net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/flow.h>
  30 #include <net/xfrm.h>
  31 #include <net/ip.h>
  32 #ifdef CONFIG_XFRM_STATISTICS
  33 #include <net/snmp.h>
  34 #endif
  35
  36 #include "xfrm_hash.h"
  37
  38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  40 #define XFRM_MAX_QUEUE_LEN      100
  41
  42 struct xfrm_flo {
  43         struct dst_entry *dst_orig;
  44         u8 flags;
  45 };
  46
  47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  48 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
  49                                                 __read_mostly;
  50
  51 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  52
  53 static void xfrm_init_pmtu(struct dst_entry *dst);
  54 static int stale_bundle(struct dst_entry *dst);
  55 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  56 static void xfrm_policy_queue_process(unsigned long arg);
  57
  58 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
  59 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  60                                                 int dir);
  61
  62 static inline bool
  63 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  64 {
  65         const struct flowi4 *fl4 = &fl->u.ip4;
  66
  67         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  68                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  69                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  70                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  71                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  72                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  73 }
  74
  75 static inline bool
  76 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  77 {
  78         const struct flowi6 *fl6 = &fl->u.ip6;
  79
  80         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  81                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  82                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  83                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  84                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  85                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  86 }
  87
  88 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  89                          unsigned short family)
  90 {
  91         switch (family) {
  92         case AF_INET:
  93                 return __xfrm4_selector_match(sel, fl);
  94         case AF_INET6:
  95                 return __xfrm6_selector_match(sel, fl);
  96         }
  97         return false;
  98 }
  99
 100 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 101 {
 102         struct xfrm_policy_afinfo *afinfo;
 103
 104         if (unlikely(family >= NPROTO))
 105                 return NULL;
 106         rcu_read_lock();
 107         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 108         if (unlikely(!afinfo))
 109                 rcu_read_unlock();
 110         return afinfo;
 111 }
 112
 113 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
 114 {
 115         rcu_read_unlock();
 116 }
 117
 118 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net,
 119                                                   int tos, int oif,
 120                                                   const xfrm_address_t *saddr,
 121                                                   const xfrm_address_t *daddr,
 122                                                   int family)
 123 {
 124         struct xfrm_policy_afinfo *afinfo;
 125         struct dst_entry *dst;
 126
 127         afinfo = xfrm_policy_get_afinfo(family);
 128         if (unlikely(afinfo == NULL))
 129                 return ERR_PTR(-EAFNOSUPPORT);
 130
 131         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr);
 132
 133         xfrm_policy_put_afinfo(afinfo);
 134
 135         return dst;
 136 }
 137
 138 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 139                                                 int tos, int oif,
 140                                                 xfrm_address_t *prev_saddr,
 141                                                 xfrm_address_t *prev_daddr,
 142                                                 int family)
 143 {
 144         struct net *net = xs_net(x);
 145         xfrm_address_t *saddr = &x->props.saddr;
 146         xfrm_address_t *daddr = &x->id.daddr;
 147         struct dst_entry *dst;
 148
 149         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 150                 saddr = x->coaddr;
 151                 daddr = prev_daddr;
 152         }
 153         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 154                 saddr = prev_saddr;
 155                 daddr = x->coaddr;
 156         }
 157
 158         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family);
 159
 160         if (!IS_ERR(dst)) {
 161                 if (prev_saddr != saddr)
 162                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 163                 if (prev_daddr != daddr)
 164                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 165         }
 166
 167         return dst;
 168 }
 169
 170 static inline unsigned long make_jiffies(long secs)
 171 {
 172         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 173                 return MAX_SCHEDULE_TIMEOUT-1;
 174         else
 175                 return secs*HZ;
 176 }
 177
 178 static void xfrm_policy_timer(unsigned long data)
 179 {
 180         struct xfrm_policy *xp = (struct xfrm_policy *)data;
 181         unsigned long now = get_seconds();
 182         long next = LONG_MAX;
 183         int warn = 0;
 184         int dir;
 185
 186         read_lock(&xp->lock);
 187
 188         if (unlikely(xp->walk.dead))
 189                 goto out;
 190
 191         dir = xfrm_policy_id2dir(xp->index);
 192
 193         if (xp->lft.hard_add_expires_seconds) {
 194                 long tmo = xp->lft.hard_add_expires_seconds +
 195                         xp->curlft.add_time - now;
 196                 if (tmo <= 0)
 197                         goto expired;
 198                 if (tmo < next)
 199                         next = tmo;
 200         }
 201         if (xp->lft.hard_use_expires_seconds) {
 202                 long tmo = xp->lft.hard_use_expires_seconds +
 203                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 204                 if (tmo <= 0)
 205                         goto expired;
 206                 if (tmo < next)
 207                         next = tmo;
 208         }
 209         if (xp->lft.soft_add_expires_seconds) {
 210                 long tmo = xp->lft.soft_add_expires_seconds +
 211                         xp->curlft.add_time - now;
 212                 if (tmo <= 0) {
 213                         warn = 1;
 214                         tmo = XFRM_KM_TIMEOUT;
 215                 }
 216                 if (tmo < next)
 217                         next = tmo;
 218         }
 219         if (xp->lft.soft_use_expires_seconds) {
 220                 long tmo = xp->lft.soft_use_expires_seconds +
 221                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 222                 if (tmo <= 0) {
 223                         warn = 1;
 224                         tmo = XFRM_KM_TIMEOUT;
 225                 }
 226                 if (tmo < next)
 227                         next = tmo;
 228         }
 229
 230         if (warn)
 231                 km_policy_expired(xp, dir, 0, 0);
 232         if (next != LONG_MAX &&
 233             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 234                 xfrm_pol_hold(xp);
 235
 236 out:
 237         read_unlock(&xp->lock);
 238         xfrm_pol_put(xp);
 239         return;
 240
 241 expired:
 242         read_unlock(&xp->lock);
 243         if (!xfrm_policy_delete(xp, dir))
 244                 km_policy_expired(xp, dir, 1, 0);
 245         xfrm_pol_put(xp);
 246 }
 247
 248 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 249 {
 250         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 251
 252         if (unlikely(pol->walk.dead))
 253                 flo = NULL;
 254         else
 255                 xfrm_pol_hold(pol);
 256
 257         return flo;
 258 }
 259
 260 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 261 {
 262         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 263
 264         return !pol->walk.dead;
 265 }
 266
 267 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 268 {
 269         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 270 }
 271
 272 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 273         .get = xfrm_policy_flo_get,
 274         .check = xfrm_policy_flo_check,
 275         .delete = xfrm_policy_flo_delete,
 276 };
 277
 278 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 279  * SPD calls.
 280  */
 281
 282 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 283 {
 284         struct xfrm_policy *policy;
 285
 286         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 287
 288         if (policy) {
 289                 write_pnet(&policy->xp_net, net);
 290                 INIT_LIST_HEAD(&policy->walk.all);
 291                 INIT_HLIST_NODE(&policy->bydst);
 292                 INIT_HLIST_NODE(&policy->byidx);
 293                 rwlock_init(&policy->lock);
 294                 atomic_set(&policy->refcnt, 1);
 295                 skb_queue_head_init(&policy->polq.hold_queue);
 296                 setup_timer(&policy->timer, xfrm_policy_timer,
 297                                 (unsigned long)policy);
 298                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 299                             (unsigned long)policy);
 300                 policy->flo.ops = &xfrm_policy_fc_ops;
 301         }
 302         return policy;
 303 }
 304 EXPORT_SYMBOL(xfrm_policy_alloc);
 305
 306 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 307 {
 308         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 309
 310         security_xfrm_policy_free(policy->security);
 311         kfree(policy);
 312 }
 313
 314 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 315
 316 void xfrm_policy_destroy(struct xfrm_policy *policy)
 317 {
 318         BUG_ON(!policy->walk.dead);
 319
 320         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 321                 BUG();
 322
 323         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 324 }
 325 EXPORT_SYMBOL(xfrm_policy_destroy);
 326
 327 /* Rule must be locked. Release descentant resources, announce
 328  * entry dead. The rule must be unlinked from lists to the moment.
 329  */
 330
 331 static void xfrm_policy_kill(struct xfrm_policy *policy)
 332 {
 333         policy->walk.dead = 1;
 334
 335         atomic_inc(&policy->genid);
 336
 337         if (del_timer(&policy->polq.hold_timer))
 338                 xfrm_pol_put(policy);
 339         skb_queue_purge(&policy->polq.hold_queue);
 340
 341         if (del_timer(&policy->timer))
 342                 xfrm_pol_put(policy);
 343
 344         xfrm_pol_put(policy);
 345 }
 346
 347 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 348
 349 static inline unsigned int idx_hash(struct net *net, u32 index)
 350 {
 351         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 352 }
 353
 354 /* calculate policy hash thresholds */
 355 static void __get_hash_thresh(struct net *net,
 356                               unsigned short family, int dir,
 357                               u8 *dbits, u8 *sbits)
 358 {
 359         switch (family) {
 360         case AF_INET:
 361                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 362                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 363                 break;
 364
 365         case AF_INET6:
 366                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 367                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 368                 break;
 369
 370         default:
 371                 *dbits = 0;
 372                 *sbits = 0;
 373         }
 374 }
 375
 376 static struct hlist_head *policy_hash_bysel(struct net *net,
 377                                             const struct xfrm_selector *sel,
 378                                             unsigned short family, int dir)
 379 {
 380         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 381         unsigned int hash;
 382         u8 dbits;
 383         u8 sbits;
 384
 385         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 386         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 387
 388         return (hash == hmask + 1 ?
