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 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  43 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
  44                                                 __read_mostly;
  45
  46 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  47
  48 static void xfrm_init_pmtu(struct dst_entry *dst);
  49 static int stale_bundle(struct dst_entry *dst);
  50 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  51 static void xfrm_policy_queue_process(unsigned long arg);
  52
  53 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  54                                                 int dir);
  55
  56 static inline bool
  57 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  58 {
  59         const struct flowi4 *fl4 = &fl->u.ip4;
  60
  61         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  62                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  63                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  64                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  65                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  66                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  67 }
  68
  69 static inline bool
  70 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  71 {
  72         const struct flowi6 *fl6 = &fl->u.ip6;
  73
  74         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  75                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  76                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  77                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  78                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  79                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  80 }
  81
  82 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  83                          unsigned short family)
  84 {
  85         switch (family) {
  86         case AF_INET:
  87                 return __xfrm4_selector_match(sel, fl);
  88         case AF_INET6:
  89                 return __xfrm6_selector_match(sel, fl);
  90         }
  91         return false;
  92 }
  93
  94 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
  95 {
  96         struct xfrm_policy_afinfo *afinfo;
  97
  98         if (unlikely(family >= NPROTO))
  99                 return NULL;
 100         rcu_read_lock();
 101         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 102         if (unlikely(!afinfo))
 103                 rcu_read_unlock();
 104         return afinfo;
 105 }
 106
 107 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
 108 {
 109         rcu_read_unlock();
 110 }
 111
 112 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
 113                                                   const xfrm_address_t *saddr,
 114                                                   const xfrm_address_t *daddr,
 115                                                   int family)
 116 {
 117         struct xfrm_policy_afinfo *afinfo;
 118         struct dst_entry *dst;
 119
 120         afinfo = xfrm_policy_get_afinfo(family);
 121         if (unlikely(afinfo == NULL))
 122                 return ERR_PTR(-EAFNOSUPPORT);
 123
 124         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 125
 126         xfrm_policy_put_afinfo(afinfo);
 127
 128         return dst;
 129 }
 130
 131 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 132                                                 xfrm_address_t *prev_saddr,
 133                                                 xfrm_address_t *prev_daddr,
 134                                                 int family)
 135 {
 136         struct net *net = xs_net(x);
 137         xfrm_address_t *saddr = &x->props.saddr;
 138         xfrm_address_t *daddr = &x->id.daddr;
 139         struct dst_entry *dst;
 140
 141         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 142                 saddr = x->coaddr;
 143                 daddr = prev_daddr;
 144         }
 145         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 146                 saddr = prev_saddr;
 147                 daddr = x->coaddr;
 148         }
 149
 150         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 151
 152         if (!IS_ERR(dst)) {
 153                 if (prev_saddr != saddr)
 154                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 155                 if (prev_daddr != daddr)
 156                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 157         }
 158
 159         return dst;
 160 }
 161
 162 static inline unsigned long make_jiffies(long secs)
 163 {
 164         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 165                 return MAX_SCHEDULE_TIMEOUT-1;
 166         else
 167                 return secs*HZ;
 168 }
 169
 170 static void xfrm_policy_timer(unsigned long data)
 171 {
 172         struct xfrm_policy *xp = (struct xfrm_policy *)data;
 173         unsigned long now = get_seconds();
 174         long next = LONG_MAX;
 175         int warn = 0;
 176         int dir;
 177
 178         read_lock(&xp->lock);
 179
 180         if (unlikely(xp->walk.dead))
 181                 goto out;
 182
 183         dir = xfrm_policy_id2dir(xp->index);
 184
 185         if (xp->lft.hard_add_expires_seconds) {
 186                 long tmo = xp->lft.hard_add_expires_seconds +
 187                         xp->curlft.add_time - now;
 188                 if (tmo <= 0)
 189                         goto expired;
 190                 if (tmo < next)
 191                         next = tmo;
 192         }
 193         if (xp->lft.hard_use_expires_seconds) {
 194                 long tmo = xp->lft.hard_use_expires_seconds +
 195                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 196                 if (tmo <= 0)
 197                         goto expired;
 198                 if (tmo < next)
 199                         next = tmo;
 200         }
 201         if (xp->lft.soft_add_expires_seconds) {
 202                 long tmo = xp->lft.soft_add_expires_seconds +
 203                         xp->curlft.add_time - now;
 204                 if (tmo <= 0) {
 205                         warn = 1;
 206                         tmo = XFRM_KM_TIMEOUT;
 207                 }
 208                 if (tmo < next)
 209                         next = tmo;
 210         }
 211         if (xp->lft.soft_use_expires_seconds) {
 212                 long tmo = xp->lft.soft_use_expires_seconds +
 213                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 214                 if (tmo <= 0) {
 215                         warn = 1;
 216                         tmo = XFRM_KM_TIMEOUT;
 217                 }
 218                 if (tmo < next)
 219                         next = tmo;
 220         }
 221
 222         if (warn)
 223                 km_policy_expired(xp, dir, 0, 0);
 224         if (next != LONG_MAX &&
 225             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 226                 xfrm_pol_hold(xp);
 227
 228 out:
 229         read_unlock(&xp->lock);
 230         xfrm_pol_put(xp);
 231         return;
 232
 233 expired:
 234         read_unlock(&xp->lock);
 235         if (!xfrm_policy_delete(xp, dir))
 236                 km_policy_expired(xp, dir, 1, 0);
 237         xfrm_pol_put(xp);
 238 }
 239
 240 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 241 {
 242         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 243
 244         if (unlikely(pol->walk.dead))
 245                 flo = NULL;
 246         else
 247                 xfrm_pol_hold(pol);
 248
 249         return flo;
 250 }
 251
 252 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 253 {
 254         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 255
 256         return !pol->walk.dead;
 257 }
 258
 259 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 260 {
 261         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 262 }
 263
 264 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 265         .get = xfrm_policy_flo_get,
 266         .check = xfrm_policy_flo_check,
 267         .delete = xfrm_policy_flo_delete,
 268 };
 269
 270 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 271  * SPD calls.
 272  */
 273
 274 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 275 {
 276         struct xfrm_policy *policy;
 277
 278         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 279
 280         if (policy) {
 281                 write_pnet(&policy->xp_net, net);
 282                 INIT_LIST_HEAD(&policy->walk.all);
 283                 INIT_HLIST_NODE(&policy->bydst);
 284                 INIT_HLIST_NODE(&policy->byidx);
 285                 rwlock_init(&policy->lock);
 286                 atomic_set(&policy->refcnt, 1);
 287                 skb_queue_head_init(&policy->polq.hold_queue);
 288                 setup_timer(&policy->timer, xfrm_policy_timer,
 289                                 (unsigned long)policy);
 290                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 291                             (unsigned long)policy);
 292                 policy->flo.ops = &xfrm_policy_fc_ops;
 293         }
 294         return policy;
 295 }
 296 EXPORT_SYMBOL(xfrm_policy_alloc);
 297
 298 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 299
 300 void xfrm_policy_destroy(struct xfrm_policy *policy)
 301 {
 302         BUG_ON(!policy->walk.dead);
 303
 304         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 305                 BUG();
 306
 307         security_xfrm_policy_free(policy->security);
 308         kfree(policy);
 309 }
 310 EXPORT_SYMBOL(xfrm_policy_destroy);
 311
 312 static void xfrm_queue_purge(struct sk_buff_head *list)
 313 {
 314         struct sk_buff *skb;
 315
 316         while ((skb = skb_dequeue(list)) != NULL)
 317                 kfree_skb(skb);
 318 }
 319
 320 /* Rule must be locked. Release descentant resources, announce
 321  * entry dead. The rule must be unlinked from lists to the moment.
 322  */
 323
 324 static void xfrm_policy_kill(struct xfrm_policy *policy)
 325 {
 326         policy->walk.dead = 1;
 327
 328         atomic_inc(&policy->genid);
 329
 330         if (del_timer(&policy->polq.hold_timer))
 331                 xfrm_pol_put(policy);
 332         xfrm_queue_purge(&policy->polq.hold_queue);
 333
 334         if (del_timer(&policy->timer))
 335                 xfrm_pol_put(policy);
 336
 337         xfrm_pol_put(policy);
 338 }
 339
 340 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 341
 342 static inline unsigned int idx_hash(struct net *net, u32 index)
 343 {
 344         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 345 }
 346
 347 static struct hlist_head *policy_hash_bysel(struct net *net,
 348                                             const struct xfrm_selector *sel,
 349                                             unsigned short family, int dir)
 350 {
 351         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 352         unsigned int hash = __sel_hash(sel, family, hmask);
 353
 354         return (hash == hmask + 1 ?
