net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/flow.h>
  30 #include <net/xfrm.h>
  31 #include <net/ip.h>
  32 #ifdef CONFIG_XFRM_STATISTICS
  33 #include <net/snmp.h>
  34 #endif
  35
  36 #include "xfrm_hash.h"
  37
  38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  40 #define XFRM_MAX_QUEUE_LEN      100
  41
  42 struct xfrm_flo {
  43         struct dst_entry *dst_orig;
  44         u8 flags;
  45 };
  46
  47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  48 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
  49                                                 __read_mostly;
  50
  51 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  52
  53 static void xfrm_init_pmtu(struct dst_entry *dst);
  54 static int stale_bundle(struct dst_entry *dst);
  55 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  56 static void xfrm_policy_queue_process(unsigned long arg);
  57
  58 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
  59 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  60                                                 int dir);
  61
  62 static inline bool
  63 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  64 {
  65         const struct flowi4 *fl4 = &fl->u.ip4;
  66
  67         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  68                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  69                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  70                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  71                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  72                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  73 }
  74
  75 static inline bool
  76 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  77 {
  78         const struct flowi6 *fl6 = &fl->u.ip6;
  79
  80         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  81                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  82                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  83                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  84                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  85                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  86 }
  87
  88 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  89                          unsigned short family)
  90 {
  91         switch (family) {
  92         case AF_INET:
  93                 return __xfrm4_selector_match(sel, fl);
  94         case AF_INET6:
  95                 return __xfrm6_selector_match(sel, fl);
  96         }
  97         return false;
  98 }
  99
 100 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 101 {
 102         struct xfrm_policy_afinfo *afinfo;
 103
 104         if (unlikely(family >= NPROTO))
 105                 return NULL;
 106         rcu_read_lock();
 107         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 108         if (unlikely(!afinfo))
 109                 rcu_read_unlock();
 110         return afinfo;
 111 }
 112
 113 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
 114 {
 115         rcu_read_unlock();
 116 }
 117
 118 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
 119                                                   const xfrm_address_t *saddr,
 120                                                   const xfrm_address_t *daddr,
 121                                                   int family)
 122 {
 123         struct xfrm_policy_afinfo *afinfo;
 124         struct dst_entry *dst;
 125
 126         afinfo = xfrm_policy_get_afinfo(family);
 127         if (unlikely(afinfo == NULL))
 128                 return ERR_PTR(-EAFNOSUPPORT);
 129
 130         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 131
 132         xfrm_policy_put_afinfo(afinfo);
 133
 134         return dst;
 135 }
 136
 137 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 138                                                 xfrm_address_t *prev_saddr,
 139                                                 xfrm_address_t *prev_daddr,
 140                                                 int family)
 141 {
 142         struct net *net = xs_net(x);
 143         xfrm_address_t *saddr = &x->props.saddr;
 144         xfrm_address_t *daddr = &x->id.daddr;
 145         struct dst_entry *dst;
 146
 147         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 148                 saddr = x->coaddr;
 149                 daddr = prev_daddr;
 150         }
 151         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 152                 saddr = prev_saddr;
 153                 daddr = x->coaddr;
 154         }
 155
 156         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 157
 158         if (!IS_ERR(dst)) {
 159                 if (prev_saddr != saddr)
 160                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 161                 if (prev_daddr != daddr)
 162                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 163         }
 164
 165         return dst;
 166 }
 167
 168 static inline unsigned long make_jiffies(long secs)
 169 {
 170         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 171                 return MAX_SCHEDULE_TIMEOUT-1;
 172         else
 173                 return secs*HZ;
 174 }
 175
 176 static void xfrm_policy_timer(unsigned long data)
 177 {
 178         struct xfrm_policy *xp = (struct xfrm_policy *)data;
 179         unsigned long now = get_seconds();
 180         long next = LONG_MAX;
 181         int warn = 0;
 182         int dir;
 183
 184         read_lock(&xp->lock);
 185
 186         if (unlikely(xp->walk.dead))
 187                 goto out;
 188
 189         dir = xfrm_policy_id2dir(xp->index);
 190
 191         if (xp->lft.hard_add_expires_seconds) {
 192                 long tmo = xp->lft.hard_add_expires_seconds +
 193                         xp->curlft.add_time - now;
 194                 if (tmo <= 0)
 195                         goto expired;
 196                 if (tmo < next)
 197                         next = tmo;
 198         }
 199         if (xp->lft.hard_use_expires_seconds) {
 200                 long tmo = xp->lft.hard_use_expires_seconds +
 201                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 202                 if (tmo <= 0)
 203                         goto expired;
 204                 if (tmo < next)
 205                         next = tmo;
 206         }
 207         if (xp->lft.soft_add_expires_seconds) {
 208                 long tmo = xp->lft.soft_add_expires_seconds +
 209                         xp->curlft.add_time - now;
 210                 if (tmo <= 0) {
 211                         warn = 1;
 212                         tmo = XFRM_KM_TIMEOUT;
 213                 }
 214                 if (tmo < next)
 215                         next = tmo;
 216         }
 217         if (xp->lft.soft_use_expires_seconds) {
 218                 long tmo = xp->lft.soft_use_expires_seconds +
 219                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 220                 if (tmo <= 0) {
 221                         warn = 1;
 222                         tmo = XFRM_KM_TIMEOUT;
 223                 }
 224                 if (tmo < next)
 225                         next = tmo;
 226         }
 227
 228         if (warn)
 229                 km_policy_expired(xp, dir, 0, 0);
 230         if (next != LONG_MAX &&
 231             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 232                 xfrm_pol_hold(xp);
 233
 234 out:
 235         read_unlock(&xp->lock);
 236         xfrm_pol_put(xp);
 237         return;
 238
 239 expired:
 240         read_unlock(&xp->lock);
 241         if (!xfrm_policy_delete(xp, dir))
 242                 km_policy_expired(xp, dir, 1, 0);
 243         xfrm_pol_put(xp);
 244 }
 245
 246 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 247 {
 248         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 249
 250         if (unlikely(pol->walk.dead))
 251                 flo = NULL;
 252         else
 253                 xfrm_pol_hold(pol);
 254
 255         return flo;
 256 }
 257
 258 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 259 {
 260         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 261
 262         return !pol->walk.dead;
 263 }
 264
 265 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 266 {
 267         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 268 }
 269
 270 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 271         .get = xfrm_policy_flo_get,
 272         .check = xfrm_policy_flo_check,
 273         .delete = xfrm_policy_flo_delete,
 274 };
 275
 276 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 277  * SPD calls.
 278  */
 279
 280 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 281 {
 282         struct xfrm_policy *policy;
 283
 284         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 285
 286         if (policy) {
 287                 write_pnet(&policy->xp_net, net);
 288                 INIT_LIST_HEAD(&policy->walk.all);
 289                 INIT_HLIST_NODE(&policy->bydst);
 290                 INIT_HLIST_NODE(&policy->byidx);
 291                 rwlock_init(&policy->lock);
 292                 atomic_set(&policy->refcnt, 1);
 293                 skb_queue_head_init(&policy->polq.hold_queue);
 294                 setup_timer(&policy->timer, xfrm_policy_timer,
 295                                 (unsigned long)policy);
 296                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 297                             (unsigned long)policy);
 298                 policy->flo.ops = &xfrm_policy_fc_ops;
 299         }
 300         return policy;
 301 }
 302 EXPORT_SYMBOL(xfrm_policy_alloc);
 303
 304 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 305
 306 void xfrm_policy_destroy(struct xfrm_policy *policy)
 307 {
 308         BUG_ON(!policy->walk.dead);
 309
 310         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 311                 BUG();
 312
 313         security_xfrm_policy_free(policy->security);
 314         kfree(policy);
 315 }
 316 EXPORT_SYMBOL(xfrm_policy_destroy);
 317
 318 static void xfrm_queue_purge(struct sk_buff_head *list)
 319 {
 320         struct sk_buff *skb;
 321
 322         while ((skb = skb_dequeue(list)) != NULL)
 323                 kfree_skb(skb);
 324 }
 325
 326 /* Rule must be locked. Release descentant resources, announce
 327  * entry dead. The rule must be unlinked from lists to the moment.
 328  */
 329
 330 static void xfrm_policy_kill(struct xfrm_policy *policy)
 331 {
 332         policy->walk.dead = 1;
 333
 334         atomic_inc(&policy->genid);
 335
 336         if (del_timer(&policy->polq.hold_timer))
 337                 xfrm_pol_put(policy);
 338         xfrm_queue_purge(&policy->polq.hold_queue);
 339
 340         if (del_timer(&policy->timer))
 341                 xfrm_pol_put(policy);
 342
 343         xfrm_pol_put(policy);
 344 }
 345
 346 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 347
 348 static inline unsigned int idx_hash(struct net *net, u32 index)
 349 {
 350         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 351 }
 352
 353 /* calculate policy hash thresholds */
 354 static void __get_hash_thresh(struct net *net,
 355                               unsigned short family, int dir,
 356                               u8 *dbits, u8 *sbits)
 357 {
 358         switch (family) {
 359         case AF_INET:
 360                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 361                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 362                 break;
 363
 364         case AF_INET6:
 365                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 366                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 367                 break;
 368
 369         default:
 370                 *dbits = 0;
 371                 *sbits = 0;
 372         }
 373 }
 374
 375 static struct hlist_head *policy_hash_bysel(struct net *net,
 376                                             const struct xfrm_selector *sel,
 377                                             unsigned short family, int dir)
 378 {
 379         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 380         unsigned int hash;
 381         u8 dbits;
 382         u8 sbits;
 383
 384         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 385         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 386
 387         return (hash == hmask + 1 ?