 389                 &net->xfrm.policy_inexact[dir] :
 390                 net->xfrm.policy_bydst[dir].table + hash);
 391 }
 392
 393 static struct hlist_head *policy_hash_direct(struct net *net,
 394                                              const xfrm_address_t *daddr,
 395                                              const xfrm_address_t *saddr,
 396                                              unsigned short family, int dir)
 397 {
 398         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 399         unsigned int hash;
 400         u8 dbits;
 401         u8 sbits;
 402
 403         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 404         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 405
 406         return net->xfrm.policy_bydst[dir].table + hash;
 407 }
 408
 409 static void xfrm_dst_hash_transfer(struct net *net,
 410                                    struct hlist_head *list,
 411                                    struct hlist_head *ndsttable,
 412                                    unsigned int nhashmask,
 413                                    int dir)
 414 {
 415         struct hlist_node *tmp, *entry0 = NULL;
 416         struct xfrm_policy *pol;
 417         unsigned int h0 = 0;
 418         u8 dbits;
 419         u8 sbits;
 420
 421 redo:
 422         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 423                 unsigned int h;
 424
 425                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 426                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 427                                 pol->family, nhashmask, dbits, sbits);
 428                 if (!entry0) {
 429                         hlist_del(&pol->bydst);
 430                         hlist_add_head(&pol->bydst, ndsttable+h);
 431                         h0 = h;
 432                 } else {
 433                         if (h != h0)
 434                                 continue;
 435                         hlist_del(&pol->bydst);
 436                         hlist_add_behind(&pol->bydst, entry0);
 437                 }
 438                 entry0 = &pol->bydst;
 439         }
 440         if (!hlist_empty(list)) {
 441                 entry0 = NULL;
 442                 goto redo;
 443         }
 444 }
 445
 446 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 447                                    struct hlist_head *nidxtable,
 448                                    unsigned int nhashmask)
 449 {
 450         struct hlist_node *tmp;
 451         struct xfrm_policy *pol;
 452
 453         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 454                 unsigned int h;
 455
 456                 h = __idx_hash(pol->index, nhashmask);
 457                 hlist_add_head(&pol->byidx, nidxtable+h);
 458         }
 459 }
 460
 461 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 462 {
 463         return ((old_hmask + 1) << 1) - 1;
 464 }
 465
 466 static void xfrm_bydst_resize(struct net *net, int dir)
 467 {
 468         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 469         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 470         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 471         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 472         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 473         int i;
 474
 475         if (!ndst)
 476                 return;
 477
 478         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 479
 480         for (i = hmask; i >= 0; i--)
 481                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 482
 483         net->xfrm.policy_bydst[dir].table = ndst;
 484         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 485
 486         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 487
 488         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 489 }
 490
 491 static void xfrm_byidx_resize(struct net *net, int total)
 492 {
 493         unsigned int hmask = net->xfrm.policy_idx_hmask;
 494         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 495         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 496         struct hlist_head *oidx = net->xfrm.policy_byidx;
 497         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 498         int i;
 499
 500         if (!nidx)
 501                 return;
 502
 503         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 504
 505         for (i = hmask; i >= 0; i--)
 506                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 507
 508         net->xfrm.policy_byidx = nidx;
 509         net->xfrm.policy_idx_hmask = nhashmask;
 510
 511         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 512
 513         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 514 }
 515
 516 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 517 {
 518         unsigned int cnt = net->xfrm.policy_count[dir];
 519         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 520
 521         if (total)
 522                 *total += cnt;
 523
 524         if ((hmask + 1) < xfrm_policy_hashmax &&
 525             cnt > hmask)
 526                 return 1;
 527
 528         return 0;
 529 }
 530
 531 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 532 {
 533         unsigned int hmask = net->xfrm.policy_idx_hmask;
 534
 535         if ((hmask + 1) < xfrm_policy_hashmax &&
 536             total > hmask)
 537                 return 1;
 538
 539         return 0;
 540 }
 541
 542 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 543 {
 544         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 545         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 546         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 547         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 548         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 549         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 550         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 551         si->spdhcnt = net->xfrm.policy_idx_hmask;
 552         si->spdhmcnt = xfrm_policy_hashmax;
 553         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
 554 }
 555 EXPORT_SYMBOL(xfrm_spd_getinfo);
 556
 557 static DEFINE_MUTEX(hash_resize_mutex);
 558 static void xfrm_hash_resize(struct work_struct *work)
 559 {
 560         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 561         int dir, total;
 562
 563         mutex_lock(&hash_resize_mutex);
 564
 565         total = 0;
 566         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 567                 if (xfrm_bydst_should_resize(net, dir, &total))
 568                         xfrm_bydst_resize(net, dir);
 569         }
 570         if (xfrm_byidx_should_resize(net, total))
 571                 xfrm_byidx_resize(net, total);
 572
 573         mutex_unlock(&hash_resize_mutex);
 574 }
 575
 576 static void xfrm_hash_rebuild(struct work_struct *work)
 577 {
 578         struct net *net = container_of(work, struct net,
 579                                        xfrm.policy_hthresh.work);
 580         unsigned int hmask;
 581         struct xfrm_policy *pol;
 582         struct xfrm_policy *policy;
 583         struct hlist_head *chain;
 584         struct hlist_head *odst;
 585         struct hlist_node *newpos;
 586         int i;
 587         int dir;
 588         unsigned seq;
 589         u8 lbits4, rbits4, lbits6, rbits6;
 590
 591         mutex_lock(&hash_resize_mutex);
 592
 593         /* read selector prefixlen thresholds */
 594         do {
 595                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
 596
 597                 lbits4 = net->xfrm.policy_hthresh.lbits4;
 598                 rbits4 = net->xfrm.policy_hthresh.rbits4;
 599                 lbits6 = net->xfrm.policy_hthresh.lbits6;
 600                 rbits6 = net->xfrm.policy_hthresh.rbits6;
 601         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
 602
 603         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 604
 605         /* reset the bydst and inexact table in all directions */
 606         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 607                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
 608                 hmask = net->xfrm.policy_bydst[dir].hmask;
 609                 odst = net->xfrm.policy_bydst[dir].table;
 610                 for (i = hmask; i >= 0; i--)
 611                         INIT_HLIST_HEAD(odst + i);
 612                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
 613                         /* dir out => dst = remote, src = local */
 614                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
 615                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
 616                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
 617                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
 618                 } else {
 619                         /* dir in/fwd => dst = local, src = remote */
 620                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
 621                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
 622                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
 623                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
 624                 }
 625         }
 626
 627         /* re-insert all policies by order of creation */
 628         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 629                 if (xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) {
 630                         /* skip socket policies */
 631                         continue;
 632                 }
 633                 newpos = NULL;
 634                 chain = policy_hash_bysel(net, &policy->selector,
 635                                           policy->family,
 636                                           xfrm_policy_id2dir(policy->index));
 637                 hlist_for_each_entry(pol, chain, bydst) {
 638                         if (policy->priority >= pol->priority)
 639                                 newpos = &pol->bydst;
 640                         else
 641                                 break;
 642                 }
 643                 if (newpos)
 644                         hlist_add_behind(&policy->bydst, newpos);
 645                 else
 646                         hlist_add_head(&policy->bydst, chain);
 647         }
 648
 649         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 650
 651         mutex_unlock(&hash_resize_mutex);
 652 }
 653
 654 void xfrm_policy_hash_rebuild(struct net *net)
 655 {
 656         schedule_work(&net->xfrm.policy_hthresh.work);
 657 }
 658 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 659
 660 /* Generate new index... KAME seems to generate them ordered by cost
 661  * of an absolute inpredictability of ordering of rules. This will not pass. */
 662 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 663 {
 664         static u32 idx_generator;
 665
 666         for (;;) {
 667                 struct hlist_head *list;
 668                 struct xfrm_policy *p;
 669                 u32 idx;
 670                 int found;
 671
 672                 if (!index) {
 673                         idx = (idx_generator | dir);
 674                         idx_generator += 8;
 675                 } else {
 676                         idx = index;
 677                         index = 0;
 678                 }
 679
 680                 if (idx == 0)
 681                         idx = 8;
 682                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 683                 found = 0;
 684                 hlist_for_each_entry(p, list, byidx) {
 685                         if (p->index == idx) {
 686                                 found = 1;
 687                                 break;
 688                         }
 689                 }
 690                 if (!found)
 691                         return idx;
 692         }
 693 }
 694
 695 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 696 {
 697         u32 *p1 = (u32 *) s1;
 698         u32 *p2 = (u32 *) s2;
 699         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 700         int i;
 701
 702         for (i = 0; i < len; i++) {
 703                 if (p1[i] != p2[i])
 704                         return 1;
 705         }
 706
 707         return 0;
 708 }
 709
 710 static void xfrm_policy_requeue(struct xfrm_policy *old,
 711                                 struct xfrm_policy *new)
 712 {
 713         struct xfrm_policy_queue *pq = &old->polq;
 714         struct sk_buff_head list;
 715
 716         if (skb_queue_empty(&pq->hold_queue))
 717                 return;
 718
 719         __skb_queue_head_init(&list);
 720
 721         spin_lock_bh(&pq->hold_queue.lock);
 722         skb_queue_splice_init(&pq->hold_queue, &list);
 723         if (del_timer(&pq->hold_timer))
 724                 xfrm_pol_put(old);
 725         spin_unlock_bh(&pq->hold_queue.lock);
 726
 727         pq = &new->polq;
 728
 729         spin_lock_bh(&pq->hold_queue.lock);
 730         skb_queue_splice(&list, &pq->hold_queue);
 731         pq->timeout = XFRM_QUEUE_TMO_MIN;
 732         if (!mod_timer(&pq->hold_timer, jiffies))
 733                 xfrm_pol_hold(new);
 734         spin_unlock_bh(&pq->hold_queue.lock);
 735 }
 736
 737 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 738                                    struct xfrm_policy *pol)
 739 {
 740         u32 mark = policy->mark.v & policy->mark.m;
 741
 742         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 743                 return true;
 744
 745         if ((mark & pol->mark.m) == pol->mark.v &&
 746             policy->priority == pol->priority)
 747                 return true;
 748
 749         return false;
 750 }
 751
 752 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 753 {
 754         struct net *net = xp_net(policy);
 755         struct xfrm_policy *pol;
 756         struct xfrm_policy *delpol;
 757         struct hlist_head *chain;
 758         struct hlist_node *newpos;
 759
 760         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 761         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 762         delpol = NULL;
 763         newpos = NULL;
 764         hlist_for_each_entry(pol, chain, bydst) {
 765                 if (pol->type == policy->type &&
 766                     !selector_cmp(&pol->selector, &policy->selector) &&
 767                     xfrm_policy_mark_match(policy, pol) &&
 768                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 769                     !WARN_ON(delpol)) {
 770                         if (excl) {
 771                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 772                                 return -EEXIST;
 773                         }
 774                         delpol = pol;
 775                         if (policy->priority > pol->priority)
 776                                 continue;
 777                 } else if (policy->priority >= pol->priority) {
 778                         newpos = &pol->bydst;
 779                         continue;
 780                 }
 781                 if (delpol)
 782                         break;
 783         }
 784         if (newpos)
 785                 hlist_add_behind(&policy->bydst, newpos);
 786         else
 787                 hlist_add_head(&policy->bydst, chain);
 788         __xfrm_policy_link(policy, dir);
 789         atomic_inc(&net->xfrm.flow_cache_genid);
 790
 791         /* After previous checking, family can either be AF_INET or AF_INET6 */
 792         if (policy->family == AF_INET)
 793                 rt_genid_bump_ipv4(net);
 794         else
 795                 rt_genid_bump_ipv6(net);
 796
 797         if (delpol) {
 798                 xfrm_policy_requeue(delpol, policy);
 799                 __xfrm_policy_unlink(delpol, dir);
 800         }
 801         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 802         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 803         policy->curlft.add_time = get_seconds();
 804         policy->curlft.use_time = 0;
 805         if (!mod_timer(&policy->timer, jiffies + HZ))
 806                 xfrm_pol_hold(policy);
 807         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 808
 809         if (delpol)
 810                 xfrm_policy_kill(delpol);
 811         else if (xfrm_bydst_should_resize(net, dir, NULL))
 812                 schedule_work(&net->xfrm.policy_hash_work);
 813
 814         return 0;
 815 }
 816 EXPORT_SYMBOL(xfrm_policy_insert);
 817
 818 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 819                                           int dir, struct xfrm_selector *sel,
 820                                           struct xfrm_sec_ctx *ctx, int delete,
 821                                           int *err)
 822 {
 823         struct xfrm_policy *pol, *ret;
 824         struct hlist_head *chain;
 825
 826         *err = 0;
 827         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 828         chain = policy_hash_bysel(net, sel, sel->family, dir);
 829         ret = NULL;
 830         hlist_for_each_entry(pol, chain, bydst) {
 831                 if (pol->type == type &&
 832                     (mark & pol->mark.m) == pol->mark.v &&
 833                     !selector_cmp(sel, &pol->selector) &&
 834                     xfrm_sec_ctx_match(ctx, pol->security)) {
 835                         xfrm_pol_hold(pol);
 836                         if (delete) {
 837                                 *err = security_xfrm_policy_delete(
 838                                                                 pol->security);
 839                                 if (*err) {
 840                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 841                                         return pol;
 842                                 }
 843                                 __xfrm_policy_unlink(pol, dir);
 844                         }
 845                         ret = pol;
 846                         break;
 847                 }
 848         }
 849         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 850
 851         if (ret && delete)
