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);
  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         const struct flowi4 *fl4 = &fl->u.ip4;
  63
  64         return  addr_match(&fl4->daddr, &sel->daddr, sel->prefixlen_d) &&
  65                 addr_match(&fl4->saddr, &sel->saddr, sel->prefixlen_s) &&
  66                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  67                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  68                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  69                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  70 }
  71
  72 static inline int
  73 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  74 {
  75         const struct flowi6 *fl6 = &fl->u.ip6;
  76
  77         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  78                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  79                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  80                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  81                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  82                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  83 }
  84
  85 int xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  86                         unsigned short family)
  87 {
  88         switch (family) {
  89         case AF_INET:
  90                 return __xfrm4_selector_match(sel, fl);
  91         case AF_INET6:
  92                 return __xfrm6_selector_match(sel, fl);
  93         }
  94         return 0;
  95 }
  96
  97 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
  98                                                   const xfrm_address_t *saddr,
  99                                                   const xfrm_address_t *daddr,
 100                                                   int family)
 101 {
 102         struct xfrm_policy_afinfo *afinfo;
 103         struct dst_entry *dst;
 104
 105         afinfo = xfrm_policy_get_afinfo(family);
 106         if (unlikely(afinfo == NULL))
 107                 return ERR_PTR(-EAFNOSUPPORT);
 108
 109         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 110
 111         xfrm_policy_put_afinfo(afinfo);
 112
 113         return dst;
 114 }
 115
 116 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 117                                                 xfrm_address_t *prev_saddr,
 118                                                 xfrm_address_t *prev_daddr,
 119                                                 int family)
 120 {
 121         struct net *net = xs_net(x);
 122         xfrm_address_t *saddr = &x->props.saddr;
 123         xfrm_address_t *daddr = &x->id.daddr;
 124         struct dst_entry *dst;
 125
 126         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 127                 saddr = x->coaddr;
 128                 daddr = prev_daddr;
 129         }
 130         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 131                 saddr = prev_saddr;
 132                 daddr = x->coaddr;
 133         }
 134
 135         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 136
 137         if (!IS_ERR(dst)) {
 138                 if (prev_saddr != saddr)
 139                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 140                 if (prev_daddr != daddr)
 141                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 142         }
 143
 144         return dst;
 145 }
 146
 147 static inline unsigned long make_jiffies(long secs)
 148 {
 149         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 150                 return MAX_SCHEDULE_TIMEOUT-1;
 151         else
 152                 return secs*HZ;
 153 }
 154
 155 static void xfrm_policy_timer(unsigned long data)
 156 {
 157         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 158         unsigned long now = get_seconds();
 159         long next = LONG_MAX;
 160         int warn = 0;
 161         int dir;
 162
 163         read_lock(&xp->lock);
 164
 165         if (unlikely(xp->walk.dead))
 166                 goto out;
 167
 168         dir = xfrm_policy_id2dir(xp->index);
 169
 170         if (xp->lft.hard_add_expires_seconds) {
 171                 long tmo = xp->lft.hard_add_expires_seconds +
 172                         xp->curlft.add_time - now;
 173                 if (tmo <= 0)
 174                         goto expired;
 175                 if (tmo < next)
 176                         next = tmo;
 177         }
 178         if (xp->lft.hard_use_expires_seconds) {
 179                 long tmo = xp->lft.hard_use_expires_seconds +
 180                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 181                 if (tmo <= 0)
 182                         goto expired;
 183                 if (tmo < next)
 184                         next = tmo;
 185         }
 186         if (xp->lft.soft_add_expires_seconds) {
 187                 long tmo = xp->lft.soft_add_expires_seconds +
 188                         xp->curlft.add_time - now;
 189                 if (tmo <= 0) {
 190                         warn = 1;
 191                         tmo = XFRM_KM_TIMEOUT;
 192                 }
 193                 if (tmo < next)
 194                         next = tmo;
 195         }
 196         if (xp->lft.soft_use_expires_seconds) {
 197                 long tmo = xp->lft.soft_use_expires_seconds +
 198                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 199                 if (tmo <= 0) {
 200                         warn = 1;
 201                         tmo = XFRM_KM_TIMEOUT;
 202                 }
 203                 if (tmo < next)
 204                         next = tmo;
 205         }
 206
 207         if (warn)
 208                 km_policy_expired(xp, dir, 0, 0);
 209         if (next != LONG_MAX &&
 210             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 211                 xfrm_pol_hold(xp);
 212
 213 out:
 214         read_unlock(&xp->lock);
 215         xfrm_pol_put(xp);
 216         return;
 217
 218 expired:
 219         read_unlock(&xp->lock);
 220         if (!xfrm_policy_delete(xp, dir))
 221                 km_policy_expired(xp, dir, 1, 0);
 222         xfrm_pol_put(xp);
 223 }
 224
 225 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 226 {
 227         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 228
 229         if (unlikely(pol->walk.dead))
 230                 flo = NULL;
 231         else
 232                 xfrm_pol_hold(pol);
 233
 234         return flo;
 235 }
 236
 237 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 238 {
 239         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 240
 241         return !pol->walk.dead;
 242 }
 243
 244 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 245 {
 246         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 247 }
 248
 249 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 250         .get = xfrm_policy_flo_get,
 251         .check = xfrm_policy_flo_check,
 252         .delete = xfrm_policy_flo_delete,
 253 };
 254
 255 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 256  * SPD calls.
 257  */
 258
 259 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 260 {
 261         struct xfrm_policy *policy;
 262
 263         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 264
 265         if (policy) {
 266                 write_pnet(&policy->xp_net, net);
 267                 INIT_LIST_HEAD(&policy->walk.all);
 268                 INIT_HLIST_NODE(&policy->bydst);
 269                 INIT_HLIST_NODE(&policy->byidx);
 270                 rwlock_init(&policy->lock);
 271                 atomic_set(&policy->refcnt, 1);
 272                 setup_timer(&policy->timer, xfrm_policy_timer,
 273                                 (unsigned long)policy);
 274                 policy->flo.ops = &xfrm_policy_fc_ops;
 275         }
 276         return policy;
 277 }
 278 EXPORT_SYMBOL(xfrm_policy_alloc);
 279
 280 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 281
 282 void xfrm_policy_destroy(struct xfrm_policy *policy)
 283 {
 284         BUG_ON(!policy->walk.dead);
 285
 286         if (del_timer(&policy->timer))
 287                 BUG();
 288
 289         security_xfrm_policy_free(policy->security);
 290         kfree(policy);
 291 }
 292 EXPORT_SYMBOL(xfrm_policy_destroy);
 293
 294 /* Rule must be locked. Release descentant resources, announce
 295  * entry dead. The rule must be unlinked from lists to the moment.
 296  */
 297
 298 static void xfrm_policy_kill(struct xfrm_policy *policy)
 299 {
 300         policy->walk.dead = 1;
 301
 302         atomic_inc(&policy->genid);
 303
 304         if (del_timer(&policy->timer))
 305                 xfrm_pol_put(policy);
 306
 307         xfrm_pol_put(policy);
 308 }
 309
 310 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 311
 312 static inline unsigned int idx_hash(struct net *net, u32 index)
 313 {
 314         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 315 }
 316
 317 static struct hlist_head *policy_hash_bysel(struct net *net,
 318                                             const struct xfrm_selector *sel,
 319                                             unsigned short family, int dir)
 320 {
 321         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 322         unsigned int hash = __sel_hash(sel, family, hmask);
 323
 324         return (hash == hmask + 1 ?
