net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <[email protected]>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <[email protected]>          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_PER_CPU(struct xfrm_dst *, xfrm_last_dst);
  49 static struct work_struct *xfrm_pcpu_work __read_mostly;
  50 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  51 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
  52                                                 __read_mostly;
  53
  54 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  55 static __read_mostly seqcount_t xfrm_policy_hash_generation;
  56
  57 static void xfrm_init_pmtu(struct dst_entry *dst);
  58 static int stale_bundle(struct dst_entry *dst);
  59 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  60 static void xfrm_policy_queue_process(struct timer_list *t);
  61
  62 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
  63 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  64                                                 int dir);
  65
  66 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
  67 {
  68         return refcount_inc_not_zero(&policy->refcnt);
  69 }
  70
  71 static inline bool
  72 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  73 {
  74         const struct flowi4 *fl4 = &fl->u.ip4;
  75
  76         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  77                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  78                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  79                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  80                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  81                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  82 }
  83
  84 static inline bool
  85 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  86 {
  87         const struct flowi6 *fl6 = &fl->u.ip6;
  88
  89         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  90                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  91                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  92                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  93                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  94                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  95 }
  96
  97 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  98                          unsigned short family)
  99 {
 100         switch (family) {
 101         case AF_INET:
 102                 return __xfrm4_selector_match(sel, fl);
 103         case AF_INET6:
 104                 return __xfrm6_selector_match(sel, fl);
 105         }
 106         return false;
 107 }
 108
 109 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 110 {
 111         const struct xfrm_policy_afinfo *afinfo;
 112
 113         if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 114                 return NULL;
 115         rcu_read_lock();
 116         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 117         if (unlikely(!afinfo))
 118                 rcu_read_unlock();
 119         return afinfo;
 120 }
 121
 122 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
 123                                     const xfrm_address_t *saddr,
 124                                     const xfrm_address_t *daddr,
 125                                     int family, u32 mark)
 126 {
 127         const struct xfrm_policy_afinfo *afinfo;
 128         struct dst_entry *dst;
 129
 130         afinfo = xfrm_policy_get_afinfo(family);
 131         if (unlikely(afinfo == NULL))
 132                 return ERR_PTR(-EAFNOSUPPORT);
 133
 134         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
 135
 136         rcu_read_unlock();
 137
 138         return dst;
 139 }
 140 EXPORT_SYMBOL(__xfrm_dst_lookup);
 141
 142 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 143                                                 int tos, int oif,
 144                                                 xfrm_address_t *prev_saddr,
 145                                                 xfrm_address_t *prev_daddr,
 146                                                 int family, u32 mark)
 147 {
 148         struct net *net = xs_net(x);
 149         xfrm_address_t *saddr = &x->props.saddr;
 150         xfrm_address_t *daddr = &x->id.daddr;
 151         struct dst_entry *dst;
 152
 153         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 154                 saddr = x->coaddr;
 155                 daddr = prev_daddr;
 156         }
 157         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 158                 saddr = prev_saddr;
 159                 daddr = x->coaddr;
 160         }
 161
 162         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
 163
 164         if (!IS_ERR(dst)) {
 165                 if (prev_saddr != saddr)
 166                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 167                 if (prev_daddr != daddr)
 168                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 169         }
 170
 171         return dst;
 172 }
 173
 174 static inline unsigned long make_jiffies(long secs)
 175 {
 176         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 177                 return MAX_SCHEDULE_TIMEOUT-1;
 178         else
 179                 return secs*HZ;
 180 }
 181
 182 static void xfrm_policy_timer(struct timer_list *t)
 183 {
 184         struct xfrm_policy *xp = from_timer(xp, t, timer);
 185         unsigned long now = get_seconds();
 186         long next = LONG_MAX;
 187         int warn = 0;
 188         int dir;
 189
 190         read_lock(&xp->lock);
 191
 192         if (unlikely(xp->walk.dead))
 193                 goto out;
 194
 195         dir = xfrm_policy_id2dir(xp->index);
 196
 197         if (xp->lft.hard_add_expires_seconds) {
 198                 long tmo = xp->lft.hard_add_expires_seconds +
 199                         xp->curlft.add_time - now;
 200                 if (tmo <= 0)
 201                         goto expired;
 202                 if (tmo < next)
 203                         next = tmo;
 204         }
 205         if (xp->lft.hard_use_expires_seconds) {
 206                 long tmo = xp->lft.hard_use_expires_seconds +
 207                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 208                 if (tmo <= 0)
 209                         goto expired;
 210                 if (tmo < next)
 211                         next = tmo;
 212         }
 213         if (xp->lft.soft_add_expires_seconds) {
 214                 long tmo = xp->lft.soft_add_expires_seconds +
 215                         xp->curlft.add_time - now;
 216                 if (tmo <= 0) {
 217                         warn = 1;
 218                         tmo = XFRM_KM_TIMEOUT;
 219                 }
 220                 if (tmo < next)
 221                         next = tmo;
 222         }
 223         if (xp->lft.soft_use_expires_seconds) {
 224                 long tmo = xp->lft.soft_use_expires_seconds +
 225                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 226                 if (tmo <= 0) {
 227                         warn = 1;
 228                         tmo = XFRM_KM_TIMEOUT;
 229                 }
 230                 if (tmo < next)
 231                         next = tmo;
 232         }
 233
 234         if (warn)
 235                 km_policy_expired(xp, dir, 0, 0);
 236         if (next != LONG_MAX &&
 237             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 238                 xfrm_pol_hold(xp);
 239
 240 out:
 241         read_unlock(&xp->lock);
 242         xfrm_pol_put(xp);
 243         return;
 244
 245 expired:
 246         read_unlock(&xp->lock);
 247         if (!xfrm_policy_delete(xp, dir))
 248                 km_policy_expired(xp, dir, 1, 0);
 249         xfrm_pol_put(xp);
 250 }
 251
 252 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 253  * SPD calls.
 254  */
 255
 256 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 257 {
 258         struct xfrm_policy *policy;
 259
 260         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 261
 262         if (policy) {
 263                 write_pnet(&policy->xp_net, net);
 264                 INIT_LIST_HEAD(&policy->walk.all);
 265                 INIT_HLIST_NODE(&policy->bydst);
 266                 INIT_HLIST_NODE(&policy->byidx);
 267                 rwlock_init(&policy->lock);
 268                 refcount_set(&policy->refcnt, 1);
 269                 skb_queue_head_init(&policy->polq.hold_queue);
 270                 timer_setup(&policy->timer, xfrm_policy_timer, 0);
 271                 timer_setup(&policy->polq.hold_timer,
 272                             xfrm_policy_queue_process, 0);
 273         }
 274         return policy;
 275 }
 276 EXPORT_SYMBOL(xfrm_policy_alloc);
 277
 278 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 279 {
 280         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 281
 282         security_xfrm_policy_free(policy->security);
 283         kfree(policy);
 284 }
 285
 286 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 287
 288 void xfrm_policy_destroy(struct xfrm_policy *policy)
 289 {
 290         BUG_ON(!policy->walk.dead);
 291
 292         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 293                 BUG();
 294
 295         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 296 }
 297 EXPORT_SYMBOL(xfrm_policy_destroy);
 298
 299 /* Rule must be locked. Release descendant resources, announce
 300  * entry dead. The rule must be unlinked from lists to the moment.
 301  */
 302
 303 static void xfrm_policy_kill(struct xfrm_policy *policy)
 304 {
 305         policy->walk.dead = 1;
 306
 307         atomic_inc(&policy->genid);
 308
 309         if (del_timer(&policy->polq.hold_timer))
 310                 xfrm_pol_put(policy);
 311         skb_queue_purge(&policy->polq.hold_queue);
 312
 313         if (del_timer(&policy->timer))
 314                 xfrm_pol_put(policy);
 315
 316         xfrm_pol_put(policy);
 317 }
 318
 319 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 320
 321 static inline unsigned int idx_hash(struct net *net, u32 index)
 322 {
 323         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 324 }
 325
 326 /* calculate policy hash thresholds */
 327 static void __get_hash_thresh(struct net *net,
 328                               unsigned short family, int dir,
 329                               u8 *dbits, u8 *sbits)
 330 {
 331         switch (family) {
 332         case AF_INET:
 333                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 334                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 335                 break;
 336
 337         case AF_INET6:
 338                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 339                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 340                 break;
 341
 342         default:
 343                 *dbits = 0;
 344                 *sbits = 0;
 345         }
 346 }
 347
 348 static struct hlist_head *policy_hash_bysel(struct net *net,
 349                                             const struct xfrm_selector *sel,
 350                                             unsigned short family, int dir)
 351 {
 352         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 353         unsigned int hash;
 354         u8 dbits;
 355         u8 sbits;
 356
 357         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 358         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 359
 360         if (hash == hmask + 1)
 361                 return &net->xfrm.policy_inexact[dir];
 362
 363         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 364                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 365 }
 366
 367 static struct hlist_head *policy_hash_direct(struct net *net,
 368                                              const xfrm_address_t *daddr,
 369                                              const xfrm_address_t *saddr,
 370                                              unsigned short family, int dir)
 371 {
 372         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 373         unsigned int hash;
 374         u8 dbits;
 375         u8 sbits;
 376
 377         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 378         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 379
 380         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 381                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 382 }
 383
 384 static void xfrm_dst_hash_transfer(struct net *net,
 385                                    struct hlist_head *list,
 386                                    struct hlist_head *ndsttable,
 387                                    unsigned int nhashmask,
 388                                    int dir)
 389 {
 390         struct hlist_node *tmp, *entry0 = NULL;
 391         struct xfrm_policy *pol;
 392         unsigned int h0 = 0;
 393         u8 dbits;
 394         u8 sbits;
 395
 396 redo:
 397         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 398                 unsigned int h;
 399
 400                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 401                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 402                                 pol->family, nhashmask, dbits, sbits);
 403                 if (!entry0) {
 404                         hlist_del_rcu(&pol->bydst);
 405                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 406                         h0 = h;
 407                 } else {
 408                         if (h != h0)
 409                                 continue;
 410                         hlist_del_rcu(&pol->bydst);
 411                         hlist_add_behind_rcu(&pol->bydst, entry0);
 412                 }
 413                 entry0 = &pol->bydst;
 414         }
 415         if (!hlist_empty(list)) {
 416                 entry0 = NULL;
 417                 goto redo;
 418         }
 419 }
 420
 421 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 422                                    struct hlist_head *nidxtable,
 423                                    unsigned int nhashmask)
 424 {
 425         struct hlist_node *tmp;
 426         struct xfrm_policy *pol;
 427
 428         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 429                 unsigned int h;
 430
 431                 h = __idx_hash(pol->index, nhashmask);
 432                 hlist_add_head(&pol->byidx, nidxtable+h);
 433         }
 434 }
 435
 436 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 437 {
 438         return ((old_hmask + 1) << 1) - 1;
 439 }
 440
 441 static void xfrm_bydst_resize(struct net *net, int dir)
 442 {
 443         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 444         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 445         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 446         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 447         struct hlist_head *odst;
 448         int i;
 449
 450         if (!ndst)
 451                 return;
 452
 453         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 454         write_seqcount_begin(&xfrm_policy_hash_generation);
 455
 456         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 457                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 458
 459         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 460                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 461
