net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/cpu.h>
  28 #include <linux/audit.h>
  29 #include <net/dst.h>
  30 #include <net/flow.h>
  31 #include <net/xfrm.h>
  32 #include <net/ip.h>
  33 #ifdef CONFIG_XFRM_STATISTICS
  34 #include <net/snmp.h>
  35 #endif
  36
  37 #include "xfrm_hash.h"
  38
  39 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  40 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  41 #define XFRM_MAX_QUEUE_LEN      100
  42
  43 struct xfrm_flo {
  44         struct dst_entry *dst_orig;
  45         u8 flags;
  46 };
  47
  48 static DEFINE_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 (policy->walk.dead ||
 613                     xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
 614                         /* skip socket policies */
 615                         continue;
 616                 }
 617                 newpos = NULL;
 618                 chain = policy_hash_bysel(net, &policy->selector,
 619                                           policy->family,
 620                                           xfrm_policy_id2dir(policy->index));
 621                 hlist_for_each_entry(pol, chain, bydst) {
 622                         if (policy->priority >= pol->priority)
 623                                 newpos = &pol->bydst;
 624                         else
 625                                 break;
 626                 }
 627                 if (newpos)
 628                         hlist_add_behind(&policy->bydst, newpos);
 629                 else
 630                         hlist_add_head(&policy->bydst, chain);
 631         }
 632
 633         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 634
 635         mutex_unlock(&hash_resize_mutex);
 636 }
 637
 638 void xfrm_policy_hash_rebuild(struct net *net)
 639 {
 640         schedule_work(&net->xfrm.policy_hthresh.work);
 641 }
 642 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 643
 644 /* Generate new index... KAME seems to generate them ordered by cost
 645  * of an absolute inpredictability of ordering of rules. This will not pass. */
 646 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 647 {
 648         static u32 idx_generator;
 649
 650         for (;;) {
 651                 struct hlist_head *list;
 652                 struct xfrm_policy *p;
 653                 u32 idx;
 654                 int found;
 655
 656                 if (!index) {
 657                         idx = (idx_generator | dir);
 658                         idx_generator += 8;
 659                 } else {
 660                         idx = index;
 661                         index = 0;
 662                 }
 663
 664                 if (idx == 0)
 665                         idx = 8;
 666                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 667                 found = 0;
 668                 hlist_for_each_entry(p, list, byidx) {
 669                         if (p->index == idx) {
 670                                 found = 1;
 671                                 break;
 672                         }
 673                 }
 674                 if (!found)
 675                         return idx;
 676         }
 677 }
 678
 679 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 680 {
 681         u32 *p1 = (u32 *) s1;
 682         u32 *p2 = (u32 *) s2;
 683         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 684         int i;
 685
 686         for (i = 0; i < len; i++) {
 687                 if (p1[i] != p2[i])
 688                         return 1;
 689         }
 690
 691         return 0;
 692 }
 693
 694 static void xfrm_policy_requeue(struct xfrm_policy *old,
 695                                 struct xfrm_policy *new)
 696 {
 697         struct xfrm_policy_queue *pq = &old->polq;
 698         struct sk_buff_head list;
 699
 700         if (skb_queue_empty(&pq->hold_queue))
 701                 return;
 702
 703         __skb_queue_head_init(&list);
 704
 705         spin_lock_bh(&pq->hold_queue.lock);
 706         skb_queue_splice_init(&pq->hold_queue, &list);
 707         if (del_timer(&pq->hold_timer))
 708                 xfrm_pol_put(old);
 709         spin_unlock_bh(&pq->hold_queue.lock);
 710
 711         pq = &new->polq;
 712
 713         spin_lock_bh(&pq->hold_queue.lock);
 714         skb_queue_splice(&list, &pq->hold_queue);
 715         pq->timeout = XFRM_QUEUE_TMO_MIN;
 716         if (!mod_timer(&pq->hold_timer, jiffies))
 717                 xfrm_pol_hold(new);
 718         spin_unlock_bh(&pq->hold_queue.lock);
 719 }
 720
 721 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 722                                    struct xfrm_policy *pol)
 723 {
 724         u32 mark = policy->mark.v & policy->mark.m;
 725
 726         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 727                 return true;
 728
 729         if ((mark & pol->mark.m) == pol->mark.v &&
 730             policy->priority == pol->priority)
 731                 return true;
 732
 733         return false;
 734 }
 735
 736 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 737 {
 738         struct net *net = xp_net(policy);
 739         struct xfrm_policy *pol;
 740         struct xfrm_policy *delpol;
 741         struct hlist_head *chain;
 742         struct hlist_node *newpos;
 743
 744         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 745         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 746         delpol = NULL;
 747         newpos = NULL;
 748         hlist_for_each_entry(pol, chain, bydst) {
 749                 if (pol->type == policy->type &&
 750                     !selector_cmp(&pol->selector, &policy->selector) &&
 751                     xfrm_policy_mark_match(policy, pol) &&
 752                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 753                     !WARN_ON(delpol)) {
 754                         if (excl) {
 755                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 756                                 return -EEXIST;
 757                         }
 758                         delpol = pol;
 759                         if (policy->priority > pol->priority)
 760                                 continue;
 761                 } else if (policy->priority >= pol->priority) {
 762                         newpos = &pol->bydst;
 763                         continue;
 764                 }
 765                 if (delpol)
 766                         break;
 767         }
 768         if (newpos)
 769                 hlist_add_behind(&policy->bydst, newpos);
 770         else
 771                 hlist_add_head(&policy->bydst, chain);
 772         __xfrm_policy_link(policy, dir);
 773
 774         /* After previous checking, family can either be AF_INET or AF_INET6 */
 775         if (policy->family == AF_INET)
 776                 rt_genid_bump_ipv4(net);
 777         else
 778                 rt_genid_bump_ipv6(net);
 779
 780         if (delpol) {
 781                 xfrm_policy_requeue(delpol, policy);
 782                 __xfrm_policy_unlink(delpol, dir);
 783         }
 784         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 785         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 786         policy->curlft.add_time = get_seconds();
 787         policy->curlft.use_time = 0;
 788         if (!mod_timer(&policy->timer, jiffies + HZ))
 789                 xfrm_pol_hold(policy);
 790         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 791
 792         if (delpol)
 793                 xfrm_policy_kill(delpol);
 794         else if (xfrm_bydst_should_resize(net, dir, NULL))
 795                 schedule_work(&net->xfrm.policy_hash_work);
 796
 797         return 0;
 798 }
 799 EXPORT_SYMBOL(xfrm_policy_insert);
 800
 801 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 802                                           int dir, struct xfrm_selector *sel,
 803                                           struct xfrm_sec_ctx *ctx, int delete,
 804                                           int *err)
 805 {
 806         struct xfrm_policy *pol, *ret;
 807         struct hlist_head *chain;
 808
 809         *err = 0;
 810         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 811         chain = policy_hash_bysel(net, sel, sel->family, dir);
 812         ret = NULL;
 813         hlist_for_each_entry(pol, chain, bydst) {
 814                 if (pol->type == type &&
 815                     (mark & pol->mark.m) == pol->mark.v &&
 816                     !selector_cmp(sel, &pol->selector) &&
 817                     xfrm_sec_ctx_match(ctx, pol->security)) {
 818                         xfrm_pol_hold(pol);
 819                         if (delete) {
 820                                 *err = security_xfrm_policy_delete(
 821                                                                 pol->security);
 822                                 if (*err) {
 823                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 824                                         return pol;
 825                                 }
 826                                 __xfrm_policy_unlink(pol, dir);
 827                         }
 828                         ret = pol;
 829                         break;
 830                 }
 831         }
 832         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 833
 834         if (ret && delete)
 835                 xfrm_policy_kill(ret);
 836         return ret;
 837 }
 838 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 839
 840 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 841                                      int dir, u32 id, int delete, int *err)
 842 {
 843         struct xfrm_policy *pol, *ret;
 844         struct hlist_head *chain;
 845
 846         *err = -ENOENT;
 847         if (xfrm_policy_id2dir(id) != dir)
 848                 return NULL;
 849
 850         *err = 0;
 851         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 852         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 853         ret = NULL;
 854         hlist_for_each_entry(pol, chain, byidx) {
 855                 if (pol->type == type && pol->index == id &&
 856                     (mark & pol->mark.m) == pol->mark.v) {
 857                         xfrm_pol_hold(pol);
 858                         if (delete) {
 859                                 *err = security_xfrm_policy_delete(
 860                                                                 pol->security);
 861                                 if (*err) {
 862                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 863                                         return pol;
 864                                 }
 865                                 __xfrm_policy_unlink(pol, dir);
 866                         }
 867                         ret = pol;
 868                         break;
 869                 }
 870         }
 871         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 872
 873         if (ret && delete)
 874                 xfrm_policy_kill(ret);
 875         return ret;
 876 }
 877 EXPORT_SYMBOL(xfrm_policy_byid);
 878
 879 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 880 static inline int
 881 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 882 {
 883         int dir, err = 0;
 884
 885         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 886                 struct xfrm_policy *pol;
 887                 int i;
 888
 889                 hlist_for_each_entry(pol,
 890                                      &net->xfrm.policy_inexact[dir], bydst) {
 891                         if (pol->type != type)
 892                                 continue;
 893                         err = security_xfrm_policy_delete(pol->security);
 894                         if (err) {
 895                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 896                                 return err;
 897                         }
 898                 }
 899                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 900                         hlist_for_each_entry(pol,
 901                                              net->xfrm.policy_bydst[dir].table + i,
 902                                              bydst) {
 903                                 if (pol->type != type)
 904                                         continue;
 905                                 err = security_xfrm_policy_delete(
 906                                                                 pol->security);
 907                                 if (err) {
 908                                         xfrm_audit_policy_delete(pol, 0,
 909                                                                  task_valid);
 910                                         return err;
 911                                 }
 912                         }
 913                 }
 914         }
 915         return err;
 916 }
 917 #else
 918 static inline int
 919 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 920 {
 921         return 0;
 922 }
 923 #endif
