1096         }
1097 }
1098
1099 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1100                                                  const struct flowi *fl)
1101 {
1102         struct xfrm_policy *pol;
1103
1104         read_lock_bh(&xfrm_policy_lock);
1105         if ((pol = sk->sk_policy[dir]) != NULL) {
1106                 bool match = xfrm_selector_match(&pol->selector, fl,
1107                                                  sk->sk_family);
1108                 int err = 0;
1109
1110                 if (match) {
1111                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1112                                 pol = NULL;
1113                                 goto out;
1114                         }
1115                         err = security_xfrm_policy_lookup(pol->security,
1116                                                       fl->flowi_secid,
1117                                                       policy_to_flow_dir(dir));
1118                         if (!err)
1119                                 xfrm_pol_hold(pol);
1120                         else if (err == -ESRCH)
1121                                 pol = NULL;
1122                         else
1123                                 pol = ERR_PTR(err);
1124                 } else
1125                         pol = NULL;
1126         }
1127 out:
1128         read_unlock_bh(&xfrm_policy_lock);
1129         return pol;
1130 }
1131
1132 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1133 {
1134         struct net *net = xp_net(pol);
1135         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1136                                                      pol->family, dir);
1137
1138         list_add(&pol->walk.all, &net->xfrm.policy_all);
1139         hlist_add_head(&pol->bydst, chain);
1140         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1141         net->xfrm.policy_count[dir]++;
1142         xfrm_pol_hold(pol);
1143
1144         if (xfrm_bydst_should_resize(net, dir, NULL))
1145                 schedule_work(&net->xfrm.policy_hash_work);
1146 }
1147
1148 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1149                                                 int dir)
1150 {
1151         struct net *net = xp_net(pol);
1152
1153         if (hlist_unhashed(&pol->bydst))
1154                 return NULL;
1155
1156         hlist_del(&pol->bydst);
1157         hlist_del(&pol->byidx);
1158         list_del(&pol->walk.all);
1159         net->xfrm.policy_count[dir]--;
1160
1161         return pol;
1162 }
1163
1164 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1165 {
1166         write_lock_bh(&xfrm_policy_lock);
1167         pol = __xfrm_policy_unlink(pol, dir);
1168         write_unlock_bh(&xfrm_policy_lock);
1169         if (pol) {
1170                 xfrm_policy_kill(pol);
1171                 return 0;
1172         }
1173         return -ENOENT;
1174 }
1175 EXPORT_SYMBOL(xfrm_policy_delete);
1176
1177 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1178 {
1179         struct net *net = xp_net(pol);
1180         struct xfrm_policy *old_pol;
1181
1182 #ifdef CONFIG_XFRM_SUB_POLICY
1183         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1184                 return -EINVAL;
1185 #endif
1186
1187         write_lock_bh(&xfrm_policy_lock);
1188         old_pol = sk->sk_policy[dir];
1189         sk->sk_policy[dir] = pol;
1190         if (pol) {
1191                 pol->curlft.add_time = get_seconds();
1192                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1193                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1194         }
1195         if (old_pol) {
1196                 if (pol)
1197                         xfrm_policy_requeue(old_pol, pol);
1198
1199                 /* Unlinking succeeds always. This is the only function
1200                  * allowed to delete or replace socket policy.
1201                  */
1202                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1203         }
1204         write_unlock_bh(&xfrm_policy_lock);
1205
1206         if (old_pol) {
1207                 xfrm_policy_kill(old_pol);
1208         }
1209         return 0;
1210 }
1211
1212 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1213 {
1214         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1215
1216         if (newp) {
1217                 newp->selector = old->selector;
1218                 if (security_xfrm_policy_clone(old->security,
1219                                                &newp->security)) {
1220                         kfree(newp);
1221                         return NULL;  /* ENOMEM */
1222                 }
1223                 newp->lft = old->lft;
1224                 newp->curlft = old->curlft;
1225                 newp->mark = old->mark;
1226                 newp->action = old->action;
1227                 newp->flags = old->flags;
1228                 newp->xfrm_nr = old->xfrm_nr;
1229                 newp->index = old->index;
1230                 newp->type = old->type;
1231                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1232                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1233                 write_lock_bh(&xfrm_policy_lock);
1234                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1235                 write_unlock_bh(&xfrm_policy_lock);
1236                 xfrm_pol_put(newp);
1237         }
1238         return newp;
1239 }
1240
1241 int __xfrm_sk_clone_policy(struct sock *sk)
1242 {
1243         struct xfrm_policy *p0 = sk->sk_policy[0],
1244                            *p1 = sk->sk_policy[1];
1245
1246         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1247         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1248                 return -ENOMEM;
1249         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1250                 return -ENOMEM;
1251         return 0;
1252 }
1253
1254 static int
1255 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1256                unsigned short family)
1257 {
1258         int err;
1259         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1260
1261         if (unlikely(afinfo == NULL))
1262                 return -EINVAL;
1263         err = afinfo->get_saddr(net, local, remote);
1264         xfrm_policy_put_afinfo(afinfo);
1265         return err;
1266 }
1267
1268 /* Resolve list of templates for the flow, given policy. */
1269
1270 static int
1271 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1272                       struct xfrm_state **xfrm, unsigned short family)
1273 {
1274         struct net *net = xp_net(policy);
1275         int nx;
1276         int i, error;
1277         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1278         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1279         xfrm_address_t tmp;
1280
1281         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1282                 struct xfrm_state *x;
1283                 xfrm_address_t *remote = daddr;
1284                 xfrm_address_t *local  = saddr;
1285                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1286
1287                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1288                     tmpl->mode == XFRM_MODE_BEET) {
1289                         remote = &tmpl->id.daddr;
1290                         local = &tmpl->saddr;
1291                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1292                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1293                                 if (error)
1294                                         goto fail;
1295                                 local = &tmp;
1296                         }
1297                 }
1298
1299                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1300
1301                 if (x && x->km.state == XFRM_STATE_VALID) {
1302                         xfrm[nx++] = x;
1303                         daddr = remote;
1304                         saddr = local;
1305                         continue;
1306                 }
1307                 if (x) {
1308                         error = (x->km.state == XFRM_STATE_ERROR ?
1309                                  -EINVAL : -EAGAIN);
1310                         xfrm_state_put(x);
1311                 }
1312                 else if (error == -ESRCH)
1313                         error = -EAGAIN;
1314
1315                 if (!tmpl->optional)
1316                         goto fail;
1317         }
1318         return nx;
1319
1320 fail:
1321         for (nx--; nx>=0; nx--)
1322                 xfrm_state_put(xfrm[nx]);
1323         return error;
1324 }
1325
1326 static int
1327 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1328                   struct xfrm_state **xfrm, unsigned short family)
1329 {
1330         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1331         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1332         int cnx = 0;
1333         int error;
1334         int ret;
1335         int i;
1336
1337         for (i = 0; i < npols; i++) {
1338                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1339                         error = -ENOBUFS;
1340                         goto fail;
1341                 }
1342
1343                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1344                 if (ret < 0) {
1345                         error = ret;
1346                         goto fail;
1347                 } else
1348                         cnx += ret;
1349         }
1350
1351         /* found states are sorted for outbound processing */
1352         if (npols > 1)
1353                 xfrm_state_sort(xfrm, tpp, cnx, family);
1354
1355         return cnx;
1356
1357  fail:
1358         for (cnx--; cnx>=0; cnx--)
1359                 xfrm_state_put(tpp[cnx]);
1360         return error;
1361
1362 }
1363
1364 /* Check that the bundle accepts the flow and its components are
1365  * still valid.
