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 inline
1534 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1535 {
1536         if (!*target) {
1537                 *target = kmalloc(size, GFP_ATOMIC);
1538                 if (!*target)
1539                         return -ENOMEM;
1540         }
1541         memcpy(*target, src, size);
1542         return 0;
1543 }
1544
1545 static int inline
1546 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1547 {
1548 #ifdef CONFIG_XFRM_SUB_POLICY
1549         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1550         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1551                                    sel, sizeof(*sel));
1552 #else
1553         return 0;
1554 #endif
1555 }
1556
1557 static int inline
1558 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1559 {
1560 #ifdef CONFIG_XFRM_SUB_POLICY
1561         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1562         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1563 #else
1564         return 0;
1565 #endif
1566 }
1567
1568 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1569                                 struct xfrm_policy **pols,
1570                                 int *num_pols, int *num_xfrms)
1571 {
1572         int i;
1573
1574         if (*num_pols == 0 || !pols[0]) {
1575                 *num_pols = 0;
1576                 *num_xfrms = 0;
1577                 return 0;
1578         }
1579         if (IS_ERR(pols[0]))
1580                 return PTR_ERR(pols[0]);
1581
1582         *num_xfrms = pols[0]->xfrm_nr;
1583
1584 #ifdef CONFIG_XFRM_SUB_POLICY
1585         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1586             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1587                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1588                                                     XFRM_POLICY_TYPE_MAIN,
1589                                                     fl, family,
1590                                                     XFRM_POLICY_OUT);
1591                 if (pols[1]) {
1592                         if (IS_ERR(pols[1])) {
1593                                 xfrm_pols_put(pols, *num_pols);
1594                                 return PTR_ERR(pols[1]);
1595                         }
1596                         (*num_pols) ++;
1597                         (*num_xfrms) += pols[1]->xfrm_nr;
1598                 }
1599         }
1600 #endif
1601         for (i = 0; i < *num_pols; i++) {
1602                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1603                         *num_xfrms = -1;
1604                         break;
1605                 }
1606         }
1607
1608         return 0;
1609
1610 }
1611
1612 static struct xfrm_dst *
1613 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1614                                const struct flowi *fl, u16 family,
1615                                struct dst_entry *dst_orig)
1616 {
1617         struct net *net = xp_net(pols[0]);
1618         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1619         struct dst_entry *dst;
1620         struct xfrm_dst *xdst;
1621         int err;
1622
1623         /* Try to instantiate a bundle */
1624         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1625         if (err <= 0) {
1626                 if (err != 0 && err != -EAGAIN)
1627                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1628                 return ERR_PTR(err);
1629         }
1630
1631         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1632         if (IS_ERR(dst)) {
1633                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1634                 return ERR_CAST(dst);
1635         }
1636
1637         xdst = (struct xfrm_dst *)dst;
1638         xdst->num_xfrms = err;
1639         if (num_pols > 1)
1640                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1641         else
1642                 err = xfrm_dst_update_origin(dst, fl);
1643         if (unlikely(err)) {
1644                 dst_free(dst);
1645                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1646                 return ERR_PTR(err);
1647         }
1648
1649         xdst->num_pols = num_pols;
1650         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1651         xdst->policy_genid = atomic_read(&pols[0]->genid);
1652
1653         return xdst;
1654 }
1655
1656 static struct flow_cache_object *
1657 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1658                    struct flow_cache_object *oldflo, void *ctx)
1659 {
1660         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1661         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1662         struct xfrm_dst *xdst, *new_xdst;
1663         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1664
1665         /* Check if the policies from old bundle are usable */
1666         xdst = NULL;
1667         if (oldflo) {
1668                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1669                 num_pols = xdst->num_pols;
1670                 num_xfrms = xdst->num_xfrms;
1671                 pol_dead = 0;
1672                 for (i = 0; i < num_pols; i++) {
1673                         pols[i] = xdst->pols[i];
1674                         pol_dead |= pols[i]->walk.dead;
1675                 }
1676                 if (pol_dead) {
1677                         dst_free(&xdst->u.dst);
1678                         xdst = NULL;
1679                         num_pols = 0;
1680                         num_xfrms = 0;
1681                         oldflo = NULL;
1682                 }
1683         }
1684
1685         /* Resolve policies to use if we couldn't get them from
1686          * previous cache entry */
1687         if (xdst == NULL) {
1688                 num_pols = 1;
1689                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1690                 err = xfrm_expand_policies(fl, family, pols,
1691                                            &num_pols, &num_xfrms);
1692                 if (err < 0)
1693                         goto inc_error;
1694                 if (num_pols == 0)
1695                         return NULL;
1696                 if (num_xfrms <= 0)
1697                         goto make_dummy_bundle;
1698         }
1699
1700         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1701         if (IS_ERR(new_xdst)) {
1702                 err = PTR_ERR(new_xdst);
1703                 if (err != -EAGAIN)
1704                         goto error;
1705                 if (oldflo == NULL)
1706                         goto make_dummy_bundle;
1707                 dst_hold(&xdst->u.dst);
1708                 return oldflo;
1709         } else if (new_xdst == NULL) {
1710                 num_xfrms = 0;
1711                 if (oldflo == NULL)
1712                         goto make_dummy_bundle;
1713                 xdst->num_xfrms = 0;