 355                 &net->xfrm.policy_inexact[dir] :
 356                 net->xfrm.policy_bydst[dir].table + hash);
 357 }
 358
 359 static struct hlist_head *policy_hash_direct(struct net *net,
 360                                              const xfrm_address_t *daddr,
 361                                              const xfrm_address_t *saddr,
 362                                              unsigned short family, int dir)
 363 {
 364         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 365         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 366
 367         return net->xfrm.policy_bydst[dir].table + hash;
 368 }
 369
 370 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 371                                    struct hlist_head *ndsttable,
 372                                    unsigned int nhashmask)
 373 {
 374         struct hlist_node *tmp, *entry0 = NULL;
 375         struct xfrm_policy *pol;
 376         unsigned int h0 = 0;
 377
 378 redo:
 379         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 380                 unsigned int h;
 381
 382                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 383                                 pol->family, nhashmask);
 384                 if (!entry0) {
 385                         hlist_del(&pol->bydst);
 386                         hlist_add_head(&pol->bydst, ndsttable+h);
 387                         h0 = h;
 388                 } else {
 389                         if (h != h0)
 390                                 continue;
 391                         hlist_del(&pol->bydst);
 392                         hlist_add_after(entry0, &pol->bydst);
 393                 }
 394                 entry0 = &pol->bydst;
 395         }
 396         if (!hlist_empty(list)) {
 397                 entry0 = NULL;
 398                 goto redo;
 399         }
 400 }
 401
 402 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 403                                    struct hlist_head *nidxtable,
 404                                    unsigned int nhashmask)
 405 {
 406         struct hlist_node *tmp;
 407         struct xfrm_policy *pol;
 408
 409         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 410                 unsigned int h;
 411
 412                 h = __idx_hash(pol->index, nhashmask);
 413                 hlist_add_head(&pol->byidx, nidxtable+h);
 414         }
 415 }
 416
 417 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 418 {
 419         return ((old_hmask + 1) << 1) - 1;
 420 }
 421
 422 static void xfrm_bydst_resize(struct net *net, int dir)
 423 {
 424         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 425         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 426         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 427         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 428         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 429         int i;
 430
 431         if (!ndst)
 432                 return;
 433
 434         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 435
 436         for (i = hmask; i >= 0; i--)
 437                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 438
 439         net->xfrm.policy_bydst[dir].table = ndst;
 440         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 441
 442         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 443
 444         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 445 }
 446
 447 static void xfrm_byidx_resize(struct net *net, int total)
 448 {
 449         unsigned int hmask = net->xfrm.policy_idx_hmask;
 450         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 451         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 452         struct hlist_head *oidx = net->xfrm.policy_byidx;
 453         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 454         int i;
 455
 456         if (!nidx)
 457                 return;
 458
 459         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 460
 461         for (i = hmask; i >= 0; i--)
 462                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 463
 464         net->xfrm.policy_byidx = nidx;
 465         net->xfrm.policy_idx_hmask = nhashmask;
 466
 467         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 468
 469         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 470 }
 471
 472 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 473 {
 474         unsigned int cnt = net->xfrm.policy_count[dir];
 475         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 476
 477         if (total)
 478                 *total += cnt;
 479
 480         if ((hmask + 1) < xfrm_policy_hashmax &&
 481             cnt > hmask)
 482                 return 1;
 483
 484         return 0;
 485 }
 486
 487 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 488 {
 489         unsigned int hmask = net->xfrm.policy_idx_hmask;
 490
 491         if ((hmask + 1) < xfrm_policy_hashmax &&
 492             total > hmask)
 493                 return 1;
 494
 495         return 0;
 496 }
 497
 498 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 499 {
 500         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 501         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 502         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 503         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 504         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 505         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 506         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 507         si->spdhcnt = net->xfrm.policy_idx_hmask;
 508         si->spdhmcnt = xfrm_policy_hashmax;
 509         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
 510 }
 511 EXPORT_SYMBOL(xfrm_spd_getinfo);
 512
 513 static DEFINE_MUTEX(hash_resize_mutex);
 514 static void xfrm_hash_resize(struct work_struct *work)
 515 {
 516         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 517         int dir, total;
 518
 519         mutex_lock(&hash_resize_mutex);
 520
 521         total = 0;
 522         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 523                 if (xfrm_bydst_should_resize(net, dir, &total))
 524                         xfrm_bydst_resize(net, dir);
 525         }
 526         if (xfrm_byidx_should_resize(net, total))
 527                 xfrm_byidx_resize(net, total);
 528
 529         mutex_unlock(&hash_resize_mutex);
 530 }
 531
 532 /* Generate new index... KAME seems to generate them ordered by cost
 533  * of an absolute inpredictability of ordering of rules. This will not pass. */
 534 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 535 {
 536         static u32 idx_generator;
 537
 538         for (;;) {
 539                 struct hlist_head *list;
 540                 struct xfrm_policy *p;
 541                 u32 idx;
 542                 int found;
 543
 544                 if (!index) {
 545                         idx = (idx_generator | dir);
 546                         idx_generator += 8;
 547                 } else {
 548                         idx = index;
 549                         index = 0;
 550                 }
 551
 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         if (del_timer(&pq->hold_timer))
 593                 xfrm_pol_put(old);
 594         spin_unlock_bh(&pq->hold_queue.lock);
 595
 596         if (skb_queue_empty(&list))
 597                 return;
 598
 599         pq = &new->polq;
 600
 601         spin_lock_bh(&pq->hold_queue.lock);
 602         skb_queue_splice(&list, &pq->hold_queue);
 603         pq->timeout = XFRM_QUEUE_TMO_MIN;
 604         if (!mod_timer(&pq->hold_timer, jiffies))
 605                 xfrm_pol_hold(new);
 606         spin_unlock_bh(&pq->hold_queue.lock);
 607 }
 608
 609 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 610                                    struct xfrm_policy *pol)
 611 {
 612         u32 mark = policy->mark.v & policy->mark.m;
 613
 614         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 615                 return true;
 616
 617         if ((mark & pol->mark.m) == pol->mark.v &&
 618             policy->priority == pol->priority)
 619                 return true;
 620
 621         return false;
 622 }
 623
 624 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 625 {
 626         struct net *net = xp_net(policy);
 627         struct xfrm_policy *pol;
 628         struct xfrm_policy *delpol;
 629         struct hlist_head *chain;
 630         struct hlist_node *newpos;
 631
 632         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 633         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 634         delpol = NULL;
 635         newpos = NULL;
 636         hlist_for_each_entry(pol, chain, bydst) {
 637                 if (pol->type == policy->type &&
 638                     !selector_cmp(&pol->selector, &policy->selector) &&
 639                     xfrm_policy_mark_match(policy, pol) &&
 640                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 641                     !WARN_ON(delpol)) {
 642                         if (excl) {
 643                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 644                                 return -EEXIST;
 645                         }
 646                         delpol = pol;
 647                         if (policy->priority > pol->priority)
 648                                 continue;
 649                 } else if (policy->priority >= pol->priority) {
 650                         newpos = &pol->bydst;
 651                         continue;
 652                 }
 653                 if (delpol)
 654                         break;
 655         }
 656         if (newpos)
 657                 hlist_add_after(newpos, &policy->bydst);
 658         else
 659                 hlist_add_head(&policy->bydst, chain);
 660         xfrm_pol_hold(policy);
 661         net->xfrm.policy_count[dir]++;
 662         atomic_inc(&net->xfrm.flow_cache_genid);
 663
 664         /* After previous checking, family can either be AF_INET or AF_INET6 */
 665         if (policy->family == AF_INET)
 666                 rt_genid_bump_ipv4(net);
 667         else
 668                 rt_genid_bump_ipv6(net);
 669
 670         if (delpol) {
 671                 xfrm_policy_requeue(delpol, policy);
 672                 __xfrm_policy_unlink(delpol, dir);
 673         }
 674         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 675         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 676         policy->curlft.add_time = get_seconds();
 677         policy->curlft.use_time = 0;
 678         if (!mod_timer(&policy->timer, jiffies + HZ))
 679                 xfrm_pol_hold(policy);
 680         list_add(&policy->walk.all, &net->xfrm.policy_all);
 681         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 682
 683         if (delpol)
 684                 xfrm_policy_kill(delpol);
 685         else if (xfrm_bydst_should_resize(net, dir, NULL))
 686                 schedule_work(&net->xfrm.policy_hash_work);
 687
 688         return 0;
 689 }
 690 EXPORT_SYMBOL(xfrm_policy_insert);
 691
 692 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 693                                           int dir, struct xfrm_selector *sel,
 694                                           struct xfrm_sec_ctx *ctx, int delete,
 695                                           int *err)
 696 {
 697         struct xfrm_policy *pol, *ret;
 698         struct hlist_head *chain;
 699
 700         *err = 0;
 701         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 702         chain = policy_hash_bysel(net, sel, sel->family, dir);
 703         ret = NULL;
 704         hlist_for_each_entry(pol, chain, bydst) {
 705                 if (pol->type == type &&
 706                     (mark & pol->mark.m) == pol->mark.v &&
 707                     !selector_cmp(sel, &pol->selector) &&
 708                     xfrm_sec_ctx_match(ctx, pol->security)) {
 709                         xfrm_pol_hold(pol);
 710                         if (delete) {
 711                                 *err = security_xfrm_policy_delete(
 712                                                                 pol->security);
 713                                 if (*err) {
 714                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 715                                         return pol;
 716                                 }
 717                                 __xfrm_policy_unlink(pol, dir);
 718                         }
 719                         ret = pol;
 720                         break;
 721                 }
 722         }
 723         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 724
 725         if (ret && delete)
 726                 xfrm_policy_kill(ret);
 727         return ret;
 728 }
 729 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 730
 731 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 732                                      int dir, u32 id, int delete, int *err)
 733 {
 734         struct xfrm_policy *pol, *ret;
 735         struct hlist_head *chain;
 736
 737         *err = -ENOENT;
 738         if (xfrm_policy_id2dir(id) != dir)
 739                 return NULL;
 740
 741         *err = 0;
 742         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 743         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 744         ret = NULL;
 745         hlist_for_each_entry(pol, chain, byidx) {
 746                 if (pol->type == type && pol->index == id &&
 747                     (mark & pol->mark.m) == pol->mark.v) {
 748                         xfrm_pol_hold(pol);
 749                         if (delete) {
 750                                 *err = security_xfrm_policy_delete(
 751                                                                 pol->security);
 752                                 if (*err) {
 753                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 754                                         return pol;
 755                                 }
 756                                 __xfrm_policy_unlink(pol, dir);
 757                         }
 758                         ret = pol;
 759                         break;
 760                 }
 761         }
 762         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 763
 764         if (ret && delete)
 765                 xfrm_policy_kill(ret);
 766         return ret;
 767 }
 768 EXPORT_SYMBOL(xfrm_policy_byid);
 769
 770 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 771 static inline int
 772 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 773 {
 774         int dir, err = 0;
 775
 776         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 777                 struct xfrm_policy *pol;
 778                 int i;
 779
 780                 hlist_for_each_entry(pol,
 781                                      &net->xfrm.policy_inexact[dir], bydst) {
 782                         if (pol->type != type)
 783                                 continue;
 784                         err = security_xfrm_policy_delete(pol->security);
 785                         if (err) {
 786                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 787                                 return err;
 788                         }
 789                 }
 790                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 791                         hlist_for_each_entry(pol,
 792                                              net->xfrm.policy_bydst[dir].table + i,
 793                                              bydst) {
 794                                 if (pol->type != type)
 795                                         continue;
 796                                 err = security_xfrm_policy_delete(
 797                                                                 pol->security);
 798                                 if (err) {
 799                                         xfrm_audit_policy_delete(pol, 0,
 800                                                                  task_valid);
 801                                         return err;
 802                                 }
 803                         }
 804                 }
 805         }
 806         return err;
 807 }
 808 #else
 809 static inline int
 810 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 811 {
 812         return 0;
 813 }
 814 #endif
 815
 816 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 817 {
 818         int dir, err = 0, cnt = 0;
 819
 820         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 821
 822         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 823         if (err)
 824                 goto out;
 825
 826         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 827                 struct xfrm_policy *pol;
 828                 int i;
 829
 830         again1:
 831                 hlist_for_each_entry(pol,
 832                                      &net->xfrm.policy_inexact[dir], bydst) {
 833                         if (pol->type != type)
 834                                 continue;
 835                         __xfrm_policy_unlink(pol, dir);
 836                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 837                         cnt++;
 838
 839                         xfrm_audit_policy_delete(pol, 1, task_valid);
 840
 841                         xfrm_policy_kill(pol);
 842
 843                         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 844                         goto again1;
 845                 }
 846
 847                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 848         again2:
 849                         hlist_for_each_entry(pol,
 850                                              net->xfrm.policy_bydst[dir].table + i,
 851                                              bydst) {
 852                                 if (pol->type != type)
 853                                         continue;
 854                                 __xfrm_policy_unlink(pol, dir);
 855                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 856                                 cnt++;
 857
 858                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 859                                 xfrm_policy_kill(pol);
 860
 861                                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
 862                                 goto again2;
 863                         }
 864                 }
 865
 866         }
 867         if (!cnt)
 868                 err = -ESRCH;
 869 out:
 870         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 871         return err;
 872 }
 873 EXPORT_SYMBOL(xfrm_policy_flush);
 874
 875 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 876                      int (*func)(struct xfrm_policy *, int, int, void*),
 877                      void *data)
 878 {
 879         struct xfrm_policy *pol;
 880         struct xfrm_policy_walk_entry *x;
 881         int error = 0;
 882
 883         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 884             walk->type != XFRM_POLICY_TYPE_ANY)
 885                 return -EINVAL;
 886
 887         if (list_empty(&walk->walk.all) && walk->seq != 0)
 888                 return 0;
 889
 890         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 891         if (list_empty(&walk->walk.all))
 892                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 893         else
 894                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 895         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 896                 if (x->dead)
 897                         continue;
 898                 pol = container_of(x, struct xfrm_policy, walk);
 899                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 900                     walk->type != pol->type)
 901                         continue;
 902                 error = func(pol, xfrm_policy_id2dir(pol->index),
 903                              walk->seq, data);
 904                 if (error) {
 905                         list_move_tail(&walk->walk.all, &x->all);
 906                         goto out;
 907                 }
 908                 walk->seq++;
 909         }
 910         if (walk->seq == 0) {
 911                 error = -ENOENT;
 912                 goto out;
 913         }
 914         list_del_init(&walk->walk.all);
 915 out:
 916         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 917         return error;
 918 }
 919 EXPORT_SYMBOL(xfrm_policy_walk);
 920
 921 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 922 {
 923         INIT_LIST_HEAD(&walk->walk.all);
 924         walk->walk.dead = 1;
 925         walk->type = type;
 926         walk->seq = 0;
 927 }
 928 EXPORT_SYMBOL(xfrm_policy_walk_init);
 929
 930 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
 931 {
 932         if (list_empty(&walk->walk.all))
 933                 return;
 934
 935         write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
 936         list_del(&walk->walk.all);
 937         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 938 }
 939 EXPORT_SYMBOL(xfrm_policy_walk_done);
 940
 941 /*
 942  * Find policy to apply to this flow.
