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/xfrm.h>
  30 #include <net/ip.h>
  31 #ifdef CONFIG_XFRM_STATISTICS
  32 #include <net/snmp.h>
  33 #endif
  34
  35 #include "xfrm_hash.h"
  36
  37 DEFINE_MUTEX(xfrm_cfg_mutex);
  38 EXPORT_SYMBOL(xfrm_cfg_mutex);
  39
  40 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
  41 static struct dst_entry *xfrm_policy_sk_bundles;
  42 static DEFINE_RWLOCK(xfrm_policy_lock);
  43
  44 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  45 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  46
  47 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  48
  49 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  50 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  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, int family);
  54
  55
  56 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  57                                                 int dir);
  58
  59 static inline int
  60 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  61 {
  62         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
  63                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
  64                 !((xfrm_flowi_dport(fl, &fl->u.ip4.uli) ^ sel->dport) & sel->dport_mask) &&
  65                 !((xfrm_flowi_sport(fl, &fl->u.ip4.uli) ^ sel->sport) & sel->sport_mask) &&
  66                 (fl->flowi_proto == sel->proto || !sel->proto) &&
  67                 (fl->flowi_oif == sel->ifindex || !sel->ifindex);
  68 }
  69
  70 static inline int
  71 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  72 {
  73         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
  74                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
  75                 !((xfrm_flowi_dport(fl, &fl->u.ip6.uli) ^ sel->dport) & sel->dport_mask) &&
  76                 !((xfrm_flowi_sport(fl, &fl->u.ip6.uli) ^ sel->sport) & sel->sport_mask) &&
  77                 (fl->flowi_proto == sel->proto || !sel->proto) &&
  78                 (fl->flowi_oif == sel->ifindex || !sel->ifindex);
  79 }
  80
  81 int xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  82                         unsigned short family)
  83 {
  84         switch (family) {
  85         case AF_INET:
  86                 return __xfrm4_selector_match(sel, fl);
  87         case AF_INET6:
  88                 return __xfrm6_selector_match(sel, fl);
  89         }
  90         return 0;
  91 }
  92
  93 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
  94                                                   const xfrm_address_t *saddr,
  95                                                   const xfrm_address_t *daddr,
  96                                                   int family)
  97 {
  98         struct xfrm_policy_afinfo *afinfo;
  99         struct dst_entry *dst;
 100
 101         afinfo = xfrm_policy_get_afinfo(family);
 102         if (unlikely(afinfo == NULL))
 103                 return ERR_PTR(-EAFNOSUPPORT);
 104
 105         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 106
 107         xfrm_policy_put_afinfo(afinfo);
 108
 109         return dst;
 110 }
 111
 112 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 113                                                 xfrm_address_t *prev_saddr,
 114                                                 xfrm_address_t *prev_daddr,
 115                                                 int family)
 116 {
 117         struct net *net = xs_net(x);
 118         xfrm_address_t *saddr = &x->props.saddr;
 119         xfrm_address_t *daddr = &x->id.daddr;
 120         struct dst_entry *dst;
 121
 122         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 123                 saddr = x->coaddr;
 124                 daddr = prev_daddr;
 125         }
 126         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 127                 saddr = prev_saddr;
 128                 daddr = x->coaddr;
 129         }
 130
 131         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 132
 133         if (!IS_ERR(dst)) {
 134                 if (prev_saddr != saddr)
 135                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 136                 if (prev_daddr != daddr)
 137                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 138         }
 139
 140         return dst;
 141 }
 142
 143 static inline unsigned long make_jiffies(long secs)
 144 {
 145         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 146                 return MAX_SCHEDULE_TIMEOUT-1;
 147         else
 148                 return secs*HZ;
 149 }
 150
 151 static void xfrm_policy_timer(unsigned long data)
 152 {
 153         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 154         unsigned long now = get_seconds();
 155         long next = LONG_MAX;
 156         int warn = 0;
 157         int dir;
 158
 159         read_lock(&xp->lock);
 160
 161         if (unlikely(xp->walk.dead))
 162                 goto out;
 163
 164         dir = xfrm_policy_id2dir(xp->index);
 165
 166         if (xp->lft.hard_add_expires_seconds) {
 167                 long tmo = xp->lft.hard_add_expires_seconds +
 168                         xp->curlft.add_time - now;
 169                 if (tmo <= 0)
 170                         goto expired;
 171                 if (tmo < next)
 172                         next = tmo;
 173         }
 174         if (xp->lft.hard_use_expires_seconds) {
 175                 long tmo = xp->lft.hard_use_expires_seconds +
 176                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 177                 if (tmo <= 0)
 178                         goto expired;
 179                 if (tmo < next)
 180                         next = tmo;
 181         }
 182         if (xp->lft.soft_add_expires_seconds) {
 183                 long tmo = xp->lft.soft_add_expires_seconds +
 184                         xp->curlft.add_time - now;
 185                 if (tmo <= 0) {
 186                         warn = 1;
 187                         tmo = XFRM_KM_TIMEOUT;
 188                 }
 189                 if (tmo < next)
 190                         next = tmo;
 191         }
 192         if (xp->lft.soft_use_expires_seconds) {
 193                 long tmo = xp->lft.soft_use_expires_seconds +
 194                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 195                 if (tmo <= 0) {
 196                         warn = 1;
 197                         tmo = XFRM_KM_TIMEOUT;
 198                 }
 199                 if (tmo < next)
 200                         next = tmo;
 201         }
 202
 203         if (warn)
 204                 km_policy_expired(xp, dir, 0, 0);
 205         if (next != LONG_MAX &&
 206             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 207                 xfrm_pol_hold(xp);
 208
 209 out:
 210         read_unlock(&xp->lock);
 211         xfrm_pol_put(xp);
 212         return;
 213
 214 expired:
 215         read_unlock(&xp->lock);
 216         if (!xfrm_policy_delete(xp, dir))
 217                 km_policy_expired(xp, dir, 1, 0);
 218         xfrm_pol_put(xp);
 219 }
 220
 221 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 222 {
 223         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 224
 225         if (unlikely(pol->walk.dead))
 226                 flo = NULL;
 227         else
 228                 xfrm_pol_hold(pol);
 229
 230         return flo;
 231 }
 232
 233 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 234 {
 235         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 236
 237         return !pol->walk.dead;
 238 }
 239
 240 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 241 {
 242         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 243 }
 244
 245 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 246         .get = xfrm_policy_flo_get,
 247         .check = xfrm_policy_flo_check,
 248         .delete = xfrm_policy_flo_delete,
 249 };
 250
 251 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 252  * SPD calls.
 253  */
 254
 255 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 256 {
 257         struct xfrm_policy *policy;
 258
 259         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 260
 261         if (policy) {
 262                 write_pnet(&policy->xp_net, net);
 263                 INIT_LIST_HEAD(&policy->walk.all);
 264                 INIT_HLIST_NODE(&policy->bydst);
 265                 INIT_HLIST_NODE(&policy->byidx);
 266                 rwlock_init(&policy->lock);
 267                 atomic_set(&policy->refcnt, 1);
 268                 setup_timer(&policy->timer, xfrm_policy_timer,
 269                                 (unsigned long)policy);
 270                 policy->flo.ops = &xfrm_policy_fc_ops;
 271         }
 272         return policy;
 273 }
 274 EXPORT_SYMBOL(xfrm_policy_alloc);
 275
 276 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 277
 278 void xfrm_policy_destroy(struct xfrm_policy *policy)
 279 {
 280         BUG_ON(!policy->walk.dead);
 281
 282         if (del_timer(&policy->timer))
 283                 BUG();
 284
 285         security_xfrm_policy_free(policy->security);
 286         kfree(policy);
 287 }
 288 EXPORT_SYMBOL(xfrm_policy_destroy);
 289
 290 /* Rule must be locked. Release descentant resources, announce
 291  * entry dead. The rule must be unlinked from lists to the moment.
 292  */
 293
 294 static void xfrm_policy_kill(struct xfrm_policy *policy)
 295 {
 296         policy->walk.dead = 1;
 297
 298         atomic_inc(&policy->genid);
 299
 300         if (del_timer(&policy->timer))
 301                 xfrm_pol_put(policy);
 302
 303         xfrm_pol_put(policy);
 304 }
 305
 306 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 307
 308 static inline unsigned int idx_hash(struct net *net, u32 index)
 309 {
 310         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 311 }
 312
 313 static struct hlist_head *policy_hash_bysel(struct net *net,
 314                                             const struct xfrm_selector *sel,
 315                                             unsigned short family, int dir)
 316 {
 317         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 318         unsigned int hash = __sel_hash(sel, family, hmask);
 319
 320         return (hash == hmask + 1 ?