 388                 &net->xfrm.policy_inexact[dir] :
 389                 net->xfrm.policy_bydst[dir].table + hash);
 390 }
 391
 392 static struct hlist_head *policy_hash_direct(struct net *net,
 393                                              const xfrm_address_t *daddr,
 394                                              const xfrm_address_t *saddr,
 395                                              unsigned short family, int dir)
 396 {
 397         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 398         unsigned int hash;
 399         u8 dbits;
 400         u8 sbits;
 401
 402         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 403         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 404
 405         return net->xfrm.policy_bydst[dir].table + hash;
 406 }
 407
 408 static void xfrm_dst_hash_transfer(struct net *net,
 409                                    struct hlist_head *list,
 410                                    struct hlist_head *ndsttable,
 411                                    unsigned int nhashmask,
 412                                    int dir)
 413 {
 414         struct hlist_node *tmp, *entry0 = NULL;
 415         struct xfrm_policy *pol;
 416         unsigned int h0 = 0;
 417         u8 dbits;
 418         u8 sbits;
 419
 420 redo:
 421         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 422                 unsigned int h;
 423
 424                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 425                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 426                                 pol->family, nhashmask, dbits, sbits);
 427                 if (!entry0) {
 428                         hlist_del(&pol->bydst);
 429                         hlist_add_head(&pol->bydst, ndsttable+h);
 430                         h0 = h;
 431                 } else {
 432                         if (h != h0)
 433                                 continue;
 434                         hlist_del(&pol->bydst);
 435                         hlist_add_behind(&pol->bydst, entry0);
 436                 }
 437                 entry0 = &pol->bydst;
 438         }
 439         if (!hlist_empty(list)) {
 440                 entry0 = NULL;
 441                 goto redo;
 442         }
 443 }
 444
 445 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 446                                    struct hlist_head *nidxtable,
 447                                    unsigned int nhashmask)
 448 {
 449         struct hlist_node *tmp;
 450         struct xfrm_policy *pol;
 451
 452         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 453                 unsigned int h;
 454
 455                 h = __idx_hash(pol->index, nhashmask);
 456                 hlist_add_head(&pol->byidx, nidxtable+h);
 457         }
 458 }
 459
 460 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 461 {
 462         return ((old_hmask + 1) << 1) - 1;
 463 }
 464
 465 static void xfrm_bydst_resize(struct net *net, int dir)
 466 {
 467         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 468         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 469         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 470         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 471         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 472         int i;
 473
 474         if (!ndst)
 475                 return;
 476
 477         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 478
 479         for (i = hmask; i >= 0; i--)
 480                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 481
 482         net->xfrm.policy_bydst[dir].table = ndst;
 483         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 484
 485         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 486
 487         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 488 }
 489
 490 static void xfrm_byidx_resize(struct net *net, int total)
 491 {
 492         unsigned int hmask = net->xfrm.policy_idx_hmask;
 493         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 494         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 495         struct hlist_head *oidx = net->xfrm.policy_byidx;
 496         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 497         int i;
 498
 499         if (!nidx)
 500                 return;
 501
 502         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 503
 504         for (i = hmask; i >= 0; i--)
 505                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 506
 507         net->xfrm.policy_byidx = nidx;
 508         net->xfrm.policy_idx_hmask = nhashmask;
 509
 510         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 511
 512         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 513 }
 514
 515 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 516 {
 517         unsigned int cnt = net->xfrm.policy_count[dir];
 518         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 519
 520         if (total)
 521                 *total += cnt;
 522
 523         if ((hmask + 1) < xfrm_policy_hashmax &&
 524             cnt > hmask)
 525                 return 1;
 526
 527         return 0;
 528 }
 529
 530 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 531 {
 532         unsigned int hmask = net->xfrm.policy_idx_hmask;
 533
 534         if ((hmask + 1) < xfrm_policy_hashmax &&
 535             total > hmask)
 536                 return 1;
 537
 538         return 0;
 539 }
 540
 541 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 542 {
 543         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 544         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 545         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 546         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 547         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 548         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 549         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 550         si->spdhcnt = net->xfrm.policy_idx_hmask;
 551         si->spdhmcnt = xfrm_policy_hashmax;
 552         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
 553 }
 554 EXPORT_SYMBOL(xfrm_spd_getinfo);
 555
 556 static DEFINE_MUTEX(hash_resize_mutex);
 557 static void xfrm_hash_resize(struct work_struct *work)
 558 {
 559         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 560         int dir, total;
 561
 562         mutex_lock(&hash_resize_mutex);
 563
 564         total = 0;
 565         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 566                 if (xfrm_bydst_should_resize(net, dir, &total))
 567                         xfrm_bydst_resize(net, dir);
 568         }
 569         if (xfrm_byidx_should_resize(net, total))
 570                 xfrm_byidx_resize(net, total);
 571
 572         mutex_unlock(&hash_resize_mutex);
 573 }
 574
 575 static void xfrm_hash_rebuild(struct work_struct *work)
 576 {
 577         struct net *net = container_of(work, struct net,
 578                                        xfrm.policy_hthresh.work);
 579         unsigned int hmask;
 580         struct xfrm_policy *pol;
 581         struct xfrm_policy *policy;
 582         struct hlist_head *chain;
 583         struct hlist_head *odst;
 584         struct hlist_node *newpos;
 585         int i;
 586         int dir;
 587         unsigned seq;
 588         u8 lbits4, rbits4, lbits6, rbits6;
 589
 590         mutex_lock(&hash_resize_mutex);
 591
 592         /* read selector prefixlen thresholds */
 593         do {
 594                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
 595
 596                 lbits4 = net->xfrm.policy_hthresh.lbits4;
 597                 rbits4 = net->xfrm.policy_hthresh.rbits4;
 598                 lbits6 = net->xfrm.policy_hthresh.lbits6;
 599                 rbits6 = net->xfrm.policy_hthresh.rbits6;
 600         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
 601
 602         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 603
 604         /* reset the bydst and inexact table in all directions */
 605         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 606                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
 607                 hmask = net->xfrm.policy_bydst[dir].hmask;
 608                 odst = net->xfrm.policy_bydst[dir].table;
 609                 for (i = hmask; i >= 0; i--)
 610                         INIT_HLIST_HEAD(odst + i);
 611                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
 612                         /* dir out => dst = remote, src = local */
 613                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
 614                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
 615                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
 616                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
 617                 } else {
 618                         /* dir in/fwd => dst = local, src = remote */
 619                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
 620                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
 621                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
 622                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
 623                 }
 624         }
 625
 626         /* re-insert all policies by order of creation */
 627         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 628                 newpos = NULL;
 629                 chain = policy_hash_bysel(net, &policy->selector,
 630                                           policy->family,
 631                                           xfrm_policy_id2dir(policy->index));
 632                 hlist_for_each_entry(pol, chain, bydst) {
 633                         if (policy->priority >= pol->priority)
 634                                 newpos = &pol->bydst;
 635                         else
 636                                 break;
 637                 }
 638                 if (newpos)
 639                         hlist_add_behind(&policy->bydst, newpos);
 640                 else
 641                         hlist_add_head(&policy->bydst, chain);
 642         }
 643
 644         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 645
 646         mutex_unlock(&hash_resize_mutex);
 647 }
 648
 649 void xfrm_policy_hash_rebuild(struct net *net)
 650 {
 651         schedule_work(&net->xfrm.policy_hthresh.work);
 652 }
 653 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
 654
 655 /* Generate new index... KAME seems to generate them ordered by cost
 656  * of an absolute inpredictability of ordering of rules. This will not pass. */
 657 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 658 {
 659         static u32 idx_generator;
 660
 661         for (;;) {
 662                 struct hlist_head *list;
 663                 struct xfrm_policy *p;
 664                 u32 idx;
 665                 int found;
 666
 667                 if (!index) {
 668                         idx = (idx_generator | dir);
 669                         idx_generator += 8;
 670                 } else {
 671                         idx = index;
 672                         index = 0;
 673                 }
 674
 675                 if (idx == 0)
 676                         idx = 8;
 677                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 678                 found = 0;
 679                 hlist_for_each_entry(p, list, byidx) {
 680                         if (p->index == idx) {
 681                                 found = 1;
 682                                 break;
 683                         }
 684                 }
 685                 if (!found)
 686                         return idx;
 687         }
 688 }
 689
 690 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 691 {
 692         u32 *p1 = (u32 *) s1;
 693         u32 *p2 = (u32 *) s2;
 694         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 695         int i;
 696
 697         for (i = 0; i < len; i++) {
 698                 if (p1[i] != p2[i])
 699                         return 1;
 700         }
 701
 702         return 0;
 703 }
 704
 705 static void xfrm_policy_requeue(struct xfrm_policy *old,
 706                                 struct xfrm_policy *new)
 707 {
 708         struct xfrm_policy_queue *pq = &old->polq;
 709         struct sk_buff_head list;
 710
 711         __skb_queue_head_init(&list);
 712
 713         spin_lock_bh(&pq->hold_queue.lock);
 714         skb_queue_splice_init(&pq->hold_queue, &list);
 715         if (del_timer(&pq->hold_timer))
 716                 xfrm_pol_put(old);
 717         spin_unlock_bh(&pq->hold_queue.lock);
 718
 719         if (skb_queue_empty(&list))
 720                 return;
 721
 722         pq = &new->polq;
 723
 724         spin_lock_bh(&pq->hold_queue.lock);
 725         skb_queue_splice(&list, &pq->hold_queue);
 726         pq->timeout = XFRM_QUEUE_TMO_MIN;
 727         if (!mod_timer(&pq->hold_timer, jiffies))
 728                 xfrm_pol_hold(new);
 729         spin_unlock_bh(&pq->hold_queue.lock);
 730 }
 731
 732 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 733                                    struct xfrm_policy *pol)
 734 {
 735         u32 mark = policy->mark.v & policy->mark.m;
 736
 737         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 738                 return true;
 739
 740         if ((mark & pol->mark.m) == pol->mark.v &&
 741             policy->priority == pol->priority)
 742                 return true;
 743
 744         return false;
 745 }
 746
 747 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 748 {
 749         struct net *net = xp_net(policy);
 750         struct xfrm_policy *pol;
 751         struct xfrm_policy *delpol;
 752         struct hlist_head *chain;
 753         struct hlist_node *newpos;
 754
 755         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 756         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 757         delpol = NULL;
 758         newpos = NULL;
 759         hlist_for_each_entry(pol, chain, bydst) {
 760                 if (pol->type == policy->type &&
 761                     !selector_cmp(&pol->selector, &policy->selector) &&
 762                     xfrm_policy_mark_match(policy, pol) &&
 763                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 764                     !WARN_ON(delpol)) {
 765                         if (excl) {
 766                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 767                                 return -EEXIST;
 768                         }
 769                         delpol = pol;
 770                         if (policy->priority > pol->priority)
 771                                 continue;
 772                 } else if (policy->priority >= pol->priority) {
 773                         newpos = &pol->bydst;
 774                         continue;
 775                 }
 776                 if (delpol)
 777                         break;
 778         }
 779         if (newpos)
 780                 hlist_add_behind(&policy->bydst, newpos);
 781         else
 782                 hlist_add_head(&policy->bydst, chain);
 783         __xfrm_policy_link(policy, 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         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 803
 804         if (delpol)
 805                 xfrm_policy_kill(delpol);
 806         else if (xfrm_bydst_should_resize(net, dir, NULL))
 807                 schedule_work(&net->xfrm.policy_hash_work);
 808
 809         return 0;
 810 }
 811 EXPORT_SYMBOL(xfrm_policy_insert);
 812
 813 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 814                                           int dir, struct xfrm_selector *sel,
 815                                           struct xfrm_sec_ctx *ctx, int delete,
 816                                           int *err)
 817 {
 818         struct xfrm_policy *pol, *ret;
 819         struct hlist_head *chain;
 820
 821         *err = 0;
 822         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 823         chain = policy_hash_bysel(net, sel, sel->family, dir);
 824         ret = NULL;
 825         hlist_for_each_entry(pol, chain, bydst) {
 826                 if (pol->type == type &&
 827                     (mark & pol->mark.m) == pol->mark.v &&
 828                     !selector_cmp(sel, &pol->selector) &&
 829                     xfrm_sec_ctx_match(ctx, pol->security)) {
 830                         xfrm_pol_hold(pol);
 831                         if (delete) {
 832                                 *err = security_xfrm_policy_delete(
 833                                                                 pol->security);
 834                                 if (*err) {
 835                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 836                                         return pol;
 837                                 }
 838                                 __xfrm_policy_unlink(pol, dir);
 839                         }
 840                         ret = pol;
 841                         break;
 842                 }
 843         }
 844         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 845
 846         if (ret && delete)
 847                 xfrm_policy_kill(ret);
 848         return ret;
 849 }
 850 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 851
 852 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 853                                      int dir, u32 id, int delete, int *err)
 854 {
 855         struct xfrm_policy *pol, *ret;