1714                 dst_hold(&xdst->u.dst);
1715                 return oldflo;
1716         }
1717
1718         /* Kill the previous bundle */
1719         if (xdst) {
1720                 /* The policies were stolen for newly generated bundle */
1721                 xdst->num_pols = 0;
1722                 dst_free(&xdst->u.dst);
1723         }
1724
1725         /* Flow cache does not have reference, it dst_free()'s,
1726          * but we do need to return one reference for original caller */
1727         dst_hold(&new_xdst->u.dst);
1728         return &new_xdst->flo;
1729
1730 make_dummy_bundle:
1731         /* We found policies, but there's no bundles to instantiate:
1732          * either because the policy blocks, has no transformations or
1733          * we could not build template (no xfrm_states).*/
1734         xdst = xfrm_alloc_dst(net, family);
1735         if (IS_ERR(xdst)) {
1736                 xfrm_pols_put(pols, num_pols);
1737                 return ERR_CAST(xdst);
1738         }
1739         xdst->num_pols = num_pols;
1740         xdst->num_xfrms = num_xfrms;
1741         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1742
1743         dst_hold(&xdst->u.dst);
1744         return &xdst->flo;
1745
1746 inc_error:
1747         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1748 error:
1749         if (xdst != NULL)
1750                 dst_free(&xdst->u.dst);
1751         else
1752                 xfrm_pols_put(pols, num_pols);
1753         return ERR_PTR(err);
1754 }
1755
1756 static struct dst_entry *make_blackhole(struct net *net, u16 family,
1757                                         struct dst_entry *dst_orig)
1758 {
1759         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1760         struct dst_entry *ret;
1761
1762         if (!afinfo) {
1763                 dst_release(dst_orig);
1764                 ret = ERR_PTR(-EINVAL);
1765         } else {
1766                 ret = afinfo->blackhole_route(net, dst_orig);
1767         }
1768         xfrm_policy_put_afinfo(afinfo);
1769
1770         return ret;
1771 }
1772
1773 /* Main function: finds/creates a bundle for given flow.
1774  *
1775  * At the moment we eat a raw IP route. Mostly to speed up lookups
1776  * on interfaces with disabled IPsec.
1777  */
1778 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
1779                               const struct flowi *fl,
1780                               struct sock *sk, int flags)
1781 {
1782         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1783         struct flow_cache_object *flo;
1784         struct xfrm_dst *xdst;
1785         struct dst_entry *dst, *route;
1786         u16 family = dst_orig->ops->family;
1787         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1788         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1789
1790 restart:
1791         dst = NULL;
1792         xdst = NULL;
1793         route = NULL;
1794
1795         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1796                 num_pols = 1;
1797                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1798                 err = xfrm_expand_policies(fl, family, pols,
1799                                            &num_pols, &num_xfrms);
1800                 if (err < 0)
1801                         goto dropdst;
1802
1803                 if (num_pols) {
1804                         if (num_xfrms <= 0) {
1805                                 drop_pols = num_pols;
1806                                 goto no_transform;
1807                         }
1808
1809                         xdst = xfrm_resolve_and_create_bundle(
1810                                         pols, num_pols, fl,
1811                                         family, dst_orig);
1812                         if (IS_ERR(xdst)) {
1813                                 xfrm_pols_put(pols, num_pols);
1814                                 err = PTR_ERR(xdst);
1815                                 goto dropdst;
1816                         } else if (xdst == NULL) {
1817                                 num_xfrms = 0;
1818                                 drop_pols = num_pols;
1819                                 goto no_transform;
1820                         }
1821
1822                         dst_hold(&xdst->u.dst);
1823
1824                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1825                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1826                         xfrm_policy_sk_bundles = &xdst->u.dst;
1827                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1828
1829                         route = xdst->route;
1830                 }
1831         }
1832
1833         if (xdst == NULL) {
1834                 /* To accelerate a bit...  */
1835                 if ((dst_orig->flags & DST_NOXFRM) ||
1836                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1837                         goto nopol;
1838
1839                 flo = flow_cache_lookup(net, fl, family, dir,
1840                                         xfrm_bundle_lookup, dst_orig);
1841                 if (flo == NULL)
1842                         goto nopol;
1843                 if (IS_ERR(flo)) {
1844                         err = PTR_ERR(flo);
1845                         goto dropdst;
1846                 }
1847                 xdst = container_of(flo, struct xfrm_dst, flo);
1848
1849                 num_pols = xdst->num_pols;
1850                 num_xfrms = xdst->num_xfrms;
1851                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1852                 route = xdst->route;
1853         }
1854
1855         dst = &xdst->u.dst;
1856         if (route == NULL && num_xfrms > 0) {
1857                 /* The only case when xfrm_bundle_lookup() returns a
1858                  * bundle with null route, is when the template could
1859                  * not be resolved. It means policies are there, but
1860                  * bundle could not be created, since we don't yet
1861                  * have the xfrm_state's. We need to wait for KM to
1862                  * negotiate new SA's or bail out with error.*/
1863                 if (net->xfrm.sysctl_larval_drop) {
1864                         /* EREMOTE tells the caller to generate
1865                          * a one-shot blackhole route. */
1866                         dst_release(dst);
1867                         xfrm_pols_put(pols, drop_pols);
1868                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1869
1870                         return make_blackhole(net, family, dst_orig);
1871                 }
1872                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
1873                         DECLARE_WAITQUEUE(wait, current);
1874
1875                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1876                         set_current_state(TASK_INTERRUPTIBLE);
1877                         schedule();
