net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/flow.h>
  30 #include <net/xfrm.h>
  31 #include <net/ip.h>
  32 #ifdef CONFIG_XFRM_STATISTICS
  33 #include <net/snmp.h>
  34 #endif
  35
  36 #include "xfrm_hash.h"
  37
  38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  40 #define XFRM_MAX_QUEUE_LEN      100
  41
  42 struct xfrm_flo {
  43         struct dst_entry *dst_orig;
  44         u8 flags;
  45 };
  46
  47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  48 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
  49                                                 __read_mostly;
  50
  51 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  52
  53 static void xfrm_init_pmtu(struct dst_entry *dst);
  54 static int stale_bundle(struct dst_entry *dst);
  55 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  56 static void xfrm_policy_queue_process(unsigned long arg);
  57
  58 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  59                                                 int dir);
  60
  61 static inline bool
  62 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  63 {
  64         const struct flowi4 *fl4 = &fl->u.ip4;
  65
  66         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  67                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  68                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  69                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  70                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  71                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  72 }
  73
  74 static inline bool
  75 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  76 {
  77         const struct flowi6 *fl6 = &fl->u.ip6;
  78
  79         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  80                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  81                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  82                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  83                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  84                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  85 }
  86
  87 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  88                          unsigned short family)
  89 {
  90         switch (family) {
  91         case AF_INET:
  92                 return __xfrm4_selector_match(sel, fl);
  93         case AF_INET6:
  94                 return __xfrm6_selector_match(sel, fl);
  95         }
  96         return false;
  97 }
  98
  99 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 100 {
 101         struct xfrm_policy_afinfo *afinfo;
 102
 103         if (unlikely(family >= NPROTO))
 104                 return NULL;
 105         rcu_read_lock();
 106         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 107         if (unlikely(!afinfo))
 108                 rcu_read_unlock();
 109         return afinfo;
 110 }
 111
 112 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
 113 {
 114         rcu_read_unlock();
 115 }
 116
 117 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
 118                                                   const xfrm_address_t *saddr,
 119                                                   const xfrm_address_t *daddr,
 120                                                   int family)
 121 {
 122         struct xfrm_policy_afinfo *afinfo;
 123         struct dst_entry *dst;
 124
 125         afinfo = xfrm_policy_get_afinfo(family);
 126         if (unlikely(afinfo == NULL))
 127                 return ERR_PTR(-EAFNOSUPPORT);
 128
 129         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 130
 131         xfrm_policy_put_afinfo(afinfo);
 132
 133         return dst;
 134 }
 135
 136 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 137                                                 xfrm_address_t *prev_saddr,
 138                                                 xfrm_address_t *prev_daddr,
 139                                                 int family)
 140 {
 141         struct net *net = xs_net(x);
 142         xfrm_address_t *saddr = &x->props.saddr;
 143         xfrm_address_t *daddr = &x->id.daddr;
 144         struct dst_entry *dst;
 145
 146         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 147                 saddr = x->coaddr;
 148                 daddr = prev_daddr;
 149         }
 150         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 151                 saddr = prev_saddr;
 152                 daddr = x->coaddr;
 153         }
 154
 155         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 156
 157         if (!IS_ERR(dst)) {
 158                 if (prev_saddr != saddr)
 159                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 160                 if (prev_daddr != daddr)
 161                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 162         }
 163
 164         return dst;
 165 }
 166
 167 static inline unsigned long make_jiffies(long secs)
 168 {
 169         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 170                 return MAX_SCHEDULE_TIMEOUT-1;
 171         else
 172                 return secs*HZ;
 173 }
 174
 175 static void xfrm_policy_timer(unsigned long data)
 176 {
 177         struct xfrm_policy *xp = (struct xfrm_policy *)data;
 178         unsigned long now = get_seconds();
 179         long next = LONG_MAX;
 180         int warn = 0;
 181         int dir;
 182
 183         read_lock(&xp->lock);
 184
 185         if (unlikely(xp->walk.dead))
 186                 goto out;
 187
 188         dir = xfrm_policy_id2dir(xp->index);
 189
 190         if (xp->lft.hard_add_expires_seconds) {
 191                 long tmo = xp->lft.hard_add_expires_seconds +
 192                         xp->curlft.add_time - now;
 193                 if (tmo <= 0)
 194                         goto expired;
 195                 if (tmo < next)
 196                         next = tmo;
 197         }
 198         if (xp->lft.hard_use_expires_seconds) {
 199                 long tmo = xp->lft.hard_use_expires_seconds +
 200                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 201                 if (tmo <= 0)
 202                         goto expired;
 203                 if (tmo < next)
 204                         next = tmo;
 205         }
 206         if (xp->lft.soft_add_expires_seconds) {
 207                 long tmo = xp->lft.soft_add_expires_seconds +
 208                         xp->curlft.add_time - now;
 209                 if (tmo <= 0) {
 210                         warn = 1;
 211                         tmo = XFRM_KM_TIMEOUT;
 212                 }
 213                 if (tmo < next)
 214                         next = tmo;
 215         }
 216         if (xp->lft.soft_use_expires_seconds) {
 217                 long tmo = xp->lft.soft_use_expires_seconds +
 218                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 219                 if (tmo <= 0) {
 220                         warn = 1;
 221                         tmo = XFRM_KM_TIMEOUT;
 222                 }
 223                 if (tmo < next)
 224                         next = tmo;
 225         }
 226
 227         if (warn)
 228                 km_policy_expired(xp, dir, 0, 0);
 229         if (next != LONG_MAX &&
 230             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 231                 xfrm_pol_hold(xp);
 232
 233 out:
 234         read_unlock(&xp->lock);
 235         xfrm_pol_put(xp);
 236         return;
 237
 238 expired:
 239         read_unlock(&xp->lock);
 240         if (!xfrm_policy_delete(xp, dir))
 241                 km_policy_expired(xp, dir, 1, 0);
 242         xfrm_pol_put(xp);
 243 }
 244
 245 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 246 {
 247         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 248
 249         if (unlikely(pol->walk.dead))
 250                 flo = NULL;
 251         else
 252                 xfrm_pol_hold(pol);
 253
 254         return flo;
 255 }
 256
 257 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 258 {
 259         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 260
 261         return !pol->walk.dead;
 262 }
 263
 264 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 265 {
 266         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 267 }
 268
 269 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 270         .get = xfrm_policy_flo_get,
 271         .check = xfrm_policy_flo_check,
 272         .delete = xfrm_policy_flo_delete,
 273 };
 274
 275 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 276  * SPD calls.
 277  */
 278
 279 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 280 {
 281         struct xfrm_policy *policy;
 282
 283         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 284
 285         if (policy) {
 286                 write_pnet(&policy->xp_net, net);
 287                 INIT_LIST_HEAD(&policy->walk.all);
 288                 INIT_HLIST_NODE(&policy->bydst);
 289                 INIT_HLIST_NODE(&policy->byidx);
 290                 rwlock_init(&policy->lock);
 291                 atomic_set(&policy->refcnt, 1);
 292                 skb_queue_head_init(&policy->polq.hold_queue);
 293                 setup_timer(&policy->timer, xfrm_policy_timer,
 294                                 (unsigned long)policy);
 295                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 296                             (unsigned long)policy);
 297                 policy->flo.ops = &xfrm_policy_fc_ops;
 298         }
 299         return policy;
 300 }
 301 EXPORT_SYMBOL(xfrm_policy_alloc);
 302
 303 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 304
 305 void xfrm_policy_destroy(struct xfrm_policy *policy)
 306 {
 307         BUG_ON(!policy->walk.dead);
 308
 309         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 310                 BUG();
 311
 312         security_xfrm_policy_free(policy->security);
 313         kfree(policy);
 314 }
 315 EXPORT_SYMBOL(xfrm_policy_destroy);
 316
 317 static void xfrm_queue_purge(struct sk_buff_head *list)
 318 {
 319         struct sk_buff *skb;
 320
 321         while ((skb = skb_dequeue(list)) != NULL)
 322                 kfree_skb(skb);
 323 }
 324
 325 /* Rule must be locked. Release descentant resources, announce
 326  * entry dead. The rule must be unlinked from lists to the moment.
 327  */
 328
 329 static void xfrm_policy_kill(struct xfrm_policy *policy)
 330 {
 331         policy->walk.dead = 1;
 332
 333         atomic_inc(&policy->genid);
 334
 335         if (del_timer(&policy->polq.hold_timer))
 336                 xfrm_pol_put(policy);
 337         xfrm_queue_purge(&policy->polq.hold_queue);
 338
 339         if (del_timer(&policy->timer))
 340                 xfrm_pol_put(policy);
 341
 342         xfrm_pol_put(policy);
 343 }
 344
 345 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 346
 347 static inline unsigned int idx_hash(struct net *net, u32 index)
 348 {
 349         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 350 }
 351
 352 /* calculate policy hash thresholds */
 353 static void __get_hash_thresh(struct net *net,
 354                               unsigned short family, int dir,
 355                               u8 *dbits, u8 *sbits)
 356 {
 357         switch (family) {
 358         case AF_INET:
 359                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 360                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 361                 break;
 362
 363         case AF_INET6:
 364                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 365                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 366                 break;
 367
 368         default:
 369                 *dbits = 0;
 370                 *sbits = 0;
 371         }
 372 }
 373
 374 static struct hlist_head *policy_hash_bysel(struct net *net,
 375                                             const struct xfrm_selector *sel,
 376                                             unsigned short family, int dir)
 377 {
 378         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 379         unsigned int hash;
 380         u8 dbits;
 381         u8 sbits;
 382
 383         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 384         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 385
 386         return (hash == hmask + 1 ?