 321                 &net->xfrm.policy_inexact[dir] :
 322                 net->xfrm.policy_bydst[dir].table + hash);
 323 }
 324
 325 static struct hlist_head *policy_hash_direct(struct net *net,
 326                                              const xfrm_address_t *daddr,
 327                                              const xfrm_address_t *saddr,
 328                                              unsigned short family, int dir)
 329 {
 330         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 331         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 332
 333         return net->xfrm.policy_bydst[dir].table + hash;
 334 }
 335
 336 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 337                                    struct hlist_head *ndsttable,
 338                                    unsigned int nhashmask)
 339 {
 340         struct hlist_node *entry, *tmp, *entry0 = NULL;
 341         struct xfrm_policy *pol;
 342         unsigned int h0 = 0;
 343
 344 redo:
 345         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 346                 unsigned int h;
 347
 348                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 349                                 pol->family, nhashmask);
 350                 if (!entry0) {
 351                         hlist_del(entry);
 352                         hlist_add_head(&pol->bydst, ndsttable+h);
 353                         h0 = h;
 354                 } else {
 355                         if (h != h0)
 356                                 continue;
 357                         hlist_del(entry);
 358                         hlist_add_after(entry0, &pol->bydst);
 359                 }
 360                 entry0 = entry;
 361         }
 362         if (!hlist_empty(list)) {
 363                 entry0 = NULL;
 364                 goto redo;
 365         }
 366 }
 367
 368 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 369                                    struct hlist_head *nidxtable,
 370                                    unsigned int nhashmask)
 371 {
 372         struct hlist_node *entry, *tmp;
 373         struct xfrm_policy *pol;
 374
 375         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 376                 unsigned int h;
 377
 378                 h = __idx_hash(pol->index, nhashmask);
 379                 hlist_add_head(&pol->byidx, nidxtable+h);
 380         }
 381 }
 382
 383 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 384 {
 385         return ((old_hmask + 1) << 1) - 1;
 386 }
 387
 388 static void xfrm_bydst_resize(struct net *net, int dir)
 389 {
 390         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 391         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 392         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 393         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 394         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 395         int i;
 396
 397         if (!ndst)
 398                 return;
 399
 400         write_lock_bh(&xfrm_policy_lock);
 401
 402         for (i = hmask; i >= 0; i--)
 403                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 404
 405         net->xfrm.policy_bydst[dir].table = ndst;
 406         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 407
 408         write_unlock_bh(&xfrm_policy_lock);
 409
 410         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 411 }
 412
 413 static void xfrm_byidx_resize(struct net *net, int total)
 414 {
 415         unsigned int hmask = net->xfrm.policy_idx_hmask;
 416         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 417         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 418         struct hlist_head *oidx = net->xfrm.policy_byidx;
 419         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 420         int i;
 421
 422         if (!nidx)
 423                 return;
 424
 425         write_lock_bh(&xfrm_policy_lock);
 426
 427         for (i = hmask; i >= 0; i--)
 428                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 429
 430         net->xfrm.policy_byidx = nidx;
 431         net->xfrm.policy_idx_hmask = nhashmask;
 432
 433         write_unlock_bh(&xfrm_policy_lock);
 434
 435         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 436 }
 437
 438 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 439 {
 440         unsigned int cnt = net->xfrm.policy_count[dir];
 441         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 442
 443         if (total)
 444                 *total += cnt;
 445
 446         if ((hmask + 1) < xfrm_policy_hashmax &&
 447             cnt > hmask)
 448                 return 1;
 449
 450         return 0;
 451 }
 452
 453 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 454 {
 455         unsigned int hmask = net->xfrm.policy_idx_hmask;
 456
 457         if ((hmask + 1) < xfrm_policy_hashmax &&
 458             total > hmask)
 459                 return 1;
 460
 461         return 0;
 462 }
 463
 464 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 465 {
 466         read_lock_bh(&xfrm_policy_lock);
 467         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 468         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 469         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 470         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 471         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 472         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 473         si->spdhcnt = net->xfrm.policy_idx_hmask;
 474         si->spdhmcnt = xfrm_policy_hashmax;
 475         read_unlock_bh(&xfrm_policy_lock);
 476 }
 477 EXPORT_SYMBOL(xfrm_spd_getinfo);
 478
 479 static DEFINE_MUTEX(hash_resize_mutex);
 480 static void xfrm_hash_resize(struct work_struct *work)
 481 {
 482         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 483         int dir, total;
 484
 485         mutex_lock(&hash_resize_mutex);
 486
 487         total = 0;
 488         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 489                 if (xfrm_bydst_should_resize(net, dir, &total))
 490                         xfrm_bydst_resize(net, dir);
 491         }
 492         if (xfrm_byidx_should_resize(net, total))
 493                 xfrm_byidx_resize(net, total);
 494
 495         mutex_unlock(&hash_resize_mutex);
 496 }
 497
 498 /* Generate new index... KAME seems to generate them ordered by cost
 499  * of an absolute inpredictability of ordering of rules. This will not pass. */
 500 static u32 xfrm_gen_index(struct net *net, int dir)
 501 {
 502         static u32 idx_generator;
 503
 504         for (;;) {
 505                 struct hlist_node *entry;
 506                 struct hlist_head *list;
 507                 struct xfrm_policy *p;
 508                 u32 idx;
 509                 int found;
 510
 511                 idx = (idx_generator | dir);
 512                 idx_generator += 8;
 513                 if (idx == 0)
 514                         idx = 8;
 515                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 516                 found = 0;
 517                 hlist_for_each_entry(p, entry, list, byidx) {
 518                         if (p->index == idx) {
 519                                 found = 1;
 520                                 break;
 521                         }
 522                 }
 523                 if (!found)
 524                         return idx;
 525         }
 526 }
 527
 528 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 529 {
 530         u32 *p1 = (u32 *) s1;
 531         u32 *p2 = (u32 *) s2;
 532         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 533         int i;
 534
 535         for (i = 0; i < len; i++) {
 536                 if (p1[i] != p2[i])
 537                         return 1;
 538         }
 539
 540         return 0;
 541 }
 542
 543 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 544 {
 545         struct net *net = xp_net(policy);
 546         struct xfrm_policy *pol;
 547         struct xfrm_policy *delpol;
 548         struct hlist_head *chain;
 549         struct hlist_node *entry, *newpos;
 550         u32 mark = policy->mark.v & policy->mark.m;
 551
 552         write_lock_bh(&xfrm_policy_lock);
 553         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 554         delpol = NULL;
 555         newpos = NULL;
 556         hlist_for_each_entry(pol, entry, chain, bydst) {
 557                 if (pol->type == policy->type &&
 558                     !selector_cmp(&pol->selector, &policy->selector) &&
 559                     (mark & pol->mark.m) == pol->mark.v &&
 560                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 561                     !WARN_ON(delpol)) {
 562                         if (excl) {
 563                                 write_unlock_bh(&xfrm_policy_lock);
 564                                 return -EEXIST;
 565                         }
 566                         delpol = pol;
 567                         if (policy->priority > pol->priority)
 568                                 continue;
 569                 } else if (policy->priority >= pol->priority) {
 570                         newpos = &pol->bydst;
 571                         continue;
 572                 }
 573                 if (delpol)
 574                         break;
 575         }
 576         if (newpos)
 577                 hlist_add_after(newpos, &policy->bydst);
 578         else
 579                 hlist_add_head(&policy->bydst, chain);
 580         xfrm_pol_hold(policy);
 581         net->xfrm.policy_count[dir]++;
 582         atomic_inc(&flow_cache_genid);
 583         if (delpol)
 584                 __xfrm_policy_unlink(delpol, dir);
 585         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 586         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 587         policy->curlft.add_time = get_seconds();
 588         policy->curlft.use_time = 0;
 589         if (!mod_timer(&policy->timer, jiffies + HZ))
 590                 xfrm_pol_hold(policy);
 591         list_add(&policy->walk.all, &net->xfrm.policy_all);
 592         write_unlock_bh(&xfrm_policy_lock);
 593
 594         if (delpol)
 595                 xfrm_policy_kill(delpol);
 596         else if (xfrm_bydst_should_resize(net, dir, NULL))
 597                 schedule_work(&net->xfrm.policy_hash_work);
 598
 599         return 0;
 600 }
 601 EXPORT_SYMBOL(xfrm_policy_insert);
 602
 603 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 604                                           int dir, struct xfrm_selector *sel,
 605                                           struct xfrm_sec_ctx *ctx, int delete,
 606                                           int *err)
 607 {
 608         struct xfrm_policy *pol, *ret;
 609         struct hlist_head *chain;
 610         struct hlist_node *entry;
 611
 612         *err = 0;
 613         write_lock_bh(&xfrm_policy_lock);
 614         chain = policy_hash_bysel(net, sel, sel->family, dir);
 615         ret = NULL;
 616         hlist_for_each_entry(pol, entry, chain, bydst) {
 617                 if (pol->type == type &&
 618                     (mark & pol->mark.m) == pol->mark.v &&
 619                     !selector_cmp(sel, &pol->selector) &&
 620                     xfrm_sec_ctx_match(ctx, pol->security)) {
 621                         xfrm_pol_hold(pol);
 622                         if (delete) {
 623                                 *err = security_xfrm_policy_delete(
 624                                                                 pol->security);
 625                                 if (*err) {
 626                                         write_unlock_bh(&xfrm_policy_lock);
 627                                         return pol;
 628                                 }
 629                                 __xfrm_policy_unlink(pol, dir);
 630                         }
 631                         ret = pol;
 632                         break;
 633                 }
 634         }
 635         write_unlock_bh(&xfrm_policy_lock);
 636
 637         if (ret && delete)
 638                 xfrm_policy_kill(ret);
 639         return ret;
 640 }
 641 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 642
 643 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 644                                      int dir, u32 id, int delete, int *err)
 645 {
 646         struct xfrm_policy *pol, *ret;
 647         struct hlist_head *chain;
 648         struct hlist_node *entry;
 649
 650         *err = -ENOENT;
 651         if (xfrm_policy_id2dir(id) != dir)
 652                 return NULL;
 653
 654         *err = 0;
 655         write_lock_bh(&xfrm_policy_lock);
 656         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 657         ret = NULL;
 658         hlist_for_each_entry(pol, entry, chain, byidx) {
 659                 if (pol->type == type && pol->index == id &&
 660                     (mark & pol->mark.m) == pol->mark.v) {
 661                         xfrm_pol_hold(pol);
 662                         if (delete) {
 663                                 *err = security_xfrm_policy_delete(
 664                                                                 pol->security);
 665                                 if (*err) {
 666                                         write_unlock_bh(&xfrm_policy_lock);
 667                                         return pol;
 668                                 }
 669                                 __xfrm_policy_unlink(pol, dir);
 670                         }
 671                         ret = pol;
 672                         break;
 673                 }
 674         }
 675         write_unlock_bh(&xfrm_policy_lock);
 676
 677         if (ret && delete)
 678                 xfrm_policy_kill(ret);
 679         return ret;
 680 }
 681 EXPORT_SYMBOL(xfrm_policy_byid);
 682
 683 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 684 static inline int
 685 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 686 {
 687         int dir, err = 0;
 688
 689         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 690                 struct xfrm_policy *pol;
 691                 struct hlist_node *entry;
 692                 int i;
 693
 694                 hlist_for_each_entry(pol, entry,
 695                                      &net->xfrm.policy_inexact[dir], bydst) {
 696                         if (pol->type != type)
 697                                 continue;
 698                         err = security_xfrm_policy_delete(pol->security);
 699                         if (err) {
 700                                 xfrm_audit_policy_delete(pol, 0,
 701                                                          audit_info->loginuid,
 702                                                          audit_info->sessionid,
 703                                                          audit_info->secid);
