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/cpu.h>
  28 #include <linux/audit.h>
  29 #include <net/dst.h>
  30 #include <net/flow.h>
  31 #include <net/xfrm.h>
  32 #include <net/ip.h>
  33 #ifdef CONFIG_XFRM_STATISTICS
  34 #include <net/snmp.h>
  35 #endif
  36
  37 #include "xfrm_hash.h"
  38
  39 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  40 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  41 #define XFRM_MAX_QUEUE_LEN      100
  42
  43 struct xfrm_flo {
  44         struct dst_entry *dst_orig;
  45         u8 flags;
  46 };
  47
  48 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
  49 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
  50
  51 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  52 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
  53                                                 __read_mostly;
  54
  55 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
  56 static __read_mostly seqcount_t xfrm_policy_hash_generation;
  57
  58 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
  59 static int stale_bundle(struct dst_entry *dst);
  60 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  61 static void xfrm_policy_queue_process(struct timer_list *t);
  62
  63 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
  64 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  65                                                 int dir);
  66
  67 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
  68 {
  69         return refcount_inc_not_zero(&policy->refcnt);
  70 }
  71
  72 static inline bool
  73 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  74 {
  75         const struct flowi4 *fl4 = &fl->u.ip4;
  76
  77         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  78                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  79                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  80                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  81                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  82                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  83 }
  84
  85 static inline bool
  86 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  87 {
  88         const struct flowi6 *fl6 = &fl->u.ip6;
  89
  90         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  91                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  92                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  93                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  94                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  95                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  96 }
  97
  98 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  99                          unsigned short family)
 100 {
 101         switch (family) {
 102         case AF_INET:
 103                 return __xfrm4_selector_match(sel, fl);
 104         case AF_INET6:
 105                 return __xfrm6_selector_match(sel, fl);
 106         }
 107         return false;
 108 }
 109
 110 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 111 {
 112         const struct xfrm_policy_afinfo *afinfo;
 113
 114         if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 115                 return NULL;
 116         rcu_read_lock();
 117         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 118         if (unlikely(!afinfo))
 119                 rcu_read_unlock();
 120         return afinfo;
 121 }
 122
 123 /* Called with rcu_read_lock(). */
 124 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
 125 {
 126         return rcu_dereference(xfrm_if_cb);
 127 }
 128
 129 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
 130                                     const xfrm_address_t *saddr,
 131                                     const xfrm_address_t *daddr,
 132                                     int family, u32 mark)
 133 {
 134         const struct xfrm_policy_afinfo *afinfo;
 135         struct dst_entry *dst;
 136
 137         afinfo = xfrm_policy_get_afinfo(family);
 138         if (unlikely(afinfo == NULL))
 139                 return ERR_PTR(-EAFNOSUPPORT);
 140
 141         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
 142
 143         rcu_read_unlock();
 144
 145         return dst;
 146 }
 147 EXPORT_SYMBOL(__xfrm_dst_lookup);
 148
 149 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 150                                                 int tos, int oif,
 151                                                 xfrm_address_t *prev_saddr,
 152                                                 xfrm_address_t *prev_daddr,
 153                                                 int family, u32 mark)
 154 {
 155         struct net *net = xs_net(x);
 156         xfrm_address_t *saddr = &x->props.saddr;
 157         xfrm_address_t *daddr = &x->id.daddr;
 158         struct dst_entry *dst;
 159
 160         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 161                 saddr = x->coaddr;
 162                 daddr = prev_daddr;
 163         }
 164         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 165                 saddr = prev_saddr;
 166                 daddr = x->coaddr;
 167         }
 168
 169         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
 170
 171         if (!IS_ERR(dst)) {
 172                 if (prev_saddr != saddr)
 173                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 174                 if (prev_daddr != daddr)
 175                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 176         }
 177
 178         return dst;
 179 }
 180
 181 static inline unsigned long make_jiffies(long secs)
 182 {
 183         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 184                 return MAX_SCHEDULE_TIMEOUT-1;
 185         else
 186                 return secs*HZ;
 187 }
 188
 189 static void xfrm_policy_timer(struct timer_list *t)
 190 {
 191         struct xfrm_policy *xp = from_timer(xp, t, timer);
 192         time64_t now = ktime_get_real_seconds();
 193         time64_t next = TIME64_MAX;
 194         int warn = 0;
 195         int dir;
 196
 197         read_lock(&xp->lock);
 198
 199         if (unlikely(xp->walk.dead))
 200                 goto out;
 201
 202         dir = xfrm_policy_id2dir(xp->index);
 203
 204         if (xp->lft.hard_add_expires_seconds) {
 205                 time64_t tmo = xp->lft.hard_add_expires_seconds +
 206                         xp->curlft.add_time - now;
 207                 if (tmo <= 0)
 208                         goto expired;
 209                 if (tmo < next)
 210                         next = tmo;
 211         }
 212         if (xp->lft.hard_use_expires_seconds) {
 213                 time64_t tmo = xp->lft.hard_use_expires_seconds +
 214                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 215                 if (tmo <= 0)
 216                         goto expired;
 217                 if (tmo < next)
 218                         next = tmo;
 219         }
 220         if (xp->lft.soft_add_expires_seconds) {
 221                 time64_t tmo = xp->lft.soft_add_expires_seconds +
 222                         xp->curlft.add_time - now;
 223                 if (tmo <= 0) {
 224                         warn = 1;
 225                         tmo = XFRM_KM_TIMEOUT;
 226                 }
 227                 if (tmo < next)
 228                         next = tmo;
 229         }
 230         if (xp->lft.soft_use_expires_seconds) {
 231                 time64_t tmo = xp->lft.soft_use_expires_seconds +
 232                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 233                 if (tmo <= 0) {
 234                         warn = 1;
 235                         tmo = XFRM_KM_TIMEOUT;
 236                 }
 237                 if (tmo < next)
 238                         next = tmo;
 239         }
 240
 241         if (warn)
 242                 km_policy_expired(xp, dir, 0, 0);
 243         if (next != TIME64_MAX &&
 244             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 245                 xfrm_pol_hold(xp);
 246
 247 out:
 248         read_unlock(&xp->lock);
 249         xfrm_pol_put(xp);
 250         return;
 251
 252 expired:
 253         read_unlock(&xp->lock);
 254         if (!xfrm_policy_delete(xp, dir))
 255                 km_policy_expired(xp, dir, 1, 0);
 256         xfrm_pol_put(xp);
 257 }
 258
 259 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 260  * SPD calls.
 261  */
 262
 263 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 264 {
 265         struct xfrm_policy *policy;
 266
 267         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 268
 269         if (policy) {
 270                 write_pnet(&policy->xp_net, net);
 271                 INIT_LIST_HEAD(&policy->walk.all);
 272                 INIT_HLIST_NODE(&policy->bydst);
 273                 INIT_HLIST_NODE(&policy->byidx);
 274                 rwlock_init(&policy->lock);
 275                 refcount_set(&policy->refcnt, 1);
 276                 skb_queue_head_init(&policy->polq.hold_queue);
 277                 timer_setup(&policy->timer, xfrm_policy_timer, 0);
 278                 timer_setup(&policy->polq.hold_timer,
 279                             xfrm_policy_queue_process, 0);
 280         }
 281         return policy;
 282 }
 283 EXPORT_SYMBOL(xfrm_policy_alloc);
 284
 285 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 286 {
 287         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 288
 289         security_xfrm_policy_free(policy->security);
 290         kfree(policy);
 291 }
 292
 293 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 294
 295 void xfrm_policy_destroy(struct xfrm_policy *policy)
 296 {
 297         BUG_ON(!policy->walk.dead);
 298
 299         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 300                 BUG();
 301
 302         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 303 }
 304 EXPORT_SYMBOL(xfrm_policy_destroy);
 305
 306 /* Rule must be locked. Release descendant resources, announce
 307  * entry dead. The rule must be unlinked from lists to the moment.
 308  */
 309
 310 static void xfrm_policy_kill(struct xfrm_policy *policy)
 311 {
 312         write_lock_bh(&policy->lock);
 313         policy->walk.dead = 1;
 314         write_unlock_bh(&policy->lock);
 315
 316         atomic_inc(&policy->genid);
 317
 318         if (del_timer(&policy->polq.hold_timer))
 319                 xfrm_pol_put(policy);
 320         skb_queue_purge(&policy->polq.hold_queue);
 321
 322         if (del_timer(&policy->timer))
 323                 xfrm_pol_put(policy);
 324
 325         xfrm_pol_put(policy);
 326 }
 327
 328 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 329
 330 static inline unsigned int idx_hash(struct net *net, u32 index)
 331 {
 332         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 333 }
 334
 335 /* calculate policy hash thresholds */
 336 static void __get_hash_thresh(struct net *net,
 337                               unsigned short family, int dir,
 338                               u8 *dbits, u8 *sbits)
 339 {
 340         switch (family) {
 341         case AF_INET:
 342                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 343                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 344                 break;
 345
 346         case AF_INET6:
 347                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 348                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 349                 break;
 350
 351         default:
 352                 *dbits = 0;
 353                 *sbits = 0;
 354         }
 355 }
 356
 357 static struct hlist_head *policy_hash_bysel(struct net *net,
 358                                             const struct xfrm_selector *sel,
 359                                             unsigned short family, int dir)
 360 {
 361         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 362         unsigned int hash;
 363         u8 dbits;
 364         u8 sbits;
 365
 366         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 367         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 368
 369         if (hash == hmask + 1)
 370                 return &net->xfrm.policy_inexact[dir];
 371
 372         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 373                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 374 }
 375
 376 static struct hlist_head *policy_hash_direct(struct net *net,
 377                                              const xfrm_address_t *daddr,
 378                                              const xfrm_address_t *saddr,
 379                                              unsigned short family, int dir)
 380 {
 381         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 382         unsigned int hash;
 383         u8 dbits;
 384         u8 sbits;
 385
 386         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 387         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 388
 389         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 390                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 391 }
 392
 393 static void xfrm_dst_hash_transfer(struct net *net,
 394                                    struct hlist_head *list,
 395                                    struct hlist_head *ndsttable,
 396                                    unsigned int nhashmask,
 397                                    int dir)
 398 {
 399         struct hlist_node *tmp, *entry0 = NULL;
 400         struct xfrm_policy *pol;
 401         unsigned int h0 = 0;
 402         u8 dbits;
 403         u8 sbits;
 404
 405 redo:
 406         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 407                 unsigned int h;
 408
 409                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 410                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 411                                 pol->family, nhashmask, dbits, sbits);
 412                 if (!entry0) {
 413                         hlist_del_rcu(&pol->bydst);
 414                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 415                         h0 = h;
 416                 } else {
 417                         if (h != h0)
 418                                 continue;
 419                         hlist_del_rcu(&pol->bydst);
 420                         hlist_add_behind_rcu(&pol->bydst, entry0);
 421                 }
 422                 entry0 = &pol->bydst;
 423         }
 424         if (!hlist_empty(list)) {
 425                 entry0 = NULL;
 426                 goto redo;
 427         }
 428 }
 429
 430 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 431                                    struct hlist_head *nidxtable,
 432                                    unsigned int nhashmask)
 433 {
 434         struct hlist_node *tmp;
 435         struct xfrm_policy *pol;
 436
 437         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 438                 unsigned int h;
 439
 440                 h = __idx_hash(pol->index, nhashmask);
 441                 hlist_add_head(&pol->byidx, nidxtable+h);
 442         }
 443 }
 444
 445 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 446 {
 447         return ((old_hmask + 1) << 1) - 1;
 448 }
 449
 450 static void xfrm_bydst_resize(struct net *net, int dir)
 451 {
 452         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 453         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 454         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 455         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 456         struct hlist_head *odst;
 457         int i;
 458
 459         if (!ndst)
 460                 return;
 461
 462         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 463         write_seqcount_begin(&xfrm_policy_hash_generation);
 464