 325                 &net->xfrm.policy_inexact[dir] :
 326                 net->xfrm.policy_bydst[dir].table + hash);
 327 }
 328
 329 static struct hlist_head *policy_hash_direct(struct net *net,
 330                                              const xfrm_address_t *daddr,
 331                                              const xfrm_address_t *saddr,
 332                                              unsigned short family, int dir)
 333 {
 334         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 335         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 336
 337         return net->xfrm.policy_bydst[dir].table + hash;
 338 }
 339
 340 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 341                                    struct hlist_head *ndsttable,
 342                                    unsigned int nhashmask)
 343 {
 344         struct hlist_node *entry, *tmp, *entry0 = NULL;
 345         struct xfrm_policy *pol;
 346         unsigned int h0 = 0;
 347
 348 redo:
 349         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 350                 unsigned int h;
 351
 352                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 353                                 pol->family, nhashmask);
 354                 if (!entry0) {
 355                         hlist_del(entry);
 356                         hlist_add_head(&pol->bydst, ndsttable+h);
 357                         h0 = h;
 358                 } else {
 359                         if (h != h0)
 360                                 continue;
 361                         hlist_del(entry);
 362                         hlist_add_after(entry0, &pol->bydst);
 363                 }
 364                 entry0 = entry;
 365         }
 366         if (!hlist_empty(list)) {
 367                 entry0 = NULL;
 368                 goto redo;
 369         }
 370 }
 371
 372 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 373                                    struct hlist_head *nidxtable,
 374                                    unsigned int nhashmask)
 375 {
 376         struct hlist_node *entry, *tmp;
 377         struct xfrm_policy *pol;
 378
 379         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 380                 unsigned int h;
 381
 382                 h = __idx_hash(pol->index, nhashmask);
 383                 hlist_add_head(&pol->byidx, nidxtable+h);
 384         }
 385 }
 386
 387 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 388 {
 389         return ((old_hmask + 1) << 1) - 1;
 390 }
 391
 392 static void xfrm_bydst_resize(struct net *net, int dir)
 393 {
 394         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 395         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 396         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 397         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 398         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 399         int i;
 400
 401         if (!ndst)
 402                 return;
 403
 404         write_lock_bh(&xfrm_policy_lock);
 405
 406         for (i = hmask; i >= 0; i--)
 407                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 408
 409         net->xfrm.policy_bydst[dir].table = ndst;
 410         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 411
 412         write_unlock_bh(&xfrm_policy_lock);
 413
 414         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 415 }
 416
 417 static void xfrm_byidx_resize(struct net *net, int total)
 418 {
 419         unsigned int hmask = net->xfrm.policy_idx_hmask;
 420         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 421         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 422         struct hlist_head *oidx = net->xfrm.policy_byidx;
 423         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 424         int i;
 425
 426         if (!nidx)
 427                 return;
 428
 429         write_lock_bh(&xfrm_policy_lock);
 430
 431         for (i = hmask; i >= 0; i--)
 432                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 433
 434         net->xfrm.policy_byidx = nidx;
 435         net->xfrm.policy_idx_hmask = nhashmask;
 436
 437         write_unlock_bh(&xfrm_policy_lock);
 438
 439         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 440 }
 441
 442 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 443 {
 444         unsigned int cnt = net->xfrm.policy_count[dir];
 445         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 446
 447         if (total)
 448                 *total += cnt;
 449
 450         if ((hmask + 1) < xfrm_policy_hashmax &&
 451             cnt > hmask)
 452                 return 1;
 453
 454         return 0;
 455 }
 456
 457 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 458 {
 459         unsigned int hmask = net->xfrm.policy_idx_hmask;
 460
 461         if ((hmask + 1) < xfrm_policy_hashmax &&
 462             total > hmask)
 463                 return 1;
 464
 465         return 0;
 466 }
 467
 468 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 469 {
 470         read_lock_bh(&xfrm_policy_lock);
 471         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 472         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 473         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 474         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 475         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 476         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 477         si->spdhcnt = net->xfrm.policy_idx_hmask;
 478         si->spdhmcnt = xfrm_policy_hashmax;
 479         read_unlock_bh(&xfrm_policy_lock);
 480 }
 481 EXPORT_SYMBOL(xfrm_spd_getinfo);
 482
 483 static DEFINE_MUTEX(hash_resize_mutex);
 484 static void xfrm_hash_resize(struct work_struct *work)
 485 {
 486         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 487         int dir, total;
 488
 489         mutex_lock(&hash_resize_mutex);
 490
 491         total = 0;
 492         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 493                 if (xfrm_bydst_should_resize(net, dir, &total))
 494                         xfrm_bydst_resize(net, dir);
 495         }
 496         if (xfrm_byidx_should_resize(net, total))
 497                 xfrm_byidx_resize(net, total);
 498
 499         mutex_unlock(&hash_resize_mutex);
 500 }
 501
 502 /* Generate new index... KAME seems to generate them ordered by cost
 503  * of an absolute inpredictability of ordering of rules. This will not pass. */
 504 static u32 xfrm_gen_index(struct net *net, int dir)
 505 {
 506         static u32 idx_generator;
 507
 508         for (;;) {
 509                 struct hlist_node *entry;
 510                 struct hlist_head *list;
 511                 struct xfrm_policy *p;
 512                 u32 idx;
 513                 int found;
 514
 515                 idx = (idx_generator | dir);
 516                 idx_generator += 8;
 517                 if (idx == 0)
 518                         idx = 8;
 519                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 520                 found = 0;
 521                 hlist_for_each_entry(p, entry, list, byidx) {
 522                         if (p->index == idx) {
 523                                 found = 1;
 524                                 break;
 525                         }
 526                 }
 527                 if (!found)
 528                         return idx;
 529         }
 530 }
 531
 532 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 533 {
 534         u32 *p1 = (u32 *) s1;
 535         u32 *p2 = (u32 *) s2;
 536         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 537         int i;
 538
 539         for (i = 0; i < len; i++) {
 540                 if (p1[i] != p2[i])
 541                         return 1;
 542         }
 543
 544         return 0;
 545 }
 546
 547 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 548 {
 549         struct net *net = xp_net(policy);
 550         struct xfrm_policy *pol;
 551         struct xfrm_policy *delpol;
 552         struct hlist_head *chain;
 553         struct hlist_node *entry, *newpos;
 554         u32 mark = policy->mark.v & policy->mark.m;
 555
 556         write_lock_bh(&xfrm_policy_lock);
 557         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 558         delpol = NULL;
 559         newpos = NULL;
 560         hlist_for_each_entry(pol, entry, chain, bydst) {
 561                 if (pol->type == policy->type &&
 562                     !selector_cmp(&pol->selector, &policy->selector) &&
 563                     (mark & pol->mark.m) == pol->mark.v &&
 564                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 565                     !WARN_ON(delpol)) {
 566                         if (excl) {
 567                                 write_unlock_bh(&xfrm_policy_lock);
 568                                 return -EEXIST;
 569                         }
 570                         delpol = pol;
 571                         if (policy->priority > pol->priority)
 572                                 continue;
 573                 } else if (policy->priority >= pol->priority) {
 574                         newpos = &pol->bydst;
 575                         continue;
 576                 }
 577                 if (delpol)
 578                         break;
 579         }
 580         if (newpos)
 581                 hlist_add_after(newpos, &policy->bydst);
 582         else
 583                 hlist_add_head(&policy->bydst, chain);
 584         xfrm_pol_hold(policy);
 585         net->xfrm.policy_count[dir]++;
 586         atomic_inc(&flow_cache_genid);
 587         if (delpol)
 588                 __xfrm_policy_unlink(delpol, dir);
 589         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 590         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 591         policy->curlft.add_time = get_seconds();
 592         policy->curlft.use_time = 0;
 593         if (!mod_timer(&policy->timer, jiffies + HZ))
 594                 xfrm_pol_hold(policy);
 595         list_add(&policy->walk.all, &net->xfrm.policy_all);
 596         write_unlock_bh(&xfrm_policy_lock);
 597
 598         if (delpol)
 599                 xfrm_policy_kill(delpol);
 600         else if (xfrm_bydst_should_resize(net, dir, NULL))
 601                 schedule_work(&net->xfrm.policy_hash_work);
 602
 603         return 0;
 604 }
 605 EXPORT_SYMBOL(xfrm_policy_insert);
 606
 607 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 608                                           int dir, struct xfrm_selector *sel,
 609                                           struct xfrm_sec_ctx *ctx, int delete,
 610                                           int *err)
 611 {
 612         struct xfrm_policy *pol, *ret;
 613         struct hlist_head *chain;
 614         struct hlist_node *entry;
 615
 616         *err = 0;
 617         write_lock_bh(&xfrm_policy_lock);
 618         chain = policy_hash_bysel(net, sel, sel->family, dir);
 619         ret = NULL;
 620         hlist_for_each_entry(pol, entry, chain, bydst) {
 621                 if (pol->type == type &&
 622                     (mark & pol->mark.m) == pol->mark.v &&
 623                     !selector_cmp(sel, &pol->selector) &&
 624                     xfrm_sec_ctx_match(ctx, pol->security)) {
 625                         xfrm_pol_hold(pol);
 626                         if (delete) {
 627                                 *err = security_xfrm_policy_delete(
 628                                                                 pol->security);
 629                                 if (*err) {
 630                                         write_unlock_bh(&xfrm_policy_lock);
 631                                         return pol;
 632                                 }
 633                                 __xfrm_policy_unlink(pol, dir);
 634                         }
 635                         ret = pol;
 636                         break;
 637                 }
 638         }
 639         write_unlock_bh(&xfrm_policy_lock);
 640
 641         if (ret && delete)
 642                 xfrm_policy_kill(ret);
 643         return ret;
 644 }
 645 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 646
 647 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 648                                      int dir, u32 id, int delete, int *err)
 649 {
 650         struct xfrm_policy *pol, *ret;
 651         struct hlist_head *chain;
 652         struct hlist_node *entry;
 653
 654         *err = -ENOENT;
 655         if (xfrm_policy_id2dir(id) != dir)
 656                 return NULL;
 657
 658         *err = 0;
 659         write_lock_bh(&xfrm_policy_lock);
 660         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 661         ret = NULL;
 662         hlist_for_each_entry(pol, entry, chain, byidx) {
 663                 if (pol->type == type && pol->index == id &&
 664                     (mark & pol->mark.m) == pol->mark.v) {
 665                         xfrm_pol_hold(pol);
 666                         if (delete) {
 667                                 *err = security_xfrm_policy_delete(
 668                                                                 pol->security);
 669                                 if (*err) {
 670                                         write_unlock_bh(&xfrm_policy_lock);
 671                                         return pol;
 672                                 }
 673                                 __xfrm_policy_unlink(pol, dir);
 674                         }
 675                         ret = pol;
 676                         break;
 677                 }
 678         }
 679         write_unlock_bh(&xfrm_policy_lock);
 680
 681         if (ret && delete)
 682                 xfrm_policy_kill(ret);
 683         return ret;
 684 }
 685 EXPORT_SYMBOL(xfrm_policy_byid);
 686
 687 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 688 static inline int
 689 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 690 {
 691         int dir, err = 0;
 692
 693         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 694                 struct xfrm_policy *pol;
 695                 struct hlist_node *entry;
 696                 int i;
 697
 698                 hlist_for_each_entry(pol, entry,
 699                                      &net->xfrm.policy_inexact[dir], bydst) {
 700                         if (pol->type != type)
 701                                 continue;
 702                         err = security_xfrm_policy_delete(pol->security);
 703                         if (err) {
 704                                 xfrm_audit_policy_delete(pol, 0,
 705                                                          audit_info->loginuid,
 706                                                          audit_info->sessionid,
 707                                                          audit_info->secid);
 708                                 return err;
 709                         }
 710                 }
 711                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 712                         hlist_for_each_entry(pol, entry,
 713                                              net->xfrm.policy_bydst[dir].table + i,
 714                                              bydst) {
 715                                 if (pol->type != type)
 716                                         continue;
 717                                 err = security_xfrm_policy_delete(
 718                                                                 pol->security);
 719                                 if (err) {
 720                                         xfrm_audit_policy_delete(pol, 0,
 721                                                         audit_info->loginuid,
 722                                                         audit_info->sessionid,
 723                                                         audit_info->secid);
 724                                         return err;
 725                                 }
 726                         }
 727                 }
 728         }
 729         return err;
 730 }
 731 #else
 732 static inline int
 733 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 734 {
 735         return 0;
 736 }
 737 #endif
 738
 739 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 740 {
 741         int dir, err = 0, cnt = 0;
 742
 743         write_lock_bh(&xfrm_policy_lock);
 744
 745         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 746         if (err)
 747                 goto out;
 748
 749         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 750                 struct xfrm_policy *pol;
 751                 struct hlist_node *entry;
 752                 int i;
 753
 754         again1:
 755                 hlist_for_each_entry(pol, entry,
 756                                      &net->xfrm.policy_inexact[dir], bydst) {
 757                         if (pol->type != type)
 758                                 continue;
 759                         __xfrm_policy_unlink(pol, dir);
 760                         write_unlock_bh(&xfrm_policy_lock);
 761                         cnt++;
 762
 763                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 764                                                  audit_info->sessionid,
 765                                                  audit_info->secid);
 766
 767                         xfrm_policy_kill(pol);
 768
 769                         write_lock_bh(&xfrm_policy_lock);
 770                         goto again1;
 771                 }
 772
 773                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 774         again2:
 775                         hlist_for_each_entry(pol, entry,
 776                                              net->xfrm.policy_bydst[dir].table + i,
 777                                              bydst) {
 778                                 if (pol->type != type)
 779                                         continue;
 780                                 __xfrm_policy_unlink(pol, dir);
 781                                 write_unlock_bh(&xfrm_policy_lock);
 782                                 cnt++;
 783
 784                                 xfrm_audit_policy_delete(pol, 1,
 785                                                          audit_info->loginuid,
 786                                                          audit_info->sessionid,
 787                                                          audit_info->secid);
 788                                 xfrm_policy_kill(pol);
 789
 790                                 write_lock_bh(&xfrm_policy_lock);
 791                                 goto again2;
 792                         }
 793                 }
 794
 795         }
 796         if (!cnt)
 797                 err = -ESRCH;
 798 out:
 799         write_unlock_bh(&xfrm_policy_lock);
 800         return err;
 801 }
 802 EXPORT_SYMBOL(xfrm_policy_flush);
 803
 804 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 805                      int (*func)(struct xfrm_policy *, int, int, void*),
 806                      void *data)
 807 {
 808         struct xfrm_policy *pol;
 809         struct xfrm_policy_walk_entry *x;
 810         int error = 0;
 811
 812         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 813             walk->type != XFRM_POLICY_TYPE_ANY)
 814                 return -EINVAL;
 815
 816         if (list_empty(&walk->walk.all) && walk->seq != 0)
 817                 return 0;
 818
 819         write_lock_bh(&xfrm_policy_lock);
 820         if (list_empty(&walk->walk.all))
 821                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 822         else
 823                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 824         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 825                 if (x->dead)
 826                         continue;
 827                 pol = container_of(x, struct xfrm_policy, walk);
 828                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 829                     walk->type != pol->type)
 830                         continue;
 831                 error = func(pol, xfrm_policy_id2dir(pol->index),
 832                              walk->seq, data);
 833                 if (error) {
 834                         list_move_tail(&walk->walk.all, &x->all);
 835                         goto out;
 836                 }
 837                 walk->seq++;
 838         }
 839         if (walk->seq == 0) {
 840                 error = -ENOENT;
 841                 goto out;
 842         }
 843         list_del_init(&walk->walk.all);
 844 out:
 845         write_unlock_bh(&xfrm_policy_lock);
 846         return error;
 847 }
 848 EXPORT_SYMBOL(xfrm_policy_walk);
 849
 850 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 851 {
 852         INIT_LIST_HEAD(&walk->walk.all);
 853         walk->walk.dead = 1;
 854         walk->type = type;
 855         walk->seq = 0;
 856 }
 857 EXPORT_SYMBOL(xfrm_policy_walk_init);
 858
 859 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 860 {
 861         if (list_empty(&walk->walk.all))
 862                 return;
 863
 864         write_lock_bh(&xfrm_policy_lock);
 865         list_del(&walk->walk.all);
 866         write_unlock_bh(&xfrm_policy_lock);
 867 }
 868 EXPORT_SYMBOL(xfrm_policy_walk_done);
 869
 870 /*
 871  * Find policy to apply to this flow.
