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, int tos,
 119                                                   const xfrm_address_t *saddr,
 120                                                   const xfrm_address_t *daddr,
 121                                                   int family)
 122 {
 123         struct xfrm_policy_afinfo *afinfo;
 124         struct dst_entry *dst;
 125
 126         afinfo = xfrm_policy_get_afinfo(family);
 127         if (unlikely(afinfo == NULL))
 128                 return ERR_PTR(-EAFNOSUPPORT);
 129
 130         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 131
 132         xfrm_policy_put_afinfo(afinfo);
 133
 134         return dst;
 135 }
 136
 137 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 138                                                 xfrm_address_t *prev_saddr,
 139                                                 xfrm_address_t *prev_daddr,
 140                                                 int family)
 141 {
 142         struct net *net = xs_net(x);
 143         xfrm_address_t *saddr = &x->props.saddr;
 144         xfrm_address_t *daddr = &x->id.daddr;
 145         struct dst_entry *dst;
 146
 147         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 148                 saddr = x->coaddr;
 149                 daddr = prev_daddr;
 150         }
 151         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 152                 saddr = prev_saddr;
 153                 daddr = x->coaddr;
 154         }
 155
 156         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 157
 158         if (!IS_ERR(dst)) {
 159                 if (prev_saddr != saddr)
 160                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 161                 if (prev_daddr != daddr)
 162                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 163         }
 164
 165         return dst;
 166 }
 167
 168 static inline unsigned long make_jiffies(long secs)
 169 {
 170         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 171                 return MAX_SCHEDULE_TIMEOUT-1;
 172         else
 173                 return secs*HZ;
 174 }
 175
 176 static void xfrm_policy_timer(unsigned long data)
 177 {
 178         struct xfrm_policy *xp = (struct xfrm_policy *)data;
 179         unsigned long now = get_seconds();
 180         long next = LONG_MAX;
 181         int warn = 0;
 182         int dir;
 183
 184         read_lock(&xp->lock);
 185
 186         if (unlikely(xp->walk.dead))
 187                 goto out;
 188
 189         dir = xfrm_policy_id2dir(xp->index);
 190
 191         if (xp->lft.hard_add_expires_seconds) {
 192                 long tmo = xp->lft.hard_add_expires_seconds +
 193                         xp->curlft.add_time - now;
 194                 if (tmo <= 0)
 195                         goto expired;
 196                 if (tmo < next)
 197                         next = tmo;
 198         }
 199         if (xp->lft.hard_use_expires_seconds) {
 200                 long tmo = xp->lft.hard_use_expires_seconds +
 201                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 202                 if (tmo <= 0)
 203                         goto expired;
 204                 if (tmo < next)
 205                         next = tmo;
 206         }
 207         if (xp->lft.soft_add_expires_seconds) {
 208                 long tmo = xp->lft.soft_add_expires_seconds +
 209                         xp->curlft.add_time - now;
 210                 if (tmo <= 0) {
 211                         warn = 1;
 212                         tmo = XFRM_KM_TIMEOUT;
 213                 }
 214                 if (tmo < next)
 215                         next = tmo;
 216         }
 217         if (xp->lft.soft_use_expires_seconds) {
 218                 long tmo = xp->lft.soft_use_expires_seconds +
 219                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 220                 if (tmo <= 0) {
 221                         warn = 1;
 222                         tmo = XFRM_KM_TIMEOUT;
 223                 }
 224                 if (tmo < next)
 225                         next = tmo;
 226         }
 227
 228         if (warn)
 229                 km_policy_expired(xp, dir, 0, 0);
 230         if (next != LONG_MAX &&
 231             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 232                 xfrm_pol_hold(xp);
 233
 234 out:
 235         read_unlock(&xp->lock);
 236         xfrm_pol_put(xp);
 237         return;
 238
 239 expired:
 240         read_unlock(&xp->lock);
 241         if (!xfrm_policy_delete(xp, dir))
 242                 km_policy_expired(xp, dir, 1, 0);
 243         xfrm_pol_put(xp);
 244 }
 245
 246 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 247 {
 248         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 249
 250         if (unlikely(pol->walk.dead))
 251                 flo = NULL;
 252         else
 253                 xfrm_pol_hold(pol);
 254
 255         return flo;
 256 }
 257
 258 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 259 {
 260         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 261
 262         return !pol->walk.dead;
 263 }
 264
 265 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 266 {
 267         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 268 }
 269
 270 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 271         .get = xfrm_policy_flo_get,
 272         .check = xfrm_policy_flo_check,
 273         .delete = xfrm_policy_flo_delete,
 274 };
 275
 276 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 277  * SPD calls.
 278  */
 279
 280 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 281 {
 282         struct xfrm_policy *policy;
 283
 284         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 285
 286         if (policy) {
 287                 write_pnet(&policy->xp_net, net);
 288                 INIT_LIST_HEAD(&policy->walk.all);
 289                 INIT_HLIST_NODE(&policy->bydst);
 290                 INIT_HLIST_NODE(&policy->byidx);
 291                 rwlock_init(&policy->lock);
 292                 atomic_set(&policy->refcnt, 1);
 293                 skb_queue_head_init(&policy->polq.hold_queue);
 294                 setup_timer(&policy->timer, xfrm_policy_timer,
 295                                 (unsigned long)policy);
 296                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 297                             (unsigned long)policy);
 298                 policy->flo.ops = &xfrm_policy_fc_ops;
 299         }
 300         return policy;
 301 }
 302 EXPORT_SYMBOL(xfrm_policy_alloc);
 303
 304 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 305
 306 void xfrm_policy_destroy(struct xfrm_policy *policy)
 307 {
 308         BUG_ON(!policy->walk.dead);
 309
 310         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 311                 BUG();
 312
 313         security_xfrm_policy_free(policy->security);
 314         kfree(policy);
 315 }
 316 EXPORT_SYMBOL(xfrm_policy_destroy);
 317
 318 /* Rule must be locked. Release descentant resources, announce
 319  * entry dead. The rule must be unlinked from lists to the moment.
 320  */
 321
 322 static void xfrm_policy_kill(struct xfrm_policy *policy)
 323 {
 324         policy->walk.dead = 1;
 325
 326         atomic_inc(&policy->genid);
 327
 328         if (del_timer(&policy->polq.hold_timer))
 329                 xfrm_pol_put(policy);
 330         skb_queue_purge(&policy->polq.hold_queue);
 331
 332         if (del_timer(&policy->timer))
 333                 xfrm_pol_put(policy);
 334
 335         xfrm_pol_put(policy);
 336 }
 337
 338 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 339
 340 static inline unsigned int idx_hash(struct net *net, u32 index)
 341 {
 342         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 343 }
 344
 345 /* calculate policy hash thresholds */
 346 static void __get_hash_thresh(struct net *net,
 347                               unsigned short family, int dir,
 348                               u8 *dbits, u8 *sbits)
 349 {
 350         switch (family) {
 351         case AF_INET:
 352                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 353                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 354                 break;
 355
 356         case AF_INET6:
 357                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 358                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 359                 break;
 360
 361         default:
 362                 *dbits = 0;
 363                 *sbits = 0;
 364         }
 365 }
 366
 367 static struct hlist_head *policy_hash_bysel(struct net *net,
 368                                             const struct xfrm_selector *sel,
 369                                             unsigned short family, int dir)
 370 {
 371         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 372         unsigned int hash;
 373         u8 dbits;
 374         u8 sbits;
 375
 376         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 377         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 378
 379         return (hash == hmask + 1 ?
 380                 &net->xfrm.policy_inexact[dir] :
 381                 net->xfrm.policy_bydst[dir].table + hash);
 382 }
 383
 384 static struct hlist_head *policy_hash_direct(struct net *net,
 385                                              const xfrm_address_t *daddr,
 386                                              const xfrm_address_t *saddr,
 387                                              unsigned short family, int dir)
 388 {
 389         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 390         unsigned int hash;
 391         u8 dbits;
 392         u8 sbits;
 393
 394         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 395         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 396
 397         return net->xfrm.policy_bydst[dir].table + hash;
 398 }
 399
 400 static void xfrm_dst_hash_transfer(struct net *net,
 401                                    struct hlist_head *list,
 402                                    struct hlist_head *ndsttable,
 403                                    unsigned int nhashmask,
 404                                    int dir)
 405 {
 406         struct hlist_node *tmp, *entry0 = NULL;
 407         struct xfrm_policy *pol;
 408         unsigned int h0 = 0;
 409         u8 dbits;
 410         u8 sbits;
 411
 412 redo:
 413         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 414                 unsigned int h;
 415
 416                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 417                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 418                                 pol->family, nhashmask, dbits, sbits);
 419                 if (!entry0) {
 420                         hlist_del(&pol->bydst);
 421                         hlist_add_head(&pol->bydst, ndsttable+h);
 422                         h0 = h;
 423                 } else {
 424                         if (h != h0)
 425                                 continue;
 426                         hlist_del(&pol->bydst);
 427                         hlist_add_behind(&pol->bydst, entry0);
 428                 }
 429                 entry0 = &pol->bydst;
 430         }
 431         if (!hlist_empty(list)) {
 432                 entry0 = NULL;
 433                 goto redo;
 434         }
 435 }
 436
 437 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 438                                    struct hlist_head *nidxtable,
 439                                    unsigned int nhashmask)
 440 {
 441         struct hlist_node *tmp;
 442         struct xfrm_policy *pol;
 443
 444         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 445                 unsigned int h;
 446
 447                 h = __idx_hash(pol->index, nhashmask);
 448                 hlist_add_head(&pol->byidx, nidxtable+h);
 449         }
 450 }
 451
 452 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 453 {
 454         return ((old_hmask + 1) << 1) - 1;
 455 }
 456
 457 static void xfrm_bydst_resize(struct net *net, int dir)
 458 {
 459         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 460         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 461         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 462         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 463         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 464         int i;
 465
 466         if (!ndst)
 467                 return;
 468
 469         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 470
 471         for (i = hmask; i >= 0; i--)
 472                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 473
 474         net->xfrm.policy_bydst[dir].table = ndst;
 475         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 476
 477         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 478
 479         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 480 }
 481
 482 static void xfrm_byidx_resize(struct net *net, int total)
 483 {
 484         unsigned int hmask = net->xfrm.policy_idx_hmask;
 485         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 486         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 487         struct hlist_head *oidx = net->xfrm.policy_byidx;
 488         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 489         int i;
 490
 491         if (!nidx)
 492                 return;
 493
 494         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 495
 496         for (i = hmask; i >= 0; i--)
 497                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 498
 499         net->xfrm.policy_byidx = nidx;
 500         net->xfrm.policy_idx_hmask = nhashmask;
 501
 502         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 503
 504         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 505 }
 506
 507 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 508 {
 509         unsigned int cnt = net->xfrm.policy_count[dir];
 510         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 511
 512         if (total)
 513                 *total += cnt;
 514
 515         if ((hmask + 1) < xfrm_policy_hashmax &&
 516             cnt > hmask)
 517                 return 1;
 518
 519         return 0;
 520 }
 521
 522 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 523 {
 524         unsigned int hmask = net->xfrm.policy_idx_hmask;
 525
 526         if ((hmask + 1) < xfrm_policy_hashmax &&
 527             total > hmask)
 528                 return 1;
 529
 530         return 0;
 531 }
 532
 533 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 534 {
 535         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 536         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 537         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 538         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 539         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 540         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 541         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 542         si->spdhcnt = net->xfrm.policy_idx_hmask;
 543         si->spdhmcnt = xfrm_policy_hashmax;
 544         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
 545 }
 546 EXPORT_SYMBOL(xfrm_spd_getinfo);
 547
 548 static DEFINE_MUTEX(hash_resize_mutex);
 549 static void xfrm_hash_resize(struct work_struct *work)
 550 {
 551         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 552         int dir, total;
 553
 554         mutex_lock(&hash_resize_mutex);
 555
 556         total = 0;
 557         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 558                 if (xfrm_bydst_should_resize(net, dir, &total))
 559                         xfrm_bydst_resize(net, dir);
 560         }
 561         if (xfrm_byidx_should_resize(net, total))
 562                 xfrm_byidx_resize(net, total);
 563
 564         mutex_unlock(&hash_resize_mutex);
 565 }
 566
 567 static void xfrm_hash_rebuild(struct work_struct *work)
 568 {
 569         struct net *net = container_of(work, struct net,
 570                                        xfrm.policy_hthresh.work);
 571         unsigned int hmask;
 572         struct xfrm_policy *pol;
 573         struct xfrm_policy *policy;
 574         struct hlist_head *chain;
 575         struct hlist_head *odst;
 576         struct hlist_node *newpos;
 577         int i;
 578         int dir;
 579         unsigned seq;
 580         u8 lbits4, rbits4, lbits6, rbits6;
 581
 582         mutex_lock(&hash_resize_mutex);
 583
 584         /* read selector prefixlen thresholds */
 585         do {
 586                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
 587
 588                 lbits4 = net->xfrm.policy_hthresh.lbits4;
 589                 rbits4 = net->xfrm.policy_hthresh.rbits4;
 590                 lbits6 = net->xfrm.policy_hthresh.lbits6;
 591                 rbits6 = net->xfrm.policy_hthresh.rbits6;
 592         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
 593
 594         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 595
 596         /* reset the bydst and inexact table in all directions */
 597         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 598                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
 599                 hmask = net->xfrm.policy_bydst[dir].hmask;
 600                 odst = net->xfrm.policy_bydst[dir].table;
 601                 for (i = hmask; i >= 0; i--)
 602                         INIT_HLIST_HEAD(odst + i);
 603                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
 604                         /* dir out => dst = remote, src = local */
 605                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
 606                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