 465         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 466                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 467
 468         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 469                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 470
 471         for (i = hmask; i >= 0; i--)
 472                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 473
 474         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 475         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 476
 477         write_seqcount_end(&xfrm_policy_hash_generation);
 478         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 479
 480         synchronize_rcu();
 481
 482         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 483 }
 484
 485 static void xfrm_byidx_resize(struct net *net, int total)
 486 {
 487         unsigned int hmask = net->xfrm.policy_idx_hmask;
 488         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 489         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 490         struct hlist_head *oidx = net->xfrm.policy_byidx;
 491         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 492         int i;
 493
 494         if (!nidx)
 495                 return;
 496
 497         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 498
 499         for (i = hmask; i >= 0; i--)
 500                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 501
 502         net->xfrm.policy_byidx = nidx;
 503         net->xfrm.policy_idx_hmask = nhashmask;
 504
 505         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 506
 507         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 508 }
 509
 510 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 511 {
 512         unsigned int cnt = net->xfrm.policy_count[dir];
 513         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 514
 515         if (total)
 516                 *total += cnt;
 517
 518         if ((hmask + 1) < xfrm_policy_hashmax &&
 519             cnt > hmask)
 520                 return 1;
 521
 522         return 0;
 523 }
 524
 525 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 526 {
 527         unsigned int hmask = net->xfrm.policy_idx_hmask;
 528
 529         if ((hmask + 1) < xfrm_policy_hashmax &&
 530             total > hmask)
 531                 return 1;
 532
 533         return 0;
 534 }
 535
 536 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 537 {
 538         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 539         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 540         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 541         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 542         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 543         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 544         si->spdhcnt = net->xfrm.policy_idx_hmask;
 545         si->spdhmcnt = xfrm_policy_hashmax;
 546 }
 547 EXPORT_SYMBOL(xfrm_spd_getinfo);
 548
 549 static DEFINE_MUTEX(hash_resize_mutex);
 550 static void xfrm_hash_resize(struct work_struct *work)
 551 {
 552         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 553         int dir, total;
 554
 555         mutex_lock(&hash_resize_mutex);
 556
 557         total = 0;
 558         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 559                 if (xfrm_bydst_should_resize(net, dir, &total))
 560                         xfrm_bydst_resize(net, dir);
 561         }
 562         if (xfrm_byidx_should_resize(net, total))
 563                 xfrm_byidx_resize(net, total);
 564
 565         mutex_unlock(&hash_resize_mutex);
 566 }
 567
 568 static void xfrm_hash_rebuild(struct work_struct *work)
 569 {
 570         struct net *net = container_of(work, struct net,
 571                                        xfrm.policy_hthresh.work);
 572         unsigned int hmask;
 573         struct xfrm_policy *pol;
 574         struct xfrm_policy *policy;
 575         struct hlist_head *chain;
 576         struct hlist_head *odst;
 577         struct hlist_node *newpos;
 578         int i;
 579         int dir;
 580         unsigned seq;
 581         u8 lbits4, rbits4, lbits6, rbits6;
 582
 583         mutex_lock(&hash_resize_mutex);
 584
 585         /* read selector prefixlen thresholds */
 586         do {
 587                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
 588
 589                 lbits4 = net->xfrm.policy_hthresh.lbits4;
 590                 rbits4 = net->xfrm.policy_hthresh.rbits4;
 591                 lbits6 = net->xfrm.policy_hthresh.lbits6;
 592                 rbits6 = net->xfrm.policy_hthresh.rbits6;
 593         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
 594
 595         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 596
 597         /* reset the bydst and inexact table in all directions */
 598         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 599                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
 600                 hmask = net->xfrm.policy_bydst[dir].hmask;
 601                 odst = net->xfrm.policy_bydst[dir].table;
 602                 for (i = hmask; i >= 0; i--)
 603                         INIT_HLIST_HEAD(odst + i);
 604                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
 605                         /* dir out => dst = remote, src = local */
 606                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
 607                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
 608                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
 609                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
 610                 } else {
 611                         /* dir in/fwd => dst = local, src = remote */
 612                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
 613                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
 614                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
 615                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
 616                 }
 617         }
 618
 619         /* re-insert all policies by order of creation */
 620         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 621                 if (policy->walk.dead ||
 622                     xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
 623                         /* skip socket policies */
 624                         continue;
 625                 }
 626                 newpos = NULL;
 627                 chain = policy_hash_bysel(net, &policy->selector,
 628                                           policy->family,
 629                                           xfrm_policy_id2dir(policy->index));
 630                 hlist_for_each_entry(pol, chain, bydst) {
 631                         if (policy->priority >= pol->priority)
 632                                 newpos = &pol->bydst;
 633                         else
 634                                 break;
 635                 }
 636                 if (newpos)
 637                         hlist_add_behind_rcu(&policy->bydst, newpos);
 638                 else
 639                         hlist_add_head_rcu(&policy->bydst, chain);
 640         }
 641
 642         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 643
 644         mutex_unlock(&hash_resize_mutex);
 645 }
 646
 647 void xfrm_policy_hash_rebuild(struct net *net)
 648 {
 649         schedule_work(&net->xfrm.policy_hthresh.work);
 650 }
 651 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 652
 653 /* Generate new index... KAME seems to generate them ordered by cost
 654  * of an absolute inpredictability of ordering of rules. This will not pass. */
 655 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 656 {
 657         static u32 idx_generator;
 658
 659         for (;;) {
 660                 struct hlist_head *list;
 661                 struct xfrm_policy *p;
 662                 u32 idx;
 663                 int found;
 664
 665                 if (!index) {
 666                         idx = (idx_generator | dir);
 667                         idx_generator += 8;
 668                 } else {
 669                         idx = index;
 670                         index = 0;
 671                 }
 672
 673                 if (idx == 0)
 674                         idx = 8;
 675                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 676                 found = 0;
 677                 hlist_for_each_entry(p, list, byidx) {
 678                         if (p->index == idx) {
 679                                 found = 1;
 680                                 break;
 681                         }
 682                 }
 683                 if (!found)
 684                         return idx;
 685         }
 686 }
 687
 688 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 689 {
 690         u32 *p1 = (u32 *) s1;
 691         u32 *p2 = (u32 *) s2;
 692         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 693         int i;
 694
 695         for (i = 0; i < len; i++) {
 696                 if (p1[i] != p2[i])
 697                         return 1;
 698         }
 699
 700         return 0;
 701 }
 702
 703 static void xfrm_policy_requeue(struct xfrm_policy *old,
 704                                 struct xfrm_policy *new)
 705 {
 706         struct xfrm_policy_queue *pq = &old->polq;
 707         struct sk_buff_head list;
 708
 709         if (skb_queue_empty(&pq->hold_queue))
 710                 return;
 711
 712         __skb_queue_head_init(&list);
 713
 714         spin_lock_bh(&pq->hold_queue.lock);
 715         skb_queue_splice_init(&pq->hold_queue, &list);
 716         if (del_timer(&pq->hold_timer))
 717                 xfrm_pol_put(old);
 718         spin_unlock_bh(&pq->hold_queue.lock);
 719
 720         pq = &new->polq;
 721
 722         spin_lock_bh(&pq->hold_queue.lock);
 723         skb_queue_splice(&list, &pq->hold_queue);
 724         pq->timeout = XFRM_QUEUE_TMO_MIN;
 725         if (!mod_timer(&pq->hold_timer, jiffies))
 726                 xfrm_pol_hold(new);
 727         spin_unlock_bh(&pq->hold_queue.lock);
 728 }
 729
 730 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
 731                                           struct xfrm_policy *pol)
 732 {
 733         return mark->v == pol->mark.v && mark->m == pol->mark.m;
 734 }
 735
 736 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 737 {
 738         struct net *net = xp_net(policy);
 739         struct xfrm_policy *pol;
 740         struct xfrm_policy *delpol;
 741         struct hlist_head *chain;
 742         struct hlist_node *newpos;
 743
 744         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 745         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 746         delpol = NULL;
 747         newpos = NULL;
 748         hlist_for_each_entry(pol, chain, bydst) {
 749                 if (pol->type == policy->type &&
 750                     pol->if_id == policy->if_id &&
 751                     !selector_cmp(&pol->selector, &policy->selector) &&
 752                     xfrm_policy_mark_match(&policy->mark, pol) &&
 753                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 754                     !WARN_ON(delpol)) {
 755                         if (excl) {
 756                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 757                                 return -EEXIST;
 758                         }
 759                         delpol = pol;
 760                         if (policy->priority > pol->priority)
 761                                 continue;
 762                 } else if (policy->priority >= pol->priority) {
 763                         newpos = &pol->bydst;
 764                         continue;
 765                 }
 766                 if (delpol)
 767                         break;
 768         }
 769         if (newpos)
 770                 hlist_add_behind_rcu(&policy->bydst, newpos);
 771         else
 772                 hlist_add_head_rcu(&policy->bydst, chain);
 773         __xfrm_policy_link(policy, dir);
 774
 775         /* After previous checking, family can either be AF_INET or AF_INET6 */
 776         if (policy->family == AF_INET)
 777                 rt_genid_bump_ipv4(net);
 778         else
 779                 rt_genid_bump_ipv6(net);
 780
 781         if (delpol) {
 782                 xfrm_policy_requeue(delpol, policy);
 783                 __xfrm_policy_unlink(delpol, dir);
 784         }
 785         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 786         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 787         policy->curlft.add_time = ktime_get_real_seconds();
 788         policy->curlft.use_time = 0;
 789         if (!mod_timer(&policy->timer, jiffies + HZ))
 790                 xfrm_pol_hold(policy);
 791         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 792
 793         if (delpol)
 794                 xfrm_policy_kill(delpol);
 795         else if (xfrm_bydst_should_resize(net, dir, NULL))
 796                 schedule_work(&net->xfrm.policy_hash_work);
 797
 798         return 0;
 799 }
 800 EXPORT_SYMBOL(xfrm_policy_insert);
 801
 802 struct xfrm_policy *
 803 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
 804                       u8 type, int dir, struct xfrm_selector *sel,
 805                       struct xfrm_sec_ctx *ctx, int delete, int *err)
 806 {
 807         struct xfrm_policy *pol, *ret;
 808         struct hlist_head *chain;
 809
 810         *err = 0;
 811         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 812         chain = policy_hash_bysel(net, sel, sel->family, dir);
 813         ret = NULL;
 814         hlist_for_each_entry(pol, chain, bydst) {
 815                 if (pol->type == type &&
 816                     pol->if_id == if_id &&
 817                     xfrm_policy_mark_match(mark, pol) &&
 818                     !selector_cmp(sel, &pol->selector) &&
 819                     xfrm_sec_ctx_match(ctx, pol->security)) {
 820                         xfrm_pol_hold(pol);
 821                         if (delete) {
 822                                 *err = security_xfrm_policy_delete(
 823                                                                 pol->security);
 824                                 if (*err) {
 825                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 826                                         return pol;
 827                                 }
 828                                 __xfrm_policy_unlink(pol, dir);
 829                         }
 830                         ret = pol;
 831                         break;
 832                 }
 833         }
 834         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 835
 836         if (ret && delete)
 837                 xfrm_policy_kill(ret);
 838         return ret;
 839 }
 840 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 841
 842 struct xfrm_policy *
 843 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
 844                  u8 type, int dir, u32 id, int delete, int *err)
 845 {
 846         struct xfrm_policy *pol, *ret;
 847         struct hlist_head *chain;
 848
 849         *err = -ENOENT;
 850         if (xfrm_policy_id2dir(id) != dir)
 851                 return NULL;
 852
 853         *err = 0;
 854         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 855         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 856         ret = NULL;
 857         hlist_for_each_entry(pol, chain, byidx) {
 858                 if (pol->type == type && pol->index == id &&
 859                     pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
 860                         xfrm_pol_hold(pol);
 861                         if (delete) {
 862                                 *err = security_xfrm_policy_delete(
 863                                                                 pol->security);
 864                                 if (*err) {