 852                 xfrm_policy_kill(ret);
 853         return ret;
 854 }
 855 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 856
 857 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 858                                      int dir, u32 id, int delete, int *err)
 859 {
 860         struct xfrm_policy *pol, *ret;
 861         struct hlist_head *chain;
 862
 863         *err = -ENOENT;
 864         if (xfrm_policy_id2dir(id) != dir)
 865                 return NULL;
 866
 867         *err = 0;
 868         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 869         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 870         ret = NULL;
 871         hlist_for_each_entry(pol, chain, byidx) {
 872                 if (pol->type == type && pol->index == id &&
 873                     (mark & pol->mark.m) == pol->mark.v) {
 874                         xfrm_pol_hold(pol);
 875                         if (delete) {
 876                                 *err = security_xfrm_policy_delete(
 877                                                                 pol->security);
 878                                 if (*err) {
 879                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 880                                         return pol;
 881                                 }
 882                                 __xfrm_policy_unlink(pol, dir);
 883                         }
 884                         ret = pol;
 885                         break;
 886                 }
 887         }
 888         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 889
 890         if (ret && delete)
 891                 xfrm_policy_kill(ret);
 892         return ret;
 893 }
 894 EXPORT_SYMBOL(xfrm_policy_byid);
 895
 896 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 897 static inline int
 898 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 899 {
 900         int dir, err = 0;
 901
 902         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 903                 struct xfrm_policy *pol;
 904                 int i;
 905
 906                 hlist_for_each_entry(pol,
 907                                      &net->xfrm.policy_inexact[dir], bydst) {
 908                         if (pol->type != type)
 909                                 continue;
 910                         err = security_xfrm_policy_delete(pol->security);
 911                         if (err) {
 912                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 913                                 return err;
 914                         }
 915                 }
 916                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 917                         hlist_for_each_entry(pol,
 918                                              net->xfrm.policy_bydst[dir].table + i,
 919                                              bydst) {
 920                                 if (pol->type != type)
 921                                         continue;
 922                                 err = security_xfrm_policy_delete(
 923                                                                 pol->security);
 924                                 if (err) {
 925                                         xfrm_audit_policy_delete(pol, 0,
 926                                                                  task_valid);
 927                                         return err;
 928                                 }
 929                         }
 930                 }
 931         }
 932         return err;
 933 }
 934 #else
 935 static inline int
 936 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 937 {
 938         return 0;
 939 }
 940 #endif
 941
 942 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 943 {
 944         int dir, err = 0, cnt = 0;
 945
 946         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 947
 948         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 949         if (err)
 950                 goto out;
 951
 952         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 953                 struct xfrm_policy *pol;
 954                 int i;
 955
 956         again1:
 957                 hlist_for_each_entry(pol,
 958                                      &net->xfrm.policy_inexact[dir], bydst) {
 959                         if (pol->type != type)
 960                                 continue;
 961                         __xfrm_policy_unlink(pol, dir);
 962                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 963                         cnt++;
 964
 965                         xfrm_audit_policy_delete(pol, 1, task_valid);
 966
 967                         xfrm_policy_kill(pol);
 968
 969                         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 970                         goto again1;
 971                 }
 972
 973                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 974         again2:
 975                         hlist_for_each_entry(pol,
 976                                              net->xfrm.policy_bydst[dir].table + i,
 977                                              bydst) {
 978                                 if (pol->type != type)
 979                                         continue;
 980                                 __xfrm_policy_unlink(pol, dir);
 981                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 982                                 cnt++;
 983
 984                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 985                                 xfrm_policy_kill(pol);
 986
 987                                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
 988                                 goto again2;
 989                         }
 990                 }
 991
 992         }
 993         if (!cnt)
 994                 err = -ESRCH;
 995 out:
 996         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 997         return err;
 998 }
 999 EXPORT_SYMBOL(xfrm_policy_flush);
1000
1001 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1002                      int (*func)(struct xfrm_policy *, int, int, void*),
1003                      void *data)
1004 {
1005         struct xfrm_policy *pol;
1006         struct xfrm_policy_walk_entry *x;
1007         int error = 0;
1008
1009         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1010             walk->type != XFRM_POLICY_TYPE_ANY)
1011                 return -EINVAL;
1012
1013         if (list_empty(&walk->walk.all) && walk->seq != 0)
1014                 return 0;
1015
1016         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1017         if (list_empty(&walk->walk.all))
1018                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1019         else
1020                 x = list_first_entry(&walk->walk.all,
1021                                      struct xfrm_policy_walk_entry, all);
1022
1023         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1024                 if (x->dead)
1025                         continue;
1026                 pol = container_of(x, struct xfrm_policy, walk);
1027                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1028                     walk->type != pol->type)
1029                         continue;
1030                 error = func(pol, xfrm_policy_id2dir(pol->index),
1031                              walk->seq, data);
1032                 if (error) {
1033                         list_move_tail(&walk->walk.all, &x->all);
1034                         goto out;
1035                 }
1036                 walk->seq++;
1037         }
1038         if (walk->seq == 0) {
1039                 error = -ENOENT;
1040                 goto out;
1041         }
1042         list_del_init(&walk->walk.all);
1043 out:
1044         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1045         return error;
1046 }
1047 EXPORT_SYMBOL(xfrm_policy_walk);
1048
1049 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1050 {
1051         INIT_LIST_HEAD(&walk->walk.all);
1052         walk->walk.dead = 1;
1053         walk->type = type;
1054         walk->seq = 0;
1055 }
1056 EXPORT_SYMBOL(xfrm_policy_walk_init);
1057
1058 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1059 {
1060         if (list_empty(&walk->walk.all))
1061                 return;
1062
1063         write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1064         list_del(&walk->walk.all);
1065         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1066 }
1067 EXPORT_SYMBOL(xfrm_policy_walk_done);
1068
1069 /*
1070  * Find policy to apply to this flow.
1071  *
1072  * Returns 0 if policy found, else an -errno.
1073  */
1074 static int xfrm_policy_match(const struct xfrm_policy *pol,
1075                              const struct flowi *fl,
1076                              u8 type, u16 family, int dir)
1077 {
1078         const struct xfrm_selector *sel = &pol->selector;
1079         int ret = -ESRCH;
1080         bool match;
1081
1082         if (pol->family != family ||
1083             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1084             pol->type != type)
1085                 return ret;
1086
1087         match = xfrm_selector_match(sel, fl, family);
1088         if (match)
1089                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1090                                                   dir);
1091
1092         return ret;
1093 }
1094
1095 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1096                                                      const struct flowi *fl,
1097                                                      u16 family, u8 dir)
1098 {
1099         int err;
1100         struct xfrm_policy *pol, *ret;
1101         const xfrm_address_t *daddr, *saddr;
1102         struct hlist_head *chain;
1103         u32 priority = ~0U;
1104
1105         daddr = xfrm_flowi_daddr(fl, family);
1106         saddr = xfrm_flowi_saddr(fl, family);
1107         if (unlikely(!daddr || !saddr))
1108                 return NULL;
1109
1110         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1111         chain = policy_hash_direct(net, daddr, saddr, family, dir);
1112         ret = NULL;
1113         hlist_for_each_entry(pol, chain, bydst) {
1114                 err = xfrm_policy_match(pol, fl, type, family, dir);
1115                 if (err) {
1116                         if (err == -ESRCH)
1117                                 continue;
1118                         else {
1119                                 ret = ERR_PTR(err);
1120                                 goto fail;
1121                         }
1122                 } else {
1123                         ret = pol;
1124                         priority = ret->priority;
1125                         break;
1126                 }
1127         }
1128         chain = &net->xfrm.policy_inexact[dir];
1129         hlist_for_each_entry(pol, chain, bydst) {
1130                 if ((pol->priority >= priority) && ret)
1131                         break;
1132
1133                 err = xfrm_policy_match(pol, fl, type, family, dir);
1134                 if (err) {
1135                         if (err == -ESRCH)
1136                                 continue;
1137                         else {
1138                                 ret = ERR_PTR(err);
1139                                 goto fail;
1140                         }
1141                 } else {
1142                         ret = pol;
1143                         break;
1144                 }
1145         }
1146
1147         xfrm_pol_hold(ret);
1148 fail:
1149         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1150
1151         return ret;
1152 }
1153
1154 static struct xfrm_policy *
1155 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1156 {
1157 #ifdef CONFIG_XFRM_SUB_POLICY
1158         struct xfrm_policy *pol;
1159
1160         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1161         if (pol != NULL)
1162                 return pol;
1163 #endif
1164         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1165 }
1166
1167 static int flow_to_policy_dir(int dir)
1168 {
1169         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1170             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1171             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1172                 return dir;
1173
1174         switch (dir) {
1175         default:
1176         case FLOW_DIR_IN:
1177                 return XFRM_POLICY_IN;
1178         case FLOW_DIR_OUT:
1179                 return XFRM_POLICY_OUT;
1180         case FLOW_DIR_FWD:
1181                 return XFRM_POLICY_FWD;
1182         }
1183 }
1184
1185 static struct flow_cache_object *
1186 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1187                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1188 {
1189         struct xfrm_policy *pol;
1190
1191         if (old_obj)
1192                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1193
1194         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1195         if (IS_ERR_OR_NULL(pol))
1196                 return ERR_CAST(pol);
1197
1198         /* Resolver returns two references:
1199          * one for cache and one for caller of flow_cache_lookup() */
1200         xfrm_pol_hold(pol);
1201
1202         return &pol->flo;
1203 }
1204
1205 static inline int policy_to_flow_dir(int dir)
1206 {
1207         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1208             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1209             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1210                 return dir;
1211         switch (dir) {
1212         default:
1213         case XFRM_POLICY_IN:
1214                 return FLOW_DIR_IN;
1215         case XFRM_POLICY_OUT:
1216                 return FLOW_DIR_OUT;
1217         case XFRM_POLICY_FWD:
1218                 return FLOW_DIR_FWD;
1219         }
1220 }
1221
1222 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1223                                                  const struct flowi *fl)
1224 {
1225         struct xfrm_policy *pol;
1226         struct net *net = sock_net(sk);
1227
1228         rcu_read_lock();
1229         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1230         pol = rcu_dereference(sk->sk_policy[dir]);
1231         if (pol != NULL) {
1232                 bool match = xfrm_selector_match(&pol->selector, fl,
1233                                                  sk->sk_family);
1234                 int err = 0;
1235
1236                 if (match) {
1237                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1238                                 pol = NULL;
1239                                 goto out;
1240                         }
1241                         err = security_xfrm_policy_lookup(pol->security,
1242                                                       fl->flowi_secid,
1243                                                       policy_to_flow_dir(dir));
1244                         if (!err)
1245                                 xfrm_pol_hold(pol);
1246                         else if (err == -ESRCH)
1247                                 pol = NULL;
1248                         else
1249                                 pol = ERR_PTR(err);
1250                 } else
1251                         pol = NULL;
1252         }
1253 out:
1254         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1255         rcu_read_unlock();
1256         return pol;
1257 }
1258
1259 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1260 {
1261         struct net *net = xp_net(pol);
1262
1263         list_add(&pol->walk.all, &net->xfrm.policy_all);
1264         net->xfrm.policy_count[dir]++;
1265         xfrm_pol_hold(pol);
1266 }
1267
1268 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1269                                                 int dir)
1270 {
1271         struct net *net = xp_net(pol);
1272
1273         if (list_empty(&pol->walk.all))
1274                 return NULL;
1275
1276         /* Socket policies are not hashed. */
1277         if (!hlist_unhashed(&pol->bydst)) {
1278                 hlist_del(&pol->bydst);
1279                 hlist_del(&pol->byidx);
1280         }
1281
1282         list_del_init(&pol->walk.all);
1283         net->xfrm.policy_count[dir]--;
1284
1285         return pol;
1286 }
1287
1288 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1289 {
1290         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1291 }
1292
1293 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1294 {
1295         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1296 }
1297
1298 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1299 {
1300         struct net *net = xp_net(pol);
1301
1302         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1303         pol = __xfrm_policy_unlink(pol, dir);
1304         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1305         if (pol) {
1306                 xfrm_policy_kill(pol);
1307                 return 0;
1308         }
1309         return -ENOENT;
1310 }
1311 EXPORT_SYMBOL(xfrm_policy_delete);
1312
1313 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1314 {
1315         struct net *net = xp_net(pol);
1316         struct xfrm_policy *old_pol;
1317
1318 #ifdef CONFIG_XFRM_SUB_POLICY
1319         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1320                 return -EINVAL;
1321 #endif
1322
1323         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1324         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
1325                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
1326         if (pol) {
1327                 pol->curlft.add_time = get_seconds();
1328                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1329                 xfrm_sk_policy_link(pol, dir);
1330         }
1331         rcu_assign_pointer(sk->sk_policy[dir], pol);
1332         if (old_pol) {
1333                 if (pol)
1334                         xfrm_policy_requeue(old_pol, pol);
1335
1336                 /* Unlinking succeeds always. This is the only function
1337                  * allowed to delete or replace socket policy.