 872  *
 873  * Returns 0 if policy found, else an -errno.
 874  */
 875 static int xfrm_policy_match(const struct xfrm_policy *pol,
 876                              const struct flowi *fl,
 877                              u8 type, u16 family, int dir)
 878 {
 879         const struct xfrm_selector *sel = &pol->selector;
 880         int match, ret = -ESRCH;
 881
 882         if (pol->family != family ||
 883             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
 884             pol->type != type)
 885                 return ret;
 886
 887         match = xfrm_selector_match(sel, fl, family);
 888         if (match)
 889                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
 890                                                   dir);
 891
 892         return ret;
 893 }
 894
 895 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 896                                                      const struct flowi *fl,
 897                                                      u16 family, u8 dir)
 898 {
 899         int err;
 900         struct xfrm_policy *pol, *ret;
 901         const xfrm_address_t *daddr, *saddr;
 902         struct hlist_node *entry;
 903         struct hlist_head *chain;
 904         u32 priority = ~0U;
 905
 906         daddr = xfrm_flowi_daddr(fl, family);
 907         saddr = xfrm_flowi_saddr(fl, family);
 908         if (unlikely(!daddr || !saddr))
 909                 return NULL;
 910
 911         read_lock_bh(&xfrm_policy_lock);
 912         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 913         ret = NULL;
 914         hlist_for_each_entry(pol, entry, chain, bydst) {
 915                 err = xfrm_policy_match(pol, fl, type, family, dir);
 916                 if (err) {
 917                         if (err == -ESRCH)
 918                                 continue;
 919                         else {
 920                                 ret = ERR_PTR(err);
 921                                 goto fail;
 922                         }
 923                 } else {
 924                         ret = pol;
 925                         priority = ret->priority;
 926                         break;
 927                 }
 928         }
 929         chain = &net->xfrm.policy_inexact[dir];
 930         hlist_for_each_entry(pol, entry, chain, bydst) {
 931                 err = xfrm_policy_match(pol, fl, type, family, dir);
 932                 if (err) {
 933                         if (err == -ESRCH)
 934                                 continue;
 935                         else {
 936                                 ret = ERR_PTR(err);
 937                                 goto fail;
 938                         }
 939                 } else if (pol->priority < priority) {
 940                         ret = pol;
 941                         break;
 942                 }
 943         }
 944         if (ret)
 945                 xfrm_pol_hold(ret);
 946 fail:
 947         read_unlock_bh(&xfrm_policy_lock);
 948
 949         return ret;
 950 }
 951
 952 static struct xfrm_policy *
 953 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
 954 {
 955 #ifdef CONFIG_XFRM_SUB_POLICY
 956         struct xfrm_policy *pol;
 957
 958         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
 959         if (pol != NULL)
 960                 return pol;
 961 #endif
 962         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 963 }
 964
 965 static struct flow_cache_object *
 966 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
 967                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
 968 {
 969         struct xfrm_policy *pol;
 970
 971         if (old_obj)
 972                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
 973
 974         pol = __xfrm_policy_lookup(net, fl, family, dir);
 975         if (IS_ERR_OR_NULL(pol))
 976                 return ERR_CAST(pol);
 977
 978         /* Resolver returns two references:
 979          * one for cache and one for caller of flow_cache_lookup() */
 980         xfrm_pol_hold(pol);
 981
 982         return &pol->flo;
 983 }
 984
 985 static inline int policy_to_flow_dir(int dir)
 986 {
 987         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 988             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 989             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 990                 return dir;
 991         switch (dir) {
 992         default:
 993         case XFRM_POLICY_IN:
 994                 return FLOW_DIR_IN;
 995         case XFRM_POLICY_OUT:
 996                 return FLOW_DIR_OUT;
 997         case XFRM_POLICY_FWD:
 998                 return FLOW_DIR_FWD;
 999         }
1000 }
1001
1002 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1003                                                  const struct flowi *fl)
1004 {
1005         struct xfrm_policy *pol;
1006
1007         read_lock_bh(&xfrm_policy_lock);
1008         if ((pol = sk->sk_policy[dir]) != NULL) {
1009                 int match = xfrm_selector_match(&pol->selector, fl,
1010                                                 sk->sk_family);
1011                 int err = 0;
1012
1013                 if (match) {
1014                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1015                                 pol = NULL;
1016                                 goto out;
1017                         }
1018                         err = security_xfrm_policy_lookup(pol->security,
1019                                                       fl->flowi_secid,
1020                                                       policy_to_flow_dir(dir));
1021                         if (!err)
1022                                 xfrm_pol_hold(pol);
1023                         else if (err == -ESRCH)
1024                                 pol = NULL;
1025                         else
1026                                 pol = ERR_PTR(err);
1027                 } else
1028                         pol = NULL;
1029         }
1030 out:
1031         read_unlock_bh(&xfrm_policy_lock);
1032         return pol;
1033 }
1034
1035 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1036 {
1037         struct net *net = xp_net(pol);
1038         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1039                                                      pol->family, dir);
1040
1041         list_add(&pol->walk.all, &net->xfrm.policy_all);
1042         hlist_add_head(&pol->bydst, chain);
1043         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1044         net->xfrm.policy_count[dir]++;
1045         xfrm_pol_hold(pol);
1046
1047         if (xfrm_bydst_should_resize(net, dir, NULL))
1048                 schedule_work(&net->xfrm.policy_hash_work);
1049 }
1050
1051 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1052                                                 int dir)
1053 {
1054         struct net *net = xp_net(pol);
1055
1056         if (hlist_unhashed(&pol->bydst))
1057                 return NULL;
1058
1059         hlist_del(&pol->bydst);
1060         hlist_del(&pol->byidx);
1061         list_del(&pol->walk.all);
1062         net->xfrm.policy_count[dir]--;
1063
1064         return pol;
1065 }
1066
1067 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1068 {
1069         write_lock_bh(&xfrm_policy_lock);
1070         pol = __xfrm_policy_unlink(pol, dir);
1071         write_unlock_bh(&xfrm_policy_lock);
1072         if (pol) {
1073                 xfrm_policy_kill(pol);
1074                 return 0;
1075         }
1076         return -ENOENT;
1077 }
1078 EXPORT_SYMBOL(xfrm_policy_delete);
1079
1080 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1081 {
1082         struct net *net = xp_net(pol);
1083         struct xfrm_policy *old_pol;
1084
1085 #ifdef CONFIG_XFRM_SUB_POLICY
1086         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1087                 return -EINVAL;
1088 #endif
1089
1090         write_lock_bh(&xfrm_policy_lock);
1091         old_pol = sk->sk_policy[dir];
1092         sk->sk_policy[dir] = pol;
1093         if (pol) {
1094                 pol->curlft.add_time = get_seconds();
1095                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1096                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1097         }
1098         if (old_pol)
1099                 /* Unlinking succeeds always. This is the only function
1100                  * allowed to delete or replace socket policy.