 865                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 866                                         return pol;
 867                                 }
 868                                 __xfrm_policy_unlink(pol, dir);
 869                         }
 870                         ret = pol;
 871                         break;
 872                 }
 873         }
 874         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 875
 876         if (ret && delete)
 877                 xfrm_policy_kill(ret);
 878         return ret;
 879 }
 880 EXPORT_SYMBOL(xfrm_policy_byid);
 881
 882 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 883 static inline int
 884 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 885 {
 886         int dir, err = 0;
 887
 888         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 889                 struct xfrm_policy *pol;
 890                 int i;
 891
 892                 hlist_for_each_entry(pol,
 893                                      &net->xfrm.policy_inexact[dir], bydst) {
 894                         if (pol->type != type)
 895                                 continue;
 896                         err = security_xfrm_policy_delete(pol->security);
 897                         if (err) {
 898                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 899                                 return err;
 900                         }
 901                 }
 902                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 903                         hlist_for_each_entry(pol,
 904                                              net->xfrm.policy_bydst[dir].table + i,
 905                                              bydst) {
 906                                 if (pol->type != type)
 907                                         continue;
 908                                 err = security_xfrm_policy_delete(
 909                                                                 pol->security);
 910                                 if (err) {
 911                                         xfrm_audit_policy_delete(pol, 0,
 912                                                                  task_valid);
 913                                         return err;
 914                                 }
 915                         }
 916                 }
 917         }
 918         return err;
 919 }
 920 #else
 921 static inline int
 922 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 923 {
 924         return 0;
 925 }
 926 #endif
 927
 928 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 929 {
 930         int dir, err = 0, cnt = 0;
 931
 932         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 933
 934         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 935         if (err)
 936                 goto out;
 937
 938         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 939                 struct xfrm_policy *pol;
 940                 int i;
 941
 942         again1:
 943                 hlist_for_each_entry(pol,
 944                                      &net->xfrm.policy_inexact[dir], bydst) {
 945                         if (pol->type != type)
 946                                 continue;
 947                         __xfrm_policy_unlink(pol, dir);
 948                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 949                         cnt++;
 950
 951                         xfrm_audit_policy_delete(pol, 1, task_valid);
 952
 953                         xfrm_policy_kill(pol);
 954
 955                         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 956                         goto again1;
 957                 }
 958
 959                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 960         again2:
 961                         hlist_for_each_entry(pol,
 962                                              net->xfrm.policy_bydst[dir].table + i,
 963                                              bydst) {
 964                                 if (pol->type != type)
 965                                         continue;
 966                                 __xfrm_policy_unlink(pol, dir);
 967                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 968                                 cnt++;
 969
 970                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 971                                 xfrm_policy_kill(pol);
 972
 973                                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 974                                 goto again2;
 975                         }
 976                 }
 977
 978         }
 979         if (!cnt)
 980                 err = -ESRCH;
 981 out:
 982         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 983         return err;
 984 }
 985 EXPORT_SYMBOL(xfrm_policy_flush);
 986
 987 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 988                      int (*func)(struct xfrm_policy *, int, int, void*),
 989                      void *data)
 990 {
 991         struct xfrm_policy *pol;
 992         struct xfrm_policy_walk_entry *x;
 993         int error = 0;
 994
 995         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 996             walk->type != XFRM_POLICY_TYPE_ANY)
 997                 return -EINVAL;
 998
 999         if (list_empty(&walk->walk.all) && walk->seq != 0)
1000                 return 0;
1001
1002         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1003         if (list_empty(&walk->walk.all))
1004                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1005         else
1006                 x = list_first_entry(&walk->walk.all,
1007                                      struct xfrm_policy_walk_entry, all);
1008
1009         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1010                 if (x->dead)
1011                         continue;
1012                 pol = container_of(x, struct xfrm_policy, walk);
1013                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1014                     walk->type != pol->type)
1015                         continue;
1016                 error = func(pol, xfrm_policy_id2dir(pol->index),
1017                              walk->seq, data);
1018                 if (error) {
1019                         list_move_tail(&walk->walk.all, &x->all);
1020                         goto out;
1021                 }
1022                 walk->seq++;
1023         }
1024         if (walk->seq == 0) {
1025                 error = -ENOENT;
1026                 goto out;
1027         }
1028         list_del_init(&walk->walk.all);
1029 out:
1030         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1031         return error;
1032 }
1033 EXPORT_SYMBOL(xfrm_policy_walk);
1034
1035 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1036 {
1037         INIT_LIST_HEAD(&walk->walk.all);
1038         walk->walk.dead = 1;
1039         walk->type = type;
1040         walk->seq = 0;
1041 }
1042 EXPORT_SYMBOL(xfrm_policy_walk_init);
1043
1044 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1045 {
1046         if (list_empty(&walk->walk.all))
1047                 return;
1048
1049         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1050         list_del(&walk->walk.all);
1051         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1052 }
1053 EXPORT_SYMBOL(xfrm_policy_walk_done);
1054
1055 /*
1056  * Find policy to apply to this flow.
1057  *
1058  * Returns 0 if policy found, else an -errno.
1059  */
1060 static int xfrm_policy_match(const struct xfrm_policy *pol,
1061                              const struct flowi *fl,
1062                              u8 type, u16 family, int dir, u32 if_id)
1063 {
1064         const struct xfrm_selector *sel = &pol->selector;
1065         int ret = -ESRCH;
1066         bool match;
1067
1068         if (pol->family != family ||
1069             pol->if_id != if_id ||
1070             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1071             pol->type != type)
1072                 return ret;
1073
1074         match = xfrm_selector_match(sel, fl, family);
1075         if (match)
1076                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1077                                                   dir);
1078
1079         return ret;
1080 }
1081
1082 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1083                                                      const struct flowi *fl,
1084                                                      u16 family, u8 dir,
1085                                                      u32 if_id)
1086 {
1087         int err;
1088         struct xfrm_policy *pol, *ret;
1089         const xfrm_address_t *daddr, *saddr;
1090         struct hlist_head *chain;
1091         unsigned int sequence;
1092         u32 priority;
1093
1094         daddr = xfrm_flowi_daddr(fl, family);
1095         saddr = xfrm_flowi_saddr(fl, family);
1096         if (unlikely(!daddr || !saddr))
1097                 return NULL;
1098
1099         rcu_read_lock();
1100  retry:
1101         do {
1102                 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1103                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1104         } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1105
1106         priority = ~0U;
1107         ret = NULL;
1108         hlist_for_each_entry_rcu(pol, chain, bydst) {
1109                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
1110                 if (err) {
1111                         if (err == -ESRCH)
1112                                 continue;
1113                         else {
1114                                 ret = ERR_PTR(err);
1115                                 goto fail;
1116                         }
1117                 } else {
1118                         ret = pol;
1119                         priority = ret->priority;
1120                         break;
1121                 }
1122         }
1123         chain = &net->xfrm.policy_inexact[dir];
1124         hlist_for_each_entry_rcu(pol, chain, bydst) {
1125                 if ((pol->priority >= priority) && ret)
1126                         break;
1127
1128                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
1129                 if (err) {
1130                         if (err == -ESRCH)
1131                                 continue;
1132                         else {
1133                                 ret = ERR_PTR(err);
1134                                 goto fail;
1135                         }
1136                 } else {
1137                         ret = pol;
1138                         break;
1139                 }
1140         }
1141
1142         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1143                 goto retry;
1144
1145         if (ret && !xfrm_pol_hold_rcu(ret))
1146                 goto retry;
1147 fail:
1148         rcu_read_unlock();
1149
1150         return ret;
1151 }
1152
1153 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
1154                                               const struct flowi *fl,
1155                                               u16 family, u8 dir, u32 if_id)
1156 {
1157 #ifdef CONFIG_XFRM_SUB_POLICY
1158         struct xfrm_policy *pol;
1159
1160         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
1161                                         dir, if_id);
1162         if (pol != NULL)
1163                 return pol;
1164 #endif
1165         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
1166                                          dir, if_id);
1167 }
1168
1169 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1170                                                  const struct flowi *fl,
1171                                                  u16 family, u32 if_id)
1172 {
1173         struct xfrm_policy *pol;
1174
1175         rcu_read_lock();
1176  again:
1177         pol = rcu_dereference(sk->sk_policy[dir]);
1178         if (pol != NULL) {
1179                 bool match;
1180                 int err = 0;
1181
1182                 if (pol->family != family) {
1183                         pol = NULL;
1184                         goto out;
1185                 }
1186
1187                 match = xfrm_selector_match(&pol->selector, fl, family);
1188                 if (match) {
1189                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
1190                             pol->if_id != if_id) {
1191                                 pol = NULL;
1192                                 goto out;
1193                         }
1194                         err = security_xfrm_policy_lookup(pol->security,
1195                                                       fl->flowi_secid,
1196                                                       dir);
1197                         if (!err) {
1198                                 if (!xfrm_pol_hold_rcu(pol))
1199                                         goto again;
1200                         } else if (err == -ESRCH) {
1201                                 pol = NULL;
1202                         } else {
1203                                 pol = ERR_PTR(err);
1204                         }
1205                 } else
1206                         pol = NULL;
1207         }
1208 out:
1209         rcu_read_unlock();
1210         return pol;
1211 }
1212
1213 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1214 {
1215         struct net *net = xp_net(pol);
1216
1217         list_add(&pol->walk.all, &net->xfrm.policy_all);
1218         net->xfrm.policy_count[dir]++;
1219         xfrm_pol_hold(pol);
1220 }
1221
1222 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1223                                                 int dir)
1224 {
1225         struct net *net = xp_net(pol);
1226
1227         if (list_empty(&pol->walk.all))
1228                 return NULL;
1229
1230         /* Socket policies are not hashed. */
1231         if (!hlist_unhashed(&pol->bydst)) {
1232                 hlist_del_rcu(&pol->bydst);
1233                 hlist_del(&pol->byidx);
1234         }
1235
1236         list_del_init(&pol->walk.all);
1237         net->xfrm.policy_count[dir]--;
1238
1239         return pol;
1240 }
1241
1242 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1243 {
1244         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1245 }
1246
1247 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1248 {
1249         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1250 }
1251
1252 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1253 {
1254         struct net *net = xp_net(pol);
1255
1256         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1257         pol = __xfrm_policy_unlink(pol, dir);
1258         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1259         if (pol) {
1260                 xfrm_policy_kill(pol);
1261                 return 0;
1262         }
1263         return -ENOENT;
1264 }
1265 EXPORT_SYMBOL(xfrm_policy_delete);
1266
1267 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1268 {
1269         struct net *net = sock_net(sk);
1270         struct xfrm_policy *old_pol;
1271
1272 #ifdef CONFIG_XFRM_SUB_POLICY
1273         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1274                 return -EINVAL;
1275 #endif
1276
1277         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1278         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1279                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1280         if (pol) {
1281                 pol->curlft.add_time = ktime_get_real_seconds();
1282                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1283                 xfrm_sk_policy_link(pol, dir);
1284         }
1285         rcu_assign_pointer(sk->sk_policy[dir], pol);
1286         if (old_pol) {
1287                 if (pol)
1288                         xfrm_policy_requeue(old_pol, pol);
1289
1290                 /* Unlinking succeeds always. This is the only function
1291                  * allowed to delete or replace socket policy.