 924
 925 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 926 {
 927         int dir, err = 0, cnt = 0;
 928
 929         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 930
 931         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 932         if (err)
 933                 goto out;
 934
 935         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 936                 struct xfrm_policy *pol;
 937                 int i;
 938
 939         again1:
 940                 hlist_for_each_entry(pol,
 941                                      &net->xfrm.policy_inexact[dir], bydst) {
 942                         if (pol->type != type)
 943                                 continue;
 944                         __xfrm_policy_unlink(pol, dir);
 945                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 946                         cnt++;
 947
 948                         xfrm_audit_policy_delete(pol, 1, task_valid);
 949
 950                         xfrm_policy_kill(pol);
 951
 952                         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 953                         goto again1;
 954                 }
 955
 956                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 957         again2:
 958                         hlist_for_each_entry(pol,
 959                                              net->xfrm.policy_bydst[dir].table + i,
 960                                              bydst) {
 961                                 if (pol->type != type)
 962                                         continue;
 963                                 __xfrm_policy_unlink(pol, dir);
 964                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 965                                 cnt++;
 966
 967                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 968                                 xfrm_policy_kill(pol);
 969
 970                                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 971                                 goto again2;
 972                         }
 973                 }
 974
 975         }
 976         if (!cnt)
 977                 err = -ESRCH;
 978 out:
 979         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 980         return err;
 981 }
 982 EXPORT_SYMBOL(xfrm_policy_flush);
 983
 984 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 985                      int (*func)(struct xfrm_policy *, int, int, void*),
 986                      void *data)
 987 {
 988         struct xfrm_policy *pol;
 989         struct xfrm_policy_walk_entry *x;
 990         int error = 0;
 991
 992         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 993             walk->type != XFRM_POLICY_TYPE_ANY)
 994                 return -EINVAL;
 995
 996         if (list_empty(&walk->walk.all) && walk->seq != 0)
 997                 return 0;
 998
 999         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1000         if (list_empty(&walk->walk.all))
1001                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1002         else
1003                 x = list_first_entry(&walk->walk.all,
1004                                      struct xfrm_policy_walk_entry, all);
1005
1006         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1007                 if (x->dead)
1008                         continue;
1009                 pol = container_of(x, struct xfrm_policy, walk);
1010                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1011                     walk->type != pol->type)
1012                         continue;
1013                 error = func(pol, xfrm_policy_id2dir(pol->index),
1014                              walk->seq, data);
1015                 if (error) {
1016                         list_move_tail(&walk->walk.all, &x->all);
1017                         goto out;
1018                 }
1019                 walk->seq++;
1020         }
1021         if (walk->seq == 0) {
1022                 error = -ENOENT;
1023                 goto out;
1024         }
1025         list_del_init(&walk->walk.all);
1026 out:
1027         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1028         return error;
1029 }
1030 EXPORT_SYMBOL(xfrm_policy_walk);
1031
1032 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1033 {
1034         INIT_LIST_HEAD(&walk->walk.all);
1035         walk->walk.dead = 1;
1036         walk->type = type;
1037         walk->seq = 0;
1038 }
1039 EXPORT_SYMBOL(xfrm_policy_walk_init);
1040
1041 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1042 {
1043         if (list_empty(&walk->walk.all))
1044                 return;
1045
1046         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1047         list_del(&walk->walk.all);
1048         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1049 }
1050 EXPORT_SYMBOL(xfrm_policy_walk_done);
1051
1052 /*
1053  * Find policy to apply to this flow.
1054  *
1055  * Returns 0 if policy found, else an -errno.
1056  */
1057 static int xfrm_policy_match(const struct xfrm_policy *pol,
1058                              const struct flowi *fl,
1059                              u8 type, u16 family, int dir)
1060 {
1061         const struct xfrm_selector *sel = &pol->selector;
1062         int ret = -ESRCH;
1063         bool match;
1064
1065         if (pol->family != family ||
1066             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1067             pol->type != type)
1068                 return ret;
1069
1070         match = xfrm_selector_match(sel, fl, family);
1071         if (match)
1072                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1073                                                   dir);
1074
1075         return ret;
1076 }
1077
1078 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1079                                                      const struct flowi *fl,
1080                                                      u16 family, u8 dir)
1081 {
1082         int err;
1083         struct xfrm_policy *pol, *ret;
1084         const xfrm_address_t *daddr, *saddr;
1085         struct hlist_head *chain;
1086         unsigned int sequence;
1087         u32 priority;
1088
1089         daddr = xfrm_flowi_daddr(fl, family);
1090         saddr = xfrm_flowi_saddr(fl, family);
1091         if (unlikely(!daddr || !saddr))
1092                 return NULL;
1093
1094         rcu_read_lock();
1095  retry:
1096         do {
1097                 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
1098                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
1099         } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
1100
1101         priority = ~0U;
1102         ret = NULL;
1103         hlist_for_each_entry_rcu(pol, chain, bydst) {
1104                 err = xfrm_policy_match(pol, fl, type, family, dir);
1105                 if (err) {
1106                         if (err == -ESRCH)
1107                                 continue;
1108                         else {
1109                                 ret = ERR_PTR(err);
1110                                 goto fail;
1111                         }
1112                 } else {
1113                         ret = pol;
1114                         priority = ret->priority;
1115                         break;
1116                 }
1117         }
1118         chain = &net->xfrm.policy_inexact[dir];
1119         hlist_for_each_entry_rcu(pol, chain, bydst) {
1120                 if ((pol->priority >= priority) && ret)
1121                         break;
1122
1123                 err = xfrm_policy_match(pol, fl, type, family, dir);
1124                 if (err) {
1125                         if (err == -ESRCH)
1126                                 continue;
1127                         else {
1128                                 ret = ERR_PTR(err);
1129                                 goto fail;
1130                         }
1131                 } else {
1132                         ret = pol;
1133                         break;
1134                 }
1135         }
1136
1137         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
1138                 goto retry;
1139
1140         if (ret && !xfrm_pol_hold_rcu(ret))
1141                 goto retry;
1142 fail:
1143         rcu_read_unlock();
1144
1145         return ret;
1146 }
1147
1148 static struct xfrm_policy *
1149 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1150 {
1151 #ifdef CONFIG_XFRM_SUB_POLICY
1152         struct xfrm_policy *pol;
1153
1154         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1155         if (pol != NULL)
1156                 return pol;
1157 #endif
1158         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1159 }
1160
1161 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1162                                                  const struct flowi *fl, u16 family)
1163 {
1164         struct xfrm_policy *pol;
1165
1166         rcu_read_lock();
1167  again:
1168         pol = rcu_dereference(sk->sk_policy[dir]);
1169         if (pol != NULL) {
1170                 bool match;
1171                 int err = 0;
1172
1173                 if (pol->family != family) {
1174                         pol = NULL;
1175                         goto out;
1176                 }
1177
1178                 match = xfrm_selector_match(&pol->selector, fl, family);
1179                 if (match) {
1180                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1181                                 pol = NULL;
1182                                 goto out;
1183                         }
1184                         err = security_xfrm_policy_lookup(pol->security,
1185                                                       fl->flowi_secid,
1186                                                       dir);
1187                         if (!err) {
1188                                 if (!xfrm_pol_hold_rcu(pol))
1189                                         goto again;
1190                         } else if (err == -ESRCH) {
1191                                 pol = NULL;
1192                         } else {
1193                                 pol = ERR_PTR(err);
1194                         }
1195                 } else
1196                         pol = NULL;
1197         }
1198 out:
1199         rcu_read_unlock();
1200         return pol;
1201 }
1202
1203 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1204 {
1205         struct net *net = xp_net(pol);
1206
1207         list_add(&pol->walk.all, &net->xfrm.policy_all);
1208         net->xfrm.policy_count[dir]++;
1209         xfrm_pol_hold(pol);
1210 }
1211
1212 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1213                                                 int dir)
1214 {
1215         struct net *net = xp_net(pol);
1216
1217         if (list_empty(&pol->walk.all))
1218                 return NULL;
1219
1220         /* Socket policies are not hashed. */
1221         if (!hlist_unhashed(&pol->bydst)) {
1222                 hlist_del_rcu(&pol->bydst);
1223                 hlist_del(&pol->byidx);
1224         }
1225
1226         list_del_init(&pol->walk.all);
1227         net->xfrm.policy_count[dir]--;
1228
1229         return pol;
1230 }
1231
1232 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1233 {
1234         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1235 }
1236
1237 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1238 {
1239         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1240 }
1241
1242 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1243 {
1244         struct net *net = xp_net(pol);
1245
1246         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1247         pol = __xfrm_policy_unlink(pol, dir);
1248         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1249         if (pol) {
1250                 xfrm_policy_kill(pol);
1251                 return 0;
1252         }
1253         return -ENOENT;
1254 }
1255 EXPORT_SYMBOL(xfrm_policy_delete);
1256
1257 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1258 {
1259         struct net *net = xp_net(pol);
1260         struct xfrm_policy *old_pol;
1261
1262 #ifdef CONFIG_XFRM_SUB_POLICY
1263         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1264                 return -EINVAL;
1265 #endif
1266
1267         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1268         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1269                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1270         if (pol) {
1271                 pol->curlft.add_time = get_seconds();
1272                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1273                 xfrm_sk_policy_link(pol, dir);
1274         }
1275         rcu_assign_pointer(sk->sk_policy[dir], pol);
1276         if (old_pol) {
1277                 if (pol)
1278                         xfrm_policy_requeue(old_pol, pol);
1279
1280                 /* Unlinking succeeds always. This is the only function
1281                  * allowed to delete or replace socket policy.