 704                                 return err;
 705                         }
 706                 }
 707                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 708                         hlist_for_each_entry(pol, entry,
 709                                              net->xfrm.policy_bydst[dir].table + i,
 710                                              bydst) {
 711                                 if (pol->type != type)
 712                                         continue;
 713                                 err = security_xfrm_policy_delete(
 714                                                                 pol->security);
 715                                 if (err) {
 716                                         xfrm_audit_policy_delete(pol, 0,
 717                                                         audit_info->loginuid,
 718                                                         audit_info->sessionid,
 719                                                         audit_info->secid);
 720                                         return err;
 721                                 }
 722                         }
 723                 }
 724         }
 725         return err;
 726 }
 727 #else
 728 static inline int
 729 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 730 {
 731         return 0;
 732 }
 733 #endif
 734
 735 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 736 {
 737         int dir, err = 0, cnt = 0;
 738
 739         write_lock_bh(&xfrm_policy_lock);
 740
 741         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 742         if (err)
 743                 goto out;
 744
 745         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 746                 struct xfrm_policy *pol;
 747                 struct hlist_node *entry;
 748                 int i;
 749
 750         again1:
 751                 hlist_for_each_entry(pol, entry,
 752                                      &net->xfrm.policy_inexact[dir], bydst) {
 753                         if (pol->type != type)
 754                                 continue;
 755                         __xfrm_policy_unlink(pol, dir);
 756                         write_unlock_bh(&xfrm_policy_lock);
 757                         cnt++;
 758
 759                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 760                                                  audit_info->sessionid,
 761                                                  audit_info->secid);
 762
 763                         xfrm_policy_kill(pol);
 764
 765                         write_lock_bh(&xfrm_policy_lock);
 766                         goto again1;
 767                 }
 768
 769                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 770         again2:
 771                         hlist_for_each_entry(pol, entry,
 772                                              net->xfrm.policy_bydst[dir].table + i,
 773                                              bydst) {
 774                                 if (pol->type != type)
 775                                         continue;
 776                                 __xfrm_policy_unlink(pol, dir);
 777                                 write_unlock_bh(&xfrm_policy_lock);
 778                                 cnt++;
 779
 780                                 xfrm_audit_policy_delete(pol, 1,
 781                                                          audit_info->loginuid,
 782                                                          audit_info->sessionid,
 783                                                          audit_info->secid);
 784                                 xfrm_policy_kill(pol);
 785
 786                                 write_lock_bh(&xfrm_policy_lock);
 787                                 goto again2;
 788                         }
 789                 }
 790
 791         }
 792         if (!cnt)
 793                 err = -ESRCH;
 794 out:
 795         write_unlock_bh(&xfrm_policy_lock);
 796         return err;
 797 }
 798 EXPORT_SYMBOL(xfrm_policy_flush);
 799
 800 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 801                      int (*func)(struct xfrm_policy *, int, int, void*),
 802                      void *data)
 803 {
 804         struct xfrm_policy *pol;
 805         struct xfrm_policy_walk_entry *x;
 806         int error = 0;
 807
 808         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 809             walk->type != XFRM_POLICY_TYPE_ANY)
 810                 return -EINVAL;
 811
 812         if (list_empty(&walk->walk.all) && walk->seq != 0)
 813                 return 0;
 814
 815         write_lock_bh(&xfrm_policy_lock);
 816         if (list_empty(&walk->walk.all))
 817                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 818         else
 819                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 820         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 821                 if (x->dead)
 822                         continue;
 823                 pol = container_of(x, struct xfrm_policy, walk);
 824                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 825                     walk->type != pol->type)
 826                         continue;
 827                 error = func(pol, xfrm_policy_id2dir(pol->index),
 828                              walk->seq, data);
 829                 if (error) {
 830                         list_move_tail(&walk->walk.all, &x->all);
 831                         goto out;
 832                 }
 833                 walk->seq++;
 834         }
 835         if (walk->seq == 0) {
 836                 error = -ENOENT;
 837                 goto out;
 838         }
 839         list_del_init(&walk->walk.all);
 840 out:
 841         write_unlock_bh(&xfrm_policy_lock);
 842         return error;
 843 }
 844 EXPORT_SYMBOL(xfrm_policy_walk);
 845
 846 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 847 {
 848         INIT_LIST_HEAD(&walk->walk.all);
 849         walk->walk.dead = 1;
 850         walk->type = type;
 851         walk->seq = 0;
 852 }
 853 EXPORT_SYMBOL(xfrm_policy_walk_init);
 854
 855 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 856 {
 857         if (list_empty(&walk->walk.all))
 858                 return;
 859
 860         write_lock_bh(&xfrm_policy_lock);
 861         list_del(&walk->walk.all);
 862         write_unlock_bh(&xfrm_policy_lock);
 863 }
 864 EXPORT_SYMBOL(xfrm_policy_walk_done);
 865
 866 /*
 867  * Find policy to apply to this flow.
 868  *
 869  * Returns 0 if policy found, else an -errno.
 870  */
 871 static int xfrm_policy_match(const struct xfrm_policy *pol,
 872                              const struct flowi *fl,
 873                              u8 type, u16 family, int dir)
 874 {
 875         const struct xfrm_selector *sel = &pol->selector;
 876         int match, ret = -ESRCH;
 877
 878         if (pol->family != family ||
 879             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
 880             pol->type != type)
 881                 return ret;
 882
 883         match = xfrm_selector_match(sel, fl, family);
 884         if (match)
 885                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
 886                                                   dir);
 887
 888         return ret;
 889 }
 890
 891 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 892                                                      const struct flowi *fl,
 893                                                      u16 family, u8 dir)
 894 {
 895         int err;
 896         struct xfrm_policy *pol, *ret;
 897         const xfrm_address_t *daddr, *saddr;
 898         struct hlist_node *entry;
 899         struct hlist_head *chain;
 900         u32 priority = ~0U;
 901
 902         daddr = xfrm_flowi_daddr(fl, family);
 903         saddr = xfrm_flowi_saddr(fl, family);
 904         if (unlikely(!daddr || !saddr))
 905                 return NULL;
 906
 907         read_lock_bh(&xfrm_policy_lock);
 908         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 909         ret = NULL;
 910         hlist_for_each_entry(pol, entry, chain, bydst) {
 911                 err = xfrm_policy_match(pol, fl, type, family, dir);
 912                 if (err) {
 913                         if (err == -ESRCH)
 914                                 continue;
 915                         else {
 916                                 ret = ERR_PTR(err);
 917                                 goto fail;
 918                         }
 919                 } else {
 920                         ret = pol;
 921                         priority = ret->priority;
 922                         break;
 923                 }
 924         }
 925         chain = &net->xfrm.policy_inexact[dir];
 926         hlist_for_each_entry(pol, entry, chain, bydst) {
 927                 err = xfrm_policy_match(pol, fl, type, family, dir);
 928                 if (err) {
 929                         if (err == -ESRCH)
 930                                 continue;
 931                         else {
 932                                 ret = ERR_PTR(err);
 933                                 goto fail;
 934                         }
 935                 } else if (pol->priority < priority) {
 936                         ret = pol;
 937                         break;
 938                 }
 939         }
 940         if (ret)
 941                 xfrm_pol_hold(ret);
 942 fail:
 943         read_unlock_bh(&xfrm_policy_lock);
 944
 945         return ret;
 946 }
 947
 948 static struct xfrm_policy *
 949 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
 950 {
 951 #ifdef CONFIG_XFRM_SUB_POLICY
 952         struct xfrm_policy *pol;
 953
 954         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
 955         if (pol != NULL)
 956                 return pol;
 957 #endif
 958         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 959 }
 960
 961 static struct flow_cache_object *
 962 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
 963                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
 964 {
 965         struct xfrm_policy *pol;
 966
 967         if (old_obj)
 968                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
 969
 970         pol = __xfrm_policy_lookup(net, fl, family, dir);
 971         if (IS_ERR_OR_NULL(pol))
 972                 return ERR_CAST(pol);
 973
 974         /* Resolver returns two references:
 975          * one for cache and one for caller of flow_cache_lookup() */
 976         xfrm_pol_hold(pol);
 977
 978         return &pol->flo;
 979 }
 980
 981 static inline int policy_to_flow_dir(int dir)
 982 {
 983         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 984             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 985             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 986                 return dir;
 987         switch (dir) {
 988         default:
 989         case XFRM_POLICY_IN:
 990                 return FLOW_DIR_IN;
 991         case XFRM_POLICY_OUT:
 992                 return FLOW_DIR_OUT;
 993         case XFRM_POLICY_FWD:
 994                 return FLOW_DIR_FWD;
 995         }
 996 }
 997
 998 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
 999                                                  const struct flowi *fl)
1000 {
1001         struct xfrm_policy *pol;
1002
1003         read_lock_bh(&xfrm_policy_lock);
1004         if ((pol = sk->sk_policy[dir]) != NULL) {
1005                 int match = xfrm_selector_match(&pol->selector, fl,
1006                                                 sk->sk_family);
1007                 int err = 0;
1008
1009                 if (match) {
1010                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1011                                 pol = NULL;
1012                                 goto out;
1013                         }
1014                         err = security_xfrm_policy_lookup(pol->security,
1015                                                       fl->flowi_secid,
1016                                                       policy_to_flow_dir(dir));
1017                         if (!err)
1018                                 xfrm_pol_hold(pol);
1019                         else if (err == -ESRCH)
1020                                 pol = NULL;
1021                         else
1022                                 pol = ERR_PTR(err);
1023                 } else
1024                         pol = NULL;
1025         }
1026 out:
1027         read_unlock_bh(&xfrm_policy_lock);
1028         return pol;
1029 }
1030
1031 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1032 {
1033         struct net *net = xp_net(pol);
1034         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1035                                                      pol->family, dir);
1036
1037         list_add(&pol->walk.all, &net->xfrm.policy_all);
1038         hlist_add_head(&pol->bydst, chain);
1039         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1040         net->xfrm.policy_count[dir]++;
1041         xfrm_pol_hold(pol);
1042
1043         if (xfrm_bydst_should_resize(net, dir, NULL))
1044                 schedule_work(&net->xfrm.policy_hash_work);
1045 }
1046
1047 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1048                                                 int dir)
1049 {
1050         struct net *net = xp_net(pol);
1051
1052         if (hlist_unhashed(&pol->bydst))
1053                 return NULL;
1054
1055         hlist_del(&pol->bydst);
1056         hlist_del(&pol->byidx);
1057         list_del(&pol->walk.all);
1058         net->xfrm.policy_count[dir]--;
1059
1060         return pol;
1061 }
1062
1063 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1064 {
1065         write_lock_bh(&xfrm_policy_lock);
1066         pol = __xfrm_policy_unlink(pol, dir);
1067         write_unlock_bh(&xfrm_policy_lock);
1068         if (pol) {
1069                 xfrm_policy_kill(pol);
1070                 return 0;
1071         }
1072         return -ENOENT;
1073 }
1074 EXPORT_SYMBOL(xfrm_policy_delete);
1075
1076 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1077 {
1078         struct net *net = xp_net(pol);
1079         struct xfrm_policy *old_pol;
1080
1081 #ifdef CONFIG_XFRM_SUB_POLICY
1082         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1083                 return -EINVAL;
1084 #endif
1085
1086         write_lock_bh(&xfrm_policy_lock);
1087         old_pol = sk->sk_policy[dir];
1088         sk->sk_policy[dir] = pol;
1089         if (pol) {
1090                 pol->curlft.add_time = get_seconds();
1091                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1092                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1093         }
1094         if (old_pol)
1095                 /* Unlinking succeeds always. This is the only function
1096                  * allowed to delete or replace socket policy.