1292                  */
1293                 xfrm_sk_policy_unlink(old_pol, dir);
1294         }
1295         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1296
1297         if (old_pol) {
1298                 xfrm_policy_kill(old_pol);
1299         }
1300         return 0;
1301 }
1302
1303 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1304 {
1305         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1306         struct net *net = xp_net(old);
1307
1308         if (newp) {
1309                 newp->selector = old->selector;
1310                 if (security_xfrm_policy_clone(old->security,
1311                                                &newp->security)) {
1312                         kfree(newp);
1313                         return NULL;  /* ENOMEM */
1314                 }
1315                 newp->lft = old->lft;
1316                 newp->curlft = old->curlft;
1317                 newp->mark = old->mark;
1318                 newp->if_id = old->if_id;
1319                 newp->action = old->action;
1320                 newp->flags = old->flags;
1321                 newp->xfrm_nr = old->xfrm_nr;
1322                 newp->index = old->index;
1323                 newp->type = old->type;
1324                 newp->family = old->family;
1325                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1326                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1327                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1328                 xfrm_sk_policy_link(newp, dir);
1329                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1330                 xfrm_pol_put(newp);
1331         }
1332         return newp;
1333 }
1334
1335 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1336 {
1337         const struct xfrm_policy *p;
1338         struct xfrm_policy *np;
1339         int i, ret = 0;
1340
1341         rcu_read_lock();
1342         for (i = 0; i < 2; i++) {
1343                 p = rcu_dereference(osk->sk_policy[i]);
1344                 if (p) {
1345                         np = clone_policy(p, i);
1346                         if (unlikely(!np)) {
1347                                 ret = -ENOMEM;
1348                                 break;
1349                         }
1350                         rcu_assign_pointer(sk->sk_policy[i], np);
1351                 }
1352         }
1353         rcu_read_unlock();
1354         return ret;
1355 }
1356
1357 static int
1358 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1359                xfrm_address_t *remote, unsigned short family, u32 mark)
1360 {
1361         int err;
1362         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1363
1364         if (unlikely(afinfo == NULL))
1365                 return -EINVAL;
1366         err = afinfo->get_saddr(net, oif, local, remote, mark);
1367         rcu_read_unlock();
1368         return err;
1369 }
1370
1371 /* Resolve list of templates for the flow, given policy. */
1372
1373 static int
1374 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1375                       struct xfrm_state **xfrm, unsigned short family)
1376 {
1377         struct net *net = xp_net(policy);
1378         int nx;
1379         int i, error;
1380         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1381         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1382         xfrm_address_t tmp;
1383
1384         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1385                 struct xfrm_state *x;
1386                 xfrm_address_t *remote = daddr;
1387                 xfrm_address_t *local  = saddr;
1388                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1389
1390                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1391                     tmpl->mode == XFRM_MODE_BEET) {
1392                         remote = &tmpl->id.daddr;
1393                         local = &tmpl->saddr;
1394                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1395                                 error = xfrm_get_saddr(net, fl->flowi_oif,
1396                                                        &tmp, remote,
1397                                                        tmpl->encap_family, 0);
1398                                 if (error)
1399                                         goto fail;
1400                                 local = &tmp;
1401                         }
1402                 }
1403
1404                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
1405                                     family, policy->if_id);
1406
1407                 if (x && x->km.state == XFRM_STATE_VALID) {
1408                         xfrm[nx++] = x;
1409                         daddr = remote;
1410                         saddr = local;
1411                         continue;
1412                 }
1413                 if (x) {
1414                         error = (x->km.state == XFRM_STATE_ERROR ?
1415                                  -EINVAL : -EAGAIN);
1416                         xfrm_state_put(x);
1417                 } else if (error == -ESRCH) {
1418                         error = -EAGAIN;
1419                 }
1420
1421                 if (!tmpl->optional)
1422                         goto fail;
1423         }
1424         return nx;
1425
1426 fail:
1427         for (nx--; nx >= 0; nx--)
1428                 xfrm_state_put(xfrm[nx]);
1429         return error;
1430 }
1431
1432 static int
1433 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1434                   struct xfrm_state **xfrm, unsigned short family)
1435 {
1436         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1437         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1438         int cnx = 0;
1439         int error;
1440         int ret;
1441         int i;
1442
1443         for (i = 0; i < npols; i++) {
1444                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1445                         error = -ENOBUFS;
1446                         goto fail;
1447                 }
1448
1449                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1450                 if (ret < 0) {
1451                         error = ret;
1452                         goto fail;
1453                 } else
1454                         cnx += ret;
1455         }
1456
1457         /* found states are sorted for outbound processing */
1458         if (npols > 1)
1459                 xfrm_state_sort(xfrm, tpp, cnx, family);
1460
1461         return cnx;
1462
1463  fail:
1464         for (cnx--; cnx >= 0; cnx--)
1465                 xfrm_state_put(tpp[cnx]);
1466         return error;
1467
1468 }
1469
1470 static int xfrm_get_tos(const struct flowi *fl, int family)
1471 {
1472         const struct xfrm_policy_afinfo *afinfo;
1473         int tos;
1474
1475         afinfo = xfrm_policy_get_afinfo(family);
1476         if (!afinfo)
1477                 return 0;
1478
1479         tos = afinfo->get_tos(fl);
1480
1481         rcu_read_unlock();
1482
1483         return tos;
1484 }
1485
1486 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1487 {
1488         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1489         struct dst_ops *dst_ops;
1490         struct xfrm_dst *xdst;
1491
1492         if (!afinfo)
1493                 return ERR_PTR(-EINVAL);
1494
1495         switch (family) {
1496         case AF_INET:
1497                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1498                 break;
1499 #if IS_ENABLED(CONFIG_IPV6)
1500         case AF_INET6:
1501                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1502                 break;
1503 #endif
1504         default:
1505                 BUG();
1506         }
1507         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1508
1509         if (likely(xdst)) {
1510                 struct dst_entry *dst = &xdst->u.dst;
1511
1512                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1513         } else
1514                 xdst = ERR_PTR(-ENOBUFS);
1515
1516         rcu_read_unlock();
1517
1518         return xdst;
1519 }
1520
1521 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1522                                  int nfheader_len)
1523 {
1524         const struct xfrm_policy_afinfo *afinfo =
1525                 xfrm_policy_get_afinfo(dst->ops->family);
1526         int err;
1527
1528         if (!afinfo)
1529                 return -EINVAL;
1530
1531         err = afinfo->init_path(path, dst, nfheader_len);
1532
1533         rcu_read_unlock();
1534
1535         return err;
1536 }
1537
1538 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1539                                 const struct flowi *fl)
1540 {
1541         const struct xfrm_policy_afinfo *afinfo =
1542                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1543         int err;
1544
1545         if (!afinfo)
1546                 return -EINVAL;
1547
1548         err = afinfo->fill_dst(xdst, dev, fl);
1549
1550         rcu_read_unlock();
1551
1552         return err;
1553 }
1554
1555
1556 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1557  * all the metrics... Shortly, bundle a bundle.
1558  */
1559
1560 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1561                                             struct xfrm_state **xfrm,
1562                                             struct xfrm_dst **bundle,
1563                                             int nx,
1564                                             const struct flowi *fl,
1565                                             struct dst_entry *dst)
1566 {
1567         struct net *net = xp_net(policy);
1568         unsigned long now = jiffies;
1569         struct net_device *dev;
1570         struct xfrm_mode *inner_mode;
1571         struct xfrm_dst *xdst_prev = NULL;
1572         struct xfrm_dst *xdst0 = NULL;
1573         int i = 0;
1574         int err;
1575         int header_len = 0;
1576         int nfheader_len = 0;
1577         int trailer_len = 0;
1578         int tos;
1579         int family = policy->selector.family;
1580         xfrm_address_t saddr, daddr;
1581
1582         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1583
1584         tos = xfrm_get_tos(fl, family);
1585
1586         dst_hold(dst);
1587
1588         for (; i < nx; i++) {
1589                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1590                 struct dst_entry *dst1 = &xdst->u.dst;
1591
1592                 err = PTR_ERR(xdst);
1593                 if (IS_ERR(xdst)) {
1594                         dst_release(dst);
1595                         goto put_states;
1596                 }
1597
1598                 bundle[i] = xdst;
1599                 if (!xdst_prev)
1600                         xdst0 = xdst;
1601                 else
1602                         /* Ref count is taken during xfrm_alloc_dst()
1603                          * No need to do dst_clone() on dst1
1604                          */
1605                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
1606
1607                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1608                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1609                                                         xfrm_af2proto(family));
1610                         if (!inner_mode) {
1611                                 err = -EAFNOSUPPORT;
1612                                 dst_release(dst);
1613                                 goto put_states;
1614                         }
1615                 } else
1616                         inner_mode = xfrm[i]->inner_mode;
1617
1618                 xdst->route = dst;
1619                 dst_copy_metrics(dst1, dst);
1620
1621                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1622                         __u32 mark = 0;
1623
1624                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
1625                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
1626
1627                         family = xfrm[i]->props.family;
1628                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1629                                               &saddr, &daddr, family, mark);
1630                         err = PTR_ERR(dst);
1631                         if (IS_ERR(dst))
1632                                 goto put_states;
1633                 } else
1634                         dst_hold(dst);
1635
1636                 dst1->xfrm = xfrm[i];
1637                 xdst->xfrm_genid = xfrm[i]->genid;
1638
1639                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1640                 dst1->flags |= DST_HOST;
1641                 dst1->lastuse = now;
1642
1643                 dst1->input = dst_discard;
1644                 dst1->output = inner_mode->afinfo->output;
1645
1646                 xdst_prev = xdst;
1647
1648                 header_len += xfrm[i]->props.header_len;
1649                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1650                         nfheader_len += xfrm[i]->props.header_len;
1651                 trailer_len += xfrm[i]->props.trailer_len;
1652         }
1653
1654         xfrm_dst_set_child(xdst_prev, dst);
1655         xdst0->path = dst;
