net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/flow.h>
  30 #include <net/xfrm.h>
  31 #include <net/ip.h>
  32 #ifdef CONFIG_XFRM_STATISTICS
  33 #include <net/snmp.h>
  34 #endif
  35
  36 #include "xfrm_hash.h"
  37
  38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  40 #define XFRM_MAX_QUEUE_LEN      100
  41
  42 struct xfrm_flo {
  43         struct dst_entry *dst_orig;
  44         u8 flags;
  45 };
  46
  47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  48 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
  49                                                 __read_mostly;
  50
  51 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  52
  53 static void xfrm_init_pmtu(struct dst_entry *dst);
  54 static int stale_bundle(struct dst_entry *dst);
  55 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  56 static void xfrm_policy_queue_process(unsigned long arg);
  57
  58 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
  59 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  60                                                 int dir);
  61
  62 static inline bool
  63 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  64 {
  65         const struct flowi4 *fl4 = &fl->u.ip4;
  66
  67         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  68                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  69                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  70                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  71                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  72                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  73 }
  74
  75 static inline bool
  76 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  77 {
  78         const struct flowi6 *fl6 = &fl->u.ip6;
  79
  80         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  81                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  82                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  83                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  84                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  85                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  86 }
  87
  88 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  89                          unsigned short family)
  90 {
  91         switch (family) {
  92         case AF_INET:
  93                 return __xfrm4_selector_match(sel, fl);
  94         case AF_INET6:
  95                 return __xfrm6_selector_match(sel, fl);
  96         }
  97         return false;
  98 }
  99
 100 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 101 {
 102         struct xfrm_policy_afinfo *afinfo;
 103
 104         if (unlikely(family >= NPROTO))
 105                 return NULL;
 106         rcu_read_lock();
 107         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 108         if (unlikely(!afinfo))
 109                 rcu_read_unlock();
 110         return afinfo;
 111 }
 112
 113 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
 114 {
 115         rcu_read_unlock();
 116 }
 117
 118 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net,
 119                                                   int tos, int oif,
 120                                                   const xfrm_address_t *saddr,
 121                                                   const xfrm_address_t *daddr,
 122                                                   int family)
 123 {
 124         struct xfrm_policy_afinfo *afinfo;
 125         struct dst_entry *dst;
 126
 127         afinfo = xfrm_policy_get_afinfo(family);
 128         if (unlikely(afinfo == NULL))
 129                 return ERR_PTR(-EAFNOSUPPORT);
 130
 131         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr);
 132
 133         xfrm_policy_put_afinfo(afinfo);
 134
 135         return dst;
 136 }
 137
 138 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 139                                                 int tos, int oif,
 140                                                 xfrm_address_t *prev_saddr,
 141                                                 xfrm_address_t *prev_daddr,
 142                                                 int family)
 143 {
 144         struct net *net = xs_net(x);
 145         xfrm_address_t *saddr = &x->props.saddr;
 146         xfrm_address_t *daddr = &x->id.daddr;
 147         struct dst_entry *dst;
 148
 149         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 150                 saddr = x->coaddr;
 151                 daddr = prev_daddr;
 152         }
 153         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 154                 saddr = prev_saddr;
 155                 daddr = x->coaddr;
 156         }
 157
 158         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family);
 159
 160         if (!IS_ERR(dst)) {
 161                 if (prev_saddr != saddr)
 162                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 163                 if (prev_daddr != daddr)
 164                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 165         }
 166
 167         return dst;
 168 }
 169
 170 static inline unsigned long make_jiffies(long secs)
 171 {
 172         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 173                 return MAX_SCHEDULE_TIMEOUT-1;
 174         else
 175                 return secs*HZ;
 176 }
 177
 178 static void xfrm_policy_timer(unsigned long data)
 179 {
 180         struct xfrm_policy *xp = (struct xfrm_policy *)data;
 181         unsigned long now = get_seconds();
 182         long next = LONG_MAX;
 183         int warn = 0;
 184         int dir;
 185
 186         read_lock(&xp->lock);
 187
 188         if (unlikely(xp->walk.dead))
 189                 goto out;
 190
 191         dir = xfrm_policy_id2dir(xp->index);
 192
 193         if (xp->lft.hard_add_expires_seconds) {
 194                 long tmo = xp->lft.hard_add_expires_seconds +
 195                         xp->curlft.add_time - now;
 196                 if (tmo <= 0)
 197                         goto expired;
 198                 if (tmo < next)
 199                         next = tmo;
 200         }
 201         if (xp->lft.hard_use_expires_seconds) {
 202                 long tmo = xp->lft.hard_use_expires_seconds +
 203                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 204                 if (tmo <= 0)
 205                         goto expired;
 206                 if (tmo < next)
 207                         next = tmo;
 208         }
 209         if (xp->lft.soft_add_expires_seconds) {
 210                 long tmo = xp->lft.soft_add_expires_seconds +
 211                         xp->curlft.add_time - now;
 212                 if (tmo <= 0) {
 213                         warn = 1;
 214                         tmo = XFRM_KM_TIMEOUT;
 215                 }
 216                 if (tmo < next)
 217                         next = tmo;
 218         }
 219         if (xp->lft.soft_use_expires_seconds) {
 220                 long tmo = xp->lft.soft_use_expires_seconds +
 221                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 222                 if (tmo <= 0) {
 223                         warn = 1;
 224                         tmo = XFRM_KM_TIMEOUT;
 225                 }
 226                 if (tmo < next)
 227                         next = tmo;
 228         }
 229
 230         if (warn)
 231                 km_policy_expired(xp, dir, 0, 0);
 232         if (next != LONG_MAX &&
 233             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 234                 xfrm_pol_hold(xp);
 235
 236 out:
 237         read_unlock(&xp->lock);
 238         xfrm_pol_put(xp);
 239         return;
 240
 241 expired:
 242         read_unlock(&xp->lock);
 243         if (!xfrm_policy_delete(xp, dir))
 244                 km_policy_expired(xp, dir, 1, 0);
 245         xfrm_pol_put(xp);
 246 }
 247
 248 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 249 {
 250         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 251
 252         if (unlikely(pol->walk.dead))
 253                 flo = NULL;
 254         else
 255                 xfrm_pol_hold(pol);
 256
 257         return flo;
 258 }
 259
 260 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 261 {
 262         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 263
 264         return !pol->walk.dead;
 265 }
 266
 267 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 268 {
 269         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 270 }
 271
 272 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 273         .get = xfrm_policy_flo_get,
 274         .check = xfrm_policy_flo_check,
 275         .delete = xfrm_policy_flo_delete,
 276 };
 277
 278 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 279  * SPD calls.
 280  */
 281
 282 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 283 {
 284         struct xfrm_policy *policy;
 285
 286         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 287
 288         if (policy) {
 289                 write_pnet(&policy->xp_net, net);
 290                 INIT_LIST_HEAD(&policy->walk.all);
 291                 INIT_HLIST_NODE(&policy->bydst);
 292                 INIT_HLIST_NODE(&policy->byidx);
 293                 rwlock_init(&policy->lock);
 294                 atomic_set(&policy->refcnt, 1);
 295                 skb_queue_head_init(&policy->polq.hold_queue);
 296                 setup_timer(&policy->timer, xfrm_policy_timer,
 297                                 (unsigned long)policy);
 298                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 299                             (unsigned long)policy);
 300                 policy->flo.ops = &xfrm_policy_fc_ops;
 301         }
 302         return policy;
 303 }
 304 EXPORT_SYMBOL(xfrm_policy_alloc);
 305
 306 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 307
 308 void xfrm_policy_destroy(struct xfrm_policy *policy)
 309 {
 310         BUG_ON(!policy->walk.dead);
 311
 312         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 313                 BUG();
 314
 315         security_xfrm_policy_free(policy->security);
 316         kfree(policy);
 317 }
 318 EXPORT_SYMBOL(xfrm_policy_destroy);
 319
 320 /* Rule must be locked. Release descentant resources, announce
 321  * entry dead. The rule must be unlinked from lists to the moment.
 322  */
 323
 324 static void xfrm_policy_kill(struct xfrm_policy *policy)
 325 {
 326         policy->walk.dead = 1;
 327
 328         atomic_inc(&policy->genid);
 329
 330         if (del_timer(&policy->polq.hold_timer))
 331                 xfrm_pol_put(policy);
 332         skb_queue_purge(&policy->polq.hold_queue);
 333
 334         if (del_timer(&policy->timer))
 335                 xfrm_pol_put(policy);
 336
 337         xfrm_pol_put(policy);
 338 }
 339
 340 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 341
 342 static inline unsigned int idx_hash(struct net *net, u32 index)
 343 {
 344         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 345 }
 346
 347 /* calculate policy hash thresholds */
 348 static void __get_hash_thresh(struct net *net,
 349                               unsigned short family, int dir,
 350                               u8 *dbits, u8 *sbits)
 351 {
 352         switch (family) {
 353         case AF_INET:
 354                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 355                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 356                 break;
 357
 358         case AF_INET6:
 359                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 360                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 361                 break;
 362
 363         default:
 364                 *dbits = 0;
 365                 *sbits = 0;
 366         }
 367 }
 368
 369 static struct hlist_head *policy_hash_bysel(struct net *net,
 370                                             const struct xfrm_selector *sel,
 371                                             unsigned short family, int dir)
 372 {
 373         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 374         unsigned int hash;
 375         u8 dbits;
 376         u8 sbits;
 377
 378         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 379         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 380
 381         return (hash == hmask + 1 ?
