net/xfrm/xfrm_policy.c

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