 943  *
 944  * Returns 0 if policy found, else an -errno.
 945  */
 946 static int xfrm_policy_match(const struct xfrm_policy *pol,
 947                              const struct flowi *fl,
 948                              u8 type, u16 family, int dir)
 949 {
 950         const struct xfrm_selector *sel = &pol->selector;
 951         int ret = -ESRCH;
 952         bool match;
 953
 954         if (pol->family != family ||
 955             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
 956             pol->type != type)
 957                 return ret;
 958
 959         match = xfrm_selector_match(sel, fl, family);
 960         if (match)
 961                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
 962                                                   dir);
 963
 964         return ret;
 965 }
 966
 967 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 968                                                      const struct flowi *fl,
 969                                                      u16 family, u8 dir)
 970 {
 971         int err;
 972         struct xfrm_policy *pol, *ret;
 973         const xfrm_address_t *daddr, *saddr;
 974         struct hlist_head *chain;
 975         u32 priority = ~0U;
 976
 977         daddr = xfrm_flowi_daddr(fl, family);
 978         saddr = xfrm_flowi_saddr(fl, family);
 979         if (unlikely(!daddr || !saddr))
 980                 return NULL;
 981
 982         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 983         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 984         ret = NULL;
 985         hlist_for_each_entry(pol, chain, bydst) {
 986                 err = xfrm_policy_match(pol, fl, type, family, dir);
 987                 if (err) {
 988                         if (err == -ESRCH)
 989                                 continue;
 990                         else {
 991                                 ret = ERR_PTR(err);
 992                                 goto fail;
 993                         }
 994                 } else {
 995                         ret = pol;
 996                         priority = ret->priority;
 997                         break;
 998                 }
 999         }
1000         chain = &net->xfrm.policy_inexact[dir];
1001         hlist_for_each_entry(pol, chain, bydst) {
1002                 err = xfrm_policy_match(pol, fl, type, family, dir);
1003                 if (err) {
1004                         if (err == -ESRCH)
1005                                 continue;
1006                         else {
1007                                 ret = ERR_PTR(err);
1008                                 goto fail;
1009                         }
1010                 } else if (pol->priority < priority) {
1011                         ret = pol;
1012                         break;
1013                 }
1014         }
1015         if (ret)
1016                 xfrm_pol_hold(ret);
1017 fail:
1018         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1019
1020         return ret;
1021 }
1022
1023 static struct xfrm_policy *
1024 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1025 {
1026 #ifdef CONFIG_XFRM_SUB_POLICY
1027         struct xfrm_policy *pol;
1028
1029         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1030         if (pol != NULL)
1031                 return pol;
1032 #endif
1033         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1034 }
1035
1036 static int flow_to_policy_dir(int dir)
1037 {
1038         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1039             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1040             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1041                 return dir;
1042
1043         switch (dir) {
1044         default:
1045         case FLOW_DIR_IN:
1046                 return XFRM_POLICY_IN;
1047         case FLOW_DIR_OUT:
1048                 return XFRM_POLICY_OUT;
1049         case FLOW_DIR_FWD:
1050                 return XFRM_POLICY_FWD;
1051         }
1052 }
1053
1054 static struct flow_cache_object *
1055 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1056                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1057 {
1058         struct xfrm_policy *pol;
1059
1060         if (old_obj)
1061                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1062
1063         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1064         if (IS_ERR_OR_NULL(pol))
1065                 return ERR_CAST(pol);
1066
1067         /* Resolver returns two references:
1068          * one for cache and one for caller of flow_cache_lookup() */
1069         xfrm_pol_hold(pol);
1070
1071         return &pol->flo;
1072 }
1073
1074 static inline int policy_to_flow_dir(int dir)
1075 {
1076         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1077             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1078             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1079                 return dir;
1080         switch (dir) {
1081         default:
1082         case XFRM_POLICY_IN:
1083                 return FLOW_DIR_IN;
1084         case XFRM_POLICY_OUT:
1085                 return FLOW_DIR_OUT;
1086         case XFRM_POLICY_FWD:
1087                 return FLOW_DIR_FWD;
1088         }
1089 }
1090
1091 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1092                                                  const struct flowi *fl)
1093 {
1094         struct xfrm_policy *pol;
1095         struct net *net = sock_net(sk);
1096
1097         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1098         if ((pol = sk->sk_policy[dir]) != NULL) {
1099                 bool match = xfrm_selector_match(&pol->selector, fl,
1100                                                  sk->sk_family);
1101                 int err = 0;
1102
1103                 if (match) {
1104                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1105                                 pol = NULL;
1106                                 goto out;
1107                         }
1108                         err = security_xfrm_policy_lookup(pol->security,
1109                                                       fl->flowi_secid,
1110                                                       policy_to_flow_dir(dir));
1111                         if (!err)
1112                                 xfrm_pol_hold(pol);
1113                         else if (err == -ESRCH)
1114                                 pol = NULL;
1115                         else
1116                                 pol = ERR_PTR(err);
1117                 } else
1118                         pol = NULL;
1119         }
1120 out:
1121         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1122         return pol;
1123 }
1124
1125 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1126 {
1127         struct net *net = xp_net(pol);
1128         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1129                                                      pol->family, dir);
1130
1131         list_add(&pol->walk.all, &net->xfrm.policy_all);
1132         hlist_add_head(&pol->bydst, chain);
1133         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1134         net->xfrm.policy_count[dir]++;
1135         xfrm_pol_hold(pol);
1136
1137         if (xfrm_bydst_should_resize(net, dir, NULL))
1138                 schedule_work(&net->xfrm.policy_hash_work);
1139 }
1140
1141 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1142                                                 int dir)
1143 {
1144         struct net *net = xp_net(pol);
1145
1146         if (hlist_unhashed(&pol->bydst))
1147                 return NULL;
1148
1149         hlist_del_init(&pol->bydst);
1150         hlist_del(&pol->byidx);
1151         list_del(&pol->walk.all);
1152         net->xfrm.policy_count[dir]--;
1153
1154         return pol;
1155 }
1156
1157 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1158 {
1159         struct net *net = xp_net(pol);
1160
1161         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1162         pol = __xfrm_policy_unlink(pol, dir);
1163         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1164         if (pol) {
1165                 xfrm_policy_kill(pol);
1166                 return 0;
1167         }
1168         return -ENOENT;
1169 }
1170 EXPORT_SYMBOL(xfrm_policy_delete);
1171
1172 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1173 {
1174         struct net *net = xp_net(pol);
1175         struct xfrm_policy *old_pol;
1176
1177 #ifdef CONFIG_XFRM_SUB_POLICY
1178         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1179                 return -EINVAL;
1180 #endif
1181
1182         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1183         old_pol = sk->sk_policy[dir];
1184         sk->sk_policy[dir] = pol;
1185         if (pol) {
1186                 pol->curlft.add_time = get_seconds();
1187                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1188                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1189         }
1190         if (old_pol) {
1191                 if (pol)
1192                         xfrm_policy_requeue(old_pol, pol);
1193
1194                 /* Unlinking succeeds always. This is the only function
1195                  * allowed to delete or replace socket policy.
1196                  */
1197                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1198         }
1199         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1200
1201         if (old_pol) {
1202                 xfrm_policy_kill(old_pol);
1203         }
1204         return 0;
1205 }
1206
1207 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1208 {
1209         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1210         struct net *net = xp_net(old);
1211
1212         if (newp) {
1213                 newp->selector = old->selector;
1214                 if (security_xfrm_policy_clone(old->security,
1215                                                &newp->security)) {
1216                         kfree(newp);
1217                         return NULL;  /* ENOMEM */
1218                 }
1219                 newp->lft = old->lft;
1220                 newp->curlft = old->curlft;
1221                 newp->mark = old->mark;
1222                 newp->action = old->action;
1223                 newp->flags = old->flags;
1224                 newp->xfrm_nr = old->xfrm_nr;
1225                 newp->index = old->index;
1226                 newp->type = old->type;
1227                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1228                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1229                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1230                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1231                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1232                 xfrm_pol_put(newp);
1233         }
1234         return newp;
1235 }
1236
1237 int __xfrm_sk_clone_policy(struct sock *sk)
1238 {
1239         struct xfrm_policy *p0 = sk->sk_policy[0],
1240                            *p1 = sk->sk_policy[1];
1241
1242         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1243         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1244                 return -ENOMEM;
1245         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1246                 return -ENOMEM;
1247         return 0;
1248 }
1249
1250 static int
1251 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1252                unsigned short family)
1253 {
1254         int err;
1255         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1256
1257         if (unlikely(afinfo == NULL))
1258                 return -EINVAL;
1259         err = afinfo->get_saddr(net, local, remote);
1260         xfrm_policy_put_afinfo(afinfo);
1261         return err;
1262 }
1263
1264 /* Resolve list of templates for the flow, given policy. */
1265
1266 static int
1267 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1268                       struct xfrm_state **xfrm, unsigned short family)
1269 {
1270         struct net *net = xp_net(policy);
1271         int nx;
1272         int i, error;
1273         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1274         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1275         xfrm_address_t tmp;
1276
1277         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1278                 struct xfrm_state *x;
1279                 xfrm_address_t *remote = daddr;
1280                 xfrm_address_t *local  = saddr;
1281                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1282
1283                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1284                     tmpl->mode == XFRM_MODE_BEET) {
1285                         remote = &tmpl->id.daddr;
1286                         local = &tmpl->saddr;
1287                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1288                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1289                                 if (error)
1290                                         goto fail;
1291                                 local = &tmp;
1292                         }
1293                 }
1294
1295                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1296
1297                 if (x && x->km.state == XFRM_STATE_VALID) {
1298                         xfrm[nx++] = x;
1299                         daddr = remote;
1300                         saddr = local;
1301                         continue;
1302                 }
1303                 if (x) {
1304                         error = (x->km.state == XFRM_STATE_ERROR ?
