net/xfrm/xfrm_policy.c

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