 462         for (i = hmask; i >= 0; i--)
 463                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 464
 465         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 466         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 467
 468         write_seqcount_end(&xfrm_policy_hash_generation);
 469         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 470
 471         synchronize_rcu();
 472
 473         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 474 }
 475
 476 static void xfrm_byidx_resize(struct net *net, int total)
 477 {
 478         unsigned int hmask = net->xfrm.policy_idx_hmask;
 479         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 480         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 481         struct hlist_head *oidx = net->xfrm.policy_byidx;
 482         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 483         int i;
 484
 485         if (!nidx)
 486                 return;
 487
 488         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 489
 490         for (i = hmask; i >= 0; i--)
 491                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 492
 493         net->xfrm.policy_byidx = nidx;
 494         net->xfrm.policy_idx_hmask = nhashmask;
 495
 496         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 497
 498         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 499 }
 500
 501 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 502 {
 503         unsigned int cnt = net->xfrm.policy_count[dir];
 504         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 505
 506         if (total)
 507                 *total += cnt;
 508
 509         if ((hmask + 1) < xfrm_policy_hashmax &&
 510             cnt > hmask)
 511                 return 1;
 512
 513         return 0;
 514 }
 515
 516 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 517 {
 518         unsigned int hmask = net->xfrm.policy_idx_hmask;
 519
 520         if ((hmask + 1) < xfrm_policy_hashmax &&
 521             total > hmask)
 522                 return 1;
 523
 524         return 0;
 525 }
 526
 527 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 528 {
 529         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 530         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 531         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 532         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 533         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 534         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 535         si->spdhcnt = net->xfrm.policy_idx_hmask;
 536         si->spdhmcnt = xfrm_policy_hashmax;
 537 }
 538 EXPORT_SYMBOL(xfrm_spd_getinfo);
 539
 540 static DEFINE_MUTEX(hash_resize_mutex);
 541 static void xfrm_hash_resize(struct work_struct *work)
 542 {
 543         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 544         int dir, total;
 545
 546         mutex_lock(&hash_resize_mutex);
 547
 548         total = 0;
 549         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 550                 if (xfrm_bydst_should_resize(net, dir, &total))
 551                         xfrm_bydst_resize(net, dir);
 552         }
 553         if (xfrm_byidx_should_resize(net, total))
 554                 xfrm_byidx_resize(net, total);
 555
 556         mutex_unlock(&hash_resize_mutex);
 557 }
 558
 559 static void xfrm_hash_rebuild(struct work_struct *work)
 560 {
 561         struct net *net = container_of(work, struct net,
 562                                        xfrm.policy_hthresh.work);
 563         unsigned int hmask;
 564         struct xfrm_policy *pol;
 565         struct xfrm_policy *policy;
 566         struct hlist_head *chain;
 567         struct hlist_head *odst;
 568         struct hlist_node *newpos;
 569         int i;
 570         int dir;
 571         unsigned seq;
 572         u8 lbits4, rbits4, lbits6, rbits6;
 573
 574         mutex_lock(&hash_resize_mutex);
 575
 576         /* read selector prefixlen thresholds */
 577         do {
 578                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
 579
 580                 lbits4 = net->xfrm.policy_hthresh.lbits4;
 581                 rbits4 = net->xfrm.policy_hthresh.rbits4;
 582                 lbits6 = net->xfrm.policy_hthresh.lbits6;
 583                 rbits6 = net->xfrm.policy_hthresh.rbits6;
 584         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
 585
 586         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 587
 588         /* reset the bydst and inexact table in all directions */
 589         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 590                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
 591                 hmask = net->xfrm.policy_bydst[dir].hmask;
 592                 odst = net->xfrm.policy_bydst[dir].table;
 593                 for (i = hmask; i >= 0; i--)
 594                         INIT_HLIST_HEAD(odst + i);
 595                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
 596                         /* dir out => dst = remote, src = local */
 597                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
 598                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
 599                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
 600                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
 601                 } else {
 602                         /* dir in/fwd => dst = local, src = remote */
 603                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
 604                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
 605                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
 606                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
 607                 }
 608         }
 609
 610         /* re-insert all policies by order of creation */
 611         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 612                 if (xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
 613                         /* skip socket policies */
 614                         continue;
 615                 }
 616                 newpos = NULL;
 617                 chain = policy_hash_bysel(net, &policy->selector,
 618                                           policy->family,
 619                                           xfrm_policy_id2dir(policy->index));
 620                 hlist_for_each_entry(pol, chain, bydst) {
 621                         if (policy->priority >= pol->priority)
 622                                 newpos = &pol->bydst;
 623                         else
 624                                 break;
 625                 }
 626                 if (newpos)
 627                         hlist_add_behind(&policy->bydst, newpos);
 628                 else
 629                         hlist_add_head(&policy->bydst, chain);
 630         }
 631
 632         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 633
 634         mutex_unlock(&hash_resize_mutex);
 635 }
 636
 637 void xfrm_policy_hash_rebuild(struct net *net)
 638 {
 639         schedule_work(&net->xfrm.policy_hthresh.work);
 640 }
 641 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 642
 643 /* Generate new index... KAME seems to generate them ordered by cost
 644  * of an absolute inpredictability of ordering of rules. This will not pass. */
 645 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 646 {
 647         static u32 idx_generator;
 648
 649         for (;;) {
 650                 struct hlist_head *list;
 651                 struct xfrm_policy *p;
 652                 u32 idx;
 653                 int found;
 654
 655                 if (!index) {
 656                         idx = (idx_generator | dir);
 657                         idx_generator += 8;
 658                 } else {
 659                         idx = index;
 660                         index = 0;
 661                 }
 662
 663                 if (idx == 0)
 664                         idx = 8;
 665                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 666                 found = 0;
 667                 hlist_for_each_entry(p, list, byidx) {
 668                         if (p->index == idx) {
 669                                 found = 1;
 670                                 break;
 671                         }
 672                 }
 673                 if (!found)
 674                         return idx;
 675         }
 676 }
 677
 678 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 679 {
 680         u32 *p1 = (u32 *) s1;
 681         u32 *p2 = (u32 *) s2;
 682         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 683         int i;
 684
 685         for (i = 0; i < len; i++) {
 686                 if (p1[i] != p2[i])
 687                         return 1;
 688         }
 689
 690         return 0;
 691 }
 692
 693 static void xfrm_policy_requeue(struct xfrm_policy *old,
 694                                 struct xfrm_policy *new)
 695 {
 696         struct xfrm_policy_queue *pq = &old->polq;
 697         struct sk_buff_head list;
 698
 699         if (skb_queue_empty(&pq->hold_queue))
 700                 return;
 701
 702         __skb_queue_head_init(&list);
 703
 704         spin_lock_bh(&pq->hold_queue.lock);
 705         skb_queue_splice_init(&pq->hold_queue, &list);
 706         if (del_timer(&pq->hold_timer))
 707                 xfrm_pol_put(old);
 708         spin_unlock_bh(&pq->hold_queue.lock);
 709
 710         pq = &new->polq;
 711
 712         spin_lock_bh(&pq->hold_queue.lock);
 713         skb_queue_splice(&list, &pq->hold_queue);
 714         pq->timeout = XFRM_QUEUE_TMO_MIN;
 715         if (!mod_timer(&pq->hold_timer, jiffies))
 716                 xfrm_pol_hold(new);
 717         spin_unlock_bh(&pq->hold_queue.lock);
 718 }
 719
 720 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 721                                    struct xfrm_policy *pol)
 722 {
 723         u32 mark = policy->mark.v & policy->mark.m;
 724
 725         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 726                 return true;
 727
 728         if ((mark & pol->mark.m) == pol->mark.v &&
 729             policy->priority == pol->priority)
 730                 return true;
 731
 732         return false;
 733 }
 734
 735 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 736 {
 737         struct net *net = xp_net(policy);
 738         struct xfrm_policy *pol;
 739         struct xfrm_policy *delpol;
 740         struct hlist_head *chain;
 741         struct hlist_node *newpos;
 742
 743         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 744         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 745         delpol = NULL;
 746         newpos = NULL;
 747         hlist_for_each_entry(pol, chain, bydst) {
 748                 if (pol->type == policy->type &&
 749                     !selector_cmp(&pol->selector, &policy->selector) &&
 750                     xfrm_policy_mark_match(policy, pol) &&
 751                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 752                     !WARN_ON(delpol)) {
 753                         if (excl) {
 754                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 755                                 return -EEXIST;
 756                         }
 757                         delpol = pol;
 758                         if (policy->priority > pol->priority)
 759                                 continue;
 760                 } else if (policy->priority >= pol->priority) {
 761                         newpos = &pol->bydst;
 762                         continue;
 763                 }
 764                 if (delpol)
 765                         break;
 766         }
 767         if (newpos)
 768                 hlist_add_behind(&policy->bydst, newpos);
 769         else
 770                 hlist_add_head(&policy->bydst, chain);
 771         __xfrm_policy_link(policy, dir);
 772
 773         /* After previous checking, family can either be AF_INET or AF_INET6 */
 774         if (policy->family == AF_INET)
 775                 rt_genid_bump_ipv4(net);
 776         else
 777                 rt_genid_bump_ipv6(net);
 778
 779         if (delpol) {
 780                 xfrm_policy_requeue(delpol, policy);
 781                 __xfrm_policy_unlink(delpol, dir);
 782         }
 783         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 784         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 785         policy->curlft.add_time = get_seconds();
 786         policy->curlft.use_time = 0;
 787         if (!mod_timer(&policy->timer, jiffies + HZ))
 788                 xfrm_pol_hold(policy);
 789         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 790
 791         if (delpol)
 792                 xfrm_policy_kill(delpol);
 793         else if (xfrm_bydst_should_resize(net, dir, NULL))
 794                 schedule_work(&net->xfrm.policy_hash_work);
 795
 796         return 0;
 797 }
 798 EXPORT_SYMBOL(xfrm_policy_insert);
 799
 800 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 801                                           int dir, struct xfrm_selector *sel,
 802                                           struct xfrm_sec_ctx *ctx, int delete,
 803                                           int *err)
 804 {
 805         struct xfrm_policy *pol, *ret;
 806         struct hlist_head *chain;
 807
 808         *err = 0;
 809         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 810         chain = policy_hash_bysel(net, sel, sel->family, dir);
 811         ret = NULL;
 812         hlist_for_each_entry(pol, chain, bydst) {
 813                 if (pol->type == type &&
 814                     (mark & pol->mark.m) == pol->mark.v &&
 815                     !selector_cmp(sel, &pol->selector) &&
 816                     xfrm_sec_ctx_match(ctx, pol->security)) {
 817                         xfrm_pol_hold(pol);
 818                         if (delete) {
 819                                 *err = security_xfrm_policy_delete(
 820                                                                 pol->security);
 821                                 if (*err) {
 822                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 823                                         return pol;
 824                                 }
 825                                 __xfrm_policy_unlink(pol, dir);
 826                         }
 827                         ret = pol;
 828                         break;
 829                 }
 830         }
 831         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 832
 833         if (ret && delete)
 834                 xfrm_policy_kill(ret);
 835         return ret;
 836 }
 837 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 838
 839 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 840                                      int dir, u32 id, int delete, int *err)
 841 {
 842         struct xfrm_policy *pol, *ret;
 843         struct hlist_head *chain;
 844
 845         *err = -ENOENT;
 846         if (xfrm_policy_id2dir(id) != dir)
 847                 return NULL;
 848
 849         *err = 0;
 850         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 851         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 852         ret = NULL;
 853         hlist_for_each_entry(pol, chain, byidx) {
 854                 if (pol->type == type && pol->index == id &&
 855                     (mark & pol->mark.m) == pol->mark.v) {
 856                         xfrm_pol_hold(pol);
 857                         if (delete) {