1305                                  -EINVAL : -EAGAIN);
1306                         xfrm_state_put(x);
1307                 } else if (error == -ESRCH) {
1308                         error = -EAGAIN;
1309                 }
1310
1311                 if (!tmpl->optional)
1312                         goto fail;
1313         }
1314         return nx;
1315
1316 fail:
1317         for (nx--; nx >= 0; nx--)
1318                 xfrm_state_put(xfrm[nx]);
1319         return error;
1320 }
1321
1322 static int
1323 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1324                   struct xfrm_state **xfrm, unsigned short family)
1325 {
1326         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1327         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1328         int cnx = 0;
1329         int error;
1330         int ret;
1331         int i;
1332
1333         for (i = 0; i < npols; i++) {
1334                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1335                         error = -ENOBUFS;
1336                         goto fail;
1337                 }
1338
1339                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1340                 if (ret < 0) {
1341                         error = ret;
1342                         goto fail;
1343                 } else
1344                         cnx += ret;
1345         }
1346
1347         /* found states are sorted for outbound processing */
1348         if (npols > 1)
1349                 xfrm_state_sort(xfrm, tpp, cnx, family);
1350
1351         return cnx;
1352
1353  fail:
1354         for (cnx--; cnx >= 0; cnx--)
1355                 xfrm_state_put(tpp[cnx]);
1356         return error;
1357
1358 }
1359
1360 /* Check that the bundle accepts the flow and its components are
1361  * still valid.
1362  */
1363
1364 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1365 {
1366         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1367         int tos;
1368
1369         if (!afinfo)
1370                 return -EINVAL;
1371
1372         tos = afinfo->get_tos(fl);
1373
1374         xfrm_policy_put_afinfo(afinfo);
1375
1376         return tos;
1377 }
1378
1379 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1380 {
1381         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1382         struct dst_entry *dst = &xdst->u.dst;
1383
1384         if (xdst->route == NULL) {
1385                 /* Dummy bundle - if it has xfrms we were not
1386                  * able to build bundle as template resolution failed.
1387                  * It means we need to try again resolving. */
1388                 if (xdst->num_xfrms > 0)
1389                         return NULL;
1390         } else if (dst->flags & DST_XFRM_QUEUE) {
1391                 return NULL;
1392         } else {
1393                 /* Real bundle */
1394                 if (stale_bundle(dst))
1395                         return NULL;
1396         }
1397
1398         dst_hold(dst);
1399         return flo;
1400 }
1401
1402 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1403 {
1404         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1405         struct dst_entry *dst = &xdst->u.dst;
1406
1407         if (!xdst->route)
1408                 return 0;
1409         if (stale_bundle(dst))
1410                 return 0;
1411
1412         return 1;
1413 }
1414
1415 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1416 {
1417         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1418         struct dst_entry *dst = &xdst->u.dst;
1419
1420         dst_free(dst);
1421 }
1422
1423 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1424         .get = xfrm_bundle_flo_get,
1425         .check = xfrm_bundle_flo_check,
1426         .delete = xfrm_bundle_flo_delete,
1427 };
1428
1429 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1430 {
1431         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1432         struct dst_ops *dst_ops;
1433         struct xfrm_dst *xdst;
1434
1435         if (!afinfo)
1436                 return ERR_PTR(-EINVAL);
1437
1438         switch (family) {
1439         case AF_INET:
1440                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1441                 break;
1442 #if IS_ENABLED(CONFIG_IPV6)
1443         case AF_INET6:
1444                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1445                 break;
1446 #endif
1447         default:
1448                 BUG();
1449         }
1450         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1451
1452         if (likely(xdst)) {
1453                 struct dst_entry *dst = &xdst->u.dst;
1454
1455                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1456                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1457                 if (afinfo->init_dst)
1458                         afinfo->init_dst(net, xdst);
1459         } else
1460                 xdst = ERR_PTR(-ENOBUFS);
1461
1462         xfrm_policy_put_afinfo(afinfo);
1463
1464         return xdst;
1465 }
1466
1467 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1468                                  int nfheader_len)
1469 {
1470         struct xfrm_policy_afinfo *afinfo =
1471                 xfrm_policy_get_afinfo(dst->ops->family);
1472         int err;
1473
1474         if (!afinfo)
1475                 return -EINVAL;
1476
1477         err = afinfo->init_path(path, dst, nfheader_len);
1478
1479         xfrm_policy_put_afinfo(afinfo);
1480
1481         return err;
1482 }
1483
1484 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1485                                 const struct flowi *fl)
1486 {
1487         struct xfrm_policy_afinfo *afinfo =
1488                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1489         int err;
1490
1491         if (!afinfo)
1492                 return -EINVAL;
1493
1494         err = afinfo->fill_dst(xdst, dev, fl);
1495
1496         xfrm_policy_put_afinfo(afinfo);
1497
1498         return err;
1499 }
1500
1501
1502 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1503  * all the metrics... Shortly, bundle a bundle.
1504  */
1505
1506 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1507                                             struct xfrm_state **xfrm, int nx,
1508                                             const struct flowi *fl,
1509                                             struct dst_entry *dst)
1510 {
1511         struct net *net = xp_net(policy);
1512         unsigned long now = jiffies;
1513         struct net_device *dev;
1514         struct xfrm_mode *inner_mode;
1515         struct dst_entry *dst_prev = NULL;
1516         struct dst_entry *dst0 = NULL;
1517         int i = 0;
1518         int err;
1519         int header_len = 0;
1520         int nfheader_len = 0;
1521         int trailer_len = 0;
1522         int tos;
1523         int family = policy->selector.family;
1524         xfrm_address_t saddr, daddr;
1525
1526         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1527
1528         tos = xfrm_get_tos(fl, family);
1529         err = tos;
1530         if (tos < 0)
1531                 goto put_states;
1532
1533         dst_hold(dst);
1534
1535         for (; i < nx; i++) {
1536                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1537                 struct dst_entry *dst1 = &xdst->u.dst;
1538
1539                 err = PTR_ERR(xdst);
1540                 if (IS_ERR(xdst)) {
1541                         dst_release(dst);
1542                         goto put_states;
1543                 }
1544
1545                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1546                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1547                                                         xfrm_af2proto(family));
1548                         if (!inner_mode) {
1549                                 err = -EAFNOSUPPORT;
1550                                 dst_release(dst);
1551                                 goto put_states;
1552                         }
1553                 } else
1554                         inner_mode = xfrm[i]->inner_mode;
1555
1556                 if (!dst_prev)
1557                         dst0 = dst1;
1558                 else {
1559                         dst_prev->child = dst_clone(dst1);
1560                         dst1->flags |= DST_NOHASH;
1561                 }
1562
1563                 xdst->route = dst;
1564                 dst_copy_metrics(dst1, dst);
1565
1566                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1567                         family = xfrm[i]->props.family;
1568                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1569                                               family);
1570                         err = PTR_ERR(dst);
1571                         if (IS_ERR(dst))
1572                                 goto put_states;
1573                 } else
1574                         dst_hold(dst);
1575
1576                 dst1->xfrm = xfrm[i];
1577                 xdst->xfrm_genid = xfrm[i]->genid;
1578
1579                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1580                 dst1->flags |= DST_HOST;
1581                 dst1->lastuse = now;
1582
1583                 dst1->input = dst_discard;
1584                 dst1->output = inner_mode->afinfo->output;
1585
1586                 dst1->next = dst_prev;
1587                 dst_prev = dst1;
1588
1589                 header_len += xfrm[i]->props.header_len;
1590                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1591                         nfheader_len += xfrm[i]->props.header_len;
1592                 trailer_len += xfrm[i]->props.trailer_len;
1593         }
1594
1595         dst_prev->child = dst;
1596         dst0->path = dst;
1597
1598         err = -ENODEV;
1599         dev = dst->dev;
1600         if (!dev)
1601                 goto free_dst;
1602
1603         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1604         xfrm_init_pmtu(dst_prev);
1605
1606         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1607                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1608
1609                 err = xfrm_fill_dst(xdst, dev, fl);
1610                 if (err)
1611                         goto free_dst;
1612
1613                 dst_prev->header_len = header_len;
1614                 dst_prev->trailer_len = trailer_len;
1615                 header_len -= xdst->u.dst.xfrm->props.header_len;
1616                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1617         }
1618
1619 out:
1620         return dst0;
1621
1622 put_states:
1623         for (; i < nx; i++)
1624                 xfrm_state_put(xfrm[i]);
1625 free_dst:
1626         if (dst0)
1627                 dst_free(dst0);
1628         dst0 = ERR_PTR(err);
1629         goto out;
1630 }
1631
1632 #ifdef CONFIG_XFRM_SUB_POLICY
1633 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1634 {
1635         if (!*target) {
1636                 *target = kmalloc(size, GFP_ATOMIC);
1637                 if (!*target)
1638                         return -ENOMEM;
1639         }
1640
1641         memcpy(*target, src, size);
1642         return 0;
1643 }
1644 #endif
1645
1646 static int xfrm_dst_update_parent(struct dst_entry *dst,
1647                                   const struct xfrm_selector *sel)
1648 {
1649 #ifdef CONFIG_XFRM_SUB_POLICY
1650         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1651         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1652                                    sel, sizeof(*sel));
1653 #else
1654         return 0;
1655 #endif
1656 }
1657
1658 static int xfrm_dst_update_origin(struct dst_entry *dst,
1659                                   const struct flowi *fl)
1660 {
1661 #ifdef CONFIG_XFRM_SUB_POLICY
1662         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1663         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1664 #else
1665         return 0;
1666 #endif
1667 }
1668
1669 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1670                                 struct xfrm_policy **pols,
1671                                 int *num_pols, int *num_xfrms)
1672 {
1673         int i;
1674
1675         if (*num_pols == 0 || !pols[0]) {
1676                 *num_pols = 0;
1677                 *num_xfrms = 0;
1678                 return 0;
1679         }
1680         if (IS_ERR(pols[0]))
1681                 return PTR_ERR(pols[0]);
1682
1683         *num_xfrms = pols[0]->xfrm_nr;
1684
1685 #ifdef CONFIG_XFRM_SUB_POLICY
1686         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1687             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1688                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1689                                                     XFRM_POLICY_TYPE_MAIN,
1690                                                     fl, family,
1691                                                     XFRM_POLICY_OUT);