 607                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
 608                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
 609                 } else {
 610                         /* dir in/fwd => dst = local, src = remote */
 611                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
 612                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
 613                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
 614                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
 615                 }
 616         }
 617
 618         /* re-insert all policies by order of creation */
 619         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 620                 newpos = NULL;
 621                 chain = policy_hash_bysel(net, &policy->selector,
 622                                           policy->family,
 623                                           xfrm_policy_id2dir(policy->index));
 624                 hlist_for_each_entry(pol, chain, bydst) {
 625                         if (policy->priority >= pol->priority)
 626                                 newpos = &pol->bydst;
 627                         else
 628                                 break;
 629                 }
 630                 if (newpos)
 631                         hlist_add_behind(&policy->bydst, newpos);
 632                 else
 633                         hlist_add_head(&policy->bydst, chain);
 634         }
 635
 636         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 637
 638         mutex_unlock(&hash_resize_mutex);
 639 }
 640
 641 void xfrm_policy_hash_rebuild(struct net *net)
 642 {
 643         schedule_work(&net->xfrm.policy_hthresh.work);
 644 }
 645 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 646
 647 /* Generate new index... KAME seems to generate them ordered by cost
 648  * of an absolute inpredictability of ordering of rules. This will not pass. */
 649 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 650 {
 651         static u32 idx_generator;
 652
 653         for (;;) {
 654                 struct hlist_head *list;
 655                 struct xfrm_policy *p;
 656                 u32 idx;
 657                 int found;
 658
 659                 if (!index) {
 660                         idx = (idx_generator | dir);
 661                         idx_generator += 8;
 662                 } else {
 663                         idx = index;
 664                         index = 0;
 665                 }
 666
 667                 if (idx == 0)
 668                         idx = 8;
 669                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 670                 found = 0;
 671                 hlist_for_each_entry(p, list, byidx) {
 672                         if (p->index == idx) {
 673                                 found = 1;
 674                                 break;
 675                         }
 676                 }
 677                 if (!found)
 678                         return idx;
 679         }
 680 }
 681
 682 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 683 {
 684         u32 *p1 = (u32 *) s1;
 685         u32 *p2 = (u32 *) s2;
 686         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 687         int i;
 688
 689         for (i = 0; i < len; i++) {
 690                 if (p1[i] != p2[i])
 691                         return 1;
 692         }
 693
 694         return 0;
 695 }
 696
 697 static void xfrm_policy_requeue(struct xfrm_policy *old,
 698                                 struct xfrm_policy *new)
 699 {
 700         struct xfrm_policy_queue *pq = &old->polq;
 701         struct sk_buff_head list;
 702
 703         if (skb_queue_empty(&pq->hold_queue))
 704                 return;
 705
 706         __skb_queue_head_init(&list);
 707
 708         spin_lock_bh(&pq->hold_queue.lock);
 709         skb_queue_splice_init(&pq->hold_queue, &list);
 710         if (del_timer(&pq->hold_timer))
 711                 xfrm_pol_put(old);
 712         spin_unlock_bh(&pq->hold_queue.lock);
 713
 714         pq = &new->polq;
 715
 716         spin_lock_bh(&pq->hold_queue.lock);
 717         skb_queue_splice(&list, &pq->hold_queue);
 718         pq->timeout = XFRM_QUEUE_TMO_MIN;
 719         if (!mod_timer(&pq->hold_timer, jiffies))
 720                 xfrm_pol_hold(new);
 721         spin_unlock_bh(&pq->hold_queue.lock);
 722 }
 723
 724 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 725                                    struct xfrm_policy *pol)
 726 {
 727         u32 mark = policy->mark.v & policy->mark.m;
 728
 729         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 730                 return true;
 731
 732         if ((mark & pol->mark.m) == pol->mark.v &&
 733             policy->priority == pol->priority)
 734                 return true;
 735
 736         return false;
 737 }
 738
 739 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 740 {
 741         struct net *net = xp_net(policy);
 742         struct xfrm_policy *pol;
 743         struct xfrm_policy *delpol;
 744         struct hlist_head *chain;
 745         struct hlist_node *newpos;
 746
 747         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 748         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 749         delpol = NULL;
 750         newpos = NULL;
 751         hlist_for_each_entry(pol, chain, bydst) {
 752                 if (pol->type == policy->type &&
 753                     !selector_cmp(&pol->selector, &policy->selector) &&
 754                     xfrm_policy_mark_match(policy, pol) &&
 755                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 756                     !WARN_ON(delpol)) {
 757                         if (excl) {
 758                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 759                                 return -EEXIST;
 760                         }
 761                         delpol = pol;
 762                         if (policy->priority > pol->priority)
 763                                 continue;
 764                 } else if (policy->priority >= pol->priority) {
 765                         newpos = &pol->bydst;
 766                         continue;
 767                 }
 768                 if (delpol)
 769                         break;
 770         }
 771         if (newpos)
 772                 hlist_add_behind(&policy->bydst, newpos);
 773         else
 774                 hlist_add_head(&policy->bydst, chain);
 775         __xfrm_policy_link(policy, dir);
 776         atomic_inc(&net->xfrm.flow_cache_genid);
 777
 778         /* After previous checking, family can either be AF_INET or AF_INET6 */
 779         if (policy->family == AF_INET)
 780                 rt_genid_bump_ipv4(net);
 781         else
 782                 rt_genid_bump_ipv6(net);
 783
 784         if (delpol) {
 785                 xfrm_policy_requeue(delpol, policy);
 786                 __xfrm_policy_unlink(delpol, dir);
 787         }
 788         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 789         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 790         policy->curlft.add_time = get_seconds();
 791         policy->curlft.use_time = 0;
 792         if (!mod_timer(&policy->timer, jiffies + HZ))
 793                 xfrm_pol_hold(policy);
 794         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 795
 796         if (delpol)
 797                 xfrm_policy_kill(delpol);
 798         else if (xfrm_bydst_should_resize(net, dir, NULL))
 799                 schedule_work(&net->xfrm.policy_hash_work);
 800
 801         return 0;
 802 }
 803 EXPORT_SYMBOL(xfrm_policy_insert);
 804
 805 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 806                                           int dir, struct xfrm_selector *sel,
 807                                           struct xfrm_sec_ctx *ctx, int delete,
 808                                           int *err)
 809 {
 810         struct xfrm_policy *pol, *ret;
 811         struct hlist_head *chain;
 812
 813         *err = 0;
 814         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 815         chain = policy_hash_bysel(net, sel, sel->family, dir);
 816         ret = NULL;
 817         hlist_for_each_entry(pol, chain, bydst) {
 818                 if (pol->type == type &&
 819                     (mark & pol->mark.m) == pol->mark.v &&
 820                     !selector_cmp(sel, &pol->selector) &&
 821                     xfrm_sec_ctx_match(ctx, pol->security)) {
 822                         xfrm_pol_hold(pol);
 823                         if (delete) {
 824                                 *err = security_xfrm_policy_delete(
 825                                                                 pol->security);
 826                                 if (*err) {
 827                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 828                                         return pol;
 829                                 }
 830                                 __xfrm_policy_unlink(pol, dir);
 831                         }
 832                         ret = pol;
 833                         break;
 834                 }
 835         }
 836         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 837
 838         if (ret && delete)
 839                 xfrm_policy_kill(ret);
 840         return ret;
 841 }
 842 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 843
 844 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 845                                      int dir, u32 id, int delete, int *err)
 846 {
 847         struct xfrm_policy *pol, *ret;
 848         struct hlist_head *chain;
 849
 850         *err = -ENOENT;
 851         if (xfrm_policy_id2dir(id) != dir)
 852                 return NULL;
 853
 854         *err = 0;
 855         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 856         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 857         ret = NULL;
 858         hlist_for_each_entry(pol, chain, byidx) {
 859                 if (pol->type == type && pol->index == id &&
 860                     (mark & pol->mark.m) == pol->mark.v) {
 861                         xfrm_pol_hold(pol);
 862                         if (delete) {
 863                                 *err = security_xfrm_policy_delete(
 864                                                                 pol->security);
 865                                 if (*err) {
 866                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 867                                         return pol;
 868                                 }
 869                                 __xfrm_policy_unlink(pol, dir);
 870                         }
 871                         ret = pol;
 872                         break;
 873                 }
 874         }
 875         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 876
 877         if (ret && delete)
 878                 xfrm_policy_kill(ret);
 879         return ret;
 880 }
 881 EXPORT_SYMBOL(xfrm_policy_byid);
 882
 883 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 884 static inline int
 885 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 886 {
 887         int dir, err = 0;
 888
 889         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 890                 struct xfrm_policy *pol;
 891                 int i;
 892
 893                 hlist_for_each_entry(pol,
 894                                      &net->xfrm.policy_inexact[dir], bydst) {
 895                         if (pol->type != type)
 896                                 continue;
 897                         err = security_xfrm_policy_delete(pol->security);
 898                         if (err) {
 899                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 900                                 return err;
 901                         }
 902                 }
 903                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 904                         hlist_for_each_entry(pol,
 905                                              net->xfrm.policy_bydst[dir].table + i,
 906                                              bydst) {
 907                                 if (pol->type != type)
 908                                         continue;
 909                                 err = security_xfrm_policy_delete(
 910                                                                 pol->security);
 911                                 if (err) {
 912                                         xfrm_audit_policy_delete(pol, 0,
 913                                                                  task_valid);
 914                                         return err;
 915                                 }
 916                         }
 917                 }
 918         }
 919         return err;
 920 }
 921 #else
 922 static inline int
 923 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 924 {
 925         return 0;
 926 }
 927 #endif
 928
 929 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 930 {
 931         int dir, err = 0, cnt = 0;
 932
 933         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 934
 935         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 936         if (err)
 937                 goto out;
 938
 939         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 940                 struct xfrm_policy *pol;
 941                 int i;
 942
 943         again1:
 944                 hlist_for_each_entry(pol,
 945                                      &net->xfrm.policy_inexact[dir], bydst) {
 946                         if (pol->type != type)
 947                                 continue;
 948                         __xfrm_policy_unlink(pol, dir);
 949                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 950                         cnt++;
 951
 952                         xfrm_audit_policy_delete(pol, 1, task_valid);
 953
 954                         xfrm_policy_kill(pol);
 955
 956                         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 957                         goto again1;
 958                 }
 959
 960                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 961         again2:
 962                         hlist_for_each_entry(pol,
 963                                              net->xfrm.policy_bydst[dir].table + i,
 964                                              bydst) {
 965                                 if (pol->type != type)
 966                                         continue;
 967                                 __xfrm_policy_unlink(pol, dir);
 968                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 969                                 cnt++;
 970
 971                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 972                                 xfrm_policy_kill(pol);
 973
 974                                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
 975                                 goto again2;
 976                         }
 977                 }
 978
 979         }
 980         if (!cnt)
 981                 err = -ESRCH;
 982 out:
 983         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 984         return err;
 985 }
 986 EXPORT_SYMBOL(xfrm_policy_flush);
 987
 988 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 989                      int (*func)(struct xfrm_policy *, int, int, void*),
 990                      void *data)
 991 {
 992         struct xfrm_policy *pol;
 993         struct xfrm_policy_walk_entry *x;
 994         int error = 0;
 995
 996         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 997             walk->type != XFRM_POLICY_TYPE_ANY)
 998                 return -EINVAL;
 999
1000         if (list_empty(&walk->walk.all) && walk->seq != 0)
1001                 return 0;
1002
1003         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1004         if (list_empty(&walk->walk.all))
1005                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1006         else
1007                 x = list_first_entry(&walk->walk.all,
1008                                      struct xfrm_policy_walk_entry, all);
1009
1010         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1011                 if (x->dead)
1012                         continue;
1013                 pol = container_of(x, struct xfrm_policy, walk);
1014                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1015                     walk->type != pol->type)
1016                         continue;
1017                 error = func(pol, xfrm_policy_id2dir(pol->index),
1018                              walk->seq, data);
1019                 if (error) {
1020                         list_move_tail(&walk->walk.all, &x->all);
1021                         goto out;
1022                 }
1023                 walk->seq++;
1024         }
1025         if (walk->seq == 0) {
1026                 error = -ENOENT;
1027                 goto out;
1028         }
1029         list_del_init(&walk->walk.all);
1030 out:
1031         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1032         return error;
1033 }
1034 EXPORT_SYMBOL(xfrm_policy_walk);
1035
1036 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1037 {
1038         INIT_LIST_HEAD(&walk->walk.all);
1039         walk->walk.dead = 1;
1040         walk->type = type;
1041         walk->seq = 0;
1042 }
1043 EXPORT_SYMBOL(xfrm_policy_walk_init);
1044
1045 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1046 {
1047         if (list_empty(&walk->walk.all))
1048                 return;
1049
1050         write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1051         list_del(&walk->walk.all);
1052         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1053 }
1054 EXPORT_SYMBOL(xfrm_policy_walk_done);
1055
1056 /*
1057  * Find policy to apply to this flow.