 858                                 *err = security_xfrm_policy_delete(
 859                                                                 pol->security);
 860                                 if (*err) {
 861                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 862                                         return pol;
 863                                 }
 864                                 __xfrm_policy_unlink(pol, dir);
 865                         }
 866                         ret = pol;
 867                         break;
 868                 }
 869         }
 870         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 871
 872         if (ret && delete)
 873                 xfrm_policy_kill(ret);
 874         return ret;
 875 }
 876 EXPORT_SYMBOL(xfrm_policy_byid);
 877
 878 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 879 static inline int
 880 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 881 {
 882         int dir, err = 0;
 883
 884         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 885                 struct xfrm_policy *pol;
 886                 int i;
 887
 888                 hlist_for_each_entry(pol,
 889                                      &net->xfrm.policy_inexact[dir], bydst) {
 890                         if (pol->type != type)
 891                                 continue;
 892                         err = security_xfrm_policy_delete(pol->security);
 893                         if (err) {
 894                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 895                                 return err;
 896                         }
 897                 }
 898                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 899                         hlist_for_each_entry(pol,
 900                                              net->xfrm.policy_bydst[dir].table + i,
 901                                              bydst) {
 902                                 if (pol->type != type)
 903                                         continue;
 904                                 err = security_xfrm_policy_delete(
 905                                                                 pol->security);
 906                                 if (err) {
 907                                         xfrm_audit_policy_delete(pol, 0,
 908                                                                  task_valid);
 909                                         return err;
 910                                 }
 911                         }
 912                 }
 913         }
 914         return err;
 915 }
 916 #else
 917 static inline int
 918 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 919 {
 920         return 0;
 921 }
 922 #endif
 923
 924 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 925 {
 926         int dir, err = 0, cnt = 0;
 927
 928         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 929
 930         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 931         if (err)
 932                 goto out;
 933
 934         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 935                 struct xfrm_policy *pol;
 936                 int i;
 937
 938         again1:
 939                 hlist_for_each_entry(pol,
 940                                      &net->xfrm.policy_inexact[dir], bydst) {
 941                         if (pol->type != type)
 942                                 continue;
 943                         __xfrm_policy_unlink(pol, dir);
 944                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 945                         cnt++;
 946
 947                         xfrm_audit_policy_delete(pol, 1, task_valid);
 948
 949                         xfrm_policy_kill(pol);
 950
 951                         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 952                         goto again1;
 953                 }
 954
 955                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 956         again2:
 957                         hlist_for_each_entry(pol,
 958                                              net->xfrm.policy_bydst[dir].table + i,
 959                                              bydst) {
 960                                 if (pol->type != type)
 961                                         continue;
 962                                 __xfrm_policy_unlink(pol, dir);
 963                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 964                                 cnt++;
 965
 966                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 967                                 xfrm_policy_kill(pol);
 968
 969                                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 970                                 goto again2;
 971                         }
 972                 }
 973
 974         }
 975         if (!cnt)
 976                 err = -ESRCH;
 977         else
 978                 xfrm_policy_cache_flush();
 979 out:
 980         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 981         return err;
 982 }
 983 EXPORT_SYMBOL(xfrm_policy_flush);
 984
 985 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 986                      int (*func)(struct xfrm_policy *, int, int, void*),
 987                      void *data)
 988 {
 989         struct xfrm_policy *pol;
 990         struct xfrm_policy_walk_entry *x;
 991         int error = 0;
 992
 993         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 994             walk->type != XFRM_POLICY_TYPE_ANY)
 995                 return -EINVAL;
 996
 997         if (list_empty(&walk->walk.all) && walk->seq != 0)
 998                 return 0;
 999
1000         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1001         if (list_empty(&walk->walk.all))
1002                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1003         else
1004                 x = list_first_entry(&walk->walk.all,
1005                                      struct xfrm_policy_walk_entry, all);
1006
1007         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1008                 if (x->dead)
1009                         continue;
1010                 pol = container_of(x, struct xfrm_policy, walk);
1011                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1012                     walk->type != pol->type)
1013                         continue;
1014                 error = func(pol, xfrm_policy_id2dir(pol->index),
1015                              walk->seq, data);
1016                 if (error) {
1017                         list_move_tail(&walk->walk.all, &x->all);
1018                         goto out;
1019                 }
1020                 walk->seq++;
1021         }
1022         if (walk->seq == 0) {
1023                 error = -ENOENT;
1024                 goto out;
1025         }
1026         list_del_init(&walk->walk.all);
1027 out:
1028         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1029         return error;
1030 }
1031 EXPORT_SYMBOL(xfrm_policy_walk);
1032
1033 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1034 {
1035         INIT_LIST_HEAD(&walk->walk.all);
1036         walk->walk.dead = 1;
1037         walk->type = type;
1038         walk->seq = 0;
1039 }
1040 EXPORT_SYMBOL(xfrm_policy_walk_init);
1041
1042 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1043 {
1044         if (list_empty(&walk->walk.all))
1045                 return;
1046
1047         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1048         list_del(&walk->walk.all);
1049         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1050 }
1051 EXPORT_SYMBOL(xfrm_policy_walk_done);
1052
1053 /*
1054  * Find policy to apply to this flow.
1055  *
1056  * Returns 0 if policy found, else an -errno.
1057  */
1058 static int xfrm_policy_match(const struct xfrm_policy *pol,
1059                              const struct flowi *fl,
1060                              u8 type, u16 family, int dir)
1061 {
1062         const struct xfrm_selector *sel = &pol->selector;
1063         int ret = -ESRCH;
1064         bool match;
1065
1066         if (pol->family != family ||
1067             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1068             pol->type != type)
1069                 return ret;
1070
1071         match = xfrm_selector_match(sel, fl, family);
1072         if (match)
1073                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1074                                                   dir);
1075
1076         return ret;
1077 }
1078
1079 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1080                                                      const struct flowi *fl,
1081                                                      u16 family, u8 dir)
1082 {
1083         int err;
1084         struct xfrm_policy *pol, *ret;
1085         const xfrm_address_t *daddr, *saddr;
1086         struct hlist_head *chain;
1087         unsigned int sequence;
1088         u32 priority;
1089
1090         daddr = xfrm_flowi_daddr(fl, family);
1091         saddr = xfrm_flowi_saddr(fl, family);
1092         if (unlikely(!daddr || !saddr))
1093                 return NULL;
1094
1095         rcu_read_lock();
1096  retry:
1097         do {
1098                 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1099                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1100         } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1101
1102         priority = ~0U;
1103         ret = NULL;
1104         hlist_for_each_entry_rcu(pol, chain, bydst) {
1105                 err = xfrm_policy_match(pol, fl, type, family, dir);
1106                 if (err) {
1107                         if (err == -ESRCH)
1108                                 continue;
1109                         else {
1110                                 ret = ERR_PTR(err);
1111                                 goto fail;
1112                         }
1113                 } else {
1114                         ret = pol;
1115                         priority = ret->priority;
1116                         break;
1117                 }
1118         }
1119         chain = &net->xfrm.policy_inexact[dir];
1120         hlist_for_each_entry_rcu(pol, chain, bydst) {
1121                 if ((pol->priority >= priority) && ret)
1122                         break;
1123
1124                 err = xfrm_policy_match(pol, fl, type, family, dir);
1125                 if (err) {
1126                         if (err == -ESRCH)
1127                                 continue;
1128                         else {
1129                                 ret = ERR_PTR(err);
1130                                 goto fail;
1131                         }
1132                 } else {
1133                         ret = pol;
1134                         break;
1135                 }
1136         }
1137
1138         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1139                 goto retry;
1140
1141         if (ret && !xfrm_pol_hold_rcu(ret))
1142                 goto retry;
1143 fail:
1144         rcu_read_unlock();
1145
1146         return ret;
1147 }
1148
1149 static struct xfrm_policy *
1150 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1151 {
1152 #ifdef CONFIG_XFRM_SUB_POLICY
1153         struct xfrm_policy *pol;
1154
1155         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1156         if (pol != NULL)
1157                 return pol;
1158 #endif
1159         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1160 }
1161
1162 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1163                                                  const struct flowi *fl, u16 family)
1164 {
1165         struct xfrm_policy *pol;
1166
1167         rcu_read_lock();
1168  again:
1169         pol = rcu_dereference(sk->sk_policy[dir]);
1170         if (pol != NULL) {
1171                 bool match = xfrm_selector_match(&pol->selector, fl, family);
1172                 int err = 0;
1173
1174                 if (match) {
1175                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1176                                 pol = NULL;
1177                                 goto out;
1178                         }
1179                         err = security_xfrm_policy_lookup(pol->security,
1180                                                       fl->flowi_secid,
1181                                                       dir);
1182                         if (!err) {
1183                                 if (!xfrm_pol_hold_rcu(pol))
1184                                         goto again;
1185                         } else if (err == -ESRCH) {
1186                                 pol = NULL;
1187                         } else {
1188                                 pol = ERR_PTR(err);
1189                         }
1190                 } else
1191                         pol = NULL;
1192         }
1193 out:
1194         rcu_read_unlock();
1195         return pol;
1196 }
1197
1198 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1199 {
1200         struct net *net = xp_net(pol);
1201
1202         list_add(&pol->walk.all, &net->xfrm.policy_all);
1203         net->xfrm.policy_count[dir]++;
1204         xfrm_pol_hold(pol);
1205 }
1206
1207 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1208                                                 int dir)
1209 {
1210         struct net *net = xp_net(pol);
1211
1212         if (list_empty(&pol->walk.all))
1213                 return NULL;
1214
1215         /* Socket policies are not hashed. */
1216         if (!hlist_unhashed(&pol->bydst)) {
1217                 hlist_del_rcu(&pol->bydst);
1218                 hlist_del(&pol->byidx);
1219         }
1220
1221         list_del_init(&pol->walk.all);
1222         net->xfrm.policy_count[dir]--;
1223
1224         return pol;
1225 }
1226
1227 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1228 {
1229         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1230 }
1231
1232 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1233 {
1234         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1235 }
1236
1237 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1238 {
1239         struct net *net = xp_net(pol);
1240
1241         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1242         pol = __xfrm_policy_unlink(pol, dir);
1243         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1244         if (pol) {
1245                 xfrm_policy_kill(pol);
1246                 return 0;
1247         }
1248         return -ENOENT;
1249 }
1250 EXPORT_SYMBOL(xfrm_policy_delete);
1251
1252 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1253 {
1254         struct net *net = xp_net(pol);
1255         struct xfrm_policy *old_pol;
1256
1257 #ifdef CONFIG_XFRM_SUB_POLICY
1258         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1259                 return -EINVAL;
1260 #endif
1261
1262         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1263         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1264                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1265         if (pol) {
1266                 pol->curlft.add_time = get_seconds();
1267                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1268                 xfrm_sk_policy_link(pol, dir);
1269         }
1270         rcu_assign_pointer(sk->sk_policy[dir], pol);
1271         if (old_pol) {
1272                 if (pol)
1273                         xfrm_policy_requeue(old_pol, pol);
1274
1275                 /* Unlinking succeeds always. This is the only function
1276                  * allowed to delete or replace socket policy.