1692                 if (pols[1]) {
1693                         if (IS_ERR(pols[1])) {
1694                                 xfrm_pols_put(pols, *num_pols);
1695                                 return PTR_ERR(pols[1]);
1696                         }
1697                         (*num_pols)++;
1698                         (*num_xfrms) += pols[1]->xfrm_nr;
1699                 }
1700         }
1701 #endif
1702         for (i = 0; i < *num_pols; i++) {
1703                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1704                         *num_xfrms = -1;
1705                         break;
1706                 }
1707         }
1708
1709         return 0;
1710
1711 }
1712
1713 static struct xfrm_dst *
1714 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1715                                const struct flowi *fl, u16 family,
1716                                struct dst_entry *dst_orig)
1717 {
1718         struct net *net = xp_net(pols[0]);
1719         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1720         struct dst_entry *dst;
1721         struct xfrm_dst *xdst;
1722         int err;
1723
1724         /* Try to instantiate a bundle */
1725         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1726         if (err <= 0) {
1727                 if (err != 0 && err != -EAGAIN)
1728                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1729                 return ERR_PTR(err);
1730         }
1731
1732         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1733         if (IS_ERR(dst)) {
1734                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1735                 return ERR_CAST(dst);
1736         }
1737
1738         xdst = (struct xfrm_dst *)dst;
1739         xdst->num_xfrms = err;
1740         if (num_pols > 1)
1741                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1742         else
1743                 err = xfrm_dst_update_origin(dst, fl);
1744         if (unlikely(err)) {
1745                 dst_free(dst);
1746                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1747                 return ERR_PTR(err);
1748         }
1749
1750         xdst->num_pols = num_pols;
1751         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1752         xdst->policy_genid = atomic_read(&pols[0]->genid);
1753
1754         return xdst;
1755 }
1756
1757 static void xfrm_policy_queue_process(unsigned long arg)
1758 {
1759         int err = 0;
1760         struct sk_buff *skb;
1761         struct sock *sk;
1762         struct dst_entry *dst;
1763         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1764         struct xfrm_policy_queue *pq = &pol->polq;
1765         struct flowi fl;
1766         struct sk_buff_head list;
1767
1768         spin_lock(&pq->hold_queue.lock);
1769         skb = skb_peek(&pq->hold_queue);
1770         if (!skb) {
1771                 spin_unlock(&pq->hold_queue.lock);
1772                 goto out;
1773         }
1774         dst = skb_dst(skb);
1775         sk = skb->sk;
1776         xfrm_decode_session(skb, &fl, dst->ops->family);
1777         spin_unlock(&pq->hold_queue.lock);
1778
1779         dst_hold(dst->path);
1780         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1781                           sk, 0);
1782         if (IS_ERR(dst))
1783                 goto purge_queue;
1784
1785         if (dst->flags & DST_XFRM_QUEUE) {
1786                 dst_release(dst);
1787
1788                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1789                         goto purge_queue;
1790
1791                 pq->timeout = pq->timeout << 1;
1792                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1793                         xfrm_pol_hold(pol);
1794         goto out;
1795         }
1796
1797         dst_release(dst);
1798
1799         __skb_queue_head_init(&list);
1800
1801         spin_lock(&pq->hold_queue.lock);
1802         pq->timeout = 0;
1803         skb_queue_splice_init(&pq->hold_queue, &list);
1804         spin_unlock(&pq->hold_queue.lock);
1805
1806         while (!skb_queue_empty(&list)) {
1807                 skb = __skb_dequeue(&list);
1808
1809                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1810                 dst_hold(skb_dst(skb)->path);
1811                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1812                                   &fl, skb->sk, 0);
1813                 if (IS_ERR(dst)) {
1814                         kfree_skb(skb);
1815                         continue;
1816                 }
1817
1818                 nf_reset(skb);
1819                 skb_dst_drop(skb);
1820                 skb_dst_set(skb, dst);
1821
1822                 err = dst_output(skb);
1823         }
1824
1825 out:
1826         xfrm_pol_put(pol);
1827         return;
1828
1829 purge_queue:
1830         pq->timeout = 0;
1831         xfrm_queue_purge(&pq->hold_queue);
1832         xfrm_pol_put(pol);
1833 }
1834
1835 static int xdst_queue_output(struct sock *sk, struct sk_buff *skb)
1836 {
1837         unsigned long sched_next;
1838         struct dst_entry *dst = skb_dst(skb);
1839         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1840         struct xfrm_policy *pol = xdst->pols[0];
1841         struct xfrm_policy_queue *pq = &pol->polq;
1842         const struct sk_buff *fclone = skb + 1;
1843
1844         if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
1845                      fclone->fclone == SKB_FCLONE_CLONE)) {
1846                 kfree_skb(skb);
1847                 return 0;
1848         }
1849
1850         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1851                 kfree_skb(skb);
1852                 return -EAGAIN;
1853         }
1854
1855         skb_dst_force(skb);
1856
1857         spin_lock_bh(&pq->hold_queue.lock);
1858
1859         if (!pq->timeout)
1860                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1861
1862         sched_next = jiffies + pq->timeout;
1863
1864         if (del_timer(&pq->hold_timer)) {
1865                 if (time_before(pq->hold_timer.expires, sched_next))
1866                         sched_next = pq->hold_timer.expires;
1867                 xfrm_pol_put(pol);
1868         }
1869
1870         __skb_queue_tail(&pq->hold_queue, skb);
1871         if (!mod_timer(&pq->hold_timer, sched_next))
1872                 xfrm_pol_hold(pol);
1873
1874         spin_unlock_bh(&pq->hold_queue.lock);
1875
1876         return 0;
1877 }
1878
1879 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1880                                                  struct dst_entry *dst,
1881                                                  const struct flowi *fl,
1882                                                  int num_xfrms,
1883                                                  u16 family)
1884 {
1885         int err;
1886         struct net_device *dev;
1887         struct dst_entry *dst1;
1888         struct xfrm_dst *xdst;
1889
1890         xdst = xfrm_alloc_dst(net, family);
1891         if (IS_ERR(xdst))
1892                 return xdst;
1893
1894         if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0)
1895                 return xdst;
1896
1897         dst1 = &xdst->u.dst;
1898         dst_hold(dst);
1899         xdst->route = dst;
1900
1901         dst_copy_metrics(dst1, dst);
1902
1903         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1904         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1905         dst1->lastuse = jiffies;
1906
1907         dst1->input = dst_discard;
1908         dst1->output = xdst_queue_output;
1909
1910         dst_hold(dst);
1911         dst1->child = dst;
1912         dst1->path = dst;
1913
1914         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1915
1916         err = -ENODEV;
1917         dev = dst->dev;
1918         if (!dev)
1919                 goto free_dst;
1920
1921         err = xfrm_fill_dst(xdst, dev, fl);
1922         if (err)
1923                 goto free_dst;
1924
1925 out:
1926         return xdst;
1927
1928 free_dst:
1929         dst_release(dst1);
1930         xdst = ERR_PTR(err);
1931         goto out;
1932 }
1933
1934 static struct flow_cache_object *
1935 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1936                    struct flow_cache_object *oldflo, void *ctx)
1937 {
1938         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1939         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1940         struct xfrm_dst *xdst, *new_xdst;
1941         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1942
1943         /* Check if the policies from old bundle are usable */
1944         xdst = NULL;
1945         if (oldflo) {
1946                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1947                 num_pols = xdst->num_pols;
1948                 num_xfrms = xdst->num_xfrms;
1949                 pol_dead = 0;
1950                 for (i = 0; i < num_pols; i++) {
1951                         pols[i] = xdst->pols[i];
1952                         pol_dead |= pols[i]->walk.dead;
1953                 }
1954                 if (pol_dead) {
1955                         dst_free(&xdst->u.dst);
1956                         xdst = NULL;
1957                         num_pols = 0;
1958                         num_xfrms = 0;
1959                         oldflo = NULL;
1960                 }
1961         }
1962
1963         /* Resolve policies to use if we couldn't get them from
1964          * previous cache entry */
1965         if (xdst == NULL) {
1966                 num_pols = 1;
1967                 pols[0] = __xfrm_policy_lookup(net, fl, family,
1968                                                flow_to_policy_dir(dir));
1969                 err = xfrm_expand_policies(fl, family, pols,
1970                                            &num_pols, &num_xfrms);
1971                 if (err < 0)
1972                         goto inc_error;
1973                 if (num_pols == 0)
1974                         return NULL;
1975                 if (num_xfrms <= 0)
1976                         goto make_dummy_bundle;
1977         }
1978
1979         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1980         if (IS_ERR(new_xdst)) {
1981                 err = PTR_ERR(new_xdst);
1982                 if (err != -EAGAIN)
1983                         goto error;
1984                 if (oldflo == NULL)
1985                         goto make_dummy_bundle;
1986                 dst_hold(&xdst->u.dst);
1987                 return oldflo;
1988         } else if (new_xdst == NULL) {
1989                 num_xfrms = 0;
1990                 if (oldflo == NULL)
1991                         goto make_dummy_bundle;
1992                 xdst->num_xfrms = 0;
1993                 dst_hold(&xdst->u.dst);
1994                 return oldflo;
1995         }
1996
1997         /* Kill the previous bundle */
1998         if (xdst) {
1999                 /* The policies were stolen for newly generated bundle */
2000                 xdst->num_pols = 0;
2001                 dst_free(&xdst->u.dst);
2002         }
2003
2004         /* Flow cache does not have reference, it dst_free()'s,
2005          * but we do need to return one reference for original caller */
2006         dst_hold(&new_xdst->u.dst);
2007         return &new_xdst->flo;
2008
2009 make_dummy_bundle:
2010         /* We found policies, but there's no bundles to instantiate:
2011          * either because the policy blocks, has no transformations or
2012          * we could not build template (no xfrm_states).*/
2013         xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
2014         if (IS_ERR(xdst)) {
2015                 xfrm_pols_put(pols, num_pols);
2016                 return ERR_CAST(xdst);
2017         }
2018         xdst->num_pols = num_pols;
2019         xdst->num_xfrms = num_xfrms;
2020         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2021
2022         dst_hold(&xdst->u.dst);
2023         return &xdst->flo;
2024
2025 inc_error:
2026         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2027 error:
2028         if (xdst != NULL)
2029                 dst_free(&xdst->u.dst);
2030         else
2031                 xfrm_pols_put(pols, num_pols);
2032         return ERR_PTR(err);
2033 }
2034
2035 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2036                                         struct dst_entry *dst_orig)
2037 {
2038         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2039         struct dst_entry *ret;
2040
2041         if (!afinfo) {
2042                 dst_release(dst_orig);
2043                 return ERR_PTR(-EINVAL);
2044         } else {
2045                 ret = afinfo->blackhole_route(net, dst_orig);
2046         }
2047         xfrm_policy_put_afinfo(afinfo);
2048
2049         return ret;
2050 }
2051
2052 /* Main function: finds/creates a bundle for given flow.
2053  *
2054  * At the moment we eat a raw IP route. Mostly to speed up lookups
2055  * on interfaces with disabled IPsec.