1878                         set_current_state(TASK_RUNNING);
1879                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1880
1881                         if (!signal_pending(current)) {
1882                                 dst_release(dst);
1883                                 goto restart;
1884                         }
1885
1886                         err = -ERESTART;
1887                 } else
1888                         err = -EAGAIN;
1889
1890                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1891                 goto error;
1892         }
1893
1894 no_transform:
1895         if (num_pols == 0)
1896                 goto nopol;
1897
1898         if ((flags & XFRM_LOOKUP_ICMP) &&
1899             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1900                 err = -ENOENT;
1901                 goto error;
1902         }
1903
1904         for (i = 0; i < num_pols; i++)
1905                 pols[i]->curlft.use_time = get_seconds();
1906
1907         if (num_xfrms < 0) {
1908                 /* Prohibit the flow */
1909                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1910                 err = -EPERM;
1911                 goto error;
1912         } else if (num_xfrms > 0) {
1913                 /* Flow transformed */
1914                 dst_release(dst_orig);
1915         } else {
1916                 /* Flow passes untransformed */
1917                 dst_release(dst);
1918                 dst = dst_orig;
1919         }
1920 ok:
1921         xfrm_pols_put(pols, drop_pols);
1922         return dst;
1923
1924 nopol:
1925         if (!(flags & XFRM_LOOKUP_ICMP)) {
1926                 dst = dst_orig;
1927                 goto ok;
1928         }
1929         err = -ENOENT;
1930 error:
1931         dst_release(dst);
1932 dropdst:
1933         dst_release(dst_orig);
1934         xfrm_pols_put(pols, drop_pols);
1935         return ERR_PTR(err);
1936 }
1937 EXPORT_SYMBOL(xfrm_lookup);
1938
1939 static inline int
1940 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
1941 {
1942         struct xfrm_state *x;
1943
1944         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1945                 return 0;
1946         x = skb->sp->xvec[idx];
1947         if (!x->type->reject)
1948                 return 0;
1949         return x->type->reject(x, skb, fl);
1950 }
1951
1952 /* When skb is transformed back to its "native" form, we have to
1953  * check policy restrictions. At the moment we make this in maximally
1954  * stupid way. Shame on me. :-) Of course, connected sockets must
1955  * have policy cached at them.
1956  */
1957
1958 static inline int
1959 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
1960               unsigned short family)
1961 {
1962         if (xfrm_state_kern(x))
1963                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1964         return  x->id.proto == tmpl->id.proto &&
1965                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1966                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1967                 x->props.mode == tmpl->mode &&
1968                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1969                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1970                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1971                   xfrm_state_addr_cmp(tmpl, x, family));
1972 }
1973
1974 /*
1975  * 0 or more than 0 is returned when validation is succeeded (either bypass
1976  * because of optional transport mode, or next index of the mathced secpath
1977  * state with the template.
1978  * -1 is returned when no matching template is found.
1979  * Otherwise "-2 - errored_index" is returned.
1980  */
1981 static inline int
1982 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
1983                unsigned short family)
1984 {
1985         int idx = start;
1986
1987         if (tmpl->optional) {
1988                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1989                         return start;
1990         } else
1991                 start = -1;
1992         for (; idx < sp->len; idx++) {
1993                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1994                         return ++idx;
1995                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1996                         if (start == -1)
1997                                 start = -2-idx;
1998                         break;
1999                 }
2000         }
2001         return start;
2002 }
2003
2004 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2005                           unsigned int family, int reverse)
2006 {
2007         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2008         int err;
2009
2010         if (unlikely(afinfo == NULL))
2011                 return -EAFNOSUPPORT;
2012
2013         afinfo->decode_session(skb, fl, reverse);
2014         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2015         xfrm_policy_put_afinfo(afinfo);
2016         return err;
2017 }
2018 EXPORT_SYMBOL(__xfrm_decode_session);
2019
2020 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2021 {
2022         for (; k < sp->len; k++) {
2023                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2024                         *idxp = k;
2025                         return 1;
2026                 }
2027         }
2028
2029         return 0;
2030 }
2031
2032 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2033                         unsigned short family)
2034 {
2035         struct net *net = dev_net(skb->dev);
2036         struct xfrm_policy *pol;
2037         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2038         int npols = 0;
2039         int xfrm_nr;
2040         int pi;
2041         int reverse;
2042         struct flowi fl;
2043         u8 fl_dir;
2044         int xerr_idx = -1;
2045
2046         reverse = dir & ~XFRM_POLICY_MASK;
2047         dir &= XFRM_POLICY_MASK;
2048         fl_dir = policy_to_flow_dir(dir);
2049
2050         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2051                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2052                 return 0;
2053         }
2054
2055         nf_nat_decode_session(skb, &fl, family);
2056
2057         /* First, check used SA against their selectors. */
2058         if (skb->sp) {
2059                 int i;
2060
2061                 for (i=skb->sp->len-1; i>=0; i--) {
2062                         struct xfrm_state *x = skb->sp->xvec[i];
2063                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2064                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2065                                 return 0;
2066                         }
2067                 }
2068         }
2069
2070         pol = NULL;
2071         if (sk && sk->sk_policy[dir]) {
2072                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2073                 if (IS_ERR(pol)) {