1277                  */
1278                 xfrm_sk_policy_unlink(old_pol, dir);
1279         }
1280         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1281
1282         if (old_pol) {
1283                 xfrm_policy_kill(old_pol);
1284         }
1285         return 0;
1286 }
1287
1288 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1289 {
1290         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1291         struct net *net = xp_net(old);
1292
1293         if (newp) {
1294                 newp->selector = old->selector;
1295                 if (security_xfrm_policy_clone(old->security,
1296                                                &newp->security)) {
1297                         kfree(newp);
1298                         return NULL;  /* ENOMEM */
1299                 }
1300                 newp->lft = old->lft;
1301                 newp->curlft = old->curlft;
1302                 newp->mark = old->mark;
1303                 newp->action = old->action;
1304                 newp->flags = old->flags;
1305                 newp->xfrm_nr = old->xfrm_nr;
1306                 newp->index = old->index;
1307                 newp->type = old->type;
1308                 newp->family = old->family;
1309                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1310                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1311                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1312                 xfrm_sk_policy_link(newp, dir);
1313                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1314                 xfrm_pol_put(newp);
1315         }
1316         return newp;
1317 }
1318
1319 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1320 {
1321         const struct xfrm_policy *p;
1322         struct xfrm_policy *np;
1323         int i, ret = 0;
1324
1325         rcu_read_lock();
1326         for (i = 0; i < 2; i++) {
1327                 p = rcu_dereference(osk->sk_policy[i]);
1328                 if (p) {
1329                         np = clone_policy(p, i);
1330                         if (unlikely(!np)) {
1331                                 ret = -ENOMEM;
1332                                 break;
1333                         }
1334                         rcu_assign_pointer(sk->sk_policy[i], np);
1335                 }
1336         }
1337         rcu_read_unlock();
1338         return ret;
1339 }
1340
1341 static int
1342 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1343                xfrm_address_t *remote, unsigned short family, u32 mark)
1344 {
1345         int err;
1346         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1347
1348         if (unlikely(afinfo == NULL))
1349                 return -EINVAL;
1350         err = afinfo->get_saddr(net, oif, local, remote, mark);
1351         rcu_read_unlock();
1352         return err;
1353 }
1354
1355 /* Resolve list of templates for the flow, given policy. */
1356
1357 static int
1358 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1359                       struct xfrm_state **xfrm, unsigned short family)
1360 {
1361         struct net *net = xp_net(policy);
1362         int nx;
1363         int i, error;
1364         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1365         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1366         xfrm_address_t tmp;
1367
1368         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1369                 struct xfrm_state *x;
1370                 xfrm_address_t *remote = daddr;
1371                 xfrm_address_t *local  = saddr;
1372                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1373
1374                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1375                     tmpl->mode == XFRM_MODE_BEET) {
1376                         remote = &tmpl->id.daddr;
1377                         local = &tmpl->saddr;
1378                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1379                                 error = xfrm_get_saddr(net, fl->flowi_oif,
1380                                                        &tmp, remote,
1381                                                        tmpl->encap_family, 0);
1382                                 if (error)
1383                                         goto fail;
1384                                 local = &tmp;
1385                         }
1386                 }
1387
1388                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1389
1390                 if (x && x->km.state == XFRM_STATE_VALID) {
1391                         xfrm[nx++] = x;
1392                         daddr = remote;
1393                         saddr = local;
1394                         continue;
1395                 }
1396                 if (x) {
1397                         error = (x->km.state == XFRM_STATE_ERROR ?
1398                                  -EINVAL : -EAGAIN);
1399                         xfrm_state_put(x);
1400                 } else if (error == -ESRCH) {
1401                         error = -EAGAIN;
1402                 }
1403
1404                 if (!tmpl->optional)
1405                         goto fail;
1406         }
1407         return nx;
1408
1409 fail:
1410         for (nx--; nx >= 0; nx--)
1411                 xfrm_state_put(xfrm[nx]);
1412         return error;
1413 }
1414
1415 static int
1416 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1417                   struct xfrm_state **xfrm, unsigned short family)
1418 {
1419         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1420         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1421         int cnx = 0;
1422         int error;
1423         int ret;
1424         int i;
1425
1426         for (i = 0; i < npols; i++) {
1427                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1428                         error = -ENOBUFS;
1429                         goto fail;
1430                 }
1431
1432                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1433                 if (ret < 0) {
1434                         error = ret;
1435                         goto fail;
1436                 } else
1437                         cnx += ret;
1438         }
1439
1440         /* found states are sorted for outbound processing */
1441         if (npols > 1)
1442                 xfrm_state_sort(xfrm, tpp, cnx, family);
1443
1444         return cnx;
1445
1446  fail:
1447         for (cnx--; cnx >= 0; cnx--)
1448                 xfrm_state_put(tpp[cnx]);
1449         return error;
1450
1451 }
1452
1453 static int xfrm_get_tos(const struct flowi *fl, int family)
1454 {
1455         const struct xfrm_policy_afinfo *afinfo;
1456         int tos = 0;
1457
1458         afinfo = xfrm_policy_get_afinfo(family);
1459         tos = afinfo ? afinfo->get_tos(fl) : 0;
1460
1461         rcu_read_unlock();
1462
1463         return tos;
1464 }
1465
1466 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1467 {
1468         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1469         struct dst_ops *dst_ops;
1470         struct xfrm_dst *xdst;
1471
1472         if (!afinfo)
1473                 return ERR_PTR(-EINVAL);
1474
1475         switch (family) {
1476         case AF_INET:
1477                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1478                 break;
1479 #if IS_ENABLED(CONFIG_IPV6)
1480         case AF_INET6:
1481                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1482                 break;
1483 #endif
1484         default:
1485                 BUG();
1486         }
1487         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1488
1489         if (likely(xdst)) {
1490                 struct dst_entry *dst = &xdst->u.dst;
1491
1492                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1493         } else
1494                 xdst = ERR_PTR(-ENOBUFS);
1495
1496         rcu_read_unlock();
1497
1498         return xdst;
1499 }
1500
1501 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1502                                  int nfheader_len)
1503 {
1504         const struct xfrm_policy_afinfo *afinfo =
1505                 xfrm_policy_get_afinfo(dst->ops->family);
1506         int err;
1507
1508         if (!afinfo)
1509                 return -EINVAL;
1510
1511         err = afinfo->init_path(path, dst, nfheader_len);
1512
1513         rcu_read_unlock();
1514
1515         return err;
1516 }
1517
1518 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1519                                 const struct flowi *fl)
1520 {
1521         const struct xfrm_policy_afinfo *afinfo =
1522                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1523         int err;
1524
1525         if (!afinfo)
1526                 return -EINVAL;
1527
1528         err = afinfo->fill_dst(xdst, dev, fl);
1529
1530         rcu_read_unlock();
1531
1532         return err;
1533 }
1534
1535
1536 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1537  * all the metrics... Shortly, bundle a bundle.
1538  */
1539
1540 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1541                                             struct xfrm_state **xfrm, int nx,
1542                                             const struct flowi *fl,
1543                                             struct dst_entry *dst)
1544 {
1545         struct net *net = xp_net(policy);
1546         unsigned long now = jiffies;
1547         struct net_device *dev;
1548         struct xfrm_mode *inner_mode;
1549         struct dst_entry *dst_prev = NULL;
1550         struct dst_entry *dst0 = NULL;
1551         int i = 0;
1552         int err;
1553         int header_len = 0;
1554         int nfheader_len = 0;
1555         int trailer_len = 0;
1556         int tos;
1557         int family = policy->selector.family;
1558         xfrm_address_t saddr, daddr;
1559
1560         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1561
1562         tos = xfrm_get_tos(fl, family);
1563
1564         dst_hold(dst);
1565
1566         for (; i < nx; i++) {
1567                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1568                 struct dst_entry *dst1 = &xdst->u.dst;
1569
1570                 err = PTR_ERR(xdst);
1571                 if (IS_ERR(xdst)) {
1572                         dst_release(dst);
1573                         goto put_states;
1574                 }
1575
1576                 if (!dst_prev)
1577                         dst0 = dst1;
1578                 else
1579                         /* Ref count is taken during xfrm_alloc_dst()
1580                          * No need to do dst_clone() on dst1
1581                          */
1582                         dst_prev->child = dst1;
1583
1584                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1585                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1586                                                         xfrm_af2proto(family));
1587                         if (!inner_mode) {
1588                                 err = -EAFNOSUPPORT;
1589                                 dst_release(dst);
1590                                 goto put_states;
1591                         }
1592                 } else
1593                         inner_mode = xfrm[i]->inner_mode;
1594
1595                 xdst->route = dst;
1596                 dst_copy_metrics(dst1, dst);
1597
1598                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1599                         family = xfrm[i]->props.family;
1600                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1601                                               &saddr, &daddr, family,
1602                                               xfrm[i]->props.output_mark);
1603                         err = PTR_ERR(dst);
1604                         if (IS_ERR(dst))
1605                                 goto put_states;
1606                 } else
1607                         dst_hold(dst);
1608
1609                 dst1->xfrm = xfrm[i];
1610                 xdst->xfrm_genid = xfrm[i]->genid;
1611
1612                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1613                 dst1->flags |= DST_HOST;
1614                 dst1->lastuse = now;
1615
1616                 dst1->input = dst_discard;
1617                 dst1->output = inner_mode->afinfo->output;
1618
1619                 dst1->next = dst_prev;
1620                 dst_prev = dst1;
1621
1622                 header_len += xfrm[i]->props.header_len;
1623                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1624                         nfheader_len += xfrm[i]->props.header_len;
1625                 trailer_len += xfrm[i]->props.trailer_len;
1626         }
1627
1628         dst_prev->child = dst;
1629         dst0->path = dst;
1630
1631         err = -ENODEV;
1632         dev = dst->dev;
1633         if (!dev)
1634                 goto free_dst;
1635
1636         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1637         xfrm_init_pmtu(dst_prev);
1638
1639         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1640                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1641
1642                 err = xfrm_fill_dst(xdst, dev, fl);
1643                 if (err)
1644                         goto free_dst;
1645
1646                 dst_prev->header_len = header_len;
1647                 dst_prev->trailer_len = trailer_len;
1648                 header_len -= xdst->u.dst.xfrm->props.header_len;
1649                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1650         }
1651
1652 out:
1653         return dst0;
1654
1655 put_states:
1656         for (; i < nx; i++)
1657                 xfrm_state_put(xfrm[i]);
1658 free_dst:
1659         if (dst0)
1660                 dst_release_immediate(dst0);
1661         dst0 = ERR_PTR(err);
1662         goto out;
1663 }
1664
1665 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1666                                 struct xfrm_policy **pols,
1667                                 int *num_pols, int *num_xfrms)
1668 {
1669         int i;
1670
1671         if (*num_pols == 0 || !pols[0]) {
1672                 *num_pols = 0;
1673                 *num_xfrms = 0;
1674                 return 0;
1675         }
1676         if (IS_ERR(pols[0]))
1677                 return PTR_ERR(pols[0]);
1678
1679         *num_xfrms = pols[0]->xfrm_nr;
1680
1681 #ifdef CONFIG_XFRM_SUB_POLICY