1656
1657         err = -ENODEV;
1658         dev = dst->dev;
1659         if (!dev)
1660                 goto free_dst;
1661
1662         xfrm_init_path(xdst0, dst, nfheader_len);
1663         xfrm_init_pmtu(bundle, nx);
1664
1665         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
1666              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
1667                 err = xfrm_fill_dst(xdst_prev, dev, fl);
1668                 if (err)
1669                         goto free_dst;
1670
1671                 xdst_prev->u.dst.header_len = header_len;
1672                 xdst_prev->u.dst.trailer_len = trailer_len;
1673                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
1674                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
1675         }
1676
1677         return &xdst0->u.dst;
1678
1679 put_states:
1680         for (; i < nx; i++)
1681                 xfrm_state_put(xfrm[i]);
1682 free_dst:
1683         if (xdst0)
1684                 dst_release_immediate(&xdst0->u.dst);
1685
1686         return ERR_PTR(err);
1687 }
1688
1689 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1690                                 struct xfrm_policy **pols,
1691                                 int *num_pols, int *num_xfrms)
1692 {
1693         int i;
1694
1695         if (*num_pols == 0 || !pols[0]) {
1696                 *num_pols = 0;
1697                 *num_xfrms = 0;
1698                 return 0;
1699         }
1700         if (IS_ERR(pols[0]))
1701                 return PTR_ERR(pols[0]);
1702
1703         *num_xfrms = pols[0]->xfrm_nr;
1704
1705 #ifdef CONFIG_XFRM_SUB_POLICY
1706         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1707             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1708                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1709                                                     XFRM_POLICY_TYPE_MAIN,
1710                                                     fl, family,
1711                                                     XFRM_POLICY_OUT,
1712                                                     pols[0]->if_id);
1713                 if (pols[1]) {
1714                         if (IS_ERR(pols[1])) {
1715                                 xfrm_pols_put(pols, *num_pols);
1716                                 return PTR_ERR(pols[1]);
1717                         }
1718                         (*num_pols)++;
1719                         (*num_xfrms) += pols[1]->xfrm_nr;
1720                 }
1721         }
1722 #endif
1723         for (i = 0; i < *num_pols; i++) {
1724                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1725                         *num_xfrms = -1;
1726                         break;
1727                 }
1728         }
1729
1730         return 0;
1731
1732 }
1733
1734 static struct xfrm_dst *
1735 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1736                                const struct flowi *fl, u16 family,
1737                                struct dst_entry *dst_orig)
1738 {
1739         struct net *net = xp_net(pols[0]);
1740         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1741         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
1742         struct xfrm_dst *xdst;
1743         struct dst_entry *dst;
1744         int err;
1745
1746         /* Try to instantiate a bundle */
1747         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1748         if (err <= 0) {
1749                 if (err == 0)
1750                         return NULL;
1751
1752                 if (err != -EAGAIN)
1753                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1754                 return ERR_PTR(err);
1755         }
1756
1757         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
1758         if (IS_ERR(dst)) {
1759                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1760                 return ERR_CAST(dst);
1761         }
1762
1763         xdst = (struct xfrm_dst *)dst;
1764         xdst->num_xfrms = err;
1765         xdst->num_pols = num_pols;
1766         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1767         xdst->policy_genid = atomic_read(&pols[0]->genid);
1768
1769         return xdst;
1770 }
1771
1772 static void xfrm_policy_queue_process(struct timer_list *t)
1773 {
1774         struct sk_buff *skb;
1775         struct sock *sk;
1776         struct dst_entry *dst;
1777         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1778         struct net *net = xp_net(pol);
1779         struct xfrm_policy_queue *pq = &pol->polq;
1780         struct flowi fl;
1781         struct sk_buff_head list;
1782
1783         spin_lock(&pq->hold_queue.lock);
1784         skb = skb_peek(&pq->hold_queue);
1785         if (!skb) {
1786                 spin_unlock(&pq->hold_queue.lock);
1787                 goto out;
1788         }
1789         dst = skb_dst(skb);
1790         sk = skb->sk;
1791         xfrm_decode_session(skb, &fl, dst->ops->family);
1792         spin_unlock(&pq->hold_queue.lock);
1793
1794         dst_hold(xfrm_dst_path(dst));
1795         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
1796         if (IS_ERR(dst))
1797                 goto purge_queue;
1798
1799         if (dst->flags & DST_XFRM_QUEUE) {
1800                 dst_release(dst);
1801
1802                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1803                         goto purge_queue;
1804
1805                 pq->timeout = pq->timeout << 1;
1806                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1807                         xfrm_pol_hold(pol);
1808         goto out;
1809         }
1810
1811         dst_release(dst);
1812
1813         __skb_queue_head_init(&list);
1814
1815         spin_lock(&pq->hold_queue.lock);
1816         pq->timeout = 0;
1817         skb_queue_splice_init(&pq->hold_queue, &list);
1818         spin_unlock(&pq->hold_queue.lock);
1819
1820         while (!skb_queue_empty(&list)) {
1821                 skb = __skb_dequeue(&list);
1822
1823                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1824                 dst_hold(xfrm_dst_path(skb_dst(skb)));
1825                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
1826                 if (IS_ERR(dst)) {
1827                         kfree_skb(skb);
1828                         continue;
1829                 }
1830
1831                 nf_reset(skb);
1832                 skb_dst_drop(skb);
1833                 skb_dst_set(skb, dst);
1834
1835                 dst_output(net, skb->sk, skb);
1836         }
1837
1838 out:
1839         xfrm_pol_put(pol);
1840         return;
1841
1842 purge_queue:
1843         pq->timeout = 0;
1844         skb_queue_purge(&pq->hold_queue);
1845         xfrm_pol_put(pol);
1846 }
1847
1848 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1849 {
1850         unsigned long sched_next;
1851         struct dst_entry *dst = skb_dst(skb);
1852         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1853         struct xfrm_policy *pol = xdst->pols[0];
1854         struct xfrm_policy_queue *pq = &pol->polq;
1855
1856         if (unlikely(skb_fclone_busy(sk, skb))) {
1857                 kfree_skb(skb);
1858                 return 0;
1859         }
1860
1861         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1862                 kfree_skb(skb);
1863                 return -EAGAIN;
1864         }
1865
1866         skb_dst_force(skb);
1867
1868         spin_lock_bh(&pq->hold_queue.lock);
1869
1870         if (!pq->timeout)
1871                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1872
1873         sched_next = jiffies + pq->timeout;
1874
1875         if (del_timer(&pq->hold_timer)) {
1876                 if (time_before(pq->hold_timer.expires, sched_next))
1877                         sched_next = pq->hold_timer.expires;
1878                 xfrm_pol_put(pol);
1879         }
1880
1881         __skb_queue_tail(&pq->hold_queue, skb);
1882         if (!mod_timer(&pq->hold_timer, sched_next))
1883                 xfrm_pol_hold(pol);
1884
1885         spin_unlock_bh(&pq->hold_queue.lock);
1886
1887         return 0;
1888 }
1889
1890 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1891                                                  struct xfrm_flo *xflo,
1892                                                  const struct flowi *fl,
1893                                                  int num_xfrms,
1894                                                  u16 family)
1895 {
1896         int err;
1897         struct net_device *dev;
1898         struct dst_entry *dst;
1899         struct dst_entry *dst1;
1900         struct xfrm_dst *xdst;
1901
1902         xdst = xfrm_alloc_dst(net, family);
1903         if (IS_ERR(xdst))
1904                 return xdst;
1905
1906         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
1907             net->xfrm.sysctl_larval_drop ||
1908             num_xfrms <= 0)
1909                 return xdst;
1910
1911         dst = xflo->dst_orig;
1912         dst1 = &xdst->u.dst;
1913         dst_hold(dst);
1914         xdst->route = dst;
1915
1916         dst_copy_metrics(dst1, dst);
1917
1918         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1919         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1920         dst1->lastuse = jiffies;
1921
1922         dst1->input = dst_discard;
1923         dst1->output = xdst_queue_output;
1924
1925         dst_hold(dst);
1926         xfrm_dst_set_child(xdst, dst);
1927         xdst->path = dst;
1928
1929         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1930
1931         err = -ENODEV;
1932         dev = dst->dev;
1933         if (!dev)
1934                 goto free_dst;
1935
1936         err = xfrm_fill_dst(xdst, dev, fl);
1937         if (err)
1938                 goto free_dst;
1939
1940 out:
1941         return xdst;
1942
1943 free_dst:
1944         dst_release(dst1);
1945         xdst = ERR_PTR(err);
1946         goto out;
1947 }
1948
1949 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
1950                                            const struct flowi *fl,
1951                                            u16 family, u8 dir,
1952                                            struct xfrm_flo *xflo, u32 if_id)
1953 {
1954         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1955         int num_pols = 0, num_xfrms = 0, err;
1956         struct xfrm_dst *xdst;
1957
1958         /* Resolve policies to use if we couldn't get them from
1959          * previous cache entry */
1960         num_pols = 1;
1961         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
1962         err = xfrm_expand_policies(fl, family, pols,
1963                                            &num_pols, &num_xfrms);
1964         if (err < 0)
1965                 goto inc_error;
1966         if (num_pols == 0)
1967                 return NULL;
1968         if (num_xfrms <= 0)
1969                 goto make_dummy_bundle;
1970
1971         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
1972                                               xflo->dst_orig);
1973         if (IS_ERR(xdst)) {
1974                 err = PTR_ERR(xdst);
1975                 if (err == -EREMOTE) {
1976                         xfrm_pols_put(pols, num_pols);
1977                         return NULL;
1978                 }
1979
1980                 if (err != -EAGAIN)
1981                         goto error;
1982                 goto make_dummy_bundle;
1983         } else if (xdst == NULL) {
1984                 num_xfrms = 0;
1985                 goto make_dummy_bundle;
1986         }
1987
1988         return xdst;
1989
1990 make_dummy_bundle:
1991         /* We found policies, but there's no bundles to instantiate:
1992          * either because the policy blocks, has no transformations or
1993          * we could not build template (no xfrm_states).*/
1994         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
1995         if (IS_ERR(xdst)) {
1996                 xfrm_pols_put(pols, num_pols);
1997                 return ERR_CAST(xdst);
1998         }
1999         xdst->num_pols = num_pols;
2000         xdst->num_xfrms = num_xfrms;
2001         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2002
2003         return xdst;
2004
2005 inc_error:
2006         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2007 error:
2008         xfrm_pols_put(pols, num_pols);
2009         return ERR_PTR(err);
2010 }
2011
2012 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2013                                         struct dst_entry *dst_orig)
2014 {
2015         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2016         struct dst_entry *ret;
2017
2018         if (!afinfo) {
2019                 dst_release(dst_orig);
2020                 return ERR_PTR(-EINVAL);
2021         } else {
2022                 ret = afinfo->blackhole_route(net, dst_orig);
2023         }
2024         rcu_read_unlock();
2025
2026         return ret;
2027 }
2028
2029 /* Finds/creates a bundle for given flow and if_id
2030  *
2031  * At the moment we eat a raw IP route. Mostly to speed up lookups
2032  * on interfaces with disabled IPsec.