1282                  */
1283                 xfrm_sk_policy_unlink(old_pol, dir);
1284         }
1285         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1286
1287         if (old_pol) {
1288                 xfrm_policy_kill(old_pol);
1289         }
1290         return 0;
1291 }
1292
1293 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1294 {
1295         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1296         struct net *net = xp_net(old);
1297
1298         if (newp) {
1299                 newp->selector = old->selector;
1300                 if (security_xfrm_policy_clone(old->security,
1301                                                &newp->security)) {
1302                         kfree(newp);
1303                         return NULL;  /* ENOMEM */
1304                 }
1305                 newp->lft = old->lft;
1306                 newp->curlft = old->curlft;
1307                 newp->mark = old->mark;
1308                 newp->action = old->action;
1309                 newp->flags = old->flags;
1310                 newp->xfrm_nr = old->xfrm_nr;
1311                 newp->index = old->index;
1312                 newp->type = old->type;
1313                 newp->family = old->family;
1314                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1315                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1316                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1317                 xfrm_sk_policy_link(newp, dir);
1318                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1319                 xfrm_pol_put(newp);
1320         }
1321         return newp;
1322 }
1323
1324 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1325 {
1326         const struct xfrm_policy *p;
1327         struct xfrm_policy *np;
1328         int i, ret = 0;
1329
1330         rcu_read_lock();
1331         for (i = 0; i < 2; i++) {
1332                 p = rcu_dereference(osk->sk_policy[i]);
1333                 if (p) {
1334                         np = clone_policy(p, i);
1335                         if (unlikely(!np)) {
1336                                 ret = -ENOMEM;
1337                                 break;
1338                         }
1339                         rcu_assign_pointer(sk->sk_policy[i], np);
1340                 }
1341         }
1342         rcu_read_unlock();
1343         return ret;
1344 }
1345
1346 static int
1347 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1348                xfrm_address_t *remote, unsigned short family, u32 mark)
1349 {
1350         int err;
1351         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1352
1353         if (unlikely(afinfo == NULL))
1354                 return -EINVAL;
1355         err = afinfo->get_saddr(net, oif, local, remote, mark);
1356         rcu_read_unlock();
1357         return err;
1358 }
1359
1360 /* Resolve list of templates for the flow, given policy. */
1361
1362 static int
1363 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1364                       struct xfrm_state **xfrm, unsigned short family)
1365 {
1366         struct net *net = xp_net(policy);
1367         int nx;
1368         int i, error;
1369         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1370         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1371         xfrm_address_t tmp;
1372
1373         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1374                 struct xfrm_state *x;
1375                 xfrm_address_t *remote = daddr;
1376                 xfrm_address_t *local  = saddr;
1377                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1378
1379                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1380                     tmpl->mode == XFRM_MODE_BEET) {
1381                         remote = &tmpl->id.daddr;
1382                         local = &tmpl->saddr;
1383                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1384                                 error = xfrm_get_saddr(net, fl->flowi_oif,
1385                                                        &tmp, remote,
1386                                                        tmpl->encap_family, 0);
1387                                 if (error)
1388                                         goto fail;
1389                                 local = &tmp;
1390                         }
1391                 }
1392
1393                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1394
1395                 if (x && x->km.state == XFRM_STATE_VALID) {
1396                         xfrm[nx++] = x;
1397                         daddr = remote;
1398                         saddr = local;
1399                         continue;
1400                 }
1401                 if (x) {
1402                         error = (x->km.state == XFRM_STATE_ERROR ?
1403                                  -EINVAL : -EAGAIN);
1404                         xfrm_state_put(x);
1405                 } else if (error == -ESRCH) {
1406                         error = -EAGAIN;
1407                 }
1408
1409                 if (!tmpl->optional)
1410                         goto fail;
1411         }
1412         return nx;
1413
1414 fail:
1415         for (nx--; nx >= 0; nx--)
1416                 xfrm_state_put(xfrm[nx]);
1417         return error;
1418 }
1419
1420 static int
1421 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1422                   struct xfrm_state **xfrm, unsigned short family)
1423 {
1424         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1425         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1426         int cnx = 0;
1427         int error;
1428         int ret;
1429         int i;
1430
1431         for (i = 0; i < npols; i++) {
1432                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1433                         error = -ENOBUFS;
1434                         goto fail;
1435                 }
1436
1437                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1438                 if (ret < 0) {
1439                         error = ret;
1440                         goto fail;
1441                 } else
1442                         cnx += ret;
1443         }
1444
1445         /* found states are sorted for outbound processing */
1446         if (npols > 1)
1447                 xfrm_state_sort(xfrm, tpp, cnx, family);
1448
1449         return cnx;
1450
1451  fail:
1452         for (cnx--; cnx >= 0; cnx--)
1453                 xfrm_state_put(tpp[cnx]);
1454         return error;
1455
1456 }
1457
1458 static int xfrm_get_tos(const struct flowi *fl, int family)
1459 {
1460         const struct xfrm_policy_afinfo *afinfo;
1461         int tos = 0;
1462
1463         afinfo = xfrm_policy_get_afinfo(family);
1464         tos = afinfo ? afinfo->get_tos(fl) : 0;
1465
1466         rcu_read_unlock();
1467
1468         return tos;
1469 }
1470
1471 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1472 {
1473         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1474         struct dst_ops *dst_ops;
1475         struct xfrm_dst *xdst;
1476
1477         if (!afinfo)
1478                 return ERR_PTR(-EINVAL);
1479
1480         switch (family) {
1481         case AF_INET:
1482                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1483                 break;
1484 #if IS_ENABLED(CONFIG_IPV6)
1485         case AF_INET6:
1486                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1487                 break;
1488 #endif
1489         default:
1490                 BUG();
1491         }
1492         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
1493
1494         if (likely(xdst)) {
1495                 struct dst_entry *dst = &xdst->u.dst;
1496
1497                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1498         } else
1499                 xdst = ERR_PTR(-ENOBUFS);
1500
1501         rcu_read_unlock();
1502
1503         return xdst;
1504 }
1505
1506 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1507                                  int nfheader_len)
1508 {
1509         const struct xfrm_policy_afinfo *afinfo =
1510                 xfrm_policy_get_afinfo(dst->ops->family);
1511         int err;
1512
1513         if (!afinfo)
1514                 return -EINVAL;
1515
1516         err = afinfo->init_path(path, dst, nfheader_len);
1517
1518         rcu_read_unlock();
1519
1520         return err;
1521 }
1522
1523 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1524                                 const struct flowi *fl)
1525 {
1526         const struct xfrm_policy_afinfo *afinfo =
1527                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1528         int err;
1529
1530         if (!afinfo)
1531                 return -EINVAL;
1532
1533         err = afinfo->fill_dst(xdst, dev, fl);
1534
1535         rcu_read_unlock();
1536
1537         return err;
1538 }
1539
1540
1541 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1542  * all the metrics... Shortly, bundle a bundle.