1058  *
1059  * Returns 0 if policy found, else an -errno.
1060  */
1061 static int xfrm_policy_match(const struct xfrm_policy *pol,
1062                              const struct flowi *fl,
1063                              u8 type, u16 family, int dir)
1064 {
1065         const struct xfrm_selector *sel = &pol->selector;
1066         int ret = -ESRCH;
1067         bool match;
1068
1069         if (pol->family != family ||
1070             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1071             pol->type != type)
1072                 return ret;
1073
1074         match = xfrm_selector_match(sel, fl, family);
1075         if (match)
1076                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1077                                                   dir);
1078
1079         return ret;
1080 }
1081
1082 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1083                                                      const struct flowi *fl,
1084                                                      u16 family, u8 dir)
1085 {
1086         int err;
1087         struct xfrm_policy *pol, *ret;
1088         const xfrm_address_t *daddr, *saddr;
1089         struct hlist_head *chain;
1090         u32 priority = ~0U;
1091
1092         daddr = xfrm_flowi_daddr(fl, family);
1093         saddr = xfrm_flowi_saddr(fl, family);
1094         if (unlikely(!daddr || !saddr))
1095                 return NULL;
1096
1097         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1098         chain = policy_hash_direct(net, daddr, saddr, family, dir);
1099         ret = NULL;
1100         hlist_for_each_entry(pol, chain, bydst) {
1101                 err = xfrm_policy_match(pol, fl, type, family, dir);
1102                 if (err) {
1103                         if (err == -ESRCH)
1104                                 continue;
1105                         else {
1106                                 ret = ERR_PTR(err);
1107                                 goto fail;
1108                         }
1109                 } else {
1110                         ret = pol;
1111                         priority = ret->priority;
1112                         break;
1113                 }
1114         }
1115         chain = &net->xfrm.policy_inexact[dir];
1116         hlist_for_each_entry(pol, chain, bydst) {
1117                 if ((pol->priority >= priority) && ret)
1118                         break;
1119
1120                 err = xfrm_policy_match(pol, fl, type, family, dir);
1121                 if (err) {
1122                         if (err == -ESRCH)
1123                                 continue;
1124                         else {
1125                                 ret = ERR_PTR(err);
1126                                 goto fail;
1127                         }
1128                 } else {
1129                         ret = pol;
1130                         break;
1131                 }
1132         }
1133
1134         xfrm_pol_hold(ret);
1135 fail:
1136         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1137
1138         return ret;
1139 }
1140
1141 static struct xfrm_policy *
1142 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1143 {
1144 #ifdef CONFIG_XFRM_SUB_POLICY
1145         struct xfrm_policy *pol;
1146
1147         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1148         if (pol != NULL)
1149                 return pol;
1150 #endif
1151         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1152 }
1153
1154 static int flow_to_policy_dir(int dir)
1155 {
1156         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1157             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1158             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1159                 return dir;
1160
1161         switch (dir) {
1162         default:
1163         case FLOW_DIR_IN:
1164                 return XFRM_POLICY_IN;
1165         case FLOW_DIR_OUT:
1166                 return XFRM_POLICY_OUT;
1167         case FLOW_DIR_FWD:
1168                 return XFRM_POLICY_FWD;
1169         }
1170 }
1171
1172 static struct flow_cache_object *
1173 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1174                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1175 {
1176         struct xfrm_policy *pol;
1177
1178         if (old_obj)
1179                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1180
1181         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1182         if (IS_ERR_OR_NULL(pol))
1183                 return ERR_CAST(pol);
1184
1185         /* Resolver returns two references:
1186          * one for cache and one for caller of flow_cache_lookup() */
1187         xfrm_pol_hold(pol);
1188
1189         return &pol->flo;
1190 }
1191
1192 static inline int policy_to_flow_dir(int dir)
1193 {
1194         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1195             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1196             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1197                 return dir;
1198         switch (dir) {
1199         default:
1200         case XFRM_POLICY_IN:
1201                 return FLOW_DIR_IN;
1202         case XFRM_POLICY_OUT:
1203                 return FLOW_DIR_OUT;
1204         case XFRM_POLICY_FWD:
1205                 return FLOW_DIR_FWD;
1206         }
1207 }
1208
1209 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1210                                                  const struct flowi *fl)
1211 {
1212         struct xfrm_policy *pol;
1213         struct net *net = sock_net(sk);
1214
1215         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1216         if ((pol = sk->sk_policy[dir]) != NULL) {
1217                 bool match = xfrm_selector_match(&pol->selector, fl,
1218                                                  sk->sk_family);
1219                 int err = 0;
1220
1221                 if (match) {
1222                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1223                                 pol = NULL;
1224                                 goto out;
1225                         }
1226                         err = security_xfrm_policy_lookup(pol->security,
1227                                                       fl->flowi_secid,
1228                                                       policy_to_flow_dir(dir));
1229                         if (!err)
1230                                 xfrm_pol_hold(pol);
1231                         else if (err == -ESRCH)
1232                                 pol = NULL;
1233                         else
1234                                 pol = ERR_PTR(err);
1235                 } else
1236                         pol = NULL;
1237         }
1238 out:
1239         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1240         return pol;
1241 }
1242
1243 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1244 {
1245         struct net *net = xp_net(pol);
1246
1247         list_add(&pol->walk.all, &net->xfrm.policy_all);
1248         net->xfrm.policy_count[dir]++;
1249         xfrm_pol_hold(pol);
1250 }
1251
1252 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1253                                                 int dir)
1254 {
1255         struct net *net = xp_net(pol);
1256
1257         if (list_empty(&pol->walk.all))
1258                 return NULL;
1259
1260         /* Socket policies are not hashed. */
1261         if (!hlist_unhashed(&pol->bydst)) {
1262                 hlist_del(&pol->bydst);
1263                 hlist_del(&pol->byidx);
1264         }
1265
1266         list_del_init(&pol->walk.all);
1267         net->xfrm.policy_count[dir]--;
1268
1269         return pol;
1270 }
1271
1272 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1273 {
1274         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1275 }
1276
1277 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1278 {
1279         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1280 }
1281
1282 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1283 {
1284         struct net *net = xp_net(pol);
1285
1286         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1287         pol = __xfrm_policy_unlink(pol, dir);
1288         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1289         if (pol) {
1290                 xfrm_policy_kill(pol);
1291                 return 0;
1292         }
1293         return -ENOENT;
1294 }
1295 EXPORT_SYMBOL(xfrm_policy_delete);
1296
1297 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1298 {
1299         struct net *net = xp_net(pol);
1300         struct xfrm_policy *old_pol;
1301
1302 #ifdef CONFIG_XFRM_SUB_POLICY
1303         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1304                 return -EINVAL;
1305 #endif
1306
1307         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1308         old_pol = sk->sk_policy[dir];
1309         sk->sk_policy[dir] = pol;
1310         if (pol) {
1311                 pol->curlft.add_time = get_seconds();
1312                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1313                 xfrm_sk_policy_link(pol, dir);
1314         }
1315         if (old_pol) {
1316                 if (pol)
1317                         xfrm_policy_requeue(old_pol, pol);
1318
1319                 /* Unlinking succeeds always. This is the only function
1320                  * allowed to delete or replace socket policy.