1366  */
1367
1368 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1369 {
1370         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1371         int tos;
1372
1373         if (!afinfo)
1374                 return -EINVAL;
1375
1376         tos = afinfo->get_tos(fl);
1377
1378         xfrm_policy_put_afinfo(afinfo);
1379
1380         return tos;
1381 }
1382
1383 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1384 {
1385         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1386         struct dst_entry *dst = &xdst->u.dst;
1387
1388         if (xdst->route == NULL) {
1389                 /* Dummy bundle - if it has xfrms we were not
1390                  * able to build bundle as template resolution failed.
1391                  * It means we need to try again resolving. */
1392                 if (xdst->num_xfrms > 0)
1393                         return NULL;
1394         } else if (dst->flags & DST_XFRM_QUEUE) {
1395                 return NULL;
1396         } else {
1397                 /* Real bundle */
1398                 if (stale_bundle(dst))
1399                         return NULL;
1400         }
1401
1402         dst_hold(dst);
1403         return flo;
1404 }
1405
1406 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1407 {
1408         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1409         struct dst_entry *dst = &xdst->u.dst;
1410
1411         if (!xdst->route)
1412                 return 0;
1413         if (stale_bundle(dst))
1414                 return 0;
1415
1416         return 1;
1417 }
1418
1419 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1420 {
1421         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1422         struct dst_entry *dst = &xdst->u.dst;
1423
1424         dst_free(dst);
1425 }
1426
1427 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1428         .get = xfrm_bundle_flo_get,
1429         .check = xfrm_bundle_flo_check,
1430         .delete = xfrm_bundle_flo_delete,
1431 };
1432
1433 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1434 {
1435         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1436         struct dst_ops *dst_ops;
1437         struct xfrm_dst *xdst;
1438
1439         if (!afinfo)
1440                 return ERR_PTR(-EINVAL);
1441
1442         switch (family) {
1443         case AF_INET:
1444                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1445                 break;
1446 #if IS_ENABLED(CONFIG_IPV6)
1447         case AF_INET6:
1448                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1449                 break;
1450 #endif
1451         default:
1452                 BUG();
1453         }
1454         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1455
1456         if (likely(xdst)) {
1457                 struct dst_entry *dst = &xdst->u.dst;
1458
1459                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1460                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1461                 if (afinfo->init_dst)
1462                         afinfo->init_dst(net, xdst);
1463         } else
1464                 xdst = ERR_PTR(-ENOBUFS);
1465
1466         xfrm_policy_put_afinfo(afinfo);
1467
1468         return xdst;
1469 }
1470
1471 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1472                                  int nfheader_len)
1473 {
1474         struct xfrm_policy_afinfo *afinfo =
1475                 xfrm_policy_get_afinfo(dst->ops->family);
1476         int err;
1477
1478         if (!afinfo)
1479                 return -EINVAL;
1480
1481         err = afinfo->init_path(path, dst, nfheader_len);
1482
1483         xfrm_policy_put_afinfo(afinfo);
1484
1485         return err;
1486 }
1487
1488 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1489                                 const struct flowi *fl)
1490 {
1491         struct xfrm_policy_afinfo *afinfo =
1492                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1493         int err;
1494
1495         if (!afinfo)
1496                 return -EINVAL;
1497
1498         err = afinfo->fill_dst(xdst, dev, fl);
1499
1500         xfrm_policy_put_afinfo(afinfo);
1501
1502         return err;
1503 }
1504
1505
1506 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1507  * all the metrics... Shortly, bundle a bundle.
1508  */
1509
1510 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1511                                             struct xfrm_state **xfrm, int nx,
1512                                             const struct flowi *fl,
1513                                             struct dst_entry *dst)
1514 {
1515         struct net *net = xp_net(policy);
1516         unsigned long now = jiffies;
1517         struct net_device *dev;
1518         struct xfrm_mode *inner_mode;
1519         struct dst_entry *dst_prev = NULL;
1520         struct dst_entry *dst0 = NULL;
1521         int i = 0;
1522         int err;
1523         int header_len = 0;
1524         int nfheader_len = 0;
1525         int trailer_len = 0;
1526         int tos;
1527         int family = policy->selector.family;
1528         xfrm_address_t saddr, daddr;
1529
1530         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1531
1532         tos = xfrm_get_tos(fl, family);
1533         err = tos;
1534         if (tos < 0)
1535                 goto put_states;
1536
1537         dst_hold(dst);
1538
1539         for (; i < nx; i++) {
1540                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1541                 struct dst_entry *dst1 = &xdst->u.dst;
1542
1543                 err = PTR_ERR(xdst);
1544                 if (IS_ERR(xdst)) {
1545                         dst_release(dst);
1546                         goto put_states;
1547                 }
1548
1549                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1550                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1551                                                         xfrm_af2proto(family));
1552                         if (!inner_mode) {
1553                                 err = -EAFNOSUPPORT;
1554                                 dst_release(dst);
1555                                 goto put_states;
1556                         }
1557                 } else
1558                         inner_mode = xfrm[i]->inner_mode;
1559
1560                 if (!dst_prev)
1561                         dst0 = dst1;
1562                 else {
1563                         dst_prev->child = dst_clone(dst1);
1564                         dst1->flags |= DST_NOHASH;
1565                 }
1566
1567                 xdst->route = dst;
1568                 dst_copy_metrics(dst1, dst);
1569
1570                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1571                         family = xfrm[i]->props.family;
1572                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1573                                               family);
1574                         err = PTR_ERR(dst);
1575                         if (IS_ERR(dst))
1576                                 goto put_states;
1577                 } else
1578                         dst_hold(dst);
1579
1580                 dst1->xfrm = xfrm[i];
1581                 xdst->xfrm_genid = xfrm[i]->genid;
1582
1583                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1584                 dst1->flags |= DST_HOST;
1585                 dst1->lastuse = now;
1586
1587                 dst1->input = dst_discard;
1588                 dst1->output = inner_mode->afinfo->output;
1589
1590                 dst1->next = dst_prev;
1591                 dst_prev = dst1;
1592
1593                 header_len += xfrm[i]->props.header_len;
1594                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1595                         nfheader_len += xfrm[i]->props.header_len;
1596                 trailer_len += xfrm[i]->props.trailer_len;
1597         }
1598
1599         dst_prev->child = dst;
1600         dst0->path = dst;
1601
1602         err = -ENODEV;
1603         dev = dst->dev;
1604         if (!dev)
1605                 goto free_dst;
1606
1607         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1608         xfrm_init_pmtu(dst_prev);
1609
1610         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1611                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1612
1613                 err = xfrm_fill_dst(xdst, dev, fl);
1614                 if (err)
1615                         goto free_dst;
1616
1617                 dst_prev->header_len = header_len;
1618                 dst_prev->trailer_len = trailer_len;
1619                 header_len -= xdst->u.dst.xfrm->props.header_len;
1620                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1621         }
1622
1623 out:
1624         return dst0;
1625
1626 put_states:
1627         for (; i < nx; i++)
1628                 xfrm_state_put(xfrm[i]);
1629 free_dst:
1630         if (dst0)
1631                 dst_free(dst0);
1632         dst0 = ERR_PTR(err);
1633         goto out;
1634 }
1635
1636 static int inline
1637 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1638 {
1639         if (!*target) {
1640                 *target = kmalloc(size, GFP_ATOMIC);
1641                 if (!*target)
1642                         return -ENOMEM;
1643         }
1644         memcpy(*target, src, size);
1645         return 0;
1646 }
1647
1648 static int inline
1649 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1650 {
1651 #ifdef CONFIG_XFRM_SUB_POLICY
1652         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1653         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1654                                    sel, sizeof(*sel));
1655 #else
1656         return 0;
1657 #endif
1658 }
1659
1660 static int inline
1661 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1662 {
1663 #ifdef CONFIG_XFRM_SUB_POLICY
1664         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1665         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1666 #else
1667         return 0;
1668 #endif
1669 }
1670
1671 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1672                                 struct xfrm_policy **pols,
1673                                 int *num_pols, int *num_xfrms)