1543  */
1544
1545 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1546                                             struct xfrm_state **xfrm, int nx,
1547                                             const struct flowi *fl,
1548                                             struct dst_entry *dst)
1549 {
1550         struct net *net = xp_net(policy);
1551         unsigned long now = jiffies;
1552         struct net_device *dev;
1553         struct xfrm_mode *inner_mode;
1554         struct dst_entry *dst_prev = NULL;
1555         struct dst_entry *dst0 = NULL;
1556         int i = 0;
1557         int err;
1558         int header_len = 0;
1559         int nfheader_len = 0;
1560         int trailer_len = 0;
1561         int tos;
1562         int family = policy->selector.family;
1563         xfrm_address_t saddr, daddr;
1564
1565         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1566
1567         tos = xfrm_get_tos(fl, family);
1568
1569         dst_hold(dst);
1570
1571         for (; i < nx; i++) {
1572                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1573                 struct dst_entry *dst1 = &xdst->u.dst;
1574
1575                 err = PTR_ERR(xdst);
1576                 if (IS_ERR(xdst)) {
1577                         dst_release(dst);
1578                         goto put_states;
1579                 }
1580
1581                 if (!dst_prev)
1582                         dst0 = dst1;
1583                 else
1584                         /* Ref count is taken during xfrm_alloc_dst()
1585                          * No need to do dst_clone() on dst1
1586                          */
1587                         dst_prev->child = dst1;
1588
1589                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1590                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1591                                                         xfrm_af2proto(family));
1592                         if (!inner_mode) {
1593                                 err = -EAFNOSUPPORT;
1594                                 dst_release(dst);
1595                                 goto put_states;
1596                         }
1597                 } else
1598                         inner_mode = xfrm[i]->inner_mode;
1599
1600                 xdst->route = dst;
1601                 dst_copy_metrics(dst1, dst);
1602
1603                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1604                         family = xfrm[i]->props.family;
1605                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1606                                               &saddr, &daddr, family,
1607                                               xfrm[i]->props.output_mark);
1608                         err = PTR_ERR(dst);
1609                         if (IS_ERR(dst))
1610                                 goto put_states;
1611                 } else
1612                         dst_hold(dst);
1613
1614                 dst1->xfrm = xfrm[i];
1615                 xdst->xfrm_genid = xfrm[i]->genid;
1616
1617                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1618                 dst1->flags |= DST_HOST;
1619                 dst1->lastuse = now;
1620
1621                 dst1->input = dst_discard;
1622                 dst1->output = inner_mode->afinfo->output;
1623
1624                 dst1->next = dst_prev;
1625                 dst_prev = dst1;
1626
1627                 header_len += xfrm[i]->props.header_len;
1628                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1629                         nfheader_len += xfrm[i]->props.header_len;
1630                 trailer_len += xfrm[i]->props.trailer_len;
1631         }
1632
1633         dst_prev->child = dst;
1634         dst0->path = dst;
1635
1636         err = -ENODEV;
1637         dev = dst->dev;
1638         if (!dev)
1639                 goto free_dst;
1640
1641         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1642         xfrm_init_pmtu(dst_prev);
1643
1644         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1645                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1646
1647                 err = xfrm_fill_dst(xdst, dev, fl);
1648                 if (err)
1649                         goto free_dst;
1650
1651                 dst_prev->header_len = header_len;
1652                 dst_prev->trailer_len = trailer_len;
1653                 header_len -= xdst->u.dst.xfrm->props.header_len;
1654                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1655         }
1656
1657 out:
1658         return dst0;
1659
1660 put_states:
1661         for (; i < nx; i++)
1662                 xfrm_state_put(xfrm[i]);
1663 free_dst:
1664         if (dst0)
1665                 dst_release_immediate(dst0);
1666         dst0 = ERR_PTR(err);
1667         goto out;
1668 }
1669
1670 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1671                                 struct xfrm_policy **pols,
1672                                 int *num_pols, int *num_xfrms)
1673 {
1674         int i;
1675
1676         if (*num_pols == 0 || !pols[0]) {
1677                 *num_pols = 0;
1678                 *num_xfrms = 0;
1679                 return 0;
1680         }
1681         if (IS_ERR(pols[0]))
1682                 return PTR_ERR(pols[0]);
1683
1684         *num_xfrms = pols[0]->xfrm_nr;
1685
1686 #ifdef CONFIG_XFRM_SUB_POLICY
1687         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1688             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1689                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1690                                                     XFRM_POLICY_TYPE_MAIN,
1691                                                     fl, family,
1692                                                     XFRM_POLICY_OUT);
1693                 if (pols[1]) {
1694                         if (IS_ERR(pols[1])) {
1695                                 xfrm_pols_put(pols, *num_pols);
1696                                 return PTR_ERR(pols[1]);
1697                         }
1698                         (*num_pols)++;
1699                         (*num_xfrms) += pols[1]->xfrm_nr;
1700                 }
1701         }
1702 #endif
1703         for (i = 0; i < *num_pols; i++) {
1704                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1705                         *num_xfrms = -1;
1706                         break;
1707                 }
1708         }
1709
1710         return 0;
1711
1712 }
1713
1714 static void xfrm_last_dst_update(struct xfrm_dst *xdst, struct xfrm_dst *old)
1715 {
1716         this_cpu_write(xfrm_last_dst, xdst);
1717         if (old)
1718                 dst_release(&old->u.dst);
1719 }
1720
1721 static void __xfrm_pcpu_work_fn(void)
1722 {
1723         struct xfrm_dst *old;
1724
1725         old = this_cpu_read(xfrm_last_dst);
1726         if (old && !xfrm_bundle_ok(old))
1727                 xfrm_last_dst_update(NULL, old);
1728 }
1729
1730 static void xfrm_pcpu_work_fn(struct work_struct *work)
1731 {
1732         local_bh_disable();
1733         rcu_read_lock();
1734         __xfrm_pcpu_work_fn();
1735         rcu_read_unlock();
1736         local_bh_enable();
1737 }
1738
1739 void xfrm_policy_cache_flush(void)
1740 {
1741         struct xfrm_dst *old;
1742         bool found = 0;
1743         int cpu;
1744
1745         might_sleep();
1746
1747         local_bh_disable();
1748         rcu_read_lock();
1749         for_each_possible_cpu(cpu) {
1750                 old = per_cpu(xfrm_last_dst, cpu);
1751                 if (old && !xfrm_bundle_ok(old)) {
1752                         if (smp_processor_id() == cpu) {
1753                                 __xfrm_pcpu_work_fn();
1754                                 continue;
1755                         }
1756                         found = true;
1757                         break;
1758                 }
1759         }
1760
1761         rcu_read_unlock();
1762         local_bh_enable();
1763
1764         if (!found)
1765                 return;
1766
1767         get_online_cpus();
1768
1769         for_each_possible_cpu(cpu) {
1770                 bool bundle_release;
1771
1772                 rcu_read_lock();
1773                 old = per_cpu(xfrm_last_dst, cpu);
1774                 bundle_release = old && !xfrm_bundle_ok(old);
1775                 rcu_read_unlock();
1776
1777                 if (!bundle_release)
1778                         continue;
1779
1780                 if (cpu_online(cpu)) {
1781                         schedule_work_on(cpu, &xfrm_pcpu_work[cpu]);
1782                         continue;
1783                 }
1784
1785                 rcu_read_lock();
1786                 old = per_cpu(xfrm_last_dst, cpu);
1787                 if (old && !xfrm_bundle_ok(old)) {
1788                         per_cpu(xfrm_last_dst, cpu) = NULL;
1789                         dst_release(&old->u.dst);
1790                 }
1791                 rcu_read_unlock();
1792         }
1793
1794         put_online_cpus();
1795 }
1796
1797 static bool xfrm_xdst_can_reuse(struct xfrm_dst *xdst,
1798                                 struct xfrm_state * const xfrm[],
1799                                 int num)
1800 {
1801         const struct dst_entry *dst = &xdst->u.dst;
1802         int i;
1803
1804         if (xdst->num_xfrms != num)
1805                 return false;
1806
1807         for (i = 0; i < num; i++) {
1808                 if (!dst || dst->xfrm != xfrm[i])
1809                         return false;
1810                 dst = dst->child;
1811         }
1812
1813         return xfrm_bundle_ok(xdst);
1814 }
1815
1816 static struct xfrm_dst *
1817 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1818                                const struct flowi *fl, u16 family,
1819                                struct dst_entry *dst_orig)
1820 {
1821         struct net *net = xp_net(pols[0]);
1822         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1823         struct xfrm_dst *xdst, *old;
1824         struct dst_entry *dst;
1825         int err;
1826
1827         /* Try to instantiate a bundle */
1828         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1829         if (err <= 0) {
1830                 if (err != 0 && err != -EAGAIN)
1831                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1832                 return ERR_PTR(err);
1833         }
1834
1835         xdst = this_cpu_read(xfrm_last_dst);
1836         if (xdst &&
1837             xdst->u.dst.dev == dst_orig->dev &&
1838             xdst->num_pols == num_pols &&
1839             memcmp(xdst->pols, pols,
1840                    sizeof(struct xfrm_policy *) * num_pols) == 0 &&
1841             xfrm_xdst_can_reuse(xdst, xfrm, err)) {
1842                 dst_hold(&xdst->u.dst);
1843                 xfrm_pols_put(pols, num_pols);
1844                 while (err > 0)
1845                         xfrm_state_put(xfrm[--err]);
1846                 return xdst;
1847         }
1848
1849         old = xdst;
1850
1851         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1852         if (IS_ERR(dst)) {
1853                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1854                 return ERR_CAST(dst);
1855         }
1856
1857         xdst = (struct xfrm_dst *)dst;
1858         xdst->num_xfrms = err;
1859         xdst->num_pols = num_pols;
1860         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1861         xdst->policy_genid = atomic_read(&pols[0]->genid);
1862
1863         atomic_set(&xdst->u.dst.__refcnt, 2);
1864         xfrm_last_dst_update(xdst, old);
1865
1866         return xdst;
1867 }
1868
1869 static void xfrm_policy_queue_process(struct timer_list *t)
1870 {
1871         struct sk_buff *skb;
1872         struct sock *sk;
1873         struct dst_entry *dst;
1874         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
1875         struct net *net = xp_net(pol);
1876         struct xfrm_policy_queue *pq = &pol->polq;
1877         struct flowi fl;
1878         struct sk_buff_head list;
1879
1880         spin_lock(&pq->hold_queue.lock);
1881         skb = skb_peek(&pq->hold_queue);
1882         if (!skb) {
1883                 spin_unlock(&pq->hold_queue.lock);
1884                 goto out;
1885         }
1886         dst = skb_dst(skb);
1887         sk = skb->sk;
1888         xfrm_decode_session(skb, &fl, dst->ops->family);
1889         spin_unlock(&pq->hold_queue.lock);
1890
1891         dst_hold(dst->path);
1892         dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1893         if (IS_ERR(dst))
1894                 goto purge_queue;
1895
1896         if (dst->flags & DST_XFRM_QUEUE) {
1897                 dst_release(dst);
1898
1899                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1900                         goto purge_queue;
1901
1902                 pq->timeout = pq->timeout << 1;
1903                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1904                         xfrm_pol_hold(pol);