 387                 &net->xfrm.policy_inexact[dir] :
 388                 net->xfrm.policy_bydst[dir].table + hash);
 389 }
 390
 391 static struct hlist_head *policy_hash_direct(struct net *net,
 392                                              const xfrm_address_t *daddr,
 393                                              const xfrm_address_t *saddr,
 394                                              unsigned short family, int dir)
 395 {
 396         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 397         unsigned int hash;
 398         u8 dbits;
 399         u8 sbits;
 400
 401         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 402         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 403
 404         return net->xfrm.policy_bydst[dir].table + hash;
 405 }
 406
 407 static void xfrm_dst_hash_transfer(struct net *net,
 408                                    struct hlist_head *list,
 409                                    struct hlist_head *ndsttable,
 410                                    unsigned int nhashmask,
 411                                    int dir)
 412 {
 413         struct hlist_node *tmp, *entry0 = NULL;
 414         struct xfrm_policy *pol;
 415         unsigned int h0 = 0;
 416         u8 dbits;
 417         u8 sbits;
 418
 419 redo:
 420         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 421                 unsigned int h;
 422
 423                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 424                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 425                                 pol->family, nhashmask, dbits, sbits);
 426                 if (!entry0) {
 427                         hlist_del(&pol->bydst);
 428                         hlist_add_head(&pol->bydst, ndsttable+h);
 429                         h0 = h;
 430                 } else {
 431                         if (h != h0)
 432                                 continue;
 433                         hlist_del(&pol->bydst);
 434                         hlist_add_behind(&pol->bydst, entry0);
 435                 }
 436                 entry0 = &pol->bydst;
 437         }
 438         if (!hlist_empty(list)) {
 439                 entry0 = NULL;
 440                 goto redo;
 441         }
 442 }
 443
 444 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 445                                    struct hlist_head *nidxtable,
 446                                    unsigned int nhashmask)
 447 {
 448         struct hlist_node *tmp;
 449         struct xfrm_policy *pol;
 450
 451         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 452                 unsigned int h;
 453
 454                 h = __idx_hash(pol->index, nhashmask);
 455                 hlist_add_head(&pol->byidx, nidxtable+h);
 456         }
 457 }
 458
 459 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 460 {
 461         return ((old_hmask + 1) << 1) - 1;
 462 }
 463
 464 static void xfrm_bydst_resize(struct net *net, int dir)
 465 {
 466         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 467         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 468         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 469         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 470         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 471         int i;
 472
 473         if (!ndst)
 474                 return;
 475
 476         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 477
 478         for (i = hmask; i >= 0; i--)
 479                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 480
 481         net->xfrm.policy_bydst[dir].table = ndst;
 482         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 483
 484         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 485
 486         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 487 }
 488
 489 static void xfrm_byidx_resize(struct net *net, int total)
 490 {
 491         unsigned int hmask = net->xfrm.policy_idx_hmask;
 492         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 493         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 494         struct hlist_head *oidx = net->xfrm.policy_byidx;
 495         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 496         int i;
 497
 498         if (!nidx)
 499                 return;
 500
 501         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 502
 503         for (i = hmask; i >= 0; i--)
 504                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 505
 506         net->xfrm.policy_byidx = nidx;
 507         net->xfrm.policy_idx_hmask = nhashmask;
 508
 509         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 510
 511         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 512 }
 513
 514 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 515 {
 516         unsigned int cnt = net->xfrm.policy_count[dir];
 517         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 518
 519         if (total)
 520                 *total += cnt;
 521
 522         if ((hmask + 1) < xfrm_policy_hashmax &&
 523             cnt > hmask)
 524                 return 1;
 525
 526         return 0;
 527 }
 528
 529 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 530 {
 531         unsigned int hmask = net->xfrm.policy_idx_hmask;
 532
 533         if ((hmask + 1) < xfrm_policy_hashmax &&
 534             total > hmask)
 535                 return 1;
 536
 537         return 0;
 538 }
 539
 540 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 541 {
 542         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 543         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 544         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 545         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 546         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 547         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 548         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 549         si->spdhcnt = net->xfrm.policy_idx_hmask;
 550         si->spdhmcnt = xfrm_policy_hashmax;
 551         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
 552 }
 553 EXPORT_SYMBOL(xfrm_spd_getinfo);
 554
 555 static DEFINE_MUTEX(hash_resize_mutex);
 556 static void xfrm_hash_resize(struct work_struct *work)
 557 {
 558         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 559         int dir, total;
 560
 561         mutex_lock(&hash_resize_mutex);
 562
 563         total = 0;
 564         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 565                 if (xfrm_bydst_should_resize(net, dir, &total))
 566                         xfrm_bydst_resize(net, dir);
 567         }
 568         if (xfrm_byidx_should_resize(net, total))
 569                 xfrm_byidx_resize(net, total);
 570
 571         mutex_unlock(&hash_resize_mutex);
 572 }
 573
 574 static void xfrm_hash_rebuild(struct work_struct *work)
 575 {
 576         struct net *net = container_of(work, struct net,
 577                                        xfrm.policy_hthresh.work);
 578         unsigned int hmask;
 579         struct xfrm_policy *pol;
 580         struct xfrm_policy *policy;
 581         struct hlist_head *chain;
 582         struct hlist_head *odst;
 583         struct hlist_node *newpos;
 584         int i;
 585         int dir;
 586         unsigned seq;
 587         u8 lbits4, rbits4, lbits6, rbits6;
 588
 589         mutex_lock(&hash_resize_mutex);
 590
 591         /* read selector prefixlen thresholds */
 592         do {
 593                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
 594
 595                 lbits4 = net->xfrm.policy_hthresh.lbits4;
 596                 rbits4 = net->xfrm.policy_hthresh.rbits4;
 597                 lbits6 = net->xfrm.policy_hthresh.lbits6;
 598                 rbits6 = net->xfrm.policy_hthresh.rbits6;
 599         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
 600
 601         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 602
 603         /* reset the bydst and inexact table in all directions */
 604         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 605                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
 606                 hmask = net->xfrm.policy_bydst[dir].hmask;
 607                 odst = net->xfrm.policy_bydst[dir].table;
 608                 for (i = hmask; i >= 0; i--)
 609                         INIT_HLIST_HEAD(odst + i);
 610                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
 611                         /* dir out => dst = remote, src = local */
 612                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
 613                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
 614                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
 615                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
 616                 } else {
 617                         /* dir in/fwd => dst = local, src = remote */
 618                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
 619                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
 620                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
 621                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
 622                 }
 623         }
 624
 625         /* re-insert all policies by order of creation */
 626         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 627                 newpos = NULL;
 628                 chain = policy_hash_bysel(net, &policy->selector,
 629                                           policy->family,
 630                                           xfrm_policy_id2dir(policy->index));
 631                 hlist_for_each_entry(pol, chain, bydst) {
 632                         if (policy->priority >= pol->priority)
 633                                 newpos = &pol->bydst;
 634                         else
 635                                 break;
 636                 }
 637                 if (newpos)
 638                         hlist_add_behind(&policy->bydst, newpos);
 639                 else
 640                         hlist_add_head(&policy->bydst, chain);
 641         }
 642
 643         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 644
 645         mutex_unlock(&hash_resize_mutex);
 646 }
 647
 648 void xfrm_policy_hash_rebuild(struct net *net)
 649 {
 650         schedule_work(&net->xfrm.policy_hthresh.work);
 651 }
 652 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 653
 654 /* Generate new index... KAME seems to generate them ordered by cost
 655  * of an absolute inpredictability of ordering of rules. This will not pass. */
 656 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 657 {
 658         static u32 idx_generator;
 659
 660         for (;;) {
 661                 struct hlist_head *list;
 662                 struct xfrm_policy *p;
 663                 u32 idx;
 664                 int found;
 665
 666                 if (!index) {
 667                         idx = (idx_generator | dir);
 668                         idx_generator += 8;
 669                 } else {
 670                         idx = index;
 671                         index = 0;
 672                 }
 673
 674                 if (idx == 0)
 675                         idx = 8;
 676                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 677                 found = 0;
 678                 hlist_for_each_entry(p, list, byidx) {
 679                         if (p->index == idx) {
 680                                 found = 1;
 681                                 break;
 682                         }
 683                 }
 684                 if (!found)
 685                         return idx;
 686         }
 687 }
 688
 689 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 690 {
 691         u32 *p1 = (u32 *) s1;
 692         u32 *p2 = (u32 *) s2;
 693         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 694         int i;
 695
 696         for (i = 0; i < len; i++) {
 697                 if (p1[i] != p2[i])
 698                         return 1;
 699         }
 700
 701         return 0;
 702 }
 703
 704 static void xfrm_policy_requeue(struct xfrm_policy *old,
 705                                 struct xfrm_policy *new)
 706 {
 707         struct xfrm_policy_queue *pq = &old->polq;
 708         struct sk_buff_head list;
 709
 710         __skb_queue_head_init(&list);
 711
 712         spin_lock_bh(&pq->hold_queue.lock);
 713         skb_queue_splice_init(&pq->hold_queue, &list);
 714         if (del_timer(&pq->hold_timer))
 715                 xfrm_pol_put(old);
 716         spin_unlock_bh(&pq->hold_queue.lock);
 717
 718         if (skb_queue_empty(&list))
 719                 return;
 720
 721         pq = &new->polq;
 722
 723         spin_lock_bh(&pq->hold_queue.lock);
 724         skb_queue_splice(&list, &pq->hold_queue);
 725         pq->timeout = XFRM_QUEUE_TMO_MIN;
 726         if (!mod_timer(&pq->hold_timer, jiffies))
 727                 xfrm_pol_hold(new);
 728         spin_unlock_bh(&pq->hold_queue.lock);
 729 }
 730
 731 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 732                                    struct xfrm_policy *pol)
 733 {
 734         u32 mark = policy->mark.v & policy->mark.m;
 735
 736         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 737                 return true;
 738
 739         if ((mark & pol->mark.m) == pol->mark.v &&
 740             policy->priority == pol->priority)
 741                 return true;
 742
 743         return false;
 744 }
 745
 746 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 747 {
 748         struct net *net = xp_net(policy);
 749         struct xfrm_policy *pol;
 750         struct xfrm_policy *delpol;
 751         struct hlist_head *chain;
 752         struct hlist_node *newpos;
 753
 754         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 755         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 756         delpol = NULL;
 757         newpos = NULL;
 758         hlist_for_each_entry(pol, chain, bydst) {
 759                 if (pol->type == policy->type &&
 760                     !selector_cmp(&pol->selector, &policy->selector) &&
 761                     xfrm_policy_mark_match(policy, pol) &&
 762                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 763                     !WARN_ON(delpol)) {
 764                         if (excl) {
 765                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 766                                 return -EEXIST;
 767                         }
 768                         delpol = pol;
 769                         if (policy->priority > pol->priority)
 770                                 continue;
 771                 } else if (policy->priority >= pol->priority) {
 772                         newpos = &pol->bydst;
 773                         continue;
 774                 }
 775                 if (delpol)
 776                         break;
 777         }
 778         if (newpos)
 779                 hlist_add_behind(&policy->bydst, newpos);
 780         else
 781                 hlist_add_head(&policy->bydst, chain);
 782         xfrm_pol_hold(policy);
 783         net->xfrm.policy_count[dir]++;
 784         atomic_inc(&net->xfrm.flow_cache_genid);
 785
 786         /* After previous checking, family can either be AF_INET or AF_INET6 */
 787         if (policy->family == AF_INET)
 788                 rt_genid_bump_ipv4(net);
 789         else
 790                 rt_genid_bump_ipv6(net);
 791
 792         if (delpol) {
 793                 xfrm_policy_requeue(delpol, policy);
 794                 __xfrm_policy_unlink(delpol, dir);
 795         }
 796         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 797         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 798         policy->curlft.add_time = get_seconds();
 799         policy->curlft.use_time = 0;
 800         if (!mod_timer(&policy->timer, jiffies + HZ))
 801                 xfrm_pol_hold(policy);
 802         list_add(&policy->walk.all, &net->xfrm.policy_all);
 803         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 804
 805         if (delpol)
 806                 xfrm_policy_kill(delpol);
 807         else if (xfrm_bydst_should_resize(net, dir, NULL))
 808                 schedule_work(&net->xfrm.policy_hash_work);
 809
 810         return 0;
 811 }
 812 EXPORT_SYMBOL(xfrm_policy_insert);
 813
 814 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 815                                           int dir, struct xfrm_selector *sel,
 816                                           struct xfrm_sec_ctx *ctx, int delete,
 817                                           int *err)
 818 {
 819         struct xfrm_policy *pol, *ret;
 820         struct hlist_head *chain;
 821
 822         *err = 0;
 823         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 824         chain = policy_hash_bysel(net, sel, sel->family, dir);
 825         ret = NULL;
 826         hlist_for_each_entry(pol, chain, bydst) {
 827                 if (pol->type == type &&
 828                     (mark & pol->mark.m) == pol->mark.v &&
 829                     !selector_cmp(sel, &pol->selector) &&
 830                     xfrm_sec_ctx_match(ctx, pol->security)) {
 831                         xfrm_pol_hold(pol);
 832                         if (delete) {
 833                                 *err = security_xfrm_policy_delete(
 834                                                                 pol->security);
 835                                 if (*err) {
 836                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 837                                         return pol;
 838                                 }
 839                                 __xfrm_policy_unlink(pol, dir);
 840                         }
 841                         ret = pol;
 842                         break;
 843                 }