1674 {
1675         int i;
1676
1677         if (*num_pols == 0 || !pols[0]) {
1678                 *num_pols = 0;
1679                 *num_xfrms = 0;
1680                 return 0;
1681         }
1682         if (IS_ERR(pols[0]))
1683                 return PTR_ERR(pols[0]);
1684
1685         *num_xfrms = pols[0]->xfrm_nr;
1686
1687 #ifdef CONFIG_XFRM_SUB_POLICY
1688         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1689             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1690                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1691                                                     XFRM_POLICY_TYPE_MAIN,
1692                                                     fl, family,
1693                                                     XFRM_POLICY_OUT);
1694                 if (pols[1]) {
1695                         if (IS_ERR(pols[1])) {
1696                                 xfrm_pols_put(pols, *num_pols);
1697                                 return PTR_ERR(pols[1]);
1698                         }
1699                         (*num_pols) ++;
1700                         (*num_xfrms) += pols[1]->xfrm_nr;
1701                 }
1702         }
1703 #endif
1704         for (i = 0; i < *num_pols; i++) {
1705                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1706                         *num_xfrms = -1;
1707                         break;
1708                 }
1709         }
1710
1711         return 0;
1712
1713 }
1714
1715 static struct xfrm_dst *
1716 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1717                                const struct flowi *fl, u16 family,
1718                                struct dst_entry *dst_orig)
1719 {
1720         struct net *net = xp_net(pols[0]);
1721         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1722         struct dst_entry *dst;
1723         struct xfrm_dst *xdst;
1724         int err;
1725
1726         /* Try to instantiate a bundle */
1727         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1728         if (err <= 0) {
1729                 if (err != 0 && err != -EAGAIN)
1730                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1731                 return ERR_PTR(err);
1732         }
1733
1734         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1735         if (IS_ERR(dst)) {
1736                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1737                 return ERR_CAST(dst);
1738         }
1739
1740         xdst = (struct xfrm_dst *)dst;
1741         xdst->num_xfrms = err;
1742         if (num_pols > 1)
1743                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1744         else
1745                 err = xfrm_dst_update_origin(dst, fl);
1746         if (unlikely(err)) {
1747                 dst_free(dst);
1748                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1749                 return ERR_PTR(err);
1750         }
1751
1752         xdst->num_pols = num_pols;
1753         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1754         xdst->policy_genid = atomic_read(&pols[0]->genid);
1755
1756         return xdst;
1757 }
1758
1759 static void xfrm_policy_queue_process(unsigned long arg)
1760 {
1761         int err = 0;
1762         struct sk_buff *skb;
1763         struct sock *sk;
1764         struct dst_entry *dst;
1765         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1766         struct xfrm_policy_queue *pq = &pol->polq;
1767         struct flowi fl;
1768         struct sk_buff_head list;
1769
1770         spin_lock(&pq->hold_queue.lock);
1771         skb = skb_peek(&pq->hold_queue);
1772         dst = skb_dst(skb);
1773         sk = skb->sk;
1774         xfrm_decode_session(skb, &fl, dst->ops->family);
1775         spin_unlock(&pq->hold_queue.lock);
1776
1777         dst_hold(dst->path);
1778         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1779                           sk, 0);
1780         if (IS_ERR(dst))
1781                 goto purge_queue;
1782
1783         if (dst->flags & DST_XFRM_QUEUE) {
1784                 dst_release(dst);
1785
1786                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1787                         goto purge_queue;
1788
1789                 pq->timeout = pq->timeout << 1;
1790                 mod_timer(&pq->hold_timer, jiffies + pq->timeout);
1791                 return;
1792         }
1793
1794         dst_release(dst);
1795
1796         __skb_queue_head_init(&list);
1797
1798         spin_lock(&pq->hold_queue.lock);
1799         pq->timeout = 0;
1800         skb_queue_splice_init(&pq->hold_queue, &list);
1801         spin_unlock(&pq->hold_queue.lock);
1802
1803         while (!skb_queue_empty(&list)) {
1804                 skb = __skb_dequeue(&list);
1805
1806                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1807                 dst_hold(skb_dst(skb)->path);
1808                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1809                                   &fl, skb->sk, 0);
1810                 if (IS_ERR(dst)) {
1811                         kfree_skb(skb);
1812                         continue;
1813                 }
1814
1815                 nf_reset(skb);
1816                 skb_dst_drop(skb);
1817                 skb_dst_set(skb, dst);
1818
1819                 err = dst_output(skb);
1820         }
1821
1822         return;
1823
1824 purge_queue:
1825         pq->timeout = 0;
1826         xfrm_queue_purge(&pq->hold_queue);
1827 }
1828
1829 static int xdst_queue_output(struct sk_buff *skb)
1830 {
1831         unsigned long sched_next;
1832         struct dst_entry *dst = skb_dst(skb);
1833         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1834         struct xfrm_policy_queue *pq = &xdst->pols[0]->polq;
1835
1836         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1837                 kfree_skb(skb);
1838                 return -EAGAIN;
1839         }
1840
1841         skb_dst_force(skb);
1842
1843         spin_lock_bh(&pq->hold_queue.lock);
1844
1845         if (!pq->timeout)
1846                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1847
1848         sched_next = jiffies + pq->timeout;
1849
1850         if (del_timer(&pq->hold_timer)) {
1851                 if (time_before(pq->hold_timer.expires, sched_next))
1852                         sched_next = pq->hold_timer.expires;
1853         }
1854
1855         __skb_queue_tail(&pq->hold_queue, skb);
1856         mod_timer(&pq->hold_timer, sched_next);
1857
1858         spin_unlock_bh(&pq->hold_queue.lock);
1859
1860         return 0;
1861 }
1862
1863 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1864                                                  struct dst_entry *dst,
1865                                                  const struct flowi *fl,
1866                                                  int num_xfrms,
1867                                                  u16 family)
1868 {
1869         int err;
1870         struct net_device *dev;
1871         struct dst_entry *dst1;
1872         struct xfrm_dst *xdst;
1873
1874         xdst = xfrm_alloc_dst(net, family);
1875         if (IS_ERR(xdst))
1876                 return xdst;
1877
1878         if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0 ||
1879             (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP))
1880                 return xdst;
1881
1882         dst1 = &xdst->u.dst;
1883         dst_hold(dst);
1884         xdst->route = dst;
1885
1886         dst_copy_metrics(dst1, dst);
1887
1888         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1889         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1890         dst1->lastuse = jiffies;
1891
1892         dst1->input = dst_discard;
1893         dst1->output = xdst_queue_output;
1894
1895         dst_hold(dst);
1896         dst1->child = dst;
1897         dst1->path = dst;
1898
1899         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1900
1901         err = -ENODEV;
1902         dev = dst->dev;
1903         if (!dev)
1904                 goto free_dst;
1905
1906         err = xfrm_fill_dst(xdst, dev, fl);
1907         if (err)
1908                 goto free_dst;
1909
1910 out:
1911         return xdst;
1912
1913 free_dst:
1914         dst_release(dst1);
1915         xdst = ERR_PTR(err);
1916         goto out;
1917 }
1918
1919 static struct flow_cache_object *
1920 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1921                    struct flow_cache_object *oldflo, void *ctx)
1922 {
1923         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1924         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1925         struct xfrm_dst *xdst, *new_xdst;
1926         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1927
1928         /* Check if the policies from old bundle are usable */
1929         xdst = NULL;
1930         if (oldflo) {
1931                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1932                 num_pols = xdst->num_pols;
1933                 num_xfrms = xdst->num_xfrms;
1934                 pol_dead = 0;
1935                 for (i = 0; i < num_pols; i++) {
1936                         pols[i] = xdst->pols[i];
1937                         pol_dead |= pols[i]->walk.dead;
1938                 }
1939                 if (pol_dead) {
1940                         dst_free(&xdst->u.dst);
1941                         xdst = NULL;
1942                         num_pols = 0;
1943                         num_xfrms = 0;
1944                         oldflo = NULL;
1945                 }
1946         }
1947
1948         /* Resolve policies to use if we couldn't get them from
1949          * previous cache entry */
1950         if (xdst == NULL) {
1951                 num_pols = 1;
1952                 pols[0] = __xfrm_policy_lookup(net, fl, family,
1953                                                flow_to_policy_dir(dir));
1954                 err = xfrm_expand_policies(fl, family, pols,
1955                                            &num_pols, &num_xfrms);
1956                 if (err < 0)
1957                         goto inc_error;
1958                 if (num_pols == 0)
1959                         return NULL;
1960                 if (num_xfrms <= 0)