 382                 &net->xfrm.policy_inexact[dir] :
 383                 net->xfrm.policy_bydst[dir].table + hash);
 384 }
 385
 386 static struct hlist_head *policy_hash_direct(struct net *net,
 387                                              const xfrm_address_t *daddr,
 388                                              const xfrm_address_t *saddr,
 389                                              unsigned short family, int dir)
 390 {
 391         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 392         unsigned int hash;
 393         u8 dbits;
 394         u8 sbits;
 395
 396         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 397         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 398
 399         return net->xfrm.policy_bydst[dir].table + hash;
 400 }
 401
 402 static void xfrm_dst_hash_transfer(struct net *net,
 403                                    struct hlist_head *list,
 404                                    struct hlist_head *ndsttable,
 405                                    unsigned int nhashmask,
 406                                    int dir)
 407 {
 408         struct hlist_node *tmp, *entry0 = NULL;
 409         struct xfrm_policy *pol;
 410         unsigned int h0 = 0;
 411         u8 dbits;
 412         u8 sbits;
 413
 414 redo:
 415         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 416                 unsigned int h;
 417
 418                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 419                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 420                                 pol->family, nhashmask, dbits, sbits);
 421                 if (!entry0) {
 422                         hlist_del(&pol->bydst);
 423                         hlist_add_head(&pol->bydst, ndsttable+h);
 424                         h0 = h;
 425                 } else {
 426                         if (h != h0)
 427                                 continue;
 428                         hlist_del(&pol->bydst);
 429                         hlist_add_behind(&pol->bydst, entry0);
 430                 }
 431                 entry0 = &pol->bydst;
 432         }
 433         if (!hlist_empty(list)) {
 434                 entry0 = NULL;
 435                 goto redo;
 436         }
 437 }
 438
 439 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 440                                    struct hlist_head *nidxtable,
 441                                    unsigned int nhashmask)
 442 {
 443         struct hlist_node *tmp;
 444         struct xfrm_policy *pol;
 445
 446         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 447                 unsigned int h;
 448
 449                 h = __idx_hash(pol->index, nhashmask);
 450                 hlist_add_head(&pol->byidx, nidxtable+h);
 451         }
 452 }
 453
 454 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 455 {
 456         return ((old_hmask + 1) << 1) - 1;
 457 }
 458
 459 static void xfrm_bydst_resize(struct net *net, int dir)
 460 {
 461         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 462         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 463         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 464         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 465         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 466         int i;
 467
 468         if (!ndst)
 469                 return;
 470
 471         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 472
 473         for (i = hmask; i >= 0; i--)
 474                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 475
 476         net->xfrm.policy_bydst[dir].table = ndst;
 477         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 478
 479         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 480
 481         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 482 }
 483
 484 static void xfrm_byidx_resize(struct net *net, int total)
 485 {
 486         unsigned int hmask = net->xfrm.policy_idx_hmask;
 487         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 488         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 489         struct hlist_head *oidx = net->xfrm.policy_byidx;
 490         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 491         int i;
 492
 493         if (!nidx)
 494                 return;
 495
 496         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 497
 498         for (i = hmask; i >= 0; i--)
 499                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 500
 501         net->xfrm.policy_byidx = nidx;
 502         net->xfrm.policy_idx_hmask = nhashmask;
 503
 504         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 505
 506         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 507 }
 508
 509 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 510 {
 511         unsigned int cnt = net->xfrm.policy_count[dir];
 512         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 513
 514         if (total)
 515                 *total += cnt;
 516
 517         if ((hmask + 1) < xfrm_policy_hashmax &&
 518             cnt > hmask)
 519                 return 1;
 520
 521         return 0;
 522 }
 523
 524 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 525 {
 526         unsigned int hmask = net->xfrm.policy_idx_hmask;
 527
 528         if ((hmask + 1) < xfrm_policy_hashmax &&
 529             total > hmask)
 530                 return 1;
 531
 532         return 0;
 533 }
 534
 535 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 536 {
 537         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 538         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 539         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 540         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 541         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 542         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 543         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 544         si->spdhcnt = net->xfrm.policy_idx_hmask;
 545         si->spdhmcnt = xfrm_policy_hashmax;
 546         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
 547 }
 548 EXPORT_SYMBOL(xfrm_spd_getinfo);
 549
 550 static DEFINE_MUTEX(hash_resize_mutex);
 551 static void xfrm_hash_resize(struct work_struct *work)
 552 {
 553         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 554         int dir, total;
 555
 556         mutex_lock(&hash_resize_mutex);
 557
 558         total = 0;
 559         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 560                 if (xfrm_bydst_should_resize(net, dir, &total))
 561                         xfrm_bydst_resize(net, dir);
 562         }
 563         if (xfrm_byidx_should_resize(net, total))
 564                 xfrm_byidx_resize(net, total);
 565
 566         mutex_unlock(&hash_resize_mutex);
 567 }
 568
 569 static void xfrm_hash_rebuild(struct work_struct *work)
 570 {
 571         struct net *net = container_of(work, struct net,
 572                                        xfrm.policy_hthresh.work);
 573         unsigned int hmask;
 574         struct xfrm_policy *pol;
 575         struct xfrm_policy *policy;
 576         struct hlist_head *chain;
 577         struct hlist_head *odst;
 578         struct hlist_node *newpos;
 579         int i;
 580         int dir;
 581         unsigned seq;
 582         u8 lbits4, rbits4, lbits6, rbits6;
 583
 584         mutex_lock(&hash_resize_mutex);
 585
 586         /* read selector prefixlen thresholds */
 587         do {
 588                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
 589
 590                 lbits4 = net->xfrm.policy_hthresh.lbits4;
 591                 rbits4 = net->xfrm.policy_hthresh.rbits4;
 592                 lbits6 = net->xfrm.policy_hthresh.lbits6;
 593                 rbits6 = net->xfrm.policy_hthresh.rbits6;
 594         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
 595
 596         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 597
 598         /* reset the bydst and inexact table in all directions */
 599         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 600                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
 601                 hmask = net->xfrm.policy_bydst[dir].hmask;
 602                 odst = net->xfrm.policy_bydst[dir].table;
 603                 for (i = hmask; i >= 0; i--)
 604                         INIT_HLIST_HEAD(odst + i);
 605                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
 606                         /* dir out => dst = remote, src = local */
 607                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
 608                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
 609                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
 610                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
 611                 } else {
 612                         /* dir in/fwd => dst = local, src = remote */
 613                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
 614                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
 615                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
 616                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
 617                 }
 618         }
 619
 620         /* re-insert all policies by order of creation */
 621         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 622                 newpos = NULL;
 623                 chain = policy_hash_bysel(net, &policy->selector,
 624                                           policy->family,
 625                                           xfrm_policy_id2dir(policy->index));
 626                 hlist_for_each_entry(pol, chain, bydst) {
 627                         if (policy->priority >= pol->priority)
 628                                 newpos = &pol->bydst;
 629                         else
 630                                 break;
 631                 }
 632                 if (newpos)
 633                         hlist_add_behind(&policy->bydst, newpos);
 634                 else
 635                         hlist_add_head(&policy->bydst, chain);
 636         }
 637
 638         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 639
 640         mutex_unlock(&hash_resize_mutex);
 641 }
 642
 643 void xfrm_policy_hash_rebuild(struct net *net)
 644 {
 645         schedule_work(&net->xfrm.policy_hthresh.work);
 646 }
 647 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 648
 649 /* Generate new index... KAME seems to generate them ordered by cost
 650  * of an absolute inpredictability of ordering of rules. This will not pass. */
 651 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 652 {
 653         static u32 idx_generator;
 654
 655         for (;;) {
 656                 struct hlist_head *list;
 657                 struct xfrm_policy *p;
 658                 u32 idx;
 659                 int found;
 660
 661                 if (!index) {
 662                         idx = (idx_generator | dir);
 663                         idx_generator += 8;
 664                 } else {
 665                         idx = index;
 666                         index = 0;
 667                 }
 668
 669                 if (idx == 0)
 670                         idx = 8;
 671                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 672                 found = 0;
 673                 hlist_for_each_entry(p, list, byidx) {
 674                         if (p->index == idx) {
 675                                 found = 1;
 676                                 break;
 677                         }
 678                 }
 679                 if (!found)
 680                         return idx;
 681         }
 682 }
 683
 684 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 685 {
 686         u32 *p1 = (u32 *) s1;
 687         u32 *p2 = (u32 *) s2;
 688         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 689         int i;
 690
 691         for (i = 0; i < len; i++) {
 692                 if (p1[i] != p2[i])
 693                         return 1;
 694         }
 695
 696         return 0;
 697 }
 698
 699 static void xfrm_policy_requeue(struct xfrm_policy *old,
 700                                 struct xfrm_policy *new)
 701 {
 702         struct xfrm_policy_queue *pq = &old->polq;
 703         struct sk_buff_head list;
 704
 705         if (skb_queue_empty(&pq->hold_queue))
 706                 return;
 707
 708         __skb_queue_head_init(&list);
 709
 710         spin_lock_bh(&pq->hold_queue.lock);
 711         skb_queue_splice_init(&pq->hold_queue, &list);
 712         if (del_timer(&pq->hold_timer))
 713                 xfrm_pol_put(old);
 714         spin_unlock_bh(&pq->hold_queue.lock);
 715
 716         pq = &new->polq;
 717
 718         spin_lock_bh(&pq->hold_queue.lock);
 719         skb_queue_splice(&list, &pq->hold_queue);
 720         pq->timeout = XFRM_QUEUE_TMO_MIN;
 721         if (!mod_timer(&pq->hold_timer, jiffies))
 722                 xfrm_pol_hold(new);
 723         spin_unlock_bh(&pq->hold_queue.lock);
 724 }
 725
 726 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 727                                    struct xfrm_policy *pol)
 728 {
 729         u32 mark = policy->mark.v & policy->mark.m;
 730
 731         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 732                 return true;
 733
 734         if ((mark & pol->mark.m) == pol->mark.v &&
 735             policy->priority == pol->priority)
 736                 return true;
 737
 738         return false;
 739 }
 740
 741 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 742 {
 743         struct net *net = xp_net(policy);
 744         struct xfrm_policy *pol;
 745         struct xfrm_policy *delpol;
 746         struct hlist_head *chain;
 747         struct hlist_node *newpos;
 748
 749         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 750         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 751         delpol = NULL;
 752         newpos = NULL;
 753         hlist_for_each_entry(pol, chain, bydst) {
 754                 if (pol->type == policy->type &&
 755                     !selector_cmp(&pol->selector, &policy->selector) &&
 756                     xfrm_policy_mark_match(policy, pol) &&
 757                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 758                     !WARN_ON(delpol)) {
 759                         if (excl) {
 760                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 761                                 return -EEXIST;
 762                         }
 763                         delpol = pol;
 764                         if (policy->priority > pol->priority)
 765                                 continue;
 766                 } else if (policy->priority >= pol->priority) {
 767                         newpos = &pol->bydst;
 768                         continue;
 769                 }
 770                 if (delpol)
 771                         break;
 772         }
 773         if (newpos)
 774                 hlist_add_behind(&policy->bydst, newpos);
 775         else
 776                 hlist_add_head(&policy->bydst, chain);
 777         __xfrm_policy_link(policy, dir);
 778         atomic_inc(&net->xfrm.flow_cache_genid);
 779
 780         /* After previous checking, family can either be AF_INET or AF_INET6 */
 781         if (policy->family == AF_INET)
 782                 rt_genid_bump_ipv4(net);
 783         else
 784                 rt_genid_bump_ipv6(net);
 785
 786         if (delpol) {
 787                 xfrm_policy_requeue(delpol, policy);
 788                 __xfrm_policy_unlink(delpol, dir);
 789         }
 790         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 791         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 792         policy->curlft.add_time = get_seconds();
 793         policy->curlft.use_time = 0;
 794         if (!mod_timer(&policy->timer, jiffies + HZ))
 795                 xfrm_pol_hold(policy);
 796         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 797
 798         if (delpol)
 799                 xfrm_policy_kill(delpol);
 800         else if (xfrm_bydst_should_resize(net, dir, NULL))
 801                 schedule_work(&net->xfrm.policy_hash_work);
 802
 803         return 0;
 804 }
 805 EXPORT_SYMBOL(xfrm_policy_insert);
 806
 807 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 808                                           int dir, struct xfrm_selector *sel,
 809                                           struct xfrm_sec_ctx *ctx, int delete,
 810                                           int *err)
 811 {
 812         struct xfrm_policy *pol, *ret;
 813         struct hlist_head *chain;
 814
 815         *err = 0;
 816         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 817         chain = policy_hash_bysel(net, sel, sel->family, dir);
 818         ret = NULL;
 819         hlist_for_each_entry(pol, chain, bydst) {
 820                 if (pol->type == type &&
 821                     (mark & pol->mark.m) == pol->mark.v &&
 822                     !selector_cmp(sel, &pol->selector) &&
 823                     xfrm_sec_ctx_match(ctx, pol->security)) {
 824                         xfrm_pol_hold(pol);
 825                         if (delete) {
 826                                 *err = security_xfrm_policy_delete(
 827                                                                 pol->security);
 828                                 if (*err) {
 829                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 830                                         return pol;
 831                                 }
 832                                 __xfrm_policy_unlink(pol, dir);
 833                         }
 834                         ret = pol;
 835                         break;
 836                 }
 837         }
 838         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 839