1321                  */
1322                 xfrm_sk_policy_unlink(old_pol, dir);
1323         }
1324         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1325
1326         if (old_pol) {
1327                 xfrm_policy_kill(old_pol);
1328         }
1329         return 0;
1330 }
1331
1332 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1333 {
1334         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1335         struct net *net = xp_net(old);
1336
1337         if (newp) {
1338                 newp->selector = old->selector;
1339                 if (security_xfrm_policy_clone(old->security,
1340                                                &newp->security)) {
1341                         kfree(newp);
1342                         return NULL;  /* ENOMEM */
1343                 }
1344                 newp->lft = old->lft;
1345                 newp->curlft = old->curlft;
1346                 newp->mark = old->mark;
1347                 newp->action = old->action;
1348                 newp->flags = old->flags;
1349                 newp->xfrm_nr = old->xfrm_nr;
1350                 newp->index = old->index;
1351                 newp->type = old->type;
1352                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1353                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1354                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1355                 xfrm_sk_policy_link(newp, dir);
1356                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1357                 xfrm_pol_put(newp);
1358         }
1359         return newp;
1360 }
1361
1362 int __xfrm_sk_clone_policy(struct sock *sk)
1363 {
1364         struct xfrm_policy *p0 = sk->sk_policy[0],
1365                            *p1 = sk->sk_policy[1];
1366
1367         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1368         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1369                 return -ENOMEM;
1370         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1371                 return -ENOMEM;
1372         return 0;
1373 }
1374
1375 static int
1376 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1377                unsigned short family)
1378 {
1379         int err;
1380         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1381
1382         if (unlikely(afinfo == NULL))
1383                 return -EINVAL;
1384         err = afinfo->get_saddr(net, local, remote);
1385         xfrm_policy_put_afinfo(afinfo);
1386         return err;
1387 }
1388
1389 /* Resolve list of templates for the flow, given policy. */
1390
1391 static int
1392 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1393                       struct xfrm_state **xfrm, unsigned short family)
1394 {
1395         struct net *net = xp_net(policy);
1396         int nx;
1397         int i, error;
1398         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1399         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1400         xfrm_address_t tmp;
1401
1402         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1403                 struct xfrm_state *x;
1404                 xfrm_address_t *remote = daddr;
1405                 xfrm_address_t *local  = saddr;
1406                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1407
1408                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1409                     tmpl->mode == XFRM_MODE_BEET) {
1410                         remote = &tmpl->id.daddr;
1411                         local = &tmpl->saddr;
1412                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1413                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1414                                 if (error)
1415                                         goto fail;
1416                                 local = &tmp;
1417                         }
1418                 }
1419
1420                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1421
1422                 if (x && x->km.state == XFRM_STATE_VALID) {
1423                         xfrm[nx++] = x;
1424                         daddr = remote;
1425                         saddr = local;
1426                         continue;
1427                 }
1428                 if (x) {
1429                         error = (x->km.state == XFRM_STATE_ERROR ?
1430                                  -EINVAL : -EAGAIN);
1431                         xfrm_state_put(x);
1432                 } else if (error == -ESRCH) {
1433                         error = -EAGAIN;
1434                 }
1435
1436                 if (!tmpl->optional)
1437                         goto fail;
1438         }
1439         return nx;
1440
1441 fail:
1442         for (nx--; nx >= 0; nx--)
1443                 xfrm_state_put(xfrm[nx]);
1444         return error;
1445 }
1446
1447 static int
1448 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1449                   struct xfrm_state **xfrm, unsigned short family)
1450 {
1451         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1452         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1453         int cnx = 0;
1454         int error;
1455         int ret;
1456         int i;
1457
1458         for (i = 0; i < npols; i++) {
1459                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1460                         error = -ENOBUFS;
1461                         goto fail;
1462                 }
1463
1464                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1465                 if (ret < 0) {
1466                         error = ret;
1467                         goto fail;
1468                 } else
1469                         cnx += ret;
1470         }
1471
1472         /* found states are sorted for outbound processing */
1473         if (npols > 1)
1474                 xfrm_state_sort(xfrm, tpp, cnx, family);
1475
1476         return cnx;
1477
1478  fail:
1479         for (cnx--; cnx >= 0; cnx--)
1480                 xfrm_state_put(tpp[cnx]);
1481         return error;
1482
1483 }
1484
1485 /* Check that the bundle accepts the flow and its components are
1486  * still valid.
1487  */
1488
1489 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1490 {
1491         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1492         int tos;
1493
1494         if (!afinfo)
1495                 return -EINVAL;
1496
1497         tos = afinfo->get_tos(fl);
1498
1499         xfrm_policy_put_afinfo(afinfo);
1500
1501         return tos;
1502 }
1503
1504 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1505 {
1506         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1507         struct dst_entry *dst = &xdst->u.dst;
1508
1509         if (xdst->route == NULL) {
1510                 /* Dummy bundle - if it has xfrms we were not
1511                  * able to build bundle as template resolution failed.
1512                  * It means we need to try again resolving. */
1513                 if (xdst->num_xfrms > 0)
1514                         return NULL;
1515         } else if (dst->flags & DST_XFRM_QUEUE) {
1516                 return NULL;
1517         } else {
1518                 /* Real bundle */
1519                 if (stale_bundle(dst))
1520                         return NULL;
1521         }
1522
1523         dst_hold(dst);
1524         return flo;
1525 }
1526
1527 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1528 {
1529         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1530         struct dst_entry *dst = &xdst->u.dst;
1531
1532         if (!xdst->route)
1533                 return 0;
1534         if (stale_bundle(dst))
1535                 return 0;
1536
1537         return 1;
1538 }
1539
1540 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1541 {
1542         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1543         struct dst_entry *dst = &xdst->u.dst;
1544
1545         dst_free(dst);
1546 }
1547
1548 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1549         .get = xfrm_bundle_flo_get,
1550         .check = xfrm_bundle_flo_check,
1551         .delete = xfrm_bundle_flo_delete,
1552 };
1553
1554 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1555 {
1556         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1557         struct dst_ops *dst_ops;
1558         struct xfrm_dst *xdst;
1559
1560         if (!afinfo)
1561                 return ERR_PTR(-EINVAL);
1562
1563         switch (family) {
1564         case AF_INET:
1565                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1566                 break;
1567 #if IS_ENABLED(CONFIG_IPV6)
1568         case AF_INET6:
1569                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1570                 break;
1571 #endif
1572         default:
1573                 BUG();
1574         }
1575         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1576
1577         if (likely(xdst)) {
1578                 struct dst_entry *dst = &xdst->u.dst;
1579
1580                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1581                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1582                 if (afinfo->init_dst)
1583                         afinfo->init_dst(net, xdst);
1584         } else
1585                 xdst = ERR_PTR(-ENOBUFS);
1586
1587         xfrm_policy_put_afinfo(afinfo);
1588
1589         return xdst;
1590 }
1591
1592 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1593                                  int nfheader_len)
1594 {
1595         struct xfrm_policy_afinfo *afinfo =
1596                 xfrm_policy_get_afinfo(dst->ops->family);
1597         int err;
1598
1599         if (!afinfo)
1600                 return -EINVAL;
1601
1602         err = afinfo->init_path(path, dst, nfheader_len);
1603
1604         xfrm_policy_put_afinfo(afinfo);
1605
1606         return err;
1607 }
1608
1609 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1610                                 const struct flowi *fl)
1611 {
1612         struct xfrm_policy_afinfo *afinfo =
1613                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1614         int err;
1615
1616         if (!afinfo)
1617                 return -EINVAL;
1618
1619         err = afinfo->fill_dst(xdst, dev, fl);
1620
1621         xfrm_policy_put_afinfo(afinfo);
1622
1623         return err;
1624 }
1625
1626
1627 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1628  * all the metrics... Shortly, bundle a bundle.
1629  */
1630
1631 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1632                                             struct xfrm_state **xfrm, int nx,
1633                                             const struct flowi *fl,
1634                                             struct dst_entry *dst)
1635 {
1636         struct net *net = xp_net(policy);
1637         unsigned long now = jiffies;
1638         struct net_device *dev;
1639         struct xfrm_mode *inner_mode;
1640         struct dst_entry *dst_prev = NULL;
1641         struct dst_entry *dst0 = NULL;
1642         int i = 0;
1643         int err;
1644         int header_len = 0;
1645         int nfheader_len = 0;
1646         int trailer_len = 0;
1647         int tos;
1648         int family = policy->selector.family;
1649         xfrm_address_t saddr, daddr;
1650
1651         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1652
1653         tos = xfrm_get_tos(fl, family);
1654         err = tos;
1655         if (tos < 0)
1656                 goto put_states;
1657
1658         dst_hold(dst);
1659
1660         for (; i < nx; i++) {
1661                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1662                 struct dst_entry *dst1 = &xdst->u.dst;
1663
1664                 err = PTR_ERR(xdst);
1665                 if (IS_ERR(xdst)) {
1666                         dst_release(dst);
1667                         goto put_states;
1668                 }
1669
1670                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1671                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1672                                                         xfrm_af2proto(family));
1673                         if (!inner_mode) {
1674                                 err = -EAFNOSUPPORT;
1675                                 dst_release(dst);
1676                                 goto put_states;
1677                         }
1678                 } else
1679                         inner_mode = xfrm[i]->inner_mode;
1680
1681                 if (!dst_prev)
1682                         dst0 = dst1;
1683                 else {
1684                         dst_prev->child = dst_clone(dst1);
1685                         dst1->flags |= DST_NOHASH;
1686                 }
1687
1688                 xdst->route = dst;
1689                 dst_copy_metrics(dst1, dst);
1690
1691                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1692                         family = xfrm[i]->props.family;
1693                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1694                                               family);
1695                         err = PTR_ERR(dst);
1696                         if (IS_ERR(dst))
1697                                 goto put_states;
1698                 } else
1699                         dst_hold(dst);
1700
1701                 dst1->xfrm = xfrm[i];
1702                 xdst->xfrm_genid = xfrm[i]->genid;
1703
1704                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1705                 dst1->flags |= DST_HOST;
1706                 dst1->lastuse = now;
1707
1708                 dst1->input = dst_discard;
1709                 dst1->output = inner_mode->afinfo->output;
1710
1711                 dst1->next = dst_prev;
1712                 dst_prev = dst1;
1713
1714                 header_len += xfrm[i]->props.header_len;
1715                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1716                         nfheader_len += xfrm[i]->props.header_len;
1717                 trailer_len += xfrm[i]->props.trailer_len;
1718         }
1719
1720         dst_prev->child = dst;
1721         dst0->path = dst;
1722
1723         err = -ENODEV;
1724         dev = dst->dev;
1725         if (!dev)
1726                 goto free_dst;
1727
1728         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1729         xfrm_init_pmtu(dst_prev);
1730
1731         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1732                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1733
1734                 err = xfrm_fill_dst(xdst, dev, fl);
1735                 if (err)
1736                         goto free_dst;
1737
1738                 dst_prev->header_len = header_len;
1739                 dst_prev->trailer_len = trailer_len;
1740                 header_len -= xdst->u.dst.xfrm->props.header_len;
1741                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1742         }
1743
1744 out:
1745         return dst0;
1746
1747 put_states:
1748         for (; i < nx; i++)
1749                 xfrm_state_put(xfrm[i]);
1750 free_dst:
1751         if (dst0)
1752                 dst_free(dst0);