1961                         goto make_dummy_bundle;
1962         }
1963
1964         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1965         if (IS_ERR(new_xdst)) {
1966                 err = PTR_ERR(new_xdst);
1967                 if (err != -EAGAIN)
1968                         goto error;
1969                 if (oldflo == NULL)
1970                         goto make_dummy_bundle;
1971                 dst_hold(&xdst->u.dst);
1972                 return oldflo;
1973         } else if (new_xdst == NULL) {
1974                 num_xfrms = 0;
1975                 if (oldflo == NULL)
1976                         goto make_dummy_bundle;
1977                 xdst->num_xfrms = 0;
1978                 dst_hold(&xdst->u.dst);
1979                 return oldflo;
1980         }
1981
1982         /* Kill the previous bundle */
1983         if (xdst) {
1984                 /* The policies were stolen for newly generated bundle */
1985                 xdst->num_pols = 0;
1986                 dst_free(&xdst->u.dst);
1987         }
1988
1989         /* Flow cache does not have reference, it dst_free()'s,
1990          * but we do need to return one reference for original caller */
1991         dst_hold(&new_xdst->u.dst);
1992         return &new_xdst->flo;
1993
1994 make_dummy_bundle:
1995         /* We found policies, but there's no bundles to instantiate:
1996          * either because the policy blocks, has no transformations or
1997          * we could not build template (no xfrm_states).*/
1998         xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
1999         if (IS_ERR(xdst)) {
2000                 xfrm_pols_put(pols, num_pols);
2001                 return ERR_CAST(xdst);
2002         }
2003         xdst->num_pols = num_pols;
2004         xdst->num_xfrms = num_xfrms;
2005         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
2006
2007         dst_hold(&xdst->u.dst);
2008         return &xdst->flo;
2009
2010 inc_error:
2011         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2012 error:
2013         if (xdst != NULL)
2014                 dst_free(&xdst->u.dst);
2015         else
2016                 xfrm_pols_put(pols, num_pols);
2017         return ERR_PTR(err);
2018 }
2019
2020 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2021                                         struct dst_entry *dst_orig)
2022 {
2023         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2024         struct dst_entry *ret;
2025
2026         if (!afinfo) {
2027                 dst_release(dst_orig);
2028                 return ERR_PTR(-EINVAL);
2029         } else {
2030                 ret = afinfo->blackhole_route(net, dst_orig);
2031         }
2032         xfrm_policy_put_afinfo(afinfo);
2033
2034         return ret;
2035 }
2036
2037 /* Main function: finds/creates a bundle for given flow.
2038  *
2039  * At the moment we eat a raw IP route. Mostly to speed up lookups
2040  * on interfaces with disabled IPsec.
2041  */
2042 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2043                               const struct flowi *fl,
2044                               struct sock *sk, int flags)
2045 {
2046         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2047         struct flow_cache_object *flo;
2048         struct xfrm_dst *xdst;
2049         struct dst_entry *dst, *route;
2050         u16 family = dst_orig->ops->family;
2051         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2052         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2053
2054 restart:
2055         dst = NULL;
2056         xdst = NULL;
2057         route = NULL;
2058
2059         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2060                 num_pols = 1;
2061                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2062                 err = xfrm_expand_policies(fl, family, pols,
2063                                            &num_pols, &num_xfrms);
2064                 if (err < 0)
2065                         goto dropdst;
2066
2067                 if (num_pols) {
2068                         if (num_xfrms <= 0) {
2069                                 drop_pols = num_pols;
2070                                 goto no_transform;
2071                         }
2072
2073                         xdst = xfrm_resolve_and_create_bundle(
2074                                         pols, num_pols, fl,
2075                                         family, dst_orig);
2076                         if (IS_ERR(xdst)) {
2077                                 xfrm_pols_put(pols, num_pols);
2078                                 err = PTR_ERR(xdst);
2079                                 goto dropdst;
2080                         } else if (xdst == NULL) {
2081                                 num_xfrms = 0;
2082                                 drop_pols = num_pols;
2083                                 goto no_transform;
2084                         }
2085
2086                         dst_hold(&xdst->u.dst);
2087
2088                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2089                         xdst->u.dst.next = xfrm_policy_sk_bundles;
2090                         xfrm_policy_sk_bundles = &xdst->u.dst;
2091                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2092
2093                         route = xdst->route;
2094                 }
2095         }
2096
2097         if (xdst == NULL) {
2098                 /* To accelerate a bit...  */
2099                 if ((dst_orig->flags & DST_NOXFRM) ||
2100                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2101                         goto nopol;
2102
2103                 flo = flow_cache_lookup(net, fl, family, dir,
2104                                         xfrm_bundle_lookup, dst_orig);
2105                 if (flo == NULL)
2106                         goto nopol;
2107                 if (IS_ERR(flo)) {
2108                         err = PTR_ERR(flo);
2109                         goto dropdst;
2110                 }
2111                 xdst = container_of(flo, struct xfrm_dst, flo);
2112
2113                 num_pols = xdst->num_pols;
2114                 num_xfrms = xdst->num_xfrms;
2115                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
2116                 route = xdst->route;
2117         }
2118
2119         dst = &xdst->u.dst;
2120         if (route == NULL && num_xfrms > 0) {
2121                 /* The only case when xfrm_bundle_lookup() returns a
2122                  * bundle with null route, is when the template could
2123                  * not be resolved. It means policies are there, but
2124                  * bundle could not be created, since we don't yet
2125                  * have the xfrm_state's. We need to wait for KM to
2126                  * negotiate new SA's or bail out with error.*/
2127                 if (net->xfrm.sysctl_larval_drop) {
2128                         dst_release(dst);
2129                         xfrm_pols_put(pols, drop_pols);
2130                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2131
2132                         return make_blackhole(net, family, dst_orig);
2133                 }
2134                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
2135                         DECLARE_WAITQUEUE(wait, current);
2136
2137                         add_wait_queue(&net->xfrm.km_waitq, &wait);
2138                         set_current_state(TASK_INTERRUPTIBLE);
2139                         schedule();
2140                         set_current_state(TASK_RUNNING);
2141                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
2142
2143                         if (!signal_pending(current)) {
2144                                 dst_release(dst);
2145                                 goto restart;
2146                         }
2147
2148                         err = -ERESTART;
2149                 } else
2150                         err = -EAGAIN;
2151
2152                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2153                 goto error;
2154         }
2155
2156 no_transform:
2157         if (num_pols == 0)
2158                 goto nopol;
2159
2160         if ((flags & XFRM_LOOKUP_ICMP) &&
2161             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2162                 err = -ENOENT;
2163                 goto error;
2164         }
2165
2166         for (i = 0; i < num_pols; i++)
2167                 pols[i]->curlft.use_time = get_seconds();
2168
2169         if (num_xfrms < 0) {
2170                 /* Prohibit the flow */
2171                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2172                 err = -EPERM;
2173                 goto error;
2174         } else if (num_xfrms > 0) {
2175                 /* Flow transformed */
2176                 dst_release(dst_orig);
2177         } else {
2178                 /* Flow passes untransformed */
2179                 dst_release(dst);
2180                 dst = dst_orig;
2181         }
2182 ok:
2183         xfrm_pols_put(pols, drop_pols);
2184         if (dst && dst->xfrm &&
2185             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2186                 dst->flags |= DST_XFRM_TUNNEL;
2187         return dst;
2188
2189 nopol:
2190         if (!(flags & XFRM_LOOKUP_ICMP)) {
2191                 dst = dst_orig;
2192                 goto ok;
2193         }
2194         err = -ENOENT;
2195 error:
2196         dst_release(dst);
2197 dropdst:
2198         dst_release(dst_orig);
2199         xfrm_pols_put(pols, drop_pols);
2200         return ERR_PTR(err);
2201 }
2202 EXPORT_SYMBOL(xfrm_lookup);
2203
2204 static inline int
2205 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2206 {
2207         struct xfrm_state *x;
2208
2209         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2210                 return 0;
2211         x = skb->sp->xvec[idx];
2212         if (!x->type->reject)
2213                 return 0;
2214         return x->type->reject(x, skb, fl);
2215 }
2216
2217 /* When skb is transformed back to its "native" form, we have to
2218  * check policy restrictions. At the moment we make this in maximally
2219  * stupid way. Shame on me. :-) Of course, connected sockets must
2220  * have policy cached at them.