 856         struct hlist_head *chain;
 857
 858         *err = -ENOENT;
 859         if (xfrm_policy_id2dir(id) != dir)
 860                 return NULL;
 861
 862         *err = 0;
 863         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 864         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 865         ret = NULL;
 866         hlist_for_each_entry(pol, chain, byidx) {
 867                 if (pol->type == type && pol->index == id &&
 868                     (mark & pol->mark.m) == pol->mark.v) {
 869                         xfrm_pol_hold(pol);
 870                         if (delete) {
 871                                 *err = security_xfrm_policy_delete(
 872                                                                 pol->security);
 873                                 if (*err) {
 874                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 875                                         return pol;
 876                                 }
 877                                 __xfrm_policy_unlink(pol, dir);
 878                         }
 879                         ret = pol;
 880                         break;
 881                 }
 882         }
 883         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 884
 885         if (ret && delete)
 886                 xfrm_policy_kill(ret);
 887         return ret;
 888 }
 889 EXPORT_SYMBOL(xfrm_policy_byid);
 890
 891 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 892 static inline int
 893 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 894 {
 895         int dir, err = 0;
 896
 897         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 898                 struct xfrm_policy *pol;
 899                 int i;
 900
 901                 hlist_for_each_entry(pol,
 902                                      &net->xfrm.policy_inexact[dir], bydst) {
 903                         if (pol->type != type)
 904                                 continue;
 905                         err = security_xfrm_policy_delete(pol->security);
 906                         if (err) {
 907                                 xfrm_audit_policy_delete(pol, 0, task_valid);
 908                                 return err;
 909                         }
 910                 }
 911                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 912                         hlist_for_each_entry(pol,
 913                                              net->xfrm.policy_bydst[dir].table + i,
 914                                              bydst) {
 915                                 if (pol->type != type)
 916                                         continue;
 917                                 err = security_xfrm_policy_delete(
 918                                                                 pol->security);
 919                                 if (err) {
 920                                         xfrm_audit_policy_delete(pol, 0,
 921                                                                  task_valid);
 922                                         return err;
 923                                 }
 924                         }
 925                 }
 926         }
 927         return err;
 928 }
 929 #else
 930 static inline int
 931 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
 932 {
 933         return 0;
 934 }
 935 #endif
 936
 937 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
 938 {
 939         int dir, err = 0, cnt = 0;
 940
 941         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 942
 943         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
 944         if (err)
 945                 goto out;
 946
 947         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 948                 struct xfrm_policy *pol;
 949                 int i;
 950
 951         again1:
 952                 hlist_for_each_entry(pol,
 953                                      &net->xfrm.policy_inexact[dir], bydst) {
 954                         if (pol->type != type)
 955                                 continue;
 956                         __xfrm_policy_unlink(pol, dir);
 957                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 958                         cnt++;
 959
 960                         xfrm_audit_policy_delete(pol, 1, task_valid);
 961
 962                         xfrm_policy_kill(pol);
 963
 964                         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 965                         goto again1;
 966                 }
 967
 968                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 969         again2:
 970                         hlist_for_each_entry(pol,
 971                                              net->xfrm.policy_bydst[dir].table + i,
 972                                              bydst) {
 973                                 if (pol->type != type)
 974                                         continue;
 975                                 __xfrm_policy_unlink(pol, dir);
 976                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 977                                 cnt++;
 978
 979                                 xfrm_audit_policy_delete(pol, 1, task_valid);
 980                                 xfrm_policy_kill(pol);
 981
 982                                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
 983                                 goto again2;
 984                         }
 985                 }
 986
 987         }
 988         if (!cnt)
 989                 err = -ESRCH;
 990 out:
 991         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 992         return err;
 993 }
 994 EXPORT_SYMBOL(xfrm_policy_flush);
 995
 996 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 997                      int (*func)(struct xfrm_policy *, int, int, void*),
 998                      void *data)
 999 {
1000         struct xfrm_policy *pol;
1001         struct xfrm_policy_walk_entry *x;
1002         int error = 0;
1003
1004         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1005             walk->type != XFRM_POLICY_TYPE_ANY)
1006                 return -EINVAL;
1007
1008         if (list_empty(&walk->walk.all) && walk->seq != 0)
1009                 return 0;
1010
1011         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1012         if (list_empty(&walk->walk.all))
1013                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1014         else
1015                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
1016         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1017                 if (x->dead)
1018                         continue;
1019                 pol = container_of(x, struct xfrm_policy, walk);
1020                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1021                     walk->type != pol->type)
1022                         continue;
1023                 error = func(pol, xfrm_policy_id2dir(pol->index),
1024                              walk->seq, data);
1025                 if (error) {
1026                         list_move_tail(&walk->walk.all, &x->all);
1027                         goto out;
1028                 }
1029                 walk->seq++;
1030         }
1031         if (walk->seq == 0) {
1032                 error = -ENOENT;
1033                 goto out;
1034         }
1035         list_del_init(&walk->walk.all);
1036 out:
1037         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1038         return error;
1039 }
1040 EXPORT_SYMBOL(xfrm_policy_walk);
1041
1042 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1043 {
1044         INIT_LIST_HEAD(&walk->walk.all);
1045         walk->walk.dead = 1;
1046         walk->type = type;
1047         walk->seq = 0;
1048 }
1049 EXPORT_SYMBOL(xfrm_policy_walk_init);
1050
1051 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1052 {
1053         if (list_empty(&walk->walk.all))
1054                 return;
1055
1056         write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1057         list_del(&walk->walk.all);
1058         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1059 }
1060 EXPORT_SYMBOL(xfrm_policy_walk_done);
1061
1062 /*
1063  * Find policy to apply to this flow.
1064  *
1065  * Returns 0 if policy found, else an -errno.
1066  */
1067 static int xfrm_policy_match(const struct xfrm_policy *pol,
1068                              const struct flowi *fl,
1069                              u8 type, u16 family, int dir)
1070 {
1071         const struct xfrm_selector *sel = &pol->selector;
1072         int ret = -ESRCH;
1073         bool match;
1074
1075         if (pol->family != family ||
1076             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1077             pol->type != type)
1078                 return ret;
1079
1080         match = xfrm_selector_match(sel, fl, family);
1081         if (match)
1082                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1083                                                   dir);
1084
1085         return ret;
1086 }
1087
1088 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
1089                                                      const struct flowi *fl,
1090                                                      u16 family, u8 dir)
1091 {
1092         int err;
1093         struct xfrm_policy *pol, *ret;
1094         const xfrm_address_t *daddr, *saddr;
1095         struct hlist_head *chain;
1096         u32 priority = ~0U;
1097
1098         daddr = xfrm_flowi_daddr(fl, family);
1099         saddr = xfrm_flowi_saddr(fl, family);
1100         if (unlikely(!daddr || !saddr))
1101                 return NULL;
1102
1103         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1104         chain = policy_hash_direct(net, daddr, saddr, family, dir);
1105         ret = NULL;
1106         hlist_for_each_entry(pol, chain, bydst) {
1107                 err = xfrm_policy_match(pol, fl, type, family, dir);
1108                 if (err) {
1109                         if (err == -ESRCH)
1110                                 continue;
1111                         else {
1112                                 ret = ERR_PTR(err);
1113                                 goto fail;
1114                         }
1115                 } else {
1116                         ret = pol;
1117                         priority = ret->priority;
1118                         break;
1119                 }
1120         }
1121         chain = &net->xfrm.policy_inexact[dir];
1122         hlist_for_each_entry(pol, chain, bydst) {
1123                 err = xfrm_policy_match(pol, fl, type, family, dir);
1124                 if (err) {
1125                         if (err == -ESRCH)
1126                                 continue;
1127                         else {
1128                                 ret = ERR_PTR(err);
1129                                 goto fail;
1130                         }
1131                 } else if (pol->priority < priority) {
1132                         ret = pol;
1133                         break;
1134                 }
1135         }
1136         if (ret)
1137                 xfrm_pol_hold(ret);
1138 fail:
1139         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1140
1141         return ret;
1142 }
1143
1144 static struct xfrm_policy *
1145 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1146 {
1147 #ifdef CONFIG_XFRM_SUB_POLICY
1148         struct xfrm_policy *pol;
1149
1150         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1151         if (pol != NULL)
1152                 return pol;
1153 #endif
1154         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1155 }
1156
1157 static int flow_to_policy_dir(int dir)
1158 {
1159         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1160             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1161             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1162                 return dir;
1163
1164         switch (dir) {
1165         default:
1166         case FLOW_DIR_IN:
1167                 return XFRM_POLICY_IN;
1168         case FLOW_DIR_OUT:
1169                 return XFRM_POLICY_OUT;
1170         case FLOW_DIR_FWD:
1171                 return XFRM_POLICY_FWD;
1172         }
1173 }
1174
1175 static struct flow_cache_object *
1176 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1177                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1178 {
1179         struct xfrm_policy *pol;
1180
1181         if (old_obj)
1182                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1183
1184         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1185         if (IS_ERR_OR_NULL(pol))
1186                 return ERR_CAST(pol);
1187
1188         /* Resolver returns two references:
1189          * one for cache and one for caller of flow_cache_lookup() */
1190         xfrm_pol_hold(pol);
1191
1192         return &pol->flo;
1193 }
1194
1195 static inline int policy_to_flow_dir(int dir)
1196 {
1197         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1198             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1199             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1200                 return dir;
1201         switch (dir) {
1202         default:
1203         case XFRM_POLICY_IN:
1204                 return FLOW_DIR_IN;
1205         case XFRM_POLICY_OUT:
1206                 return FLOW_DIR_OUT;
1207         case XFRM_POLICY_FWD:
1208                 return FLOW_DIR_FWD;
1209         }
1210 }
1211
1212 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1213                                                  const struct flowi *fl)
1214 {
1215         struct xfrm_policy *pol;
1216         struct net *net = sock_net(sk);
1217
1218         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1219         if ((pol = sk->sk_policy[dir]) != NULL) {
1220                 bool match = xfrm_selector_match(&pol->selector, fl,
1221                                                  sk->sk_family);
1222                 int err = 0;
1223
1224                 if (match) {
1225                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1226                                 pol = NULL;
1227                                 goto out;
1228                         }
1229                         err = security_xfrm_policy_lookup(pol->security,
1230                                                       fl->flowi_secid,
1231                                                       policy_to_flow_dir(dir));
1232                         if (!err)
1233                                 xfrm_pol_hold(pol);
1234                         else if (err == -ESRCH)
1235                                 pol = NULL;
1236                         else
1237                                 pol = ERR_PTR(err);
1238                 } else
1239                         pol = NULL;
1240         }
1241 out:
1242         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1243         return pol;
1244 }
1245
1246 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1247 {
1248         struct net *net = xp_net(pol);
1249
1250         list_add(&pol->walk.all, &net->xfrm.policy_all);
1251         net->xfrm.policy_count[dir]++;
1252         xfrm_pol_hold(pol);
1253 }
1254
1255 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1256                                                 int dir)
1257 {
1258         struct net *net = xp_net(pol);
1259
1260         if (list_empty(&pol->walk.all))
1261                 return NULL;
1262
1263         /* Socket policies are not hashed. */
1264         if (!hlist_unhashed(&pol->bydst)) {
1265                 hlist_del(&pol->bydst);
1266                 hlist_del(&pol->byidx);
1267         }
1268
1269         list_del_init(&pol->walk.all);
1270         net->xfrm.policy_count[dir]--;
1271
1272         return pol;
1273 }
1274
1275 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
1276 {
1277         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
1278 }
1279
1280 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
1281 {
1282         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
1283 }
1284
1285 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1286 {
1287         struct net *net = xp_net(pol);
1288
1289         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1290         pol = __xfrm_policy_unlink(pol, dir);
1291         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1292         if (pol) {
1293                 xfrm_policy_kill(pol);
1294                 return 0;
1295         }
1296         return -ENOENT;
1297 }
1298 EXPORT_SYMBOL(xfrm_policy_delete);
1299
1300 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1301 {
1302         struct net *net = xp_net(pol);
1303         struct xfrm_policy *old_pol;
1304
1305 #ifdef CONFIG_XFRM_SUB_POLICY
1306         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1307                 return -EINVAL;
1308 #endif
1309
1310         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1311         old_pol = sk->sk_policy[dir];
1312         sk->sk_policy[dir] = pol;
1313         if (pol) {
1314                 pol->curlft.add_time = get_seconds();
1315                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1316                 xfrm_sk_policy_link(pol, dir);
1317         }
1318         if (old_pol) {
1319                 if (pol)
1320                         xfrm_policy_requeue(old_pol, pol);
1321
1322                 /* Unlinking succeeds always. This is the only function
1323                  * allowed to delete or replace socket policy.