 840         if (ret && delete)
 841                 xfrm_policy_kill(ret);
 842         return ret;
 843 }
 844 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 845
 846 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 847                                      int dir, u32 id, int delete, int *err)
 848 {
 849         struct xfrm_policy *pol, *ret;
 850         struct hlist_head *chain;
 851
 852         *err = -ENOENT;
 853         if (xfrm_policy_id2dir(id) != dir)
 854                 return NULL;
 855
 856         *err = 0;
 857         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 858         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 859         ret = NULL;
 860         hlist_for_each_entry(pol, chain, byidx) {
 861                 if (pol->type == type && pol->index == id &&
 862                     (mark & pol->mark.m) == pol->mark.v) {
 863                         xfrm_pol_hold(pol);
 864                         if (delete) {
 865                                 *err = security_xfrm_policy_delete(
 866                                                                 pol->security);
 867                                 if (*err) {
 868                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 869                                         return pol;
 870                                 }
 871                                 __xfrm_policy_unlink(pol, dir);
 872                         }
 873                         ret = pol;
 874                         break;
 875                 }
 876         }
 877         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 878
 879         if (ret && delete)
 880                 xfrm_policy_kill(ret);
 881         return ret;
 882 }
 883 EXPORT_SYMBOL(xfrm_policy_byid);
 884
 885 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 886 static inline int
 887 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 888 {
 889         int dir, err = 0;
 890
 891         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 892                 struct xfrm_policy *pol;
 893                 int i;
 894
 895                 hlist_for_each_entry(pol,
 896                                      &net->xfrm.policy_inexact[dir], bydst) {
 897                         if (pol->type != type)
 898                                 continue;
 899                         err = security_xfrm_policy_delete(pol->security);
 900                         if (err) {
 901                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 902                                 return err;
 903                         }
 904                 }
 905                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 906                         hlist_for_each_entry(pol,
 907                                              net->xfrm.policy_bydst[dir].table + i,
 908                                              bydst) {
 909                                 if (pol->type != type)
 910                                         continue;
 911                                 err = security_xfrm_policy_delete(
 912                                                                 pol->security);
 913                                 if (err) {
 914                                         xfrm_audit_policy_delete(pol, 0,
 915                                                                  task_valid);
 916                                         return err;
 917                                 }
 918                         }
 919                 }
 920         }
 921         return err;
 922 }
 923 #else
 924 static inline int
 925 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 926 {
 927         return 0;
 928 }
 929 #endif
 930
 931 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 932 {
 933         int dir, err = 0, cnt = 0;
 934
 935         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 936
 937         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 938         if (err)
 939                 goto out;
 940
 941         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 942                 struct xfrm_policy *pol;
 943                 int i;
 944
 945         again1:
 946                 hlist_for_each_entry(pol,
 947                                      &net->xfrm.policy_inexact[dir], bydst) {
 948                         if (pol->type != type)
 949                                 continue;
 950                         __xfrm_policy_unlink(pol, dir);
 951                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 952                         cnt++;
 953
 954                         xfrm_audit_policy_delete(pol, 1, task_valid);
 955
 956                         xfrm_policy_kill(pol);
 957
 958                         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 959                         goto again1;
 960                 }
 961
 962                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 963         again2:
 964                         hlist_for_each_entry(pol,
 965                                              net->xfrm.policy_bydst[dir].table + i,
 966                                              bydst) {
 967                                 if (pol->type != type)
 968                                         continue;
 969                                 __xfrm_policy_unlink(pol, dir);
 970                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 971                                 cnt++;
 972
 973                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 974                                 xfrm_policy_kill(pol);
 975
 976                                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
 977                                 goto again2;
 978                         }
 979                 }
 980
 981         }
 982         if (!cnt)
 983                 err = -ESRCH;
 984 out:
 985         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 986         return err;
 987 }
 988 EXPORT_SYMBOL(xfrm_policy_flush);
 989
 990 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 991                      int (*func)(struct xfrm_policy *, int, int, void*),
 992                      void *data)
 993 {
 994         struct xfrm_policy *pol;
 995         struct xfrm_policy_walk_entry *x;
 996         int error = 0;
 997
 998         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 999             walk->type != XFRM_POLICY_TYPE_ANY)
1000                 return -EINVAL;
1001
1002         if (list_empty(&walk->walk.all) && walk->seq != 0)
1003                 return 0;
1004
1005         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1006         if (list_empty(&walk->walk.all))
1007                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1008         else
1009                 x = list_first_entry(&walk->walk.all,
1010                                      struct xfrm_policy_walk_entry, all);
1011
1012         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1013                 if (x->dead)
1014                         continue;
1015                 pol = container_of(x, struct xfrm_policy, walk);
1016                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1017                     walk->type != pol->type)
1018                         continue;
1019                 error = func(pol, xfrm_policy_id2dir(pol->index),
1020                              walk->seq, data);
1021                 if (error) {
1022                         list_move_tail(&walk->walk.all, &x->all);
1023                         goto out;
1024                 }
1025                 walk->seq++;
1026         }
1027         if (walk->seq == 0) {
1028                 error = -ENOENT;
1029                 goto out;
1030         }
1031         list_del_init(&walk->walk.all);
1032 out:
1033         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1034         return error;
1035 }
1036 EXPORT_SYMBOL(xfrm_policy_walk);
1037
1038 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1039 {
1040         INIT_LIST_HEAD(&walk->walk.all);
1041         walk->walk.dead = 1;
1042         walk->type = type;
1043         walk->seq = 0;
1044 }
1045 EXPORT_SYMBOL(xfrm_policy_walk_init);
1046
1047 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1048 {
1049         if (list_empty(&walk->walk.all))
1050                 return;
1051
1052         write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1053         list_del(&walk->walk.all);
1054         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1055 }
1056 EXPORT_SYMBOL(xfrm_policy_walk_done);
1057
1058 /*
1059  * Find policy to apply to this flow.
1060  *
1061  * Returns 0 if policy found, else an -errno.
1062  */
1063 static int xfrm_policy_match(const struct xfrm_policy *pol,
1064                              const struct flowi *fl,
1065                              u8 type, u16 family, int dir)
1066 {
1067         const struct xfrm_selector *sel = &pol->selector;
1068         int ret = -ESRCH;
1069         bool match;
1070
1071         if (pol->family != family ||
1072             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1073             pol->type != type)
1074                 return ret;
1075
1076         match = xfrm_selector_match(sel, fl, family);
1077         if (match)
1078                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1079                                                   dir);
1080
1081         return ret;
1082 }
1083
1084 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1085                                                      const struct flowi *fl,
1086                                                      u16 family, u8 dir)
1087 {
1088         int err;
1089         struct xfrm_policy *pol, *ret;
1090         const xfrm_address_t *daddr, *saddr;
1091         struct hlist_head *chain;
1092         u32 priority = ~0U;
1093
1094         daddr = xfrm_flowi_daddr(fl, family);
1095         saddr = xfrm_flowi_saddr(fl, family);
1096         if (unlikely(!daddr || !saddr))
1097                 return NULL;
1098
1099         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1100         chain = policy_hash_direct(net, daddr, saddr, family, dir);
1101         ret = NULL;
1102         hlist_for_each_entry(pol, chain, bydst) {
1103                 err = xfrm_policy_match(pol, fl, type, family, dir);
1104                 if (err) {
1105                         if (err == -ESRCH)
1106                                 continue;
1107                         else {
1108                                 ret = ERR_PTR(err);
1109                                 goto fail;
1110                         }
1111                 } else {
1112                         ret = pol;
1113                         priority = ret->priority;
1114                         break;
1115                 }
1116         }
1117         chain = &net->xfrm.policy_inexact[dir];
1118         hlist_for_each_entry(pol, chain, bydst) {
1119                 if ((pol->priority >= priority) && ret)
1120                         break;
1121
1122                 err = xfrm_policy_match(pol, fl, type, family, dir);
1123                 if (err) {
1124                         if (err == -ESRCH)
1125                                 continue;
1126                         else {
1127                                 ret = ERR_PTR(err);
1128                                 goto fail;
1129                         }
1130                 } else {
1131                         ret = pol;
1132                         break;
1133                 }
1134         }
1135
1136         xfrm_pol_hold(ret);
1137 fail:
1138         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1139
1140         return ret;
1141 }
1142
1143 static struct xfrm_policy *
1144 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1145 {
1146 #ifdef CONFIG_XFRM_SUB_POLICY
1147         struct xfrm_policy *pol;
1148
1149         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1150         if (pol != NULL)
1151                 return pol;
1152 #endif
1153         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1154 }
1155
1156 static int flow_to_policy_dir(int dir)
1157 {
1158         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1159             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1160             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1161                 return dir;
1162
1163         switch (dir) {
1164         default:
1165         case FLOW_DIR_IN:
1166                 return XFRM_POLICY_IN;
1167         case FLOW_DIR_OUT:
1168                 return XFRM_POLICY_OUT;
1169         case FLOW_DIR_FWD:
1170                 return XFRM_POLICY_FWD;
1171         }
1172 }
1173
1174 static struct flow_cache_object *
1175 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1176                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1177 {
1178         struct xfrm_policy *pol;
1179
1180         if (old_obj)
1181                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1182
1183         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1184         if (IS_ERR_OR_NULL(pol))
1185                 return ERR_CAST(pol);
1186
1187         /* Resolver returns two references:
1188          * one for cache and one for caller of flow_cache_lookup() */
1189         xfrm_pol_hold(pol);
1190
1191         return &pol->flo;
1192 }
1193
1194 static inline int policy_to_flow_dir(int dir)
1195 {
1196         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1197             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1198             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1199                 return dir;
1200         switch (dir) {
1201         default:
1202         case XFRM_POLICY_IN:
1203                 return FLOW_DIR_IN;
1204         case XFRM_POLICY_OUT:
1205                 return FLOW_DIR_OUT;
1206         case XFRM_POLICY_FWD:
1207                 return FLOW_DIR_FWD;
1208         }
1209 }
1210
1211 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
1212                                                  const struct flowi *fl)
1213 {
1214         struct xfrm_policy *pol;
1215         struct net *net = sock_net(sk);
1216
1217         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1218         if ((pol = sk->sk_policy[dir]) != NULL) {
1219                 bool match = xfrm_selector_match(&pol->selector, fl,
1220                                                  sk->sk_family);
1221                 int err = 0;
1222
1223                 if (match) {
1224                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1225                                 pol = NULL;
1226                                 goto out;
1227                         }
1228                         err = security_xfrm_policy_lookup(pol->security,
1229                                                       fl->flowi_secid,
1230                                                       policy_to_flow_dir(dir));
1231                         if (!err)
1232                                 xfrm_pol_hold(pol);
1233                         else if (err == -ESRCH)
1234                                 pol = NULL;
1235                         else
1236                                 pol = ERR_PTR(err);
1237                 } else
1238                         pol = NULL;
1239         }
1240 out:
1241         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1242         return pol;
1243 }
1244
1245 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1246 {
1247         struct net *net = xp_net(pol);
1248
1249         list_add(&pol->walk.all, &net->xfrm.policy_all);
1250         net->xfrm.policy_count[dir]++;
1251         xfrm_pol_hold(pol);
1252 }
1253
1254 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1255                                                 int dir)
1256 {
1257         struct net *net = xp_net(pol);
1258
1259         if (list_empty(&pol->walk.all))
1260                 return NULL;
1261
1262         /* Socket policies are not hashed. */
1263         if (!hlist_unhashed(&pol->bydst)) {
1264                 hlist_del(&pol->bydst);
1265                 hlist_del(&pol->byidx);
1266         }
1267
1268         list_del_init(&pol->walk.all);
1269         net->xfrm.policy_count[dir]--;
1270
1271         return pol;
1272 }
1273
1274 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1275 {
1276         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1277 }
1278
1279 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1280 {
1281         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1282 }
1283
1284 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1285 {
1286         struct net *net = xp_net(pol);
1287
1288         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1289         pol = __xfrm_policy_unlink(pol, dir);
1290         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1291         if (pol) {
1292                 xfrm_policy_kill(pol);
1293                 return 0;
1294         }
1295         return -ENOENT;
1296 }
1297 EXPORT_SYMBOL(xfrm_policy_delete);
1298
1299 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1300 {
1301         struct net *net = xp_net(pol);
1302         struct xfrm_policy *old_pol;
1303
1304 #ifdef CONFIG_XFRM_SUB_POLICY
1305         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1306                 return -EINVAL;
1307 #endif
1308
1309         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1310         old_pol = sk->sk_policy[dir];
1311         sk->sk_policy[dir] = pol;
1312         if (pol) {
1313                 pol->curlft.add_time = get_seconds();
1314                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1315                 xfrm_sk_policy_link(pol, dir);
1316         }
1317         if (old_pol) {
1318                 if (pol)
1319                         xfrm_policy_requeue(old_pol, pol);
1320
1321                 /* Unlinking succeeds always. This is the only function
1322                  * allowed to delete or replace socket policy.