2074                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2075                         return 0;
2076                 }
2077         }
2078
2079         if (!pol) {
2080                 struct flow_cache_object *flo;
2081
2082                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2083                                         xfrm_policy_lookup, NULL);
2084                 if (IS_ERR_OR_NULL(flo))
2085                         pol = ERR_CAST(flo);
2086                 else
2087                         pol = container_of(flo, struct xfrm_policy, flo);
2088         }
2089
2090         if (IS_ERR(pol)) {
2091                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2092                 return 0;
2093         }
2094
2095         if (!pol) {
2096                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2097                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2098                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2099                         return 0;
2100                 }
2101                 return 1;
2102         }
2103
2104         pol->curlft.use_time = get_seconds();
2105
2106         pols[0] = pol;
2107         npols ++;
2108 #ifdef CONFIG_XFRM_SUB_POLICY
2109         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2110                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2111                                                     &fl, family,
2112                                                     XFRM_POLICY_IN);
2113                 if (pols[1]) {
2114                         if (IS_ERR(pols[1])) {
2115                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2116                                 return 0;
2117                         }
2118                         pols[1]->curlft.use_time = get_seconds();
2119                         npols ++;
2120                 }
2121         }
2122 #endif
2123
2124         if (pol->action == XFRM_POLICY_ALLOW) {
2125                 struct sec_path *sp;
2126                 static struct sec_path dummy;
2127                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2128                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2129                 struct xfrm_tmpl **tpp = tp;
2130                 int ti = 0;
2131                 int i, k;
2132
2133                 if ((sp = skb->sp) == NULL)
2134                         sp = &dummy;
2135
2136                 for (pi = 0; pi < npols; pi++) {
2137                         if (pols[pi] != pol &&
2138                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2139                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2140                                 goto reject;
2141                         }
2142                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2143                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2144                                 goto reject_error;
2145                         }
2146                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2147                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2148                 }
2149                 xfrm_nr = ti;
2150                 if (npols > 1) {
2151                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2152                         tpp = stp;
2153                 }
2154
2155                 /* For each tunnel xfrm, find the first matching tmpl.
2156                  * For each tmpl before that, find corresponding xfrm.
2157                  * Order is _important_. Later we will implement
2158                  * some barriers, but at the moment barriers
2159                  * are implied between each two transformations.
2160                  */
2161                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2162                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2163                         if (k < 0) {
2164                                 if (k < -1)
2165                                         /* "-2 - errored_index" returned */
2166                                         xerr_idx = -(2+k);
2167                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2168                                 goto reject;
2169                         }
2170                 }
2171
2172                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2173                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2174                         goto reject;
2175                 }
2176
2177                 xfrm_pols_put(pols, npols);
2178                 return 1;
2179         }
2180         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2181
2182 reject:
2183         xfrm_secpath_reject(xerr_idx, skb, &fl);
2184 reject_error:
2185         xfrm_pols_put(pols, npols);
2186         return 0;
2187 }
2188 EXPORT_SYMBOL(__xfrm_policy_check);
2189
2190 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2191 {
2192         struct net *net = dev_net(skb->dev);
2193         struct flowi fl;
2194         struct dst_entry *dst;
2195         int res = 1;
2196
2197         if (xfrm_decode_session(skb, &fl, family) < 0) {
2198                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2199                 return 0;
2200         }
2201
2202         skb_dst_force(skb);
2203
2204         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2205         if (IS_ERR(dst)) {
2206                 res = 0;
2207                 dst = NULL;
2208         }
2209         skb_dst_set(skb, dst);
2210         return res;
2211 }
2212 EXPORT_SYMBOL(__xfrm_route_forward);
2213
2214 /* Optimize later using cookies and generation ids. */
2215
2216 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2217 {
2218         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2219          * to "-1" to force all XFRM destinations to get validated by
2220          * dst_ops->check on every use.  We do this because when a
2221          * normal route referenced by an XFRM dst is obsoleted we do
2222          * not go looking around for all parent referencing XFRM dsts
2223          * so that we can invalidate them.  It is just too much work.
2224          * Instead we make the checks here on every use.  For example:
2225          *
2226          *      XFRM dst A --> IPv4 dst X
2227          *
2228          * X is the "xdst->route" of A (X is also the "dst->path" of A
2229          * in this example).  If X is marked obsolete, "A" will not
2230          * notice.  That's what we are validating here via the
2231          * stale_bundle() check.
2232          *
2233          * When a policy's bundle is pruned, we dst_free() the XFRM
2234          * dst which causes it's ->obsolete field to be set to a
2235          * positive non-zero integer.  If an XFRM dst has been pruned
2236          * like this, we want to force a new route lookup.