1338                  */
1339                 xfrm_sk_policy_unlink(old_pol, dir);
1340         }
1341         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1342
1343         if (old_pol) {
1344                 xfrm_policy_kill(old_pol);
1345         }
1346         return 0;
1347 }
1348
1349 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1350 {
1351         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1352         struct net *net = xp_net(old);
1353
1354         if (newp) {
1355                 newp->selector = old->selector;
1356                 if (security_xfrm_policy_clone(old->security,
1357                                                &newp->security)) {
1358                         kfree(newp);
1359                         return NULL;  /* ENOMEM */
1360                 }
1361                 newp->lft = old->lft;
1362                 newp->curlft = old->curlft;
1363                 newp->mark = old->mark;
1364                 newp->action = old->action;
1365                 newp->flags = old->flags;
1366                 newp->xfrm_nr = old->xfrm_nr;
1367                 newp->index = old->index;
1368                 newp->type = old->type;
1369                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1370                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1371                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1372                 xfrm_sk_policy_link(newp, dir);
1373                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1374                 xfrm_pol_put(newp);
1375         }
1376         return newp;
1377 }
1378
1379 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
1380 {
1381         const struct xfrm_policy *p;
1382         struct xfrm_policy *np;
1383         int i, ret = 0;
1384
1385         rcu_read_lock();
1386         for (i = 0; i < 2; i++) {
1387                 p = rcu_dereference(osk->sk_policy[i]);
1388                 if (p) {
1389                         np = clone_policy(p, i);
1390                         if (unlikely(!np)) {
1391                                 ret = -ENOMEM;
1392                                 break;
1393                         }
1394                         rcu_assign_pointer(sk->sk_policy[i], np);
1395                 }
1396         }
1397         rcu_read_unlock();
1398         return ret;
1399 }
1400
1401 static int
1402 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1403                xfrm_address_t *remote, unsigned short family)
1404 {
1405         int err;
1406         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1407
1408         if (unlikely(afinfo == NULL))
1409                 return -EINVAL;
1410         err = afinfo->get_saddr(net, oif, local, remote);
1411         xfrm_policy_put_afinfo(afinfo);
1412         return err;
1413 }
1414
1415 /* Resolve list of templates for the flow, given policy. */
1416
1417 static int
1418 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1419                       struct xfrm_state **xfrm, unsigned short family)
1420 {
1421         struct net *net = xp_net(policy);
1422         int nx;
1423         int i, error;
1424         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1425         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1426         xfrm_address_t tmp;
1427
1428         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1429                 struct xfrm_state *x;
1430                 xfrm_address_t *remote = daddr;
1431                 xfrm_address_t *local  = saddr;
1432                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1433
1434                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1435                     tmpl->mode == XFRM_MODE_BEET) {
1436                         remote = &tmpl->id.daddr;
1437                         local = &tmpl->saddr;
1438                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1439                                 error = xfrm_get_saddr(net, fl->flowi_oif,
1440                                                        &tmp, remote,
1441                                                        tmpl->encap_family);
1442                                 if (error)
1443                                         goto fail;
1444                                 local = &tmp;
1445                         }
1446                 }
1447
1448                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1449
1450                 if (x && x->km.state == XFRM_STATE_VALID) {
1451                         xfrm[nx++] = x;
1452                         daddr = remote;
1453                         saddr = local;
1454                         continue;
1455                 }
1456                 if (x) {
1457                         error = (x->km.state == XFRM_STATE_ERROR ?
1458                                  -EINVAL : -EAGAIN);
1459                         xfrm_state_put(x);
1460                 } else if (error == -ESRCH) {
1461                         error = -EAGAIN;
1462                 }
1463
1464                 if (!tmpl->optional)
1465                         goto fail;
1466         }
1467         return nx;
1468
1469 fail:
1470         for (nx--; nx >= 0; nx--)
1471                 xfrm_state_put(xfrm[nx]);
1472         return error;
1473 }
1474
1475 static int
1476 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1477                   struct xfrm_state **xfrm, unsigned short family)
1478 {
1479         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1480         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1481         int cnx = 0;
1482         int error;
1483         int ret;
1484         int i;
1485
1486         for (i = 0; i < npols; i++) {
1487                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1488                         error = -ENOBUFS;
1489                         goto fail;
1490                 }
1491
1492                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1493                 if (ret < 0) {
1494                         error = ret;
1495                         goto fail;
1496                 } else
1497                         cnx += ret;
1498         }
1499
1500         /* found states are sorted for outbound processing */
1501         if (npols > 1)
1502                 xfrm_state_sort(xfrm, tpp, cnx, family);
1503
1504         return cnx;
1505
1506  fail:
1507         for (cnx--; cnx >= 0; cnx--)
1508                 xfrm_state_put(tpp[cnx]);
1509         return error;
1510
1511 }
1512
1513 /* Check that the bundle accepts the flow and its components are
1514  * still valid.
1515  */
1516
1517 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1518 {
1519         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1520         int tos;
1521
1522         if (!afinfo)
1523                 return -EINVAL;
1524
1525         tos = afinfo->get_tos(fl);
1526
1527         xfrm_policy_put_afinfo(afinfo);
1528
1529         return tos;
1530 }
1531
1532 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1533 {
1534         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1535         struct dst_entry *dst = &xdst->u.dst;
1536
1537         if (xdst->route == NULL) {
1538                 /* Dummy bundle - if it has xfrms we were not
1539                  * able to build bundle as template resolution failed.
1540                  * It means we need to try again resolving. */
1541                 if (xdst->num_xfrms > 0)
1542                         return NULL;
1543         } else if (dst->flags & DST_XFRM_QUEUE) {
1544                 return NULL;
1545         } else {
1546                 /* Real bundle */
1547                 if (stale_bundle(dst))
1548                         return NULL;
1549         }
1550
1551         dst_hold(dst);
1552         return flo;
1553 }
1554
1555 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1556 {
1557         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1558         struct dst_entry *dst = &xdst->u.dst;
1559
1560         if (!xdst->route)
1561                 return 0;
1562         if (stale_bundle(dst))
1563                 return 0;
1564
1565         return 1;
1566 }
1567
1568 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1569 {
1570         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1571         struct dst_entry *dst = &xdst->u.dst;
1572
1573         dst_free(dst);
1574 }
1575
1576 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1577         .get = xfrm_bundle_flo_get,
1578         .check = xfrm_bundle_flo_check,
1579         .delete = xfrm_bundle_flo_delete,
1580 };
1581
1582 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1583 {
1584         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1585         struct dst_ops *dst_ops;
1586         struct xfrm_dst *xdst;
1587
1588         if (!afinfo)
1589                 return ERR_PTR(-EINVAL);
1590
1591         switch (family) {
1592         case AF_INET:
1593                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1594                 break;
1595 #if IS_ENABLED(CONFIG_IPV6)
1596         case AF_INET6:
1597                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1598                 break;
1599 #endif
1600         default:
1601                 BUG();
1602         }
1603         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1604
1605         if (likely(xdst)) {
1606                 struct dst_entry *dst = &xdst->u.dst;
1607
1608                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1609                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1610         } else
1611                 xdst = ERR_PTR(-ENOBUFS);
1612
1613         xfrm_policy_put_afinfo(afinfo);
1614
1615         return xdst;
1616 }
1617
1618 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1619                                  int nfheader_len)
1620 {
1621         struct xfrm_policy_afinfo *afinfo =
1622                 xfrm_policy_get_afinfo(dst->ops->family);
1623         int err;
1624
1625         if (!afinfo)
1626                 return -EINVAL;
1627
1628         err = afinfo->init_path(path, dst, nfheader_len);
1629
1630         xfrm_policy_put_afinfo(afinfo);
1631
1632         return err;
1633 }
1634
1635 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1636                                 const struct flowi *fl)
1637 {
1638         struct xfrm_policy_afinfo *afinfo =
1639                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1640         int err;
1641
1642         if (!afinfo)
1643                 return -EINVAL;
1644
1645         err = afinfo->fill_dst(xdst, dev, fl);
1646
1647         xfrm_policy_put_afinfo(afinfo);
1648
1649         return err;
1650 }
1651
1652
1653 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1654  * all the metrics... Shortly, bundle a bundle.