1097                  */
1098                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1099         write_unlock_bh(&xfrm_policy_lock);
1100
1101         if (old_pol) {
1102                 xfrm_policy_kill(old_pol);
1103         }
1104         return 0;
1105 }
1106
1107 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1108 {
1109         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1110
1111         if (newp) {
1112                 newp->selector = old->selector;
1113                 if (security_xfrm_policy_clone(old->security,
1114                                                &newp->security)) {
1115                         kfree(newp);
1116                         return NULL;  /* ENOMEM */
1117                 }
1118                 newp->lft = old->lft;
1119                 newp->curlft = old->curlft;
1120                 newp->mark = old->mark;
1121                 newp->action = old->action;
1122                 newp->flags = old->flags;
1123                 newp->xfrm_nr = old->xfrm_nr;
1124                 newp->index = old->index;
1125                 newp->type = old->type;
1126                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1127                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1128                 write_lock_bh(&xfrm_policy_lock);
1129                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1130                 write_unlock_bh(&xfrm_policy_lock);
1131                 xfrm_pol_put(newp);
1132         }
1133         return newp;
1134 }
1135
1136 int __xfrm_sk_clone_policy(struct sock *sk)
1137 {
1138         struct xfrm_policy *p0 = sk->sk_policy[0],
1139                            *p1 = sk->sk_policy[1];
1140
1141         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1142         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1143                 return -ENOMEM;
1144         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1145                 return -ENOMEM;
1146         return 0;
1147 }
1148
1149 static int
1150 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1151                unsigned short family)
1152 {
1153         int err;
1154         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1155
1156         if (unlikely(afinfo == NULL))
1157                 return -EINVAL;
1158         err = afinfo->get_saddr(net, local, remote);
1159         xfrm_policy_put_afinfo(afinfo);
1160         return err;
1161 }
1162
1163 /* Resolve list of templates for the flow, given policy. */
1164
1165 static int
1166 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1167                       struct xfrm_state **xfrm, unsigned short family)
1168 {
1169         struct net *net = xp_net(policy);
1170         int nx;
1171         int i, error;
1172         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1173         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1174         xfrm_address_t tmp;
1175
1176         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1177                 struct xfrm_state *x;
1178                 xfrm_address_t *remote = daddr;
1179                 xfrm_address_t *local  = saddr;
1180                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1181
1182                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1183                     tmpl->mode == XFRM_MODE_BEET) {
1184                         remote = &tmpl->id.daddr;
1185                         local = &tmpl->saddr;
1186                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1187                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1188                                 if (error)
1189                                         goto fail;
1190                                 local = &tmp;
1191                         }
1192                 }
1193
1194                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1195
1196                 if (x && x->km.state == XFRM_STATE_VALID) {
1197                         xfrm[nx++] = x;
1198                         daddr = remote;
1199                         saddr = local;
1200                         continue;
1201                 }
1202                 if (x) {
1203                         error = (x->km.state == XFRM_STATE_ERROR ?
1204                                  -EINVAL : -EAGAIN);
1205                         xfrm_state_put(x);
1206                 }
1207                 else if (error == -ESRCH)
1208                         error = -EAGAIN;
1209
1210                 if (!tmpl->optional)
1211                         goto fail;
1212         }
1213         return nx;
1214
1215 fail:
1216         for (nx--; nx>=0; nx--)
1217                 xfrm_state_put(xfrm[nx]);
1218         return error;
1219 }
1220
1221 static int
1222 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1223                   struct xfrm_state **xfrm, unsigned short family)
1224 {
1225         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1226         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1227         int cnx = 0;
1228         int error;
1229         int ret;
1230         int i;
1231
1232         for (i = 0; i < npols; i++) {
1233                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1234                         error = -ENOBUFS;
1235                         goto fail;
1236                 }
1237
1238                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1239                 if (ret < 0) {
1240                         error = ret;
1241                         goto fail;
1242                 } else
1243                         cnx += ret;
1244         }
1245
1246         /* found states are sorted for outbound processing */
1247         if (npols > 1)
1248                 xfrm_state_sort(xfrm, tpp, cnx, family);
1249
1250         return cnx;
1251
1252  fail:
1253         for (cnx--; cnx>=0; cnx--)
1254                 xfrm_state_put(tpp[cnx]);
1255         return error;
1256
1257 }
1258
1259 /* Check that the bundle accepts the flow and its components are
1260  * still valid.
1261  */
1262
1263 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1264 {
1265         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1266         int tos;
1267
1268         if (!afinfo)
1269                 return -EINVAL;
1270
1271         tos = afinfo->get_tos(fl);
1272
1273         xfrm_policy_put_afinfo(afinfo);
1274
1275         return tos;
1276 }
1277
1278 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1279 {
1280         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1281         struct dst_entry *dst = &xdst->u.dst;
1282
1283         if (xdst->route == NULL) {
1284                 /* Dummy bundle - if it has xfrms we were not
1285                  * able to build bundle as template resolution failed.
1286                  * It means we need to try again resolving. */
1287                 if (xdst->num_xfrms > 0)
1288                         return NULL;
1289         } else {
1290                 /* Real bundle */
1291                 if (stale_bundle(dst))
1292                         return NULL;
1293         }
1294
1295         dst_hold(dst);
1296         return flo;
1297 }
1298
1299 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1300 {
1301         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1302         struct dst_entry *dst = &xdst->u.dst;
1303
1304         if (!xdst->route)
1305                 return 0;
1306         if (stale_bundle(dst))
1307                 return 0;
1308
1309         return 1;
1310 }
1311
1312 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1313 {
1314         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1315         struct dst_entry *dst = &xdst->u.dst;
1316
1317         dst_free(dst);
1318 }
1319
1320 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1321         .get = xfrm_bundle_flo_get,
1322         .check = xfrm_bundle_flo_check,
1323         .delete = xfrm_bundle_flo_delete,
1324 };
1325
1326 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1327 {
1328         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1329         struct dst_ops *dst_ops;
1330         struct xfrm_dst *xdst;
1331
1332         if (!afinfo)
1333                 return ERR_PTR(-EINVAL);
1334
1335         switch (family) {
1336         case AF_INET:
1337                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1338                 break;
1339 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1340         case AF_INET6:
1341                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1342                 break;
1343 #endif
1344         default:
1345                 BUG();
1346         }
1347         xdst = dst_alloc(dst_ops, 0);
1348         xfrm_policy_put_afinfo(afinfo);
1349
1350         if (likely(xdst))
1351                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1352         else
1353                 xdst = ERR_PTR(-ENOBUFS);
1354
1355         return xdst;
1356 }
1357
1358 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1359                                  int nfheader_len)
1360 {
1361         struct xfrm_policy_afinfo *afinfo =
1362                 xfrm_policy_get_afinfo(dst->ops->family);
1363         int err;
1364
1365         if (!afinfo)
1366                 return -EINVAL;
1367
1368         err = afinfo->init_path(path, dst, nfheader_len);
1369
1370         xfrm_policy_put_afinfo(afinfo);
1371
1372         return err;
1373 }
1374
1375 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1376                                 const struct flowi *fl)
1377 {
1378         struct xfrm_policy_afinfo *afinfo =
1379                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1380         int err;
1381
1382         if (!afinfo)
1383                 return -EINVAL;
1384
1385         err = afinfo->fill_dst(xdst, dev, fl);
1386
1387         xfrm_policy_put_afinfo(afinfo);
1388
1389         return err;
1390 }
1391
1392
1393 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1394  * all the metrics... Shortly, bundle a bundle.