 844         }
 845         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 846
 847         if (ret && delete)
 848                 xfrm_policy_kill(ret);
 849         return ret;
 850 }
 851 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 852
 853 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 854                                      int dir, u32 id, int delete, int *err)
 855 {
 856         struct xfrm_policy *pol, *ret;
 857         struct hlist_head *chain;
 858
 859         *err = -ENOENT;
 860         if (xfrm_policy_id2dir(id) != dir)
 861                 return NULL;
 862
 863         *err = 0;
 864         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 865         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 866         ret = NULL;
 867         hlist_for_each_entry(pol, chain, byidx) {
 868                 if (pol->type == type && pol->index == id &&
 869                     (mark & pol->mark.m) == pol->mark.v) {
 870                         xfrm_pol_hold(pol);
 871                         if (delete) {
 872                                 *err = security_xfrm_policy_delete(
 873                                                                 pol->security);
 874                                 if (*err) {
 875                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 876                                         return pol;
 877                                 }
 878                                 __xfrm_policy_unlink(pol, dir);
 879                         }
 880                         ret = pol;
 881                         break;
 882                 }
 883         }
 884         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 885
 886         if (ret && delete)
 887                 xfrm_policy_kill(ret);
 888         return ret;
 889 }
 890 EXPORT_SYMBOL(xfrm_policy_byid);
 891
 892 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 893 static inline int
 894 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 895 {
 896         int dir, err = 0;
 897
 898         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 899                 struct xfrm_policy *pol;
 900                 int i;
 901
 902                 hlist_for_each_entry(pol,
 903                                      &net->xfrm.policy_inexact[dir], bydst) {
 904                         if (pol->type != type)
 905                                 continue;
 906                         err = security_xfrm_policy_delete(pol->security);
 907                         if (err) {
 908                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 909                                 return err;
 910                         }
 911                 }
 912                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 913                         hlist_for_each_entry(pol,
 914                                              net->xfrm.policy_bydst[dir].table + i,
 915                                              bydst) {
 916                                 if (pol->type != type)
 917                                         continue;
 918                                 err = security_xfrm_policy_delete(
 919                                                                 pol->security);
 920                                 if (err) {
 921                                         xfrm_audit_policy_delete(pol, 0,
 922                                                                  task_valid);
 923                                         return err;
 924                                 }
 925                         }
 926                 }
 927         }
 928         return err;
 929 }
 930 #else
 931 static inline int
 932 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 933 {
 934         return 0;
 935 }
 936 #endif
 937
 938 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 939 {
 940         int dir, err = 0, cnt = 0;
 941
 942         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 943
 944         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 945         if (err)
 946                 goto out;
 947
 948         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 949                 struct xfrm_policy *pol;
 950                 int i;
 951
 952         again1:
 953                 hlist_for_each_entry(pol,
 954                                      &net->xfrm.policy_inexact[dir], bydst) {
 955                         if (pol->type != type)
 956                                 continue;
 957                         __xfrm_policy_unlink(pol, dir);
 958                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 959                         cnt++;
 960
 961                         xfrm_audit_policy_delete(pol, 1, task_valid);
 962
 963                         xfrm_policy_kill(pol);
 964
 965                         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 966                         goto again1;
 967                 }
 968
 969                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 970         again2:
 971                         hlist_for_each_entry(pol,
 972                                              net->xfrm.policy_bydst[dir].table + i,
 973                                              bydst) {
 974                                 if (pol->type != type)
 975                                         continue;
 976                                 __xfrm_policy_unlink(pol, dir);
 977                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 978                                 cnt++;
 979
 980                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 981                                 xfrm_policy_kill(pol);
 982
 983                                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
 984                                 goto again2;
 985                         }
 986                 }
 987
 988         }
 989         if (!cnt)
 990                 err = -ESRCH;
 991 out:
 992         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 993         return err;
 994 }
 995 EXPORT_SYMBOL(xfrm_policy_flush);
 996
 997 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 998                      int (*func)(struct xfrm_policy *, int, int, void*),
 999                      void *data)
1000 {
1001         struct xfrm_policy *pol;
1002         struct xfrm_policy_walk_entry *x;
1003         int error = 0;
1004
1005         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1006             walk->type != XFRM_POLICY_TYPE_ANY)
1007                 return -EINVAL;
1008
1009         if (list_empty(&walk->walk.all) && walk->seq != 0)
1010                 return 0;
1011
1012         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1013         if (list_empty(&walk->walk.all))
1014                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1015         else
1016                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
1017         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1018                 if (x->dead)
1019                         continue;
1020                 pol = container_of(x, struct xfrm_policy, walk);
1021                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1022                     walk->type != pol->type)
1023                         continue;
1024                 error = func(pol, xfrm_policy_id2dir(pol->index),
1025                              walk->seq, data);
1026                 if (error) {
1027                         list_move_tail(&walk->walk.all, &x->all);
1028                         goto out;
1029                 }
1030                 walk->seq++;
1031         }
1032         if (walk->seq == 0) {
1033                 error = -ENOENT;
1034                 goto out;
1035         }
1036         list_del_init(&walk->walk.all);
1037 out:
1038         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1039         return error;
1040 }
1041 EXPORT_SYMBOL(xfrm_policy_walk);
1042
1043 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1044 {
1045         INIT_LIST_HEAD(&walk->walk.all);
1046         walk->walk.dead = 1;
1047         walk->type = type;
1048         walk->seq = 0;
1049 }
1050 EXPORT_SYMBOL(xfrm_policy_walk_init);
1051
1052 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1053 {
1054         if (list_empty(&walk->walk.all))
1055                 return;
1056
1057         write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1058         list_del(&walk->walk.all);
1059         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1060 }
1061 EXPORT_SYMBOL(xfrm_policy_walk_done);
1062
1063 /*
1064  * Find policy to apply to this flow.
1065  *
1066  * Returns 0 if policy found, else an -errno.
1067  */
1068 static int xfrm_policy_match(const struct xfrm_policy *pol,
1069                              const struct flowi *fl,
1070                              u8 type, u16 family, int dir)
1071 {
1072         const struct xfrm_selector *sel = &pol->selector;
1073         int ret = -ESRCH;
1074         bool match;
1075
1076         if (pol->family != family ||
1077             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1078             pol->type != type)
1079                 return ret;
1080
1081         match = xfrm_selector_match(sel, fl, family);
1082         if (match)
1083                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1084                                                   dir);
1085
1086         return ret;
1087 }
1088
1089 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1090                                                      const struct flowi *fl,
1091                                                      u16 family, u8 dir)
1092 {
1093         int err;
1094         struct xfrm_policy *pol, *ret;
1095         const xfrm_address_t *daddr, *saddr;
1096         struct hlist_head *chain;
1097         u32 priority = ~0U;
1098
1099         daddr = xfrm_flowi_daddr(fl, family);
1100         saddr = xfrm_flowi_saddr(fl, family);
1101         if (unlikely(!daddr || !saddr))
1102                 return NULL;
1103
1104         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1105         chain = policy_hash_direct(net, daddr, saddr, family, dir);
1106         ret = NULL;
1107         hlist_for_each_entry(pol, chain, bydst) {
1108                 err = xfrm_policy_match(pol, fl, type, family, dir);
1109                 if (err) {
1110                         if (err == -ESRCH)
1111                                 continue;
1112                         else {
1113                                 ret = ERR_PTR(err);
1114                                 goto fail;
1115                         }
1116                 } else {
1117                         ret = pol;
1118                         priority = ret->priority;
1119                         break;
1120                 }
1121         }
1122         chain = &net->xfrm.policy_inexact[dir];
1123         hlist_for_each_entry(pol, chain, bydst) {
1124                 err = xfrm_policy_match(pol, fl, type, family, dir);
1125                 if (err) {
1126                         if (err == -ESRCH)
1127                                 continue;
1128                         else {
1129                                 ret = ERR_PTR(err);
1130                                 goto fail;
1131                         }
1132                 } else if (pol->priority < priority) {
1133                         ret = pol;
1134                         break;
1135                 }
1136         }
1137         if (ret)
1138                 xfrm_pol_hold(ret);
1139 fail:
1140         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1141
1142         return ret;
1143 }
1144
1145 static struct xfrm_policy *
1146 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1147 {
1148 #ifdef CONFIG_XFRM_SUB_POLICY
1149         struct xfrm_policy *pol;
1150
1151         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1152         if (pol != NULL)
1153                 return pol;
1154 #endif
1155         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1156 }
1157
1158 static int flow_to_policy_dir(int dir)
1159 {
1160         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1161             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1162             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1163                 return dir;
1164
1165         switch (dir) {
1166         default:
1167         case FLOW_DIR_IN:
1168                 return XFRM_POLICY_IN;
1169         case FLOW_DIR_OUT:
1170                 return XFRM_POLICY_OUT;
1171         case FLOW_DIR_FWD:
1172                 return XFRM_POLICY_FWD;
1173         }
1174 }
1175
1176 static struct flow_cache_object *
1177 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1178                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1179 {
1180         struct xfrm_policy *pol;
1181
1182         if (old_obj)
1183                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1184
1185         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1186         if (IS_ERR_OR_NULL(pol))
1187                 return ERR_CAST(pol);
1188
1189         /* Resolver returns two references:
1190          * one for cache and one for caller of flow_cache_lookup() */
1191         xfrm_pol_hold(pol);
1192
1193         return &pol->flo;
1194 }
1195
1196 static inline int policy_to_flow_dir(int dir)
1197 {
1198         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1199             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1200             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1201                 return dir;
1202         switch (dir) {
1203         default:
1204         case XFRM_POLICY_IN:
1205                 return FLOW_DIR_IN;
1206         case XFRM_POLICY_OUT:
1207                 return FLOW_DIR_OUT;
1208         case XFRM_POLICY_FWD:
1209                 return FLOW_DIR_FWD;
1210         }
1211 }
1212
1213 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1214                                                  const struct flowi *fl)
1215 {
1216         struct xfrm_policy *pol;
1217         struct net *net = sock_net(sk);
1218
1219         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1220         if ((pol = sk->sk_policy[dir]) != NULL) {
1221                 bool match = xfrm_selector_match(&pol->selector, fl,
1222                                                  sk->sk_family);
1223                 int err = 0;
1224
1225                 if (match) {
1226                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1227                                 pol = NULL;
1228                                 goto out;
1229                         }
1230                         err = security_xfrm_policy_lookup(pol->security,
1231                                                       fl->flowi_secid,
1232                                                       policy_to_flow_dir(dir));
1233                         if (!err)
1234                                 xfrm_pol_hold(pol);
1235                         else if (err == -ESRCH)
1236                                 pol = NULL;
1237                         else
1238                                 pol = ERR_PTR(err);
1239                 } else
1240                         pol = NULL;
1241         }
1242 out:
1243         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1244         return pol;
1245 }
1246
1247 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1248 {
1249         struct net *net = xp_net(pol);
1250         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1251                                                      pol->family, dir);
1252
1253         list_add(&pol->walk.all, &net->xfrm.policy_all);
1254         hlist_add_head(&pol->bydst, chain);
1255         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1256         net->xfrm.policy_count[dir]++;
1257         xfrm_pol_hold(pol);
1258
1259         if (xfrm_bydst_should_resize(net, dir, NULL))
1260                 schedule_work(&net->xfrm.policy_hash_work);
1261 }
1262
1263 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1264                                                 int dir)
1265 {
1266         struct net *net = xp_net(pol);
1267
1268         if (hlist_unhashed(&pol->bydst))
1269                 return NULL;
1270
1271         hlist_del_init(&pol->bydst);
1272         hlist_del(&pol->byidx);
1273         list_del(&pol->walk.all);
1274         net->xfrm.policy_count[dir]--;
1275
1276         return pol;
1277 }
1278
1279 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1280 {
1281         struct net *net = xp_net(pol);
1282
1283         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1284         pol = __xfrm_policy_unlink(pol, dir);
1285         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1286         if (pol) {
1287                 xfrm_policy_kill(pol);
1288                 return 0;
1289         }
1290         return -ENOENT;
1291 }
1292 EXPORT_SYMBOL(xfrm_policy_delete);
1293
1294 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1295 {
1296         struct net *net = xp_net(pol);
1297         struct xfrm_policy *old_pol;
1298
1299 #ifdef CONFIG_XFRM_SUB_POLICY
1300         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1301                 return -EINVAL;
1302 #endif
1303
1304         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1305         old_pol = sk->sk_policy[dir];
1306         sk->sk_policy[dir] = pol;
1307         if (pol) {
1308                 pol->curlft.add_time = get_seconds();
1309                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1310                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1311         }
1312         if (old_pol) {
1313                 if (pol)
1314                         xfrm_policy_requeue(old_pol, pol);
1315
1316                 /* Unlinking succeeds always. This is the only function
1317                  * allowed to delete or replace socket policy.