1101                  */
1102                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1103         write_unlock_bh(&xfrm_policy_lock);
1104
1105         if (old_pol) {
1106                 xfrm_policy_kill(old_pol);
1107         }
1108         return 0;
1109 }
1110
1111 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1112 {
1113         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1114
1115         if (newp) {
1116                 newp->selector = old->selector;
1117                 if (security_xfrm_policy_clone(old->security,
1118                                                &newp->security)) {
1119                         kfree(newp);
1120                         return NULL;  /* ENOMEM */
1121                 }
1122                 newp->lft = old->lft;
1123                 newp->curlft = old->curlft;
1124                 newp->mark = old->mark;
1125                 newp->action = old->action;
1126                 newp->flags = old->flags;
1127                 newp->xfrm_nr = old->xfrm_nr;
1128                 newp->index = old->index;
1129                 newp->type = old->type;
1130                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1131                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1132                 write_lock_bh(&xfrm_policy_lock);
1133                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1134                 write_unlock_bh(&xfrm_policy_lock);
1135                 xfrm_pol_put(newp);
1136         }
1137         return newp;
1138 }
1139
1140 int __xfrm_sk_clone_policy(struct sock *sk)
1141 {
1142         struct xfrm_policy *p0 = sk->sk_policy[0],
1143                            *p1 = sk->sk_policy[1];
1144
1145         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1146         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1147                 return -ENOMEM;
1148         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1149                 return -ENOMEM;
1150         return 0;
1151 }
1152
1153 static int
1154 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1155                unsigned short family)
1156 {
1157         int err;
1158         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1159
1160         if (unlikely(afinfo == NULL))
1161                 return -EINVAL;
1162         err = afinfo->get_saddr(net, local, remote);
1163         xfrm_policy_put_afinfo(afinfo);
1164         return err;
1165 }
1166
1167 /* Resolve list of templates for the flow, given policy. */
1168
1169 static int
1170 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1171                       struct xfrm_state **xfrm, unsigned short family)
1172 {
1173         struct net *net = xp_net(policy);
1174         int nx;
1175         int i, error;
1176         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1177         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1178         xfrm_address_t tmp;
1179
1180         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1181                 struct xfrm_state *x;
1182                 xfrm_address_t *remote = daddr;
1183                 xfrm_address_t *local  = saddr;
1184                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1185
1186                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1187                     tmpl->mode == XFRM_MODE_BEET) {
1188                         remote = &tmpl->id.daddr;
1189                         local = &tmpl->saddr;
1190                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1191                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1192                                 if (error)
1193                                         goto fail;
1194                                 local = &tmp;
1195                         }
1196                 }
1197
1198                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1199
1200                 if (x && x->km.state == XFRM_STATE_VALID) {
1201                         xfrm[nx++] = x;
1202                         daddr = remote;
1203                         saddr = local;
1204                         continue;
1205                 }
1206                 if (x) {
1207                         error = (x->km.state == XFRM_STATE_ERROR ?
1208                                  -EINVAL : -EAGAIN);
1209                         xfrm_state_put(x);
1210                 }
1211                 else if (error == -ESRCH)
1212                         error = -EAGAIN;
1213
1214                 if (!tmpl->optional)
1215                         goto fail;
1216         }
1217         return nx;
1218
1219 fail:
1220         for (nx--; nx>=0; nx--)
1221                 xfrm_state_put(xfrm[nx]);
1222         return error;
1223 }
1224
1225 static int
1226 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1227                   struct xfrm_state **xfrm, unsigned short family)
1228 {
1229         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1230         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1231         int cnx = 0;
1232         int error;
1233         int ret;
1234         int i;
1235
1236         for (i = 0; i < npols; i++) {
1237                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1238                         error = -ENOBUFS;
1239                         goto fail;
1240                 }
1241
1242                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1243                 if (ret < 0) {
1244                         error = ret;
1245                         goto fail;
1246                 } else
1247                         cnx += ret;
1248         }
1249
1250         /* found states are sorted for outbound processing */
1251         if (npols > 1)
1252                 xfrm_state_sort(xfrm, tpp, cnx, family);
1253
1254         return cnx;
1255
1256  fail:
1257         for (cnx--; cnx>=0; cnx--)
1258                 xfrm_state_put(tpp[cnx]);
1259         return error;
1260
1261 }
1262
1263 /* Check that the bundle accepts the flow and its components are
1264  * still valid.
1265  */
1266
1267 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1268 {
1269         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1270         int tos;
1271
1272         if (!afinfo)
1273                 return -EINVAL;
1274
1275         tos = afinfo->get_tos(fl);
1276
1277         xfrm_policy_put_afinfo(afinfo);
1278
1279         return tos;
1280 }
1281
1282 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1283 {
1284         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1285         struct dst_entry *dst = &xdst->u.dst;
1286
1287         if (xdst->route == NULL) {
1288                 /* Dummy bundle - if it has xfrms we were not
1289                  * able to build bundle as template resolution failed.
1290                  * It means we need to try again resolving. */
1291                 if (xdst->num_xfrms > 0)
1292                         return NULL;
1293         } else {
1294                 /* Real bundle */
1295                 if (stale_bundle(dst))
1296                         return NULL;
1297         }
1298
1299         dst_hold(dst);
1300         return flo;
1301 }
1302
1303 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1304 {
1305         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1306         struct dst_entry *dst = &xdst->u.dst;
1307
1308         if (!xdst->route)
1309                 return 0;
1310         if (stale_bundle(dst))
1311                 return 0;
1312
1313         return 1;
1314 }
1315
1316 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1317 {
1318         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1319         struct dst_entry *dst = &xdst->u.dst;
1320
1321         dst_free(dst);
1322 }
1323
1324 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1325         .get = xfrm_bundle_flo_get,
1326         .check = xfrm_bundle_flo_check,
1327         .delete = xfrm_bundle_flo_delete,
1328 };
1329
1330 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1331 {
1332         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1333         struct dst_ops *dst_ops;
1334         struct xfrm_dst *xdst;
1335
1336         if (!afinfo)
1337                 return ERR_PTR(-EINVAL);
1338
1339         switch (family) {
1340         case AF_INET:
1341                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1342                 break;
1343 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1344         case AF_INET6:
1345                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1346                 break;
1347 #endif
1348         default:
1349                 BUG();
1350         }
1351         xdst = dst_alloc(dst_ops, NULL, 0, 0, 0);
1352
1353         if (likely(xdst)) {
1354                 memset(&xdst->u.rt6.rt6i_table, 0,
1355                         sizeof(*xdst) - sizeof(struct dst_entry));
1356                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1357         } else
1358                 xdst = ERR_PTR(-ENOBUFS);
1359
1360         xfrm_policy_put_afinfo(afinfo);
1361
1362         return xdst;
1363 }
1364
1365 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1366                                  int nfheader_len)
1367 {
1368         struct xfrm_policy_afinfo *afinfo =
1369                 xfrm_policy_get_afinfo(dst->ops->family);
1370         int err;
1371
1372         if (!afinfo)
1373                 return -EINVAL;
1374
1375         err = afinfo->init_path(path, dst, nfheader_len);
1376
1377         xfrm_policy_put_afinfo(afinfo);
1378
1379         return err;
1380 }
1381
1382 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1383                                 const struct flowi *fl)
1384 {
1385         struct xfrm_policy_afinfo *afinfo =
1386                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1387         int err;
1388
1389         if (!afinfo)
1390                 return -EINVAL;
1391
1392         err = afinfo->fill_dst(xdst, dev, fl);
1393
1394         xfrm_policy_put_afinfo(afinfo);
1395
1396         return err;
1397 }
1398
1399
1400 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1401  * all the metrics... Shortly, bundle a bundle.
1402  */
1403
1404 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1405                                             struct xfrm_state **xfrm, int nx,
1406                                             const struct flowi *fl,
1407                                             struct dst_entry *dst)
1408 {
1409         struct net *net = xp_net(policy);
1410         unsigned long now = jiffies;
1411         struct net_device *dev;
1412         struct xfrm_mode *inner_mode;
1413         struct dst_entry *dst_prev = NULL;
1414         struct dst_entry *dst0 = NULL;
1415         int i = 0;
1416         int err;
1417         int header_len = 0;
1418         int nfheader_len = 0;
1419         int trailer_len = 0;
1420         int tos;
1421         int family = policy->selector.family;
1422         xfrm_address_t saddr, daddr;
1423
1424         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1425
1426         tos = xfrm_get_tos(fl, family);
1427         err = tos;
1428         if (tos < 0)
1429                 goto put_states;
1430
1431         dst_hold(dst);
1432
1433         for (; i < nx; i++) {
1434                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1435                 struct dst_entry *dst1 = &xdst->u.dst;
1436
1437                 err = PTR_ERR(xdst);
1438                 if (IS_ERR(xdst)) {
1439                         dst_release(dst);
1440                         goto put_states;
1441                 }
1442
1443                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1444                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1445                                                         xfrm_af2proto(family));
1446                         if (!inner_mode) {
1447                                 err = -EAFNOSUPPORT;
1448                                 dst_release(dst);
1449                                 goto put_states;
1450                         }
1451                 } else
1452                         inner_mode = xfrm[i]->inner_mode;
1453
1454                 if (!dst_prev)
1455                         dst0 = dst1;
1456                 else {
1457                         dst_prev->child = dst_clone(dst1);
1458                         dst1->flags |= DST_NOHASH;
1459                 }
1460
1461                 xdst->route = dst;
1462                 dst_copy_metrics(dst1, dst);
1463
1464                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1465                         family = xfrm[i]->props.family;
1466                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1467                                               family);
1468                         err = PTR_ERR(dst);
1469                         if (IS_ERR(dst))
1470                                 goto put_states;
1471                 } else
1472                         dst_hold(dst);
1473
1474                 dst1->xfrm = xfrm[i];
1475                 xdst->xfrm_genid = xfrm[i]->genid;
1476
1477                 dst1->obsolete = -1;
1478                 dst1->flags |= DST_HOST;
1479                 dst1->lastuse = now;
1480
1481                 dst1->input = dst_discard;
1482                 dst1->output = inner_mode->afinfo->output;
1483
1484                 dst1->next = dst_prev;
1485                 dst_prev = dst1;
1486
1487                 header_len += xfrm[i]->props.header_len;
1488                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1489                         nfheader_len += xfrm[i]->props.header_len;
1490                 trailer_len += xfrm[i]->props.trailer_len;
1491         }
1492
1493         dst_prev->child = dst;
1494         dst0->path = dst;
1495
1496         err = -ENODEV;
1497         dev = dst->dev;
1498         if (!dev)
1499                 goto free_dst;
1500
1501         /* Copy neighbour for reachability confirmation */
1502         dst_set_neighbour(dst0, neigh_clone(dst_get_neighbour(dst)));
1503
1504         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1505         xfrm_init_pmtu(dst_prev);
1506
1507         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1508                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1509
1510                 err = xfrm_fill_dst(xdst, dev, fl);
1511                 if (err)
1512                         goto free_dst;
1513
1514                 dst_prev->header_len = header_len;
1515                 dst_prev->trailer_len = trailer_len;
1516                 header_len -= xdst->u.dst.xfrm->props.header_len;
1517                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1518         }
1519
1520 out:
1521         return dst0;
1522
1523 put_states:
1524         for (; i < nx; i++)
1525                 xfrm_state_put(xfrm[i]);
1526 free_dst:
1527         if (dst0)
1528                 dst_free(dst0);
1529         dst0 = ERR_PTR(err);
1530         goto out;
1531 }
1532
1533 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1534                                 struct xfrm_policy **pols,