1753         dst0 = ERR_PTR(err);
1754         goto out;
1755 }
1756
1757 #ifdef CONFIG_XFRM_SUB_POLICY
1758 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1759 {
1760         if (!*target) {
1761                 *target = kmalloc(size, GFP_ATOMIC);
1762                 if (!*target)
1763                         return -ENOMEM;
1764         }
1765
1766         memcpy(*target, src, size);
1767         return 0;
1768 }
1769 #endif
1770
1771 static int xfrm_dst_update_parent(struct dst_entry *dst,
1772                                   const struct xfrm_selector *sel)
1773 {
1774 #ifdef CONFIG_XFRM_SUB_POLICY
1775         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1776         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1777                                    sel, sizeof(*sel));
1778 #else
1779         return 0;
1780 #endif
1781 }
1782
1783 static int xfrm_dst_update_origin(struct dst_entry *dst,
1784                                   const struct flowi *fl)
1785 {
1786 #ifdef CONFIG_XFRM_SUB_POLICY
1787         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1788         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1789 #else
1790         return 0;
1791 #endif
1792 }
1793
1794 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1795                                 struct xfrm_policy **pols,
1796                                 int *num_pols, int *num_xfrms)
1797 {
1798         int i;
1799
1800         if (*num_pols == 0 || !pols[0]) {
1801                 *num_pols = 0;
1802                 *num_xfrms = 0;
1803                 return 0;
1804         }
1805         if (IS_ERR(pols[0]))
1806                 return PTR_ERR(pols[0]);
1807
1808         *num_xfrms = pols[0]->xfrm_nr;
1809
1810 #ifdef CONFIG_XFRM_SUB_POLICY
1811         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1812             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1813                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1814                                                     XFRM_POLICY_TYPE_MAIN,
1815                                                     fl, family,
1816                                                     XFRM_POLICY_OUT);
1817                 if (pols[1]) {
1818                         if (IS_ERR(pols[1])) {
1819                                 xfrm_pols_put(pols, *num_pols);
1820                                 return PTR_ERR(pols[1]);
1821                         }
1822                         (*num_pols)++;
1823                         (*num_xfrms) += pols[1]->xfrm_nr;
1824                 }
1825         }
1826 #endif
1827         for (i = 0; i < *num_pols; i++) {
1828                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1829                         *num_xfrms = -1;
1830                         break;
1831                 }
1832         }
1833
1834         return 0;
1835
1836 }
1837
1838 static struct xfrm_dst *
1839 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1840                                const struct flowi *fl, u16 family,
1841                                struct dst_entry *dst_orig)
1842 {
1843         struct net *net = xp_net(pols[0]);
1844         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1845         struct dst_entry *dst;
1846         struct xfrm_dst *xdst;
1847         int err;
1848
1849         /* Try to instantiate a bundle */
1850         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1851         if (err <= 0) {
1852                 if (err != 0 && err != -EAGAIN)
1853                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1854                 return ERR_PTR(err);
1855         }
1856
1857         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1858         if (IS_ERR(dst)) {
1859                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1860                 return ERR_CAST(dst);
1861         }
1862
1863         xdst = (struct xfrm_dst *)dst;
1864         xdst->num_xfrms = err;
1865         if (num_pols > 1)
1866                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1867         else
1868                 err = xfrm_dst_update_origin(dst, fl);
1869         if (unlikely(err)) {
1870                 dst_free(dst);
1871                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1872                 return ERR_PTR(err);
1873         }
1874
1875         xdst->num_pols = num_pols;
1876         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1877         xdst->policy_genid = atomic_read(&pols[0]->genid);
1878
1879         return xdst;
1880 }
1881
1882 static void xfrm_policy_queue_process(unsigned long arg)
1883 {
1884         struct sk_buff *skb;
1885         struct sock *sk;
1886         struct dst_entry *dst;
1887         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1888         struct xfrm_policy_queue *pq = &pol->polq;
1889         struct flowi fl;
1890         struct sk_buff_head list;
1891
1892         spin_lock(&pq->hold_queue.lock);
1893         skb = skb_peek(&pq->hold_queue);
1894         if (!skb) {
1895                 spin_unlock(&pq->hold_queue.lock);
1896                 goto out;
1897         }
1898         dst = skb_dst(skb);
1899         sk = skb->sk;
1900         xfrm_decode_session(skb, &fl, dst->ops->family);
1901         spin_unlock(&pq->hold_queue.lock);
1902
1903         dst_hold(dst->path);
1904         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1905                           sk, 0);
1906         if (IS_ERR(dst))
1907                 goto purge_queue;
1908
1909         if (dst->flags & DST_XFRM_QUEUE) {
1910                 dst_release(dst);
1911
1912                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1913                         goto purge_queue;
1914
1915                 pq->timeout = pq->timeout << 1;
1916                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1917                         xfrm_pol_hold(pol);
1918         goto out;
1919         }
1920
1921         dst_release(dst);
1922
1923         __skb_queue_head_init(&list);
1924
1925         spin_lock(&pq->hold_queue.lock);
1926         pq->timeout = 0;
1927         skb_queue_splice_init(&pq->hold_queue, &list);
1928         spin_unlock(&pq->hold_queue.lock);
1929
1930         while (!skb_queue_empty(&list)) {
1931                 skb = __skb_dequeue(&list);
1932
1933                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1934                 dst_hold(skb_dst(skb)->path);
1935                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1936                                   &fl, skb->sk, 0);
1937                 if (IS_ERR(dst)) {
1938                         kfree_skb(skb);
1939                         continue;
1940                 }
1941
1942                 nf_reset(skb);
1943                 skb_dst_drop(skb);
1944                 skb_dst_set(skb, dst);
1945
1946                 dst_output(skb);
1947         }
1948
1949 out:
1950         xfrm_pol_put(pol);
1951         return;
1952
1953 purge_queue:
1954         pq->timeout = 0;
1955         skb_queue_purge(&pq->hold_queue);
1956         xfrm_pol_put(pol);
1957 }
1958
1959 static int xdst_queue_output(struct sock *sk, struct sk_buff *skb)
1960 {
1961         unsigned long sched_next;
1962         struct dst_entry *dst = skb_dst(skb);
1963         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1964         struct xfrm_policy *pol = xdst->pols[0];
1965         struct xfrm_policy_queue *pq = &pol->polq;
1966
1967         if (unlikely(skb_fclone_busy(sk, skb))) {
1968                 kfree_skb(skb);
1969                 return 0;
1970         }
1971
1972         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1973                 kfree_skb(skb);
1974                 return -EAGAIN;
1975         }
1976
1977         skb_dst_force(skb);
1978
1979         spin_lock_bh(&pq->hold_queue.lock);
1980
1981         if (!pq->timeout)
1982                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1983
1984         sched_next = jiffies + pq->timeout;
1985
1986         if (del_timer(&pq->hold_timer)) {
1987                 if (time_before(pq->hold_timer.expires, sched_next))
1988                         sched_next = pq->hold_timer.expires;
1989                 xfrm_pol_put(pol);
1990         }
1991
1992         __skb_queue_tail(&pq->hold_queue, skb);
1993         if (!mod_timer(&pq->hold_timer, sched_next))
1994                 xfrm_pol_hold(pol);
1995
1996         spin_unlock_bh(&pq->hold_queue.lock);
1997
1998         return 0;
1999 }
2000
2001 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2002                                                  struct xfrm_flo *xflo,
2003                                                  const struct flowi *fl,
2004                                                  int num_xfrms,
2005                                                  u16 family)
2006 {
2007         int err;
2008         struct net_device *dev;
2009         struct dst_entry *dst;
2010         struct dst_entry *dst1;
2011         struct xfrm_dst *xdst;
2012
2013         xdst = xfrm_alloc_dst(net, family);
2014         if (IS_ERR(xdst))
2015                 return xdst;
2016
2017         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2018             net->xfrm.sysctl_larval_drop ||
2019             num_xfrms <= 0)
2020                 return xdst;
2021
2022         dst = xflo->dst_orig;
2023         dst1 = &xdst->u.dst;
2024         dst_hold(dst);
2025         xdst->route = dst;
2026
2027         dst_copy_metrics(dst1, dst);
2028
2029         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2030         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2031         dst1->lastuse = jiffies;
2032
2033         dst1->input = dst_discard;
2034         dst1->output = xdst_queue_output;
2035
2036         dst_hold(dst);
2037         dst1->child = dst;
2038         dst1->path = dst;
2039
2040         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2041
2042         err = -ENODEV;
2043         dev = dst->dev;
2044         if (!dev)
2045                 goto free_dst;
2046
2047         err = xfrm_fill_dst(xdst, dev, fl);
2048         if (err)
2049                 goto free_dst;
2050
2051 out:
2052         return xdst;
2053
2054 free_dst:
2055         dst_release(dst1);
2056         xdst = ERR_PTR(err);
2057         goto out;
2058 }
2059
2060 static struct flow_cache_object *
2061 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2062                    struct flow_cache_object *oldflo, void *ctx)
2063 {
2064         struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2065         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2066         struct xfrm_dst *xdst, *new_xdst;
2067         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2068
2069         /* Check if the policies from old bundle are usable */
2070         xdst = NULL;
2071         if (oldflo) {
2072                 xdst = container_of(oldflo, struct xfrm_dst, flo);
2073                 num_pols = xdst->num_pols;
2074                 num_xfrms = xdst->num_xfrms;
2075                 pol_dead = 0;
2076                 for (i = 0; i < num_pols; i++) {
2077                         pols[i] = xdst->pols[i];
2078                         pol_dead |= pols[i]->walk.dead;
2079                 }
2080                 if (pol_dead) {
2081                         dst_free(&xdst->u.dst);
2082                         xdst = NULL;
2083                         num_pols = 0;
2084                         num_xfrms = 0;
2085                         oldflo = NULL;
2086                 }
2087         }
2088
2089         /* Resolve policies to use if we couldn't get them from
2090          * previous cache entry */
2091         if (xdst == NULL) {
2092                 num_pols = 1;
2093                 pols[0] = __xfrm_policy_lookup(net, fl, family,
2094                                                flow_to_policy_dir(dir));
2095                 err = xfrm_expand_policies(fl, family, pols,
2096                                            &num_pols, &num_xfrms);
2097                 if (err < 0)
2098                         goto inc_error;
2099                 if (num_pols == 0)
2100                         return NULL;
2101                 if (num_xfrms <= 0)
2102                         goto make_dummy_bundle;
2103         }
2104
2105         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2106                                                   xflo->dst_orig);
2107         if (IS_ERR(new_xdst)) {
2108                 err = PTR_ERR(new_xdst);
2109                 if (err != -EAGAIN)
2110                         goto error;
2111                 if (oldflo == NULL)
2112                         goto make_dummy_bundle;
2113                 dst_hold(&xdst->u.dst);
2114                 return oldflo;
2115         } else if (new_xdst == NULL) {
2116                 num_xfrms = 0;
2117                 if (oldflo == NULL)
2118                         goto make_dummy_bundle;
2119                 xdst->num_xfrms = 0;
2120                 dst_hold(&xdst->u.dst);
2121                 return oldflo;
2122         }
2123
2124         /* Kill the previous bundle */
2125         if (xdst) {
2126                 /* The policies were stolen for newly generated bundle */
2127                 xdst->num_pols = 0;
2128                 dst_free(&xdst->u.dst);
2129         }
2130
2131         /* Flow cache does not have reference, it dst_free()'s,
2132          * but we do need to return one reference for original caller */
2133         dst_hold(&new_xdst->u.dst);
2134         return &new_xdst->flo;
2135
2136 make_dummy_bundle:
2137         /* We found policies, but there's no bundles to instantiate:
2138          * either because the policy blocks, has no transformations or
2139          * we could not build template (no xfrm_states).*/
2140         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2141         if (IS_ERR(xdst)) {
2142                 xfrm_pols_put(pols, num_pols);
2143                 return ERR_CAST(xdst);
2144         }
2145         xdst->num_pols = num_pols;
2146         xdst->num_xfrms = num_xfrms;
2147         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2148
2149         dst_hold(&xdst->u.dst);
2150         return &xdst->flo;
2151
2152 inc_error:
2153         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2154 error:
2155         if (xdst != NULL)
2156                 dst_free(&xdst->u.dst);
2157         else
2158                 xfrm_pols_put(pols, num_pols);
2159         return ERR_PTR(err);
2160 }
2161
2162 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2163                                         struct dst_entry *dst_orig)
2164 {
2165         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2166         struct dst_entry *ret;
2167
2168         if (!afinfo) {
2169                 dst_release(dst_orig);
2170                 return ERR_PTR(-EINVAL);
2171         } else {
2172                 ret = afinfo->blackhole_route(net, dst_orig);
2173         }
2174         xfrm_policy_put_afinfo(afinfo);
2175
2176         return ret;
2177 }
2178
2179 /* Main function: finds/creates a bundle for given flow.