1318                  */
1319                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1320         }
1321         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1322
1323         if (old_pol) {
1324                 xfrm_policy_kill(old_pol);
1325         }
1326         return 0;
1327 }
1328
1329 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1330 {
1331         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1332         struct net *net = xp_net(old);
1333
1334         if (newp) {
1335                 newp->selector = old->selector;
1336                 if (security_xfrm_policy_clone(old->security,
1337                                                &newp->security)) {
1338                         kfree(newp);
1339                         return NULL;  /* ENOMEM */
1340                 }
1341                 newp->lft = old->lft;
1342                 newp->curlft = old->curlft;
1343                 newp->mark = old->mark;
1344                 newp->action = old->action;
1345                 newp->flags = old->flags;
1346                 newp->xfrm_nr = old->xfrm_nr;
1347                 newp->index = old->index;
1348                 newp->type = old->type;
1349                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1350                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1351                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1352                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1353                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1354                 xfrm_pol_put(newp);
1355         }
1356         return newp;
1357 }
1358
1359 int __xfrm_sk_clone_policy(struct sock *sk)
1360 {
1361         struct xfrm_policy *p0 = sk->sk_policy[0],
1362                            *p1 = sk->sk_policy[1];
1363
1364         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1365         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1366                 return -ENOMEM;
1367         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1368                 return -ENOMEM;
1369         return 0;
1370 }
1371
1372 static int
1373 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1374                unsigned short family)
1375 {
1376         int err;
1377         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1378
1379         if (unlikely(afinfo == NULL))
1380                 return -EINVAL;
1381         err = afinfo->get_saddr(net, local, remote);
1382         xfrm_policy_put_afinfo(afinfo);
1383         return err;
1384 }
1385
1386 /* Resolve list of templates for the flow, given policy. */
1387
1388 static int
1389 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1390                       struct xfrm_state **xfrm, unsigned short family)
1391 {
1392         struct net *net = xp_net(policy);
1393         int nx;
1394         int i, error;
1395         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1396         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1397         xfrm_address_t tmp;
1398
1399         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1400                 struct xfrm_state *x;
1401                 xfrm_address_t *remote = daddr;
1402                 xfrm_address_t *local  = saddr;
1403                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1404
1405                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1406                     tmpl->mode == XFRM_MODE_BEET) {
1407                         remote = &tmpl->id.daddr;
1408                         local = &tmpl->saddr;
1409                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1410                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1411                                 if (error)
1412                                         goto fail;
1413                                 local = &tmp;
1414                         }
1415                 }
1416
1417                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1418
1419                 if (x && x->km.state == XFRM_STATE_VALID) {
1420                         xfrm[nx++] = x;
1421                         daddr = remote;
1422                         saddr = local;
1423                         continue;
1424                 }
1425                 if (x) {
1426                         error = (x->km.state == XFRM_STATE_ERROR ?
1427                                  -EINVAL : -EAGAIN);
1428                         xfrm_state_put(x);
1429                 } else if (error == -ESRCH) {
1430                         error = -EAGAIN;
1431                 }
1432
1433                 if (!tmpl->optional)
1434                         goto fail;
1435         }
1436         return nx;
1437
1438 fail:
1439         for (nx--; nx >= 0; nx--)
1440                 xfrm_state_put(xfrm[nx]);
1441         return error;
1442 }
1443
1444 static int
1445 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1446                   struct xfrm_state **xfrm, unsigned short family)
1447 {
1448         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1449         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1450         int cnx = 0;
1451         int error;
1452         int ret;
1453         int i;
1454
1455         for (i = 0; i < npols; i++) {
1456                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1457                         error = -ENOBUFS;
1458                         goto fail;
1459                 }
1460
1461                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1462                 if (ret < 0) {
1463                         error = ret;
1464                         goto fail;
1465                 } else
1466                         cnx += ret;
1467         }
1468
1469         /* found states are sorted for outbound processing */
1470         if (npols > 1)
1471                 xfrm_state_sort(xfrm, tpp, cnx, family);
1472
1473         return cnx;
1474
1475  fail:
1476         for (cnx--; cnx >= 0; cnx--)
1477                 xfrm_state_put(tpp[cnx]);
1478         return error;
1479
1480 }
1481
1482 /* Check that the bundle accepts the flow and its components are
1483  * still valid.
1484  */
1485
1486 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1487 {
1488         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1489         int tos;
1490
1491         if (!afinfo)
1492                 return -EINVAL;
1493
1494         tos = afinfo->get_tos(fl);
1495
1496         xfrm_policy_put_afinfo(afinfo);
1497
1498         return tos;
1499 }
1500
1501 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1502 {
1503         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1504         struct dst_entry *dst = &xdst->u.dst;
1505
1506         if (xdst->route == NULL) {
1507                 /* Dummy bundle - if it has xfrms we were not
1508                  * able to build bundle as template resolution failed.
1509                  * It means we need to try again resolving. */
1510                 if (xdst->num_xfrms > 0)
1511                         return NULL;
1512         } else if (dst->flags & DST_XFRM_QUEUE) {
1513                 return NULL;
1514         } else {
1515                 /* Real bundle */
1516                 if (stale_bundle(dst))
1517                         return NULL;
1518         }
1519
1520         dst_hold(dst);
1521         return flo;
1522 }
1523
1524 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1525 {
1526         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1527         struct dst_entry *dst = &xdst->u.dst;
1528
1529         if (!xdst->route)
1530                 return 0;
1531         if (stale_bundle(dst))
1532                 return 0;
1533
1534         return 1;
1535 }
1536
1537 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1538 {
1539         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1540         struct dst_entry *dst = &xdst->u.dst;
1541
1542         dst_free(dst);
1543 }
1544
1545 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1546         .get = xfrm_bundle_flo_get,
1547         .check = xfrm_bundle_flo_check,
1548         .delete = xfrm_bundle_flo_delete,
1549 };
1550
1551 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1552 {
1553         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1554         struct dst_ops *dst_ops;
1555         struct xfrm_dst *xdst;
1556
1557         if (!afinfo)
1558                 return ERR_PTR(-EINVAL);
1559
1560         switch (family) {
1561         case AF_INET:
1562                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1563                 break;
1564 #if IS_ENABLED(CONFIG_IPV6)
1565         case AF_INET6:
1566                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1567                 break;
1568 #endif
1569         default:
1570                 BUG();
1571         }
1572         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1573
1574         if (likely(xdst)) {
1575                 struct dst_entry *dst = &xdst->u.dst;
1576
1577                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1578                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1579                 if (afinfo->init_dst)
1580                         afinfo->init_dst(net, xdst);
1581         } else
1582                 xdst = ERR_PTR(-ENOBUFS);
1583
1584         xfrm_policy_put_afinfo(afinfo);
1585
1586         return xdst;
1587 }
1588
1589 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1590                                  int nfheader_len)
1591 {
1592         struct xfrm_policy_afinfo *afinfo =
1593                 xfrm_policy_get_afinfo(dst->ops->family);
1594         int err;
1595
1596         if (!afinfo)
1597                 return -EINVAL;
1598
1599         err = afinfo->init_path(path, dst, nfheader_len);
1600
1601         xfrm_policy_put_afinfo(afinfo);
1602
1603         return err;
1604 }
1605
1606 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1607                                 const struct flowi *fl)
1608 {
1609         struct xfrm_policy_afinfo *afinfo =
1610                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1611         int err;
1612
1613         if (!afinfo)
1614                 return -EINVAL;
1615
1616         err = afinfo->fill_dst(xdst, dev, fl);
1617
1618         xfrm_policy_put_afinfo(afinfo);
1619
1620         return err;
1621 }
1622
1623
1624 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1625  * all the metrics... Shortly, bundle a bundle.
1626  */
1627
1628 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1629                                             struct xfrm_state **xfrm, int nx,
1630                                             const struct flowi *fl,
1631                                             struct dst_entry *dst)
1632 {
1633         struct net *net = xp_net(policy);
1634         unsigned long now = jiffies;
1635         struct net_device *dev;
1636         struct xfrm_mode *inner_mode;
1637         struct dst_entry *dst_prev = NULL;
1638         struct dst_entry *dst0 = NULL;
1639         int i = 0;
1640         int err;
1641         int header_len = 0;
1642         int nfheader_len = 0;
1643         int trailer_len = 0;
1644         int tos;
1645         int family = policy->selector.family;
1646         xfrm_address_t saddr, daddr;
1647
1648         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1649
1650         tos = xfrm_get_tos(fl, family);
1651         err = tos;
1652         if (tos < 0)
1653                 goto put_states;
1654
1655         dst_hold(dst);
1656
1657         for (; i < nx; i++) {
1658                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1659                 struct dst_entry *dst1 = &xdst->u.dst;
1660
1661                 err = PTR_ERR(xdst);
1662                 if (IS_ERR(xdst)) {
1663                         dst_release(dst);
1664                         goto put_states;
1665                 }
1666
1667                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1668                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1669                                                         xfrm_af2proto(family));
1670                         if (!inner_mode) {
1671                                 err = -EAFNOSUPPORT;
1672                                 dst_release(dst);
1673                                 goto put_states;
1674                         }
1675                 } else
1676                         inner_mode = xfrm[i]->inner_mode;
1677
1678                 if (!dst_prev)
1679                         dst0 = dst1;
1680                 else {
1681                         dst_prev->child = dst_clone(dst1);
1682                         dst1->flags |= DST_NOHASH;
1683                 }
1684
1685                 xdst->route = dst;
1686                 dst_copy_metrics(dst1, dst);
1687
1688                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1689                         family = xfrm[i]->props.family;
1690                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1691                                               family);
1692                         err = PTR_ERR(dst);
1693                         if (IS_ERR(dst))
1694                                 goto put_states;
1695                 } else
1696                         dst_hold(dst);
1697
1698                 dst1->xfrm = xfrm[i];
1699                 xdst->xfrm_genid = xfrm[i]->genid;
1700
1701                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1702                 dst1->flags |= DST_HOST;
1703                 dst1->lastuse = now;
1704
1705                 dst1->input = dst_discard;
1706                 dst1->output = inner_mode->afinfo->output;
1707
1708                 dst1->next = dst_prev;
1709                 dst_prev = dst1;
1710
1711                 header_len += xfrm[i]->props.header_len;
1712                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1713                         nfheader_len += xfrm[i]->props.header_len;
1714                 trailer_len += xfrm[i]->props.trailer_len;
1715         }
1716
1717         dst_prev->child = dst;
1718         dst0->path = dst;
1719
1720         err = -ENODEV;
1721         dev = dst->dev;
1722         if (!dev)
1723                 goto free_dst;
1724
1725         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1726         xfrm_init_pmtu(dst_prev);
1727
1728         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1729                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1730
1731                 err = xfrm_fill_dst(xdst, dev, fl);
1732                 if (err)
1733                         goto free_dst;
1734
1735                 dst_prev->header_len = header_len;
1736                 dst_prev->trailer_len = trailer_len;
1737                 header_len -= xdst->u.dst.xfrm->props.header_len;
1738                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1739         }
1740
1741 out:
1742         return dst0;
1743
1744 put_states:
1745         for (; i < nx; i++)
1746                 xfrm_state_put(xfrm[i]);
1747 free_dst:
1748         if (dst0)
1749                 dst_free(dst0);