1324                  */
1325                 xfrm_sk_policy_unlink(old_pol, dir);
1326         }
1327         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1328
1329         if (old_pol) {
1330                 xfrm_policy_kill(old_pol);
1331         }
1332         return 0;
1333 }
1334
1335 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1336 {
1337         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1338         struct net *net = xp_net(old);
1339
1340         if (newp) {
1341                 newp->selector = old->selector;
1342                 if (security_xfrm_policy_clone(old->security,
1343                                                &newp->security)) {
1344                         kfree(newp);
1345                         return NULL;  /* ENOMEM */
1346                 }
1347                 newp->lft = old->lft;
1348                 newp->curlft = old->curlft;
1349                 newp->mark = old->mark;
1350                 newp->action = old->action;
1351                 newp->flags = old->flags;
1352                 newp->xfrm_nr = old->xfrm_nr;
1353                 newp->index = old->index;
1354                 newp->type = old->type;
1355                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1356                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1357                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1358                 xfrm_sk_policy_link(newp, dir);
1359                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1360                 xfrm_pol_put(newp);
1361         }
1362         return newp;
1363 }
1364
1365 int __xfrm_sk_clone_policy(struct sock *sk)
1366 {
1367         struct xfrm_policy *p0 = sk->sk_policy[0],
1368                            *p1 = sk->sk_policy[1];
1369
1370         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1371         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1372                 return -ENOMEM;
1373         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1374                 return -ENOMEM;
1375         return 0;
1376 }
1377
1378 static int
1379 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1380                unsigned short family)
1381 {
1382         int err;
1383         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1384
1385         if (unlikely(afinfo == NULL))
1386                 return -EINVAL;
1387         err = afinfo->get_saddr(net, local, remote);
1388         xfrm_policy_put_afinfo(afinfo);
1389         return err;
1390 }
1391
1392 /* Resolve list of templates for the flow, given policy. */
1393
1394 static int
1395 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1396                       struct xfrm_state **xfrm, unsigned short family)
1397 {
1398         struct net *net = xp_net(policy);
1399         int nx;
1400         int i, error;
1401         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1402         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1403         xfrm_address_t tmp;
1404
1405         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
1406                 struct xfrm_state *x;
1407                 xfrm_address_t *remote = daddr;
1408                 xfrm_address_t *local  = saddr;
1409                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1410
1411                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1412                     tmpl->mode == XFRM_MODE_BEET) {
1413                         remote = &tmpl->id.daddr;
1414                         local = &tmpl->saddr;
1415                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1416                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1417                                 if (error)
1418                                         goto fail;
1419                                 local = &tmp;
1420                         }
1421                 }
1422
1423                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1424
1425                 if (x && x->km.state == XFRM_STATE_VALID) {
1426                         xfrm[nx++] = x;
1427                         daddr = remote;
1428                         saddr = local;
1429                         continue;
1430                 }
1431                 if (x) {
1432                         error = (x->km.state == XFRM_STATE_ERROR ?
1433                                  -EINVAL : -EAGAIN);
1434                         xfrm_state_put(x);
1435                 } else if (error == -ESRCH) {
1436                         error = -EAGAIN;
1437                 }
1438
1439                 if (!tmpl->optional)
1440                         goto fail;
1441         }
1442         return nx;
1443
1444 fail:
1445         for (nx--; nx >= 0; nx--)
1446                 xfrm_state_put(xfrm[nx]);
1447         return error;
1448 }
1449
1450 static int
1451 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1452                   struct xfrm_state **xfrm, unsigned short family)
1453 {
1454         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1455         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1456         int cnx = 0;
1457         int error;
1458         int ret;
1459         int i;
1460
1461         for (i = 0; i < npols; i++) {
1462                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1463                         error = -ENOBUFS;
1464                         goto fail;
1465                 }
1466
1467                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1468                 if (ret < 0) {
1469                         error = ret;
1470                         goto fail;
1471                 } else
1472                         cnx += ret;
1473         }
1474
1475         /* found states are sorted for outbound processing */
1476         if (npols > 1)
1477                 xfrm_state_sort(xfrm, tpp, cnx, family);
1478
1479         return cnx;
1480
1481  fail:
1482         for (cnx--; cnx >= 0; cnx--)
1483                 xfrm_state_put(tpp[cnx]);
1484         return error;
1485
1486 }
1487
1488 /* Check that the bundle accepts the flow and its components are
1489  * still valid.
1490  */
1491
1492 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1493 {
1494         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1495         int tos;
1496
1497         if (!afinfo)
1498                 return -EINVAL;
1499
1500         tos = afinfo->get_tos(fl);
1501
1502         xfrm_policy_put_afinfo(afinfo);
1503
1504         return tos;
1505 }
1506
1507 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1508 {
1509         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1510         struct dst_entry *dst = &xdst->u.dst;
1511
1512         if (xdst->route == NULL) {
1513                 /* Dummy bundle - if it has xfrms we were not
1514                  * able to build bundle as template resolution failed.
1515                  * It means we need to try again resolving. */
1516                 if (xdst->num_xfrms > 0)
1517                         return NULL;
1518         } else if (dst->flags & DST_XFRM_QUEUE) {
1519                 return NULL;
1520         } else {
1521                 /* Real bundle */
1522                 if (stale_bundle(dst))
1523                         return NULL;
1524         }
1525
1526         dst_hold(dst);
1527         return flo;
1528 }
1529
1530 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1531 {
1532         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1533         struct dst_entry *dst = &xdst->u.dst;
1534
1535         if (!xdst->route)
1536                 return 0;
1537         if (stale_bundle(dst))
1538                 return 0;
1539
1540         return 1;
1541 }
1542
1543 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1544 {
1545         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1546         struct dst_entry *dst = &xdst->u.dst;
1547
1548         dst_free(dst);
1549 }
1550
1551 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1552         .get = xfrm_bundle_flo_get,
1553         .check = xfrm_bundle_flo_check,
1554         .delete = xfrm_bundle_flo_delete,
1555 };
1556
1557 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1558 {
1559         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1560         struct dst_ops *dst_ops;
1561         struct xfrm_dst *xdst;
1562
1563         if (!afinfo)
1564                 return ERR_PTR(-EINVAL);
1565
1566         switch (family) {
1567         case AF_INET:
1568                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1569                 break;
1570 #if IS_ENABLED(CONFIG_IPV6)
1571         case AF_INET6:
1572                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1573                 break;
1574 #endif
1575         default:
1576                 BUG();
1577         }
1578         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1579
1580         if (likely(xdst)) {
1581                 struct dst_entry *dst = &xdst->u.dst;
1582
1583                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1584                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1585                 if (afinfo->init_dst)
1586                         afinfo->init_dst(net, xdst);
1587         } else
1588                 xdst = ERR_PTR(-ENOBUFS);
1589
1590         xfrm_policy_put_afinfo(afinfo);
1591
1592         return xdst;
1593 }
1594
1595 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1596                                  int nfheader_len)
1597 {
1598         struct xfrm_policy_afinfo *afinfo =
1599                 xfrm_policy_get_afinfo(dst->ops->family);
1600         int err;
1601
1602         if (!afinfo)
1603                 return -EINVAL;
1604
1605         err = afinfo->init_path(path, dst, nfheader_len);
1606
1607         xfrm_policy_put_afinfo(afinfo);
1608
1609         return err;
1610 }
1611
1612 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1613                                 const struct flowi *fl)
1614 {
1615         struct xfrm_policy_afinfo *afinfo =
1616                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1617         int err;
1618
1619         if (!afinfo)
1620                 return -EINVAL;
1621
1622         err = afinfo->fill_dst(xdst, dev, fl);
1623
1624         xfrm_policy_put_afinfo(afinfo);
1625
1626         return err;
1627 }
1628
1629
1630 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1631  * all the metrics... Shortly, bundle a bundle.
1632  */
1633
1634 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1635                                             struct xfrm_state **xfrm, int nx,
1636                                             const struct flowi *fl,
1637                                             struct dst_entry *dst)
1638 {
1639         struct net *net = xp_net(policy);
1640         unsigned long now = jiffies;
1641         struct net_device *dev;
1642         struct xfrm_mode *inner_mode;
1643         struct dst_entry *dst_prev = NULL;
1644         struct dst_entry *dst0 = NULL;
1645         int i = 0;
1646         int err;
1647         int header_len = 0;
1648         int nfheader_len = 0;
1649         int trailer_len = 0;
1650         int tos;
1651         int family = policy->selector.family;
1652         xfrm_address_t saddr, daddr;
1653
1654         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1655
1656         tos = xfrm_get_tos(fl, family);
1657         err = tos;
1658         if (tos < 0)
1659                 goto put_states;
1660
1661         dst_hold(dst);
1662
1663         for (; i < nx; i++) {
1664                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1665                 struct dst_entry *dst1 = &xdst->u.dst;
1666
1667                 err = PTR_ERR(xdst);
1668                 if (IS_ERR(xdst)) {
1669                         dst_release(dst);
1670                         goto put_states;
1671                 }
1672
1673                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1674                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1675                                                         xfrm_af2proto(family));
1676                         if (!inner_mode) {
1677                                 err = -EAFNOSUPPORT;
1678                                 dst_release(dst);
1679                                 goto put_states;
1680                         }
1681                 } else
1682                         inner_mode = xfrm[i]->inner_mode;
1683
1684                 if (!dst_prev)
1685                         dst0 = dst1;
1686                 else {
1687                         dst_prev->child = dst_clone(dst1);
1688                         dst1->flags |= DST_NOHASH;
1689                 }
1690
1691                 xdst->route = dst;
1692                 dst_copy_metrics(dst1, dst);
1693
1694                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1695                         family = xfrm[i]->props.family;
1696                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1697                                               family);
1698                         err = PTR_ERR(dst);
1699                         if (IS_ERR(dst))
1700                                 goto put_states;
1701                 } else
1702                         dst_hold(dst);
1703
1704                 dst1->xfrm = xfrm[i];
1705                 xdst->xfrm_genid = xfrm[i]->genid;
1706
1707                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1708                 dst1->flags |= DST_HOST;
1709                 dst1->lastuse = now;
1710
1711                 dst1->input = dst_discard;
1712                 dst1->output = inner_mode->afinfo->output;
1713
1714                 dst1->next = dst_prev;
1715                 dst_prev = dst1;
1716
1717                 header_len += xfrm[i]->props.header_len;
1718                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1719                         nfheader_len += xfrm[i]->props.header_len;
1720                 trailer_len += xfrm[i]->props.trailer_len;
1721         }
1722
1723         dst_prev->child = dst;
1724         dst0->path = dst;
1725
1726         err = -ENODEV;
1727         dev = dst->dev;
1728         if (!dev)
1729                 goto free_dst;
1730
1731         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1732         xfrm_init_pmtu(dst_prev);
1733
1734         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1735                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1736
1737                 err = xfrm_fill_dst(xdst, dev, fl);
1738                 if (err)
1739                         goto free_dst;
1740
1741                 dst_prev->header_len = header_len;
1742                 dst_prev->trailer_len = trailer_len;
1743                 header_len -= xdst->u.dst.xfrm->props.header_len;
1744                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1745         }
1746
1747 out:
1748         return dst0;
1749
1750 put_states:
1751         for (; i < nx; i++)
1752                 xfrm_state_put(xfrm[i]);
1753 free_dst:
1754         if (dst0)
1755                 dst_free(dst0);