2180  *
2181  * At the moment we eat a raw IP route. Mostly to speed up lookups
2182  * on interfaces with disabled IPsec.
2183  */
2184 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2185                               const struct flowi *fl,
2186                               struct sock *sk, int flags)
2187 {
2188         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2189         struct flow_cache_object *flo;
2190         struct xfrm_dst *xdst;
2191         struct dst_entry *dst, *route;
2192         u16 family = dst_orig->ops->family;
2193         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2194         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2195
2196         dst = NULL;
2197         xdst = NULL;
2198         route = NULL;
2199
2200         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2201                 num_pols = 1;
2202                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2203                 err = xfrm_expand_policies(fl, family, pols,
2204                                            &num_pols, &num_xfrms);
2205                 if (err < 0)
2206                         goto dropdst;
2207
2208                 if (num_pols) {
2209                         if (num_xfrms <= 0) {
2210                                 drop_pols = num_pols;
2211                                 goto no_transform;
2212                         }
2213
2214                         xdst = xfrm_resolve_and_create_bundle(
2215                                         pols, num_pols, fl,
2216                                         family, dst_orig);
2217                         if (IS_ERR(xdst)) {
2218                                 xfrm_pols_put(pols, num_pols);
2219                                 err = PTR_ERR(xdst);
2220                                 goto dropdst;
2221                         } else if (xdst == NULL) {
2222                                 num_xfrms = 0;
2223                                 drop_pols = num_pols;
2224                                 goto no_transform;
2225                         }
2226
2227                         dst_hold(&xdst->u.dst);
2228                         xdst->u.dst.flags |= DST_NOCACHE;
2229                         route = xdst->route;
2230                 }
2231         }
2232
2233         if (xdst == NULL) {
2234                 struct xfrm_flo xflo;
2235
2236                 xflo.dst_orig = dst_orig;
2237                 xflo.flags = flags;
2238
2239                 /* To accelerate a bit...  */
2240                 if ((dst_orig->flags & DST_NOXFRM) ||
2241                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2242                         goto nopol;
2243
2244                 flo = flow_cache_lookup(net, fl, family, dir,
2245                                         xfrm_bundle_lookup, &xflo);
2246                 if (flo == NULL)
2247                         goto nopol;
2248                 if (IS_ERR(flo)) {
2249                         err = PTR_ERR(flo);
2250                         goto dropdst;
2251                 }
2252                 xdst = container_of(flo, struct xfrm_dst, flo);
2253
2254                 num_pols = xdst->num_pols;
2255                 num_xfrms = xdst->num_xfrms;
2256                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2257                 route = xdst->route;
2258         }
2259
2260         dst = &xdst->u.dst;
2261         if (route == NULL && num_xfrms > 0) {
2262                 /* The only case when xfrm_bundle_lookup() returns a
2263                  * bundle with null route, is when the template could
2264                  * not be resolved. It means policies are there, but
2265                  * bundle could not be created, since we don't yet
2266                  * have the xfrm_state's. We need to wait for KM to
2267                  * negotiate new SA's or bail out with error.*/
2268                 if (net->xfrm.sysctl_larval_drop) {
2269                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2270                         err = -EREMOTE;
2271                         goto error;
2272                 }
2273
2274                 err = -EAGAIN;
2275
2276                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2277                 goto error;
2278         }
2279
2280 no_transform:
2281         if (num_pols == 0)
2282                 goto nopol;
2283
2284         if ((flags & XFRM_LOOKUP_ICMP) &&
2285             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2286                 err = -ENOENT;
2287                 goto error;
2288         }
2289
2290         for (i = 0; i < num_pols; i++)
2291                 pols[i]->curlft.use_time = get_seconds();
2292
2293         if (num_xfrms < 0) {
2294                 /* Prohibit the flow */
2295                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2296                 err = -EPERM;
2297                 goto error;
2298         } else if (num_xfrms > 0) {
2299                 /* Flow transformed */
2300                 dst_release(dst_orig);
2301         } else {
2302                 /* Flow passes untransformed */
2303                 dst_release(dst);
2304                 dst = dst_orig;
2305         }
2306 ok:
2307         xfrm_pols_put(pols, drop_pols);
2308         if (dst && dst->xfrm &&
2309             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2310                 dst->flags |= DST_XFRM_TUNNEL;
2311         return dst;
2312
2313 nopol:
2314         if (!(flags & XFRM_LOOKUP_ICMP)) {
2315                 dst = dst_orig;
2316                 goto ok;
2317         }
2318         err = -ENOENT;
2319 error:
2320         dst_release(dst);
2321 dropdst:
2322         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2323                 dst_release(dst_orig);
2324         xfrm_pols_put(pols, drop_pols);
2325         return ERR_PTR(err);
2326 }
2327 EXPORT_SYMBOL(xfrm_lookup);
2328
2329 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2330  * Otherwise we may send out blackholed packets.
2331  */
2332 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2333                                     const struct flowi *fl,
2334                                     struct sock *sk, int flags)
2335 {
2336         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2337                                             flags | XFRM_LOOKUP_QUEUE |
2338                                             XFRM_LOOKUP_KEEP_DST_REF);
2339
2340         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2341                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2342
2343         return dst;
2344 }
2345 EXPORT_SYMBOL(xfrm_lookup_route);
2346
2347 static inline int
2348 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2349 {
2350         struct xfrm_state *x;
2351
2352         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2353                 return 0;
2354         x = skb->sp->xvec[idx];
2355         if (!x->type->reject)
2356                 return 0;
2357         return x->type->reject(x, skb, fl);
2358 }
2359
2360 /* When skb is transformed back to its "native" form, we have to
2361  * check policy restrictions. At the moment we make this in maximally
2362  * stupid way. Shame on me. :-) Of course, connected sockets must
2363  * have policy cached at them.
2364  */
2365
2366 static inline int
2367 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2368               unsigned short family)
2369 {
2370         if (xfrm_state_kern(x))
2371                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2372         return  x->id.proto == tmpl->id.proto &&
2373                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2374                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2375                 x->props.mode == tmpl->mode &&
2376                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2377                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2378                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2379                   xfrm_state_addr_cmp(tmpl, x, family));
2380 }
2381
2382 /*
2383  * 0 or more than 0 is returned when validation is succeeded (either bypass
2384  * because of optional transport mode, or next index of the mathced secpath
2385  * state with the template.
2386  * -1 is returned when no matching template is found.
2387  * Otherwise "-2 - errored_index" is returned.
2388  */
2389 static inline int
2390 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2391                unsigned short family)
2392 {
2393         int idx = start;
2394
2395         if (tmpl->optional) {
2396                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2397                         return start;
2398         } else
2399                 start = -1;
2400         for (; idx < sp->len; idx++) {
2401                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2402                         return ++idx;
2403                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2404                         if (start == -1)
2405                                 start = -2-idx;
2406                         break;
2407                 }
2408         }
2409         return start;
2410 }
2411
2412 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2413                           unsigned int family, int reverse)
2414 {
2415         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2416         int err;
2417
2418         if (unlikely(afinfo == NULL))
2419                 return -EAFNOSUPPORT;
2420
2421         afinfo->decode_session(skb, fl, reverse);
2422         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2423         xfrm_policy_put_afinfo(afinfo);
2424         return err;
2425 }
2426 EXPORT_SYMBOL(__xfrm_decode_session);
2427
2428 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2429 {
2430         for (; k < sp->len; k++) {
2431                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2432                         *idxp = k;
2433                         return 1;
2434                 }
2435         }
2436
2437         return 0;
2438 }
2439
2440 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2441                         unsigned short family)
2442 {
2443         struct net *net = dev_net(skb->dev);
2444         struct xfrm_policy *pol;
2445         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2446         int npols = 0;
2447         int xfrm_nr;
2448         int pi;
2449         int reverse;
2450         struct flowi fl;
2451         u8 fl_dir;
2452         int xerr_idx = -1;
2453
2454         reverse = dir & ~XFRM_POLICY_MASK;
2455         dir &= XFRM_POLICY_MASK;
2456         fl_dir = policy_to_flow_dir(dir);
2457
2458         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2459                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2460                 return 0;
2461         }
2462
2463         nf_nat_decode_session(skb, &fl, family);
2464
2465         /* First, check used SA against their selectors. */
2466         if (skb->sp) {
2467                 int i;
2468
2469                 for (i = skb->sp->len-1; i >= 0; i--) {
2470                         struct xfrm_state *x = skb->sp->xvec[i];
2471                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2472                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2473                                 return 0;
2474                         }
2475                 }
2476         }
2477
2478         pol = NULL;
2479         if (sk && sk->sk_policy[dir]) {
2480                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2481                 if (IS_ERR(pol)) {
2482                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2483                         return 0;
2484                 }
2485         }
2486
2487         if (!pol) {
2488                 struct flow_cache_object *flo;
2489
2490                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2491                                         xfrm_policy_lookup, NULL);
2492                 if (IS_ERR_OR_NULL(flo))
2493                         pol = ERR_CAST(flo);
2494                 else
2495                         pol = container_of(flo, struct xfrm_policy, flo);
2496         }
2497
2498         if (IS_ERR(pol)) {
2499                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2500                 return 0;
2501         }
2502
2503         if (!pol) {
2504                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2505                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2506                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2507                         return 0;
2508                 }
2509                 return 1;
2510         }
2511
2512         pol->curlft.use_time = get_seconds();
2513
2514         pols[0] = pol;
2515         npols++;
2516 #ifdef CONFIG_XFRM_SUB_POLICY
2517         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2518                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2519                                                     &fl, family,
2520                                                     XFRM_POLICY_IN);
2521                 if (pols[1]) {
2522                         if (IS_ERR(pols[1])) {
2523                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2524                                 return 0;
2525                         }
2526                         pols[1]->curlft.use_time = get_seconds();
2527                         npols++;
2528                 }
2529         }
2530 #endif
2531
2532         if (pol->action == XFRM_POLICY_ALLOW) {
2533                 struct sec_path *sp;
2534                 static struct sec_path dummy;
2535                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2536                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2537                 struct xfrm_tmpl **tpp = tp;
2538                 int ti = 0;
2539                 int i, k;
2540
2541                 if ((sp = skb->sp) == NULL)
2542                         sp = &dummy;
2543
2544                 for (pi = 0; pi < npols; pi++) {
2545                         if (pols[pi] != pol &&
2546                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2547                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2548                                 goto reject;
2549                         }
2550                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2551                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2552                                 goto reject_error;
2553                         }
2554                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2555                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2556                 }
2557                 xfrm_nr = ti;
2558                 if (npols > 1) {
2559                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2560                         tpp = stp;
2561                 }
2562
2563                 /* For each tunnel xfrm, find the first matching tmpl.
2564                  * For each tmpl before that, find corresponding xfrm.
2565                  * Order is _important_. Later we will implement
2566                  * some barriers, but at the moment barriers
2567                  * are implied between each two transformations.