1750         dst0 = ERR_PTR(err);
1751         goto out;
1752 }
1753
1754 #ifdef CONFIG_XFRM_SUB_POLICY
1755 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1756 {
1757         if (!*target) {
1758                 *target = kmalloc(size, GFP_ATOMIC);
1759                 if (!*target)
1760                         return -ENOMEM;
1761         }
1762
1763         memcpy(*target, src, size);
1764         return 0;
1765 }
1766 #endif
1767
1768 static int xfrm_dst_update_parent(struct dst_entry *dst,
1769                                   const struct xfrm_selector *sel)
1770 {
1771 #ifdef CONFIG_XFRM_SUB_POLICY
1772         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1773         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1774                                    sel, sizeof(*sel));
1775 #else
1776         return 0;
1777 #endif
1778 }
1779
1780 static int xfrm_dst_update_origin(struct dst_entry *dst,
1781                                   const struct flowi *fl)
1782 {
1783 #ifdef CONFIG_XFRM_SUB_POLICY
1784         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1785         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1786 #else
1787         return 0;
1788 #endif
1789 }
1790
1791 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1792                                 struct xfrm_policy **pols,
1793                                 int *num_pols, int *num_xfrms)
1794 {
1795         int i;
1796
1797         if (*num_pols == 0 || !pols[0]) {
1798                 *num_pols = 0;
1799                 *num_xfrms = 0;
1800                 return 0;
1801         }
1802         if (IS_ERR(pols[0]))
1803                 return PTR_ERR(pols[0]);
1804
1805         *num_xfrms = pols[0]->xfrm_nr;
1806
1807 #ifdef CONFIG_XFRM_SUB_POLICY
1808         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1809             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1810                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1811                                                     XFRM_POLICY_TYPE_MAIN,
1812                                                     fl, family,
1813                                                     XFRM_POLICY_OUT);
1814                 if (pols[1]) {
1815                         if (IS_ERR(pols[1])) {
1816                                 xfrm_pols_put(pols, *num_pols);
1817                                 return PTR_ERR(pols[1]);
1818                         }
1819                         (*num_pols)++;
1820                         (*num_xfrms) += pols[1]->xfrm_nr;
1821                 }
1822         }
1823 #endif
1824         for (i = 0; i < *num_pols; i++) {
1825                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1826                         *num_xfrms = -1;
1827                         break;
1828                 }
1829         }
1830
1831         return 0;
1832
1833 }
1834
1835 static struct xfrm_dst *
1836 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1837                                const struct flowi *fl, u16 family,
1838                                struct dst_entry *dst_orig)
1839 {
1840         struct net *net = xp_net(pols[0]);
1841         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1842         struct dst_entry *dst;
1843         struct xfrm_dst *xdst;
1844         int err;
1845
1846         /* Try to instantiate a bundle */
1847         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1848         if (err <= 0) {
1849                 if (err != 0 && err != -EAGAIN)
1850                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1851                 return ERR_PTR(err);
1852         }
1853
1854         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1855         if (IS_ERR(dst)) {
1856                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1857                 return ERR_CAST(dst);
1858         }
1859
1860         xdst = (struct xfrm_dst *)dst;
1861         xdst->num_xfrms = err;
1862         if (num_pols > 1)
1863                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1864         else
1865                 err = xfrm_dst_update_origin(dst, fl);
1866         if (unlikely(err)) {
1867                 dst_free(dst);
1868                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1869                 return ERR_PTR(err);
1870         }
1871
1872         xdst->num_pols = num_pols;
1873         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1874         xdst->policy_genid = atomic_read(&pols[0]->genid);
1875
1876         return xdst;
1877 }
1878
1879 static void xfrm_policy_queue_process(unsigned long arg)
1880 {
1881         int err = 0;
1882         struct sk_buff *skb;
1883         struct sock *sk;
1884         struct dst_entry *dst;
1885         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1886         struct xfrm_policy_queue *pq = &pol->polq;
1887         struct flowi fl;
1888         struct sk_buff_head list;
1889
1890         spin_lock(&pq->hold_queue.lock);
1891         skb = skb_peek(&pq->hold_queue);
1892         if (!skb) {
1893                 spin_unlock(&pq->hold_queue.lock);
1894                 goto out;
1895         }
1896         dst = skb_dst(skb);
1897         sk = skb->sk;
1898         xfrm_decode_session(skb, &fl, dst->ops->family);
1899         spin_unlock(&pq->hold_queue.lock);
1900
1901         dst_hold(dst->path);
1902         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1903                           sk, 0);
1904         if (IS_ERR(dst))
1905                 goto purge_queue;
1906
1907         if (dst->flags & DST_XFRM_QUEUE) {
1908                 dst_release(dst);
1909
1910                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1911                         goto purge_queue;
1912
1913                 pq->timeout = pq->timeout << 1;
1914                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1915                         xfrm_pol_hold(pol);
1916         goto out;
1917         }
1918
1919         dst_release(dst);
1920
1921         __skb_queue_head_init(&list);
1922
1923         spin_lock(&pq->hold_queue.lock);
1924         pq->timeout = 0;
1925         skb_queue_splice_init(&pq->hold_queue, &list);
1926         spin_unlock(&pq->hold_queue.lock);
1927
1928         while (!skb_queue_empty(&list)) {
1929                 skb = __skb_dequeue(&list);
1930
1931                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1932                 dst_hold(skb_dst(skb)->path);
1933                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1934                                   &fl, skb->sk, 0);
1935                 if (IS_ERR(dst)) {
1936                         kfree_skb(skb);
1937                         continue;
1938                 }
1939
1940                 nf_reset(skb);
1941                 skb_dst_drop(skb);
1942                 skb_dst_set(skb, dst);
1943
1944                 err = dst_output(skb);
1945         }
1946
1947 out:
1948         xfrm_pol_put(pol);
1949         return;
1950
1951 purge_queue:
1952         pq->timeout = 0;
1953         xfrm_queue_purge(&pq->hold_queue);
1954         xfrm_pol_put(pol);
1955 }
1956
1957 static int xdst_queue_output(struct sock *sk, struct sk_buff *skb)
1958 {
1959         unsigned long sched_next;
1960         struct dst_entry *dst = skb_dst(skb);
1961         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1962         struct xfrm_policy *pol = xdst->pols[0];
1963         struct xfrm_policy_queue *pq = &pol->polq;
1964
1965         if (unlikely(skb_fclone_busy(skb))) {
1966                 kfree_skb(skb);
1967                 return 0;
1968         }
1969
1970         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1971                 kfree_skb(skb);
1972                 return -EAGAIN;
1973         }
1974
1975         skb_dst_force(skb);
1976
1977         spin_lock_bh(&pq->hold_queue.lock);
1978
1979         if (!pq->timeout)
1980                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1981
1982         sched_next = jiffies + pq->timeout;
1983
1984         if (del_timer(&pq->hold_timer)) {
1985                 if (time_before(pq->hold_timer.expires, sched_next))
1986                         sched_next = pq->hold_timer.expires;
1987                 xfrm_pol_put(pol);
1988         }
1989
1990         __skb_queue_tail(&pq->hold_queue, skb);
1991         if (!mod_timer(&pq->hold_timer, sched_next))
1992                 xfrm_pol_hold(pol);
1993
1994         spin_unlock_bh(&pq->hold_queue.lock);
1995
1996         return 0;
1997 }
1998
1999 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2000                                                  struct xfrm_flo *xflo,
2001                                                  const struct flowi *fl,
2002                                                  int num_xfrms,
2003                                                  u16 family)
2004 {
2005         int err;
2006         struct net_device *dev;
2007         struct dst_entry *dst;
2008         struct dst_entry *dst1;
2009         struct xfrm_dst *xdst;
2010
2011         xdst = xfrm_alloc_dst(net, family);
2012         if (IS_ERR(xdst))
2013                 return xdst;
2014
2015         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2016             net->xfrm.sysctl_larval_drop ||
2017             num_xfrms <= 0)
2018                 return xdst;
2019
2020         dst = xflo->dst_orig;
2021         dst1 = &xdst->u.dst;
2022         dst_hold(dst);
2023         xdst->route = dst;
2024
2025         dst_copy_metrics(dst1, dst);
2026
2027         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2028         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2029         dst1->lastuse = jiffies;
2030
2031         dst1->input = dst_discard;
2032         dst1->output = xdst_queue_output;
2033
2034         dst_hold(dst);
2035         dst1->child = dst;
2036         dst1->path = dst;
2037
2038         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2039
2040         err = -ENODEV;
2041         dev = dst->dev;
2042         if (!dev)
2043                 goto free_dst;
2044
2045         err = xfrm_fill_dst(xdst, dev, fl);
2046         if (err)
2047                 goto free_dst;
2048
2049 out:
2050         return xdst;
2051
2052 free_dst:
2053         dst_release(dst1);
2054         xdst = ERR_PTR(err);
2055         goto out;
2056 }
2057
2058 static struct flow_cache_object *
2059 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2060                    struct flow_cache_object *oldflo, void *ctx)
2061 {
2062         struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2063         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2064         struct xfrm_dst *xdst, *new_xdst;
2065         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2066
2067         /* Check if the policies from old bundle are usable */
2068         xdst = NULL;
2069         if (oldflo) {
2070                 xdst = container_of(oldflo, struct xfrm_dst, flo);
2071                 num_pols = xdst->num_pols;
2072                 num_xfrms = xdst->num_xfrms;
2073                 pol_dead = 0;
2074                 for (i = 0; i < num_pols; i++) {
2075                         pols[i] = xdst->pols[i];
2076                         pol_dead |= pols[i]->walk.dead;
2077                 }
2078                 if (pol_dead) {
2079                         dst_free(&xdst->u.dst);
2080                         xdst = NULL;
2081                         num_pols = 0;
2082                         num_xfrms = 0;
2083                         oldflo = NULL;
2084                 }
2085         }
2086
2087         /* Resolve policies to use if we couldn't get them from
2088          * previous cache entry */
2089         if (xdst == NULL) {
2090                 num_pols = 1;
2091                 pols[0] = __xfrm_policy_lookup(net, fl, family,
2092                                                flow_to_policy_dir(dir));
2093                 err = xfrm_expand_policies(fl, family, pols,
2094                                            &num_pols, &num_xfrms);
2095                 if (err < 0)
2096                         goto inc_error;
2097                 if (num_pols == 0)
2098                         return NULL;
2099                 if (num_xfrms <= 0)
2100                         goto make_dummy_bundle;
2101         }
2102
2103         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2104                                                   xflo->dst_orig);
2105         if (IS_ERR(new_xdst)) {
2106                 err = PTR_ERR(new_xdst);
2107                 if (err != -EAGAIN)
2108                         goto error;
2109                 if (oldflo == NULL)
2110                         goto make_dummy_bundle;
2111                 dst_hold(&xdst->u.dst);
2112                 return oldflo;
2113         } else if (new_xdst == NULL) {
2114                 num_xfrms = 0;
2115                 if (oldflo == NULL)
2116                         goto make_dummy_bundle;
2117                 xdst->num_xfrms = 0;
2118                 dst_hold(&xdst->u.dst);
2119                 return oldflo;
2120         }
2121
2122         /* Kill the previous bundle */
2123         if (xdst) {
2124                 /* The policies were stolen for newly generated bundle */
2125                 xdst->num_pols = 0;
2126                 dst_free(&xdst->u.dst);
2127         }
2128
2129         /* Flow cache does not have reference, it dst_free()'s,
2130          * but we do need to return one reference for original caller */
2131         dst_hold(&new_xdst->u.dst);
2132         return &new_xdst->flo;
2133
2134 make_dummy_bundle:
2135         /* We found policies, but there's no bundles to instantiate:
2136          * either because the policy blocks, has no transformations or
2137          * we could not build template (no xfrm_states).*/
2138         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2139         if (IS_ERR(xdst)) {
2140                 xfrm_pols_put(pols, num_pols);
2141                 return ERR_CAST(xdst);
2142         }
2143         xdst->num_pols = num_pols;
2144         xdst->num_xfrms = num_xfrms;
2145         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2146
2147         dst_hold(&xdst->u.dst);
2148         return &xdst->flo;
2149
2150 inc_error:
2151         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2152 error:
2153         if (xdst != NULL)
2154                 dst_free(&xdst->u.dst);
2155         else
2156                 xfrm_pols_put(pols, num_pols);
2157         return ERR_PTR(err);
2158 }
2159
2160 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2161                                         struct dst_entry *dst_orig)
2162 {
2163         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2164         struct dst_entry *ret;
2165
2166         if (!afinfo) {
2167                 dst_release(dst_orig);
2168                 return ERR_PTR(-EINVAL);
2169         } else {
2170                 ret = afinfo->blackhole_route(net, dst_orig);
2171         }
2172         xfrm_policy_put_afinfo(afinfo);
2173
2174         return ret;
2175 }
2176
2177 /* Main function: finds/creates a bundle for given flow.