1756         dst0 = ERR_PTR(err);
1757         goto out;
1758 }
1759
1760 #ifdef CONFIG_XFRM_SUB_POLICY
1761 static int xfrm_dst_alloc_copy(void **target, const void *src, int size)
1762 {
1763         if (!*target) {
1764                 *target = kmalloc(size, GFP_ATOMIC);
1765                 if (!*target)
1766                         return -ENOMEM;
1767         }
1768
1769         memcpy(*target, src, size);
1770         return 0;
1771 }
1772 #endif
1773
1774 static int xfrm_dst_update_parent(struct dst_entry *dst,
1775                                   const struct xfrm_selector *sel)
1776 {
1777 #ifdef CONFIG_XFRM_SUB_POLICY
1778         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1779         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1780                                    sel, sizeof(*sel));
1781 #else
1782         return 0;
1783 #endif
1784 }
1785
1786 static int xfrm_dst_update_origin(struct dst_entry *dst,
1787                                   const struct flowi *fl)
1788 {
1789 #ifdef CONFIG_XFRM_SUB_POLICY
1790         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1791         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1792 #else
1793         return 0;
1794 #endif
1795 }
1796
1797 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1798                                 struct xfrm_policy **pols,
1799                                 int *num_pols, int *num_xfrms)
1800 {
1801         int i;
1802
1803         if (*num_pols == 0 || !pols[0]) {
1804                 *num_pols = 0;
1805                 *num_xfrms = 0;
1806                 return 0;
1807         }
1808         if (IS_ERR(pols[0]))
1809                 return PTR_ERR(pols[0]);
1810
1811         *num_xfrms = pols[0]->xfrm_nr;
1812
1813 #ifdef CONFIG_XFRM_SUB_POLICY
1814         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1815             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1816                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1817                                                     XFRM_POLICY_TYPE_MAIN,
1818                                                     fl, family,
1819                                                     XFRM_POLICY_OUT);
1820                 if (pols[1]) {
1821                         if (IS_ERR(pols[1])) {
1822                                 xfrm_pols_put(pols, *num_pols);
1823                                 return PTR_ERR(pols[1]);
1824                         }
1825                         (*num_pols)++;
1826                         (*num_xfrms) += pols[1]->xfrm_nr;
1827                 }
1828         }
1829 #endif
1830         for (i = 0; i < *num_pols; i++) {
1831                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1832                         *num_xfrms = -1;
1833                         break;
1834                 }
1835         }
1836
1837         return 0;
1838
1839 }
1840
1841 static struct xfrm_dst *
1842 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1843                                const struct flowi *fl, u16 family,
1844                                struct dst_entry *dst_orig)
1845 {
1846         struct net *net = xp_net(pols[0]);
1847         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1848         struct dst_entry *dst;
1849         struct xfrm_dst *xdst;
1850         int err;
1851
1852         /* Try to instantiate a bundle */
1853         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1854         if (err <= 0) {
1855                 if (err != 0 && err != -EAGAIN)
1856                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1857                 return ERR_PTR(err);
1858         }
1859
1860         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1861         if (IS_ERR(dst)) {
1862                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1863                 return ERR_CAST(dst);
1864         }
1865
1866         xdst = (struct xfrm_dst *)dst;
1867         xdst->num_xfrms = err;
1868         if (num_pols > 1)
1869                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1870         else
1871                 err = xfrm_dst_update_origin(dst, fl);
1872         if (unlikely(err)) {
1873                 dst_free(dst);
1874                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1875                 return ERR_PTR(err);
1876         }
1877
1878         xdst->num_pols = num_pols;
1879         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
1880         xdst->policy_genid = atomic_read(&pols[0]->genid);
1881
1882         return xdst;
1883 }
1884
1885 static void xfrm_policy_queue_process(unsigned long arg)
1886 {
1887         struct sk_buff *skb;
1888         struct sock *sk;
1889         struct dst_entry *dst;
1890         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1891         struct xfrm_policy_queue *pq = &pol->polq;
1892         struct flowi fl;
1893         struct sk_buff_head list;
1894
1895         spin_lock(&pq->hold_queue.lock);
1896         skb = skb_peek(&pq->hold_queue);
1897         if (!skb) {
1898                 spin_unlock(&pq->hold_queue.lock);
1899                 goto out;
1900         }
1901         dst = skb_dst(skb);
1902         sk = skb->sk;
1903         xfrm_decode_session(skb, &fl, dst->ops->family);
1904         spin_unlock(&pq->hold_queue.lock);
1905
1906         dst_hold(dst->path);
1907         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1908                           sk, 0);
1909         if (IS_ERR(dst))
1910                 goto purge_queue;
1911
1912         if (dst->flags & DST_XFRM_QUEUE) {
1913                 dst_release(dst);
1914
1915                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1916                         goto purge_queue;
1917
1918                 pq->timeout = pq->timeout << 1;
1919                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1920                         xfrm_pol_hold(pol);
1921         goto out;
1922         }
1923
1924         dst_release(dst);
1925
1926         __skb_queue_head_init(&list);
1927
1928         spin_lock(&pq->hold_queue.lock);
1929         pq->timeout = 0;
1930         skb_queue_splice_init(&pq->hold_queue, &list);
1931         spin_unlock(&pq->hold_queue.lock);
1932
1933         while (!skb_queue_empty(&list)) {
1934                 skb = __skb_dequeue(&list);
1935
1936                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1937                 dst_hold(skb_dst(skb)->path);
1938                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1939                                   &fl, skb->sk, 0);
1940                 if (IS_ERR(dst)) {
1941                         kfree_skb(skb);
1942                         continue;
1943                 }
1944
1945                 nf_reset(skb);
1946                 skb_dst_drop(skb);
1947                 skb_dst_set(skb, dst);
1948
1949                 dst_output(skb);
1950         }
1951
1952 out:
1953         xfrm_pol_put(pol);
1954         return;
1955
1956 purge_queue:
1957         pq->timeout = 0;
1958         xfrm_queue_purge(&pq->hold_queue);
1959         xfrm_pol_put(pol);
1960 }
1961
1962 static int xdst_queue_output(struct sock *sk, struct sk_buff *skb)
1963 {
1964         unsigned long sched_next;
1965         struct dst_entry *dst = skb_dst(skb);
1966         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1967         struct xfrm_policy *pol = xdst->pols[0];
1968         struct xfrm_policy_queue *pq = &pol->polq;
1969
1970         if (unlikely(skb_fclone_busy(sk, skb))) {
1971                 kfree_skb(skb);
1972                 return 0;
1973         }
1974
1975         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1976                 kfree_skb(skb);
1977                 return -EAGAIN;
1978         }
1979
1980         skb_dst_force(skb);
1981
1982         spin_lock_bh(&pq->hold_queue.lock);
1983
1984         if (!pq->timeout)
1985                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1986
1987         sched_next = jiffies + pq->timeout;
1988
1989         if (del_timer(&pq->hold_timer)) {
1990                 if (time_before(pq->hold_timer.expires, sched_next))
1991                         sched_next = pq->hold_timer.expires;
1992                 xfrm_pol_put(pol);
1993         }
1994
1995         __skb_queue_tail(&pq->hold_queue, skb);
1996         if (!mod_timer(&pq->hold_timer, sched_next))
1997                 xfrm_pol_hold(pol);
1998
1999         spin_unlock_bh(&pq->hold_queue.lock);
2000
2001         return 0;
2002 }
2003
2004 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2005                                                  struct xfrm_flo *xflo,
2006                                                  const struct flowi *fl,
2007                                                  int num_xfrms,
2008                                                  u16 family)
2009 {
2010         int err;
2011         struct net_device *dev;
2012         struct dst_entry *dst;
2013         struct dst_entry *dst1;
2014         struct xfrm_dst *xdst;
2015
2016         xdst = xfrm_alloc_dst(net, family);
2017         if (IS_ERR(xdst))
2018                 return xdst;
2019
2020         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2021             net->xfrm.sysctl_larval_drop ||
2022             num_xfrms <= 0)
2023                 return xdst;
2024
2025         dst = xflo->dst_orig;
2026         dst1 = &xdst->u.dst;
2027         dst_hold(dst);
2028         xdst->route = dst;
2029
2030         dst_copy_metrics(dst1, dst);
2031
2032         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2033         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2034         dst1->lastuse = jiffies;
2035
2036         dst1->input = dst_discard;
2037         dst1->output = xdst_queue_output;
2038
2039         dst_hold(dst);
2040         dst1->child = dst;
2041         dst1->path = dst;
2042
2043         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2044
2045         err = -ENODEV;
2046         dev = dst->dev;
2047         if (!dev)
2048                 goto free_dst;
2049
2050         err = xfrm_fill_dst(xdst, dev, fl);
2051         if (err)
2052                 goto free_dst;
2053
2054 out:
2055         return xdst;
2056
2057 free_dst:
2058         dst_release(dst1);
2059         xdst = ERR_PTR(err);
2060         goto out;
2061 }
2062
2063 static struct flow_cache_object *
2064 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
2065                    struct flow_cache_object *oldflo, void *ctx)
2066 {
2067         struct xfrm_flo *xflo = (struct xfrm_flo *)ctx;
2068         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2069         struct xfrm_dst *xdst, *new_xdst;
2070         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
2071
2072         /* Check if the policies from old bundle are usable */
2073         xdst = NULL;
2074         if (oldflo) {
2075                 xdst = container_of(oldflo, struct xfrm_dst, flo);
2076                 num_pols = xdst->num_pols;
2077                 num_xfrms = xdst->num_xfrms;
2078                 pol_dead = 0;
2079                 for (i = 0; i < num_pols; i++) {
2080                         pols[i] = xdst->pols[i];
2081                         pol_dead |= pols[i]->walk.dead;
2082                 }
2083                 if (pol_dead) {
2084                         dst_free(&xdst->u.dst);
2085                         xdst = NULL;
2086                         num_pols = 0;
2087                         num_xfrms = 0;
2088                         oldflo = NULL;
2089                 }
2090         }
2091
2092         /* Resolve policies to use if we couldn't get them from
2093          * previous cache entry */
2094         if (xdst == NULL) {
2095                 num_pols = 1;
2096                 pols[0] = __xfrm_policy_lookup(net, fl, family,
2097                                                flow_to_policy_dir(dir));
2098                 err = xfrm_expand_policies(fl, family, pols,
2099                                            &num_pols, &num_xfrms);
2100                 if (err < 0)
2101                         goto inc_error;
2102                 if (num_pols == 0)
2103                         return NULL;
2104                 if (num_xfrms <= 0)
2105                         goto make_dummy_bundle;
2106         }
2107
2108         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2109                                                   xflo->dst_orig);
2110         if (IS_ERR(new_xdst)) {
2111                 err = PTR_ERR(new_xdst);
2112                 if (err != -EAGAIN)
2113                         goto error;
2114                 if (oldflo == NULL)
2115                         goto make_dummy_bundle;
2116                 dst_hold(&xdst->u.dst);
2117                 return oldflo;
2118         } else if (new_xdst == NULL) {
2119                 num_xfrms = 0;
2120                 if (oldflo == NULL)
2121                         goto make_dummy_bundle;
2122                 xdst->num_xfrms = 0;
2123                 dst_hold(&xdst->u.dst);
2124                 return oldflo;
2125         }
2126
2127         /* Kill the previous bundle */
2128         if (xdst) {
2129                 /* The policies were stolen for newly generated bundle */
2130                 xdst->num_pols = 0;
2131                 dst_free(&xdst->u.dst);
2132         }
2133
2134         /* Flow cache does not have reference, it dst_free()'s,
2135          * but we do need to return one reference for original caller */
2136         dst_hold(&new_xdst->u.dst);
2137         return &new_xdst->flo;
2138
2139 make_dummy_bundle:
2140         /* We found policies, but there's no bundles to instantiate:
2141          * either because the policy blocks, has no transformations or
2142          * we could not build template (no xfrm_states).*/
2143         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2144         if (IS_ERR(xdst)) {
2145                 xfrm_pols_put(pols, num_pols);
2146                 return ERR_CAST(xdst);
2147         }
2148         xdst->num_pols = num_pols;
2149         xdst->num_xfrms = num_xfrms;
2150         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2151
2152         dst_hold(&xdst->u.dst);
2153         return &xdst->flo;
2154
2155 inc_error:
2156         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2157 error:
2158         if (xdst != NULL)
2159                 dst_free(&xdst->u.dst);
2160         else
2161                 xfrm_pols_put(pols, num_pols);
2162         return ERR_PTR(err);
2163 }
2164
2165 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2166                                         struct dst_entry *dst_orig)
2167 {
2168         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2169         struct dst_entry *ret;
2170
2171         if (!afinfo) {
2172                 dst_release(dst_orig);
2173                 return ERR_PTR(-EINVAL);
2174         } else {
2175                 ret = afinfo->blackhole_route(net, dst_orig);
2176         }
2177         xfrm_policy_put_afinfo(afinfo);
2178
2179         return ret;
2180 }
2181
2182 /* Main function: finds/creates a bundle for given flow.