2221  */
2222
2223 static inline int
2224 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2225               unsigned short family)
2226 {
2227         if (xfrm_state_kern(x))
2228                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2229         return  x->id.proto == tmpl->id.proto &&
2230                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2231                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2232                 x->props.mode == tmpl->mode &&
2233                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2234                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2235                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2236                   xfrm_state_addr_cmp(tmpl, x, family));
2237 }
2238
2239 /*
2240  * 0 or more than 0 is returned when validation is succeeded (either bypass
2241  * because of optional transport mode, or next index of the mathced secpath
2242  * state with the template.
2243  * -1 is returned when no matching template is found.
2244  * Otherwise "-2 - errored_index" is returned.
2245  */
2246 static inline int
2247 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2248                unsigned short family)
2249 {
2250         int idx = start;
2251
2252         if (tmpl->optional) {
2253                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2254                         return start;
2255         } else
2256                 start = -1;
2257         for (; idx < sp->len; idx++) {
2258                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2259                         return ++idx;
2260                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2261                         if (start == -1)
2262                                 start = -2-idx;
2263                         break;
2264                 }
2265         }
2266         return start;
2267 }
2268
2269 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2270                           unsigned int family, int reverse)
2271 {
2272         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2273         int err;
2274
2275         if (unlikely(afinfo == NULL))
2276                 return -EAFNOSUPPORT;
2277
2278         afinfo->decode_session(skb, fl, reverse);
2279         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2280         xfrm_policy_put_afinfo(afinfo);
2281         return err;
2282 }
2283 EXPORT_SYMBOL(__xfrm_decode_session);
2284
2285 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2286 {
2287         for (; k < sp->len; k++) {
2288                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2289                         *idxp = k;
2290                         return 1;
2291                 }
2292         }
2293
2294         return 0;
2295 }
2296
2297 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2298                         unsigned short family)
2299 {
2300         struct net *net = dev_net(skb->dev);
2301         struct xfrm_policy *pol;
2302         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2303         int npols = 0;
2304         int xfrm_nr;
2305         int pi;
2306         int reverse;
2307         struct flowi fl;
2308         u8 fl_dir;
2309         int xerr_idx = -1;
2310
2311         reverse = dir & ~XFRM_POLICY_MASK;
2312         dir &= XFRM_POLICY_MASK;
2313         fl_dir = policy_to_flow_dir(dir);
2314
2315         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2316                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2317                 return 0;
2318         }
2319
2320         nf_nat_decode_session(skb, &fl, family);
2321
2322         /* First, check used SA against their selectors. */
2323         if (skb->sp) {
2324                 int i;
2325
2326                 for (i=skb->sp->len-1; i>=0; i--) {
2327                         struct xfrm_state *x = skb->sp->xvec[i];
2328                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2329                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2330                                 return 0;
2331                         }
2332                 }
2333         }
2334
2335         pol = NULL;
2336         if (sk && sk->sk_policy[dir]) {
2337                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2338                 if (IS_ERR(pol)) {
2339                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2340                         return 0;
2341                 }
2342         }
2343
2344         if (!pol) {
2345                 struct flow_cache_object *flo;
2346
2347                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2348                                         xfrm_policy_lookup, NULL);
2349                 if (IS_ERR_OR_NULL(flo))
2350                         pol = ERR_CAST(flo);
2351                 else
2352                         pol = container_of(flo, struct xfrm_policy, flo);
2353         }
2354
2355         if (IS_ERR(pol)) {
2356                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2357                 return 0;
2358         }
2359
2360         if (!pol) {
2361                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2362                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2363                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2364                         return 0;
2365                 }
2366                 return 1;
2367         }
2368
2369         pol->curlft.use_time = get_seconds();
2370
2371         pols[0] = pol;
2372         npols ++;
2373 #ifdef CONFIG_XFRM_SUB_POLICY
2374         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2375                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2376                                                     &fl, family,
2377                                                     XFRM_POLICY_IN);
2378                 if (pols[1]) {
2379                         if (IS_ERR(pols[1])) {
2380                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2381                                 return 0;
2382                         }
2383                         pols[1]->curlft.use_time = get_seconds();
2384                         npols ++;
2385                 }
2386         }
2387 #endif
2388
2389         if (pol->action == XFRM_POLICY_ALLOW) {
2390                 struct sec_path *sp;
2391                 static struct sec_path dummy;
2392                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2393                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2394                 struct xfrm_tmpl **tpp = tp;
2395                 int ti = 0;
2396                 int i, k;
2397
2398                 if ((sp = skb->sp) == NULL)
2399                         sp = &dummy;
2400
2401                 for (pi = 0; pi < npols; pi++) {
2402                         if (pols[pi] != pol &&
2403                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2404                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2405                                 goto reject;
2406                         }
2407                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2408                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2409                                 goto reject_error;
2410                         }
2411                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2412                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2413                 }
2414                 xfrm_nr = ti;
2415                 if (npols > 1) {
2416                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2417                         tpp = stp;
2418                 }
2419
2420                 /* For each tunnel xfrm, find the first matching tmpl.
2421                  * For each tmpl before that, find corresponding xfrm.
2422                  * Order is _important_. Later we will implement
2423                  * some barriers, but at the moment barriers
2424                  * are implied between each two transformations.