1323                  */
1324                 xfrm_sk_policy_unlink(old_pol, dir);
1325         }
1326         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1327
1328         if (old_pol) {
1329                 xfrm_policy_kill(old_pol);
1330         }
1331         return 0;
1332 }
1333
1334 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1335 {
1336         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1337         struct net *net = xp_net(old);
1338
1339         if (newp) {
1340                 newp->selector = old->selector;
1341                 if (security_xfrm_policy_clone(old->security,
1342                                                &newp->security)) {
1343                         kfree(newp);
1344                         return NULL;  /* ENOMEM */
1345                 }
1346                 newp->lft = old->lft;
1347                 newp->curlft = old->curlft;
1348                 newp->mark = old->mark;
1349                 newp->action = old->action;
1350                 newp->flags = old->flags;
1351                 newp->xfrm_nr = old->xfrm_nr;
1352                 newp->index = old->index;
1353                 newp->type = old->type;
1354                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1355                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1356                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1357                 xfrm_sk_policy_link(newp, dir);
1358                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1359                 xfrm_pol_put(newp);
1360         }
1361         return newp;
1362 }
1363
1364 int __xfrm_sk_clone_policy(struct sock *sk)
1365 {
1366         struct xfrm_policy *p0 = sk->sk_policy[0],
1367                            *p1 = sk->sk_policy[1];
1368
1369         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1370         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1371                 return -ENOMEM;
1372         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1373                 return -ENOMEM;
1374         return 0;
1375 }
1376
1377 static int
1378 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
1379                xfrm_address_t *remote, unsigned short family)
1380 {
1381         int err;
1382         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1383
1384         if (unlikely(afinfo == NULL))
1385                 return -EINVAL;
1386         err = afinfo->get_saddr(net, oif, local, remote);
1387         xfrm_policy_put_afinfo(afinfo);
1388         return err;
1389 }
1390
1391 /* Resolve list of templates for the flow, given policy. */
1392
1393 static int
1394 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1395                       struct xfrm_state **xfrm, unsigned short family)
1396 {
1397         struct net *net = xp_net(policy);
1398         int nx;
1399         int i, error;
1400         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1401         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1402         xfrm_address_t tmp;
1403
1404         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1405                 struct xfrm_state *x;
1406                 xfrm_address_t *remote = daddr;
1407                 xfrm_address_t *local  = saddr;
1408                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1409
1410                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1411                     tmpl->mode == XFRM_MODE_BEET) {
1412                         remote = &tmpl->id.daddr;
1413                         local = &tmpl->saddr;
1414                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1415                                 error = xfrm_get_saddr(net, fl->flowi_oif,
1416                                                        &tmp, remote,
1417                                                        tmpl->encap_family);
1418                                 if (error)
1419                                         goto fail;
1420                                 local = &tmp;
1421                         }
1422                 }
1423
1424                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1425
1426                 if (x && x->km.state == XFRM_STATE_VALID) {
1427                         xfrm[nx++] = x;
1428                         daddr = remote;
1429                         saddr = local;
1430                         continue;
1431                 }
1432                 if (x) {
1433                         error = (x->km.state == XFRM_STATE_ERROR ?
1434                                  -EINVAL : -EAGAIN);
1435                         xfrm_state_put(x);
1436                 } else if (error == -ESRCH) {
1437                         error = -EAGAIN;
1438                 }
1439
1440                 if (!tmpl->optional)
1441                         goto fail;
1442         }
1443         return nx;
1444
1445 fail:
1446         for (nx--; nx >= 0; nx--)
1447                 xfrm_state_put(xfrm[nx]);
1448         return error;
1449 }
1450
1451 static int
1452 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1453                   struct xfrm_state **xfrm, unsigned short family)
1454 {
1455         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1456         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1457         int cnx = 0;
1458         int error;
1459         int ret;
1460         int i;
1461
1462         for (i = 0; i < npols; i++) {
1463                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1464                         error = -ENOBUFS;
1465                         goto fail;
1466                 }
1467
1468                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1469                 if (ret < 0) {
1470                         error = ret;
1471                         goto fail;
1472                 } else
1473                         cnx += ret;
1474         }
1475
1476         /* found states are sorted for outbound processing */
1477         if (npols > 1)
1478                 xfrm_state_sort(xfrm, tpp, cnx, family);
1479
1480         return cnx;
1481
1482  fail:
1483         for (cnx--; cnx >= 0; cnx--)
1484                 xfrm_state_put(tpp[cnx]);
1485         return error;
1486
1487 }
1488
1489 /* Check that the bundle accepts the flow and its components are
1490  * still valid.
1491  */
1492
1493 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1494 {
1495         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1496         int tos;
1497
1498         if (!afinfo)
1499                 return -EINVAL;
1500
1501         tos = afinfo->get_tos(fl);
1502
1503         xfrm_policy_put_afinfo(afinfo);
1504
1505         return tos;
1506 }
1507
1508 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1509 {
1510         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1511         struct dst_entry *dst = &xdst->u.dst;
1512
1513         if (xdst->route == NULL) {
1514                 /* Dummy bundle - if it has xfrms we were not
1515                  * able to build bundle as template resolution failed.
1516                  * It means we need to try again resolving. */
1517                 if (xdst->num_xfrms > 0)
1518                         return NULL;
1519         } else if (dst->flags & DST_XFRM_QUEUE) {
1520                 return NULL;
1521         } else {
1522                 /* Real bundle */
1523                 if (stale_bundle(dst))
1524                         return NULL;
1525         }
1526
1527         dst_hold(dst);
1528         return flo;
1529 }
1530
1531 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1532 {
1533         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1534         struct dst_entry *dst = &xdst->u.dst;
1535
1536         if (!xdst->route)
1537                 return 0;
1538         if (stale_bundle(dst))
1539                 return 0;
1540
1541         return 1;
1542 }
1543
1544 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1545 {
1546         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1547         struct dst_entry *dst = &xdst->u.dst;
1548
1549         dst_free(dst);
1550 }
1551
1552 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1553         .get = xfrm_bundle_flo_get,
1554         .check = xfrm_bundle_flo_check,
1555         .delete = xfrm_bundle_flo_delete,
1556 };
1557
1558 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1559 {
1560         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1561         struct dst_ops *dst_ops;
1562         struct xfrm_dst *xdst;
1563
1564         if (!afinfo)
1565                 return ERR_PTR(-EINVAL);
1566
1567         switch (family) {
1568         case AF_INET:
1569                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1570                 break;
1571 #if IS_ENABLED(CONFIG_IPV6)
1572         case AF_INET6:
1573                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1574                 break;
1575 #endif
1576         default:
1577                 BUG();
1578         }
1579         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1580
1581         if (likely(xdst)) {
1582                 struct dst_entry *dst = &xdst->u.dst;
1583
1584                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1585                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1586         } else
1587                 xdst = ERR_PTR(-ENOBUFS);
1588
1589         xfrm_policy_put_afinfo(afinfo);
1590
1591         return xdst;
1592 }
1593
1594 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1595                                  int nfheader_len)
1596 {
1597         struct xfrm_policy_afinfo *afinfo =
1598                 xfrm_policy_get_afinfo(dst->ops->family);
1599         int err;
1600
1601         if (!afinfo)
1602                 return -EINVAL;
1603
1604         err = afinfo->init_path(path, dst, nfheader_len);
1605
1606         xfrm_policy_put_afinfo(afinfo);
1607
1608         return err;
1609 }
1610
1611 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1612                                 const struct flowi *fl)
1613 {
1614         struct xfrm_policy_afinfo *afinfo =
1615                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1616         int err;
1617
1618         if (!afinfo)
1619                 return -EINVAL;
1620
1621         err = afinfo->fill_dst(xdst, dev, fl);
1622
1623         xfrm_policy_put_afinfo(afinfo);
1624
1625         return err;
1626 }
1627
1628
1629 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1630  * all the metrics... Shortly, bundle a bundle.