2183  *
2184  * At the moment we eat a raw IP route. Mostly to speed up lookups
2185  * on interfaces with disabled IPsec.
2186  */
2187 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2188                               const struct flowi *fl,
2189                               struct sock *sk, int flags)
2190 {
2191         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2192         struct flow_cache_object *flo;
2193         struct xfrm_dst *xdst;
2194         struct dst_entry *dst, *route;
2195         u16 family = dst_orig->ops->family;
2196         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2197         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2198
2199         dst = NULL;
2200         xdst = NULL;
2201         route = NULL;
2202
2203         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2204                 num_pols = 1;
2205                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2206                 err = xfrm_expand_policies(fl, family, pols,
2207                                            &num_pols, &num_xfrms);
2208                 if (err < 0)
2209                         goto dropdst;
2210
2211                 if (num_pols) {
2212                         if (num_xfrms <= 0) {
2213                                 drop_pols = num_pols;
2214                                 goto no_transform;
2215                         }
2216
2217                         xdst = xfrm_resolve_and_create_bundle(
2218                                         pols, num_pols, fl,
2219                                         family, dst_orig);
2220                         if (IS_ERR(xdst)) {
2221                                 xfrm_pols_put(pols, num_pols);
2222                                 err = PTR_ERR(xdst);
2223                                 goto dropdst;
2224                         } else if (xdst == NULL) {
2225                                 num_xfrms = 0;
2226                                 drop_pols = num_pols;
2227                                 goto no_transform;
2228                         }
2229
2230                         dst_hold(&xdst->u.dst);
2231                         xdst->u.dst.flags |= DST_NOCACHE;
2232                         route = xdst->route;
2233                 }
2234         }
2235
2236         if (xdst == NULL) {
2237                 struct xfrm_flo xflo;
2238
2239                 xflo.dst_orig = dst_orig;
2240                 xflo.flags = flags;
2241
2242                 /* To accelerate a bit...  */
2243                 if ((dst_orig->flags & DST_NOXFRM) ||
2244                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2245                         goto nopol;
2246
2247                 flo = flow_cache_lookup(net, fl, family, dir,
2248                                         xfrm_bundle_lookup, &xflo);
2249                 if (flo == NULL)
2250                         goto nopol;
2251                 if (IS_ERR(flo)) {
2252                         err = PTR_ERR(flo);
2253                         goto dropdst;
2254                 }
2255                 xdst = container_of(flo, struct xfrm_dst, flo);
2256
2257                 num_pols = xdst->num_pols;
2258                 num_xfrms = xdst->num_xfrms;
2259                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
2260                 route = xdst->route;
2261         }
2262
2263         dst = &xdst->u.dst;
2264         if (route == NULL && num_xfrms > 0) {
2265                 /* The only case when xfrm_bundle_lookup() returns a
2266                  * bundle with null route, is when the template could
2267                  * not be resolved. It means policies are there, but
2268                  * bundle could not be created, since we don't yet
2269                  * have the xfrm_state's. We need to wait for KM to
2270                  * negotiate new SA's or bail out with error.*/
2271                 if (net->xfrm.sysctl_larval_drop) {
2272                         dst_release(dst);
2273                         xfrm_pols_put(pols, drop_pols);
2274                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2275
2276                         return ERR_PTR(-EREMOTE);
2277                 }
2278
2279                 err = -EAGAIN;
2280
2281                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2282                 goto error;
2283         }
2284
2285 no_transform:
2286         if (num_pols == 0)
2287                 goto nopol;
2288
2289         if ((flags & XFRM_LOOKUP_ICMP) &&
2290             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2291                 err = -ENOENT;
2292                 goto error;
2293         }
2294
2295         for (i = 0; i < num_pols; i++)
2296                 pols[i]->curlft.use_time = get_seconds();
2297
2298         if (num_xfrms < 0) {
2299                 /* Prohibit the flow */
2300                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2301                 err = -EPERM;
2302                 goto error;
2303         } else if (num_xfrms > 0) {
2304                 /* Flow transformed */
2305                 dst_release(dst_orig);
2306         } else {
2307                 /* Flow passes untransformed */
2308                 dst_release(dst);
2309                 dst = dst_orig;
2310         }
2311 ok:
2312         xfrm_pols_put(pols, drop_pols);
2313         if (dst && dst->xfrm &&
2314             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2315                 dst->flags |= DST_XFRM_TUNNEL;
2316         return dst;
2317
2318 nopol:
2319         if (!(flags & XFRM_LOOKUP_ICMP)) {
2320                 dst = dst_orig;
2321                 goto ok;
2322         }
2323         err = -ENOENT;
2324 error:
2325         dst_release(dst);
2326 dropdst:
2327         dst_release(dst_orig);
2328         xfrm_pols_put(pols, drop_pols);
2329         return ERR_PTR(err);
2330 }
2331 EXPORT_SYMBOL(xfrm_lookup);
2332
2333 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
2334  * Otherwise we may send out blackholed packets.
2335  */
2336 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
2337                                     const struct flowi *fl,
2338                                     struct sock *sk, int flags)
2339 {
2340         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
2341                                             flags | XFRM_LOOKUP_QUEUE);
2342
2343         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
2344                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
2345
2346         return dst;
2347 }
2348 EXPORT_SYMBOL(xfrm_lookup_route);
2349
2350 static inline int
2351 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2352 {
2353         struct xfrm_state *x;
2354
2355         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2356                 return 0;
2357         x = skb->sp->xvec[idx];
2358         if (!x->type->reject)
2359                 return 0;
2360         return x->type->reject(x, skb, fl);
2361 }
2362
2363 /* When skb is transformed back to its "native" form, we have to
2364  * check policy restrictions. At the moment we make this in maximally
2365  * stupid way. Shame on me. :-) Of course, connected sockets must
2366  * have policy cached at them.
2367  */
2368
2369 static inline int
2370 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2371               unsigned short family)
2372 {
2373         if (xfrm_state_kern(x))
2374                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2375         return  x->id.proto == tmpl->id.proto &&
2376                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2377                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2378                 x->props.mode == tmpl->mode &&
2379                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2380                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2381                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2382                   xfrm_state_addr_cmp(tmpl, x, family));
2383 }
2384
2385 /*
2386  * 0 or more than 0 is returned when validation is succeeded (either bypass
2387  * because of optional transport mode, or next index of the mathced secpath
2388  * state with the template.
2389  * -1 is returned when no matching template is found.
2390  * Otherwise "-2 - errored_index" is returned.
2391  */
2392 static inline int
2393 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2394                unsigned short family)
2395 {
2396         int idx = start;
2397
2398         if (tmpl->optional) {
2399                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2400                         return start;
2401         } else
2402                 start = -1;
2403         for (; idx < sp->len; idx++) {
2404                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2405                         return ++idx;
2406                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2407                         if (start == -1)
2408                                 start = -2-idx;
2409                         break;
2410                 }
2411         }
2412         return start;
2413 }
2414
2415 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2416                           unsigned int family, int reverse)
2417 {
2418         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2419         int err;
2420
2421         if (unlikely(afinfo == NULL))
2422                 return -EAFNOSUPPORT;
2423
2424         afinfo->decode_session(skb, fl, reverse);
2425         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2426         xfrm_policy_put_afinfo(afinfo);
2427         return err;
2428 }
2429 EXPORT_SYMBOL(__xfrm_decode_session);
2430
2431 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2432 {
2433         for (; k < sp->len; k++) {
2434                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2435                         *idxp = k;
2436                         return 1;
2437                 }
2438         }
2439
2440         return 0;
2441 }
2442
2443 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2444                         unsigned short family)
2445 {
2446         struct net *net = dev_net(skb->dev);
2447         struct xfrm_policy *pol;
2448         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2449         int npols = 0;
2450         int xfrm_nr;
2451         int pi;
2452         int reverse;
2453         struct flowi fl;
2454         u8 fl_dir;
2455         int xerr_idx = -1;
2456
2457         reverse = dir & ~XFRM_POLICY_MASK;
2458         dir &= XFRM_POLICY_MASK;
2459         fl_dir = policy_to_flow_dir(dir);
2460
2461         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2462                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2463                 return 0;
2464         }
2465
2466         nf_nat_decode_session(skb, &fl, family);
2467
2468         /* First, check used SA against their selectors. */
2469         if (skb->sp) {
2470                 int i;
2471
2472                 for (i = skb->sp->len-1; i >= 0; i--) {
2473                         struct xfrm_state *x = skb->sp->xvec[i];
2474                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2475                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2476                                 return 0;
2477                         }
2478                 }
2479         }
2480
2481         pol = NULL;
2482         if (sk && sk->sk_policy[dir]) {
2483                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2484                 if (IS_ERR(pol)) {
2485                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2486                         return 0;
2487                 }
2488         }
2489
2490         if (!pol) {
2491                 struct flow_cache_object *flo;
2492
2493                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2494                                         xfrm_policy_lookup, NULL);
2495                 if (IS_ERR_OR_NULL(flo))
2496                         pol = ERR_CAST(flo);
2497                 else
2498                         pol = container_of(flo, struct xfrm_policy, flo);
2499         }
2500
2501         if (IS_ERR(pol)) {
2502                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2503                 return 0;
2504         }
2505
2506         if (!pol) {
2507                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2508                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2509                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2510                         return 0;
2511                 }
2512                 return 1;
2513         }
2514
2515         pol->curlft.use_time = get_seconds();
2516
2517         pols[0] = pol;
2518         npols++;
2519 #ifdef CONFIG_XFRM_SUB_POLICY
2520         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2521                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2522                                                     &fl, family,
2523                                                     XFRM_POLICY_IN);
2524                 if (pols[1]) {
2525                         if (IS_ERR(pols[1])) {
2526                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2527                                 return 0;
2528                         }
2529                         pols[1]->curlft.use_time = get_seconds();
2530                         npols++;
2531                 }
2532         }
2533 #endif
2534
2535         if (pol->action == XFRM_POLICY_ALLOW) {
2536                 struct sec_path *sp;
2537                 static struct sec_path dummy;
2538                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2539                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2540                 struct xfrm_tmpl **tpp = tp;
2541                 int ti = 0;
2542                 int i, k;
2543
2544                 if ((sp = skb->sp) == NULL)
2545                         sp = &dummy;
2546
2547                 for (pi = 0; pi < npols; pi++) {
2548                         if (pols[pi] != pol &&
2549                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2550                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2551                                 goto reject;
2552                         }
2553                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2554                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2555                                 goto reject_error;
2556                         }
2557                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2558                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2559                 }
2560                 xfrm_nr = ti;
2561                 if (npols > 1) {
2562                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2563                         tpp = stp;
2564                 }
2565
2566                 /* For each tunnel xfrm, find the first matching tmpl.
2567                  * For each tmpl before that, find corresponding xfrm.
2568                  * Order is _important_. Later we will implement
2569                  * some barriers, but at the moment barriers
2570                  * are implied between each two transformations.