2056  */
2057 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2058                               const struct flowi *fl,
2059                               struct sock *sk, int flags)
2060 {
2061         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2062         struct flow_cache_object *flo;
2063         struct xfrm_dst *xdst;
2064         struct dst_entry *dst, *route;
2065         u16 family = dst_orig->ops->family;
2066         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2067         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2068
2069         dst = NULL;
2070         xdst = NULL;
2071         route = NULL;
2072
2073         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2074                 num_pols = 1;
2075                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2076                 err = xfrm_expand_policies(fl, family, pols,
2077                                            &num_pols, &num_xfrms);
2078                 if (err < 0)
2079                         goto dropdst;
2080
2081                 if (num_pols) {
2082                         if (num_xfrms <= 0) {
2083                                 drop_pols = num_pols;
2084                                 goto no_transform;
2085                         }
2086
2087                         xdst = xfrm_resolve_and_create_bundle(
2088                                         pols, num_pols, fl,
2089                                         family, dst_orig);
2090                         if (IS_ERR(xdst)) {
2091                                 xfrm_pols_put(pols, num_pols);
2092                                 err = PTR_ERR(xdst);
2093                                 goto dropdst;
2094                         } else if (xdst == NULL) {
2095                                 num_xfrms = 0;
2096                                 drop_pols = num_pols;
2097                                 goto no_transform;
2098                         }
2099
2100                         route = xdst->route;
2101                 }
2102         }
2103
2104         if (xdst == NULL) {
2105                 /* To accelerate a bit...  */
2106                 if ((dst_orig->flags & DST_NOXFRM) ||
2107                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2108                         goto nopol;
2109
2110                 flo = flow_cache_lookup(net, fl, family, dir,
2111                                         xfrm_bundle_lookup, dst_orig);
2112                 if (flo == NULL)
2113                         goto nopol;
2114                 if (IS_ERR(flo)) {
2115                         err = PTR_ERR(flo);
2116                         goto dropdst;
2117                 }
2118                 xdst = container_of(flo, struct xfrm_dst, flo);
2119
2120                 num_pols = xdst->num_pols;
2121                 num_xfrms = xdst->num_xfrms;
2122                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2123                 route = xdst->route;
2124         }
2125
2126         dst = &xdst->u.dst;
2127         if (route == NULL && num_xfrms > 0) {
2128                 /* The only case when xfrm_bundle_lookup() returns a
2129                  * bundle with null route, is when the template could
2130                  * not be resolved. It means policies are there, but
2131                  * bundle could not be created, since we don't yet
2132                  * have the xfrm_state's. We need to wait for KM to
2133                  * negotiate new SA's or bail out with error.*/
2134                 if (net->xfrm.sysctl_larval_drop) {
2135                         dst_release(dst);
2136                         xfrm_pols_put(pols, drop_pols);
2137                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2138
2139                         return make_blackhole(net, family, dst_orig);
2140                 }
2141
2142                 err = -EAGAIN;
2143
2144                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2145                 goto error;
2146         }
2147
2148 no_transform:
2149         if (num_pols == 0)
2150                 goto nopol;
2151
2152         if ((flags & XFRM_LOOKUP_ICMP) &&
2153             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2154                 err = -ENOENT;
2155                 goto error;
2156         }
2157
2158         for (i = 0; i < num_pols; i++)
2159                 pols[i]->curlft.use_time = get_seconds();
2160
2161         if (num_xfrms < 0) {
2162                 /* Prohibit the flow */
2163                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2164                 err = -EPERM;
2165                 goto error;
2166         } else if (num_xfrms > 0) {
2167                 /* Flow transformed */
2168                 dst_release(dst_orig);
2169         } else {
2170                 /* Flow passes untransformed */
2171                 dst_release(dst);
2172                 dst = dst_orig;
2173         }
2174 ok:
2175         xfrm_pols_put(pols, drop_pols);
2176         if (dst && dst->xfrm &&
2177             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2178                 dst->flags |= DST_XFRM_TUNNEL;
2179         return dst;
2180
2181 nopol:
2182         if (!(flags & XFRM_LOOKUP_ICMP)) {
2183                 dst = dst_orig;
2184                 goto ok;
2185         }
2186         err = -ENOENT;
2187 error:
2188         dst_release(dst);
2189 dropdst:
2190         dst_release(dst_orig);
2191         xfrm_pols_put(pols, drop_pols);
2192         return ERR_PTR(err);
2193 }
2194 EXPORT_SYMBOL(xfrm_lookup);
2195
2196 static inline int
2197 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2198 {
2199         struct xfrm_state *x;
2200
2201         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2202                 return 0;
2203         x = skb->sp->xvec[idx];
2204         if (!x->type->reject)
2205                 return 0;
2206         return x->type->reject(x, skb, fl);
2207 }
2208
2209 /* When skb is transformed back to its "native" form, we have to
2210  * check policy restrictions. At the moment we make this in maximally
2211  * stupid way. Shame on me. :-) Of course, connected sockets must
2212  * have policy cached at them.
2213  */
2214
2215 static inline int
2216 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2217               unsigned short family)
2218 {
2219         if (xfrm_state_kern(x))
2220                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2221         return  x->id.proto == tmpl->id.proto &&
2222                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2223                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2224                 x->props.mode == tmpl->mode &&
2225                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2226                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2227                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2228                   xfrm_state_addr_cmp(tmpl, x, family));
2229 }
2230
2231 /*
2232  * 0 or more than 0 is returned when validation is succeeded (either bypass
2233  * because of optional transport mode, or next index of the mathced secpath
2234  * state with the template.
2235  * -1 is returned when no matching template is found.
2236  * Otherwise "-2 - errored_index" is returned.
2237  */
2238 static inline int
2239 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2240                unsigned short family)
2241 {
2242         int idx = start;
2243
2244         if (tmpl->optional) {
2245                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2246                         return start;
2247         } else
2248                 start = -1;
2249         for (; idx < sp->len; idx++) {
2250                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2251                         return ++idx;
2252                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2253                         if (start == -1)
2254                                 start = -2-idx;
2255                         break;
2256                 }
2257         }
2258         return start;
2259 }
2260
2261 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2262                           unsigned int family, int reverse)
2263 {
2264         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2265         int err;
2266
2267         if (unlikely(afinfo == NULL))
2268                 return -EAFNOSUPPORT;
2269
2270         afinfo->decode_session(skb, fl, reverse);
2271         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2272         xfrm_policy_put_afinfo(afinfo);
2273         return err;
2274 }
2275 EXPORT_SYMBOL(__xfrm_decode_session);
2276
2277 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2278 {
2279         for (; k < sp->len; k++) {
2280                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2281                         *idxp = k;
2282                         return 1;
2283                 }
2284         }
2285
2286         return 0;
2287 }
2288
2289 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2290                         unsigned short family)
2291 {
2292         struct net *net = dev_net(skb->dev);
2293         struct xfrm_policy *pol;
2294         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2295         int npols = 0;
2296         int xfrm_nr;
2297         int pi;
2298         int reverse;
2299         struct flowi fl;
2300         u8 fl_dir;
2301         int xerr_idx = -1;
2302
2303         reverse = dir & ~XFRM_POLICY_MASK;
2304         dir &= XFRM_POLICY_MASK;
2305         fl_dir = policy_to_flow_dir(dir);
2306
2307         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2308                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2309                 return 0;
2310         }
2311
2312         nf_nat_decode_session(skb, &fl, family);
2313
2314         /* First, check used SA against their selectors. */
2315         if (skb->sp) {
2316                 int i;
2317
2318                 for (i = skb->sp->len-1; i >= 0; i--) {
2319                         struct xfrm_state *x = skb->sp->xvec[i];
2320                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2321                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2322                                 return 0;
2323                         }
2324                 }
2325         }
2326
2327         pol = NULL;
2328         if (sk && sk->sk_policy[dir]) {
2329                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2330                 if (IS_ERR(pol)) {
2331                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2332                         return 0;
2333                 }
2334         }
2335
2336         if (!pol) {
2337                 struct flow_cache_object *flo;
2338
2339                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2340                                         xfrm_policy_lookup, NULL);
2341                 if (IS_ERR_OR_NULL(flo))
2342                         pol = ERR_CAST(flo);
2343                 else
2344                         pol = container_of(flo, struct xfrm_policy, flo);
2345         }
2346
2347         if (IS_ERR(pol)) {
2348                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2349                 return 0;
2350         }
2351
2352         if (!pol) {
2353                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2354                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2355                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2356                         return 0;
2357                 }
2358                 return 1;
2359         }
2360
2361         pol->curlft.use_time = get_seconds();
2362
2363         pols[0] = pol;
2364         npols++;
2365 #ifdef CONFIG_XFRM_SUB_POLICY
2366         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2367                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2368                                                     &fl, family,
2369                                                     XFRM_POLICY_IN);
2370                 if (pols[1]) {
2371                         if (IS_ERR(pols[1])) {
2372                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2373                                 return 0;
2374                         }
2375                         pols[1]->curlft.use_time = get_seconds();
2376                         npols++;
2377                 }
2378         }
2379 #endif
2380
2381         if (pol->action == XFRM_POLICY_ALLOW) {
2382                 struct sec_path *sp;
2383                 static struct sec_path dummy;
2384                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2385                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2386                 struct xfrm_tmpl **tpp = tp;
2387                 int ti = 0;
2388                 int i, k;
2389
2390                 if ((sp = skb->sp) == NULL)
2391                         sp = &dummy;
2392
2393                 for (pi = 0; pi < npols; pi++) {
2394                         if (pols[pi] != pol &&
2395                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2396                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2397                                 goto reject;
2398                         }
2399                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2400                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2401                                 goto reject_error;
2402                         }
2403                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2404                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2405                 }
2406                 xfrm_nr = ti;
2407                 if (npols > 1) {
2408                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2409                         tpp = stp;
2410                 }
2411
2412                 /* For each tunnel xfrm, find the first matching tmpl.
2413                  * For each tmpl before that, find corresponding xfrm.
2414                  * Order is _important_. Later we will implement
2415                  * some barriers, but at the moment barriers
2416                  * are implied between each two transformations.
2417                  */
2418                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2419                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2420                         if (k < 0) {
2421                                 if (k < -1)
2422                                         /* "-2 - errored_index" returned */
2423                                         xerr_idx = -(2+k);
2424                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2425                                 goto reject;
2426                         }
2427                 }
2428
2429                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2430                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2431                         goto reject;
2432                 }
2433
2434                 xfrm_pols_put(pols, npols);
2435                 return 1;
2436         }
2437         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2438
2439 reject:
2440         xfrm_secpath_reject(xerr_idx, skb, &fl);
2441 reject_error:
2442         xfrm_pols_put(pols, npols);
2443         return 0;
2444 }
2445 EXPORT_SYMBOL(__xfrm_policy_check);
2446
2447 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2448 {
2449         struct net *net = dev_net(skb->dev);
2450         struct flowi fl;
2451         struct dst_entry *dst;
2452         int res = 1;
2453
2454         if (xfrm_decode_session(skb, &fl, family) < 0) {
2455                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2456                 return 0;
2457         }
2458
2459         skb_dst_force(skb);
2460
2461         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2462         if (IS_ERR(dst)) {
2463                 res = 0;
2464                 dst = NULL;
2465         }
2466         skb_dst_set(skb, dst);
2467         return res;
2468 }
2469 EXPORT_SYMBOL(__xfrm_route_forward);
2470
2471 /* Optimize later using cookies and generation ids. */
2472
2473 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2474 {
2475         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2476          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2477          * get validated by dst_ops->check on every use.  We do this
2478          * because when a normal route referenced by an XFRM dst is
2479          * obsoleted we do not go looking around for all parent
2480          * referencing XFRM dsts so that we can invalidate them.  It
2481          * is just too much work.  Instead we make the checks here on
2482          * every use.  For example:
2483          *
2484          *      XFRM dst A --> IPv4 dst X
2485          *
2486          * X is the "xdst->route" of A (X is also the "dst->path" of A
2487          * in this example).  If X is marked obsolete, "A" will not
2488          * notice.  That's what we are validating here via the
2489          * stale_bundle() check.