1535                                 int *num_pols, int *num_xfrms)
1536 {
1537         int i;
1538
1539         if (*num_pols == 0 || !pols[0]) {
1540                 *num_pols = 0;
1541                 *num_xfrms = 0;
1542                 return 0;
1543         }
1544         if (IS_ERR(pols[0]))
1545                 return PTR_ERR(pols[0]);
1546
1547         *num_xfrms = pols[0]->xfrm_nr;
1548
1549 #ifdef CONFIG_XFRM_SUB_POLICY
1550         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1551             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1552                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1553                                                     XFRM_POLICY_TYPE_MAIN,
1554                                                     fl, family,
1555                                                     XFRM_POLICY_OUT);
1556                 if (pols[1]) {
1557                         if (IS_ERR(pols[1])) {
1558                                 xfrm_pols_put(pols, *num_pols);
1559                                 return PTR_ERR(pols[1]);
1560                         }
1561                         (*num_pols) ++;
1562                         (*num_xfrms) += pols[1]->xfrm_nr;
1563                 }
1564         }
1565 #endif
1566         for (i = 0; i < *num_pols; i++) {
1567                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1568                         *num_xfrms = -1;
1569                         break;
1570                 }
1571         }
1572
1573         return 0;
1574
1575 }
1576
1577 static struct xfrm_dst *
1578 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1579                                const struct flowi *fl, u16 family,
1580                                struct dst_entry *dst_orig)
1581 {
1582         struct net *net = xp_net(pols[0]);
1583         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1584         struct dst_entry *dst;
1585         struct xfrm_dst *xdst;
1586         int err;
1587
1588         /* Try to instantiate a bundle */
1589         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1590         if (err <= 0) {
1591                 if (err != 0 && err != -EAGAIN)
1592                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1593                 return ERR_PTR(err);
1594         }
1595
1596         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1597         if (IS_ERR(dst)) {
1598                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1599                 return ERR_CAST(dst);
1600         }
1601
1602         xdst = (struct xfrm_dst *)dst;
1603         xdst->num_xfrms = err;
1604         xdst->num_pols = num_pols;
1605         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1606         xdst->policy_genid = atomic_read(&pols[0]->genid);
1607
1608         return xdst;
1609 }
1610
1611 static struct flow_cache_object *
1612 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1613                    struct flow_cache_object *oldflo, void *ctx)
1614 {
1615         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1616         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1617         struct xfrm_dst *xdst, *new_xdst;
1618         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1619
1620         /* Check if the policies from old bundle are usable */
1621         xdst = NULL;
1622         if (oldflo) {
1623                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1624                 num_pols = xdst->num_pols;
1625                 num_xfrms = xdst->num_xfrms;
1626                 pol_dead = 0;
1627                 for (i = 0; i < num_pols; i++) {
1628                         pols[i] = xdst->pols[i];
1629                         pol_dead |= pols[i]->walk.dead;
1630                 }
1631                 if (pol_dead) {
1632                         dst_free(&xdst->u.dst);
1633                         xdst = NULL;
1634                         num_pols = 0;
1635                         num_xfrms = 0;
1636                         oldflo = NULL;
1637                 }
1638         }
1639
1640         /* Resolve policies to use if we couldn't get them from
1641          * previous cache entry */
1642         if (xdst == NULL) {
1643                 num_pols = 1;
1644                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1645                 err = xfrm_expand_policies(fl, family, pols,
1646                                            &num_pols, &num_xfrms);
1647                 if (err < 0)
1648                         goto inc_error;
1649                 if (num_pols == 0)
1650                         return NULL;
1651                 if (num_xfrms <= 0)
1652                         goto make_dummy_bundle;
1653         }
1654
1655         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1656         if (IS_ERR(new_xdst)) {
1657                 err = PTR_ERR(new_xdst);
1658                 if (err != -EAGAIN)
1659                         goto error;
1660                 if (oldflo == NULL)
1661                         goto make_dummy_bundle;
1662                 dst_hold(&xdst->u.dst);
1663                 return oldflo;
1664         } else if (new_xdst == NULL) {
1665                 num_xfrms = 0;
1666                 if (oldflo == NULL)
1667                         goto make_dummy_bundle;
1668                 xdst->num_xfrms = 0;
1669                 dst_hold(&xdst->u.dst);
1670                 return oldflo;
1671         }
1672
1673         /* Kill the previous bundle */
1674         if (xdst) {
1675                 /* The policies were stolen for newly generated bundle */
1676                 xdst->num_pols = 0;
1677                 dst_free(&xdst->u.dst);
1678         }
1679
1680         /* Flow cache does not have reference, it dst_free()'s,
1681          * but we do need to return one reference for original caller */
1682         dst_hold(&new_xdst->u.dst);
1683         return &new_xdst->flo;
1684
1685 make_dummy_bundle:
1686         /* We found policies, but there's no bundles to instantiate:
1687          * either because the policy blocks, has no transformations or
1688          * we could not build template (no xfrm_states).*/
1689         xdst = xfrm_alloc_dst(net, family);
1690         if (IS_ERR(xdst)) {
1691                 xfrm_pols_put(pols, num_pols);
1692                 return ERR_CAST(xdst);
1693         }
1694         xdst->num_pols = num_pols;
1695         xdst->num_xfrms = num_xfrms;
1696         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1697
1698         dst_hold(&xdst->u.dst);
1699         return &xdst->flo;
1700
1701 inc_error:
1702         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1703 error:
1704         if (xdst != NULL)
1705                 dst_free(&xdst->u.dst);
1706         else
1707                 xfrm_pols_put(pols, num_pols);
1708         return ERR_PTR(err);
1709 }
1710
1711 static struct dst_entry *make_blackhole(struct net *net, u16 family,
1712                                         struct dst_entry *dst_orig)
1713 {
1714         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1715         struct dst_entry *ret;
1716
1717         if (!afinfo) {
1718                 dst_release(dst_orig);
1719                 return ERR_PTR(-EINVAL);
1720         } else {
1721                 ret = afinfo->blackhole_route(net, dst_orig);
1722         }
1723         xfrm_policy_put_afinfo(afinfo);
1724
1725         return ret;
1726 }
1727
1728 /* Main function: finds/creates a bundle for given flow.
1729  *
1730  * At the moment we eat a raw IP route. Mostly to speed up lookups
1731  * on interfaces with disabled IPsec.
1732  */
1733 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
1734                               const struct flowi *fl,
1735                               struct sock *sk, int flags)
1736 {
1737         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1738         struct flow_cache_object *flo;
1739         struct xfrm_dst *xdst;
1740         struct dst_entry *dst, *route;
1741         u16 family = dst_orig->ops->family;
1742         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1743         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1744
1745 restart:
1746         dst = NULL;
1747         xdst = NULL;
1748         route = NULL;
1749
1750         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1751                 num_pols = 1;
1752                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1753                 err = xfrm_expand_policies(fl, family, pols,
1754                                            &num_pols, &num_xfrms);
1755                 if (err < 0)
1756                         goto dropdst;
1757
1758                 if (num_pols) {
1759                         if (num_xfrms <= 0) {
1760                                 drop_pols = num_pols;
1761                                 goto no_transform;
1762                         }
1763
1764                         xdst = xfrm_resolve_and_create_bundle(
1765                                         pols, num_pols, fl,
1766                                         family, dst_orig);
1767                         if (IS_ERR(xdst)) {
1768                                 xfrm_pols_put(pols, num_pols);
1769                                 err = PTR_ERR(xdst);
1770                                 goto dropdst;
1771                         } else if (xdst == NULL) {
1772                                 num_xfrms = 0;
1773                                 drop_pols = num_pols;
1774                                 goto no_transform;
1775                         }
1776
1777                         dst_hold(&xdst->u.dst);
1778
1779                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1780                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1781                         xfrm_policy_sk_bundles = &xdst->u.dst;
1782                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1783
1784                         route = xdst->route;
1785                 }
1786         }
1787
1788         if (xdst == NULL) {
1789                 /* To accelerate a bit...  */
1790                 if ((dst_orig->flags & DST_NOXFRM) ||
1791                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1792                         goto nopol;
1793
1794                 flo = flow_cache_lookup(net, fl, family, dir,
1795                                         xfrm_bundle_lookup, dst_orig);
1796                 if (flo == NULL)
1797                         goto nopol;
1798                 if (IS_ERR(flo)) {
1799                         err = PTR_ERR(flo);
1800                         goto dropdst;
1801                 }
1802                 xdst = container_of(flo, struct xfrm_dst, flo);
1803
1804                 num_pols = xdst->num_pols;
1805                 num_xfrms = xdst->num_xfrms;
1806                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1807                 route = xdst->route;
1808         }
1809
1810         dst = &xdst->u.dst;
1811         if (route == NULL && num_xfrms > 0) {
1812                 /* The only case when xfrm_bundle_lookup() returns a
1813                  * bundle with null route, is when the template could
1814                  * not be resolved. It means policies are there, but
1815                  * bundle could not be created, since we don't yet
1816                  * have the xfrm_state's. We need to wait for KM to
1817                  * negotiate new SA's or bail out with error.*/
1818                 if (net->xfrm.sysctl_larval_drop) {
1819                         /* EREMOTE tells the caller to generate
1820                          * a one-shot blackhole route. */
1821                         dst_release(dst);
1822                         xfrm_pols_put(pols, drop_pols);
1823                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1824
1825                         return make_blackhole(net, family, dst_orig);
1826                 }
1827                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
1828                         DECLARE_WAITQUEUE(wait, current);
1829
1830                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1831                         set_current_state(TASK_INTERRUPTIBLE);
1832                         schedule();
1833                         set_current_state(TASK_RUNNING);
1834                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1835
1836                         if (!signal_pending(current)) {
1837                                 dst_release(dst);
1838                                 goto restart;
1839                         }
1840
1841                         err = -ERESTART;
1842                 } else
1843                         err = -EAGAIN;
1844
1845                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1846                 goto error;
1847         }
1848
1849 no_transform:
1850         if (num_pols == 0)
1851                 goto nopol;
1852
1853         if ((flags & XFRM_LOOKUP_ICMP) &&
1854             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1855                 err = -ENOENT;
1856                 goto error;
1857         }
1858
1859         for (i = 0; i < num_pols; i++)
1860                 pols[i]->curlft.use_time = get_seconds();
1861
1862         if (num_xfrms < 0) {
1863                 /* Prohibit the flow */
1864                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1865                 err = -EPERM;
1866                 goto error;
1867         } else if (num_xfrms > 0) {
1868                 /* Flow transformed */
1869                 dst_release(dst_orig);
1870         } else {
1871                 /* Flow passes untransformed */
1872                 dst_release(dst);
1873                 dst = dst_orig;
1874         }
1875 ok:
1876         xfrm_pols_put(pols, drop_pols);
1877         if (dst && dst->xfrm &&
1878             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
1879                 dst->flags |= DST_XFRM_TUNNEL;
1880         return dst;
1881
1882 nopol:
1883         if (!(flags & XFRM_LOOKUP_ICMP)) {
1884                 dst = dst_orig;
1885                 goto ok;
1886         }
1887         err = -ENOENT;
1888 error:
1889         dst_release(dst);
1890 dropdst:
1891         dst_release(dst_orig);
1892         xfrm_pols_put(pols, drop_pols);
1893         return ERR_PTR(err);
1894 }
1895 EXPORT_SYMBOL(xfrm_lookup);
1896
1897 static inline int
1898 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
1899 {
1900         struct xfrm_state *x;
1901
1902         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1903                 return 0;
1904         x = skb->sp->xvec[idx];
1905         if (!x->type->reject)
1906                 return 0;
1907         return x->type->reject(x, skb, fl);
1908 }
1909
1910 /* When skb is transformed back to its "native" form, we have to
1911  * check policy restrictions. At the moment we make this in maximally
1912  * stupid way. Shame on me. :-) Of course, connected sockets must
1913  * have policy cached at them.