1395  */
1396
1397 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1398                                             struct xfrm_state **xfrm, int nx,
1399                                             const struct flowi *fl,
1400                                             struct dst_entry *dst)
1401 {
1402         struct net *net = xp_net(policy);
1403         unsigned long now = jiffies;
1404         struct net_device *dev;
1405         struct dst_entry *dst_prev = NULL;
1406         struct dst_entry *dst0 = NULL;
1407         int i = 0;
1408         int err;
1409         int header_len = 0;
1410         int nfheader_len = 0;
1411         int trailer_len = 0;
1412         int tos;
1413         int family = policy->selector.family;
1414         xfrm_address_t saddr, daddr;
1415
1416         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1417
1418         tos = xfrm_get_tos(fl, family);
1419         err = tos;
1420         if (tos < 0)
1421                 goto put_states;
1422
1423         dst_hold(dst);
1424
1425         for (; i < nx; i++) {
1426                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1427                 struct dst_entry *dst1 = &xdst->u.dst;
1428
1429                 err = PTR_ERR(xdst);
1430                 if (IS_ERR(xdst)) {
1431                         dst_release(dst);
1432                         goto put_states;
1433                 }
1434
1435                 if (!dst_prev)
1436                         dst0 = dst1;
1437                 else {
1438                         dst_prev->child = dst_clone(dst1);
1439                         dst1->flags |= DST_NOHASH;
1440                 }
1441
1442                 xdst->route = dst;
1443                 dst_copy_metrics(dst1, dst);
1444
1445                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1446                         family = xfrm[i]->props.family;
1447                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1448                                               family);
1449                         err = PTR_ERR(dst);
1450                         if (IS_ERR(dst))
1451                                 goto put_states;
1452                 } else
1453                         dst_hold(dst);
1454
1455                 dst1->xfrm = xfrm[i];
1456                 xdst->xfrm_genid = xfrm[i]->genid;
1457
1458                 dst1->obsolete = -1;
1459                 dst1->flags |= DST_HOST;
1460                 dst1->lastuse = now;
1461
1462                 dst1->input = dst_discard;
1463                 dst1->output = xfrm[i]->outer_mode->afinfo->output;
1464
1465                 dst1->next = dst_prev;
1466                 dst_prev = dst1;
1467
1468                 header_len += xfrm[i]->props.header_len;
1469                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1470                         nfheader_len += xfrm[i]->props.header_len;
1471                 trailer_len += xfrm[i]->props.trailer_len;
1472         }
1473
1474         dst_prev->child = dst;
1475         dst0->path = dst;
1476
1477         err = -ENODEV;
1478         dev = dst->dev;
1479         if (!dev)
1480                 goto free_dst;
1481
1482         /* Copy neighbour for reachability confirmation */
1483         dst0->neighbour = neigh_clone(dst->neighbour);
1484
1485         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1486         xfrm_init_pmtu(dst_prev);
1487
1488         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1489                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1490
1491                 err = xfrm_fill_dst(xdst, dev, fl);
1492                 if (err)
1493                         goto free_dst;
1494
1495                 dst_prev->header_len = header_len;
1496                 dst_prev->trailer_len = trailer_len;
1497                 header_len -= xdst->u.dst.xfrm->props.header_len;
1498                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1499         }
1500
1501 out:
1502         return dst0;
1503
1504 put_states:
1505         for (; i < nx; i++)
1506                 xfrm_state_put(xfrm[i]);
1507 free_dst:
1508         if (dst0)
1509                 dst_free(dst0);
1510         dst0 = ERR_PTR(err);
1511         goto out;
1512 }
1513
1514 static int inline
1515 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1516 {
1517         if (!*target) {
1518                 *target = kmalloc(size, GFP_ATOMIC);
1519                 if (!*target)
1520                         return -ENOMEM;
1521         }
1522         memcpy(*target, src, size);
1523         return 0;
1524 }
1525
1526 static int inline
1527 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1528 {
1529 #ifdef CONFIG_XFRM_SUB_POLICY
1530         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1531         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1532                                    sel, sizeof(*sel));
1533 #else
1534         return 0;
1535 #endif
1536 }
1537
1538 static int inline
1539 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1540 {
1541 #ifdef CONFIG_XFRM_SUB_POLICY
1542         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1543         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1544 #else
1545         return 0;
1546 #endif
1547 }
1548
1549 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1550                                 struct xfrm_policy **pols,
1551                                 int *num_pols, int *num_xfrms)
1552 {
1553         int i;
1554
1555         if (*num_pols == 0 || !pols[0]) {
1556                 *num_pols = 0;
1557                 *num_xfrms = 0;
1558                 return 0;
1559         }
1560         if (IS_ERR(pols[0]))
1561                 return PTR_ERR(pols[0]);
1562
1563         *num_xfrms = pols[0]->xfrm_nr;
1564
1565 #ifdef CONFIG_XFRM_SUB_POLICY
1566         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1567             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1568                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1569                                                     XFRM_POLICY_TYPE_MAIN,
1570                                                     fl, family,
1571                                                     XFRM_POLICY_OUT);
1572                 if (pols[1]) {
1573                         if (IS_ERR(pols[1])) {
1574                                 xfrm_pols_put(pols, *num_pols);
1575                                 return PTR_ERR(pols[1]);
1576                         }
1577                         (*num_pols) ++;
1578                         (*num_xfrms) += pols[1]->xfrm_nr;
1579                 }
1580         }
1581 #endif
1582         for (i = 0; i < *num_pols; i++) {
1583                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1584                         *num_xfrms = -1;
1585                         break;
1586                 }
1587         }
1588
1589         return 0;
1590
1591 }
1592
1593 static struct xfrm_dst *
1594 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1595                                const struct flowi *fl, u16 family,
1596                                struct dst_entry *dst_orig)
1597 {
1598         struct net *net = xp_net(pols[0]);
1599         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1600         struct dst_entry *dst;
1601         struct xfrm_dst *xdst;
1602         int err;
1603
1604         /* Try to instantiate a bundle */
1605         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1606         if (err <= 0) {
1607                 if (err != 0 && err != -EAGAIN)
1608                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1609                 return ERR_PTR(err);
1610         }
1611
1612         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1613         if (IS_ERR(dst)) {
1614                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1615                 return ERR_CAST(dst);
1616         }
1617
1618         xdst = (struct xfrm_dst *)dst;
1619         xdst->num_xfrms = err;
1620         if (num_pols > 1)
1621                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1622         else
1623                 err = xfrm_dst_update_origin(dst, fl);
1624         if (unlikely(err)) {
1625                 dst_free(dst);
1626                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1627                 return ERR_PTR(err);
1628         }
1629
1630         xdst->num_pols = num_pols;
1631         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1632         xdst->policy_genid = atomic_read(&pols[0]->genid);
1633
1634         return xdst;
1635 }
1636
1637 static struct flow_cache_object *
1638 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1639                    struct flow_cache_object *oldflo, void *ctx)
1640 {
1641         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1642         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1643         struct xfrm_dst *xdst, *new_xdst;
1644         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1645
1646         /* Check if the policies from old bundle are usable */
1647         xdst = NULL;
1648         if (oldflo) {
1649                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1650                 num_pols = xdst->num_pols;
1651                 num_xfrms = xdst->num_xfrms;
1652                 pol_dead = 0;
1653                 for (i = 0; i < num_pols; i++) {
1654                         pols[i] = xdst->pols[i];
1655                         pol_dead |= pols[i]->walk.dead;
1656                 }
1657                 if (pol_dead) {
1658                         dst_free(&xdst->u.dst);
1659                         xdst = NULL;
1660                         num_pols = 0;
1661                         num_xfrms = 0;
1662                         oldflo = NULL;
1663                 }
1664         }
1665
1666         /* Resolve policies to use if we couldn't get them from
1667          * previous cache entry */
1668         if (xdst == NULL) {
1669                 num_pols = 1;
1670                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1671                 err = xfrm_expand_policies(fl, family, pols,
1672                                            &num_pols, &num_xfrms);
1673                 if (err < 0)
1674                         goto inc_error;
1675                 if (num_pols == 0)
1676                         return NULL;
1677                 if (num_xfrms <= 0)
1678                         goto make_dummy_bundle;
1679         }
1680
1681         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1682         if (IS_ERR(new_xdst)) {
1683                 err = PTR_ERR(new_xdst);
1684                 if (err != -EAGAIN)
1685                         goto error;
1686                 if (oldflo == NULL)
1687                         goto make_dummy_bundle;
1688                 dst_hold(&xdst->u.dst);
1689                 return oldflo;
1690         } else if (new_xdst == NULL) {
1691                 num_xfrms = 0;
1692                 if (oldflo == NULL)
1693                         goto make_dummy_bundle;
1694                 xdst->num_xfrms = 0;
1695                 dst_hold(&xdst->u.dst);
1696                 return oldflo;
1697         }
1698
1699         /* Kill the previous bundle */
1700         if (xdst) {
1701                 /* The policies were stolen for newly generated bundle */
1702                 xdst->num_pols = 0;
1703                 dst_free(&xdst->u.dst);
1704         }
1705
1706         /* Flow cache does not have reference, it dst_free()'s,
1707          * but we do need to return one reference for original caller */
1708         dst_hold(&new_xdst->u.dst);
1709         return &new_xdst->flo;
1710
1711 make_dummy_bundle:
1712         /* We found policies, but there's no bundles to instantiate:
1713          * either because the policy blocks, has no transformations or
1714          * we could not build template (no xfrm_states).*/
1715         xdst = xfrm_alloc_dst(net, family);
1716         if (IS_ERR(xdst)) {
1717                 xfrm_pols_put(pols, num_pols);
1718                 return ERR_CAST(xdst);
1719         }
1720         xdst->num_pols = num_pols;
1721         xdst->num_xfrms = num_xfrms;
1722         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1723
1724         dst_hold(&xdst->u.dst);
1725         return &xdst->flo;
1726
1727 inc_error:
1728         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1729 error:
1730         if (xdst != NULL)
1731                 dst_free(&xdst->u.dst);
1732         else
1733                 xfrm_pols_put(pols, num_pols);
1734         return ERR_PTR(err);
1735 }
1736
1737 static struct dst_entry *make_blackhole(struct net *net, u16 family,
1738                                         struct dst_entry *dst_orig)
1739 {
1740         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1741         struct dst_entry *ret;
1742
1743         if (!afinfo) {
1744                 dst_release(dst_orig);
1745                 ret = ERR_PTR(-EINVAL);
1746         } else {
1747                 ret = afinfo->blackhole_route(net, dst_orig);
1748         }
1749         xfrm_policy_put_afinfo(afinfo);
1750
1751         return ret;
1752 }
1753
1754 /* Main function: finds/creates a bundle for given flow.
1755  *
1756  * At the moment we eat a raw IP route. Mostly to speed up lookups
1757  * on interfaces with disabled IPsec.
1758  */
1759 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
1760                               const struct flowi *fl,
1761                               struct sock *sk, int flags)
1762 {
1763         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1764         struct flow_cache_object *flo;
1765         struct xfrm_dst *xdst;
1766         struct dst_entry *dst, *route;
1767         u16 family = dst_orig->ops->family;
1768         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1769         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1770
1771 restart:
1772         dst = NULL;
1773         xdst = NULL;
1774         route = NULL;
1775
1776         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1777                 num_pols = 1;
1778                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1779                 err = xfrm_expand_policies(fl, family, pols,
1780                                            &num_pols, &num_xfrms);
1781                 if (err < 0)
1782                         goto dropdst;
1783
1784                 if (num_pols) {
1785                         if (num_xfrms <= 0) {
1786                                 drop_pols = num_pols;
1787                                 goto no_transform;
1788                         }
1789
1790                         xdst = xfrm_resolve_and_create_bundle(
1791                                         pols, num_pols, fl,
1792                                         family, dst_orig);
1793                         if (IS_ERR(xdst)) {
1794                                 xfrm_pols_put(pols, num_pols);
1795                                 err = PTR_ERR(xdst);
1796                                 goto dropdst;
1797                         } else if (xdst == NULL) {
1798                                 num_xfrms = 0;
1799                                 drop_pols = num_pols;
1800                                 goto no_transform;
1801                         }
1802
1803                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1804                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1805                         xfrm_policy_sk_bundles = &xdst->u.dst;
1806                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1807
1808                         route = xdst->route;
1809                 }
1810         }
1811
1812         if (xdst == NULL) {
1813                 /* To accelerate a bit...  */
1814                 if ((dst_orig->flags & DST_NOXFRM) ||
1815                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1816                         goto nopol;
1817
1818                 flo = flow_cache_lookup(net, fl, family, dir,
1819                                         xfrm_bundle_lookup, dst_orig);
1820                 if (flo == NULL)
1821                         goto nopol;
1822                 if (IS_ERR(flo)) {
1823                         err = PTR_ERR(flo);
1824                         goto dropdst;
1825                 }
1826                 xdst = container_of(flo, struct xfrm_dst, flo);
1827
1828                 num_pols = xdst->num_pols;
1829                 num_xfrms = xdst->num_xfrms;
1830                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1831                 route = xdst->route;
1832         }
1833
1834         dst = &xdst->u.dst;
1835         if (route == NULL && num_xfrms > 0) {
1836                 /* The only case when xfrm_bundle_lookup() returns a
1837                  * bundle with null route, is when the template could
1838                  * not be resolved. It means policies are there, but
1839                  * bundle could not be created, since we don't yet
1840                  * have the xfrm_state's. We need to wait for KM to
1841                  * negotiate new SA's or bail out with error.*/
1842                 if (net->xfrm.sysctl_larval_drop) {
1843                         /* EREMOTE tells the caller to generate
1844                          * a one-shot blackhole route. */
1845                         dst_release(dst);
1846                         xfrm_pols_put(pols, drop_pols);
1847                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1848
1849                         return make_blackhole(net, family, dst_orig);
1850                 }
1851                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
1852                         DECLARE_WAITQUEUE(wait, current);
1853
1854                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1855                         set_current_state(TASK_INTERRUPTIBLE);
1856                         schedule();
1857                         set_current_state(TASK_RUNNING);
1858                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1859
1860                         if (!signal_pending(current)) {
1861                                 dst_release(dst);
1862                                 goto restart;
1863                         }
1864
1865                         err = -ERESTART;
1866                 } else
1867                         err = -EAGAIN;
1868
1869                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1870                 goto error;
1871         }
1872
1873 no_transform:
1874         if (num_pols == 0)
1875                 goto nopol;
1876
1877         if ((flags & XFRM_LOOKUP_ICMP) &&
1878             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1879                 err = -ENOENT;
1880                 goto error;
1881         }
1882
1883         for (i = 0; i < num_pols; i++)
1884                 pols[i]->curlft.use_time = get_seconds();
1885
1886         if (num_xfrms < 0) {
1887                 /* Prohibit the flow */
1888                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1889                 err = -EPERM;
1890                 goto error;
1891         } else if (num_xfrms > 0) {
1892                 /* Flow transformed */
1893                 dst_release(dst_orig);
1894         } else {
1895                 /* Flow passes untransformed */
1896                 dst_release(dst);
1897                 dst = dst_orig;
1898         }
1899 ok:
1900         xfrm_pols_put(pols, drop_pols);
1901         return dst;
1902
1903 nopol:
1904         if (!(flags & XFRM_LOOKUP_ICMP)) {
1905                 dst = dst_orig;
1906                 goto ok;
1907         }
1908         err = -ENOENT;
1909 error:
1910         dst_release(dst);
1911 dropdst:
1912         dst_release(dst_orig);
1913         xfrm_pols_put(pols, drop_pols);
1914         return ERR_PTR(err);
1915 }
1916 EXPORT_SYMBOL(xfrm_lookup);
1917
1918 static inline int
1919 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
1920 {
1921         struct xfrm_state *x;
1922
1923         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1924                 return 0;
1925         x = skb->sp->xvec[idx];
1926         if (!x->type->reject)
1927                 return 0;
1928         return x->type->reject(x, skb, fl);
1929 }
1930
1931 /* When skb is transformed back to its "native" form, we have to
1932  * check policy restrictions. At the moment we make this in maximally
1933  * stupid way. Shame on me. :-) Of course, connected sockets must
1934  * have policy cached at them.