1631  */
1632
1633 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1634                                             struct xfrm_state **xfrm, int nx,
1635                                             const struct flowi *fl,
1636                                             struct dst_entry *dst)
1637 {
1638         struct net *net = xp_net(policy);
1639         unsigned long now = jiffies;
1640         struct net_device *dev;
1641         struct xfrm_mode *inner_mode;
1642         struct dst_entry *dst_prev = NULL;
1643         struct dst_entry *dst0 = NULL;
1644         int i = 0;
1645         int err;
1646         int header_len = 0;
1647         int nfheader_len = 0;
1648         int trailer_len = 0;
1649         int tos;
1650         int family = policy->selector.family;
1651         xfrm_address_t saddr, daddr;
1652
1653         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1654
1655         tos = xfrm_get_tos(fl, family);
1656         err = tos;
1657         if (tos < 0)
1658                 goto put_states;
1659
1660         dst_hold(dst);
1661
1662         for (; i < nx; i++) {
1663                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1664                 struct dst_entry *dst1 = &xdst->u.dst;
1665
1666                 err = PTR_ERR(xdst);
1667                 if (IS_ERR(xdst)) {
1668                         dst_release(dst);
1669                         goto put_states;
1670                 }
1671
1672                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1673                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1674                                                         xfrm_af2proto(family));
1675                         if (!inner_mode) {
1676                                 err = -EAFNOSUPPORT;
1677                                 dst_release(dst);
1678                                 goto put_states;
1679                         }
1680                 } else
1681                         inner_mode = xfrm[i]->inner_mode;
1682
1683                 if (!dst_prev)
1684                         dst0 = dst1;
1685                 else {
1686                         dst_prev->child = dst_clone(dst1);
1687                         dst1->flags |= DST_NOHASH;
1688                 }
1689
1690                 xdst->route = dst;
1691                 dst_copy_metrics(dst1, dst);
1692
1693                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1694                         family = xfrm[i]->props.family;
1695                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
1696                                               &saddr, &daddr, family);
1697                         err = PTR_ERR(dst);
1698                         if (IS_ERR(dst))
1699                                 goto put_states;
1700                 } else
1701                         dst_hold(dst);
1702
1703                 dst1->xfrm = xfrm[i];
1704                 xdst->xfrm_genid = xfrm[i]->genid;
1705
1706                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1707                 dst1->flags |= DST_HOST;
1708                 dst1->lastuse = now;
1709
1710                 dst1->input = dst_discard;
1711                 dst1->output = inner_mode->afinfo->output;
1712
1713                 dst1->next = dst_prev;
1714                 dst_prev = dst1;
1715
1716                 header_len += xfrm[i]->props.header_len;
1717                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1718                         nfheader_len += xfrm[i]->props.header_len;
1719                 trailer_len += xfrm[i]->props.trailer_len;
1720         }
1721
1722         dst_prev->child = dst;
1723         dst0->path = dst;
1724
1725         err = -ENODEV;
1726         dev = dst->dev;
1727         if (!dev)
1728                 goto free_dst;
1729
1730         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1731         xfrm_init_pmtu(dst_prev);
1732
1733         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1734                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1735
1736                 err = xfrm_fill_dst(xdst, dev, fl);
1737                 if (err)
1738                         goto free_dst;
1739
1740                 dst_prev->header_len = header_len;
1741                 dst_prev->trailer_len = trailer_len;
1742                 header_len -= xdst->u.dst.xfrm->props.header_len;
1743                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1744         }
1745
1746 out:
1747         return dst0;
1748
1749 put_states:
1750         for (; i < nx; i++)
1751                 xfrm_state_put(xfrm[i]);
1752 free_dst:
1753         if (dst0)
1754                 dst_free(dst0);
1755         dst0 = ERR_PTR(err);
1756         goto out;
1757 }
1758
1759 #ifdef CONFIG_XFRM_SUB_POLICY
1760 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1761 {
1762         if (!*target) {
1763                 *target = kmalloc(size, GFP_ATOMIC);
1764                 if (!*target)
1765                         return -ENOMEM;
1766         }
1767
1768         memcpy(*target, src, size);
1769         return 0;
1770 }
1771 #endif
1772
1773 static int xfrm_dst_update_parent(struct dst_entry *dst,
1774                                   const struct xfrm_selector *sel)
1775 {
1776 #ifdef CONFIG_XFRM_SUB_POLICY
1777         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1778         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1779                                    sel, sizeof(*sel));
1780 #else
1781         return 0;
1782 #endif
1783 }
1784
1785 static int xfrm_dst_update_origin(struct dst_entry *dst,
1786                                   const struct flowi *fl)
1787 {
1788 #ifdef CONFIG_XFRM_SUB_POLICY
1789         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1790         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1791 #else
1792         return 0;
1793 #endif
1794 }
1795
1796 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1797                                 struct xfrm_policy **pols,
1798                                 int *num_pols, int *num_xfrms)
1799 {
1800         int i;
1801
1802         if (*num_pols == 0 || !pols[0]) {
1803                 *num_pols = 0;
1804                 *num_xfrms = 0;
1805                 return 0;
1806         }
1807         if (IS_ERR(pols[0]))
1808                 return PTR_ERR(pols[0]);
1809
1810         *num_xfrms = pols[0]->xfrm_nr;
1811
1812 #ifdef CONFIG_XFRM_SUB_POLICY
1813         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1814             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1815                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1816                                                     XFRM_POLICY_TYPE_MAIN,
1817                                                     fl, family,
1818                                                     XFRM_POLICY_OUT);
1819                 if (pols[1]) {
1820                         if (IS_ERR(pols[1])) {
1821                                 xfrm_pols_put(pols, *num_pols);
1822                                 return PTR_ERR(pols[1]);
1823                         }
1824                         (*num_pols)++;
1825                         (*num_xfrms) += pols[1]->xfrm_nr;
1826                 }
1827         }
1828 #endif
1829         for (i = 0; i < *num_pols; i++) {
1830                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1831                         *num_xfrms = -1;
1832                         break;
1833                 }
1834         }
1835
1836         return 0;
1837
1838 }
1839
1840 static struct xfrm_dst *
1841 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1842                                const struct flowi *fl, u16 family,
1843                                struct dst_entry *dst_orig)
1844 {
1845         struct net *net = xp_net(pols[0]);
1846         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1847         struct dst_entry *dst;
1848         struct xfrm_dst *xdst;
1849         int err;
1850
1851         /* Try to instantiate a bundle */
1852         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1853         if (err <= 0) {
1854                 if (err != 0 && err != -EAGAIN)
1855                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1856                 return ERR_PTR(err);
1857         }
1858
1859         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1860         if (IS_ERR(dst)) {
1861                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1862                 return ERR_CAST(dst);
1863         }
1864
1865         xdst = (struct xfrm_dst *)dst;
1866         xdst->num_xfrms = err;
1867         if (num_pols > 1)
1868                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1869         else
1870                 err = xfrm_dst_update_origin(dst, fl);
1871         if (unlikely(err)) {
1872                 dst_free(dst);
1873                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1874                 return ERR_PTR(err);
1875         }
1876
1877         xdst->num_pols = num_pols;
1878         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1879         xdst->policy_genid = atomic_read(&pols[0]->genid);
1880
1881         return xdst;
1882 }
1883
1884 static void xfrm_policy_queue_process(unsigned long arg)
1885 {
1886         struct sk_buff *skb;
1887         struct sock *sk;
1888         struct dst_entry *dst;
1889         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1890         struct net *net = xp_net(pol);
1891         struct xfrm_policy_queue *pq = &pol->polq;
1892         struct flowi fl;
1893         struct sk_buff_head list;
1894
1895         spin_lock(&pq->hold_queue.lock);
1896         skb = skb_peek(&pq->hold_queue);
1897         if (!skb) {
1898                 spin_unlock(&pq->hold_queue.lock);
1899                 goto out;
1900         }
1901         dst = skb_dst(skb);
1902         sk = skb->sk;
1903         xfrm_decode_session(skb, &fl, dst->ops->family);
1904         spin_unlock(&pq->hold_queue.lock);
1905
1906         dst_hold(dst->path);
1907         dst = xfrm_lookup(net, dst->path, &fl, sk, 0);
1908         if (IS_ERR(dst))
1909                 goto purge_queue;
1910
1911         if (dst->flags & DST_XFRM_QUEUE) {
1912                 dst_release(dst);
1913
1914                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1915                         goto purge_queue;
1916
1917                 pq->timeout = pq->timeout << 1;
1918                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1919                         xfrm_pol_hold(pol);
1920         goto out;
1921         }
1922
1923         dst_release(dst);
1924
1925         __skb_queue_head_init(&list);
1926
1927         spin_lock(&pq->hold_queue.lock);
1928         pq->timeout = 0;
1929         skb_queue_splice_init(&pq->hold_queue, &list);
1930         spin_unlock(&pq->hold_queue.lock);
1931
1932         while (!skb_queue_empty(&list)) {
1933                 skb = __skb_dequeue(&list);
1934
1935                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1936                 dst_hold(skb_dst(skb)->path);
1937                 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0);
1938                 if (IS_ERR(dst)) {
1939                         kfree_skb(skb);
1940                         continue;
1941                 }
1942
1943                 nf_reset(skb);
1944                 skb_dst_drop(skb);
1945                 skb_dst_set(skb, dst);
1946
1947                 dst_output(net, skb->sk, skb);
1948         }
1949
1950 out:
1951         xfrm_pol_put(pol);
1952         return;
1953
1954 purge_queue:
1955         pq->timeout = 0;
1956         skb_queue_purge(&pq->hold_queue);
1957         xfrm_pol_put(pol);
1958 }
1959
1960 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1961 {
1962         unsigned long sched_next;
1963         struct dst_entry *dst = skb_dst(skb);
1964         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1965         struct xfrm_policy *pol = xdst->pols[0];
1966         struct xfrm_policy_queue *pq = &pol->polq;
1967
1968         if (unlikely(skb_fclone_busy(sk, skb))) {
1969                 kfree_skb(skb);
1970                 return 0;
1971         }
1972
1973         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1974                 kfree_skb(skb);
1975                 return -EAGAIN;
1976         }
1977
1978         skb_dst_force(skb);
1979
1980         spin_lock_bh(&pq->hold_queue.lock);
1981
1982         if (!pq->timeout)
1983                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1984
1985         sched_next = jiffies + pq->timeout;
1986
1987         if (del_timer(&pq->hold_timer)) {
1988                 if (time_before(pq->hold_timer.expires, sched_next))
1989                         sched_next = pq->hold_timer.expires;
1990                 xfrm_pol_put(pol);
1991         }
1992
1993         __skb_queue_tail(&pq->hold_queue, skb);
1994         if (!mod_timer(&pq->hold_timer, sched_next))
1995                 xfrm_pol_hold(pol);
1996
1997         spin_unlock_bh(&pq->hold_queue.lock);
1998
1999         return 0;
2000 }
2001
2002 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2003                                                  struct xfrm_flo *xflo,
2004                                                  const struct flowi *fl,
2005                                                  int num_xfrms,
2006                                                  u16 family)
2007 {
2008         int err;
2009         struct net_device *dev;
2010         struct dst_entry *dst;
2011         struct dst_entry *dst1;
2012         struct xfrm_dst *xdst;
2013
2014         xdst = xfrm_alloc_dst(net, family);
2015         if (IS_ERR(xdst))
2016                 return xdst;
2017
2018         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2019             net->xfrm.sysctl_larval_drop ||
2020             num_xfrms <= 0)
2021                 return xdst;
2022
2023         dst = xflo->dst_orig;
2024         dst1 = &xdst->u.dst;
2025         dst_hold(dst);
2026         xdst->route = dst;
2027
2028         dst_copy_metrics(dst1, dst);
2029
2030         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2031         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2032         dst1->lastuse = jiffies;
2033
2034         dst1->input = dst_discard;
2035         dst1->output = xdst_queue_output;
2036
2037         dst_hold(dst);
2038         dst1->child = dst;
2039         dst1->path = dst;
2040
2041         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2042
2043         err = -ENODEV;
2044         dev = dst->dev;
2045         if (!dev)
2046                 goto free_dst;
2047
2048         err = xfrm_fill_dst(xdst, dev, fl);
2049         if (err)
2050                 goto free_dst;
2051
2052 out:
2053         return xdst;
2054
2055 free_dst:
2056         dst_release(dst1);
2057         xdst = ERR_PTR(err);
2058         goto out;
2059 }
2060
2061 static struct flow_cache_object *
2062 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2063                    struct flow_cache_object *oldflo, void *ctx)
2064 {
2065         struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2066         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2067         struct xfrm_dst *xdst, *new_xdst;
2068         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2069
2070         /* Check if the policies from old bundle are usable */
2071         xdst = NULL;
2072         if (oldflo) {
2073                 xdst = container_of(oldflo, struct xfrm_dst, flo);
2074                 num_pols = xdst->num_pols;
2075                 num_xfrms = xdst->num_xfrms;
2076                 pol_dead = 0;
2077                 for (i = 0; i < num_pols; i++) {
2078                         pols[i] = xdst->pols[i];
2079                         pol_dead |= pols[i]->walk.dead;
2080                 }
2081                 if (pol_dead) {
2082                         dst_free(&xdst->u.dst);
2083                         xdst = NULL;
2084                         num_pols = 0;
2085                         num_xfrms = 0;
2086                         oldflo = NULL;
2087                 }
2088         }
2089
2090         /* Resolve policies to use if we couldn't get them from
2091          * previous cache entry */
2092         if (xdst == NULL) {
2093                 num_pols = 1;
2094                 pols[0] = __xfrm_policy_lookup(net, fl, family,
2095                                                flow_to_policy_dir(dir));
2096                 err = xfrm_expand_policies(fl, family, pols,
2097                                            &num_pols, &num_xfrms);
2098                 if (err < 0)
2099                         goto inc_error;
2100                 if (num_pols == 0)
2101                         return NULL;
2102                 if (num_xfrms <= 0)
2103                         goto make_dummy_bundle;
2104         }
2105
2106         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2107                                                   xflo->dst_orig);
2108         if (IS_ERR(new_xdst)) {
2109                 err = PTR_ERR(new_xdst);
2110                 if (err != -EAGAIN)
2111                         goto error;
2112                 if (oldflo == NULL)
2113                         goto make_dummy_bundle;
2114                 dst_hold(&xdst->u.dst);
2115                 return oldflo;
2116         } else if (new_xdst == NULL) {
2117                 num_xfrms = 0;
2118                 if (oldflo == NULL)
2119                         goto make_dummy_bundle;
2120                 xdst->num_xfrms = 0;
2121                 dst_hold(&xdst->u.dst);
2122                 return oldflo;
2123         }
2124
2125         /* Kill the previous bundle */
2126         if (xdst) {
2127                 /* The policies were stolen for newly generated bundle */
2128                 xdst->num_pols = 0;
2129                 dst_free(&xdst->u.dst);
2130         }
2131
2132         /* Flow cache does not have reference, it dst_free()'s,
2133          * but we do need to return one reference for original caller */
2134         dst_hold(&new_xdst->u.dst);
2135         return &new_xdst->flo;
2136
2137 make_dummy_bundle:
2138         /* We found policies, but there's no bundles to instantiate:
2139          * either because the policy blocks, has no transformations or
2140          * we could not build template (no xfrm_states).*/
2141         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2142         if (IS_ERR(xdst)) {
2143                 xfrm_pols_put(pols, num_pols);
2144                 return ERR_CAST(xdst);
2145         }
2146         xdst->num_pols = num_pols;
2147         xdst->num_xfrms = num_xfrms;
2148         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2149
2150         dst_hold(&xdst->u.dst);
2151         return &xdst->flo;
2152
2153 inc_error:
2154         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2155 error:
2156         if (xdst != NULL)
2157                 dst_free(&xdst->u.dst);
2158         else
2159                 xfrm_pols_put(pols, num_pols);
2160         return ERR_PTR(err);
2161 }
2162
2163 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2164                                         struct dst_entry *dst_orig)
2165 {
2166         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2167         struct dst_entry *ret;
2168
2169         if (!afinfo) {
2170                 dst_release(dst_orig);
2171                 return ERR_PTR(-EINVAL);
2172         } else {
2173                 ret = afinfo->blackhole_route(net, dst_orig);
2174         }
2175         xfrm_policy_put_afinfo(afinfo);
2176
2177         return ret;
2178 }
2179
2180 /* Main function: finds/creates a bundle for given flow.