2490          *
2491          * When a policy's bundle is pruned, we dst_free() the XFRM
2492          * dst which causes it's ->obsolete field to be set to
2493          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2494          * this, we want to force a new route lookup.
2495          */
2496         if (dst->obsolete < 0 && !stale_bundle(dst))
2497                 return dst;
2498
2499         return NULL;
2500 }
2501
2502 static int stale_bundle(struct dst_entry *dst)
2503 {
2504         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2505 }
2506
2507 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2508 {
2509         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2510                 dst->dev = dev_net(dev)->loopback_dev;
2511                 dev_hold(dst->dev);
2512                 dev_put(dev);
2513         }
2514 }
2515 EXPORT_SYMBOL(xfrm_dst_ifdown);
2516
2517 static void xfrm_link_failure(struct sk_buff *skb)
2518 {
2519         /* Impossible. Such dst must be popped before reaches point of failure. */
2520 }
2521
2522 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2523 {
2524         if (dst) {
2525                 if (dst->obsolete) {
2526                         dst_release(dst);
2527                         dst = NULL;
2528                 }
2529         }
2530         return dst;
2531 }
2532
2533 void xfrm_garbage_collect(struct net *net)
2534 {
2535         flow_cache_flush(net);
2536 }
2537 EXPORT_SYMBOL(xfrm_garbage_collect);
2538
2539 static void xfrm_garbage_collect_deferred(struct net *net)
2540 {
2541         flow_cache_flush_deferred(net);
2542 }
2543
2544 static void xfrm_init_pmtu(struct dst_entry *dst)
2545 {
2546         do {
2547                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2548                 u32 pmtu, route_mtu_cached;
2549
2550                 pmtu = dst_mtu(dst->child);
2551                 xdst->child_mtu_cached = pmtu;
2552
2553                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2554
2555                 route_mtu_cached = dst_mtu(xdst->route);
2556                 xdst->route_mtu_cached = route_mtu_cached;
2557
2558                 if (pmtu > route_mtu_cached)
2559                         pmtu = route_mtu_cached;
2560
2561                 dst_metric_set(dst, RTAX_MTU, pmtu);
2562         } while ((dst = dst->next));
2563 }
2564
2565 /* Check that the bundle accepts the flow and its components are
2566  * still valid.
2567  */
2568
2569 static int xfrm_bundle_ok(struct xfrm_dst *first)
2570 {
2571         struct dst_entry *dst = &first->u.dst;
2572         struct xfrm_dst *last;
2573         u32 mtu;
2574
2575         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2576             (dst->dev && !netif_running(dst->dev)))
2577                 return 0;
2578
2579         if (dst->flags & DST_XFRM_QUEUE)
2580                 return 1;
2581
2582         last = NULL;
2583
2584         do {
2585                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2586
2587                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2588                         return 0;
2589                 if (xdst->xfrm_genid != dst->xfrm->genid)
2590                         return 0;
2591                 if (xdst->num_pols > 0 &&
2592                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2593                         return 0;
2594
2595                 mtu = dst_mtu(dst->child);
2596                 if (xdst->child_mtu_cached != mtu) {
2597                         last = xdst;
2598                         xdst->child_mtu_cached = mtu;
2599                 }
2600
2601                 if (!dst_check(xdst->route, xdst->route_cookie))
2602                         return 0;
2603                 mtu = dst_mtu(xdst->route);
2604                 if (xdst->route_mtu_cached != mtu) {
2605                         last = xdst;
2606                         xdst->route_mtu_cached = mtu;
2607                 }
2608
2609                 dst = dst->child;
2610         } while (dst->xfrm);
2611
2612         if (likely(!last))
2613                 return 1;
2614
2615         mtu = last->child_mtu_cached;
2616         for (;;) {
2617                 dst = &last->u.dst;
2618
2619                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2620                 if (mtu > last->route_mtu_cached)
2621                         mtu = last->route_mtu_cached;
2622                 dst_metric_set(dst, RTAX_MTU, mtu);
2623
2624                 if (last == first)
2625                         break;
2626
2627                 last = (struct xfrm_dst *)last->u.dst.next;
2628                 last->child_mtu_cached = mtu;
2629         }
2630
2631         return 1;
2632 }
2633
2634 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2635 {
2636         return dst_metric_advmss(dst->path);
2637 }
2638
2639 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2640 {
2641         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2642
2643         return mtu ? : dst_mtu(dst->path);
2644 }
2645
2646 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2647                                            struct sk_buff *skb,
2648                                            const void *daddr)
2649 {
2650         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2651 }
2652
2653 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2654 {
2655         struct net *net;
2656         int err = 0;
2657         if (unlikely(afinfo == NULL))
2658                 return -EINVAL;
2659         if (unlikely(afinfo->family >= NPROTO))
2660                 return -EAFNOSUPPORT;
2661         spin_lock(&xfrm_policy_afinfo_lock);
2662         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2663                 err = -ENOBUFS;
2664         else {
2665                 struct dst_ops *dst_ops = afinfo->dst_ops;
2666                 if (likely(dst_ops->kmem_cachep == NULL))
2667                         dst_ops->kmem_cachep = xfrm_dst_cache;
2668                 if (likely(dst_ops->check == NULL))
2669                         dst_ops->check = xfrm_dst_check;
2670                 if (likely(dst_ops->default_advmss == NULL))
2671                         dst_ops->default_advmss = xfrm_default_advmss;
2672                 if (likely(dst_ops->mtu == NULL))
2673                         dst_ops->mtu = xfrm_mtu;
2674                 if (likely(dst_ops->negative_advice == NULL))
2675                         dst_ops->negative_advice = xfrm_negative_advice;
2676                 if (likely(dst_ops->link_failure == NULL))
2677                         dst_ops->link_failure = xfrm_link_failure;
2678                 if (likely(dst_ops->neigh_lookup == NULL))
2679                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2680                 if (likely(afinfo->garbage_collect == NULL))
2681                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2682                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2683         }
2684         spin_unlock(&xfrm_policy_afinfo_lock);
2685
2686         rtnl_lock();
2687         for_each_net(net) {
2688                 struct dst_ops *xfrm_dst_ops;
2689
2690                 switch (afinfo->family) {
2691                 case AF_INET:
2692                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2693                         break;
2694 #if IS_ENABLED(CONFIG_IPV6)
2695                 case AF_INET6:
2696                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2697                         break;
2698 #endif
2699                 default:
2700                         BUG();
2701                 }
2702                 *xfrm_dst_ops = *afinfo->dst_ops;
2703         }
2704         rtnl_unlock();
2705
2706         return err;
2707 }
2708 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2709
2710 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2711 {
2712         int err = 0;
2713         if (unlikely(afinfo == NULL))
2714                 return -EINVAL;
2715         if (unlikely(afinfo->family >= NPROTO))
2716                 return -EAFNOSUPPORT;
2717         spin_lock(&xfrm_policy_afinfo_lock);
2718         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2719                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2720                         err = -EINVAL;
2721                 else
2722                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2723                                          NULL);
2724         }
2725         spin_unlock(&xfrm_policy_afinfo_lock);
2726         if (!err) {
2727                 struct dst_ops *dst_ops = afinfo->dst_ops;
2728
2729                 synchronize_rcu();
2730
2731                 dst_ops->kmem_cachep = NULL;
2732                 dst_ops->check = NULL;
2733                 dst_ops->negative_advice = NULL;
2734                 dst_ops->link_failure = NULL;
2735                 afinfo->garbage_collect = NULL;
2736         }
2737         return err;
2738 }
2739 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2740
2741 static void __net_init xfrm_dst_ops_init(struct net *net)
2742 {
2743         struct xfrm_policy_afinfo *afinfo;
2744
2745         rcu_read_lock();
2746         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2747         if (afinfo)
2748                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2749 #if IS_ENABLED(CONFIG_IPV6)
2750         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2751         if (afinfo)
2752                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2753 #endif
2754         rcu_read_unlock();
2755 }
2756
2757 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2758 {
2759         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2760
2761         switch (event) {
2762         case NETDEV_DOWN:
2763                 xfrm_garbage_collect(dev_net(dev));
2764         }
2765         return NOTIFY_DONE;
2766 }
2767
2768 static struct notifier_block xfrm_dev_notifier = {
2769         .notifier_call  = xfrm_dev_event,
2770 };
2771
2772 #ifdef CONFIG_XFRM_STATISTICS
2773 static int __net_init xfrm_statistics_init(struct net *net)
2774 {
2775         int rv;
2776         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2777         if (!net->mib.xfrm_statistics)
2778                 return -ENOMEM;
2779         rv = xfrm_proc_init(net);
2780         if (rv < 0)
2781                 free_percpu(net->mib.xfrm_statistics);
2782         return rv;
2783 }
2784
2785 static void xfrm_statistics_fini(struct net *net)
2786 {
2787         xfrm_proc_fini(net);
2788         free_percpu(net->mib.xfrm_statistics);
2789 }
2790 #else
2791 static int __net_init xfrm_statistics_init(struct net *net)
2792 {
2793         return 0;
2794 }
2795
2796 static void xfrm_statistics_fini(struct net *net)
2797 {
2798 }
2799 #endif
2800
2801 static int __net_init xfrm_policy_init(struct net *net)
2802 {
2803         unsigned int hmask, sz;
2804         int dir;
2805
2806         if (net_eq(net, &init_net))
2807                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2808                                            sizeof(struct xfrm_dst),
2809                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2810                                            NULL);
2811
2812         hmask = 8 - 1;
2813         sz = (hmask+1) * sizeof(struct hlist_head);
2814
2815         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2816         if (!net->xfrm.policy_byidx)
2817                 goto out_byidx;
2818         net->xfrm.policy_idx_hmask = hmask;
2819
2820         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2821                 struct xfrm_policy_hash *htab;
2822
2823                 net->xfrm.policy_count[dir] = 0;
2824                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2825
2826                 htab = &net->xfrm.policy_bydst[dir];
2827                 htab->table = xfrm_hash_alloc(sz);
2828                 if (!htab->table)
2829                         goto out_bydst;
2830                 htab->hmask = hmask;
2831         }
2832
2833         INIT_LIST_HEAD(&net->xfrm.policy_all);
2834         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2835         if (net_eq(net, &init_net))
2836                 register_netdevice_notifier(&xfrm_dev_notifier);
2837         return 0;
2838
2839 out_bydst:
2840         for (dir--; dir >= 0; dir--) {
2841                 struct xfrm_policy_hash *htab;
2842
2843                 htab = &net->xfrm.policy_bydst[dir];
2844                 xfrm_hash_free(htab->table, sz);
2845         }
2846         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2847 out_byidx:
2848         return -ENOMEM;
2849 }
2850
2851 static void xfrm_policy_fini(struct net *net)
2852 {
2853         unsigned int sz;
2854         int dir;
2855
2856         flush_work(&net->xfrm.policy_hash_work);
2857 #ifdef CONFIG_XFRM_SUB_POLICY
2858         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2859 #endif
2860         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2861
2862         WARN_ON(!list_empty(&net->xfrm.policy_all));
2863
2864         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2865                 struct xfrm_policy_hash *htab;
2866
2867                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2868
2869                 htab = &net->xfrm.policy_bydst[dir];
2870                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2871                 WARN_ON(!hlist_empty(htab->table));
2872                 xfrm_hash_free(htab->table, sz);
2873         }
2874
2875         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2876         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2877         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2878 }