2425                  */
2426                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2427                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2428                         if (k < 0) {
2429                                 if (k < -1)
2430                                         /* "-2 - errored_index" returned */
2431                                         xerr_idx = -(2+k);
2432                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2433                                 goto reject;
2434                         }
2435                 }
2436
2437                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2438                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2439                         goto reject;
2440                 }
2441
2442                 xfrm_pols_put(pols, npols);
2443                 return 1;
2444         }
2445         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2446
2447 reject:
2448         xfrm_secpath_reject(xerr_idx, skb, &fl);
2449 reject_error:
2450         xfrm_pols_put(pols, npols);
2451         return 0;
2452 }
2453 EXPORT_SYMBOL(__xfrm_policy_check);
2454
2455 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2456 {
2457         struct net *net = dev_net(skb->dev);
2458         struct flowi fl;
2459         struct dst_entry *dst;
2460         int res = 1;
2461
2462         if (xfrm_decode_session(skb, &fl, family) < 0) {
2463                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2464                 return 0;
2465         }
2466
2467         skb_dst_force(skb);
2468
2469         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2470         if (IS_ERR(dst)) {
2471                 res = 0;
2472                 dst = NULL;
2473         }
2474         skb_dst_set(skb, dst);
2475         return res;
2476 }
2477 EXPORT_SYMBOL(__xfrm_route_forward);
2478
2479 /* Optimize later using cookies and generation ids. */
2480
2481 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2482 {
2483         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2484          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2485          * get validated by dst_ops->check on every use.  We do this
2486          * because when a normal route referenced by an XFRM dst is
2487          * obsoleted we do not go looking around for all parent
2488          * referencing XFRM dsts so that we can invalidate them.  It
2489          * is just too much work.  Instead we make the checks here on
2490          * every use.  For example:
2491          *
2492          *      XFRM dst A --> IPv4 dst X
2493          *
2494          * X is the "xdst->route" of A (X is also the "dst->path" of A
2495          * in this example).  If X is marked obsolete, "A" will not
2496          * notice.  That's what we are validating here via the
2497          * stale_bundle() check.
2498          *
2499          * When a policy's bundle is pruned, we dst_free() the XFRM
2500          * dst which causes it's ->obsolete field to be set to
2501          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2502          * this, we want to force a new route lookup.
2503          */
2504         if (dst->obsolete < 0 && !stale_bundle(dst))
2505                 return dst;
2506
2507         return NULL;
2508 }
2509
2510 static int stale_bundle(struct dst_entry *dst)
2511 {
2512         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2513 }
2514
2515 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2516 {
2517         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2518                 dst->dev = dev_net(dev)->loopback_dev;
2519                 dev_hold(dst->dev);
2520                 dev_put(dev);
2521         }
2522 }
2523 EXPORT_SYMBOL(xfrm_dst_ifdown);
2524
2525 static void xfrm_link_failure(struct sk_buff *skb)
2526 {
2527         /* Impossible. Such dst must be popped before reaches point of failure. */
2528 }
2529
2530 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2531 {
2532         if (dst) {
2533                 if (dst->obsolete) {
2534                         dst_release(dst);
2535                         dst = NULL;
2536                 }
2537         }
2538         return dst;
2539 }
2540
2541 static void __xfrm_garbage_collect(struct net *net)
2542 {
2543         struct dst_entry *head, *next;
2544
2545         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2546         head = xfrm_policy_sk_bundles;
2547         xfrm_policy_sk_bundles = NULL;
2548         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2549
2550         while (head) {
2551                 next = head->next;
2552                 dst_free(head);
2553                 head = next;
2554         }
2555 }
2556
2557 void xfrm_garbage_collect(struct net *net)
2558 {
2559         flow_cache_flush();
2560         __xfrm_garbage_collect(net);
2561 }
2562 EXPORT_SYMBOL(xfrm_garbage_collect);
2563
2564 static void xfrm_garbage_collect_deferred(struct net *net)
2565 {
2566         flow_cache_flush_deferred();
2567         __xfrm_garbage_collect(net);
2568 }
2569
2570 static void xfrm_init_pmtu(struct dst_entry *dst)
2571 {
2572         do {
2573                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2574                 u32 pmtu, route_mtu_cached;
2575
2576                 pmtu = dst_mtu(dst->child);
2577                 xdst->child_mtu_cached = pmtu;
2578
2579                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2580
2581                 route_mtu_cached = dst_mtu(xdst->route);
2582                 xdst->route_mtu_cached = route_mtu_cached;
2583
2584                 if (pmtu > route_mtu_cached)
2585                         pmtu = route_mtu_cached;
2586
2587                 dst_metric_set(dst, RTAX_MTU, pmtu);
2588         } while ((dst = dst->next));
2589 }
2590
2591 /* Check that the bundle accepts the flow and its components are
2592  * still valid.
2593  */
2594
2595 static int xfrm_bundle_ok(struct xfrm_dst *first)
2596 {
2597         struct dst_entry *dst = &first->u.dst;
2598         struct xfrm_dst *last;
2599         u32 mtu;
2600
2601         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2602             (dst->dev && !netif_running(dst->dev)))
2603                 return 0;
2604
2605         if (dst->flags & DST_XFRM_QUEUE)
2606                 return 1;
2607
2608         last = NULL;
2609
2610         do {
2611                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2612
2613                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2614                         return 0;
2615                 if (xdst->xfrm_genid != dst->xfrm->genid)
2616                         return 0;
2617                 if (xdst->num_pols > 0 &&
2618                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2619                         return 0;
2620
2621                 mtu = dst_mtu(dst->child);
2622                 if (xdst->child_mtu_cached != mtu) {
2623                         last = xdst;
2624                         xdst->child_mtu_cached = mtu;
2625                 }
2626
2627                 if (!dst_check(xdst->route, xdst->route_cookie))
2628                         return 0;
2629                 mtu = dst_mtu(xdst->route);
2630                 if (xdst->route_mtu_cached != mtu) {
2631                         last = xdst;
2632                         xdst->route_mtu_cached = mtu;
2633                 }
2634
2635                 dst = dst->child;
2636         } while (dst->xfrm);
2637
2638         if (likely(!last))
2639                 return 1;
2640
2641         mtu = last->child_mtu_cached;
2642         for (;;) {
2643                 dst = &last->u.dst;
2644
2645                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2646                 if (mtu > last->route_mtu_cached)
2647                         mtu = last->route_mtu_cached;
2648                 dst_metric_set(dst, RTAX_MTU, mtu);
2649
2650                 if (last == first)
2651                         break;
2652
2653                 last = (struct xfrm_dst *)last->u.dst.next;
2654                 last->child_mtu_cached = mtu;
2655         }
2656
2657         return 1;
2658 }
2659
2660 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2661 {
2662         return dst_metric_advmss(dst->path);
2663 }
2664
2665 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2666 {
2667         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2668
2669         return mtu ? : dst_mtu(dst->path);
2670 }
2671
2672 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2673                                            struct sk_buff *skb,
2674                                            const void *daddr)
2675 {
2676         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2677 }
2678
2679 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2680 {
2681         struct net *net;
2682         int err = 0;
2683         if (unlikely(afinfo == NULL))
2684                 return -EINVAL;
2685         if (unlikely(afinfo->family >= NPROTO))
2686                 return -EAFNOSUPPORT;
2687         spin_lock(&xfrm_policy_afinfo_lock);
2688         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2689                 err = -ENOBUFS;
2690         else {
2691                 struct dst_ops *dst_ops = afinfo->dst_ops;
2692                 if (likely(dst_ops->kmem_cachep == NULL))
2693                         dst_ops->kmem_cachep = xfrm_dst_cache;
2694                 if (likely(dst_ops->check == NULL))
2695                         dst_ops->check = xfrm_dst_check;
2696                 if (likely(dst_ops->default_advmss == NULL))
2697                         dst_ops->default_advmss = xfrm_default_advmss;
2698                 if (likely(dst_ops->mtu == NULL))
2699                         dst_ops->mtu = xfrm_mtu;
2700                 if (likely(dst_ops->negative_advice == NULL))
2701                         dst_ops->negative_advice = xfrm_negative_advice;
2702                 if (likely(dst_ops->link_failure == NULL))
2703                         dst_ops->link_failure = xfrm_link_failure;
2704                 if (likely(dst_ops->neigh_lookup == NULL))
2705                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2706                 if (likely(afinfo->garbage_collect == NULL))