2237          */
2238         if (dst->obsolete < 0 && !stale_bundle(dst))
2239                 return dst;
2240
2241         return NULL;
2242 }
2243
2244 static int stale_bundle(struct dst_entry *dst)
2245 {
2246         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2247 }
2248
2249 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2250 {
2251         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2252                 dst->dev = dev_net(dev)->loopback_dev;
2253                 dev_hold(dst->dev);
2254                 dev_put(dev);
2255         }
2256 }
2257 EXPORT_SYMBOL(xfrm_dst_ifdown);
2258
2259 static void xfrm_link_failure(struct sk_buff *skb)
2260 {
2261         /* Impossible. Such dst must be popped before reaches point of failure. */
2262 }
2263
2264 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2265 {
2266         if (dst) {
2267                 if (dst->obsolete) {
2268                         dst_release(dst);
2269                         dst = NULL;
2270                 }
2271         }
2272         return dst;
2273 }
2274
2275 static void __xfrm_garbage_collect(struct net *net)
2276 {
2277         struct dst_entry *head, *next;
2278
2279         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2280         head = xfrm_policy_sk_bundles;
2281         xfrm_policy_sk_bundles = NULL;
2282         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2283
2284         while (head) {
2285                 next = head->next;
2286                 dst_free(head);
2287                 head = next;
2288         }
2289 }
2290
2291 static void xfrm_garbage_collect(struct net *net)
2292 {
2293         flow_cache_flush();
2294         __xfrm_garbage_collect(net);
2295 }
2296
2297 static void xfrm_garbage_collect_deferred(struct net *net)
2298 {
2299         flow_cache_flush_deferred();
2300         __xfrm_garbage_collect(net);
2301 }
2302
2303 static void xfrm_init_pmtu(struct dst_entry *dst)
2304 {
2305         do {
2306                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2307                 u32 pmtu, route_mtu_cached;
2308
2309                 pmtu = dst_mtu(dst->child);
2310                 xdst->child_mtu_cached = pmtu;
2311
2312                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2313
2314                 route_mtu_cached = dst_mtu(xdst->route);
2315                 xdst->route_mtu_cached = route_mtu_cached;
2316
2317                 if (pmtu > route_mtu_cached)
2318                         pmtu = route_mtu_cached;
2319
2320                 dst_metric_set(dst, RTAX_MTU, pmtu);
2321         } while ((dst = dst->next));
2322 }
2323
2324 /* Check that the bundle accepts the flow and its components are
2325  * still valid.
2326  */
2327
2328 static int xfrm_bundle_ok(struct xfrm_dst *first)
2329 {
2330         struct dst_entry *dst = &first->u.dst;
2331         struct xfrm_dst *last;
2332         u32 mtu;
2333
2334         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2335             (dst->dev && !netif_running(dst->dev)))
2336                 return 0;
2337
2338         last = NULL;
2339
2340         do {
2341                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2342
2343                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2344                         return 0;
2345                 if (xdst->xfrm_genid != dst->xfrm->genid)
2346                         return 0;
2347                 if (xdst->num_pols > 0 &&
2348                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2349                         return 0;
2350
2351                 mtu = dst_mtu(dst->child);
2352                 if (xdst->child_mtu_cached != mtu) {
2353                         last = xdst;
2354                         xdst->child_mtu_cached = mtu;
2355                 }
2356
2357                 if (!dst_check(xdst->route, xdst->route_cookie))
2358                         return 0;
2359                 mtu = dst_mtu(xdst->route);
2360                 if (xdst->route_mtu_cached != mtu) {
2361                         last = xdst;
2362                         xdst->route_mtu_cached = mtu;
2363                 }
2364
2365                 dst = dst->child;
2366         } while (dst->xfrm);
2367
2368         if (likely(!last))
2369                 return 1;
2370
2371         mtu = last->child_mtu_cached;
2372         for (;;) {
2373                 dst = &last->u.dst;
2374
2375                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2376                 if (mtu > last->route_mtu_cached)
2377                         mtu = last->route_mtu_cached;
2378                 dst_metric_set(dst, RTAX_MTU, mtu);
2379
2380                 if (last == first)
2381                         break;
2382
2383                 last = (struct xfrm_dst *)last->u.dst.next;
2384                 last->child_mtu_cached = mtu;
2385         }
2386
2387         return 1;
2388 }
2389
2390 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2391 {
2392         return dst_metric_advmss(dst->path);
2393 }
2394
2395 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2396 {
2397         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2398
2399         return mtu ? : dst_mtu(dst->path);
2400 }
2401
2402 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, const void *daddr)
2403 {
2404         return dst_neigh_lookup(dst->path, daddr);
2405 }
2406
2407 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2408 {
2409         struct net *net;
2410         int err = 0;
2411         if (unlikely(afinfo == NULL))
2412                 return -EINVAL;
2413         if (unlikely(afinfo->family >= NPROTO))
2414                 return -EAFNOSUPPORT;
2415         write_lock_bh(&xfrm_policy_afinfo_lock);
2416         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2417                 err = -ENOBUFS;
2418         else {
2419                 struct dst_ops *dst_ops = afinfo->dst_ops;
2420                 if (likely(dst_ops->kmem_cachep == NULL))
2421                         dst_ops->kmem_cachep = xfrm_dst_cache;
2422                 if (likely(dst_ops->check == NULL))
2423                         dst_ops->check = xfrm_dst_check;
2424                 if (likely(dst_ops->default_advmss == NULL))
2425                         dst_ops->default_advmss = xfrm_default_advmss;
2426                 if (likely(dst_ops->mtu == NULL))
2427                         dst_ops->mtu = xfrm_mtu;
2428                 if (likely(dst_ops->negative_advice == NULL))
2429                         dst_ops->negative_advice = xfrm_negative_advice;
2430                 if (likely(dst_ops->link_failure == NULL))
2431                         dst_ops->link_failure = xfrm_link_failure;
2432                 if (likely(dst_ops->neigh_lookup == NULL))
2433                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2434                 if (likely(afinfo->garbage_collect == NULL))
2435                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2436                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2437         }