2879
2880 static int __net_init xfrm_net_init(struct net *net)
2881 {
2882         int rv;
2883
2884         rv = xfrm_statistics_init(net);
2885         if (rv < 0)
2886                 goto out_statistics;
2887         rv = xfrm_state_init(net);
2888         if (rv < 0)
2889                 goto out_state;
2890         rv = xfrm_policy_init(net);
2891         if (rv < 0)
2892                 goto out_policy;
2893         xfrm_dst_ops_init(net);
2894         rv = xfrm_sysctl_init(net);
2895         if (rv < 0)
2896                 goto out_sysctl;
2897         rv = flow_cache_init(net);
2898         if (rv < 0)
2899                 goto out;
2900
2901         /* Initialize the per-net locks here */
2902         spin_lock_init(&net->xfrm.xfrm_state_lock);
2903         rwlock_init(&net->xfrm.xfrm_policy_lock);
2904         mutex_init(&net->xfrm.xfrm_cfg_mutex);
2905
2906         return 0;
2907
2908 out:
2909         xfrm_sysctl_fini(net);
2910 out_sysctl:
2911         xfrm_policy_fini(net);
2912 out_policy:
2913         xfrm_state_fini(net);
2914 out_state:
2915         xfrm_statistics_fini(net);
2916 out_statistics:
2917         return rv;
2918 }
2919
2920 static void __net_exit xfrm_net_exit(struct net *net)
2921 {
2922         flow_cache_fini(net);
2923         xfrm_sysctl_fini(net);
2924         xfrm_policy_fini(net);
2925         xfrm_state_fini(net);
2926         xfrm_statistics_fini(net);
2927 }
2928
2929 static struct pernet_operations __net_initdata xfrm_net_ops = {
2930         .init = xfrm_net_init,
2931         .exit = xfrm_net_exit,
2932 };
2933
2934 void __init xfrm_init(void)
2935 {
2936         register_pernet_subsys(&xfrm_net_ops);
2937         xfrm_input_init();
2938 }
2939
2940 #ifdef CONFIG_AUDITSYSCALL
2941 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2942                                          struct audit_buffer *audit_buf)
2943 {
2944         struct xfrm_sec_ctx *ctx = xp->security;
2945         struct xfrm_selector *sel = &xp->selector;
2946
2947         if (ctx)
2948                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2949                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2950
2951         switch (sel->family) {
2952         case AF_INET:
2953                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2954                 if (sel->prefixlen_s != 32)
2955                         audit_log_format(audit_buf, " src_prefixlen=%d",
2956                                          sel->prefixlen_s);
2957                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2958                 if (sel->prefixlen_d != 32)
2959                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2960                                          sel->prefixlen_d);
2961                 break;
2962         case AF_INET6:
2963                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2964                 if (sel->prefixlen_s != 128)
2965                         audit_log_format(audit_buf, " src_prefixlen=%d",
2966                                          sel->prefixlen_s);
2967                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2968                 if (sel->prefixlen_d != 128)
2969                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2970                                          sel->prefixlen_d);
2971                 break;
2972         }
2973 }
2974
2975 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
2976 {
2977         struct audit_buffer *audit_buf;
2978
2979         audit_buf = xfrm_audit_start("SPD-add");
2980         if (audit_buf == NULL)
2981                 return;
2982         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2983         audit_log_format(audit_buf, " res=%u", result);
2984         xfrm_audit_common_policyinfo(xp, audit_buf);
2985         audit_log_end(audit_buf);
2986 }
2987 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2988
2989 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2990                               bool task_valid)
2991 {
2992         struct audit_buffer *audit_buf;
2993
2994         audit_buf = xfrm_audit_start("SPD-delete");
2995         if (audit_buf == NULL)
2996                 return;
2997         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2998         audit_log_format(audit_buf, " res=%u", result);
2999         xfrm_audit_common_policyinfo(xp, audit_buf);
3000         audit_log_end(audit_buf);
3001 }
3002 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3003 #endif
3004
3005 #ifdef CONFIG_XFRM_MIGRATE
3006 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3007                                         const struct xfrm_selector *sel_tgt)
3008 {
3009         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3010                 if (sel_tgt->family == sel_cmp->family &&
3011                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3012                                     sel_cmp->family) &&
3013                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3014                                     sel_cmp->family) &&
3015                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3016                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3017                         return true;
3018                 }
3019         } else {
3020                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3021                         return true;
3022                 }
3023         }
3024         return false;
3025 }
3026
3027 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3028                                                     u8 dir, u8 type, struct net *net)
3029 {
3030         struct xfrm_policy *pol, *ret = NULL;
3031         struct hlist_head *chain;
3032         u32 priority = ~0U;
3033
3034         read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3035         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3036         hlist_for_each_entry(pol, chain, bydst) {
3037                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3038                     pol->type == type) {
3039                         ret = pol;
3040                         priority = ret->priority;
3041                         break;
3042                 }
3043         }
3044         chain = &net->xfrm.policy_inexact[dir];
3045         hlist_for_each_entry(pol, chain, bydst) {
3046                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3047                     pol->type == type &&
3048                     pol->priority < priority) {
3049                         ret = pol;
3050                         break;
3051                 }
3052         }
3053
3054         if (ret)
3055                 xfrm_pol_hold(ret);
3056
3057         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3058
3059         return ret;
3060 }
3061
3062 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3063 {
3064         int match = 0;
3065
3066         if (t->mode == m->mode && t->id.proto == m->proto &&
3067             (m->reqid == 0 || t->reqid == m->reqid)) {
3068                 switch (t->mode) {
3069                 case XFRM_MODE_TUNNEL:
3070                 case XFRM_MODE_BEET:
3071                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3072                                             m->old_family) &&
3073                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3074                                             m->old_family)) {
3075                                 match = 1;
3076                         }
3077                         break;
3078                 case XFRM_MODE_TRANSPORT:
3079                         /* in case of transport mode, template does not store
3080                            any IP addresses, hence we just compare mode and
3081                            protocol */
3082                         match = 1;
3083                         break;
3084                 default:
3085                         break;
3086                 }
3087         }
3088         return match;
3089 }
3090
3091 /* update endpoint address(es) of template(s) */
3092 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3093                                struct xfrm_migrate *m, int num_migrate)
3094 {
3095         struct xfrm_migrate *mp;
3096         int i, j, n = 0;
3097
3098         write_lock_bh(&pol->lock);
3099         if (unlikely(pol->walk.dead)) {
3100                 /* target policy has been deleted */
3101                 write_unlock_bh(&pol->lock);
3102                 return -ENOENT;
3103         }
3104
3105         for (i = 0; i < pol->xfrm_nr; i++) {
3106                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3107                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3108                                 continue;
3109                         n++;
3110                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3111                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3112                                 continue;
3113                         /* update endpoints */
3114                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3115                                sizeof(pol->xfrm_vec[i].id.daddr));
3116                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3117                                sizeof(pol->xfrm_vec[i].saddr));
3118                         pol->xfrm_vec[i].encap_family = mp->new_family;
3119                         /* flush bundles */
3120                         atomic_inc(&pol->genid);
3121                 }
3122         }
3123
3124         write_unlock_bh(&pol->lock);
3125
3126         if (!n)
3127                 return -ENODATA;
3128
3129         return 0;
3130 }
3131
3132 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3133 {
3134         int i, j;
3135
3136         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3137                 return -EINVAL;
3138
3139         for (i = 0; i < num_migrate; i++) {
3140                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3141                                     m[i].old_family) &&
3142                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3143                                     m[i].old_family))
3144                         return -EINVAL;
3145                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3146                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3147                         return -EINVAL;
3148
3149                 /* check if there is any duplicated entry */
3150                 for (j = i + 1; j < num_migrate; j++) {
3151                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3152                                     sizeof(m[i].old_daddr)) &&
3153                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3154                                     sizeof(m[i].old_saddr)) &&
3155                             m[i].proto == m[j].proto &&
3156                             m[i].mode == m[j].mode &&
3157                             m[i].reqid == m[j].reqid &&
3158                             m[i].old_family == m[j].old_family)
3159                                 return -EINVAL;
3160                 }
3161         }
3162
3163         return 0;
3164 }
3165
3166 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3167                  struct xfrm_migrate *m, int num_migrate,
3168                  struct xfrm_kmaddress *k, struct net *net)
3169 {
3170         int i, err, nx_cur = 0, nx_new = 0;
3171         struct xfrm_policy *pol = NULL;
3172         struct xfrm_state *x, *xc;
3173         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3174         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3175         struct xfrm_migrate *mp;
3176
3177         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3178                 goto out;
3179
3180         /* Stage 1 - find policy */
3181         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3182                 err = -ENOENT;
3183                 goto out;
3184         }
3185
3186         /* Stage 2 - find and update state(s) */
3187         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3188                 if ((x = xfrm_migrate_state_find(mp, net))) {
3189                         x_cur[nx_cur] = x;
3190                         nx_cur++;
3191                         if ((xc = xfrm_state_migrate(x, mp))) {
3192                                 x_new[nx_new] = xc;
3193                                 nx_new++;
3194                         } else {
3195                                 err = -ENODATA;
3196                                 goto restore_state;
3197                         }
3198                 }
3199         }
3200
3201         /* Stage 3 - update policy */
3202         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3203                 goto restore_state;
3204
3205         /* Stage 4 - delete old state(s) */
3206         if (nx_cur) {
3207                 xfrm_states_put(x_cur, nx_cur);
3208                 xfrm_states_delete(x_cur, nx_cur);
3209         }
3210
3211         /* Stage 5 - announce */
3212         km_migrate(sel, dir, type, m, num_migrate, k);
3213
3214         xfrm_pol_put(pol);
3215
3216         return 0;
3217 out:
3218         return err;
3219
3220 restore_state:
3221         if (pol)
3222                 xfrm_pol_put(pol);
3223         if (nx_cur)
3224                 xfrm_states_put(x_cur, nx_cur);
3225         if (nx_new)
3226                 xfrm_states_delete(x_new, nx_new);
3227
3228         return err;
3229 }
3230 EXPORT_SYMBOL(xfrm_migrate);
3231 #endif