2178  *
2179  * At the moment we eat a raw IP route. Mostly to speed up lookups
2180  * on interfaces with disabled IPsec.
2181  */
2182 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2183                               const struct flowi *fl,
2184                               struct sock *sk, int flags)
2185 {
2186         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2187         struct flow_cache_object *flo;
2188         struct xfrm_dst *xdst;
2189         struct dst_entry *dst, *route;
2190         u16 family = dst_orig->ops->family;
2191         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2192         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2193
2194         dst = NULL;
2195         xdst = NULL;
2196         route = NULL;
2197
2198         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2199                 num_pols = 1;
2200                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2201                 err = xfrm_expand_policies(fl, family, pols,
2202                                            &num_pols, &num_xfrms);
2203                 if (err < 0)
2204                         goto dropdst;
2205
2206                 if (num_pols) {
2207                         if (num_xfrms <= 0) {
2208                                 drop_pols = num_pols;
2209                                 goto no_transform;
2210                         }
2211
2212                         xdst = xfrm_resolve_and_create_bundle(
2213                                         pols, num_pols, fl,
2214                                         family, dst_orig);
2215                         if (IS_ERR(xdst)) {
2216                                 xfrm_pols_put(pols, num_pols);
2217                                 err = PTR_ERR(xdst);
2218                                 goto dropdst;
2219                         } else if (xdst == NULL) {
2220                                 num_xfrms = 0;
2221                                 drop_pols = num_pols;
2222                                 goto no_transform;
2223                         }
2224
2225                         dst_hold(&xdst->u.dst);
2226                         xdst->u.dst.flags |= DST_NOCACHE;
2227                         route = xdst->route;
2228                 }
2229         }
2230
2231         if (xdst == NULL) {
2232                 struct xfrm_flo xflo;
2233
2234                 xflo.dst_orig = dst_orig;
2235                 xflo.flags = flags;
2236
2237                 /* To accelerate a bit...  */
2238                 if ((dst_orig->flags & DST_NOXFRM) ||
2239                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2240                         goto nopol;
2241
2242                 flo = flow_cache_lookup(net, fl, family, dir,
2243                                         xfrm_bundle_lookup, &xflo);
2244                 if (flo == NULL)
2245                         goto nopol;
2246                 if (IS_ERR(flo)) {
2247                         err = PTR_ERR(flo);
2248                         goto dropdst;
2249                 }
2250                 xdst = container_of(flo, struct xfrm_dst, flo);
2251
2252                 num_pols = xdst->num_pols;
2253                 num_xfrms = xdst->num_xfrms;
2254                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2255                 route = xdst->route;
2256         }
2257
2258         dst = &xdst->u.dst;
2259         if (route == NULL && num_xfrms > 0) {
2260                 /* The only case when xfrm_bundle_lookup() returns a
2261                  * bundle with null route, is when the template could
2262                  * not be resolved. It means policies are there, but
2263                  * bundle could not be created, since we don't yet
2264                  * have the xfrm_state's. We need to wait for KM to
2265                  * negotiate new SA's or bail out with error.*/
2266                 if (net->xfrm.sysctl_larval_drop) {
2267                         dst_release(dst);
2268                         xfrm_pols_put(pols, drop_pols);
2269                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2270
2271                         return ERR_PTR(-EREMOTE);
2272                 }
2273
2274                 err = -EAGAIN;
2275
2276                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2277                 goto error;
2278         }
2279
2280 no_transform:
2281         if (num_pols == 0)
2282                 goto nopol;
2283
2284         if ((flags & XFRM_LOOKUP_ICMP) &&
2285             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2286                 err = -ENOENT;
2287                 goto error;
2288         }
2289
2290         for (i = 0; i < num_pols; i++)
2291                 pols[i]->curlft.use_time = get_seconds();
2292
2293         if (num_xfrms < 0) {
2294                 /* Prohibit the flow */
2295                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2296                 err = -EPERM;
2297                 goto error;
2298         } else if (num_xfrms > 0) {
2299                 /* Flow transformed */
2300                 dst_release(dst_orig);
2301         } else {
2302                 /* Flow passes untransformed */
2303                 dst_release(dst);
2304                 dst = dst_orig;
2305         }
2306 ok:
2307         xfrm_pols_put(pols, drop_pols);
2308         if (dst && dst->xfrm &&
2309             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2310                 dst->flags |= DST_XFRM_TUNNEL;
2311         return dst;
2312
2313 nopol:
2314         if (!(flags & XFRM_LOOKUP_ICMP)) {
2315                 dst = dst_orig;
2316                 goto ok;
2317         }
2318         err = -ENOENT;
2319 error:
2320         dst_release(dst);
2321 dropdst:
2322         dst_release(dst_orig);
2323         xfrm_pols_put(pols, drop_pols);
2324         return ERR_PTR(err);
2325 }
2326 EXPORT_SYMBOL(xfrm_lookup);
2327
2328 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2329  * Otherwise we may send out blackholed packets.
2330  */
2331 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2332                                     const struct flowi *fl,
2333                                     struct sock *sk, int flags)
2334 {
2335         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2336                                             flags | XFRM_LOOKUP_QUEUE);
2337
2338         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2339                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2340
2341         return dst;
2342 }
2343 EXPORT_SYMBOL(xfrm_lookup_route);
2344
2345 static inline int
2346 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2347 {
2348         struct xfrm_state *x;
2349
2350         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2351                 return 0;
2352         x = skb->sp->xvec[idx];
2353         if (!x->type->reject)
2354                 return 0;
2355         return x->type->reject(x, skb, fl);
2356 }
2357
2358 /* When skb is transformed back to its "native" form, we have to
2359  * check policy restrictions. At the moment we make this in maximally
2360  * stupid way. Shame on me. :-) Of course, connected sockets must
2361  * have policy cached at them.
2362  */
2363
2364 static inline int
2365 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2366               unsigned short family)
2367 {
2368         if (xfrm_state_kern(x))
2369                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2370         return  x->id.proto == tmpl->id.proto &&
2371                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2372                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2373                 x->props.mode == tmpl->mode &&
2374                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2375                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2376                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2377                   xfrm_state_addr_cmp(tmpl, x, family));
2378 }
2379
2380 /*
2381  * 0 or more than 0 is returned when validation is succeeded (either bypass
2382  * because of optional transport mode, or next index of the mathced secpath
2383  * state with the template.
2384  * -1 is returned when no matching template is found.
2385  * Otherwise "-2 - errored_index" is returned.
2386  */
2387 static inline int
2388 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2389                unsigned short family)
2390 {
2391         int idx = start;
2392
2393         if (tmpl->optional) {
2394                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2395                         return start;
2396         } else
2397                 start = -1;
2398         for (; idx < sp->len; idx++) {
2399                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2400                         return ++idx;
2401                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2402                         if (start == -1)
2403                                 start = -2-idx;
2404                         break;
2405                 }
2406         }
2407         return start;
2408 }
2409
2410 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2411                           unsigned int family, int reverse)
2412 {
2413         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2414         int err;
2415
2416         if (unlikely(afinfo == NULL))
2417                 return -EAFNOSUPPORT;
2418
2419         afinfo->decode_session(skb, fl, reverse);
2420         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2421         xfrm_policy_put_afinfo(afinfo);
2422         return err;
2423 }
2424 EXPORT_SYMBOL(__xfrm_decode_session);
2425
2426 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2427 {
2428         for (; k < sp->len; k++) {
2429                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2430                         *idxp = k;
2431                         return 1;
2432                 }
2433         }
2434
2435         return 0;
2436 }
2437
2438 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2439                         unsigned short family)
2440 {
2441         struct net *net = dev_net(skb->dev);
2442         struct xfrm_policy *pol;
2443         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2444         int npols = 0;
2445         int xfrm_nr;
2446         int pi;
2447         int reverse;
2448         struct flowi fl;
2449         u8 fl_dir;
2450         int xerr_idx = -1;
2451
2452         reverse = dir & ~XFRM_POLICY_MASK;
2453         dir &= XFRM_POLICY_MASK;
2454         fl_dir = policy_to_flow_dir(dir);
2455
2456         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2457                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2458                 return 0;
2459         }
2460
2461         nf_nat_decode_session(skb, &fl, family);
2462
2463         /* First, check used SA against their selectors. */
2464         if (skb->sp) {
2465                 int i;
2466
2467                 for (i = skb->sp->len-1; i >= 0; i--) {
2468                         struct xfrm_state *x = skb->sp->xvec[i];
2469                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2470                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2471                                 return 0;
2472                         }
2473                 }
2474         }
2475
2476         pol = NULL;
2477         if (sk && sk->sk_policy[dir]) {
2478                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2479                 if (IS_ERR(pol)) {
2480                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2481                         return 0;
2482                 }
2483         }
2484
2485         if (!pol) {
2486                 struct flow_cache_object *flo;
2487
2488                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2489                                         xfrm_policy_lookup, NULL);
2490                 if (IS_ERR_OR_NULL(flo))
2491                         pol = ERR_CAST(flo);
2492                 else
2493                         pol = container_of(flo, struct xfrm_policy, flo);
2494         }
2495
2496         if (IS_ERR(pol)) {
2497                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2498                 return 0;
2499         }
2500
2501         if (!pol) {
2502                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2503                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2504                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2505                         return 0;
2506                 }
2507                 return 1;
2508         }
2509
2510         pol->curlft.use_time = get_seconds();
2511
2512         pols[0] = pol;
2513         npols++;
2514 #ifdef CONFIG_XFRM_SUB_POLICY
2515         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2516                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2517                                                     &fl, family,
2518                                                     XFRM_POLICY_IN);
2519                 if (pols[1]) {
2520                         if (IS_ERR(pols[1])) {
2521                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2522                                 return 0;
2523                         }
2524                         pols[1]->curlft.use_time = get_seconds();
2525                         npols++;
2526                 }
2527         }
2528 #endif
2529
2530         if (pol->action == XFRM_POLICY_ALLOW) {
2531                 struct sec_path *sp;
2532                 static struct sec_path dummy;
2533                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2534                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2535                 struct xfrm_tmpl **tpp = tp;
2536                 int ti = 0;
2537                 int i, k;
2538
2539                 if ((sp = skb->sp) == NULL)
2540                         sp = &dummy;
2541
2542                 for (pi = 0; pi < npols; pi++) {
2543                         if (pols[pi] != pol &&
2544                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2545                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2546                                 goto reject;
2547                         }
2548                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2549                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2550                                 goto reject_error;
2551                         }
2552                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2553                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2554                 }
2555                 xfrm_nr = ti;
2556                 if (npols > 1) {
2557                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2558                         tpp = stp;
2559                 }
2560
2561                 /* For each tunnel xfrm, find the first matching tmpl.
2562                  * For each tmpl before that, find corresponding xfrm.
2563                  * Order is _important_. Later we will implement
2564                  * some barriers, but at the moment barriers
2565                  * are implied between each two transformations.