1935  */
1936
1937 static inline int
1938 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
1939               unsigned short family)
1940 {
1941         if (xfrm_state_kern(x))
1942                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1943         return  x->id.proto == tmpl->id.proto &&
1944                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1945                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1946                 x->props.mode == tmpl->mode &&
1947                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1948                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1949                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1950                   xfrm_state_addr_cmp(tmpl, x, family));
1951 }
1952
1953 /*
1954  * 0 or more than 0 is returned when validation is succeeded (either bypass
1955  * because of optional transport mode, or next index of the mathced secpath
1956  * state with the template.
1957  * -1 is returned when no matching template is found.
1958  * Otherwise "-2 - errored_index" is returned.
1959  */
1960 static inline int
1961 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
1962                unsigned short family)
1963 {
1964         int idx = start;
1965
1966         if (tmpl->optional) {
1967                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1968                         return start;
1969         } else
1970                 start = -1;
1971         for (; idx < sp->len; idx++) {
1972                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1973                         return ++idx;
1974                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1975                         if (start == -1)
1976                                 start = -2-idx;
1977                         break;
1978                 }
1979         }
1980         return start;
1981 }
1982
1983 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1984                           unsigned int family, int reverse)
1985 {
1986         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1987         int err;
1988
1989         if (unlikely(afinfo == NULL))
1990                 return -EAFNOSUPPORT;
1991
1992         afinfo->decode_session(skb, fl, reverse);
1993         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
1994         xfrm_policy_put_afinfo(afinfo);
1995         return err;
1996 }
1997 EXPORT_SYMBOL(__xfrm_decode_session);
1998
1999 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2000 {
2001         for (; k < sp->len; k++) {
2002                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2003                         *idxp = k;
2004                         return 1;
2005                 }
2006         }
2007
2008         return 0;
2009 }
2010
2011 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2012                         unsigned short family)
2013 {
2014         struct net *net = dev_net(skb->dev);
2015         struct xfrm_policy *pol;
2016         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2017         int npols = 0;
2018         int xfrm_nr;
2019         int pi;
2020         int reverse;
2021         struct flowi fl;
2022         u8 fl_dir;
2023         int xerr_idx = -1;
2024
2025         reverse = dir & ~XFRM_POLICY_MASK;
2026         dir &= XFRM_POLICY_MASK;
2027         fl_dir = policy_to_flow_dir(dir);
2028
2029         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2030                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2031                 return 0;
2032         }
2033
2034         nf_nat_decode_session(skb, &fl, family);
2035
2036         /* First, check used SA against their selectors. */
2037         if (skb->sp) {
2038                 int i;
2039
2040                 for (i=skb->sp->len-1; i>=0; i--) {
2041                         struct xfrm_state *x = skb->sp->xvec[i];
2042                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2043                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2044                                 return 0;
2045                         }
2046                 }
2047         }
2048
2049         pol = NULL;
2050         if (sk && sk->sk_policy[dir]) {
2051                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2052                 if (IS_ERR(pol)) {
2053                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2054                         return 0;
2055                 }
2056         }
2057
2058         if (!pol) {
2059                 struct flow_cache_object *flo;
2060
2061                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2062                                         xfrm_policy_lookup, NULL);
2063                 if (IS_ERR_OR_NULL(flo))
2064                         pol = ERR_CAST(flo);
2065                 else
2066                         pol = container_of(flo, struct xfrm_policy, flo);
2067         }
2068
2069         if (IS_ERR(pol)) {
2070                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2071                 return 0;
2072         }
2073
2074         if (!pol) {
2075                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2076                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2077                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2078                         return 0;
2079                 }
2080                 return 1;
2081         }
2082
2083         pol->curlft.use_time = get_seconds();
2084
2085         pols[0] = pol;
2086         npols ++;
2087 #ifdef CONFIG_XFRM_SUB_POLICY
2088         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2089                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2090                                                     &fl, family,
2091                                                     XFRM_POLICY_IN);
2092                 if (pols[1]) {
2093                         if (IS_ERR(pols[1])) {
2094                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2095                                 return 0;
2096                         }
2097                         pols[1]->curlft.use_time = get_seconds();
2098                         npols ++;
2099                 }
2100         }
2101 #endif
2102
2103         if (pol->action == XFRM_POLICY_ALLOW) {
2104                 struct sec_path *sp;
2105                 static struct sec_path dummy;
2106                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2107                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2108                 struct xfrm_tmpl **tpp = tp;
2109                 int ti = 0;
2110                 int i, k;
2111
2112                 if ((sp = skb->sp) == NULL)
2113                         sp = &dummy;
2114
2115                 for (pi = 0; pi < npols; pi++) {
2116                         if (pols[pi] != pol &&
2117                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2118                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2119                                 goto reject;
2120                         }
2121                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2122                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2123                                 goto reject_error;
2124                         }
2125                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2126                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2127                 }
2128                 xfrm_nr = ti;
2129                 if (npols > 1) {
2130                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2131                         tpp = stp;
2132                 }
2133
2134                 /* For each tunnel xfrm, find the first matching tmpl.
2135                  * For each tmpl before that, find corresponding xfrm.
2136                  * Order is _important_. Later we will implement
2137                  * some barriers, but at the moment barriers
2138                  * are implied between each two transformations.
2139                  */
2140                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2141                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2142                         if (k < 0) {
2143                                 if (k < -1)
2144                                         /* "-2 - errored_index" returned */
2145                                         xerr_idx = -(2+k);
2146                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2147                                 goto reject;
2148                         }
2149                 }
2150
2151                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2152                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2153                         goto reject;
2154                 }
2155
2156                 xfrm_pols_put(pols, npols);
2157                 return 1;
2158         }
2159         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2160
2161 reject:
2162         xfrm_secpath_reject(xerr_idx, skb, &fl);
2163 reject_error:
2164         xfrm_pols_put(pols, npols);
2165         return 0;
2166 }
2167 EXPORT_SYMBOL(__xfrm_policy_check);
2168
2169 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2170 {
2171         struct net *net = dev_net(skb->dev);
2172         struct flowi fl;
2173         struct dst_entry *dst;
2174         int res = 0;
2175
2176         if (xfrm_decode_session(skb, &fl, family) < 0) {
2177                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2178                 return 0;
2179         }
2180
2181         skb_dst_force(skb);
2182
2183         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2184         if (IS_ERR(dst)) {
2185                 res = 1;
2186                 dst = NULL;
2187         }
2188         skb_dst_set(skb, dst);
2189         return res;
2190 }
2191 EXPORT_SYMBOL(__xfrm_route_forward);
2192
2193 /* Optimize later using cookies and generation ids. */
2194
2195 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2196 {
2197         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2198          * to "-1" to force all XFRM destinations to get validated by
2199          * dst_ops->check on every use.  We do this because when a
2200          * normal route referenced by an XFRM dst is obsoleted we do
2201          * not go looking around for all parent referencing XFRM dsts
2202          * so that we can invalidate them.  It is just too much work.
2203          * Instead we make the checks here on every use.  For example:
2204          *
2205          *      XFRM dst A --> IPv4 dst X
2206          *
2207          * X is the "xdst->route" of A (X is also the "dst->path" of A
2208          * in this example).  If X is marked obsolete, "A" will not
2209          * notice.  That's what we are validating here via the
2210          * stale_bundle() check.
2211          *
2212          * When a policy's bundle is pruned, we dst_free() the XFRM
2213          * dst which causes it's ->obsolete field to be set to a
2214          * positive non-zero integer.  If an XFRM dst has been pruned
2215          * like this, we want to force a new route lookup.