1914  */
1915
1916 static inline int
1917 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
1918               unsigned short family)
1919 {
1920         if (xfrm_state_kern(x))
1921                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1922         return  x->id.proto == tmpl->id.proto &&
1923                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1924                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1925                 x->props.mode == tmpl->mode &&
1926                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1927                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1928                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1929                   xfrm_state_addr_cmp(tmpl, x, family));
1930 }
1931
1932 /*
1933  * 0 or more than 0 is returned when validation is succeeded (either bypass
1934  * because of optional transport mode, or next index of the mathced secpath
1935  * state with the template.
1936  * -1 is returned when no matching template is found.
1937  * Otherwise "-2 - errored_index" is returned.
1938  */
1939 static inline int
1940 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
1941                unsigned short family)
1942 {
1943         int idx = start;
1944
1945         if (tmpl->optional) {
1946                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1947                         return start;
1948         } else
1949                 start = -1;
1950         for (; idx < sp->len; idx++) {
1951                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1952                         return ++idx;
1953                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1954                         if (start == -1)
1955                                 start = -2-idx;
1956                         break;
1957                 }
1958         }
1959         return start;
1960 }
1961
1962 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
1963                           unsigned int family, int reverse)
1964 {
1965         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1966         int err;
1967
1968         if (unlikely(afinfo == NULL))
1969                 return -EAFNOSUPPORT;
1970
1971         afinfo->decode_session(skb, fl, reverse);
1972         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
1973         xfrm_policy_put_afinfo(afinfo);
1974         return err;
1975 }
1976 EXPORT_SYMBOL(__xfrm_decode_session);
1977
1978 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
1979 {
1980         for (; k < sp->len; k++) {
1981                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1982                         *idxp = k;
1983                         return 1;
1984                 }
1985         }
1986
1987         return 0;
1988 }
1989
1990 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
1991                         unsigned short family)
1992 {
1993         struct net *net = dev_net(skb->dev);
1994         struct xfrm_policy *pol;
1995         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1996         int npols = 0;
1997         int xfrm_nr;
1998         int pi;
1999         int reverse;
2000         struct flowi fl;
2001         u8 fl_dir;
2002         int xerr_idx = -1;
2003
2004         reverse = dir & ~XFRM_POLICY_MASK;
2005         dir &= XFRM_POLICY_MASK;
2006         fl_dir = policy_to_flow_dir(dir);
2007
2008         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2009                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2010                 return 0;
2011         }
2012
2013         nf_nat_decode_session(skb, &fl, family);
2014
2015         /* First, check used SA against their selectors. */
2016         if (skb->sp) {
2017                 int i;
2018
2019                 for (i=skb->sp->len-1; i>=0; i--) {
2020                         struct xfrm_state *x = skb->sp->xvec[i];
2021                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2022                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2023                                 return 0;
2024                         }
2025                 }
2026         }
2027
2028         pol = NULL;
2029         if (sk && sk->sk_policy[dir]) {
2030                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2031                 if (IS_ERR(pol)) {
2032                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2033                         return 0;
2034                 }
2035         }
2036
2037         if (!pol) {
2038                 struct flow_cache_object *flo;
2039
2040                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2041                                         xfrm_policy_lookup, NULL);
2042                 if (IS_ERR_OR_NULL(flo))
2043                         pol = ERR_CAST(flo);
2044                 else
2045                         pol = container_of(flo, struct xfrm_policy, flo);
2046         }
2047
2048         if (IS_ERR(pol)) {
2049                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2050                 return 0;
2051         }
2052
2053         if (!pol) {
2054                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2055                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2056                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2057                         return 0;
2058                 }
2059                 return 1;
2060         }
2061
2062         pol->curlft.use_time = get_seconds();
2063
2064         pols[0] = pol;
2065         npols ++;
2066 #ifdef CONFIG_XFRM_SUB_POLICY
2067         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2068                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2069                                                     &fl, family,
2070                                                     XFRM_POLICY_IN);
2071                 if (pols[1]) {
2072                         if (IS_ERR(pols[1])) {
2073                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2074                                 return 0;
2075                         }
2076                         pols[1]->curlft.use_time = get_seconds();
2077                         npols ++;
2078                 }
2079         }
2080 #endif
2081
2082         if (pol->action == XFRM_POLICY_ALLOW) {
2083                 struct sec_path *sp;
2084                 static struct sec_path dummy;
2085                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2086                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2087                 struct xfrm_tmpl **tpp = tp;
2088                 int ti = 0;
2089                 int i, k;
2090
2091                 if ((sp = skb->sp) == NULL)
2092                         sp = &dummy;
2093
2094                 for (pi = 0; pi < npols; pi++) {
2095                         if (pols[pi] != pol &&
2096                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2097                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2098                                 goto reject;
2099                         }
2100                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2101                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2102                                 goto reject_error;
2103                         }
2104                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2105                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2106                 }
2107                 xfrm_nr = ti;
2108                 if (npols > 1) {
2109                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2110                         tpp = stp;
2111                 }
2112
2113                 /* For each tunnel xfrm, find the first matching tmpl.
2114                  * For each tmpl before that, find corresponding xfrm.
2115                  * Order is _important_. Later we will implement
2116                  * some barriers, but at the moment barriers
2117                  * are implied between each two transformations.
2118                  */
2119                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2120                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2121                         if (k < 0) {
2122                                 if (k < -1)
2123                                         /* "-2 - errored_index" returned */
2124                                         xerr_idx = -(2+k);
2125                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2126                                 goto reject;
2127                         }
2128                 }
2129
2130                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2131                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2132                         goto reject;
2133                 }
2134
2135                 xfrm_pols_put(pols, npols);
2136                 return 1;
2137         }
2138         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2139
2140 reject:
2141         xfrm_secpath_reject(xerr_idx, skb, &fl);
2142 reject_error:
2143         xfrm_pols_put(pols, npols);
2144         return 0;
2145 }
2146 EXPORT_SYMBOL(__xfrm_policy_check);
2147
2148 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2149 {
2150         struct net *net = dev_net(skb->dev);
2151         struct flowi fl;
2152         struct dst_entry *dst;
2153         int res = 1;
2154
2155         if (xfrm_decode_session(skb, &fl, family) < 0) {
2156                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2157                 return 0;
2158         }
2159
2160         skb_dst_force(skb);
2161
2162         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2163         if (IS_ERR(dst)) {
2164                 res = 0;
2165                 dst = NULL;
2166         }
2167         skb_dst_set(skb, dst);
2168         return res;
2169 }
2170 EXPORT_SYMBOL(__xfrm_route_forward);
2171
2172 /* Optimize later using cookies and generation ids. */
2173
2174 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2175 {
2176         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2177          * to "-1" to force all XFRM destinations to get validated by
2178          * dst_ops->check on every use.  We do this because when a
2179          * normal route referenced by an XFRM dst is obsoleted we do
2180          * not go looking around for all parent referencing XFRM dsts
2181          * so that we can invalidate them.  It is just too much work.
2182          * Instead we make the checks here on every use.  For example:
2183          *
2184          *      XFRM dst A --> IPv4 dst X
2185          *
2186          * X is the "xdst->route" of A (X is also the "dst->path" of A
2187          * in this example).  If X is marked obsolete, "A" will not
2188          * notice.  That's what we are validating here via the
2189          * stale_bundle() check.
2190          *
2191          * When a policy's bundle is pruned, we dst_free() the XFRM
2192          * dst which causes it's ->obsolete field to be set to a
2193          * positive non-zero integer.  If an XFRM dst has been pruned
2194          * like this, we want to force a new route lookup.
2195          */
2196         if (dst->obsolete < 0 && !stale_bundle(dst))
2197                 return dst;
2198
2199         return NULL;
2200 }
2201
2202 static int stale_bundle(struct dst_entry *dst)
2203 {
2204         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2205 }
2206
2207 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2208 {
2209         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2210                 dst->dev = dev_net(dev)->loopback_dev;
2211                 dev_hold(dst->dev);
2212                 dev_put(dev);
2213         }
2214 }
2215 EXPORT_SYMBOL(xfrm_dst_ifdown);
2216
2217 static void xfrm_link_failure(struct sk_buff *skb)
2218 {
2219         /* Impossible. Such dst must be popped before reaches point of failure. */
2220 }
2221
2222 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2223 {
2224         if (dst) {
2225                 if (dst->obsolete) {
2226                         dst_release(dst);
2227                         dst = NULL;
2228                 }
2229         }
2230         return dst;
2231 }
2232
2233 static void __xfrm_garbage_collect(struct net *net)
2234 {
2235         struct dst_entry *head, *next;
2236
2237         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2238         head = xfrm_policy_sk_bundles;
2239         xfrm_policy_sk_bundles = NULL;
2240         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2241
2242         while (head) {
2243                 next = head->next;
2244                 dst_free(head);
2245                 head = next;
2246         }
2247 }
2248
2249 static void xfrm_garbage_collect(struct net *net)
2250 {
2251         flow_cache_flush();
2252         __xfrm_garbage_collect(net);
2253 }
2254
2255 static void xfrm_garbage_collect_deferred(struct net *net)
2256 {
2257         flow_cache_flush_deferred();
2258         __xfrm_garbage_collect(net);
2259 }
2260
2261 static void xfrm_init_pmtu(struct dst_entry *dst)
2262 {
2263         do {
2264                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2265                 u32 pmtu, route_mtu_cached;
2266
2267                 pmtu = dst_mtu(dst->child);
2268                 xdst->child_mtu_cached = pmtu;
2269
2270                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2271
2272                 route_mtu_cached = dst_mtu(xdst->route);
2273                 xdst->route_mtu_cached = route_mtu_cached;
2274
2275                 if (pmtu > route_mtu_cached)
2276                         pmtu = route_mtu_cached;
2277
2278                 dst_metric_set(dst, RTAX_MTU, pmtu);
2279         } while ((dst = dst->next));
2280 }
2281
2282 /* Check that the bundle accepts the flow and its components are
2283  * still valid.