2566                  */
2567                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2568                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2569                         if (k < 0) {
2570                                 if (k < -1)
2571                                         /* "-2 - errored_index" returned */
2572                                         xerr_idx = -(2+k);
2573                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2574                                 goto reject;
2575                         }
2576                 }
2577
2578                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2579                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2580                         goto reject;
2581                 }
2582
2583                 xfrm_pols_put(pols, npols);
2584                 return 1;
2585         }
2586         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2587
2588 reject:
2589         xfrm_secpath_reject(xerr_idx, skb, &fl);
2590 reject_error:
2591         xfrm_pols_put(pols, npols);
2592         return 0;
2593 }
2594 EXPORT_SYMBOL(__xfrm_policy_check);
2595
2596 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2597 {
2598         struct net *net = dev_net(skb->dev);
2599         struct flowi fl;
2600         struct dst_entry *dst;
2601         int res = 1;
2602
2603         if (xfrm_decode_session(skb, &fl, family) < 0) {
2604                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2605                 return 0;
2606         }
2607
2608         skb_dst_force(skb);
2609
2610         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2611         if (IS_ERR(dst)) {
2612                 res = 0;
2613                 dst = NULL;
2614         }
2615         skb_dst_set(skb, dst);
2616         return res;
2617 }
2618 EXPORT_SYMBOL(__xfrm_route_forward);
2619
2620 /* Optimize later using cookies and generation ids. */
2621
2622 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2623 {
2624         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2625          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2626          * get validated by dst_ops->check on every use.  We do this
2627          * because when a normal route referenced by an XFRM dst is
2628          * obsoleted we do not go looking around for all parent
2629          * referencing XFRM dsts so that we can invalidate them.  It
2630          * is just too much work.  Instead we make the checks here on
2631          * every use.  For example:
2632          *
2633          *      XFRM dst A --> IPv4 dst X
2634          *
2635          * X is the "xdst->route" of A (X is also the "dst->path" of A
2636          * in this example).  If X is marked obsolete, "A" will not
2637          * notice.  That's what we are validating here via the
2638          * stale_bundle() check.
2639          *
2640          * When a policy's bundle is pruned, we dst_free() the XFRM
2641          * dst which causes it's ->obsolete field to be set to
2642          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2643          * this, we want to force a new route lookup.
2644          */
2645         if (dst->obsolete < 0 && !stale_bundle(dst))
2646                 return dst;
2647
2648         return NULL;
2649 }
2650
2651 static int stale_bundle(struct dst_entry *dst)
2652 {
2653         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2654 }
2655
2656 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2657 {
2658         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2659                 dst->dev = dev_net(dev)->loopback_dev;
2660                 dev_hold(dst->dev);
2661                 dev_put(dev);
2662         }
2663 }
2664 EXPORT_SYMBOL(xfrm_dst_ifdown);
2665
2666 static void xfrm_link_failure(struct sk_buff *skb)
2667 {
2668         /* Impossible. Such dst must be popped before reaches point of failure. */
2669 }
2670
2671 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2672 {
2673         if (dst) {
2674                 if (dst->obsolete) {
2675                         dst_release(dst);
2676                         dst = NULL;
2677                 }
2678         }
2679         return dst;
2680 }
2681
2682 void xfrm_garbage_collect(struct net *net)
2683 {
2684         flow_cache_flush(net);
2685 }
2686 EXPORT_SYMBOL(xfrm_garbage_collect);
2687
2688 static void xfrm_garbage_collect_deferred(struct net *net)
2689 {
2690         flow_cache_flush_deferred(net);
2691 }
2692
2693 static void xfrm_init_pmtu(struct dst_entry *dst)
2694 {
2695         do {
2696                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2697                 u32 pmtu, route_mtu_cached;
2698
2699                 pmtu = dst_mtu(dst->child);
2700                 xdst->child_mtu_cached = pmtu;
2701
2702                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2703
2704                 route_mtu_cached = dst_mtu(xdst->route);
2705                 xdst->route_mtu_cached = route_mtu_cached;
2706
2707                 if (pmtu > route_mtu_cached)
2708                         pmtu = route_mtu_cached;
2709
2710                 dst_metric_set(dst, RTAX_MTU, pmtu);
2711         } while ((dst = dst->next));
2712 }
2713
2714 /* Check that the bundle accepts the flow and its components are
2715  * still valid.
2716  */
2717
2718 static int xfrm_bundle_ok(struct xfrm_dst *first)
2719 {
2720         struct dst_entry *dst = &first->u.dst;
2721         struct xfrm_dst *last;
2722         u32 mtu;
2723
2724         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2725             (dst->dev && !netif_running(dst->dev)))
2726                 return 0;
2727
2728         if (dst->flags & DST_XFRM_QUEUE)
2729                 return 1;
2730
2731         last = NULL;
2732
2733         do {
2734                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2735
2736                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2737                         return 0;
2738                 if (xdst->xfrm_genid != dst->xfrm->genid)
2739                         return 0;
2740                 if (xdst->num_pols > 0 &&
2741                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2742                         return 0;
2743
2744                 mtu = dst_mtu(dst->child);
2745                 if (xdst->child_mtu_cached != mtu) {
2746                         last = xdst;
2747                         xdst->child_mtu_cached = mtu;
2748                 }
2749
2750                 if (!dst_check(xdst->route, xdst->route_cookie))
2751                         return 0;
2752                 mtu = dst_mtu(xdst->route);
2753                 if (xdst->route_mtu_cached != mtu) {
2754                         last = xdst;
2755                         xdst->route_mtu_cached = mtu;
2756                 }
2757
2758                 dst = dst->child;
2759         } while (dst->xfrm);
2760
2761         if (likely(!last))
2762                 return 1;
2763
2764         mtu = last->child_mtu_cached;
2765         for (;;) {
2766                 dst = &last->u.dst;
2767
2768                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2769                 if (mtu > last->route_mtu_cached)
2770                         mtu = last->route_mtu_cached;
2771                 dst_metric_set(dst, RTAX_MTU, mtu);
2772
2773                 if (last == first)
2774                         break;
2775
2776                 last = (struct xfrm_dst *)last->u.dst.next;
2777                 last->child_mtu_cached = mtu;
2778         }
2779
2780         return 1;
2781 }
2782
2783 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2784 {
2785         return dst_metric_advmss(dst->path);
2786 }
2787
2788 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2789 {
2790         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2791
2792         return mtu ? : dst_mtu(dst->path);
2793 }
2794
2795 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2796                                            struct sk_buff *skb,
2797                                            const void *daddr)
2798 {
2799         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2800 }
2801
2802 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2803 {
2804         struct net *net;
2805         int err = 0;
2806         if (unlikely(afinfo == NULL))
2807                 return -EINVAL;
2808         if (unlikely(afinfo->family >= NPROTO))
2809                 return -EAFNOSUPPORT;
2810         spin_lock(&xfrm_policy_afinfo_lock);
2811         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2812                 err = -ENOBUFS;
2813         else {
2814                 struct dst_ops *dst_ops = afinfo->dst_ops;
2815                 if (likely(dst_ops->kmem_cachep == NULL))
2816                         dst_ops->kmem_cachep = xfrm_dst_cache;
2817                 if (likely(dst_ops->check == NULL))
2818                         dst_ops->check = xfrm_dst_check;
2819                 if (likely(dst_ops->default_advmss == NULL))
2820                         dst_ops->default_advmss = xfrm_default_advmss;
2821                 if (likely(dst_ops->mtu == NULL))
2822                         dst_ops->mtu = xfrm_mtu;
2823                 if (likely(dst_ops->negative_advice == NULL))
2824                         dst_ops->negative_advice = xfrm_negative_advice;
2825                 if (likely(dst_ops->link_failure == NULL))
2826                         dst_ops->link_failure = xfrm_link_failure;
2827                 if (likely(dst_ops->neigh_lookup == NULL))
2828                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2829                 if (likely(afinfo->garbage_collect == NULL))
2830                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2831                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2832         }
2833         spin_unlock(&xfrm_policy_afinfo_lock);
2834
2835         rtnl_lock();
2836         for_each_net(net) {
2837                 struct dst_ops *xfrm_dst_ops;
2838
2839                 switch (afinfo->family) {
2840                 case AF_INET:
2841                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2842                         break;
2843 #if IS_ENABLED(CONFIG_IPV6)
2844                 case AF_INET6:
2845                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2846                         break;
2847 #endif
2848                 default:
2849                         BUG();
2850                 }
2851                 *xfrm_dst_ops = *afinfo->dst_ops;
2852         }
2853         rtnl_unlock();
2854
2855         return err;
2856 }
2857 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2858
2859 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2860 {
2861         int err = 0;
2862         if (unlikely(afinfo == NULL))
2863                 return -EINVAL;
2864         if (unlikely(afinfo->family >= NPROTO))
2865                 return -EAFNOSUPPORT;
2866         spin_lock(&xfrm_policy_afinfo_lock);
2867         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2868                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2869                         err = -EINVAL;
2870                 else
2871                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2872                                          NULL);
2873         }
2874         spin_unlock(&xfrm_policy_afinfo_lock);
2875         if (!err) {
2876                 struct dst_ops *dst_ops = afinfo->dst_ops;
2877
2878                 synchronize_rcu();
2879
2880                 dst_ops->kmem_cachep = NULL;
2881                 dst_ops->check = NULL;
2882                 dst_ops->negative_advice = NULL;
2883                 dst_ops->link_failure = NULL;
2884                 afinfo->garbage_collect = NULL;
2885         }
2886         return err;
2887 }
2888 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2889
2890 static void __net_init xfrm_dst_ops_init(struct net *net)
2891 {
2892         struct xfrm_policy_afinfo *afinfo;
2893
2894         rcu_read_lock();
2895         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2896         if (afinfo)
2897                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2898 #if IS_ENABLED(CONFIG_IPV6)
2899         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2900         if (afinfo)
2901                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2902 #endif
2903         rcu_read_unlock();
2904 }
2905
2906 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2907 {
2908         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2909
2910         switch (event) {
2911         case NETDEV_DOWN:
2912                 xfrm_garbage_collect(dev_net(dev));
2913         }
2914         return NOTIFY_DONE;
2915 }
2916
2917 static struct notifier_block xfrm_dev_notifier = {
2918         .notifier_call  = xfrm_dev_event,
2919 };
2920
2921 #ifdef CONFIG_XFRM_STATISTICS
2922 static int __net_init xfrm_statistics_init(struct net *net)
2923 {
2924         int rv;
2925         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2926         if (!net->mib.xfrm_statistics)
2927                 return -ENOMEM;
2928         rv = xfrm_proc_init(net);
2929         if (rv < 0)
2930                 free_percpu(net->mib.xfrm_statistics);
2931         return rv;
2932 }
2933
2934 static void xfrm_statistics_fini(struct net *net)
2935 {
2936         xfrm_proc_fini(net);
2937         free_percpu(net->mib.xfrm_statistics);
2938 }
2939 #else
2940 static int __net_init xfrm_statistics_init(struct net *net)
2941 {
2942         return 0;
2943 }
2944
2945 static void xfrm_statistics_fini(struct net *net)
2946 {
2947 }
2948 #endif
2949
2950 static int __net_init xfrm_policy_init(struct net *net)
2951 {
2952         unsigned int hmask, sz;
2953         int dir;
2954
2955         if (net_eq(net, &init_net))
2956                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2957                                            sizeof(struct xfrm_dst),
2958                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2959                                            NULL);
2960
2961         hmask = 8 - 1;
2962         sz = (hmask+1) * sizeof(struct hlist_head);
2963
2964         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2965         if (!net->xfrm.policy_byidx)
2966                 goto out_byidx;
2967         net->xfrm.policy_idx_hmask = hmask;
2968
2969         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2970                 struct xfrm_policy_hash *htab;
2971
2972                 net->xfrm.policy_count[dir] = 0;
2973                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2974
2975                 htab = &net->xfrm.policy_bydst[dir];
2976                 htab->table = xfrm_hash_alloc(sz);
2977                 if (!htab->table)
2978                         goto out_bydst;
2979                 htab->hmask = hmask;
2980                 htab->dbits4 = 32;
2981                 htab->sbits4 = 32;
2982                 htab->dbits6 = 128;
2983                 htab->sbits6 = 128;
2984         }
2985         net->xfrm.policy_hthresh.lbits4 = 32;
2986         net->xfrm.policy_hthresh.rbits4 = 32;
2987         net->xfrm.policy_hthresh.lbits6 = 128;
2988         net->xfrm.policy_hthresh.rbits6 = 128;
2989
2990         seqlock_init(&net->xfrm.policy_hthresh.lock);