1682         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1683             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1684                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1685                                                     XFRM_POLICY_TYPE_MAIN,
1686                                                     fl, family,
1687                                                     XFRM_POLICY_OUT);
1688                 if (pols[1]) {
1689                         if (IS_ERR(pols[1])) {
1690                                 xfrm_pols_put(pols, *num_pols);
1691                                 return PTR_ERR(pols[1]);
1692                         }
1693                         (*num_pols)++;
1694                         (*num_xfrms) += pols[1]->xfrm_nr;
1695                 }
1696         }
1697 #endif
1698         for (i = 0; i < *num_pols; i++) {
1699                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1700                         *num_xfrms = -1;
1701                         break;
1702                 }
1703         }
1704
1705         return 0;
1706
1707 }
1708
1709 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1710 {
1711         this_cpu_write(xfrm_last_dst, xdst);
1712         if (old)
1713                 dst_release(&old->u.dst);
1714 }
1715
1716 static void __xfrm_pcpu_work_fn(void)
1717 {
1718         struct xfrm_dst *old;
1719
1720         old = this_cpu_read(xfrm_last_dst);
1721         if (old && !xfrm_bundle_ok(old))
1722                 xfrm_last_dst_update(NULL, old);
1723 }
1724
1725 static void xfrm_pcpu_work_fn(struct work_struct *work)
1726 {
1727         local_bh_disable();
1728         rcu_read_lock();
1729         __xfrm_pcpu_work_fn();
1730         rcu_read_unlock();
1731         local_bh_enable();
1732 }
1733
1734 void xfrm_policy_cache_flush(void)
1735 {
1736         struct xfrm_dst *old;
1737         bool found = 0;
1738         int cpu;
1739
1740         local_bh_disable();
1741         rcu_read_lock();
1742         for_each_possible_cpu(cpu) {
1743                 old = per_cpu(xfrm_last_dst, cpu);
1744                 if (old && !xfrm_bundle_ok(old)) {
1745                         if (smp_processor_id() == cpu) {
1746                                 __xfrm_pcpu_work_fn();
1747                                 continue;
1748                         }
1749                         found = true;
1750                         break;
1751                 }
1752         }
1753
1754         rcu_read_unlock();
1755         local_bh_enable();
1756
1757         if (!found)
1758                 return;
1759
1760         get_online_cpus();
1761
1762         for_each_possible_cpu(cpu) {
1763                 bool bundle_release;
1764
1765                 rcu_read_lock();
1766                 old = per_cpu(xfrm_last_dst, cpu);
1767                 bundle_release = old && !xfrm_bundle_ok(old);
1768                 rcu_read_unlock();
1769
1770                 if (!bundle_release)
1771                         continue;
1772
1773                 if (cpu_online(cpu)) {
1774                         schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1775                         continue;
1776                 }
1777
1778                 rcu_read_lock();
1779                 old = per_cpu(xfrm_last_dst, cpu);
1780                 if (old && !xfrm_bundle_ok(old)) {
1781                         per_cpu(xfrm_last_dst, cpu) = NULL;
1782                         dst_release(&old->u.dst);
1783                 }
1784                 rcu_read_unlock();
1785         }
1786
1787         put_online_cpus();
1788 }
1789
1790 static bool xfrm_xdst_can_reuse(struct xfrm_dst *xdst,
1791                                 struct xfrm_state * const xfrm[],
1792                                 int num)
1793 {
1794         const struct dst_entry *dst = &xdst->u.dst;
1795         int i;
1796
1797         if (xdst->num_xfrms != num)
1798                 return false;
1799
1800         for (i = 0; i < num; i++) {
1801                 if (!dst || dst->xfrm != xfrm[i])
1802                         return false;
1803                 dst = dst->child;
1804         }
1805
1806         return xfrm_bundle_ok(xdst);
1807 }
1808
1809 static struct xfrm_dst *
1810 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1811                                const struct flowi *fl, u16 family,
1812                                struct dst_entry *dst_orig)
1813 {
1814         struct net *net = xp_net(pols[0]);
1815         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1816         struct xfrm_dst *xdst, *old;
1817         struct dst_entry *dst;
1818         int err;
1819
1820         /* Try to instantiate a bundle */
1821         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1822         if (err <= 0) {
1823                 if (err != 0 && err != -EAGAIN)
1824                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1825                 return ERR_PTR(err);
1826         }
1827
1828         xdst = this_cpu_read(xfrm_last_dst);
1829         if (xdst &&
1830             xdst->u.dst.dev == dst_orig->dev &&
1831             xdst->num_pols == num_pols &&
1832             memcmp(xdst->pols, pols,
1833                    sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1834             xfrm_xdst_can_reuse(xdst, xfrm, err)) {
1835                 dst_hold(&xdst->u.dst);
1836                 while (err > 0)
1837                         xfrm_state_put(xfrm[--err]);
1838                 return xdst;
1839         }
1840
1841         old = xdst;
1842
1843         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1844         if (IS_ERR(dst)) {
1845                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1846                 return ERR_CAST(dst);
1847         }
1848
1849         xdst = (struct xfrm_dst *)dst;
1850         xdst->num_xfrms = err;
1851         xdst->num_pols = num_pols;
1852         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1853         xdst->policy_genid = atomic_read(&pols[0]->genid);
1854
1855         atomic_set(&xdst->u.dst.__refcnt, 2);
1856         xfrm_last_dst_update(xdst, old);
1857
1858         return xdst;
1859 }
1860
1861 static void xfrm_policy_queue_process(struct timer_list *t)
1862 {
1863         struct sk_buff *skb;
1864         struct sock *sk;
1865         struct dst_entry *dst;
1866         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1867         struct net *net = xp_net(pol);
1868         struct xfrm_policy_queue *pq = &pol->polq;
1869         struct flowi fl;
1870         struct sk_buff_head list;
1871
1872         spin_lock(&pq->hold_queue.lock);
1873         skb = skb_peek(&pq->hold_queue);
1874         if (!skb) {
1875                 spin_unlock(&pq->hold_queue.lock);
1876                 goto out;
1877         }
1878         dst = skb_dst(skb);
1879         sk = skb->sk;
1880         xfrm_decode_session(skb, &fl, dst->ops->family);
1881         spin_unlock(&pq->hold_queue.lock);
1882
1883         dst_hold(dst->path);
1884         dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1885         if (IS_ERR(dst))
1886                 goto purge_queue;
1887
1888         if (dst->flags & DST_XFRM_QUEUE) {
1889                 dst_release(dst);
1890
1891                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1892                         goto purge_queue;
1893
1894                 pq->timeout = pq->timeout << 1;
1895                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1896                         xfrm_pol_hold(pol);
1897         goto out;
1898         }
1899
1900         dst_release(dst);
1901
1902         __skb_queue_head_init(&list);
1903
1904         spin_lock(&pq->hold_queue.lock);
1905         pq->timeout = 0;
1906         skb_queue_splice_init(&pq->hold_queue, &list);
1907         spin_unlock(&pq->hold_queue.lock);
1908
1909         while (!skb_queue_empty(&list)) {
1910                 skb = __skb_dequeue(&list);
1911
1912                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1913                 dst_hold(skb_dst(skb)->path);
1914                 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1915                 if (IS_ERR(dst)) {
1916                         kfree_skb(skb);
1917                         continue;
1918                 }
1919
1920                 nf_reset(skb);
1921                 skb_dst_drop(skb);
1922                 skb_dst_set(skb, dst);
1923
1924                 dst_output(net, skb->sk, skb);
1925         }
1926
1927 out:
1928         xfrm_pol_put(pol);
1929         return;
1930
1931 purge_queue:
1932         pq->timeout = 0;
1933         skb_queue_purge(&pq->hold_queue);
1934         xfrm_pol_put(pol);
1935 }
1936
1937 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1938 {
1939         unsigned long sched_next;
1940         struct dst_entry *dst = skb_dst(skb);
1941         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1942         struct xfrm_policy *pol = xdst->pols[0];
1943         struct xfrm_policy_queue *pq = &pol->polq;
1944
1945         if (unlikely(skb_fclone_busy(sk, skb))) {
1946                 kfree_skb(skb);
1947                 return 0;
1948         }
1949
1950         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1951                 kfree_skb(skb);
1952                 return -EAGAIN;
1953         }
1954
1955         skb_dst_force(skb);
1956
1957         spin_lock_bh(&pq->hold_queue.lock);
1958
1959         if (!pq->timeout)
1960                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1961
1962         sched_next = jiffies + pq->timeout;
1963
1964         if (del_timer(&pq->hold_timer)) {
1965                 if (time_before(pq->hold_timer.expires, sched_next))
1966                         sched_next = pq->hold_timer.expires;
1967                 xfrm_pol_put(pol);
1968         }
1969
1970         __skb_queue_tail(&pq->hold_queue, skb);
1971         if (!mod_timer(&pq->hold_timer, sched_next))
1972                 xfrm_pol_hold(pol);
1973
1974         spin_unlock_bh(&pq->hold_queue.lock);
1975
1976         return 0;
1977 }
1978
1979 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1980                                                  struct xfrm_flo *xflo,
1981                                                  const struct flowi *fl,
1982                                                  int num_xfrms,
1983                                                  u16 family)
1984 {
1985         int err;
1986         struct net_device *dev;
1987         struct dst_entry *dst;
1988         struct dst_entry *dst1;
1989         struct xfrm_dst *xdst;
1990
1991         xdst = xfrm_alloc_dst(net, family);
1992         if (IS_ERR(xdst))
1993                 return xdst;
1994
1995         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
1996             net->xfrm.sysctl_larval_drop ||
1997             num_xfrms <= 0)
1998                 return xdst;
1999
2000         dst = xflo->dst_orig;
2001         dst1 = &xdst->u.dst;
2002         dst_hold(dst);
2003         xdst->route = dst;
2004
2005         dst_copy_metrics(dst1, dst);
2006
2007         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2008         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2009         dst1->lastuse = jiffies;
2010
2011         dst1->input = dst_discard;
2012         dst1->output = xdst_queue_output;
2013
2014         dst_hold(dst);
2015         dst1->child = dst;
2016         dst1->path = dst;
2017
2018         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2019
2020         err = -ENODEV;
2021         dev = dst->dev;
2022         if (!dev)
2023                 goto free_dst;
2024
2025         err = xfrm_fill_dst(xdst, dev, fl);
2026         if (err)
2027                 goto free_dst;
2028
2029 out:
2030         return xdst;
2031
2032 free_dst:
2033         dst_release(dst1);
2034         xdst = ERR_PTR(err);
2035         goto out;
2036 }
2037
2038 static struct xfrm_dst *
2039 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2040 {
2041         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2042         int num_pols = 0, num_xfrms = 0, err;
2043         struct xfrm_dst *xdst;
2044
2045         /* Resolve policies to use if we couldn't get them from
2046          * previous cache entry */
2047         num_pols = 1;
2048         pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2049         err = xfrm_expand_policies(fl, family, pols,
2050                                            &num_pols, &num_xfrms);
2051         if (err < 0)
2052                 goto inc_error;
2053         if (num_pols == 0)
2054                 return NULL;
2055         if (num_xfrms <= 0)
2056                 goto make_dummy_bundle;
2057
2058         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2059                                                   xflo->dst_orig);
2060         if (IS_ERR(xdst)) {
2061                 err = PTR_ERR(xdst);
2062                 if (err != -EAGAIN)
2063                         goto error;
2064                 goto make_dummy_bundle;
2065         } else if (xdst == NULL) {
2066                 num_xfrms = 0;
2067                 goto make_dummy_bundle;
2068         }
2069
2070         return xdst;
2071
2072 make_dummy_bundle:
2073         /* We found policies, but there's no bundles to instantiate:
2074          * either because the policy blocks, has no transformations or
2075          * we could not build template (no xfrm_states).*/
2076         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2077         if (IS_ERR(xdst)) {
2078                 xfrm_pols_put(pols, num_pols);
2079                 return ERR_CAST(xdst);
2080         }
2081         xdst->num_pols = num_pols;
2082         xdst->num_xfrms = num_xfrms;
2083         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2084
2085         return xdst;
2086
2087 inc_error:
2088         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2089 error:
2090         xfrm_pols_put(pols, num_pols);
2091         return ERR_PTR(err);
2092 }
2093
2094 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2095                                         struct dst_entry *dst_orig)
2096 {
2097         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2098         struct dst_entry *ret;
2099
2100         if (!afinfo) {
2101                 dst_release(dst_orig);
2102                 return ERR_PTR(-EINVAL);
2103         } else {
2104                 ret = afinfo->blackhole_route(net, dst_orig);
2105         }
2106         rcu_read_unlock();
2107
2108         return ret;
2109 }
2110
2111 /* Main function: finds/creates a bundle for given flow.
