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