2571                  */
2572                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2573                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2574                         if (k < 0) {
2575                                 if (k < -1)
2576                                         /* "-2 - errored_index" returned */
2577                                         xerr_idx = -(2+k);
2578                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2579                                 goto reject;
2580                         }
2581                 }
2582
2583                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2584                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2585                         goto reject;
2586                 }
2587
2588                 xfrm_pols_put(pols, npols);
2589                 return 1;
2590         }
2591         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2592
2593 reject:
2594         xfrm_secpath_reject(xerr_idx, skb, &fl);
2595 reject_error:
2596         xfrm_pols_put(pols, npols);
2597         return 0;
2598 }
2599 EXPORT_SYMBOL(__xfrm_policy_check);
2600
2601 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2602 {
2603         struct net *net = dev_net(skb->dev);
2604         struct flowi fl;
2605         struct dst_entry *dst;
2606         int res = 1;
2607
2608         if (xfrm_decode_session(skb, &fl, family) < 0) {
2609                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2610                 return 0;
2611         }
2612
2613         skb_dst_force(skb);
2614
2615         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
2616         if (IS_ERR(dst)) {
2617                 res = 0;
2618                 dst = NULL;
2619         }
2620         skb_dst_set(skb, dst);
2621         return res;
2622 }
2623 EXPORT_SYMBOL(__xfrm_route_forward);
2624
2625 /* Optimize later using cookies and generation ids. */
2626
2627 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2628 {
2629         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2630          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2631          * get validated by dst_ops->check on every use.  We do this
2632          * because when a normal route referenced by an XFRM dst is
2633          * obsoleted we do not go looking around for all parent
2634          * referencing XFRM dsts so that we can invalidate them.  It
2635          * is just too much work.  Instead we make the checks here on
2636          * every use.  For example:
2637          *
2638          *      XFRM dst A --> IPv4 dst X
2639          *
2640          * X is the "xdst->route" of A (X is also the "dst->path" of A
2641          * in this example).  If X is marked obsolete, "A" will not
2642          * notice.  That's what we are validating here via the
2643          * stale_bundle() check.
2644          *
2645          * When a policy's bundle is pruned, we dst_free() the XFRM
2646          * dst which causes it's ->obsolete field to be set to
2647          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2648          * this, we want to force a new route lookup.
2649          */
2650         if (dst->obsolete < 0 && !stale_bundle(dst))
2651                 return dst;
2652
2653         return NULL;
2654 }
2655
2656 static int stale_bundle(struct dst_entry *dst)
2657 {
2658         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2659 }
2660
2661 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2662 {
2663         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2664                 dst->dev = dev_net(dev)->loopback_dev;
2665                 dev_hold(dst->dev);
2666                 dev_put(dev);
2667         }
2668 }
2669 EXPORT_SYMBOL(xfrm_dst_ifdown);
2670
2671 static void xfrm_link_failure(struct sk_buff *skb)
2672 {
2673         /* Impossible. Such dst must be popped before reaches point of failure. */
2674 }
2675
2676 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2677 {
2678         if (dst) {
2679                 if (dst->obsolete) {
2680                         dst_release(dst);
2681                         dst = NULL;
2682                 }
2683         }
2684         return dst;
2685 }
2686
2687 void xfrm_garbage_collect(struct net *net)
2688 {
2689         flow_cache_flush(net);
2690 }
2691 EXPORT_SYMBOL(xfrm_garbage_collect);
2692
2693 static void xfrm_garbage_collect_deferred(struct net *net)
2694 {
2695         flow_cache_flush_deferred(net);
2696 }
2697
2698 static void xfrm_init_pmtu(struct dst_entry *dst)
2699 {
2700         do {
2701                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2702                 u32 pmtu, route_mtu_cached;
2703
2704                 pmtu = dst_mtu(dst->child);
2705                 xdst->child_mtu_cached = pmtu;
2706
2707                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2708
2709                 route_mtu_cached = dst_mtu(xdst->route);
2710                 xdst->route_mtu_cached = route_mtu_cached;
2711
2712                 if (pmtu > route_mtu_cached)
2713                         pmtu = route_mtu_cached;
2714
2715                 dst_metric_set(dst, RTAX_MTU, pmtu);
2716         } while ((dst = dst->next));
2717 }
2718
2719 /* Check that the bundle accepts the flow and its components are
2720  * still valid.
2721  */
2722
2723 static int xfrm_bundle_ok(struct xfrm_dst *first)
2724 {
2725         struct dst_entry *dst = &first->u.dst;
2726         struct xfrm_dst *last;
2727         u32 mtu;
2728
2729         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2730             (dst->dev && !netif_running(dst->dev)))
2731                 return 0;
2732
2733         if (dst->flags & DST_XFRM_QUEUE)
2734                 return 1;
2735
2736         last = NULL;
2737
2738         do {
2739                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2740
2741                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2742                         return 0;
2743                 if (xdst->xfrm_genid != dst->xfrm->genid)
2744                         return 0;
2745                 if (xdst->num_pols > 0 &&
2746                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2747                         return 0;
2748
2749                 mtu = dst_mtu(dst->child);
2750                 if (xdst->child_mtu_cached != mtu) {
2751                         last = xdst;
2752                         xdst->child_mtu_cached = mtu;
2753                 }
2754
2755                 if (!dst_check(xdst->route, xdst->route_cookie))
2756                         return 0;
2757                 mtu = dst_mtu(xdst->route);
2758                 if (xdst->route_mtu_cached != mtu) {
2759                         last = xdst;
2760                         xdst->route_mtu_cached = mtu;
2761                 }
2762
2763                 dst = dst->child;
2764         } while (dst->xfrm);
2765
2766         if (likely(!last))
2767                 return 1;
2768
2769         mtu = last->child_mtu_cached;
2770         for (;;) {
2771                 dst = &last->u.dst;
2772
2773                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2774                 if (mtu > last->route_mtu_cached)
2775                         mtu = last->route_mtu_cached;
2776                 dst_metric_set(dst, RTAX_MTU, mtu);
2777
2778                 if (last == first)
2779                         break;
2780
2781                 last = (struct xfrm_dst *)last->u.dst.next;
2782                 last->child_mtu_cached = mtu;
2783         }
2784
2785         return 1;
2786 }
2787
2788 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2789 {
2790         return dst_metric_advmss(dst->path);
2791 }
2792
2793 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2794 {
2795         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2796
2797         return mtu ? : dst_mtu(dst->path);
2798 }
2799
2800 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2801                                            struct sk_buff *skb,
2802                                            const void *daddr)
2803 {
2804         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2805 }
2806
2807 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2808 {
2809         struct net *net;
2810         int err = 0;
2811         if (unlikely(afinfo == NULL))
2812                 return -EINVAL;
2813         if (unlikely(afinfo->family >= NPROTO))
2814                 return -EAFNOSUPPORT;
2815         spin_lock(&xfrm_policy_afinfo_lock);
2816         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2817                 err = -ENOBUFS;
2818         else {
2819                 struct dst_ops *dst_ops = afinfo->dst_ops;
2820                 if (likely(dst_ops->kmem_cachep == NULL))
2821                         dst_ops->kmem_cachep = xfrm_dst_cache;
2822                 if (likely(dst_ops->check == NULL))
2823                         dst_ops->check = xfrm_dst_check;
2824                 if (likely(dst_ops->default_advmss == NULL))
2825                         dst_ops->default_advmss = xfrm_default_advmss;
2826                 if (likely(dst_ops->mtu == NULL))
2827                         dst_ops->mtu = xfrm_mtu;
2828                 if (likely(dst_ops->negative_advice == NULL))
2829                         dst_ops->negative_advice = xfrm_negative_advice;
2830                 if (likely(dst_ops->link_failure == NULL))
2831                         dst_ops->link_failure = xfrm_link_failure;
2832                 if (likely(dst_ops->neigh_lookup == NULL))
2833                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2834                 if (likely(afinfo->garbage_collect == NULL))
2835                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2836                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2837         }
2838         spin_unlock(&xfrm_policy_afinfo_lock);
2839
2840         rtnl_lock();
2841         for_each_net(net) {
2842                 struct dst_ops *xfrm_dst_ops;
2843
2844                 switch (afinfo->family) {
2845                 case AF_INET:
2846                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2847                         break;
2848 #if IS_ENABLED(CONFIG_IPV6)
2849                 case AF_INET6:
2850                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2851                         break;
2852 #endif
2853                 default:
2854                         BUG();
2855                 }
2856                 *xfrm_dst_ops = *afinfo->dst_ops;
2857         }
2858         rtnl_unlock();
2859
2860         return err;
2861 }
2862 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2863
2864 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2865 {
2866         int err = 0;
2867         if (unlikely(afinfo == NULL))
2868                 return -EINVAL;
2869         if (unlikely(afinfo->family >= NPROTO))
2870                 return -EAFNOSUPPORT;
2871         spin_lock(&xfrm_policy_afinfo_lock);
2872         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2873                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2874                         err = -EINVAL;
2875                 else
2876                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2877                                          NULL);
2878         }
2879         spin_unlock(&xfrm_policy_afinfo_lock);
2880         if (!err) {
2881                 struct dst_ops *dst_ops = afinfo->dst_ops;
2882
2883                 synchronize_rcu();
2884
2885                 dst_ops->kmem_cachep = NULL;
2886                 dst_ops->check = NULL;
2887                 dst_ops->negative_advice = NULL;
2888                 dst_ops->link_failure = NULL;
2889                 afinfo->garbage_collect = NULL;
2890         }
2891         return err;
2892 }
2893 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2894
2895 static void __net_init xfrm_dst_ops_init(struct net *net)
2896 {
2897         struct xfrm_policy_afinfo *afinfo;
2898
2899         rcu_read_lock();
2900         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2901         if (afinfo)
2902                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2903 #if IS_ENABLED(CONFIG_IPV6)
2904         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2905         if (afinfo)
2906                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2907 #endif
2908         rcu_read_unlock();
2909 }
2910
2911 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2912 {
2913         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2914
2915         switch (event) {
2916         case NETDEV_DOWN:
2917                 xfrm_garbage_collect(dev_net(dev));
2918         }
2919         return NOTIFY_DONE;
2920 }
2921
2922 static struct notifier_block xfrm_dev_notifier = {
2923         .notifier_call  = xfrm_dev_event,
2924 };
2925
2926 #ifdef CONFIG_XFRM_STATISTICS
2927 static int __net_init xfrm_statistics_init(struct net *net)
2928 {
2929         int rv;
2930         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
2931         if (!net->mib.xfrm_statistics)
2932                 return -ENOMEM;
2933         rv = xfrm_proc_init(net);
2934         if (rv < 0)
2935                 free_percpu(net->mib.xfrm_statistics);
2936         return rv;
2937 }
2938
2939 static void xfrm_statistics_fini(struct net *net)
2940 {
2941         xfrm_proc_fini(net);
2942         free_percpu(net->mib.xfrm_statistics);
2943 }
2944 #else
2945 static int __net_init xfrm_statistics_init(struct net *net)
2946 {
2947         return 0;
2948 }
2949
2950 static void xfrm_statistics_fini(struct net *net)
2951 {
2952 }
2953 #endif
2954
2955 static int __net_init xfrm_policy_init(struct net *net)
2956 {
2957         unsigned int hmask, sz;
2958         int dir;
2959
2960         if (net_eq(net, &init_net))
2961                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2962                                            sizeof(struct xfrm_dst),
2963                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2964                                            NULL);
2965
2966         hmask = 8 - 1;
2967         sz = (hmask+1) * sizeof(struct hlist_head);
2968
2969         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2970         if (!net->xfrm.policy_byidx)
2971                 goto out_byidx;
2972         net->xfrm.policy_idx_hmask = hmask;
2973
2974         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
2975                 struct xfrm_policy_hash *htab;
2976
2977                 net->xfrm.policy_count[dir] = 0;
2978                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
2979                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2980
2981                 htab = &net->xfrm.policy_bydst[dir];
2982                 htab->table = xfrm_hash_alloc(sz);
2983                 if (!htab->table)
2984                         goto out_bydst;
2985                 htab->hmask = hmask;
2986                 htab->dbits4 = 32;
2987                 htab->sbits4 = 32;
2988                 htab->dbits6 = 128;
2989                 htab->sbits6 = 128;
2990         }
2991         net->xfrm.policy_hthresh.lbits4 = 32;
2992         net->xfrm.policy_hthresh.rbits4 = 32;
2993         net->xfrm.policy_hthresh.lbits6 = 128;
2994         net->xfrm.policy_hthresh.rbits6 = 128;
2995