2033  *
2034  * xfrm_lookup uses an if_id of 0 by default, and is provided for
2035  * compatibility
2036  */
2037 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
2038                                         struct dst_entry *dst_orig,
2039                                         const struct flowi *fl,
2040                                         const struct sock *sk,
2041                                         int flags, u32 if_id)
2042 {
2043         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2044         struct xfrm_dst *xdst;
2045         struct dst_entry *dst, *route;
2046         u16 family = dst_orig->ops->family;
2047         u8 dir = XFRM_POLICY_OUT;
2048         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2049
2050         dst = NULL;
2051         xdst = NULL;
2052         route = NULL;
2053
2054         sk = sk_const_to_full_sk(sk);
2055         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2056                 num_pols = 1;
2057                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
2058                                                 if_id);
2059                 err = xfrm_expand_policies(fl, family, pols,
2060                                            &num_pols, &num_xfrms);
2061                 if (err < 0)
2062                         goto dropdst;
2063
2064                 if (num_pols) {
2065                         if (num_xfrms <= 0) {
2066                                 drop_pols = num_pols;
2067                                 goto no_transform;
2068                         }
2069
2070                         xdst = xfrm_resolve_and_create_bundle(
2071                                         pols, num_pols, fl,
2072                                         family, dst_orig);
2073
2074                         if (IS_ERR(xdst)) {
2075                                 xfrm_pols_put(pols, num_pols);
2076                                 err = PTR_ERR(xdst);
2077                                 if (err == -EREMOTE)
2078                                         goto nopol;
2079
2080                                 goto dropdst;
2081                         } else if (xdst == NULL) {
2082                                 num_xfrms = 0;
2083                                 drop_pols = num_pols;
2084                                 goto no_transform;
2085                         }
2086
2087                         route = xdst->route;
2088                 }
2089         }
2090
2091         if (xdst == NULL) {
2092                 struct xfrm_flo xflo;
2093
2094                 xflo.dst_orig = dst_orig;
2095                 xflo.flags = flags;
2096
2097                 /* To accelerate a bit...  */
2098                 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
2099                                !net->xfrm.policy_count[XFRM_POLICY_OUT]))
2100                         goto nopol;
2101
2102                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
2103                 if (xdst == NULL)
2104                         goto nopol;
2105                 if (IS_ERR(xdst)) {
2106                         err = PTR_ERR(xdst);
2107                         goto dropdst;
2108                 }
2109
2110                 num_pols = xdst->num_pols;
2111                 num_xfrms = xdst->num_xfrms;
2112                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2113                 route = xdst->route;
2114         }
2115
2116         dst = &xdst->u.dst;
2117         if (route == NULL && num_xfrms > 0) {
2118                 /* The only case when xfrm_bundle_lookup() returns a
2119                  * bundle with null route, is when the template could
2120                  * not be resolved. It means policies are there, but
2121                  * bundle could not be created, since we don't yet
2122                  * have the xfrm_state's. We need to wait for KM to
2123                  * negotiate new SA's or bail out with error.*/
2124                 if (net->xfrm.sysctl_larval_drop) {
2125                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2126                         err = -EREMOTE;
2127                         goto error;
2128                 }
2129
2130                 err = -EAGAIN;
2131
2132                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2133                 goto error;
2134         }
2135
2136 no_transform:
2137         if (num_pols == 0)
2138                 goto nopol;
2139
2140         if ((flags & XFRM_LOOKUP_ICMP) &&
2141             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2142                 err = -ENOENT;
2143                 goto error;
2144         }
2145
2146         for (i = 0; i < num_pols; i++)
2147                 pols[i]->curlft.use_time = ktime_get_real_seconds();
2148
2149         if (num_xfrms < 0) {
2150                 /* Prohibit the flow */
2151                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2152                 err = -EPERM;
2153                 goto error;
2154         } else if (num_xfrms > 0) {
2155                 /* Flow transformed */
2156                 dst_release(dst_orig);
2157         } else {
2158                 /* Flow passes untransformed */
2159                 dst_release(dst);
2160                 dst = dst_orig;
2161         }
2162 ok:
2163         xfrm_pols_put(pols, drop_pols);
2164         if (dst && dst->xfrm &&
2165             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2166                 dst->flags |= DST_XFRM_TUNNEL;
2167         return dst;
2168
2169 nopol:
2170         if (!(flags & XFRM_LOOKUP_ICMP)) {
2171                 dst = dst_orig;
2172                 goto ok;
2173         }
2174         err = -ENOENT;
2175 error:
2176         dst_release(dst);
2177 dropdst:
2178         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2179                 dst_release(dst_orig);
2180         xfrm_pols_put(pols, drop_pols);
2181         return ERR_PTR(err);
2182 }
2183 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
2184
2185 /* Main function: finds/creates a bundle for given flow.
2186  *
2187  * At the moment we eat a raw IP route. Mostly to speed up lookups
2188  * on interfaces with disabled IPsec.
2189  */
2190 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2191                               const struct flowi *fl, const struct sock *sk,
2192                               int flags)
2193 {
2194         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
2195 }
2196 EXPORT_SYMBOL(xfrm_lookup);
2197
2198 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2199  * Otherwise we may send out blackholed packets.
2200  */
2201 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2202                                     const struct flowi *fl,
2203                                     const struct sock *sk, int flags)
2204 {
2205         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2206                                             flags | XFRM_LOOKUP_QUEUE |
2207                                             XFRM_LOOKUP_KEEP_DST_REF);
2208
2209         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2210                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2211
2212         if (IS_ERR(dst))
2213                 dst_release(dst_orig);
2214
2215         return dst;
2216 }
2217 EXPORT_SYMBOL(xfrm_lookup_route);
2218
2219 static inline int
2220 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2221 {
2222         struct xfrm_state *x;
2223
2224         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2225                 return 0;
2226         x = skb->sp->xvec[idx];
2227         if (!x->type->reject)
2228                 return 0;
2229         return x->type->reject(x, skb, fl);
2230 }
2231
2232 /* When skb is transformed back to its "native" form, we have to
2233  * check policy restrictions. At the moment we make this in maximally
2234  * stupid way. Shame on me. :-) Of course, connected sockets must
2235  * have policy cached at them.
2236  */
2237
2238 static inline int
2239 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2240               unsigned short family)
2241 {
2242         if (xfrm_state_kern(x))
2243                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2244         return  x->id.proto == tmpl->id.proto &&
2245                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2246                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2247                 x->props.mode == tmpl->mode &&
2248                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2249                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2250                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2251                   xfrm_state_addr_cmp(tmpl, x, family));
2252 }
2253
2254 /*
2255  * 0 or more than 0 is returned when validation is succeeded (either bypass
2256  * because of optional transport mode, or next index of the mathced secpath
2257  * state with the template.
2258  * -1 is returned when no matching template is found.
2259  * Otherwise "-2 - errored_index" is returned.
2260  */
2261 static inline int
2262 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2263                unsigned short family)
2264 {
2265         int idx = start;
2266
2267         if (tmpl->optional) {
2268                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2269                         return start;
2270         } else
2271                 start = -1;
2272         for (; idx < sp->len; idx++) {
2273                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2274                         return ++idx;
2275                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2276                         if (start == -1)
2277                                 start = -2-idx;
2278                         break;
2279                 }
2280         }
2281         return start;
2282 }
2283
2284 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2285                           unsigned int family, int reverse)
2286 {
2287         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2288         int err;
2289
2290         if (unlikely(afinfo == NULL))
2291                 return -EAFNOSUPPORT;
2292
2293         afinfo->decode_session(skb, fl, reverse);
2294
2295         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2296         rcu_read_unlock();
2297         return err;
2298 }
2299 EXPORT_SYMBOL(__xfrm_decode_session);
2300
2301 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2302 {
2303         for (; k < sp->len; k++) {
2304                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2305                         *idxp = k;
2306                         return 1;
2307                 }
2308         }
2309
2310         return 0;
2311 }
2312
2313 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2314                         unsigned short family)
2315 {
2316         struct net *net = dev_net(skb->dev);
2317         struct xfrm_policy *pol;
2318         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2319         int npols = 0;
2320         int xfrm_nr;
2321         int pi;
2322         int reverse;
2323         struct flowi fl;
2324         int xerr_idx = -1;
2325         const struct xfrm_if_cb *ifcb;
2326         struct xfrm_if *xi;
2327         u32 if_id = 0;
2328
2329         rcu_read_lock();
2330         ifcb = xfrm_if_get_cb();
2331
2332         if (ifcb) {
2333                 xi = ifcb->decode_session(skb, family);
2334                 if (xi) {
2335                         if_id = xi->p.if_id;
2336                         net = xi->net;
2337                 }
2338         }
2339         rcu_read_unlock();
2340
2341         reverse = dir & ~XFRM_POLICY_MASK;
2342         dir &= XFRM_POLICY_MASK;
2343
2344         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2345                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2346                 return 0;
2347         }
2348
2349         nf_nat_decode_session(skb, &fl, family);
2350
2351         /* First, check used SA against their selectors. */
2352         if (skb->sp) {
2353                 int i;
2354
2355                 for (i = skb->sp->len-1; i >= 0; i--) {
2356                         struct xfrm_state *x = skb->sp->xvec[i];
2357                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2358                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2359                                 return 0;
2360                         }
2361                 }
2362         }
2363
2364         pol = NULL;
2365         sk = sk_to_full_sk(sk);
2366         if (sk && sk->sk_policy[dir]) {
2367                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
2368                 if (IS_ERR(pol)) {
2369                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2370                         return 0;
2371                 }
2372         }
2373
2374         if (!pol)
2375                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
2376
2377         if (IS_ERR(pol)) {
2378                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2379                 return 0;
2380         }
2381
2382         if (!pol) {
2383                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2384                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2385                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2386                         return 0;
2387                 }
2388                 return 1;
2389         }
2390
2391         pol->curlft.use_time = ktime_get_real_seconds();
2392
2393         pols[0] = pol;
2394         npols++;
2395 #ifdef CONFIG_XFRM_SUB_POLICY
2396         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2397                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2398                                                     &fl, family,
2399                                                     XFRM_POLICY_IN, if_id);
2400                 if (pols[1]) {
2401                         if (IS_ERR(pols[1])) {
2402                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2403                                 return 0;
2404                         }
2405                         pols[1]->curlft.use_time = ktime_get_real_seconds();
2406                         npols++;
2407                 }
2408         }
2409 #endif
2410
2411         if (pol->action == XFRM_POLICY_ALLOW) {
2412                 struct sec_path *sp;
2413                 static struct sec_path dummy;
2414                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2415                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2416                 struct xfrm_tmpl **tpp = tp;
2417                 int ti = 0;
2418                 int i, k;
2419
2420                 if ((sp = skb->sp) == NULL)
2421                         sp = &dummy;
2422
2423                 for (pi = 0; pi < npols; pi++) {
2424                         if (pols[pi] != pol &&
2425                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2426                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2427                                 goto reject;
2428                         }
2429                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2430                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2431                                 goto reject_error;
2432                         }
2433                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2434                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2435                 }
2436                 xfrm_nr = ti;
2437                 if (npols > 1) {
2438                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2439                         tpp = stp;
2440                 }
2441
2442                 /* For each tunnel xfrm, find the first matching tmpl.
2443                  * For each tmpl before that, find corresponding xfrm.
2444                  * Order is _important_. Later we will implement
2445                  * some barriers, but at the moment barriers
2446                  * are implied between each two transformations.
2447                  */
2448                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2449                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2450                         if (k < 0) {
2451                                 if (k < -1)
2452                                         /* "-2 - errored_index" returned */
2453                                         xerr_idx = -(2+k);
2454                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2455                                 goto reject;
2456                         }
2457                 }
2458
2459                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2460                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2461                         goto reject;
2462                 }
2463
2464                 xfrm_pols_put(pols, npols);
2465                 return 1;
2466         }
2467         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2468
2469 reject:
2470         xfrm_secpath_reject(xerr_idx, skb, &fl);
2471 reject_error:
2472         xfrm_pols_put(pols, npols);
2473         return 0;
2474 }
2475 EXPORT_SYMBOL(__xfrm_policy_check);
2476
2477 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2478 {
2479         struct net *net = dev_net(skb->dev);
2480         struct flowi fl;
2481         struct dst_entry *dst;
2482         int res = 1;
2483
2484         if (xfrm_decode_session(skb, &fl, family) < 0) {
2485                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2486                 return 0;
2487         }
2488
2489         skb_dst_force(skb);
2490         if (!skb_dst(skb)) {
2491                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2492                 return 0;
2493         }
2494
2495         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2496         if (IS_ERR(dst)) {
2497                 res = 0;
2498                 dst = NULL;
2499         }
2500         skb_dst_set(skb, dst);
2501         return res;
2502 }
2503 EXPORT_SYMBOL(__xfrm_route_forward);
2504
2505 /* Optimize later using cookies and generation ids. */
2506
2507 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2508 {
2509         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2510          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2511          * get validated by dst_ops->check on every use.  We do this
2512          * because when a normal route referenced by an XFRM dst is
2513          * obsoleted we do not go looking around for all parent
2514          * referencing XFRM dsts so that we can invalidate them.  It
2515          * is just too much work.  Instead we make the checks here on
2516          * every use.  For example:
2517          *
2518          *      XFRM dst A --> IPv4 dst X
2519          *
2520          * X is the "xdst->route" of A (X is also the "dst->path" of A
2521          * in this example).  If X is marked obsolete, "A" will not
2522          * notice.  That's what we are validating here via the
2523          * stale_bundle() check.