2438         write_unlock_bh(&xfrm_policy_afinfo_lock);
2439
2440         rtnl_lock();
2441         for_each_net(net) {
2442                 struct dst_ops *xfrm_dst_ops;
2443
2444                 switch (afinfo->family) {
2445                 case AF_INET:
2446                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2447                         break;
2448 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2449                 case AF_INET6:
2450                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2451                         break;
2452 #endif
2453                 default:
2454                         BUG();
2455                 }
2456                 *xfrm_dst_ops = *afinfo->dst_ops;
2457         }
2458         rtnl_unlock();
2459
2460         return err;
2461 }
2462 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2463
2464 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2465 {
2466         int err = 0;
2467         if (unlikely(afinfo == NULL))
2468                 return -EINVAL;
2469         if (unlikely(afinfo->family >= NPROTO))
2470                 return -EAFNOSUPPORT;
2471         write_lock_bh(&xfrm_policy_afinfo_lock);
2472         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2473                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2474                         err = -EINVAL;
2475                 else {
2476                         struct dst_ops *dst_ops = afinfo->dst_ops;
2477                         xfrm_policy_afinfo[afinfo->family] = NULL;
2478                         dst_ops->kmem_cachep = NULL;
2479                         dst_ops->check = NULL;
2480                         dst_ops->negative_advice = NULL;
2481                         dst_ops->link_failure = NULL;
2482                         afinfo->garbage_collect = NULL;
2483                 }
2484         }
2485         write_unlock_bh(&xfrm_policy_afinfo_lock);
2486         return err;
2487 }
2488 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2489
2490 static void __net_init xfrm_dst_ops_init(struct net *net)
2491 {
2492         struct xfrm_policy_afinfo *afinfo;
2493
2494         read_lock_bh(&xfrm_policy_afinfo_lock);
2495         afinfo = xfrm_policy_afinfo[AF_INET];
2496         if (afinfo)
2497                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2498 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2499         afinfo = xfrm_policy_afinfo[AF_INET6];
2500         if (afinfo)
2501                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2502 #endif
2503         read_unlock_bh(&xfrm_policy_afinfo_lock);
2504 }
2505
2506 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2507 {
2508         struct xfrm_policy_afinfo *afinfo;
2509         if (unlikely(family >= NPROTO))
2510                 return NULL;
2511         read_lock(&xfrm_policy_afinfo_lock);
2512         afinfo = xfrm_policy_afinfo[family];
2513         if (unlikely(!afinfo))
2514                 read_unlock(&xfrm_policy_afinfo_lock);
2515         return afinfo;
2516 }
2517
2518 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2519 {
2520         read_unlock(&xfrm_policy_afinfo_lock);
2521 }
2522
2523 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2524 {
2525         struct net_device *dev = ptr;
2526
2527         switch (event) {
2528         case NETDEV_DOWN:
2529                 xfrm_garbage_collect(dev_net(dev));
2530         }
2531         return NOTIFY_DONE;
2532 }
2533
2534 static struct notifier_block xfrm_dev_notifier = {
2535         .notifier_call  = xfrm_dev_event,
2536 };
2537
2538 #ifdef CONFIG_XFRM_STATISTICS
2539 static int __net_init xfrm_statistics_init(struct net *net)
2540 {
2541         int rv;
2542
2543         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2544                           sizeof(struct linux_xfrm_mib),
2545                           __alignof__(struct linux_xfrm_mib)) < 0)
2546                 return -ENOMEM;
2547         rv = xfrm_proc_init(net);
2548         if (rv < 0)
2549                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2550         return rv;
2551 }
2552
2553 static void xfrm_statistics_fini(struct net *net)
2554 {
2555         xfrm_proc_fini(net);
2556         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2557 }
2558 #else
2559 static int __net_init xfrm_statistics_init(struct net *net)
2560 {
2561         return 0;
2562 }
2563
2564 static void xfrm_statistics_fini(struct net *net)
2565 {
2566 }
2567 #endif
2568
2569 static int __net_init xfrm_policy_init(struct net *net)
2570 {
2571         unsigned int hmask, sz;
2572         int dir;
2573
2574         if (net_eq(net, &init_net))
2575                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2576                                            sizeof(struct xfrm_dst),
2577                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2578                                            NULL);
2579
2580         hmask = 8 - 1;
2581         sz = (hmask+1) * sizeof(struct hlist_head);
2582
2583         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2584         if (!net->xfrm.policy_byidx)
2585                 goto out_byidx;
2586         net->xfrm.policy_idx_hmask = hmask;
2587
2588         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2589                 struct xfrm_policy_hash *htab;
2590
2591                 net->xfrm.policy_count[dir] = 0;
2592                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2593
2594                 htab = &net->xfrm.policy_bydst[dir];
2595                 htab->table = xfrm_hash_alloc(sz);
2596                 if (!htab->table)
2597                         goto out_bydst;
2598                 htab->hmask = hmask;
2599         }
2600
2601         INIT_LIST_HEAD(&net->xfrm.policy_all);
2602         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2603         if (net_eq(net, &init_net))
2604                 register_netdevice_notifier(&xfrm_dev_notifier);
2605         return 0;
2606
2607 out_bydst:
2608         for (dir--; dir >= 0; dir--) {
2609                 struct xfrm_policy_hash *htab;
2610
2611                 htab = &net->xfrm.policy_bydst[dir];
2612                 xfrm_hash_free(htab->table, sz);
2613         }
2614         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2615 out_byidx:
2616         return -ENOMEM;
2617 }
2618
2619 static void xfrm_policy_fini(struct net *net)
2620 {
2621         struct xfrm_audit audit_info;
2622         unsigned int sz;
2623         int dir;
2624
2625         flush_work(&net->xfrm.policy_hash_work);
2626 #ifdef CONFIG_XFRM_SUB_POLICY
2627         audit_info.loginuid = -1;
2628         audit_info.sessionid = -1;
2629         audit_info.secid = 0;
2630         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2631 #endif
2632         audit_info.loginuid = -1;
2633         audit_info.sessionid = -1;
2634         audit_info.secid = 0;
2635         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2636