1905         goto out;
1906         }
1907
1908         dst_release(dst);
1909
1910         __skb_queue_head_init(&list);
1911
1912         spin_lock(&pq->hold_queue.lock);
1913         pq->timeout = 0;
1914         skb_queue_splice_init(&pq->hold_queue, &list);
1915         spin_unlock(&pq->hold_queue.lock);
1916
1917         while (!skb_queue_empty(&list)) {
1918                 skb = __skb_dequeue(&list);
1919
1920                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1921                 dst_hold(skb_dst(skb)->path);
1922                 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1923                 if (IS_ERR(dst)) {
1924                         kfree_skb(skb);
1925                         continue;
1926                 }
1927
1928                 nf_reset(skb);
1929                 skb_dst_drop(skb);
1930                 skb_dst_set(skb, dst);
1931
1932                 dst_output(net, skb->sk, skb);
1933         }
1934
1935 out:
1936         xfrm_pol_put(pol);
1937         return;
1938
1939 purge_queue:
1940         pq->timeout = 0;
1941         skb_queue_purge(&pq->hold_queue);
1942         xfrm_pol_put(pol);
1943 }
1944
1945 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1946 {
1947         unsigned long sched_next;
1948         struct dst_entry *dst = skb_dst(skb);
1949         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1950         struct xfrm_policy *pol = xdst->pols[0];
1951         struct xfrm_policy_queue *pq = &pol->polq;
1952
1953         if (unlikely(skb_fclone_busy(sk, skb))) {
1954                 kfree_skb(skb);
1955                 return 0;
1956         }
1957
1958         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1959                 kfree_skb(skb);
1960                 return -EAGAIN;
1961         }
1962
1963         skb_dst_force(skb);
1964
1965         spin_lock_bh(&pq->hold_queue.lock);
1966
1967         if (!pq->timeout)
1968                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1969
1970         sched_next = jiffies + pq->timeout;
1971
1972         if (del_timer(&pq->hold_timer)) {
1973                 if (time_before(pq->hold_timer.expires, sched_next))
1974                         sched_next = pq->hold_timer.expires;
1975                 xfrm_pol_put(pol);
1976         }
1977
1978         __skb_queue_tail(&pq->hold_queue, skb);
1979         if (!mod_timer(&pq->hold_timer, sched_next))
1980                 xfrm_pol_hold(pol);
1981
1982         spin_unlock_bh(&pq->hold_queue.lock);
1983
1984         return 0;
1985 }
1986
1987 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1988                                                  struct xfrm_flo *xflo,
1989                                                  const struct flowi *fl,
1990                                                  int num_xfrms,
1991                                                  u16 family)
1992 {
1993         int err;
1994         struct net_device *dev;
1995         struct dst_entry *dst;
1996         struct dst_entry *dst1;
1997         struct xfrm_dst *xdst;
1998
1999         xdst = xfrm_alloc_dst(net, family);
2000         if (IS_ERR(xdst))
2001                 return xdst;
2002
2003         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2004             net->xfrm.sysctl_larval_drop ||
2005             num_xfrms <= 0)
2006                 return xdst;
2007
2008         dst = xflo->dst_orig;
2009         dst1 = &xdst->u.dst;
2010         dst_hold(dst);
2011         xdst->route = dst;
2012
2013         dst_copy_metrics(dst1, dst);
2014
2015         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2016         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2017         dst1->lastuse = jiffies;
2018
2019         dst1->input = dst_discard;
2020         dst1->output = xdst_queue_output;
2021
2022         dst_hold(dst);
2023         dst1->child = dst;
2024         dst1->path = dst;
2025
2026         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2027
2028         err = -ENODEV;
2029         dev = dst->dev;
2030         if (!dev)
2031                 goto free_dst;
2032
2033         err = xfrm_fill_dst(xdst, dev, fl);
2034         if (err)
2035                 goto free_dst;
2036
2037 out:
2038         return xdst;
2039
2040 free_dst:
2041         dst_release(dst1);
2042         xdst = ERR_PTR(err);
2043         goto out;
2044 }
2045
2046 static struct xfrm_dst *
2047 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, struct xfrm_flo *xflo)
2048 {
2049         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2050         int num_pols = 0, num_xfrms = 0, err;
2051         struct xfrm_dst *xdst;
2052
2053         /* Resolve policies to use if we couldn't get them from
2054          * previous cache entry */
2055         num_pols = 1;
2056         pols[0] = xfrm_policy_lookup(net, fl, family, dir);
2057         err = xfrm_expand_policies(fl, family, pols,
2058                                            &num_pols, &num_xfrms);
2059         if (err < 0)
2060                 goto inc_error;
2061         if (num_pols == 0)
2062                 return NULL;
2063         if (num_xfrms <= 0)
2064                 goto make_dummy_bundle;
2065
2066         local_bh_disable();
2067         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2068                                               xflo->dst_orig);
2069         local_bh_enable();
2070
2071         if (IS_ERR(xdst)) {
2072                 err = PTR_ERR(xdst);
2073                 if (err != -EAGAIN)
2074                         goto error;
2075                 goto make_dummy_bundle;
2076         } else if (xdst == NULL) {
2077                 num_xfrms = 0;
2078                 goto make_dummy_bundle;
2079         }
2080
2081         return xdst;
2082
2083 make_dummy_bundle:
2084         /* We found policies, but there's no bundles to instantiate:
2085          * either because the policy blocks, has no transformations or
2086          * we could not build template (no xfrm_states).*/
2087         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2088         if (IS_ERR(xdst)) {
2089                 xfrm_pols_put(pols, num_pols);
2090                 return ERR_CAST(xdst);
2091         }
2092         xdst->num_pols = num_pols;
2093         xdst->num_xfrms = num_xfrms;
2094         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2095
2096         return xdst;
2097
2098 inc_error:
2099         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2100 error:
2101         xfrm_pols_put(pols, num_pols);
2102         return ERR_PTR(err);
2103 }
2104
2105 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2106                                         struct dst_entry *dst_orig)
2107 {
2108         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2109         struct dst_entry *ret;
2110
2111         if (!afinfo) {
2112                 dst_release(dst_orig);
2113                 return ERR_PTR(-EINVAL);
2114         } else {
2115                 ret = afinfo->blackhole_route(net, dst_orig);
2116         }
2117         rcu_read_unlock();
2118
2119         return ret;
2120 }
2121
2122 /* Main function: finds/creates a bundle for given flow.
2123  *
2124  * At the moment we eat a raw IP route. Mostly to speed up lookups
2125  * on interfaces with disabled IPsec.
2126  */
2127 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2128                               const struct flowi *fl,
2129                               const struct sock *sk, int flags)
2130 {
2131         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2132         struct xfrm_dst *xdst;
2133         struct dst_entry *dst, *route;
2134         u16 family = dst_orig->ops->family;
2135         u8 dir = XFRM_POLICY_OUT;
2136         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2137
2138         dst = NULL;
2139         xdst = NULL;
2140         route = NULL;
2141
2142         sk = sk_const_to_full_sk(sk);
2143         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2144                 num_pols = 1;
2145                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family);
2146                 err = xfrm_expand_policies(fl, family, pols,
2147                                            &num_pols, &num_xfrms);
2148                 if (err < 0)
2149                         goto dropdst;
2150
2151                 if (num_pols) {
2152                         if (num_xfrms <= 0) {
2153                                 drop_pols = num_pols;
2154                                 goto no_transform;
2155                         }
2156
2157                         local_bh_disable();
2158                         xdst = xfrm_resolve_and_create_bundle(
2159                                         pols, num_pols, fl,
2160                                         family, dst_orig);
2161                         local_bh_enable();
2162
2163                         if (IS_ERR(xdst)) {
2164                                 xfrm_pols_put(pols, num_pols);
2165                                 err = PTR_ERR(xdst);
2166                                 goto dropdst;
2167                         } else if (xdst == NULL) {
2168                                 num_xfrms = 0;
2169                                 drop_pols = num_pols;
2170                                 goto no_transform;
2171                         }
2172
2173                         route = xdst->route;
2174                 }
2175         }
2176
2177         if (xdst == NULL) {
2178                 struct xfrm_flo xflo;
2179
2180                 xflo.dst_orig = dst_orig;
2181                 xflo.flags = flags;
2182
2183                 /* To accelerate a bit...  */
2184                 if ((dst_orig->flags & DST_NOXFRM) ||
2185                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2186                         goto nopol;
2187
2188                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo);
2189                 if (xdst == NULL)
2190                         goto nopol;
2191                 if (IS_ERR(xdst)) {
2192                         err = PTR_ERR(xdst);
2193                         goto dropdst;
2194                 }
2195
2196                 num_pols = xdst->num_pols;
2197                 num_xfrms = xdst->num_xfrms;
2198                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2199                 route = xdst->route;
2200         }
2201
2202         dst = &xdst->u.dst;
2203         if (route == NULL && num_xfrms > 0) {
2204                 /* The only case when xfrm_bundle_lookup() returns a
2205                  * bundle with null route, is when the template could
2206                  * not be resolved. It means policies are there, but
2207                  * bundle could not be created, since we don't yet
2208                  * have the xfrm_state's. We need to wait for KM to
2209                  * negotiate new SA's or bail out with error.*/
2210                 if (net->xfrm.sysctl_larval_drop) {
2211                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2212                         err = -EREMOTE;
2213                         goto error;
2214                 }
2215
2216                 err = -EAGAIN;
2217
2218                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2219                 goto error;
2220         }
2221
2222 no_transform:
2223         if (num_pols == 0)
2224                 goto nopol;
2225
2226         if ((flags & XFRM_LOOKUP_ICMP) &&
2227             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2228                 err = -ENOENT;
2229                 goto error;
2230         }
2231
2232         for (i = 0; i < num_pols; i++)
2233                 pols[i]->curlft.use_time = get_seconds();
2234
2235         if (num_xfrms < 0) {
2236                 /* Prohibit the flow */
2237                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2238                 err = -EPERM;
2239                 goto error;
2240         } else if (num_xfrms > 0) {
2241                 /* Flow transformed */
2242                 dst_release(dst_orig);
2243         } else {
2244                 /* Flow passes untransformed */
2245                 dst_release(dst);
2246                 dst = dst_orig;
2247         }
2248 ok:
2249         xfrm_pols_put(pols, drop_pols);
2250         if (dst && dst->xfrm &&
2251             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2252                 dst->flags |= DST_XFRM_TUNNEL;
2253         return dst;
2254
2255 nopol:
2256         if (!(flags & XFRM_LOOKUP_ICMP)) {
2257                 dst = dst_orig;
2258                 goto ok;
2259         }
2260         err = -ENOENT;
2261 error:
2262         dst_release(dst);
2263 dropdst:
2264         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2265                 dst_release(dst_orig);
2266         xfrm_pols_put(pols, drop_pols);
2267         return ERR_PTR(err);
2268 }
2269 EXPORT_SYMBOL(xfrm_lookup);
2270
2271 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2272  * Otherwise we may send out blackholed packets.