2707                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2708                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2709         }
2710         spin_unlock(&xfrm_policy_afinfo_lock);
2711
2712         rtnl_lock();
2713         for_each_net(net) {
2714                 struct dst_ops *xfrm_dst_ops;
2715
2716                 switch (afinfo->family) {
2717                 case AF_INET:
2718                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2719                         break;
2720 #if IS_ENABLED(CONFIG_IPV6)
2721                 case AF_INET6:
2722                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2723                         break;
2724 #endif
2725                 default:
2726                         BUG();
2727                 }
2728                 *xfrm_dst_ops = *afinfo->dst_ops;
2729         }
2730         rtnl_unlock();
2731
2732         return err;
2733 }
2734 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2735
2736 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2737 {
2738         int err = 0;
2739         if (unlikely(afinfo == NULL))
2740                 return -EINVAL;
2741         if (unlikely(afinfo->family >= NPROTO))
2742                 return -EAFNOSUPPORT;
2743         spin_lock(&xfrm_policy_afinfo_lock);
2744         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2745                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2746                         err = -EINVAL;
2747                 else
2748                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2749                                          NULL);
2750         }
2751         spin_unlock(&xfrm_policy_afinfo_lock);
2752         if (!err) {
2753                 struct dst_ops *dst_ops = afinfo->dst_ops;
2754
2755                 synchronize_rcu();
2756
2757                 dst_ops->kmem_cachep = NULL;
2758                 dst_ops->check = NULL;
2759                 dst_ops->negative_advice = NULL;
2760                 dst_ops->link_failure = NULL;
2761                 afinfo->garbage_collect = NULL;
2762         }
2763         return err;
2764 }
2765 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2766
2767 static void __net_init xfrm_dst_ops_init(struct net *net)
2768 {
2769         struct xfrm_policy_afinfo *afinfo;
2770
2771         rcu_read_lock();
2772         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2773         if (afinfo)
2774                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2775 #if IS_ENABLED(CONFIG_IPV6)
2776         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2777         if (afinfo)
2778                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2779 #endif
2780         rcu_read_unlock();
2781 }
2782
2783 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2784 {
2785         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2786
2787         switch (event) {
2788         case NETDEV_DOWN:
2789                 xfrm_garbage_collect(dev_net(dev));
2790         }
2791         return NOTIFY_DONE;
2792 }
2793
2794 static struct notifier_block xfrm_dev_notifier = {
2795         .notifier_call  = xfrm_dev_event,
2796 };
2797
2798 #ifdef CONFIG_XFRM_STATISTICS
2799 static int __net_init xfrm_statistics_init(struct net *net)
2800 {
2801         int rv;
2802
2803         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2804                           sizeof(struct linux_xfrm_mib),
2805                           __alignof__(struct linux_xfrm_mib)) < 0)
2806                 return -ENOMEM;
2807         rv = xfrm_proc_init(net);
2808         if (rv < 0)
2809                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2810         return rv;
2811 }
2812
2813 static void xfrm_statistics_fini(struct net *net)
2814 {
2815         xfrm_proc_fini(net);
2816         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2817 }
2818 #else
2819 static int __net_init xfrm_statistics_init(struct net *net)
2820 {
2821         return 0;
2822 }
2823
2824 static void xfrm_statistics_fini(struct net *net)
2825 {
2826 }
2827 #endif
2828
2829 static int __net_init xfrm_policy_init(struct net *net)
2830 {
2831         unsigned int hmask, sz;
2832         int dir;
2833
2834         if (net_eq(net, &init_net))
2835                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2836                                            sizeof(struct xfrm_dst),
2837                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2838                                            NULL);
2839
2840         hmask = 8 - 1;
2841         sz = (hmask+1) * sizeof(struct hlist_head);
2842
2843         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2844         if (!net->xfrm.policy_byidx)
2845                 goto out_byidx;
2846         net->xfrm.policy_idx_hmask = hmask;
2847
2848         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2849                 struct xfrm_policy_hash *htab;
2850
2851                 net->xfrm.policy_count[dir] = 0;
2852                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2853
2854                 htab = &net->xfrm.policy_bydst[dir];
2855                 htab->table = xfrm_hash_alloc(sz);
2856                 if (!htab->table)
2857                         goto out_bydst;
2858                 htab->hmask = hmask;
2859         }
2860
2861         INIT_LIST_HEAD(&net->xfrm.policy_all);
2862         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2863         if (net_eq(net, &init_net))
2864                 register_netdevice_notifier(&xfrm_dev_notifier);
2865         return 0;
2866
2867 out_bydst:
2868         for (dir--; dir >= 0; dir--) {
2869                 struct xfrm_policy_hash *htab;
2870
2871                 htab = &net->xfrm.policy_bydst[dir];
2872                 xfrm_hash_free(htab->table, sz);
2873         }
2874         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2875 out_byidx:
2876         return -ENOMEM;
2877 }
2878
2879 static void xfrm_policy_fini(struct net *net)
2880 {
2881         struct xfrm_audit audit_info;
2882         unsigned int sz;
2883         int dir;
2884
2885         flush_work(&net->xfrm.policy_hash_work);
2886 #ifdef CONFIG_XFRM_SUB_POLICY
2887         audit_info.loginuid = INVALID_UID;
2888         audit_info.sessionid = -1;
2889         audit_info.secid = 0;
2890         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2891 #endif
2892         audit_info.loginuid = INVALID_UID;
2893         audit_info.sessionid = -1;
2894         audit_info.secid = 0;
2895         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2896
2897         WARN_ON(!list_empty(&net->xfrm.policy_all));
2898
2899         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2900                 struct xfrm_policy_hash *htab;
2901
2902                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2903
2904                 htab = &net->xfrm.policy_bydst[dir];
2905                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2906                 WARN_ON(!hlist_empty(htab->table));
2907                 xfrm_hash_free(htab->table, sz);
2908         }
2909
2910         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2911         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2912         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2913 }
2914
2915 static int __net_init xfrm_net_init(struct net *net)
2916 {
2917         int rv;
2918
2919         rv = xfrm_statistics_init(net);
2920         if (rv < 0)
2921                 goto out_statistics;
2922         rv = xfrm_state_init(net);
2923         if (rv < 0)
2924                 goto out_state;
2925         rv = xfrm_policy_init(net);
2926         if (rv < 0)
2927                 goto out_policy;
2928         xfrm_dst_ops_init(net);
2929         rv = xfrm_sysctl_init(net);
2930         if (rv < 0)
2931                 goto out_sysctl;
2932         return 0;
2933
2934 out_sysctl:
2935         xfrm_policy_fini(net);
2936 out_policy:
2937         xfrm_state_fini(net);
2938 out_state:
2939         xfrm_statistics_fini(net);
2940 out_statistics:
2941         return rv;
2942 }
2943
2944 static void __net_exit xfrm_net_exit(struct net *net)
2945 {
2946         xfrm_sysctl_fini(net);
2947         xfrm_policy_fini(net);
2948         xfrm_state_fini(net);
2949         xfrm_statistics_fini(net);
2950 }
2951
2952 static struct pernet_operations __net_initdata xfrm_net_ops = {
2953         .init = xfrm_net_init,
2954         .exit = xfrm_net_exit,
2955 };
2956
2957 void __init xfrm_init(void)
2958 {
2959         register_pernet_subsys(&xfrm_net_ops);
2960         xfrm_input_init();
2961 }
2962
2963 #ifdef CONFIG_AUDITSYSCALL
2964 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2965                                          struct audit_buffer *audit_buf)
2966 {
2967         struct xfrm_sec_ctx *ctx = xp->security;
2968         struct xfrm_selector *sel = &xp->selector;
2969
2970         if (ctx)
2971                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2972                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2973
2974         switch(sel->family) {
2975         case AF_INET:
2976                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2977                 if (sel->prefixlen_s != 32)
2978                         audit_log_format(audit_buf, " src_prefixlen=%d",
2979                                          sel->prefixlen_s);
2980                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2981                 if (sel->prefixlen_d != 32)
2982                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2983                                          sel->prefixlen_d);
2984                 break;
2985         case AF_INET6:
2986                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2987                 if (sel->prefixlen_s != 128)
2988                         audit_log_format(audit_buf, " src_prefixlen=%d",
2989                                          sel->prefixlen_s);
2990                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2991                 if (sel->prefixlen_d != 128)
2992                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2993                                          sel->prefixlen_d);