2996         seqlock_init(&net->xfrm.policy_hthresh.lock);
2997
2998         INIT_LIST_HEAD(&net->xfrm.policy_all);
2999         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
3000         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
3001         if (net_eq(net, &init_net))
3002                 register_netdevice_notifier(&xfrm_dev_notifier);
3003         return 0;
3004
3005 out_bydst:
3006         for (dir--; dir >= 0; dir--) {
3007                 struct xfrm_policy_hash *htab;
3008
3009                 htab = &net->xfrm.policy_bydst[dir];
3010                 xfrm_hash_free(htab->table, sz);
3011         }
3012         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3013 out_byidx:
3014         return -ENOMEM;
3015 }
3016
3017 static void xfrm_policy_fini(struct net *net)
3018 {
3019         unsigned int sz;
3020         int dir;
3021
3022         flush_work(&net->xfrm.policy_hash_work);
3023 #ifdef CONFIG_XFRM_SUB_POLICY
3024         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3025 #endif
3026         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3027
3028         WARN_ON(!list_empty(&net->xfrm.policy_all));
3029
3030         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3031                 struct xfrm_policy_hash *htab;
3032
3033                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3034
3035                 htab = &net->xfrm.policy_bydst[dir];
3036                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3037                 WARN_ON(!hlist_empty(htab->table));
3038                 xfrm_hash_free(htab->table, sz);
3039         }
3040
3041         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3042         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3043         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3044 }
3045
3046 static int __net_init xfrm_net_init(struct net *net)
3047 {
3048         int rv;
3049
3050         rv = xfrm_statistics_init(net);
3051         if (rv < 0)
3052                 goto out_statistics;
3053         rv = xfrm_state_init(net);
3054         if (rv < 0)
3055                 goto out_state;
3056         rv = xfrm_policy_init(net);
3057         if (rv < 0)
3058                 goto out_policy;
3059         xfrm_dst_ops_init(net);
3060         rv = xfrm_sysctl_init(net);
3061         if (rv < 0)
3062                 goto out_sysctl;
3063         rv = flow_cache_init(net);
3064         if (rv < 0)
3065                 goto out;
3066
3067         /* Initialize the per-net locks here */
3068         spin_lock_init(&net->xfrm.xfrm_state_lock);
3069         rwlock_init(&net->xfrm.xfrm_policy_lock);
3070         mutex_init(&net->xfrm.xfrm_cfg_mutex);
3071
3072         return 0;
3073
3074 out:
3075         xfrm_sysctl_fini(net);
3076 out_sysctl:
3077         xfrm_policy_fini(net);
3078 out_policy:
3079         xfrm_state_fini(net);
3080 out_state:
3081         xfrm_statistics_fini(net);
3082 out_statistics:
3083         return rv;
3084 }
3085
3086 static void __net_exit xfrm_net_exit(struct net *net)
3087 {
3088         flow_cache_fini(net);
3089         xfrm_sysctl_fini(net);
3090         xfrm_policy_fini(net);
3091         xfrm_state_fini(net);
3092         xfrm_statistics_fini(net);
3093 }
3094
3095 static struct pernet_operations __net_initdata xfrm_net_ops = {
3096         .init = xfrm_net_init,
3097         .exit = xfrm_net_exit,
3098 };
3099
3100 void __init xfrm_init(void)
3101 {
3102         register_pernet_subsys(&xfrm_net_ops);
3103         xfrm_input_init();
3104 }
3105
3106 #ifdef CONFIG_AUDITSYSCALL
3107 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3108                                          struct audit_buffer *audit_buf)
3109 {
3110         struct xfrm_sec_ctx *ctx = xp->security;
3111         struct xfrm_selector *sel = &xp->selector;
3112
3113         if (ctx)
3114                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3115                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3116
3117         switch (sel->family) {
3118         case AF_INET:
3119                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3120                 if (sel->prefixlen_s != 32)
3121                         audit_log_format(audit_buf, " src_prefixlen=%d",
3122                                          sel->prefixlen_s);
3123                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3124                 if (sel->prefixlen_d != 32)
3125                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3126                                          sel->prefixlen_d);
3127                 break;
3128         case AF_INET6:
3129                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3130                 if (sel->prefixlen_s != 128)
3131                         audit_log_format(audit_buf, " src_prefixlen=%d",
3132                                          sel->prefixlen_s);
3133                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3134                 if (sel->prefixlen_d != 128)
3135                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3136                                          sel->prefixlen_d);
3137                 break;
3138         }
3139 }
3140
3141 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3142 {
3143         struct audit_buffer *audit_buf;
3144
3145         audit_buf = xfrm_audit_start("SPD-add");
3146         if (audit_buf == NULL)
3147                 return;
3148         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3149         audit_log_format(audit_buf, " res=%u", result);
3150         xfrm_audit_common_policyinfo(xp, audit_buf);
3151         audit_log_end(audit_buf);
3152 }
3153 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3154
3155 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3156                               bool task_valid)
3157 {
3158         struct audit_buffer *audit_buf;
3159
3160         audit_buf = xfrm_audit_start("SPD-delete");
3161         if (audit_buf == NULL)
3162                 return;
3163         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3164         audit_log_format(audit_buf, " res=%u", result);
3165         xfrm_audit_common_policyinfo(xp, audit_buf);
3166         audit_log_end(audit_buf);
3167 }
3168 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3169 #endif
3170
3171 #ifdef CONFIG_XFRM_MIGRATE
3172 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3173                                         const struct xfrm_selector *sel_tgt)
3174 {
3175         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3176                 if (sel_tgt->family == sel_cmp->family &&
3177                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3178                                     sel_cmp->family) &&
3179                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3180                                     sel_cmp->family) &&
3181                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3182                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3183                         return true;
3184                 }
3185         } else {
3186                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3187                         return true;
3188                 }
3189         }
3190         return false;
3191 }
3192
3193 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3194                                                     u8 dir, u8 type, struct net *net)
3195 {
3196         struct xfrm_policy *pol, *ret = NULL;
3197         struct hlist_head *chain;
3198         u32 priority = ~0U;
3199
3200         read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3201         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3202         hlist_for_each_entry(pol, chain, bydst) {
3203                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3204                     pol->type == type) {
3205                         ret = pol;
3206                         priority = ret->priority;
3207                         break;
3208                 }
3209         }
3210         chain = &net->xfrm.policy_inexact[dir];
3211         hlist_for_each_entry(pol, chain, bydst) {
3212                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3213                     pol->type == type &&
3214                     pol->priority < priority) {
3215                         ret = pol;
3216                         break;
3217                 }
3218         }
3219
3220         if (ret)
3221                 xfrm_pol_hold(ret);
3222
3223         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3224
3225         return ret;
3226 }
3227
3228 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3229 {
3230         int match = 0;
3231
3232         if (t->mode == m->mode && t->id.proto == m->proto &&
3233             (m->reqid == 0 || t->reqid == m->reqid)) {
3234                 switch (t->mode) {
3235                 case XFRM_MODE_TUNNEL:
3236                 case XFRM_MODE_BEET:
3237                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3238                                             m->old_family) &&
3239                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3240                                             m->old_family)) {
3241                                 match = 1;
3242                         }
3243                         break;
3244                 case XFRM_MODE_TRANSPORT:
3245                         /* in case of transport mode, template does not store
3246                            any IP addresses, hence we just compare mode and
3247                            protocol */
3248                         match = 1;
3249                         break;
3250                 default:
3251                         break;
3252                 }
3253         }
3254         return match;
3255 }
3256
3257 /* update endpoint address(es) of template(s) */
3258 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3259                                struct xfrm_migrate *m, int num_migrate)
3260 {
3261         struct xfrm_migrate *mp;
3262         int i, j, n = 0;
3263
3264         write_lock_bh(&pol->lock);
3265         if (unlikely(pol->walk.dead)) {
3266                 /* target policy has been deleted */
3267                 write_unlock_bh(&pol->lock);
3268                 return -ENOENT;
3269         }
3270
3271         for (i = 0; i < pol->xfrm_nr; i++) {
3272                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3273                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3274                                 continue;
3275                         n++;
3276                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3277                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3278                                 continue;
3279                         /* update endpoints */
3280                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3281                                sizeof(pol->xfrm_vec[i].id.daddr));
3282                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3283                                sizeof(pol->xfrm_vec[i].saddr));
3284                         pol->xfrm_vec[i].encap_family = mp->new_family;
3285                         /* flush bundles */
3286                         atomic_inc(&pol->genid);
3287                 }
3288         }
3289
3290         write_unlock_bh(&pol->lock);
3291
3292         if (!n)
3293                 return -ENODATA;
3294
3295         return 0;
3296 }
3297
3298 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3299 {
3300         int i, j;
3301
3302         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3303                 return -EINVAL;
3304
3305         for (i = 0; i < num_migrate; i++) {
3306                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3307                                     m[i].old_family) &&
3308                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3309                                     m[i].old_family))
3310                         return -EINVAL;
3311                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3312                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3313                         return -EINVAL;
3314
3315                 /* check if there is any duplicated entry */
3316                 for (j = i + 1; j < num_migrate; j++) {
3317                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3318                                     sizeof(m[i].old_daddr)) &&
3319                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3320                                     sizeof(m[i].old_saddr)) &&
3321                             m[i].proto == m[j].proto &&
3322                             m[i].mode == m[j].mode &&
3323                             m[i].reqid == m[j].reqid &&
3324                             m[i].old_family == m[j].old_family)
3325                                 return -EINVAL;
3326                 }
3327         }
3328
3329         return 0;
3330 }
3331
3332 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3333                  struct xfrm_migrate *m, int num_migrate,
3334                  struct xfrm_kmaddress *k, struct net *net)
3335 {
3336         int i, err, nx_cur = 0, nx_new = 0;
3337         struct xfrm_policy *pol = NULL;
3338         struct xfrm_state *x, *xc;
3339         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3340         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3341         struct xfrm_migrate *mp;
3342
3343         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3344                 goto out;
3345
3346         /* Stage 1 - find policy */
3347         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3348                 err = -ENOENT;
3349                 goto out;
3350         }
3351
3352         /* Stage 2 - find and update state(s) */
3353         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3354                 if ((x = xfrm_migrate_state_find(mp, net))) {
3355                         x_cur[nx_cur] = x;
3356                         nx_cur++;
3357                         if ((xc = xfrm_state_migrate(x, mp))) {
3358                                 x_new[nx_new] = xc;
3359                                 nx_new++;
3360                         } else {
3361                                 err = -ENODATA;
3362                                 goto restore_state;
3363                         }
3364                 }
3365         }
3366
3367         /* Stage 3 - update policy */
3368         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3369                 goto restore_state;
3370
3371         /* Stage 4 - delete old state(s) */
3372         if (nx_cur) {
3373                 xfrm_states_put(x_cur, nx_cur);
3374                 xfrm_states_delete(x_cur, nx_cur);
3375         }
3376
3377         /* Stage 5 - announce */
3378         km_migrate(sel, dir, type, m, num_migrate, k);
3379
3380         xfrm_pol_put(pol);
3381
3382         return 0;
3383 out:
3384         return err;
3385
3386 restore_state:
3387         if (pol)
3388                 xfrm_pol_put(pol);
3389         if (nx_cur)
3390                 xfrm_states_put(x_cur, nx_cur);
3391         if (nx_new)
3392                 xfrm_states_delete(x_new, nx_new);
3393
3394         return err;
3395 }
3396 EXPORT_SYMBOL(xfrm_migrate);
3397 #endif