1655  */
1656
1657 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1658                                             struct xfrm_state **xfrm, int nx,
1659                                             const struct flowi *fl,
1660                                             struct dst_entry *dst)
1661 {
1662         struct net *net = xp_net(policy);
1663         unsigned long now = jiffies;
1664         struct net_device *dev;
1665         struct xfrm_mode *inner_mode;
1666         struct dst_entry *dst_prev = NULL;
1667         struct dst_entry *dst0 = NULL;
1668         int i = 0;
1669         int err;
1670         int header_len = 0;
1671         int nfheader_len = 0;
1672         int trailer_len = 0;
1673         int tos;
1674         int family = policy->selector.family;
1675         xfrm_address_t saddr, daddr;
1676
1677         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1678
1679         tos = xfrm_get_tos(fl, family);
1680         err = tos;
1681         if (tos < 0)
1682                 goto put_states;
1683
1684         dst_hold(dst);
1685
1686         for (; i < nx; i++) {
1687                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1688                 struct dst_entry *dst1 = &xdst->u.dst;
1689
1690                 err = PTR_ERR(xdst);
1691                 if (IS_ERR(xdst)) {
1692                         dst_release(dst);
1693                         goto put_states;
1694                 }
1695
1696                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1697                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1698                                                         xfrm_af2proto(family));
1699                         if (!inner_mode) {
1700                                 err = -EAFNOSUPPORT;
1701                                 dst_release(dst);
1702                                 goto put_states;
1703                         }
1704                 } else
1705                         inner_mode = xfrm[i]->inner_mode;
1706
1707                 if (!dst_prev)
1708                         dst0 = dst1;
1709                 else {
1710                         dst_prev->child = dst_clone(dst1);
1711                         dst1->flags |= DST_NOHASH;
1712                 }
1713
1714                 xdst->route = dst;
1715                 dst_copy_metrics(dst1, dst);
1716
1717                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1718                         family = xfrm[i]->props.family;
1719                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1720                                               &saddr, &daddr, family);
1721                         err = PTR_ERR(dst);
1722                         if (IS_ERR(dst))
1723                                 goto put_states;
1724                 } else
1725                         dst_hold(dst);
1726
1727                 dst1->xfrm = xfrm[i];
1728                 xdst->xfrm_genid = xfrm[i]->genid;
1729
1730                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1731                 dst1->flags |= DST_HOST;
1732                 dst1->lastuse = now;
1733
1734                 dst1->input = dst_discard;
1735                 dst1->output = inner_mode->afinfo->output;
1736
1737                 dst1->next = dst_prev;
1738                 dst_prev = dst1;
1739
1740                 header_len += xfrm[i]->props.header_len;
1741                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1742                         nfheader_len += xfrm[i]->props.header_len;
1743                 trailer_len += xfrm[i]->props.trailer_len;
1744         }
1745
1746         dst_prev->child = dst;
1747         dst0->path = dst;
1748
1749         err = -ENODEV;
1750         dev = dst->dev;
1751         if (!dev)
1752                 goto free_dst;
1753
1754         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1755         xfrm_init_pmtu(dst_prev);
1756
1757         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1758                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1759
1760                 err = xfrm_fill_dst(xdst, dev, fl);
1761                 if (err)
1762                         goto free_dst;
1763
1764                 dst_prev->header_len = header_len;
1765                 dst_prev->trailer_len = trailer_len;
1766                 header_len -= xdst->u.dst.xfrm->props.header_len;
1767                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1768         }
1769
1770 out:
1771         return dst0;
1772
1773 put_states:
1774         for (; i < nx; i++)
1775                 xfrm_state_put(xfrm[i]);
1776 free_dst:
1777         if (dst0)
1778                 dst_free(dst0);
1779         dst0 = ERR_PTR(err);
1780         goto out;
1781 }
1782
1783 #ifdef CONFIG_XFRM_SUB_POLICY
1784 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1785 {
1786         if (!*target) {
1787                 *target = kmalloc(size, GFP_ATOMIC);
1788                 if (!*target)
1789                         return -ENOMEM;
1790         }
1791
1792         memcpy(*target, src, size);
1793         return 0;
1794 }
1795 #endif
1796
1797 static int xfrm_dst_update_parent(struct dst_entry *dst,
1798                                   const struct xfrm_selector *sel)
1799 {
1800 #ifdef CONFIG_XFRM_SUB_POLICY
1801         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1802         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1803                                    sel, sizeof(*sel));
1804 #else
1805         return 0;
1806 #endif
1807 }
1808
1809 static int xfrm_dst_update_origin(struct dst_entry *dst,
1810                                   const struct flowi *fl)
1811 {
1812 #ifdef CONFIG_XFRM_SUB_POLICY
1813         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1814         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1815 #else
1816         return 0;
1817 #endif
1818 }
1819
1820 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1821                                 struct xfrm_policy **pols,
1822                                 int *num_pols, int *num_xfrms)
1823 {
1824         int i;
1825
1826         if (*num_pols == 0 || !pols[0]) {
1827                 *num_pols = 0;
1828                 *num_xfrms = 0;
1829                 return 0;
1830         }
1831         if (IS_ERR(pols[0]))
1832                 return PTR_ERR(pols[0]);
1833
1834         *num_xfrms = pols[0]->xfrm_nr;
1835
1836 #ifdef CONFIG_XFRM_SUB_POLICY
1837         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1838             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1839                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1840                                                     XFRM_POLICY_TYPE_MAIN,
1841                                                     fl, family,
1842                                                     XFRM_POLICY_OUT);
1843                 if (pols[1]) {
1844                         if (IS_ERR(pols[1])) {
1845                                 xfrm_pols_put(pols, *num_pols);
1846                                 return PTR_ERR(pols[1]);
1847                         }
1848                         (*num_pols)++;
1849                         (*num_xfrms) += pols[1]->xfrm_nr;
1850                 }
1851         }
1852 #endif
1853         for (i = 0; i < *num_pols; i++) {
1854                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1855                         *num_xfrms = -1;
1856                         break;
1857                 }
1858         }
1859
1860         return 0;
1861
1862 }
1863
1864 static struct xfrm_dst *
1865 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1866                                const struct flowi *fl, u16 family,
1867                                struct dst_entry *dst_orig)
1868 {
1869         struct net *net = xp_net(pols[0]);
1870         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1871         struct dst_entry *dst;
1872         struct xfrm_dst *xdst;
1873         int err;
1874
1875         /* Try to instantiate a bundle */
1876         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1877         if (err <= 0) {
1878                 if (err != 0 && err != -EAGAIN)
1879                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1880                 return ERR_PTR(err);
1881         }
1882
1883         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1884         if (IS_ERR(dst)) {
1885                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1886                 return ERR_CAST(dst);
1887         }
1888
1889         xdst = (struct xfrm_dst *)dst;
1890         xdst->num_xfrms = err;
1891         if (num_pols > 1)
1892                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1893         else
1894                 err = xfrm_dst_update_origin(dst, fl);
1895         if (unlikely(err)) {
1896                 dst_free(dst);
1897                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1898                 return ERR_PTR(err);
1899         }
1900
1901         xdst->num_pols = num_pols;
1902         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1903         xdst->policy_genid = atomic_read(&pols[0]->genid);
1904
1905         return xdst;
1906 }
1907
1908 static void xfrm_policy_queue_process(unsigned long arg)
1909 {
1910         struct sk_buff *skb;
1911         struct sock *sk;
1912         struct dst_entry *dst;
1913         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1914         struct net *net = xp_net(pol);
1915         struct xfrm_policy_queue *pq = &pol->polq;
1916         struct flowi fl;
1917         struct sk_buff_head list;
1918
1919         spin_lock(&pq->hold_queue.lock);
1920         skb = skb_peek(&pq->hold_queue);
1921         if (!skb) {
1922                 spin_unlock(&pq->hold_queue.lock);
1923                 goto out;
1924         }
1925         dst = skb_dst(skb);
1926         sk = skb->sk;
1927         xfrm_decode_session(skb, &fl, dst->ops->family);
1928         spin_unlock(&pq->hold_queue.lock);
1929
1930         dst_hold(dst->path);
1931         dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1932         if (IS_ERR(dst))
1933                 goto purge_queue;
1934
1935         if (dst->flags & DST_XFRM_QUEUE) {
1936                 dst_release(dst);
1937
1938                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1939                         goto purge_queue;
1940
1941                 pq->timeout = pq->timeout << 1;
1942                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1943                         xfrm_pol_hold(pol);
1944         goto out;
1945         }
1946
1947         dst_release(dst);
1948
1949         __skb_queue_head_init(&list);
1950
1951         spin_lock(&pq->hold_queue.lock);
1952         pq->timeout = 0;
1953         skb_queue_splice_init(&pq->hold_queue, &list);
1954         spin_unlock(&pq->hold_queue.lock);
1955
1956         while (!skb_queue_empty(&list)) {
1957                 skb = __skb_dequeue(&list);
1958
1959                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1960                 dst_hold(skb_dst(skb)->path);
1961                 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1962                 if (IS_ERR(dst)) {
1963                         kfree_skb(skb);
1964                         continue;
1965                 }
1966
1967                 nf_reset(skb);
1968                 skb_dst_drop(skb);
1969                 skb_dst_set(skb, dst);
1970
1971                 dst_output(net, skb->sk, skb);
1972         }
1973
1974 out:
1975         xfrm_pol_put(pol);
1976         return;
1977
1978 purge_queue:
1979         pq->timeout = 0;
1980         skb_queue_purge(&pq->hold_queue);
1981         xfrm_pol_put(pol);
1982 }
1983
1984 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1985 {
1986         unsigned long sched_next;
1987         struct dst_entry *dst = skb_dst(skb);
1988         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1989         struct xfrm_policy *pol = xdst->pols[0];
1990         struct xfrm_policy_queue *pq = &pol->polq;
1991
1992         if (unlikely(skb_fclone_busy(sk, skb))) {
1993                 kfree_skb(skb);
1994                 return 0;
1995         }
1996
1997         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1998                 kfree_skb(skb);
1999                 return -EAGAIN;
2000         }
2001
2002         skb_dst_force(skb);
2003
2004         spin_lock_bh(&pq->hold_queue.lock);
2005
2006         if (!pq->timeout)
2007                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2008
2009         sched_next = jiffies + pq->timeout;
2010
2011         if (del_timer(&pq->hold_timer)) {
2012                 if (time_before(pq->hold_timer.expires, sched_next))
2013                         sched_next = pq->hold_timer.expires;
2014                 xfrm_pol_put(pol);
2015         }
2016
2017         __skb_queue_tail(&pq->hold_queue, skb);
2018         if (!mod_timer(&pq->hold_timer, sched_next))
2019                 xfrm_pol_hold(pol);
2020
2021         spin_unlock_bh(&pq->hold_queue.lock);
2022
2023         return 0;
2024 }
2025
2026 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2027                                                  struct xfrm_flo *xflo,
2028                                                  const struct flowi *fl,
2029                                                  int num_xfrms,
2030                                                  u16 family)
2031 {
2032         int err;
2033         struct net_device *dev;
2034         struct dst_entry *dst;
2035         struct dst_entry *dst1;
2036         struct xfrm_dst *xdst;
2037
2038         xdst = xfrm_alloc_dst(net, family);
2039         if (IS_ERR(xdst))
2040                 return xdst;
2041
2042         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2043             net->xfrm.sysctl_larval_drop ||
2044             num_xfrms <= 0)
2045                 return xdst;
2046
2047         dst = xflo->dst_orig;
2048         dst1 = &xdst->u.dst;
2049         dst_hold(dst);
2050         xdst->route = dst;
2051
2052         dst_copy_metrics(dst1, dst);
2053
2054         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2055         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2056         dst1->lastuse = jiffies;
2057
2058         dst1->input = dst_discard;
2059         dst1->output = xdst_queue_output;
2060
2061         dst_hold(dst);
2062         dst1->child = dst;
2063         dst1->path = dst;
2064
2065         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2066
2067         err = -ENODEV;
2068         dev = dst->dev;
2069         if (!dev)
2070                 goto free_dst;
2071
2072         err = xfrm_fill_dst(xdst, dev, fl);
2073         if (err)
2074                 goto free_dst;
2075
2076 out:
2077         return xdst;
2078
2079 free_dst:
2080         dst_release(dst1);
2081         xdst = ERR_PTR(err);
2082         goto out;
2083 }
2084
2085 static struct flow_cache_object *
2086 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2087                    struct flow_cache_object *oldflo, void *ctx)
2088 {
2089         struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2090         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2091         struct xfrm_dst *xdst, *new_xdst;
2092         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2093
2094         /* Check if the policies from old bundle are usable */
2095         xdst = NULL;
2096         if (oldflo) {
2097                 xdst = container_of(oldflo, struct xfrm_dst, flo);
2098                 num_pols = xdst->num_pols;
2099                 num_xfrms = xdst->num_xfrms;
2100                 pol_dead = 0;
2101                 for (i = 0; i < num_pols; i++) {
2102                         pols[i] = xdst->pols[i];
2103                         pol_dead |= pols[i]->walk.dead;
2104                 }
2105                 if (pol_dead) {
2106                         dst_free(&xdst->u.dst);
2107                         xdst = NULL;
2108                         num_pols = 0;
2109                         num_xfrms = 0;
2110                         oldflo = NULL;
2111                 }
2112         }
2113
2114         /* Resolve policies to use if we couldn't get them from
2115          * previous cache entry */
2116         if (xdst == NULL) {
2117                 num_pols = 1;
2118                 pols[0] = __xfrm_policy_lookup(net, fl, family,
2119                                                flow_to_policy_dir(dir));
2120                 err = xfrm_expand_policies(fl, family, pols,
2121                                            &num_pols, &num_xfrms);
2122                 if (err < 0)
2123                         goto inc_error;
2124                 if (num_pols == 0)
2125                         return NULL;
2126                 if (num_xfrms <= 0)
2127                         goto make_dummy_bundle;
2128         }
2129
2130         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2131                                                   xflo->dst_orig);
2132         if (IS_ERR(new_xdst)) {
2133                 err = PTR_ERR(new_xdst);
2134                 if (err != -EAGAIN)
2135                         goto error;
2136                 if (oldflo == NULL)
2137                         goto make_dummy_bundle;
2138                 dst_hold(&xdst->u.dst);
2139                 return oldflo;
2140         } else if (new_xdst == NULL) {
2141                 num_xfrms = 0;
2142                 if (oldflo == NULL)
2143                         goto make_dummy_bundle;
2144                 xdst->num_xfrms = 0;
2145                 dst_hold(&xdst->u.dst);
2146                 return oldflo;
2147         }
2148
2149         /* Kill the previous bundle */
2150         if (xdst) {
2151                 /* The policies were stolen for newly generated bundle */
2152                 xdst->num_pols = 0;
2153                 dst_free(&xdst->u.dst);
2154         }
2155
2156         /* Flow cache does not have reference, it dst_free()'s,
2157          * but we do need to return one reference for original caller */
2158         dst_hold(&new_xdst->u.dst);
2159         return &new_xdst->flo;
2160
2161 make_dummy_bundle:
2162         /* We found policies, but there's no bundles to instantiate:
2163          * either because the policy blocks, has no transformations or
2164          * we could not build template (no xfrm_states).*/
2165         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2166         if (IS_ERR(xdst)) {
2167                 xfrm_pols_put(pols, num_pols);
2168                 return ERR_CAST(xdst);
2169         }
2170         xdst->num_pols = num_pols;
2171         xdst->num_xfrms = num_xfrms;
2172         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2173
2174         dst_hold(&xdst->u.dst);
2175         return &xdst->flo;
2176
2177 inc_error:
2178         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2179 error:
2180         if (xdst != NULL)
2181                 dst_free(&xdst->u.dst);
2182         else
2183                 xfrm_pols_put(pols, num_pols);
2184         return ERR_PTR(err);
2185 }
2186
2187 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2188                                         struct dst_entry *dst_orig)
2189 {
2190         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2191         struct dst_entry *ret;
2192
2193         if (!afinfo) {
2194                 dst_release(dst_orig);
2195                 return ERR_PTR(-EINVAL);
2196         } else {
2197                 ret = afinfo->blackhole_route(net, dst_orig);
2198         }
2199         xfrm_policy_put_afinfo(afinfo);
2200
2201         return ret;
2202 }
2203
2204 /* Main function: finds/creates a bundle for given flow.