2216          */
2217         if (dst->obsolete < 0 && !stale_bundle(dst))
2218                 return dst;
2219
2220         return NULL;
2221 }
2222
2223 static int stale_bundle(struct dst_entry *dst)
2224 {
2225         return !xfrm_bundle_ok((struct xfrm_dst *)dst, AF_UNSPEC);
2226 }
2227
2228 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2229 {
2230         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2231                 dst->dev = dev_net(dev)->loopback_dev;
2232                 dev_hold(dst->dev);
2233                 dev_put(dev);
2234         }
2235 }
2236 EXPORT_SYMBOL(xfrm_dst_ifdown);
2237
2238 static void xfrm_link_failure(struct sk_buff *skb)
2239 {
2240         /* Impossible. Such dst must be popped before reaches point of failure. */
2241 }
2242
2243 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2244 {
2245         if (dst) {
2246                 if (dst->obsolete) {
2247                         dst_release(dst);
2248                         dst = NULL;
2249                 }
2250         }
2251         return dst;
2252 }
2253
2254 static void __xfrm_garbage_collect(struct net *net)
2255 {
2256         struct dst_entry *head, *next;
2257
2258         flow_cache_flush();
2259
2260         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2261         head = xfrm_policy_sk_bundles;
2262         xfrm_policy_sk_bundles = NULL;
2263         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2264
2265         while (head) {
2266                 next = head->next;
2267                 dst_free(head);
2268                 head = next;
2269         }
2270 }
2271
2272 static void xfrm_init_pmtu(struct dst_entry *dst)
2273 {
2274         do {
2275                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2276                 u32 pmtu, route_mtu_cached;
2277
2278                 pmtu = dst_mtu(dst->child);
2279                 xdst->child_mtu_cached = pmtu;
2280
2281                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2282
2283                 route_mtu_cached = dst_mtu(xdst->route);
2284                 xdst->route_mtu_cached = route_mtu_cached;
2285
2286                 if (pmtu > route_mtu_cached)
2287                         pmtu = route_mtu_cached;
2288
2289                 dst_metric_set(dst, RTAX_MTU, pmtu);
2290         } while ((dst = dst->next));
2291 }
2292
2293 /* Check that the bundle accepts the flow and its components are
2294  * still valid.
2295  */
2296
2297 static int xfrm_bundle_ok(struct xfrm_dst *first, int family)
2298 {
2299         struct dst_entry *dst = &first->u.dst;
2300         struct xfrm_dst *last;
2301         u32 mtu;
2302
2303         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2304             (dst->dev && !netif_running(dst->dev)))
2305                 return 0;
2306
2307         last = NULL;
2308
2309         do {
2310                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2311
2312                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2313                         return 0;
2314                 if (xdst->xfrm_genid != dst->xfrm->genid)
2315                         return 0;
2316                 if (xdst->num_pols > 0 &&
2317                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2318                         return 0;
2319
2320                 mtu = dst_mtu(dst->child);
2321                 if (xdst->child_mtu_cached != mtu) {
2322                         last = xdst;
2323                         xdst->child_mtu_cached = mtu;
2324                 }
2325
2326                 if (!dst_check(xdst->route, xdst->route_cookie))
2327                         return 0;
2328                 mtu = dst_mtu(xdst->route);
2329                 if (xdst->route_mtu_cached != mtu) {
2330                         last = xdst;
2331                         xdst->route_mtu_cached = mtu;
2332                 }
2333
2334                 dst = dst->child;
2335         } while (dst->xfrm);
2336
2337         if (likely(!last))
2338                 return 1;
2339
2340         mtu = last->child_mtu_cached;
2341         for (;;) {
2342                 dst = &last->u.dst;
2343
2344                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2345                 if (mtu > last->route_mtu_cached)
2346                         mtu = last->route_mtu_cached;
2347                 dst_metric_set(dst, RTAX_MTU, mtu);
2348
2349                 if (last == first)
2350                         break;
2351
2352                 last = (struct xfrm_dst *)last->u.dst.next;
2353                 last->child_mtu_cached = mtu;
2354         }
2355
2356         return 1;
2357 }
2358
2359 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2360 {
2361         return dst_metric_advmss(dst->path);
2362 }
2363
2364 static unsigned int xfrm_default_mtu(const struct dst_entry *dst)
2365 {
2366         return dst_mtu(dst->path);
2367 }
2368
2369 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2370 {
2371         struct net *net;
2372         int err = 0;
2373         if (unlikely(afinfo == NULL))
2374                 return -EINVAL;
2375         if (unlikely(afinfo->family >= NPROTO))
2376                 return -EAFNOSUPPORT;
2377         write_lock_bh(&xfrm_policy_afinfo_lock);
2378         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2379                 err = -ENOBUFS;
2380         else {
2381                 struct dst_ops *dst_ops = afinfo->dst_ops;
2382                 if (likely(dst_ops->kmem_cachep == NULL))
2383                         dst_ops->kmem_cachep = xfrm_dst_cache;
2384                 if (likely(dst_ops->check == NULL))
2385                         dst_ops->check = xfrm_dst_check;
2386                 if (likely(dst_ops->default_advmss == NULL))
2387                         dst_ops->default_advmss = xfrm_default_advmss;
2388                 if (likely(dst_ops->default_mtu == NULL))
2389                         dst_ops->default_mtu = xfrm_default_mtu;
2390                 if (likely(dst_ops->negative_advice == NULL))
2391                         dst_ops->negative_advice = xfrm_negative_advice;
2392                 if (likely(dst_ops->link_failure == NULL))
2393                         dst_ops->link_failure = xfrm_link_failure;
2394                 if (likely(afinfo->garbage_collect == NULL))
2395                         afinfo->garbage_collect = __xfrm_garbage_collect;
2396                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2397         }
2398         write_unlock_bh(&xfrm_policy_afinfo_lock);
2399
2400         rtnl_lock();
2401         for_each_net(net) {
2402                 struct dst_ops *xfrm_dst_ops;
2403
2404                 switch (afinfo->family) {
2405                 case AF_INET:
2406                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2407                         break;
2408 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2409                 case AF_INET6:
2410                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2411                         break;
2412 #endif
2413                 default:
2414                         BUG();
2415                 }
2416                 *xfrm_dst_ops = *afinfo->dst_ops;
2417         }
2418         rtnl_unlock();
2419
2420         return err;
2421 }
2422 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2423
2424 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2425 {
2426         int err = 0;
2427         if (unlikely(afinfo == NULL))
2428                 return -EINVAL;
2429         if (unlikely(afinfo->family >= NPROTO))
2430                 return -EAFNOSUPPORT;
2431         write_lock_bh(&xfrm_policy_afinfo_lock);
2432         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2433                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2434                         err = -EINVAL;
2435                 else {
2436                         struct dst_ops *dst_ops = afinfo->dst_ops;
2437                         xfrm_policy_afinfo[afinfo->family] = NULL;
2438                         dst_ops->kmem_cachep = NULL;
2439                         dst_ops->check = NULL;
2440                         dst_ops->negative_advice = NULL;
2441                         dst_ops->link_failure = NULL;
2442                         afinfo->garbage_collect = NULL;
2443                 }
2444         }
2445         write_unlock_bh(&xfrm_policy_afinfo_lock);
2446         return err;
2447 }
2448 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2449
2450 static void __net_init xfrm_dst_ops_init(struct net *net)
2451 {
2452         struct xfrm_policy_afinfo *afinfo;
2453
2454         read_lock_bh(&xfrm_policy_afinfo_lock);
2455         afinfo = xfrm_policy_afinfo[AF_INET];
2456         if (afinfo)
2457                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2458 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2459         afinfo = xfrm_policy_afinfo[AF_INET6];
2460         if (afinfo)
2461                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2462 #endif
2463         read_unlock_bh(&xfrm_policy_afinfo_lock);
2464 }
2465
2466 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2467 {
2468         struct xfrm_policy_afinfo *afinfo;
2469         if (unlikely(family >= NPROTO))
2470                 return NULL;
2471         read_lock(&xfrm_policy_afinfo_lock);
2472         afinfo = xfrm_policy_afinfo[family];
2473         if (unlikely(!afinfo))
2474                 read_unlock(&xfrm_policy_afinfo_lock);
2475         return afinfo;
2476 }
2477
2478 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2479 {
2480         read_unlock(&xfrm_policy_afinfo_lock);
2481 }
2482
2483 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2484 {
2485         struct net_device *dev = ptr;
2486
2487         switch (event) {
2488         case NETDEV_DOWN:
2489                 __xfrm_garbage_collect(dev_net(dev));
2490         }
2491         return NOTIFY_DONE;
2492 }
2493
2494 static struct notifier_block xfrm_dev_notifier = {
2495         .notifier_call  = xfrm_dev_event,
2496 };
2497
2498 #ifdef CONFIG_XFRM_STATISTICS
2499 static int __net_init xfrm_statistics_init(struct net *net)
2500 {
2501         int rv;
2502
2503         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2504                           sizeof(struct linux_xfrm_mib),
2505                           __alignof__(struct linux_xfrm_mib)) < 0)
2506                 return -ENOMEM;
2507         rv = xfrm_proc_init(net);
2508         if (rv < 0)
2509                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2510         return rv;
2511 }
2512
2513 static void xfrm_statistics_fini(struct net *net)
2514 {
2515         xfrm_proc_fini(net);
2516         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2517 }
2518 #else
2519 static int __net_init xfrm_statistics_init(struct net *net)
2520 {
2521         return 0;
2522 }
2523
2524 static void xfrm_statistics_fini(struct net *net)
2525 {
2526 }
2527 #endif
2528
2529 static int __net_init xfrm_policy_init(struct net *net)
2530 {
2531         unsigned int hmask, sz;
2532         int dir;
2533
2534         if (net_eq(net, &init_net))
2535                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2536                                            sizeof(struct xfrm_dst),
2537                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2538                                            NULL);
2539
2540         hmask = 8 - 1;
2541         sz = (hmask+1) * sizeof(struct hlist_head);
2542
2543         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2544         if (!net->xfrm.policy_byidx)
2545                 goto out_byidx;
2546         net->xfrm.policy_idx_hmask = hmask;
2547
2548         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2549                 struct xfrm_policy_hash *htab;
2550
2551                 net->xfrm.policy_count[dir] = 0;
2552                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2553
2554                 htab = &net->xfrm.policy_bydst[dir];
2555                 htab->table = xfrm_hash_alloc(sz);
2556                 if (!htab->table)
2557                         goto out_bydst;
2558                 htab->hmask = hmask;
2559         }
2560
2561         INIT_LIST_HEAD(&net->xfrm.policy_all);
2562         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2563         if (net_eq(net, &init_net))
2564                 register_netdevice_notifier(&xfrm_dev_notifier);
2565         return 0;
2566
2567 out_bydst:
2568         for (dir--; dir >= 0; dir--) {
2569                 struct xfrm_policy_hash *htab;
2570
2571                 htab = &net->xfrm.policy_bydst[dir];
2572                 xfrm_hash_free(htab->table, sz);
2573         }
2574         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2575 out_byidx:
2576         return -ENOMEM;
2577 }
2578
2579 static void xfrm_policy_fini(struct net *net)
2580 {
2581         struct xfrm_audit audit_info;
2582         unsigned int sz;
2583         int dir;
2584
2585         flush_work(&net->xfrm.policy_hash_work);
2586 #ifdef CONFIG_XFRM_SUB_POLICY
2587         audit_info.loginuid = -1;
2588         audit_info.sessionid = -1;
2589         audit_info.secid = 0;
2590         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2591 #endif
2592         audit_info.loginuid = -1;
2593         audit_info.sessionid = -1;
2594         audit_info.secid = 0;
2595         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2596
2597         WARN_ON(!list_empty(&net->xfrm.policy_all));
2598
2599         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2600                 struct xfrm_policy_hash *htab;
2601
2602                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2603