2637         WARN_ON(!list_empty(&net->xfrm.policy_all));
2638
2639         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2640                 struct xfrm_policy_hash *htab;
2641
2642                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2643
2644                 htab = &net->xfrm.policy_bydst[dir];
2645                 sz = (htab->hmask + 1);
2646                 WARN_ON(!hlist_empty(htab->table));
2647                 xfrm_hash_free(htab->table, sz);
2648         }
2649
2650         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2651         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2652         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2653 }
2654
2655 static int __net_init xfrm_net_init(struct net *net)
2656 {
2657         int rv;
2658
2659         rv = xfrm_statistics_init(net);
2660         if (rv < 0)
2661                 goto out_statistics;
2662         rv = xfrm_state_init(net);
2663         if (rv < 0)
2664                 goto out_state;
2665         rv = xfrm_policy_init(net);
2666         if (rv < 0)
2667                 goto out_policy;
2668         xfrm_dst_ops_init(net);
2669         rv = xfrm_sysctl_init(net);
2670         if (rv < 0)
2671                 goto out_sysctl;
2672         return 0;
2673
2674 out_sysctl:
2675         xfrm_policy_fini(net);
2676 out_policy:
2677         xfrm_state_fini(net);
2678 out_state:
2679         xfrm_statistics_fini(net);
2680 out_statistics:
2681         return rv;
2682 }
2683
2684 static void __net_exit xfrm_net_exit(struct net *net)
2685 {
2686         xfrm_sysctl_fini(net);
2687         xfrm_policy_fini(net);
2688         xfrm_state_fini(net);
2689         xfrm_statistics_fini(net);
2690 }
2691
2692 static struct pernet_operations __net_initdata xfrm_net_ops = {
2693         .init = xfrm_net_init,
2694         .exit = xfrm_net_exit,
2695 };
2696
2697 void __init xfrm_init(void)
2698 {
2699         register_pernet_subsys(&xfrm_net_ops);
2700         xfrm_input_init();
2701 }
2702
2703 #ifdef CONFIG_AUDITSYSCALL
2704 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2705                                          struct audit_buffer *audit_buf)
2706 {
2707         struct xfrm_sec_ctx *ctx = xp->security;
2708         struct xfrm_selector *sel = &xp->selector;
2709
2710         if (ctx)
2711                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2712                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2713
2714         switch(sel->family) {
2715         case AF_INET:
2716                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2717                 if (sel->prefixlen_s != 32)
2718                         audit_log_format(audit_buf, " src_prefixlen=%d",
2719                                          sel->prefixlen_s);
2720                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2721                 if (sel->prefixlen_d != 32)
2722                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2723                                          sel->prefixlen_d);
2724                 break;
2725         case AF_INET6:
2726                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2727                 if (sel->prefixlen_s != 128)
2728                         audit_log_format(audit_buf, " src_prefixlen=%d",
2729                                          sel->prefixlen_s);
2730                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2731                 if (sel->prefixlen_d != 128)
2732                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2733                                          sel->prefixlen_d);
2734                 break;
2735         }
2736 }
2737
2738 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2739                            uid_t auid, u32 sessionid, u32 secid)
2740 {
2741         struct audit_buffer *audit_buf;
2742
2743         audit_buf = xfrm_audit_start("SPD-add");
2744         if (audit_buf == NULL)
2745                 return;
2746         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2747         audit_log_format(audit_buf, " res=%u", result);
2748         xfrm_audit_common_policyinfo(xp, audit_buf);
2749         audit_log_end(audit_buf);
2750 }
2751 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2752
2753 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2754                               uid_t auid, u32 sessionid, u32 secid)
2755 {
2756         struct audit_buffer *audit_buf;
2757
2758         audit_buf = xfrm_audit_start("SPD-delete");
2759         if (audit_buf == NULL)
2760                 return;
2761         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2762         audit_log_format(audit_buf, " res=%u", result);
2763         xfrm_audit_common_policyinfo(xp, audit_buf);
2764         audit_log_end(audit_buf);
2765 }
2766 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2767 #endif
2768
2769 #ifdef CONFIG_XFRM_MIGRATE
2770 static int xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
2771                                        const struct xfrm_selector *sel_tgt)
2772 {
2773         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2774                 if (sel_tgt->family == sel_cmp->family &&
2775                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2776                                   sel_cmp->family) == 0 &&
2777                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2778                                   sel_cmp->family) == 0 &&
2779                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2780                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2781                         return 1;
2782                 }
2783         } else {
2784                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2785                         return 1;
2786                 }
2787         }
2788         return 0;
2789 }
2790
2791 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
2792                                                      u8 dir, u8 type)
2793 {
2794         struct xfrm_policy *pol, *ret = NULL;
2795         struct hlist_node *entry;
2796         struct hlist_head *chain;
2797         u32 priority = ~0U;
2798
2799         read_lock_bh(&xfrm_policy_lock);
2800         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2801         hlist_for_each_entry(pol, entry, chain, bydst) {
2802                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2803                     pol->type == type) {
2804                         ret = pol;
2805                         priority = ret->priority;
2806                         break;
2807                 }
2808         }
2809         chain = &init_net.xfrm.policy_inexact[dir];
2810         hlist_for_each_entry(pol, entry, chain, bydst) {
2811                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2812                     pol->type == type &&
2813                     pol->priority < priority) {
2814                         ret = pol;