2112  *
2113  * At the moment we eat a raw IP route. Mostly to speed up lookups
2114  * on interfaces with disabled IPsec.
2115  */
2116 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2117                               const struct flowi *fl,
2118                               const struct sock *sk, int flags)
2119 {
2120         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2121         struct xfrm_dst *xdst;
2122         struct dst_entry *dst, *route;
2123         u16 family = dst_orig->ops->family;
2124         u8 dir = XFRM_POLICY_OUT;
2125         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2126
2127         dst = NULL;
2128         xdst = NULL;
2129         route = NULL;
2130
2131         sk = sk_const_to_full_sk(sk);
2132         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2133                 num_pols = 1;
2134                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2135                 err = xfrm_expand_policies(fl, family, pols,
2136                                            &num_pols, &num_xfrms);
2137                 if (err < 0)
2138                         goto dropdst;
2139
2140                 if (num_pols) {
2141                         if (num_xfrms <= 0) {
2142                                 drop_pols = num_pols;
2143                                 goto no_transform;
2144                         }
2145
2146                         xdst = xfrm_resolve_and_create_bundle(
2147                                         pols, num_pols, fl,
2148                                         family, dst_orig);
2149                         if (IS_ERR(xdst)) {
2150                                 xfrm_pols_put(pols, num_pols);
2151                                 err = PTR_ERR(xdst);
2152                                 goto dropdst;
2153                         } else if (xdst == NULL) {
2154                                 num_xfrms = 0;
2155                                 drop_pols = num_pols;
2156                                 goto no_transform;
2157                         }
2158
2159                         route = xdst->route;
2160                 }
2161         }
2162
2163         if (xdst == NULL) {
2164                 struct xfrm_flo xflo;
2165
2166                 xflo.dst_orig = dst_orig;
2167                 xflo.flags = flags;
2168
2169                 /* To accelerate a bit...  */
2170                 if ((dst_orig->flags & DST_NOXFRM) ||
2171                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2172                         goto nopol;
2173
2174                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2175                 if (xdst == NULL)
2176                         goto nopol;
2177                 if (IS_ERR(xdst)) {
2178                         err = PTR_ERR(xdst);
2179                         goto dropdst;
2180                 }
2181
2182                 num_pols = xdst->num_pols;
2183                 num_xfrms = xdst->num_xfrms;
2184                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2185                 route = xdst->route;
2186         }
2187
2188         dst = &xdst->u.dst;
2189         if (route == NULL && num_xfrms > 0) {
2190                 /* The only case when xfrm_bundle_lookup() returns a
2191                  * bundle with null route, is when the template could
2192                  * not be resolved. It means policies are there, but
2193                  * bundle could not be created, since we don't yet
2194                  * have the xfrm_state's. We need to wait for KM to
2195                  * negotiate new SA's or bail out with error.*/
2196                 if (net->xfrm.sysctl_larval_drop) {
2197                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2198                         err = -EREMOTE;
2199                         goto error;
2200                 }
2201
2202                 err = -EAGAIN;
2203
2204                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2205                 goto error;
2206         }
2207
2208 no_transform:
2209         if (num_pols == 0)
2210                 goto nopol;
2211
2212         if ((flags & XFRM_LOOKUP_ICMP) &&
2213             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2214                 err = -ENOENT;
2215                 goto error;
2216         }
2217
2218         for (i = 0; i < num_pols; i++)
2219                 pols[i]->curlft.use_time = get_seconds();
2220
2221         if (num_xfrms < 0) {
2222                 /* Prohibit the flow */
2223                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2224                 err = -EPERM;
2225                 goto error;
2226         } else if (num_xfrms > 0) {
2227                 /* Flow transformed */
2228                 dst_release(dst_orig);
2229         } else {
2230                 /* Flow passes untransformed */
2231                 dst_release(dst);
2232                 dst = dst_orig;
2233         }
2234 ok:
2235         xfrm_pols_put(pols, drop_pols);
2236         if (dst && dst->xfrm &&
2237             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2238                 dst->flags |= DST_XFRM_TUNNEL;
2239         return dst;
2240
2241 nopol:
2242         if (!(flags & XFRM_LOOKUP_ICMP)) {
2243                 dst = dst_orig;
2244                 goto ok;
2245         }
2246         err = -ENOENT;
2247 error:
2248         dst_release(dst);
2249 dropdst:
2250         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2251                 dst_release(dst_orig);
2252         xfrm_pols_put(pols, drop_pols);
2253         return ERR_PTR(err);
2254 }
2255 EXPORT_SYMBOL(xfrm_lookup);
2256
2257 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2258  * Otherwise we may send out blackholed packets.
2259  */
2260 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2261                                     const struct flowi *fl,
2262                                     const struct sock *sk, int flags)
2263 {
2264         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2265                                             flags | XFRM_LOOKUP_QUEUE |
2266                                             XFRM_LOOKUP_KEEP_DST_REF);
2267
2268         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2269                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2270
2271         return dst;
2272 }
2273 EXPORT_SYMBOL(xfrm_lookup_route);
2274
2275 static inline int
2276 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2277 {
2278         struct xfrm_state *x;
2279
2280         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2281                 return 0;
2282         x = skb->sp->xvec[idx];
2283         if (!x->type->reject)
2284                 return 0;
2285         return x->type->reject(x, skb, fl);
2286 }
2287
2288 /* When skb is transformed back to its "native" form, we have to
2289  * check policy restrictions. At the moment we make this in maximally
2290  * stupid way. Shame on me. :-) Of course, connected sockets must
2291  * have policy cached at them.
2292  */
2293
2294 static inline int
2295 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2296               unsigned short family)
2297 {
2298         if (xfrm_state_kern(x))
2299                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2300         return  x->id.proto == tmpl->id.proto &&
2301                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2302                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2303                 x->props.mode == tmpl->mode &&
2304                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2305                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2306                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2307                   xfrm_state_addr_cmp(tmpl, x, family));
2308 }
2309
2310 /*
2311  * 0 or more than 0 is returned when validation is succeeded (either bypass
2312  * because of optional transport mode, or next index of the mathced secpath
2313  * state with the template.
2314  * -1 is returned when no matching template is found.
2315  * Otherwise "-2 - errored_index" is returned.
2316  */
2317 static inline int
2318 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2319                unsigned short family)
2320 {
2321         int idx = start;
2322
2323         if (tmpl->optional) {
2324                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2325                         return start;
2326         } else
2327                 start = -1;
2328         for (; idx < sp->len; idx++) {
2329                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2330                         return ++idx;
2331                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2332                         if (start == -1)
2333                                 start = -2-idx;
2334                         break;
2335                 }
2336         }
2337         return start;
2338 }
2339
2340 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2341                           unsigned int family, int reverse)
2342 {
2343         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2344         int err;
2345
2346         if (unlikely(afinfo == NULL))
2347                 return -EAFNOSUPPORT;
2348
2349         afinfo->decode_session(skb, fl, reverse);
2350         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2351         rcu_read_unlock();
2352         return err;
2353 }
2354 EXPORT_SYMBOL(__xfrm_decode_session);
2355
2356 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2357 {
2358         for (; k < sp->len; k++) {
2359                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2360                         *idxp = k;
2361                         return 1;
2362                 }
2363         }
2364
2365         return 0;
2366 }
2367
2368 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2369                         unsigned short family)
2370 {
2371         struct net *net = dev_net(skb->dev);
2372         struct xfrm_policy *pol;
2373         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2374         int npols = 0;
2375         int xfrm_nr;
2376         int pi;
2377         int reverse;
2378         struct flowi fl;
2379         int xerr_idx = -1;
2380
2381         reverse = dir & ~XFRM_POLICY_MASK;
2382         dir &= XFRM_POLICY_MASK;
2383
2384         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2385                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2386                 return 0;
2387         }
2388
2389         nf_nat_decode_session(skb, &fl, family);
2390
2391         /* First, check used SA against their selectors. */
2392         if (skb->sp) {
2393                 int i;
2394
2395                 for (i = skb->sp->len-1; i >= 0; i--) {
2396                         struct xfrm_state *x = skb->sp->xvec[i];
2397                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2398                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2399                                 return 0;
2400                         }
2401                 }
2402         }
2403
2404         pol = NULL;
2405         sk = sk_to_full_sk(sk);
2406         if (sk && sk->sk_policy[dir]) {
2407                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2408                 if (IS_ERR(pol)) {
2409                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2410                         return 0;
2411                 }
2412         }
2413
2414         if (!pol)
2415                 pol = xfrm_policy_lookup(net, &fl, family, dir);
2416
2417         if (IS_ERR(pol)) {
2418                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2419                 return 0;
2420         }
2421
2422         if (!pol) {
2423                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2424                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2425                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2426                         return 0;
2427                 }
2428                 return 1;
2429         }
2430
2431         pol->curlft.use_time = get_seconds();
2432
2433         pols[0] = pol;
2434         npols++;
2435 #ifdef CONFIG_XFRM_SUB_POLICY
2436         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2437                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2438                                                     &fl, family,
2439                                                     XFRM_POLICY_IN);
2440                 if (pols[1]) {
2441                         if (IS_ERR(pols[1])) {
2442                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2443                                 return 0;
2444                         }
2445                         pols[1]->curlft.use_time = get_seconds();
2446                         npols++;
2447                 }
2448         }
2449 #endif
2450
2451         if (pol->action == XFRM_POLICY_ALLOW) {
2452                 struct sec_path *sp;
2453                 static struct sec_path dummy;
2454                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2455                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2456                 struct xfrm_tmpl **tpp = tp;
2457                 int ti = 0;
2458                 int i, k;
2459
2460                 if ((sp = skb->sp) == NULL)
2461                         sp = &dummy;
2462
2463                 for (pi = 0; pi < npols; pi++) {
2464                         if (pols[pi] != pol &&
2465                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2466                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2467                                 goto reject;
2468                         }
2469                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2470                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2471                                 goto reject_error;
2472                         }
2473                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2474                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2475                 }
2476                 xfrm_nr = ti;
2477                 if (npols > 1) {
2478                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2479                         tpp = stp;
2480                 }
2481
2482                 /* For each tunnel xfrm, find the first matching tmpl.
2483                  * For each tmpl before that, find corresponding xfrm.
2484                  * Order is _important_. Later we will implement
2485                  * some barriers, but at the moment barriers
2486                  * are implied between each two transformations.