2284  */
2285
2286 static int xfrm_bundle_ok(struct xfrm_dst *first)
2287 {
2288         struct dst_entry *dst = &first->u.dst;
2289         struct xfrm_dst *last;
2290         u32 mtu;
2291
2292         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2293             (dst->dev && !netif_running(dst->dev)))
2294                 return 0;
2295
2296         last = NULL;
2297
2298         do {
2299                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2300
2301                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2302                         return 0;
2303                 if (xdst->xfrm_genid != dst->xfrm->genid)
2304                         return 0;
2305                 if (xdst->num_pols > 0 &&
2306                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2307                         return 0;
2308
2309                 mtu = dst_mtu(dst->child);
2310                 if (xdst->child_mtu_cached != mtu) {
2311                         last = xdst;
2312                         xdst->child_mtu_cached = mtu;
2313                 }
2314
2315                 if (!dst_check(xdst->route, xdst->route_cookie))
2316                         return 0;
2317                 mtu = dst_mtu(xdst->route);
2318                 if (xdst->route_mtu_cached != mtu) {
2319                         last = xdst;
2320                         xdst->route_mtu_cached = mtu;
2321                 }
2322
2323                 dst = dst->child;
2324         } while (dst->xfrm);
2325
2326         if (likely(!last))
2327                 return 1;
2328
2329         mtu = last->child_mtu_cached;
2330         for (;;) {
2331                 dst = &last->u.dst;
2332
2333                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2334                 if (mtu > last->route_mtu_cached)
2335                         mtu = last->route_mtu_cached;
2336                 dst_metric_set(dst, RTAX_MTU, mtu);
2337
2338                 if (last == first)
2339                         break;
2340
2341                 last = (struct xfrm_dst *)last->u.dst.next;
2342                 last->child_mtu_cached = mtu;
2343         }
2344
2345         return 1;
2346 }
2347
2348 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2349 {
2350         return dst_metric_advmss(dst->path);
2351 }
2352
2353 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2354 {
2355         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2356
2357         return mtu ? : dst_mtu(dst->path);
2358 }
2359
2360 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, const void *daddr)
2361 {
2362         return dst_neigh_lookup(dst->path, daddr);
2363 }
2364
2365 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2366 {
2367         struct net *net;
2368         int err = 0;
2369         if (unlikely(afinfo == NULL))
2370                 return -EINVAL;
2371         if (unlikely(afinfo->family >= NPROTO))
2372                 return -EAFNOSUPPORT;
2373         write_lock_bh(&xfrm_policy_afinfo_lock);
2374         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2375                 err = -ENOBUFS;
2376         else {
2377                 struct dst_ops *dst_ops = afinfo->dst_ops;
2378                 if (likely(dst_ops->kmem_cachep == NULL))
2379                         dst_ops->kmem_cachep = xfrm_dst_cache;
2380                 if (likely(dst_ops->check == NULL))
2381                         dst_ops->check = xfrm_dst_check;
2382                 if (likely(dst_ops->default_advmss == NULL))
2383                         dst_ops->default_advmss = xfrm_default_advmss;
2384                 if (likely(dst_ops->mtu == NULL))
2385                         dst_ops->mtu = xfrm_mtu;
2386                 if (likely(dst_ops->negative_advice == NULL))
2387                         dst_ops->negative_advice = xfrm_negative_advice;
2388                 if (likely(dst_ops->link_failure == NULL))
2389                         dst_ops->link_failure = xfrm_link_failure;
2390                 if (likely(dst_ops->neigh_lookup == NULL))
2391                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2392                 if (likely(afinfo->garbage_collect == NULL))
2393                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2394                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2395         }
2396         write_unlock_bh(&xfrm_policy_afinfo_lock);
2397
2398         rtnl_lock();
2399         for_each_net(net) {
2400                 struct dst_ops *xfrm_dst_ops;
2401
2402                 switch (afinfo->family) {
2403                 case AF_INET:
2404                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2405                         break;
2406 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2407                 case AF_INET6:
2408                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2409                         break;
2410 #endif
2411                 default:
2412                         BUG();
2413                 }
2414                 *xfrm_dst_ops = *afinfo->dst_ops;
2415         }
2416         rtnl_unlock();
2417
2418         return err;
2419 }
2420 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2421
2422 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2423 {
2424         int err = 0;
2425         if (unlikely(afinfo == NULL))
2426                 return -EINVAL;
2427         if (unlikely(afinfo->family >= NPROTO))
2428                 return -EAFNOSUPPORT;
2429         write_lock_bh(&xfrm_policy_afinfo_lock);
2430         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2431                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2432                         err = -EINVAL;
2433                 else {
2434                         struct dst_ops *dst_ops = afinfo->dst_ops;
2435                         xfrm_policy_afinfo[afinfo->family] = NULL;
2436                         dst_ops->kmem_cachep = NULL;
2437                         dst_ops->check = NULL;
2438                         dst_ops->negative_advice = NULL;
2439                         dst_ops->link_failure = NULL;
2440                         afinfo->garbage_collect = NULL;
2441                 }
2442         }
2443         write_unlock_bh(&xfrm_policy_afinfo_lock);
2444         return err;
2445 }
2446 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2447
2448 static void __net_init xfrm_dst_ops_init(struct net *net)
2449 {
2450         struct xfrm_policy_afinfo *afinfo;
2451
2452         read_lock_bh(&xfrm_policy_afinfo_lock);
2453         afinfo = xfrm_policy_afinfo[AF_INET];
2454         if (afinfo)
2455                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2456 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2457         afinfo = xfrm_policy_afinfo[AF_INET6];
2458         if (afinfo)
2459                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2460 #endif
2461         read_unlock_bh(&xfrm_policy_afinfo_lock);
2462 }
2463
2464 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2465 {
2466         struct xfrm_policy_afinfo *afinfo;
2467         if (unlikely(family >= NPROTO))
2468                 return NULL;
2469         read_lock(&xfrm_policy_afinfo_lock);
2470         afinfo = xfrm_policy_afinfo[family];
2471         if (unlikely(!afinfo))
2472                 read_unlock(&xfrm_policy_afinfo_lock);
2473         return afinfo;
2474 }
2475
2476 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2477 {
2478         read_unlock(&xfrm_policy_afinfo_lock);
2479 }
2480
2481 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2482 {
2483         struct net_device *dev = ptr;
2484
2485         switch (event) {
2486         case NETDEV_DOWN:
2487                 xfrm_garbage_collect(dev_net(dev));
2488         }
2489         return NOTIFY_DONE;
2490 }
2491
2492 static struct notifier_block xfrm_dev_notifier = {
2493         .notifier_call  = xfrm_dev_event,
2494 };
2495
2496 #ifdef CONFIG_XFRM_STATISTICS
2497 static int __net_init xfrm_statistics_init(struct net *net)
2498 {
2499         int rv;
2500
2501         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2502                           sizeof(struct linux_xfrm_mib),
2503                           __alignof__(struct linux_xfrm_mib)) < 0)
2504                 return -ENOMEM;
2505         rv = xfrm_proc_init(net);
2506         if (rv < 0)
2507                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2508         return rv;
2509 }
2510
2511 static void xfrm_statistics_fini(struct net *net)
2512 {
2513         xfrm_proc_fini(net);
2514         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2515 }
2516 #else
2517 static int __net_init xfrm_statistics_init(struct net *net)
2518 {
2519         return 0;
2520 }
2521
2522 static void xfrm_statistics_fini(struct net *net)
2523 {
2524 }
2525 #endif
2526
2527 static int __net_init xfrm_policy_init(struct net *net)
2528 {
2529         unsigned int hmask, sz;
2530         int dir;
2531
2532         if (net_eq(net, &init_net))
2533                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2534                                            sizeof(struct xfrm_dst),
2535                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2536                                            NULL);
2537
2538         hmask = 8 - 1;
2539         sz = (hmask+1) * sizeof(struct hlist_head);
2540
2541         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2542         if (!net->xfrm.policy_byidx)
2543                 goto out_byidx;
2544         net->xfrm.policy_idx_hmask = hmask;
2545
2546         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2547                 struct xfrm_policy_hash *htab;
2548
2549                 net->xfrm.policy_count[dir] = 0;
2550                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2551
2552                 htab = &net->xfrm.policy_bydst[dir];
2553                 htab->table = xfrm_hash_alloc(sz);
2554                 if (!htab->table)
2555                         goto out_bydst;
2556                 htab->hmask = hmask;
2557         }
2558
2559         INIT_LIST_HEAD(&net->xfrm.policy_all);
2560         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2561         if (net_eq(net, &init_net))
2562                 register_netdevice_notifier(&xfrm_dev_notifier);
2563         return 0;
2564
2565 out_bydst:
2566         for (dir--; dir >= 0; dir--) {
2567                 struct xfrm_policy_hash *htab;
2568
2569                 htab = &net->xfrm.policy_bydst[dir];
2570                 xfrm_hash_free(htab->table, sz);
2571         }
2572         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2573 out_byidx:
2574         return -ENOMEM;
2575 }
2576
2577 static void xfrm_policy_fini(struct net *net)
2578 {
2579         struct xfrm_audit audit_info;
2580         unsigned int sz;
2581         int dir;
2582
2583         flush_work(&net->xfrm.policy_hash_work);
2584 #ifdef CONFIG_XFRM_SUB_POLICY
2585         audit_info.loginuid = -1;
2586         audit_info.sessionid = -1;
2587         audit_info.secid = 0;
2588         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2589 #endif
2590         audit_info.loginuid = -1;
2591         audit_info.sessionid = -1;
2592         audit_info.secid = 0;
2593         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2594
2595         WARN_ON(!list_empty(&net->xfrm.policy_all));
2596
2597         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2598                 struct xfrm_policy_hash *htab;
2599
2600                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2601
2602                 htab = &net->xfrm.policy_bydst[dir];
2603                 sz = (htab->hmask + 1);
2604                 WARN_ON(!hlist_empty(htab->table));
2605                 xfrm_hash_free(htab->table, sz);
2606         }
2607
2608         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2609         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2610         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2611 }
2612
2613 static int __net_init xfrm_net_init(struct net *net)
2614 {
2615         int rv;
2616
2617         rv = xfrm_statistics_init(net);
2618         if (rv < 0)
2619                 goto out_statistics;
2620         rv = xfrm_state_init(net);
2621         if (rv < 0)
2622                 goto out_state;
2623         rv = xfrm_policy_init(net);
2624         if (rv < 0)
2625                 goto out_policy;
2626         xfrm_dst_ops_init(net);
2627         rv = xfrm_sysctl_init(net);
2628         if (rv < 0)
2629                 goto out_sysctl;
2630         return 0;
2631
2632 out_sysctl:
2633         xfrm_policy_fini(net);
2634 out_policy:
2635         xfrm_state_fini(net);
2636 out_state:
2637         xfrm_statistics_fini(net);