2205  *
2206  * At the moment we eat a raw IP route. Mostly to speed up lookups
2207  * on interfaces with disabled IPsec.
2208  */
2209 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2210                               const struct flowi *fl,
2211                               const struct sock *sk, int flags)
2212 {
2213         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2214         struct flow_cache_object *flo;
2215         struct xfrm_dst *xdst;
2216         struct dst_entry *dst, *route;
2217         u16 family = dst_orig->ops->family;
2218         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2219         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2220
2221         dst = NULL;
2222         xdst = NULL;
2223         route = NULL;
2224
2225         sk = sk_const_to_full_sk(sk);
2226         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2227                 num_pols = 1;
2228                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2229                 err = xfrm_expand_policies(fl, family, pols,
2230                                            &num_pols, &num_xfrms);
2231                 if (err < 0)
2232                         goto dropdst;
2233
2234                 if (num_pols) {
2235                         if (num_xfrms <= 0) {
2236                                 drop_pols = num_pols;
2237                                 goto no_transform;
2238                         }
2239
2240                         xdst = xfrm_resolve_and_create_bundle(
2241                                         pols, num_pols, fl,
2242                                         family, dst_orig);
2243                         if (IS_ERR(xdst)) {
2244                                 xfrm_pols_put(pols, num_pols);
2245                                 err = PTR_ERR(xdst);
2246                                 goto dropdst;
2247                         } else if (xdst == NULL) {
2248                                 num_xfrms = 0;
2249                                 drop_pols = num_pols;
2250                                 goto no_transform;
2251                         }
2252
2253                         dst_hold(&xdst->u.dst);
2254                         xdst->u.dst.flags |= DST_NOCACHE;
2255                         route = xdst->route;
2256                 }
2257         }
2258
2259         if (xdst == NULL) {
2260                 struct xfrm_flo xflo;
2261
2262                 xflo.dst_orig = dst_orig;
2263                 xflo.flags = flags;
2264
2265                 /* To accelerate a bit...  */
2266                 if ((dst_orig->flags & DST_NOXFRM) ||
2267                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2268                         goto nopol;
2269
2270                 flo = flow_cache_lookup(net, fl, family, dir,
2271                                         xfrm_bundle_lookup, &xflo);
2272                 if (flo == NULL)
2273                         goto nopol;
2274                 if (IS_ERR(flo)) {
2275                         err = PTR_ERR(flo);
2276                         goto dropdst;
2277                 }
2278                 xdst = container_of(flo, struct xfrm_dst, flo);
2279
2280                 num_pols = xdst->num_pols;
2281                 num_xfrms = xdst->num_xfrms;
2282                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2283                 route = xdst->route;
2284         }
2285
2286         dst = &xdst->u.dst;
2287         if (route == NULL && num_xfrms > 0) {
2288                 /* The only case when xfrm_bundle_lookup() returns a
2289                  * bundle with null route, is when the template could
2290                  * not be resolved. It means policies are there, but
2291                  * bundle could not be created, since we don't yet
2292                  * have the xfrm_state's. We need to wait for KM to
2293                  * negotiate new SA's or bail out with error.*/
2294                 if (net->xfrm.sysctl_larval_drop) {
2295                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2296                         err = -EREMOTE;
2297                         goto error;
2298                 }
2299
2300                 err = -EAGAIN;
2301
2302                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2303                 goto error;
2304         }
2305
2306 no_transform:
2307         if (num_pols == 0)
2308                 goto nopol;
2309
2310         if ((flags & XFRM_LOOKUP_ICMP) &&
2311             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2312                 err = -ENOENT;
2313                 goto error;
2314         }
2315
2316         for (i = 0; i < num_pols; i++)
2317                 pols[i]->curlft.use_time = get_seconds();
2318
2319         if (num_xfrms < 0) {
2320                 /* Prohibit the flow */
2321                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2322                 err = -EPERM;
2323                 goto error;
2324         } else if (num_xfrms > 0) {
2325                 /* Flow transformed */
2326                 dst_release(dst_orig);
2327         } else {
2328                 /* Flow passes untransformed */
2329                 dst_release(dst);
2330                 dst = dst_orig;
2331         }
2332 ok:
2333         xfrm_pols_put(pols, drop_pols);
2334         if (dst && dst->xfrm &&
2335             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2336                 dst->flags |= DST_XFRM_TUNNEL;
2337         return dst;
2338
2339 nopol:
2340         if (!(flags & XFRM_LOOKUP_ICMP)) {
2341                 dst = dst_orig;
2342                 goto ok;
2343         }
2344         err = -ENOENT;
2345 error:
2346         dst_release(dst);
2347 dropdst:
2348         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2349                 dst_release(dst_orig);
2350         xfrm_pols_put(pols, drop_pols);
2351         return ERR_PTR(err);
2352 }
2353 EXPORT_SYMBOL(xfrm_lookup);
2354
2355 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2356  * Otherwise we may send out blackholed packets.
2357  */
2358 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2359                                     const struct flowi *fl,
2360                                     const struct sock *sk, int flags)
2361 {
2362         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2363                                             flags | XFRM_LOOKUP_QUEUE |
2364                                             XFRM_LOOKUP_KEEP_DST_REF);
2365
2366         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2367                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2368
2369         return dst;
2370 }
2371 EXPORT_SYMBOL(xfrm_lookup_route);
2372
2373 static inline int
2374 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2375 {
2376         struct xfrm_state *x;
2377
2378         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2379                 return 0;
2380         x = skb->sp->xvec[idx];
2381         if (!x->type->reject)
2382                 return 0;
2383         return x->type->reject(x, skb, fl);
2384 }
2385
2386 /* When skb is transformed back to its "native" form, we have to
2387  * check policy restrictions. At the moment we make this in maximally
2388  * stupid way. Shame on me. :-) Of course, connected sockets must
2389  * have policy cached at them.
2390  */
2391
2392 static inline int
2393 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2394               unsigned short family)
2395 {
2396         if (xfrm_state_kern(x))
2397                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2398         return  x->id.proto == tmpl->id.proto &&
2399                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2400                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2401                 x->props.mode == tmpl->mode &&
2402                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2403                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2404                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2405                   xfrm_state_addr_cmp(tmpl, x, family));
2406 }
2407
2408 /*
2409  * 0 or more than 0 is returned when validation is succeeded (either bypass
2410  * because of optional transport mode, or next index of the mathced secpath
2411  * state with the template.
2412  * -1 is returned when no matching template is found.
2413  * Otherwise "-2 - errored_index" is returned.
2414  */
2415 static inline int
2416 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2417                unsigned short family)
2418 {
2419         int idx = start;
2420
2421         if (tmpl->optional) {
2422                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2423                         return start;
2424         } else
2425                 start = -1;
2426         for (; idx < sp->len; idx++) {
2427                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2428                         return ++idx;
2429                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2430                         if (start == -1)
2431                                 start = -2-idx;
2432                         break;
2433                 }
2434         }
2435         return start;
2436 }
2437
2438 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2439                           unsigned int family, int reverse)
2440 {
2441         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2442         int err;
2443
2444         if (unlikely(afinfo == NULL))
2445                 return -EAFNOSUPPORT;
2446
2447         afinfo->decode_session(skb, fl, reverse);
2448         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2449         xfrm_policy_put_afinfo(afinfo);
2450         return err;
2451 }
2452 EXPORT_SYMBOL(__xfrm_decode_session);
2453
2454 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2455 {
2456         for (; k < sp->len; k++) {
2457                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2458                         *idxp = k;
2459                         return 1;
2460                 }
2461         }
2462
2463         return 0;
2464 }
2465
2466 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2467                         unsigned short family)
2468 {
2469         struct net *net = dev_net(skb->dev);
2470         struct xfrm_policy *pol;
2471         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2472         int npols = 0;
2473         int xfrm_nr;
2474         int pi;
2475         int reverse;
2476         struct flowi fl;
2477         u8 fl_dir;
2478         int xerr_idx = -1;
2479
2480         reverse = dir & ~XFRM_POLICY_MASK;
2481         dir &= XFRM_POLICY_MASK;
2482         fl_dir = policy_to_flow_dir(dir);
2483
2484         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2485                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2486                 return 0;
2487         }
2488
2489         nf_nat_decode_session(skb, &fl, family);
2490
2491         /* First, check used SA against their selectors. */
2492         if (skb->sp) {
2493                 int i;
2494
2495                 for (i = skb->sp->len-1; i >= 0; i--) {
2496                         struct xfrm_state *x = skb->sp->xvec[i];
2497                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2498                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2499                                 return 0;
2500                         }
2501                 }
2502         }
2503
2504         pol = NULL;
2505         sk = sk_to_full_sk(sk);
2506         if (sk && sk->sk_policy[dir]) {
2507                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2508                 if (IS_ERR(pol)) {
2509                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2510                         return 0;
2511                 }
2512         }
2513
2514         if (!pol) {
2515                 struct flow_cache_object *flo;
2516
2517                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2518                                         xfrm_policy_lookup, NULL);
2519                 if (IS_ERR_OR_NULL(flo))
2520                         pol = ERR_CAST(flo);
2521                 else
2522                         pol = container_of(flo, struct xfrm_policy, flo);
2523         }
2524
2525         if (IS_ERR(pol)) {
2526                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2527                 return 0;
2528         }
2529
2530         if (!pol) {
2531                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2532                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2533                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2534                         return 0;
2535                 }
2536                 return 1;
2537         }
2538
2539         pol->curlft.use_time = get_seconds();
2540
2541         pols[0] = pol;
2542         npols++;
2543 #ifdef CONFIG_XFRM_SUB_POLICY
2544         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2545                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2546                                                     &fl, family,
2547                                                     XFRM_POLICY_IN);
2548                 if (pols[1]) {
2549                         if (IS_ERR(pols[1])) {
2550                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2551                                 return 0;
2552                         }
2553                         pols[1]->curlft.use_time = get_seconds();
2554                         npols++;
2555                 }
2556         }
2557 #endif
2558
2559         if (pol->action == XFRM_POLICY_ALLOW) {
2560                 struct sec_path *sp;
2561                 static struct sec_path dummy;
2562                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2563                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2564                 struct xfrm_tmpl **tpp = tp;
2565                 int ti = 0;
2566                 int i, k;
2567
2568                 if ((sp = skb->sp) == NULL)
2569                         sp = &dummy;
2570
2571                 for (pi = 0; pi < npols; pi++) {
2572                         if (pols[pi] != pol &&
2573                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2574                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2575                                 goto reject;
2576                         }
2577                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2578                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2579                                 goto reject_error;
2580                         }
2581                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2582                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2583                 }
2584                 xfrm_nr = ti;
2585                 if (npols > 1) {
2586                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2587                         tpp = stp;
2588                 }
2589
2590                 /* For each tunnel xfrm, find the first matching tmpl.