2181  *
2182  * At the moment we eat a raw IP route. Mostly to speed up lookups
2183  * on interfaces with disabled IPsec.
2184  */
2185 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2186                               const struct flowi *fl,
2187                               const struct sock *sk, int flags)
2188 {
2189         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2190         struct flow_cache_object *flo;
2191         struct xfrm_dst *xdst;
2192         struct dst_entry *dst, *route;
2193         u16 family = dst_orig->ops->family;
2194         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2195         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2196
2197         dst = NULL;
2198         xdst = NULL;
2199         route = NULL;
2200
2201         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2202                 num_pols = 1;
2203                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2204                 err = xfrm_expand_policies(fl, family, pols,
2205                                            &num_pols, &num_xfrms);
2206                 if (err < 0)
2207                         goto dropdst;
2208
2209                 if (num_pols) {
2210                         if (num_xfrms <= 0) {
2211                                 drop_pols = num_pols;
2212                                 goto no_transform;
2213                         }
2214
2215                         xdst = xfrm_resolve_and_create_bundle(
2216                                         pols, num_pols, fl,
2217                                         family, dst_orig);
2218                         if (IS_ERR(xdst)) {
2219                                 xfrm_pols_put(pols, num_pols);
2220                                 err = PTR_ERR(xdst);
2221                                 goto dropdst;
2222                         } else if (xdst == NULL) {
2223                                 num_xfrms = 0;
2224                                 drop_pols = num_pols;
2225                                 goto no_transform;
2226                         }
2227
2228                         dst_hold(&xdst->u.dst);
2229                         xdst->u.dst.flags |= DST_NOCACHE;
2230                         route = xdst->route;
2231                 }
2232         }
2233
2234         if (xdst == NULL) {
2235                 struct xfrm_flo xflo;
2236
2237                 xflo.dst_orig = dst_orig;
2238                 xflo.flags = flags;
2239
2240                 /* To accelerate a bit...  */
2241                 if ((dst_orig->flags & DST_NOXFRM) ||
2242                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2243                         goto nopol;
2244
2245                 flo = flow_cache_lookup(net, fl, family, dir,
2246                                         xfrm_bundle_lookup, &xflo);
2247                 if (flo == NULL)
2248                         goto nopol;
2249                 if (IS_ERR(flo)) {
2250                         err = PTR_ERR(flo);
2251                         goto dropdst;
2252                 }
2253                 xdst = container_of(flo, struct xfrm_dst, flo);
2254
2255                 num_pols = xdst->num_pols;
2256                 num_xfrms = xdst->num_xfrms;
2257                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2258                 route = xdst->route;
2259         }
2260
2261         dst = &xdst->u.dst;
2262         if (route == NULL && num_xfrms > 0) {
2263                 /* The only case when xfrm_bundle_lookup() returns a
2264                  * bundle with null route, is when the template could
2265                  * not be resolved. It means policies are there, but
2266                  * bundle could not be created, since we don't yet
2267                  * have the xfrm_state's. We need to wait for KM to
2268                  * negotiate new SA's or bail out with error.*/
2269                 if (net->xfrm.sysctl_larval_drop) {
2270                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2271                         err = -EREMOTE;
2272                         goto error;
2273                 }
2274
2275                 err = -EAGAIN;
2276
2277                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2278                 goto error;
2279         }
2280
2281 no_transform:
2282         if (num_pols == 0)
2283                 goto nopol;
2284
2285         if ((flags & XFRM_LOOKUP_ICMP) &&
2286             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2287                 err = -ENOENT;
2288                 goto error;
2289         }
2290
2291         for (i = 0; i < num_pols; i++)
2292                 pols[i]->curlft.use_time = get_seconds();
2293
2294         if (num_xfrms < 0) {
2295                 /* Prohibit the flow */
2296                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2297                 err = -EPERM;
2298                 goto error;
2299         } else if (num_xfrms > 0) {
2300                 /* Flow transformed */
2301                 dst_release(dst_orig);
2302         } else {
2303                 /* Flow passes untransformed */
2304                 dst_release(dst);
2305                 dst = dst_orig;
2306         }
2307 ok:
2308         xfrm_pols_put(pols, drop_pols);
2309         if (dst && dst->xfrm &&
2310             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2311                 dst->flags |= DST_XFRM_TUNNEL;
2312         return dst;
2313
2314 nopol:
2315         if (!(flags & XFRM_LOOKUP_ICMP)) {
2316                 dst = dst_orig;
2317                 goto ok;
2318         }
2319         err = -ENOENT;
2320 error:
2321         dst_release(dst);
2322 dropdst:
2323         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
2324                 dst_release(dst_orig);
2325         xfrm_pols_put(pols, drop_pols);
2326         return ERR_PTR(err);
2327 }
2328 EXPORT_SYMBOL(xfrm_lookup);
2329
2330 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2331  * Otherwise we may send out blackholed packets.
2332  */
2333 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2334                                     const struct flowi *fl,
2335                                     const struct sock *sk, int flags)
2336 {
2337         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2338                                             flags | XFRM_LOOKUP_QUEUE |
2339                                             XFRM_LOOKUP_KEEP_DST_REF);
2340
2341         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2342                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2343
2344         return dst;
2345 }
2346 EXPORT_SYMBOL(xfrm_lookup_route);
2347
2348 static inline int
2349 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2350 {
2351         struct xfrm_state *x;
2352
2353         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2354                 return 0;
2355         x = skb->sp->xvec[idx];
2356         if (!x->type->reject)
2357                 return 0;
2358         return x->type->reject(x, skb, fl);
2359 }
2360
2361 /* When skb is transformed back to its "native" form, we have to
2362  * check policy restrictions. At the moment we make this in maximally
2363  * stupid way. Shame on me. :-) Of course, connected sockets must
2364  * have policy cached at them.
2365  */
2366
2367 static inline int
2368 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2369               unsigned short family)
2370 {
2371         if (xfrm_state_kern(x))
2372                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2373         return  x->id.proto == tmpl->id.proto &&
2374                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2375                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2376                 x->props.mode == tmpl->mode &&
2377                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2378                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2379                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2380                   xfrm_state_addr_cmp(tmpl, x, family));
2381 }
2382
2383 /*
2384  * 0 or more than 0 is returned when validation is succeeded (either bypass
2385  * because of optional transport mode, or next index of the mathced secpath
2386  * state with the template.
2387  * -1 is returned when no matching template is found.
2388  * Otherwise "-2 - errored_index" is returned.
2389  */
2390 static inline int
2391 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2392                unsigned short family)
2393 {
2394         int idx = start;
2395
2396         if (tmpl->optional) {
2397                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2398                         return start;
2399         } else
2400                 start = -1;
2401         for (; idx < sp->len; idx++) {
2402                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2403                         return ++idx;
2404                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2405                         if (start == -1)
2406                                 start = -2-idx;
2407                         break;
2408                 }
2409         }
2410         return start;
2411 }
2412
2413 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2414                           unsigned int family, int reverse)
2415 {
2416         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2417         int err;
2418
2419         if (unlikely(afinfo == NULL))
2420                 return -EAFNOSUPPORT;
2421
2422         afinfo->decode_session(skb, fl, reverse);
2423         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2424         xfrm_policy_put_afinfo(afinfo);
2425         return err;
2426 }
2427 EXPORT_SYMBOL(__xfrm_decode_session);
2428
2429 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2430 {
2431         for (; k < sp->len; k++) {
2432                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2433                         *idxp = k;
2434                         return 1;
2435                 }
2436         }
2437
2438         return 0;
2439 }
2440
2441 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2442                         unsigned short family)
2443 {
2444         struct net *net = dev_net(skb->dev);
2445         struct xfrm_policy *pol;
2446         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2447         int npols = 0;
2448         int xfrm_nr;
2449         int pi;
2450         int reverse;
2451         struct flowi fl;
2452         u8 fl_dir;
2453         int xerr_idx = -1;
2454
2455         reverse = dir & ~XFRM_POLICY_MASK;
2456         dir &= XFRM_POLICY_MASK;
2457         fl_dir = policy_to_flow_dir(dir);
2458
2459         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2460                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2461                 return 0;
2462         }
2463
2464         nf_nat_decode_session(skb, &fl, family);
2465
2466         /* First, check used SA against their selectors. */
2467         if (skb->sp) {
2468                 int i;
2469
2470                 for (i = skb->sp->len-1; i >= 0; i--) {
2471                         struct xfrm_state *x = skb->sp->xvec[i];
2472                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2473                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2474                                 return 0;
2475                         }
2476                 }
2477         }
2478
2479         pol = NULL;
2480         if (sk && sk->sk_policy[dir]) {
2481                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2482                 if (IS_ERR(pol)) {
2483                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2484                         return 0;
2485                 }
2486         }
2487
2488         if (!pol) {
2489                 struct flow_cache_object *flo;
2490
2491                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2492                                         xfrm_policy_lookup, NULL);
2493                 if (IS_ERR_OR_NULL(flo))
2494                         pol = ERR_CAST(flo);
2495                 else
2496                         pol = container_of(flo, struct xfrm_policy, flo);
2497         }
2498
2499         if (IS_ERR(pol)) {
2500                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2501                 return 0;
2502         }
2503
2504         if (!pol) {
2505                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2506                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2507                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2508                         return 0;
2509                 }
2510                 return 1;
2511         }
2512
2513         pol->curlft.use_time = get_seconds();
2514
2515         pols[0] = pol;
2516         npols++;
2517 #ifdef CONFIG_XFRM_SUB_POLICY
2518         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2519                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2520                                                     &fl, family,
2521                                                     XFRM_POLICY_IN);
2522                 if (pols[1]) {
2523                         if (IS_ERR(pols[1])) {
2524                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2525                                 return 0;
2526                         }
2527                         pols[1]->curlft.use_time = get_seconds();
2528                         npols++;
2529                 }
2530         }
2531 #endif
2532
2533         if (pol->action == XFRM_POLICY_ALLOW) {
2534                 struct sec_path *sp;
2535                 static struct sec_path dummy;
2536                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2537                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2538                 struct xfrm_tmpl **tpp = tp;
2539                 int ti = 0;
2540                 int i, k;
2541
2542                 if ((sp = skb->sp) == NULL)
2543                         sp = &dummy;
2544
2545                 for (pi = 0; pi < npols; pi++) {
2546                         if (pols[pi] != pol &&
2547                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2548                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2549                                 goto reject;
2550                         }
2551                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2552                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2553                                 goto reject_error;
2554                         }
2555                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2556                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2557                 }
2558                 xfrm_nr = ti;
2559                 if (npols > 1) {
2560                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2561                         tpp = stp;
2562                 }
2563
2564                 /* For each tunnel xfrm, find the first matching tmpl.