2524          *
2525          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2526          * be marked on it.
2527          * This will force stale_bundle() to fail on any xdst bundle with
2528          * this dst linked in it.
2529          */
2530         if (dst->obsolete < 0 && !stale_bundle(dst))
2531                 return dst;
2532
2533         return NULL;
2534 }
2535
2536 static int stale_bundle(struct dst_entry *dst)
2537 {
2538         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2539 }
2540
2541 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2542 {
2543         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
2544                 dst->dev = dev_net(dev)->loopback_dev;
2545                 dev_hold(dst->dev);
2546                 dev_put(dev);
2547         }
2548 }
2549 EXPORT_SYMBOL(xfrm_dst_ifdown);
2550
2551 static void xfrm_link_failure(struct sk_buff *skb)
2552 {
2553         /* Impossible. Such dst must be popped before reaches point of failure. */
2554 }
2555
2556 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2557 {
2558         if (dst) {
2559                 if (dst->obsolete) {
2560                         dst_release(dst);
2561                         dst = NULL;
2562                 }
2563         }
2564         return dst;
2565 }
2566
2567 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
2568 {
2569         while (nr--) {
2570                 struct xfrm_dst *xdst = bundle[nr];
2571                 u32 pmtu, route_mtu_cached;
2572                 struct dst_entry *dst;
2573
2574                 dst = &xdst->u.dst;
2575                 pmtu = dst_mtu(xfrm_dst_child(dst));
2576                 xdst->child_mtu_cached = pmtu;
2577
2578                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2579
2580                 route_mtu_cached = dst_mtu(xdst->route);
2581                 xdst->route_mtu_cached = route_mtu_cached;
2582
2583                 if (pmtu > route_mtu_cached)
2584                         pmtu = route_mtu_cached;
2585
2586                 dst_metric_set(dst, RTAX_MTU, pmtu);
2587         }
2588 }
2589
2590 /* Check that the bundle accepts the flow and its components are
2591  * still valid.
2592  */
2593
2594 static int xfrm_bundle_ok(struct xfrm_dst *first)
2595 {
2596         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2597         struct dst_entry *dst = &first->u.dst;
2598         struct xfrm_dst *xdst;
2599         int start_from, nr;
2600         u32 mtu;
2601
2602         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
2603             (dst->dev && !netif_running(dst->dev)))
2604                 return 0;
2605
2606         if (dst->flags & DST_XFRM_QUEUE)
2607                 return 1;
2608
2609         start_from = nr = 0;
2610         do {
2611                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2612
2613                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2614                         return 0;
2615                 if (xdst->xfrm_genid != dst->xfrm->genid)
2616                         return 0;
2617                 if (xdst->num_pols > 0 &&
2618                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2619                         return 0;
2620
2621                 bundle[nr++] = xdst;
2622
2623                 mtu = dst_mtu(xfrm_dst_child(dst));
2624                 if (xdst->child_mtu_cached != mtu) {
2625                         start_from = nr;
2626                         xdst->child_mtu_cached = mtu;
2627                 }
2628
2629                 if (!dst_check(xdst->route, xdst->route_cookie))
2630                         return 0;
2631                 mtu = dst_mtu(xdst->route);
2632                 if (xdst->route_mtu_cached != mtu) {
2633                         start_from = nr;
2634                         xdst->route_mtu_cached = mtu;
2635                 }
2636
2637                 dst = xfrm_dst_child(dst);
2638         } while (dst->xfrm);
2639
2640         if (likely(!start_from))
2641                 return 1;
2642
2643         xdst = bundle[start_from - 1];
2644         mtu = xdst->child_mtu_cached;
2645         while (start_from--) {
2646                 dst = &xdst->u.dst;
2647
2648                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2649                 if (mtu > xdst->route_mtu_cached)
2650                         mtu = xdst->route_mtu_cached;
2651                 dst_metric_set(dst, RTAX_MTU, mtu);
2652                 if (!start_from)
2653                         break;
2654
2655                 xdst = bundle[start_from - 1];
2656                 xdst->child_mtu_cached = mtu;
2657         }
2658
2659         return 1;
2660 }
2661
2662 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2663 {
2664         return dst_metric_advmss(xfrm_dst_path(dst));
2665 }
2666
2667 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2668 {
2669         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2670
2671         return mtu ? : dst_mtu(xfrm_dst_path(dst));
2672 }
2673
2674 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2675                                         const void *daddr)
2676 {
2677         while (dst->xfrm) {
2678                 const struct xfrm_state *xfrm = dst->xfrm;
2679
2680                 dst = xfrm_dst_child(dst);
2681
2682                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2683                         continue;
2684                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2685                         daddr = xfrm->coaddr;
2686                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2687                         daddr = &xfrm->id.daddr;
2688         }
2689         return daddr;
2690 }
2691
2692 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2693                                            struct sk_buff *skb,
2694                                            const void *daddr)
2695 {
2696         const struct dst_entry *path = xfrm_dst_path(dst);
2697
2698         if (!skb)
2699                 daddr = xfrm_get_dst_nexthop(dst, daddr);
2700         return path->ops->neigh_lookup(path, skb, daddr);
2701 }
2702
2703 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2704 {
2705         const struct dst_entry *path = xfrm_dst_path(dst);
2706
2707         daddr = xfrm_get_dst_nexthop(dst, daddr);
2708         path->ops->confirm_neigh(path, daddr);
2709 }
2710
2711 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2712 {
2713         int err = 0;
2714
2715         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2716                 return -EAFNOSUPPORT;
2717
2718         spin_lock(&xfrm_policy_afinfo_lock);
2719         if (unlikely(xfrm_policy_afinfo[family] != NULL))
2720                 err = -EEXIST;
2721         else {
2722                 struct dst_ops *dst_ops = afinfo->dst_ops;
2723                 if (likely(dst_ops->kmem_cachep == NULL))
2724                         dst_ops->kmem_cachep = xfrm_dst_cache;
2725                 if (likely(dst_ops->check == NULL))
2726                         dst_ops->check = xfrm_dst_check;
2727                 if (likely(dst_ops->default_advmss == NULL))
2728                         dst_ops->default_advmss = xfrm_default_advmss;
2729                 if (likely(dst_ops->mtu == NULL))
2730                         dst_ops->mtu = xfrm_mtu;
2731                 if (likely(dst_ops->negative_advice == NULL))
2732                         dst_ops->negative_advice = xfrm_negative_advice;
2733                 if (likely(dst_ops->link_failure == NULL))
2734                         dst_ops->link_failure = xfrm_link_failure;
2735                 if (likely(dst_ops->neigh_lookup == NULL))
2736                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2737                 if (likely(!dst_ops->confirm_neigh))
2738                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
2739                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2740         }
2741         spin_unlock(&xfrm_policy_afinfo_lock);
2742
2743         return err;
2744 }
2745 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2746
2747 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2748 {
2749         struct dst_ops *dst_ops = afinfo->dst_ops;
2750         int i;
2751
2752         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2753                 if (xfrm_policy_afinfo[i] != afinfo)
2754                         continue;
2755                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2756                 break;
2757         }
2758
2759         synchronize_rcu();
2760
2761         dst_ops->kmem_cachep = NULL;
2762         dst_ops->check = NULL;
2763         dst_ops->negative_advice = NULL;
2764         dst_ops->link_failure = NULL;
2765 }
2766 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2767
2768 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
2769 {
2770         spin_lock(&xfrm_if_cb_lock);
2771         rcu_assign_pointer(xfrm_if_cb, ifcb);
2772         spin_unlock(&xfrm_if_cb_lock);
2773 }
2774 EXPORT_SYMBOL(xfrm_if_register_cb);
2775
2776 void xfrm_if_unregister_cb(void)
2777 {
2778         RCU_INIT_POINTER(xfrm_if_cb, NULL);
2779         synchronize_rcu();
2780 }
2781 EXPORT_SYMBOL(xfrm_if_unregister_cb);
2782
2783 #ifdef CONFIG_XFRM_STATISTICS
2784 static int __net_init xfrm_statistics_init(struct net *net)
2785 {
2786         int rv;
2787         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2788         if (!net->mib.xfrm_statistics)
2789                 return -ENOMEM;
2790         rv = xfrm_proc_init(net);
2791         if (rv < 0)
2792                 free_percpu(net->mib.xfrm_statistics);
2793         return rv;
2794 }
2795
2796 static void xfrm_statistics_fini(struct net *net)
2797 {
2798         xfrm_proc_fini(net);
2799         free_percpu(net->mib.xfrm_statistics);
2800 }
2801 #else
2802 static int __net_init xfrm_statistics_init(struct net *net)
2803 {
2804         return 0;
2805 }
2806
2807 static void xfrm_statistics_fini(struct net *net)
2808 {
2809 }
2810 #endif
2811
2812 static int __net_init xfrm_policy_init(struct net *net)
2813 {
2814         unsigned int hmask, sz;
2815         int dir;
2816
2817         if (net_eq(net, &init_net))
2818                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2819                                            sizeof(struct xfrm_dst),
2820                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2821                                            NULL);
2822
2823         hmask = 8 - 1;
2824         sz = (hmask+1) * sizeof(struct hlist_head);
2825
2826         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2827         if (!net->xfrm.policy_byidx)
2828                 goto out_byidx;
2829         net->xfrm.policy_idx_hmask = hmask;
2830
2831         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2832                 struct xfrm_policy_hash *htab;
2833
2834                 net->xfrm.policy_count[dir] = 0;
2835                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2836                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2837
2838                 htab = &net->xfrm.policy_bydst[dir];
2839                 htab->table = xfrm_hash_alloc(sz);
2840                 if (!htab->table)
2841                         goto out_bydst;
2842                 htab->hmask = hmask;
2843                 htab->dbits4 = 32;
2844                 htab->sbits4 = 32;
2845                 htab->dbits6 = 128;
2846                 htab->sbits6 = 128;
2847         }
2848         net->xfrm.policy_hthresh.lbits4 = 32;
2849         net->xfrm.policy_hthresh.rbits4 = 32;
2850         net->xfrm.policy_hthresh.lbits6 = 128;
2851         net->xfrm.policy_hthresh.rbits6 = 128;
2852
2853         seqlock_init(&net->xfrm.policy_hthresh.lock);
2854
2855         INIT_LIST_HEAD(&net->xfrm.policy_all);
2856         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2857         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2858         return 0;
2859
2860 out_bydst:
2861         for (dir--; dir >= 0; dir--) {
2862                 struct xfrm_policy_hash *htab;
2863
2864                 htab = &net->xfrm.policy_bydst[dir];
2865                 xfrm_hash_free(htab->table, sz);
2866         }
2867         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2868 out_byidx:
2869         return -ENOMEM;
2870 }
2871
2872 static void xfrm_policy_fini(struct net *net)
2873 {
2874         unsigned int sz;
2875         int dir;
2876
2877         flush_work(&net->xfrm.policy_hash_work);
2878 #ifdef CONFIG_XFRM_SUB_POLICY
2879         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2880 #endif
2881         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2882
2883         WARN_ON(!list_empty(&net->xfrm.policy_all));
2884
2885         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2886                 struct xfrm_policy_hash *htab;
2887
2888                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2889
2890                 htab = &net->xfrm.policy_bydst[dir];
2891                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2892                 WARN_ON(!hlist_empty(htab->table));
2893                 xfrm_hash_free(htab->table, sz);
2894         }
2895
2896         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2897         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2898         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2899 }
2900
2901 static int __net_init xfrm_net_init(struct net *net)
2902 {
2903         int rv;
2904
2905         /* Initialize the per-net locks here */