2591                  * For each tmpl before that, find corresponding xfrm.
2592                  * Order is _important_. Later we will implement
2593                  * some barriers, but at the moment barriers
2594                  * are implied between each two transformations.
2595                  */
2596                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2597                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2598                         if (k < 0) {
2599                                 if (k < -1)
2600                                         /* "-2 - errored_index" returned */
2601                                         xerr_idx = -(2+k);
2602                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2603                                 goto reject;
2604                         }
2605                 }
2606
2607                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2608                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2609                         goto reject;
2610                 }
2611
2612                 xfrm_pols_put(pols, npols);
2613                 return 1;
2614         }
2615         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2616
2617 reject:
2618         xfrm_secpath_reject(xerr_idx, skb, &fl);
2619 reject_error:
2620         xfrm_pols_put(pols, npols);
2621         return 0;
2622 }
2623 EXPORT_SYMBOL(__xfrm_policy_check);
2624
2625 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2626 {
2627         struct net *net = dev_net(skb->dev);
2628         struct flowi fl;
2629         struct dst_entry *dst;
2630         int res = 1;
2631
2632         if (xfrm_decode_session(skb, &fl, family) < 0) {
2633                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2634                 return 0;
2635         }
2636
2637         skb_dst_force(skb);
2638
2639         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2640         if (IS_ERR(dst)) {
2641                 res = 0;
2642                 dst = NULL;
2643         }
2644         skb_dst_set(skb, dst);
2645         return res;
2646 }
2647 EXPORT_SYMBOL(__xfrm_route_forward);
2648
2649 /* Optimize later using cookies and generation ids. */
2650
2651 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2652 {
2653         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2654          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2655          * get validated by dst_ops->check on every use.  We do this
2656          * because when a normal route referenced by an XFRM dst is
2657          * obsoleted we do not go looking around for all parent
2658          * referencing XFRM dsts so that we can invalidate them.  It
2659          * is just too much work.  Instead we make the checks here on
2660          * every use.  For example:
2661          *
2662          *      XFRM dst A --> IPv4 dst X
2663          *
2664          * X is the "xdst->route" of A (X is also the "dst->path" of A
2665          * in this example).  If X is marked obsolete, "A" will not
2666          * notice.  That's what we are validating here via the
2667          * stale_bundle() check.
2668          *
2669          * When a policy's bundle is pruned, we dst_free() the XFRM
2670          * dst which causes it's ->obsolete field to be set to
2671          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2672          * this, we want to force a new route lookup.
2673          */
2674         if (dst->obsolete < 0 && !stale_bundle(dst))
2675                 return dst;
2676
2677         return NULL;
2678 }
2679
2680 static int stale_bundle(struct dst_entry *dst)
2681 {
2682         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2683 }
2684
2685 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2686 {
2687         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2688                 dst->dev = dev_net(dev)->loopback_dev;
2689                 dev_hold(dst->dev);
2690                 dev_put(dev);
2691         }
2692 }
2693 EXPORT_SYMBOL(xfrm_dst_ifdown);
2694
2695 static void xfrm_link_failure(struct sk_buff *skb)
2696 {
2697         /* Impossible. Such dst must be popped before reaches point of failure. */
2698 }
2699
2700 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2701 {
2702         if (dst) {
2703                 if (dst->obsolete) {
2704                         dst_release(dst);
2705                         dst = NULL;
2706                 }
2707         }
2708         return dst;
2709 }
2710
2711 void xfrm_garbage_collect(struct net *net)
2712 {
2713         flow_cache_flush(net);
2714 }
2715 EXPORT_SYMBOL(xfrm_garbage_collect);
2716
2717 static void xfrm_garbage_collect_deferred(struct net *net)
2718 {
2719         flow_cache_flush_deferred(net);
2720 }
2721
2722 static void xfrm_init_pmtu(struct dst_entry *dst)
2723 {
2724         do {
2725                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2726                 u32 pmtu, route_mtu_cached;
2727
2728                 pmtu = dst_mtu(dst->child);
2729                 xdst->child_mtu_cached = pmtu;
2730
2731                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2732
2733                 route_mtu_cached = dst_mtu(xdst->route);
2734                 xdst->route_mtu_cached = route_mtu_cached;
2735
2736                 if (pmtu > route_mtu_cached)
2737                         pmtu = route_mtu_cached;
2738
2739                 dst_metric_set(dst, RTAX_MTU, pmtu);
2740         } while ((dst = dst->next));
2741 }
2742
2743 /* Check that the bundle accepts the flow and its components are
2744  * still valid.
2745  */
2746
2747 static int xfrm_bundle_ok(struct xfrm_dst *first)
2748 {
2749         struct dst_entry *dst = &first->u.dst;
2750         struct xfrm_dst *last;
2751         u32 mtu;
2752
2753         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2754             (dst->dev && !netif_running(dst->dev)))
2755                 return 0;
2756
2757         if (dst->flags & DST_XFRM_QUEUE)
2758                 return 1;
2759
2760         last = NULL;
2761
2762         do {
2763                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2764
2765                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2766                         return 0;
2767                 if (xdst->xfrm_genid != dst->xfrm->genid)
2768                         return 0;
2769                 if (xdst->num_pols > 0 &&
2770                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2771                         return 0;
2772
2773                 mtu = dst_mtu(dst->child);
2774                 if (xdst->child_mtu_cached != mtu) {
2775                         last = xdst;
2776                         xdst->child_mtu_cached = mtu;
2777                 }
2778
2779                 if (!dst_check(xdst->route, xdst->route_cookie))
2780                         return 0;
2781                 mtu = dst_mtu(xdst->route);
2782                 if (xdst->route_mtu_cached != mtu) {
2783                         last = xdst;
2784                         xdst->route_mtu_cached = mtu;
2785                 }
2786
2787                 dst = dst->child;
2788         } while (dst->xfrm);
2789
2790         if (likely(!last))
2791                 return 1;
2792
2793         mtu = last->child_mtu_cached;
2794         for (;;) {
2795                 dst = &last->u.dst;
2796
2797                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2798                 if (mtu > last->route_mtu_cached)
2799                         mtu = last->route_mtu_cached;
2800                 dst_metric_set(dst, RTAX_MTU, mtu);
2801
2802                 if (last == first)
2803                         break;
2804
2805                 last = (struct xfrm_dst *)last->u.dst.next;
2806                 last->child_mtu_cached = mtu;
2807         }
2808
2809         return 1;
2810 }
2811
2812 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2813 {
2814         return dst_metric_advmss(dst->path);
2815 }
2816
2817 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2818 {
2819         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2820
2821         return mtu ? : dst_mtu(dst->path);
2822 }
2823
2824 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2825                                            struct sk_buff *skb,
2826                                            const void *daddr)
2827 {
2828         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2829 }
2830
2831 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2832 {
2833         int err = 0;
2834         if (unlikely(afinfo == NULL))
2835                 return -EINVAL;
2836         if (unlikely(afinfo->family >= NPROTO))
2837                 return -EAFNOSUPPORT;
2838         spin_lock(&xfrm_policy_afinfo_lock);
2839         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2840                 err = -EEXIST;
2841         else {
2842                 struct dst_ops *dst_ops = afinfo->dst_ops;
2843                 if (likely(dst_ops->kmem_cachep == NULL))
2844                         dst_ops->kmem_cachep = xfrm_dst_cache;
2845                 if (likely(dst_ops->check == NULL))
2846                         dst_ops->check = xfrm_dst_check;
2847                 if (likely(dst_ops->default_advmss == NULL))
2848                         dst_ops->default_advmss = xfrm_default_advmss;
2849                 if (likely(dst_ops->mtu == NULL))
2850                         dst_ops->mtu = xfrm_mtu;
2851                 if (likely(dst_ops->negative_advice == NULL))
2852                         dst_ops->negative_advice = xfrm_negative_advice;
2853                 if (likely(dst_ops->link_failure == NULL))
2854                         dst_ops->link_failure = xfrm_link_failure;
2855                 if (likely(dst_ops->neigh_lookup == NULL))
2856                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2857                 if (likely(afinfo->garbage_collect == NULL))
2858                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2859                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2860         }
2861         spin_unlock(&xfrm_policy_afinfo_lock);
2862
2863         return err;
2864 }
2865 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2866
2867 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2868 {
2869         int err = 0;
2870         if (unlikely(afinfo == NULL))
2871                 return -EINVAL;
2872         if (unlikely(afinfo->family >= NPROTO))
2873                 return -EAFNOSUPPORT;
2874         spin_lock(&xfrm_policy_afinfo_lock);
2875         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2876                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2877                         err = -EINVAL;
2878                 else
2879                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2880                                          NULL);
2881         }
2882         spin_unlock(&xfrm_policy_afinfo_lock);
2883         if (!err) {
2884                 struct dst_ops *dst_ops = afinfo->dst_ops;
2885
2886                 synchronize_rcu();
2887
2888                 dst_ops->kmem_cachep = NULL;
2889                 dst_ops->check = NULL;
2890                 dst_ops->negative_advice = NULL;
2891                 dst_ops->link_failure = NULL;
2892                 afinfo->garbage_collect = NULL;
2893         }
2894         return err;
2895 }
2896 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2897
2898 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2899 {
2900         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2901
2902         switch (event) {
2903         case NETDEV_DOWN:
2904                 xfrm_garbage_collect(dev_net(dev));
2905         }
2906         return NOTIFY_DONE;
2907 }
2908
2909 static struct notifier_block xfrm_dev_notifier = {
2910         .notifier_call  = xfrm_dev_event,
2911 };
2912
2913 #ifdef CONFIG_XFRM_STATISTICS
2914 static int __net_init xfrm_statistics_init(struct net *net)
2915 {
2916         int rv;
2917         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2918         if (!net->mib.xfrm_statistics)
2919                 return -ENOMEM;
2920         rv = xfrm_proc_init(net);
2921         if (rv < 0)
2922                 free_percpu(net->mib.xfrm_statistics);
2923         return rv;
2924 }
2925
2926 static void xfrm_statistics_fini(struct net *net)
2927 {
2928         xfrm_proc_fini(net);
2929         free_percpu(net->mib.xfrm_statistics);
2930 }
2931 #else
2932 static int __net_init xfrm_statistics_init(struct net *net)
2933 {
2934         return 0;
2935 }
2936
2937 static void xfrm_statistics_fini(struct net *net)
2938 {
2939 }
2940 #endif
2941
2942 static int __net_init xfrm_policy_init(struct net *net)
2943 {
2944         unsigned int hmask, sz;
2945         int dir;
2946
2947         if (net_eq(net, &init_net))
2948                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2949                                            sizeof(struct xfrm_dst),
2950                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2951                                            NULL);
2952
2953         hmask = 8 - 1;
2954         sz = (hmask+1) * sizeof(struct hlist_head);
2955
2956         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2957         if (!net->xfrm.policy_byidx)
2958                 goto out_byidx;
2959         net->xfrm.policy_idx_hmask = hmask;
2960
2961         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2962                 struct xfrm_policy_hash *htab;
2963
2964                 net->xfrm.policy_count[dir] = 0;
2965                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2966                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2967
2968                 htab = &net->xfrm.policy_bydst[dir];
2969                 htab->table = xfrm_hash_alloc(sz);
2970                 if (!htab->table)
2971                         goto out_bydst;
2972                 htab->hmask = hmask;
2973                 htab->dbits4 = 32;
2974                 htab->sbits4 = 32;
2975                 htab->dbits6 = 128;
2976                 htab->sbits6 = 128;
2977         }
2978         net->xfrm.policy_hthresh.lbits4 = 32;
2979         net->xfrm.policy_hthresh.rbits4 = 32;
2980         net->xfrm.policy_hthresh.lbits6 = 128;
2981         net->xfrm.policy_hthresh.rbits6 = 128;
2982
2983         seqlock_init(&net->xfrm.policy_hthresh.lock);
2984
2985         INIT_LIST_HEAD(&net->xfrm.policy_all);
2986         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2987         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2988         if (net_eq(net, &init_net))
2989                 register_netdevice_notifier(&xfrm_dev_notifier);
2990         return 0;
2991
2992 out_bydst:
2993         for (dir--; dir >= 0; dir--) {
2994                 struct xfrm_policy_hash *htab;
2995