2273  */
2274 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2275                                     const struct flowi *fl,
2276                                     const struct sock *sk, int flags)
2277 {
2278         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2279                                             flags | XFRM_LOOKUP_QUEUE |
2280                                             XFRM_LOOKUP_KEEP_DST_REF);
2281
2282         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2283                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2284
2285         return dst;
2286 }
2287 EXPORT_SYMBOL(xfrm_lookup_route);
2288
2289 static inline int
2290 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2291 {
2292         struct xfrm_state *x;
2293
2294         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2295                 return 0;
2296         x = skb->sp->xvec[idx];
2297         if (!x->type->reject)
2298                 return 0;
2299         return x->type->reject(x, skb, fl);
2300 }
2301
2302 /* When skb is transformed back to its "native" form, we have to
2303  * check policy restrictions. At the moment we make this in maximally
2304  * stupid way. Shame on me. :-) Of course, connected sockets must
2305  * have policy cached at them.
2306  */
2307
2308 static inline int
2309 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2310               unsigned short family)
2311 {
2312         if (xfrm_state_kern(x))
2313                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2314         return  x->id.proto == tmpl->id.proto &&
2315                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2316                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2317                 x->props.mode == tmpl->mode &&
2318                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2319                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2320                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2321                   xfrm_state_addr_cmp(tmpl, x, family));
2322 }
2323
2324 /*
2325  * 0 or more than 0 is returned when validation is succeeded (either bypass
2326  * because of optional transport mode, or next index of the mathced secpath
2327  * state with the template.
2328  * -1 is returned when no matching template is found.
2329  * Otherwise "-2 - errored_index" is returned.
2330  */
2331 static inline int
2332 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2333                unsigned short family)
2334 {
2335         int idx = start;
2336
2337         if (tmpl->optional) {
2338                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2339                         return start;
2340         } else
2341                 start = -1;
2342         for (; idx < sp->len; idx++) {
2343                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2344                         return ++idx;
2345                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2346                         if (start == -1)
2347                                 start = -2-idx;
2348                         break;
2349                 }
2350         }
2351         return start;
2352 }
2353
2354 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2355                           unsigned int family, int reverse)
2356 {
2357         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2358         int err;
2359
2360         if (unlikely(afinfo == NULL))
2361                 return -EAFNOSUPPORT;
2362
2363         afinfo->decode_session(skb, fl, reverse);
2364         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2365         rcu_read_unlock();
2366         return err;
2367 }
2368 EXPORT_SYMBOL(__xfrm_decode_session);
2369
2370 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2371 {
2372         for (; k < sp->len; k++) {
2373                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2374                         *idxp = k;
2375                         return 1;
2376                 }
2377         }
2378
2379         return 0;
2380 }
2381
2382 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2383                         unsigned short family)
2384 {
2385         struct net *net = dev_net(skb->dev);
2386         struct xfrm_policy *pol;
2387         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2388         int npols = 0;
2389         int xfrm_nr;
2390         int pi;
2391         int reverse;
2392         struct flowi fl;
2393         int xerr_idx = -1;
2394
2395         reverse = dir & ~XFRM_POLICY_MASK;
2396         dir &= XFRM_POLICY_MASK;
2397
2398         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2399                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2400                 return 0;
2401         }
2402
2403         nf_nat_decode_session(skb, &fl, family);
2404
2405         /* First, check used SA against their selectors. */
2406         if (skb->sp) {
2407                 int i;
2408
2409                 for (i = skb->sp->len-1; i >= 0; i--) {
2410                         struct xfrm_state *x = skb->sp->xvec[i];
2411                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2412                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2413                                 return 0;
2414                         }
2415                 }
2416         }
2417
2418         pol = NULL;
2419         sk = sk_to_full_sk(sk);
2420         if (sk && sk->sk_policy[dir]) {
2421                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family);
2422                 if (IS_ERR(pol)) {
2423                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2424                         return 0;
2425                 }
2426         }
2427
2428         if (!pol)
2429                 pol = xfrm_policy_lookup(net, &fl, family, dir);
2430
2431         if (IS_ERR(pol)) {
2432                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2433                 return 0;
2434         }
2435
2436         if (!pol) {
2437                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2438                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2439                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2440                         return 0;
2441                 }
2442                 return 1;
2443         }
2444
2445         pol->curlft.use_time = get_seconds();
2446
2447         pols[0] = pol;
2448         npols++;
2449 #ifdef CONFIG_XFRM_SUB_POLICY
2450         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2451                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2452                                                     &fl, family,
2453                                                     XFRM_POLICY_IN);
2454                 if (pols[1]) {
2455                         if (IS_ERR(pols[1])) {
2456                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2457                                 return 0;
2458                         }
2459                         pols[1]->curlft.use_time = get_seconds();
2460                         npols++;
2461                 }
2462         }
2463 #endif
2464
2465         if (pol->action == XFRM_POLICY_ALLOW) {
2466                 struct sec_path *sp;
2467                 static struct sec_path dummy;
2468                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2469                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2470                 struct xfrm_tmpl **tpp = tp;
2471                 int ti = 0;
2472                 int i, k;
2473
2474                 if ((sp = skb->sp) == NULL)
2475                         sp = &dummy;
2476
2477                 for (pi = 0; pi < npols; pi++) {
2478                         if (pols[pi] != pol &&
2479                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2480                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2481                                 goto reject;
2482                         }
2483                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2484                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2485                                 goto reject_error;
2486                         }
2487                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2488                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2489                 }
2490                 xfrm_nr = ti;
2491                 if (npols > 1) {
2492                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2493                         tpp = stp;
2494                 }
2495
2496                 /* For each tunnel xfrm, find the first matching tmpl.
2497                  * For each tmpl before that, find corresponding xfrm.
2498                  * Order is _important_. Later we will implement
2499                  * some barriers, but at the moment barriers
2500                  * are implied between each two transformations.
2501                  */
2502                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2503                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2504                         if (k < 0) {
2505                                 if (k < -1)
2506                                         /* "-2 - errored_index" returned */
2507                                         xerr_idx = -(2+k);
2508                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2509                                 goto reject;
2510                         }
2511                 }
2512
2513                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2514                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2515                         goto reject;
2516                 }
2517
2518                 xfrm_pols_put(pols, npols);
2519                 return 1;
2520         }
2521         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2522
2523 reject:
2524         xfrm_secpath_reject(xerr_idx, skb, &fl);
2525 reject_error:
2526         xfrm_pols_put(pols, npols);
2527         return 0;
2528 }
2529 EXPORT_SYMBOL(__xfrm_policy_check);
2530
2531 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2532 {
2533         struct net *net = dev_net(skb->dev);
2534         struct flowi fl;
2535         struct dst_entry *dst;
2536         int res = 1;
2537
2538         if (xfrm_decode_session(skb, &fl, family) < 0) {
2539                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2540                 return 0;
2541         }
2542
2543         skb_dst_force(skb);
2544
2545         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2546         if (IS_ERR(dst)) {
2547                 res = 0;
2548                 dst = NULL;
2549         }
2550         skb_dst_set(skb, dst);
2551         return res;
2552 }
2553 EXPORT_SYMBOL(__xfrm_route_forward);
2554
2555 /* Optimize later using cookies and generation ids. */
2556
2557 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2558 {
2559         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2560          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2561          * get validated by dst_ops->check on every use.  We do this
2562          * because when a normal route referenced by an XFRM dst is
2563          * obsoleted we do not go looking around for all parent
2564          * referencing XFRM dsts so that we can invalidate them.  It
2565          * is just too much work.  Instead we make the checks here on
2566          * every use.  For example:
2567          *
2568          *      XFRM dst A --> IPv4 dst X
2569          *
2570          * X is the "xdst->route" of A (X is also the "dst->path" of A
2571          * in this example).  If X is marked obsolete, "A" will not
2572          * notice.  That's what we are validating here via the
2573          * stale_bundle() check.
2574          *
2575          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
2576          * be marked on it.
2577          * This will force stale_bundle() to fail on any xdst bundle with
2578          * this dst linked in it.
2579          */
2580         if (dst->obsolete < 0 && !stale_bundle(dst))
2581                 return dst;
2582
2583         return NULL;
2584 }
2585
2586 static int stale_bundle(struct dst_entry *dst)
2587 {
2588         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2589 }
2590
2591 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2592 {
2593         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2594                 dst->dev = dev_net(dev)->loopback_dev;
2595                 dev_hold(dst->dev);
2596                 dev_put(dev);
2597         }
2598 }
2599 EXPORT_SYMBOL(xfrm_dst_ifdown);
2600
2601 static void xfrm_link_failure(struct sk_buff *skb)
2602 {
2603         /* Impossible. Such dst must be popped before reaches point of failure. */
2604 }
2605
2606 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2607 {
2608         if (dst) {
2609                 if (dst->obsolete) {
2610                         dst_release(dst);
2611                         dst = NULL;
2612                 }
2613         }
2614         return dst;
2615 }
2616
2617 static void xfrm_init_pmtu(struct dst_entry *dst)
2618 {
2619         do {
2620                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2621                 u32 pmtu, route_mtu_cached;
2622
2623                 pmtu = dst_mtu(dst->child);
2624                 xdst->child_mtu_cached = pmtu;
2625
2626                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2627
2628                 route_mtu_cached = dst_mtu(xdst->route);
2629                 xdst->route_mtu_cached = route_mtu_cached;
2630
2631                 if (pmtu > route_mtu_cached)
2632                         pmtu = route_mtu_cached;
2633
2634                 dst_metric_set(dst, RTAX_MTU, pmtu);
2635         } while ((dst = dst->next));
2636 }
2637
2638 /* Check that the bundle accepts the flow and its components are
2639  * still valid.