2991
2992         INIT_LIST_HEAD(&net->xfrm.policy_all);
2993         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2994         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
2995         if (net_eq(net, &init_net))
2996                 register_netdevice_notifier(&xfrm_dev_notifier);
2997         return 0;
2998
2999 out_bydst:
3000         for (dir--; dir >= 0; dir--) {
3001                 struct xfrm_policy_hash *htab;
3002
3003                 htab = &net->xfrm.policy_bydst[dir];
3004                 xfrm_hash_free(htab->table, sz);
3005         }
3006         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3007 out_byidx:
3008         return -ENOMEM;
3009 }
3010
3011 static void xfrm_policy_fini(struct net *net)
3012 {
3013         unsigned int sz;
3014         int dir;
3015
3016         flush_work(&net->xfrm.policy_hash_work);
3017 #ifdef CONFIG_XFRM_SUB_POLICY
3018         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3019 #endif
3020         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3021
3022         WARN_ON(!list_empty(&net->xfrm.policy_all));
3023
3024         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
3025                 struct xfrm_policy_hash *htab;
3026
3027                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3028
3029                 htab = &net->xfrm.policy_bydst[dir];
3030                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3031                 WARN_ON(!hlist_empty(htab->table));
3032                 xfrm_hash_free(htab->table, sz);
3033         }
3034
3035         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3036         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3037         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3038 }
3039
3040 static int __net_init xfrm_net_init(struct net *net)
3041 {
3042         int rv;
3043
3044         rv = xfrm_statistics_init(net);
3045         if (rv < 0)
3046                 goto out_statistics;
3047         rv = xfrm_state_init(net);
3048         if (rv < 0)
3049                 goto out_state;
3050         rv = xfrm_policy_init(net);
3051         if (rv < 0)
3052                 goto out_policy;
3053         xfrm_dst_ops_init(net);
3054         rv = xfrm_sysctl_init(net);
3055         if (rv < 0)
3056                 goto out_sysctl;
3057         rv = flow_cache_init(net);
3058         if (rv < 0)
3059                 goto out;
3060
3061         /* Initialize the per-net locks here */
3062         spin_lock_init(&net->xfrm.xfrm_state_lock);
3063         rwlock_init(&net->xfrm.xfrm_policy_lock);
3064         mutex_init(&net->xfrm.xfrm_cfg_mutex);
3065
3066         return 0;
3067
3068 out:
3069         xfrm_sysctl_fini(net);
3070 out_sysctl:
3071         xfrm_policy_fini(net);
3072 out_policy:
3073         xfrm_state_fini(net);
3074 out_state:
3075         xfrm_statistics_fini(net);
3076 out_statistics:
3077         return rv;
3078 }
3079
3080 static void __net_exit xfrm_net_exit(struct net *net)
3081 {
3082         flow_cache_fini(net);
3083         xfrm_sysctl_fini(net);
3084         xfrm_policy_fini(net);
3085         xfrm_state_fini(net);
3086         xfrm_statistics_fini(net);
3087 }
3088
3089 static struct pernet_operations __net_initdata xfrm_net_ops = {
3090         .init = xfrm_net_init,
3091         .exit = xfrm_net_exit,
3092 };
3093
3094 void __init xfrm_init(void)
3095 {
3096         register_pernet_subsys(&xfrm_net_ops);
3097         xfrm_input_init();
3098 }
3099
3100 #ifdef CONFIG_AUDITSYSCALL
3101 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3102                                          struct audit_buffer *audit_buf)
3103 {
3104         struct xfrm_sec_ctx *ctx = xp->security;
3105         struct xfrm_selector *sel = &xp->selector;
3106
3107         if (ctx)
3108                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3109                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3110
3111         switch (sel->family) {
3112         case AF_INET:
3113                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3114                 if (sel->prefixlen_s != 32)
3115                         audit_log_format(audit_buf, " src_prefixlen=%d",
3116                                          sel->prefixlen_s);
3117                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3118                 if (sel->prefixlen_d != 32)
3119                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3120                                          sel->prefixlen_d);
3121                 break;
3122         case AF_INET6:
3123                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3124                 if (sel->prefixlen_s != 128)
3125                         audit_log_format(audit_buf, " src_prefixlen=%d",
3126                                          sel->prefixlen_s);
3127                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3128                 if (sel->prefixlen_d != 128)
3129                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3130                                          sel->prefixlen_d);
3131                 break;
3132         }
3133 }
3134
3135 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3136 {
3137         struct audit_buffer *audit_buf;
3138
3139         audit_buf = xfrm_audit_start("SPD-add");
3140         if (audit_buf == NULL)
3141                 return;
3142         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3143         audit_log_format(audit_buf, " res=%u", result);
3144         xfrm_audit_common_policyinfo(xp, audit_buf);
3145         audit_log_end(audit_buf);
3146 }
3147 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3148
3149 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3150                               bool task_valid)
3151 {
3152         struct audit_buffer *audit_buf;
3153
3154         audit_buf = xfrm_audit_start("SPD-delete");
3155         if (audit_buf == NULL)
3156                 return;
3157         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3158         audit_log_format(audit_buf, " res=%u", result);
3159         xfrm_audit_common_policyinfo(xp, audit_buf);
3160         audit_log_end(audit_buf);
3161 }
3162 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3163 #endif
3164
3165 #ifdef CONFIG_XFRM_MIGRATE
3166 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3167                                         const struct xfrm_selector *sel_tgt)
3168 {
3169         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3170                 if (sel_tgt->family == sel_cmp->family &&
3171                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3172                                     sel_cmp->family) &&
3173                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3174                                     sel_cmp->family) &&
3175                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3176                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3177                         return true;
3178                 }
3179         } else {
3180                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3181                         return true;
3182                 }
3183         }
3184         return false;
3185 }
3186
3187 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3188                                                     u8 dir, u8 type, struct net *net)
3189 {
3190         struct xfrm_policy *pol, *ret = NULL;
3191         struct hlist_head *chain;
3192         u32 priority = ~0U;
3193
3194         read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3195         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3196         hlist_for_each_entry(pol, chain, bydst) {
3197                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3198                     pol->type == type) {
3199                         ret = pol;
3200                         priority = ret->priority;
3201                         break;
3202                 }
3203         }
3204         chain = &net->xfrm.policy_inexact[dir];
3205         hlist_for_each_entry(pol, chain, bydst) {
3206                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3207                     pol->type == type &&
3208                     pol->priority < priority) {
3209                         ret = pol;
3210                         break;
3211                 }
3212         }
3213
3214         if (ret)
3215                 xfrm_pol_hold(ret);
3216
3217         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3218
3219         return ret;
3220 }
3221
3222 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3223 {
3224         int match = 0;
3225
3226         if (t->mode == m->mode && t->id.proto == m->proto &&
3227             (m->reqid == 0 || t->reqid == m->reqid)) {
3228                 switch (t->mode) {
3229                 case XFRM_MODE_TUNNEL:
3230                 case XFRM_MODE_BEET:
3231                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3232                                             m->old_family) &&
3233                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3234                                             m->old_family)) {
3235                                 match = 1;
3236                         }
3237                         break;
3238                 case XFRM_MODE_TRANSPORT:
3239                         /* in case of transport mode, template does not store
3240                            any IP addresses, hence we just compare mode and
3241                            protocol */
3242                         match = 1;
3243                         break;
3244                 default:
3245                         break;
3246                 }
3247         }
3248         return match;
3249 }
3250
3251 /* update endpoint address(es) of template(s) */
3252 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3253                                struct xfrm_migrate *m, int num_migrate)
3254 {
3255         struct xfrm_migrate *mp;
3256         int i, j, n = 0;
3257
3258         write_lock_bh(&pol->lock);
3259         if (unlikely(pol->walk.dead)) {
3260                 /* target policy has been deleted */
3261                 write_unlock_bh(&pol->lock);
3262                 return -ENOENT;
3263         }
3264
3265         for (i = 0; i < pol->xfrm_nr; i++) {
3266                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3267                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3268                                 continue;
3269                         n++;
3270                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3271                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3272                                 continue;
3273                         /* update endpoints */
3274                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3275                                sizeof(pol->xfrm_vec[i].id.daddr));
3276                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3277                                sizeof(pol->xfrm_vec[i].saddr));
3278                         pol->xfrm_vec[i].encap_family = mp->new_family;
3279                         /* flush bundles */
3280                         atomic_inc(&pol->genid);
3281                 }
3282         }
3283
3284         write_unlock_bh(&pol->lock);
3285
3286         if (!n)
3287                 return -ENODATA;
3288
3289         return 0;
3290 }
3291
3292 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3293 {
3294         int i, j;
3295
3296         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3297                 return -EINVAL;
3298
3299         for (i = 0; i < num_migrate; i++) {
3300                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3301                                     m[i].old_family) &&
3302                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3303                                     m[i].old_family))
3304                         return -EINVAL;
3305                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3306                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3307                         return -EINVAL;
3308
3309                 /* check if there is any duplicated entry */
3310                 for (j = i + 1; j < num_migrate; j++) {
3311                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3312                                     sizeof(m[i].old_daddr)) &&
3313                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3314                                     sizeof(m[i].old_saddr)) &&
3315                             m[i].proto == m[j].proto &&
3316                             m[i].mode == m[j].mode &&
3317                             m[i].reqid == m[j].reqid &&
3318                             m[i].old_family == m[j].old_family)
3319                                 return -EINVAL;
3320                 }
3321         }
3322
3323         return 0;
3324 }
3325
3326 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3327                  struct xfrm_migrate *m, int num_migrate,
3328                  struct xfrm_kmaddress *k, struct net *net)
3329 {
3330         int i, err, nx_cur = 0, nx_new = 0;
3331         struct xfrm_policy *pol = NULL;
3332         struct xfrm_state *x, *xc;
3333         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3334         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3335         struct xfrm_migrate *mp;
3336
3337         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3338                 goto out;
3339
3340         /* Stage 1 - find policy */
3341         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3342                 err = -ENOENT;
3343                 goto out;
3344         }
3345
3346         /* Stage 2 - find and update state(s) */
3347         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3348                 if ((x = xfrm_migrate_state_find(mp, net))) {
3349                         x_cur[nx_cur] = x;
3350                         nx_cur++;
3351                         if ((xc = xfrm_state_migrate(x, mp))) {
3352                                 x_new[nx_new] = xc;
3353                                 nx_new++;
3354                         } else {
3355                                 err = -ENODATA;
3356                                 goto restore_state;
3357                         }
3358                 }
3359         }
3360
3361         /* Stage 3 - update policy */
3362         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3363                 goto restore_state;
3364
3365         /* Stage 4 - delete old state(s) */
3366         if (nx_cur) {
3367                 xfrm_states_put(x_cur, nx_cur);
3368                 xfrm_states_delete(x_cur, nx_cur);
3369         }
3370
3371         /* Stage 5 - announce */
3372         km_migrate(sel, dir, type, m, num_migrate, k);
3373
3374         xfrm_pol_put(pol);
3375
3376         return 0;
3377 out:
3378         return err;
3379
3380 restore_state:
3381         if (pol)
3382                 xfrm_pol_put(pol);
3383         if (nx_cur)
3384                 xfrm_states_put(x_cur, nx_cur);
3385         if (nx_new)
3386                 xfrm_states_delete(x_new, nx_new);
3387
3388         return err;
3389 }
3390 EXPORT_SYMBOL(xfrm_migrate);
3391 #endif