2906         spin_lock_init(&net->xfrm.xfrm_state_lock);
2907         spin_lock_init(&net->xfrm.xfrm_policy_lock);
2908         mutex_init(&net->xfrm.xfrm_cfg_mutex);
2909
2910         rv = xfrm_statistics_init(net);
2911         if (rv < 0)
2912                 goto out_statistics;
2913         rv = xfrm_state_init(net);
2914         if (rv < 0)
2915                 goto out_state;
2916         rv = xfrm_policy_init(net);
2917         if (rv < 0)
2918                 goto out_policy;
2919         rv = xfrm_sysctl_init(net);
2920         if (rv < 0)
2921                 goto out_sysctl;
2922
2923         return 0;
2924
2925 out_sysctl:
2926         xfrm_policy_fini(net);
2927 out_policy:
2928         xfrm_state_fini(net);
2929 out_state:
2930         xfrm_statistics_fini(net);
2931 out_statistics:
2932         return rv;
2933 }
2934
2935 static void __net_exit xfrm_net_exit(struct net *net)
2936 {
2937         xfrm_sysctl_fini(net);
2938         xfrm_policy_fini(net);
2939         xfrm_state_fini(net);
2940         xfrm_statistics_fini(net);
2941 }
2942
2943 static struct pernet_operations __net_initdata xfrm_net_ops = {
2944         .init = xfrm_net_init,
2945         .exit = xfrm_net_exit,
2946 };
2947
2948 void __init xfrm_init(void)
2949 {
2950         register_pernet_subsys(&xfrm_net_ops);
2951         xfrm_dev_init();
2952         seqcount_init(&xfrm_policy_hash_generation);
2953         xfrm_input_init();
2954
2955         RCU_INIT_POINTER(xfrm_if_cb, NULL);
2956         synchronize_rcu();
2957 }
2958
2959 #ifdef CONFIG_AUDITSYSCALL
2960 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2961                                          struct audit_buffer *audit_buf)
2962 {
2963         struct xfrm_sec_ctx *ctx = xp->security;
2964         struct xfrm_selector *sel = &xp->selector;
2965
2966         if (ctx)
2967                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2968                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2969
2970         switch (sel->family) {
2971         case AF_INET:
2972                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2973                 if (sel->prefixlen_s != 32)
2974                         audit_log_format(audit_buf, " src_prefixlen=%d",
2975                                          sel->prefixlen_s);
2976                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2977                 if (sel->prefixlen_d != 32)
2978                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2979                                          sel->prefixlen_d);
2980                 break;
2981         case AF_INET6:
2982                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2983                 if (sel->prefixlen_s != 128)
2984                         audit_log_format(audit_buf, " src_prefixlen=%d",
2985                                          sel->prefixlen_s);
2986                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2987                 if (sel->prefixlen_d != 128)
2988                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2989                                          sel->prefixlen_d);
2990                 break;
2991         }
2992 }
2993
2994 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
2995 {
2996         struct audit_buffer *audit_buf;
2997
2998         audit_buf = xfrm_audit_start("SPD-add");
2999         if (audit_buf == NULL)
3000                 return;
3001         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3002         audit_log_format(audit_buf, " res=%u", result);
3003         xfrm_audit_common_policyinfo(xp, audit_buf);
3004         audit_log_end(audit_buf);
3005 }
3006 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3007
3008 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3009                               bool task_valid)
3010 {
3011         struct audit_buffer *audit_buf;
3012
3013         audit_buf = xfrm_audit_start("SPD-delete");
3014         if (audit_buf == NULL)
3015                 return;
3016         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3017         audit_log_format(audit_buf, " res=%u", result);
3018         xfrm_audit_common_policyinfo(xp, audit_buf);
3019         audit_log_end(audit_buf);
3020 }
3021 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3022 #endif
3023
3024 #ifdef CONFIG_XFRM_MIGRATE
3025 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3026                                         const struct xfrm_selector *sel_tgt)
3027 {
3028         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3029                 if (sel_tgt->family == sel_cmp->family &&
3030                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3031                                     sel_cmp->family) &&
3032                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3033                                     sel_cmp->family) &&
3034                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3035                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3036                         return true;
3037                 }
3038         } else {
3039                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3040                         return true;
3041                 }
3042         }
3043         return false;
3044 }
3045
3046 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3047                                                     u8 dir, u8 type, struct net *net, u32 if_id)
3048 {
3049         struct xfrm_policy *pol, *ret = NULL;
3050         struct hlist_head *chain;
3051         u32 priority = ~0U;
3052
3053         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3054         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3055         hlist_for_each_entry(pol, chain, bydst) {
3056                 if ((if_id == 0 || pol->if_id == if_id) &&
3057                     xfrm_migrate_selector_match(sel, &pol->selector) &&
3058                     pol->type == type) {
3059                         ret = pol;
3060                         priority = ret->priority;
3061                         break;
3062                 }
3063         }
3064         chain = &net->xfrm.policy_inexact[dir];
3065         hlist_for_each_entry(pol, chain, bydst) {
3066                 if ((pol->priority >= priority) && ret)
3067                         break;
3068
3069                 if ((if_id == 0 || pol->if_id == if_id) &&
3070                     xfrm_migrate_selector_match(sel, &pol->selector) &&
3071                     pol->type == type) {
3072                         ret = pol;
3073                         break;
3074                 }
3075         }
3076
3077         xfrm_pol_hold(ret);
3078
3079         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3080
3081         return ret;
3082 }
3083
3084 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3085 {
3086         int match = 0;
3087
3088         if (t->mode == m->mode && t->id.proto == m->proto &&
3089             (m->reqid == 0 || t->reqid == m->reqid)) {
3090                 switch (t->mode) {
3091                 case XFRM_MODE_TUNNEL:
3092                 case XFRM_MODE_BEET:
3093                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3094                                             m->old_family) &&
3095                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3096                                             m->old_family)) {
3097                                 match = 1;
3098                         }
3099                         break;
3100                 case XFRM_MODE_TRANSPORT:
3101                         /* in case of transport mode, template does not store
3102                            any IP addresses, hence we just compare mode and
3103                            protocol */
3104                         match = 1;
3105                         break;
3106                 default:
3107                         break;
3108                 }
3109         }
3110         return match;
3111 }
3112
3113 /* update endpoint address(es) of template(s) */
3114 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3115                                struct xfrm_migrate *m, int num_migrate)
3116 {
3117         struct xfrm_migrate *mp;
3118         int i, j, n = 0;
3119
3120         write_lock_bh(&pol->lock);
3121         if (unlikely(pol->walk.dead)) {
3122                 /* target policy has been deleted */
3123                 write_unlock_bh(&pol->lock);
3124                 return -ENOENT;
3125         }
3126
3127         for (i = 0; i < pol->xfrm_nr; i++) {
3128                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3129                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3130                                 continue;
3131                         n++;
3132                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3133                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3134                                 continue;
3135                         /* update endpoints */
3136                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3137                                sizeof(pol->xfrm_vec[i].id.daddr));
3138                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3139                                sizeof(pol->xfrm_vec[i].saddr));
3140                         pol->xfrm_vec[i].encap_family = mp->new_family;
3141                         /* flush bundles */
3142                         atomic_inc(&pol->genid);
3143                 }
3144         }
3145
3146         write_unlock_bh(&pol->lock);
3147
3148         if (!n)
3149                 return -ENODATA;
3150
3151         return 0;
3152 }
3153
3154 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3155 {
3156         int i, j;
3157
3158         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3159                 return -EINVAL;
3160
3161         for (i = 0; i < num_migrate; i++) {
3162                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3163                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3164                         return -EINVAL;
3165
3166                 /* check if there is any duplicated entry */
3167                 for (j = i + 1; j < num_migrate; j++) {
3168                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3169                                     sizeof(m[i].old_daddr)) &&
3170                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3171                                     sizeof(m[i].old_saddr)) &&
3172                             m[i].proto == m[j].proto &&
3173                             m[i].mode == m[j].mode &&
3174                             m[i].reqid == m[j].reqid &&
3175                             m[i].old_family == m[j].old_family)
3176                                 return -EINVAL;
3177                 }
3178         }
3179
3180         return 0;
3181 }
3182
3183 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3184                  struct xfrm_migrate *m, int num_migrate,
3185                  struct xfrm_kmaddress *k, struct net *net,
3186                  struct xfrm_encap_tmpl *encap, u32 if_id)
3187 {
3188         int i, err, nx_cur = 0, nx_new = 0;
3189         struct xfrm_policy *pol = NULL;
3190         struct xfrm_state *x, *xc;
3191         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3192         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3193         struct xfrm_migrate *mp;
3194
3195         /* Stage 0 - sanity checks */
3196         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3197                 goto out;
3198
3199         if (dir >= XFRM_POLICY_MAX) {
3200                 err = -EINVAL;
3201                 goto out;
3202         }
3203
3204         /* Stage 1 - find policy */
3205         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id)) == NULL) {
3206                 err = -ENOENT;
3207                 goto out;
3208         }
3209
3210         /* Stage 2 - find and update state(s) */
3211         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3212                 if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
3213                         x_cur[nx_cur] = x;
3214                         nx_cur++;
3215                         xc = xfrm_state_migrate(x, mp, encap);
3216                         if (xc) {
3217                                 x_new[nx_new] = xc;
3218                                 nx_new++;
3219                         } else {
3220                                 err = -ENODATA;
3221                                 goto restore_state;
3222                         }
3223                 }
3224         }
3225
3226         /* Stage 3 - update policy */
3227         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3228                 goto restore_state;
3229
3230         /* Stage 4 - delete old state(s) */
3231         if (nx_cur) {
3232                 xfrm_states_put(x_cur, nx_cur);
3233                 xfrm_states_delete(x_cur, nx_cur);
3234         }
3235
3236         /* Stage 5 - announce */
3237         km_migrate(sel, dir, type, m, num_migrate, k, encap);
3238
3239         xfrm_pol_put(pol);
3240
3241         return 0;
3242 out:
3243         return err;
3244
3245 restore_state:
3246         if (pol)
3247                 xfrm_pol_put(pol);
3248         if (nx_cur)
3249                 xfrm_states_put(x_cur, nx_cur);
3250         if (nx_new)
3251                 xfrm_states_delete(x_new, nx_new);
3252
3253         return err;
3254 }
3255 EXPORT_SYMBOL(xfrm_migrate);
3256 #endif