2994                 break;
2995         }
2996 }
2997
2998 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2999                            kuid_t auid, u32 sessionid, u32 secid)
3000 {
3001         struct audit_buffer *audit_buf;
3002
3003         audit_buf = xfrm_audit_start("SPD-add");
3004         if (audit_buf == NULL)
3005                 return;
3006         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3007         audit_log_format(audit_buf, " res=%u", result);
3008         xfrm_audit_common_policyinfo(xp, audit_buf);
3009         audit_log_end(audit_buf);
3010 }
3011 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3012
3013 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3014                               kuid_t auid, u32 sessionid, u32 secid)
3015 {
3016         struct audit_buffer *audit_buf;
3017
3018         audit_buf = xfrm_audit_start("SPD-delete");
3019         if (audit_buf == NULL)
3020                 return;
3021         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3022         audit_log_format(audit_buf, " res=%u", result);
3023         xfrm_audit_common_policyinfo(xp, audit_buf);
3024         audit_log_end(audit_buf);
3025 }
3026 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3027 #endif
3028
3029 #ifdef CONFIG_XFRM_MIGRATE
3030 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3031                                         const struct xfrm_selector *sel_tgt)
3032 {
3033         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3034                 if (sel_tgt->family == sel_cmp->family &&
3035                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3036                                     sel_cmp->family) &&
3037                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3038                                     sel_cmp->family) &&
3039                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3040                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3041                         return true;
3042                 }
3043         } else {
3044                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3045                         return true;
3046                 }
3047         }
3048         return false;
3049 }
3050
3051 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3052                                                      u8 dir, u8 type)
3053 {
3054         struct xfrm_policy *pol, *ret = NULL;
3055         struct hlist_head *chain;
3056         u32 priority = ~0U;
3057
3058         read_lock_bh(&xfrm_policy_lock);
3059         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
3060         hlist_for_each_entry(pol, chain, bydst) {
3061                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3062                     pol->type == type) {
3063                         ret = pol;
3064                         priority = ret->priority;
3065                         break;
3066                 }
3067         }
3068         chain = &init_net.xfrm.policy_inexact[dir];
3069         hlist_for_each_entry(pol, chain, bydst) {
3070                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3071                     pol->type == type &&
3072                     pol->priority < priority) {
3073                         ret = pol;
3074                         break;
3075                 }
3076         }
3077
3078         if (ret)
3079                 xfrm_pol_hold(ret);
3080
3081         read_unlock_bh(&xfrm_policy_lock);
3082
3083         return ret;
3084 }
3085
3086 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3087 {
3088         int match = 0;
3089
3090         if (t->mode == m->mode && t->id.proto == m->proto &&
3091             (m->reqid == 0 || t->reqid == m->reqid)) {
3092                 switch (t->mode) {
3093                 case XFRM_MODE_TUNNEL:
3094                 case XFRM_MODE_BEET:
3095                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3096                                             m->old_family) &&
3097                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3098                                             m->old_family)) {
3099                                 match = 1;
3100                         }
3101                         break;
3102                 case XFRM_MODE_TRANSPORT:
3103                         /* in case of transport mode, template does not store
3104                            any IP addresses, hence we just compare mode and
3105                            protocol */
3106                         match = 1;
3107                         break;
3108                 default:
3109                         break;
3110                 }
3111         }
3112         return match;
3113 }
3114
3115 /* update endpoint address(es) of template(s) */
3116 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3117                                struct xfrm_migrate *m, int num_migrate)
3118 {
3119         struct xfrm_migrate *mp;
3120         int i, j, n = 0;
3121
3122         write_lock_bh(&pol->lock);
3123         if (unlikely(pol->walk.dead)) {
3124                 /* target policy has been deleted */
3125                 write_unlock_bh(&pol->lock);
3126                 return -ENOENT;
3127         }
3128
3129         for (i = 0; i < pol->xfrm_nr; i++) {
3130                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3131                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3132                                 continue;
3133                         n++;
3134                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3135                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3136                                 continue;
3137                         /* update endpoints */
3138                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3139                                sizeof(pol->xfrm_vec[i].id.daddr));
3140                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3141                                sizeof(pol->xfrm_vec[i].saddr));
3142                         pol->xfrm_vec[i].encap_family = mp->new_family;
3143                         /* flush bundles */
3144                         atomic_inc(&pol->genid);
3145                 }
3146         }
3147
3148         write_unlock_bh(&pol->lock);
3149
3150         if (!n)
3151                 return -ENODATA;
3152
3153         return 0;
3154 }
3155
3156 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3157 {
3158         int i, j;
3159
3160         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3161                 return -EINVAL;
3162
3163         for (i = 0; i < num_migrate; i++) {
3164                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3165                                     m[i].old_family) &&
3166                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3167                                     m[i].old_family))
3168                         return -EINVAL;
3169                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3170                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3171                         return -EINVAL;
3172
3173                 /* check if there is any duplicated entry */
3174                 for (j = i + 1; j < num_migrate; j++) {
3175                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3176                                     sizeof(m[i].old_daddr)) &&
3177                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3178                                     sizeof(m[i].old_saddr)) &&
3179                             m[i].proto == m[j].proto &&
3180                             m[i].mode == m[j].mode &&
3181                             m[i].reqid == m[j].reqid &&
3182                             m[i].old_family == m[j].old_family)
3183                                 return -EINVAL;
3184                 }
3185         }
3186
3187         return 0;
3188 }
3189
3190 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3191                  struct xfrm_migrate *m, int num_migrate,
3192                  struct xfrm_kmaddress *k)
3193 {
3194         int i, err, nx_cur = 0, nx_new = 0;
3195         struct xfrm_policy *pol = NULL;
3196         struct xfrm_state *x, *xc;
3197         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3198         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3199         struct xfrm_migrate *mp;
3200
3201         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3202                 goto out;
3203
3204         /* Stage 1 - find policy */
3205         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
3206                 err = -ENOENT;
3207                 goto out;
3208         }
3209
3210         /* Stage 2 - find and update state(s) */
3211         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3212                 if ((x = xfrm_migrate_state_find(mp))) {
3213                         x_cur[nx_cur] = x;
3214                         nx_cur++;
3215                         if ((xc = xfrm_state_migrate(x, mp))) {
3216                                 x_new[nx_new] = xc;
3217                                 nx_new++;
3218                         } else {
3219                                 err = -ENODATA;
3220                                 goto restore_state;
3221                         }
3222                 }
3223         }
3224
3225         /* Stage 3 - update policy */
3226         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3227                 goto restore_state;
3228
3229         /* Stage 4 - delete old state(s) */
3230         if (nx_cur) {
3231                 xfrm_states_put(x_cur, nx_cur);
3232                 xfrm_states_delete(x_cur, nx_cur);
3233         }
3234
3235         /* Stage 5 - announce */
3236         km_migrate(sel, dir, type, m, num_migrate, k);
3237
3238         xfrm_pol_put(pol);
3239
3240         return 0;
3241 out:
3242         return err;
3243
3244 restore_state:
3245         if (pol)
3246                 xfrm_pol_put(pol);
3247         if (nx_cur)
3248                 xfrm_states_put(x_cur, nx_cur);
3249         if (nx_new)
3250                 xfrm_states_delete(x_new, nx_new);
3251
3252         return err;
3253 }
3254 EXPORT_SYMBOL(xfrm_migrate);
3255 #endif