2568                  */
2569                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2570                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2571                         if (k < 0) {
2572                                 if (k < -1)
2573                                         /* "-2 - errored_index" returned */
2574                                         xerr_idx = -(2+k);
2575                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2576                                 goto reject;
2577                         }
2578                 }
2579
2580                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2581                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2582                         goto reject;
2583                 }
2584
2585                 xfrm_pols_put(pols, npols);
2586                 return 1;
2587         }
2588         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2589
2590 reject:
2591         xfrm_secpath_reject(xerr_idx, skb, &fl);
2592 reject_error:
2593         xfrm_pols_put(pols, npols);
2594         return 0;
2595 }
2596 EXPORT_SYMBOL(__xfrm_policy_check);
2597
2598 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2599 {
2600         struct net *net = dev_net(skb->dev);
2601         struct flowi fl;
2602         struct dst_entry *dst;
2603         int res = 1;
2604
2605         if (xfrm_decode_session(skb, &fl, family) < 0) {
2606                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2607                 return 0;
2608         }
2609
2610         skb_dst_force(skb);
2611
2612         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2613         if (IS_ERR(dst)) {
2614                 res = 0;
2615                 dst = NULL;
2616         }
2617         skb_dst_set(skb, dst);
2618         return res;
2619 }
2620 EXPORT_SYMBOL(__xfrm_route_forward);
2621
2622 /* Optimize later using cookies and generation ids. */
2623
2624 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2625 {
2626         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2627          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2628          * get validated by dst_ops->check on every use.  We do this
2629          * because when a normal route referenced by an XFRM dst is
2630          * obsoleted we do not go looking around for all parent
2631          * referencing XFRM dsts so that we can invalidate them.  It
2632          * is just too much work.  Instead we make the checks here on
2633          * every use.  For example:
2634          *
2635          *      XFRM dst A --> IPv4 dst X
2636          *
2637          * X is the "xdst->route" of A (X is also the "dst->path" of A
2638          * in this example).  If X is marked obsolete, "A" will not
2639          * notice.  That's what we are validating here via the
2640          * stale_bundle() check.
2641          *
2642          * When a policy's bundle is pruned, we dst_free() the XFRM
2643          * dst which causes it's ->obsolete field to be set to
2644          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2645          * this, we want to force a new route lookup.
2646          */
2647         if (dst->obsolete < 0 && !stale_bundle(dst))
2648                 return dst;
2649
2650         return NULL;
2651 }
2652
2653 static int stale_bundle(struct dst_entry *dst)
2654 {
2655         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2656 }
2657
2658 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2659 {
2660         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2661                 dst->dev = dev_net(dev)->loopback_dev;
2662                 dev_hold(dst->dev);
2663                 dev_put(dev);
2664         }
2665 }
2666 EXPORT_SYMBOL(xfrm_dst_ifdown);
2667
2668 static void xfrm_link_failure(struct sk_buff *skb)
2669 {
2670         /* Impossible. Such dst must be popped before reaches point of failure. */
2671 }
2672
2673 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2674 {
2675         if (dst) {
2676                 if (dst->obsolete) {
2677                         dst_release(dst);
2678                         dst = NULL;
2679                 }
2680         }
2681         return dst;
2682 }
2683
2684 void xfrm_garbage_collect(struct net *net)
2685 {
2686         flow_cache_flush(net);
2687 }
2688 EXPORT_SYMBOL(xfrm_garbage_collect);
2689
2690 static void xfrm_garbage_collect_deferred(struct net *net)
2691 {
2692         flow_cache_flush_deferred(net);
2693 }
2694
2695 static void xfrm_init_pmtu(struct dst_entry *dst)
2696 {
2697         do {
2698                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2699                 u32 pmtu, route_mtu_cached;
2700
2701                 pmtu = dst_mtu(dst->child);
2702                 xdst->child_mtu_cached = pmtu;
2703
2704                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2705
2706                 route_mtu_cached = dst_mtu(xdst->route);
2707                 xdst->route_mtu_cached = route_mtu_cached;
2708
2709                 if (pmtu > route_mtu_cached)
2710                         pmtu = route_mtu_cached;
2711
2712                 dst_metric_set(dst, RTAX_MTU, pmtu);
2713         } while ((dst = dst->next));
2714 }
2715
2716 /* Check that the bundle accepts the flow and its components are
2717  * still valid.
2718  */
2719
2720 static int xfrm_bundle_ok(struct xfrm_dst *first)
2721 {
2722         struct dst_entry *dst = &first->u.dst;
2723         struct xfrm_dst *last;
2724         u32 mtu;
2725
2726         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2727             (dst->dev && !netif_running(dst->dev)))
2728                 return 0;
2729
2730         if (dst->flags & DST_XFRM_QUEUE)
2731                 return 1;
2732
2733         last = NULL;
2734
2735         do {
2736                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2737
2738                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2739                         return 0;
2740                 if (xdst->xfrm_genid != dst->xfrm->genid)
2741                         return 0;
2742                 if (xdst->num_pols > 0 &&
2743                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2744                         return 0;
2745
2746                 mtu = dst_mtu(dst->child);
2747                 if (xdst->child_mtu_cached != mtu) {
2748                         last = xdst;
2749                         xdst->child_mtu_cached = mtu;
2750                 }
2751
2752                 if (!dst_check(xdst->route, xdst->route_cookie))
2753                         return 0;
2754                 mtu = dst_mtu(xdst->route);
2755                 if (xdst->route_mtu_cached != mtu) {
2756                         last = xdst;
2757                         xdst->route_mtu_cached = mtu;
2758                 }
2759
2760                 dst = dst->child;
2761         } while (dst->xfrm);
2762
2763         if (likely(!last))
2764                 return 1;
2765
2766         mtu = last->child_mtu_cached;
2767         for (;;) {
2768                 dst = &last->u.dst;
2769
2770                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2771                 if (mtu > last->route_mtu_cached)
2772                         mtu = last->route_mtu_cached;
2773                 dst_metric_set(dst, RTAX_MTU, mtu);
2774
2775                 if (last == first)
2776                         break;
2777
2778                 last = (struct xfrm_dst *)last->u.dst.next;
2779                 last->child_mtu_cached = mtu;
2780         }
2781
2782         return 1;
2783 }
2784
2785 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2786 {
2787         return dst_metric_advmss(dst->path);
2788 }
2789
2790 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2791 {
2792         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2793
2794         return mtu ? : dst_mtu(dst->path);
2795 }
2796
2797 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2798                                            struct sk_buff *skb,
2799                                            const void *daddr)
2800 {
2801         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2802 }
2803
2804 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2805 {
2806         struct net *net;
2807         int err = 0;
2808         if (unlikely(afinfo == NULL))
2809                 return -EINVAL;
2810         if (unlikely(afinfo->family >= NPROTO))
2811                 return -EAFNOSUPPORT;
2812         spin_lock(&xfrm_policy_afinfo_lock);
2813         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2814                 err = -EEXIST;
2815         else {
2816                 struct dst_ops *dst_ops = afinfo->dst_ops;
2817                 if (likely(dst_ops->kmem_cachep == NULL))
2818                         dst_ops->kmem_cachep = xfrm_dst_cache;
2819                 if (likely(dst_ops->check == NULL))
2820                         dst_ops->check = xfrm_dst_check;
2821                 if (likely(dst_ops->default_advmss == NULL))
2822                         dst_ops->default_advmss = xfrm_default_advmss;
2823                 if (likely(dst_ops->mtu == NULL))
2824                         dst_ops->mtu = xfrm_mtu;
2825                 if (likely(dst_ops->negative_advice == NULL))
2826                         dst_ops->negative_advice = xfrm_negative_advice;
2827                 if (likely(dst_ops->link_failure == NULL))
2828                         dst_ops->link_failure = xfrm_link_failure;
2829                 if (likely(dst_ops->neigh_lookup == NULL))
2830                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2831                 if (likely(afinfo->garbage_collect == NULL))
2832                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2833                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2834         }
2835         spin_unlock(&xfrm_policy_afinfo_lock);
2836
2837         rtnl_lock();
2838         for_each_net(net) {
2839                 struct dst_ops *xfrm_dst_ops;
2840
2841                 switch (afinfo->family) {
2842                 case AF_INET:
2843                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2844                         break;
2845 #if IS_ENABLED(CONFIG_IPV6)
2846                 case AF_INET6:
2847                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2848                         break;
2849 #endif
2850                 default:
2851                         BUG();
2852                 }
2853                 *xfrm_dst_ops = *afinfo->dst_ops;
2854         }
2855         rtnl_unlock();
2856
2857         return err;
2858 }
2859 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2860
2861 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2862 {
2863         int err = 0;
2864         if (unlikely(afinfo == NULL))
2865                 return -EINVAL;
2866         if (unlikely(afinfo->family >= NPROTO))
2867                 return -EAFNOSUPPORT;
2868         spin_lock(&xfrm_policy_afinfo_lock);
2869         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2870                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2871                         err = -EINVAL;
2872                 else
2873                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2874                                          NULL);
2875         }
2876         spin_unlock(&xfrm_policy_afinfo_lock);
2877         if (!err) {
2878                 struct dst_ops *dst_ops = afinfo->dst_ops;
2879
2880                 synchronize_rcu();
2881
2882                 dst_ops->kmem_cachep = NULL;
2883                 dst_ops->check = NULL;
2884                 dst_ops->negative_advice = NULL;
2885                 dst_ops->link_failure = NULL;
2886                 afinfo->garbage_collect = NULL;
2887         }
2888         return err;
2889 }
2890 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2891
2892 static void __net_init xfrm_dst_ops_init(struct net *net)
2893 {
2894         struct xfrm_policy_afinfo *afinfo;
2895
2896         rcu_read_lock();
2897         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2898         if (afinfo)
2899                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2900 #if IS_ENABLED(CONFIG_IPV6)
2901         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2902         if (afinfo)
2903                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2904 #endif
2905         rcu_read_unlock();
2906 }
2907
2908 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2909 {
2910         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2911
2912         switch (event) {
2913         case NETDEV_DOWN:
2914                 xfrm_garbage_collect(dev_net(dev));
2915         }
2916         return NOTIFY_DONE;
2917 }
2918
2919 static struct notifier_block xfrm_dev_notifier = {
2920         .notifier_call  = xfrm_dev_event,
2921 };
2922
2923 #ifdef CONFIG_XFRM_STATISTICS
2924 static int __net_init xfrm_statistics_init(struct net *net)
2925 {
2926         int rv;
2927         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2928         if (!net->mib.xfrm_statistics)
2929                 return -ENOMEM;
2930         rv = xfrm_proc_init(net);
2931         if (rv < 0)
2932                 free_percpu(net->mib.xfrm_statistics);
2933         return rv;
2934 }
2935
2936 static void xfrm_statistics_fini(struct net *net)
2937 {
2938         xfrm_proc_fini(net);
2939         free_percpu(net->mib.xfrm_statistics);
2940 }
2941 #else
2942 static int __net_init xfrm_statistics_init(struct net *net)
2943 {
2944         return 0;
2945 }
2946
2947 static void xfrm_statistics_fini(struct net *net)
2948 {
2949 }
2950 #endif
2951
2952 static int __net_init xfrm_policy_init(struct net *net)
2953 {
2954         unsigned int hmask, sz;
2955         int dir;
2956
2957         if (net_eq(net, &init_net))
2958                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2959                                            sizeof(struct xfrm_dst),
2960                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2961                                            NULL);
2962
2963         hmask = 8 - 1;
2964         sz = (hmask+1) * sizeof(struct hlist_head);
2965
2966         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2967         if (!net->xfrm.policy_byidx)
2968                 goto out_byidx;
2969         net->xfrm.policy_idx_hmask = hmask;
2970
2971         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2972                 struct xfrm_policy_hash *htab;
2973
2974                 net->xfrm.policy_count[dir] = 0;
2975                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2976                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2977
2978                 htab = &net->xfrm.policy_bydst[dir];
2979                 htab->table = xfrm_hash_alloc(sz);
2980                 if (!htab->table)
2981                         goto out_bydst;
2982                 htab->hmask = hmask;
2983                 htab->dbits4 = 32;
2984                 htab->sbits4 = 32;
2985                 htab->dbits6 = 128;
2986                 htab->sbits6 = 128;
2987         }
2988         net->xfrm.policy_hthresh.lbits4 = 32;
2989         net->xfrm.policy_hthresh.rbits4 = 32;
2990         net->xfrm.policy_hthresh.lbits6 = 128;
2991         net->xfrm.policy_hthresh.rbits6 = 128;
2992
2993         seqlock_init(&net->xfrm.policy_hthresh.lock);