2640  */
2641
2642 static int xfrm_bundle_ok(struct xfrm_dst *first)
2643 {
2644         struct dst_entry *dst = &first->u.dst;
2645         struct xfrm_dst *last;
2646         u32 mtu;
2647
2648         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2649             (dst->dev && !netif_running(dst->dev)))
2650                 return 0;
2651
2652         if (dst->flags & DST_XFRM_QUEUE)
2653                 return 1;
2654
2655         last = NULL;
2656
2657         do {
2658                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2659
2660                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2661                         return 0;
2662                 if (xdst->xfrm_genid != dst->xfrm->genid)
2663                         return 0;
2664                 if (xdst->num_pols > 0 &&
2665                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2666                         return 0;
2667
2668                 mtu = dst_mtu(dst->child);
2669                 if (xdst->child_mtu_cached != mtu) {
2670                         last = xdst;
2671                         xdst->child_mtu_cached = mtu;
2672                 }
2673
2674                 if (!dst_check(xdst->route, xdst->route_cookie))
2675                         return 0;
2676                 mtu = dst_mtu(xdst->route);
2677                 if (xdst->route_mtu_cached != mtu) {
2678                         last = xdst;
2679                         xdst->route_mtu_cached = mtu;
2680                 }
2681
2682                 dst = dst->child;
2683         } while (dst->xfrm);
2684
2685         if (likely(!last))
2686                 return 1;
2687
2688         mtu = last->child_mtu_cached;
2689         for (;;) {
2690                 dst = &last->u.dst;
2691
2692                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2693                 if (mtu > last->route_mtu_cached)
2694                         mtu = last->route_mtu_cached;
2695                 dst_metric_set(dst, RTAX_MTU, mtu);
2696
2697                 if (last == first)
2698                         break;
2699
2700                 last = (struct xfrm_dst *)last->u.dst.next;
2701                 last->child_mtu_cached = mtu;
2702         }
2703
2704         return 1;
2705 }
2706
2707 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2708 {
2709         return dst_metric_advmss(dst->path);
2710 }
2711
2712 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2713 {
2714         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2715
2716         return mtu ? : dst_mtu(dst->path);
2717 }
2718
2719 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
2720                                         const void *daddr)
2721 {
2722         const struct dst_entry *path = dst->path;
2723
2724         for (; dst != path; dst = dst->child) {
2725                 const struct xfrm_state *xfrm = dst->xfrm;
2726
2727                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
2728                         continue;
2729                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
2730                         daddr = xfrm->coaddr;
2731                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
2732                         daddr = &xfrm->id.daddr;
2733         }
2734         return daddr;
2735 }
2736
2737 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2738                                            struct sk_buff *skb,
2739                                            const void *daddr)
2740 {
2741         const struct dst_entry *path = dst->path;
2742
2743         if (!skb)
2744                 daddr = xfrm_get_dst_nexthop(dst, daddr);
2745         return path->ops->neigh_lookup(path, skb, daddr);
2746 }
2747
2748 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
2749 {
2750         const struct dst_entry *path = dst->path;
2751
2752         daddr = xfrm_get_dst_nexthop(dst, daddr);
2753         path->ops->confirm_neigh(path, daddr);
2754 }
2755
2756 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
2757 {
2758         int err = 0;
2759
2760         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
2761                 return -EAFNOSUPPORT;
2762
2763         spin_lock(&xfrm_policy_afinfo_lock);
2764         if (unlikely(xfrm_policy_afinfo[family] != NULL))
2765                 err = -EEXIST;
2766         else {
2767                 struct dst_ops *dst_ops = afinfo->dst_ops;
2768                 if (likely(dst_ops->kmem_cachep == NULL))
2769                         dst_ops->kmem_cachep = xfrm_dst_cache;
2770                 if (likely(dst_ops->check == NULL))
2771                         dst_ops->check = xfrm_dst_check;
2772                 if (likely(dst_ops->default_advmss == NULL))
2773                         dst_ops->default_advmss = xfrm_default_advmss;
2774                 if (likely(dst_ops->mtu == NULL))
2775                         dst_ops->mtu = xfrm_mtu;
2776                 if (likely(dst_ops->negative_advice == NULL))
2777                         dst_ops->negative_advice = xfrm_negative_advice;
2778                 if (likely(dst_ops->link_failure == NULL))
2779                         dst_ops->link_failure = xfrm_link_failure;
2780                 if (likely(dst_ops->neigh_lookup == NULL))
2781                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2782                 if (likely(!dst_ops->confirm_neigh))
2783                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
2784                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
2785         }
2786         spin_unlock(&xfrm_policy_afinfo_lock);
2787
2788         return err;
2789 }
2790 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2791
2792 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
2793 {
2794         struct dst_ops *dst_ops = afinfo->dst_ops;
2795         int i;
2796
2797         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
2798                 if (xfrm_policy_afinfo[i] != afinfo)
2799                         continue;
2800                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
2801                 break;
2802         }
2803
2804         synchronize_rcu();
2805
2806         dst_ops->kmem_cachep = NULL;
2807         dst_ops->check = NULL;
2808         dst_ops->negative_advice = NULL;
2809         dst_ops->link_failure = NULL;
2810 }
2811 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2812
2813 #ifdef CONFIG_XFRM_STATISTICS
2814 static int __net_init xfrm_statistics_init(struct net *net)
2815 {
2816         int rv;
2817         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2818         if (!net->mib.xfrm_statistics)
2819                 return -ENOMEM;
2820         rv = xfrm_proc_init(net);
2821         if (rv < 0)
2822                 free_percpu(net->mib.xfrm_statistics);
2823         return rv;
2824 }
2825
2826 static void xfrm_statistics_fini(struct net *net)
2827 {
2828         xfrm_proc_fini(net);
2829         free_percpu(net->mib.xfrm_statistics);
2830 }
2831 #else
2832 static int __net_init xfrm_statistics_init(struct net *net)
2833 {
2834         return 0;
2835 }
2836
2837 static void xfrm_statistics_fini(struct net *net)
2838 {
2839 }
2840 #endif
2841
2842 static int __net_init xfrm_policy_init(struct net *net)
2843 {
2844         unsigned int hmask, sz;
2845         int dir;
2846
2847         if (net_eq(net, &init_net))
2848                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2849                                            sizeof(struct xfrm_dst),
2850                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2851                                            NULL);
2852
2853         hmask = 8 - 1;
2854         sz = (hmask+1) * sizeof(struct hlist_head);
2855
2856         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2857         if (!net->xfrm.policy_byidx)
2858                 goto out_byidx;
2859         net->xfrm.policy_idx_hmask = hmask;
2860
2861         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2862                 struct xfrm_policy_hash *htab;
2863
2864                 net->xfrm.policy_count[dir] = 0;
2865                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2866                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2867
2868                 htab = &net->xfrm.policy_bydst[dir];
2869                 htab->table = xfrm_hash_alloc(sz);
2870                 if (!htab->table)
2871                         goto out_bydst;
2872                 htab->hmask = hmask;
2873                 htab->dbits4 = 32;
2874                 htab->sbits4 = 32;
2875                 htab->dbits6 = 128;
2876                 htab->sbits6 = 128;
2877         }
2878         net->xfrm.policy_hthresh.lbits4 = 32;
2879         net->xfrm.policy_hthresh.rbits4 = 32;
2880         net->xfrm.policy_hthresh.lbits6 = 128;
2881         net->xfrm.policy_hthresh.rbits6 = 128;
2882
2883         seqlock_init(&net->xfrm.policy_hthresh.lock);
2884
2885         INIT_LIST_HEAD(&net->xfrm.policy_all);
2886         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2887         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2888         if (net_eq(net, &init_net))
2889                 xfrm_dev_init();
2890         return 0;
2891
2892 out_bydst:
2893         for (dir--; dir >= 0; dir--) {
2894                 struct xfrm_policy_hash *htab;
2895
2896                 htab = &net->xfrm.policy_bydst[dir];
2897                 xfrm_hash_free(htab->table, sz);
2898         }
2899         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2900 out_byidx:
2901         return -ENOMEM;
2902 }
2903
2904 static void xfrm_policy_fini(struct net *net)
2905 {
2906         unsigned int sz;
2907         int dir;
2908
2909         flush_work(&net->xfrm.policy_hash_work);
2910 #ifdef CONFIG_XFRM_SUB_POLICY
2911         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2912 #endif
2913         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2914
2915         WARN_ON(!list_empty(&net->xfrm.policy_all));
2916
2917         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2918                 struct xfrm_policy_hash *htab;
2919
2920                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2921
2922                 htab = &net->xfrm.policy_bydst[dir];
2923                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2924                 WARN_ON(!hlist_empty(htab->table));
2925                 xfrm_hash_free(htab->table, sz);
2926         }
2927
2928         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2929         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2930         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2931 }
2932
2933 static int __net_init xfrm_net_init(struct net *net)
2934 {
2935         int rv;
2936
2937         /* Initialize the per-net locks here */
2938         spin_lock_init(&net->xfrm.xfrm_state_lock);
2939         spin_lock_init(&net->xfrm.xfrm_policy_lock);
2940         mutex_init(&net->xfrm.xfrm_cfg_mutex);
2941
2942         rv = xfrm_statistics_init(net);
2943         if (rv < 0)
2944                 goto out_statistics;
2945         rv = xfrm_state_init(net);
2946         if (rv < 0)
2947                 goto out_state;
2948         rv = xfrm_policy_init(net);
2949         if (rv < 0)
2950                 goto out_policy;
2951         rv = xfrm_sysctl_init(net);
2952         if (rv < 0)
2953                 goto out_sysctl;
2954
2955         return 0;
2956
2957 out_sysctl:
2958         xfrm_policy_fini(net);
2959 out_policy:
2960         xfrm_state_fini(net);
2961 out_state:
2962         xfrm_statistics_fini(net);
2963 out_statistics:
2964         return rv;
2965 }
2966
2967 static void __net_exit xfrm_net_exit(struct net *net)
2968 {
2969         xfrm_sysctl_fini(net);
2970         xfrm_policy_fini(net);
2971         xfrm_state_fini(net);
2972         xfrm_statistics_fini(net);
2973 }
2974
2975 static struct pernet_operations __net_initdata xfrm_net_ops = {
2976         .init = xfrm_net_init,
2977         .exit = xfrm_net_exit,
2978 };
2979
2980 void __init xfrm_init(void)
2981 {
2982         int i;
2983
2984         xfrm_pcpu_work = kmalloc_array(NR_CPUS, sizeof(*xfrm_pcpu_work),
2985                                        GFP_KERNEL);