2565                  * For each tmpl before that, find corresponding xfrm.
2566                  * Order is _important_. Later we will implement
2567                  * some barriers, but at the moment barriers
2568                  * are implied between each two transformations.
2569                  */
2570                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2571                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2572                         if (k < 0) {
2573                                 if (k < -1)
2574                                         /* "-2 - errored_index" returned */
2575                                         xerr_idx = -(2+k);
2576                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2577                                 goto reject;
2578                         }
2579                 }
2580
2581                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2582                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2583                         goto reject;
2584                 }
2585
2586                 xfrm_pols_put(pols, npols);
2587                 return 1;
2588         }
2589         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2590
2591 reject:
2592         xfrm_secpath_reject(xerr_idx, skb, &fl);
2593 reject_error:
2594         xfrm_pols_put(pols, npols);
2595         return 0;
2596 }
2597 EXPORT_SYMBOL(__xfrm_policy_check);
2598
2599 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2600 {
2601         struct net *net = dev_net(skb->dev);
2602         struct flowi fl;
2603         struct dst_entry *dst;
2604         int res = 1;
2605
2606         if (xfrm_decode_session(skb, &fl, family) < 0) {
2607                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2608                 return 0;
2609         }
2610
2611         skb_dst_force(skb);
2612
2613         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2614         if (IS_ERR(dst)) {
2615                 res = 0;
2616                 dst = NULL;
2617         }
2618         skb_dst_set(skb, dst);
2619         return res;
2620 }
2621 EXPORT_SYMBOL(__xfrm_route_forward);
2622
2623 /* Optimize later using cookies and generation ids. */
2624
2625 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2626 {
2627         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2628          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2629          * get validated by dst_ops->check on every use.  We do this
2630          * because when a normal route referenced by an XFRM dst is
2631          * obsoleted we do not go looking around for all parent
2632          * referencing XFRM dsts so that we can invalidate them.  It
2633          * is just too much work.  Instead we make the checks here on
2634          * every use.  For example:
2635          *
2636          *      XFRM dst A --> IPv4 dst X
2637          *
2638          * X is the "xdst->route" of A (X is also the "dst->path" of A
2639          * in this example).  If X is marked obsolete, "A" will not
2640          * notice.  That's what we are validating here via the
2641          * stale_bundle() check.
2642          *
2643          * When a policy's bundle is pruned, we dst_free() the XFRM
2644          * dst which causes it's ->obsolete field to be set to
2645          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2646          * this, we want to force a new route lookup.
2647          */
2648         if (dst->obsolete < 0 && !stale_bundle(dst))
2649                 return dst;
2650
2651         return NULL;
2652 }
2653
2654 static int stale_bundle(struct dst_entry *dst)
2655 {
2656         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2657 }
2658
2659 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2660 {
2661         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2662                 dst->dev = dev_net(dev)->loopback_dev;
2663                 dev_hold(dst->dev);
2664                 dev_put(dev);
2665         }
2666 }
2667 EXPORT_SYMBOL(xfrm_dst_ifdown);
2668
2669 static void xfrm_link_failure(struct sk_buff *skb)
2670 {
2671         /* Impossible. Such dst must be popped before reaches point of failure. */
2672 }
2673
2674 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2675 {
2676         if (dst) {
2677                 if (dst->obsolete) {
2678                         dst_release(dst);
2679                         dst = NULL;
2680                 }
2681         }
2682         return dst;
2683 }
2684
2685 void xfrm_garbage_collect(struct net *net)
2686 {
2687         flow_cache_flush(net);
2688 }
2689 EXPORT_SYMBOL(xfrm_garbage_collect);
2690
2691 static void xfrm_garbage_collect_deferred(struct net *net)
2692 {
2693         flow_cache_flush_deferred(net);
2694 }
2695
2696 static void xfrm_init_pmtu(struct dst_entry *dst)
2697 {
2698         do {
2699                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2700                 u32 pmtu, route_mtu_cached;
2701
2702                 pmtu = dst_mtu(dst->child);
2703                 xdst->child_mtu_cached = pmtu;
2704
2705                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2706
2707                 route_mtu_cached = dst_mtu(xdst->route);
2708                 xdst->route_mtu_cached = route_mtu_cached;
2709
2710                 if (pmtu > route_mtu_cached)
2711                         pmtu = route_mtu_cached;
2712
2713                 dst_metric_set(dst, RTAX_MTU, pmtu);
2714         } while ((dst = dst->next));
2715 }
2716
2717 /* Check that the bundle accepts the flow and its components are
2718  * still valid.
2719  */
2720
2721 static int xfrm_bundle_ok(struct xfrm_dst *first)
2722 {
2723         struct dst_entry *dst = &first->u.dst;
2724         struct xfrm_dst *last;
2725         u32 mtu;
2726
2727         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2728             (dst->dev && !netif_running(dst->dev)))
2729                 return 0;
2730
2731         if (dst->flags & DST_XFRM_QUEUE)
2732                 return 1;
2733
2734         last = NULL;
2735
2736         do {
2737                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2738
2739                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2740                         return 0;
2741                 if (xdst->xfrm_genid != dst->xfrm->genid)
2742                         return 0;
2743                 if (xdst->num_pols > 0 &&
2744                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2745                         return 0;
2746
2747                 mtu = dst_mtu(dst->child);
2748                 if (xdst->child_mtu_cached != mtu) {
2749                         last = xdst;
2750                         xdst->child_mtu_cached = mtu;
2751                 }
2752
2753                 if (!dst_check(xdst->route, xdst->route_cookie))
2754                         return 0;
2755                 mtu = dst_mtu(xdst->route);
2756                 if (xdst->route_mtu_cached != mtu) {
2757                         last = xdst;
2758                         xdst->route_mtu_cached = mtu;
2759                 }
2760
2761                 dst = dst->child;
2762         } while (dst->xfrm);
2763
2764         if (likely(!last))
2765                 return 1;
2766
2767         mtu = last->child_mtu_cached;
2768         for (;;) {
2769                 dst = &last->u.dst;
2770
2771                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2772                 if (mtu > last->route_mtu_cached)
2773                         mtu = last->route_mtu_cached;
2774                 dst_metric_set(dst, RTAX_MTU, mtu);
2775
2776                 if (last == first)
2777                         break;
2778
2779                 last = (struct xfrm_dst *)last->u.dst.next;
2780                 last->child_mtu_cached = mtu;
2781         }
2782
2783         return 1;
2784 }
2785
2786 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2787 {
2788         return dst_metric_advmss(dst->path);
2789 }
2790
2791 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2792 {
2793         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2794
2795         return mtu ? : dst_mtu(dst->path);
2796 }
2797
2798 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2799                                            struct sk_buff *skb,
2800                                            const void *daddr)
2801 {
2802         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2803 }
2804
2805 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2806 {
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         return err;
2838 }
2839 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2840
2841 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2842 {
2843         int err = 0;
2844         if (unlikely(afinfo == NULL))
2845                 return -EINVAL;
2846         if (unlikely(afinfo->family >= NPROTO))
2847                 return -EAFNOSUPPORT;
2848         spin_lock(&xfrm_policy_afinfo_lock);
2849         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2850                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2851                         err = -EINVAL;
2852                 else
2853                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2854                                          NULL);
2855         }
2856         spin_unlock(&xfrm_policy_afinfo_lock);
2857         if (!err) {
2858                 struct dst_ops *dst_ops = afinfo->dst_ops;
2859
2860                 synchronize_rcu();
2861
2862                 dst_ops->kmem_cachep = NULL;
2863                 dst_ops->check = NULL;
2864                 dst_ops->negative_advice = NULL;
2865                 dst_ops->link_failure = NULL;
2866                 afinfo->garbage_collect = NULL;
2867         }
2868         return err;
2869 }
2870 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2871
2872 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2873 {
2874         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2875
2876         switch (event) {
2877         case NETDEV_DOWN:
2878                 xfrm_garbage_collect(dev_net(dev));
2879         }
2880         return NOTIFY_DONE;
2881 }
2882
2883 static struct notifier_block xfrm_dev_notifier = {
2884         .notifier_call  = xfrm_dev_event,
2885 };
2886
2887 #ifdef CONFIG_XFRM_STATISTICS
2888 static int __net_init xfrm_statistics_init(struct net *net)
2889 {
2890         int rv;
2891         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2892         if (!net->mib.xfrm_statistics)
2893                 return -ENOMEM;
2894         rv = xfrm_proc_init(net);
2895         if (rv < 0)
2896                 free_percpu(net->mib.xfrm_statistics);
2897         return rv;
2898 }
2899
2900 static void xfrm_statistics_fini(struct net *net)
2901 {
2902         xfrm_proc_fini(net);
2903         free_percpu(net->mib.xfrm_statistics);
2904 }
2905 #else
2906 static int __net_init xfrm_statistics_init(struct net *net)
2907 {
2908         return 0;
2909 }
2910
2911 static void xfrm_statistics_fini(struct net *net)
2912 {
2913 }
2914 #endif
2915
2916 static int __net_init xfrm_policy_init(struct net *net)
2917 {
2918         unsigned int hmask, sz;
2919         int dir;
2920
2921         if (net_eq(net, &init_net))
2922                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2923                                            sizeof(struct xfrm_dst),
2924                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2925                                            NULL);
2926
2927         hmask = 8 - 1;
2928         sz = (hmask+1) * sizeof(struct hlist_head);
2929
2930         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2931         if (!net->xfrm.policy_byidx)
2932                 goto out_byidx;
2933         net->xfrm.policy_idx_hmask = hmask;
2934
2935         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2936                 struct xfrm_policy_hash *htab;
2937
2938                 net->xfrm.policy_count[dir] = 0;
2939                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2940                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2941
2942                 htab = &net->xfrm.policy_bydst[dir];
2943                 htab->table = xfrm_hash_alloc(sz);
2944                 if (!htab->table)
2945                         goto out_bydst;
2946                 htab->hmask = hmask;
2947                 htab->dbits4 = 32;
2948                 htab->sbits4 = 32;
2949                 htab->dbits6 = 128;
2950                 htab->sbits6 = 128;
2951         }
2952         net->xfrm.policy_hthresh.lbits4 = 32;
2953         net->xfrm.policy_hthresh.rbits4 = 32;
2954         net->xfrm.policy_hthresh.lbits6 = 128;
2955         net->xfrm.policy_hthresh.rbits6 = 128;
2956
2957         seqlock_init(&net->xfrm.policy_hthresh.lock);
2958
2959         INIT_LIST_HEAD(&net->xfrm.policy_all);
2960         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2961         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2962         if (net_eq(net, &init_net))
2963                 register_netdevice_notifier(&xfrm_dev_notifier);
2964         return 0;
2965
2966 out_bydst:
2967         for (dir--; dir >= 0; dir--) {
2968                 struct xfrm_policy_hash *htab;
2969
2970                 htab = &net->xfrm.policy_bydst[dir];
2971                 xfrm_hash_free(htab->table, sz);
2972         }
2973         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2974 out_byidx:
2975         return -ENOMEM;
2976 }
2977
2978 static void xfrm_policy_fini(struct net *net)
2979 {
2980         unsigned int sz;
2981         int dir;
2982
2983         flush_work(&net->xfrm.policy_hash_work);
2984 #ifdef CONFIG_XFRM_SUB_POLICY
2985         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
2986 #endif
2987         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
2988
2989         WARN_ON(!list_empty(&net->xfrm.policy_all));
2990
2991         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2992                 struct xfrm_policy_hash *htab;
2993