2604                 htab = &net->xfrm.policy_bydst[dir];
2605                 sz = (htab->hmask + 1);
2606                 WARN_ON(!hlist_empty(htab->table));
2607                 xfrm_hash_free(htab->table, sz);
2608         }
2609
2610         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2611         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2612         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2613 }
2614
2615 static int __net_init xfrm_net_init(struct net *net)
2616 {
2617         int rv;
2618
2619         rv = xfrm_statistics_init(net);
2620         if (rv < 0)
2621                 goto out_statistics;
2622         rv = xfrm_state_init(net);
2623         if (rv < 0)
2624                 goto out_state;
2625         rv = xfrm_policy_init(net);
2626         if (rv < 0)
2627                 goto out_policy;
2628         xfrm_dst_ops_init(net);
2629         rv = xfrm_sysctl_init(net);
2630         if (rv < 0)
2631                 goto out_sysctl;
2632         return 0;
2633
2634 out_sysctl:
2635         xfrm_policy_fini(net);
2636 out_policy:
2637         xfrm_state_fini(net);
2638 out_state:
2639         xfrm_statistics_fini(net);
2640 out_statistics:
2641         return rv;
2642 }
2643
2644 static void __net_exit xfrm_net_exit(struct net *net)
2645 {
2646         xfrm_sysctl_fini(net);
2647         xfrm_policy_fini(net);
2648         xfrm_state_fini(net);
2649         xfrm_statistics_fini(net);
2650 }
2651
2652 static struct pernet_operations __net_initdata xfrm_net_ops = {
2653         .init = xfrm_net_init,
2654         .exit = xfrm_net_exit,
2655 };
2656
2657 void __init xfrm_init(void)
2658 {
2659         register_pernet_subsys(&xfrm_net_ops);
2660         xfrm_input_init();
2661 }
2662
2663 #ifdef CONFIG_AUDITSYSCALL
2664 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2665                                          struct audit_buffer *audit_buf)
2666 {
2667         struct xfrm_sec_ctx *ctx = xp->security;
2668         struct xfrm_selector *sel = &xp->selector;
2669
2670         if (ctx)
2671                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2672                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2673
2674         switch(sel->family) {
2675         case AF_INET:
2676                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2677                 if (sel->prefixlen_s != 32)
2678                         audit_log_format(audit_buf, " src_prefixlen=%d",
2679                                          sel->prefixlen_s);
2680                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2681                 if (sel->prefixlen_d != 32)
2682                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2683                                          sel->prefixlen_d);
2684                 break;
2685         case AF_INET6:
2686                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2687                 if (sel->prefixlen_s != 128)
2688                         audit_log_format(audit_buf, " src_prefixlen=%d",
2689                                          sel->prefixlen_s);
2690                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2691                 if (sel->prefixlen_d != 128)
2692                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2693                                          sel->prefixlen_d);
2694                 break;
2695         }
2696 }
2697
2698 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2699                            uid_t auid, u32 sessionid, u32 secid)
2700 {
2701         struct audit_buffer *audit_buf;
2702
2703         audit_buf = xfrm_audit_start("SPD-add");
2704         if (audit_buf == NULL)
2705                 return;
2706         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2707         audit_log_format(audit_buf, " res=%u", result);
2708         xfrm_audit_common_policyinfo(xp, audit_buf);
2709         audit_log_end(audit_buf);
2710 }
2711 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2712
2713 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2714                               uid_t auid, u32 sessionid, u32 secid)
2715 {
2716         struct audit_buffer *audit_buf;
2717
2718         audit_buf = xfrm_audit_start("SPD-delete");
2719         if (audit_buf == NULL)
2720                 return;
2721         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2722         audit_log_format(audit_buf, " res=%u", result);
2723         xfrm_audit_common_policyinfo(xp, audit_buf);
2724         audit_log_end(audit_buf);
2725 }
2726 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2727 #endif
2728
2729 #ifdef CONFIG_XFRM_MIGRATE
2730 static int xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
2731                                        const struct xfrm_selector *sel_tgt)
2732 {
2733         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2734                 if (sel_tgt->family == sel_cmp->family &&
2735                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2736                                   sel_cmp->family) == 0 &&
2737                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2738                                   sel_cmp->family) == 0 &&
2739                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2740                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2741                         return 1;
2742                 }
2743         } else {
2744                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2745                         return 1;
2746                 }
2747         }
2748         return 0;
2749 }
2750
2751 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
2752                                                      u8 dir, u8 type)
2753 {
2754         struct xfrm_policy *pol, *ret = NULL;
2755         struct hlist_node *entry;
2756         struct hlist_head *chain;
2757         u32 priority = ~0U;
2758
2759         read_lock_bh(&xfrm_policy_lock);
2760         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2761         hlist_for_each_entry(pol, entry, chain, bydst) {
2762                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2763                     pol->type == type) {
2764                         ret = pol;
2765                         priority = ret->priority;
2766                         break;
2767                 }
2768         }
2769         chain = &init_net.xfrm.policy_inexact[dir];
2770         hlist_for_each_entry(pol, entry, chain, bydst) {
2771                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2772                     pol->type == type &&
2773                     pol->priority < priority) {
2774                         ret = pol;
2775                         break;
2776                 }
2777         }
2778
2779         if (ret)
2780                 xfrm_pol_hold(ret);
2781
2782         read_unlock_bh(&xfrm_policy_lock);
2783
2784         return ret;
2785 }
2786
2787 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2788 {
2789         int match = 0;
2790
2791         if (t->mode == m->mode && t->id.proto == m->proto &&
2792             (m->reqid == 0 || t->reqid == m->reqid)) {
2793                 switch (t->mode) {
2794                 case XFRM_MODE_TUNNEL:
2795                 case XFRM_MODE_BEET:
2796                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2797                                           m->old_family) == 0 &&
2798                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2799                                           m->old_family) == 0) {
2800                                 match = 1;
2801                         }
2802                         break;
2803                 case XFRM_MODE_TRANSPORT:
2804                         /* in case of transport mode, template does not store
2805                            any IP addresses, hence we just compare mode and
2806                            protocol */
2807                         match = 1;
2808                         break;
2809                 default:
2810                         break;
2811                 }
2812         }
2813         return match;
2814 }
2815
2816 /* update endpoint address(es) of template(s) */
2817 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2818                                struct xfrm_migrate *m, int num_migrate)
2819 {
2820         struct xfrm_migrate *mp;
2821         int i, j, n = 0;
2822
2823         write_lock_bh(&pol->lock);
2824         if (unlikely(pol->walk.dead)) {
2825                 /* target policy has been deleted */
2826                 write_unlock_bh(&pol->lock);
2827                 return -ENOENT;
2828         }
2829
2830         for (i = 0; i < pol->xfrm_nr; i++) {
2831                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2832                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2833                                 continue;
2834                         n++;
2835                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2836                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2837                                 continue;
2838                         /* update endpoints */
2839                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2840                                sizeof(pol->xfrm_vec[i].id.daddr));
2841                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2842                                sizeof(pol->xfrm_vec[i].saddr));
2843                         pol->xfrm_vec[i].encap_family = mp->new_family;
2844                         /* flush bundles */
2845                         atomic_inc(&pol->genid);
2846                 }
2847         }
2848
2849         write_unlock_bh(&pol->lock);
2850
2851         if (!n)
2852                 return -ENODATA;
2853
2854         return 0;
2855 }
2856
2857 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
2858 {
2859         int i, j;
2860
2861         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2862                 return -EINVAL;
2863
2864         for (i = 0; i < num_migrate; i++) {
2865                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2866                                    m[i].old_family) == 0) &&
2867                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2868                                    m[i].old_family) == 0))
2869                         return -EINVAL;
2870                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2871                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2872                         return -EINVAL;
2873
2874                 /* check if there is any duplicated entry */
2875                 for (j = i + 1; j < num_migrate; j++) {
2876                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2877                                     sizeof(m[i].old_daddr)) &&
2878                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2879                                     sizeof(m[i].old_saddr)) &&
2880                             m[i].proto == m[j].proto &&
2881                             m[i].mode == m[j].mode &&
2882                             m[i].reqid == m[j].reqid &&
2883                             m[i].old_family == m[j].old_family)
2884                                 return -EINVAL;
2885                 }
2886         }
2887
2888         return 0;
2889 }
2890
2891 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
2892                  struct xfrm_migrate *m, int num_migrate,
2893                  struct xfrm_kmaddress *k)
2894 {
2895         int i, err, nx_cur = 0, nx_new = 0;
2896         struct xfrm_policy *pol = NULL;
2897         struct xfrm_state *x, *xc;
2898         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2899         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2900         struct xfrm_migrate *mp;
2901
2902         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2903                 goto out;
2904
2905         /* Stage 1 - find policy */
2906         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2907                 err = -ENOENT;
2908                 goto out;
2909         }
2910
2911         /* Stage 2 - find and update state(s) */
2912         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2913                 if ((x = xfrm_migrate_state_find(mp))) {
2914                         x_cur[nx_cur] = x;
2915                         nx_cur++;
2916                         if ((xc = xfrm_state_migrate(x, mp))) {
2917                                 x_new[nx_new] = xc;
2918                                 nx_new++;
2919                         } else {
2920                                 err = -ENODATA;
2921                                 goto restore_state;
2922                         }
2923                 }
2924         }
2925
2926         /* Stage 3 - update policy */
2927         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2928                 goto restore_state;
2929
2930         /* Stage 4 - delete old state(s) */
2931         if (nx_cur) {
2932                 xfrm_states_put(x_cur, nx_cur);
2933                 xfrm_states_delete(x_cur, nx_cur);
2934         }
2935
2936         /* Stage 5 - announce */
2937         km_migrate(sel, dir, type, m, num_migrate, k);
2938
2939         xfrm_pol_put(pol);
2940
2941         return 0;
2942 out:
2943         return err;
2944
2945 restore_state:
2946         if (pol)
2947                 xfrm_pol_put(pol);
2948         if (nx_cur)
2949                 xfrm_states_put(x_cur, nx_cur);
2950         if (nx_new)
2951                 xfrm_states_delete(x_new, nx_new);
2952
2953         return err;
2954 }
2955 EXPORT_SYMBOL(xfrm_migrate);
2956 #endif