2638 out_statistics:
2639         return rv;
2640 }
2641
2642 static void __net_exit xfrm_net_exit(struct net *net)
2643 {
2644         xfrm_sysctl_fini(net);
2645         xfrm_policy_fini(net);
2646         xfrm_state_fini(net);
2647         xfrm_statistics_fini(net);
2648 }
2649
2650 static struct pernet_operations __net_initdata xfrm_net_ops = {
2651         .init = xfrm_net_init,
2652         .exit = xfrm_net_exit,
2653 };
2654
2655 void __init xfrm_init(void)
2656 {
2657         register_pernet_subsys(&xfrm_net_ops);
2658         xfrm_input_init();
2659 }
2660
2661 #ifdef CONFIG_AUDITSYSCALL
2662 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2663                                          struct audit_buffer *audit_buf)
2664 {
2665         struct xfrm_sec_ctx *ctx = xp->security;
2666         struct xfrm_selector *sel = &xp->selector;
2667
2668         if (ctx)
2669                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2670                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2671
2672         switch(sel->family) {
2673         case AF_INET:
2674                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2675                 if (sel->prefixlen_s != 32)
2676                         audit_log_format(audit_buf, " src_prefixlen=%d",
2677                                          sel->prefixlen_s);
2678                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2679                 if (sel->prefixlen_d != 32)
2680                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2681                                          sel->prefixlen_d);
2682                 break;
2683         case AF_INET6:
2684                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2685                 if (sel->prefixlen_s != 128)
2686                         audit_log_format(audit_buf, " src_prefixlen=%d",
2687                                          sel->prefixlen_s);
2688                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2689                 if (sel->prefixlen_d != 128)
2690                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2691                                          sel->prefixlen_d);
2692                 break;
2693         }
2694 }
2695
2696 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2697                            uid_t auid, u32 sessionid, u32 secid)
2698 {
2699         struct audit_buffer *audit_buf;
2700
2701         audit_buf = xfrm_audit_start("SPD-add");
2702         if (audit_buf == NULL)
2703                 return;
2704         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2705         audit_log_format(audit_buf, " res=%u", result);
2706         xfrm_audit_common_policyinfo(xp, audit_buf);
2707         audit_log_end(audit_buf);
2708 }
2709 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2710
2711 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2712                               uid_t auid, u32 sessionid, u32 secid)
2713 {
2714         struct audit_buffer *audit_buf;
2715
2716         audit_buf = xfrm_audit_start("SPD-delete");
2717         if (audit_buf == NULL)
2718                 return;
2719         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2720         audit_log_format(audit_buf, " res=%u", result);
2721         xfrm_audit_common_policyinfo(xp, audit_buf);
2722         audit_log_end(audit_buf);
2723 }
2724 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2725 #endif
2726
2727 #ifdef CONFIG_XFRM_MIGRATE
2728 static int xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
2729                                        const struct xfrm_selector *sel_tgt)
2730 {
2731         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2732                 if (sel_tgt->family == sel_cmp->family &&
2733                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2734                                   sel_cmp->family) == 0 &&
2735                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2736                                   sel_cmp->family) == 0 &&
2737                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2738                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2739                         return 1;
2740                 }
2741         } else {
2742                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2743                         return 1;
2744                 }
2745         }
2746         return 0;
2747 }
2748
2749 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
2750                                                      u8 dir, u8 type)
2751 {
2752         struct xfrm_policy *pol, *ret = NULL;
2753         struct hlist_node *entry;
2754         struct hlist_head *chain;
2755         u32 priority = ~0U;
2756
2757         read_lock_bh(&xfrm_policy_lock);
2758         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2759         hlist_for_each_entry(pol, entry, chain, bydst) {
2760                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2761                     pol->type == type) {
2762                         ret = pol;
2763                         priority = ret->priority;
2764                         break;
2765                 }
2766         }
2767         chain = &init_net.xfrm.policy_inexact[dir];
2768         hlist_for_each_entry(pol, entry, chain, bydst) {
2769                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2770                     pol->type == type &&
2771                     pol->priority < priority) {
2772                         ret = pol;
2773                         break;
2774                 }
2775         }
2776
2777         if (ret)
2778                 xfrm_pol_hold(ret);
2779
2780         read_unlock_bh(&xfrm_policy_lock);
2781
2782         return ret;
2783 }
2784
2785 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2786 {
2787         int match = 0;
2788
2789         if (t->mode == m->mode && t->id.proto == m->proto &&
2790             (m->reqid == 0 || t->reqid == m->reqid)) {
2791                 switch (t->mode) {
2792                 case XFRM_MODE_TUNNEL:
2793                 case XFRM_MODE_BEET:
2794                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2795                                           m->old_family) == 0 &&
2796                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2797                                           m->old_family) == 0) {
2798                                 match = 1;
2799                         }
2800                         break;
2801                 case XFRM_MODE_TRANSPORT:
2802                         /* in case of transport mode, template does not store
2803                            any IP addresses, hence we just compare mode and
2804                            protocol */
2805                         match = 1;
2806                         break;
2807                 default:
2808                         break;
2809                 }
2810         }
2811         return match;
2812 }
2813
2814 /* update endpoint address(es) of template(s) */
2815 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2816                                struct xfrm_migrate *m, int num_migrate)
2817 {
2818         struct xfrm_migrate *mp;
2819         int i, j, n = 0;
2820
2821         write_lock_bh(&pol->lock);
2822         if (unlikely(pol->walk.dead)) {
2823                 /* target policy has been deleted */
2824                 write_unlock_bh(&pol->lock);
2825                 return -ENOENT;
2826         }
2827
2828         for (i = 0; i < pol->xfrm_nr; i++) {
2829                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2830                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2831                                 continue;
2832                         n++;
2833                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2834                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2835                                 continue;
2836                         /* update endpoints */
2837                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2838                                sizeof(pol->xfrm_vec[i].id.daddr));
2839                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2840                                sizeof(pol->xfrm_vec[i].saddr));
2841                         pol->xfrm_vec[i].encap_family = mp->new_family;
2842                         /* flush bundles */
2843                         atomic_inc(&pol->genid);
2844                 }
2845         }
2846
2847         write_unlock_bh(&pol->lock);
2848
2849         if (!n)
2850                 return -ENODATA;
2851
2852         return 0;
2853 }
2854
2855 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
2856 {
2857         int i, j;
2858
2859         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2860                 return -EINVAL;
2861
2862         for (i = 0; i < num_migrate; i++) {
2863                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2864                                    m[i].old_family) == 0) &&
2865                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2866                                    m[i].old_family) == 0))
2867                         return -EINVAL;
2868                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2869                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2870                         return -EINVAL;
2871
2872                 /* check if there is any duplicated entry */
2873                 for (j = i + 1; j < num_migrate; j++) {
2874                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2875                                     sizeof(m[i].old_daddr)) &&
2876                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2877                                     sizeof(m[i].old_saddr)) &&
2878                             m[i].proto == m[j].proto &&
2879                             m[i].mode == m[j].mode &&
2880                             m[i].reqid == m[j].reqid &&
2881                             m[i].old_family == m[j].old_family)
2882                                 return -EINVAL;
2883                 }
2884         }
2885
2886         return 0;
2887 }
2888
2889 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
2890                  struct xfrm_migrate *m, int num_migrate,
2891                  struct xfrm_kmaddress *k)
2892 {
2893         int i, err, nx_cur = 0, nx_new = 0;
2894         struct xfrm_policy *pol = NULL;
2895         struct xfrm_state *x, *xc;
2896         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2897         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2898         struct xfrm_migrate *mp;
2899
2900         /* Stage 0 - sanity checks */
2901         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2902                 goto out;
2903
2904         if (dir >= XFRM_POLICY_MAX) {
2905                 err = -EINVAL;
2906                 goto out;
2907         }
2908
2909         /* Stage 1 - find policy */
2910         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2911                 err = -ENOENT;
2912                 goto out;
2913         }
2914
2915         /* Stage 2 - find and update state(s) */
2916         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2917                 if ((x = xfrm_migrate_state_find(mp))) {
2918                         x_cur[nx_cur] = x;
2919                         nx_cur++;
2920                         if ((xc = xfrm_state_migrate(x, mp))) {
2921                                 x_new[nx_new] = xc;
2922                                 nx_new++;
2923                         } else {
2924                                 err = -ENODATA;
2925                                 goto restore_state;
2926                         }
2927                 }
2928         }
2929
2930         /* Stage 3 - update policy */
2931         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2932                 goto restore_state;
2933
2934         /* Stage 4 - delete old state(s) */
2935         if (nx_cur) {
2936                 xfrm_states_put(x_cur, nx_cur);
2937                 xfrm_states_delete(x_cur, nx_cur);
2938         }
2939
2940         /* Stage 5 - announce */
2941         km_migrate(sel, dir, type, m, num_migrate, k);
2942
2943         xfrm_pol_put(pol);
2944
2945         return 0;
2946 out:
2947         return err;
2948
2949 restore_state:
2950         if (pol)
2951                 xfrm_pol_put(pol);
2952         if (nx_cur)
2953                 xfrm_states_put(x_cur, nx_cur);
2954         if (nx_new)
2955                 xfrm_states_delete(x_new, nx_new);
2956
2957         return err;
2958 }
2959 EXPORT_SYMBOL(xfrm_migrate);
2960 #endif