2996                 htab = &net->xfrm.policy_bydst[dir];
2997                 xfrm_hash_free(htab->table, sz);
2998         }
2999         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3000 out_byidx:
3001         return -ENOMEM;
3002 }
3003
3004 static void xfrm_policy_fini(struct net *net)
3005 {
3006         unsigned int sz;
3007         int dir;
3008
3009         flush_work(&net->xfrm.policy_hash_work);
3010 #ifdef CONFIG_XFRM_SUB_POLICY
3011         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3012 #endif
3013         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3014
3015         WARN_ON(!list_empty(&net->xfrm.policy_all));
3016
3017         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3018                 struct xfrm_policy_hash *htab;
3019
3020                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3021
3022                 htab = &net->xfrm.policy_bydst[dir];
3023                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3024                 WARN_ON(!hlist_empty(htab->table));
3025                 xfrm_hash_free(htab->table, sz);
3026         }
3027
3028         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3029         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3030         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3031 }
3032
3033 static int __net_init xfrm_net_init(struct net *net)
3034 {
3035         int rv;
3036
3037         rv = xfrm_statistics_init(net);
3038         if (rv < 0)
3039                 goto out_statistics;
3040         rv = xfrm_state_init(net);
3041         if (rv < 0)
3042                 goto out_state;
3043         rv = xfrm_policy_init(net);
3044         if (rv < 0)
3045                 goto out_policy;
3046         rv = xfrm_sysctl_init(net);
3047         if (rv < 0)
3048                 goto out_sysctl;
3049         rv = flow_cache_init(net);
3050         if (rv < 0)
3051                 goto out;
3052
3053         /* Initialize the per-net locks here */
3054         spin_lock_init(&net->xfrm.xfrm_state_lock);
3055         rwlock_init(&net->xfrm.xfrm_policy_lock);
3056         mutex_init(&net->xfrm.xfrm_cfg_mutex);
3057
3058         return 0;
3059
3060 out:
3061         xfrm_sysctl_fini(net);
3062 out_sysctl:
3063         xfrm_policy_fini(net);
3064 out_policy:
3065         xfrm_state_fini(net);
3066 out_state:
3067         xfrm_statistics_fini(net);
3068 out_statistics:
3069         return rv;
3070 }
3071
3072 static void __net_exit xfrm_net_exit(struct net *net)
3073 {
3074         flow_cache_fini(net);
3075         xfrm_sysctl_fini(net);
3076         xfrm_policy_fini(net);
3077         xfrm_state_fini(net);
3078         xfrm_statistics_fini(net);
3079 }
3080
3081 static struct pernet_operations __net_initdata xfrm_net_ops = {
3082         .init = xfrm_net_init,
3083         .exit = xfrm_net_exit,
3084 };
3085
3086 void __init xfrm_init(void)
3087 {
3088         register_pernet_subsys(&xfrm_net_ops);
3089         xfrm_input_init();
3090 }
3091
3092 #ifdef CONFIG_AUDITSYSCALL
3093 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3094                                          struct audit_buffer *audit_buf)
3095 {
3096         struct xfrm_sec_ctx *ctx = xp->security;
3097         struct xfrm_selector *sel = &xp->selector;
3098
3099         if (ctx)
3100                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3101                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3102
3103         switch (sel->family) {
3104         case AF_INET:
3105                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3106                 if (sel->prefixlen_s != 32)
3107                         audit_log_format(audit_buf, " src_prefixlen=%d",
3108                                          sel->prefixlen_s);
3109                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3110                 if (sel->prefixlen_d != 32)
3111                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3112                                          sel->prefixlen_d);
3113                 break;
3114         case AF_INET6:
3115                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3116                 if (sel->prefixlen_s != 128)
3117                         audit_log_format(audit_buf, " src_prefixlen=%d",
3118                                          sel->prefixlen_s);
3119                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3120                 if (sel->prefixlen_d != 128)
3121                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3122                                          sel->prefixlen_d);
3123                 break;
3124         }
3125 }
3126
3127 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3128 {
3129         struct audit_buffer *audit_buf;
3130
3131         audit_buf = xfrm_audit_start("SPD-add");
3132         if (audit_buf == NULL)
3133                 return;
3134         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3135         audit_log_format(audit_buf, " res=%u", result);
3136         xfrm_audit_common_policyinfo(xp, audit_buf);
3137         audit_log_end(audit_buf);
3138 }
3139 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3140
3141 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3142                               bool task_valid)
3143 {
3144         struct audit_buffer *audit_buf;
3145
3146         audit_buf = xfrm_audit_start("SPD-delete");
3147         if (audit_buf == NULL)
3148                 return;
3149         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3150         audit_log_format(audit_buf, " res=%u", result);
3151         xfrm_audit_common_policyinfo(xp, audit_buf);
3152         audit_log_end(audit_buf);
3153 }
3154 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3155 #endif
3156
3157 #ifdef CONFIG_XFRM_MIGRATE
3158 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3159                                         const struct xfrm_selector *sel_tgt)
3160 {
3161         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3162                 if (sel_tgt->family == sel_cmp->family &&
3163                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3164                                     sel_cmp->family) &&
3165                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3166                                     sel_cmp->family) &&
3167                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3168                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3169                         return true;
3170                 }
3171         } else {
3172                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3173                         return true;
3174                 }
3175         }
3176         return false;
3177 }
3178
3179 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3180                                                     u8 dir, u8 type, struct net *net)
3181 {
3182         struct xfrm_policy *pol, *ret = NULL;
3183         struct hlist_head *chain;
3184         u32 priority = ~0U;
3185
3186         read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3187         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3188         hlist_for_each_entry(pol, chain, bydst) {
3189                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3190                     pol->type == type) {
3191                         ret = pol;
3192                         priority = ret->priority;
3193                         break;
3194                 }
3195         }
3196         chain = &net->xfrm.policy_inexact[dir];
3197         hlist_for_each_entry(pol, chain, bydst) {
3198                 if ((pol->priority >= priority) && ret)
3199                         break;
3200
3201                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3202                     pol->type == type) {
3203                         ret = pol;
3204                         break;
3205                 }
3206         }
3207
3208         xfrm_pol_hold(ret);
3209
3210         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3211
3212         return ret;
3213 }
3214
3215 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3216 {
3217         int match = 0;
3218
3219         if (t->mode == m->mode && t->id.proto == m->proto &&
3220             (m->reqid == 0 || t->reqid == m->reqid)) {
3221                 switch (t->mode) {
3222                 case XFRM_MODE_TUNNEL:
3223                 case XFRM_MODE_BEET:
3224                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3225                                             m->old_family) &&
3226                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3227                                             m->old_family)) {
3228                                 match = 1;
3229                         }
3230                         break;
3231                 case XFRM_MODE_TRANSPORT:
3232                         /* in case of transport mode, template does not store
3233                            any IP addresses, hence we just compare mode and
3234                            protocol */
3235                         match = 1;
3236                         break;
3237                 default:
3238                         break;
3239                 }
3240         }
3241         return match;
3242 }
3243
3244 /* update endpoint address(es) of template(s) */
3245 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3246                                struct xfrm_migrate *m, int num_migrate)
3247 {
3248         struct xfrm_migrate *mp;
3249         int i, j, n = 0;
3250
3251         write_lock_bh(&pol->lock);
3252         if (unlikely(pol->walk.dead)) {
3253                 /* target policy has been deleted */
3254                 write_unlock_bh(&pol->lock);
3255                 return -ENOENT;
3256         }
3257
3258         for (i = 0; i < pol->xfrm_nr; i++) {
3259                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3260                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3261                                 continue;
3262                         n++;
3263                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3264                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3265                                 continue;
3266                         /* update endpoints */
3267                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3268                                sizeof(pol->xfrm_vec[i].id.daddr));
3269                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3270                                sizeof(pol->xfrm_vec[i].saddr));
3271                         pol->xfrm_vec[i].encap_family = mp->new_family;
3272                         /* flush bundles */
3273                         atomic_inc(&pol->genid);
3274                 }
3275         }
3276
3277         write_unlock_bh(&pol->lock);
3278
3279         if (!n)
3280                 return -ENODATA;
3281
3282         return 0;
3283 }
3284
3285 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3286 {
3287         int i, j;
3288
3289         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3290                 return -EINVAL;
3291
3292         for (i = 0; i < num_migrate; i++) {
3293                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3294                                     m[i].old_family) &&
3295                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3296                                     m[i].old_family))
3297                         return -EINVAL;
3298                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3299                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3300                         return -EINVAL;
3301
3302                 /* check if there is any duplicated entry */
3303                 for (j = i + 1; j < num_migrate; j++) {
3304                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3305                                     sizeof(m[i].old_daddr)) &&
3306                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3307                                     sizeof(m[i].old_saddr)) &&
3308                             m[i].proto == m[j].proto &&
3309                             m[i].mode == m[j].mode &&
3310                             m[i].reqid == m[j].reqid &&
3311                             m[i].old_family == m[j].old_family)
3312                                 return -EINVAL;
3313                 }
3314         }
3315
3316         return 0;
3317 }
3318
3319 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3320                  struct xfrm_migrate *m, int num_migrate,
3321                  struct xfrm_kmaddress *k, struct net *net)
3322 {
3323         int i, err, nx_cur = 0, nx_new = 0;
3324         struct xfrm_policy *pol = NULL;
3325         struct xfrm_state *x, *xc;
3326         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3327         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3328         struct xfrm_migrate *mp;
3329
3330         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3331                 goto out;
3332
3333         /* Stage 1 - find policy */
3334         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3335                 err = -ENOENT;
3336                 goto out;
3337         }
3338
3339         /* Stage 2 - find and update state(s) */
3340         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3341                 if ((x = xfrm_migrate_state_find(mp, net))) {
3342                         x_cur[nx_cur] = x;
3343                         nx_cur++;
3344                         if ((xc = xfrm_state_migrate(x, mp))) {
3345                                 x_new[nx_new] = xc;
3346                                 nx_new++;
3347                         } else {
3348                                 err = -ENODATA;
3349                                 goto restore_state;
3350                         }
3351                 }
3352         }
3353
3354         /* Stage 3 - update policy */
3355         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3356                 goto restore_state;
3357
3358         /* Stage 4 - delete old state(s) */
3359         if (nx_cur) {
3360                 xfrm_states_put(x_cur, nx_cur);
3361                 xfrm_states_delete(x_cur, nx_cur);
3362         }
3363
3364         /* Stage 5 - announce */
3365         km_migrate(sel, dir, type, m, num_migrate, k);
3366
3367         xfrm_pol_put(pol);
3368
3369         return 0;
3370 out:
3371         return err;
3372
3373 restore_state:
3374         if (pol)
3375                 xfrm_pol_put(pol);
3376         if (nx_cur)
3377                 xfrm_states_put(x_cur, nx_cur);
3378         if (nx_new)
3379                 xfrm_states_delete(x_new, nx_new);
3380
3381         return err;
3382 }
3383 EXPORT_SYMBOL(xfrm_migrate);
3384 #endif