2487                  */
2488                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2489                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2490                         if (k < 0) {
2491                                 if (k < -1)
2492                                         /* "-2 - errored_index" returned */
2493                                         xerr_idx = -(2+k);
2494                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2495                                 goto reject;
2496                         }
2497                 }
2498
2499                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2500                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2501                         goto reject;
2502                 }
2503
2504                 xfrm_pols_put(pols, npols);
2505                 return 1;
2506         }
2507         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2508
2509 reject:
2510         xfrm_secpath_reject(xerr_idx, skb, &fl);
2511 reject_error:
2512         xfrm_pols_put(pols, npols);
2513         return 0;
2514 }
2515 EXPORT_SYMBOL(__xfrm_policy_check);
2516
2517 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2518 {
2519         struct net *net = dev_net(skb->dev);
2520         struct flowi fl;
2521         struct dst_entry *dst;
2522         int res = 1;
2523
2524         if (xfrm_decode_session(skb, &fl, family) < 0) {
2525                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2526                 return 0;
2527         }
2528
2529         skb_dst_force(skb);
2530
2531         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2532         if (IS_ERR(dst)) {
2533                 res = 0;
2534                 dst = NULL;
2535         }
2536         skb_dst_set(skb, dst);
2537         return res;
2538 }
2539 EXPORT_SYMBOL(__xfrm_route_forward);
2540
2541 /* Optimize later using cookies and generation ids. */
2542
2543 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2544 {
2545         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2546          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2547          * get validated by dst_ops->check on every use.  We do this
2548          * because when a normal route referenced by an XFRM dst is
2549          * obsoleted we do not go looking around for all parent
2550          * referencing XFRM dsts so that we can invalidate them.  It
2551          * is just too much work.  Instead we make the checks here on
2552          * every use.  For example:
2553          *
2554          *      XFRM dst A --> IPv4 dst X
2555          *
2556          * X is the "xdst->route" of A (X is also the "dst->path" of A
2557          * in this example).  If X is marked obsolete, "A" will not
2558          * notice.  That's what we are validating here via the
2559          * stale_bundle() check.
2560          *
2561          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2562          * be marked on it.
2563          * This will force stale_bundle() to fail on any xdst bundle with
2564          * this dst linked in it.
2565          */
2566         if (dst->obsolete < 0 && !stale_bundle(dst))
2567                 return dst;
2568
2569         return NULL;
2570 }
2571
2572 static int stale_bundle(struct dst_entry *dst)
2573 {
2574         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2575 }
2576
2577 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2578 {
2579         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2580                 dst->dev = dev_net(dev)->loopback_dev;
2581                 dev_hold(dst->dev);
2582                 dev_put(dev);
2583         }
2584 }
2585 EXPORT_SYMBOL(xfrm_dst_ifdown);
2586
2587 static void xfrm_link_failure(struct sk_buff *skb)
2588 {
2589         /* Impossible. Such dst must be popped before reaches point of failure. */
2590 }
2591
2592 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2593 {
2594         if (dst) {
2595                 if (dst->obsolete) {
2596                         dst_release(dst);
2597                         dst = NULL;
2598                 }
2599         }
2600         return dst;
2601 }
2602
2603 static void xfrm_init_pmtu(struct dst_entry *dst)
2604 {
2605         do {
2606                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2607                 u32 pmtu, route_mtu_cached;
2608
2609                 pmtu = dst_mtu(dst->child);
2610                 xdst->child_mtu_cached = pmtu;
2611
2612                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2613
2614                 route_mtu_cached = dst_mtu(xdst->route);
2615                 xdst->route_mtu_cached = route_mtu_cached;
2616
2617                 if (pmtu > route_mtu_cached)
2618                         pmtu = route_mtu_cached;
2619
2620                 dst_metric_set(dst, RTAX_MTU, pmtu);
2621         } while ((dst = dst->next));
2622 }
2623
2624 /* Check that the bundle accepts the flow and its components are
2625  * still valid.
2626  */
2627
2628 static int xfrm_bundle_ok(struct xfrm_dst *first)
2629 {
2630         struct dst_entry *dst = &first->u.dst;
2631         struct xfrm_dst *last;
2632         u32 mtu;
2633
2634         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2635             (dst->dev && !netif_running(dst->dev)))
2636                 return 0;
2637
2638         if (dst->flags & DST_XFRM_QUEUE)
2639                 return 1;
2640
2641         last = NULL;
2642
2643         do {
2644                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2645
2646                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2647                         return 0;
2648                 if (xdst->xfrm_genid != dst->xfrm->genid)
2649                         return 0;
2650                 if (xdst->num_pols > 0 &&
2651                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2652                         return 0;
2653
2654                 mtu = dst_mtu(dst->child);
2655                 if (xdst->child_mtu_cached != mtu) {
2656                         last = xdst;
2657                         xdst->child_mtu_cached = mtu;
2658                 }
2659
2660                 if (!dst_check(xdst->route, xdst->route_cookie))
2661                         return 0;
2662                 mtu = dst_mtu(xdst->route);
2663                 if (xdst->route_mtu_cached != mtu) {
2664                         last = xdst;
2665                         xdst->route_mtu_cached = mtu;
2666                 }
2667
2668                 dst = dst->child;
2669         } while (dst->xfrm);
2670
2671         if (likely(!last))
2672                 return 1;
2673
2674         mtu = last->child_mtu_cached;
2675         for (;;) {
2676                 dst = &last->u.dst;
2677
2678                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2679                 if (mtu > last->route_mtu_cached)
2680                         mtu = last->route_mtu_cached;
2681                 dst_metric_set(dst, RTAX_MTU, mtu);
2682
2683                 if (last == first)
2684                         break;
2685
2686                 last = (struct xfrm_dst *)last->u.dst.next;
2687                 last->child_mtu_cached = mtu;
2688         }
2689
2690         return 1;
2691 }
2692
2693 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2694 {
2695         return dst_metric_advmss(dst->path);
2696 }
2697
2698 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2699 {
2700         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2701
2702         return mtu ? : dst_mtu(dst->path);
2703 }
2704
2705 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2706                                         const void *daddr)
2707 {
2708         const struct dst_entry *path = dst->path;
2709
2710         for (; dst != path; dst = dst->child) {
2711                 const struct xfrm_state *xfrm = dst->xfrm;
2712
2713                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2714                         continue;
2715                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2716                         daddr = xfrm->coaddr;
2717                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2718                         daddr = &xfrm->id.daddr;
2719         }
2720         return daddr;
2721 }
2722
2723 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2724                                            struct sk_buff *skb,
2725                                            const void *daddr)
2726 {
2727         const struct dst_entry *path = dst->path;
2728
2729         if (!skb)
2730                 daddr = xfrm_get_dst_nexthop(dst, daddr);
2731         return path->ops->neigh_lookup(path, skb, daddr);
2732 }
2733
2734 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2735 {
2736         const struct dst_entry *path = dst->path;
2737
2738         daddr = xfrm_get_dst_nexthop(dst, daddr);
2739         path->ops->confirm_neigh(path, daddr);
2740 }
2741
2742 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2743 {
2744         int err = 0;
2745
2746         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2747                 return -EAFNOSUPPORT;
2748
2749         spin_lock(&xfrm_policy_afinfo_lock);
2750         if (unlikely(xfrm_policy_afinfo[family] != NULL))
2751                 err = -EEXIST;
2752         else {
2753                 struct dst_ops *dst_ops = afinfo->dst_ops;
2754                 if (likely(dst_ops->kmem_cachep == NULL))
2755                         dst_ops->kmem_cachep = xfrm_dst_cache;
2756                 if (likely(dst_ops->check == NULL))
2757                         dst_ops->check = xfrm_dst_check;
2758                 if (likely(dst_ops->default_advmss == NULL))
2759                         dst_ops->default_advmss = xfrm_default_advmss;
2760                 if (likely(dst_ops->mtu == NULL))
2761                         dst_ops->mtu = xfrm_mtu;
2762                 if (likely(dst_ops->negative_advice == NULL))
2763                         dst_ops->negative_advice = xfrm_negative_advice;
2764                 if (likely(dst_ops->link_failure == NULL))
2765                         dst_ops->link_failure = xfrm_link_failure;
2766                 if (likely(dst_ops->neigh_lookup == NULL))
2767                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2768                 if (likely(!dst_ops->confirm_neigh))
2769                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
2770                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2771         }
2772         spin_unlock(&xfrm_policy_afinfo_lock);
2773
2774         return err;
2775 }
2776 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2777
2778 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2779 {
2780         struct dst_ops *dst_ops = afinfo->dst_ops;
2781         int i;
2782
2783         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2784                 if (xfrm_policy_afinfo[i] != afinfo)
2785                         continue;
2786                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2787                 break;
2788         }
2789
2790         synchronize_rcu();
2791
2792         dst_ops->kmem_cachep = NULL;
2793         dst_ops->check = NULL;
2794         dst_ops->negative_advice = NULL;
2795         dst_ops->link_failure = NULL;
2796 }
2797 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2798
2799 #ifdef CONFIG_XFRM_STATISTICS
2800 static int __net_init xfrm_statistics_init(struct net *net)
2801 {
2802         int rv;
2803         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2804         if (!net->mib.xfrm_statistics)
2805                 return -ENOMEM;
2806         rv = xfrm_proc_init(net);
2807         if (rv < 0)
2808                 free_percpu(net->mib.xfrm_statistics);
2809         return rv;
2810 }
2811
2812 static void xfrm_statistics_fini(struct net *net)
2813 {
2814         xfrm_proc_fini(net);
2815         free_percpu(net->mib.xfrm_statistics);
2816 }
2817 #else
2818 static int __net_init xfrm_statistics_init(struct net *net)
2819 {
2820         return 0;
2821 }
2822
2823 static void xfrm_statistics_fini(struct net *net)
2824 {
2825 }
2826 #endif
2827
2828 static int __net_init xfrm_policy_init(struct net *net)
2829 {
2830         unsigned int hmask, sz;
2831         int dir;
2832
2833         if (net_eq(net, &init_net))
2834                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2835                                            sizeof(struct xfrm_dst),
2836                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2837                                            NULL);
2838
2839         hmask = 8 - 1;
2840         sz = (hmask+1) * sizeof(struct hlist_head);
2841
2842         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2843         if (!net->xfrm.policy_byidx)
2844                 goto out_byidx;
2845         net->xfrm.policy_idx_hmask = hmask;
2846
2847         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2848                 struct xfrm_policy_hash *htab;
2849
2850                 net->xfrm.policy_count[dir] = 0;
2851                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2852                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2853
2854                 htab = &net->xfrm.policy_bydst[dir];
2855                 htab->table = xfrm_hash_alloc(sz);
2856                 if (!htab->table)
2857                         goto out_bydst;
2858                 htab->hmask = hmask;
2859                 htab->dbits4 = 32;
2860                 htab->sbits4 = 32;
2861                 htab->dbits6 = 128;
2862                 htab->sbits6 = 128;
2863         }
2864         net->xfrm.policy_hthresh.lbits4 = 32;
2865         net->xfrm.policy_hthresh.rbits4 = 32;
2866         net->xfrm.policy_hthresh.lbits6 = 128;
2867         net->xfrm.policy_hthresh.rbits6 = 128;
2868
2869         seqlock_init(&net->xfrm.policy_hthresh.lock);
2870
2871         INIT_LIST_HEAD(&net->xfrm.policy_all);
2872         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2873         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2874         if (net_eq(net, &init_net))
2875                 xfrm_dev_init();
2876         return 0;
2877
2878 out_bydst:
2879         for (dir--; dir >= 0; dir--) {
2880                 struct xfrm_policy_hash *htab;
2881
2882                 htab = &net->xfrm.policy_bydst[dir];
2883                 xfrm_hash_free(htab->table, sz);
2884         }
2885         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2886 out_byidx:
2887         return -ENOMEM;
2888 }
2889
2890 static void xfrm_policy_fini(struct net *net)
2891 {
2892         unsigned int sz;
2893         int dir;