2994
2995         INIT_LIST_HEAD(&net->xfrm.policy_all);
2996         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2997         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2998         if (net_eq(net, &init_net))
2999                 register_netdevice_notifier(&xfrm_dev_notifier);
3000         return 0;
3001
3002 out_bydst:
3003         for (dir--; dir >= 0; dir--) {
3004                 struct xfrm_policy_hash *htab;
3005
3006                 htab = &net->xfrm.policy_bydst[dir];
3007                 xfrm_hash_free(htab->table, sz);
3008         }
3009         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3010 out_byidx:
3011         return -ENOMEM;
3012 }
3013
3014 static void xfrm_policy_fini(struct net *net)
3015 {
3016         unsigned int sz;
3017         int dir;
3018
3019         flush_work(&net->xfrm.policy_hash_work);
3020 #ifdef CONFIG_XFRM_SUB_POLICY
3021         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3022 #endif
3023         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3024
3025         WARN_ON(!list_empty(&net->xfrm.policy_all));
3026
3027         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3028                 struct xfrm_policy_hash *htab;
3029
3030                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3031
3032                 htab = &net->xfrm.policy_bydst[dir];
3033                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3034                 WARN_ON(!hlist_empty(htab->table));
3035                 xfrm_hash_free(htab->table, sz);
3036         }
3037
3038         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3039         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3040         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3041 }
3042
3043 static int __net_init xfrm_net_init(struct net *net)
3044 {
3045         int rv;
3046
3047         rv = xfrm_statistics_init(net);
3048         if (rv < 0)
3049                 goto out_statistics;
3050         rv = xfrm_state_init(net);
3051         if (rv < 0)
3052                 goto out_state;
3053         rv = xfrm_policy_init(net);
3054         if (rv < 0)
3055                 goto out_policy;
3056         xfrm_dst_ops_init(net);
3057         rv = xfrm_sysctl_init(net);
3058         if (rv < 0)
3059                 goto out_sysctl;
3060         rv = flow_cache_init(net);
3061         if (rv < 0)
3062                 goto out;
3063
3064         /* Initialize the per-net locks here */
3065         spin_lock_init(&net->xfrm.xfrm_state_lock);
3066         rwlock_init(&net->xfrm.xfrm_policy_lock);
3067         mutex_init(&net->xfrm.xfrm_cfg_mutex);
3068
3069         return 0;
3070
3071 out:
3072         xfrm_sysctl_fini(net);
3073 out_sysctl:
3074         xfrm_policy_fini(net);
3075 out_policy:
3076         xfrm_state_fini(net);
3077 out_state:
3078         xfrm_statistics_fini(net);
3079 out_statistics:
3080         return rv;
3081 }
3082
3083 static void __net_exit xfrm_net_exit(struct net *net)
3084 {
3085         flow_cache_fini(net);
3086         xfrm_sysctl_fini(net);
3087         xfrm_policy_fini(net);
3088         xfrm_state_fini(net);
3089         xfrm_statistics_fini(net);
3090 }
3091
3092 static struct pernet_operations __net_initdata xfrm_net_ops = {
3093         .init = xfrm_net_init,
3094         .exit = xfrm_net_exit,
3095 };
3096
3097 void __init xfrm_init(void)
3098 {
3099         register_pernet_subsys(&xfrm_net_ops);
3100         xfrm_input_init();
3101 }
3102
3103 #ifdef CONFIG_AUDITSYSCALL
3104 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3105                                          struct audit_buffer *audit_buf)
3106 {
3107         struct xfrm_sec_ctx *ctx = xp->security;
3108         struct xfrm_selector *sel = &xp->selector;
3109
3110         if (ctx)
3111                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3112                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3113
3114         switch (sel->family) {
3115         case AF_INET:
3116                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3117                 if (sel->prefixlen_s != 32)
3118                         audit_log_format(audit_buf, " src_prefixlen=%d",
3119                                          sel->prefixlen_s);
3120                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3121                 if (sel->prefixlen_d != 32)
3122                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3123                                          sel->prefixlen_d);
3124                 break;
3125         case AF_INET6:
3126                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3127                 if (sel->prefixlen_s != 128)
3128                         audit_log_format(audit_buf, " src_prefixlen=%d",
3129                                          sel->prefixlen_s);
3130                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3131                 if (sel->prefixlen_d != 128)
3132                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3133                                          sel->prefixlen_d);
3134                 break;
3135         }
3136 }
3137
3138 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3139 {
3140         struct audit_buffer *audit_buf;
3141
3142         audit_buf = xfrm_audit_start("SPD-add");
3143         if (audit_buf == NULL)
3144                 return;
3145         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3146         audit_log_format(audit_buf, " res=%u", result);
3147         xfrm_audit_common_policyinfo(xp, audit_buf);
3148         audit_log_end(audit_buf);
3149 }
3150 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3151
3152 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3153                               bool task_valid)
3154 {
3155         struct audit_buffer *audit_buf;
3156
3157         audit_buf = xfrm_audit_start("SPD-delete");
3158         if (audit_buf == NULL)
3159                 return;
3160         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3161         audit_log_format(audit_buf, " res=%u", result);
3162         xfrm_audit_common_policyinfo(xp, audit_buf);
3163         audit_log_end(audit_buf);
3164 }
3165 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3166 #endif
3167
3168 #ifdef CONFIG_XFRM_MIGRATE
3169 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3170                                         const struct xfrm_selector *sel_tgt)
3171 {
3172         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3173                 if (sel_tgt->family == sel_cmp->family &&
3174                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3175                                     sel_cmp->family) &&
3176                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3177                                     sel_cmp->family) &&
3178                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3179                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3180                         return true;
3181                 }
3182         } else {
3183                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3184                         return true;
3185                 }
3186         }
3187         return false;
3188 }
3189
3190 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3191                                                     u8 dir, u8 type, struct net *net)
3192 {
3193         struct xfrm_policy *pol, *ret = NULL;
3194         struct hlist_head *chain;
3195         u32 priority = ~0U;
3196
3197         read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3198         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3199         hlist_for_each_entry(pol, chain, bydst) {
3200                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3201                     pol->type == type) {
3202                         ret = pol;
3203                         priority = ret->priority;
3204                         break;
3205                 }
3206         }
3207         chain = &net->xfrm.policy_inexact[dir];
3208         hlist_for_each_entry(pol, chain, bydst) {
3209                 if ((pol->priority >= priority) && ret)
3210                         break;
3211
3212                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3213                     pol->type == type) {
3214                         ret = pol;
3215                         break;
3216                 }
3217         }
3218
3219         xfrm_pol_hold(ret);
3220
3221         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3222
3223         return ret;
3224 }
3225
3226 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3227 {
3228         int match = 0;
3229
3230         if (t->mode == m->mode && t->id.proto == m->proto &&
3231             (m->reqid == 0 || t->reqid == m->reqid)) {
3232                 switch (t->mode) {
3233                 case XFRM_MODE_TUNNEL:
3234                 case XFRM_MODE_BEET:
3235                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3236                                             m->old_family) &&
3237                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3238                                             m->old_family)) {
3239                                 match = 1;
3240                         }
3241                         break;
3242                 case XFRM_MODE_TRANSPORT:
3243                         /* in case of transport mode, template does not store
3244                            any IP addresses, hence we just compare mode and
3245                            protocol */
3246                         match = 1;
3247                         break;
3248                 default:
3249                         break;
3250                 }
3251         }
3252         return match;
3253 }
3254
3255 /* update endpoint address(es) of template(s) */
3256 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3257                                struct xfrm_migrate *m, int num_migrate)
3258 {
3259         struct xfrm_migrate *mp;
3260         int i, j, n = 0;
3261
3262         write_lock_bh(&pol->lock);
3263         if (unlikely(pol->walk.dead)) {
3264                 /* target policy has been deleted */
3265                 write_unlock_bh(&pol->lock);
3266                 return -ENOENT;
3267         }
3268
3269         for (i = 0; i < pol->xfrm_nr; i++) {
3270                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3271                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3272                                 continue;
3273                         n++;
3274                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3275                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3276                                 continue;
3277                         /* update endpoints */
3278                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3279                                sizeof(pol->xfrm_vec[i].id.daddr));
3280                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3281                                sizeof(pol->xfrm_vec[i].saddr));
3282                         pol->xfrm_vec[i].encap_family = mp->new_family;
3283                         /* flush bundles */
3284                         atomic_inc(&pol->genid);
3285                 }
3286         }
3287
3288         write_unlock_bh(&pol->lock);
3289
3290         if (!n)
3291                 return -ENODATA;
3292
3293         return 0;
3294 }
3295
3296 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3297 {
3298         int i, j;
3299
3300         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3301                 return -EINVAL;
3302
3303         for (i = 0; i < num_migrate; i++) {
3304                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3305                                     m[i].old_family) &&
3306                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3307                                     m[i].old_family))
3308                         return -EINVAL;
3309                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3310                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3311                         return -EINVAL;
3312
3313                 /* check if there is any duplicated entry */
3314                 for (j = i + 1; j < num_migrate; j++) {
3315                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3316                                     sizeof(m[i].old_daddr)) &&
3317                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3318                                     sizeof(m[i].old_saddr)) &&
3319                             m[i].proto == m[j].proto &&
3320                             m[i].mode == m[j].mode &&
3321                             m[i].reqid == m[j].reqid &&
3322                             m[i].old_family == m[j].old_family)
3323                                 return -EINVAL;
3324                 }
3325         }
3326
3327         return 0;
3328 }
3329
3330 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3331                  struct xfrm_migrate *m, int num_migrate,
3332                  struct xfrm_kmaddress *k, struct net *net)
3333 {
3334         int i, err, nx_cur = 0, nx_new = 0;
3335         struct xfrm_policy *pol = NULL;
3336         struct xfrm_state *x, *xc;
3337         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3338         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3339         struct xfrm_migrate *mp;
3340
3341         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3342                 goto out;
3343
3344         /* Stage 1 - find policy */
3345         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3346                 err = -ENOENT;
3347                 goto out;
3348         }
3349
3350         /* Stage 2 - find and update state(s) */
3351         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3352                 if ((x = xfrm_migrate_state_find(mp, net))) {
3353                         x_cur[nx_cur] = x;
3354                         nx_cur++;
3355                         if ((xc = xfrm_state_migrate(x, mp))) {
3356                                 x_new[nx_new] = xc;
3357                                 nx_new++;
3358                         } else {
3359                                 err = -ENODATA;
3360                                 goto restore_state;
3361                         }
3362                 }
3363         }
3364
3365         /* Stage 3 - update policy */
3366         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3367                 goto restore_state;
3368
3369         /* Stage 4 - delete old state(s) */
3370         if (nx_cur) {
3371                 xfrm_states_put(x_cur, nx_cur);
3372                 xfrm_states_delete(x_cur, nx_cur);
3373         }
3374
3375         /* Stage 5 - announce */
3376         km_migrate(sel, dir, type, m, num_migrate, k);
3377
3378         xfrm_pol_put(pol);
3379
3380         return 0;
3381 out:
3382         return err;
3383
3384 restore_state:
3385         if (pol)
3386                 xfrm_pol_put(pol);
3387         if (nx_cur)
3388                 xfrm_states_put(x_cur, nx_cur);
3389         if (nx_new)
3390                 xfrm_states_delete(x_new, nx_new);
3391
3392         return err;
3393 }
3394 EXPORT_SYMBOL(xfrm_migrate);
3395 #endif