2994                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2995
2996                 htab = &net->xfrm.policy_bydst[dir];
2997                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2998                 WARN_ON(!hlist_empty(htab->table));
2999                 xfrm_hash_free(htab->table, sz);
3000         }
3001
3002         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3003         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3004         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3005 }
3006
3007 static int __net_init xfrm_net_init(struct net *net)
3008 {
3009         int rv;
3010
3011         rv = xfrm_statistics_init(net);
3012         if (rv < 0)
3013                 goto out_statistics;
3014         rv = xfrm_state_init(net);
3015         if (rv < 0)
3016                 goto out_state;
3017         rv = xfrm_policy_init(net);
3018         if (rv < 0)
3019                 goto out_policy;
3020         rv = xfrm_sysctl_init(net);
3021         if (rv < 0)
3022                 goto out_sysctl;
3023         rv = flow_cache_init(net);
3024         if (rv < 0)
3025                 goto out;
3026
3027         /* Initialize the per-net locks here */
3028         spin_lock_init(&net->xfrm.xfrm_state_lock);
3029         rwlock_init(&net->xfrm.xfrm_policy_lock);
3030         mutex_init(&net->xfrm.xfrm_cfg_mutex);
3031
3032         return 0;
3033
3034 out:
3035         xfrm_sysctl_fini(net);
3036 out_sysctl:
3037         xfrm_policy_fini(net);
3038 out_policy:
3039         xfrm_state_fini(net);
3040 out_state:
3041         xfrm_statistics_fini(net);
3042 out_statistics:
3043         return rv;
3044 }
3045
3046 static void __net_exit xfrm_net_exit(struct net *net)
3047 {
3048         flow_cache_fini(net);
3049         xfrm_sysctl_fini(net);
3050         xfrm_policy_fini(net);
3051         xfrm_state_fini(net);
3052         xfrm_statistics_fini(net);
3053 }
3054
3055 static struct pernet_operations __net_initdata xfrm_net_ops = {
3056         .init = xfrm_net_init,
3057         .exit = xfrm_net_exit,
3058 };
3059
3060 void __init xfrm_init(void)
3061 {
3062         register_pernet_subsys(&xfrm_net_ops);
3063         xfrm_input_init();
3064 }
3065
3066 #ifdef CONFIG_AUDITSYSCALL
3067 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3068                                          struct audit_buffer *audit_buf)
3069 {
3070         struct xfrm_sec_ctx *ctx = xp->security;
3071         struct xfrm_selector *sel = &xp->selector;
3072
3073         if (ctx)
3074                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3075                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3076
3077         switch (sel->family) {
3078         case AF_INET:
3079                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3080                 if (sel->prefixlen_s != 32)
3081                         audit_log_format(audit_buf, " src_prefixlen=%d",
3082                                          sel->prefixlen_s);
3083                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3084                 if (sel->prefixlen_d != 32)
3085                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3086                                          sel->prefixlen_d);
3087                 break;
3088         case AF_INET6:
3089                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3090                 if (sel->prefixlen_s != 128)
3091                         audit_log_format(audit_buf, " src_prefixlen=%d",
3092                                          sel->prefixlen_s);
3093                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3094                 if (sel->prefixlen_d != 128)
3095                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3096                                          sel->prefixlen_d);
3097                 break;
3098         }
3099 }
3100
3101 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3102 {
3103         struct audit_buffer *audit_buf;
3104
3105         audit_buf = xfrm_audit_start("SPD-add");
3106         if (audit_buf == NULL)
3107                 return;
3108         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3109         audit_log_format(audit_buf, " res=%u", result);
3110         xfrm_audit_common_policyinfo(xp, audit_buf);
3111         audit_log_end(audit_buf);
3112 }
3113 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3114
3115 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3116                               bool task_valid)
3117 {
3118         struct audit_buffer *audit_buf;
3119
3120         audit_buf = xfrm_audit_start("SPD-delete");
3121         if (audit_buf == NULL)
3122                 return;
3123         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3124         audit_log_format(audit_buf, " res=%u", result);
3125         xfrm_audit_common_policyinfo(xp, audit_buf);
3126         audit_log_end(audit_buf);
3127 }
3128 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3129 #endif
3130
3131 #ifdef CONFIG_XFRM_MIGRATE
3132 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3133                                         const struct xfrm_selector *sel_tgt)
3134 {
3135         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3136                 if (sel_tgt->family == sel_cmp->family &&
3137                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3138                                     sel_cmp->family) &&
3139                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3140                                     sel_cmp->family) &&
3141                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3142                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3143                         return true;
3144                 }
3145         } else {
3146                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3147                         return true;
3148                 }
3149         }
3150         return false;
3151 }
3152
3153 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3154                                                     u8 dir, u8 type, struct net *net)
3155 {
3156         struct xfrm_policy *pol, *ret = NULL;
3157         struct hlist_head *chain;
3158         u32 priority = ~0U;
3159
3160         read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3161         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3162         hlist_for_each_entry(pol, chain, bydst) {
3163                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3164                     pol->type == type) {
3165                         ret = pol;
3166                         priority = ret->priority;
3167                         break;
3168                 }
3169         }
3170         chain = &net->xfrm.policy_inexact[dir];
3171         hlist_for_each_entry(pol, chain, bydst) {
3172                 if ((pol->priority >= priority) && ret)
3173                         break;
3174
3175                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3176                     pol->type == type) {
3177                         ret = pol;
3178                         break;
3179                 }
3180         }
3181
3182         xfrm_pol_hold(ret);
3183
3184         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3185
3186         return ret;
3187 }
3188
3189 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3190 {
3191         int match = 0;
3192
3193         if (t->mode == m->mode && t->id.proto == m->proto &&
3194             (m->reqid == 0 || t->reqid == m->reqid)) {
3195                 switch (t->mode) {
3196                 case XFRM_MODE_TUNNEL:
3197                 case XFRM_MODE_BEET:
3198                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3199                                             m->old_family) &&
3200                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3201                                             m->old_family)) {
3202                                 match = 1;
3203                         }
3204                         break;
3205                 case XFRM_MODE_TRANSPORT:
3206                         /* in case of transport mode, template does not store
3207                            any IP addresses, hence we just compare mode and
3208                            protocol */
3209                         match = 1;
3210                         break;
3211                 default:
3212                         break;
3213                 }
3214         }
3215         return match;
3216 }
3217
3218 /* update endpoint address(es) of template(s) */
3219 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3220                                struct xfrm_migrate *m, int num_migrate)
3221 {
3222         struct xfrm_migrate *mp;
3223         int i, j, n = 0;
3224
3225         write_lock_bh(&pol->lock);
3226         if (unlikely(pol->walk.dead)) {
3227                 /* target policy has been deleted */
3228                 write_unlock_bh(&pol->lock);
3229                 return -ENOENT;
3230         }
3231
3232         for (i = 0; i < pol->xfrm_nr; i++) {
3233                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3234                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3235                                 continue;
3236                         n++;
3237                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3238                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3239                                 continue;
3240                         /* update endpoints */
3241                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3242                                sizeof(pol->xfrm_vec[i].id.daddr));
3243                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3244                                sizeof(pol->xfrm_vec[i].saddr));
3245                         pol->xfrm_vec[i].encap_family = mp->new_family;
3246                         /* flush bundles */
3247                         atomic_inc(&pol->genid);
3248                 }
3249         }
3250
3251         write_unlock_bh(&pol->lock);
3252
3253         if (!n)
3254                 return -ENODATA;
3255
3256         return 0;
3257 }
3258
3259 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3260 {
3261         int i, j;
3262
3263         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3264                 return -EINVAL;
3265
3266         for (i = 0; i < num_migrate; i++) {
3267                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3268                                     m[i].old_family) &&
3269                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3270                                     m[i].old_family))
3271                         return -EINVAL;
3272                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3273                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3274                         return -EINVAL;
3275
3276                 /* check if there is any duplicated entry */
3277                 for (j = i + 1; j < num_migrate; j++) {
3278                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3279                                     sizeof(m[i].old_daddr)) &&
3280                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3281                                     sizeof(m[i].old_saddr)) &&
3282                             m[i].proto == m[j].proto &&
3283                             m[i].mode == m[j].mode &&
3284                             m[i].reqid == m[j].reqid &&
3285                             m[i].old_family == m[j].old_family)
3286                                 return -EINVAL;
3287                 }
3288         }
3289
3290         return 0;
3291 }
3292
3293 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3294                  struct xfrm_migrate *m, int num_migrate,
3295                  struct xfrm_kmaddress *k, struct net *net)
3296 {
3297         int i, err, nx_cur = 0, nx_new = 0;
3298         struct xfrm_policy *pol = NULL;
3299         struct xfrm_state *x, *xc;
3300         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3301         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3302         struct xfrm_migrate *mp;
3303
3304         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3305                 goto out;
3306
3307         /* Stage 1 - find policy */
3308         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3309                 err = -ENOENT;
3310                 goto out;
3311         }
3312
3313         /* Stage 2 - find and update state(s) */
3314         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3315                 if ((x = xfrm_migrate_state_find(mp, net))) {
3316                         x_cur[nx_cur] = x;
3317                         nx_cur++;
3318                         if ((xc = xfrm_state_migrate(x, mp))) {
3319                                 x_new[nx_new] = xc;
3320                                 nx_new++;
3321                         } else {
3322                                 err = -ENODATA;
3323                                 goto restore_state;
3324                         }
3325                 }
3326         }
3327
3328         /* Stage 3 - update policy */
3329         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3330                 goto restore_state;
3331
3332         /* Stage 4 - delete old state(s) */
3333         if (nx_cur) {
3334                 xfrm_states_put(x_cur, nx_cur);
3335                 xfrm_states_delete(x_cur, nx_cur);
3336         }
3337
3338         /* Stage 5 - announce */
3339         km_migrate(sel, dir, type, m, num_migrate, k);
3340
3341         xfrm_pol_put(pol);
3342
3343         return 0;
3344 out:
3345         return err;
3346
3347 restore_state:
3348         if (pol)
3349                 xfrm_pol_put(pol);
3350         if (nx_cur)
3351                 xfrm_states_put(x_cur, nx_cur);
3352         if (nx_new)
3353                 xfrm_states_delete(x_new, nx_new);
3354
3355         return err;
3356 }
3357 EXPORT_SYMBOL(xfrm_migrate);
3358 #endif