2894
2895         flush_work(&net->xfrm.policy_hash_work);
2896 #ifdef CONFIG_XFRM_SUB_POLICY
2897         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2898 #endif
2899         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2900
2901         WARN_ON(!list_empty(&net->xfrm.policy_all));
2902
2903         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2904                 struct xfrm_policy_hash *htab;
2905
2906                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2907
2908                 htab = &net->xfrm.policy_bydst[dir];
2909                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2910                 WARN_ON(!hlist_empty(htab->table));
2911                 xfrm_hash_free(htab->table, sz);
2912         }
2913
2914         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2915         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2916         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2917 }
2918
2919 static int __net_init xfrm_net_init(struct net *net)
2920 {
2921         int rv;
2922
2923         /* Initialize the per-net locks here */
2924         spin_lock_init(&net->xfrm.xfrm_state_lock);
2925         spin_lock_init(&net->xfrm.xfrm_policy_lock);
2926         mutex_init(&net->xfrm.xfrm_cfg_mutex);
2927
2928         rv = xfrm_statistics_init(net);
2929         if (rv < 0)
2930                 goto out_statistics;
2931         rv = xfrm_state_init(net);
2932         if (rv < 0)
2933                 goto out_state;
2934         rv = xfrm_policy_init(net);
2935         if (rv < 0)
2936                 goto out_policy;
2937         rv = xfrm_sysctl_init(net);
2938         if (rv < 0)
2939                 goto out_sysctl;
2940
2941         return 0;
2942
2943 out_sysctl:
2944         xfrm_policy_fini(net);
2945 out_policy:
2946         xfrm_state_fini(net);
2947 out_state:
2948         xfrm_statistics_fini(net);
2949 out_statistics:
2950         return rv;
2951 }
2952
2953 static void __net_exit xfrm_net_exit(struct net *net)
2954 {
2955         xfrm_sysctl_fini(net);
2956         xfrm_policy_fini(net);
2957         xfrm_state_fini(net);
2958         xfrm_statistics_fini(net);
2959 }
2960
2961 static struct pernet_operations __net_initdata xfrm_net_ops = {
2962         .init = xfrm_net_init,
2963         .exit = xfrm_net_exit,
2964 };
2965
2966 void __init xfrm_init(void)
2967 {
2968         int i;
2969
2970         xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2971                                        GFP_KERNEL);
2972         BUG_ON(!xfrm_pcpu_work);
2973
2974         for (i = 0; i < NR_CPUS; i++)
2975                 INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
2976
2977         register_pernet_subsys(&xfrm_net_ops);
2978         seqcount_init(&xfrm_policy_hash_generation);
2979         xfrm_input_init();
2980 }
2981
2982 #ifdef CONFIG_AUDITSYSCALL
2983 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2984                                          struct audit_buffer *audit_buf)
2985 {
2986         struct xfrm_sec_ctx *ctx = xp->security;
2987         struct xfrm_selector *sel = &xp->selector;
2988
2989         if (ctx)
2990                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2991                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2992
2993         switch (sel->family) {
2994         case AF_INET:
2995                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2996                 if (sel->prefixlen_s != 32)
2997                         audit_log_format(audit_buf, " src_prefixlen=%d",
2998                                          sel->prefixlen_s);
2999                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3000                 if (sel->prefixlen_d != 32)
3001                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3002                                          sel->prefixlen_d);
3003                 break;
3004         case AF_INET6:
3005                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3006                 if (sel->prefixlen_s != 128)
3007                         audit_log_format(audit_buf, " src_prefixlen=%d",
3008                                          sel->prefixlen_s);
3009                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3010                 if (sel->prefixlen_d != 128)
3011                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3012                                          sel->prefixlen_d);
3013                 break;
3014         }
3015 }
3016
3017 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3018 {
3019         struct audit_buffer *audit_buf;
3020
3021         audit_buf = xfrm_audit_start("SPD-add");
3022         if (audit_buf == NULL)
3023                 return;
3024         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3025         audit_log_format(audit_buf, " res=%u", result);
3026         xfrm_audit_common_policyinfo(xp, audit_buf);
3027         audit_log_end(audit_buf);
3028 }
3029 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3030
3031 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3032                               bool task_valid)
3033 {
3034         struct audit_buffer *audit_buf;
3035
3036         audit_buf = xfrm_audit_start("SPD-delete");
3037         if (audit_buf == NULL)
3038                 return;
3039         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3040         audit_log_format(audit_buf, " res=%u", result);
3041         xfrm_audit_common_policyinfo(xp, audit_buf);
3042         audit_log_end(audit_buf);
3043 }
3044 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3045 #endif
3046
3047 #ifdef CONFIG_XFRM_MIGRATE
3048 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3049                                         const struct xfrm_selector *sel_tgt)
3050 {
3051         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3052                 if (sel_tgt->family == sel_cmp->family &&
3053                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3054                                     sel_cmp->family) &&
3055                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3056                                     sel_cmp->family) &&
3057                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3058                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3059                         return true;
3060                 }
3061         } else {
3062                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3063                         return true;
3064                 }
3065         }
3066         return false;
3067 }
3068
3069 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3070                                                     u8 dir, u8 type, struct net *net)
3071 {
3072         struct xfrm_policy *pol, *ret = NULL;
3073         struct hlist_head *chain;
3074         u32 priority = ~0U;
3075
3076         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3077         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3078         hlist_for_each_entry(pol, chain, bydst) {
3079                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3080                     pol->type == type) {
3081                         ret = pol;
3082                         priority = ret->priority;
3083                         break;
3084                 }
3085         }
3086         chain = &net->xfrm.policy_inexact[dir];
3087         hlist_for_each_entry(pol, chain, bydst) {
3088                 if ((pol->priority >= priority) && ret)
3089                         break;
3090
3091                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3092                     pol->type == type) {
3093                         ret = pol;
3094                         break;
3095                 }
3096         }
3097
3098         xfrm_pol_hold(ret);
3099
3100         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3101
3102         return ret;
3103 }
3104
3105 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3106 {
3107         int match = 0;
3108
3109         if (t->mode == m->mode && t->id.proto == m->proto &&
3110             (m->reqid == 0 || t->reqid == m->reqid)) {
3111                 switch (t->mode) {
3112                 case XFRM_MODE_TUNNEL:
3113                 case XFRM_MODE_BEET:
3114                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3115                                             m->old_family) &&
3116                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3117                                             m->old_family)) {
3118                                 match = 1;
3119                         }
3120                         break;
3121                 case XFRM_MODE_TRANSPORT:
3122                         /* in case of transport mode, template does not store
3123                            any IP addresses, hence we just compare mode and
3124                            protocol */
3125                         match = 1;
3126                         break;
3127                 default:
3128                         break;
3129                 }
3130         }
3131         return match;
3132 }
3133
3134 /* update endpoint address(es) of template(s) */
3135 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3136                                struct xfrm_migrate *m, int num_migrate)
3137 {
3138         struct xfrm_migrate *mp;
3139         int i, j, n = 0;
3140
3141         write_lock_bh(&pol->lock);
3142         if (unlikely(pol->walk.dead)) {
3143                 /* target policy has been deleted */
3144                 write_unlock_bh(&pol->lock);
3145                 return -ENOENT;
3146         }
3147
3148         for (i = 0; i < pol->xfrm_nr; i++) {
3149                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3150                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3151                                 continue;
3152                         n++;
3153                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3154                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3155                                 continue;
3156                         /* update endpoints */
3157                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3158                                sizeof(pol->xfrm_vec[i].id.daddr));
3159                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3160                                sizeof(pol->xfrm_vec[i].saddr));
3161                         pol->xfrm_vec[i].encap_family = mp->new_family;
3162                         /* flush bundles */
3163                         atomic_inc(&pol->genid);
3164                 }
3165         }
3166
3167         write_unlock_bh(&pol->lock);
3168
3169         if (!n)
3170                 return -ENODATA;
3171
3172         return 0;
3173 }
3174
3175 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3176 {
3177         int i, j;
3178
3179         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3180                 return -EINVAL;
3181
3182         for (i = 0; i < num_migrate; i++) {
3183                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3184                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3185                         return -EINVAL;
3186
3187                 /* check if there is any duplicated entry */
3188                 for (j = i + 1; j < num_migrate; j++) {
3189                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3190                                     sizeof(m[i].old_daddr)) &&
3191                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3192                                     sizeof(m[i].old_saddr)) &&
3193                             m[i].proto == m[j].proto &&
3194                             m[i].mode == m[j].mode &&
3195                             m[i].reqid == m[j].reqid &&
3196                             m[i].old_family == m[j].old_family)
3197                                 return -EINVAL;
3198                 }
3199         }
3200
3201         return 0;
3202 }
3203
3204 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3205                  struct xfrm_migrate *m, int num_migrate,
3206                  struct xfrm_kmaddress *k, struct net *net,
3207                  struct xfrm_encap_tmpl *encap)
3208 {
3209         int i, err, nx_cur = 0, nx_new = 0;
3210         struct xfrm_policy *pol = NULL;
3211         struct xfrm_state *x, *xc;
3212         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3213         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3214         struct xfrm_migrate *mp;
3215
3216         /* Stage 0 - sanity checks */
3217         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3218                 goto out;
3219
3220         if (dir >= XFRM_POLICY_MAX) {
3221                 err = -EINVAL;
3222                 goto out;
3223         }
3224
3225         /* Stage 1 - find policy */
3226         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3227                 err = -ENOENT;
3228                 goto out;
3229         }
3230
3231         /* Stage 2 - find and update state(s) */
3232         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3233                 if ((x = xfrm_migrate_state_find(mp, net))) {
3234                         x_cur[nx_cur] = x;
3235                         nx_cur++;
3236                         xc = xfrm_state_migrate(x, mp, encap);
3237                         if (xc) {
3238                                 x_new[nx_new] = xc;
3239                                 nx_new++;
3240                         } else {
3241                                 err = -ENODATA;
3242                                 goto restore_state;
3243                         }
3244                 }
3245         }
3246
3247         /* Stage 3 - update policy */
3248         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3249                 goto restore_state;
3250
3251         /* Stage 4 - delete old state(s) */
3252         if (nx_cur) {
3253                 xfrm_states_put(x_cur, nx_cur);
3254                 xfrm_states_delete(x_cur, nx_cur);
3255         }
3256
3257         /* Stage 5 - announce */
3258         km_migrate(sel, dir, type, m, num_migrate, k, encap);
3259
3260         xfrm_pol_put(pol);
3261
3262         return 0;
3263 out:
3264         return err;
3265
3266 restore_state:
3267         if (pol)
3268                 xfrm_pol_put(pol);
3269         if (nx_cur)
3270                 xfrm_states_put(x_cur, nx_cur);
3271         if (nx_new)
3272                 xfrm_states_delete(x_new, nx_new);
3273
3274         return err;
3275 }
3276 EXPORT_SYMBOL(xfrm_migrate);
3277 #endif