2815                         break;
2816                 }
2817         }
2818
2819         if (ret)
2820                 xfrm_pol_hold(ret);
2821
2822         read_unlock_bh(&xfrm_policy_lock);
2823
2824         return ret;
2825 }
2826
2827 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2828 {
2829         int match = 0;
2830
2831         if (t->mode == m->mode && t->id.proto == m->proto &&
2832             (m->reqid == 0 || t->reqid == m->reqid)) {
2833                 switch (t->mode) {
2834                 case XFRM_MODE_TUNNEL:
2835                 case XFRM_MODE_BEET:
2836                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2837                                           m->old_family) == 0 &&
2838                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2839                                           m->old_family) == 0) {
2840                                 match = 1;
2841                         }
2842                         break;
2843                 case XFRM_MODE_TRANSPORT:
2844                         /* in case of transport mode, template does not store
2845                            any IP addresses, hence we just compare mode and
2846                            protocol */
2847                         match = 1;
2848                         break;
2849                 default:
2850                         break;
2851                 }
2852         }
2853         return match;
2854 }
2855
2856 /* update endpoint address(es) of template(s) */
2857 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2858                                struct xfrm_migrate *m, int num_migrate)
2859 {
2860         struct xfrm_migrate *mp;
2861         int i, j, n = 0;
2862
2863         write_lock_bh(&pol->lock);
2864         if (unlikely(pol->walk.dead)) {
2865                 /* target policy has been deleted */
2866                 write_unlock_bh(&pol->lock);
2867                 return -ENOENT;
2868         }
2869
2870         for (i = 0; i < pol->xfrm_nr; i++) {
2871                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2872                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2873                                 continue;
2874                         n++;
2875                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2876                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2877                                 continue;
2878                         /* update endpoints */
2879                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2880                                sizeof(pol->xfrm_vec[i].id.daddr));
2881                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2882                                sizeof(pol->xfrm_vec[i].saddr));
2883                         pol->xfrm_vec[i].encap_family = mp->new_family;
2884                         /* flush bundles */
2885                         atomic_inc(&pol->genid);
2886                 }
2887         }
2888
2889         write_unlock_bh(&pol->lock);
2890
2891         if (!n)
2892                 return -ENODATA;
2893
2894         return 0;
2895 }
2896
2897 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
2898 {
2899         int i, j;
2900
2901         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2902                 return -EINVAL;
2903
2904         for (i = 0; i < num_migrate; i++) {
2905                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2906                                    m[i].old_family) == 0) &&
2907                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2908                                    m[i].old_family) == 0))
2909                         return -EINVAL;
2910                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2911                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2912                         return -EINVAL;
2913
2914                 /* check if there is any duplicated entry */
2915                 for (j = i + 1; j < num_migrate; j++) {
2916                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2917                                     sizeof(m[i].old_daddr)) &&
2918                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2919                                     sizeof(m[i].old_saddr)) &&
2920                             m[i].proto == m[j].proto &&
2921                             m[i].mode == m[j].mode &&
2922                             m[i].reqid == m[j].reqid &&
2923                             m[i].old_family == m[j].old_family)
2924                                 return -EINVAL;
2925                 }
2926         }
2927
2928         return 0;
2929 }
2930
2931 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
2932                  struct xfrm_migrate *m, int num_migrate,
2933                  struct xfrm_kmaddress *k)
2934 {
2935         int i, err, nx_cur = 0, nx_new = 0;
2936         struct xfrm_policy *pol = NULL;
2937         struct xfrm_state *x, *xc;
2938         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2939         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2940         struct xfrm_migrate *mp;
2941
2942         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2943                 goto out;
2944
2945         /* Stage 1 - find policy */
2946         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2947                 err = -ENOENT;
2948                 goto out;
2949         }
2950
2951         /* Stage 2 - find and update state(s) */
2952         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2953                 if ((x = xfrm_migrate_state_find(mp))) {
2954                         x_cur[nx_cur] = x;
2955                         nx_cur++;
2956                         if ((xc = xfrm_state_migrate(x, mp))) {
2957                                 x_new[nx_new] = xc;
2958                                 nx_new++;
2959                         } else {
2960                                 err = -ENODATA;
2961                                 goto restore_state;
2962                         }
2963                 }
2964         }
2965
2966         /* Stage 3 - update policy */
2967         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2968                 goto restore_state;
2969
2970         /* Stage 4 - delete old state(s) */
2971         if (nx_cur) {
2972                 xfrm_states_put(x_cur, nx_cur);
2973                 xfrm_states_delete(x_cur, nx_cur);
2974         }
2975
2976         /* Stage 5 - announce */
2977         km_migrate(sel, dir, type, m, num_migrate, k);
2978
2979         xfrm_pol_put(pol);
2980
2981         return 0;
2982 out:
2983         return err;
2984
2985 restore_state:
2986         if (pol)
2987                 xfrm_pol_put(pol);
2988         if (nx_cur)
2989                 xfrm_states_put(x_cur, nx_cur);
2990         if (nx_new)
2991                 xfrm_states_delete(x_new, nx_new);
2992
2993         return err;
2994 }
2995 EXPORT_SYMBOL(xfrm_migrate);
2996 #endif