2986         BUG_ON(!xfrm_pcpu_work);
2987
2988         for (i = 0; i < NR_CPUS; i++)
2989                 INIT_WORK(&xfrm_pcpu_work[i], xfrm_pcpu_work_fn);
2990
2991         register_pernet_subsys(&xfrm_net_ops);
2992         seqcount_init(&xfrm_policy_hash_generation);
2993         xfrm_input_init();
2994 }
2995
2996 #ifdef CONFIG_AUDITSYSCALL
2997 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2998                                          struct audit_buffer *audit_buf)
2999 {
3000         struct xfrm_sec_ctx *ctx = xp->security;
3001         struct xfrm_selector *sel = &xp->selector;
3002
3003         if (ctx)
3004                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3005                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3006
3007         switch (sel->family) {
3008         case AF_INET:
3009                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3010                 if (sel->prefixlen_s != 32)
3011                         audit_log_format(audit_buf, " src_prefixlen=%d",
3012                                          sel->prefixlen_s);
3013                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3014                 if (sel->prefixlen_d != 32)
3015                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3016                                          sel->prefixlen_d);
3017                 break;
3018         case AF_INET6:
3019                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3020                 if (sel->prefixlen_s != 128)
3021                         audit_log_format(audit_buf, " src_prefixlen=%d",
3022                                          sel->prefixlen_s);
3023                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3024                 if (sel->prefixlen_d != 128)
3025                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3026                                          sel->prefixlen_d);
3027                 break;
3028         }
3029 }
3030
3031 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3032 {
3033         struct audit_buffer *audit_buf;
3034
3035         audit_buf = xfrm_audit_start("SPD-add");
3036         if (audit_buf == NULL)
3037                 return;
3038         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3039         audit_log_format(audit_buf, " res=%u", result);
3040         xfrm_audit_common_policyinfo(xp, audit_buf);
3041         audit_log_end(audit_buf);
3042 }
3043 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3044
3045 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3046                               bool task_valid)
3047 {
3048         struct audit_buffer *audit_buf;
3049
3050         audit_buf = xfrm_audit_start("SPD-delete");
3051         if (audit_buf == NULL)
3052                 return;
3053         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3054         audit_log_format(audit_buf, " res=%u", result);
3055         xfrm_audit_common_policyinfo(xp, audit_buf);
3056         audit_log_end(audit_buf);
3057 }
3058 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3059 #endif
3060
3061 #ifdef CONFIG_XFRM_MIGRATE
3062 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3063                                         const struct xfrm_selector *sel_tgt)
3064 {
3065         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3066                 if (sel_tgt->family == sel_cmp->family &&
3067                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3068                                     sel_cmp->family) &&
3069                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3070                                     sel_cmp->family) &&
3071                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3072                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3073                         return true;
3074                 }
3075         } else {
3076                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3077                         return true;
3078                 }
3079         }
3080         return false;
3081 }
3082
3083 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3084                                                     u8 dir, u8 type, struct net *net)
3085 {
3086         struct xfrm_policy *pol, *ret = NULL;
3087         struct hlist_head *chain;
3088         u32 priority = ~0U;
3089
3090         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3091         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3092         hlist_for_each_entry(pol, chain, bydst) {
3093                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3094                     pol->type == type) {
3095                         ret = pol;
3096                         priority = ret->priority;
3097                         break;
3098                 }
3099         }
3100         chain = &net->xfrm.policy_inexact[dir];
3101         hlist_for_each_entry(pol, chain, bydst) {
3102                 if ((pol->priority >= priority) && ret)
3103                         break;
3104
3105                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3106                     pol->type == type) {
3107                         ret = pol;
3108                         break;
3109                 }
3110         }
3111
3112         xfrm_pol_hold(ret);
3113
3114         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3115
3116         return ret;
3117 }
3118
3119 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3120 {
3121         int match = 0;
3122
3123         if (t->mode == m->mode && t->id.proto == m->proto &&
3124             (m->reqid == 0 || t->reqid == m->reqid)) {
3125                 switch (t->mode) {
3126                 case XFRM_MODE_TUNNEL:
3127                 case XFRM_MODE_BEET:
3128                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3129                                             m->old_family) &&
3130                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3131                                             m->old_family)) {
3132                                 match = 1;
3133                         }
3134                         break;
3135                 case XFRM_MODE_TRANSPORT:
3136                         /* in case of transport mode, template does not store
3137                            any IP addresses, hence we just compare mode and
3138                            protocol */
3139                         match = 1;
3140                         break;
3141                 default:
3142                         break;
3143                 }
3144         }
3145         return match;
3146 }
3147
3148 /* update endpoint address(es) of template(s) */
3149 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3150                                struct xfrm_migrate *m, int num_migrate)
3151 {
3152         struct xfrm_migrate *mp;
3153         int i, j, n = 0;
3154
3155         write_lock_bh(&pol->lock);
3156         if (unlikely(pol->walk.dead)) {
3157                 /* target policy has been deleted */
3158                 write_unlock_bh(&pol->lock);
3159                 return -ENOENT;
3160         }
3161
3162         for (i = 0; i < pol->xfrm_nr; i++) {
3163                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3164                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3165                                 continue;
3166                         n++;
3167                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3168                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3169                                 continue;
3170                         /* update endpoints */
3171                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3172                                sizeof(pol->xfrm_vec[i].id.daddr));
3173                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3174                                sizeof(pol->xfrm_vec[i].saddr));
3175                         pol->xfrm_vec[i].encap_family = mp->new_family;
3176                         /* flush bundles */
3177                         atomic_inc(&pol->genid);
3178                 }
3179         }
3180
3181         write_unlock_bh(&pol->lock);
3182
3183         if (!n)
3184                 return -ENODATA;
3185
3186         return 0;
3187 }
3188
3189 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3190 {
3191         int i, j;
3192
3193         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3194                 return -EINVAL;
3195
3196         for (i = 0; i < num_migrate; i++) {
3197                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3198                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3199                         return -EINVAL;
3200
3201                 /* check if there is any duplicated entry */
3202                 for (j = i + 1; j < num_migrate; j++) {
3203                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3204                                     sizeof(m[i].old_daddr)) &&
3205                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3206                                     sizeof(m[i].old_saddr)) &&
3207                             m[i].proto == m[j].proto &&
3208                             m[i].mode == m[j].mode &&
3209                             m[i].reqid == m[j].reqid &&
3210                             m[i].old_family == m[j].old_family)
3211                                 return -EINVAL;
3212                 }
3213         }
3214
3215         return 0;
3216 }
3217
3218 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3219                  struct xfrm_migrate *m, int num_migrate,
3220                  struct xfrm_kmaddress *k, struct net *net,
3221                  struct xfrm_encap_tmpl *encap)
3222 {
3223         int i, err, nx_cur = 0, nx_new = 0;
3224         struct xfrm_policy *pol = NULL;
3225         struct xfrm_state *x, *xc;
3226         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3227         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3228         struct xfrm_migrate *mp;
3229
3230         /* Stage 0 - sanity checks */
3231         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3232                 goto out;
3233
3234         if (dir >= XFRM_POLICY_MAX) {
3235                 err = -EINVAL;
3236                 goto out;
3237         }
3238
3239         /* Stage 1 - find policy */
3240         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3241                 err = -ENOENT;
3242                 goto out;
3243         }
3244
3245         /* Stage 2 - find and update state(s) */
3246         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3247                 if ((x = xfrm_migrate_state_find(mp, net))) {
3248                         x_cur[nx_cur] = x;
3249                         nx_cur++;
3250                         xc = xfrm_state_migrate(x, mp, encap);
3251                         if (xc) {
3252                                 x_new[nx_new] = xc;
3253                                 nx_new++;
3254                         } else {
3255                                 err = -ENODATA;
3256                                 goto restore_state;
3257                         }
3258                 }
3259         }
3260
3261         /* Stage 3 - update policy */
3262         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3263                 goto restore_state;
3264
3265         /* Stage 4 - delete old state(s) */
3266         if (nx_cur) {
3267                 xfrm_states_put(x_cur, nx_cur);
3268                 xfrm_states_delete(x_cur, nx_cur);
3269         }
3270
3271         /* Stage 5 - announce */
3272         km_migrate(sel, dir, type, m, num_migrate, k, encap);
3273
3274         xfrm_pol_put(pol);
3275
3276         return 0;
3277 out:
3278         return err;
3279
3280 restore_state:
3281         if (pol)
3282                 xfrm_pol_put(pol);
3283         if (nx_cur)
3284                 xfrm_states_put(x_cur, nx_cur);
3285         if (nx_new)
3286                 xfrm_states_delete(x_new, nx_new);
3287
3288         return err;
3289 }
3290 EXPORT_SYMBOL(xfrm_migrate);
3291 #endif